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

   1 /*******************************************************************************
   2
   3   Intel 10 Gigabit PCI Express Linux driver
   4   Copyright(c) 1999 - 2015 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include <linux/types.h>
  30 #include <linux/module.h>
  31 #include <linux/pci.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/string.h>
  35 #include <linux/in.h>
  36 #include <linux/interrupt.h>
  37 #include <linux/ip.h>
  38 #include <linux/tcp.h>
  39 #include <linux/sctp.h>
  40 #include <linux/pkt_sched.h>
  41 #include <linux/ipv6.h>
  42 #include <linux/slab.h>
  43 #include <net/checksum.h>
  44 #include <net/ip6_checksum.h>
  45 #include <linux/etherdevice.h>
  46 #include <linux/ethtool.h>
  47 #include <linux/if.h>
  48 #include <linux/if_vlan.h>
  49 #include <linux/if_macvlan.h>
  50 #include <linux/if_bridge.h>
  51 #include <linux/prefetch.h>
  52 #include <scsi/fc/fc_fcoe.h>
  53 #include <net/vxlan.h>
  54
  55 #ifdef CONFIG_OF
  56 #include <linux/of_net.h>
  57 #endif
  58
  59 #ifdef CONFIG_SPARC
  60 #include <asm/idprom.h>
  61 #include <asm/prom.h>
  62 #endif
  63
  64 #include "ixgbe.h"
  65 #include "ixgbe_common.h"
  66 #include "ixgbe_dcb_82599.h"
  67 #include "ixgbe_sriov.h"
  68 #ifdef CONFIG_IXGBE_VXLAN
  69 #include <net/vxlan.h>
  70 #endif
  71
  72 char ixgbe_driver_name[] = "ixgbe";
  73 static const char ixgbe_driver_string[] =
  74                               "Intel(R) 10 Gigabit PCI Express Network Driver";
  75 #ifdef IXGBE_FCOE
  76 char ixgbe_default_device_descr[] =
  77                               "Intel(R) 10 Gigabit Network Connection";
  78 #else
  79 static char ixgbe_default_device_descr[] =
  80                               "Intel(R) 10 Gigabit Network Connection";
  81 #endif
  82 #define DRV_VERSION "4.0.1-k"
  83 const char ixgbe_driver_version[] = DRV_VERSION;
  84 static const char ixgbe_copyright[] =
  85                                 "Copyright (c) 1999-2015 Intel Corporation.";
  86
  87 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";
  88
  89 static const struct ixgbe_info *ixgbe_info_tbl[] = {
  90         [board_82598]           = &ixgbe_82598_info,
  91         [board_82599]           = &ixgbe_82599_info,
  92         [board_X540]            = &ixgbe_X540_info,
  93         [board_X550]            = &ixgbe_X550_info,
  94         [board_X550EM_x]        = &ixgbe_X550EM_x_info,
  95 };
  96
  97 /* ixgbe_pci_tbl - PCI Device ID Table
  98  *
  99  * Wildcard entries (PCI_ANY_ID) should come last
 100  * Last entry must be all 0s
 101  *
 102  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
 103  *   Class, Class Mask, private data (not used) }
 104  */
 105 static const struct pci_device_id ixgbe_pci_tbl[] = {
 106         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598), board_82598 },
 107         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT), board_82598 },
 108         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT), board_82598 },
 109         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT), board_82598 },
 110         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT2), board_82598 },
 111         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4), board_82598 },
 112         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT), board_82598 },
 113         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_DA_DUAL_PORT), board_82598 },
 114         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_SR_DUAL_PORT_EM), board_82598 },
 115         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR), board_82598 },
 116         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_SFP_LOM), board_82598 },
 117         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_BX), board_82598 },
 118         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4), board_82599 },
 119         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_XAUI_LOM), board_82599 },
 120         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KR), board_82599 },
 121         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP), board_82599 },
 122         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_EM), board_82599 },
 123         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_KX4_MEZZ), board_82599 },
 124         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_CX4), board_82599 },
 125         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_BACKPLANE_FCOE), board_82599 },
 126         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_FCOE), board_82599 },
 127         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_T3_LOM), board_82599 },
 128         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_COMBO_BACKPLANE), board_82599 },
 129         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T), board_X540 },
 130         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF2), board_82599 },
 131         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_LS), board_82599 },
 132         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_QSFP_SF_QP), board_82599 },
 133         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599EN_SFP), board_82599 },
 134         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_SFP_SF_QP), board_82599 },
 135         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540T1), board_X540 },
 136         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550T), board_X550},
 137         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KX4), board_X550EM_x},
 138         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_KR), board_X550EM_x},
 139         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_10G_T), board_X550EM_x},
 140         /* required last entry */
 141         {0, }
 142 };
 143 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
 144
 145 #ifdef CONFIG_IXGBE_DCA
 146 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 147                             void *p);
 148 static struct notifier_block dca_notifier = {
 149         .notifier_call = ixgbe_notify_dca,
 150         .next          = NULL,
 151         .priority      = 0
 152 };
 153 #endif
 154
 155 #ifdef CONFIG_PCI_IOV
 156 static unsigned int max_vfs;
 157 module_param(max_vfs, uint, 0);
 158 MODULE_PARM_DESC(max_vfs,
 159                  "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
 160 #endif /* CONFIG_PCI_IOV */
 161
 162 static unsigned int allow_unsupported_sfp;
 163 module_param(allow_unsupported_sfp, uint, 0);
 164 MODULE_PARM_DESC(allow_unsupported_sfp,
 165                  "Allow unsupported and untested SFP+ modules on 82599-based adapters");
 166
 167 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 168 static int debug = -1;
 169 module_param(debug, int, 0);
 170 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 171
 172 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 173 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 174 MODULE_LICENSE("GPL");
 175 MODULE_VERSION(DRV_VERSION);
 176
 177 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
 178
 179 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
 180                                           u32 reg, u16 *value)
 181 {
 182         struct pci_dev *parent_dev;
 183         struct pci_bus *parent_bus;
 184
 185         parent_bus = adapter->pdev->bus->parent;
 186         if (!parent_bus)
 187                 return -1;
 188
 189         parent_dev = parent_bus->self;
 190         if (!parent_dev)
 191                 return -1;
 192
 193         if (!pci_is_pcie(parent_dev))
 194                 return -1;
 195
 196         pcie_capability_read_word(parent_dev, reg, value);
 197         if (*value == IXGBE_FAILED_READ_CFG_WORD &&
 198             ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
 199                 return -1;
 200         return 0;
 201 }
 202
 203 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
 204 {
 205         struct ixgbe_hw *hw = &adapter->hw;
 206         u16 link_status = 0;
 207         int err;
 208
 209         hw->bus.type = ixgbe_bus_type_pci_express;
 210
 211         /* Get the negotiated link width and speed from PCI config space of the
 212          * parent, as this device is behind a switch
 213          */
 214         err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
 215
 216         /* assume caller will handle error case */
 217         if (err)
 218                 return err;
 219
 220         hw->bus.width = ixgbe_convert_bus_width(link_status);
 221         hw->bus.speed = ixgbe_convert_bus_speed(link_status);
 222
 223         return 0;
 224 }
 225
 226 /**
 227  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
 228  * @hw: hw specific details
 229  *
 230  * This function is used by probe to determine whether a device's PCI-Express
 231  * bandwidth details should be gathered from the parent bus instead of from the
 232  * device. Used to ensure that various locations all have the correct device ID
 233  * checks.
 234  */
 235 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
 236 {
 237         switch (hw->device_id) {
 238         case IXGBE_DEV_ID_82599_SFP_SF_QP:
 239         case IXGBE_DEV_ID_82599_QSFP_SF_QP:
 240                 return true;
 241         default:
 242                 return false;
 243         }
 244 }
 245
 246 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
 247                                      int expected_gts)
 248 {
 249         struct ixgbe_hw *hw = &adapter->hw;
 250         int max_gts = 0;
 251         enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
 252         enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
 253         struct pci_dev *pdev;
 254
 255         /* Some devices are not connected over PCIe and thus do not negotiate
 256          * speed. These devices do not have valid bus info, and thus any report
 257          * we generate may not be correct.
 258          */
 259         if (hw->bus.type == ixgbe_bus_type_internal)
 260                 return;
 261
 262         /* determine whether to use the parent device */
 263         if (ixgbe_pcie_from_parent(&adapter->hw))
 264                 pdev = adapter->pdev->bus->parent->self;
 265         else
 266                 pdev = adapter->pdev;
 267
 268         if (pcie_get_minimum_link(pdev, &speed, &width) ||
 269             speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
 270                 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
 271                 return;
 272         }
 273
 274         switch (speed) {
 275         case PCIE_SPEED_2_5GT:
 276                 /* 8b/10b encoding reduces max throughput by 20% */
 277                 max_gts = 2 * width;
 278                 break;
 279         case PCIE_SPEED_5_0GT:
 280                 /* 8b/10b encoding reduces max throughput by 20% */
 281                 max_gts = 4 * width;
 282                 break;
 283         case PCIE_SPEED_8_0GT:
 284                 /* 128b/130b encoding reduces throughput by less than 2% */
 285                 max_gts = 8 * width;
 286                 break;
 287         default:
 288                 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
 289                 return;
 290         }
 291
 292         e_dev_info("PCI Express bandwidth of %dGT/s available\n",
 293                    max_gts);
 294         e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
 295                    (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
 296                     speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
 297                     speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
 298                     "Unknown"),
 299                    width,
 300                    (speed == PCIE_SPEED_2_5GT ? "20%" :
 301                     speed == PCIE_SPEED_5_0GT ? "20%" :
 302                     speed == PCIE_SPEED_8_0GT ? "<2%" :
 303                     "Unknown"));
 304
 305         if (max_gts < expected_gts) {
 306                 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
 307                 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
 308                         expected_gts);
 309                 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
 310         }
 311 }
 312
 313 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
 314 {
 315         if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
 316             !test_bit(__IXGBE_REMOVING, &adapter->state) &&
 317             !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
 318                 schedule_work(&adapter->service_task);
 319 }
 320
 321 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
 322 {
 323         struct ixgbe_adapter *adapter = hw->back;
 324
 325         if (!hw->hw_addr)
 326                 return;
 327         hw->hw_addr = NULL;
 328         e_dev_err("Adapter removed\n");
 329         if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
 330                 ixgbe_service_event_schedule(adapter);
 331 }
 332
 333 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
 334 {
 335         u32 value;
 336
 337         /* The following check not only optimizes a bit by not
 338          * performing a read on the status register when the
 339          * register just read was a status register read that
 340          * returned IXGBE_FAILED_READ_REG. It also blocks any
 341          * potential recursion.
 342          */
 343         if (reg == IXGBE_STATUS) {
 344                 ixgbe_remove_adapter(hw);
 345                 return;
 346         }
 347         value = ixgbe_read_reg(hw, IXGBE_STATUS);
 348         if (value == IXGBE_FAILED_READ_REG)
 349                 ixgbe_remove_adapter(hw);
 350 }
 351
 352 /**
 353  * ixgbe_read_reg - Read from device register
 354  * @hw: hw specific details
 355  * @reg: offset of register to read
 356  *
 357  * Returns : value read or IXGBE_FAILED_READ_REG if removed
 358  *
 359  * This function is used to read device registers. It checks for device
 360  * removal by confirming any read that returns all ones by checking the
 361  * status register value for all ones. This function avoids reading from
 362  * the hardware if a removal was previously detected in which case it
 363  * returns IXGBE_FAILED_READ_REG (all ones).
 364  */
 365 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
 366 {
 367         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 368         u32 value;
 369
 370         if (ixgbe_removed(reg_addr))
 371                 return IXGBE_FAILED_READ_REG;
 372         value = readl(reg_addr + reg);
 373         if (unlikely(value == IXGBE_FAILED_READ_REG))
 374                 ixgbe_check_remove(hw, reg);
 375         return value;
 376 }
 377
 378 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
 379 {
 380         u16 value;
 381
 382         pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
 383         if (value == IXGBE_FAILED_READ_CFG_WORD) {
 384                 ixgbe_remove_adapter(hw);
 385                 return true;
 386         }
 387         return false;
 388 }
 389
 390 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
 391 {
 392         struct ixgbe_adapter *adapter = hw->back;
 393         u16 value;
 394
 395         if (ixgbe_removed(hw->hw_addr))
 396                 return IXGBE_FAILED_READ_CFG_WORD;
 397         pci_read_config_word(adapter->pdev, reg, &value);
 398         if (value == IXGBE_FAILED_READ_CFG_WORD &&
 399             ixgbe_check_cfg_remove(hw, adapter->pdev))
 400                 return IXGBE_FAILED_READ_CFG_WORD;
 401         return value;
 402 }
 403
 404 #ifdef CONFIG_PCI_IOV
 405 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
 406 {
 407         struct ixgbe_adapter *adapter = hw->back;
 408         u32 value;
 409
 410         if (ixgbe_removed(hw->hw_addr))
 411                 return IXGBE_FAILED_READ_CFG_DWORD;
 412         pci_read_config_dword(adapter->pdev, reg, &value);
 413         if (value == IXGBE_FAILED_READ_CFG_DWORD &&
 414             ixgbe_check_cfg_remove(hw, adapter->pdev))
 415                 return IXGBE_FAILED_READ_CFG_DWORD;
 416         return value;
 417 }
 418 #endif /* CONFIG_PCI_IOV */
 419
 420 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
 421 {
 422         struct ixgbe_adapter *adapter = hw->back;
 423
 424         if (ixgbe_removed(hw->hw_addr))
 425                 return;
 426         pci_write_config_word(adapter->pdev, reg, value);
 427 }
 428
 429 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
 430 {
 431         BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
 432
 433         /* flush memory to make sure state is correct before next watchdog */
 434         smp_mb__before_atomic();
 435         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
 436 }
 437
 438 struct ixgbe_reg_info {
 439         u32 ofs;
 440         char *name;
 441 };
 442
 443 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
 444
 445         /* General Registers */
 446         {IXGBE_CTRL, "CTRL"},
 447         {IXGBE_STATUS, "STATUS"},
 448         {IXGBE_CTRL_EXT, "CTRL_EXT"},
 449
 450         /* Interrupt Registers */
 451         {IXGBE_EICR, "EICR"},
 452
 453         /* RX Registers */
 454         {IXGBE_SRRCTL(0), "SRRCTL"},
 455         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
 456         {IXGBE_RDLEN(0), "RDLEN"},
 457         {IXGBE_RDH(0), "RDH"},
 458         {IXGBE_RDT(0), "RDT"},
 459         {IXGBE_RXDCTL(0), "RXDCTL"},
 460         {IXGBE_RDBAL(0), "RDBAL"},
 461         {IXGBE_RDBAH(0), "RDBAH"},
 462
 463         /* TX Registers */
 464         {IXGBE_TDBAL(0), "TDBAL"},
 465         {IXGBE_TDBAH(0), "TDBAH"},
 466         {IXGBE_TDLEN(0), "TDLEN"},
 467         {IXGBE_TDH(0), "TDH"},
 468         {IXGBE_TDT(0), "TDT"},
 469         {IXGBE_TXDCTL(0), "TXDCTL"},
 470
 471         /* List Terminator */
 472         { .name = NULL }
 473 };
 474
 475
 476 /*
 477  * ixgbe_regdump - register printout routine
 478  */
 479 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
 480 {
 481         int i = 0, j = 0;
 482         char rname[16];
 483         u32 regs[64];
 484
 485         switch (reginfo->ofs) {
 486         case IXGBE_SRRCTL(0):
 487                 for (i = 0; i < 64; i++)
 488                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
 489                 break;
 490         case IXGBE_DCA_RXCTRL(0):
 491                 for (i = 0; i < 64; i++)
 492                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
 493                 break;
 494         case IXGBE_RDLEN(0):
 495                 for (i = 0; i < 64; i++)
 496                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
 497                 break;
 498         case IXGBE_RDH(0):
 499                 for (i = 0; i < 64; i++)
 500                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
 501                 break;
 502         case IXGBE_RDT(0):
 503                 for (i = 0; i < 64; i++)
 504                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
 505                 break;
 506         case IXGBE_RXDCTL(0):
 507                 for (i = 0; i < 64; i++)
 508                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
 509                 break;
 510         case IXGBE_RDBAL(0):
 511                 for (i = 0; i < 64; i++)
 512                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
 513                 break;
 514         case IXGBE_RDBAH(0):
 515                 for (i = 0; i < 64; i++)
 516                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
 517                 break;
 518         case IXGBE_TDBAL(0):
 519                 for (i = 0; i < 64; i++)
 520                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
 521                 break;
 522         case IXGBE_TDBAH(0):
 523                 for (i = 0; i < 64; i++)
 524                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
 525                 break;
 526         case IXGBE_TDLEN(0):
 527                 for (i = 0; i < 64; i++)
 528                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
 529                 break;
 530         case IXGBE_TDH(0):
 531                 for (i = 0; i < 64; i++)
 532                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
 533                 break;
 534         case IXGBE_TDT(0):
 535                 for (i = 0; i < 64; i++)
 536                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
 537                 break;
 538         case IXGBE_TXDCTL(0):
 539                 for (i = 0; i < 64; i++)
 540                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
 541                 break;
 542         default:
 543                 pr_info("%-15s %08x\n", reginfo->name,
 544                         IXGBE_READ_REG(hw, reginfo->ofs));
 545                 return;
 546         }
 547
 548         for (i = 0; i < 8; i++) {
 549                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
 550                 pr_err("%-15s", rname);
 551                 for (j = 0; j < 8; j++)
 552                         pr_cont(" %08x", regs[i*8+j]);
 553                 pr_cont("\n");
 554         }
 555
 556 }
 557
 558 /*
 559  * ixgbe_dump - Print registers, tx-rings and rx-rings
 560  */
 561 static void ixgbe_dump(struct ixgbe_adapter *adapter)
 562 {
 563         struct net_device *netdev = adapter->netdev;
 564         struct ixgbe_hw *hw = &adapter->hw;
 565         struct ixgbe_reg_info *reginfo;
 566         int n = 0;
 567         struct ixgbe_ring *tx_ring;
 568         struct ixgbe_tx_buffer *tx_buffer;
 569         union ixgbe_adv_tx_desc *tx_desc;
 570         struct my_u0 { u64 a; u64 b; } *u0;
 571         struct ixgbe_ring *rx_ring;
 572         union ixgbe_adv_rx_desc *rx_desc;
 573         struct ixgbe_rx_buffer *rx_buffer_info;
 574         u32 staterr;
 575         int i = 0;
 576
 577         if (!netif_msg_hw(adapter))
 578                 return;
 579
 580         /* Print netdevice Info */
 581         if (netdev) {
 582                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 583                 pr_info("Device Name     state            "
 584                         "trans_start      last_rx\n");
 585                 pr_info("%-15s %016lX %016lX %016lX\n",
 586                         netdev->name,
 587                         netdev->state,
 588                         netdev->trans_start,
 589                         netdev->last_rx);
 590         }
 591
 592         /* Print Registers */
 593         dev_info(&adapter->pdev->dev, "Register Dump\n");
 594         pr_info(" Register Name   Value\n");
 595         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
 596              reginfo->name; reginfo++) {
 597                 ixgbe_regdump(hw, reginfo);
 598         }
 599
 600         /* Print TX Ring Summary */
 601         if (!netdev || !netif_running(netdev))
 602                 return;
 603
 604         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 605         pr_info(" %s     %s              %s        %s\n",
 606                 "Queue [NTU] [NTC] [bi(ntc)->dma  ]",
 607                 "leng", "ntw", "timestamp");
 608         for (n = 0; n < adapter->num_tx_queues; n++) {
 609                 tx_ring = adapter->tx_ring[n];
 610                 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 611                 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
 612                            n, tx_ring->next_to_use, tx_ring->next_to_clean,
 613                            (u64)dma_unmap_addr(tx_buffer, dma),
 614                            dma_unmap_len(tx_buffer, len),
 615                            tx_buffer->next_to_watch,
 616                            (u64)tx_buffer->time_stamp);
 617         }
 618
 619         /* Print TX Rings */
 620         if (!netif_msg_tx_done(adapter))
 621                 goto rx_ring_summary;
 622
 623         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 624
 625         /* Transmit Descriptor Formats
 626          *
 627          * 82598 Advanced Transmit Descriptor
 628          *   +--------------------------------------------------------------+
 629          * 0 |         Buffer Address [63:0]                                |
 630          *   +--------------------------------------------------------------+
 631          * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 632          *   +--------------------------------------------------------------+
 633          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 634          *
 635          * 82598 Advanced Transmit Descriptor (Write-Back Format)
 636          *   +--------------------------------------------------------------+
 637          * 0 |                          RSV [63:0]                          |
 638          *   +--------------------------------------------------------------+
 639          * 8 |            RSV           |  STA  |          NXTSEQ           |
 640          *   +--------------------------------------------------------------+
 641          *   63                       36 35   32 31                         0
 642          *
 643          * 82599+ Advanced Transmit Descriptor
 644          *   +--------------------------------------------------------------+
 645          * 0 |         Buffer Address [63:0]                                |
 646          *   +--------------------------------------------------------------+
 647          * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
 648          *   +--------------------------------------------------------------+
 649          *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
 650          *
 651          * 82599+ Advanced Transmit Descriptor (Write-Back Format)
 652          *   +--------------------------------------------------------------+
 653          * 0 |                          RSV [63:0]                          |
 654          *   +--------------------------------------------------------------+
 655          * 8 |            RSV           |  STA  |           RSV             |
 656          *   +--------------------------------------------------------------+
 657          *   63                       36 35   32 31                         0
