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 <scsi/fc/fc_fcoe.h>
  32 #include <net/udp_tunnel.h>
  33 #include <net/pkt_cls.h>
  34 #include <net/tc_act/tc_gact.h>
  35 #include <net/tc_act/tc_mirred.h>
  36 #include <net/vxlan.h>
  37 #include <net/mpls.h>
  38 #include <net/xdp_sock.h>
  39
  40 #include "ixgbe.h"
  41 #include "ixgbe_common.h"
  42 #include "ixgbe_dcb_82599.h"
  43 #include "ixgbe_phy.h"
  44 #include "ixgbe_sriov.h"
  45 #include "ixgbe_model.h"
  46 #include "ixgbe_txrx_common.h"
  47
  48 char ixgbe_driver_name[] = "ixgbe";
  49 static const char ixgbe_driver_string[] =
  50                               "Intel(R) 10 Gigabit PCI Express Network Driver";
  51 #ifdef IXGBE_FCOE
  52 char ixgbe_default_device_descr[] =
  53                               "Intel(R) 10 Gigabit Network Connection";
  54 #else
  55 static char ixgbe_default_device_descr[] =
  56                               "Intel(R) 10 Gigabit Network Connection";
  57 #endif
  58 #define DRV_VERSION "5.1.0-k"
  59 const char ixgbe_driver_version[] = DRV_VERSION;
  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 MODULE_VERSION(DRV_VERSION);
 168
 169 static struct workqueue_struct *ixgbe_wq;
 170
 171 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
 172 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *);
 173
 174 static const struct net_device_ops ixgbe_netdev_ops;
 175
 176 static bool netif_is_ixgbe(struct net_device *dev)
 177 {
 178         return dev && (dev->netdev_ops == &ixgbe_netdev_ops);
 179 }
 180
 181 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
 182                                           u32 reg, u16 *value)
 183 {
 184         struct pci_dev *parent_dev;
 185         struct pci_bus *parent_bus;
 186
 187         parent_bus = adapter->pdev->bus->parent;
 188         if (!parent_bus)
 189                 return -1;
 190
 191         parent_dev = parent_bus->self;
 192         if (!parent_dev)
 193                 return -1;
 194
 195         if (!pci_is_pcie(parent_dev))
 196                 return -1;
 197
 198         pcie_capability_read_word(parent_dev, reg, value);
 199         if (*value == IXGBE_FAILED_READ_CFG_WORD &&
 200             ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
 201                 return -1;
 202         return 0;
 203 }
 204
 205 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
 206 {
 207         struct ixgbe_hw *hw = &adapter->hw;
 208         u16 link_status = 0;
 209         int err;
 210
 211         hw->bus.type = ixgbe_bus_type_pci_express;
 212
 213         /* Get the negotiated link width and speed from PCI config space of the
 214          * parent, as this device is behind a switch
 215          */
 216         err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
 217
 218         /* assume caller will handle error case */
 219         if (err)
 220                 return err;
 221
 222         hw->bus.width = ixgbe_convert_bus_width(link_status);
 223         hw->bus.speed = ixgbe_convert_bus_speed(link_status);
 224
 225         return 0;
 226 }
 227
 228 /**
 229  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
 230  * @hw: hw specific details
 231  *
 232  * This function is used by probe to determine whether a device's PCI-Express
 233  * bandwidth details should be gathered from the parent bus instead of from the
 234  * device. Used to ensure that various locations all have the correct device ID
 235  * checks.
 236  */
 237 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
 238 {
 239         switch (hw->device_id) {
 240         case IXGBE_DEV_ID_82599_SFP_SF_QP:
 241         case IXGBE_DEV_ID_82599_QSFP_SF_QP:
 242                 return true;
 243         default:
 244                 return false;
 245         }
 246 }
 247
 248 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
 249                                      int expected_gts)
 250 {
 251         struct ixgbe_hw *hw = &adapter->hw;
 252         struct pci_dev *pdev;
 253
 254         /* Some devices are not connected over PCIe and thus do not negotiate
 255          * speed. These devices do not have valid bus info, and thus any report
 256          * we generate may not be correct.
 257          */
 258         if (hw->bus.type == ixgbe_bus_type_internal)
 259                 return;
 260
 261         /* determine whether to use the parent device */
 262         if (ixgbe_pcie_from_parent(&adapter->hw))
 263                 pdev = adapter->pdev->bus->parent->self;
 264         else
 265                 pdev = adapter->pdev;
 266
 267         pcie_print_link_status(pdev);
 268 }
 269
 270 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
 271 {
 272         if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
 273             !test_bit(__IXGBE_REMOVING, &adapter->state) &&
 274             !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
 275                 queue_work(ixgbe_wq, &adapter->service_task);
 276 }
 277
 278 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
 279 {
 280         struct ixgbe_adapter *adapter = hw->back;
 281
 282         if (!hw->hw_addr)
 283                 return;
 284         hw->hw_addr = NULL;
 285         e_dev_err("Adapter removed\n");
 286         if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
 287                 ixgbe_service_event_schedule(adapter);
 288 }
 289
 290 static u32 ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
 291 {
 292         u8 __iomem *reg_addr;
 293         u32 value;
 294         int i;
 295
 296         reg_addr = READ_ONCE(hw->hw_addr);
 297         if (ixgbe_removed(reg_addr))
 298                 return IXGBE_FAILED_READ_REG;
 299
 300         /* Register read of 0xFFFFFFF can indicate the adapter has been removed,
 301          * so perform several status register reads to determine if the adapter
 302          * has been removed.
 303          */
 304         for (i = 0; i < IXGBE_FAILED_READ_RETRIES; i++) {
 305                 value = readl(reg_addr + IXGBE_STATUS);
 306                 if (value != IXGBE_FAILED_READ_REG)
 307                         break;
 308                 mdelay(3);
 309         }
 310
 311         if (value == IXGBE_FAILED_READ_REG)
 312                 ixgbe_remove_adapter(hw);
 313         else
 314                 value = readl(reg_addr + reg);
 315         return value;
 316 }
 317
 318 /**
 319  * ixgbe_read_reg - Read from device register
 320  * @hw: hw specific details
 321  * @reg: offset of register to read
 322  *
 323  * Returns : value read or IXGBE_FAILED_READ_REG if removed
 324  *
 325  * This function is used to read device registers. It checks for device
 326  * removal by confirming any read that returns all ones by checking the
 327  * status register value for all ones. This function avoids reading from
 328  * the hardware if a removal was previously detected in which case it
 329  * returns IXGBE_FAILED_READ_REG (all ones).
 330  */
 331 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
 332 {
 333         u8 __iomem *reg_addr = READ_ONCE(hw->hw_addr);
 334         u32 value;
 335
 336         if (ixgbe_removed(reg_addr))
 337                 return IXGBE_FAILED_READ_REG;
 338         if (unlikely(hw->phy.nw_mng_if_sel &
 339                      IXGBE_NW_MNG_IF_SEL_SGMII_ENABLE)) {
 340                 struct ixgbe_adapter *adapter;
 341                 int i;
 342
 343                 for (i = 0; i < 200; ++i) {
 344                         value = readl(reg_addr + IXGBE_MAC_SGMII_BUSY);
 345                         if (likely(!value))
 346                                 goto writes_completed;
 347                         if (value == IXGBE_FAILED_READ_REG) {
 348                                 ixgbe_remove_adapter(hw);
 349                                 return IXGBE_FAILED_READ_REG;
 350                         }
 351                         udelay(5);
 352                 }
 353
 354                 adapter = hw->back;
 355                 e_warn(hw, "register writes incomplete %08x\n", value);
 356         }
 357
 358 writes_completed:
 359         value = readl(reg_addr + reg);
 360         if (unlikely(value == IXGBE_FAILED_READ_REG))
 361                 value = ixgbe_check_remove(hw, reg);
 362         return value;
 363 }
 364
 365 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
 366 {
 367         u16 value;
 368
 369         pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
 370         if (value == IXGBE_FAILED_READ_CFG_WORD) {
 371                 ixgbe_remove_adapter(hw);
 372                 return true;
 373         }
 374         return false;
 375 }
 376
 377 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
 378 {
 379         struct ixgbe_adapter *adapter = hw->back;
 380         u16 value;
 381
 382         if (ixgbe_removed(hw->hw_addr))
 383                 return IXGBE_FAILED_READ_CFG_WORD;
 384         pci_read_config_word(adapter->pdev, reg, &value);
 385         if (value == IXGBE_FAILED_READ_CFG_WORD &&
 386             ixgbe_check_cfg_remove(hw, adapter->pdev))
 387                 return IXGBE_FAILED_READ_CFG_WORD;
 388         return value;
 389 }
 390
 391 #ifdef CONFIG_PCI_IOV
 392 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
 393 {
 394         struct ixgbe_adapter *adapter = hw->back;
 395         u32 value;
 396
 397         if (ixgbe_removed(hw->hw_addr))
 398                 return IXGBE_FAILED_READ_CFG_DWORD;
 399         pci_read_config_dword(adapter->pdev, reg, &value);
 400         if (value == IXGBE_FAILED_READ_CFG_DWORD &&
 401             ixgbe_check_cfg_remove(hw, adapter->pdev))
 402                 return IXGBE_FAILED_READ_CFG_DWORD;
 403         return value;
 404 }
 405 #endif /* CONFIG_PCI_IOV */
 406
 407 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
 408 {
 409         struct ixgbe_adapter *adapter = hw->back;
 410
 411         if (ixgbe_removed(hw->hw_addr))
 412                 return;
 413         pci_write_config_word(adapter->pdev, reg, value);
 414 }
 415
 416 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
 417 {
 418         BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
 419
 420         /* flush memory to make sure state is correct before next watchdog */
 421         smp_mb__before_atomic();
 422         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
 423 }
 424
 425 struct ixgbe_reg_info {
 426         u32 ofs;
 427         char *name;
 428 };
 429
 430 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
 431
 432         /* General Registers */
 433         {IXGBE_CTRL, "CTRL"},
 434         {IXGBE_STATUS, "STATUS"},
 435         {IXGBE_CTRL_EXT, "CTRL_EXT"},
 436
 437         /* Interrupt Registers */
 438         {IXGBE_EICR, "EICR"},
 439
 440         /* RX Registers */
 441         {IXGBE_SRRCTL(0), "SRRCTL"},
 442         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
 443         {IXGBE_RDLEN(0), "RDLEN"},
 444         {IXGBE_RDH(0), "RDH"},
 445         {IXGBE_RDT(0), "RDT"},
 446         {IXGBE_RXDCTL(0), "RXDCTL"},
 447         {IXGBE_RDBAL(0), "RDBAL"},
 448         {IXGBE_RDBAH(0), "RDBAH"},
 449
 450         /* TX Registers */
 451         {IXGBE_TDBAL(0), "TDBAL"},
 452         {IXGBE_TDBAH(0), "TDBAH"},
 453         {IXGBE_TDLEN(0), "TDLEN"},
 454         {IXGBE_TDH(0), "TDH"},
 455         {IXGBE_TDT(0), "TDT"},
 456         {IXGBE_TXDCTL(0), "TXDCTL"},
 457
 458         /* List Terminator */
 459         { .name = NULL }
 460 };
 461
 462
 463 /*
 464  * ixgbe_regdump - register printout routine
 465  */
 466 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
 467 {
 468         int i;
 469         char rname[16];
 470         u32 regs[64];
 471
 472         switch (reginfo->ofs) {
 473         case IXGBE_SRRCTL(0):
 474                 for (i = 0; i < 64; i++)
 475                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
 476                 break;
 477         case IXGBE_DCA_RXCTRL(0):
 478                 for (i = 0; i < 64; i++)
 479                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
 480                 break;
 481         case IXGBE_RDLEN(0):
 482                 for (i = 0; i < 64; i++)
 483                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
 484                 break;
 485         case IXGBE_RDH(0):
 486                 for (i = 0; i < 64; i++)
 487                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
 488                 break;
 489         case IXGBE_RDT(0):
 490                 for (i = 0; i < 64; i++)
 491                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
 492                 break;
 493         case IXGBE_RXDCTL(0):
 494                 for (i = 0; i < 64; i++)
 495                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
 496                 break;
 497         case IXGBE_RDBAL(0):
 498                 for (i = 0; i < 64; i++)
 499                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
 500                 break;
 501         case IXGBE_RDBAH(0):
 502                 for (i = 0; i < 64; i++)
 503                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
 504                 break;
 505         case IXGBE_TDBAL(0):
 506                 for (i = 0; i < 64; i++)
 507                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
 508                 break;
 509         case IXGBE_TDBAH(0):
 510                 for (i = 0; i < 64; i++)
 511                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
 512                 break;
 513         case IXGBE_TDLEN(0):
 514                 for (i = 0; i < 64; i++)
 515                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
 516                 break;
 517         case IXGBE_TDH(0):
 518                 for (i = 0; i < 64; i++)
 519                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
 520                 break;
 521         case IXGBE_TDT(0):
 522                 for (i = 0; i < 64; i++)
 523                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
 524                 break;
 525         case IXGBE_TXDCTL(0):
 526                 for (i = 0; i < 64; i++)
 527                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
 528                 break;
 529         default:
 530                 pr_info("%-15s %08x\n",
 531                         reginfo->name, IXGBE_READ_REG(hw, reginfo->ofs));
 532                 return;
 533         }
 534
 535         i = 0;
 536         while (i < 64) {
 537                 int j;
 538                 char buf[9 * 8 + 1];
 539                 char *p = buf;
 540
 541                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i, i + 7);
 542                 for (j = 0; j < 8; j++)
 543                         p += sprintf(p, " %08x", regs[i++]);
 544                 pr_err("%-15s%s\n", rname, buf);
 545         }
 546
 547 }
 548
 549 static void ixgbe_print_buffer(struct ixgbe_ring *ring, int n)
 550 {
 551         struct ixgbe_tx_buffer *tx_buffer;
 552
 553         tx_buffer = &ring->tx_buffer_info[ring->next_to_clean];
 554         pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
 555                 n, ring->next_to_use, ring->next_to_clean,
 556                 (u64)dma_unmap_addr(tx_buffer, dma),
 557                 dma_unmap_len(tx_buffer, len),
 558                 tx_buffer->next_to_watch,
 559                 (u64)tx_buffer->time_stamp);
 560 }
 561
 562 /*
 563  * ixgbe_dump - Print registers, tx-rings and rx-rings
 564  */
 565 static void ixgbe_dump(struct ixgbe_adapter *adapter)
 566 {
 567         struct net_device *netdev = adapter->netdev;
 568         struct ixgbe_hw *hw = &adapter->hw;
 569         struct ixgbe_reg_info *reginfo;
 570         int n = 0;
 571         struct ixgbe_ring *ring;
 572         struct ixgbe_tx_buffer *tx_buffer;
 573         union ixgbe_adv_tx_desc *tx_desc;
 574         struct my_u0 { u64 a; u64 b; } *u0;
 575         struct ixgbe_ring *rx_ring;
 576         union ixgbe_adv_rx_desc *rx_desc;
 577         struct ixgbe_rx_buffer *rx_buffer_info;
 578         int i = 0;
 579
 580         if (!netif_msg_hw(adapter))
 581                 return;
 582
 583         /* Print netdevice Info */
 584         if (netdev) {
 585                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 586                 pr_info("Device Name     state            "
 587                         "trans_start\n");
 588                 pr_info("%-15s %016lX %016lX\n",
 589                         netdev->name,
 590                         netdev->state,
 591                         dev_trans_start(netdev));
 592         }
 593
 594         /* Print Registers */
 595         dev_info(&adapter->pdev->dev, "Register Dump\n");
 596         pr_info(" Register Name   Value\n");
 597         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
 598              reginfo->name; reginfo++) {
 599                 ixgbe_regdump(hw, reginfo);
 600         }
 601
 602         /* Print TX Ring Summary */
 603         if (!netdev || !netif_running(netdev))
 604                 return;
 605
 606         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 607         pr_info(" %s     %s              %s        %s\n",
 608                 "Queue [NTU] [NTC] [bi(ntc)->dma  ]",
 609                 "leng", "ntw", "timestamp");
 610         for (n = 0; n < adapter->num_tx_queues; n++) {
 611                 ring = adapter->tx_ring[n];
 612                 ixgbe_print_buffer(ring, n);
 613         }
 614
 615         for (n = 0; n < adapter->num_xdp_queues; n++) {
 616                 ring = adapter->xdp_ring[n];
 617                 ixgbe_print_buffer(ring, n);
 618         }
 619
 620         /* Print TX Rings */
 621         if (!netif_msg_tx_done(adapter))
 622                 goto rx_ring_summary;
 623
 624         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 625
 626         /* Transmit Descriptor Formats
 627          *
 628          * 82598 Advanced Transmit Descriptor
 629          *   +--------------------------------------------------------------+
 630          * 0 |         Buffer Address [63:0]                                |
 631          *   +--------------------------------------------------------------+
 632          * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 633          *   +--------------------------------------------------------------+
 634          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 635          *
 636          * 82598 Advanced Transmit Descriptor (Write-Back Format)
 637          *   +--------------------------------------------------------------+
 638          * 0 |                          RSV [63:0]                          |
 639          *   +--------------------------------------------------------------+
 640          * 8 |            RSV           |  STA  |          NXTSEQ           |
 641          *   +--------------------------------------------------------------+
 642          *   63                       36 35   32 31                         0
 643          *
 644          * 82599+ Advanced Transmit Descriptor
 645          *   +--------------------------------------------------------------+
 646          * 0 |         Buffer Address [63:0]                                |
 647          *   +--------------------------------------------------------------+
 648          * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
 649          *   +--------------------------------------------------------------+
 650          *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
 651          *
 652          * 82599+ Advanced Transmit Descriptor (Write-Back Format)
 653          *   +--------------------------------------------------------------+
 654          * 0 |                          RSV [63:0]                          |
 655          *   +--------------------------------------------------------------+
 656          * 8 |            RSV           |  STA  |           RSV             |
 657          *   +--------------------------------------------------------------+
 658          *   63                       36 35   32 31                         0
 659          */
 660
 661         for (n = 0; n < adapter->num_tx_queues; n++) {
 662                 ring = adapter->tx_ring[n];
 663                 pr_info("------------------------------------\n");
 664                 pr_info("TX QUEUE INDEX = %d\n", ring->queue_index);
 665                 pr_info("------------------------------------\n");
 666                 pr_info("%s%s    %s              %s        %s          %s\n",
 667                         "T [desc]     [address 63:0  ] ",
 668                         "[PlPOIdStDDt Ln] [bi->dma       ] ",
 669                         "leng", "ntw", "timestamp", "bi->skb");
 670
 671                 for (i = 0; ring->desc && (i < ring->count); i++) {
 672                         tx_desc = IXGBE_TX_DESC(ring, i);
 673                         tx_buffer = &ring->tx_buffer_info[i];
 674                         u0 = (struct my_u0 *)tx_desc;
 675                         if (dma_unmap_len(tx_buffer, len) > 0) {
 676                                 const char *ring_desc;
 677
 678                                 if (i == ring->next_to_use &&
 679                                     i == ring->next_to_clean)
 680                                         ring_desc = " NTC/U";
 681                                 else if (i == ring->next_to_use)
 682                                         ring_desc = " NTU";
 683                                 else if (i == ring->next_to_clean)
 684                                         ring_desc = " NTC";
 685                                 else
 686                                         ring_desc = "";
 687                                 pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p%s",
 688                                         i,
 689                                         le64_to_cpu((__force __le64)u0->a),
 690                                         le64_to_cpu((__force __le64)u0->b),
 691                                         (u64)dma_unmap_addr(tx_buffer, dma),
 692                                         dma_unmap_len(tx_buffer, len),
 693                                         tx_buffer->next_to_watch,
 694                                         (u64)tx_buffer->time_stamp,
 695                                         tx_buffer->skb,
 696                                         ring_desc);
 697
 698                                 if (netif_msg_pktdata(adapter) &&
 699                                     tx_buffer->skb)
 700                                         print_hex_dump(KERN_INFO, "",
 701                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 702                                                 tx_buffer->skb->data,
 703                                                 dma_unmap_len(tx_buffer, len),
 704                                                 true);
 705                         }
 706                 }
 707         }
 708
 709         /* Print RX Rings Summary */
 710 rx_ring_summary:
 711         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 712         pr_info("Queue [NTU] [NTC]\n");
 713         for (n = 0; n < adapter->num_rx_queues; n++) {
 714                 rx_ring = adapter->rx_ring[n];
 715                 pr_info("%5d %5X %5X\n",
 716                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 717         }
 718
 719         /* Print RX Rings */
 720         if (!netif_msg_rx_status(adapter))
 721                 return;
 722
 723         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 724
 725         /* Receive Descriptor Formats
 726          *
 727          * 82598 Advanced Receive Descriptor (Read) Format
 728          *    63                                           1        0
 729          *    +-----------------------------------------------------+
 730          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 731          *    +----------------------------------------------+------+
 732          *  8 |       Header Buffer Address [63:1]           |  DD  |
 733          *    +-----------------------------------------------------+
 734          *
 735          *
 736          * 82598 Advanced Receive Descriptor (Write-Back) Format
 737          *
 738          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 739          *   +------------------------------------------------------+
 740          * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
 741          *   | Packet   | IP     |   |          |     | Type | Type |
 742          *   | Checksum | Ident  |   |          |     |      |      |
 743          *   +------------------------------------------------------+
 744          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 745          *   +------------------------------------------------------+
 746          *   63       48 47    32 31            20 19               0
 747          *
 748          * 82599+ Advanced Receive Descriptor (Read) Format
 749          *    63                                           1        0
 750          *    +-----------------------------------------------------+
 751          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 752          *    +----------------------------------------------+------+
 753          *  8 |       Header Buffer Address [63:1]           |  DD  |
 754          *    +-----------------------------------------------------+
 755          *
 756          *
 757          * 82599+ Advanced Receive Descriptor (Write-Back) Format
 758          *
 759          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 760          *   +------------------------------------------------------+
 761          * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
 762          *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
 763          *   |/ Flow Dir Flt ID  |   |          |     |      |      |
 764          *   +------------------------------------------------------+
 765          * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
 766          *   +------------------------------------------------------+
 767          *   63       48 47    32 31          20 19                 0
 768          */
 769
 770         for (n = 0; n < adapter->num_rx_queues; n++) {
 771                 rx_ring = adapter->rx_ring[n];
 772                 pr_info("------------------------------------\n");
 773                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 774                 pr_info("------------------------------------\n");
 775                 pr_info("%s%s%s\n",
 776                         "R  [desc]      [ PktBuf     A0] ",
 777                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
 778                         "<-- Adv Rx Read format");
 779                 pr_info("%s%s%s\n",
 780                         "RWB[desc]      [PcsmIpSHl PtRs] ",
 781                         "[vl er S cks ln] ---------------- [bi->skb       ] ",
 782                         "<-- Adv Rx Write-Back format");
 783
 784                 for (i = 0; i < rx_ring->count; i++) {
 785                         const char *ring_desc;
 786
 787                         if (i == rx_ring->next_to_use)
 788                                 ring_desc = " NTU";
 789                         else if (i == rx_ring->next_to_clean)
 790                                 ring_desc = " NTC";
 791                         else
 792                                 ring_desc = "";
 793
 794                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 795                         rx_desc = IXGBE_RX_DESC(rx_ring, i);
 796                         u0 = (struct my_u0 *)rx_desc;
 797                         if (rx_desc->wb.upper.length) {
 798                                 /* Descriptor Done */
 799                                 pr_info("RWB[0x%03X]     %016llX %016llX ---------------- %p%s\n",
 800                                         i,
 801                                         le64_to_cpu((__force __le64)u0->a),
 802                                         le64_to_cpu((__force __le64)u0->b),
 803                                         rx_buffer_info->skb,
 804                                         ring_desc);
 805                         } else {
 806                                 pr_info("R  [0x%03X]     %016llX %016llX %016llX %p%s\n",
 807                                         i,
 808                                         le64_to_cpu((__force __le64)u0->a),
 809                                         le64_to_cpu((__force __le64)u0->b),
 810                                         (u64)rx_buffer_info->dma,
 811                                         rx_buffer_info->skb,
 812                                         ring_desc);
 813
 814                                 if (netif_msg_pktdata(adapter) &&
 815                                     rx_buffer_info->dma) {
 816                                         print_hex_dump(KERN_INFO, "",
 817                                            DUMP_PREFIX_ADDRESS, 16, 1,
 818                                            page_address(rx_buffer_info->page) +
 819                                                     rx_buffer_info->page_offset,
 820                                            ixgbe_rx_bufsz(rx_ring), true);
 821                                 }
 822                         }
 823                 }
 824         }
 825 }
 826
 827 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 828 {
 829         u32 ctrl_ext;
 830
 831         /* Let firmware take over control of h/w */
 832         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 833         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 834                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 835 }
 836
 837 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 838 {
 839         u32 ctrl_ext;
 840
 841         /* Let firmware know the driver has taken over */
 842         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 843         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 844                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 845 }
 846
 847 /**
 848  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 849  * @adapter: pointer to adapter struct
 850  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 851  * @queue: queue to map the corresponding interrupt to
 852  * @msix_vector: the vector to map to the corresponding queue
 853  *
 854  */
 855 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 856                            u8 queue, u8 msix_vector)
 857 {
 858         u32 ivar, index;
 859         struct ixgbe_hw *hw = &adapter->hw;
 860         switch (hw->mac.type) {
 861         case ixgbe_mac_82598EB:
 862                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 863                 if (direction == -1)
 864                         direction = 0;
 865                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 866                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 867                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 868                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 869                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 870                 break;
 871         case ixgbe_mac_82599EB:
 872         case ixgbe_mac_X540:
 873         case ixgbe_mac_X550:
 874         case ixgbe_mac_X550EM_x:
 875         case ixgbe_mac_x550em_a:
 876                 if (direction == -1) {
 877                         /* other causes */
 878                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 879                         index = ((queue & 1) * 8);
 880                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 881                         ivar &= ~(0xFF << index);
 882                         ivar |= (msix_vector << index);
 883                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 884                         break;
 885                 } else {
 886                         /* tx or rx causes */
 887                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 888                         index = ((16 * (queue & 1)) + (8 * direction));
 889                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 890                         ivar &= ~(0xFF << index);
 891                         ivar |= (msix_vector << index);
 892                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 893                         break;
 894                 }
 895         default:
 896                 break;
 897         }
 898 }
 899
 900 void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 901                             u64 qmask)
 902 {
 903         u32 mask;
 904
 905         switch (adapter->hw.mac.type) {
 906         case ixgbe_mac_82598EB:
 907                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 908                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 909                 break;
 910         case ixgbe_mac_82599EB:
 911         case ixgbe_mac_X540:
 912         case ixgbe_mac_X550:
 913         case ixgbe_mac_X550EM_x:
 914         case ixgbe_mac_x550em_a:
 915                 mask = (qmask & 0xFFFFFFFF);
 916                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 917                 mask = (qmask >> 32);
 918                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 919                 break;
 920         default:
 921                 break;
 922         }
 923 }
 924
 925 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
 926 {
 927         struct ixgbe_hw *hw = &adapter->hw;
 928         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 929         int i;
 930         u32 data;
 931
 932         if ((hw->fc.current_mode != ixgbe_fc_full) &&
 933             (hw->fc.current_mode != ixgbe_fc_rx_pause))
 934                 return;
 935
 936         switch (hw->mac.type) {
 937         case ixgbe_mac_82598EB:
 938                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
 939                 break;
 940         default:
 941                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
 942         }
 943         hwstats->lxoffrxc += data;
 944
 945         /* refill credits (no tx hang) if we received xoff */
 946         if (!data)
 947                 return;
 948
 949         for (i = 0; i < adapter->num_tx_queues; i++)
 950                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
 951                           &adapter->tx_ring[i]->state);
 952
 953         for (i = 0; i < adapter->num_xdp_queues; i++)
 954                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
 955                           &adapter->xdp_ring[i]->state);
 956 }
 957
 958 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
 959 {
 960         struct ixgbe_hw *hw = &adapter->hw;
 961         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 962         u32 xoff[8] = {0};
 963         u8 tc;
 964         int i;
 965         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
 966
 967         if (adapter->ixgbe_ieee_pfc)
 968                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
 969
 970         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
 971                 ixgbe_update_xoff_rx_lfc(adapter);
 972                 return;
 973         }
 974
 975         /* update stats for each tc, only valid with PFC enabled */
 976         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
 977                 u32 pxoffrxc;
 978
 979                 switch (hw->mac.type) {
 980                 case ixgbe_mac_82598EB:
 981                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
 982                         break;
 983                 default:
 984                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
 985                 }
 986                 hwstats->pxoffrxc[i] += pxoffrxc;
 987                 /* Get the TC for given UP */
 988                 tc = netdev_get_prio_tc_map(adapter->netdev, i);
 989                 xoff[tc] += pxoffrxc;
 990         }
 991
 992         /* disarm tx queues that have received xoff frames */
 993         for (i = 0; i < adapter->num_tx_queues; i++) {
 994                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
 995
 996                 tc = tx_ring->dcb_tc;
 997                 if (xoff[tc])
 998                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
 999         }
1000
1001         for (i = 0; i < adapter->num_xdp_queues; i++) {
1002                 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
1003
1004                 tc = xdp_ring->dcb_tc;
1005                 if (xoff[tc])
1006                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &xdp_ring->state);
1007         }
1008 }
1009
1010 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1011 {
1012         return ring->stats.packets;
1013 }
1014
1015 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1016 {
1017         unsigned int head, tail;
1018
1019         head = ring->next_to_clean;
1020         tail = ring->next_to_use;
1021
1022         return ((head <= tail) ? tail : tail + ring->count) - head;
1023 }
1024
1025 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1026 {
1027         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1028         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1029         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1030
1031         clear_check_for_tx_hang(tx_ring);
1032
1033         /*
1034          * Check for a hung queue, but be thorough. This verifies
1035          * that a transmit has been completed since the previous
1036          * check AND there is at least one packet pending. The
1037          * ARMED bit is set to indicate a potential hang. The
1038          * bit is cleared if a pause frame is received to remove
1039          * false hang detection due to PFC or 802.3x frames. By
1040          * requiring this to fail twice we avoid races with
1041          * pfc clearing the ARMED bit and conditions where we
1042          * run the check_tx_hang logic with a transmit completion
1043          * pending but without time to complete it yet.
1044          */
1045         if (tx_done_old == tx_done && tx_pending)
1046                 /* make sure it is true for two checks in a row */
1047                 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1048                                         &tx_ring->state);
1049         /* update completed stats and continue */
1050         tx_ring->tx_stats.tx_done_old = tx_done;
1051         /* reset the countdown */
1052         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1053
1054         return false;
1055 }
1056
1057 /**
1058  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1059  * @adapter: driver private struct
1060  **/
1061 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1062 {
1063
1064         /* Do the reset outside of interrupt context */
1065         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1066                 set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
1067                 e_warn(drv, "initiating reset due to tx timeout\n");
1068                 ixgbe_service_event_schedule(adapter);
1069         }
1070 }
1071
1072 /**
1073  * ixgbe_tx_maxrate - callback to set the maximum per-queue bitrate
1074  * @netdev: network interface device structure
1075  * @queue_index: Tx queue to set
1076  * @maxrate: desired maximum transmit bitrate
1077  **/
1078 static int ixgbe_tx_maxrate(struct net_device *netdev,
1079                             int queue_index, u32 maxrate)
1080 {
1081         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1082         struct ixgbe_hw *hw = &adapter->hw;
1083         u32 bcnrc_val = ixgbe_link_mbps(adapter);
1084
1085         if (!maxrate)
1086                 return 0;
1087
1088         /* Calculate the rate factor values to set */
1089         bcnrc_val <<= IXGBE_RTTBCNRC_RF_INT_SHIFT;
1090         bcnrc_val /= maxrate;
1091
1092         /* clear everything but the rate factor */
1093         bcnrc_val &= IXGBE_RTTBCNRC_RF_INT_MASK |
1094         IXGBE_RTTBCNRC_RF_DEC_MASK;
1095
1096         /* enable the rate scheduler */
1097         bcnrc_val |= IXGBE_RTTBCNRC_RS_ENA;
1098
1099         IXGBE_WRITE_REG(hw, IXGBE_RTTDQSEL, queue_index);
1100         IXGBE_WRITE_REG(hw, IXGBE_RTTBCNRC, bcnrc_val);
1101
1102         return 0;
1103 }
1104
1105 /**
1106  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1107  * @q_vector: structure containing interrupt and ring information
1108  * @tx_ring: tx ring to clean
1109  * @napi_budget: Used to determine if we are in netpoll
1110  **/
1111 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1112                                struct ixgbe_ring *tx_ring, int napi_budget)
1113 {
1114         struct ixgbe_adapter *adapter = q_vector->adapter;
1115         struct ixgbe_tx_buffer *tx_buffer;
1116         union ixgbe_adv_tx_desc *tx_desc;
1117         unsigned int total_bytes = 0, total_packets = 0, total_ipsec = 0;
1118         unsigned int budget = q_vector->tx.work_limit;
1119         unsigned int i = tx_ring->next_to_clean;
1120
1121         if (test_bit(__IXGBE_DOWN, &adapter->state))
1122                 return true;
1123
1124         tx_buffer = &tx_ring->tx_buffer_info[i];
1125         tx_desc = IXGBE_TX_DESC(tx_ring, i);
1126         i -= tx_ring->count;
1127
1128         do {
1129                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1130
1131                 /* if next_to_watch is not set then there is no work pending */
1132                 if (!eop_desc)
1133                         break;
1134
1135                 /* prevent any other reads prior to eop_desc */
1136                 smp_rmb();
1137
1138                 /* if DD is not set pending work has not been completed */
1139                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1140                         break;
1141
1142                 /* clear next_to_watch to prevent false hangs */
1143                 tx_buffer->next_to_watch = NULL;
1144
1145                 /* update the statistics for this packet */
1146                 total_bytes += tx_buffer->bytecount;
1147                 total_packets += tx_buffer->gso_segs;
1148                 if (tx_buffer->tx_flags & IXGBE_TX_FLAGS_IPSEC)
1149                         total_ipsec++;
1150
1151                 /* free the skb */
1152                 if (ring_is_xdp(tx_ring))
1153                         xdp_return_frame(tx_buffer->xdpf);
1154                 else
1155                         napi_consume_skb(tx_buffer->skb, napi_budget);
1156
1157                 /* unmap skb header data */
1158                 dma_unmap_single(tx_ring->dev,
1159                                  dma_unmap_addr(tx_buffer, dma),
1160                                  dma_unmap_len(tx_buffer, len),
1161                                  DMA_TO_DEVICE);
1162
1163                 /* clear tx_buffer data */
1164                 dma_unmap_len_set(tx_buffer, len, 0);
1165
1166                 /* unmap remaining buffers */
1167                 while (tx_desc != eop_desc) {
1168                         tx_buffer++;
1169                         tx_desc++;
1170                         i++;
1171                         if (unlikely(!i)) {
1172                                 i -= tx_ring->count;
1173                                 tx_buffer = tx_ring->tx_buffer_info;
1174                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1175                         }
1176
1177                         /* unmap any remaining paged data */
1178                         if (dma_unmap_len(tx_buffer, len)) {
1179                                 dma_unmap_page(tx_ring->dev,
1180                                                dma_unmap_addr(tx_buffer, dma),
1181                                                dma_unmap_len(tx_buffer, len),
1182                                                DMA_TO_DEVICE);
1183                                 dma_unmap_len_set(tx_buffer, len, 0);
1184                         }
1185                 }
1186
1187                 /* move us one more past the eop_desc for start of next pkt */
1188                 tx_buffer++;
1189                 tx_desc++;
1190                 i++;
1191                 if (unlikely(!i)) {
1192                         i -= tx_ring->count;
1193                         tx_buffer = tx_ring->tx_buffer_info;
1194                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1195                 }
1196
1197                 /* issue prefetch for next Tx descriptor */
1198                 prefetch(tx_desc);
1199
1200                 /* update budget accounting */
1201                 budget--;
1202         } while (likely(budget));
1203
1204         i += tx_ring->count;
1205         tx_ring->next_to_clean = i;
1206         u64_stats_update_begin(&tx_ring->syncp);
1207         tx_ring->stats.bytes += total_bytes;
1208         tx_ring->stats.packets += total_packets;
1209         u64_stats_update_end(&tx_ring->syncp);
1210         q_vector->tx.total_bytes += total_bytes;
1211         q_vector->tx.total_packets += total_packets;
1212         adapter->tx_ipsec += total_ipsec;
1213
1214         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1215                 /* schedule immediate reset if we believe we hung */
1216                 struct ixgbe_hw *hw = &adapter->hw;
1217                 e_err(drv, "Detected Tx Unit Hang %s\n"
1218                         "  Tx Queue             <%d>\n"
1219                         "  TDH, TDT             <%x>, <%x>\n"
1220                         "  next_to_use          <%x>\n"
1221                         "  next_to_clean        <%x>\n"
1222                         "tx_buffer_info[next_to_clean]\n"
1223                         "  time_stamp           <%lx>\n"
1224                         "  jiffies              <%lx>\n",
1225                         ring_is_xdp(tx_ring) ? "(XDP)" : "",
1226                         tx_ring->queue_index,
1227                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1228                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1229                         tx_ring->next_to_use, i,
1230                         tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1231
1232                 if (!ring_is_xdp(tx_ring))
1233                         netif_stop_subqueue(tx_ring->netdev,
1234                                             tx_ring->queue_index);
1235
1236                 e_info(probe,
1237                        "tx hang %d detected on queue %d, resetting adapter\n",
1238                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
1239
1240                 /* schedule immediate reset if we believe we hung */
1241                 ixgbe_tx_timeout_reset(adapter);
1242
1243                 /* the adapter is about to reset, no point in enabling stuff */
1244                 return true;
1245         }
1246
1247         if (ring_is_xdp(tx_ring))
1248                 return !!budget;
1249
1250         netdev_tx_completed_queue(txring_txq(tx_ring),
1251                                   total_packets, total_bytes);
1252
1253 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1254         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1255                      (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1256                 /* Make sure that anybody stopping the queue after this
1257                  * sees the new next_to_clean.
1258                  */
1259                 smp_mb();
1260                 if (__netif_subqueue_stopped(tx_ring->netdev,
1261                                              tx_ring->queue_index)
1262                     && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1263                         netif_wake_subqueue(tx_ring->netdev,
1264                                             tx_ring->queue_index);
1265                         ++tx_ring->tx_stats.restart_queue;
1266                 }
1267         }
1268
1269         return !!budget;
1270 }
1271
1272 #ifdef CONFIG_IXGBE_DCA
1273 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1274                                 struct ixgbe_ring *tx_ring,
1275                                 int cpu)
1276 {
1277         struct ixgbe_hw *hw = &adapter->hw;
1278         u32 txctrl = 0;
1279         u16 reg_offset;
1280
1281         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1282                 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1283
1284         switch (hw->mac.type) {
1285         case ixgbe_mac_82598EB:
1286                 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1287                 break;
1288         case ixgbe_mac_82599EB:
1289         case ixgbe_mac_X540:
1290                 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1291                 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1292                 break;
1293         default:
1294                 /* for unknown hardware do not write register */
1295                 return;
1296         }
1297
1298         /*
1299          * We can enable relaxed ordering for reads, but not writes when
1300          * DCA is enabled.  This is due to a known issue in some chipsets
1301          * which will cause the DCA tag to be cleared.
1302          */
1303         txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1304                   IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1305                   IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1306
1307         IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1308 }
1309
1310 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1311                                 struct ixgbe_ring *rx_ring,
1312                                 int cpu)
1313 {
1314         struct ixgbe_hw *hw = &adapter->hw;
1315         u32 rxctrl = 0;
1316         u8 reg_idx = rx_ring->reg_idx;
1317
1318         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1319                 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1320
1321         switch (hw->mac.type) {
1322         case ixgbe_mac_82599EB:
1323         case ixgbe_mac_X540:
1324                 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1325                 break;
1326         default:
1327                 break;
1328         }
1329
1330         /*
1331          * We can enable relaxed ordering for reads, but not writes when
1332          * DCA is enabled.  This is due to a known issue in some chipsets
1333          * which will cause the DCA tag to be cleared.
1334          */
1335         rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1336                   IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1337                   IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1338
1339         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1340 }
1341
1342 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1343 {
1344         struct ixgbe_adapter *adapter = q_vector->adapter;
1345         struct ixgbe_ring *ring;
1346         int cpu = get_cpu();
1347
1348         if (q_vector->cpu == cpu)
1349                 goto out_no_update;
1350
1351         ixgbe_for_each_ring(ring, q_vector->tx)
1352                 ixgbe_update_tx_dca(adapter, ring, cpu);
1353
1354         ixgbe_for_each_ring(ring, q_vector->rx)
1355                 ixgbe_update_rx_dca(adapter, ring, cpu);
1356
1357         q_vector->cpu = cpu;
1358 out_no_update:
1359         put_cpu();
1360 }
1361
1362 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1363 {
1364         int i;
1365
1366         /* always use CB2 mode, difference is masked in the CB driver */
1367         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1368                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1369                                 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1370         else
1371                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1372                                 IXGBE_DCA_CTRL_DCA_DISABLE);
1373
1374         for (i = 0; i < adapter->num_q_vectors; i++) {
1375                 adapter->q_vector[i]->cpu = -1;
1376                 ixgbe_update_dca(adapter->q_vector[i]);
1377         }
1378 }
1379
1380 static int __ixgbe_notify_dca(struct device *dev, void *data)
1381 {
1382         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1383         unsigned long event = *(unsigned long *)data;
1384
1385         if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1386                 return 0;
1387
1388         switch (event) {
1389         case DCA_PROVIDER_ADD:
1390                 /* if we're already enabled, don't do it again */
1391                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1392                         break;
1393                 if (dca_add_requester(dev) == 0) {
1394                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1395                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1396                                         IXGBE_DCA_CTRL_DCA_MODE_CB2);
1397                         break;
1398                 }
1399                 /* fall through - DCA is disabled. */
1400         case DCA_PROVIDER_REMOVE:
1401                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1402                         dca_remove_requester(dev);
1403                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1404                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1405                                         IXGBE_DCA_CTRL_DCA_DISABLE);
1406                 }
1407                 break;
1408         }
1409
1410         return 0;
1411 }
1412
1413 #endif /* CONFIG_IXGBE_DCA */
1414
1415 #define IXGBE_RSS_L4_TYPES_MASK \
1416         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1417          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1418          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1419          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1420
1421 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1422                                  union ixgbe_adv_rx_desc *rx_desc,
1423                                  struct sk_buff *skb)
1424 {
1425         u16 rss_type;
1426
1427         if (!(ring->netdev->features & NETIF_F_RXHASH))
1428                 return;
1429
1430         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1431                    IXGBE_RXDADV_RSSTYPE_MASK;
1432
1433         if (!rss_type)
1434                 return;
1435
1436         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1437                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1438                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1439 }
1440
1441 #ifdef IXGBE_FCOE
1442 /**
1443  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1444  * @ring: structure containing ring specific data
1445  * @rx_desc: advanced rx descriptor
1446  *
1447  * Returns : true if it is FCoE pkt
1448  */
1449 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1450                                     union ixgbe_adv_rx_desc *rx_desc)
1451 {
1452         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1453
1454         return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1455                ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1456                 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1457                              IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1458 }
1459
1460 #endif /* IXGBE_FCOE */
1461 /**
1462  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1463  * @ring: structure containing ring specific data
1464  * @rx_desc: current Rx descriptor being processed
1465  * @skb: skb currently being received and modified
1466  **/
1467 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1468                                      union ixgbe_adv_rx_desc *rx_desc,
1469                                      struct sk_buff *skb)
1470 {
1471         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1472         bool encap_pkt = false;
1473
1474         skb_checksum_none_assert(skb);
1475
1476         /* Rx csum disabled */
1477         if (!(ring->netdev->features & NETIF_F_RXCSUM))
1478                 return;
1479
1480         /* check for VXLAN and Geneve packets */
1481         if (pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) {
1482                 encap_pkt = true;
1483                 skb->encapsulation = 1;
1484         }
1485
1486         /* if IP and error */
1487         if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1488             ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1489                 ring->rx_stats.csum_err++;
1490                 return;
1491         }
1492
1493         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1494                 return;
1495
1496         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1497                 /*
1498                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1499                  * checksum errors.
1500                  */
1501                 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1502                     test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1503                         return;
1504
1505                 ring->rx_stats.csum_err++;
1506                 return;
1507         }
1508
1509         /* It must be a TCP or UDP packet with a valid checksum */
1510         skb->ip_summed = CHECKSUM_UNNECESSARY;
1511         if (encap_pkt) {
1512                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1513                         return;
1514
1515                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1516                         skb->ip_summed = CHECKSUM_NONE;
1517                         return;
1518                 }
1519                 /* If we checked the outer header let the stack know */
1520                 skb->csum_level = 1;
1521         }
1522 }
1523
1524 static inline unsigned int ixgbe_rx_offset(struct ixgbe_ring *rx_ring)
1525 {
1526         return ring_uses_build_skb(rx_ring) ? IXGBE_SKB_PAD : 0;
1527 }
1528
1529 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1530                                     struct ixgbe_rx_buffer *bi)
1531 {
1532         struct page *page = bi->page;
1533         dma_addr_t dma;
1534
1535         /* since we are recycling buffers we should seldom need to alloc */
1536         if (likely(page))
1537                 return true;
1538
1539         /* alloc new page for storage */
1540         page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1541         if (unlikely(!page)) {
1542                 rx_ring->rx_stats.alloc_rx_page_failed++;
1543                 return false;
1544         }
1545
1546         /* map page for use */
1547         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
1548                                  ixgbe_rx_pg_size(rx_ring),
1549                                  DMA_FROM_DEVICE,
1550                                  IXGBE_RX_DMA_ATTR);
1551
1552         /*
1553          * if mapping failed free memory back to system since
1554          * there isn't much point in holding memory we can't use
1555          */
1556         if (dma_mapping_error(rx_ring->dev, dma)) {
1557                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1558
1559                 rx_ring->rx_stats.alloc_rx_page_failed++;
1560                 return false;
1561         }
1562
1563         bi->dma = dma;
1564         bi->page = page;
1565         bi->page_offset = ixgbe_rx_offset(rx_ring);
1566         page_ref_add(page, USHRT_MAX - 1);
1567         bi->pagecnt_bias = USHRT_MAX;
1568         rx_ring->rx_stats.alloc_rx_page++;
1569
1570         return true;
1571 }
1572
1573 /**
1574  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1575  * @rx_ring: ring to place buffers on
1576  * @cleaned_count: number of buffers to replace
1577  **/
1578 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1579 {
1580         union ixgbe_adv_rx_desc *rx_desc;
1581         struct ixgbe_rx_buffer *bi;
1582         u16 i = rx_ring->next_to_use;
1583         u16 bufsz;
1584
1585         /* nothing to do */
1586         if (!cleaned_count)
1587                 return;
1588
1589         rx_desc = IXGBE_RX_DESC(rx_ring, i);
1590         bi = &rx_ring->rx_buffer_info[i];
1591         i -= rx_ring->count;
1592
1593         bufsz = ixgbe_rx_bufsz(rx_ring);
1594
1595         do {
1596                 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1597                         break;
1598
1599                 /* sync the buffer for use by the device */
1600                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
1601                                                  bi->page_offset, bufsz,
1602                                                  DMA_FROM_DEVICE);
1603
1604                 /*
1605                  * Refresh the desc even if buffer_addrs didn't change
1606                  * because each write-back erases this info.
1607                  */
1608                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1609
1610                 rx_desc++;
1611                 bi++;
1612                 i++;
1613                 if (unlikely(!i)) {
1614                         rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1615                         bi = rx_ring->rx_buffer_info;
1616                         i -= rx_ring->count;
1617                 }
1618
1619                 /* clear the length for the next_to_use descriptor */
1620                 rx_desc->wb.upper.length = 0;
1621
1622                 cleaned_count--;
1623         } while (cleaned_count);
1624
1625         i += rx_ring->count;
1626
1627         if (rx_ring->next_to_use != i) {
1628                 rx_ring->next_to_use = i;
1629
1630                 /* update next to alloc since we have filled the ring */
1631                 rx_ring->next_to_alloc = i;
1632
1633                 /* Force memory writes to complete before letting h/w
1634                  * know there are new descriptors to fetch.  (Only
1635                  * applicable for weak-ordered memory model archs,
1636                  * such as IA-64).
1637                  */
1638                 wmb();
1639                 writel(i, rx_ring->tail);
1640         }
1641 }
1642
1643 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1644                                    struct sk_buff *skb)
1645 {
1646         u16 hdr_len = skb_headlen(skb);
1647
1648         /* set gso_size to avoid messing up TCP MSS */
1649         skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1650                                                  IXGBE_CB(skb)->append_cnt);
1651         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1652 }
1653
1654 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1655                                    struct sk_buff *skb)
1656 {
1657         /* if append_cnt is 0 then frame is not RSC */
1658         if (!IXGBE_CB(skb)->append_cnt)
1659                 return;
1660
1661         rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1662         rx_ring->rx_stats.rsc_flush++;
1663
1664         ixgbe_set_rsc_gso_size(rx_ring, skb);
1665
1666         /* gso_size is computed using append_cnt so always clear it last */
1667         IXGBE_CB(skb)->append_cnt = 0;
1668 }
1669
1670 /**
1671  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1672  * @rx_ring: rx descriptor ring packet is being transacted on
1673  * @rx_desc: pointer to the EOP Rx descriptor
1674  * @skb: pointer to current skb being populated
1675  *
1676  * This function checks the ring, descriptor, and packet information in
1677  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1678  * other fields within the skb.
1679  **/
1680 void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1681                               union ixgbe_adv_rx_desc *rx_desc,
1682                               struct sk_buff *skb)
1683 {
1684         struct net_device *dev = rx_ring->netdev;
1685         u32 flags = rx_ring->q_vector->adapter->flags;
1686
1687         ixgbe_update_rsc_stats(rx_ring, skb);
1688
1689         ixgbe_rx_hash(rx_ring, rx_desc, skb);
1690
1691         ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1692
1693         if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1694                 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1695
1696         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1697             ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1698                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1699                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1700         }
1701
1702         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_SECP))
1703                 ixgbe_ipsec_rx(rx_ring, rx_desc, skb);
1704
1705         /* record Rx queue, or update MACVLAN statistics */
1706         if (netif_is_ixgbe(dev))
1707                 skb_record_rx_queue(skb, rx_ring->queue_index);
1708         else
1709                 macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
1710                                  false);
1711
1712         skb->protocol = eth_type_trans(skb, dev);
1713 }
1714
1715 void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1716                   struct sk_buff *skb)
1717 {
1718         napi_gro_receive(&q_vector->napi, skb);
1719 }
1720
1721 /**
1722  * ixgbe_is_non_eop - process handling of non-EOP buffers
1723  * @rx_ring: Rx ring being processed
1724  * @rx_desc: Rx descriptor for current buffer
1725  * @skb: Current socket buffer containing buffer in progress
1726  *
1727  * This function updates next to clean.  If the buffer is an EOP buffer
1728  * this function exits returning false, otherwise it will place the
1729  * sk_buff in the next buffer to be chained and return true indicating
1730  * that this is in fact a non-EOP buffer.
1731  **/
1732 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1733                              union ixgbe_adv_rx_desc *rx_desc,
1734                              struct sk_buff *skb)
1735 {
1736         u32 ntc = rx_ring->next_to_clean + 1;
1737
1738         /* fetch, update, and store next to clean */
1739         ntc = (ntc < rx_ring->count) ? ntc : 0;
1740         rx_ring->next_to_clean = ntc;
1741
1742         prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1743
1744         /* update RSC append count if present */
1745         if (ring_is_rsc_enabled(rx_ring)) {
1746                 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1747                                      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1748
1749                 if (unlikely(rsc_enabled)) {
1750                         u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1751
1752                         rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1753                         IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1754
1755                         /* update ntc based on RSC value */
1756                         ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1757                         ntc &= IXGBE_RXDADV_NEXTP_MASK;
1758                         ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1759                 }
1760         }
1761
1762         /* if we are the last buffer then there is nothing else to do */
1763         if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1764                 return false;
1765
1766         /* place skb in next buffer to be received */
1767         rx_ring->rx_buffer_info[ntc].skb = skb;
1768         rx_ring->rx_stats.non_eop_descs++;
1769
1770         return true;
1771 }
1772
1773 /**
1774  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1775  * @rx_ring: rx descriptor ring packet is being transacted on
1776  * @skb: pointer to current skb being adjusted
1777  *
1778  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1779  * main difference between this version and the original function is that
1780  * this function can make several assumptions about the state of things
1781  * that allow for significant optimizations versus the standard function.
1782  * As a result we can do things like drop a frag and maintain an accurate
1783  * truesize for the skb.
1784  */
1785 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1786                             struct sk_buff *skb)
1787 {
1788         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1789         unsigned char *va;
1790         unsigned int pull_len;
1791
1792         /*
1793          * it is valid to use page_address instead of kmap since we are
1794          * working with pages allocated out of the lomem pool per
1795          * alloc_page(GFP_ATOMIC)
1796          */
1797         va = skb_frag_address(frag);
1798
1799         /*
1800          * we need the header to contain the greater of either ETH_HLEN or
1801          * 60 bytes if the skb->len is less than 60 for skb_pad.
1802          */
1803         pull_len = eth_get_headlen(skb->dev, va, IXGBE_RX_HDR_SIZE);
1804
1805         /* align pull length to size of long to optimize memcpy performance */
1806         skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1807
1808         /* update all of the pointers */
1809         skb_frag_size_sub(frag, pull_len);
1810         frag->page_offset += pull_len;
1811         skb->data_len -= pull_len;
1812         skb->tail += pull_len;
1813 }
1814
1815 /**
1816  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1817  * @rx_ring: rx descriptor ring packet is being transacted on
1818  * @skb: pointer to current skb being updated
1819  *
1820  * This function provides a basic DMA sync up for the first fragment of an
1821  * skb.  The reason for doing this is that the first fragment cannot be
1822  * unmapped until we have reached the end of packet descriptor for a buffer
1823  * chain.
1824  */
1825 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1826                                 struct sk_buff *skb)
1827 {
1828         /* if the page was released unmap it, else just sync our portion */
1829         if (unlikely(IXGBE_CB(skb)->page_released)) {
1830                 dma_unmap_page_attrs(rx_ring->dev, IXGBE_CB(skb)->dma,
1831                                      ixgbe_rx_pg_size(rx_ring),
1832                                      DMA_FROM_DEVICE,
1833                                      IXGBE_RX_DMA_ATTR);
1834         } else if (ring_uses_build_skb(rx_ring)) {
1835                 unsigned long offset = (unsigned long)(skb->data) & ~PAGE_MASK;
1836
1837                 dma_sync_single_range_for_cpu(rx_ring->dev,
1838                                               IXGBE_CB(skb)->dma,
1839                                               offset,
1840                                               skb_headlen(skb),
1841                                               DMA_FROM_DEVICE);
1842         } else {
1843                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1844
1845                 dma_sync_single_range_for_cpu(rx_ring->dev,
1846                                               IXGBE_CB(skb)->dma,
1847                                               frag->page_offset,
1848                                               skb_frag_size(frag),
1849                                               DMA_FROM_DEVICE);
1850         }
1851 }
1852
1853 /**
1854  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1855  * @rx_ring: rx descriptor ring packet is being transacted on
1856  * @rx_desc: pointer to the EOP Rx descriptor
1857  * @skb: pointer to current skb being fixed
1858  *
1859  * Check if the skb is valid in the XDP case it will be an error pointer.
1860  * Return true in this case to abort processing and advance to next
1861  * descriptor.
1862  *
1863  * Check for corrupted packet headers caused by senders on the local L2
1864  * embedded NIC switch not setting up their Tx Descriptors right.  These
1865  * should be very rare.
1866  *
1867  * Also address the case where we are pulling data in on pages only
1868  * and as such no data is present in the skb header.
1869  *
1870  * In addition if skb is not at least 60 bytes we need to pad it so that
1871  * it is large enough to qualify as a valid Ethernet frame.
1872  *
1873  * Returns true if an error was encountered and skb was freed.
1874  **/
1875 bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1876                            union ixgbe_adv_rx_desc *rx_desc,
1877                            struct sk_buff *skb)
1878 {
1879         struct net_device *netdev = rx_ring->netdev;
1880
1881         /* XDP packets use error pointer so abort at this point */
1882         if (IS_ERR(skb))
1883                 return true;
1884
1885         /* Verify netdev is present, and that packet does not have any
1886          * errors that would be unacceptable to the netdev.
1887          */
1888         if (!netdev ||
1889             (unlikely(ixgbe_test_staterr(rx_desc,
1890                                          IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1891              !(netdev->features & NETIF_F_RXALL)))) {
1892                 dev_kfree_skb_any(skb);
1893                 return true;
1894         }
1895
1896         /* place header in linear portion of buffer */
1897         if (!skb_headlen(skb))
1898                 ixgbe_pull_tail(rx_ring, skb);
1899
1900 #ifdef IXGBE_FCOE
1901         /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1902         if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1903                 return false;
1904
1905 #endif
1906         /* if eth_skb_pad returns an error the skb was freed */
1907         if (eth_skb_pad(skb))
1908                 return true;
1909
1910         return false;
1911 }
1912
1913 /**
1914  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1915  * @rx_ring: rx descriptor ring to store buffers on
1916  * @old_buff: donor buffer to have page reused
1917  *
1918  * Synchronizes page for reuse by the adapter
1919  **/
1920 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1921                                 struct ixgbe_rx_buffer *old_buff)
1922 {
1923         struct ixgbe_rx_buffer *new_buff;
1924         u16 nta = rx_ring->next_to_alloc;
1925
1926         new_buff = &rx_ring->rx_buffer_info[nta];
1927
1928         /* update, and store next to alloc */
1929         nta++;
1930         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1931
1932         /* Transfer page from old buffer to new buffer.
1933          * Move each member individually to avoid possible store
1934          * forwarding stalls and unnecessary copy of skb.
1935          */
1936         new_buff->dma           = old_buff->dma;
1937         new_buff->page          = old_buff->page;
1938         new_buff->page_offset   = old_buff->page_offset;
1939         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
1940 }
1941
1942 static inline bool ixgbe_page_is_reserved(struct page *page)
1943 {
1944         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1945 }
1946
1947 static bool ixgbe_can_reuse_rx_page(struct ixgbe_rx_buffer *rx_buffer)
1948 {
1949         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
1950         struct page *page = rx_buffer->page;
1951
1952         /* avoid re-using remote pages */
1953         if (unlikely(ixgbe_page_is_reserved(page)))
1954                 return false;
1955
1956 #if (PAGE_SIZE < 8192)
1957         /* if we are only owner of page we can reuse it */
1958         if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
1959                 return false;
1960 #else
1961         /* The last offset is a bit aggressive in that we assume the
1962          * worst case of FCoE being enabled and using a 3K buffer.
1963          * However this should have minimal impact as the 1K extra is
1964          * still less than one buffer in size.
1965          */
1966 #define IXGBE_LAST_OFFSET \
1967         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IXGBE_RXBUFFER_3K)
1968         if (rx_buffer->page_offset > IXGBE_LAST_OFFSET)
1969                 return false;
1970 #endif
1971
1972         /* If we have drained the page fragment pool we need to update
1973          * the pagecnt_bias and page count so that we fully restock the
1974          * number of references the driver holds.
1975          */
1976         if (unlikely(pagecnt_bias == 1)) {
1977                 page_ref_add(page, USHRT_MAX - 1);
1978                 rx_buffer->pagecnt_bias = USHRT_MAX;
1979         }
1980
1981         return true;
1982 }
1983
1984 /**
1985  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1986  * @rx_ring: rx descriptor ring to transact packets on
1987  * @rx_buffer: buffer containing page to add
1988  * @skb: sk_buff to place the data into
1989  * @size: size of data in rx_buffer
1990  *
1991  * This function will add the data contained in rx_buffer->page to the skb.
1992  * This is done either through a direct copy if the data in the buffer is
1993  * less than the skb header size, otherwise it will just attach the page as
1994  * a frag to the skb.
1995  *
1996  * The function will then update the page offset if necessary and return
1997  * true if the buffer can be reused by the adapter.
1998  **/
1999 static void ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
2000                               struct ixgbe_rx_buffer *rx_buffer,
2001                               struct sk_buff *skb,
2002                               unsigned int size)
2003 {
2004 #if (PAGE_SIZE < 8192)
2005         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2006 #else
2007         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2008                                 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2009                                 SKB_DATA_ALIGN(size);
2010 #endif
2011         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
2012                         rx_buffer->page_offset, size, truesize);
2013 #if (PAGE_SIZE < 8192)
2014         rx_buffer->page_offset ^= truesize;
2015 #else
2016         rx_buffer->page_offset += truesize;
2017 #endif
2018 }
2019
2020 static struct ixgbe_rx_buffer *ixgbe_get_rx_buffer(struct ixgbe_ring *rx_ring,
2021                                                    union ixgbe_adv_rx_desc *rx_desc,
2022                                                    struct sk_buff **skb,
2023                                                    const unsigned int size)
2024 {
2025         struct ixgbe_rx_buffer *rx_buffer;
2026
2027         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
2028         prefetchw(rx_buffer->page);
2029         *skb = rx_buffer->skb;
2030
2031         /* Delay unmapping of the first packet. It carries the header
2032          * information, HW may still access the header after the writeback.
2033          * Only unmap it when EOP is reached
2034          */
2035         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)) {
2036                 if (!*skb)
2037                         goto skip_sync;
2038         } else {
2039                 if (*skb)
2040                         ixgbe_dma_sync_frag(rx_ring, *skb);
2041         }
2042
2043         /* we are reusing so sync this buffer for CPU use */
2044         dma_sync_single_range_for_cpu(rx_ring->dev,
2045                                       rx_buffer->dma,
2046                                       rx_buffer->page_offset,
2047                                       size,
2048                                       DMA_FROM_DEVICE);
2049 skip_sync:
2050         rx_buffer->pagecnt_bias--;
2051
2052         return rx_buffer;
2053 }
2054
2055 static void ixgbe_put_rx_buffer(struct ixgbe_ring *rx_ring,
2056                                 struct ixgbe_rx_buffer *rx_buffer,
2057                                 struct sk_buff *skb)
2058 {
2059         if (ixgbe_can_reuse_rx_page(rx_buffer)) {
2060                 /* hand second half of page back to the ring */
2061                 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2062         } else {
2063                 if (!IS_ERR(skb) && IXGBE_CB(skb)->dma == rx_buffer->dma) {
2064                         /* the page has been released from the ring */
2065                         IXGBE_CB(skb)->page_released = true;
2066                 } else {
2067                         /* we are not reusing the buffer so unmap it */
2068                         dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
2069                                              ixgbe_rx_pg_size(rx_ring),
2070                                              DMA_FROM_DEVICE,
2071                                              IXGBE_RX_DMA_ATTR);
2072                 }
2073                 __page_frag_cache_drain(rx_buffer->page,
2074                                         rx_buffer->pagecnt_bias);
2075         }
2076
2077         /* clear contents of rx_buffer */
2078         rx_buffer->page = NULL;
2079         rx_buffer->skb = NULL;
2080 }
2081
2082 static struct sk_buff *ixgbe_construct_skb(struct ixgbe_ring *rx_ring,
2083                                            struct ixgbe_rx_buffer *rx_buffer,
2084                                            struct xdp_buff *xdp,
2085                                            union ixgbe_adv_rx_desc *rx_desc)
2086 {
2087         unsigned int size = xdp->data_end - xdp->data;
2088 #if (PAGE_SIZE < 8192)
2089         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2090 #else
2091         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
2092                                                xdp->data_hard_start);
2093 #endif
2094         struct sk_buff *skb;
2095
2096         /* prefetch first cache line of first page */
2097         prefetch(xdp->data);
2098 #if L1_CACHE_BYTES < 128
2099         prefetch(xdp->data + L1_CACHE_BYTES);
2100 #endif
2101         /* Note, we get here by enabling legacy-rx via:
2102          *
2103          *    ethtool --set-priv-flags <dev> legacy-rx on
2104          *
2105          * In this mode, we currently get 0 extra XDP headroom as
2106          * opposed to having legacy-rx off, where we process XDP
2107          * packets going to stack via ixgbe_build_skb(). The latter
2108          * provides us currently with 192 bytes of headroom.
2109          *
2110          * For ixgbe_construct_skb() mode it means that the
2111          * xdp->data_meta will always point to xdp->data, since
2112          * the helper cannot expand the head. Should this ever
2113          * change in future for legacy-rx mode on, then lets also
2114          * add xdp->data_meta handling here.
2115          */
2116
2117         /* allocate a skb to store the frags */
2118         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IXGBE_RX_HDR_SIZE);
2119         if (unlikely(!skb))
2120                 return NULL;
2121
2122         if (size > IXGBE_RX_HDR_SIZE) {
2123                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2124                         IXGBE_CB(skb)->dma = rx_buffer->dma;
2125
2126                 skb_add_rx_frag(skb, 0, rx_buffer->page,
2127                                 xdp->data - page_address(rx_buffer->page),
2128                                 size, truesize);
2129 #if (PAGE_SIZE < 8192)
2130                 rx_buffer->page_offset ^= truesize;
2131 #else
2132                 rx_buffer->page_offset += truesize;
2133 #endif
2134         } else {
2135                 memcpy(__skb_put(skb, size),
2136                        xdp->data, ALIGN(size, sizeof(long)));
2137                 rx_buffer->pagecnt_bias++;
2138         }
2139
2140         return skb;
2141 }
2142
2143 static struct sk_buff *ixgbe_build_skb(struct ixgbe_ring *rx_ring,
2144                                        struct ixgbe_rx_buffer *rx_buffer,
2145                                        struct xdp_buff *xdp,
2146                                        union ixgbe_adv_rx_desc *rx_desc)
2147 {
2148         unsigned int metasize = xdp->data - xdp->data_meta;
2149 #if (PAGE_SIZE < 8192)
2150         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2151 #else
2152         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
2153                                 SKB_DATA_ALIGN(xdp->data_end -
2154                                                xdp->data_hard_start);
2155 #endif
2156         struct sk_buff *skb;
2157
2158         /* Prefetch first cache line of first page. If xdp->data_meta
2159          * is unused, this points extactly as xdp->data, otherwise we
2160          * likely have a consumer accessing first few bytes of meta
2161          * data, and then actual data.
2162          */
2163         prefetch(xdp->data_meta);
2164 #if L1_CACHE_BYTES < 128
2165         prefetch(xdp->data_meta + L1_CACHE_BYTES);
2166 #endif
2167
2168         /* build an skb to around the page buffer */
2169         skb = build_skb(xdp->data_hard_start, truesize);
2170         if (unlikely(!skb))
2171                 return NULL;
2172
2173         /* update pointers within the skb to store the data */
2174         skb_reserve(skb, xdp->data - xdp->data_hard_start);
2175         __skb_put(skb, xdp->data_end - xdp->data);
2176         if (metasize)
2177                 skb_metadata_set(skb, metasize);
2178
2179         /* record DMA address if this is the start of a chain of buffers */
2180         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2181                 IXGBE_CB(skb)->dma = rx_buffer->dma;
2182
2183         /* update buffer offset */
2184 #if (PAGE_SIZE < 8192)
2185         rx_buffer->page_offset ^= truesize;
2186 #else
2187         rx_buffer->page_offset += truesize;
2188 #endif
2189
2190         return skb;
2191 }
2192
2193 static struct sk_buff *ixgbe_run_xdp(struct ixgbe_adapter *adapter,
2194                                      struct ixgbe_ring *rx_ring,
2195                                      struct xdp_buff *xdp)
2196 {
2197         int err, result = IXGBE_XDP_PASS;
2198         struct bpf_prog *xdp_prog;
2199         struct xdp_frame *xdpf;
2200         u32 act;
2201
2202         rcu_read_lock();
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 = convert_to_xdp_frame(xdp);
2216                 if (unlikely(!xdpf)) {
2217                         result = IXGBE_XDP_CONSUMED;
2218                         break;
2219                 }
2220                 result = ixgbe_xmit_xdp_ring(adapter, xdpf);
2221                 break;
2222         case XDP_REDIRECT:
2223                 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
2224                 if (!err)
2225                         result = IXGBE_XDP_REDIR;
2226                 else
2227                         result = IXGBE_XDP_CONSUMED;
2228                 break;
2229         default:
2230                 bpf_warn_invalid_xdp_action(act);
2231                 /* fallthrough */
2232         case XDP_ABORTED:
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         rcu_read_unlock();
2241         return ERR_PTR(-result);
2242 }
2243
2244 static void ixgbe_rx_buffer_flip(struct ixgbe_ring *rx_ring,
2245                                  struct ixgbe_rx_buffer *rx_buffer,
2246                                  unsigned int size)
2247 {
2248 #if (PAGE_SIZE < 8192)
2249         unsigned int truesize = ixgbe_rx_pg_size(rx_ring) / 2;
2250
2251         rx_buffer->page_offset ^= truesize;
2252 #else
2253         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
2254                                 SKB_DATA_ALIGN(IXGBE_SKB_PAD + size) :
2255                                 SKB_DATA_ALIGN(size);
2256
2257         rx_buffer->page_offset += truesize;
2258 #endif
2259 }
2260
2261 /**
2262  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2263  * @q_vector: structure containing interrupt and ring information
2264  * @rx_ring: rx descriptor ring to transact packets on
2265  * @budget: Total limit on number of packets to process
2266  *
2267  * This function provides a "bounce buffer" approach to Rx interrupt
2268  * processing.  The advantage to this is that on systems that have
2269  * expensive overhead for IOMMU access this provides a means of avoiding
2270  * it by maintaining the mapping of the page to the syste.
2271  *
2272  * Returns amount of work completed
2273  **/
2274 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2275                                struct ixgbe_ring *rx_ring,
2276                                const int budget)
2277 {
2278         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2279         struct ixgbe_adapter *adapter = q_vector->adapter;
2280 #ifdef IXGBE_FCOE
2281         int ddp_bytes;
2282         unsigned int mss = 0;
2283 #endif /* IXGBE_FCOE */
2284         u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2285         unsigned int xdp_xmit = 0;
2286         struct xdp_buff xdp;
2287
2288         xdp.rxq = &rx_ring->xdp_rxq;
2289
2290         while (likely(total_rx_packets < budget)) {
2291                 union ixgbe_adv_rx_desc *rx_desc;
2292                 struct ixgbe_rx_buffer *rx_buffer;
2293                 struct sk_buff *skb;
2294                 unsigned int size;
2295
2296                 /* return some buffers to hardware, one at a time is too slow */
2297                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2298                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2299                         cleaned_count = 0;
2300                 }
2301
2302                 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2303                 size = le16_to_cpu(rx_desc->wb.upper.length);
2304                 if (!size)
2305                         break;
2306
2307                 /* This memory barrier is needed to keep us from reading
2308                  * any other fields out of the rx_desc until we know the
2309                  * descriptor has been written back
2310                  */
2311                 dma_rmb();
2312
2313                 rx_buffer = ixgbe_get_rx_buffer(rx_ring, rx_desc, &skb, size);
2314
2315                 /* retrieve a buffer from the ring */
2316                 if (!skb) {
2317                         xdp.data = page_address(rx_buffer->page) +
2318                                    rx_buffer->page_offset;
2319                         xdp.data_meta = xdp.data;
2320                         xdp.data_hard_start = xdp.data -
2321                                               ixgbe_rx_offset(rx_ring);
2322                         xdp.data_end = xdp.data + size;
2323
2324                         skb = ixgbe_run_xdp(adapter, rx_ring, &xdp);
2325                 }
2326
2327                 if (IS_ERR(skb)) {
2328                         unsigned int xdp_res = -PTR_ERR(skb);
2329
2330                         if (xdp_res & (IXGBE_XDP_TX | IXGBE_XDP_REDIR)) {
2331                                 xdp_xmit |= xdp_res;
2332                                 ixgbe_rx_buffer_flip(rx_ring, rx_buffer, size);
2333                         } else {
2334                                 rx_buffer->pagecnt_bias++;
2335                         }
2336                         total_rx_packets++;
2337                         total_rx_bytes += size;
2338                 } else if (skb) {
2339                         ixgbe_add_rx_frag(rx_ring, rx_buffer, skb, size);
2340                 } else if (ring_uses_build_skb(rx_ring)) {
2341                         skb = ixgbe_build_skb(rx_ring, rx_buffer,
2342                                               &xdp, rx_desc);
2343                 } else {
2344                         skb = ixgbe_construct_skb(rx_ring, rx_buffer,
2345                                                   &xdp, rx_desc);
2346                 }
2347
2348                 /* exit if we failed to retrieve a buffer */
2349                 if (!skb) {
2350                         rx_ring->rx_stats.alloc_rx_buff_failed++;
2351                         rx_buffer->pagecnt_bias++;
2352                         break;
2353                 }
2354
2355                 ixgbe_put_rx_buffer(rx_ring, rx_buffer, skb);
2356                 cleaned_count++;
2357
2358                 /* place incomplete frames back on ring for completion */
2359                 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2360                         continue;
2361
2362                 /* verify the packet layout is correct */
2363                 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2364                         continue;
2365
2366                 /* probably a little skewed due to removing CRC */
2367                 total_rx_bytes += skb->len;
2368
2369                 /* populate checksum, timestamp, VLAN, and protocol */
2370                 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2371
2372 #ifdef IXGBE_FCOE
2373                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2374                 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2375                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2376                         /* include DDPed FCoE data */
2377                         if (ddp_bytes > 0) {
2378                                 if (!mss) {
2379                                         mss = rx_ring->netdev->mtu -
2380                                                 sizeof(struct fcoe_hdr) -
2381                                                 sizeof(struct fc_frame_header) -
2382                                                 sizeof(struct fcoe_crc_eof);
2383                                         if (mss > 512)
2384                                                 mss &= ~511;
2385                                 }
2386                                 total_rx_bytes += ddp_bytes;
2387                                 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2388                                                                  mss);
2389                         }
2390                         if (!ddp_bytes) {
2391                                 dev_kfree_skb_any(skb);
2392                                 continue;
2393                         }
2394                 }
2395
2396 #endif /* IXGBE_FCOE */
2397                 ixgbe_rx_skb(q_vector, skb);
2398
2399                 /* update budget accounting */
2400                 total_rx_packets++;
2401         }
2402
2403         if (xdp_xmit & IXGBE_XDP_REDIR)
2404                 xdp_do_flush_map();
2405
2406         if (xdp_xmit & IXGBE_XDP_TX) {
2407                 struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
2408
2409                 /* Force memory writes to complete before letting h/w
2410                  * know there are new descriptors to fetch.
2411                  */
2412                 wmb();
2413                 writel(ring->next_to_use, ring->tail);
2414         }
2415
2416         u64_stats_update_begin(&rx_ring->syncp);
2417         rx_ring->stats.packets += total_rx_packets;
2418         rx_ring->stats.bytes += total_rx_bytes;
2419         u64_stats_update_end(&rx_ring->syncp);
2420         q_vector->rx.total_packets += total_rx_packets;
2421         q_vector->rx.total_bytes += total_rx_bytes;
2422
2423         return total_rx_packets;
2424 }
2425
2426 /**
2427  * ixgbe_configure_msix - Configure MSI-X hardware
2428  * @adapter: board private structure
2429  *
2430  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2431  * interrupts.
2432  **/
2433 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2434 {
2435         struct ixgbe_q_vector *q_vector;
2436         int v_idx;
2437         u32 mask;
2438
2439         /* Populate MSIX to EITR Select */
2440         if (adapter->num_vfs > 32) {
2441                 u32 eitrsel = BIT(adapter->num_vfs - 32) - 1;
2442                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2443         }
2444
2445         /*
2446          * Populate the IVAR table and set the ITR values to the
2447          * corresponding register.
2448          */
2449         for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2450                 struct ixgbe_ring *ring;
2451                 q_vector = adapter->q_vector[v_idx];
2452
2453                 ixgbe_for_each_ring(ring, q_vector->rx)
2454                         ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2455
2456                 ixgbe_for_each_ring(ring, q_vector->tx)
2457                         ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2458
2459                 ixgbe_write_eitr(q_vector);
2460         }
2461
2462         switch (adapter->hw.mac.type) {
2463         case ixgbe_mac_82598EB:
2464                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2465                                v_idx);
2466                 break;
2467         case ixgbe_mac_82599EB:
2468         case ixgbe_mac_X540:
2469         case ixgbe_mac_X550:
2470         case ixgbe_mac_X550EM_x:
2471         case ixgbe_mac_x550em_a:
2472                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2473                 break;
2474         default:
2475                 break;
2476         }
2477         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2478
2479         /* set up to autoclear timer, and the vectors */
2480         mask = IXGBE_EIMS_ENABLE_MASK;
2481         mask &= ~(IXGBE_EIMS_OTHER |
2482                   IXGBE_EIMS_MAILBOX |
2483                   IXGBE_EIMS_LSC);
2484
2485         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2486 }
2487
2488 /**
2489  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2490  * @q_vector: structure containing interrupt and ring information
2491  * @ring_container: structure containing ring performance data
2492  *
2493  *      Stores a new ITR value based on packets and byte
2494  *      counts during the last interrupt.  The advantage of per interrupt
2495  *      computation is faster updates and more accurate ITR for the current
2496  *      traffic pattern.  Constants in this function were computed
2497  *      based on theoretical maximum wire speed and thresholds were set based
2498  *      on testing data as well as attempting to minimize response time
2499  *      while increasing bulk throughput.
2500  **/
2501 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2502                              struct ixgbe_ring_container *ring_container)
2503 {
2504         unsigned int itr = IXGBE_ITR_ADAPTIVE_MIN_USECS |
2505                            IXGBE_ITR_ADAPTIVE_LATENCY;
2506         unsigned int avg_wire_size, packets, bytes;
2507         unsigned long next_update = jiffies;
2508
2509         /* If we don't have any rings just leave ourselves set for maximum
2510          * possible latency so we take ourselves out of the equation.
2511          */
2512         if (!ring_container->ring)
2513                 return;
2514
2515         /* If we didn't update within up to 1 - 2 jiffies we can assume
2516          * that either packets are coming in so slow there hasn't been
2517          * any work, or that there is so much work that NAPI is dealing
2518          * with interrupt moderation and we don't need to do anything.
2519          */
2520         if (time_after(next_update, ring_container->next_update))
2521                 goto clear_counts;
2522
2523         packets = ring_container->total_packets;
2524
2525         /* We have no packets to actually measure against. This means
2526          * either one of the other queues on this vector is active or
2527          * we are a Tx queue doing TSO with too high of an interrupt rate.
2528          *
2529          * When this occurs just tick up our delay by the minimum value
2530          * and hope that this extra delay will prevent us from being called
2531          * without any work on our queue.
2532          */
2533         if (!packets) {
2534                 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2535                 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2536                         itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2537                 itr += ring_container->itr & IXGBE_ITR_ADAPTIVE_LATENCY;
2538                 goto clear_counts;
2539         }
2540
2541         bytes = ring_container->total_bytes;
2542
2543         /* If packets are less than 4 or bytes are less than 9000 assume
2544          * insufficient data to use bulk rate limiting approach. We are
2545          * likely latency driven.
2546          */
2547         if (packets < 4 && bytes < 9000) {
2548                 itr = IXGBE_ITR_ADAPTIVE_LATENCY;
2549                 goto adjust_by_size;
2550         }
2551
2552         /* Between 4 and 48 we can assume that our current interrupt delay
2553          * is only slightly too low. As such we should increase it by a small
2554          * fixed amount.
2555          */
2556         if (packets < 48) {
2557                 itr = (q_vector->itr >> 2) + IXGBE_ITR_ADAPTIVE_MIN_INC;
2558                 if (itr > IXGBE_ITR_ADAPTIVE_MAX_USECS)
2559                         itr = IXGBE_ITR_ADAPTIVE_MAX_USECS;
2560                 goto clear_counts;
2561         }
2562
2563         /* Between 48 and 96 is our "goldilocks" zone where we are working
2564          * out "just right". Just report that our current ITR is good for us.
2565          */
2566         if (packets < 96) {
2567                 itr = q_vector->itr >> 2;
2568                 goto clear_counts;
2569         }
2570
2571         /* If packet count is 96 or greater we are likely looking at a slight
2572          * overrun of the delay we want. Try halving our delay to see if that
2573          * will cut the number of packets in half per interrupt.
2574          */
2575         if (packets < 256) {
2576                 itr = q_vector->itr >> 3;
2577                 if (itr < IXGBE_ITR_ADAPTIVE_MIN_USECS)
2578                         itr = IXGBE_ITR_ADAPTIVE_MIN_USECS;
2579                 goto clear_counts;
2580         }
2581
2582         /* The paths below assume we are dealing with a bulk ITR since number
2583          * of packets is 256 or greater. We are just going to have to compute
2584          * a value and try to bring the count under control, though for smaller
2585          * packet sizes there isn't much we can do as NAPI polling will likely
2586          * be kicking in sooner rather than later.
2587          */
2588         itr = IXGBE_ITR_ADAPTIVE_BULK;
2589
2590 adjust_by_size:
2591         /* If packet counts are 256 or greater we can assume we have a gross
2592          * overestimation of what the rate should be. Instead of trying to fine
2593          * tune it just use the formula below to try and dial in an exact value
2594          * give the current packet size of the frame.
2595          */
2596         avg_wire_size = bytes / packets;
2597
2598         /* The following is a crude approximation of:
2599          *  wmem_default / (size + overhead) = desired_pkts_per_int
2600          *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
2601          *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
2602          *
2603          * Assuming wmem_default is 212992 and overhead is 640 bytes per
2604          * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
2605          * formula down to
2606          *
2607          *  (170 * (size + 24)) / (size + 640) = ITR
2608          *
2609          * We first do some math on the packet size and then finally bitshift
2610          * by 8 after rounding up. We also have to account for PCIe link speed
2611          * difference as ITR scales based on this.
2612          */
2613         if (avg_wire_size <= 60) {
2614                 /* Start at 50k ints/sec */
2615                 avg_wire_size = 5120;
2616         } else if (avg_wire_size <= 316) {
2617                 /* 50K ints/sec to 16K ints/sec */
2618                 avg_wire_size *= 40;
2619                 avg_wire_size += 2720;
2620         } else if (avg_wire_size <= 1084) {
2621                 /* 16K ints/sec to 9.2K ints/sec */
2622                 avg_wire_size *= 15;
2623                 avg_wire_size += 11452;
2624         } else if (avg_wire_size <= 1980) {
2625                 /* 9.2K ints/sec to 8K ints/sec */
2626                 avg_wire_size *= 5;
2627                 avg_wire_size += 22420;
2628         } else {
2629                 /* plateau at a limit of 8K ints/sec */
2630                 avg_wire_size = 32256;
2631         }
2632
2633         /* If we are in low latency mode half our delay which doubles the rate
2634          * to somewhere between 100K to 16K ints/sec
2635          */
2636         if (itr & IXGBE_ITR_ADAPTIVE_LATENCY)
2637                 avg_wire_size >>= 1;
2638
2639         /* Resultant value is 256 times larger than it needs to be. This
2640          * gives us room to adjust the value as needed to either increase
2641          * or decrease the value based on link speeds of 10G, 2.5G, 1G, etc.
2642          *
2643          * Use addition as we have already recorded the new latency flag
2644          * for the ITR value.
2645          */
2646         switch (q_vector->adapter->link_speed) {
2647         case IXGBE_LINK_SPEED_10GB_FULL:
2648         case IXGBE_LINK_SPEED_100_FULL:
2649         default:
2650                 itr += DIV_ROUND_UP(avg_wire_size,
2651                                     IXGBE_ITR_ADAPTIVE_MIN_INC * 256) *
2652                        IXGBE_ITR_ADAPTIVE_MIN_INC;
2653                 break;
2654         case IXGBE_LINK_SPEED_2_5GB_FULL:
2655         case IXGBE_LINK_SPEED_1GB_FULL:
2656         case IXGBE_LINK_SPEED_10_FULL:
2657                 itr += DIV_ROUND_UP(avg_wire_size,
2658                                     IXGBE_ITR_ADAPTIVE_MIN_INC * 64) *
2659                        IXGBE_ITR_ADAPTIVE_MIN_INC;
2660                 break;
2661         }
2662
2663 clear_counts:
2664         /* write back value */
2665         ring_container->itr = itr;
2666
2667         /* next update should occur within next jiffy */
2668         ring_container->next_update = next_update + 1;
2669
2670         ring_container->total_bytes = 0;
2671         ring_container->total_packets = 0;
2672 }
2673
2674 /**
2675  * ixgbe_write_eitr - write EITR register in hardware specific way
2676  * @q_vector: structure containing interrupt and ring information
2677  *
2678  * This function is made to be called by ethtool and by the driver
2679  * when it needs to update EITR registers at runtime.  Hardware
2680  * specific quirks/differences are taken care of here.
2681  */
2682 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2683 {
2684         struct ixgbe_adapter *adapter = q_vector->adapter;
2685         struct ixgbe_hw *hw = &adapter->hw;
2686         int v_idx = q_vector->v_idx;
2687         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2688
2689         switch (adapter->hw.mac.type) {
2690         case ixgbe_mac_82598EB:
2691                 /* must write high and low 16 bits to reset counter */
2692                 itr_reg |= (itr_reg << 16);
2693                 break;
2694         case ixgbe_mac_82599EB:
2695         case ixgbe_mac_X540:
2696         case ixgbe_mac_X550:
2697         case ixgbe_mac_X550EM_x:
2698         case ixgbe_mac_x550em_a:
2699                 /*
2700                  * set the WDIS bit to not clear the timer bits and cause an
2701                  * immediate assertion of the interrupt
2702                  */
2703                 itr_reg |= IXGBE_EITR_CNT_WDIS;
2704                 break;
2705         default:
2706                 break;
2707         }
2708         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2709 }
2710
2711 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2712 {
2713         u32 new_itr;
2714
2715         ixgbe_update_itr(q_vector, &q_vector->tx);
2716         ixgbe_update_itr(q_vector, &q_vector->rx);
2717
2718         /* use the smallest value of new ITR delay calculations */
2719         new_itr = min(q_vector->rx.itr, q_vector->tx.itr);
2720
2721         /* Clear latency flag if set, shift into correct position */
2722         new_itr &= ~IXGBE_ITR_ADAPTIVE_LATENCY;
2723         new_itr <<= 2;
2724
2725         if (new_itr != q_vector->itr) {
2726                 /* save the algorithm value here */
2727                 q_vector->itr = new_itr;
2728
2729                 ixgbe_write_eitr(q_vector);
2730         }
2731 }
2732
2733 /**
2734  * ixgbe_check_overtemp_subtask - check for over temperature
2735  * @adapter: pointer to adapter
2736  **/
2737 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2738 {
2739         struct ixgbe_hw *hw = &adapter->hw;
2740         u32 eicr = adapter->interrupt_event;
2741         s32 rc;
2742
2743         if (test_bit(__IXGBE_DOWN, &adapter->state))
2744                 return;
2745
2746         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2747                 return;
2748
2749         adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2750
2751         switch (hw->device_id) {
2752         case IXGBE_DEV_ID_82599_T3_LOM:
2753                 /*
2754                  * Since the warning interrupt is for both ports
2755                  * we don't have to check if:
2756                  *  - This interrupt wasn't for our port.
2757                  *  - We may have missed the interrupt so always have to
2758                  *    check if we  got a LSC
2759                  */
2760                 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2761                     !(eicr & IXGBE_EICR_LSC))
2762                         return;
2763
2764                 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2765                         u32 speed;
2766                         bool link_up = false;
2767
2768                         hw->mac.ops.check_link(hw, &speed, &link_up, false);
2769
2770                         if (link_up)
2771                                 return;
2772                 }
2773
2774                 /* Check if this is not due to overtemp */
2775                 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2776                         return;
2777
2778                 break;
2779         case IXGBE_DEV_ID_X550EM_A_1G_T:
2780         case IXGBE_DEV_ID_X550EM_A_1G_T_L:
2781                 rc = hw->phy.ops.check_overtemp(hw);
2782                 if (rc != IXGBE_ERR_OVERTEMP)
2783                         return;
2784                 break;
2785         default:
2786                 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2787                         return;
2788                 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2789                         return;
2790                 break;
2791         }
2792         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2793
2794         adapter->interrupt_event = 0;
2795 }
2796
2797 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2798 {
2799         struct ixgbe_hw *hw = &adapter->hw;
2800
2801         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2802             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2803                 e_crit(probe, "Fan has stopped, replace the adapter\n");
2804                 /* write to clear the interrupt */
2805                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2806         }
2807 }
2808
2809 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2810 {
2811         struct ixgbe_hw *hw = &adapter->hw;
2812
2813         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2814                 return;
2815
2816         switch (adapter->hw.mac.type) {
2817         case ixgbe_mac_82599EB:
2818                 /*
2819                  * Need to check link state so complete overtemp check
2820                  * on service task
2821                  */
2822                 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2823                      (eicr & IXGBE_EICR_LSC)) &&
2824                     (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2825                         adapter->interrupt_event = eicr;
2826                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2827                         ixgbe_service_event_schedule(adapter);
2828                         return;
2829                 }
2830                 return;
2831         case ixgbe_mac_x550em_a:
2832                 if (eicr & IXGBE_EICR_GPI_SDP0_X550EM_a) {
2833                         adapter->interrupt_event = eicr;
2834                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2835                         ixgbe_service_event_schedule(adapter);
2836                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC,
2837                                         IXGBE_EICR_GPI_SDP0_X550EM_a);
2838                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICR,
2839                                         IXGBE_EICR_GPI_SDP0_X550EM_a);
2840                 }
2841                 return;
2842         case ixgbe_mac_X550:
2843         case ixgbe_mac_X540:
2844                 if (!(eicr & IXGBE_EICR_TS))
2845                         return;
2846                 break;
2847         default:
2848                 return;
2849         }
2850
2851         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2852 }
2853
2854 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2855 {
2856         switch (hw->mac.type) {
2857         case ixgbe_mac_82598EB:
2858                 if (hw->phy.type == ixgbe_phy_nl)
2859                         return true;
2860                 return false;
2861         case ixgbe_mac_82599EB:
2862         case ixgbe_mac_X550EM_x:
2863         case ixgbe_mac_x550em_a:
2864                 switch (hw->mac.ops.get_media_type(hw)) {
2865                 case ixgbe_media_type_fiber:
2866                 case ixgbe_media_type_fiber_qsfp:
2867                         return true;
2868                 default:
2869                         return false;
2870                 }
2871         default:
2872                 return false;
2873         }
2874 }
2875
2876 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2877 {
2878         struct ixgbe_hw *hw = &adapter->hw;
2879         u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2880
2881         if (!ixgbe_is_sfp(hw))
2882                 return;
2883
2884         /* Later MAC's use different SDP */
2885         if (hw->mac.type >= ixgbe_mac_X540)
2886                 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2887
2888         if (eicr & eicr_mask) {
2889                 /* Clear the interrupt */
2890                 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2891                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2892                         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2893                         adapter->sfp_poll_time = 0;
2894                         ixgbe_service_event_schedule(adapter);
2895                 }
2896         }
2897
2898         if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2899             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2900                 /* Clear the interrupt */
2901                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2902                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2903                         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2904                         ixgbe_service_event_schedule(adapter);
2905                 }
2906         }
2907 }
2908
2909 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2910 {
2911         struct ixgbe_hw *hw = &adapter->hw;
2912
2913         adapter->lsc_int++;
2914         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2915         adapter->link_check_timeout = jiffies;
2916         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2917                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2918                 IXGBE_WRITE_FLUSH(hw);
2919                 ixgbe_service_event_schedule(adapter);
2920         }
2921 }
2922
2923 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2924                                            u64 qmask)
2925 {
2926         u32 mask;
2927         struct ixgbe_hw *hw = &adapter->hw;
2928
2929         switch (hw->mac.type) {
2930         case ixgbe_mac_82598EB:
2931                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2932                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2933                 break;
2934         case ixgbe_mac_82599EB:
2935         case ixgbe_mac_X540:
2936         case ixgbe_mac_X550:
2937         case ixgbe_mac_X550EM_x:
2938         case ixgbe_mac_x550em_a:
2939                 mask = (qmask & 0xFFFFFFFF);
2940                 if (mask)
2941                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2942                 mask = (qmask >> 32);
2943                 if (mask)
2944                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2945                 break;
2946         default:
2947                 break;
2948         }
2949         /* skip the flush */
2950 }
2951
2952 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2953                                             u64 qmask)
2954 {
2955         u32 mask;
2956         struct ixgbe_hw *hw = &adapter->hw;
2957
2958         switch (hw->mac.type) {
2959         case ixgbe_mac_82598EB:
2960                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2961                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2962                 break;
2963         case ixgbe_mac_82599EB:
2964         case ixgbe_mac_X540:
2965         case ixgbe_mac_X550:
2966         case ixgbe_mac_X550EM_x:
2967         case ixgbe_mac_x550em_a:
2968                 mask = (qmask & 0xFFFFFFFF);
2969                 if (mask)
2970                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2971                 mask = (qmask >> 32);
2972                 if (mask)
2973                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2974                 break;
2975         default:
2976                 break;
2977         }
2978         /* skip the flush */
2979 }
2980
2981 /**
2982  * ixgbe_irq_enable - Enable default interrupt generation settings
2983  * @adapter: board private structure
2984  * @queues: enable irqs for queues
2985  * @flush: flush register write
2986  **/
2987 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2988                                     bool flush)
2989 {
2990         struct ixgbe_hw *hw = &adapter->hw;
2991         u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2992
2993         /* don't reenable LSC while waiting for link */
2994         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2995                 mask &= ~IXGBE_EIMS_LSC;
2996
2997         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2998                 switch (adapter->hw.mac.type) {
2999                 case ixgbe_mac_82599EB:
3000                         mask |= IXGBE_EIMS_GPI_SDP0(hw);
3001                         break;
3002                 case ixgbe_mac_X540:
3003                 case ixgbe_mac_X550:
3004                 case ixgbe_mac_X550EM_x:
3005                 case ixgbe_mac_x550em_a:
3006                         mask |= IXGBE_EIMS_TS;
3007                         break;
3008                 default:
3009                         break;
3010                 }
3011         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
3012                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3013         switch (adapter->hw.mac.type) {
3014         case ixgbe_mac_82599EB:
3015                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
3016                 mask |= IXGBE_EIMS_GPI_SDP2(hw);
3017                 /* fall through */
3018         case ixgbe_mac_X540:
3019         case ixgbe_mac_X550:
3020         case ixgbe_mac_X550EM_x:
3021         case ixgbe_mac_x550em_a:
3022                 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP ||
3023                     adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP ||
3024                     adapter->hw.device_id == IXGBE_DEV_ID_X550EM_A_SFP_N)
3025                         mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
3026                 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
3027                         mask |= IXGBE_EICR_GPI_SDP0_X540;
3028                 mask |= IXGBE_EIMS_ECC;
3029                 mask |= IXGBE_EIMS_MAILBOX;
3030                 break;
3031         default:
3032                 break;
3033         }
3034
3035         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
3036             !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
3037                 mask |= IXGBE_EIMS_FLOW_DIR;
3038
3039         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
3040         if (queues)
3041                 ixgbe_irq_enable_queues(adapter, ~0);
3042         if (flush)
3043                 IXGBE_WRITE_FLUSH(&adapter->hw);
3044 }
3045
3046 static irqreturn_t ixgbe_msix_other(int irq, void *data)
3047 {
3048         struct ixgbe_adapter *adapter = data;
3049         struct ixgbe_hw *hw = &adapter->hw;
3050         u32 eicr;
3051
3052         /*
3053          * Workaround for Silicon errata.  Use clear-by-write instead
3054          * of clear-by-read.  Reading with EICS will return the
3055          * interrupt causes without clearing, which later be done
3056          * with the write to EICR.