 658          */
 659
 660         for (n = 0; n < adapter->num_tx_queues; n++) {
 661                 tx_ring = adapter->tx_ring[n];
 662                 pr_info("------------------------------------\n");
 663                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 664                 pr_info("------------------------------------\n");
 665                 pr_info("%s%s    %s              %s        %s          %s\n",
 666                         "T [desc]     [address 63:0  ] ",
 667                         "[PlPOIdStDDt Ln] [bi->dma       ] ",
 668                         "leng", "ntw", "timestamp", "bi->skb");
 669
 670                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 671                         tx_desc = IXGBE_TX_DESC(tx_ring, i);
 672                         tx_buffer = &tx_ring->tx_buffer_info[i];
 673                         u0 = (struct my_u0 *)tx_desc;
 674                         if (dma_unmap_len(tx_buffer, len) > 0) {
 675                                 pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p",
 676                                         i,
 677                                         le64_to_cpu(u0->a),
 678                                         le64_to_cpu(u0->b),
 679                                         (u64)dma_unmap_addr(tx_buffer, dma),
 680                                         dma_unmap_len(tx_buffer, len),
 681                                         tx_buffer->next_to_watch,
 682                                         (u64)tx_buffer->time_stamp,
 683                                         tx_buffer->skb);
 684                                 if (i == tx_ring->next_to_use &&
 685                                         i == tx_ring->next_to_clean)
 686                                         pr_cont(" NTC/U\n");
 687                                 else if (i == tx_ring->next_to_use)
 688                                         pr_cont(" NTU\n");
 689                                 else if (i == tx_ring->next_to_clean)
 690                                         pr_cont(" NTC\n");
 691                                 else
 692                                         pr_cont("\n");
 693
 694                                 if (netif_msg_pktdata(adapter) &&
 695                                     tx_buffer->skb)
 696                                         print_hex_dump(KERN_INFO, "",
 697                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 698                                                 tx_buffer->skb->data,
 699                                                 dma_unmap_len(tx_buffer, len),
 700                                                 true);
 701                         }
 702                 }
 703         }
 704
 705         /* Print RX Rings Summary */
 706 rx_ring_summary:
 707         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 708         pr_info("Queue [NTU] [NTC]\n");
 709         for (n = 0; n < adapter->num_rx_queues; n++) {
 710                 rx_ring = adapter->rx_ring[n];
 711                 pr_info("%5d %5X %5X\n",
 712                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 713         }
 714
 715         /* Print RX Rings */
 716         if (!netif_msg_rx_status(adapter))
 717                 return;
 718
 719         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 720
 721         /* Receive Descriptor Formats
 722          *
 723          * 82598 Advanced Receive Descriptor (Read) Format
 724          *    63                                           1        0
 725          *    +-----------------------------------------------------+
 726          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 727          *    +----------------------------------------------+------+
 728          *  8 |       Header Buffer Address [63:1]           |  DD  |
 729          *    +-----------------------------------------------------+
 730          *
 731          *
 732          * 82598 Advanced Receive Descriptor (Write-Back) Format
 733          *
 734          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 735          *   +------------------------------------------------------+
 736          * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
 737          *   | Packet   | IP     |   |          |     | Type | Type |
 738          *   | Checksum | Ident  |   |          |     |      |      |
 739          *   +------------------------------------------------------+
 740          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 741          *   +------------------------------------------------------+
 742          *   63       48 47    32 31            20 19               0
 743          *
 744          * 82599+ Advanced Receive Descriptor (Read) Format
 745          *    63                                           1        0
 746          *    +-----------------------------------------------------+
 747          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 748          *    +----------------------------------------------+------+
 749          *  8 |       Header Buffer Address [63:1]           |  DD  |
 750          *    +-----------------------------------------------------+
 751          *
 752          *
 753          * 82599+ Advanced Receive Descriptor (Write-Back) Format
 754          *
 755          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 756          *   +------------------------------------------------------+
 757          * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
 758          *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
 759          *   |/ Flow Dir Flt ID  |   |          |     |      |      |
 760          *   +------------------------------------------------------+
 761          * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
 762          *   +------------------------------------------------------+
 763          *   63       48 47    32 31          20 19                 0
 764          */
 765
 766         for (n = 0; n < adapter->num_rx_queues; n++) {
 767                 rx_ring = adapter->rx_ring[n];
 768                 pr_info("------------------------------------\n");
 769                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 770                 pr_info("------------------------------------\n");
 771                 pr_info("%s%s%s",
 772                         "R  [desc]      [ PktBuf     A0] ",
 773                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
 774                         "<-- Adv Rx Read format\n");
 775                 pr_info("%s%s%s",
 776                         "RWB[desc]      [PcsmIpSHl PtRs] ",
 777                         "[vl er S cks ln] ---------------- [bi->skb       ] ",
 778                         "<-- Adv Rx Write-Back format\n");
 779
 780                 for (i = 0; i < rx_ring->count; i++) {
 781                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 782                         rx_desc = IXGBE_RX_DESC(rx_ring, i);
 783                         u0 = (struct my_u0 *)rx_desc;
 784                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 785                         if (staterr & IXGBE_RXD_STAT_DD) {
 786                                 /* Descriptor Done */
 787                                 pr_info("RWB[0x%03X]     %016llX "
 788                                         "%016llX ---------------- %p", i,
 789                                         le64_to_cpu(u0->a),
 790                                         le64_to_cpu(u0->b),
 791                                         rx_buffer_info->skb);
 792                         } else {
 793                                 pr_info("R  [0x%03X]     %016llX "
 794                                         "%016llX %016llX %p", i,
 795                                         le64_to_cpu(u0->a),
 796                                         le64_to_cpu(u0->b),
 797                                         (u64)rx_buffer_info->dma,
 798                                         rx_buffer_info->skb);
 799
 800                                 if (netif_msg_pktdata(adapter) &&
 801                                     rx_buffer_info->dma) {
 802                                         print_hex_dump(KERN_INFO, "",
 803                                            DUMP_PREFIX_ADDRESS, 16, 1,
 804                                            page_address(rx_buffer_info->page) +
 805                                                     rx_buffer_info->page_offset,
 806                                            ixgbe_rx_bufsz(rx_ring), true);
 807                                 }
 808                         }
 809
 810                         if (i == rx_ring->next_to_use)
 811                                 pr_cont(" NTU\n");
 812                         else if (i == rx_ring->next_to_clean)
 813                                 pr_cont(" NTC\n");
 814                         else
 815                                 pr_cont("\n");
 816
 817                 }
 818         }
 819 }
 820
 821 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 822 {
 823         u32 ctrl_ext;
 824
 825         /* Let firmware take over control of h/w */
 826         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 827         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 828                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 829 }
 830
 831 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 832 {
 833         u32 ctrl_ext;
 834
 835         /* Let firmware know the driver has taken over */
 836         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 837         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 838                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 839 }
 840
 841 /**
 842  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 843  * @adapter: pointer to adapter struct
 844  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 845  * @queue: queue to map the corresponding interrupt to
 846  * @msix_vector: the vector to map to the corresponding queue
 847  *
 848  */
 849 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 850                            u8 queue, u8 msix_vector)
 851 {
 852         u32 ivar, index;
 853         struct ixgbe_hw *hw = &adapter->hw;
 854         switch (hw->mac.type) {
 855         case ixgbe_mac_82598EB:
 856                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 857                 if (direction == -1)
 858                         direction = 0;
 859                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 860                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 861                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 862                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 863                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 864                 break;
 865         case ixgbe_mac_82599EB:
 866         case ixgbe_mac_X540:
 867         case ixgbe_mac_X550:
 868         case ixgbe_mac_X550EM_x:
 869                 if (direction == -1) {
 870                         /* other causes */
 871                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 872                         index = ((queue & 1) * 8);
 873                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 874                         ivar &= ~(0xFF << index);
 875                         ivar |= (msix_vector << index);
 876                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 877                         break;
 878                 } else {
 879                         /* tx or rx causes */
 880                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 881                         index = ((16 * (queue & 1)) + (8 * direction));
 882                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 883                         ivar &= ~(0xFF << index);
 884                         ivar |= (msix_vector << index);
 885                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 886                         break;
 887                 }
 888         default:
 889                 break;
 890         }
 891 }
 892
 893 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 894                                           u64 qmask)
 895 {
 896         u32 mask;
 897
 898         switch (adapter->hw.mac.type) {
 899         case ixgbe_mac_82598EB:
 900                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 901                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 902                 break;
 903         case ixgbe_mac_82599EB:
 904         case ixgbe_mac_X540:
 905         case ixgbe_mac_X550:
 906         case ixgbe_mac_X550EM_x:
 907                 mask = (qmask & 0xFFFFFFFF);
 908                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 909                 mask = (qmask >> 32);
 910                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 911                 break;
 912         default:
 913                 break;
 914         }
 915 }
 916
 917 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
 918                                       struct ixgbe_tx_buffer *tx_buffer)
 919 {
 920         if (tx_buffer->skb) {
 921                 dev_kfree_skb_any(tx_buffer->skb);
 922                 if (dma_unmap_len(tx_buffer, len))
 923                         dma_unmap_single(ring->dev,
 924                                          dma_unmap_addr(tx_buffer, dma),
 925                                          dma_unmap_len(tx_buffer, len),
 926                                          DMA_TO_DEVICE);
 927         } else if (dma_unmap_len(tx_buffer, len)) {
 928                 dma_unmap_page(ring->dev,
 929                                dma_unmap_addr(tx_buffer, dma),
 930                                dma_unmap_len(tx_buffer, len),
 931                                DMA_TO_DEVICE);
 932         }
 933         tx_buffer->next_to_watch = NULL;
 934         tx_buffer->skb = NULL;
 935         dma_unmap_len_set(tx_buffer, len, 0);
 936         /* tx_buffer must be completely set up in the transmit path */
 937 }
 938
 939 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
 940 {
 941         struct ixgbe_hw *hw = &adapter->hw;
 942         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 943         int i;
 944         u32 data;
 945
 946         if ((hw->fc.current_mode != ixgbe_fc_full) &&
 947             (hw->fc.current_mode != ixgbe_fc_rx_pause))
 948                 return;
 949
 950         switch (hw->mac.type) {
 951         case ixgbe_mac_82598EB:
 952                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
 953                 break;
 954         default:
 955                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
 956         }
 957         hwstats->lxoffrxc += data;
 958
 959         /* refill credits (no tx hang) if we received xoff */
 960         if (!data)
 961                 return;
 962
 963         for (i = 0; i < adapter->num_tx_queues; i++)
 964                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
 965                           &adapter->tx_ring[i]->state);
 966 }
 967
 968 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
 969 {
 970         struct ixgbe_hw *hw = &adapter->hw;
 971         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 972         u32 xoff[8] = {0};
 973         u8 tc;
 974         int i;
 975         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
 976
 977         if (adapter->ixgbe_ieee_pfc)
 978                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
 979
 980         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
 981                 ixgbe_update_xoff_rx_lfc(adapter);
 982                 return;
 983         }
 984
 985         /* update stats for each tc, only valid with PFC enabled */
 986         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
 987                 u32 pxoffrxc;
 988
 989                 switch (hw->mac.type) {
 990                 case ixgbe_mac_82598EB:
 991                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
 992                         break;
 993                 default:
 994                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
 995                 }
 996                 hwstats->pxoffrxc[i] += pxoffrxc;
 997                 /* Get the TC for given UP */
 998                 tc = netdev_get_prio_tc_map(adapter->netdev, i);
 999                 xoff[tc] += pxoffrxc;
1000         }
1001
1002         /* disarm tx queues that have received xoff frames */
1003         for (i = 0; i < adapter->num_tx_queues; i++) {
1004                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
1005
1006                 tc = tx_ring->dcb_tc;
1007                 if (xoff[tc])
1008                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1009         }
1010 }
1011
1012 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1013 {
1014         return ring->stats.packets;
1015 }
1016
1017 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1018 {
1019         struct ixgbe_adapter *adapter;
1020         struct ixgbe_hw *hw;
1021         u32 head, tail;
1022
1023         if (ring->l2_accel_priv)
1024                 adapter = ring->l2_accel_priv->real_adapter;
1025         else
1026                 adapter = netdev_priv(ring->netdev);
1027
1028         hw = &adapter->hw;
1029         head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1030         tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1031
1032         if (head != tail)
1033                 return (head < tail) ?
1034                         tail - head : (tail + ring->count - head);
1035
1036         return 0;
1037 }
1038
1039 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1040 {
1041         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1042         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1043         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1044
1045         clear_check_for_tx_hang(tx_ring);
1046
1047         /*
1048          * Check for a hung queue, but be thorough. This verifies
1049          * that a transmit has been completed since the previous
1050          * check AND there is at least one packet pending. The
1051          * ARMED bit is set to indicate a potential hang. The
1052          * bit is cleared if a pause frame is received to remove
1053          * false hang detection due to PFC or 802.3x frames. By
1054          * requiring this to fail twice we avoid races with
1055          * pfc clearing the ARMED bit and conditions where we
1056          * run the check_tx_hang logic with a transmit completion
1057          * pending but without time to complete it yet.
1058          */
1059         if (tx_done_old == tx_done && tx_pending)
1060                 /* make sure it is true for two checks in a row */
1061                 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1062                                         &tx_ring->state);
1063         /* update completed stats and continue */
1064         tx_ring->tx_stats.tx_done_old = tx_done;
1065         /* reset the countdown */
1066         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1067
1068         return false;
1069 }
1070
1071 /**
1072  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1073  * @adapter: driver private struct
1074  **/
1075 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1076 {
1077
1078         /* Do the reset outside of interrupt context */
1079         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1080                 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
1081                 e_warn(drv, "initiating reset due to tx timeout\n");
1082                 ixgbe_service_event_schedule(adapter);
1083         }
1084 }
1085
1086 /**
1087  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1088  * @q_vector: structure containing interrupt and ring information
1089  * @tx_ring: tx ring to clean
1090  **/
1091 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1092                                struct ixgbe_ring *tx_ring)
1093 {
1094         struct ixgbe_adapter *adapter = q_vector->adapter;
1095         struct ixgbe_tx_buffer *tx_buffer;
1096         union ixgbe_adv_tx_desc *tx_desc;
1097         unsigned int total_bytes = 0, total_packets = 0;
1098         unsigned int budget = q_vector->tx.work_limit;
1099         unsigned int i = tx_ring->next_to_clean;
1100
1101         if (test_bit(__IXGBE_DOWN, &adapter->state))
1102                 return true;
1103
1104         tx_buffer = &tx_ring->tx_buffer_info[i];
1105         tx_desc = IXGBE_TX_DESC(tx_ring, i);
1106         i -= tx_ring->count;
1107
1108         do {
1109                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1110
1111                 /* if next_to_watch is not set then there is no work pending */
1112                 if (!eop_desc)
1113                         break;
1114
1115                 /* prevent any other reads prior to eop_desc */
1116                 read_barrier_depends();
1117
1118                 /* if DD is not set pending work has not been completed */
1119                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1120                         break;
1121
1122                 /* clear next_to_watch to prevent false hangs */
1123                 tx_buffer->next_to_watch = NULL;
1124
1125                 /* update the statistics for this packet */
1126                 total_bytes += tx_buffer->bytecount;
1127                 total_packets += tx_buffer->gso_segs;
1128
1129                 /* free the skb */
1130                 dev_consume_skb_any(tx_buffer->skb);
1131
1132                 /* unmap skb header data */
1133                 dma_unmap_single(tx_ring->dev,
1134                                  dma_unmap_addr(tx_buffer, dma),
1135                                  dma_unmap_len(tx_buffer, len),
1136                                  DMA_TO_DEVICE);
1137
1138                 /* clear tx_buffer data */
1139                 tx_buffer->skb = NULL;
1140                 dma_unmap_len_set(tx_buffer, len, 0);
1141
1142                 /* unmap remaining buffers */
1143                 while (tx_desc != eop_desc) {
1144                         tx_buffer++;
1145                         tx_desc++;
1146                         i++;
1147                         if (unlikely(!i)) {
1148                                 i -= tx_ring->count;
1149                                 tx_buffer = tx_ring->tx_buffer_info;
1150                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1151                         }
1152
1153                         /* unmap any remaining paged data */
1154                         if (dma_unmap_len(tx_buffer, len)) {
1155                                 dma_unmap_page(tx_ring->dev,
1156                                                dma_unmap_addr(tx_buffer, dma),
1157                                                dma_unmap_len(tx_buffer, len),
1158                                                DMA_TO_DEVICE);
1159                                 dma_unmap_len_set(tx_buffer, len, 0);
1160                         }
1161                 }
1162
1163                 /* move us one more past the eop_desc for start of next pkt */
1164                 tx_buffer++;
1165                 tx_desc++;
1166                 i++;
1167                 if (unlikely(!i)) {
1168                         i -= tx_ring->count;
1169                         tx_buffer = tx_ring->tx_buffer_info;
1170                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1171                 }
1172
1173                 /* issue prefetch for next Tx descriptor */
1174                 prefetch(tx_desc);
1175
1176                 /* update budget accounting */
1177                 budget--;
1178         } while (likely(budget));
1179
1180         i += tx_ring->count;
1181         tx_ring->next_to_clean = i;
1182         u64_stats_update_begin(&tx_ring->syncp);
1183         tx_ring->stats.bytes += total_bytes;
1184         tx_ring->stats.packets += total_packets;
1185         u64_stats_update_end(&tx_ring->syncp);
1186         q_vector->tx.total_bytes += total_bytes;
1187         q_vector->tx.total_packets += total_packets;
1188
1189         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1190                 /* schedule immediate reset if we believe we hung */
1191                 struct ixgbe_hw *hw = &adapter->hw;
1192                 e_err(drv, "Detected Tx Unit Hang\n"
1193                         "  Tx Queue             <%d>\n"
1194                         "  TDH, TDT             <%x>, <%x>\n"
1195                         "  next_to_use          <%x>\n"
1196                         "  next_to_clean        <%x>\n"
1197                         "tx_buffer_info[next_to_clean]\n"
1198                         "  time_stamp           <%lx>\n"
1199                         "  jiffies              <%lx>\n",
1200                         tx_ring->queue_index,
1201                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1202                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1203                         tx_ring->next_to_use, i,
1204                         tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1205
1206                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1207
1208                 e_info(probe,
1209                        "tx hang %d detected on queue %d, resetting adapter\n",
1210                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
1211
1212                 /* schedule immediate reset if we believe we hung */
1213                 ixgbe_tx_timeout_reset(adapter);
1214
1215                 /* the adapter is about to reset, no point in enabling stuff */
1216                 return true;
1217         }
1218
1219         netdev_tx_completed_queue(txring_txq(tx_ring),
1220                                   total_packets, total_bytes);
1221
1222 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1223         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1224                      (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1225                 /* Make sure that anybody stopping the queue after this
1226                  * sees the new next_to_clean.
1227                  */
1228                 smp_mb();
1229                 if (__netif_subqueue_stopped(tx_ring->netdev,
1230                                              tx_ring->queue_index)
1231                     && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1232                         netif_wake_subqueue(tx_ring->netdev,
1233                                             tx_ring->queue_index);
1234                         ++tx_ring->tx_stats.restart_queue;
1235                 }
1236         }
1237
1238         return !!budget;
1239 }
1240
1241 #ifdef CONFIG_IXGBE_DCA
1242 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1243                                 struct ixgbe_ring *tx_ring,
1244                                 int cpu)
1245 {
1246         struct ixgbe_hw *hw = &adapter->hw;
1247         u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1248         u16 reg_offset;
1249
1250         switch (hw->mac.type) {
1251         case ixgbe_mac_82598EB:
1252                 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1253                 break;
1254         case ixgbe_mac_82599EB:
1255         case ixgbe_mac_X540:
1256                 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1257                 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1258                 break;
1259         default:
1260                 /* for unknown hardware do not write register */
1261                 return;
1262         }
1263
1264         /*
1265          * We can enable relaxed ordering for reads, but not writes when
1266          * DCA is enabled.  This is due to a known issue in some chipsets
1267          * which will cause the DCA tag to be cleared.
1268          */
1269         txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1270                   IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1271                   IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1272
1273         IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1274 }
1275
1276 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1277                                 struct ixgbe_ring *rx_ring,
1278                                 int cpu)
1279 {
1280         struct ixgbe_hw *hw = &adapter->hw;
1281         u32 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1282         u8 reg_idx = rx_ring->reg_idx;
1283
1284
1285         switch (hw->mac.type) {
1286         case ixgbe_mac_82599EB:
1287         case ixgbe_mac_X540:
1288                 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1289                 break;
1290         default:
1291                 break;
1292         }
1293
1294         /*
1295          * We can enable relaxed ordering for reads, but not writes when
1296          * DCA is enabled.  This is due to a known issue in some chipsets
1297          * which will cause the DCA tag to be cleared.
1298          */
1299         rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1300                   IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1301
1302         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1303 }
1304
1305 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1306 {
1307         struct ixgbe_adapter *adapter = q_vector->adapter;
1308         struct ixgbe_ring *ring;
1309         int cpu = get_cpu();
1310
1311         if (q_vector->cpu == cpu)
1312                 goto out_no_update;
1313
1314         ixgbe_for_each_ring(ring, q_vector->tx)
1315                 ixgbe_update_tx_dca(adapter, ring, cpu);
1316
1317         ixgbe_for_each_ring(ring, q_vector->rx)
1318                 ixgbe_update_rx_dca(adapter, ring, cpu);
1319
1320         q_vector->cpu = cpu;
1321 out_no_update:
1322         put_cpu();
1323 }
1324
1325 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1326 {
1327         int i;
1328
1329         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
1330                 return;
1331
1332         /* always use CB2 mode, difference is masked in the CB driver */
1333         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
1334
1335         for (i = 0; i < adapter->num_q_vectors; i++) {
1336                 adapter->q_vector[i]->cpu = -1;
1337                 ixgbe_update_dca(adapter->q_vector[i]);
1338         }
1339 }
1340
1341 static int __ixgbe_notify_dca(struct device *dev, void *data)
1342 {
1343         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1344         unsigned long event = *(unsigned long *)data;
1345
1346         if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1347                 return 0;
1348
1349         switch (event) {
1350         case DCA_PROVIDER_ADD:
1351                 /* if we're already enabled, don't do it again */
1352                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1353                         break;
1354                 if (dca_add_requester(dev) == 0) {
1355                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1356                         ixgbe_setup_dca(adapter);
1357                         break;
1358                 }
1359                 /* Fall Through since DCA is disabled. */
1360         case DCA_PROVIDER_REMOVE:
1361                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1362                         dca_remove_requester(dev);
1363                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1364                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
1365                 }
1366                 break;
1367         }
1368
1369         return 0;
1370 }
1371
1372 #endif /* CONFIG_IXGBE_DCA */
1373
1374 #define IXGBE_RSS_L4_TYPES_MASK \
1375         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1376          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1377          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1378          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1379
1380 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1381                                  union ixgbe_adv_rx_desc *rx_desc,
1382                                  struct sk_buff *skb)
1383 {
1384         u16 rss_type;
1385
1386         if (!(ring->netdev->features & NETIF_F_RXHASH))
1387                 return;
1388
1389         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1390                    IXGBE_RXDADV_RSSTYPE_MASK;
1391
1392         if (!rss_type)
1393                 return;
1394
1395         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1396                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1397                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1398 }
1399
1400 #ifdef IXGBE_FCOE
1401 /**
1402  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1403  * @ring: structure containing ring specific data
1404  * @rx_desc: advanced rx descriptor
1405  *
1406  * Returns : true if it is FCoE pkt
1407  */
1408 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1409                                     union ixgbe_adv_rx_desc *rx_desc)
1410 {
1411         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1412
1413         return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1414                ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1415                 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1416                              IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1417 }
1418
1419 #endif /* IXGBE_FCOE */
1420 /**
1421  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1422  * @ring: structure containing ring specific data
1423  * @rx_desc: current Rx descriptor being processed
1424  * @skb: skb currently being received and modified
1425  **/
1426 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1427                                      union ixgbe_adv_rx_desc *rx_desc,
1428                                      struct sk_buff *skb)
1429 {
1430         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1431         __le16 hdr_info = rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1432         bool encap_pkt = false;
1433
1434         skb_checksum_none_assert(skb);
1435
1436         /* Rx csum disabled */
1437         if (!(ring->netdev->features & NETIF_F_RXCSUM))
1438                 return;
1439
1440         if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) &&
1441             (hdr_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_TUNNEL >> 16))) {
1442                 encap_pkt = true;
1443                 skb->encapsulation = 1;
1444         }
1445
1446         /* if IP and error */
1447         if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1448             ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1449                 ring->rx_stats.csum_err++;
1450                 return;
1451         }
1452
1453         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1454                 return;
1455
1456         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1457                 /*
1458                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1459                  * checksum errors.
1460                  */
1461                 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1462                     test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1463                         return;
1464
1465                 ring->rx_stats.csum_err++;
1466                 return;
1467         }
1468
1469         /* It must be a TCP or UDP packet with a valid checksum */
1470         skb->ip_summed = CHECKSUM_UNNECESSARY;
1471         if (encap_pkt) {
1472                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1473                         return;
1474
1475                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1476                         ring->rx_stats.csum_err++;
1477                         return;
1478                 }
1479                 /* If we checked the outer header let the stack know */
1480                 skb->csum_level = 1;
1481         }
1482 }
1483
1484 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1485                                     struct ixgbe_rx_buffer *bi)
1486 {
1487         struct page *page = bi->page;
1488         dma_addr_t dma;
1489
1490         /* since we are recycling buffers we should seldom need to alloc */
1491         if (likely(page))
1492                 return true;
1493
1494         /* alloc new page for storage */
1495         page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1496         if (unlikely(!page)) {
1497                 rx_ring->rx_stats.alloc_rx_page_failed++;
1498                 return false;
1499         }
1500
1501         /* map page for use */
1502         dma = dma_map_page(rx_ring->dev, page, 0,
1503                            ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1504
1505         /*
1506          * if mapping failed free memory back to system since
1507          * there isn't much point in holding memory we can't use
1508          */
1509         if (dma_mapping_error(rx_ring->dev, dma)) {
1510                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1511
1512                 rx_ring->rx_stats.alloc_rx_page_failed++;
1513                 return false;
1514         }
1515
1516         bi->dma = dma;
1517         bi->page = page;
1518         bi->page_offset = 0;
1519
1520         return true;
1521 }
1522
1523 /**
1524  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1525  * @rx_ring: ring to place buffers on
1526  * @cleaned_count: number of buffers to replace
1527  **/
1528 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1529 {
1530         union ixgbe_adv_rx_desc *rx_desc;
1531         struct ixgbe_rx_buffer *bi;
1532         u16 i = rx_ring->next_to_use;
1533
1534         /* nothing to do */
1535         if (!cleaned_count)
1536                 return;
1537
1538         rx_desc = IXGBE_RX_DESC(rx_ring, i);
1539         bi = &rx_ring->rx_buffer_info[i];
1540         i -= rx_ring->count;
1541
1542         do {
1543                 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1544                         break;
1545
1546                 /*
1547                  * Refresh the desc even if buffer_addrs didn't change
1548                  * because each write-back erases this info.
1549                  */
1550                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1551
1552                 rx_desc++;
1553                 bi++;
1554                 i++;
1555                 if (unlikely(!i)) {
1556                         rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1557                         bi = rx_ring->rx_buffer_info;
1558                         i -= rx_ring->count;
1559                 }
1560
1561                 /* clear the status bits for the next_to_use descriptor */
1562                 rx_desc->wb.upper.status_error = 0;
1563
1564                 cleaned_count--;
1565         } while (cleaned_count);
1566
1567         i += rx_ring->count;
1568
1569         if (rx_ring->next_to_use != i) {
1570                 rx_ring->next_to_use = i;
1571
1572                 /* update next to alloc since we have filled the ring */
1573                 rx_ring->next_to_alloc = i;
1574
1575                 /* Force memory writes to complete before letting h/w
1576                  * know there are new descriptors to fetch.  (Only
1577                  * applicable for weak-ordered memory model archs,
1578                  * such as IA-64).
1579                  */
1580                 wmb();
1581                 writel(i, rx_ring->tail);
1582         }
1583 }
1584
1585 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1586                                    struct sk_buff *skb)
1587 {
1588         u16 hdr_len = skb_headlen(skb);
1589
1590         /* set gso_size to avoid messing up TCP MSS */
1591         skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1592                                                  IXGBE_CB(skb)->append_cnt);
1593         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1594 }
1595
1596 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1597                                    struct sk_buff *skb)
1598 {
1599         /* if append_cnt is 0 then frame is not RSC */
1600         if (!IXGBE_CB(skb)->append_cnt)
1601                 return;
1602
1603         rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1604         rx_ring->rx_stats.rsc_flush++;
1605
1606         ixgbe_set_rsc_gso_size(rx_ring, skb);
1607
1608         /* gso_size is computed using append_cnt so always clear it last */
1609         IXGBE_CB(skb)->append_cnt = 0;
1610 }
1611
1612 /**
1613  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1614  * @rx_ring: rx descriptor ring packet is being transacted on
1615  * @rx_desc: pointer to the EOP Rx descriptor
1616  * @skb: pointer to current skb being populated
1617  *
1618  * This function checks the ring, descriptor, and packet information in
1619  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1620  * other fields within the skb.
1621  **/
1622 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1623                                      union ixgbe_adv_rx_desc *rx_desc,
1624                                      struct sk_buff *skb)
1625 {
1626         struct net_device *dev = rx_ring->netdev;
1627
1628         ixgbe_update_rsc_stats(rx_ring, skb);
1629
1630         ixgbe_rx_hash(rx_ring, rx_desc, skb);
1631
1632         ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1633
1634         if (unlikely(ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_STAT_TS)))
1635                 ixgbe_ptp_rx_hwtstamp(rx_ring->q_vector->adapter, skb);
1636
1637         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1638             ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1639                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1640                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1641         }
1642
1643         skb_record_rx_queue(skb, rx_ring->queue_index);
1644
1645         skb->protocol = eth_type_trans(skb, dev);
1646 }
1647
1648 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1649                          struct sk_buff *skb)
1650 {
1651         if (ixgbe_qv_busy_polling(q_vector))
1652                 netif_receive_skb(skb);
1653         else
1654                 napi_gro_receive(&q_vector->napi, skb);
1655 }
1656
1657 /**
1658  * ixgbe_is_non_eop - process handling of non-EOP buffers
1659  * @rx_ring: Rx ring being processed
1660  * @rx_desc: Rx descriptor for current buffer
1661  * @skb: Current socket buffer containing buffer in progress
1662  *
1663  * This function updates next to clean.  If the buffer is an EOP buffer
1664  * this function exits returning false, otherwise it will place the
1665  * sk_buff in the next buffer to be chained and return true indicating
1666  * that this is in fact a non-EOP buffer.
1667  **/
1668 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1669                              union ixgbe_adv_rx_desc *rx_desc,
1670                              struct sk_buff *skb)
1671 {
1672         u32 ntc = rx_ring->next_to_clean + 1;
1673
1674         /* fetch, update, and store next to clean */
1675         ntc = (ntc < rx_ring->count) ? ntc : 0;
1676         rx_ring->next_to_clean = ntc;
1677
1678         prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1679
1680         /* update RSC append count if present */
1681         if (ring_is_rsc_enabled(rx_ring)) {
1682                 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1683                                      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1684
1685                 if (unlikely(rsc_enabled)) {
1686                         u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1687
1688                         rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1689                         IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1690
1691                         /* update ntc based on RSC value */
1692                         ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1693                         ntc &= IXGBE_RXDADV_NEXTP_MASK;
1694                         ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1695                 }
1696         }
1697
1698         /* if we are the last buffer then there is nothing else to do */
1699         if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1700                 return false;
1701
1702         /* place skb in next buffer to be received */
1703         rx_ring->rx_buffer_info[ntc].skb = skb;
1704         rx_ring->rx_stats.non_eop_descs++;
1705
1706         return true;
1707 }
1708
1709 /**
1710  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1711  * @rx_ring: rx descriptor ring packet is being transacted on
1712  * @skb: pointer to current skb being adjusted
1713  *
1714  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1715  * main difference between this version and the original function is that
1716  * this function can make several assumptions about the state of things
1717  * that allow for significant optimizations versus the standard function.
1718  * As a result we can do things like drop a frag and maintain an accurate
1719  * truesize for the skb.
1720  */
1721 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1722                             struct sk_buff *skb)
1723 {
1724         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1725         unsigned char *va;
1726         unsigned int pull_len;
1727
1728         /*
1729          * it is valid to use page_address instead of kmap since we are
1730          * working with pages allocated out of the lomem pool per
1731          * alloc_page(GFP_ATOMIC)
1732          */
1733         va = skb_frag_address(frag);
1734
1735         /*
1736          * we need the header to contain the greater of either ETH_HLEN or
1737          * 60 bytes if the skb->len is less than 60 for skb_pad.
1738          */
1739         pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1740
1741         /* align pull length to size of long to optimize memcpy performance */
1742         skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1743
1744         /* update all of the pointers */
1745         skb_frag_size_sub(frag, pull_len);
1746         frag->page_offset += pull_len;
1747         skb->data_len -= pull_len;
1748         skb->tail += pull_len;
1749 }
1750
1751 /**
1752  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1753  * @rx_ring: rx descriptor ring packet is being transacted on
1754  * @skb: pointer to current skb being updated
1755  *
1756  * This function provides a basic DMA sync up for the first fragment of an
1757  * skb.  The reason for doing this is that the first fragment cannot be
1758  * unmapped until we have reached the end of packet descriptor for a buffer
1759  * chain.
1760  */
1761 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1762                                 struct sk_buff *skb)
1763 {
1764         /* if the page was released unmap it, else just sync our portion */
1765         if (unlikely(IXGBE_CB(skb)->page_released)) {
1766                 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1767                                ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1768                 IXGBE_CB(skb)->page_released = false;
1769         } else {
1770                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1771
1772                 dma_sync_single_range_for_cpu(rx_ring->dev,
1773                                               IXGBE_CB(skb)->dma,
1774                                               frag->page_offset,
1775                                               ixgbe_rx_bufsz(rx_ring),
1776                                               DMA_FROM_DEVICE);
1777         }
1778         IXGBE_CB(skb)->dma = 0;
1779 }
1780
1781 /**
1782  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1783  * @rx_ring: rx descriptor ring packet is being transacted on
1784  * @rx_desc: pointer to the EOP Rx descriptor
1785  * @skb: pointer to current skb being fixed
1786  *
1787  * Check for corrupted packet headers caused by senders on the local L2
1788  * embedded NIC switch not setting up their Tx Descriptors right.  These
1789  * should be very rare.
1790  *
1791  * Also address the case where we are pulling data in on pages only
1792  * and as such no data is present in the skb header.
1793  *
1794  * In addition if skb is not at least 60 bytes we need to pad it so that
1795  * it is large enough to qualify as a valid Ethernet frame.
1796  *
1797  * Returns true if an error was encountered and skb was freed.
1798  **/
1799 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1800                                   union ixgbe_adv_rx_desc *rx_desc,
1801                                   struct sk_buff *skb)
1802 {
1803         struct net_device *netdev = rx_ring->netdev;
1804
1805         /* verify that the packet does not have any known errors */
1806         if (unlikely(ixgbe_test_staterr(rx_desc,
1807                                         IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1808             !(netdev->features & NETIF_F_RXALL))) {
1809                 dev_kfree_skb_any(skb);
1810                 return true;
1811         }
1812
1813         /* place header in linear portion of buffer */
1814         if (skb_is_nonlinear(skb))
1815                 ixgbe_pull_tail(rx_ring, skb);
1816
1817 #ifdef IXGBE_FCOE
1818         /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1819         if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1820                 return false;
1821
1822 #endif
1823         /* if eth_skb_pad returns an error the skb was freed */
1824         if (eth_skb_pad(skb))
1825                 return true;
1826
1827         return false;
1828 }
1829
1830 /**
1831  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1832  * @rx_ring: rx descriptor ring to store buffers on
1833  * @old_buff: donor buffer to have page reused
1834  *
1835  * Synchronizes page for reuse by the adapter
1836  **/
1837 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1838                                 struct ixgbe_rx_buffer *old_buff)
1839 {
1840         struct ixgbe_rx_buffer *new_buff;
1841         u16 nta = rx_ring->next_to_alloc;
1842
1843         new_buff = &rx_ring->rx_buffer_info[nta];
1844
1845         /* update, and store next to alloc */
1846         nta++;
1847         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1848
1849         /* transfer page from old buffer to new buffer */
1850         *new_buff = *old_buff;
1851
1852         /* sync the buffer for use by the device */
1853         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1854                                          new_buff->page_offset,
1855                                          ixgbe_rx_bufsz(rx_ring),
1856                                          DMA_FROM_DEVICE);
1857 }
1858
1859 static inline bool ixgbe_page_is_reserved(struct page *page)
1860 {
1861         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1862 }
1863
1864 /**
1865  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1866  * @rx_ring: rx descriptor ring to transact packets on
1867  * @rx_buffer: buffer containing page to add
1868  * @rx_desc: descriptor containing length of buffer written by hardware
1869  * @skb: sk_buff to place the data into
1870  *
1871  * This function will add the data contained in rx_buffer->page to the skb.
1872  * This is done either through a direct copy if the data in the buffer is
1873  * less than the skb header size, otherwise it will just attach the page as
1874  * a frag to the skb.
1875  *
1876  * The function will then update the page offset if necessary and return
1877  * true if the buffer can be reused by the adapter.
1878  **/
1879 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1880                               struct ixgbe_rx_buffer *rx_buffer,
1881                               union ixgbe_adv_rx_desc *rx_desc,
1882                               struct sk_buff *skb)
1883 {
1884         struct page *page = rx_buffer->page;
1885         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1886 #if (PAGE_SIZE < 8192)
1887         unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1888 #else
1889         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1890         unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1891                                    ixgbe_rx_bufsz(rx_ring);
1892 #endif
1893
1894         if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1895                 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1896
1897                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1898
1899                 /* page is not reserved, we can reuse buffer as-is */
1900                 if (likely(!ixgbe_page_is_reserved(page)))
1901                         return true;
1902
1903                 /* this page cannot be reused so discard it */
1904                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1905                 return false;
1906         }
1907
1908         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1909                         rx_buffer->page_offset, size, truesize);
1910
1911         /* avoid re-using remote pages */
1912         if (unlikely(ixgbe_page_is_reserved(page)))
1913                 return false;
1914
1915 #if (PAGE_SIZE < 8192)
1916         /* if we are only owner of page we can reuse it */
1917         if (unlikely(page_count(page) != 1))
1918                 return false;
1919
1920         /* flip page offset to other buffer */
1921         rx_buffer->page_offset ^= truesize;
1922 #else
1923         /* move offset up to the next cache line */
1924         rx_buffer->page_offset += truesize;
1925
1926         if (rx_buffer->page_offset > last_offset)
1927                 return false;
1928 #endif
1929
1930         /* Even if we own the page, we are not allowed to use atomic_set()
1931          * This would break get_page_unless_zero() users.