3057          */
3058         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
3059
3060         /* The lower 16bits of the EICR register are for the queue interrupts
3061          * which should be masked here in order to not accidentally clear them if
3062          * the bits are high when ixgbe_msix_other is called. There is a race
3063          * condition otherwise which results in possible performance loss
3064          * especially if the ixgbe_msix_other interrupt is triggering
3065          * consistently (as it would when PPS is turned on for the X540 device)
3066          */
3067         eicr &= 0xFFFF0000;
3068
3069         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
3070
3071         if (eicr & IXGBE_EICR_LSC)
3072                 ixgbe_check_lsc(adapter);
3073
3074         if (eicr & IXGBE_EICR_MAILBOX)
3075                 ixgbe_msg_task(adapter);
3076
3077         switch (hw->mac.type) {
3078         case ixgbe_mac_82599EB:
3079         case ixgbe_mac_X540:
3080         case ixgbe_mac_X550:
3081         case ixgbe_mac_X550EM_x:
3082         case ixgbe_mac_x550em_a:
3083                 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
3084                     (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
3085                         adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
3086                         ixgbe_service_event_schedule(adapter);
3087                         IXGBE_WRITE_REG(hw, IXGBE_EICR,
3088                                         IXGBE_EICR_GPI_SDP0_X540);
3089                 }
3090                 if (eicr & IXGBE_EICR_ECC) {
3091                         e_info(link, "Received ECC Err, initiating reset\n");
3092                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3093                         ixgbe_service_event_schedule(adapter);
3094                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3095                 }
3096                 /* Handle Flow Director Full threshold interrupt */
3097                 if (eicr & IXGBE_EICR_FLOW_DIR) {
3098                         int reinit_count = 0;
3099                         int i;
3100                         for (i = 0; i < adapter->num_tx_queues; i++) {
3101                                 struct ixgbe_ring *ring = adapter->tx_ring[i];
3102                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
3103                                                        &ring->state))
3104                                         reinit_count++;
3105                         }
3106                         if (reinit_count) {
3107                                 /* no more flow director interrupts until after init */
3108                                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
3109                                 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
3110                                 ixgbe_service_event_schedule(adapter);
3111                         }
3112                 }
3113                 ixgbe_check_sfp_event(adapter, eicr);
3114                 ixgbe_check_overtemp_event(adapter, eicr);
3115                 break;
3116         default:
3117                 break;
3118         }
3119
3120         ixgbe_check_fan_failure(adapter, eicr);
3121
3122         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3123                 ixgbe_ptp_check_pps_event(adapter);
3124
3125         /* re-enable the original interrupt state, no lsc, no queues */
3126         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3127                 ixgbe_irq_enable(adapter, false, false);
3128
3129         return IRQ_HANDLED;
3130 }
3131
3132 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
3133 {
3134         struct ixgbe_q_vector *q_vector = data;
3135
3136         /* EIAM disabled interrupts (on this vector) for us */
3137
3138         if (q_vector->rx.ring || q_vector->tx.ring)
3139                 napi_schedule_irqoff(&q_vector->napi);
3140
3141         return IRQ_HANDLED;
3142 }
3143
3144 /**
3145  * ixgbe_poll - NAPI Rx polling callback
3146  * @napi: structure for representing this polling device
3147  * @budget: how many packets driver is allowed to clean
3148  *
3149  * This function is used for legacy and MSI, NAPI mode
3150  **/
3151 int ixgbe_poll(struct napi_struct *napi, int budget)
3152 {
3153         struct ixgbe_q_vector *q_vector =
3154                                 container_of(napi, struct ixgbe_q_vector, napi);
3155         struct ixgbe_adapter *adapter = q_vector->adapter;
3156         struct ixgbe_ring *ring;
3157         int per_ring_budget, work_done = 0;
3158         bool clean_complete = true;
3159
3160 #ifdef CONFIG_IXGBE_DCA
3161         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
3162                 ixgbe_update_dca(q_vector);
3163 #endif
3164
3165         ixgbe_for_each_ring(ring, q_vector->tx) {
3166                 bool wd = ring->xsk_umem ?
3167                           ixgbe_clean_xdp_tx_irq(q_vector, ring, budget) :
3168                           ixgbe_clean_tx_irq(q_vector, ring, budget);
3169
3170                 if (!wd)
3171                         clean_complete = false;
3172         }
3173
3174         /* Exit if we are called by netpoll */
3175         if (budget <= 0)
3176                 return budget;
3177
3178         /* attempt to distribute budget to each queue fairly, but don't allow
3179          * the budget to go below 1 because we'll exit polling */
3180         if (q_vector->rx.count > 1)
3181                 per_ring_budget = max(budget/q_vector->rx.count, 1);
3182         else
3183                 per_ring_budget = budget;
3184
3185         ixgbe_for_each_ring(ring, q_vector->rx) {
3186                 int cleaned = ring->xsk_umem ?
3187                               ixgbe_clean_rx_irq_zc(q_vector, ring,
3188                                                     per_ring_budget) :
3189                               ixgbe_clean_rx_irq(q_vector, ring,
3190                                                  per_ring_budget);
3191
3192                 work_done += cleaned;
3193                 if (cleaned >= per_ring_budget)
3194                         clean_complete = false;
3195         }
3196
3197         /* If all work not completed, return budget and keep polling */
3198         if (!clean_complete)
3199                 return budget;
3200
3201         /* all work done, exit the polling mode */
3202         if (likely(napi_complete_done(napi, work_done))) {
3203                 if (adapter->rx_itr_setting & 1)
3204                         ixgbe_set_itr(q_vector);
3205                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3206                         ixgbe_irq_enable_queues(adapter,
3207                                                 BIT_ULL(q_vector->v_idx));
3208         }
3209
3210         return min(work_done, budget - 1);
3211 }
3212
3213 /**
3214  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
3215  * @adapter: board private structure
3216  *
3217  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
3218  * interrupts from the kernel.
3219  **/
3220 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
3221 {
3222         struct net_device *netdev = adapter->netdev;
3223         unsigned int ri = 0, ti = 0;
3224         int vector, err;
3225
3226         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3227                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3228                 struct msix_entry *entry = &adapter->msix_entries[vector];
3229
3230                 if (q_vector->tx.ring && q_vector->rx.ring) {
3231                         snprintf(q_vector->name, sizeof(q_vector->name),
3232                                  "%s-TxRx-%u", netdev->name, ri++);
3233                         ti++;
3234                 } else if (q_vector->rx.ring) {
3235                         snprintf(q_vector->name, sizeof(q_vector->name),
3236                                  "%s-rx-%u", netdev->name, ri++);
3237                 } else if (q_vector->tx.ring) {
3238                         snprintf(q_vector->name, sizeof(q_vector->name),
3239                                  "%s-tx-%u", netdev->name, ti++);
3240                 } else {
3241                         /* skip this unused q_vector */
3242                         continue;
3243                 }
3244                 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
3245                                   q_vector->name, q_vector);
3246                 if (err) {
3247                         e_err(probe, "request_irq failed for MSIX interrupt "
3248                               "Error: %d\n", err);
3249                         goto free_queue_irqs;
3250                 }
3251                 /* If Flow Director is enabled, set interrupt affinity */
3252                 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3253                         /* assign the mask for this irq */
3254                         irq_set_affinity_hint(entry->vector,
3255                                               &q_vector->affinity_mask);
3256                 }
3257         }
3258
3259         err = request_irq(adapter->msix_entries[vector].vector,
3260                           ixgbe_msix_other, 0, netdev->name, adapter);
3261         if (err) {
3262                 e_err(probe, "request_irq for msix_other failed: %d\n", err);
3263                 goto free_queue_irqs;
3264         }
3265
3266         return 0;
3267
3268 free_queue_irqs:
3269         while (vector) {
3270                 vector--;
3271                 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
3272                                       NULL);
3273                 free_irq(adapter->msix_entries[vector].vector,
3274                          adapter->q_vector[vector]);
3275         }
3276         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
3277         pci_disable_msix(adapter->pdev);
3278         kfree(adapter->msix_entries);
3279         adapter->msix_entries = NULL;
3280         return err;
3281 }
3282
3283 /**
3284  * ixgbe_intr - legacy mode Interrupt Handler
3285  * @irq: interrupt number
3286  * @data: pointer to a network interface device structure
3287  **/
3288 static irqreturn_t ixgbe_intr(int irq, void *data)
3289 {
3290         struct ixgbe_adapter *adapter = data;
3291         struct ixgbe_hw *hw = &adapter->hw;
3292         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3293         u32 eicr;
3294
3295         /*
3296          * Workaround for silicon errata #26 on 82598.  Mask the interrupt
3297          * before the read of EICR.
3298          */
3299         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
3300
3301         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
3302          * therefore no explicit interrupt disable is necessary */
3303         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
3304         if (!eicr) {
3305                 /*
3306                  * shared interrupt alert!
3307                  * make sure interrupts are enabled because the read will
3308                  * have disabled interrupts due to EIAM
3309                  * finish the workaround of silicon errata on 82598.  Unmask
3310                  * the interrupt that we masked before the EICR read.
3311                  */
3312                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
3313                         ixgbe_irq_enable(adapter, true, true);
3314                 return IRQ_NONE;        /* Not our interrupt */
3315         }
3316
3317         if (eicr & IXGBE_EICR_LSC)
3318                 ixgbe_check_lsc(adapter);
3319
3320         switch (hw->mac.type) {
3321         case ixgbe_mac_82599EB:
3322                 ixgbe_check_sfp_event(adapter, eicr);
3323                 /* Fall through */
3324         case ixgbe_mac_X540:
3325         case ixgbe_mac_X550:
3326         case ixgbe_mac_X550EM_x:
3327         case ixgbe_mac_x550em_a:
3328                 if (eicr & IXGBE_EICR_ECC) {
3329                         e_info(link, "Received ECC Err, initiating reset\n");
3330                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
3331                         ixgbe_service_event_schedule(adapter);
3332                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
3333                 }
3334                 ixgbe_check_overtemp_event(adapter, eicr);
3335                 break;
3336         default:
3337                 break;
3338         }
3339
3340         ixgbe_check_fan_failure(adapter, eicr);
3341         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
3342                 ixgbe_ptp_check_pps_event(adapter);
3343
3344         /* would disable interrupts here but EIAM disabled it */
3345         napi_schedule_irqoff(&q_vector->napi);
3346
3347         /*
3348          * re-enable link(maybe) and non-queue interrupts, no flush.
3349          * ixgbe_poll will re-enable the queue interrupts
3350          */
3351         if (!test_bit(__IXGBE_DOWN, &adapter->state))
3352                 ixgbe_irq_enable(adapter, false, false);
3353
3354         return IRQ_HANDLED;
3355 }
3356
3357 /**
3358  * ixgbe_request_irq - initialize interrupts
3359  * @adapter: board private structure
3360  *
3361  * Attempts to configure interrupts using the best available
3362  * capabilities of the hardware and kernel.
3363  **/
3364 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
3365 {
3366         struct net_device *netdev = adapter->netdev;
3367         int err;
3368
3369         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
3370                 err = ixgbe_request_msix_irqs(adapter);
3371         else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
3372                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
3373                                   netdev->name, adapter);
3374         else
3375                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
3376                                   netdev->name, adapter);
3377
3378         if (err)
3379                 e_err(probe, "request_irq failed, Error %d\n", err);
3380
3381         return err;
3382 }
3383
3384 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
3385 {
3386         int vector;
3387
3388         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3389                 free_irq(adapter->pdev->irq, adapter);
3390                 return;
3391         }
3392
3393         if (!adapter->msix_entries)
3394                 return;
3395
3396         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3397                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3398                 struct msix_entry *entry = &adapter->msix_entries[vector];
3399
3400                 /* free only the irqs that were actually requested */
3401                 if (!q_vector->rx.ring && !q_vector->tx.ring)
3402                         continue;
3403
3404                 /* clear the affinity_mask in the IRQ descriptor */
3405                 irq_set_affinity_hint(entry->vector, NULL);
3406
3407                 free_irq(entry->vector, q_vector);
3408         }
3409
3410         free_irq(adapter->msix_entries[vector].vector, adapter);
3411 }
3412
3413 /**
3414  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3415  * @adapter: board private structure
3416  **/
3417 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3418 {
3419         switch (adapter->hw.mac.type) {
3420         case ixgbe_mac_82598EB:
3421                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3422                 break;
3423         case ixgbe_mac_82599EB:
3424         case ixgbe_mac_X540:
3425         case ixgbe_mac_X550:
3426         case ixgbe_mac_X550EM_x:
3427         case ixgbe_mac_x550em_a:
3428                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3429                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3430                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3431                 break;
3432         default:
3433                 break;
3434         }
3435         IXGBE_WRITE_FLUSH(&adapter->hw);
3436         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3437                 int vector;
3438
3439                 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3440                         synchronize_irq(adapter->msix_entries[vector].vector);
3441
3442                 synchronize_irq(adapter->msix_entries[vector++].vector);
3443         } else {
3444                 synchronize_irq(adapter->pdev->irq);
3445         }
3446 }
3447
3448 /**
3449  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3450  * @adapter: board private structure
3451  *
3452  **/
3453 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3454 {
3455         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3456
3457         ixgbe_write_eitr(q_vector);
3458
3459         ixgbe_set_ivar(adapter, 0, 0, 0);
3460         ixgbe_set_ivar(adapter, 1, 0, 0);
3461
3462         e_info(hw, "Legacy interrupt IVAR setup done\n");
3463 }
3464
3465 /**
3466  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3467  * @adapter: board private structure
3468  * @ring: structure containing ring specific data
3469  *
3470  * Configure the Tx descriptor ring after a reset.
3471  **/
3472 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3473                              struct ixgbe_ring *ring)
3474 {
3475         struct ixgbe_hw *hw = &adapter->hw;
3476         u64 tdba = ring->dma;
3477         int wait_loop = 10;
3478         u32 txdctl = IXGBE_TXDCTL_ENABLE;
3479         u8 reg_idx = ring->reg_idx;
3480
3481         ring->xsk_umem = NULL;
3482         if (ring_is_xdp(ring))
3483                 ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
3484
3485         /* disable queue to avoid issues while updating state */
3486         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3487         IXGBE_WRITE_FLUSH(hw);
3488
3489         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3490                         (tdba & DMA_BIT_MASK(32)));
3491         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3492         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3493                         ring->count * sizeof(union ixgbe_adv_tx_desc));
3494         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3495         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3496         ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3497
3498         /*
3499          * set WTHRESH to encourage burst writeback, it should not be set
3500          * higher than 1 when:
3501          * - ITR is 0 as it could cause false TX hangs
3502          * - ITR is set to > 100k int/sec and BQL is enabled
3503          *
3504          * In order to avoid issues WTHRESH + PTHRESH should always be equal
3505          * to or less than the number of on chip descriptors, which is
3506          * currently 40.
3507          */
3508         if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3509                 txdctl |= 1u << 16;     /* WTHRESH = 1 */
3510         else
3511                 txdctl |= 8u << 16;     /* WTHRESH = 8 */
3512
3513         /*
3514          * Setting PTHRESH to 32 both improves performance
3515          * and avoids a TX hang with DFP enabled
3516          */
3517         txdctl |= (1u << 8) |   /* HTHRESH = 1 */
3518                    32;          /* PTHRESH = 32 */
3519
3520         /* reinitialize flowdirector state */
3521         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3522                 ring->atr_sample_rate = adapter->atr_sample_rate;
3523                 ring->atr_count = 0;
3524                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3525         } else {
3526                 ring->atr_sample_rate = 0;
3527         }
3528
3529         /* initialize XPS */
3530         if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3531                 struct ixgbe_q_vector *q_vector = ring->q_vector;
3532
3533                 if (q_vector)
3534                         netif_set_xps_queue(ring->netdev,
3535                                             &q_vector->affinity_mask,
3536                                             ring->queue_index);
3537         }
3538
3539         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3540
3541         /* reinitialize tx_buffer_info */
3542         memset(ring->tx_buffer_info, 0,
3543                sizeof(struct ixgbe_tx_buffer) * ring->count);
3544
3545         /* enable queue */
3546         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3547
3548         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3549         if (hw->mac.type == ixgbe_mac_82598EB &&
3550             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3551                 return;
3552
3553         /* poll to verify queue is enabled */
3554         do {
3555                 usleep_range(1000, 2000);
3556                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3557         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3558         if (!wait_loop)
3559                 hw_dbg(hw, "Could not enable Tx Queue %d\n", reg_idx);
3560 }
3561
3562 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3563 {
3564         struct ixgbe_hw *hw = &adapter->hw;
3565         u32 rttdcs, mtqc;
3566         u8 tcs = adapter->hw_tcs;
3567
3568         if (hw->mac.type == ixgbe_mac_82598EB)
3569                 return;
3570
3571         /* disable the arbiter while setting MTQC */
3572         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3573         rttdcs |= IXGBE_RTTDCS_ARBDIS;
3574         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3575
3576         /* set transmit pool layout */
3577         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3578                 mtqc = IXGBE_MTQC_VT_ENA;
3579                 if (tcs > 4)
3580                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3581                 else if (tcs > 1)
3582                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3583                 else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3584                          IXGBE_82599_VMDQ_4Q_MASK)
3585                         mtqc |= IXGBE_MTQC_32VF;
3586                 else
3587                         mtqc |= IXGBE_MTQC_64VF;
3588         } else {
3589                 if (tcs > 4) {
3590                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3591                 } else if (tcs > 1) {
3592                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3593                 } else {
3594                         u8 max_txq = adapter->num_tx_queues +
3595                                 adapter->num_xdp_queues;
3596                         if (max_txq > 63)
3597                                 mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3598                         else
3599                                 mtqc = IXGBE_MTQC_64Q_1PB;
3600                 }
3601         }
3602
3603         IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3604
3605         /* Enable Security TX Buffer IFG for multiple pb */
3606         if (tcs) {
3607                 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3608                 sectx |= IXGBE_SECTX_DCB;
3609                 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3610         }
3611
3612         /* re-enable the arbiter */
3613         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3614         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3615 }
3616
3617 /**
3618  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3619  * @adapter: board private structure
3620  *
3621  * Configure the Tx unit of the MAC after a reset.
3622  **/
3623 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3624 {
3625         struct ixgbe_hw *hw = &adapter->hw;
3626         u32 dmatxctl;
3627         u32 i;
3628
3629         ixgbe_setup_mtqc(adapter);
3630
3631         if (hw->mac.type != ixgbe_mac_82598EB) {
3632                 /* DMATXCTL.EN must be before Tx queues are enabled */
3633                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3634                 dmatxctl |= IXGBE_DMATXCTL_TE;
3635                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3636         }
3637
3638         /* Setup the HW Tx Head and Tail descriptor pointers */
3639         for (i = 0; i < adapter->num_tx_queues; i++)
3640                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3641         for (i = 0; i < adapter->num_xdp_queues; i++)
3642                 ixgbe_configure_tx_ring(adapter, adapter->xdp_ring[i]);
3643 }
3644
3645 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3646                                  struct ixgbe_ring *ring)
3647 {
3648         struct ixgbe_hw *hw = &adapter->hw;
3649         u8 reg_idx = ring->reg_idx;
3650         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3651
3652         srrctl |= IXGBE_SRRCTL_DROP_EN;
3653
3654         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3655 }
3656
3657 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3658                                   struct ixgbe_ring *ring)
3659 {
3660         struct ixgbe_hw *hw = &adapter->hw;
3661         u8 reg_idx = ring->reg_idx;
3662         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3663
3664         srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3665
3666         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3667 }
3668
3669 #ifdef CONFIG_IXGBE_DCB
3670 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3671 #else
3672 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3673 #endif
3674 {
3675         int i;
3676         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3677
3678         if (adapter->ixgbe_ieee_pfc)
3679                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3680
3681         /*
3682          * We should set the drop enable bit if:
3683          *  SR-IOV is enabled
3684          *   or
3685          *  Number of Rx queues > 1 and flow control is disabled
3686          *
3687          *  This allows us to avoid head of line blocking for security
3688          *  and performance reasons.
3689          */
3690         if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3691             !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3692                 for (i = 0; i < adapter->num_rx_queues; i++)
3693                         ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3694         } else {
3695                 for (i = 0; i < adapter->num_rx_queues; i++)
3696                         ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3697         }
3698 }
3699
3700 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3701
3702 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3703                                    struct ixgbe_ring *rx_ring)
3704 {
3705         struct ixgbe_hw *hw = &adapter->hw;
3706         u32 srrctl;
3707         u8 reg_idx = rx_ring->reg_idx;
3708
3709         if (hw->mac.type == ixgbe_mac_82598EB) {
3710                 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3711
3712                 /*
3713                  * if VMDq is not active we must program one srrctl register
3714                  * per RSS queue since we have enabled RDRXCTL.MVMEN
3715                  */
3716                 reg_idx &= mask;
3717         }
3718
3719         /* configure header buffer length, needed for RSC */
3720         srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3721
3722         /* configure the packet buffer length */
3723         if (rx_ring->xsk_umem) {
3724                 u32 xsk_buf_len = rx_ring->xsk_umem->chunk_size_nohr -
3725                                   XDP_PACKET_HEADROOM;
3726
3727                 /* If the MAC support setting RXDCTL.RLPML, the
3728                  * SRRCTL[n].BSIZEPKT is set to PAGE_SIZE and
3729                  * RXDCTL.RLPML is set to the actual UMEM buffer
3730                  * size. If not, then we are stuck with a 1k buffer
3731                  * size resolution. In this case frames larger than
3732                  * the UMEM buffer size viewed in a 1k resolution will
3733                  * be dropped.
3734                  */
3735                 if (hw->mac.type != ixgbe_mac_82599EB)
3736                         srrctl |= PAGE_SIZE >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3737                 else
3738                         srrctl |= xsk_buf_len >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3739         } else if (test_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state)) {
3740                 srrctl |= IXGBE_RXBUFFER_3K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3741         } else {
3742                 srrctl |= IXGBE_RXBUFFER_2K >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3743         }
3744
3745         /* configure descriptor type */
3746         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3747
3748         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3749 }
3750
3751 /**
3752  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3753  * @adapter: device handle
3754  *
3755  *  - 82598/82599/X540:     128
3756  *  - X550(non-SRIOV mode): 512
3757  *  - X550(SRIOV mode):     64
3758  */
3759 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3760 {
3761         if (adapter->hw.mac.type < ixgbe_mac_X550)
3762                 return 128;
3763         else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3764                 return 64;
3765         else
3766                 return 512;
3767 }
3768
3769 /**
3770  * ixgbe_store_key - Write the RSS key to HW
3771  * @adapter: device handle
3772  *
3773  * Write the RSS key stored in adapter.rss_key to HW.
3774  */
3775 void ixgbe_store_key(struct ixgbe_adapter *adapter)
3776 {
3777         struct ixgbe_hw *hw = &adapter->hw;
3778         int i;
3779
3780         for (i = 0; i < 10; i++)
3781                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3782 }
3783
3784 /**
3785  * ixgbe_init_rss_key - Initialize adapter RSS key
3786  * @adapter: device handle
3787  *
3788  * Allocates and initializes the RSS key if it is not allocated.
3789  **/
3790 static inline int ixgbe_init_rss_key(struct ixgbe_adapter *adapter)
3791 {
3792         u32 *rss_key;
3793
3794         if (!adapter->rss_key) {
3795                 rss_key = kzalloc(IXGBE_RSS_KEY_SIZE, GFP_KERNEL);
3796                 if (unlikely(!rss_key))
3797                         return -ENOMEM;
3798
3799                 netdev_rss_key_fill(rss_key, IXGBE_RSS_KEY_SIZE);
3800                 adapter->rss_key = rss_key;
3801         }
3802
3803         return 0;
3804 }
3805
3806 /**
3807  * ixgbe_store_reta - Write the RETA table to HW
3808  * @adapter: device handle
3809  *
3810  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3811  */
3812 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3813 {
3814         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3815         struct ixgbe_hw *hw = &adapter->hw;
3816         u32 reta = 0;
3817         u32 indices_multi;
3818         u8 *indir_tbl = adapter->rss_indir_tbl;
3819
3820         /* Fill out the redirection table as follows:
3821          *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3822          *    indices.
3823          *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3824          *  - X550:       8 bit wide entries containing 6 bit RSS index
3825          */
3826         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3827                 indices_multi = 0x11;
3828         else
3829                 indices_multi = 0x1;
3830
3831         /* Write redirection table to HW */
3832         for (i = 0; i < reta_entries; i++) {
3833                 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3834                 if ((i & 3) == 3) {
3835                         if (i < 128)
3836                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3837                         else
3838                                 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3839                                                 reta);
3840                         reta = 0;
3841                 }
3842         }
3843 }
3844
3845 /**
3846  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3847  * @adapter: device handle
3848  *
3849  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3850  */
3851 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3852 {
3853         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3854         struct ixgbe_hw *hw = &adapter->hw;
3855         u32 vfreta = 0;
3856
3857         /* Write redirection table to HW */
3858         for (i = 0; i < reta_entries; i++) {
3859                 u16 pool = adapter->num_rx_pools;
3860
3861                 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3862                 if ((i & 3) != 3)
3863                         continue;
3864
3865                 while (pool--)
3866                         IXGBE_WRITE_REG(hw,
3867                                         IXGBE_PFVFRETA(i >> 2, VMDQ_P(pool)),
3868                                         vfreta);
3869                 vfreta = 0;
3870         }
3871 }
3872
3873 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3874 {
3875         u32 i, j;
3876         u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3877         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3878
3879         /* Program table for at least 4 queues w/ SR-IOV so that VFs can
3880          * make full use of any rings they may have.  We will use the
3881          * PSRTYPE register to control how many rings we use within the PF.
3882          */
3883         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 4))
3884                 rss_i = 4;
3885
3886         /* Fill out hash function seeds */
3887         ixgbe_store_key(adapter);
3888
3889         /* Fill out redirection table */
3890         memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3891
3892         for (i = 0, j = 0; i < reta_entries; i++, j++) {
3893                 if (j == rss_i)
3894                         j = 0;
3895
3896                 adapter->rss_indir_tbl[i] = j;
3897         }
3898
3899         ixgbe_store_reta(adapter);
3900 }
3901
3902 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3903 {
3904         struct ixgbe_hw *hw = &adapter->hw;
3905         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3906         int i, j;
3907
3908         /* Fill out hash function seeds */
3909         for (i = 0; i < 10; i++) {
3910                 u16 pool = adapter->num_rx_pools;
3911
3912                 while (pool--)
3913                         IXGBE_WRITE_REG(hw,
3914                                         IXGBE_PFVFRSSRK(i, VMDQ_P(pool)),
3915                                         *(adapter->rss_key + i));
3916         }
3917
3918         /* Fill out the redirection table */
3919         for (i = 0, j = 0; i < 64; i++, j++) {
3920                 if (j == rss_i)
3921                         j = 0;
3922
3923                 adapter->rss_indir_tbl[i] = j;
3924         }
3925
3926         ixgbe_store_vfreta(adapter);
3927 }
3928
3929 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3930 {
3931         struct ixgbe_hw *hw = &adapter->hw;
3932         u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3933         u32 rxcsum;
3934
3935         /* Disable indicating checksum in descriptor, enables RSS hash */
3936         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3937         rxcsum |= IXGBE_RXCSUM_PCSD;
3938         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3939
3940         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3941                 if (adapter->ring_feature[RING_F_RSS].mask)
3942                         mrqc = IXGBE_MRQC_RSSEN;
3943         } else {
3944                 u8 tcs = adapter->hw_tcs;
3945
3946                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3947                         if (tcs > 4)
3948                                 mrqc = IXGBE_MRQC_VMDQRT8TCEN;  /* 8 TCs */
3949                         else if (tcs > 1)
3950                                 mrqc = IXGBE_MRQC_VMDQRT4TCEN;  /* 4 TCs */
3951                         else if (adapter->ring_feature[RING_F_VMDQ].mask ==
3952                                  IXGBE_82599_VMDQ_4Q_MASK)
3953                                 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3954                         else
3955                                 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3956
3957                         /* Enable L3/L4 for Tx Switched packets only for X550,
3958                          * older devices do not support this feature
3959                          */
3960                         if (hw->mac.type >= ixgbe_mac_X550)
3961                                 mrqc |= IXGBE_MRQC_L3L4TXSWEN;
3962                 } else {
3963                         if (tcs > 4)
3964                                 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3965                         else if (tcs > 1)
3966                                 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3967                         else
3968                                 mrqc = IXGBE_MRQC_RSSEN;
3969                 }
3970         }
3971
3972         /* Perform hash on these packet types */
3973         rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3974                      IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3975                      IXGBE_MRQC_RSS_FIELD_IPV6 |
3976                      IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3977
3978         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3979                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3980         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3981                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3982
3983         if ((hw->mac.type >= ixgbe_mac_X550) &&
3984             (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3985                 u16 pool = adapter->num_rx_pools;
3986
3987                 /* Enable VF RSS mode */
3988                 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3989                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3990
3991                 /* Setup RSS through the VF registers */
3992                 ixgbe_setup_vfreta(adapter);
3993                 vfmrqc = IXGBE_MRQC_RSSEN;
3994                 vfmrqc |= rss_field;
3995
3996                 while (pool--)
3997                         IXGBE_WRITE_REG(hw,
3998                                         IXGBE_PFVFMRQC(VMDQ_P(pool)),
3999                                         vfmrqc);
4000         } else {
4001                 ixgbe_setup_reta(adapter);
4002                 mrqc |= rss_field;
4003                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
4004         }
4005 }
4006
4007 /**
4008  * ixgbe_configure_rscctl - enable RSC for the indicated ring
4009  * @adapter: address of board private structure
4010  * @ring: structure containing ring specific data
4011  **/
4012 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
4013                                    struct ixgbe_ring *ring)
4014 {
4015         struct ixgbe_hw *hw = &adapter->hw;
4016         u32 rscctrl;
4017         u8 reg_idx = ring->reg_idx;
4018
4019         if (!ring_is_rsc_enabled(ring))
4020                 return;
4021
4022         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
4023         rscctrl |= IXGBE_RSCCTL_RSCEN;
4024         /*
4025          * we must limit the number of descriptors so that the
4026          * total size of max desc * buf_len is not greater
4027          * than 65536
4028          */
4029         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
4030         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
4031 }
4032
4033 #define IXGBE_MAX_RX_DESC_POLL 10
4034 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
4035                                        struct ixgbe_ring *ring)
4036 {
4037         struct ixgbe_hw *hw = &adapter->hw;
4038         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
4039         u32 rxdctl;
4040         u8 reg_idx = ring->reg_idx;
4041
4042         if (ixgbe_removed(hw->hw_addr))
4043                 return;
4044         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
4045         if (hw->mac.type == ixgbe_mac_82598EB &&
4046             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
4047                 return;
4048
4049         do {
4050                 usleep_range(1000, 2000);
4051                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4052         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
4053
4054         if (!wait_loop) {
4055                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
4056                       "the polling period\n", reg_idx);
4057         }
4058 }
4059
4060 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
4061                              struct ixgbe_ring *ring)
4062 {
4063         struct ixgbe_hw *hw = &adapter->hw;
4064         union ixgbe_adv_rx_desc *rx_desc;
4065         u64 rdba = ring->dma;
4066         u32 rxdctl;
4067         u8 reg_idx = ring->reg_idx;
4068
4069         xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4070         ring->xsk_umem = ixgbe_xsk_umem(adapter, ring);
4071         if (ring->xsk_umem) {
4072                 ring->zca.free = ixgbe_zca_free;
4073                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4074                                                    MEM_TYPE_ZERO_COPY,
4075                                                    &ring->zca));
4076
4077         } else {
4078                 WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4079                                                    MEM_TYPE_PAGE_SHARED, NULL));
4080         }
4081
4082         /* disable queue to avoid use of these values while updating state */
4083         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
4084         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
4085
4086         /* write value back with RXDCTL.ENABLE bit cleared */
4087         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4088         IXGBE_WRITE_FLUSH(hw);
4089
4090         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
4091         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
4092         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
4093                         ring->count * sizeof(union ixgbe_adv_rx_desc));
4094         /* Force flushing of IXGBE_RDLEN to prevent MDD */
4095         IXGBE_WRITE_FLUSH(hw);
4096
4097         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
4098         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
4099         ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
4100
4101         ixgbe_configure_srrctl(adapter, ring);
4102         ixgbe_configure_rscctl(adapter, ring);
4103
4104         if (hw->mac.type == ixgbe_mac_82598EB) {
4105                 /*
4106                  * enable cache line friendly hardware writes:
4107                  * PTHRESH=32 descriptors (half the internal cache),
4108                  * this also removes ugly rx_no_buffer_count increment
4109                  * HTHRESH=4 descriptors (to minimize latency on fetch)
4110                  * WTHRESH=8 burst writeback up to two cache lines
4111                  */
4112                 rxdctl &= ~0x3FFFFF;
4113                 rxdctl |=  0x080420;
4114 #if (PAGE_SIZE < 8192)
4115         /* RXDCTL.RLPML does not work on 82599 */
4116         } else if (hw->mac.type != ixgbe_mac_82599EB) {
4117                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4118                             IXGBE_RXDCTL_RLPML_EN);
4119
4120                 /* Limit the maximum frame size so we don't overrun the skb.
4121                  * This can happen in SRIOV mode when the MTU of the VF is
4122                  * higher than the MTU of the PF.
4123                  */
4124                 if (ring_uses_build_skb(ring) &&
4125                     !test_bit(__IXGBE_RX_3K_BUFFER, &ring->state))
4126                         rxdctl |= IXGBE_MAX_2K_FRAME_BUILD_SKB |
4127                                   IXGBE_RXDCTL_RLPML_EN;
4128 #endif
4129         }
4130
4131         if (ring->xsk_umem && hw->mac.type != ixgbe_mac_82599EB) {
4132                 u32 xsk_buf_len = ring->xsk_umem->chunk_size_nohr -
4133                                   XDP_PACKET_HEADROOM;
4134
4135                 rxdctl &= ~(IXGBE_RXDCTL_RLPMLMASK |
4136                             IXGBE_RXDCTL_RLPML_EN);
4137                 rxdctl |= xsk_buf_len | IXGBE_RXDCTL_RLPML_EN;
4138
4139                 ring->rx_buf_len = xsk_buf_len;
4140         }
4141
4142         /* initialize rx_buffer_info */
4143         memset(ring->rx_buffer_info, 0,
4144                sizeof(struct ixgbe_rx_buffer) * ring->count);
4145
4146         /* initialize Rx descriptor 0 */
4147         rx_desc = IXGBE_RX_DESC(ring, 0);
4148         rx_desc->wb.upper.length = 0;
4149
4150         /* enable receive descriptor ring */
4151         rxdctl |= IXGBE_RXDCTL_ENABLE;
4152         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
4153
4154         ixgbe_rx_desc_queue_enable(adapter, ring);
4155         if (ring->xsk_umem)
4156                 ixgbe_alloc_rx_buffers_zc(ring, ixgbe_desc_unused(ring));
4157         else
4158                 ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
4159 }
4160
4161 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
4162 {
4163         struct ixgbe_hw *hw = &adapter->hw;
4164         int rss_i = adapter->ring_feature[RING_F_RSS].indices;
4165         u16 pool = adapter->num_rx_pools;
4166
4167         /* PSRTYPE must be initialized in non 82598 adapters */
4168         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4169                       IXGBE_PSRTYPE_UDPHDR |
4170                       IXGBE_PSRTYPE_IPV4HDR |
4171                       IXGBE_PSRTYPE_L2HDR |
4172                       IXGBE_PSRTYPE_IPV6HDR;
4173
4174         if (hw->mac.type == ixgbe_mac_82598EB)
4175                 return;
4176
4177         if (rss_i > 3)
4178                 psrtype |= 2u << 29;
4179         else if (rss_i > 1)
4180                 psrtype |= 1u << 29;
4181
4182         while (pool--)
4183                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4184 }
4185
4186 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
4187 {
4188         struct ixgbe_hw *hw = &adapter->hw;
4189         u16 pool = adapter->num_rx_pools;
4190         u32 reg_offset, vf_shift, vmolr;
4191         u32 gcr_ext, vmdctl;
4192         int i;
4193
4194         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
4195                 return;
4196
4197         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
4198         vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
4199         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
4200         vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
4201         vmdctl |= IXGBE_VT_CTL_REPLEN;
4202         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
4203
4204         /* accept untagged packets until a vlan tag is
4205          * specifically set for the VMDQ queue/pool
4206          */
4207         vmolr = IXGBE_VMOLR_AUPE;
4208         while (pool--)
4209                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(pool)), vmolr);
4210
4211         vf_shift = VMDQ_P(0) % 32;
4212         reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
4213
4214         /* Enable only the PF's pool for Tx/Rx */
4215         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), GENMASK(31, vf_shift));
4216         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
4217         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), GENMASK(31, vf_shift));
4218         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
4219         if (adapter->bridge_mode == BRIDGE_MODE_VEB)
4220                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
4221
4222         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
4223         hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
4224
4225         /* clear VLAN promisc flag so VFTA will be updated if necessary */
4226         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4227
4228         /*
4229          * Set up VF register offsets for selected VT Mode,
4230          * i.e. 32 or 64 VFs for SR-IOV
4231          */
4232         switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4233         case IXGBE_82599_VMDQ_8Q_MASK:
4234                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
4235                 break;
4236         case IXGBE_82599_VMDQ_4Q_MASK:
4237                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
4238                 break;
4239         default:
4240                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
4241                 break;
4242         }
4243
4244         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
4245
4246         for (i = 0; i < adapter->num_vfs; i++) {
4247                 /* configure spoof checking */
4248                 ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i,
4249                                           adapter->vfinfo[i].spoofchk_enabled);
4250
4251                 /* Enable/Disable RSS query feature  */
4252                 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
4253                                           adapter->vfinfo[i].rss_query_enabled);
4254         }
4255 }
4256
4257 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
4258 {
4259         struct ixgbe_hw *hw = &adapter->hw;
4260         struct net_device *netdev = adapter->netdev;
4261         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4262         struct ixgbe_ring *rx_ring;
4263         int i;
4264         u32 mhadd, hlreg0;
4265
4266 #ifdef IXGBE_FCOE
4267         /* adjust max frame to be able to do baby jumbo for FCoE */
4268         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
4269             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
4270                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4271
4272 #endif /* IXGBE_FCOE */
4273
4274         /* adjust max frame to be at least the size of a standard frame */
4275         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4276                 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
4277
4278         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
4279         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
4280                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
4281                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
4282
4283                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
4284         }
4285
4286         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
4287         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
4288         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
4289         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
4290
4291         /*
4292          * Setup the HW Rx Head and Tail Descriptor Pointers and
4293          * the Base and Length of the Rx Descriptor Ring
4294          */
4295         for (i = 0; i < adapter->num_rx_queues; i++) {
4296                 rx_ring = adapter->rx_ring[i];
4297
4298                 clear_ring_rsc_enabled(rx_ring);
4299                 clear_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4300                 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4301
4302                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4303                         set_ring_rsc_enabled(rx_ring);
4304
4305                 if (test_bit(__IXGBE_RX_FCOE, &rx_ring->state))
4306                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4307
4308                 clear_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4309                 if (adapter->flags2 & IXGBE_FLAG2_RX_LEGACY)
4310                         continue;
4311
4312                 set_bit(__IXGBE_RX_BUILD_SKB_ENABLED, &rx_ring->state);
4313
4314 #if (PAGE_SIZE < 8192)
4315                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
4316                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4317
4318                 if (IXGBE_2K_TOO_SMALL_WITH_PADDING ||
4319                     (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
4320                         set_bit(__IXGBE_RX_3K_BUFFER, &rx_ring->state);
4321 #endif
4322         }
4323 }
4324
4325 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
4326 {
4327         struct ixgbe_hw *hw = &adapter->hw;
4328         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
4329
4330         switch (hw->mac.type) {
4331         case ixgbe_mac_82598EB:
4332                 /*
4333                  * For VMDq support of different descriptor types or
4334                  * buffer sizes through the use of multiple SRRCTL
4335                  * registers, RDRXCTL.MVMEN must be set to 1
4336                  *
4337                  * also, the manual doesn't mention it clearly but DCA hints
4338                  * will only use queue 0's tags unless this bit is set.  Side
4339                  * effects of setting this bit are only that SRRCTL must be
4340                  * fully programmed [0..15]
4341                  */
4342                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
4343                 break;
4344         case ixgbe_mac_X550:
4345         case ixgbe_mac_X550EM_x:
4346         case ixgbe_mac_x550em_a:
4347                 if (adapter->num_vfs)
4348                         rdrxctl |= IXGBE_RDRXCTL_PSP;
4349                 /* fall through */
4350         case ixgbe_mac_82599EB:
4351         case ixgbe_mac_X540:
4352                 /* Disable RSC for ACK packets */
4353                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
4354                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
4355                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
4356                 /* hardware requires some bits to be set by default */
4357                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
4358                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
4359                 break;
4360         default:
4361                 /* We should do nothing since we don't know this hardware */
4362                 return;
4363         }
4364
4365         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
4366 }
4367
4368 /**
4369  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
4370  * @adapter: board private structure
4371  *
4372  * Configure the Rx unit of the MAC after a reset.
4373  **/
4374 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
4375 {
4376         struct ixgbe_hw *hw = &adapter->hw;
4377         int i;
4378         u32 rxctrl, rfctl;
4379
4380         /* disable receives while setting up the descriptors */
4381         hw->mac.ops.disable_rx(hw);
4382
4383         ixgbe_setup_psrtype(adapter);
4384         ixgbe_setup_rdrxctl(adapter);
4385
4386         /* RSC Setup */
4387         rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
4388         rfctl &= ~IXGBE_RFCTL_RSC_DIS;
4389         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
4390                 rfctl |= IXGBE_RFCTL_RSC_DIS;
4391
4392         /* disable NFS filtering */
4393         rfctl |= (IXGBE_RFCTL_NFSW_DIS | IXGBE_RFCTL_NFSR_DIS);
4394         IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
4395
4396         /* Program registers for the distribution of queues */
4397         ixgbe_setup_mrqc(adapter);
4398
4399         /* set_rx_buffer_len must be called before ring initialization */
4400         ixgbe_set_rx_buffer_len(adapter);
4401
4402         /*
4403          * Setup the HW Rx Head and Tail Descriptor Pointers and
4404          * the Base and Length of the Rx Descriptor Ring
4405          */
4406         for (i = 0; i < adapter->num_rx_queues; i++)
4407                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
4408
4409         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
4410         /* disable drop enable for 82598 parts */
4411         if (hw->mac.type == ixgbe_mac_82598EB)
4412                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
4413
4414         /* enable all receives */
4415         rxctrl |= IXGBE_RXCTRL_RXEN;
4416         hw->mac.ops.enable_rx_dma(hw, rxctrl);
4417 }
4418
4419 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
4420                                  __be16 proto, u16 vid)
4421 {
4422         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4423         struct ixgbe_hw *hw = &adapter->hw;
4424
4425         /* add VID to filter table */
4426         if (!vid || !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4427                 hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, !!vid);
4428
4429         set_bit(vid, adapter->active_vlans);
4430
4431         return 0;
4432 }
4433
4434 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
4435 {
4436         u32 vlvf;
4437         int idx;
4438
4439         /* short cut the special case */
4440         if (vlan == 0)
4441                 return 0;
4442
4443         /* Search for the vlan id in the VLVF entries */
4444         for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
4445                 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
4446                 if ((vlvf & VLAN_VID_MASK) == vlan)
4447                         break;
4448         }
4449
4450         return idx;
4451 }
4452
4453 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
4454 {
4455         struct ixgbe_hw *hw = &adapter->hw;
4456         u32 bits, word;
4457         int idx;
4458
4459         idx = ixgbe_find_vlvf_entry(hw, vid);
4460         if (!idx)
4461                 return;
4462
4463         /* See if any other pools are set for this VLAN filter
4464          * entry other than the PF.