1932          */
1933         atomic_inc(&page->_count);
1934
1935         return true;
1936 }
1937
1938 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1939                                              union ixgbe_adv_rx_desc *rx_desc)
1940 {
1941         struct ixgbe_rx_buffer *rx_buffer;
1942         struct sk_buff *skb;
1943         struct page *page;
1944
1945         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1946         page = rx_buffer->page;
1947         prefetchw(page);
1948
1949         skb = rx_buffer->skb;
1950
1951         if (likely(!skb)) {
1952                 void *page_addr = page_address(page) +
1953                                   rx_buffer->page_offset;
1954
1955                 /* prefetch first cache line of first page */
1956                 prefetch(page_addr);
1957 #if L1_CACHE_BYTES < 128
1958                 prefetch(page_addr + L1_CACHE_BYTES);
1959 #endif
1960
1961                 /* allocate a skb to store the frags */
1962                 skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1963                                      IXGBE_RX_HDR_SIZE);
1964                 if (unlikely(!skb)) {
1965                         rx_ring->rx_stats.alloc_rx_buff_failed++;
1966                         return NULL;
1967                 }
1968
1969                 /*
1970                  * we will be copying header into skb->data in
1971                  * pskb_may_pull so it is in our interest to prefetch
1972                  * it now to avoid a possible cache miss
1973                  */
1974                 prefetchw(skb->data);
1975
1976                 /*
1977                  * Delay unmapping of the first packet. It carries the
1978                  * header information, HW may still access the header
1979                  * after the writeback.  Only unmap it when EOP is
1980                  * reached
1981                  */
1982                 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1983                         goto dma_sync;
1984
1985                 IXGBE_CB(skb)->dma = rx_buffer->dma;
1986         } else {
1987                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
1988                         ixgbe_dma_sync_frag(rx_ring, skb);
1989
1990 dma_sync:
1991                 /* we are reusing so sync this buffer for CPU use */
1992                 dma_sync_single_range_for_cpu(rx_ring->dev,
1993                                               rx_buffer->dma,
1994                                               rx_buffer->page_offset,
1995                                               ixgbe_rx_bufsz(rx_ring),
1996                                               DMA_FROM_DEVICE);
1997
1998                 rx_buffer->skb = NULL;
1999         }
2000
2001         /* pull page into skb */
2002         if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
2003                 /* hand second half of page back to the ring */
2004                 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2005         } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
2006                 /* the page has been released from the ring */
2007                 IXGBE_CB(skb)->page_released = true;
2008         } else {
2009                 /* we are not reusing the buffer so unmap it */
2010                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
2011                                ixgbe_rx_pg_size(rx_ring),
2012                                DMA_FROM_DEVICE);
2013         }
2014
2015         /* clear contents of buffer_info */
2016         rx_buffer->page = NULL;
2017
2018         return skb;
2019 }
2020
2021 /**
2022  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2023  * @q_vector: structure containing interrupt and ring information
2024  * @rx_ring: rx descriptor ring to transact packets on
2025  * @budget: Total limit on number of packets to process
2026  *
2027  * This function provides a "bounce buffer" approach to Rx interrupt
2028  * processing.  The advantage to this is that on systems that have
2029  * expensive overhead for IOMMU access this provides a means of avoiding
2030  * it by maintaining the mapping of the page to the syste.
2031  *
2032  * Returns amount of work completed
2033  **/
2034 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2035                                struct ixgbe_ring *rx_ring,
2036                                const int budget)
2037 {
2038         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2039 #ifdef IXGBE_FCOE
2040         struct ixgbe_adapter *adapter = q_vector->adapter;
2041         int ddp_bytes;
2042         unsigned int mss = 0;
2043 #endif /* IXGBE_FCOE */
2044         u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2045
2046         while (likely(total_rx_packets < budget)) {
2047                 union ixgbe_adv_rx_desc *rx_desc;
2048                 struct sk_buff *skb;
2049
2050                 /* return some buffers to hardware, one at a time is too slow */
2051                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2052                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2053                         cleaned_count = 0;
2054                 }
2055
2056                 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2057
2058                 if (!rx_desc->wb.upper.status_error)
2059                         break;
2060
2061                 /* This memory barrier is needed to keep us from reading
2062                  * any other fields out of the rx_desc until we know the
2063                  * descriptor has been written back
2064                  */
2065                 dma_rmb();
2066
2067                 /* retrieve a buffer from the ring */
2068                 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2069
2070                 /* exit if we failed to retrieve a buffer */
2071                 if (!skb)
2072                         break;
2073
2074                 cleaned_count++;
2075
2076                 /* place incomplete frames back on ring for completion */
2077                 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2078                         continue;
2079
2080                 /* verify the packet layout is correct */
2081                 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2082                         continue;
2083
2084                 /* probably a little skewed due to removing CRC */
2085                 total_rx_bytes += skb->len;
2086
2087                 /* populate checksum, timestamp, VLAN, and protocol */
2088                 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2089
2090 #ifdef IXGBE_FCOE
2091                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2092                 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2093                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2094                         /* include DDPed FCoE data */
2095                         if (ddp_bytes > 0) {
2096                                 if (!mss) {
2097                                         mss = rx_ring->netdev->mtu -
2098                                                 sizeof(struct fcoe_hdr) -
2099                                                 sizeof(struct fc_frame_header) -
2100                                                 sizeof(struct fcoe_crc_eof);
2101                                         if (mss > 512)
2102                                                 mss &= ~511;
2103                                 }
2104                                 total_rx_bytes += ddp_bytes;
2105                                 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2106                                                                  mss);
2107                         }
2108                         if (!ddp_bytes) {
2109                                 dev_kfree_skb_any(skb);
2110                                 continue;
2111                         }
2112                 }
2113
2114 #endif /* IXGBE_FCOE */
2115                 skb_mark_napi_id(skb, &q_vector->napi);
2116                 ixgbe_rx_skb(q_vector, skb);
2117
2118                 /* update budget accounting */
2119                 total_rx_packets++;
2120         }
2121
2122         u64_stats_update_begin(&rx_ring->syncp);
2123         rx_ring->stats.packets += total_rx_packets;
2124         rx_ring->stats.bytes += total_rx_bytes;
2125         u64_stats_update_end(&rx_ring->syncp);
2126         q_vector->rx.total_packets += total_rx_packets;
2127         q_vector->rx.total_bytes += total_rx_bytes;
2128
2129         return total_rx_packets;
2130 }
2131
2132 #ifdef CONFIG_NET_RX_BUSY_POLL
2133 /* must be called with local_bh_disable()d */
2134 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2135 {
2136         struct ixgbe_q_vector *q_vector =
2137                         container_of(napi, struct ixgbe_q_vector, napi);
2138         struct ixgbe_adapter *adapter = q_vector->adapter;
2139         struct ixgbe_ring  *ring;
2140         int found = 0;
2141
2142         if (test_bit(__IXGBE_DOWN, &adapter->state))
2143                 return LL_FLUSH_FAILED;
2144
2145         if (!ixgbe_qv_lock_poll(q_vector))
2146                 return LL_FLUSH_BUSY;
2147
2148         ixgbe_for_each_ring(ring, q_vector->rx) {
2149                 found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2150 #ifdef BP_EXTENDED_STATS
2151                 if (found)
2152                         ring->stats.cleaned += found;
2153                 else
2154                         ring->stats.misses++;
2155 #endif
2156                 if (found)
2157                         break;
2158         }
2159
2160         ixgbe_qv_unlock_poll(q_vector);
2161
2162         return found;
2163 }
2164 #endif  /* CONFIG_NET_RX_BUSY_POLL */
2165
2166 /**
2167  * ixgbe_configure_msix - Configure MSI-X hardware
2168  * @adapter: board private structure
2169  *
2170  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2171  * interrupts.
2172  **/
2173 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2174 {
2175         struct ixgbe_q_vector *q_vector;
2176         int v_idx;
2177         u32 mask;
2178
2179         /* Populate MSIX to EITR Select */
2180         if (adapter->num_vfs > 32) {
2181                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2182                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2183         }
2184
2185         /*
2186          * Populate the IVAR table and set the ITR values to the
2187          * corresponding register.
2188          */
2189         for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2190                 struct ixgbe_ring *ring;
2191                 q_vector = adapter->q_vector[v_idx];
2192
2193                 ixgbe_for_each_ring(ring, q_vector->rx)
2194                         ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2195
2196                 ixgbe_for_each_ring(ring, q_vector->tx)
2197                         ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2198
2199                 ixgbe_write_eitr(q_vector);
2200         }
2201
2202         switch (adapter->hw.mac.type) {
2203         case ixgbe_mac_82598EB:
2204                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2205                                v_idx);
2206                 break;
2207         case ixgbe_mac_82599EB:
2208         case ixgbe_mac_X540:
2209         case ixgbe_mac_X550:
2210         case ixgbe_mac_X550EM_x:
2211                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2212                 break;
2213         default:
2214                 break;
2215         }
2216         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2217
2218         /* set up to autoclear timer, and the vectors */
2219         mask = IXGBE_EIMS_ENABLE_MASK;
2220         mask &= ~(IXGBE_EIMS_OTHER |
2221                   IXGBE_EIMS_MAILBOX |
2222                   IXGBE_EIMS_LSC);
2223
2224         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2225 }
2226
2227 enum latency_range {
2228         lowest_latency = 0,
2229         low_latency = 1,
2230         bulk_latency = 2,
2231         latency_invalid = 255
2232 };
2233
2234 /**
2235  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2236  * @q_vector: structure containing interrupt and ring information
2237  * @ring_container: structure containing ring performance data
2238  *
2239  *      Stores a new ITR value based on packets and byte
2240  *      counts during the last interrupt.  The advantage of per interrupt
2241  *      computation is faster updates and more accurate ITR for the current
2242  *      traffic pattern.  Constants in this function were computed
2243  *      based on theoretical maximum wire speed and thresholds were set based
2244  *      on testing data as well as attempting to minimize response time
2245  *      while increasing bulk throughput.
2246  *      this functionality is controlled by the InterruptThrottleRate module
2247  *      parameter (see ixgbe_param.c)
2248  **/
2249 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2250                              struct ixgbe_ring_container *ring_container)
2251 {
2252         int bytes = ring_container->total_bytes;
2253         int packets = ring_container->total_packets;
2254         u32 timepassed_us;
2255         u64 bytes_perint;
2256         u8 itr_setting = ring_container->itr;
2257
2258         if (packets == 0)
2259                 return;
2260
2261         /* simple throttlerate management
2262          *   0-10MB/s   lowest (100000 ints/s)
2263          *  10-20MB/s   low    (20000 ints/s)
2264          *  20-1249MB/s bulk   (8000 ints/s)
2265          */
2266         /* what was last interrupt timeslice? */
2267         timepassed_us = q_vector->itr >> 2;
2268         if (timepassed_us == 0)
2269                 return;
2270
2271         bytes_perint = bytes / timepassed_us; /* bytes/usec */
2272
2273         switch (itr_setting) {
2274         case lowest_latency:
2275                 if (bytes_perint > 10)
2276                         itr_setting = low_latency;
2277                 break;
2278         case low_latency:
2279                 if (bytes_perint > 20)
2280                         itr_setting = bulk_latency;
2281                 else if (bytes_perint <= 10)
2282                         itr_setting = lowest_latency;
2283                 break;
2284         case bulk_latency:
2285                 if (bytes_perint <= 20)
2286                         itr_setting = low_latency;
2287                 break;
2288         }
2289
2290         /* clear work counters since we have the values we need */
2291         ring_container->total_bytes = 0;
2292         ring_container->total_packets = 0;
2293
2294         /* write updated itr to ring container */
2295         ring_container->itr = itr_setting;
2296 }
2297
2298 /**
2299  * ixgbe_write_eitr - write EITR register in hardware specific way
2300  * @q_vector: structure containing interrupt and ring information
2301  *
2302  * This function is made to be called by ethtool and by the driver
2303  * when it needs to update EITR registers at runtime.  Hardware
2304  * specific quirks/differences are taken care of here.
2305  */
2306 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2307 {
2308         struct ixgbe_adapter *adapter = q_vector->adapter;
2309         struct ixgbe_hw *hw = &adapter->hw;
2310         int v_idx = q_vector->v_idx;
2311         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2312
2313         switch (adapter->hw.mac.type) {
2314         case ixgbe_mac_82598EB:
2315                 /* must write high and low 16 bits to reset counter */
2316                 itr_reg |= (itr_reg << 16);
2317                 break;
2318         case ixgbe_mac_82599EB:
2319         case ixgbe_mac_X540:
2320         case ixgbe_mac_X550:
2321         case ixgbe_mac_X550EM_x:
2322                 /*
2323                  * set the WDIS bit to not clear the timer bits and cause an
2324                  * immediate assertion of the interrupt
2325                  */
2326                 itr_reg |= IXGBE_EITR_CNT_WDIS;
2327                 break;
2328         default:
2329                 break;
2330         }
2331         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2332 }
2333
2334 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2335 {
2336         u32 new_itr = q_vector->itr;
2337         u8 current_itr;
2338
2339         ixgbe_update_itr(q_vector, &q_vector->tx);
2340         ixgbe_update_itr(q_vector, &q_vector->rx);
2341
2342         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2343
2344         switch (current_itr) {
2345         /* counts and packets in update_itr are dependent on these numbers */
2346         case lowest_latency:
2347                 new_itr = IXGBE_100K_ITR;
2348                 break;
2349         case low_latency:
2350                 new_itr = IXGBE_20K_ITR;
2351                 break;
2352         case bulk_latency:
2353                 new_itr = IXGBE_8K_ITR;
2354                 break;
2355         default:
2356                 break;
2357         }
2358
2359         if (new_itr != q_vector->itr) {
2360                 /* do an exponential smoothing */
2361                 new_itr = (10 * new_itr * q_vector->itr) /
2362                           ((9 * new_itr) + q_vector->itr);
2363
2364                 /* save the algorithm value here */
2365                 q_vector->itr = new_itr;
2366
2367                 ixgbe_write_eitr(q_vector);
2368         }
2369 }
2370
2371 /**
2372  * ixgbe_check_overtemp_subtask - check for over temperature
2373  * @adapter: pointer to adapter
2374  **/
2375 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2376 {
2377         struct ixgbe_hw *hw = &adapter->hw;
2378         u32 eicr = adapter->interrupt_event;
2379
2380         if (test_bit(__IXGBE_DOWN, &adapter->state))
2381                 return;
2382
2383         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2384             !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2385                 return;
2386
2387         adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2388
2389         switch (hw->device_id) {
2390         case IXGBE_DEV_ID_82599_T3_LOM:
2391                 /*
2392                  * Since the warning interrupt is for both ports
2393                  * we don't have to check if:
2394                  *  - This interrupt wasn't for our port.
2395                  *  - We may have missed the interrupt so always have to
2396                  *    check if we  got a LSC
2397                  */
2398                 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2399                     !(eicr & IXGBE_EICR_LSC))
2400                         return;
2401
2402                 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2403                         u32 speed;
2404                         bool link_up = false;
2405
2406                         hw->mac.ops.check_link(hw, &speed, &link_up, false);
2407
2408                         if (link_up)
2409                                 return;
2410                 }
2411
2412                 /* Check if this is not due to overtemp */
2413                 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2414                         return;
2415
2416                 break;
2417         default:
2418                 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2419                         return;
2420                 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2421                         return;
2422                 break;
2423         }
2424         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2425
2426         adapter->interrupt_event = 0;
2427 }
2428
2429 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2430 {
2431         struct ixgbe_hw *hw = &adapter->hw;
2432
2433         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2434             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2435                 e_crit(probe, "Fan has stopped, replace the adapter\n");
2436                 /* write to clear the interrupt */
2437                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2438         }
2439 }
2440
2441 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2442 {
2443         struct ixgbe_hw *hw = &adapter->hw;
2444
2445         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2446                 return;
2447
2448         switch (adapter->hw.mac.type) {
2449         case ixgbe_mac_82599EB:
2450                 /*
2451                  * Need to check link state so complete overtemp check
2452                  * on service task
2453                  */
2454                 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2455                      (eicr & IXGBE_EICR_LSC)) &&
2456                     (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2457                         adapter->interrupt_event = eicr;
2458                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2459                         ixgbe_service_event_schedule(adapter);
2460                         return;
2461                 }
2462                 return;
2463         case ixgbe_mac_X540:
2464                 if (!(eicr & IXGBE_EICR_TS))
2465                         return;
2466                 break;
2467         default:
2468                 return;
2469         }
2470
2471         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2472 }
2473
2474 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2475 {
2476         switch (hw->mac.type) {
2477         case ixgbe_mac_82598EB:
2478                 if (hw->phy.type == ixgbe_phy_nl)
2479                         return true;
2480                 return false;
2481         case ixgbe_mac_82599EB:
2482         case ixgbe_mac_X550EM_x:
2483                 switch (hw->mac.ops.get_media_type(hw)) {
2484                 case ixgbe_media_type_fiber:
2485                 case ixgbe_media_type_fiber_qsfp:
2486                         return true;
2487                 default:
2488                         return false;
2489                 }
2490         default:
2491                 return false;
2492         }
2493 }
2494
2495 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2496 {
2497         struct ixgbe_hw *hw = &adapter->hw;
2498
2499         if (eicr & IXGBE_EICR_GPI_SDP2(hw)) {
2500                 /* Clear the interrupt */
2501                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP2(hw));
2502                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2503                         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2504                         ixgbe_service_event_schedule(adapter);
2505                 }
2506         }
2507
2508         if (eicr & IXGBE_EICR_GPI_SDP1(hw)) {
2509                 /* Clear the interrupt */
2510                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2511                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2512                         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2513                         ixgbe_service_event_schedule(adapter);
2514                 }
2515         }
2516 }
2517
2518 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2519 {
2520         struct ixgbe_hw *hw = &adapter->hw;
2521
2522         adapter->lsc_int++;
2523         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2524         adapter->link_check_timeout = jiffies;
2525         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2526                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2527                 IXGBE_WRITE_FLUSH(hw);
2528                 ixgbe_service_event_schedule(adapter);
2529         }
2530 }
2531
2532 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2533                                            u64 qmask)
2534 {
2535         u32 mask;
2536         struct ixgbe_hw *hw = &adapter->hw;
2537
2538         switch (hw->mac.type) {
2539         case ixgbe_mac_82598EB:
2540                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2541                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2542                 break;
2543         case ixgbe_mac_82599EB:
2544         case ixgbe_mac_X540:
2545         case ixgbe_mac_X550:
2546         case ixgbe_mac_X550EM_x:
2547                 mask = (qmask & 0xFFFFFFFF);
2548                 if (mask)
2549                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2550                 mask = (qmask >> 32);
2551                 if (mask)
2552                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2553                 break;
2554         default:
2555                 break;
2556         }
2557         /* skip the flush */
2558 }
2559
2560 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2561                                             u64 qmask)
2562 {
2563         u32 mask;
2564         struct ixgbe_hw *hw = &adapter->hw;
2565
2566         switch (hw->mac.type) {
2567         case ixgbe_mac_82598EB:
2568                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2569                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2570                 break;
2571         case ixgbe_mac_82599EB:
2572         case ixgbe_mac_X540:
2573         case ixgbe_mac_X550:
2574         case ixgbe_mac_X550EM_x:
2575                 mask = (qmask & 0xFFFFFFFF);
2576                 if (mask)
2577                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2578                 mask = (qmask >> 32);
2579                 if (mask)
2580                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2581                 break;
2582         default:
2583                 break;
2584         }
2585         /* skip the flush */
2586 }
2587
2588 /**
2589  * ixgbe_irq_enable - Enable default interrupt generation settings
2590  * @adapter: board private structure
2591  **/
2592 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2593                                     bool flush)
2594 {
2595         struct ixgbe_hw *hw = &adapter->hw;
2596         u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2597
2598         /* don't reenable LSC while waiting for link */
2599         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2600                 mask &= ~IXGBE_EIMS_LSC;
2601
2602         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2603                 switch (adapter->hw.mac.type) {
2604                 case ixgbe_mac_82599EB:
2605                         mask |= IXGBE_EIMS_GPI_SDP0(hw);
2606                         break;
2607                 case ixgbe_mac_X540:
2608                 case ixgbe_mac_X550:
2609                 case ixgbe_mac_X550EM_x:
2610                         mask |= IXGBE_EIMS_TS;
2611                         break;
2612                 default:
2613                         break;
2614                 }
2615         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2616                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2617         switch (adapter->hw.mac.type) {
2618         case ixgbe_mac_82599EB:
2619                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2620                 mask |= IXGBE_EIMS_GPI_SDP2(hw);
2621                 /* fall through */
2622         case ixgbe_mac_X540:
2623         case ixgbe_mac_X550:
2624         case ixgbe_mac_X550EM_x:
2625                 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
2626                         mask |= IXGBE_EICR_GPI_SDP0_X540;
2627                 mask |= IXGBE_EIMS_ECC;
2628                 mask |= IXGBE_EIMS_MAILBOX;
2629                 break;
2630         default:
2631                 break;
2632         }
2633
2634         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2635             !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2636                 mask |= IXGBE_EIMS_FLOW_DIR;
2637
2638         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2639         if (queues)
2640                 ixgbe_irq_enable_queues(adapter, ~0);
2641         if (flush)
2642                 IXGBE_WRITE_FLUSH(&adapter->hw);
2643 }
2644
2645 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2646 {
2647         struct ixgbe_adapter *adapter = data;
2648         struct ixgbe_hw *hw = &adapter->hw;
2649         u32 eicr;
2650
2651         /*
2652          * Workaround for Silicon errata.  Use clear-by-write instead
2653          * of clear-by-read.  Reading with EICS will return the
2654          * interrupt causes without clearing, which later be done
2655          * with the write to EICR.
2656          */
2657         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2658
2659         /* The lower 16bits of the EICR register are for the queue interrupts
2660          * which should be masked here in order to not accidentally clear them if
2661          * the bits are high when ixgbe_msix_other is called. There is a race
2662          * condition otherwise which results in possible performance loss
2663          * especially if the ixgbe_msix_other interrupt is triggering
2664          * consistently (as it would when PPS is turned on for the X540 device)
2665          */
2666         eicr &= 0xFFFF0000;
2667
2668         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2669
2670         if (eicr & IXGBE_EICR_LSC)
2671                 ixgbe_check_lsc(adapter);
2672
2673         if (eicr & IXGBE_EICR_MAILBOX)
2674                 ixgbe_msg_task(adapter);
2675
2676         switch (hw->mac.type) {
2677         case ixgbe_mac_82599EB:
2678         case ixgbe_mac_X540:
2679         case ixgbe_mac_X550:
2680         case ixgbe_mac_X550EM_x:
2681                 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
2682                     (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
2683                         adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
2684                         ixgbe_service_event_schedule(adapter);
2685                         IXGBE_WRITE_REG(hw, IXGBE_EICR,
2686                                         IXGBE_EICR_GPI_SDP0_X540);
2687                 }
2688                 if (eicr & IXGBE_EICR_ECC) {
2689                         e_info(link, "Received ECC Err, initiating reset\n");
2690                         adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2691                         ixgbe_service_event_schedule(adapter);
2692                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2693                 }
2694                 /* Handle Flow Director Full threshold interrupt */
2695                 if (eicr & IXGBE_EICR_FLOW_DIR) {
2696                         int reinit_count = 0;
2697                         int i;
2698                         for (i = 0; i < adapter->num_tx_queues; i++) {
2699                                 struct ixgbe_ring *ring = adapter->tx_ring[i];
2700                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2701                                                        &ring->state))
2702                                         reinit_count++;
2703                         }
2704                         if (reinit_count) {
2705                                 /* no more flow director interrupts until after init */
2706                                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2707                                 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2708                                 ixgbe_service_event_schedule(adapter);
2709                         }
2710                 }
2711                 ixgbe_check_sfp_event(adapter, eicr);
2712                 ixgbe_check_overtemp_event(adapter, eicr);
2713                 break;
2714         default:
2715                 break;
2716         }
2717
2718         ixgbe_check_fan_failure(adapter, eicr);
2719
2720         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2721                 ixgbe_ptp_check_pps_event(adapter, eicr);
2722
2723         /* re-enable the original interrupt state, no lsc, no queues */
2724         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2725                 ixgbe_irq_enable(adapter, false, false);
2726
2727         return IRQ_HANDLED;
2728 }
2729
2730 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2731 {
2732         struct ixgbe_q_vector *q_vector = data;
2733
2734         /* EIAM disabled interrupts (on this vector) for us */
2735
2736         if (q_vector->rx.ring || q_vector->tx.ring)
2737                 napi_schedule(&q_vector->napi);
2738
2739         return IRQ_HANDLED;
2740 }
2741
2742 /**
2743  * ixgbe_poll - NAPI Rx polling callback
2744  * @napi: structure for representing this polling device
2745  * @budget: how many packets driver is allowed to clean
2746  *
2747  * This function is used for legacy and MSI, NAPI mode
2748  **/
2749 int ixgbe_poll(struct napi_struct *napi, int budget)
2750 {
2751         struct ixgbe_q_vector *q_vector =
2752                                 container_of(napi, struct ixgbe_q_vector, napi);
2753         struct ixgbe_adapter *adapter = q_vector->adapter;
2754         struct ixgbe_ring *ring;
2755         int per_ring_budget;
2756         bool clean_complete = true;
2757
2758 #ifdef CONFIG_IXGBE_DCA
2759         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2760                 ixgbe_update_dca(q_vector);
2761 #endif
2762
2763         ixgbe_for_each_ring(ring, q_vector->tx)
2764                 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring);
2765
2766         if (!ixgbe_qv_lock_napi(q_vector))
2767                 return budget;
2768
2769         /* attempt to distribute budget to each queue fairly, but don't allow
2770          * the budget to go below 1 because we'll exit polling */
2771         if (q_vector->rx.count > 1)
2772                 per_ring_budget = max(budget/q_vector->rx.count, 1);
2773         else
2774                 per_ring_budget = budget;
2775
2776         ixgbe_for_each_ring(ring, q_vector->rx)
2777                 clean_complete &= (ixgbe_clean_rx_irq(q_vector, ring,
2778                                    per_ring_budget) < per_ring_budget);
2779
2780         ixgbe_qv_unlock_napi(q_vector);
2781         /* If all work not completed, return budget and keep polling */
2782         if (!clean_complete)
2783                 return budget;
2784
2785         /* all work done, exit the polling mode */
2786         napi_complete(napi);
2787         if (adapter->rx_itr_setting & 1)
2788                 ixgbe_set_itr(q_vector);
2789         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2790                 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2791
2792         return 0;
2793 }
2794
2795 /**
2796  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2797  * @adapter: board private structure
2798  *
2799  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2800  * interrupts from the kernel.
2801  **/
2802 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2803 {
2804         struct net_device *netdev = adapter->netdev;
2805         int vector, err;
2806         int ri = 0, ti = 0;
2807
2808         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2809                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2810                 struct msix_entry *entry = &adapter->msix_entries[vector];
2811
2812                 if (q_vector->tx.ring && q_vector->rx.ring) {
2813                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2814                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
2815                         ti++;
2816                 } else if (q_vector->rx.ring) {
2817                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2818                                  "%s-%s-%d", netdev->name, "rx", ri++);
2819                 } else if (q_vector->tx.ring) {
2820                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2821                                  "%s-%s-%d", netdev->name, "tx", ti++);
2822                 } else {
2823                         /* skip this unused q_vector */
2824                         continue;
2825                 }
2826                 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2827                                   q_vector->name, q_vector);
2828                 if (err) {
2829                         e_err(probe, "request_irq failed for MSIX interrupt "
2830                               "Error: %d\n", err);
2831                         goto free_queue_irqs;
2832                 }
2833                 /* If Flow Director is enabled, set interrupt affinity */
2834                 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2835                         /* assign the mask for this irq */
2836                         irq_set_affinity_hint(entry->vector,
2837                                               &q_vector->affinity_mask);
2838                 }
2839         }
2840
2841         err = request_irq(adapter->msix_entries[vector].vector,
2842                           ixgbe_msix_other, 0, netdev->name, adapter);
2843         if (err) {
2844                 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2845                 goto free_queue_irqs;
2846         }
2847
2848         return 0;
2849
2850 free_queue_irqs:
2851         while (vector) {
2852                 vector--;
2853                 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2854                                       NULL);
2855                 free_irq(adapter->msix_entries[vector].vector,
2856                          adapter->q_vector[vector]);
2857         }
2858         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2859         pci_disable_msix(adapter->pdev);
2860         kfree(adapter->msix_entries);
2861         adapter->msix_entries = NULL;
2862         return err;
2863 }
2864
2865 /**
2866  * ixgbe_intr - legacy mode Interrupt Handler
2867  * @irq: interrupt number
2868  * @data: pointer to a network interface device structure
2869  **/
2870 static irqreturn_t ixgbe_intr(int irq, void *data)
2871 {
2872         struct ixgbe_adapter *adapter = data;
2873         struct ixgbe_hw *hw = &adapter->hw;
2874         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2875         u32 eicr;
2876
2877         /*
2878          * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2879          * before the read of EICR.
2880          */
2881         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2882
2883         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2884          * therefore no explicit interrupt disable is necessary */
2885         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2886         if (!eicr) {
2887                 /*
2888                  * shared interrupt alert!
2889                  * make sure interrupts are enabled because the read will
2890                  * have disabled interrupts due to EIAM
2891                  * finish the workaround of silicon errata on 82598.  Unmask
2892                  * the interrupt that we masked before the EICR read.
2893                  */
2894                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2895                         ixgbe_irq_enable(adapter, true, true);
2896                 return IRQ_NONE;        /* Not our interrupt */
2897         }
2898
2899         if (eicr & IXGBE_EICR_LSC)
2900                 ixgbe_check_lsc(adapter);
2901
2902         switch (hw->mac.type) {
2903         case ixgbe_mac_82599EB:
2904                 ixgbe_check_sfp_event(adapter, eicr);
2905                 /* Fall through */
2906         case ixgbe_mac_X540:
2907         case ixgbe_mac_X550:
2908         case ixgbe_mac_X550EM_x:
2909                 if (eicr & IXGBE_EICR_ECC) {
2910                         e_info(link, "Received ECC Err, initiating reset\n");
2911                         adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2912                         ixgbe_service_event_schedule(adapter);
2913                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2914                 }
2915                 ixgbe_check_overtemp_event(adapter, eicr);
2916                 break;
2917         default:
2918                 break;
2919         }
2920
2921         ixgbe_check_fan_failure(adapter, eicr);
2922         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2923                 ixgbe_ptp_check_pps_event(adapter, eicr);
2924
2925         /* would disable interrupts here but EIAM disabled it */
2926         napi_schedule(&q_vector->napi);
2927
2928         /*
2929          * re-enable link(maybe) and non-queue interrupts, no flush.
2930          * ixgbe_poll will re-enable the queue interrupts
2931          */
2932         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2933                 ixgbe_irq_enable(adapter, false, false);
2934
2935         return IRQ_HANDLED;
2936 }
2937
2938 /**
2939  * ixgbe_request_irq - initialize interrupts
2940  * @adapter: board private structure
2941  *
2942  * Attempts to configure interrupts using the best available
2943  * capabilities of the hardware and kernel.
2944  **/
2945 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2946 {
2947         struct net_device *netdev = adapter->netdev;
2948         int err;
2949
2950         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2951                 err = ixgbe_request_msix_irqs(adapter);
2952         else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2953                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2954                                   netdev->name, adapter);
2955         else
2956                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2957                                   netdev->name, adapter);
2958
2959         if (err)
2960                 e_err(probe, "request_irq failed, Error %d\n", err);
2961
2962         return err;
2963 }
2964
2965 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2966 {
2967         int vector;
2968
2969         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
2970                 free_irq(adapter->pdev->irq, adapter);
2971                 return;
2972         }
2973
2974         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2975                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2976                 struct msix_entry *entry = &adapter->msix_entries[vector];
2977
2978                 /* free only the irqs that were actually requested */
2979                 if (!q_vector->rx.ring && !q_vector->tx.ring)
2980                         continue;
2981
2982                 /* clear the affinity_mask in the IRQ descriptor */
2983                 irq_set_affinity_hint(entry->vector, NULL);
2984
2985                 free_irq(entry->vector, q_vector);
2986         }
2987
2988         free_irq(adapter->msix_entries[vector++].vector, adapter);
2989 }
2990
2991 /**
2992  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
2993  * @adapter: board private structure
2994  **/
2995 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
2996 {
2997         switch (adapter->hw.mac.type) {
2998         case ixgbe_mac_82598EB:
2999                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3000                 break;
3001         case ixgbe_mac_82599EB:
3002         case ixgbe_mac_X540:
3003         case ixgbe_mac_X550:
3004         case ixgbe_mac_X550EM_x:
3005                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3006                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3007                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3008                 break;
3009         default:
3010                 break;
3011         }
3012         IXGBE_WRITE_FLUSH(&adapter->hw);
3013         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3014                 int vector;
3015
3016                 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3017                         synchronize_irq(adapter->msix_entries[vector].vector);
3018
3019                 synchronize_irq(adapter->msix_entries[vector++].vector);
3020         } else {
3021                 synchronize_irq(adapter->pdev->irq);
3022         }
3023 }
3024
3025 /**
3026  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3027  *
3028  **/
3029 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3030 {
3031         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3032
3033         ixgbe_write_eitr(q_vector);
3034
3035         ixgbe_set_ivar(adapter, 0, 0, 0);
3036         ixgbe_set_ivar(adapter, 1, 0, 0);
3037
3038         e_info(hw, "Legacy interrupt IVAR setup done\n");
3039 }
3040
3041 /**
3042  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3043  * @adapter: board private structure
3044  * @ring: structure containing ring specific data
3045  *
3046  * Configure the Tx descriptor ring after a reset.
3047  **/
3048 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3049                              struct ixgbe_ring *ring)
3050 {
3051         struct ixgbe_hw *hw = &adapter->hw;
3052         u64 tdba = ring->dma;
3053         int wait_loop = 10;
3054         u32 txdctl = IXGBE_TXDCTL_ENABLE;
3055         u8 reg_idx = ring->reg_idx;
3056
3057         /* disable queue to avoid issues while updating state */
3058         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3059         IXGBE_WRITE_FLUSH(hw);
3060
3061         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3062                         (tdba & DMA_BIT_MASK(32)));
3063         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3064         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3065                         ring->count * sizeof(union ixgbe_adv_tx_desc));
3066         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3067         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3068         ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3069
3070         /*
3071          * set WTHRESH to encourage burst writeback, it should not be set
3072          * higher than 1 when:
3073          * - ITR is 0 as it could cause false TX hangs
3074          * - ITR is set to > 100k int/sec and BQL is enabled
3075          *
3076          * In order to avoid issues WTHRESH + PTHRESH should always be equal
3077          * to or less than the number of on chip descriptors, which is
3078          * currently 40.
3079          */
3080         if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3081                 txdctl |= (1 << 16);    /* WTHRESH = 1 */
3082         else
3083                 txdctl |= (8 << 16);    /* WTHRESH = 8 */
3084
3085         /*
3086          * Setting PTHRESH to 32 both improves performance
3087          * and avoids a TX hang with DFP enabled
3088          */
3089         txdctl |= (1 << 8) |    /* HTHRESH = 1 */
3090                    32;          /* PTHRESH = 32 */
3091
3092         /* reinitialize flowdirector state */
3093         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3094                 ring->atr_sample_rate = adapter->atr_sample_rate;
3095                 ring->atr_count = 0;
3096                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3097         } else {
3098                 ring->atr_sample_rate = 0;
3099         }
3100
3101         /* initialize XPS */
3102         if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3103                 struct ixgbe_q_vector *q_vector = ring->q_vector;
3104
3105                 if (q_vector)
3106                         netif_set_xps_queue(ring->netdev,
3107                                             &q_vector->affinity_mask,
3108                                             ring->queue_index);
3109         }
3110
3111         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3112
3113         /* enable queue */
3114         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3115
3116         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3117         if (hw->mac.type == ixgbe_mac_82598EB &&
3118             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3119                 return;
3120
3121         /* poll to verify queue is enabled */
3122         do {
3123                 usleep_range(1000, 2000);
3124                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3125         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3126         if (!wait_loop)
3127                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3128 }
3129
3130 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3131 {
3132         struct ixgbe_hw *hw = &adapter->hw;
3133         u32 rttdcs, mtqc;
3134         u8 tcs = netdev_get_num_tc(adapter->netdev);
3135
3136         if (hw->mac.type == ixgbe_mac_82598EB)
3137                 return;
3138
3139         /* disable the arbiter while setting MTQC */
3140         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3141         rttdcs |= IXGBE_RTTDCS_ARBDIS;
3142         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3143
3144         /* set transmit pool layout */
3145         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3146                 mtqc = IXGBE_MTQC_VT_ENA;
3147                 if (tcs > 4)
3148                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3149                 else if (tcs > 1)
3150                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3151                 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3152                         mtqc |= IXGBE_MTQC_32VF;
3153                 else
3154                         mtqc |= IXGBE_MTQC_64VF;
3155         } else {
3156                 if (tcs > 4)
3157                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3158                 else if (tcs > 1)
3159                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3160                 else
3161                         mtqc = IXGBE_MTQC_64Q_1PB;
3162         }
3163
3164         IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3165
3166         /* Enable Security TX Buffer IFG for multiple pb */
3167         if (tcs) {
3168                 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3169                 sectx |= IXGBE_SECTX_DCB;
3170                 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3171         }
3172
3173         /* re-enable the arbiter */
3174         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3175         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3176 }
3177
3178 /**
3179  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3180  * @adapter: board private structure
3181  *
3182  * Configure the Tx unit of the MAC after a reset.
3183  **/
3184 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3185 {
3186         struct ixgbe_hw *hw = &adapter->hw;
3187         u32 dmatxctl;
3188         u32 i;
3189
3190         ixgbe_setup_mtqc(adapter);
3191
3192         if (hw->mac.type != ixgbe_mac_82598EB) {
3193                 /* DMATXCTL.EN must be before Tx queues are enabled */
3194                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3195                 dmatxctl |= IXGBE_DMATXCTL_TE;
3196                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3197         }
3198
3199         /* Setup the HW Tx Head and Tail descriptor pointers */
3200         for (i = 0; i < adapter->num_tx_queues; i++)
3201                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3202 }
3203
3204 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3205                                  struct ixgbe_ring *ring)
3206 {
3207         struct ixgbe_hw *hw = &adapter->hw;
3208         u8 reg_idx = ring->reg_idx;
3209         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3210
3211         srrctl |= IXGBE_SRRCTL_DROP_EN;
3212
3213         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3214 }
3215
3216 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3217                                   struct ixgbe_ring *ring)
3218 {
3219         struct ixgbe_hw *hw = &adapter->hw;
3220         u8 reg_idx = ring->reg_idx;
3221         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3222
3223         srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3224
3225         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3226 }
3227
3228 #ifdef CONFIG_IXGBE_DCB
3229 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3230 #else
3231 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3232 #endif
3233 {
3234         int i;
3235         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3236
3237         if (adapter->ixgbe_ieee_pfc)
3238                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3239
3240         /*
3241          * We should set the drop enable bit if:
3242          *  SR-IOV is enabled
3243          *   or
3244          *  Number of Rx queues > 1 and flow control is disabled
3245          *
3246          *  This allows us to avoid head of line blocking for security
3247          *  and performance reasons.
3248          */
3249         if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3250             !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3251                 for (i = 0; i < adapter->num_rx_queues; i++)
3252                         ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3253         } else {
3254                 for (i = 0; i < adapter->num_rx_queues; i++)
3255                         ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3256         }
3257 }
3258
3259 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3260
3261 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3262                                    struct ixgbe_ring *rx_ring)
3263 {
3264         struct ixgbe_hw *hw = &adapter->hw;
3265         u32 srrctl;
3266         u8 reg_idx = rx_ring->reg_idx;
3267
3268         if (hw->mac.type == ixgbe_mac_82598EB) {
3269                 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3270
3271                 /*
3272                  * if VMDq is not active we must program one srrctl register
3273                  * per RSS queue since we have enabled RDRXCTL.MVMEN
3274                  */
3275                 reg_idx &= mask;
3276         }
3277
3278         /* configure header buffer length, needed for RSC */
3279         srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3280
3281         /* configure the packet buffer length */
3282         srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3283
3284         /* configure descriptor type */
3285         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3286
3287         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3288 }
3289
3290 /**
3291  * Return a number of entries in the RSS indirection table
3292  *
3293  * @adapter: device handle
3294  *
3295  *  - 82598/82599/X540:     128
3296  *  - X550(non-SRIOV mode): 512
3297  *  - X550(SRIOV mode):     64
3298  */
3299 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3300 {
3301         if (adapter->hw.mac.type < ixgbe_mac_X550)
3302                 return 128;
3303         else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3304                 return 64;
3305         else
3306                 return 512;
3307 }
3308
3309 /**
3310  * Write the RETA table to HW
3311  *
3312  * @adapter: device handle
3313  *
3314  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3315  */
3316 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3317 {
3318         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3319         struct ixgbe_hw *hw = &adapter->hw;
3320         u32 reta = 0;
3321         u32 indices_multi;
3322         u8 *indir_tbl = adapter->rss_indir_tbl;
3323
3324         /* Fill out the redirection table as follows:
3325          *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3326          *    indices.