4465          */
4466         word = idx * 2 + (VMDQ_P(0) / 32);
4467         bits = ~BIT(VMDQ_P(0) % 32);
4468         bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4469
4470         /* Disable the filter so this falls into the default pool. */
4471         if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
4472                 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4473                         IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
4474                 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
4475         }
4476 }
4477
4478 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
4479                                   __be16 proto, u16 vid)
4480 {
4481         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4482         struct ixgbe_hw *hw = &adapter->hw;
4483
4484         /* remove VID from filter table */
4485         if (vid && !(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4486                 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
4487
4488         clear_bit(vid, adapter->active_vlans);
4489
4490         return 0;
4491 }
4492
4493 /**
4494  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
4495  * @adapter: driver data
4496  */
4497 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
4498 {
4499         struct ixgbe_hw *hw = &adapter->hw;
4500         u32 vlnctrl;
4501         int i, j;
4502
4503         switch (hw->mac.type) {
4504         case ixgbe_mac_82598EB:
4505                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4506                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
4507                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4508                 break;
4509         case ixgbe_mac_82599EB:
4510         case ixgbe_mac_X540:
4511         case ixgbe_mac_X550:
4512         case ixgbe_mac_X550EM_x:
4513         case ixgbe_mac_x550em_a:
4514                 for (i = 0; i < adapter->num_rx_queues; i++) {
4515                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4516
4517                         if (!netif_is_ixgbe(ring->netdev))
4518                                 continue;
4519
4520                         j = ring->reg_idx;
4521                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4522                         vlnctrl &= ~IXGBE_RXDCTL_VME;
4523                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4524                 }
4525                 break;
4526         default:
4527                 break;
4528         }
4529 }
4530
4531 /**
4532  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4533  * @adapter: driver data
4534  */
4535 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4536 {
4537         struct ixgbe_hw *hw = &adapter->hw;
4538         u32 vlnctrl;
4539         int i, j;
4540
4541         switch (hw->mac.type) {
4542         case ixgbe_mac_82598EB:
4543                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4544                 vlnctrl |= IXGBE_VLNCTRL_VME;
4545                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4546                 break;
4547         case ixgbe_mac_82599EB:
4548         case ixgbe_mac_X540:
4549         case ixgbe_mac_X550:
4550         case ixgbe_mac_X550EM_x:
4551         case ixgbe_mac_x550em_a:
4552                 for (i = 0; i < adapter->num_rx_queues; i++) {
4553                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4554
4555                         if (!netif_is_ixgbe(ring->netdev))
4556                                 continue;
4557
4558                         j = ring->reg_idx;
4559                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4560                         vlnctrl |= IXGBE_RXDCTL_VME;
4561                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4562                 }
4563                 break;
4564         default:
4565                 break;
4566         }
4567 }
4568
4569 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4570 {
4571         struct ixgbe_hw *hw = &adapter->hw;
4572         u32 vlnctrl, i;
4573
4574         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4575
4576         if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
4577         /* For VMDq and SR-IOV we must leave VLAN filtering enabled */
4578                 vlnctrl |= IXGBE_VLNCTRL_VFE;
4579                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4580         } else {
4581                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
4582                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4583                 return;
4584         }
4585
4586         /* Nothing to do for 82598 */
4587         if (hw->mac.type == ixgbe_mac_82598EB)
4588                 return;
4589
4590         /* We are already in VLAN promisc, nothing to do */
4591         if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4592                 return;
4593
4594         /* Set flag so we don't redo unnecessary work */
4595         adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4596
4597         /* Add PF to all active pools */
4598         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4599                 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4600                 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4601
4602                 vlvfb |= BIT(VMDQ_P(0) % 32);
4603                 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4604         }
4605
4606         /* Set all bits in the VLAN filter table array */
4607         for (i = hw->mac.vft_size; i--;)
4608                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4609 }
4610
4611 #define VFTA_BLOCK_SIZE 8
4612 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4613 {
4614         struct ixgbe_hw *hw = &adapter->hw;
4615         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4616         u32 vid_start = vfta_offset * 32;
4617         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4618         u32 i, vid, word, bits;
4619
4620         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4621                 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4622
4623                 /* pull VLAN ID from VLVF */
4624                 vid = vlvf & VLAN_VID_MASK;
4625
4626                 /* only concern outselves with a certain range */
4627                 if (vid < vid_start || vid >= vid_end)
4628                         continue;
4629
4630                 if (vlvf) {
4631                         /* record VLAN ID in VFTA */
4632                         vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4633
4634                         /* if PF is part of this then continue */
4635                         if (test_bit(vid, adapter->active_vlans))
4636                                 continue;
4637                 }
4638
4639                 /* remove PF from the pool */
4640                 word = i * 2 + VMDQ_P(0) / 32;
4641                 bits = ~BIT(VMDQ_P(0) % 32);
4642                 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4643                 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4644         }
4645
4646         /* extract values from active_vlans and write back to VFTA */
4647         for (i = VFTA_BLOCK_SIZE; i--;) {
4648                 vid = (vfta_offset + i) * 32;
4649                 word = vid / BITS_PER_LONG;
4650                 bits = vid % BITS_PER_LONG;
4651
4652                 vfta[i] |= adapter->active_vlans[word] >> bits;
4653
4654                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4655         }
4656 }
4657
4658 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4659 {
4660         struct ixgbe_hw *hw = &adapter->hw;
4661         u32 vlnctrl, i;
4662
4663         /* Set VLAN filtering to enabled */
4664         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4665         vlnctrl |= IXGBE_VLNCTRL_VFE;
4666         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4667
4668         if (!(adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) ||
4669             hw->mac.type == ixgbe_mac_82598EB)
4670                 return;
4671
4672         /* We are not in VLAN promisc, nothing to do */
4673         if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4674                 return;
4675
4676         /* Set flag so we don't redo unnecessary work */
4677         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4678
4679         for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4680                 ixgbe_scrub_vfta(adapter, i);
4681 }
4682
4683 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4684 {
4685         u16 vid = 1;
4686
4687         ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4688
4689         for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
4690                 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4691 }
4692
4693 /**
4694  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4695  * @netdev: network interface device structure
4696  *
4697  * Writes multicast address list to the MTA hash table.
4698  * Returns: -ENOMEM on failure
4699  *                0 on no addresses written
4700  *                X on writing X addresses to MTA
4701  **/
4702 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4703 {
4704         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4705         struct ixgbe_hw *hw = &adapter->hw;
4706
4707         if (!netif_running(netdev))
4708                 return 0;
4709
4710         if (hw->mac.ops.update_mc_addr_list)
4711                 hw->mac.ops.update_mc_addr_list(hw, netdev);
4712         else
4713                 return -ENOMEM;
4714
4715 #ifdef CONFIG_PCI_IOV
4716         ixgbe_restore_vf_multicasts(adapter);
4717 #endif
4718
4719         return netdev_mc_count(netdev);
4720 }
4721
4722 #ifdef CONFIG_PCI_IOV
4723 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4724 {
4725         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4726         struct ixgbe_hw *hw = &adapter->hw;
4727         int i;
4728
4729         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4730                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4731
4732                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4733                         hw->mac.ops.set_rar(hw, i,
4734                                             mac_table->addr,
4735                                             mac_table->pool,
4736                                             IXGBE_RAH_AV);
4737                 else
4738                         hw->mac.ops.clear_rar(hw, i);
4739         }
4740 }
4741
4742 #endif
4743 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4744 {
4745         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4746         struct ixgbe_hw *hw = &adapter->hw;
4747         int i;
4748
4749         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4750                 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4751                         continue;
4752
4753                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4754
4755                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4756                         hw->mac.ops.set_rar(hw, i,
4757                                             mac_table->addr,
4758                                             mac_table->pool,
4759                                             IXGBE_RAH_AV);
4760                 else
4761                         hw->mac.ops.clear_rar(hw, i);
4762         }
4763 }
4764
4765 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4766 {
4767         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4768         struct ixgbe_hw *hw = &adapter->hw;
4769         int i;
4770
4771         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4772                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4773                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4774         }
4775
4776         ixgbe_sync_mac_table(adapter);
4777 }
4778
4779 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4780 {
4781         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4782         struct ixgbe_hw *hw = &adapter->hw;
4783         int i, count = 0;
4784
4785         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4786                 /* do not count default RAR as available */
4787                 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4788                         continue;
4789
4790                 /* only count unused and addresses that belong to us */
4791                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4792                         if (mac_table->pool != pool)
4793                                 continue;
4794                 }
4795
4796                 count++;
4797         }
4798
4799         return count;
4800 }
4801
4802 /* this function destroys the first RAR entry */
4803 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4804 {
4805         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4806         struct ixgbe_hw *hw = &adapter->hw;
4807
4808         memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4809         mac_table->pool = VMDQ_P(0);
4810
4811         mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4812
4813         hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4814                             IXGBE_RAH_AV);
4815 }
4816
4817 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4818                          const u8 *addr, u16 pool)
4819 {
4820         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4821         struct ixgbe_hw *hw = &adapter->hw;
4822         int i;
4823
4824         if (is_zero_ether_addr(addr))
4825                 return -EINVAL;
4826
4827         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4828                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4829                         continue;
4830
4831                 ether_addr_copy(mac_table->addr, addr);
4832                 mac_table->pool = pool;
4833
4834                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4835                                     IXGBE_MAC_STATE_IN_USE;
4836
4837                 ixgbe_sync_mac_table(adapter);
4838
4839                 return i;
4840         }
4841
4842         return -ENOMEM;
4843 }
4844
4845 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4846                          const u8 *addr, u16 pool)
4847 {
4848         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4849         struct ixgbe_hw *hw = &adapter->hw;
4850         int i;
4851
4852         if (is_zero_ether_addr(addr))
4853                 return -EINVAL;
4854
4855         /* search table for addr, if found clear IN_USE flag and sync */
4856         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4857                 /* we can only delete an entry if it is in use */
4858                 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4859                         continue;
4860                 /* we only care about entries that belong to the given pool */
4861                 if (mac_table->pool != pool)
4862                         continue;
4863                 /* we only care about a specific MAC address */
4864                 if (!ether_addr_equal(addr, mac_table->addr))
4865                         continue;
4866
4867                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4868                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4869
4870                 ixgbe_sync_mac_table(adapter);
4871
4872                 return 0;
4873         }
4874
4875         return -ENOMEM;
4876 }
4877
4878 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4879 {
4880         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4881         int ret;
4882
4883         ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4884
4885         return min_t(int, ret, 0);
4886 }
4887
4888 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4889 {
4890         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4891
4892         ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4893
4894         return 0;
4895 }
4896
4897 /**
4898  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4899  * @netdev: network interface device structure
4900  *
4901  * The set_rx_method entry point is called whenever the unicast/multicast
4902  * address list or the network interface flags are updated.  This routine is
4903  * responsible for configuring the hardware for proper unicast, multicast and
4904  * promiscuous mode.
4905  **/
4906 void ixgbe_set_rx_mode(struct net_device *netdev)
4907 {
4908         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4909         struct ixgbe_hw *hw = &adapter->hw;
4910         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4911         netdev_features_t features = netdev->features;
4912         int count;
4913
4914         /* Check for Promiscuous and All Multicast modes */
4915         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4916
4917         /* set all bits that we expect to always be set */
4918         fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4919         fctrl |= IXGBE_FCTRL_BAM;
4920         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4921         fctrl |= IXGBE_FCTRL_PMCF;
4922
4923         /* clear the bits we are changing the status of */
4924         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4925         if (netdev->flags & IFF_PROMISC) {
4926                 hw->addr_ctrl.user_set_promisc = true;
4927                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4928                 vmolr |= IXGBE_VMOLR_MPE;
4929                 features &= ~NETIF_F_HW_VLAN_CTAG_FILTER;
4930         } else {
4931                 if (netdev->flags & IFF_ALLMULTI) {
4932                         fctrl |= IXGBE_FCTRL_MPE;
4933                         vmolr |= IXGBE_VMOLR_MPE;
4934                 }
4935                 hw->addr_ctrl.user_set_promisc = false;
4936         }
4937
4938         /*
4939          * Write addresses to available RAR registers, if there is not
4940          * sufficient space to store all the addresses then enable
4941          * unicast promiscuous mode
4942          */
4943         if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4944                 fctrl |= IXGBE_FCTRL_UPE;
4945                 vmolr |= IXGBE_VMOLR_ROPE;
4946         }
4947
4948         /* Write addresses to the MTA, if the attempt fails
4949          * then we should just turn on promiscuous mode so
4950          * that we can at least receive multicast traffic
4951          */
4952         count = ixgbe_write_mc_addr_list(netdev);
4953         if (count < 0) {
4954                 fctrl |= IXGBE_FCTRL_MPE;
4955                 vmolr |= IXGBE_VMOLR_MPE;
4956         } else if (count) {
4957                 vmolr |= IXGBE_VMOLR_ROMPE;
4958         }
4959
4960         if (hw->mac.type != ixgbe_mac_82598EB) {
4961                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4962                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4963                            IXGBE_VMOLR_ROPE);
4964                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4965         }
4966
4967         /* This is useful for sniffing bad packets. */
4968         if (features & NETIF_F_RXALL) {
4969                 /* UPE and MPE will be handled by normal PROMISC logic
4970                  * in e1000e_set_rx_mode */
4971                 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4972                           IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4973                           IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4974
4975                 fctrl &= ~(IXGBE_FCTRL_DPF);
4976                 /* NOTE:  VLAN filtering is disabled by setting PROMISC */
4977         }
4978
4979         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4980
4981         if (features & NETIF_F_HW_VLAN_CTAG_RX)
4982                 ixgbe_vlan_strip_enable(adapter);
4983         else
4984                 ixgbe_vlan_strip_disable(adapter);
4985
4986         if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
4987                 ixgbe_vlan_promisc_disable(adapter);
4988         else
4989                 ixgbe_vlan_promisc_enable(adapter);
4990 }
4991
4992 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4993 {
4994         int q_idx;
4995
4996         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
4997                 napi_enable(&adapter->q_vector[q_idx]->napi);
4998 }
4999
5000 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
5001 {
5002         int q_idx;
5003
5004         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++)
5005                 napi_disable(&adapter->q_vector[q_idx]->napi);
5006 }
5007
5008 static void ixgbe_clear_udp_tunnel_port(struct ixgbe_adapter *adapter, u32 mask)
5009 {
5010         struct ixgbe_hw *hw = &adapter->hw;
5011         u32 vxlanctrl;
5012
5013         if (!(adapter->flags & (IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE |
5014                                 IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)))
5015                 return;
5016
5017         vxlanctrl = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) & ~mask;
5018         IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, vxlanctrl);
5019
5020         if (mask & IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK)
5021                 adapter->vxlan_port = 0;
5022
5023         if (mask & IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK)
5024                 adapter->geneve_port = 0;
5025 }
5026
5027 #ifdef CONFIG_IXGBE_DCB
5028 /**
5029  * ixgbe_configure_dcb - Configure DCB hardware
5030  * @adapter: ixgbe adapter struct
5031  *
5032  * This is called by the driver on open to configure the DCB hardware.
5033  * This is also called by the gennetlink interface when reconfiguring
5034  * the DCB state.
5035  */
5036 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
5037 {
5038         struct ixgbe_hw *hw = &adapter->hw;
5039         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
5040
5041         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
5042                 if (hw->mac.type == ixgbe_mac_82598EB)
5043                         netif_set_gso_max_size(adapter->netdev, 65536);
5044                 return;
5045         }
5046
5047         if (hw->mac.type == ixgbe_mac_82598EB)
5048                 netif_set_gso_max_size(adapter->netdev, 32768);
5049
5050 #ifdef IXGBE_FCOE
5051         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
5052                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
5053 #endif
5054
5055         /* reconfigure the hardware */
5056         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
5057                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5058                                                 DCB_TX_CONFIG);
5059                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
5060                                                 DCB_RX_CONFIG);
5061                 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
5062         } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
5063                 ixgbe_dcb_hw_ets(&adapter->hw,
5064                                  adapter->ixgbe_ieee_ets,
5065                                  max_frame);
5066                 ixgbe_dcb_hw_pfc_config(&adapter->hw,
5067                                         adapter->ixgbe_ieee_pfc->pfc_en,
5068                                         adapter->ixgbe_ieee_ets->prio_tc);
5069         }
5070
5071         /* Enable RSS Hash per TC */
5072         if (hw->mac.type != ixgbe_mac_82598EB) {
5073                 u32 msb = 0;
5074                 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
5075
5076                 while (rss_i) {
5077                         msb++;
5078                         rss_i >>= 1;
5079                 }
5080
5081                 /* write msb to all 8 TCs in one write */
5082                 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
5083         }
5084 }
5085 #endif
5086
5087 /* Additional bittime to account for IXGBE framing */
5088 #define IXGBE_ETH_FRAMING 20
5089
5090 /**
5091  * ixgbe_hpbthresh - calculate high water mark for flow control
5092  *
5093  * @adapter: board private structure to calculate for
5094  * @pb: packet buffer to calculate
5095  */
5096 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
5097 {
5098         struct ixgbe_hw *hw = &adapter->hw;
5099         struct net_device *dev = adapter->netdev;
5100         int link, tc, kb, marker;
5101         u32 dv_id, rx_pba;
5102
5103         /* Calculate max LAN frame size */
5104         tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
5105
5106 #ifdef IXGBE_FCOE
5107         /* FCoE traffic class uses FCOE jumbo frames */
5108         if ((dev->features & NETIF_F_FCOE_MTU) &&
5109             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5110             (pb == ixgbe_fcoe_get_tc(adapter)))
5111                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5112 #endif
5113
5114         /* Calculate delay value for device */
5115         switch (hw->mac.type) {
5116         case ixgbe_mac_X540:
5117         case ixgbe_mac_X550:
5118         case ixgbe_mac_X550EM_x:
5119         case ixgbe_mac_x550em_a:
5120                 dv_id = IXGBE_DV_X540(link, tc);
5121                 break;
5122         default:
5123                 dv_id = IXGBE_DV(link, tc);
5124                 break;
5125         }
5126
5127         /* Loopback switch introduces additional latency */
5128         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5129                 dv_id += IXGBE_B2BT(tc);
5130
5131         /* Delay value is calculated in bit times convert to KB */
5132         kb = IXGBE_BT2KB(dv_id);
5133         rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
5134
5135         marker = rx_pba - kb;
5136
5137         /* It is possible that the packet buffer is not large enough
5138          * to provide required headroom. In this case throw an error
5139          * to user and a do the best we can.
5140          */
5141         if (marker < 0) {
5142                 e_warn(drv, "Packet Buffer(%i) can not provide enough"
5143                             "headroom to support flow control."
5144                             "Decrease MTU or number of traffic classes\n", pb);
5145                 marker = tc + 1;
5146         }
5147
5148         return marker;
5149 }
5150
5151 /**
5152  * ixgbe_lpbthresh - calculate low water mark for for flow control
5153  *
5154  * @adapter: board private structure to calculate for
5155  * @pb: packet buffer to calculate
5156  */
5157 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
5158 {
5159         struct ixgbe_hw *hw = &adapter->hw;
5160         struct net_device *dev = adapter->netdev;
5161         int tc;
5162         u32 dv_id;
5163
5164         /* Calculate max LAN frame size */
5165         tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
5166
5167 #ifdef IXGBE_FCOE
5168         /* FCoE traffic class uses FCOE jumbo frames */
5169         if ((dev->features & NETIF_F_FCOE_MTU) &&
5170             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
5171             (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
5172                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
5173 #endif
5174
5175         /* Calculate delay value for device */
5176         switch (hw->mac.type) {
5177         case ixgbe_mac_X540:
5178         case ixgbe_mac_X550:
5179         case ixgbe_mac_X550EM_x:
5180         case ixgbe_mac_x550em_a:
5181                 dv_id = IXGBE_LOW_DV_X540(tc);
5182                 break;
5183         default:
5184                 dv_id = IXGBE_LOW_DV(tc);
5185                 break;
5186         }
5187
5188         /* Delay value is calculated in bit times convert to KB */
5189         return IXGBE_BT2KB(dv_id);
5190 }
5191
5192 /*
5193  * ixgbe_pbthresh_setup - calculate and setup high low water marks
5194  */
5195 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
5196 {
5197         struct ixgbe_hw *hw = &adapter->hw;
5198         int num_tc = adapter->hw_tcs;
5199         int i;
5200
5201         if (!num_tc)
5202                 num_tc = 1;
5203
5204         for (i = 0; i < num_tc; i++) {
5205                 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
5206                 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
5207
5208                 /* Low water marks must not be larger than high water marks */
5209                 if (hw->fc.low_water[i] > hw->fc.high_water[i])
5210                         hw->fc.low_water[i] = 0;
5211         }
5212
5213         for (; i < MAX_TRAFFIC_CLASS; i++)
5214                 hw->fc.high_water[i] = 0;
5215 }
5216
5217 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
5218 {
5219         struct ixgbe_hw *hw = &adapter->hw;
5220         int hdrm;
5221         u8 tc = adapter->hw_tcs;
5222
5223         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
5224             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
5225                 hdrm = 32 << adapter->fdir_pballoc;
5226         else
5227                 hdrm = 0;
5228
5229         hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
5230         ixgbe_pbthresh_setup(adapter);
5231 }
5232
5233 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
5234 {
5235         struct ixgbe_hw *hw = &adapter->hw;
5236         struct hlist_node *node2;
5237         struct ixgbe_fdir_filter *filter;
5238         u64 action;
5239
5240         spin_lock(&adapter->fdir_perfect_lock);
5241
5242         if (!hlist_empty(&adapter->fdir_filter_list))
5243                 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
5244
5245         hlist_for_each_entry_safe(filter, node2,
5246                                   &adapter->fdir_filter_list, fdir_node) {
5247                 action = filter->action;
5248                 if (action != IXGBE_FDIR_DROP_QUEUE && action != 0)
5249                         action =
5250                         (action >> ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF) - 1;
5251
5252                 ixgbe_fdir_write_perfect_filter_82599(hw,
5253                                 &filter->filter,
5254                                 filter->sw_idx,
5255                                 (action == IXGBE_FDIR_DROP_QUEUE) ?
5256                                 IXGBE_FDIR_DROP_QUEUE :
5257                                 adapter->rx_ring[action]->reg_idx);
5258         }
5259
5260         spin_unlock(&adapter->fdir_perfect_lock);
5261 }
5262
5263 /**
5264  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
5265  * @rx_ring: ring to free buffers from
5266  **/
5267 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
5268 {
5269         u16 i = rx_ring->next_to_clean;
5270         struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
5271
5272         if (rx_ring->xsk_umem) {
5273                 ixgbe_xsk_clean_rx_ring(rx_ring);
5274                 goto skip_free;
5275         }
5276
5277         /* Free all the Rx ring sk_buffs */
5278         while (i != rx_ring->next_to_alloc) {
5279                 if (rx_buffer->skb) {
5280                         struct sk_buff *skb = rx_buffer->skb;
5281                         if (IXGBE_CB(skb)->page_released)
5282                                 dma_unmap_page_attrs(rx_ring->dev,
5283                                                      IXGBE_CB(skb)->dma,
5284                                                      ixgbe_rx_pg_size(rx_ring),
5285                                                      DMA_FROM_DEVICE,
5286                                                      IXGBE_RX_DMA_ATTR);
5287                         dev_kfree_skb(skb);
5288                 }
5289
5290                 /* Invalidate cache lines that may have been written to by
5291                  * device so that we avoid corrupting memory.
5292                  */
5293                 dma_sync_single_range_for_cpu(rx_ring->dev,
5294                                               rx_buffer->dma,
5295                                               rx_buffer->page_offset,
5296                                               ixgbe_rx_bufsz(rx_ring),
5297                                               DMA_FROM_DEVICE);
5298
5299                 /* free resources associated with mapping */
5300                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
5301                                      ixgbe_rx_pg_size(rx_ring),
5302                                      DMA_FROM_DEVICE,
5303                                      IXGBE_RX_DMA_ATTR);
5304                 __page_frag_cache_drain(rx_buffer->page,
5305                                         rx_buffer->pagecnt_bias);
5306
5307                 i++;
5308                 rx_buffer++;
5309                 if (i == rx_ring->count) {
5310                         i = 0;
5311                         rx_buffer = rx_ring->rx_buffer_info;
5312                 }
5313         }
5314
5315 skip_free:
5316         rx_ring->next_to_alloc = 0;
5317         rx_ring->next_to_clean = 0;
5318         rx_ring->next_to_use = 0;
5319 }
5320
5321 static int ixgbe_fwd_ring_up(struct ixgbe_adapter *adapter,
5322                              struct ixgbe_fwd_adapter *accel)
5323 {
5324         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
5325         int num_tc = netdev_get_num_tc(adapter->netdev);
5326         struct net_device *vdev = accel->netdev;
5327         int i, baseq, err;
5328
5329         baseq = accel->pool * adapter->num_rx_queues_per_pool;
5330         netdev_dbg(vdev, "pool %i:%i queues %i:%i\n",
5331                    accel->pool, adapter->num_rx_pools,
5332                    baseq, baseq + adapter->num_rx_queues_per_pool);
5333
5334         accel->rx_base_queue = baseq;
5335         accel->tx_base_queue = baseq;
5336
5337         /* record configuration for macvlan interface in vdev */
5338         for (i = 0; i < num_tc; i++)
5339                 netdev_bind_sb_channel_queue(adapter->netdev, vdev,
5340                                              i, rss_i, baseq + (rss_i * i));
5341
5342         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5343                 adapter->rx_ring[baseq + i]->netdev = vdev;
5344
5345         /* Guarantee all rings are updated before we update the
5346          * MAC address filter.
5347          */
5348         wmb();
5349
5350         /* ixgbe_add_mac_filter will return an index if it succeeds, so we
5351          * need to only treat it as an error value if it is negative.
5352          */
5353         err = ixgbe_add_mac_filter(adapter, vdev->dev_addr,
5354                                    VMDQ_P(accel->pool));
5355         if (err >= 0)
5356                 return 0;
5357
5358         /* if we cannot add the MAC rule then disable the offload */
5359         macvlan_release_l2fw_offload(vdev);
5360
5361         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
5362                 adapter->rx_ring[baseq + i]->netdev = NULL;
5363
5364         netdev_err(vdev, "L2FW offload disabled due to L2 filter error\n");
5365
5366         /* unbind the queues and drop the subordinate channel config */
5367         netdev_unbind_sb_channel(adapter->netdev, vdev);
5368         netdev_set_sb_channel(vdev, 0);
5369
5370         clear_bit(accel->pool, adapter->fwd_bitmask);
5371         kfree(accel);
5372
5373         return err;
5374 }
5375
5376 static int ixgbe_macvlan_up(struct net_device *vdev, void *data)
5377 {
5378         struct ixgbe_adapter *adapter = data;
5379         struct ixgbe_fwd_adapter *accel;
5380
5381         if (!netif_is_macvlan(vdev))
5382                 return 0;
5383
5384         accel = macvlan_accel_priv(vdev);
5385         if (!accel)
5386                 return 0;
5387
5388         ixgbe_fwd_ring_up(adapter, accel);
5389
5390         return 0;
5391 }
5392
5393 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
5394 {
5395         netdev_walk_all_upper_dev_rcu(adapter->netdev,
5396                                       ixgbe_macvlan_up, adapter);
5397 }
5398
5399 static void ixgbe_configure(struct ixgbe_adapter *adapter)
5400 {
5401         struct ixgbe_hw *hw = &adapter->hw;
5402
5403         ixgbe_configure_pb(adapter);
5404 #ifdef CONFIG_IXGBE_DCB
5405         ixgbe_configure_dcb(adapter);
5406 #endif
5407         /*
5408          * We must restore virtualization before VLANs or else
5409          * the VLVF registers will not be populated
5410          */
5411         ixgbe_configure_virtualization(adapter);
5412
5413         ixgbe_set_rx_mode(adapter->netdev);
5414         ixgbe_restore_vlan(adapter);
5415         ixgbe_ipsec_restore(adapter);
5416
5417         switch (hw->mac.type) {
5418         case ixgbe_mac_82599EB:
5419         case ixgbe_mac_X540:
5420                 hw->mac.ops.disable_rx_buff(hw);
5421                 break;
5422         default:
5423                 break;
5424         }
5425
5426         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5427                 ixgbe_init_fdir_signature_82599(&adapter->hw,
5428                                                 adapter->fdir_pballoc);
5429         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5430                 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5431                                               adapter->fdir_pballoc);
5432                 ixgbe_fdir_filter_restore(adapter);
5433         }
5434
5435         switch (hw->mac.type) {
5436         case ixgbe_mac_82599EB:
5437         case ixgbe_mac_X540:
5438                 hw->mac.ops.enable_rx_buff(hw);
5439                 break;
5440         default:
5441                 break;
5442         }
5443
5444 #ifdef CONFIG_IXGBE_DCA
5445         /* configure DCA */
5446         if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5447                 ixgbe_setup_dca(adapter);
5448 #endif /* CONFIG_IXGBE_DCA */
5449
5450 #ifdef IXGBE_FCOE
5451         /* configure FCoE L2 filters, redirection table, and Rx control */
5452         ixgbe_configure_fcoe(adapter);
5453
5454 #endif /* IXGBE_FCOE */
5455         ixgbe_configure_tx(adapter);
5456         ixgbe_configure_rx(adapter);
5457         ixgbe_configure_dfwd(adapter);
5458 }
5459
5460 /**
5461  * ixgbe_sfp_link_config - set up SFP+ link
5462  * @adapter: pointer to private adapter struct
5463  **/
5464 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5465 {
5466         /*
5467          * We are assuming the worst case scenario here, and that
5468          * is that an SFP was inserted/removed after the reset
5469          * but before SFP detection was enabled.  As such the best
5470          * solution is to just start searching as soon as we start
5471          */
5472         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5473                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5474
5475         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5476         adapter->sfp_poll_time = 0;
5477 }
5478
5479 /**
5480  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5481  * @hw: pointer to private hardware struct
5482  *
5483  * Returns 0 on success, negative on failure
5484  **/
5485 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5486 {
5487         u32 speed;
5488         bool autoneg, link_up = false;
5489         int ret = IXGBE_ERR_LINK_SETUP;
5490
5491         if (hw->mac.ops.check_link)
5492                 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5493
5494         if (ret)
5495                 return ret;
5496
5497         speed = hw->phy.autoneg_advertised;
5498         if ((!speed) && (hw->mac.ops.get_link_capabilities))
5499                 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5500                                                         &autoneg);
5501         if (ret)
5502                 return ret;
5503
5504         if (hw->mac.ops.setup_link)
5505                 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5506
5507         return ret;
5508 }
5509
5510 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5511 {
5512         struct ixgbe_hw *hw = &adapter->hw;
5513         u32 gpie = 0;
5514
5515         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5516                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5517                        IXGBE_GPIE_OCD;
5518                 gpie |= IXGBE_GPIE_EIAME;
5519                 /*
5520                  * use EIAM to auto-mask when MSI-X interrupt is asserted
5521                  * this saves a register write for every interrupt
5522                  */
5523                 switch (hw->mac.type) {
5524                 case ixgbe_mac_82598EB:
5525                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5526                         break;
5527                 case ixgbe_mac_82599EB:
5528                 case ixgbe_mac_X540:
5529                 case ixgbe_mac_X550:
5530                 case ixgbe_mac_X550EM_x:
5531                 case ixgbe_mac_x550em_a:
5532                 default:
5533                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5534                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5535                         break;
5536                 }
5537         } else {
5538                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5539                  * specifically only auto mask tx and rx interrupts */
5540                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5541         }
5542
5543         /* XXX: to interrupt immediately for EICS writes, enable this */
5544         /* gpie |= IXGBE_GPIE_EIMEN; */
5545
5546         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5547                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5548
5549                 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5550                 case IXGBE_82599_VMDQ_8Q_MASK:
5551                         gpie |= IXGBE_GPIE_VTMODE_16;
5552                         break;
5553                 case IXGBE_82599_VMDQ_4Q_MASK:
5554                         gpie |= IXGBE_GPIE_VTMODE_32;
5555                         break;
5556                 default:
5557                         gpie |= IXGBE_GPIE_VTMODE_64;
5558                         break;
5559                 }
5560         }
5561
5562         /* Enable Thermal over heat sensor interrupt */
5563         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5564                 switch (adapter->hw.mac.type) {
5565                 case ixgbe_mac_82599EB:
5566                         gpie |= IXGBE_SDP0_GPIEN_8259X;
5567                         break;
5568                 default:
5569                         break;
5570                 }
5571         }
5572
5573         /* Enable fan failure interrupt */
5574         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5575                 gpie |= IXGBE_SDP1_GPIEN(hw);
5576
5577         switch (hw->mac.type) {
5578         case ixgbe_mac_82599EB:
5579                 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5580                 break;
5581         case ixgbe_mac_X550EM_x:
5582         case ixgbe_mac_x550em_a:
5583                 gpie |= IXGBE_SDP0_GPIEN_X540;
5584                 break;
5585         default:
5586                 break;
5587         }
5588
5589         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5590 }
5591
5592 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5593 {
5594         struct ixgbe_hw *hw = &adapter->hw;
5595         int err;
5596         u32 ctrl_ext;
5597
5598         ixgbe_get_hw_control(adapter);
5599         ixgbe_setup_gpie(adapter);
5600
5601         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5602                 ixgbe_configure_msix(adapter);
5603         else
5604                 ixgbe_configure_msi_and_legacy(adapter);
5605
5606         /* enable the optics for 82599 SFP+ fiber */
5607         if (hw->mac.ops.enable_tx_laser)
5608                 hw->mac.ops.enable_tx_laser(hw);
5609
5610         if (hw->phy.ops.set_phy_power)
5611                 hw->phy.ops.set_phy_power(hw, true);
5612
5613         smp_mb__before_atomic();
5614         clear_bit(__IXGBE_DOWN, &adapter->state);
5615         ixgbe_napi_enable_all(adapter);
5616
5617         if (ixgbe_is_sfp(hw)) {
5618                 ixgbe_sfp_link_config(adapter);
5619         } else {
5620                 err = ixgbe_non_sfp_link_config(hw);
5621                 if (err)
5622                         e_err(probe, "link_config FAILED %d\n", err);
5623         }
5624
5625         /* clear any pending interrupts, may auto mask */
5626         IXGBE_READ_REG(hw, IXGBE_EICR);
5627         ixgbe_irq_enable(adapter, true, true);
5628
5629         /*
5630          * If this adapter has a fan, check to see if we had a failure
5631          * before we enabled the interrupt.
5632          */
5633         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5634                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5635                 if (esdp & IXGBE_ESDP_SDP1)
5636                         e_crit(drv, "Fan has stopped, replace the adapter\n");
5637         }
5638
5639         /* bring the link up in the watchdog, this could race with our first
5640          * link up interrupt but shouldn't be a problem */
5641         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5642         adapter->link_check_timeout = jiffies;
5643         mod_timer(&adapter->service_timer, jiffies);
5644
5645         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5646         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5647         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5648         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5649 }
5650
5651 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5652 {
5653         WARN_ON(in_interrupt());
5654         /* put off any impending NetWatchDogTimeout */
5655         netif_trans_update(adapter->netdev);
5656
5657         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5658                 usleep_range(1000, 2000);
5659         if (adapter->hw.phy.type == ixgbe_phy_fw)
5660                 ixgbe_watchdog_link_is_down(adapter);
5661         ixgbe_down(adapter);
5662         /*
5663          * If SR-IOV enabled then wait a bit before bringing the adapter
5664          * back up to give the VFs time to respond to the reset.  The
5665          * two second wait is based upon the watchdog timer cycle in
5666          * the VF driver.
5667          */
5668         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5669                 msleep(2000);
5670         ixgbe_up(adapter);
5671         clear_bit(__IXGBE_RESETTING, &adapter->state);
5672 }
5673
5674 void ixgbe_up(struct ixgbe_adapter *adapter)
5675 {
5676         /* hardware has been reset, we need to reload some things */
5677         ixgbe_configure(adapter);
5678
5679         ixgbe_up_complete(adapter);
5680 }
5681
5682 static unsigned long ixgbe_get_completion_timeout(struct ixgbe_adapter *adapter)
5683 {
5684         u16 devctl2;
5685
5686         pcie_capability_read_word(adapter->pdev, PCI_EXP_DEVCTL2, &devctl2);
5687
5688         switch (devctl2 & IXGBE_PCIDEVCTRL2_TIMEO_MASK) {
5689         case IXGBE_PCIDEVCTRL2_17_34s:
5690         case IXGBE_PCIDEVCTRL2_4_8s:
5691                 /* For now we cap the upper limit on delay to 2 seconds
5692                  * as we end up going up to 34 seconds of delay in worst
5693                  * case timeout value.
5694                  */
5695         case IXGBE_PCIDEVCTRL2_1_2s:
5696                 return 2000000ul;       /* 2.0 s */
5697         case IXGBE_PCIDEVCTRL2_260_520ms:
5698                 return 520000ul;        /* 520 ms */
5699         case IXGBE_PCIDEVCTRL2_65_130ms:
5700                 return 130000ul;        /* 130 ms */
5701         case IXGBE_PCIDEVCTRL2_16_32ms:
5702                 return 32000ul;         /* 32 ms */
5703         case IXGBE_PCIDEVCTRL2_1_2ms:
5704                 return 2000ul;          /* 2 ms */
5705         case IXGBE_PCIDEVCTRL2_50_100us:
5706                 return 100ul;           /* 100 us */
5707         case IXGBE_PCIDEVCTRL2_16_32ms_def:
5708                 return 32000ul;         /* 32 ms */
5709         default:
5710                 break;
5711         }
5712
5713         /* We shouldn't need to hit this path, but just in case default as
5714          * though completion timeout is not supported and support 32ms.
5715          */
5716         return 32000ul;
5717 }
5718
5719 void ixgbe_disable_rx(struct ixgbe_adapter *adapter)
5720 {
5721         unsigned long wait_delay, delay_interval;
5722         struct ixgbe_hw *hw = &adapter->hw;
5723         int i, wait_loop;
5724         u32 rxdctl;
5725
5726         /* disable receives */
5727         hw->mac.ops.disable_rx(hw);
5728
5729         if (ixgbe_removed(hw->hw_addr))
5730                 return;
5731
5732         /* disable all enabled Rx queues */
5733         for (i = 0; i < adapter->num_rx_queues; i++) {
5734                 struct ixgbe_ring *ring = adapter->rx_ring[i];
5735                 u8 reg_idx = ring->reg_idx;
5736
5737                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5738                 rxdctl &= ~IXGBE_RXDCTL_ENABLE;
5739                 rxdctl |= IXGBE_RXDCTL_SWFLSH;
5740
5741                 /* write value back with RXDCTL.ENABLE bit cleared */
5742                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
5743         }
5744
5745         /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
5746         if (hw->mac.type == ixgbe_mac_82598EB &&
5747             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5748                 return;
5749
5750         /* Determine our minimum delay interval. We will increase this value
5751          * with each subsequent test. This way if the device returns quickly
5752          * we should spend as little time as possible waiting, however as
5753          * the time increases we will wait for larger periods of time.
5754          *
5755          * The trick here is that we increase the interval using the
5756          * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5757          * of that wait is that it totals up to 100x whatever interval we
5758          * choose. Since our minimum wait is 100us we can just divide the
5759          * total timeout by 100 to get our minimum delay interval.
5760          */
5761         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5762
5763         wait_loop = IXGBE_MAX_RX_DESC_POLL;
5764         wait_delay = delay_interval;
5765
5766         while (wait_loop--) {
5767                 usleep_range(wait_delay, wait_delay + 10);
5768                 wait_delay += delay_interval * 2;
5769                 rxdctl = 0;
5770
5771                 /* OR together the reading of all the active RXDCTL registers,
5772                  * and then test the result. We need the disable to complete
5773                  * before we start freeing the memory and invalidating the
5774                  * DMA mappings.
5775                  */
5776                 for (i = 0; i < adapter->num_rx_queues; i++) {
5777                         struct ixgbe_ring *ring = adapter->rx_ring[i];
5778                         u8 reg_idx = ring->reg_idx;
5779
5780                         rxdctl |= IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
5781                 }
5782
5783                 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
5784                         return;
5785         }
5786
5787         e_err(drv,
5788               "RXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5789 }
5790
5791 void ixgbe_disable_tx(struct ixgbe_adapter *adapter)
5792 {
5793         unsigned long wait_delay, delay_interval;
5794         struct ixgbe_hw *hw = &adapter->hw;
5795         int i, wait_loop;
5796         u32 txdctl;
5797
5798         if (ixgbe_removed(hw->hw_addr))
5799                 return;
5800
5801         /* disable all enabled Tx queues */
5802         for (i = 0; i < adapter->num_tx_queues; i++) {
5803                 struct ixgbe_ring *ring = adapter->tx_ring[i];
5804                 u8 reg_idx = ring->reg_idx;
5805
5806                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5807         }
5808
5809         /* disable all enabled XDP Tx queues */
5810         for (i = 0; i < adapter->num_xdp_queues; i++) {
5811                 struct ixgbe_ring *ring = adapter->xdp_ring[i];
5812                 u8 reg_idx = ring->reg_idx;
5813
5814                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5815         }
5816
5817         /* If the link is not up there shouldn't be much in the way of
5818          * pending transactions. Those that are left will be flushed out
5819          * when the reset logic goes through the flush sequence to clean out
5820          * the pending Tx transactions.
5821          */
5822         if (!(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
5823                 goto dma_engine_disable;
5824
5825         /* Determine our minimum delay interval. We will increase this value
5826          * with each subsequent test. This way if the device returns quickly
5827          * we should spend as little time as possible waiting, however as
5828          * the time increases we will wait for larger periods of time.
5829          *
5830          * The trick here is that we increase the interval using the
5831          * following pattern: 1x 3x 5x 7x 9x 11x 13x 15x 17x 19x. The result
5832          * of that wait is that it totals up to 100x whatever interval we
5833          * choose. Since our minimum wait is 100us we can just divide the
5834          * total timeout by 100 to get our minimum delay interval.
5835          */
5836         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
5837
5838         wait_loop = IXGBE_MAX_RX_DESC_POLL;
5839         wait_delay = delay_interval;
5840
5841         while (wait_loop--) {
5842                 usleep_range(wait_delay, wait_delay + 10);
5843                 wait_delay += delay_interval * 2;
5844                 txdctl = 0;
5845
5846                 /* OR together the reading of all the active TXDCTL registers,
5847                  * and then test the result. We need the disable to complete
5848                  * before we start freeing the memory and invalidating the
5849                  * DMA mappings.