3327          *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3328          *  - X550:       8 bit wide entries containing 6 bit RSS index
3329          */
3330         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3331                 indices_multi = 0x11;
3332         else
3333                 indices_multi = 0x1;
3334
3335         /* Write redirection table to HW */
3336         for (i = 0; i < reta_entries; i++) {
3337                 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3338                 if ((i & 3) == 3) {
3339                         if (i < 128)
3340                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3341                         else
3342                                 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3343                                                 reta);
3344                         reta = 0;
3345                 }
3346         }
3347 }
3348
3349 /**
3350  * Write the RETA table to HW (for x550 devices in SRIOV mode)
3351  *
3352  * @adapter: device handle
3353  *
3354  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3355  */
3356 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3357 {
3358         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3359         struct ixgbe_hw *hw = &adapter->hw;
3360         u32 vfreta = 0;
3361         unsigned int pf_pool = adapter->num_vfs;
3362
3363         /* Write redirection table to HW */
3364         for (i = 0; i < reta_entries; i++) {
3365                 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3366                 if ((i & 3) == 3) {
3367                         IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3368                                         vfreta);
3369                         vfreta = 0;
3370                 }
3371         }
3372 }
3373
3374 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3375 {
3376         struct ixgbe_hw *hw = &adapter->hw;
3377         u32 i, j;
3378         u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3379         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3380
3381         /* Program table for at least 2 queues w/ SR-IOV so that VFs can
3382          * make full use of any rings they may have.  We will use the
3383          * PSRTYPE register to control how many rings we use within the PF.
3384          */
3385         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3386                 rss_i = 2;
3387
3388         /* Fill out hash function seeds */
3389         for (i = 0; i < 10; i++)
3390                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3391
3392         /* Fill out redirection table */
3393         memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3394
3395         for (i = 0, j = 0; i < reta_entries; i++, j++) {
3396                 if (j == rss_i)
3397                         j = 0;
3398
3399                 adapter->rss_indir_tbl[i] = j;
3400         }
3401
3402         ixgbe_store_reta(adapter);
3403 }
3404
3405 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3406 {
3407         struct ixgbe_hw *hw = &adapter->hw;
3408         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3409         unsigned int pf_pool = adapter->num_vfs;
3410         int i, j;
3411
3412         /* Fill out hash function seeds */
3413         for (i = 0; i < 10; i++)
3414                 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3415                                 adapter->rss_key[i]);
3416
3417         /* Fill out the redirection table */
3418         for (i = 0, j = 0; i < 64; i++, j++) {
3419                 if (j == rss_i)
3420                         j = 0;
3421
3422                 adapter->rss_indir_tbl[i] = j;
3423         }
3424
3425         ixgbe_store_vfreta(adapter);
3426 }
3427
3428 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3429 {
3430         struct ixgbe_hw *hw = &adapter->hw;
3431         u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3432         u32 rxcsum;
3433
3434         /* Disable indicating checksum in descriptor, enables RSS hash */
3435         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3436         rxcsum |= IXGBE_RXCSUM_PCSD;
3437         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3438
3439         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3440                 if (adapter->ring_feature[RING_F_RSS].mask)
3441                         mrqc = IXGBE_MRQC_RSSEN;
3442         } else {
3443                 u8 tcs = netdev_get_num_tc(adapter->netdev);
3444
3445                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3446                         if (tcs > 4)
3447                                 mrqc = IXGBE_MRQC_VMDQRT8TCEN;  /* 8 TCs */
3448                         else if (tcs > 1)
3449                                 mrqc = IXGBE_MRQC_VMDQRT4TCEN;  /* 4 TCs */
3450                         else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3451                                 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3452                         else
3453                                 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3454                 } else {
3455                         if (tcs > 4)
3456                                 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3457                         else if (tcs > 1)
3458                                 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3459                         else
3460                                 mrqc = IXGBE_MRQC_RSSEN;
3461                 }
3462         }
3463
3464         /* Perform hash on these packet types */
3465         rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3466                      IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3467                      IXGBE_MRQC_RSS_FIELD_IPV6 |
3468                      IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3469
3470         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3471                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3472         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3473                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3474
3475         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
3476         if ((hw->mac.type >= ixgbe_mac_X550) &&
3477             (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3478                 unsigned int pf_pool = adapter->num_vfs;
3479
3480                 /* Enable VF RSS mode */
3481                 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3482                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3483
3484                 /* Setup RSS through the VF registers */
3485                 ixgbe_setup_vfreta(adapter);
3486                 vfmrqc = IXGBE_MRQC_RSSEN;
3487                 vfmrqc |= rss_field;
3488                 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3489         } else {
3490                 ixgbe_setup_reta(adapter);
3491                 mrqc |= rss_field;
3492                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3493         }
3494 }
3495
3496 /**
3497  * ixgbe_configure_rscctl - enable RSC for the indicated ring
3498  * @adapter:    address of board private structure
3499  * @index:      index of ring to set
3500  **/
3501 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3502                                    struct ixgbe_ring *ring)
3503 {
3504         struct ixgbe_hw *hw = &adapter->hw;
3505         u32 rscctrl;
3506         u8 reg_idx = ring->reg_idx;
3507
3508         if (!ring_is_rsc_enabled(ring))
3509                 return;
3510
3511         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3512         rscctrl |= IXGBE_RSCCTL_RSCEN;
3513         /*
3514          * we must limit the number of descriptors so that the
3515          * total size of max desc * buf_len is not greater
3516          * than 65536
3517          */
3518         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3519         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3520 }
3521
3522 #define IXGBE_MAX_RX_DESC_POLL 10
3523 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3524                                        struct ixgbe_ring *ring)
3525 {
3526         struct ixgbe_hw *hw = &adapter->hw;
3527         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3528         u32 rxdctl;
3529         u8 reg_idx = ring->reg_idx;
3530
3531         if (ixgbe_removed(hw->hw_addr))
3532                 return;
3533         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3534         if (hw->mac.type == ixgbe_mac_82598EB &&
3535             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3536                 return;
3537
3538         do {
3539                 usleep_range(1000, 2000);
3540                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3541         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3542
3543         if (!wait_loop) {
3544                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3545                       "the polling period\n", reg_idx);
3546         }
3547 }
3548
3549 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3550                             struct ixgbe_ring *ring)
3551 {
3552         struct ixgbe_hw *hw = &adapter->hw;
3553         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3554         u32 rxdctl;
3555         u8 reg_idx = ring->reg_idx;
3556
3557         if (ixgbe_removed(hw->hw_addr))
3558                 return;
3559         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3560         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3561
3562         /* write value back with RXDCTL.ENABLE bit cleared */
3563         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3564
3565         if (hw->mac.type == ixgbe_mac_82598EB &&
3566             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3567                 return;
3568
3569         /* the hardware may take up to 100us to really disable the rx queue */
3570         do {
3571                 udelay(10);
3572                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3573         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3574
3575         if (!wait_loop) {
3576                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3577                       "the polling period\n", reg_idx);
3578         }
3579 }
3580
3581 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3582                              struct ixgbe_ring *ring)
3583 {
3584         struct ixgbe_hw *hw = &adapter->hw;
3585         u64 rdba = ring->dma;
3586         u32 rxdctl;
3587         u8 reg_idx = ring->reg_idx;
3588
3589         /* disable queue to avoid issues while updating state */
3590         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3591         ixgbe_disable_rx_queue(adapter, ring);
3592
3593         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3594         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3595         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3596                         ring->count * sizeof(union ixgbe_adv_rx_desc));
3597         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3598         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3599         ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3600
3601         ixgbe_configure_srrctl(adapter, ring);
3602         ixgbe_configure_rscctl(adapter, ring);
3603
3604         if (hw->mac.type == ixgbe_mac_82598EB) {
3605                 /*
3606                  * enable cache line friendly hardware writes:
3607                  * PTHRESH=32 descriptors (half the internal cache),
3608                  * this also removes ugly rx_no_buffer_count increment
3609                  * HTHRESH=4 descriptors (to minimize latency on fetch)
3610                  * WTHRESH=8 burst writeback up to two cache lines
3611                  */
3612                 rxdctl &= ~0x3FFFFF;
3613                 rxdctl |=  0x080420;
3614         }
3615
3616         /* enable receive descriptor ring */
3617         rxdctl |= IXGBE_RXDCTL_ENABLE;
3618         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3619
3620         ixgbe_rx_desc_queue_enable(adapter, ring);
3621         ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3622 }
3623
3624 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3625 {
3626         struct ixgbe_hw *hw = &adapter->hw;
3627         int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3628         u16 pool;
3629
3630         /* PSRTYPE must be initialized in non 82598 adapters */
3631         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3632                       IXGBE_PSRTYPE_UDPHDR |
3633                       IXGBE_PSRTYPE_IPV4HDR |
3634                       IXGBE_PSRTYPE_L2HDR |
3635                       IXGBE_PSRTYPE_IPV6HDR;
3636
3637         if (hw->mac.type == ixgbe_mac_82598EB)
3638                 return;
3639
3640         if (rss_i > 3)
3641                 psrtype |= 2 << 29;
3642         else if (rss_i > 1)
3643                 psrtype |= 1 << 29;
3644
3645         for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3646                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3647 }
3648
3649 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3650 {
3651         struct ixgbe_hw *hw = &adapter->hw;
3652         u32 reg_offset, vf_shift;
3653         u32 gcr_ext, vmdctl;
3654         int i;
3655
3656         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3657                 return;
3658
3659         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3660         vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3661         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3662         vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3663         vmdctl |= IXGBE_VT_CTL_REPLEN;
3664         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3665
3666         vf_shift = VMDQ_P(0) % 32;
3667         reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3668
3669         /* Enable only the PF's pool for Tx/Rx */
3670         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3671         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3672         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3673         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3674         if (adapter->bridge_mode == BRIDGE_MODE_VEB)
3675                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3676
3677         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3678         hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3679
3680         /*
3681          * Set up VF register offsets for selected VT Mode,
3682          * i.e. 32 or 64 VFs for SR-IOV
3683          */
3684         switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3685         case IXGBE_82599_VMDQ_8Q_MASK:
3686                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3687                 break;
3688         case IXGBE_82599_VMDQ_4Q_MASK:
3689                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3690                 break;
3691         default:
3692                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3693                 break;
3694         }
3695
3696         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3697
3698
3699         /* Enable MAC Anti-Spoofing */
3700         hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3701                                           adapter->num_vfs);
3702
3703         /* Ensure LLDP is set for Ethertype Antispoofing if we will be
3704          * calling set_ethertype_anti_spoofing for each VF in loop below
3705          */
3706         if (hw->mac.ops.set_ethertype_anti_spoofing)
3707                 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_LLDP),
3708                                 (IXGBE_ETQF_FILTER_EN    | /* enable filter */
3709                                  IXGBE_ETQF_TX_ANTISPOOF | /* tx antispoof */
3710                                  IXGBE_ETH_P_LLDP));       /* LLDP eth type */
3711
3712         /* For VFs that have spoof checking turned off */
3713         for (i = 0; i < adapter->num_vfs; i++) {
3714                 if (!adapter->vfinfo[i].spoofchk_enabled)
3715                         ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3716
3717                 /* enable ethertype anti spoofing if hw supports it */
3718                 if (hw->mac.ops.set_ethertype_anti_spoofing)
3719                         hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
3720
3721                 /* Enable/Disable RSS query feature  */
3722                 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
3723                                           adapter->vfinfo[i].rss_query_enabled);
3724         }
3725 }
3726
3727 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3728 {
3729         struct ixgbe_hw *hw = &adapter->hw;
3730         struct net_device *netdev = adapter->netdev;
3731         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3732         struct ixgbe_ring *rx_ring;
3733         int i;
3734         u32 mhadd, hlreg0;
3735
3736 #ifdef IXGBE_FCOE
3737         /* adjust max frame to be able to do baby jumbo for FCoE */
3738         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3739             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3740                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3741
3742 #endif /* IXGBE_FCOE */
3743
3744         /* adjust max frame to be at least the size of a standard frame */
3745         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3746                 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3747
3748         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3749         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3750                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3751                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3752
3753                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3754         }
3755
3756         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3757         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3758         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3759         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3760
3761         /*
3762          * Setup the HW Rx Head and Tail Descriptor Pointers and
3763          * the Base and Length of the Rx Descriptor Ring
3764          */
3765         for (i = 0; i < adapter->num_rx_queues; i++) {
3766                 rx_ring = adapter->rx_ring[i];
3767                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3768                         set_ring_rsc_enabled(rx_ring);
3769                 else
3770                         clear_ring_rsc_enabled(rx_ring);
3771         }
3772 }
3773
3774 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3775 {
3776         struct ixgbe_hw *hw = &adapter->hw;
3777         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3778
3779         switch (hw->mac.type) {
3780         case ixgbe_mac_X550:
3781         case ixgbe_mac_X550EM_x:
3782         case ixgbe_mac_82598EB:
3783                 /*
3784                  * For VMDq support of different descriptor types or
3785                  * buffer sizes through the use of multiple SRRCTL
3786                  * registers, RDRXCTL.MVMEN must be set to 1
3787                  *
3788                  * also, the manual doesn't mention it clearly but DCA hints
3789                  * will only use queue 0's tags unless this bit is set.  Side
3790                  * effects of setting this bit are only that SRRCTL must be
3791                  * fully programmed [0..15]
3792                  */
3793                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3794                 break;
3795         case ixgbe_mac_82599EB:
3796         case ixgbe_mac_X540:
3797                 /* Disable RSC for ACK packets */
3798                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3799                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3800                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3801                 /* hardware requires some bits to be set by default */
3802                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3803                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3804                 break;
3805         default:
3806                 /* We should do nothing since we don't know this hardware */
3807                 return;
3808         }
3809
3810         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3811 }
3812
3813 /**
3814  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3815  * @adapter: board private structure
3816  *
3817  * Configure the Rx unit of the MAC after a reset.
3818  **/
3819 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3820 {
3821         struct ixgbe_hw *hw = &adapter->hw;
3822         int i;
3823         u32 rxctrl, rfctl;
3824
3825         /* disable receives while setting up the descriptors */
3826         hw->mac.ops.disable_rx(hw);
3827
3828         ixgbe_setup_psrtype(adapter);
3829         ixgbe_setup_rdrxctl(adapter);
3830
3831         /* RSC Setup */
3832         rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3833         rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3834         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3835                 rfctl |= IXGBE_RFCTL_RSC_DIS;
3836         IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3837
3838         /* Program registers for the distribution of queues */
3839         ixgbe_setup_mrqc(adapter);
3840
3841         /* set_rx_buffer_len must be called before ring initialization */
3842         ixgbe_set_rx_buffer_len(adapter);
3843
3844         /*
3845          * Setup the HW Rx Head and Tail Descriptor Pointers and
3846          * the Base and Length of the Rx Descriptor Ring
3847          */
3848         for (i = 0; i < adapter->num_rx_queues; i++)
3849                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3850
3851         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3852         /* disable drop enable for 82598 parts */
3853         if (hw->mac.type == ixgbe_mac_82598EB)
3854                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3855
3856         /* enable all receives */
3857         rxctrl |= IXGBE_RXCTRL_RXEN;
3858         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3859 }
3860
3861 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3862                                  __be16 proto, u16 vid)
3863 {
3864         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3865         struct ixgbe_hw *hw = &adapter->hw;
3866
3867         /* add VID to filter table */
3868         hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true);
3869         set_bit(vid, adapter->active_vlans);
3870
3871         return 0;
3872 }
3873
3874 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3875                                   __be16 proto, u16 vid)
3876 {
3877         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3878         struct ixgbe_hw *hw = &adapter->hw;
3879
3880         /* remove VID from filter table */
3881         hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), false);
3882         clear_bit(vid, adapter->active_vlans);
3883
3884         return 0;
3885 }
3886
3887 /**
3888  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3889  * @adapter: driver data
3890  */
3891 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3892 {
3893         struct ixgbe_hw *hw = &adapter->hw;
3894         u32 vlnctrl;
3895         int i, j;
3896
3897         switch (hw->mac.type) {
3898         case ixgbe_mac_82598EB:
3899                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3900                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3901                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3902                 break;
3903         case ixgbe_mac_82599EB:
3904         case ixgbe_mac_X540:
3905         case ixgbe_mac_X550:
3906         case ixgbe_mac_X550EM_x:
3907                 for (i = 0; i < adapter->num_rx_queues; i++) {
3908                         struct ixgbe_ring *ring = adapter->rx_ring[i];
3909
3910                         if (ring->l2_accel_priv)
3911                                 continue;
3912                         j = ring->reg_idx;
3913                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3914                         vlnctrl &= ~IXGBE_RXDCTL_VME;
3915                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3916                 }
3917                 break;
3918         default:
3919                 break;
3920         }
3921 }
3922
3923 /**
3924  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
3925  * @adapter: driver data
3926  */
3927 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
3928 {
3929         struct ixgbe_hw *hw = &adapter->hw;
3930         u32 vlnctrl;
3931         int i, j;
3932
3933         switch (hw->mac.type) {
3934         case ixgbe_mac_82598EB:
3935                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3936                 vlnctrl |= IXGBE_VLNCTRL_VME;
3937                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3938                 break;
3939         case ixgbe_mac_82599EB:
3940         case ixgbe_mac_X540:
3941         case ixgbe_mac_X550:
3942         case ixgbe_mac_X550EM_x:
3943                 for (i = 0; i < adapter->num_rx_queues; i++) {
3944                         struct ixgbe_ring *ring = adapter->rx_ring[i];
3945
3946                         if (ring->l2_accel_priv)
3947                                 continue;
3948                         j = ring->reg_idx;
3949                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
3950                         vlnctrl |= IXGBE_RXDCTL_VME;
3951                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
3952                 }
3953                 break;
3954         default:
3955                 break;
3956         }
3957 }
3958
3959 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
3960 {
3961         u16 vid;
3962
3963         ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
3964
3965         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
3966                 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
3967 }
3968
3969 /**
3970  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
3971  * @netdev: network interface device structure
3972  *
3973  * Writes multicast address list to the MTA hash table.
3974  * Returns: -ENOMEM on failure
3975  *                0 on no addresses written
3976  *                X on writing X addresses to MTA
3977  **/
3978 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
3979 {
3980         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3981         struct ixgbe_hw *hw = &adapter->hw;
3982
3983         if (!netif_running(netdev))
3984                 return 0;
3985
3986         if (hw->mac.ops.update_mc_addr_list)
3987                 hw->mac.ops.update_mc_addr_list(hw, netdev);
3988         else
3989                 return -ENOMEM;
3990
3991 #ifdef CONFIG_PCI_IOV
3992         ixgbe_restore_vf_multicasts(adapter);
3993 #endif
3994
3995         return netdev_mc_count(netdev);
3996 }
3997
3998 #ifdef CONFIG_PCI_IOV
3999 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4000 {
4001         struct ixgbe_hw *hw = &adapter->hw;
4002         int i;
4003         for (i = 0; i < hw->mac.num_rar_entries; i++) {
4004                 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4005                         hw->mac.ops.set_rar(hw, i, adapter->mac_table[i].addr,
4006                                             adapter->mac_table[i].queue,
4007                                             IXGBE_RAH_AV);
4008                 else
4009                         hw->mac.ops.clear_rar(hw, i);
4010
4011                 adapter->mac_table[i].state &= ~(IXGBE_MAC_STATE_MODIFIED);
4012         }
4013 }
4014 #endif
4015
4016 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4017 {
4018         struct ixgbe_hw *hw = &adapter->hw;
4019         int i;
4020         for (i = 0; i < hw->mac.num_rar_entries; i++) {
4021                 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_MODIFIED) {
4022                         if (adapter->mac_table[i].state &
4023                             IXGBE_MAC_STATE_IN_USE)
4024                                 hw->mac.ops.set_rar(hw, i,
4025                                                 adapter->mac_table[i].addr,
4026                                                 adapter->mac_table[i].queue,
4027                                                 IXGBE_RAH_AV);
4028                         else
4029                                 hw->mac.ops.clear_rar(hw, i);
4030
4031                         adapter->mac_table[i].state &=
4032                                                 ~(IXGBE_MAC_STATE_MODIFIED);
4033                 }
4034         }
4035 }
4036
4037 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4038 {
4039         int i;
4040         struct ixgbe_hw *hw = &adapter->hw;
4041
4042         for (i = 0; i < hw->mac.num_rar_entries; i++) {
4043                 adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4044                 adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4045                 eth_zero_addr(adapter->mac_table[i].addr);
4046                 adapter->mac_table[i].queue = 0;
4047         }
4048         ixgbe_sync_mac_table(adapter);
4049 }
4050
4051 static int ixgbe_available_rars(struct ixgbe_adapter *adapter)
4052 {
4053         struct ixgbe_hw *hw = &adapter->hw;
4054         int i, count = 0;
4055
4056         for (i = 0; i < hw->mac.num_rar_entries; i++) {
4057                 if (adapter->mac_table[i].state == 0)
4058                         count++;
4059         }
4060         return count;
4061 }
4062
4063 /* this function destroys the first RAR entry */
4064 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter,
4065                                          u8 *addr)
4066 {
4067         struct ixgbe_hw *hw = &adapter->hw;
4068
4069         memcpy(&adapter->mac_table[0].addr, addr, ETH_ALEN);
4070         adapter->mac_table[0].queue = VMDQ_P(0);
4071         adapter->mac_table[0].state = (IXGBE_MAC_STATE_DEFAULT |
4072                                        IXGBE_MAC_STATE_IN_USE);
4073         hw->mac.ops.set_rar(hw, 0, adapter->mac_table[0].addr,
4074                             adapter->mac_table[0].queue,
4075                             IXGBE_RAH_AV);
4076 }
4077
4078 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4079 {
4080         struct ixgbe_hw *hw = &adapter->hw;
4081         int i;
4082
4083         if (is_zero_ether_addr(addr))
4084                 return -EINVAL;
4085
4086         for (i = 0; i < hw->mac.num_rar_entries; i++) {
4087                 if (adapter->mac_table[i].state & IXGBE_MAC_STATE_IN_USE)
4088                         continue;
4089                 adapter->mac_table[i].state |= (IXGBE_MAC_STATE_MODIFIED |
4090                                                 IXGBE_MAC_STATE_IN_USE);
4091                 ether_addr_copy(adapter->mac_table[i].addr, addr);
4092                 adapter->mac_table[i].queue = queue;
4093                 ixgbe_sync_mac_table(adapter);
4094                 return i;
4095         }
4096         return -ENOMEM;
4097 }
4098
4099 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter, u8 *addr, u16 queue)
4100 {
4101         /* search table for addr, if found, set to 0 and sync */
4102         int i;
4103         struct ixgbe_hw *hw = &adapter->hw;
4104
4105         if (is_zero_ether_addr(addr))
4106                 return -EINVAL;
4107
4108         for (i = 0; i < hw->mac.num_rar_entries; i++) {
4109                 if (ether_addr_equal(addr, adapter->mac_table[i].addr) &&
4110                     adapter->mac_table[i].queue == queue) {
4111                         adapter->mac_table[i].state |= IXGBE_MAC_STATE_MODIFIED;
4112                         adapter->mac_table[i].state &= ~IXGBE_MAC_STATE_IN_USE;
4113                         eth_zero_addr(adapter->mac_table[i].addr);
4114                         adapter->mac_table[i].queue = 0;
4115                         ixgbe_sync_mac_table(adapter);
4116                         return 0;
4117                 }
4118         }
4119         return -ENOMEM;
4120 }
4121 /**
4122  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4123  * @netdev: network interface device structure
4124  *
4125  * Writes unicast address list to the RAR table.
4126  * Returns: -ENOMEM on failure/insufficient address space
4127  *                0 on no addresses written
4128  *                X on writing X addresses to the RAR table
4129  **/
4130 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4131 {
4132         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4133         int count = 0;
4134
4135         /* return ENOMEM indicating insufficient memory for addresses */
4136         if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter))
4137                 return -ENOMEM;
4138
4139         if (!netdev_uc_empty(netdev)) {
4140                 struct netdev_hw_addr *ha;
4141                 netdev_for_each_uc_addr(ha, netdev) {
4142                         ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4143                         ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4144                         count++;
4145                 }
4146         }
4147         return count;
4148 }
4149
4150 /**
4151  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4152  * @netdev: network interface device structure
4153  *
4154  * The set_rx_method entry point is called whenever the unicast/multicast
4155  * address list or the network interface flags are updated.  This routine is
4156  * responsible for configuring the hardware for proper unicast, multicast and
4157  * promiscuous mode.
4158  **/
4159 void ixgbe_set_rx_mode(struct net_device *netdev)
4160 {
4161         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4162         struct ixgbe_hw *hw = &adapter->hw;
4163         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4164         u32 vlnctrl;
4165         int count;
4166
4167         /* Check for Promiscuous and All Multicast modes */
4168         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4169         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4170
4171         /* set all bits that we expect to always be set */
4172         fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4173         fctrl |= IXGBE_FCTRL_BAM;
4174         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4175         fctrl |= IXGBE_FCTRL_PMCF;
4176
4177         /* clear the bits we are changing the status of */
4178         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4179         vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4180         if (netdev->flags & IFF_PROMISC) {
4181                 hw->addr_ctrl.user_set_promisc = true;
4182                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4183                 vmolr |= IXGBE_VMOLR_MPE;
4184                 /* Only disable hardware filter vlans in promiscuous mode
4185                  * if SR-IOV and VMDQ are disabled - otherwise ensure
4186                  * that hardware VLAN filters remain enabled.
4187                  */
4188                 if (adapter->flags & (IXGBE_FLAG_VMDQ_ENABLED |
4189                                       IXGBE_FLAG_SRIOV_ENABLED))
4190                         vlnctrl |= (IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4191         } else {
4192                 if (netdev->flags & IFF_ALLMULTI) {
4193                         fctrl |= IXGBE_FCTRL_MPE;
4194                         vmolr |= IXGBE_VMOLR_MPE;
4195                 }
4196                 vlnctrl |= IXGBE_VLNCTRL_VFE;
4197                 hw->addr_ctrl.user_set_promisc = false;
4198         }
4199
4200         /*
4201          * Write addresses to available RAR registers, if there is not
4202          * sufficient space to store all the addresses then enable
4203          * unicast promiscuous mode
4204          */
4205         count = ixgbe_write_uc_addr_list(netdev, VMDQ_P(0));
4206         if (count < 0) {
4207                 fctrl |= IXGBE_FCTRL_UPE;
4208                 vmolr |= IXGBE_VMOLR_ROPE;
4209         }
4210
4211         /* Write addresses to the MTA, if the attempt fails
4212          * then we should just turn on promiscuous mode so
4213          * that we can at least receive multicast traffic
4214          */
4215         count = ixgbe_write_mc_addr_list(netdev);
4216         if (count < 0) {
4217                 fctrl |= IXGBE_FCTRL_MPE;
4218                 vmolr |= IXGBE_VMOLR_MPE;
4219         } else if (count) {
4220                 vmolr |= IXGBE_VMOLR_ROMPE;
4221         }
4222
4223         if (hw->mac.type != ixgbe_mac_82598EB) {
4224                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4225                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4226                            IXGBE_VMOLR_ROPE);
4227                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4228         }
4229
4230         /* This is useful for sniffing bad packets. */
4231         if (adapter->netdev->features & NETIF_F_RXALL) {
4232                 /* UPE and MPE will be handled by normal PROMISC logic
4233                  * in e1000e_set_rx_mode */
4234                 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4235                           IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4236                           IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4237
4238                 fctrl &= ~(IXGBE_FCTRL_DPF);
4239                 /* NOTE:  VLAN filtering is disabled by setting PROMISC */
4240         }
4241
4242         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4243         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4244
4245         if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
4246                 ixgbe_vlan_strip_enable(adapter);
4247         else
4248                 ixgbe_vlan_strip_disable(adapter);
4249 }
4250
4251 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4252 {
4253         int q_idx;
4254
4255         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4256                 ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4257                 napi_enable(&adapter->q_vector[q_idx]->napi);
4258         }
4259 }
4260
4261 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4262 {
4263         int q_idx;
4264
4265         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4266                 napi_disable(&adapter->q_vector[q_idx]->napi);
4267                 while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4268                         pr_info("QV %d locked\n", q_idx);
4269                         usleep_range(1000, 20000);
4270                 }
4271         }
4272 }
4273
4274 static void ixgbe_clear_vxlan_port(struct ixgbe_adapter *adapter)
4275 {
4276         switch (adapter->hw.mac.type) {
4277         case ixgbe_mac_X550:
4278         case ixgbe_mac_X550EM_x:
4279                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VXLANCTRL, 0);
4280 #ifdef CONFIG_IXGBE_VXLAN
4281                 adapter->vxlan_port = 0;
4282 #endif
4283                 break;
4284         default:
4285                 break;
4286         }
4287 }
4288
4289 #ifdef CONFIG_IXGBE_DCB
4290 /**
4291  * ixgbe_configure_dcb - Configure DCB hardware
4292  * @adapter: ixgbe adapter struct
4293  *
4294  * This is called by the driver on open to configure the DCB hardware.
4295  * This is also called by the gennetlink interface when reconfiguring
4296  * the DCB state.
4297  */
4298 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4299 {
4300         struct ixgbe_hw *hw = &adapter->hw;
4301         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4302
4303         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4304                 if (hw->mac.type == ixgbe_mac_82598EB)
4305                         netif_set_gso_max_size(adapter->netdev, 65536);
4306                 return;
4307         }
4308
4309         if (hw->mac.type == ixgbe_mac_82598EB)
4310                 netif_set_gso_max_size(adapter->netdev, 32768);
4311
4312 #ifdef IXGBE_FCOE
4313         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4314                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4315 #endif
4316
4317         /* reconfigure the hardware */
4318         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4319                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4320                                                 DCB_TX_CONFIG);
4321                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4322                                                 DCB_RX_CONFIG);
4323                 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4324         } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4325                 ixgbe_dcb_hw_ets(&adapter->hw,
4326                                  adapter->ixgbe_ieee_ets,
4327                                  max_frame);
4328                 ixgbe_dcb_hw_pfc_config(&adapter->hw,
4329                                         adapter->ixgbe_ieee_pfc->pfc_en,
4330                                         adapter->ixgbe_ieee_ets->prio_tc);
4331         }
4332
4333         /* Enable RSS Hash per TC */
4334         if (hw->mac.type != ixgbe_mac_82598EB) {
4335                 u32 msb = 0;
4336                 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4337
4338                 while (rss_i) {
4339                         msb++;
4340                         rss_i >>= 1;
4341                 }
4342
4343                 /* write msb to all 8 TCs in one write */
4344                 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4345         }
4346 }
4347 #endif
4348
4349 /* Additional bittime to account for IXGBE framing */
4350 #define IXGBE_ETH_FRAMING 20
4351
4352 /**
4353  * ixgbe_hpbthresh - calculate high water mark for flow control
4354  *
4355  * @adapter: board private structure to calculate for
4356  * @pb: packet buffer to calculate
4357  */
4358 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4359 {
4360         struct ixgbe_hw *hw = &adapter->hw;
4361         struct net_device *dev = adapter->netdev;
4362         int link, tc, kb, marker;
4363         u32 dv_id, rx_pba;
4364
4365         /* Calculate max LAN frame size */
4366         tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4367
4368 #ifdef IXGBE_FCOE
4369         /* FCoE traffic class uses FCOE jumbo frames */
4370         if ((dev->features & NETIF_F_FCOE_MTU) &&
4371             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4372             (pb == ixgbe_fcoe_get_tc(adapter)))
4373                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4374 #endif
4375
4376         /* Calculate delay value for device */
4377         switch (hw->mac.type) {
4378         case ixgbe_mac_X540:
4379         case ixgbe_mac_X550:
4380         case ixgbe_mac_X550EM_x:
4381                 dv_id = IXGBE_DV_X540(link, tc);
4382                 break;
4383         default:
4384                 dv_id = IXGBE_DV(link, tc);
4385                 break;
4386         }
4387
4388         /* Loopback switch introduces additional latency */
4389         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4390                 dv_id += IXGBE_B2BT(tc);
4391
4392         /* Delay value is calculated in bit times convert to KB */
4393         kb = IXGBE_BT2KB(dv_id);
4394         rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4395
4396         marker = rx_pba - kb;
4397
4398         /* It is possible that the packet buffer is not large enough
4399          * to provide required headroom. In this case throw an error
4400          * to user and a do the best we can.
4401          */
4402         if (marker < 0) {
4403                 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4404                             "headroom to support flow control."
4405                             "Decrease MTU or number of traffic classes\n", pb);
4406                 marker = tc + 1;
4407         }
4408
4409         return marker;
4410 }
4411
4412 /**
4413  * ixgbe_lpbthresh - calculate low water mark for for flow control
4414  *
4415  * @adapter: board private structure to calculate for
4416  * @pb: packet buffer to calculate
4417  */
4418 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4419 {
4420         struct ixgbe_hw *hw = &adapter->hw;
4421         struct net_device *dev = adapter->netdev;
4422         int tc;
4423         u32 dv_id;
4424
4425         /* Calculate max LAN frame size */
4426         tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4427
4428 #ifdef IXGBE_FCOE
4429         /* FCoE traffic class uses FCOE jumbo frames */
4430         if ((dev->features & NETIF_F_FCOE_MTU) &&
4431             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4432             (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4433                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4434 #endif
4435
4436         /* Calculate delay value for device */
4437         switch (hw->mac.type) {
4438         case ixgbe_mac_X540:
4439         case ixgbe_mac_X550:
4440         case ixgbe_mac_X550EM_x:
4441                 dv_id = IXGBE_LOW_DV_X540(tc);
4442                 break;
4443         default:
4444                 dv_id = IXGBE_LOW_DV(tc);
4445                 break;
4446         }
4447
4448         /* Delay value is calculated in bit times convert to KB */
4449         return IXGBE_BT2KB(dv_id);
4450 }
4451
4452 /*
4453  * ixgbe_pbthresh_setup - calculate and setup high low water marks
4454  */
4455 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4456 {
4457         struct ixgbe_hw *hw = &adapter->hw;
4458         int num_tc = netdev_get_num_tc(adapter->netdev);
4459         int i;
4460
4461         if (!num_tc)
4462                 num_tc = 1;
4463
4464         for (i = 0; i < num_tc; i++) {
4465                 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4466                 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4467
4468                 /* Low water marks must not be larger than high water marks */
4469                 if (hw->fc.low_water[i] > hw->fc.high_water[i])
4470                         hw->fc.low_water[i] = 0;
4471         }
4472
4473         for (; i < MAX_TRAFFIC_CLASS; i++)
4474                 hw->fc.high_water[i] = 0;
4475 }
4476
4477 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4478 {
4479         struct ixgbe_hw *hw = &adapter->hw;
4480         int hdrm;
4481         u8 tc = netdev_get_num_tc(adapter->netdev);
4482
4483         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4484             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4485                 hdrm = 32 << adapter->fdir_pballoc;
4486         else
4487                 hdrm = 0;
4488
4489         hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4490         ixgbe_pbthresh_setup(adapter);
4491 }
4492
4493 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4494 {
4495         struct ixgbe_hw *hw = &adapter->hw;
4496         struct hlist_node *node2;
4497         struct ixgbe_fdir_filter *filter;
4498
4499         spin_lock(&adapter->fdir_perfect_lock);
4500
4501         if (!hlist_empty(&adapter->fdir_filter_list))
4502                 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4503
4504         hlist_for_each_entry_safe(filter, node2,
4505                                   &adapter->fdir_filter_list, fdir_node) {
4506                 ixgbe_fdir_write_perfect_filter_82599(hw,
4507                                 &filter->filter,
4508                                 filter->sw_idx,
4509                                 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4510                                 IXGBE_FDIR_DROP_QUEUE :
4511                                 adapter->rx_ring[filter->action]->reg_idx);
4512         }
4513
4514         spin_unlock(&adapter->fdir_perfect_lock);
4515 }
4516
4517 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4518                                       struct ixgbe_adapter *adapter)
4519 {
4520         struct ixgbe_hw *hw = &adapter->hw;
4521         u32 vmolr;
4522
4523         /* No unicast promiscuous support for VMDQ devices. */
4524         vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4525         vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4526
4527         /* clear the affected bit */
4528         vmolr &= ~IXGBE_VMOLR_MPE;
4529
4530         if (dev->flags & IFF_ALLMULTI) {
4531                 vmolr |= IXGBE_VMOLR_MPE;
4532         } else {
4533                 vmolr |= IXGBE_VMOLR_ROMPE;
4534                 hw->mac.ops.update_mc_addr_list(hw, dev);
4535         }
4536         ixgbe_write_uc_addr_list(adapter->netdev, pool);
4537         IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4538 }
4539
4540 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4541 {
4542         struct ixgbe_adapter *adapter = vadapter->real_adapter;
4543         int rss_i = adapter->num_rx_queues_per_pool;
4544         struct ixgbe_hw *hw = &adapter->hw;
4545         u16 pool = vadapter->pool;
4546         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4547                       IXGBE_PSRTYPE_UDPHDR |
4548                       IXGBE_PSRTYPE_IPV4HDR |
4549                       IXGBE_PSRTYPE_L2HDR |
4550                       IXGBE_PSRTYPE_IPV6HDR;
4551
4552         if (hw->mac.type == ixgbe_mac_82598EB)
4553                 return;
4554
4555         if (rss_i > 3)
4556                 psrtype |= 2 << 29;
4557         else if (rss_i > 1)
4558                 psrtype |= 1 << 29;
4559
4560         IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4561 }
4562
4563 /**
4564  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4565  * @rx_ring: ring to free buffers from
4566  **/
4567 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4568 {
4569         struct device *dev = rx_ring->dev;
4570         unsigned long size;
4571         u16 i;
4572
4573         /* ring already cleared, nothing to do */
4574         if (!rx_ring->rx_buffer_info)
4575                 return;
4576
4577         /* Free all the Rx ring sk_buffs */
4578         for (i = 0; i < rx_ring->count; i++) {
4579                 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4580
4581                 if (rx_buffer->skb) {
4582                         struct sk_buff *skb = rx_buffer->skb;
4583                         if (IXGBE_CB(skb)->page_released)
4584                                 dma_unmap_page(dev,
4585                                                IXGBE_CB(skb)->dma,
4586                                                ixgbe_rx_bufsz(rx_ring),
4587                                                DMA_FROM_DEVICE);
4588                         dev_kfree_skb(skb);
4589                         rx_buffer->skb = NULL;
4590                 }
4591
4592                 if (!rx_buffer->page)
4593                         continue;
4594
4595                 dma_unmap_page(dev, rx_buffer->dma,
4596                                ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4597                 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4598
4599                 rx_buffer->page = NULL;
4600         }
4601
4602         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4603         memset(rx_ring->rx_buffer_info, 0, size);
4604
4605         /* Zero out the descriptor ring */
4606         memset(rx_ring->desc, 0, rx_ring->size);
4607
4608         rx_ring->next_to_alloc = 0;
4609         rx_ring->next_to_clean = 0;
4610         rx_ring->next_to_use = 0;
4611 }
4612
4613 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4614                                    struct ixgbe_ring *rx_ring)
4615 {
4616         struct ixgbe_adapter *adapter = vadapter->real_adapter;
4617         int index = rx_ring->queue_index + vadapter->rx_base_queue;
4618
4619         /* shutdown specific queue receive and wait for dma to settle */
4620         ixgbe_disable_rx_queue(adapter, rx_ring);
4621         usleep_range(10000, 20000);
4622         ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4623         ixgbe_clean_rx_ring(rx_ring);
4624         rx_ring->l2_accel_priv = NULL;
4625 }
4626
4627 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4628                                struct ixgbe_fwd_adapter *accel)