5850                  */
5851                 for (i = 0; i < adapter->num_tx_queues; i++) {
5852                         struct ixgbe_ring *ring = adapter->tx_ring[i];
5853                         u8 reg_idx = ring->reg_idx;
5854
5855                         txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5856                 }
5857                 for (i = 0; i < adapter->num_xdp_queues; i++) {
5858                         struct ixgbe_ring *ring = adapter->xdp_ring[i];
5859                         u8 reg_idx = ring->reg_idx;
5860
5861                         txdctl |= IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
5862                 }
5863
5864                 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
5865                         goto dma_engine_disable;
5866         }
5867
5868         e_err(drv,
5869               "TXDCTL.ENABLE for one or more queues not cleared within the polling period\n");
5870
5871 dma_engine_disable:
5872         /* Disable the Tx DMA engine on 82599 and later MAC */
5873         switch (hw->mac.type) {
5874         case ixgbe_mac_82599EB:
5875         case ixgbe_mac_X540:
5876         case ixgbe_mac_X550:
5877         case ixgbe_mac_X550EM_x:
5878         case ixgbe_mac_x550em_a:
5879                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5880                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5881                                  ~IXGBE_DMATXCTL_TE));
5882                 /* fall through */
5883         default:
5884                 break;
5885         }
5886 }
5887
5888 void ixgbe_reset(struct ixgbe_adapter *adapter)
5889 {
5890         struct ixgbe_hw *hw = &adapter->hw;
5891         struct net_device *netdev = adapter->netdev;
5892         int err;
5893
5894         if (ixgbe_removed(hw->hw_addr))
5895                 return;
5896         /* lock SFP init bit to prevent race conditions with the watchdog */
5897         while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5898                 usleep_range(1000, 2000);
5899
5900         /* clear all SFP and link config related flags while holding SFP_INIT */
5901         adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5902                              IXGBE_FLAG2_SFP_NEEDS_RESET);
5903         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5904
5905         err = hw->mac.ops.init_hw(hw);
5906         switch (err) {
5907         case 0:
5908         case IXGBE_ERR_SFP_NOT_PRESENT:
5909         case IXGBE_ERR_SFP_NOT_SUPPORTED:
5910                 break;
5911         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5912                 e_dev_err("master disable timed out\n");
5913                 break;
5914         case IXGBE_ERR_EEPROM_VERSION:
5915                 /* We are running on a pre-production device, log a warning */
5916                 e_dev_warn("This device is a pre-production adapter/LOM. "
5917                            "Please be aware there may be issues associated with "
5918                            "your hardware.  If you are experiencing problems "
5919                            "please contact your Intel or hardware "
5920                            "representative who provided you with this "
5921                            "hardware.\n");
5922                 break;
5923         default:
5924                 e_dev_err("Hardware Error: %d\n", err);
5925         }
5926
5927         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5928
5929         /* flush entries out of MAC table */
5930         ixgbe_flush_sw_mac_table(adapter);
5931         __dev_uc_unsync(netdev, NULL);
5932
5933         /* do not flush user set addresses */
5934         ixgbe_mac_set_default_filter(adapter);
5935
5936         /* update SAN MAC vmdq pool selection */
5937         if (hw->mac.san_mac_rar_index)
5938                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5939
5940         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5941                 ixgbe_ptp_reset(adapter);
5942
5943         if (hw->phy.ops.set_phy_power) {
5944                 if (!netif_running(adapter->netdev) && !adapter->wol)
5945                         hw->phy.ops.set_phy_power(hw, false);
5946                 else
5947                         hw->phy.ops.set_phy_power(hw, true);
5948         }
5949 }
5950
5951 /**
5952  * ixgbe_clean_tx_ring - Free Tx Buffers
5953  * @tx_ring: ring to be cleaned
5954  **/
5955 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5956 {
5957         u16 i = tx_ring->next_to_clean;
5958         struct ixgbe_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
5959
5960         if (tx_ring->xsk_umem) {
5961                 ixgbe_xsk_clean_tx_ring(tx_ring);
5962                 goto out;
5963         }
5964
5965         while (i != tx_ring->next_to_use) {
5966                 union ixgbe_adv_tx_desc *eop_desc, *tx_desc;
5967
5968                 /* Free all the Tx ring sk_buffs */
5969                 if (ring_is_xdp(tx_ring))
5970                         xdp_return_frame(tx_buffer->xdpf);
5971                 else
5972                         dev_kfree_skb_any(tx_buffer->skb);
5973
5974                 /* unmap skb header data */
5975                 dma_unmap_single(tx_ring->dev,
5976                                  dma_unmap_addr(tx_buffer, dma),
5977                                  dma_unmap_len(tx_buffer, len),
5978                                  DMA_TO_DEVICE);
5979
5980                 /* check for eop_desc to determine the end of the packet */
5981                 eop_desc = tx_buffer->next_to_watch;
5982                 tx_desc = IXGBE_TX_DESC(tx_ring, i);
5983
5984                 /* unmap remaining buffers */
5985                 while (tx_desc != eop_desc) {
5986                         tx_buffer++;
5987                         tx_desc++;
5988                         i++;
5989                         if (unlikely(i == tx_ring->count)) {
5990                                 i = 0;
5991                                 tx_buffer = tx_ring->tx_buffer_info;
5992                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
5993                         }
5994
5995                         /* unmap any remaining paged data */
5996                         if (dma_unmap_len(tx_buffer, len))
5997                                 dma_unmap_page(tx_ring->dev,
5998                                                dma_unmap_addr(tx_buffer, dma),
5999                                                dma_unmap_len(tx_buffer, len),
6000                                                DMA_TO_DEVICE);
6001                 }
6002
6003                 /* move us one more past the eop_desc for start of next pkt */
6004                 tx_buffer++;
6005                 i++;
6006                 if (unlikely(i == tx_ring->count)) {
6007                         i = 0;
6008                         tx_buffer = tx_ring->tx_buffer_info;
6009                 }
6010         }
6011
6012         /* reset BQL for queue */
6013         if (!ring_is_xdp(tx_ring))
6014                 netdev_tx_reset_queue(txring_txq(tx_ring));
6015
6016 out:
6017         /* reset next_to_use and next_to_clean */
6018         tx_ring->next_to_use = 0;
6019         tx_ring->next_to_clean = 0;
6020 }
6021
6022 /**
6023  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
6024  * @adapter: board private structure
6025  **/
6026 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
6027 {
6028         int i;
6029
6030         for (i = 0; i < adapter->num_rx_queues; i++)
6031                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
6032 }
6033
6034 /**
6035  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
6036  * @adapter: board private structure
6037  **/
6038 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
6039 {
6040         int i;
6041
6042         for (i = 0; i < adapter->num_tx_queues; i++)
6043                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
6044         for (i = 0; i < adapter->num_xdp_queues; i++)
6045                 ixgbe_clean_tx_ring(adapter->xdp_ring[i]);
6046 }
6047
6048 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
6049 {
6050         struct hlist_node *node2;
6051         struct ixgbe_fdir_filter *filter;
6052
6053         spin_lock(&adapter->fdir_perfect_lock);
6054
6055         hlist_for_each_entry_safe(filter, node2,
6056                                   &adapter->fdir_filter_list, fdir_node) {
6057                 hlist_del(&filter->fdir_node);
6058                 kfree(filter);
6059         }
6060         adapter->fdir_filter_count = 0;
6061
6062         spin_unlock(&adapter->fdir_perfect_lock);
6063 }
6064
6065 void ixgbe_down(struct ixgbe_adapter *adapter)
6066 {
6067         struct net_device *netdev = adapter->netdev;
6068         struct ixgbe_hw *hw = &adapter->hw;
6069         int i;
6070
6071         /* signal that we are down to the interrupt handler */
6072         if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
6073                 return; /* do nothing if already down */
6074
6075         /* Shut off incoming Tx traffic */
6076         netif_tx_stop_all_queues(netdev);
6077
6078         /* call carrier off first to avoid false dev_watchdog timeouts */
6079         netif_carrier_off(netdev);
6080         netif_tx_disable(netdev);
6081
6082         /* Disable Rx */
6083         ixgbe_disable_rx(adapter);
6084
6085         /* synchronize_rcu() needed for pending XDP buffers to drain */
6086         if (adapter->xdp_ring[0])
6087                 synchronize_rcu();
6088
6089         ixgbe_irq_disable(adapter);
6090
6091         ixgbe_napi_disable_all(adapter);
6092
6093         clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
6094         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6095         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6096
6097         del_timer_sync(&adapter->service_timer);
6098
6099         if (adapter->num_vfs) {
6100                 /* Clear EITR Select mapping */
6101                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
6102
6103                 /* Mark all the VFs as inactive */
6104                 for (i = 0 ; i < adapter->num_vfs; i++)
6105                         adapter->vfinfo[i].clear_to_send = false;
6106
6107                 /* ping all the active vfs to let them know we are going down */
6108                 ixgbe_ping_all_vfs(adapter);
6109
6110                 /* Disable all VFTE/VFRE TX/RX */
6111                 ixgbe_disable_tx_rx(adapter);
6112         }
6113
6114         /* disable transmits in the hardware now that interrupts are off */
6115         ixgbe_disable_tx(adapter);
6116
6117         if (!pci_channel_offline(adapter->pdev))
6118                 ixgbe_reset(adapter);
6119
6120         /* power down the optics for 82599 SFP+ fiber */
6121         if (hw->mac.ops.disable_tx_laser)
6122                 hw->mac.ops.disable_tx_laser(hw);
6123
6124         ixgbe_clean_all_tx_rings(adapter);
6125         ixgbe_clean_all_rx_rings(adapter);
6126 }
6127
6128 /**
6129  * ixgbe_eee_capable - helper function to determine EEE support on X550
6130  * @adapter: board private structure
6131  */
6132 static void ixgbe_set_eee_capable(struct ixgbe_adapter *adapter)
6133 {
6134         struct ixgbe_hw *hw = &adapter->hw;
6135
6136         switch (hw->device_id) {
6137         case IXGBE_DEV_ID_X550EM_A_1G_T:
6138         case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6139                 if (!hw->phy.eee_speeds_supported)
6140                         break;
6141                 adapter->flags2 |= IXGBE_FLAG2_EEE_CAPABLE;
6142                 if (!hw->phy.eee_speeds_advertised)
6143                         break;
6144                 adapter->flags2 |= IXGBE_FLAG2_EEE_ENABLED;
6145                 break;
6146         default:
6147                 adapter->flags2 &= ~IXGBE_FLAG2_EEE_CAPABLE;
6148                 adapter->flags2 &= ~IXGBE_FLAG2_EEE_ENABLED;
6149                 break;
6150         }
6151 }
6152
6153 /**
6154  * ixgbe_tx_timeout - Respond to a Tx Hang
6155  * @netdev: network interface device structure
6156  **/
6157 static void ixgbe_tx_timeout(struct net_device *netdev)
6158 {
6159         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6160
6161         /* Do the reset outside of interrupt context */
6162         ixgbe_tx_timeout_reset(adapter);
6163 }
6164
6165 #ifdef CONFIG_IXGBE_DCB
6166 static void ixgbe_init_dcb(struct ixgbe_adapter *adapter)
6167 {
6168         struct ixgbe_hw *hw = &adapter->hw;
6169         struct tc_configuration *tc;
6170         int j;
6171
6172         switch (hw->mac.type) {
6173         case ixgbe_mac_82598EB:
6174         case ixgbe_mac_82599EB:
6175                 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
6176                 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
6177                 break;
6178         case ixgbe_mac_X540:
6179         case ixgbe_mac_X550:
6180                 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
6181                 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
6182                 break;
6183         case ixgbe_mac_X550EM_x:
6184         case ixgbe_mac_x550em_a:
6185         default:
6186                 adapter->dcb_cfg.num_tcs.pg_tcs = DEF_TRAFFIC_CLASS;
6187                 adapter->dcb_cfg.num_tcs.pfc_tcs = DEF_TRAFFIC_CLASS;
6188                 break;
6189         }
6190
6191         /* Configure DCB traffic classes */
6192         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
6193                 tc = &adapter->dcb_cfg.tc_config[j];
6194                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
6195                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
6196                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
6197                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
6198                 tc->dcb_pfc = pfc_disabled;
6199         }
6200
6201         /* Initialize default user to priority mapping, UPx->TC0 */
6202         tc = &adapter->dcb_cfg.tc_config[0];
6203         tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
6204         tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
6205
6206         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
6207         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
6208         adapter->dcb_cfg.pfc_mode_enable = false;
6209         adapter->dcb_set_bitmap = 0x00;
6210         if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
6211                 adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
6212         memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
6213                sizeof(adapter->temp_dcb_cfg));
6214 }
6215 #endif
6216
6217 /**
6218  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
6219  * @adapter: board private structure to initialize
6220  * @ii: pointer to ixgbe_info for device
6221  *
6222  * ixgbe_sw_init initializes the Adapter private data structure.
6223  * Fields are initialized based on PCI device information and
6224  * OS network device settings (MTU size).
6225  **/
6226 static int ixgbe_sw_init(struct ixgbe_adapter *adapter,
6227                          const struct ixgbe_info *ii)
6228 {
6229         struct ixgbe_hw *hw = &adapter->hw;
6230         struct pci_dev *pdev = adapter->pdev;
6231         unsigned int rss, fdir;
6232         u32 fwsm;
6233         int i;
6234
6235         /* PCI config space info */
6236
6237         hw->vendor_id = pdev->vendor;
6238         hw->device_id = pdev->device;
6239         hw->revision_id = pdev->revision;
6240         hw->subsystem_vendor_id = pdev->subsystem_vendor;
6241         hw->subsystem_device_id = pdev->subsystem_device;
6242
6243         /* get_invariants needs the device IDs */
6244         ii->get_invariants(hw);
6245
6246         /* Set common capability flags and settings */
6247         rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
6248         adapter->ring_feature[RING_F_RSS].limit = rss;
6249         adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
6250         adapter->max_q_vectors = MAX_Q_VECTORS_82599;
6251         adapter->atr_sample_rate = 20;
6252         fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
6253         adapter->ring_feature[RING_F_FDIR].limit = fdir;
6254         adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
6255         adapter->ring_feature[RING_F_VMDQ].limit = 1;
6256 #ifdef CONFIG_IXGBE_DCA
6257         adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
6258 #endif
6259 #ifdef CONFIG_IXGBE_DCB
6260         adapter->flags |= IXGBE_FLAG_DCB_CAPABLE;
6261         adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
6262 #endif
6263 #ifdef IXGBE_FCOE
6264         adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
6265         adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6266 #ifdef CONFIG_IXGBE_DCB
6267         /* Default traffic class to use for FCoE */
6268         adapter->fcoe.up = IXGBE_FCOE_DEFTC;
6269 #endif /* CONFIG_IXGBE_DCB */
6270 #endif /* IXGBE_FCOE */
6271
6272         /* initialize static ixgbe jump table entries */
6273         adapter->jump_tables[0] = kzalloc(sizeof(*adapter->jump_tables[0]),
6274                                           GFP_KERNEL);
6275         if (!adapter->jump_tables[0])
6276                 return -ENOMEM;
6277         adapter->jump_tables[0]->mat = ixgbe_ipv4_fields;
6278
6279         for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++)
6280                 adapter->jump_tables[i] = NULL;
6281
6282         adapter->mac_table = kcalloc(hw->mac.num_rar_entries,
6283                                      sizeof(struct ixgbe_mac_addr),
6284                                      GFP_KERNEL);
6285         if (!adapter->mac_table)
6286                 return -ENOMEM;
6287
6288         if (ixgbe_init_rss_key(adapter))
6289                 return -ENOMEM;
6290
6291         adapter->af_xdp_zc_qps = bitmap_zalloc(MAX_XDP_QUEUES, GFP_KERNEL);
6292         if (!adapter->af_xdp_zc_qps)
6293                 return -ENOMEM;
6294
6295         /* Set MAC specific capability flags and exceptions */
6296         switch (hw->mac.type) {
6297         case ixgbe_mac_82598EB:
6298                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
6299
6300                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
6301                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
6302
6303                 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
6304                 adapter->ring_feature[RING_F_FDIR].limit = 0;
6305                 adapter->atr_sample_rate = 0;
6306                 adapter->fdir_pballoc = 0;
6307 #ifdef IXGBE_FCOE
6308                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6309                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
6310 #ifdef CONFIG_IXGBE_DCB
6311                 adapter->fcoe.up = 0;
6312 #endif /* IXGBE_DCB */
6313 #endif /* IXGBE_FCOE */
6314                 break;
6315         case ixgbe_mac_82599EB:
6316                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
6317                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6318                 break;
6319         case ixgbe_mac_X540:
6320                 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
6321                 if (fwsm & IXGBE_FWSM_TS_ENABLED)
6322                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6323                 break;
6324         case ixgbe_mac_x550em_a:
6325                 adapter->flags |= IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE;
6326                 switch (hw->device_id) {
6327                 case IXGBE_DEV_ID_X550EM_A_1G_T:
6328                 case IXGBE_DEV_ID_X550EM_A_1G_T_L:
6329                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6330                         break;
6331                 default:
6332                         break;
6333                 }
6334         /* fall through */
6335         case ixgbe_mac_X550EM_x:
6336 #ifdef CONFIG_IXGBE_DCB
6337                 adapter->flags &= ~IXGBE_FLAG_DCB_CAPABLE;
6338 #endif
6339 #ifdef IXGBE_FCOE
6340                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
6341 #ifdef CONFIG_IXGBE_DCB
6342                 adapter->fcoe.up = 0;
6343 #endif /* IXGBE_DCB */
6344 #endif /* IXGBE_FCOE */
6345         /* Fall Through */
6346         case ixgbe_mac_X550:
6347                 if (hw->mac.type == ixgbe_mac_X550)
6348                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
6349 #ifdef CONFIG_IXGBE_DCA
6350                 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
6351 #endif
6352                 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
6353                 break;
6354         default:
6355                 break;
6356         }
6357
6358 #ifdef IXGBE_FCOE
6359         /* FCoE support exists, always init the FCoE lock */
6360         spin_lock_init(&adapter->fcoe.lock);
6361
6362 #endif
6363         /* n-tuple support exists, always init our spinlock */
6364         spin_lock_init(&adapter->fdir_perfect_lock);
6365
6366 #ifdef CONFIG_IXGBE_DCB
6367         ixgbe_init_dcb(adapter);
6368 #endif
6369         ixgbe_init_ipsec_offload(adapter);
6370
6371         /* default flow control settings */
6372         hw->fc.requested_mode = ixgbe_fc_full;
6373         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
6374         ixgbe_pbthresh_setup(adapter);
6375         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
6376         hw->fc.send_xon = true;
6377         hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
6378
6379 #ifdef CONFIG_PCI_IOV
6380         if (max_vfs > 0)
6381                 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
6382
6383         /* assign number of SR-IOV VFs */
6384         if (hw->mac.type != ixgbe_mac_82598EB) {
6385                 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
6386                         max_vfs = 0;
6387                         e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
6388                 }
6389         }
6390 #endif /* CONFIG_PCI_IOV */
6391
6392         /* enable itr by default in dynamic mode */
6393         adapter->rx_itr_setting = 1;
6394         adapter->tx_itr_setting = 1;
6395
6396         /* set default ring sizes */
6397         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
6398         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
6399
6400         /* set default work limits */
6401         adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
6402
6403         /* initialize eeprom parameters */
6404         if (ixgbe_init_eeprom_params_generic(hw)) {
6405                 e_dev_err("EEPROM initialization failed\n");
6406                 return -EIO;
6407         }
6408
6409         /* PF holds first pool slot */
6410         set_bit(0, adapter->fwd_bitmask);
6411         set_bit(__IXGBE_DOWN, &adapter->state);
6412
6413         return 0;
6414 }
6415
6416 /**
6417  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
6418  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
6419  *
6420  * Return 0 on success, negative on failure
6421  **/
6422 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
6423 {
6424         struct device *dev = tx_ring->dev;
6425         int orig_node = dev_to_node(dev);
6426         int ring_node = NUMA_NO_NODE;
6427         int size;
6428
6429         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
6430
6431         if (tx_ring->q_vector)
6432                 ring_node = tx_ring->q_vector->numa_node;
6433
6434         tx_ring->tx_buffer_info = vmalloc_node(size, ring_node);
6435         if (!tx_ring->tx_buffer_info)
6436                 tx_ring->tx_buffer_info = vmalloc(size);
6437         if (!tx_ring->tx_buffer_info)
6438                 goto err;
6439
6440         /* round up to nearest 4K */
6441         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
6442         tx_ring->size = ALIGN(tx_ring->size, 4096);
6443
6444         set_dev_node(dev, ring_node);
6445         tx_ring->desc = dma_alloc_coherent(dev,
6446                                            tx_ring->size,
6447                                            &tx_ring->dma,
6448                                            GFP_KERNEL);
6449         set_dev_node(dev, orig_node);
6450         if (!tx_ring->desc)
6451                 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
6452                                                    &tx_ring->dma, GFP_KERNEL);
6453         if (!tx_ring->desc)
6454                 goto err;
6455
6456         tx_ring->next_to_use = 0;
6457         tx_ring->next_to_clean = 0;
6458         return 0;
6459
6460 err:
6461         vfree(tx_ring->tx_buffer_info);
6462         tx_ring->tx_buffer_info = NULL;
6463         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
6464         return -ENOMEM;
6465 }
6466
6467 /**
6468  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
6469  * @adapter: board private structure
6470  *
6471  * If this function returns with an error, then it's possible one or
6472  * more of the rings is populated (while the rest are not).  It is the
6473  * callers duty to clean those orphaned rings.
6474  *
6475  * Return 0 on success, negative on failure
6476  **/
6477 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
6478 {
6479         int i, j = 0, err = 0;
6480
6481         for (i = 0; i < adapter->num_tx_queues; i++) {
6482                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
6483                 if (!err)
6484                         continue;
6485
6486                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
6487                 goto err_setup_tx;
6488         }
6489         for (j = 0; j < adapter->num_xdp_queues; j++) {
6490                 err = ixgbe_setup_tx_resources(adapter->xdp_ring[j]);
6491                 if (!err)
6492                         continue;
6493
6494                 e_err(probe, "Allocation for Tx Queue %u failed\n", j);
6495                 goto err_setup_tx;
6496         }
6497
6498         return 0;
6499 err_setup_tx:
6500         /* rewind the index freeing the rings as we go */
6501         while (j--)
6502                 ixgbe_free_tx_resources(adapter->xdp_ring[j]);
6503         while (i--)
6504                 ixgbe_free_tx_resources(adapter->tx_ring[i]);
6505         return err;
6506 }
6507
6508 /**
6509  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
6510  * @adapter: pointer to ixgbe_adapter
6511  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
6512  *
6513  * Returns 0 on success, negative on failure
6514  **/
6515 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
6516                              struct ixgbe_ring *rx_ring)
6517 {
6518         struct device *dev = rx_ring->dev;
6519         int orig_node = dev_to_node(dev);
6520         int ring_node = NUMA_NO_NODE;
6521         int size;
6522
6523         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
6524
6525         if (rx_ring->q_vector)
6526                 ring_node = rx_ring->q_vector->numa_node;
6527
6528         rx_ring->rx_buffer_info = vmalloc_node(size, ring_node);
6529         if (!rx_ring->rx_buffer_info)
6530                 rx_ring->rx_buffer_info = vmalloc(size);
6531         if (!rx_ring->rx_buffer_info)
6532                 goto err;
6533
6534         /* Round up to nearest 4K */
6535         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
6536         rx_ring->size = ALIGN(rx_ring->size, 4096);
6537
6538         set_dev_node(dev, ring_node);
6539         rx_ring->desc = dma_alloc_coherent(dev,
6540                                            rx_ring->size,
6541                                            &rx_ring->dma,
6542                                            GFP_KERNEL);
6543         set_dev_node(dev, orig_node);
6544         if (!rx_ring->desc)
6545                 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
6546                                                    &rx_ring->dma, GFP_KERNEL);
6547         if (!rx_ring->desc)
6548                 goto err;
6549
6550         rx_ring->next_to_clean = 0;
6551         rx_ring->next_to_use = 0;
6552
6553         /* XDP RX-queue info */
6554         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, adapter->netdev,
6555                              rx_ring->queue_index) < 0)
6556                 goto err;
6557
6558         rx_ring->xdp_prog = adapter->xdp_prog;
6559
6560         return 0;
6561 err:
6562         vfree(rx_ring->rx_buffer_info);
6563         rx_ring->rx_buffer_info = NULL;
6564         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
6565         return -ENOMEM;
6566 }
6567
6568 /**
6569  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
6570  * @adapter: board private structure
6571  *
6572  * If this function returns with an error, then it's possible one or
6573  * more of the rings is populated (while the rest are not).  It is the
6574  * callers duty to clean those orphaned rings.
6575  *
6576  * Return 0 on success, negative on failure
6577  **/
6578 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
6579 {
6580         int i, err = 0;
6581
6582         for (i = 0; i < adapter->num_rx_queues; i++) {
6583                 err = ixgbe_setup_rx_resources(adapter, adapter->rx_ring[i]);
6584                 if (!err)
6585                         continue;
6586
6587                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
6588                 goto err_setup_rx;
6589         }
6590
6591 #ifdef IXGBE_FCOE
6592         err = ixgbe_setup_fcoe_ddp_resources(adapter);
6593         if (!err)
6594 #endif
6595                 return 0;
6596 err_setup_rx:
6597         /* rewind the index freeing the rings as we go */
6598         while (i--)
6599                 ixgbe_free_rx_resources(adapter->rx_ring[i]);
6600         return err;
6601 }
6602
6603 /**
6604  * ixgbe_free_tx_resources - Free Tx Resources per Queue
6605  * @tx_ring: Tx descriptor ring for a specific queue
6606  *
6607  * Free all transmit software resources
6608  **/
6609 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
6610 {
6611         ixgbe_clean_tx_ring(tx_ring);
6612
6613         vfree(tx_ring->tx_buffer_info);
6614         tx_ring->tx_buffer_info = NULL;
6615
6616         /* if not set, then don't free */
6617         if (!tx_ring->desc)
6618                 return;
6619
6620         dma_free_coherent(tx_ring->dev, tx_ring->size,
6621                           tx_ring->desc, tx_ring->dma);
6622
6623         tx_ring->desc = NULL;
6624 }
6625
6626 /**
6627  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
6628  * @adapter: board private structure
6629  *
6630  * Free all transmit software resources
6631  **/
6632 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
6633 {
6634         int i;
6635
6636         for (i = 0; i < adapter->num_tx_queues; i++)
6637                 if (adapter->tx_ring[i]->desc)
6638                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
6639         for (i = 0; i < adapter->num_xdp_queues; i++)
6640                 if (adapter->xdp_ring[i]->desc)
6641                         ixgbe_free_tx_resources(adapter->xdp_ring[i]);
6642 }
6643
6644 /**
6645  * ixgbe_free_rx_resources - Free Rx Resources
6646  * @rx_ring: ring to clean the resources from
6647  *
6648  * Free all receive software resources
6649  **/
6650 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
6651 {
6652         ixgbe_clean_rx_ring(rx_ring);
6653
6654         rx_ring->xdp_prog = NULL;
6655         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
6656         vfree(rx_ring->rx_buffer_info);
6657         rx_ring->rx_buffer_info = NULL;
6658
6659         /* if not set, then don't free */
6660         if (!rx_ring->desc)
6661                 return;
6662
6663         dma_free_coherent(rx_ring->dev, rx_ring->size,
6664                           rx_ring->desc, rx_ring->dma);
6665
6666         rx_ring->desc = NULL;
6667 }
6668
6669 /**
6670  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
6671  * @adapter: board private structure
6672  *
6673  * Free all receive software resources
6674  **/
6675 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
6676 {
6677         int i;
6678
6679 #ifdef IXGBE_FCOE
6680         ixgbe_free_fcoe_ddp_resources(adapter);
6681
6682 #endif
6683         for (i = 0; i < adapter->num_rx_queues; i++)
6684                 if (adapter->rx_ring[i]->desc)
6685                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
6686 }
6687
6688 /**
6689  * ixgbe_change_mtu - Change the Maximum Transfer Unit
6690  * @netdev: network interface device structure
6691  * @new_mtu: new value for maximum frame size
6692  *
6693  * Returns 0 on success, negative on failure
6694  **/
6695 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
6696 {
6697         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6698
6699         if (adapter->xdp_prog) {
6700                 int new_frame_size = new_mtu + ETH_HLEN + ETH_FCS_LEN +
6701                                      VLAN_HLEN;
6702                 int i;
6703
6704                 for (i = 0; i < adapter->num_rx_queues; i++) {
6705                         struct ixgbe_ring *ring = adapter->rx_ring[i];
6706
6707                         if (new_frame_size > ixgbe_rx_bufsz(ring)) {
6708                                 e_warn(probe, "Requested MTU size is not supported with XDP\n");
6709                                 return -EINVAL;
6710                         }
6711                 }
6712         }
6713
6714         /*
6715          * For 82599EB we cannot allow legacy VFs to enable their receive
6716          * paths when MTU greater than 1500 is configured.  So display a
6717          * warning that legacy VFs will be disabled.
6718          */
6719         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
6720             (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
6721             (new_mtu > ETH_DATA_LEN))
6722                 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
6723
6724         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
6725
6726         /* must set new MTU before calling down or up */
6727         netdev->mtu = new_mtu;
6728
6729         if (netif_running(netdev))
6730                 ixgbe_reinit_locked(adapter);
6731
6732         return 0;
6733 }
6734
6735 /**
6736  * ixgbe_open - Called when a network interface is made active
6737  * @netdev: network interface device structure
6738  *
6739  * Returns 0 on success, negative value on failure
6740  *
6741  * The open entry point is called when a network interface is made
6742  * active by the system (IFF_UP).  At this point all resources needed
6743  * for transmit and receive operations are allocated, the interrupt
6744  * handler is registered with the OS, the watchdog timer is started,
6745  * and the stack is notified that the interface is ready.
6746  **/
6747 int ixgbe_open(struct net_device *netdev)
6748 {
6749         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6750         struct ixgbe_hw *hw = &adapter->hw;
6751         int err, queues;
6752
6753         /* disallow open during test */
6754         if (test_bit(__IXGBE_TESTING, &adapter->state))
6755                 return -EBUSY;
6756
6757         netif_carrier_off(netdev);
6758
6759         /* allocate transmit descriptors */
6760         err = ixgbe_setup_all_tx_resources(adapter);
6761         if (err)
6762                 goto err_setup_tx;
6763
6764         /* allocate receive descriptors */
6765         err = ixgbe_setup_all_rx_resources(adapter);
6766         if (err)
6767                 goto err_setup_rx;
6768
6769         ixgbe_configure(adapter);
6770
6771         err = ixgbe_request_irq(adapter);
6772         if (err)
6773                 goto err_req_irq;
6774
6775         /* Notify the stack of the actual queue counts. */
6776         queues = adapter->num_tx_queues;
6777         err = netif_set_real_num_tx_queues(netdev, queues);
6778         if (err)
6779                 goto err_set_queues;
6780
6781         queues = adapter->num_rx_queues;
6782         err = netif_set_real_num_rx_queues(netdev, queues);
6783         if (err)
6784                 goto err_set_queues;
6785
6786         ixgbe_ptp_init(adapter);
6787
6788         ixgbe_up_complete(adapter);
6789
6790         ixgbe_clear_udp_tunnel_port(adapter, IXGBE_VXLANCTRL_ALL_UDPPORT_MASK);
6791         udp_tunnel_get_rx_info(netdev);
6792
6793         return 0;
6794
6795 err_set_queues:
6796         ixgbe_free_irq(adapter);
6797 err_req_irq:
6798         ixgbe_free_all_rx_resources(adapter);
6799         if (hw->phy.ops.set_phy_power && !adapter->wol)
6800                 hw->phy.ops.set_phy_power(&adapter->hw, false);
6801 err_setup_rx:
6802         ixgbe_free_all_tx_resources(adapter);
6803 err_setup_tx:
6804         ixgbe_reset(adapter);
6805
6806         return err;
6807 }
6808
6809 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6810 {
6811         ixgbe_ptp_suspend(adapter);
6812
6813         if (adapter->hw.phy.ops.enter_lplu) {
6814                 adapter->hw.phy.reset_disable = true;
6815                 ixgbe_down(adapter);
6816                 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6817                 adapter->hw.phy.reset_disable = false;
6818         } else {
6819                 ixgbe_down(adapter);
6820         }
6821
6822         ixgbe_free_irq(adapter);
6823
6824         ixgbe_free_all_tx_resources(adapter);
6825         ixgbe_free_all_rx_resources(adapter);
6826 }
6827
6828 /**
6829  * ixgbe_close - Disables a network interface
6830  * @netdev: network interface device structure
6831  *
6832  * Returns 0, this is not allowed to fail
6833  *
6834  * The close entry point is called when an interface is de-activated
6835  * by the OS.  The hardware is still under the drivers control, but
6836  * needs to be disabled.  A global MAC reset is issued to stop the
6837  * hardware, and all transmit and receive resources are freed.
6838  **/
6839 int ixgbe_close(struct net_device *netdev)
6840 {
6841         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6842
6843         ixgbe_ptp_stop(adapter);
6844
6845         if (netif_device_present(netdev))
6846                 ixgbe_close_suspend(adapter);
6847
6848         ixgbe_fdir_filter_exit(adapter);
6849
6850         ixgbe_release_hw_control(adapter);
6851
6852         return 0;
6853 }
6854
6855 #ifdef CONFIG_PM
6856 static int ixgbe_resume(struct pci_dev *pdev)
6857 {
6858         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6859         struct net_device *netdev = adapter->netdev;
6860         u32 err;
6861
6862         adapter->hw.hw_addr = adapter->io_addr;
6863         pci_set_power_state(pdev, PCI_D0);
6864         pci_restore_state(pdev);
6865         /*
6866          * pci_restore_state clears dev->state_saved so call
6867          * pci_save_state to restore it.
6868          */
6869         pci_save_state(pdev);
6870
6871         err = pci_enable_device_mem(pdev);
6872         if (err) {
6873                 e_dev_err("Cannot enable PCI device from suspend\n");
6874                 return err;
6875         }
6876         smp_mb__before_atomic();
6877         clear_bit(__IXGBE_DISABLED, &adapter->state);
6878         pci_set_master(pdev);
6879
6880         pci_wake_from_d3(pdev, false);
6881
6882         ixgbe_reset(adapter);
6883
6884         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6885
6886         rtnl_lock();
6887         err = ixgbe_init_interrupt_scheme(adapter);
6888         if (!err && netif_running(netdev))
6889                 err = ixgbe_open(netdev);
6890
6891
6892         if (!err)
6893                 netif_device_attach(netdev);
6894         rtnl_unlock();
6895
6896         return err;
6897 }
6898 #endif /* CONFIG_PM */
6899
6900 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6901 {
6902         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6903         struct net_device *netdev = adapter->netdev;
6904         struct ixgbe_hw *hw = &adapter->hw;
6905         u32 ctrl;
6906         u32 wufc = adapter->wol;
6907 #ifdef CONFIG_PM
6908         int retval = 0;
6909 #endif
6910
6911         rtnl_lock();
6912         netif_device_detach(netdev);
6913
6914         if (netif_running(netdev))
6915                 ixgbe_close_suspend(adapter);
6916
6917         ixgbe_clear_interrupt_scheme(adapter);
6918         rtnl_unlock();
6919
6920 #ifdef CONFIG_PM
6921         retval = pci_save_state(pdev);
6922         if (retval)
6923                 return retval;
6924
6925 #endif
6926         if (hw->mac.ops.stop_link_on_d3)
6927                 hw->mac.ops.stop_link_on_d3(hw);
6928
6929         if (wufc) {
6930                 u32 fctrl;
6931
6932                 ixgbe_set_rx_mode(netdev);
6933
6934                 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6935                 if (hw->mac.ops.enable_tx_laser)
6936                         hw->mac.ops.enable_tx_laser(hw);
6937
6938                 /* enable the reception of multicast packets */
6939                 fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6940                 fctrl |= IXGBE_FCTRL_MPE;
6941                 IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6942
6943                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6944                 ctrl |= IXGBE_CTRL_GIO_DIS;
6945                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6946
6947                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6948         } else {
6949                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6950                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6951         }
6952
6953         switch (hw->mac.type) {
6954         case ixgbe_mac_82598EB:
6955                 pci_wake_from_d3(pdev, false);
6956                 break;
6957         case ixgbe_mac_82599EB:
6958         case ixgbe_mac_X540:
6959         case ixgbe_mac_X550:
6960         case ixgbe_mac_X550EM_x:
6961         case ixgbe_mac_x550em_a:
6962                 pci_wake_from_d3(pdev, !!wufc);
6963                 break;
6964         default:
6965                 break;
6966         }
6967
6968         *enable_wake = !!wufc;
6969         if (hw->phy.ops.set_phy_power && !*enable_wake)
6970                 hw->phy.ops.set_phy_power(hw, false);
6971
6972         ixgbe_release_hw_control(adapter);
6973
6974         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6975                 pci_disable_device(pdev);
6976
6977         return 0;
6978 }
6979
6980 #ifdef CONFIG_PM
6981 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6982 {
6983         int retval;
6984         bool wake;
6985
6986         retval = __ixgbe_shutdown(pdev, &wake);
6987         if (retval)
6988                 return retval;
6989
6990         if (wake) {
6991                 pci_prepare_to_sleep(pdev);
6992         } else {
6993                 pci_wake_from_d3(pdev, false);
6994                 pci_set_power_state(pdev, PCI_D3hot);
6995         }
6996
6997         return 0;
6998 }
6999 #endif /* CONFIG_PM */
7000
7001 static void ixgbe_shutdown(struct pci_dev *pdev)
7002 {
7003         bool wake;
7004
7005         __ixgbe_shutdown(pdev, &wake);
7006
7007         if (system_state == SYSTEM_POWER_OFF) {
7008                 pci_wake_from_d3(pdev, wake);
7009                 pci_set_power_state(pdev, PCI_D3hot);
7010         }
7011 }
7012
7013 /**
7014  * ixgbe_update_stats - Update the board statistics counters.
7015  * @adapter: board private structure
7016  **/
7017 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
7018 {
7019         struct net_device *netdev = adapter->netdev;
7020         struct ixgbe_hw *hw = &adapter->hw;
7021         struct ixgbe_hw_stats *hwstats = &adapter->stats;
7022         u64 total_mpc = 0;
7023         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
7024         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
7025         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
7026         u64 alloc_rx_page = 0;
7027         u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
7028
7029         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7030             test_bit(__IXGBE_RESETTING, &adapter->state))
7031                 return;
7032
7033         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
7034                 u64 rsc_count = 0;
7035                 u64 rsc_flush = 0;
7036                 for (i = 0; i < adapter->num_rx_queues; i++) {
7037                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
7038                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
7039                 }
7040                 adapter->rsc_total_count = rsc_count;
7041                 adapter->rsc_total_flush = rsc_flush;
7042         }
7043
7044         for (i = 0; i < adapter->num_rx_queues; i++) {
7045                 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
7046                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
7047                 alloc_rx_page += rx_ring->rx_stats.alloc_rx_page;
7048                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
7049                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
7050                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
7051                 bytes += rx_ring->stats.bytes;
7052                 packets += rx_ring->stats.packets;
7053         }
7054         adapter->non_eop_descs = non_eop_descs;
7055         adapter->alloc_rx_page = alloc_rx_page;
7056         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
7057         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
7058         adapter->hw_csum_rx_error = hw_csum_rx_error;
7059         netdev->stats.rx_bytes = bytes;
7060         netdev->stats.rx_packets = packets;
7061
7062         bytes = 0;
7063         packets = 0;
7064         /* gather some stats to the adapter struct that are per queue */
7065         for (i = 0; i < adapter->num_tx_queues; i++) {
7066                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7067                 restart_queue += tx_ring->tx_stats.restart_queue;
7068                 tx_busy += tx_ring->tx_stats.tx_busy;
7069                 bytes += tx_ring->stats.bytes;
7070                 packets += tx_ring->stats.packets;
7071         }
7072         for (i = 0; i < adapter->num_xdp_queues; i++) {
7073                 struct ixgbe_ring *xdp_ring = adapter->xdp_ring[i];
7074
7075                 restart_queue += xdp_ring->tx_stats.restart_queue;
7076                 tx_busy += xdp_ring->tx_stats.tx_busy;
7077                 bytes += xdp_ring->stats.bytes;
7078                 packets += xdp_ring->stats.packets;
7079         }
7080         adapter->restart_queue = restart_queue;
7081         adapter->tx_busy = tx_busy;
7082         netdev->stats.tx_bytes = bytes;
7083         netdev->stats.tx_packets = packets;
7084
7085         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
7086
7087         /* 8 register reads */
7088         for (i = 0; i < 8; i++) {
7089                 /* for packet buffers not used, the register should read 0 */
7090                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
7091                 missed_rx += mpc;
7092                 hwstats->mpc[i] += mpc;
7093                 total_mpc += hwstats->mpc[i];
7094                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
7095                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
7096                 switch (hw->mac.type) {
7097                 case ixgbe_mac_82598EB:
7098                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
7099                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
7100                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
7101                         hwstats->pxonrxc[i] +=
7102                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
7103                         break;
7104                 case ixgbe_mac_82599EB:
7105                 case ixgbe_mac_X540:
7106                 case ixgbe_mac_X550:
7107                 case ixgbe_mac_X550EM_x:
7108                 case ixgbe_mac_x550em_a:
7109                         hwstats->pxonrxc[i] +=
7110                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
7111                         break;
7112                 default:
7113                         break;
7114                 }
7115         }
7116
7117         /*16 register reads */
7118         for (i = 0; i < 16; i++) {
7119                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
7120                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
7121                 if ((hw->mac.type == ixgbe_mac_82599EB) ||
7122                     (hw->mac.type == ixgbe_mac_X540) ||
7123                     (hw->mac.type == ixgbe_mac_X550) ||
7124                     (hw->mac.type == ixgbe_mac_X550EM_x) ||
7125                     (hw->mac.type == ixgbe_mac_x550em_a)) {
7126                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
7127                         IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
7128                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
7129                         IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
7130                 }
7131         }
7132
7133         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
7134         /* work around hardware counting issue */
7135         hwstats->gprc -= missed_rx;
7136
7137         ixgbe_update_xoff_received(adapter);
7138
7139         /* 82598 hardware only has a 32 bit counter in the high register */
7140         switch (hw->mac.type) {
7141         case ixgbe_mac_82598EB:
7142                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
7143                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
7144                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
7145                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
7146                 break;
7147         case ixgbe_mac_X540:
7148         case ixgbe_mac_X550:
7149         case ixgbe_mac_X550EM_x:
7150         case ixgbe_mac_x550em_a:
7151                 /* OS2BMC stats are X540 and later */
7152                 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
7153                 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
7154                 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
7155                 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
7156                 /* fall through */
7157         case ixgbe_mac_82599EB:
7158                 for (i = 0; i < 16; i++)
7159                         adapter->hw_rx_no_dma_resources +=
7160                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
7161                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
7162                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
7163                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
7164                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
7165                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
7166                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
7167                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
7168                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
7169                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
7170 #ifdef IXGBE_FCOE
7171                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
7172                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
7173                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
7174                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
7175                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
7176                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
7177                 /* Add up per cpu counters for total ddp aloc fail */
7178                 if (adapter->fcoe.ddp_pool) {
7179                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
7180                         struct ixgbe_fcoe_ddp_pool *ddp_pool;
7181                         unsigned int cpu;
7182                         u64 noddp = 0, noddp_ext_buff = 0;
7183                         for_each_possible_cpu(cpu) {
7184                                 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
7185                                 noddp += ddp_pool->noddp;
7186                                 noddp_ext_buff += ddp_pool->noddp_ext_buff;
7187                         }
7188                         hwstats->fcoe_noddp = noddp;
7189                         hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
7190                 }
7191 #endif /* IXGBE_FCOE */
7192                 break;
7193         default:
7194                 break;
7195         }
7196         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
7197         hwstats->bprc += bprc;
7198         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
7199         if (hw->mac.type == ixgbe_mac_82598EB)
7200                 hwstats->mprc -= bprc;
7201         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
7202         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
7203         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
7204         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
7205         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
7206         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
7207         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
7208         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
7209         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
7210         hwstats->lxontxc += lxon;
7211         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
7212         hwstats->lxofftxc += lxoff;
7213         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
7214         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
7215         /*
7216          * 82598 errata - tx of flow control packets is included in tx counters
7217          */
7218         xon_off_tot = lxon + lxoff;
7219         hwstats->gptc -= xon_off_tot;
7220         hwstats->mptc -= xon_off_tot;
7221         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
7222         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
7223         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
7224         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
7225         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
7226         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
7227         hwstats->ptc64 -= xon_off_tot;
7228         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
7229         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
7230         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
7231         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
7232         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
7233         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
7234
7235         /* Fill out the OS statistics structure */
7236         netdev->stats.multicast = hwstats->mprc;
7237
7238         /* Rx Errors */
7239         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
7240         netdev->stats.rx_dropped = 0;
7241         netdev->stats.rx_length_errors = hwstats->rlec;
7242         netdev->stats.rx_crc_errors = hwstats->crcerrs;
7243         netdev->stats.rx_missed_errors = total_mpc;
7244 }
7245
7246 /**
7247  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
7248  * @adapter: pointer to the device adapter structure
7249  **/
7250 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
7251 {
7252         struct ixgbe_hw *hw = &adapter->hw;
7253         int i;
7254
7255         if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
7256                 return;
7257
7258         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
7259
7260         /* if interface is down do nothing */
7261         if (test_bit(__IXGBE_DOWN, &adapter->state))
7262                 return;
7263
7264         /* do nothing if we are not using signature filters */
7265         if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
7266                 return;
7267
7268         adapter->fdir_overflow++;
7269
7270         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
7271                 for (i = 0; i < adapter->num_tx_queues; i++)
7272                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7273                                 &(adapter->tx_ring[i]->state));
7274                 for (i = 0; i < adapter->num_xdp_queues; i++)
7275                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
7276                                 &adapter->xdp_ring[i]->state);
7277                 /* re-enable flow director interrupts */
7278                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
7279         } else {
7280                 e_err(probe, "failed to finish FDIR re-initialization, "
7281                       "ignored adding FDIR ATR filters\n");
7282         }
7283 }
7284
7285 /**
7286  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
7287  * @adapter: pointer to the device adapter structure
7288  *
7289  * This function serves two purposes.  First it strobes the interrupt lines
7290  * in order to make certain interrupts are occurring.  Secondly it sets the
7291  * bits needed to check for TX hangs.  As a result we should immediately
7292  * determine if a hang has occurred.
7293  */
7294 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
7295 {
7296         struct ixgbe_hw *hw = &adapter->hw;
7297         u64 eics = 0;
7298         int i;
7299
7300         /* If we're down, removing or resetting, just bail */
7301         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7302             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7303             test_bit(__IXGBE_RESETTING, &adapter->state))
7304                 return;
7305
7306         /* Force detection of hung controller */
7307         if (netif_carrier_ok(adapter->netdev)) {
7308                 for (i = 0; i < adapter->num_tx_queues; i++)
7309                         set_check_for_tx_hang(adapter->tx_ring[i]);
7310                 for (i = 0; i < adapter->num_xdp_queues; i++)
7311                         set_check_for_tx_hang(adapter->xdp_ring[i]);
7312         }
7313
7314         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
7315                 /*
7316                  * for legacy and MSI interrupts don't set any bits
7317                  * that are enabled for EIAM, because this operation
7318                  * would set *both* EIMS and EICS for any bit in EIAM
7319                  */
7320                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
7321                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
7322         } else {
7323                 /* get one bit for every active tx/rx interrupt vector */
7324                 for (i = 0; i < adapter->num_q_vectors; i++) {
7325                         struct ixgbe_q_vector *qv = adapter->q_vector[i];
7326                         if (qv->rx.ring || qv->tx.ring)
7327                                 eics |= BIT_ULL(i);
7328                 }
7329         }
7330
7331         /* Cause software interrupt to ensure rings are cleaned */
7332         ixgbe_irq_rearm_queues(adapter, eics);
7333 }
7334
7335 /**
7336  * ixgbe_watchdog_update_link - update the link status
7337  * @adapter: pointer to the device adapter structure
7338  **/
7339 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
7340 {
7341         struct ixgbe_hw *hw = &adapter->hw;
7342         u32 link_speed = adapter->link_speed;
7343         bool link_up = adapter->link_up;
7344         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
7345
7346         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
7347                 return;
7348
7349         if (hw->mac.ops.check_link) {
7350                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
7351         } else {
7352                 /* always assume link is up, if no check link function */
7353                 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
7354                 link_up = true;
7355         }
7356
7357         if (adapter->ixgbe_ieee_pfc)
7358                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
7359
7360         if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
7361                 hw->mac.ops.fc_enable(hw);
7362                 ixgbe_set_rx_drop_en(adapter);
7363         }
7364
7365         if (link_up ||
7366             time_after(jiffies, (adapter->link_check_timeout +
7367                                  IXGBE_TRY_LINK_TIMEOUT))) {
7368                 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
7369                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
7370                 IXGBE_WRITE_FLUSH(hw);
7371         }
7372
7373         adapter->link_up = link_up;
7374         adapter->link_speed = link_speed;
7375 }
7376
7377 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
7378 {
7379 #ifdef CONFIG_IXGBE_DCB
7380         struct net_device *netdev = adapter->netdev;
7381         struct dcb_app app = {
7382                               .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
7383                               .protocol = 0,
7384                              };
7385         u8 up = 0;
7386
7387         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
7388                 up = dcb_ieee_getapp_mask(netdev, &app);
7389
7390         adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
7391 #endif
7392 }
7393
7394 /**
7395  * ixgbe_watchdog_link_is_up - update netif_carrier status and
7396  *                             print link up message
7397  * @adapter: pointer to the device adapter structure
7398  **/
7399 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
7400 {
7401         struct net_device *netdev = adapter->netdev;
7402         struct ixgbe_hw *hw = &adapter->hw;
7403         u32 link_speed = adapter->link_speed;
7404         const char *speed_str;
7405         bool flow_rx, flow_tx;
7406
7407         /* only continue if link was previously down */
7408         if (netif_carrier_ok(netdev))
7409                 return;
7410
7411         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
7412
7413         switch (hw->mac.type) {
7414         case ixgbe_mac_82598EB: {
7415                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
7416                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
7417                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
7418                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
7419         }
7420                 break;
7421         case ixgbe_mac_X540:
7422         case ixgbe_mac_X550:
7423         case ixgbe_mac_X550EM_x:
7424         case ixgbe_mac_x550em_a:
7425         case ixgbe_mac_82599EB: {
7426                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
7427                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
7428                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
7429                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
7430         }
7431                 break;
7432         default:
7433                 flow_tx = false;
7434                 flow_rx = false;
7435                 break;
7436         }
7437
7438         adapter->last_rx_ptp_check = jiffies;
7439
7440         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7441                 ixgbe_ptp_start_cyclecounter(adapter);
7442
7443         switch (link_speed) {
7444         case IXGBE_LINK_SPEED_10GB_FULL:
7445                 speed_str = "10 Gbps";
7446                 break;
7447         case IXGBE_LINK_SPEED_5GB_FULL:
7448                 speed_str = "5 Gbps";
7449                 break;
7450         case IXGBE_LINK_SPEED_2_5GB_FULL:
7451                 speed_str = "2.5 Gbps";
7452                 break;
7453         case IXGBE_LINK_SPEED_1GB_FULL:
7454                 speed_str = "1 Gbps";
7455                 break;
7456         case IXGBE_LINK_SPEED_100_FULL:
7457                 speed_str = "100 Mbps";
7458                 break;
7459         case IXGBE_LINK_SPEED_10_FULL:
7460                 speed_str = "10 Mbps";
7461                 break;
7462         default:
7463                 speed_str = "unknown speed";
7464                 break;
7465         }
7466         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
7467                ((flow_rx && flow_tx) ? "RX/TX" :
7468                (flow_rx ? "RX" :
7469                (flow_tx ? "TX" : "None"))));
7470
7471         netif_carrier_on(netdev);
7472         ixgbe_check_vf_rate_limit(adapter);
7473
7474         /* enable transmits */
7475         netif_tx_wake_all_queues(adapter->netdev);
7476
7477         /* update the default user priority for VFs */
7478         ixgbe_update_default_up(adapter);
7479
7480         /* ping all the active vfs to let them know link has changed */
7481         ixgbe_ping_all_vfs(adapter);
7482 }
7483
7484 /**
7485  * ixgbe_watchdog_link_is_down - update netif_carrier status and
7486  *                               print link down message
7487  * @adapter: pointer to the adapter structure
7488  **/
7489 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
7490 {
7491         struct net_device *netdev = adapter->netdev;
7492         struct ixgbe_hw *hw = &adapter->hw;
7493
7494         adapter->link_up = false;
7495         adapter->link_speed = 0;
7496
7497         /* only continue if link was up previously */
7498         if (!netif_carrier_ok(netdev))
7499                 return;
7500
7501         /* poll for SFP+ cable when link is down */
7502         if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
7503                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
7504
7505         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
7506                 ixgbe_ptp_start_cyclecounter(adapter);
7507
7508         e_info(drv, "NIC Link is Down\n");
7509         netif_carrier_off(netdev);
7510
7511         /* ping all the active vfs to let them know link has changed */
7512         ixgbe_ping_all_vfs(adapter);
7513 }
7514
7515 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
7516 {
7517         int i;
7518
7519         for (i = 0; i < adapter->num_tx_queues; i++) {
7520                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
7521
7522                 if (tx_ring->next_to_use != tx_ring->next_to_clean)
7523                         return true;
7524         }
7525
7526         for (i = 0; i < adapter->num_xdp_queues; i++) {
7527                 struct ixgbe_ring *ring = adapter->xdp_ring[i];
7528
7529                 if (ring->next_to_use != ring->next_to_clean)
7530                         return true;
7531         }
7532
7533         return false;
7534 }
7535
7536 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
7537 {
7538         struct ixgbe_hw *hw = &adapter->hw;
7539         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
7540         u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
7541
7542         int i, j;
7543
7544         if (!adapter->num_vfs)
7545                 return false;
7546
7547         /* resetting the PF is only needed for MAC before X550 */
7548         if (hw->mac.type >= ixgbe_mac_X550)
7549                 return false;
7550
7551         for (i = 0; i < adapter->num_vfs; i++) {
7552                 for (j = 0; j < q_per_pool; j++) {
7553                         u32 h, t;
7554
7555                         h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
7556                         t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
7557
7558                         if (h != t)
7559                                 return true;
7560                 }
7561         }
7562
7563         return false;
7564 }
7565
7566 /**
7567  * ixgbe_watchdog_flush_tx - flush queues on link down
7568  * @adapter: pointer to the device adapter structure
7569  **/
7570 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
7571 {
7572         if (!netif_carrier_ok(adapter->netdev)) {
7573                 if (ixgbe_ring_tx_pending(adapter) ||
7574                     ixgbe_vf_tx_pending(adapter)) {
7575                         /* We've lost link, so the controller stops DMA,
7576                          * but we've got queued Tx work that's never going
7577                          * to get done, so reset controller to flush Tx.
7578                          * (Do the reset outside of interrupt context).
7579                          */
7580                         e_warn(drv, "initiating reset to clear Tx work after link loss\n");
7581                         set_bit(__IXGBE_RESET_REQUESTED, &adapter->state);
7582                 }
7583         }
7584 }
7585
7586 #ifdef CONFIG_PCI_IOV
7587 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
7588 {
7589         struct ixgbe_hw *hw = &adapter->hw;
7590         struct pci_dev *pdev = adapter->pdev;
7591         unsigned int vf;
7592         u32 gpc;
7593
7594         if (!(netif_carrier_ok(adapter->netdev)))
7595                 return;
7596
7597         gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
7598         if (gpc) /* If incrementing then no need for the check below */
7599                 return;
7600         /* Check to see if a bad DMA write target from an errant or
7601          * malicious VF has caused a PCIe error.  If so then we can
7602          * issue a VFLR to the offending VF(s) and then resume without
7603          * requesting a full slot reset.
7604          */
7605
7606         if (!pdev)
7607                 return;
7608
7609         /* check status reg for all VFs owned by this PF */
7610         for (vf = 0; vf < adapter->num_vfs; ++vf) {
7611                 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
7612                 u16 status_reg;
7613
7614                 if (!vfdev)
7615                         continue;
7616                 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
7617                 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
7618                     status_reg & PCI_STATUS_REC_MASTER_ABORT)
7619                         pcie_flr(vfdev);
7620         }
7621 }
7622
7623 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
7624 {
7625         u32 ssvpc;
7626
7627         /* Do not perform spoof check for 82598 or if not in IOV mode */
7628         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
7629             adapter->num_vfs == 0)
7630                 return;
7631
7632         ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
7633
7634         /*
7635          * ssvpc register is cleared on read, if zero then no
7636          * spoofed packets in the last interval.
7637          */
7638         if (!ssvpc)
7639                 return;
7640
7641         e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
7642 }
7643 #else
7644 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
7645 {
7646 }
7647
7648 static void
7649 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
7650 {
7651 }
7652 #endif /* CONFIG_PCI_IOV */
7653
7654
7655 /**
7656  * ixgbe_watchdog_subtask - check and bring link up
7657  * @adapter: pointer to the device adapter structure
7658  **/
7659 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
7660 {
7661         /* if interface is down, removing or resetting, do nothing */
7662         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7663             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7664             test_bit(__IXGBE_RESETTING, &adapter->state))
7665                 return;
7666
7667         ixgbe_watchdog_update_link(adapter);
7668
7669         if (adapter->link_up)
7670                 ixgbe_watchdog_link_is_up(adapter);
7671         else
7672                 ixgbe_watchdog_link_is_down(adapter);
7673
7674         ixgbe_check_for_bad_vf(adapter);
7675         ixgbe_spoof_check(adapter);
7676         ixgbe_update_stats(adapter);
7677
7678         ixgbe_watchdog_flush_tx(adapter);
7679 }
7680
7681 /**
7682  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
7683  * @adapter: the ixgbe adapter structure
7684  **/
7685 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
7686 {
7687         struct ixgbe_hw *hw = &adapter->hw;
7688         s32 err;
7689
7690         /* not searching for SFP so there is nothing to do here */
7691         if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
7692             !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7693                 return;
7694
7695         if (adapter->sfp_poll_time &&
7696             time_after(adapter->sfp_poll_time, jiffies))
7697                 return; /* If not yet time to poll for SFP */
7698
7699         /* someone else is in init, wait until next service event */
7700         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7701                 return;
7702
7703         adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
7704
7705         err = hw->phy.ops.identify_sfp(hw);
7706         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7707                 goto sfp_out;
7708
7709         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
7710                 /* If no cable is present, then we need to reset
7711                  * the next time we find a good cable. */
7712                 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
7713         }
7714
7715         /* exit on error */
7716         if (err)
7717                 goto sfp_out;
7718
7719         /* exit if reset not needed */
7720         if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
7721                 goto sfp_out;
7722
7723         adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
7724
7725         /*
7726          * A module may be identified correctly, but the EEPROM may not have
7727          * support for that module.  setup_sfp() will fail in that case, so
7728          * we should not allow that module to load.
7729          */
7730         if (hw->mac.type == ixgbe_mac_82598EB)
7731                 err = hw->phy.ops.reset(hw);
7732         else
7733                 err = hw->mac.ops.setup_sfp(hw);
7734
7735         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
7736                 goto sfp_out;
7737
7738         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
7739         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
7740
7741 sfp_out:
7742         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7743
7744         if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
7745             (adapter->netdev->reg_state == NETREG_REGISTERED)) {
7746                 e_dev_err("failed to initialize because an unsupported "
7747                           "SFP+ module type was detected.\n");
7748                 e_dev_err("Reload the driver after installing a "
7749                           "supported module.\n");
7750                 unregister_netdev(adapter->netdev);
7751         }
7752 }
7753
7754 /**
7755  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7756  * @adapter: the ixgbe adapter structure
7757  **/
7758 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7759 {
7760         struct ixgbe_hw *hw = &adapter->hw;
7761         u32 cap_speed;
7762         u32 speed;
7763         bool autoneg = false;
7764
7765         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7766                 return;
7767
7768         /* someone else is in init, wait until next service event */
7769         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7770                 return;
7771
7772         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7773
7774         hw->mac.ops.get_link_capabilities(hw, &cap_speed, &autoneg);
7775
7776         /* advertise highest capable link speed */
7777         if (!autoneg && (cap_speed & IXGBE_LINK_SPEED_10GB_FULL))
7778                 speed = IXGBE_LINK_SPEED_10GB_FULL;
7779         else
7780                 speed = cap_speed & (IXGBE_LINK_SPEED_10GB_FULL |
7781                                      IXGBE_LINK_SPEED_1GB_FULL);
7782
7783         if (hw->mac.ops.setup_link)
7784                 hw->mac.ops.setup_link(hw, speed, true);
7785
7786         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7787         adapter->link_check_timeout = jiffies;
7788         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7789 }
7790
7791 /**
7792  * ixgbe_service_timer - Timer Call-back
7793  * @t: pointer to timer_list structure
7794  **/
7795 static void ixgbe_service_timer(struct timer_list *t)
7796 {
7797         struct ixgbe_adapter *adapter = from_timer(adapter, t, service_timer);
7798         unsigned long next_event_offset;
7799
7800         /* poll faster when waiting for link */
7801         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7802                 next_event_offset = HZ / 10;
7803         else
7804                 next_event_offset = HZ * 2;
7805
7806         /* Reset the timer */
7807         mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7808
7809         ixgbe_service_event_schedule(adapter);
7810 }
7811
7812 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7813 {
7814         struct ixgbe_hw *hw = &adapter->hw;
7815         u32 status;
7816
7817         if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7818                 return;
7819
7820         adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7821
7822         if (!hw->phy.ops.handle_lasi)
7823                 return;
7824
7825         status = hw->phy.ops.handle_lasi(&adapter->hw);
7826         if (status != IXGBE_ERR_OVERTEMP)
7827                 return;
7828
7829         e_crit(drv, "%s\n", ixgbe_overheat_msg);
7830 }
7831
7832 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7833 {
7834         if (!test_and_clear_bit(__IXGBE_RESET_REQUESTED, &adapter->state))
7835                 return;
7836
7837         rtnl_lock();
7838         /* If we're already down, removing or resetting, just bail */
7839         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7840             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7841             test_bit(__IXGBE_RESETTING, &adapter->state)) {
7842                 rtnl_unlock();
7843                 return;
7844         }
7845
7846         ixgbe_dump(adapter);
7847         netdev_err(adapter->netdev, "Reset adapter\n");
7848         adapter->tx_timeout_count++;
7849
7850         ixgbe_reinit_locked(adapter);
7851         rtnl_unlock();
7852 }
7853
7854 /**
7855  * ixgbe_check_fw_error - Check firmware for errors
7856  * @adapter: the adapter private structure
7857  *
7858  * Check firmware errors in register FWSM
7859  */
7860 static bool ixgbe_check_fw_error(struct ixgbe_adapter *adapter)
7861 {
7862         struct ixgbe_hw *hw = &adapter->hw;
7863         u32 fwsm;
7864
7865         /* read fwsm.ext_err_ind register and log errors */
7866         fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
7867
7868         if (fwsm & IXGBE_FWSM_EXT_ERR_IND_MASK ||
7869             !(fwsm & IXGBE_FWSM_FW_VAL_BIT))
7870                 e_dev_warn("Warning firmware error detected FWSM: 0x%08X\n",
7871                            fwsm);
7872
7873         if (hw->mac.ops.fw_recovery_mode && hw->mac.ops.fw_recovery_mode(hw)) {
7874                 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");
7875                 return true;
7876         }
7877
7878         return false;
7879 }
7880
7881 /**
7882  * ixgbe_service_task - manages and runs subtasks
7883  * @work: pointer to work_struct containing our data
7884  **/
7885 static void ixgbe_service_task(struct work_struct *work)
7886 {
7887         struct ixgbe_adapter *adapter = container_of(work,
7888                                                      struct ixgbe_adapter,
7889                                                      service_task);
7890         if (ixgbe_removed(adapter->hw.hw_addr)) {
7891                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7892                         rtnl_lock();
7893                         ixgbe_down(adapter);
7894                         rtnl_unlock();
7895                 }
7896                 ixgbe_service_event_complete(adapter);
7897                 return;
7898         }
7899         if (ixgbe_check_fw_error(adapter)) {
7900                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
7901                         unregister_netdev(adapter->netdev);
7902                 ixgbe_service_event_complete(adapter);
7903                 return;
7904         }
7905         if (adapter->flags2 & IXGBE_FLAG2_UDP_TUN_REREG_NEEDED) {
7906                 rtnl_lock();
7907                 adapter->flags2 &= ~IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
7908                 udp_tunnel_get_rx_info(adapter->netdev);
7909                 rtnl_unlock();
7910         }
7911         ixgbe_reset_subtask(adapter);
7912         ixgbe_phy_interrupt_subtask(adapter);
7913         ixgbe_sfp_detection_subtask(adapter);
7914         ixgbe_sfp_link_config_subtask(adapter);
7915         ixgbe_check_overtemp_subtask(adapter);
7916         ixgbe_watchdog_subtask(adapter);
7917         ixgbe_fdir_reinit_subtask(adapter);
7918         ixgbe_check_hang_subtask(adapter);
7919
7920         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7921                 ixgbe_ptp_overflow_check(adapter);
7922                 if (adapter->flags & IXGBE_FLAG_RX_HWTSTAMP_IN_REGISTER)
7923                         ixgbe_ptp_rx_hang(adapter);
7924                 ixgbe_ptp_tx_hang(adapter);
7925         }
7926
7927         ixgbe_service_event_complete(adapter);
7928 }
7929
7930 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7931                      struct ixgbe_tx_buffer *first,
7932                      u8 *hdr_len,
7933                      struct ixgbe_ipsec_tx_data *itd)
7934 {
7935         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
7936         struct sk_buff *skb = first->skb;
7937         union {
7938                 struct iphdr *v4;
7939                 struct ipv6hdr *v6;
7940                 unsigned char *hdr;
7941         } ip;
7942         union {
7943                 struct tcphdr *tcp;
7944                 unsigned char *hdr;
7945         } l4;
7946         u32 paylen, l4_offset;
7947         u32 fceof_saidx = 0;
7948         int err;
7949
7950         if (skb->ip_summed != CHECKSUM_PARTIAL)
7951                 return 0;
7952
7953         if (!skb_is_gso(skb))
7954                 return 0;
7955
7956         err = skb_cow_head(skb, 0);
7957         if (err < 0)
7958                 return err;
7959
7960         if (eth_p_mpls(first->protocol))
7961                 ip.hdr = skb_inner_network_header(skb);
7962         else
7963                 ip.hdr = skb_network_header(skb);
7964         l4.hdr = skb_checksum_start(skb);
7965
7966         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7967         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7968
7969         /* initialize outer IP header fields */
7970         if (ip.v4->version == 4) {
7971                 unsigned char *csum_start = skb_checksum_start(skb);
7972                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
7973                 int len = csum_start - trans_start;
7974
7975                 /* IP header will have to cancel out any data that
7976                  * is not a part of the outer IP header, so set to
7977                  * a reverse csum if needed, else init check to 0.
7978                  */
7979                 ip.v4->check = (skb_shinfo(skb)->gso_type & SKB_GSO_PARTIAL) ?
7980                                            csum_fold(csum_partial(trans_start,
7981                                                                   len, 0)) : 0;
7982                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7983
7984                 ip.v4->tot_len = 0;
7985                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7986                                    IXGBE_TX_FLAGS_CSUM |
7987                                    IXGBE_TX_FLAGS_IPV4;
7988         } else {
7989                 ip.v6->payload_len = 0;
7990                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7991                                    IXGBE_TX_FLAGS_CSUM;
7992         }
7993
7994         /* determine offset of inner transport header */
7995         l4_offset = l4.hdr - skb->data;
7996
7997         /* compute length of segmentation header */
7998         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
7999
8000         /* remove payload length from inner checksum */
8001         paylen = skb->len - l4_offset;
8002         csum_replace_by_diff(&l4.tcp->check, (__force __wsum)htonl(paylen));
8003
8004         /* update gso size and bytecount with header size */
8005         first->gso_segs = skb_shinfo(skb)->gso_segs;
8006         first->bytecount += (first->gso_segs - 1) * *hdr_len;
8007
8008         /* mss_l4len_id: use 0 as index for TSO */
8009         mss_l4len_idx = (*hdr_len - l4_offset) << IXGBE_ADVTXD_L4LEN_SHIFT;
8010         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
8011
8012         fceof_saidx |= itd->sa_idx;
8013         type_tucmd |= itd->flags | itd->trailer_len;
8014
8015         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
8016         vlan_macip_lens = l4.hdr - ip.hdr;
8017         vlan_macip_lens |= (ip.hdr - skb->data) << IXGBE_ADVTXD_MACLEN_SHIFT;
8018         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8019
8020         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd,
8021                           mss_l4len_idx);
8022
8023         return 1;
8024 }
8025
8026 static inline bool ixgbe_ipv6_csum_is_sctp(struct sk_buff *skb)
8027 {
8028         unsigned int offset = 0;
8029
8030         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
8031
8032         return offset == skb_checksum_start_offset(skb);
8033 }
8034
8035 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
8036                           struct ixgbe_tx_buffer *first,
8037                           struct ixgbe_ipsec_tx_data *itd)
8038 {
8039         struct sk_buff *skb = first->skb;
8040         u32 vlan_macip_lens = 0;
8041         u32 fceof_saidx = 0;
8042         u32 type_tucmd = 0;
8043
8044         if (skb->ip_summed != CHECKSUM_PARTIAL) {
8045 csum_failed:
8046                 if (!(first->tx_flags & (IXGBE_TX_FLAGS_HW_VLAN |
8047                                          IXGBE_TX_FLAGS_CC)))
8048                         return;
8049                 goto no_csum;
8050         }
8051
8052         switch (skb->csum_offset) {
8053         case offsetof(struct tcphdr, check):
8054                 type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
8055                 /* fall through */
8056         case offsetof(struct udphdr, check):
8057                 break;
8058         case offsetof(struct sctphdr, checksum):
8059                 /* validate that this is actually an SCTP request */
8060                 if (((first->protocol == htons(ETH_P_IP)) &&
8061                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
8062                     ((first->protocol == htons(ETH_P_IPV6)) &&
8063                      ixgbe_ipv6_csum_is_sctp(skb))) {
8064                         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_SCTP;
8065                         break;
8066                 }
8067                 /* fall through */
8068         default:
8069                 skb_checksum_help(skb);
8070                 goto csum_failed;
8071         }
8072
8073         /* update TX checksum flag */
8074         first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
8075         vlan_macip_lens = skb_checksum_start_offset(skb) -
8076                           skb_network_offset(skb);
8077 no_csum:
8078         /* vlan_macip_lens: MACLEN, VLAN tag */
8079         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
8080         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
8081
8082         fceof_saidx |= itd->sa_idx;
8083         type_tucmd |= itd->flags | itd->trailer_len;
8084
8085         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, fceof_saidx, type_tucmd, 0);
8086 }
8087
8088 #define IXGBE_SET_FLAG(_input, _flag, _result) \
8089         ((_flag <= _result) ? \
8090          ((u32)(_input & _flag) * (_result / _flag)) : \
8091          ((u32)(_input & _flag) / (_flag / _result)))
8092
8093 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
8094 {
8095         /* set type for advanced descriptor with frame checksum insertion */
8096         u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8097                        IXGBE_ADVTXD_DCMD_DEXT |
8098                        IXGBE_ADVTXD_DCMD_IFCS;
8099
8100         /* set HW vlan bit if vlan is present */
8101         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
8102                                    IXGBE_ADVTXD_DCMD_VLE);
8103
8104         /* set segmentation enable bits for TSO/FSO */
8105         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
8106                                    IXGBE_ADVTXD_DCMD_TSE);
8107
8108         /* set timestamp bit if present */
8109         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
8110                                    IXGBE_ADVTXD_MAC_TSTAMP);
8111
8112         /* insert frame checksum */
8113         cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
8114
8115         return cmd_type;
8116 }
8117
8118 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
8119                                    u32 tx_flags, unsigned int paylen)
8120 {
8121         u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
8122
8123         /* enable L4 checksum for TSO and TX checksum offload */
8124         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8125                                         IXGBE_TX_FLAGS_CSUM,
8126                                         IXGBE_ADVTXD_POPTS_TXSM);
8127
8128         /* enable IPv4 checksum for TSO */
8129         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8130                                         IXGBE_TX_FLAGS_IPV4,
8131                                         IXGBE_ADVTXD_POPTS_IXSM);
8132
8133         /* enable IPsec */
8134         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8135                                         IXGBE_TX_FLAGS_IPSEC,
8136                                         IXGBE_ADVTXD_POPTS_IPSEC);
8137
8138         /*
8139          * Check Context must be set if Tx switch is enabled, which it
8140          * always is for case where virtual functions are running
8141          */
8142         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
8143                                         IXGBE_TX_FLAGS_CC,
8144                                         IXGBE_ADVTXD_CC);
8145
8146         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
8147 }
8148
8149 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8150 {
8151         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
8152
8153         /* Herbert's original patch had:
8154          *  smp_mb__after_netif_stop_queue();
8155          * but since that doesn't exist yet, just open code it.
8156          */
8157         smp_mb();
8158
8159         /* We need to check again in a case another CPU has just
8160          * made room available.
8161          */
8162         if (likely(ixgbe_desc_unused(tx_ring) < size))
8163                 return -EBUSY;
8164
8165         /* A reprieve! - use start_queue because it doesn't call schedule */
8166         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
8167         ++tx_ring->tx_stats.restart_queue;
8168         return 0;
8169 }
8170
8171 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
8172 {
8173         if (likely(ixgbe_desc_unused(tx_ring) >= size))
8174                 return 0;
8175
8176         return __ixgbe_maybe_stop_tx(tx_ring, size);
8177 }
8178
8179 static int ixgbe_tx_map(struct ixgbe_ring *tx_ring,
8180                         struct ixgbe_tx_buffer *first,
8181                         const u8 hdr_len)
8182 {
8183         struct sk_buff *skb = first->skb;
8184         struct ixgbe_tx_buffer *tx_buffer;
8185         union ixgbe_adv_tx_desc *tx_desc;
8186         struct skb_frag_struct *frag;
8187         dma_addr_t dma;
8188         unsigned int data_len, size;
8189         u32 tx_flags = first->tx_flags;
8190         u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
8191         u16 i = tx_ring->next_to_use;
8192
8193         tx_desc = IXGBE_TX_DESC(tx_ring, i);
8194
8195         ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
8196
8197         size = skb_headlen(skb);
8198         data_len = skb->data_len;
8199
8200 #ifdef IXGBE_FCOE
8201         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
8202                 if (data_len < sizeof(struct fcoe_crc_eof)) {
8203                         size -= sizeof(struct fcoe_crc_eof) - data_len;
8204                         data_len = 0;
8205                 } else {
8206                         data_len -= sizeof(struct fcoe_crc_eof);
8207                 }
8208         }
8209
8210 #endif
8211         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
8212
8213         tx_buffer = first;
8214
8215         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
8216                 if (dma_mapping_error(tx_ring->dev, dma))
8217                         goto dma_error;
8218
8219                 /* record length, and DMA address */
8220                 dma_unmap_len_set(tx_buffer, len, size);
8221                 dma_unmap_addr_set(tx_buffer, dma, dma);
8222
8223                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
8224
8225                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
8226                         tx_desc->read.cmd_type_len =
8227                                 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
8228
8229                         i++;
8230                         tx_desc++;
8231                         if (i == tx_ring->count) {
8232                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8233                                 i = 0;
8234                         }
8235                         tx_desc->read.olinfo_status = 0;
8236
8237                         dma += IXGBE_MAX_DATA_PER_TXD;
8238                         size -= IXGBE_MAX_DATA_PER_TXD;
8239
8240                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
8241                 }
8242
8243                 if (likely(!data_len))
8244                         break;
8245
8246                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
8247
8248                 i++;
8249                 tx_desc++;
8250                 if (i == tx_ring->count) {
8251                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
8252                         i = 0;
8253                 }
8254                 tx_desc->read.olinfo_status = 0;
8255
8256 #ifdef IXGBE_FCOE
8257                 size = min_t(unsigned int, data_len, skb_frag_size(frag));
8258 #else
8259                 size = skb_frag_size(frag);
8260 #endif
8261                 data_len -= size;
8262
8263                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
8264                                        DMA_TO_DEVICE);
8265
8266                 tx_buffer = &tx_ring->tx_buffer_info[i];
8267         }
8268
8269         /* write last descriptor with RS and EOP bits */
8270         cmd_type |= size | IXGBE_TXD_CMD;
8271         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8272
8273         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
8274
8275         /* set the timestamp */
8276         first->time_stamp = jiffies;
8277
8278         skb_tx_timestamp(skb);
8279
8280         /*
8281          * Force memory writes to complete before letting h/w know there
8282          * are new descriptors to fetch.  (Only applicable for weak-ordered
8283          * memory model archs, such as IA-64).
8284          *
8285          * We also need this memory barrier to make certain all of the
8286          * status bits have been updated before next_to_watch is written.
8287          */
8288         wmb();
8289
8290         /* set next_to_watch value indicating a packet is present */
8291         first->next_to_watch = tx_desc;
8292
8293         i++;
8294         if (i == tx_ring->count)
8295                 i = 0;
8296
8297         tx_ring->next_to_use = i;
8298
8299         ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
8300
8301         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
8302                 writel(i, tx_ring->tail);
8303         }
8304
8305         return 0;
8306 dma_error:
8307         dev_err(tx_ring->dev, "TX DMA map failed\n");
8308
8309         /* clear dma mappings for failed tx_buffer_info map */
8310         for (;;) {
8311                 tx_buffer = &tx_ring->tx_buffer_info[i];
8312                 if (dma_unmap_len(tx_buffer, len))
8313                         dma_unmap_page(tx_ring->dev,
8314                                        dma_unmap_addr(tx_buffer, dma),
8315                                        dma_unmap_len(tx_buffer, len),
8316                                        DMA_TO_DEVICE);
8317                 dma_unmap_len_set(tx_buffer, len, 0);
8318                 if (tx_buffer == first)
8319                         break;
8320                 if (i == 0)
8321                         i += tx_ring->count;
8322                 i--;
8323         }
8324
8325         dev_kfree_skb_any(first->skb);
8326         first->skb = NULL;
8327
8328         tx_ring->next_to_use = i;
8329
8330         return -1;
8331 }
8332
8333 static void ixgbe_atr(struct ixgbe_ring *ring,
8334                       struct ixgbe_tx_buffer *first)
8335 {
8336         struct ixgbe_q_vector *q_vector = ring->q_vector;
8337         union ixgbe_atr_hash_dword input = { .dword = 0 };
8338         union ixgbe_atr_hash_dword common = { .dword = 0 };
8339         union {
8340                 unsigned char *network;
8341                 struct iphdr *ipv4;
8342                 struct ipv6hdr *ipv6;
8343         } hdr;
8344         struct tcphdr *th;
8345         unsigned int hlen;
8346         struct sk_buff *skb;
8347         __be16 vlan_id;
8348         int l4_proto;
8349
8350         /* if ring doesn't have a interrupt vector, cannot perform ATR */
8351         if (!q_vector)
8352                 return;
8353
8354         /* do nothing if sampling is disabled */
8355         if (!ring->atr_sample_rate)
8356                 return;
8357
8358         ring->atr_count++;
8359
8360         /* currently only IPv4/IPv6 with TCP is supported */
8361         if ((first->protocol != htons(ETH_P_IP)) &&
8362             (first->protocol != htons(ETH_P_IPV6)))
8363                 return;
8364
8365         /* snag network header to get L4 type and address */
8366         skb = first->skb;
8367         hdr.network = skb_network_header(skb);
8368         if (unlikely(hdr.network <= skb->data))
8369                 return;
8370         if (skb->encapsulation &&
8371             first->protocol == htons(ETH_P_IP) &&
8372             hdr.ipv4->protocol == IPPROTO_UDP) {
8373                 struct ixgbe_adapter *adapter = q_vector->adapter;
8374
8375                 if (unlikely(skb_tail_pointer(skb) < hdr.network +
8376                              VXLAN_HEADROOM))
8377                         return;
8378
8379                 /* verify the port is recognized as VXLAN */
8380                 if (adapter->vxlan_port &&
8381                     udp_hdr(skb)->dest == adapter->vxlan_port)
8382                         hdr.network = skb_inner_network_header(skb);
8383
8384                 if (adapter->geneve_port &&
8385                     udp_hdr(skb)->dest == adapter->geneve_port)
8386                         hdr.network = skb_inner_network_header(skb);
8387         }
8388
8389         /* Make sure we have at least [minimum IPv4 header + TCP]
8390          * or [IPv6 header] bytes
8391          */
8392         if (unlikely(skb_tail_pointer(skb) < hdr.network + 40))
8393                 return;
8394
8395         /* Currently only IPv4/IPv6 with TCP is supported */
8396         switch (hdr.ipv4->version) {
8397         case IPVERSION:
8398                 /* access ihl as u8 to avoid unaligned access on ia64 */
8399                 hlen = (hdr.network[0] & 0x0F) << 2;
8400                 l4_proto = hdr.ipv4->protocol;
8401                 break;
8402         case 6:
8403                 hlen = hdr.network - skb->data;
8404                 l4_proto = ipv6_find_hdr(skb, &hlen, IPPROTO_TCP, NULL, NULL);
8405                 hlen -= hdr.network - skb->data;
8406                 break;
8407         default:
8408                 return;
8409         }
8410
8411         if (l4_proto != IPPROTO_TCP)
8412                 return;
8413
8414         if (unlikely(skb_tail_pointer(skb) < hdr.network +
8415                      hlen + sizeof(struct tcphdr)))
8416                 return;
8417
8418         th = (struct tcphdr *)(hdr.network + hlen);
8419
8420         /* skip this packet since the socket is closing */
8421         if (th->fin)
8422                 return;
8423
8424         /* sample on all syn packets or once every atr sample count */
8425         if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
8426                 return;
8427
8428         /* reset sample count */
8429         ring->atr_count = 0;
8430
8431         vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
8432
8433         /*
8434          * src and dst are inverted, think how the receiver sees them
8435          *
8436          * The input is broken into two sections, a non-compressed section
8437          * containing vm_pool, vlan_id, and flow_type.  The rest of the data
8438          * is XORed together and stored in the compressed dword.
8439          */
8440         input.formatted.vlan_id = vlan_id;
8441
8442         /*
8443          * since src port and flex bytes occupy the same word XOR them together
8444          * and write the value to source port portion of compressed dword
8445          */
8446         if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
8447                 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
8448         else
8449                 common.port.src ^= th->dest ^ first->protocol;
8450         common.port.dst ^= th->source;
8451
8452         switch (hdr.ipv4->version) {
8453         case IPVERSION:
8454                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
8455                 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
8456                 break;
8457         case 6:
8458                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
8459                 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
8460                              hdr.ipv6->saddr.s6_addr32[1] ^
8461                              hdr.ipv6->saddr.s6_addr32[2] ^
8462                              hdr.ipv6->saddr.s6_addr32[3] ^
8463                              hdr.ipv6->daddr.s6_addr32[0] ^
8464                              hdr.ipv6->daddr.s6_addr32[1] ^
8465                              hdr.ipv6->daddr.s6_addr32[2] ^
8466                              hdr.ipv6->daddr.s6_addr32[3];
8467                 break;
8468         default:
8469                 break;
8470         }
8471
8472         if (hdr.network != skb_network_header(skb))
8473                 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
8474
8475         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
8476         ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
8477                                               input, common, ring->queue_index);
8478 }
8479
8480 #ifdef IXGBE_FCOE
8481 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
8482                               struct net_device *sb_dev)
8483 {
8484         struct ixgbe_adapter *adapter;
8485         struct ixgbe_ring_feature *f;
8486         int txq;
8487
8488         if (sb_dev) {
8489                 u8 tc = netdev_get_prio_tc_map(dev, skb->priority);
8490                 struct net_device *vdev = sb_dev;
8491
8492                 txq = vdev->tc_to_txq[tc].offset;
8493                 txq += reciprocal_scale(skb_get_hash(skb),
8494                                         vdev->tc_to_txq[tc].count);
8495
8496                 return txq;
8497         }
8498
8499         /*
8500          * only execute the code below if protocol is FCoE
8501          * or FIP and we have FCoE enabled on the adapter
8502          */
8503         switch (vlan_get_protocol(skb)) {
8504         case htons(ETH_P_FCOE):
8505         case htons(ETH_P_FIP):
8506                 adapter = netdev_priv(dev);
8507
8508                 if (!sb_dev && (adapter->flags & IXGBE_FLAG_FCOE_ENABLED))
8509                         break;
8510                 /* fall through */
8511         default:
8512                 return netdev_pick_tx(dev, skb, sb_dev);
8513         }
8514
8515         f = &adapter->ring_feature[RING_F_FCOE];
8516
8517         txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
8518                                            smp_processor_id();
8519
8520         while (txq >= f->indices)
8521                 txq -= f->indices;
8522
8523         return txq + f->offset;
8524 }
8525
8526 #endif
8527 int ixgbe_xmit_xdp_ring(struct ixgbe_adapter *adapter,
8528                         struct xdp_frame *xdpf)
8529 {
8530         struct ixgbe_ring *ring = adapter->xdp_ring[smp_processor_id()];
8531         struct ixgbe_tx_buffer *tx_buffer;
8532         union ixgbe_adv_tx_desc *tx_desc;
8533         u32 len, cmd_type;
8534         dma_addr_t dma;
8535         u16 i;
8536
8537         len = xdpf->len;
8538
8539         if (unlikely(!ixgbe_desc_unused(ring)))
8540                 return IXGBE_XDP_CONSUMED;
8541
8542         dma = dma_map_single(ring->dev, xdpf->data, len, DMA_TO_DEVICE);
8543         if (dma_mapping_error(ring->dev, dma))
8544                 return IXGBE_XDP_CONSUMED;
8545
8546         /* record the location of the first descriptor for this packet */
8547         tx_buffer = &ring->tx_buffer_info[ring->next_to_use];
8548         tx_buffer->bytecount = len;
8549         tx_buffer->gso_segs = 1;
8550         tx_buffer->protocol = 0;
8551
8552         i = ring->next_to_use;
8553         tx_desc = IXGBE_TX_DESC(ring, i);
8554
8555         dma_unmap_len_set(tx_buffer, len, len);
8556         dma_unmap_addr_set(tx_buffer, dma, dma);
8557         tx_buffer->xdpf = xdpf;
8558
8559         tx_desc->read.buffer_addr = cpu_to_le64(dma);
8560
8561         /* put descriptor type bits */
8562         cmd_type = IXGBE_ADVTXD_DTYP_DATA |
8563                    IXGBE_ADVTXD_DCMD_DEXT |
8564                    IXGBE_ADVTXD_DCMD_IFCS;
8565         cmd_type |= len | IXGBE_TXD_CMD;
8566         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
8567         tx_desc->read.olinfo_status =
8568                 cpu_to_le32(len << IXGBE_ADVTXD_PAYLEN_SHIFT);
8569
8570         /* Avoid any potential race with xdp_xmit and cleanup */
8571         smp_wmb();
8572
8573         /* set next_to_watch value indicating a packet is present */
8574         i++;
8575         if (i == ring->count)
8576                 i = 0;
8577
8578         tx_buffer->next_to_watch = tx_desc;
8579         ring->next_to_use = i;
8580
8581         return IXGBE_XDP_TX;
8582 }
8583
8584 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
8585                           struct ixgbe_adapter *adapter,
8586                           struct ixgbe_ring *tx_ring)
8587 {
8588         struct ixgbe_tx_buffer *first;
8589         int tso;
8590         u32 tx_flags = 0;
8591         unsigned short f;
8592         u16 count = TXD_USE_COUNT(skb_headlen(skb));
8593         struct ixgbe_ipsec_tx_data ipsec_tx = { 0 };
8594         __be16 protocol = skb->protocol;
8595         u8 hdr_len = 0;
8596
8597         /*
8598          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
8599          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
8600          *       + 2 desc gap to keep tail from touching head,
8601          *       + 1 desc for context descriptor,
8602          * otherwise try next time
8603          */
8604         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
8605                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
8606
8607         if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
8608                 tx_ring->tx_stats.tx_busy++;
8609                 return NETDEV_TX_BUSY;
8610         }
8611
8612         /* record the location of the first descriptor for this packet */
8613         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
8614         first->skb = skb;
8615         first->bytecount = skb->len;
8616         first->gso_segs = 1;
8617
8618         /* if we have a HW VLAN tag being added default to the HW one */
8619         if (skb_vlan_tag_present(skb)) {
8620                 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
8621                 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8622         /* else if it is a SW VLAN check the next protocol and store the tag */
8623         } else if (protocol == htons(ETH_P_8021Q)) {
8624                 struct vlan_hdr *vhdr, _vhdr;
8625                 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
8626                 if (!vhdr)
8627                         goto out_drop;
8628
8629                 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
8630                                   IXGBE_TX_FLAGS_VLAN_SHIFT;
8631                 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
8632         }
8633         protocol = vlan_get_protocol(skb);
8634
8635         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
8636             adapter->ptp_clock) {
8637                 if (!test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
8638                                            &adapter->state)) {
8639                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
8640                         tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
8641
8642                         /* schedule check for Tx timestamp */
8643                         adapter->ptp_tx_skb = skb_get(skb);
8644                         adapter->ptp_tx_start = jiffies;
8645                         schedule_work(&adapter->ptp_tx_work);
8646                 } else {
8647                         adapter->tx_hwtstamp_skipped++;
8648                 }
8649         }
8650
8651 #ifdef CONFIG_PCI_IOV
8652         /*
8653          * Use the l2switch_enable flag - would be false if the DMA
8654          * Tx switch had been disabled.
8655          */
8656         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
8657                 tx_flags |= IXGBE_TX_FLAGS_CC;
8658
8659 #endif
8660         /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
8661         if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8662             ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
8663              (skb->priority != TC_PRIO_CONTROL))) {
8664                 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
8665                 tx_flags |= (skb->priority & 0x7) <<
8666                                         IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
8667                 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
8668                         struct vlan_ethhdr *vhdr;
8669
8670                         if (skb_cow_head(skb, 0))
8671                                 goto out_drop;
8672                         vhdr = (struct vlan_ethhdr *)skb->data;
8673                         vhdr->h_vlan_TCI = htons(tx_flags >>
8674                                                  IXGBE_TX_FLAGS_VLAN_SHIFT);
8675                 } else {
8676                         tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
8677                 }
8678         }
8679
8680         /* record initial flags and protocol */
8681         first->tx_flags = tx_flags;
8682         first->protocol = protocol;
8683
8684 #ifdef IXGBE_FCOE
8685         /* setup tx offload for FCoE */
8686         if ((protocol == htons(ETH_P_FCOE)) &&
8687             (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
8688                 tso = ixgbe_fso(tx_ring, first, &hdr_len);
8689                 if (tso < 0)
8690                         goto out_drop;
8691
8692                 goto xmit_fcoe;
8693         }
8694
8695 #endif /* IXGBE_FCOE */
8696
8697 #ifdef CONFIG_IXGBE_IPSEC
8698         if (secpath_exists(skb) &&
8699             !ixgbe_ipsec_tx(tx_ring, first, &ipsec_tx))
8700                 goto out_drop;
8701 #endif
8702         tso = ixgbe_tso(tx_ring, first, &hdr_len, &ipsec_tx);
8703         if (tso < 0)
8704                 goto out_drop;
8705         else if (!tso)
8706                 ixgbe_tx_csum(tx_ring, first, &ipsec_tx);
8707
8708         /* add the ATR filter if ATR is on */
8709         if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
8710                 ixgbe_atr(tx_ring, first);
8711
8712 #ifdef IXGBE_FCOE
8713 xmit_fcoe:
8714 #endif /* IXGBE_FCOE */
8715         if (ixgbe_tx_map(tx_ring, first, hdr_len))
8716                 goto cleanup_tx_timestamp;
8717
8718         return NETDEV_TX_OK;
8719
8720 out_drop:
8721         dev_kfree_skb_any(first->skb);
8722         first->skb = NULL;
8723 cleanup_tx_timestamp:
8724         if (unlikely(tx_flags & IXGBE_TX_FLAGS_TSTAMP)) {
8725                 dev_kfree_skb_any(adapter->ptp_tx_skb);
8726                 adapter->ptp_tx_skb = NULL;
8727                 cancel_work_sync(&adapter->ptp_tx_work);
8728                 clear_bit_unlock(__IXGBE_PTP_TX_IN_PROGRESS, &adapter->state);
8729         }
8730
8731         return NETDEV_TX_OK;
8732 }
8733
8734 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
8735                                       struct net_device *netdev,
8736                                       struct ixgbe_ring *ring)
8737 {
8738         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8739         struct ixgbe_ring *tx_ring;
8740
8741         /*
8742          * The minimum packet size for olinfo paylen is 17 so pad the skb
8743          * in order to meet this minimum size requirement.
8744          */
8745         if (skb_put_padto(skb, 17))
8746                 return NETDEV_TX_OK;
8747
8748         tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
8749         if (unlikely(test_bit(__IXGBE_TX_DISABLED, &tx_ring->state)))
8750                 return NETDEV_TX_BUSY;
8751
8752         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
8753 }
8754
8755 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
8756                                     struct net_device *netdev)
8757 {
8758         return __ixgbe_xmit_frame(skb, netdev, NULL);
8759 }
8760
8761 /**
8762  * ixgbe_set_mac - Change the Ethernet Address of the NIC
8763  * @netdev: network interface device structure
8764  * @p: pointer to an address structure
8765  *
8766  * Returns 0 on success, negative on failure
8767  **/
8768 static int ixgbe_set_mac(struct net_device *netdev, void *p)
8769 {
8770         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8771         struct ixgbe_hw *hw = &adapter->hw;
8772         struct sockaddr *addr = p;
8773
8774         if (!is_valid_ether_addr(addr->sa_data))
8775                 return -EADDRNOTAVAIL;
8776
8777         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
8778         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
8779
8780         ixgbe_mac_set_default_filter(adapter);
8781
8782         return 0;
8783 }
8784
8785 static int
8786 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
8787 {
8788         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8789         struct ixgbe_hw *hw = &adapter->hw;
8790         u16 value;
8791         int rc;
8792
8793         if (adapter->mii_bus) {
8794                 int regnum = addr;
8795
8796                 if (devad != MDIO_DEVAD_NONE)
8797                         regnum |= (devad << 16) | MII_ADDR_C45;
8798
8799                 return mdiobus_read(adapter->mii_bus, prtad, regnum);
8800         }
8801
8802         if (prtad != hw->phy.mdio.prtad)
8803                 return -EINVAL;
8804         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
8805         if (!rc)
8806                 rc = value;
8807         return rc;
8808 }
8809
8810 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
8811                             u16 addr, u16 value)
8812 {
8813         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8814         struct ixgbe_hw *hw = &adapter->hw;
8815
8816         if (adapter->mii_bus) {
8817                 int regnum = addr;
8818
8819                 if (devad != MDIO_DEVAD_NONE)
8820                         regnum |= (devad << 16) | MII_ADDR_C45;
8821
8822                 return mdiobus_write(adapter->mii_bus, prtad, regnum, value);
8823         }
8824
8825         if (prtad != hw->phy.mdio.prtad)
8826                 return -EINVAL;
8827         return hw->phy.ops.write_reg(hw, addr, devad, value);
8828 }
8829
8830 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
8831 {
8832         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8833
8834         switch (cmd) {
8835         case SIOCSHWTSTAMP:
8836                 return ixgbe_ptp_set_ts_config(adapter, req);
8837         case SIOCGHWTSTAMP:
8838                 return ixgbe_ptp_get_ts_config(adapter, req);
8839         case SIOCGMIIPHY:
8840                 if (!adapter->hw.phy.ops.read_reg)
8841                         return -EOPNOTSUPP;
8842                 /* fall through */
8843         default:
8844                 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
8845         }
8846 }
8847
8848 /**
8849  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
8850  * netdev->dev_addrs
8851  * @dev: network interface device structure
8852  *
8853  * Returns non-zero on failure
8854  **/
8855 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
8856 {
8857         int err = 0;
8858         struct ixgbe_adapter *adapter = netdev_priv(dev);
8859         struct ixgbe_hw *hw = &adapter->hw;
8860
8861         if (is_valid_ether_addr(hw->mac.san_addr)) {
8862                 rtnl_lock();
8863                 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
8864                 rtnl_unlock();
8865
8866                 /* update SAN MAC vmdq pool selection */
8867                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
8868         }
8869         return err;
8870 }
8871
8872 /**
8873  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
8874  * netdev->dev_addrs
8875  * @dev: network interface device structure
8876  *
8877  * Returns non-zero on failure
8878  **/
8879 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
8880 {
8881         int err = 0;
8882         struct ixgbe_adapter *adapter = netdev_priv(dev);
8883         struct ixgbe_mac_info *mac = &adapter->hw.mac;
8884
8885         if (is_valid_ether_addr(mac->san_addr)) {
8886                 rtnl_lock();
8887                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8888                 rtnl_unlock();
8889         }
8890         return err;
8891 }
8892
8893 static void ixgbe_get_ring_stats64(struct rtnl_link_stats64 *stats,
8894                                    struct ixgbe_ring *ring)
8895 {
8896         u64 bytes, packets;
8897         unsigned int start;
8898
8899         if (ring) {
8900                 do {
8901                         start = u64_stats_fetch_begin_irq(&ring->syncp);
8902                         packets = ring->stats.packets;
8903                         bytes   = ring->stats.bytes;
8904                 } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8905                 stats->tx_packets += packets;
8906                 stats->tx_bytes   += bytes;
8907         }
8908 }
8909
8910 static void ixgbe_get_stats64(struct net_device *netdev,
8911                               struct rtnl_link_stats64 *stats)
8912 {
8913         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8914         int i;
8915
8916         rcu_read_lock();
8917         for (i = 0; i < adapter->num_rx_queues; i++) {
8918                 struct ixgbe_ring *ring = READ_ONCE(adapter->rx_ring[i]);
8919                 u64 bytes, packets;
8920                 unsigned int start;
8921
8922                 if (ring) {
8923                         do {
8924                                 start = u64_stats_fetch_begin_irq(&ring->syncp);
8925                                 packets = ring->stats.packets;
8926                                 bytes   = ring->stats.bytes;
8927                         } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8928                         stats->rx_packets += packets;
8929                         stats->rx_bytes   += bytes;
8930                 }
8931         }
8932
8933         for (i = 0; i < adapter->num_tx_queues; i++) {
8934                 struct ixgbe_ring *ring = READ_ONCE(adapter->tx_ring[i]);
8935
8936                 ixgbe_get_ring_stats64(stats, ring);
8937         }
8938         for (i = 0; i < adapter->num_xdp_queues; i++) {
8939                 struct ixgbe_ring *ring = READ_ONCE(adapter->xdp_ring[i]);
8940
8941                 ixgbe_get_ring_stats64(stats, ring);
8942         }
8943         rcu_read_unlock();
8944
8945         /* following stats updated by ixgbe_watchdog_task() */
8946         stats->multicast        = netdev->stats.multicast;
8947         stats->rx_errors        = netdev->stats.rx_errors;
8948         stats->rx_length_errors = netdev->stats.rx_length_errors;
8949         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
8950         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8951 }
8952
8953 #ifdef CONFIG_IXGBE_DCB
8954 /**
8955  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8956  * @adapter: pointer to ixgbe_adapter
8957  * @tc: number of traffic classes currently enabled
8958  *
8959  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8960  * 802.1Q priority maps to a packet buffer that exists.
8961  */
8962 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8963 {
8964         struct ixgbe_hw *hw = &adapter->hw;
8965         u32 reg, rsave;
8966         int i;
8967
8968         /* 82598 have a static priority to TC mapping that can not
8969          * be changed so no validation is needed.
8970          */
8971         if (hw->mac.type == ixgbe_mac_82598EB)
8972                 return;
8973
8974         reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8975         rsave = reg;
8976
8977         for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8978                 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8979
8980                 /* If up2tc is out of bounds default to zero */
8981                 if (up2tc > tc)
8982                         reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8983         }
8984
8985         if (reg != rsave)
8986                 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8987
8988         return;
8989 }
8990
8991 /**
8992  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8993  * @adapter: Pointer to adapter struct
8994  *
8995  * Populate the netdev user priority to tc map
8996  */
8997 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
8998 {
8999         struct net_device *dev = adapter->netdev;
9000         struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
9001         struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
9002         u8 prio;
9003
9004         for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
9005                 u8 tc = 0;
9006
9007                 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
9008                         tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
9009                 else if (ets)
9010                         tc = ets->prio_tc[prio];
9011
9012                 netdev_set_prio_tc_map(dev, prio, tc);
9013         }
9014 }
9015
9016 #endif /* CONFIG_IXGBE_DCB */
9017 static int ixgbe_reassign_macvlan_pool(struct net_device *vdev, void *data)
9018 {
9019         struct ixgbe_adapter *adapter = data;
9020         struct ixgbe_fwd_adapter *accel;
9021         int pool;
9022
9023         /* we only care about macvlans... */
9024         if (!netif_is_macvlan(vdev))
9025                 return 0;
9026
9027         /* that have hardware offload enabled... */
9028         accel = macvlan_accel_priv(vdev);
9029         if (!accel)
9030                 return 0;
9031
9032         /* If we can relocate to a different bit do so */
9033         pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
9034         if (pool < adapter->num_rx_pools) {
9035                 set_bit(pool, adapter->fwd_bitmask);
9036                 accel->pool = pool;
9037                 return 0;
9038         }
9039
9040         /* if we cannot find a free pool then disable the offload */
9041         netdev_err(vdev, "L2FW offload disabled due to lack of queue resources\n");
9042         macvlan_release_l2fw_offload(vdev);
9043
9044         /* unbind the queues and drop the subordinate channel config */
9045         netdev_unbind_sb_channel(adapter->netdev, vdev);
9046         netdev_set_sb_channel(vdev, 0);
9047
9048         kfree(accel);
9049
9050         return 0;
9051 }
9052
9053 static void ixgbe_defrag_macvlan_pools(struct net_device *dev)
9054 {
9055         struct ixgbe_adapter *adapter = netdev_priv(dev);
9056
9057         /* flush any stale bits out of the fwd bitmask */
9058         bitmap_clear(adapter->fwd_bitmask, 1, 63);
9059
9060         /* walk through upper devices reassigning pools */
9061         netdev_walk_all_upper_dev_rcu(dev, ixgbe_reassign_macvlan_pool,
9062                                       adapter);
9063 }
9064
9065 /**
9066  * ixgbe_setup_tc - configure net_device for multiple traffic classes
9067  *
9068  * @dev: net device to configure
9069  * @tc: number of traffic classes to enable
9070  */
9071 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
9072 {
9073         struct ixgbe_adapter *adapter = netdev_priv(dev);
9074         struct ixgbe_hw *hw = &adapter->hw;
9075
9076         /* Hardware supports up to 8 traffic classes */
9077         if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
9078                 return -EINVAL;
9079
9080         if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
9081                 return -EINVAL;
9082
9083         /* Hardware has to reinitialize queues and interrupts to
9084          * match packet buffer alignment. Unfortunately, the
9085          * hardware is not flexible enough to do this dynamically.
9086          */
9087         if (netif_running(dev))
9088                 ixgbe_close(dev);
9089         else
9090                 ixgbe_reset(adapter);
9091
9092         ixgbe_clear_interrupt_scheme(adapter);
9093
9094 #ifdef CONFIG_IXGBE_DCB
9095         if (tc) {
9096                 if (adapter->xdp_prog) {
9097                         e_warn(probe, "DCB is not supported with XDP\n");
9098
9099                         ixgbe_init_interrupt_scheme(adapter);
9100                         if (netif_running(dev))
9101                                 ixgbe_open(dev);
9102                         return -EINVAL;
9103                 }
9104
9105                 netdev_set_num_tc(dev, tc);
9106                 ixgbe_set_prio_tc_map(adapter);
9107
9108                 adapter->hw_tcs = tc;
9109                 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
9110
9111                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
9112                         adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
9113                         adapter->hw.fc.requested_mode = ixgbe_fc_none;
9114                 }
9115         } else {
9116                 netdev_reset_tc(dev);
9117
9118                 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9119                         adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
9120
9121                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
9122                 adapter->hw_tcs = tc;
9123
9124                 adapter->temp_dcb_cfg.pfc_mode_enable = false;
9125                 adapter->dcb_cfg.pfc_mode_enable = false;
9126         }
9127
9128         ixgbe_validate_rtr(adapter, tc);
9129
9130 #endif /* CONFIG_IXGBE_DCB */
9131         ixgbe_init_interrupt_scheme(adapter);
9132
9133         ixgbe_defrag_macvlan_pools(dev);
9134
9135         if (netif_running(dev))
9136                 return ixgbe_open(dev);
9137
9138         return 0;
9139 }
9140
9141 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
9142                                struct tc_cls_u32_offload *cls)
9143 {
9144         u32 hdl = cls->knode.handle;
9145         u32 uhtid = TC_U32_USERHTID(cls->knode.handle);
9146         u32 loc = cls->knode.handle & 0xfffff;
9147         int err = 0, i, j;
9148         struct ixgbe_jump_table *jump = NULL;
9149
9150         if (loc > IXGBE_MAX_HW_ENTRIES)
9151                 return -EINVAL;
9152
9153         if ((uhtid != 0x800) && (uhtid >= IXGBE_MAX_LINK_HANDLE))
9154                 return -EINVAL;
9155
9156         /* Clear this filter in the link data it is associated with */
9157         if (uhtid != 0x800) {
9158                 jump = adapter->jump_tables[uhtid];
9159                 if (!jump)
9160                         return -EINVAL;
9161                 if (!test_bit(loc - 1, jump->child_loc_map))
9162                         return -EINVAL;
9163                 clear_bit(loc - 1, jump->child_loc_map);
9164         }
9165
9166         /* Check if the filter being deleted is a link */
9167         for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9168                 jump = adapter->jump_tables[i];
9169                 if (jump && jump->link_hdl == hdl) {
9170                         /* Delete filters in the hardware in the child hash
9171                          * table associated with this link
9172                          */
9173                         for (j = 0; j < IXGBE_MAX_HW_ENTRIES; j++) {
9174                                 if (!test_bit(j, jump->child_loc_map))
9175                                         continue;
9176                                 spin_lock(&adapter->fdir_perfect_lock);
9177                                 err = ixgbe_update_ethtool_fdir_entry(adapter,
9178                                                                       NULL,
9179                                                                       j + 1);
9180                                 spin_unlock(&adapter->fdir_perfect_lock);
9181                                 clear_bit(j, jump->child_loc_map);
9182                         }
9183                         /* Remove resources for this link */
9184                         kfree(jump->input);
9185                         kfree(jump->mask);
9186                         kfree(jump);
9187                         adapter->jump_tables[i] = NULL;
9188                         return err;
9189                 }
9190         }
9191
9192         spin_lock(&adapter->fdir_perfect_lock);
9193         err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, loc);
9194         spin_unlock(&adapter->fdir_perfect_lock);
9195         return err;
9196 }
9197
9198 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
9199                                             struct tc_cls_u32_offload *cls)
9200 {
9201         u32 uhtid = TC_U32_USERHTID(cls->hnode.handle);
9202
9203         if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9204                 return -EINVAL;
9205
9206         /* This ixgbe devices do not support hash tables at the moment
9207          * so abort when given hash tables.
9208          */
9209         if (cls->hnode.divisor > 0)
9210                 return -EINVAL;
9211
9212         set_bit(uhtid - 1, &adapter->tables);
9213         return 0;
9214 }
9215
9216 static int ixgbe_configure_clsu32_del_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         clear_bit(uhtid - 1, &adapter->tables);
9225         return 0;
9226 }
9227
9228 #ifdef CONFIG_NET_CLS_ACT
9229 struct upper_walk_data {
9230         struct ixgbe_adapter *adapter;
9231         u64 action;
9232         int ifindex;
9233         u8 queue;
9234 };
9235
9236 static int get_macvlan_queue(struct net_device *upper, void *_data)
9237 {
9238         if (netif_is_macvlan(upper)) {
9239                 struct ixgbe_fwd_adapter *vadapter = macvlan_accel_priv(upper);
9240                 struct upper_walk_data *data = _data;
9241                 struct ixgbe_adapter *adapter = data->adapter;
9242                 int ifindex = data->ifindex;
9243
9244                 if (vadapter && upper->ifindex == ifindex) {
9245                         data->queue = adapter->rx_ring[vadapter->rx_base_queue]->reg_idx;
9246                         data->action = data->queue;
9247                         return 1;
9248                 }
9249         }
9250
9251         return 0;
9252 }
9253
9254 static int handle_redirect_action(struct ixgbe_adapter *adapter, int ifindex,
9255                                   u8 *queue, u64 *action)
9256 {
9257         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
9258         unsigned int num_vfs = adapter->num_vfs, vf;
9259         struct upper_walk_data data;
9260         struct net_device *upper;
9261
9262         /* redirect to a SRIOV VF */
9263         for (vf = 0; vf < num_vfs; ++vf) {
9264                 upper = pci_get_drvdata(adapter->vfinfo[vf].vfdev);
9265                 if (upper->ifindex == ifindex) {
9266                         *queue = vf * __ALIGN_MASK(1, ~vmdq->mask);
9267                         *action = vf + 1;
9268                         *action <<= ETHTOOL_RX_FLOW_SPEC_RING_VF_OFF;
9269                         return 0;
9270                 }
9271         }
9272
9273         /* redirect to a offloaded macvlan netdev */
9274         data.adapter = adapter;
9275         data.ifindex = ifindex;
9276         data.action = 0;
9277         data.queue = 0;
9278         if (netdev_walk_all_upper_dev_rcu(adapter->netdev,
9279                                           get_macvlan_queue, &data)) {
9280                 *action = data.action;
9281                 *queue = data.queue;
9282
9283                 return 0;
9284         }
9285
9286         return -EINVAL;
9287 }
9288
9289 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9290                             struct tcf_exts *exts, u64 *action, u8 *queue)
9291 {
9292         const struct tc_action *a;
9293         int i;
9294
9295         if (!tcf_exts_has_actions(exts))
9296                 return -EINVAL;
9297
9298         tcf_exts_for_each_action(i, a, exts) {
9299                 /* Drop action */
9300                 if (is_tcf_gact_shot(a)) {
9301                         *action = IXGBE_FDIR_DROP_QUEUE;
9302                         *queue = IXGBE_FDIR_DROP_QUEUE;
9303                         return 0;
9304                 }
9305
9306                 /* Redirect to a VF or a offloaded macvlan */
9307                 if (is_tcf_mirred_egress_redirect(a)) {
9308                         struct net_device *dev = tcf_mirred_dev(a);
9309
9310                         if (!dev)
9311                                 return -EINVAL;
9312                         return handle_redirect_action(adapter, dev->ifindex,
9313                                                       queue, action);
9314                 }
9315
9316                 return -EINVAL;
9317         }
9318
9319         return -EINVAL;
9320 }
9321 #else
9322 static int parse_tc_actions(struct ixgbe_adapter *adapter,
9323                             struct tcf_exts *exts, u64 *action, u8 *queue)
9324 {
9325         return -EINVAL;
9326 }
9327 #endif /* CONFIG_NET_CLS_ACT */
9328
9329 static int ixgbe_clsu32_build_input(struct ixgbe_fdir_filter *input,
9330                                     union ixgbe_atr_input *mask,
9331                                     struct tc_cls_u32_offload *cls,
9332                                     struct ixgbe_mat_field *field_ptr,
9333                                     struct ixgbe_nexthdr *nexthdr)
9334 {
9335         int i, j, off;
9336         __be32 val, m;
9337         bool found_entry = false, found_jump_field = false;
9338
9339         for (i = 0; i < cls->knode.sel->nkeys; i++) {
9340                 off = cls->knode.sel->keys[i].off;
9341                 val = cls->knode.sel->keys[i].val;
9342                 m = cls->knode.sel->keys[i].mask;
9343
9344                 for (j = 0; field_ptr[j].val; j++) {
9345                         if (field_ptr[j].off == off) {
9346                                 field_ptr[j].val(input, mask, (__force u32)val,
9347                                                  (__force u32)m);
9348                                 input->filter.formatted.flow_type |=
9349                                         field_ptr[j].type;
9350                                 found_entry = true;
9351                                 break;
9352                         }
9353                 }
9354                 if (nexthdr) {
9355                         if (nexthdr->off == cls->knode.sel->keys[i].off &&
9356                             nexthdr->val ==
9357                             (__force u32)cls->knode.sel->keys[i].val &&
9358                             nexthdr->mask ==
9359                             (__force u32)cls->knode.sel->keys[i].mask)
9360                                 found_jump_field = true;
9361                         else
9362                                 continue;
9363                 }
9364         }
9365
9366         if (nexthdr && !found_jump_field)
9367                 return -EINVAL;
9368
9369         if (!found_entry)
9370                 return 0;
9371
9372         mask->formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
9373                                     IXGBE_ATR_L4TYPE_MASK;
9374
9375         if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
9376                 mask->formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
9377
9378         return 0;
9379 }
9380
9381 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
9382                                   struct tc_cls_u32_offload *cls)
9383 {
9384         __be16 protocol = cls->common.protocol;
9385         u32 loc = cls->knode.handle & 0xfffff;
9386         struct ixgbe_hw *hw = &adapter->hw;
9387         struct ixgbe_mat_field *field_ptr;
9388         struct ixgbe_fdir_filter *input = NULL;
9389         union ixgbe_atr_input *mask = NULL;
9390         struct ixgbe_jump_table *jump = NULL;
9391         int i, err = -EINVAL;
9392         u8 queue;
9393         u32 uhtid, link_uhtid;
9394
9395         uhtid = TC_U32_USERHTID(cls->knode.handle);
9396         link_uhtid = TC_U32_USERHTID(cls->knode.link_handle);
9397
9398         /* At the moment cls_u32 jumps to network layer and skips past
9399          * L2 headers. The canonical method to match L2 frames is to use
9400          * negative values. However this is error prone at best but really
9401          * just broken because there is no way to "know" what sort of hdr
9402          * is in front of the network layer. Fix cls_u32 to support L2
9403          * headers when needed.
9404          */
9405         if (protocol != htons(ETH_P_IP))
9406                 return err;
9407
9408         if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
9409                 e_err(drv, "Location out of range\n");
9410                 return err;
9411         }
9412
9413         /* cls u32 is a graph starting at root node 0x800. The driver tracks
9414          * links and also the fields used to advance the parser across each
9415          * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
9416          * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
9417          * To add support for new nodes update ixgbe_model.h parse structures
9418          * this function _should_ be generic try not to hardcode values here.
9419          */
9420         if (uhtid == 0x800) {
9421                 field_ptr = (adapter->jump_tables[0])->mat;
9422         } else {
9423                 if (uhtid >= IXGBE_MAX_LINK_HANDLE)
9424                         return err;
9425                 if (!adapter->jump_tables[uhtid])
9426                         return err;
9427                 field_ptr = (adapter->jump_tables[uhtid])->mat;
9428         }
9429
9430         if (!field_ptr)
9431                 return err;
9432
9433         /* At this point we know the field_ptr is valid and need to either
9434          * build cls_u32 link or attach filter. Because adding a link to
9435          * a handle that does not exist is invalid and the same for adding
9436          * rules to handles that don't exist.
9437          */
9438
9439         if (link_uhtid) {
9440                 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
9441
9442                 if (link_uhtid >= IXGBE_MAX_LINK_HANDLE)
9443                         return err;
9444
9445                 if (!test_bit(link_uhtid - 1, &adapter->tables))
9446                         return err;
9447
9448                 /* Multiple filters as links to the same hash table are not
9449                  * supported. To add a new filter with the same next header
9450                  * but different match/jump conditions, create a new hash table
9451                  * and link to it.
9452                  */
9453                 if (adapter->jump_tables[link_uhtid] &&
9454                     (adapter->jump_tables[link_uhtid])->link_hdl) {
9455                         e_err(drv, "Link filter exists for link: %x\n",
9456                               link_uhtid);
9457                         return err;
9458                 }
9459
9460                 for (i = 0; nexthdr[i].jump; i++) {
9461                         if (nexthdr[i].o != cls->knode.sel->offoff ||
9462                             nexthdr[i].s != cls->knode.sel->offshift ||
9463                             nexthdr[i].m !=
9464                             (__force u32)cls->knode.sel->offmask)
9465                                 return err;
9466
9467                         jump = kzalloc(sizeof(*jump), GFP_KERNEL);
9468                         if (!jump)
9469                                 return -ENOMEM;
9470                         input = kzalloc(sizeof(*input), GFP_KERNEL);
9471                         if (!input) {
9472                                 err = -ENOMEM;
9473                                 goto free_jump;
9474                         }
9475                         mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9476                         if (!mask) {
9477                                 err = -ENOMEM;
9478                                 goto free_input;
9479                         }
9480                         jump->input = input;
9481                         jump->mask = mask;
9482                         jump->link_hdl = cls->knode.handle;
9483
9484                         err = ixgbe_clsu32_build_input(input, mask, cls,
9485                                                        field_ptr, &nexthdr[i]);
9486                         if (!err) {
9487                                 jump->mat = nexthdr[i].jump;
9488                                 adapter->jump_tables[link_uhtid] = jump;
9489                                 break;
9490                         }
9491                 }
9492                 return 0;
9493         }
9494
9495         input = kzalloc(sizeof(*input), GFP_KERNEL);
9496         if (!input)
9497                 return -ENOMEM;
9498         mask = kzalloc(sizeof(*mask), GFP_KERNEL);
9499         if (!mask) {
9500                 err = -ENOMEM;
9501                 goto free_input;
9502         }
9503
9504         if ((uhtid != 0x800) && (adapter->jump_tables[uhtid])) {
9505                 if ((adapter->jump_tables[uhtid])->input)
9506                         memcpy(input, (adapter->jump_tables[uhtid])->input,
9507                                sizeof(*input));
9508                 if ((adapter->jump_tables[uhtid])->mask)
9509                         memcpy(mask, (adapter->jump_tables[uhtid])->mask,
9510                                sizeof(*mask));
9511
9512                 /* Lookup in all child hash tables if this location is already
9513                  * filled with a filter
9514                  */
9515                 for (i = 1; i < IXGBE_MAX_LINK_HANDLE; i++) {
9516                         struct ixgbe_jump_table *link = adapter->jump_tables[i];
9517
9518                         if (link && (test_bit(loc - 1, link->child_loc_map))) {
9519                                 e_err(drv, "Filter exists in location: %x\n",
9520                                       loc);
9521                                 err = -EINVAL;
9522                                 goto err_out;
9523                         }
9524                 }
9525         }
9526         err = ixgbe_clsu32_build_input(input, mask, cls, field_ptr, NULL);
9527         if (err)
9528                 goto err_out;
9529
9530         err = parse_tc_actions(adapter, cls->knode.exts, &input->action,
9531                                &queue);
9532         if (err < 0)
9533                 goto err_out;
9534
9535         input->sw_idx = loc;
9536
9537         spin_lock(&adapter->fdir_perfect_lock);
9538
9539         if (hlist_empty(&adapter->fdir_filter_list)) {
9540                 memcpy(&adapter->fdir_mask, mask, sizeof(*mask));
9541                 err = ixgbe_fdir_set_input_mask_82599(hw, mask);
9542                 if (err)
9543                         goto err_out_w_lock;
9544         } else if (memcmp(&adapter->fdir_mask, mask, sizeof(*mask))) {
9545                 err = -EINVAL;
9546                 goto err_out_w_lock;
9547         }
9548
9549         ixgbe_atr_compute_perfect_hash_82599(&input->filter, mask);
9550         err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
9551                                                     input->sw_idx, queue);
9552         if (!err)
9553                 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
9554         spin_unlock(&adapter->fdir_perfect_lock);
9555
9556         if ((uhtid != 0x800) && (adapter->jump_tables[uhtid]))
9557                 set_bit(loc - 1, (adapter->jump_tables[uhtid])->child_loc_map);
9558
9559         kfree(mask);
9560         return err;
9561 err_out_w_lock:
9562         spin_unlock(&adapter->fdir_perfect_lock);
9563 err_out:
9564         kfree(mask);
9565 free_input:
9566         kfree(input);
9567 free_jump:
9568         kfree(jump);
9569         return err;
9570 }
9571
9572 static int ixgbe_setup_tc_cls_u32(struct ixgbe_adapter *adapter,
9573                                   struct tc_cls_u32_offload *cls_u32)
9574 {
9575         switch (cls_u32->command) {
9576         case TC_CLSU32_NEW_KNODE:
9577         case TC_CLSU32_REPLACE_KNODE:
9578                 return ixgbe_configure_clsu32(adapter, cls_u32);
9579         case TC_CLSU32_DELETE_KNODE:
9580                 return ixgbe_delete_clsu32(adapter, cls_u32);
9581         case TC_CLSU32_NEW_HNODE:
9582         case TC_CLSU32_REPLACE_HNODE:
9583                 return ixgbe_configure_clsu32_add_hnode(adapter, cls_u32);
9584         case TC_CLSU32_DELETE_HNODE:
9585                 return ixgbe_configure_clsu32_del_hnode(adapter, cls_u32);
9586         default:
9587                 return -EOPNOTSUPP;
9588         }
9589 }
9590
9591 static int ixgbe_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
9592                                    void *cb_priv)
9593 {
9594         struct ixgbe_adapter *adapter = cb_priv;
9595
9596         if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
9597                 return -EOPNOTSUPP;
9598
9599         switch (type) {
9600         case TC_SETUP_CLSU32:
9601                 return ixgbe_setup_tc_cls_u32(adapter, type_data);
9602         default:
9603                 return -EOPNOTSUPP;
9604         }
9605 }
9606
9607 static int ixgbe_setup_tc_mqprio(struct net_device *dev,
9608                                  struct tc_mqprio_qopt *mqprio)
9609 {
9610         mqprio->hw = TC_MQPRIO_HW_OFFLOAD_TCS;
9611         return ixgbe_setup_tc(dev, mqprio->num_tc);
9612 }
9613
9614 static LIST_HEAD(ixgbe_block_cb_list);
9615
9616 static int __ixgbe_setup_tc(struct net_device *dev, enum tc_setup_type type,
9617                             void *type_data)
9618 {
9619         struct ixgbe_adapter *adapter = netdev_priv(dev);
9620
9621         switch (type) {
9622         case TC_SETUP_BLOCK:
9623                 return flow_block_cb_setup_simple(type_data,
9624                                                   &ixgbe_block_cb_list,
9625                                                   ixgbe_setup_tc_block_cb,
9626                                                   adapter, adapter, true);
9627         case TC_SETUP_QDISC_MQPRIO:
9628                 return ixgbe_setup_tc_mqprio(dev, type_data);
9629         default:
9630                 return -EOPNOTSUPP;
9631         }
9632 }
9633
9634 #ifdef CONFIG_PCI_IOV
9635 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
9636 {
9637         struct net_device *netdev = adapter->netdev;
9638
9639         rtnl_lock();
9640         ixgbe_setup_tc(netdev, adapter->hw_tcs);
9641         rtnl_unlock();
9642 }
9643
9644 #endif
9645 void ixgbe_do_reset(struct net_device *netdev)
9646 {
9647         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9648
9649         if (netif_running(netdev))
9650                 ixgbe_reinit_locked(adapter);
9651         else
9652                 ixgbe_reset(adapter);
9653 }
9654
9655 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
9656                                             netdev_features_t features)
9657 {
9658         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9659
9660         /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
9661         if (!(features & NETIF_F_RXCSUM))
9662                 features &= ~NETIF_F_LRO;
9663
9664         /* Turn off LRO if not RSC capable */
9665         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
9666                 features &= ~NETIF_F_LRO;
9667
9668         if (adapter->xdp_prog && (features & NETIF_F_LRO)) {
9669                 e_dev_err("LRO is not supported with XDP\n");
9670                 features &= ~NETIF_F_LRO;
9671         }
9672
9673         return features;
9674 }
9675
9676 static void ixgbe_reset_l2fw_offload(struct ixgbe_adapter *adapter)
9677 {
9678         int rss = min_t(int, ixgbe_max_rss_indices(adapter),
9679                         num_online_cpus());
9680
9681         /* go back to full RSS if we're not running SR-IOV */
9682         if (!adapter->ring_feature[RING_F_VMDQ].offset)
9683                 adapter->flags &= ~(IXGBE_FLAG_VMDQ_ENABLED |
9684                                     IXGBE_FLAG_SRIOV_ENABLED);
9685
9686         adapter->ring_feature[RING_F_RSS].limit = rss;
9687         adapter->ring_feature[RING_F_VMDQ].limit = 1;
9688
9689         ixgbe_setup_tc(adapter->netdev, adapter->hw_tcs);
9690 }
9691
9692 static int ixgbe_set_features(struct net_device *netdev,
9693                               netdev_features_t features)
9694 {
9695         struct ixgbe_adapter *adapter = netdev_priv(netdev);
9696         netdev_features_t changed = netdev->features ^ features;
9697         bool need_reset = false;
9698
9699         /* Make sure RSC matches LRO, reset if change */
9700         if (!(features & NETIF_F_LRO)) {
9701                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9702                         need_reset = true;
9703                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
9704         } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
9705                    !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
9706                 if (adapter->rx_itr_setting == 1 ||
9707                     adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
9708                         adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
9709                         need_reset = true;
9710                 } else if ((changed ^ features) & NETIF_F_LRO) {
9711                         e_info(probe, "rx-usecs set too low, "
9712                                "disabling RSC\n");
9713                 }
9714         }
9715
9716         /*
9717          * Check if Flow Director n-tuple support or hw_tc support was
9718          * enabled or disabled.  If the state changed, we need to reset.
9719          */
9720         if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
9721                 /* turn off ATR, enable perfect filters and reset */
9722                 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
9723                         need_reset = true;
9724
9725                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
9726                 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9727         } else {
9728                 /* turn off perfect filters, enable ATR and reset */
9729                 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
9730                         need_reset = true;
9731
9732                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
9733
9734                 /* We cannot enable ATR if SR-IOV is enabled */
9735                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
9736                     /* We cannot enable ATR if we have 2 or more tcs */
9737                     (adapter->hw_tcs > 1) ||
9738                     /* We cannot enable ATR if RSS is disabled */
9739                     (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
9740                     /* A sample rate of 0 indicates ATR disabled */
9741                     (!adapter->atr_sample_rate))
9742                         ; /* do nothing not supported */
9743                 else /* otherwise supported and set the flag */
9744                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
9745         }
9746
9747         if (changed & NETIF_F_RXALL)
9748                 need_reset = true;
9749
9750         netdev->features = features;
9751
9752         if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
9753                 if (features & NETIF_F_RXCSUM) {
9754                         adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9755                 } else {
9756                         u32 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9757
9758                         ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9759                 }
9760         }
9761
9762         if ((adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE)) {
9763                 if (features & NETIF_F_RXCSUM) {
9764                         adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9765                 } else {
9766                         u32 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9767
9768                         ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9769                 }
9770         }
9771
9772         if ((changed & NETIF_F_HW_L2FW_DOFFLOAD) && adapter->num_rx_pools > 1)
9773                 ixgbe_reset_l2fw_offload(adapter);
9774         else if (need_reset)
9775                 ixgbe_do_reset(netdev);
9776         else if (changed & (NETIF_F_HW_VLAN_CTAG_RX |
9777                             NETIF_F_HW_VLAN_CTAG_FILTER))
9778                 ixgbe_set_rx_mode(netdev);
9779
9780         return 1;
9781 }
9782
9783 /**
9784  * ixgbe_add_udp_tunnel_port - Get notifications about adding UDP tunnel ports
9785  * @dev: The port's netdev
9786  * @ti: Tunnel endpoint information
9787  **/
9788 static void ixgbe_add_udp_tunnel_port(struct net_device *dev,
9789                                       struct udp_tunnel_info *ti)
9790 {
9791         struct ixgbe_adapter *adapter = netdev_priv(dev);
9792         struct ixgbe_hw *hw = &adapter->hw;
9793         __be16 port = ti->port;
9794         u32 port_shift = 0;
9795         u32 reg;
9796
9797         if (ti->sa_family != AF_INET)
9798                 return;
9799
9800         switch (ti->type) {
9801         case UDP_TUNNEL_TYPE_VXLAN:
9802                 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9803                         return;
9804
9805                 if (adapter->vxlan_port == port)
9806                         return;
9807
9808                 if (adapter->vxlan_port) {
9809                         netdev_info(dev,
9810                                     "VXLAN port %d set, not adding port %d\n",
9811                                     ntohs(adapter->vxlan_port),
9812                                     ntohs(port));
9813                         return;
9814                 }
9815
9816                 adapter->vxlan_port = port;
9817                 break;
9818         case UDP_TUNNEL_TYPE_GENEVE:
9819                 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9820                         return;
9821
9822                 if (adapter->geneve_port == port)
9823                         return;
9824
9825                 if (adapter->geneve_port) {
9826                         netdev_info(dev,
9827                                     "GENEVE port %d set, not adding port %d\n",
9828                                     ntohs(adapter->geneve_port),
9829                                     ntohs(port));
9830                         return;
9831                 }
9832
9833                 port_shift = IXGBE_VXLANCTRL_GENEVE_UDPPORT_SHIFT;
9834                 adapter->geneve_port = port;
9835                 break;
9836         default:
9837                 return;
9838         }
9839
9840         reg = IXGBE_READ_REG(hw, IXGBE_VXLANCTRL) | ntohs(port) << port_shift;
9841         IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, reg);
9842 }
9843
9844 /**
9845  * ixgbe_del_udp_tunnel_port - Get notifications about removing UDP tunnel ports
9846  * @dev: The port's netdev
9847  * @ti: Tunnel endpoint information
9848  **/
9849 static void ixgbe_del_udp_tunnel_port(struct net_device *dev,
9850                                       struct udp_tunnel_info *ti)
9851 {
9852         struct ixgbe_adapter *adapter = netdev_priv(dev);
9853         u32 port_mask;
9854
9855         if (ti->type != UDP_TUNNEL_TYPE_VXLAN &&
9856             ti->type != UDP_TUNNEL_TYPE_GENEVE)
9857                 return;
9858
9859         if (ti->sa_family != AF_INET)
9860                 return;
9861
9862         switch (ti->type) {
9863         case UDP_TUNNEL_TYPE_VXLAN:
9864                 if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
9865                         return;
9866
9867                 if (adapter->vxlan_port != ti->port) {
9868                         netdev_info(dev, "VXLAN port %d not found\n",
9869                                     ntohs(ti->port));
9870                         return;
9871                 }
9872
9873                 port_mask = IXGBE_VXLANCTRL_VXLAN_UDPPORT_MASK;
9874                 break;
9875         case UDP_TUNNEL_TYPE_GENEVE:
9876                 if (!(adapter->flags & IXGBE_FLAG_GENEVE_OFFLOAD_CAPABLE))
9877                         return;
9878
9879                 if (adapter->geneve_port != ti->port) {
9880                         netdev_info(dev, "GENEVE port %d not found\n",
9881                                     ntohs(ti->port));
9882                         return;
9883                 }
9884
9885                 port_mask = IXGBE_VXLANCTRL_GENEVE_UDPPORT_MASK;
9886                 break;
9887         default:
9888                 return;
9889         }
9890
9891         ixgbe_clear_udp_tunnel_port(adapter, port_mask);
9892         adapter->flags2 |= IXGBE_FLAG2_UDP_TUN_REREG_NEEDED;
9893 }
9894
9895 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
9896                              struct net_device *dev,
9897                              const unsigned char *addr, u16 vid,
9898                              u16 flags,
9899                              struct netlink_ext_ack *extack)
9900 {
9901         /* guarantee we can provide a unique filter for the unicast address */
9902         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
9903                 struct ixgbe_adapter *adapter = netdev_priv(dev);
9904                 u16 pool = VMDQ_P(0);
9905
9906                 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
9907                         return -ENOMEM;
9908         }
9909
9910         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
9911 }
9912
9913 /**
9914  * ixgbe_configure_bridge_mode - set various bridge modes
9915  * @adapter: the private structure
9916  * @mode: requested bridge mode
9917  *
9918  * Configure some settings require for various bridge modes.
9919  **/
9920 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
9921                                        __u16 mode)
9922 {
9923         struct ixgbe_hw *hw = &adapter->hw;
9924         unsigned int p, num_pools;
9925         u32 vmdctl;
9926
9927         switch (mode) {
9928         case BRIDGE_MODE_VEPA:
9929                 /* disable Tx loopback, rely on switch hairpin mode */
9930                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
9931
9932                 /* must enable Rx switching replication to allow multicast
9933                  * packet reception on all VFs, and to enable source address
9934                  * pruning.
9935                  */
9936                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9937                 vmdctl |= IXGBE_VT_CTL_REPLEN;
9938                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9939
9940                 /* enable Rx source address pruning. Note, this requires
9941                  * replication to be enabled or else it does nothing.
9942                  */
9943                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9944                 for (p = 0; p < num_pools; p++) {
9945                         if (hw->mac.ops.set_source_address_pruning)
9946                                 hw->mac.ops.set_source_address_pruning(hw,
9947                                                                        true,
9948                                                                        p);
9949                 }
9950                 break;
9951         case BRIDGE_MODE_VEB:
9952                 /* enable Tx loopback for internal VF/PF communication */
9953                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
9954                                 IXGBE_PFDTXGSWC_VT_LBEN);
9955
9956                 /* disable Rx switching replication unless we have SR-IOV
9957                  * virtual functions
9958                  */
9959                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
9960                 if (!adapter->num_vfs)
9961                         vmdctl &= ~IXGBE_VT_CTL_REPLEN;
9962                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
9963
9964                 /* disable Rx source address pruning, since we don't expect to
9965                  * be receiving external loopback of our transmitted frames.
9966                  */
9967                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
9968                 for (p = 0; p < num_pools; p++) {
9969                         if (hw->mac.ops.set_source_address_pruning)
9970                                 hw->mac.ops.set_source_address_pruning(hw,
9971                                                                        false,
9972                                                                        p);
9973                 }
9974                 break;
9975         default:
9976                 return -EINVAL;
9977         }
9978
9979         adapter->bridge_mode = mode;
9980
9981         e_info(drv, "enabling bridge mode: %s\n",
9982                mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
9983
9984         return 0;
9985 }
9986
9987 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
9988                                     struct nlmsghdr *nlh, u16 flags,
9989                                     struct netlink_ext_ack *extack)
9990 {
9991         struct ixgbe_adapter *adapter = netdev_priv(dev);
9992         struct nlattr *attr, *br_spec;
9993         int rem;
9994
9995         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
9996                 return -EOPNOTSUPP;
9997
9998         br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
9999         if (!br_spec)
10000                 return -EINVAL;
10001
10002         nla_for_each_nested(attr, br_spec, rem) {
10003                 int status;
10004                 __u16 mode;
10005
10006                 if (nla_type(attr) != IFLA_BRIDGE_MODE)
10007                         continue;
10008
10009                 if (nla_len(attr) < sizeof(mode))
10010                         return -EINVAL;
10011
10012                 mode = nla_get_u16(attr);
10013                 status = ixgbe_configure_bridge_mode(adapter, mode);
10014                 if (status)
10015                         return status;
10016
10017                 break;
10018         }
10019
10020         return 0;
10021 }
10022
10023 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
10024                                     struct net_device *dev,
10025                                     u32 filter_mask, int nlflags)
10026 {
10027         struct ixgbe_adapter *adapter = netdev_priv(dev);
10028
10029         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
10030                 return 0;
10031
10032         return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
10033                                        adapter->bridge_mode, 0, 0, nlflags,
10034                                        filter_mask, NULL);
10035 }
10036
10037 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
10038 {
10039         struct ixgbe_adapter *adapter = netdev_priv(pdev);
10040         struct ixgbe_fwd_adapter *accel;
10041         int tcs = adapter->hw_tcs ? : 1;
10042         int pool, err;
10043
10044         if (adapter->xdp_prog) {
10045                 e_warn(probe, "L2FW offload is not supported with XDP\n");
10046                 return ERR_PTR(-EINVAL);
10047         }
10048
10049         /* The hardware supported by ixgbe only filters on the destination MAC
10050          * address. In order to avoid issues we only support offloading modes
10051          * where the hardware can actually provide the functionality.
10052          */
10053         if (!macvlan_supports_dest_filter(vdev))
10054                 return ERR_PTR(-EMEDIUMTYPE);
10055
10056         /* We need to lock down the macvlan to be a single queue device so that
10057          * we can reuse the tc_to_txq field in the macvlan netdev to represent
10058          * the queue mapping to our netdev.
10059          */
10060         if (netif_is_multiqueue(vdev))
10061                 return ERR_PTR(-ERANGE);
10062
10063         pool = find_first_zero_bit(adapter->fwd_bitmask, adapter->num_rx_pools);
10064         if (pool == adapter->num_rx_pools) {
10065                 u16 used_pools = adapter->num_vfs + adapter->num_rx_pools;
10066                 u16 reserved_pools;
10067
10068                 if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
10069                      adapter->num_rx_pools >= (MAX_TX_QUEUES / tcs)) ||
10070                     adapter->num_rx_pools > IXGBE_MAX_MACVLANS)
10071                         return ERR_PTR(-EBUSY);
10072
10073                 /* Hardware has a limited number of available pools. Each VF,
10074                  * and the PF require a pool. Check to ensure we don't
10075                  * attempt to use more then the available number of pools.
10076                  */
10077                 if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
10078                         return ERR_PTR(-EBUSY);
10079
10080                 /* Enable VMDq flag so device will be set in VM mode */
10081                 adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED |
10082                                   IXGBE_FLAG_SRIOV_ENABLED;
10083
10084                 /* Try to reserve as many queues per pool as possible,
10085                  * we start with the configurations that support 4 queues
10086                  * per pools, followed by 2, and then by just 1 per pool.
10087                  */
10088                 if (used_pools < 32 && adapter->num_rx_pools < 16)
10089                         reserved_pools = min_t(u16,
10090                                                32 - used_pools,
10091                                                16 - adapter->num_rx_pools);
10092                 else if (adapter->num_rx_pools < 32)
10093                         reserved_pools = min_t(u16,
10094                                                64 - used_pools,
10095                                                32 - adapter->num_rx_pools);
10096                 else
10097                         reserved_pools = 64 - used_pools;
10098
10099
10100                 if (!reserved_pools)
10101                         return ERR_PTR(-EBUSY);
10102
10103                 adapter->ring_feature[RING_F_VMDQ].limit += reserved_pools;
10104
10105                 /* Force reinit of ring allocation with VMDQ enabled */
10106                 err = ixgbe_setup_tc(pdev, adapter->hw_tcs);
10107                 if (err)
10108                         return ERR_PTR(err);
10109
10110                 if (pool >= adapter->num_rx_pools)
10111                         return ERR_PTR(-ENOMEM);
10112         }
10113
10114         accel = kzalloc(sizeof(*accel), GFP_KERNEL);
10115         if (!accel)
10116                 return ERR_PTR(-ENOMEM);
10117
10118         set_bit(pool, adapter->fwd_bitmask);
10119         netdev_set_sb_channel(vdev, pool);
10120         accel->pool = pool;
10121         accel->netdev = vdev;
10122
10123         if (!netif_running(pdev))
10124                 return accel;
10125
10126         err = ixgbe_fwd_ring_up(adapter, accel);
10127         if (err)
10128                 return ERR_PTR(err);
10129
10130         return accel;
10131 }
10132
10133 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
10134 {
10135         struct ixgbe_fwd_adapter *accel = priv;
10136         struct ixgbe_adapter *adapter = netdev_priv(pdev);
10137         unsigned int rxbase = accel->rx_base_queue;
10138         unsigned int i;
10139
10140         /* delete unicast filter associated with offloaded interface */
10141         ixgbe_del_mac_filter(adapter, accel->netdev->dev_addr,
10142                              VMDQ_P(accel->pool));
10143
10144         /* Allow remaining Rx packets to get flushed out of the
10145          * Rx FIFO before we drop the netdev for the ring.
10146          */
10147         usleep_range(10000, 20000);
10148
10149         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
10150                 struct ixgbe_ring *ring = adapter->rx_ring[rxbase + i];
10151                 struct ixgbe_q_vector *qv = ring->q_vector;
10152
10153                 /* Make sure we aren't processing any packets and clear
10154                  * netdev to shut down the ring.
10155                  */
10156                 if (netif_running(adapter->netdev))
10157                         napi_synchronize(&qv->napi);
10158                 ring->netdev = NULL;
10159         }
10160
10161         /* unbind the queues and drop the subordinate channel config */
10162         netdev_unbind_sb_channel(pdev, accel->netdev);
10163         netdev_set_sb_channel(accel->netdev, 0);
10164
10165         clear_bit(accel->pool, adapter->fwd_bitmask);
10166         kfree(accel);
10167 }
10168
10169 #define IXGBE_MAX_MAC_HDR_LEN           127
10170 #define IXGBE_MAX_NETWORK_HDR_LEN       511
10171
10172 static netdev_features_t
10173 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
10174                      netdev_features_t features)
10175 {
10176         unsigned int network_hdr_len, mac_hdr_len;
10177
10178         /* Make certain the headers can be described by a context descriptor */
10179         mac_hdr_len = skb_network_header(skb) - skb->data;
10180         if (unlikely(mac_hdr_len > IXGBE_MAX_MAC_HDR_LEN))
10181                 return features & ~(NETIF_F_HW_CSUM |
10182                                     NETIF_F_SCTP_CRC |
10183                                     NETIF_F_HW_VLAN_CTAG_TX |
10184                                     NETIF_F_TSO |
10185                                     NETIF_F_TSO6);
10186
10187         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
10188         if (unlikely(network_hdr_len >  IXGBE_MAX_NETWORK_HDR_LEN))
10189                 return features & ~(NETIF_F_HW_CSUM |
10190                                     NETIF_F_SCTP_CRC |
10191                                     NETIF_F_TSO |
10192                                     NETIF_F_TSO6);
10193
10194         /* We can only support IPV4 TSO in tunnels if we can mangle the
10195          * inner IP ID field, so strip TSO if MANGLEID is not supported.
10196          * IPsec offoad sets skb->encapsulation but still can handle
10197          * the TSO, so it's the exception.
10198          */
10199         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID)) {
10200 #ifdef CONFIG_IXGBE_IPSEC
10201                 if (!secpath_exists(skb))
10202 #endif
10203                         features &= ~NETIF_F_TSO;
10204         }
10205
10206         return features;
10207 }
10208
10209 static int ixgbe_xdp_setup(struct net_device *dev, struct bpf_prog *prog)
10210 {
10211         int i, frame_size = dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
10212         struct ixgbe_adapter *adapter = netdev_priv(dev);
10213         struct bpf_prog *old_prog;
10214         bool need_reset;
10215
10216         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
10217                 return -EINVAL;
10218
10219         if (adapter->flags & IXGBE_FLAG_DCB_ENABLED)
10220                 return -EINVAL;
10221
10222         /* verify ixgbe ring attributes are sufficient for XDP */
10223         for (i = 0; i < adapter->num_rx_queues; i++) {
10224                 struct ixgbe_ring *ring = adapter->rx_ring[i];
10225
10226                 if (ring_is_rsc_enabled(ring))
10227                         return -EINVAL;
10228
10229                 if (frame_size > ixgbe_rx_bufsz(ring))
10230                         return -EINVAL;
10231         }
10232
10233         if (nr_cpu_ids > MAX_XDP_QUEUES)
10234                 return -ENOMEM;
10235
10236         old_prog = xchg(&adapter->xdp_prog, prog);
10237         need_reset = (!!prog != !!old_prog);
10238
10239         /* If transitioning XDP modes reconfigure rings */
10240         if (need_reset) {
10241                 int err = ixgbe_setup_tc(dev, adapter->hw_tcs);
10242
10243                 if (err) {
10244                         rcu_assign_pointer(adapter->xdp_prog, old_prog);
10245                         return -EINVAL;
10246                 }
10247         } else {
10248                 for (i = 0; i < adapter->num_rx_queues; i++)
10249                         (void)xchg(&adapter->rx_ring[i]->xdp_prog,
10250                             adapter->xdp_prog);
10251         }
10252
10253         if (old_prog)
10254                 bpf_prog_put(old_prog);
10255
10256         /* Kick start the NAPI context if there is an AF_XDP socket open
10257          * on that queue id. This so that receiving will start.
10258          */
10259         if (need_reset && prog)
10260                 for (i = 0; i < adapter->num_rx_queues; i++)
10261                         if (adapter->xdp_ring[i]->xsk_umem)
10262                                 (void)ixgbe_xsk_async_xmit(adapter->netdev, i);
10263
10264         return 0;
10265 }
10266
10267 static int ixgbe_xdp(struct net_device *dev, struct netdev_bpf *xdp)
10268 {
10269         struct ixgbe_adapter *adapter = netdev_priv(dev);
10270
10271         switch (xdp->command) {
10272         case XDP_SETUP_PROG:
10273                 return ixgbe_xdp_setup(dev, xdp->prog);
10274         case XDP_QUERY_PROG:
10275                 xdp->prog_id = adapter->xdp_prog ?
10276                         adapter->xdp_prog->aux->id : 0;
10277                 return 0;
10278         case XDP_SETUP_XSK_UMEM:
10279                 return ixgbe_xsk_umem_setup(adapter, xdp->xsk.umem,
10280                                             xdp->xsk.queue_id);
10281
10282         default:
10283                 return -EINVAL;
10284         }
10285 }
10286
10287 void ixgbe_xdp_ring_update_tail(struct ixgbe_ring *ring)
10288 {
10289         /* Force memory writes to complete before letting h/w know there
10290          * are new descriptors to fetch.
10291          */
10292         wmb();
10293         writel(ring->next_to_use, ring->tail);
10294 }
10295
10296 static int ixgbe_xdp_xmit(struct net_device *dev, int n,
10297                           struct xdp_frame **frames, u32 flags)
10298 {
10299         struct ixgbe_adapter *adapter = netdev_priv(dev);
10300         struct ixgbe_ring *ring;
10301         int drops = 0;
10302         int i;
10303
10304         if (unlikely(test_bit(__IXGBE_DOWN, &adapter->state)))
10305                 return -ENETDOWN;
10306
10307         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
10308                 return -EINVAL;
10309
10310         /* During program transitions its possible adapter->xdp_prog is assigned
10311          * but ring has not been configured yet. In this case simply abort xmit.
10312          */
10313         ring = adapter->xdp_prog ? adapter->xdp_ring[smp_processor_id()] : NULL;
10314         if (unlikely(!ring))
10315                 return -ENXIO;
10316
10317         if (unlikely(test_bit(__IXGBE_TX_DISABLED, &ring->state)))
10318                 return -ENXIO;
10319
10320         for (i = 0; i < n; i++) {
10321                 struct xdp_frame *xdpf = frames[i];
10322                 int err;
10323
10324                 err = ixgbe_xmit_xdp_ring(adapter, xdpf);
10325                 if (err != IXGBE_XDP_TX) {
10326                         xdp_return_frame_rx_napi(xdpf);
10327                         drops++;
10328                 }
10329         }
10330
10331         if (unlikely(flags & XDP_XMIT_FLUSH))
10332                 ixgbe_xdp_ring_update_tail(ring);
10333
10334         return n - drops;
10335 }
10336
10337 static const struct net_device_ops ixgbe_netdev_ops = {
10338         .ndo_open               = ixgbe_open,
10339         .ndo_stop               = ixgbe_close,
10340         .ndo_start_xmit         = ixgbe_xmit_frame,
10341         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
10342         .ndo_validate_addr      = eth_validate_addr,
10343         .ndo_set_mac_address    = ixgbe_set_mac,
10344         .ndo_change_mtu         = ixgbe_change_mtu,
10345         .ndo_tx_timeout         = ixgbe_tx_timeout,
10346         .ndo_set_tx_maxrate     = ixgbe_tx_maxrate,
10347         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
10348         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
10349         .ndo_do_ioctl           = ixgbe_ioctl,
10350         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
10351         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
10352         .ndo_set_vf_rate        = ixgbe_ndo_set_vf_bw,
10353         .ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
10354         .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
10355         .ndo_set_vf_trust       = ixgbe_ndo_set_vf_trust,
10356         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
10357         .ndo_get_stats64        = ixgbe_get_stats64,
10358         .ndo_setup_tc           = __ixgbe_setup_tc,
10359 #ifdef IXGBE_FCOE
10360         .ndo_select_queue       = ixgbe_select_queue,
10361         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
10362         .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
10363         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
10364         .ndo_fcoe_enable = ixgbe_fcoe_enable,
10365         .ndo_fcoe_disable = ixgbe_fcoe_disable,
10366         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
10367         .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
10368 #endif /* IXGBE_FCOE */
10369         .ndo_set_features = ixgbe_set_features,
10370         .ndo_fix_features = ixgbe_fix_features,
10371         .ndo_fdb_add            = ixgbe_ndo_fdb_add,
10372         .ndo_bridge_setlink     = ixgbe_ndo_bridge_setlink,
10373         .ndo_bridge_getlink     = ixgbe_ndo_bridge_getlink,
10374         .ndo_dfwd_add_station   = ixgbe_fwd_add,
10375         .ndo_dfwd_del_station   = ixgbe_fwd_del,
10376         .ndo_udp_tunnel_add     = ixgbe_add_udp_tunnel_port,
10377         .ndo_udp_tunnel_del     = ixgbe_del_udp_tunnel_port,
10378         .ndo_features_check     = ixgbe_features_check,
10379         .ndo_bpf                = ixgbe_xdp,
10380         .ndo_xdp_xmit           = ixgbe_xdp_xmit,
10381         .ndo_xsk_async_xmit     = ixgbe_xsk_async_xmit,
10382 };
10383
10384 static void ixgbe_disable_txr_hw(struct ixgbe_adapter *adapter,
10385                                  struct ixgbe_ring *tx_ring)
10386 {
10387         unsigned long wait_delay, delay_interval;
10388         struct ixgbe_hw *hw = &adapter->hw;
10389         u8 reg_idx = tx_ring->reg_idx;
10390         int wait_loop;
10391         u32 txdctl;
10392
10393         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
10394
10395         /* delay mechanism from ixgbe_disable_tx */
10396         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10397
10398         wait_loop = IXGBE_MAX_RX_DESC_POLL;
10399         wait_delay = delay_interval;
10400
10401         while (wait_loop--) {
10402                 usleep_range(wait_delay, wait_delay + 10);
10403                 wait_delay += delay_interval * 2;
10404                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
10405
10406                 if (!(txdctl & IXGBE_TXDCTL_ENABLE))
10407                         return;
10408         }
10409
10410         e_err(drv, "TXDCTL.ENABLE not cleared within the polling period\n");
10411 }
10412
10413 static void ixgbe_disable_txr(struct ixgbe_adapter *adapter,
10414                               struct ixgbe_ring *tx_ring)
10415 {
10416         set_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10417         ixgbe_disable_txr_hw(adapter, tx_ring);
10418 }
10419
10420 static void ixgbe_disable_rxr_hw(struct ixgbe_adapter *adapter,
10421                                  struct ixgbe_ring *rx_ring)
10422 {
10423         unsigned long wait_delay, delay_interval;
10424         struct ixgbe_hw *hw = &adapter->hw;
10425         u8 reg_idx = rx_ring->reg_idx;
10426         int wait_loop;
10427         u32 rxdctl;
10428
10429         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10430         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
10431         rxdctl |= IXGBE_RXDCTL_SWFLSH;
10432
10433         /* write value back with RXDCTL.ENABLE bit cleared */
10434         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
10435
10436         /* RXDCTL.EN may not change on 82598 if link is down, so skip it */
10437         if (hw->mac.type == ixgbe_mac_82598EB &&
10438             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
10439                 return;
10440
10441         /* delay mechanism from ixgbe_disable_rx */
10442         delay_interval = ixgbe_get_completion_timeout(adapter) / 100;
10443
10444         wait_loop = IXGBE_MAX_RX_DESC_POLL;
10445         wait_delay = delay_interval;
10446
10447         while (wait_loop--) {
10448                 usleep_range(wait_delay, wait_delay + 10);
10449                 wait_delay += delay_interval * 2;
10450                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
10451
10452                 if (!(rxdctl & IXGBE_RXDCTL_ENABLE))
10453                         return;
10454         }
10455
10456         e_err(drv, "RXDCTL.ENABLE not cleared within the polling period\n");
10457 }
10458
10459 static void ixgbe_reset_txr_stats(struct ixgbe_ring *tx_ring)
10460 {
10461         memset(&tx_ring->stats, 0, sizeof(tx_ring->stats));
10462         memset(&tx_ring->tx_stats, 0, sizeof(tx_ring->tx_stats));
10463 }
10464
10465 static void ixgbe_reset_rxr_stats(struct ixgbe_ring *rx_ring)
10466 {
10467         memset(&rx_ring->stats, 0, sizeof(rx_ring->stats));
10468         memset(&rx_ring->rx_stats, 0, sizeof(rx_ring->rx_stats));
10469 }
10470
10471 /**
10472  * ixgbe_txrx_ring_disable - Disable Rx/Tx/XDP Tx rings
10473  * @adapter: adapter structure
10474  * @ring: ring index
10475  *
10476  * This function disables a certain Rx/Tx/XDP Tx ring. The function
10477  * assumes that the netdev is running.
10478  **/
10479 void ixgbe_txrx_ring_disable(struct ixgbe_adapter *adapter, int ring)
10480 {
10481         struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10482
10483         rx_ring = adapter->rx_ring[ring];
10484         tx_ring = adapter->tx_ring[ring];
10485         xdp_ring = adapter->xdp_ring[ring];
10486
10487         ixgbe_disable_txr(adapter, tx_ring);
10488         if (xdp_ring)
10489                 ixgbe_disable_txr(adapter, xdp_ring);
10490         ixgbe_disable_rxr_hw(adapter, rx_ring);
10491
10492         if (xdp_ring)
10493                 synchronize_rcu();
10494
10495         /* Rx/Tx/XDP Tx share the same napi context. */
10496         napi_disable(&rx_ring->q_vector->napi);
10497
10498         ixgbe_clean_tx_ring(tx_ring);
10499         if (xdp_ring)
10500                 ixgbe_clean_tx_ring(xdp_ring);
10501         ixgbe_clean_rx_ring(rx_ring);
10502
10503         ixgbe_reset_txr_stats(tx_ring);
10504         if (xdp_ring)
10505                 ixgbe_reset_txr_stats(xdp_ring);
10506         ixgbe_reset_rxr_stats(rx_ring);
10507 }
10508
10509 /**
10510  * ixgbe_txrx_ring_enable - Enable Rx/Tx/XDP Tx rings
10511  * @adapter: adapter structure
10512  * @ring: ring index
10513  *
10514  * This function enables a certain Rx/Tx/XDP Tx ring. The function
10515  * assumes that the netdev is running.
10516  **/
10517 void ixgbe_txrx_ring_enable(struct ixgbe_adapter *adapter, int ring)
10518 {
10519         struct ixgbe_ring *rx_ring, *tx_ring, *xdp_ring;
10520
10521         rx_ring = adapter->rx_ring[ring];
10522         tx_ring = adapter->tx_ring[ring];
10523         xdp_ring = adapter->xdp_ring[ring];
10524
10525         /* Rx/Tx/XDP Tx share the same napi context. */
10526         napi_enable(&rx_ring->q_vector->napi);
10527
10528         ixgbe_configure_tx_ring(adapter, tx_ring);
10529         if (xdp_ring)
10530                 ixgbe_configure_tx_ring(adapter, xdp_ring);
10531         ixgbe_configure_rx_ring(adapter, rx_ring);
10532
10533         clear_bit(__IXGBE_TX_DISABLED, &tx_ring->state);
10534         if (xdp_ring)
10535                 clear_bit(__IXGBE_TX_DISABLED, &xdp_ring->state);
10536 }
10537
10538 /**
10539  * ixgbe_enumerate_functions - Get the number of ports this device has
10540  * @adapter: adapter structure
10541  *
10542  * This function enumerates the phsyical functions co-located on a single slot,
10543  * in order to determine how many ports a device has. This is most useful in
10544  * determining the required GT/s of PCIe bandwidth necessary for optimal
10545  * performance.
10546  **/
10547 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
10548 {
10549         struct pci_dev *entry, *pdev = adapter->pdev;
10550         int physfns = 0;
10551
10552         /* Some cards can not use the generic count PCIe functions method,
10553          * because they are behind a parent switch, so we hardcode these with
10554          * the correct number of functions.
10555          */
10556         if (ixgbe_pcie_from_parent(&adapter->hw))
10557                 physfns = 4;
10558
10559         list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
10560                 /* don't count virtual functions */
10561                 if (entry->is_virtfn)
10562                         continue;
10563
10564                 /* When the devices on the bus don't all match our device ID,
10565                  * we can't reliably determine the correct number of
10566                  * functions. This can occur if a function has been direct
10567                  * attached to a virtual machine using VT-d, for example. In
10568                  * this case, simply return -1 to indicate this.
10569                  */
10570                 if ((entry->vendor != pdev->vendor) ||
10571                     (entry->device != pdev->device))
10572                         return -1;
10573
10574                 physfns++;
10575         }
10576
10577         return physfns;
10578 }
10579
10580 /**
10581  * ixgbe_wol_supported - Check whether device supports WoL
10582  * @adapter: the adapter private structure
10583  * @device_id: the device ID
10584  * @subdevice_id: the subsystem device ID
10585  *
10586  * This function is used by probe and ethtool to determine
10587  * which devices have WoL support
10588  *
10589  **/
10590 bool ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
10591                          u16 subdevice_id)
10592 {
10593         struct ixgbe_hw *hw = &adapter->hw;
10594         u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
10595
10596         /* WOL not supported on 82598 */
10597         if (hw->mac.type == ixgbe_mac_82598EB)
10598                 return false;
10599
10600         /* check eeprom to see if WOL is enabled for X540 and newer */
10601         if (hw->mac.type >= ixgbe_mac_X540) {
10602                 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
10603                     ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
10604                      (hw->bus.func == 0)))
10605                         return true;
10606         }
10607
10608         /* WOL is determined based on device IDs for 82599 MACs */
10609         switch (device_id) {
10610         case IXGBE_DEV_ID_82599_SFP:
10611                 /* Only these subdevices could supports WOL */
10612                 switch (subdevice_id) {
10613                 case IXGBE_SUBDEV_ID_82599_560FLR:
10614                 case IXGBE_SUBDEV_ID_82599_LOM_SNAP6:
10615                 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
10616                 case IXGBE_SUBDEV_ID_82599_SFP_2OCP:
10617                         /* only support first port */
10618                         if (hw->bus.func != 0)
10619                                 break;
10620                         /* fall through */
10621                 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
10622                 case IXGBE_SUBDEV_ID_82599_SFP:
10623                 case IXGBE_SUBDEV_ID_82599_RNDC:
10624                 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
10625                 case IXGBE_SUBDEV_ID_82599_SFP_1OCP:
10626                 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM1:
10627                 case IXGBE_SUBDEV_ID_82599_SFP_LOM_OEM2:
10628                         return true;
10629                 }
10630                 break;
10631         case IXGBE_DEV_ID_82599EN_SFP:
10632                 /* Only these subdevices support WOL */
10633                 switch (subdevice_id) {
10634                 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
10635                         return true;
10636                 }
10637                 break;
10638         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
10639                 /* All except this subdevice support WOL */
10640                 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
10641                         return true;
10642                 break;
10643         case IXGBE_DEV_ID_82599_KX4:
10644                 return  true;
10645         default:
10646                 break;
10647         }
10648
10649         return false;
10650 }
10651
10652 /**
10653  * ixgbe_set_fw_version - Set FW version
10654  * @adapter: the adapter private structure
10655  *
10656  * This function is used by probe and ethtool to determine the FW version to
10657  * format to display. The FW version is taken from the EEPROM/NVM.
10658  */
10659 static void ixgbe_set_fw_version(struct ixgbe_adapter *adapter)
10660 {
10661         struct ixgbe_hw *hw = &adapter->hw;
10662         struct ixgbe_nvm_version nvm_ver;
10663
10664         ixgbe_get_oem_prod_version(hw, &nvm_ver);
10665         if (nvm_ver.oem_valid) {
10666                 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10667                          "%x.%x.%x", nvm_ver.oem_major, nvm_ver.oem_minor,
10668                          nvm_ver.oem_release);
10669                 return;
10670         }
10671
10672         ixgbe_get_etk_id(hw, &nvm_ver);
10673         ixgbe_get_orom_version(hw, &nvm_ver);
10674
10675         if (nvm_ver.or_valid) {
10676                 snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10677                          "0x%08x, %d.%d.%d", nvm_ver.etk_id, nvm_ver.or_major,
10678                          nvm_ver.or_build, nvm_ver.or_patch);
10679                 return;
10680         }
10681
10682         /* Set ETrack ID format */
10683         snprintf(adapter->eeprom_id, sizeof(adapter->eeprom_id),
10684                  "0x%08x", nvm_ver.etk_id);
10685 }
10686
10687 /**
10688  * ixgbe_probe - Device Initialization Routine
10689  * @pdev: PCI device information struct
10690  * @ent: entry in ixgbe_pci_tbl
10691  *
10692  * Returns 0 on success, negative on failure
10693  *
10694  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
10695  * The OS initialization, configuring of the adapter private structure,
10696  * and a hardware reset occur.
10697  **/
10698 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
10699 {
10700         struct net_device *netdev;
10701         struct ixgbe_adapter *adapter = NULL;
10702         struct ixgbe_hw *hw;
10703         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
10704         int i, err, pci_using_dac, expected_gts;
10705         unsigned int indices = MAX_TX_QUEUES;
10706         u8 part_str[IXGBE_PBANUM_LENGTH];
10707         bool disable_dev = false;
10708 #ifdef IXGBE_FCOE
10709         u16 device_caps;
10710 #endif
10711         u32 eec;
10712
10713         /* Catch broken hardware that put the wrong VF device ID in
10714          * the PCIe SR-IOV capability.
10715          */
10716         if (pdev->is_virtfn) {
10717                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
10718                      pci_name(pdev), pdev->vendor, pdev->device);
10719                 return -EINVAL;
10720         }
10721
10722         err = pci_enable_device_mem(pdev);
10723         if (err)
10724                 return err;
10725
10726         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
10727                 pci_using_dac = 1;
10728         } else {
10729                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
10730                 if (err) {
10731                         dev_err(&pdev->dev,
10732                                 "No usable DMA configuration, aborting\n");
10733                         goto err_dma;
10734                 }
10735                 pci_using_dac = 0;
10736         }
10737
10738         err = pci_request_mem_regions(pdev, ixgbe_driver_name);
10739         if (err) {
10740                 dev_err(&pdev->dev,
10741                         "pci_request_selected_regions failed 0x%x\n", err);
10742                 goto err_pci_reg;
10743         }
10744
10745         pci_enable_pcie_error_reporting(pdev);
10746
10747         pci_set_master(pdev);
10748         pci_save_state(pdev);
10749
10750         if (ii->mac == ixgbe_mac_82598EB) {
10751 #ifdef CONFIG_IXGBE_DCB
10752                 /* 8 TC w/ 4 queues per TC */
10753                 indices = 4 * MAX_TRAFFIC_CLASS;
10754 #else
10755                 indices = IXGBE_MAX_RSS_INDICES;
10756 #endif
10757         }
10758
10759         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
10760         if (!netdev) {
10761                 err = -ENOMEM;
10762                 goto err_alloc_etherdev;
10763         }
10764
10765         SET_NETDEV_DEV(netdev, &pdev->dev);
10766
10767         adapter = netdev_priv(netdev);
10768
10769         adapter->netdev = netdev;
10770         adapter->pdev = pdev;
10771         hw = &adapter->hw;
10772         hw->back = adapter;
10773         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
10774
10775         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
10776                               pci_resource_len(pdev, 0));
10777         adapter->io_addr = hw->hw_addr;
10778         if (!hw->hw_addr) {
10779                 err = -EIO;
10780                 goto err_ioremap;
10781         }
10782
10783         netdev->netdev_ops = &ixgbe_netdev_ops;
10784         ixgbe_set_ethtool_ops(netdev);
10785         netdev->watchdog_timeo = 5 * HZ;
10786         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
10787
10788         /* Setup hw api */
10789         hw->mac.ops   = *ii->mac_ops;
10790         hw->mac.type  = ii->mac;
10791         hw->mvals     = ii->mvals;
10792         if (ii->link_ops)
10793                 hw->link.ops  = *ii->link_ops;
10794
10795         /* EEPROM */
10796         hw->eeprom.ops = *ii->eeprom_ops;
10797         eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
10798         if (ixgbe_removed(hw->hw_addr)) {
10799                 err = -EIO;
10800                 goto err_ioremap;
10801         }
10802         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
10803         if (!(eec & BIT(8)))
10804                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
10805
10806         /* PHY */
10807         hw->phy.ops = *ii->phy_ops;
10808         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
10809         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
10810         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
10811         hw->phy.mdio.mmds = 0;
10812         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
10813         hw->phy.mdio.dev = netdev;
10814         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
10815         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
10816
10817         /* setup the private structure */
10818         err = ixgbe_sw_init(adapter, ii);
10819         if (err)
10820                 goto err_sw_init;
10821
10822         /* Make sure the SWFW semaphore is in a valid state */
10823         if (hw->mac.ops.init_swfw_sync)
10824                 hw->mac.ops.init_swfw_sync(hw);
10825
10826         /* Make it possible the adapter to be woken up via WOL */
10827         switch (adapter->hw.mac.type) {
10828         case ixgbe_mac_82599EB:
10829         case ixgbe_mac_X540:
10830         case ixgbe_mac_X550:
10831         case ixgbe_mac_X550EM_x:
10832         case ixgbe_mac_x550em_a:
10833                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
10834                 break;
10835         default:
10836                 break;
10837         }
10838
10839         /*
10840          * If there is a fan on this device and it has failed log the
10841          * failure.
10842          */
10843         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
10844                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
10845                 if (esdp & IXGBE_ESDP_SDP1)
10846                         e_crit(probe, "Fan has stopped, replace the adapter\n");
10847         }
10848
10849         if (allow_unsupported_sfp)
10850                 hw->allow_unsupported_sfp = allow_unsupported_sfp;
10851
10852         /* reset_hw fills in the perm_addr as well */
10853         hw->phy.reset_if_overtemp = true;
10854         err = hw->mac.ops.reset_hw(hw);
10855         hw->phy.reset_if_overtemp = false;
10856         ixgbe_set_eee_capable(adapter);
10857         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
10858                 err = 0;
10859         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
10860                 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
10861                 e_dev_err("Reload the driver after installing a supported module.\n");
10862                 goto err_sw_init;
10863         } else if (err) {
10864                 e_dev_err("HW Init failed: %d\n", err);
10865                 goto err_sw_init;
10866         }
10867
10868 #ifdef CONFIG_PCI_IOV
10869         /* SR-IOV not supported on the 82598 */
10870         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
10871                 goto skip_sriov;
10872         /* Mailbox */
10873         ixgbe_init_mbx_params_pf(hw);
10874         hw->mbx.ops = ii->mbx_ops;
10875         pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
10876         ixgbe_enable_sriov(adapter, max_vfs);
10877 skip_sriov:
10878
10879 #endif
10880         netdev->features = NETIF_F_SG |
10881                            NETIF_F_TSO |
10882                            NETIF_F_TSO6 |
10883                            NETIF_F_RXHASH |
10884                            NETIF_F_RXCSUM |
10885                            NETIF_F_HW_CSUM;
10886
10887 #define IXGBE_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
10888                                     NETIF_F_GSO_GRE_CSUM | \
10889                                     NETIF_F_GSO_IPXIP4 | \
10890                                     NETIF_F_GSO_IPXIP6 | \
10891                                     NETIF_F_GSO_UDP_TUNNEL | \
10892                                     NETIF_F_GSO_UDP_TUNNEL_CSUM)
10893
10894         netdev->gso_partial_features = IXGBE_GSO_PARTIAL_FEATURES;
10895         netdev->features |= NETIF_F_GSO_PARTIAL |
10896                             IXGBE_GSO_PARTIAL_FEATURES;
10897
10898         if (hw->mac.type >= ixgbe_mac_82599EB)
10899                 netdev->features |= NETIF_F_SCTP_CRC;
10900
10901 #ifdef CONFIG_IXGBE_IPSEC
10902 #define IXGBE_ESP_FEATURES      (NETIF_F_HW_ESP | \
10903                                  NETIF_F_HW_ESP_TX_CSUM | \
10904                                  NETIF_F_GSO_ESP)
10905
10906         if (adapter->ipsec)
10907                 netdev->features |= IXGBE_ESP_FEATURES;
10908 #endif
10909         /* copy netdev features into list of user selectable features */
10910         netdev->hw_features |= netdev->features |
10911                                NETIF_F_HW_VLAN_CTAG_FILTER |
10912                                NETIF_F_HW_VLAN_CTAG_RX |
10913                                NETIF_F_HW_VLAN_CTAG_TX |
10914                                NETIF_F_RXALL |
10915                                NETIF_F_HW_L2FW_DOFFLOAD;
10916
10917         if (hw->mac.type >= ixgbe_mac_82599EB)
10918                 netdev->hw_features |= NETIF_F_NTUPLE |
10919                                        NETIF_F_HW_TC;
10920
10921         if (pci_using_dac)
10922                 netdev->features |= NETIF_F_HIGHDMA;
10923
10924         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
10925         netdev->hw_enc_features |= netdev->vlan_features;
10926         netdev->mpls_features |= NETIF_F_SG |
10927                                  NETIF_F_TSO |
10928                                  NETIF_F_TSO6 |
10929                                  NETIF_F_HW_CSUM;
10930         netdev->mpls_features |= IXGBE_GSO_PARTIAL_FEATURES;
10931
10932         /* set this bit last since it cannot be part of vlan_features */
10933         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
10934                             NETIF_F_HW_VLAN_CTAG_RX |
10935                             NETIF_F_HW_VLAN_CTAG_TX;
10936
10937         netdev->priv_flags |= IFF_UNICAST_FLT;
10938         netdev->priv_flags |= IFF_SUPP_NOFCS;
10939
10940         /* MTU range: 68 - 9710 */
10941         netdev->min_mtu = ETH_MIN_MTU;
10942         netdev->max_mtu = IXGBE_MAX_JUMBO_FRAME_SIZE - (ETH_HLEN + ETH_FCS_LEN);
10943
10944 #ifdef CONFIG_IXGBE_DCB
10945         if (adapter->flags & IXGBE_FLAG_DCB_CAPABLE)
10946                 netdev->dcbnl_ops = &ixgbe_dcbnl_ops;
10947 #endif
10948
10949 #ifdef IXGBE_FCOE
10950         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
10951                 unsigned int fcoe_l;
10952
10953                 if (hw->mac.ops.get_device_caps) {
10954                         hw->mac.ops.get_device_caps(hw, &device_caps);
10955                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
10956                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
10957                 }
10958
10959
10960                 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
10961                 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
10962
10963                 netdev->features |= NETIF_F_FSO |
10964                                     NETIF_F_FCOE_CRC;
10965
10966                 netdev->vlan_features |= NETIF_F_FSO |
10967                                          NETIF_F_FCOE_CRC |
10968                                          NETIF_F_FCOE_MTU;
10969         }
10970 #endif /* IXGBE_FCOE */
10971         if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
10972                 netdev->hw_features |= NETIF_F_LRO;
10973         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
10974                 netdev->features |= NETIF_F_LRO;
10975
10976         if (ixgbe_check_fw_error(adapter)) {
10977                 err = -EIO;
10978                 goto err_sw_init;
10979         }
10980
10981         /* make sure the EEPROM is good */
10982         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
10983                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
10984                 err = -EIO;
10985                 goto err_sw_init;
10986         }
10987
10988         eth_platform_get_mac_address(&adapter->pdev->dev,
10989                                      adapter->hw.mac.perm_addr);
10990
10991         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
10992
10993         if (!is_valid_ether_addr(netdev->dev_addr)) {
10994                 e_dev_err("invalid MAC address\n");
10995                 err = -EIO;
10996                 goto err_sw_init;
10997         }
10998
10999         /* Set hw->mac.addr to permanent MAC address */
11000         ether_addr_copy(hw->mac.addr, hw->mac.perm_addr);
11001         ixgbe_mac_set_default_filter(adapter);
11002
11003         timer_setup(&adapter->service_timer, ixgbe_service_timer, 0);
11004
11005         if (ixgbe_removed(hw->hw_addr)) {
11006                 err = -EIO;
11007                 goto err_sw_init;
11008         }
11009         INIT_WORK(&adapter->service_task, ixgbe_service_task);
11010         set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
11011         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
11012
11013         err = ixgbe_init_interrupt_scheme(adapter);
11014         if (err)
11015                 goto err_sw_init;
11016
11017         for (i = 0; i < adapter->num_rx_queues; i++)
11018                 u64_stats_init(&adapter->rx_ring[i]->syncp);
11019         for (i = 0; i < adapter->num_tx_queues; i++)
11020                 u64_stats_init(&adapter->tx_ring[i]->syncp);
11021         for (i = 0; i < adapter->num_xdp_queues; i++)
11022                 u64_stats_init(&adapter->xdp_ring[i]->syncp);
11023
11024         /* WOL not supported for all devices */
11025         adapter->wol = 0;
11026         hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
11027         hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
11028                                                 pdev->subsystem_device);
11029         if (hw->wol_enabled)
11030                 adapter->wol = IXGBE_WUFC_MAG;
11031
11032         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
11033
11034         /* save off EEPROM version number */
11035         ixgbe_set_fw_version(adapter);
11036
11037         /* pick up the PCI bus settings for reporting later */
11038         if (ixgbe_pcie_from_parent(hw))
11039                 ixgbe_get_parent_bus_info(adapter);
11040         else
11041                  hw->mac.ops.get_bus_info(hw);
11042
11043         /* calculate the expected PCIe bandwidth required for optimal
11044          * performance. Note that some older parts will never have enough
11045          * bandwidth due to being older generation PCIe parts. We clamp these
11046          * parts to ensure no warning is displayed if it can't be fixed.
11047          */
11048         switch (hw->mac.type) {
11049         case ixgbe_mac_82598EB:
11050                 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
11051                 break;
11052         default:
11053                 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
11054                 break;
11055         }
11056
11057         /* don't check link if we failed to enumerate functions */
11058         if (expected_gts > 0)
11059                 ixgbe_check_minimum_link(adapter, expected_gts);
11060
11061         err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
11062         if (err)
11063                 strlcpy(part_str, "Unknown", sizeof(part_str));
11064         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
11065                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
11066                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
11067                            part_str);
11068         else
11069                 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
11070                            hw->mac.type, hw->phy.type, part_str);
11071
11072         e_dev_info("%pM\n", netdev->dev_addr);
11073
11074         /* reset the hardware with the new settings */
11075         err = hw->mac.ops.start_hw(hw);
11076         if (err == IXGBE_ERR_EEPROM_VERSION) {
11077                 /* We are running on a pre-production device, log a warning */
11078                 e_dev_warn("This device is a pre-production adapter/LOM. "
11079                            "Please be aware there may be issues associated "
11080                            "with your hardware.  If you are experiencing "
11081                            "problems please contact your Intel or hardware "
11082                            "representative who provided you with this "
11083                            "hardware.\n");
11084         }
11085         strcpy(netdev->name, "eth%d");
11086         pci_set_drvdata(pdev, adapter);
11087         err = register_netdev(netdev);
11088         if (err)
11089                 goto err_register;
11090
11091
11092         /* power down the optics for 82599 SFP+ fiber */
11093         if (hw->mac.ops.disable_tx_laser)
11094                 hw->mac.ops.disable_tx_laser(hw);
11095
11096         /* carrier off reporting is important to ethtool even BEFORE open */
11097         netif_carrier_off(netdev);
11098
11099 #ifdef CONFIG_IXGBE_DCA
11100         if (dca_add_requester(&pdev->dev) == 0) {
11101                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
11102                 ixgbe_setup_dca(adapter);
11103         }
11104 #endif
11105         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
11106                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
11107                 for (i = 0; i < adapter->num_vfs; i++)
11108                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
11109         }
11110
11111         /* firmware requires driver version to be 0xFFFFFFFF
11112          * since os does not support feature
11113          */
11114         if (hw->mac.ops.set_fw_drv_ver)
11115                 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF, 0xFF,
11116                                            sizeof(ixgbe_driver_version) - 1,
11117                                            ixgbe_driver_version);
11118
11119         /* add san mac addr to netdev */
11120         ixgbe_add_sanmac_netdev(netdev);
11121
11122         e_dev_info("%s\n", ixgbe_default_device_descr);
11123
11124 #ifdef CONFIG_IXGBE_HWMON
11125         if (ixgbe_sysfs_init(adapter))
11126                 e_err(probe, "failed to allocate sysfs resources\n");
11127 #endif /* CONFIG_IXGBE_HWMON */
11128
11129         ixgbe_dbg_adapter_init(adapter);
11130
11131         /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
11132         if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
11133                 hw->mac.ops.setup_link(hw,
11134                         IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
11135                         true);
11136
11137         ixgbe_mii_bus_init(hw);
11138
11139         return 0;
11140
11141 err_register:
11142         ixgbe_release_hw_control(adapter);
11143         ixgbe_clear_interrupt_scheme(adapter);
11144 err_sw_init:
11145         ixgbe_disable_sriov(adapter);
11146         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
11147         iounmap(adapter->io_addr);
11148         kfree(adapter->jump_tables[0]);
11149         kfree(adapter->mac_table);
11150         kfree(adapter->rss_key);
11151         bitmap_free(adapter->af_xdp_zc_qps);
11152 err_ioremap:
11153         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11154         free_netdev(netdev);
11155 err_alloc_etherdev:
11156         pci_release_mem_regions(pdev);
11157 err_pci_reg:
11158 err_dma:
11159         if (!adapter || disable_dev)
11160                 pci_disable_device(pdev);
11161         return err;
11162 }
11163
11164 /**
11165  * ixgbe_remove - Device Removal Routine
11166  * @pdev: PCI device information struct
11167  *
11168  * ixgbe_remove is called by the PCI subsystem to alert the driver
11169  * that it should release a PCI device.  The could be caused by a
11170  * Hot-Plug event, or because the driver is going to be removed from
11171  * memory.
11172  **/
11173 static void ixgbe_remove(struct pci_dev *pdev)
11174 {
11175         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11176         struct net_device *netdev;
11177         bool disable_dev;
11178         int i;
11179
11180         /* if !adapter then we already cleaned up in probe */
11181         if (!adapter)
11182                 return;
11183
11184         netdev  = adapter->netdev;
11185         ixgbe_dbg_adapter_exit(adapter);
11186
11187         set_bit(__IXGBE_REMOVING, &adapter->state);
11188         cancel_work_sync(&adapter->service_task);
11189
11190         if (adapter->mii_bus)
11191                 mdiobus_unregister(adapter->mii_bus);
11192
11193 #ifdef CONFIG_IXGBE_DCA
11194         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
11195                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
11196                 dca_remove_requester(&pdev->dev);
11197                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
11198                                 IXGBE_DCA_CTRL_DCA_DISABLE);
11199         }
11200
11201 #endif
11202 #ifdef CONFIG_IXGBE_HWMON
11203         ixgbe_sysfs_exit(adapter);
11204 #endif /* CONFIG_IXGBE_HWMON */
11205
11206         /* remove the added san mac */
11207         ixgbe_del_sanmac_netdev(netdev);
11208
11209 #ifdef CONFIG_PCI_IOV
11210         ixgbe_disable_sriov(adapter);
11211 #endif
11212         if (netdev->reg_state == NETREG_REGISTERED)
11213                 unregister_netdev(netdev);
11214
11215         ixgbe_stop_ipsec_offload(adapter);
11216         ixgbe_clear_interrupt_scheme(adapter);
11217
11218         ixgbe_release_hw_control(adapter);
11219
11220 #ifdef CONFIG_DCB
11221         kfree(adapter->ixgbe_ieee_pfc);
11222         kfree(adapter->ixgbe_ieee_ets);
11223
11224 #endif
11225         iounmap(adapter->io_addr);
11226         pci_release_mem_regions(pdev);
11227
11228         e_dev_info("complete\n");
11229
11230         for (i = 0; i < IXGBE_MAX_LINK_HANDLE; i++) {
11231                 if (adapter->jump_tables[i]) {
11232                         kfree(adapter->jump_tables[i]->input);
11233                         kfree(adapter->jump_tables[i]->mask);
11234                 }
11235                 kfree(adapter->jump_tables[i]);
11236         }
11237
11238         kfree(adapter->mac_table);
11239         kfree(adapter->rss_key);
11240         bitmap_free(adapter->af_xdp_zc_qps);
11241         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
11242         free_netdev(netdev);
11243
11244         pci_disable_pcie_error_reporting(pdev);
11245
11246         if (disable_dev)
11247                 pci_disable_device(pdev);
11248 }
11249
11250 /**
11251  * ixgbe_io_error_detected - called when PCI error is detected
11252  * @pdev: Pointer to PCI device
11253  * @state: The current pci connection state
11254  *
11255  * This function is called after a PCI bus error affecting
11256  * this device has been detected.
11257  */
11258 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
11259                                                 pci_channel_state_t state)
11260 {
11261         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11262         struct net_device *netdev = adapter->netdev;
11263
11264 #ifdef CONFIG_PCI_IOV
11265         struct ixgbe_hw *hw = &adapter->hw;
11266         struct pci_dev *bdev, *vfdev;
11267         u32 dw0, dw1, dw2, dw3;
11268         int vf, pos;
11269         u16 req_id, pf_func;
11270
11271         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
11272             adapter->num_vfs == 0)
11273                 goto skip_bad_vf_detection;
11274
11275         bdev = pdev->bus->self;
11276         while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
11277                 bdev = bdev->bus->self;
11278
11279         if (!bdev)
11280                 goto skip_bad_vf_detection;
11281
11282         pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
11283         if (!pos)
11284                 goto skip_bad_vf_detection;
11285
11286         dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
11287         dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
11288         dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
11289         dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
11290         if (ixgbe_removed(hw->hw_addr))
11291                 goto skip_bad_vf_detection;
11292
11293         req_id = dw1 >> 16;
11294         /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
11295         if (!(req_id & 0x0080))
11296                 goto skip_bad_vf_detection;
11297
11298         pf_func = req_id & 0x01;
11299         if ((pf_func & 1) == (pdev->devfn & 1)) {
11300                 unsigned int device_id;
11301
11302                 vf = (req_id & 0x7F) >> 1;
11303                 e_dev_err("VF %d has caused a PCIe error\n", vf);
11304                 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
11305                                 "%8.8x\tdw3: %8.8x\n",
11306                 dw0, dw1, dw2, dw3);
11307                 switch (adapter->hw.mac.type) {
11308                 case ixgbe_mac_82599EB:
11309                         device_id = IXGBE_82599_VF_DEVICE_ID;
11310                         break;
11311                 case ixgbe_mac_X540:
11312                         device_id = IXGBE_X540_VF_DEVICE_ID;
11313                         break;
11314                 case ixgbe_mac_X550:
11315                         device_id = IXGBE_DEV_ID_X550_VF;
11316                         break;
11317                 case ixgbe_mac_X550EM_x:
11318                         device_id = IXGBE_DEV_ID_X550EM_X_VF;
11319                         break;
11320                 case ixgbe_mac_x550em_a:
11321                         device_id = IXGBE_DEV_ID_X550EM_A_VF;
11322                         break;
11323                 default:
11324                         device_id = 0;
11325                         break;
11326                 }
11327
11328                 /* Find the pci device of the offending VF */
11329                 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
11330                 while (vfdev) {
11331                         if (vfdev->devfn == (req_id & 0xFF))
11332                                 break;
11333                         vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
11334                                                device_id, vfdev);
11335                 }
11336                 /*
11337                  * There's a slim chance the VF could have been hot plugged,
11338                  * so if it is no longer present we don't need to issue the
11339                  * VFLR.  Just clean up the AER in that case.
11340                  */
11341                 if (vfdev) {
11342                         pcie_flr(vfdev);
11343                         /* Free device reference count */
11344                         pci_dev_put(vfdev);
11345                 }
11346         }
11347
11348         /*
11349          * Even though the error may have occurred on the other port
11350          * we still need to increment the vf error reference count for
11351          * both ports because the I/O resume function will be called
11352          * for both of them.
11353          */
11354         adapter->vferr_refcount++;
11355
11356         return PCI_ERS_RESULT_RECOVERED;
11357
11358 skip_bad_vf_detection:
11359 #endif /* CONFIG_PCI_IOV */
11360         if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
11361                 return PCI_ERS_RESULT_DISCONNECT;
11362
11363         if (!netif_device_present(netdev))
11364                 return PCI_ERS_RESULT_DISCONNECT;
11365
11366         rtnl_lock();
11367         netif_device_detach(netdev);
11368
11369         if (netif_running(netdev))
11370                 ixgbe_close_suspend(adapter);
11371
11372         if (state == pci_channel_io_perm_failure) {
11373                 rtnl_unlock();
11374                 return PCI_ERS_RESULT_DISCONNECT;
11375         }
11376
11377         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
11378                 pci_disable_device(pdev);
11379         rtnl_unlock();
11380
11381         /* Request a slot reset. */
11382         return PCI_ERS_RESULT_NEED_RESET;
11383 }
11384
11385 /**
11386  * ixgbe_io_slot_reset - called after the pci bus has been reset.
11387  * @pdev: Pointer to PCI device
11388  *
11389  * Restart the card from scratch, as if from a cold-boot.
11390  */
11391 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
11392 {
11393         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11394         pci_ers_result_t result;
11395
11396         if (pci_enable_device_mem(pdev)) {
11397                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
11398                 result = PCI_ERS_RESULT_DISCONNECT;
11399         } else {
11400                 smp_mb__before_atomic();
11401                 clear_bit(__IXGBE_DISABLED, &adapter->state);
11402                 adapter->hw.hw_addr = adapter->io_addr;
11403                 pci_set_master(pdev);
11404                 pci_restore_state(pdev);
11405                 pci_save_state(pdev);
11406
11407                 pci_wake_from_d3(pdev, false);
11408
11409                 ixgbe_reset(adapter);
11410                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
11411                 result = PCI_ERS_RESULT_RECOVERED;
11412         }
11413
11414         return result;
11415 }
11416
11417 /**
11418  * ixgbe_io_resume - called when traffic can start flowing again.
11419  * @pdev: Pointer to PCI device
11420  *
11421  * This callback is called when the error recovery driver tells us that
11422  * its OK to resume normal operation.
11423  */
11424 static void ixgbe_io_resume(struct pci_dev *pdev)
11425 {
11426         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
11427         struct net_device *netdev = adapter->netdev;
11428
11429 #ifdef CONFIG_PCI_IOV
11430         if (adapter->vferr_refcount) {
11431                 e_info(drv, "Resuming after VF err\n");
11432                 adapter->vferr_refcount--;
11433                 return;
11434         }
11435
11436 #endif
11437         rtnl_lock();
11438         if (netif_running(netdev))
11439                 ixgbe_open(netdev);
11440
11441         netif_device_attach(netdev);
11442         rtnl_unlock();
11443 }
11444
11445 static const struct pci_error_handlers ixgbe_err_handler = {
11446         .error_detected = ixgbe_io_error_detected,
11447         .slot_reset = ixgbe_io_slot_reset,
11448         .resume = ixgbe_io_resume,
11449 };
11450
11451 static struct pci_driver ixgbe_driver = {
11452         .name     = ixgbe_driver_name,
11453         .id_table = ixgbe_pci_tbl,
11454         .probe    = ixgbe_probe,
11455         .remove   = ixgbe_remove,
11456 #ifdef CONFIG_PM
11457         .suspend  = ixgbe_suspend,
11458         .resume   = ixgbe_resume,
11459 #endif
11460         .shutdown = ixgbe_shutdown,
11461         .sriov_configure = ixgbe_pci_sriov_configure,
11462         .err_handler = &ixgbe_err_handler
11463 };
11464
11465 /**
11466  * ixgbe_init_module - Driver Registration Routine
11467  *
11468  * ixgbe_init_module is the first routine called when the driver is
11469  * loaded. All it does is register with the PCI subsystem.
11470  **/
11471 static int __init ixgbe_init_module(void)
11472 {
11473         int ret;
11474         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
11475         pr_info("%s\n", ixgbe_copyright);
11476
11477         ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
11478         if (!ixgbe_wq) {
11479                 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
11480                 return -ENOMEM;
11481         }
11482
11483         ixgbe_dbg_init();
11484
11485         ret = pci_register_driver(&ixgbe_driver);
11486         if (ret) {
11487                 destroy_workqueue(ixgbe_wq);
11488                 ixgbe_dbg_exit();
11489                 return ret;
11490         }
11491
11492 #ifdef CONFIG_IXGBE_DCA
11493         dca_register_notify(&dca_notifier);
11494 #endif
11495
11496         return 0;
11497 }
11498
11499 module_init(ixgbe_init_module);
11500
11501 /**
11502  * ixgbe_exit_module - Driver Exit Cleanup Routine
11503  *
11504  * ixgbe_exit_module is called just before the driver is removed
11505  * from memory.
11506  **/
11507 static void __exit ixgbe_exit_module(void)
11508 {
11509 #ifdef CONFIG_IXGBE_DCA
11510         dca_unregister_notify(&dca_notifier);
11511 #endif
11512         pci_unregister_driver(&ixgbe_driver);
11513
11514         ixgbe_dbg_exit();
11515         if (ixgbe_wq) {
11516                 destroy_workqueue(ixgbe_wq);
11517                 ixgbe_wq = NULL;
11518         }
11519 }
11520
11521 #ifdef CONFIG_IXGBE_DCA
11522 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
11523                             void *p)
11524 {
11525         int ret_val;
11526
11527         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
11528                                          __ixgbe_notify_dca);
11529
11530         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
11531 }
11532
11533 #endif /* CONFIG_IXGBE_DCA */
11534
11535 module_exit(ixgbe_exit_module);
11536
11537 /* ixgbe_main.c */