4629 {
4630         struct ixgbe_adapter *adapter = accel->real_adapter;
4631         unsigned int rxbase = accel->rx_base_queue;
4632         unsigned int txbase = accel->tx_base_queue;
4633         int i;
4634
4635         netif_tx_stop_all_queues(vdev);
4636
4637         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4638                 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4639                 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4640         }
4641
4642         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4643                 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4644                 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4645         }
4646
4647
4648         return 0;
4649 }
4650
4651 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4652                              struct ixgbe_fwd_adapter *accel)
4653 {
4654         struct ixgbe_adapter *adapter = accel->real_adapter;
4655         unsigned int rxbase, txbase, queues;
4656         int i, baseq, err = 0;
4657
4658         if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4659                 return 0;
4660
4661         baseq = accel->pool * adapter->num_rx_queues_per_pool;
4662         netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4663                    accel->pool, adapter->num_rx_pools,
4664                    baseq, baseq + adapter->num_rx_queues_per_pool,
4665                    adapter->fwd_bitmask);
4666
4667         accel->netdev = vdev;
4668         accel->rx_base_queue = rxbase = baseq;
4669         accel->tx_base_queue = txbase = baseq;
4670
4671         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4672                 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4673
4674         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4675                 adapter->rx_ring[rxbase + i]->netdev = vdev;
4676                 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4677                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4678         }
4679
4680         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4681                 adapter->tx_ring[txbase + i]->netdev = vdev;
4682                 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4683         }
4684
4685         queues = min_t(unsigned int,
4686                        adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4687         err = netif_set_real_num_tx_queues(vdev, queues);
4688         if (err)
4689                 goto fwd_queue_err;
4690
4691         err = netif_set_real_num_rx_queues(vdev, queues);
4692         if (err)
4693                 goto fwd_queue_err;
4694
4695         if (is_valid_ether_addr(vdev->dev_addr))
4696                 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4697
4698         ixgbe_fwd_psrtype(accel);
4699         ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4700         return err;
4701 fwd_queue_err:
4702         ixgbe_fwd_ring_down(vdev, accel);
4703         return err;
4704 }
4705
4706 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4707 {
4708         struct net_device *upper;
4709         struct list_head *iter;
4710         int err;
4711
4712         netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4713                 if (netif_is_macvlan(upper)) {
4714                         struct macvlan_dev *dfwd = netdev_priv(upper);
4715                         struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4716
4717                         if (dfwd->fwd_priv) {
4718                                 err = ixgbe_fwd_ring_up(upper, vadapter);
4719                                 if (err)
4720                                         continue;
4721                         }
4722                 }
4723         }
4724 }
4725
4726 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4727 {
4728         struct ixgbe_hw *hw = &adapter->hw;
4729
4730         ixgbe_configure_pb(adapter);
4731 #ifdef CONFIG_IXGBE_DCB
4732         ixgbe_configure_dcb(adapter);
4733 #endif
4734         /*
4735          * We must restore virtualization before VLANs or else
4736          * the VLVF registers will not be populated
4737          */
4738         ixgbe_configure_virtualization(adapter);
4739
4740         ixgbe_set_rx_mode(adapter->netdev);
4741         ixgbe_restore_vlan(adapter);
4742
4743         switch (hw->mac.type) {
4744         case ixgbe_mac_82599EB:
4745         case ixgbe_mac_X540:
4746                 hw->mac.ops.disable_rx_buff(hw);
4747                 break;
4748         default:
4749                 break;
4750         }
4751
4752         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
4753                 ixgbe_init_fdir_signature_82599(&adapter->hw,
4754                                                 adapter->fdir_pballoc);
4755         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
4756                 ixgbe_init_fdir_perfect_82599(&adapter->hw,
4757                                               adapter->fdir_pballoc);
4758                 ixgbe_fdir_filter_restore(adapter);
4759         }
4760
4761         switch (hw->mac.type) {
4762         case ixgbe_mac_82599EB:
4763         case ixgbe_mac_X540:
4764                 hw->mac.ops.enable_rx_buff(hw);
4765                 break;
4766         default:
4767                 break;
4768         }
4769
4770 #ifdef IXGBE_FCOE
4771         /* configure FCoE L2 filters, redirection table, and Rx control */
4772         ixgbe_configure_fcoe(adapter);
4773
4774 #endif /* IXGBE_FCOE */
4775         ixgbe_configure_tx(adapter);
4776         ixgbe_configure_rx(adapter);
4777         ixgbe_configure_dfwd(adapter);
4778 }
4779
4780 /**
4781  * ixgbe_sfp_link_config - set up SFP+ link
4782  * @adapter: pointer to private adapter struct
4783  **/
4784 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
4785 {
4786         /*
4787          * We are assuming the worst case scenario here, and that
4788          * is that an SFP was inserted/removed after the reset
4789          * but before SFP detection was enabled.  As such the best
4790          * solution is to just start searching as soon as we start
4791          */
4792         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
4793                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
4794
4795         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
4796 }
4797
4798 /**
4799  * ixgbe_non_sfp_link_config - set up non-SFP+ link
4800  * @hw: pointer to private hardware struct
4801  *
4802  * Returns 0 on success, negative on failure
4803  **/
4804 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
4805 {
4806         u32 speed;
4807         bool autoneg, link_up = false;
4808         int ret = IXGBE_ERR_LINK_SETUP;
4809
4810         if (hw->mac.ops.check_link)
4811                 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
4812
4813         if (ret)
4814                 return ret;
4815
4816         speed = hw->phy.autoneg_advertised;
4817         if ((!speed) && (hw->mac.ops.get_link_capabilities))
4818                 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
4819                                                         &autoneg);
4820         if (ret)
4821                 return ret;
4822
4823         if (hw->mac.ops.setup_link)
4824                 ret = hw->mac.ops.setup_link(hw, speed, link_up);
4825
4826         return ret;
4827 }
4828
4829 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
4830 {
4831         struct ixgbe_hw *hw = &adapter->hw;
4832         u32 gpie = 0;
4833
4834         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
4835                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
4836                        IXGBE_GPIE_OCD;
4837                 gpie |= IXGBE_GPIE_EIAME;
4838                 /*
4839                  * use EIAM to auto-mask when MSI-X interrupt is asserted
4840                  * this saves a register write for every interrupt
4841                  */
4842                 switch (hw->mac.type) {
4843                 case ixgbe_mac_82598EB:
4844                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4845                         break;
4846                 case ixgbe_mac_82599EB:
4847                 case ixgbe_mac_X540:
4848                 case ixgbe_mac_X550:
4849                 case ixgbe_mac_X550EM_x:
4850                 default:
4851                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
4852                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
4853                         break;
4854                 }
4855         } else {
4856                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
4857                  * specifically only auto mask tx and rx interrupts */
4858                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
4859         }
4860
4861         /* XXX: to interrupt immediately for EICS writes, enable this */
4862         /* gpie |= IXGBE_GPIE_EIMEN; */
4863
4864         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
4865                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
4866
4867                 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
4868                 case IXGBE_82599_VMDQ_8Q_MASK:
4869                         gpie |= IXGBE_GPIE_VTMODE_16;
4870                         break;
4871                 case IXGBE_82599_VMDQ_4Q_MASK:
4872                         gpie |= IXGBE_GPIE_VTMODE_32;
4873                         break;
4874                 default:
4875                         gpie |= IXGBE_GPIE_VTMODE_64;
4876                         break;
4877                 }
4878         }
4879
4880         /* Enable Thermal over heat sensor interrupt */
4881         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
4882                 switch (adapter->hw.mac.type) {
4883                 case ixgbe_mac_82599EB:
4884                         gpie |= IXGBE_SDP0_GPIEN_8259X;
4885                         break;
4886                 case ixgbe_mac_X540:
4887                         gpie |= IXGBE_EIMS_TS;
4888                         break;
4889                 default:
4890                         break;
4891                 }
4892         }
4893
4894         /* Enable fan failure interrupt */
4895         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
4896                 gpie |= IXGBE_SDP1_GPIEN(hw);
4897
4898         if (hw->mac.type == ixgbe_mac_82599EB) {
4899                 gpie |= IXGBE_SDP1_GPIEN_8259X;
4900                 gpie |= IXGBE_SDP2_GPIEN_8259X;
4901         }
4902
4903         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
4904 }
4905
4906 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
4907 {
4908         struct ixgbe_hw *hw = &adapter->hw;
4909         int err;
4910         u32 ctrl_ext;
4911
4912         ixgbe_get_hw_control(adapter);
4913         ixgbe_setup_gpie(adapter);
4914
4915         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
4916                 ixgbe_configure_msix(adapter);
4917         else
4918                 ixgbe_configure_msi_and_legacy(adapter);
4919
4920         /* enable the optics for 82599 SFP+ fiber */
4921         if (hw->mac.ops.enable_tx_laser)
4922                 hw->mac.ops.enable_tx_laser(hw);
4923
4924         if (hw->phy.ops.set_phy_power)
4925                 hw->phy.ops.set_phy_power(hw, true);
4926
4927         smp_mb__before_atomic();
4928         clear_bit(__IXGBE_DOWN, &adapter->state);
4929         ixgbe_napi_enable_all(adapter);
4930
4931         if (ixgbe_is_sfp(hw)) {
4932                 ixgbe_sfp_link_config(adapter);
4933         } else {
4934                 err = ixgbe_non_sfp_link_config(hw);
4935                 if (err)
4936                         e_err(probe, "link_config FAILED %d\n", err);
4937         }
4938
4939         /* clear any pending interrupts, may auto mask */
4940         IXGBE_READ_REG(hw, IXGBE_EICR);
4941         ixgbe_irq_enable(adapter, true, true);
4942
4943         /*
4944          * If this adapter has a fan, check to see if we had a failure
4945          * before we enabled the interrupt.
4946          */
4947         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
4948                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
4949                 if (esdp & IXGBE_ESDP_SDP1)
4950                         e_crit(drv, "Fan has stopped, replace the adapter\n");
4951         }
4952
4953         /* bring the link up in the watchdog, this could race with our first
4954          * link up interrupt but shouldn't be a problem */
4955         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
4956         adapter->link_check_timeout = jiffies;
4957         mod_timer(&adapter->service_timer, jiffies);
4958
4959         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
4960         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
4961         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
4962         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
4963 }
4964
4965 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
4966 {
4967         WARN_ON(in_interrupt());
4968         /* put off any impending NetWatchDogTimeout */
4969         adapter->netdev->trans_start = jiffies;
4970
4971         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
4972                 usleep_range(1000, 2000);
4973         ixgbe_down(adapter);
4974         /*
4975          * If SR-IOV enabled then wait a bit before bringing the adapter
4976          * back up to give the VFs time to respond to the reset.  The
4977          * two second wait is based upon the watchdog timer cycle in
4978          * the VF driver.
4979          */
4980         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4981                 msleep(2000);
4982         ixgbe_up(adapter);
4983         clear_bit(__IXGBE_RESETTING, &adapter->state);
4984 }
4985
4986 void ixgbe_up(struct ixgbe_adapter *adapter)
4987 {
4988         /* hardware has been reset, we need to reload some things */
4989         ixgbe_configure(adapter);
4990
4991         ixgbe_up_complete(adapter);
4992 }
4993
4994 void ixgbe_reset(struct ixgbe_adapter *adapter)
4995 {
4996         struct ixgbe_hw *hw = &adapter->hw;
4997         struct net_device *netdev = adapter->netdev;
4998         int err;
4999         u8 old_addr[ETH_ALEN];
5000
5001         if (ixgbe_removed(hw->hw_addr))
5002                 return;
5003         /* lock SFP init bit to prevent race conditions with the watchdog */
5004         while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5005                 usleep_range(1000, 2000);
5006
5007         /* clear all SFP and link config related flags while holding SFP_INIT */
5008         adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5009                              IXGBE_FLAG2_SFP_NEEDS_RESET);
5010         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5011
5012         err = hw->mac.ops.init_hw(hw);
5013         switch (err) {
5014         case 0:
5015         case IXGBE_ERR_SFP_NOT_PRESENT:
5016         case IXGBE_ERR_SFP_NOT_SUPPORTED:
5017                 break;
5018         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5019                 e_dev_err("master disable timed out\n");
5020                 break;
5021         case IXGBE_ERR_EEPROM_VERSION:
5022                 /* We are running on a pre-production device, log a warning */
5023                 e_dev_warn("This device is a pre-production adapter/LOM. "
5024                            "Please be aware there may be issues associated with "
5025                            "your hardware.  If you are experiencing problems "
5026                            "please contact your Intel or hardware "
5027                            "representative who provided you with this "
5028                            "hardware.\n");
5029                 break;
5030         default:
5031                 e_dev_err("Hardware Error: %d\n", err);
5032         }
5033
5034         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5035         /* do not flush user set addresses */
5036         memcpy(old_addr, &adapter->mac_table[0].addr, netdev->addr_len);
5037         ixgbe_flush_sw_mac_table(adapter);
5038         ixgbe_mac_set_default_filter(adapter, old_addr);
5039
5040         /* update SAN MAC vmdq pool selection */
5041         if (hw->mac.san_mac_rar_index)
5042                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5043
5044         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5045                 ixgbe_ptp_reset(adapter);
5046
5047         if (hw->phy.ops.set_phy_power) {
5048                 if (!netif_running(adapter->netdev) && !adapter->wol)
5049                         hw->phy.ops.set_phy_power(hw, false);
5050                 else
5051                         hw->phy.ops.set_phy_power(hw, true);
5052         }
5053 }
5054
5055 /**
5056  * ixgbe_clean_tx_ring - Free Tx Buffers
5057  * @tx_ring: ring to be cleaned
5058  **/
5059 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5060 {
5061         struct ixgbe_tx_buffer *tx_buffer_info;
5062         unsigned long size;
5063         u16 i;
5064
5065         /* ring already cleared, nothing to do */
5066         if (!tx_ring->tx_buffer_info)
5067                 return;
5068
5069         /* Free all the Tx ring sk_buffs */
5070         for (i = 0; i < tx_ring->count; i++) {
5071                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5072                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
5073         }
5074
5075         netdev_tx_reset_queue(txring_txq(tx_ring));
5076
5077         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5078         memset(tx_ring->tx_buffer_info, 0, size);
5079
5080         /* Zero out the descriptor ring */
5081         memset(tx_ring->desc, 0, tx_ring->size);
5082
5083         tx_ring->next_to_use = 0;
5084         tx_ring->next_to_clean = 0;
5085 }
5086
5087 /**
5088  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5089  * @adapter: board private structure
5090  **/
5091 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5092 {
5093         int i;
5094
5095         for (i = 0; i < adapter->num_rx_queues; i++)
5096                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5097 }
5098
5099 /**
5100  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5101  * @adapter: board private structure
5102  **/
5103 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5104 {
5105         int i;
5106
5107         for (i = 0; i < adapter->num_tx_queues; i++)
5108                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5109 }
5110
5111 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5112 {
5113         struct hlist_node *node2;
5114         struct ixgbe_fdir_filter *filter;
5115
5116         spin_lock(&adapter->fdir_perfect_lock);
5117
5118         hlist_for_each_entry_safe(filter, node2,
5119                                   &adapter->fdir_filter_list, fdir_node) {
5120                 hlist_del(&filter->fdir_node);
5121                 kfree(filter);
5122         }
5123         adapter->fdir_filter_count = 0;
5124
5125         spin_unlock(&adapter->fdir_perfect_lock);
5126 }
5127
5128 void ixgbe_down(struct ixgbe_adapter *adapter)
5129 {
5130         struct net_device *netdev = adapter->netdev;
5131         struct ixgbe_hw *hw = &adapter->hw;
5132         struct net_device *upper;
5133         struct list_head *iter;
5134         int i;
5135
5136         /* signal that we are down to the interrupt handler */
5137         if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5138                 return; /* do nothing if already down */
5139
5140         /* disable receives */
5141         hw->mac.ops.disable_rx(hw);
5142
5143         /* disable all enabled rx queues */
5144         for (i = 0; i < adapter->num_rx_queues; i++)
5145                 /* this call also flushes the previous write */
5146                 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5147
5148         usleep_range(10000, 20000);
5149
5150         netif_tx_stop_all_queues(netdev);
5151
5152         /* call carrier off first to avoid false dev_watchdog timeouts */
5153         netif_carrier_off(netdev);
5154         netif_tx_disable(netdev);
5155
5156         /* disable any upper devices */
5157         netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
5158                 if (netif_is_macvlan(upper)) {
5159                         struct macvlan_dev *vlan = netdev_priv(upper);
5160
5161                         if (vlan->fwd_priv) {
5162                                 netif_tx_stop_all_queues(upper);
5163                                 netif_carrier_off(upper);
5164                                 netif_tx_disable(upper);
5165                         }
5166                 }
5167         }
5168
5169         ixgbe_irq_disable(adapter);
5170
5171         ixgbe_napi_disable_all(adapter);
5172
5173         adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
5174                              IXGBE_FLAG2_RESET_REQUESTED);
5175         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5176
5177         del_timer_sync(&adapter->service_timer);
5178
5179         if (adapter->num_vfs) {
5180                 /* Clear EITR Select mapping */
5181                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5182
5183                 /* Mark all the VFs as inactive */
5184                 for (i = 0 ; i < adapter->num_vfs; i++)
5185                         adapter->vfinfo[i].clear_to_send = false;
5186
5187                 /* ping all the active vfs to let them know we are going down */
5188                 ixgbe_ping_all_vfs(adapter);
5189
5190                 /* Disable all VFTE/VFRE TX/RX */
5191                 ixgbe_disable_tx_rx(adapter);
5192         }
5193
5194         /* disable transmits in the hardware now that interrupts are off */
5195         for (i = 0; i < adapter->num_tx_queues; i++) {
5196                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5197                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5198         }
5199
5200         /* Disable the Tx DMA engine on 82599 and later MAC */
5201         switch (hw->mac.type) {
5202         case ixgbe_mac_82599EB:
5203         case ixgbe_mac_X540:
5204         case ixgbe_mac_X550:
5205         case ixgbe_mac_X550EM_x:
5206                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5207                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5208                                  ~IXGBE_DMATXCTL_TE));
5209                 break;
5210         default:
5211                 break;
5212         }
5213
5214         if (!pci_channel_offline(adapter->pdev))
5215                 ixgbe_reset(adapter);
5216
5217         /* power down the optics for 82599 SFP+ fiber */
5218         if (hw->mac.ops.disable_tx_laser)
5219                 hw->mac.ops.disable_tx_laser(hw);
5220
5221         ixgbe_clean_all_tx_rings(adapter);
5222         ixgbe_clean_all_rx_rings(adapter);
5223
5224 #ifdef CONFIG_IXGBE_DCA
5225         /* since we reset the hardware DCA settings were cleared */
5226         ixgbe_setup_dca(adapter);
5227 #endif
5228 }
5229
5230 /**
5231  * ixgbe_tx_timeout - Respond to a Tx Hang
5232  * @netdev: network interface device structure
5233  **/
5234 static void ixgbe_tx_timeout(struct net_device *netdev)
5235 {
5236         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5237
5238         /* Do the reset outside of interrupt context */
5239         ixgbe_tx_timeout_reset(adapter);
5240 }
5241
5242 /**
5243  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5244  * @adapter: board private structure to initialize
5245  *
5246  * ixgbe_sw_init initializes the Adapter private data structure.
5247  * Fields are initialized based on PCI device information and
5248  * OS network device settings (MTU size).
5249  **/
5250 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
5251 {
5252         struct ixgbe_hw *hw = &adapter->hw;
5253         struct pci_dev *pdev = adapter->pdev;
5254         unsigned int rss, fdir;
5255         u32 fwsm;
5256 #ifdef CONFIG_IXGBE_DCB
5257         int j;
5258         struct tc_configuration *tc;
5259 #endif
5260
5261         /* PCI config space info */
5262
5263         hw->vendor_id = pdev->vendor;
5264         hw->device_id = pdev->device;
5265         hw->revision_id = pdev->revision;
5266         hw->subsystem_vendor_id = pdev->subsystem_vendor;
5267         hw->subsystem_device_id = pdev->subsystem_device;
5268
5269         /* Set common capability flags and settings */
5270         rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5271         adapter->ring_feature[RING_F_RSS].limit = rss;
5272         adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5273         adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
5274         adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5275         adapter->atr_sample_rate = 20;
5276         fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5277         adapter->ring_feature[RING_F_FDIR].limit = fdir;
5278         adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5279 #ifdef CONFIG_IXGBE_DCA
5280         adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5281 #endif
5282 #ifdef IXGBE_FCOE
5283         adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5284         adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5285 #ifdef CONFIG_IXGBE_DCB
5286         /* Default traffic class to use for FCoE */
5287         adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5288 #endif /* CONFIG_IXGBE_DCB */
5289 #endif /* IXGBE_FCOE */
5290
5291         adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5292                                      hw->mac.num_rar_entries,
5293                                      GFP_ATOMIC);
5294
5295         /* Set MAC specific capability flags and exceptions */
5296         switch (hw->mac.type) {
5297         case ixgbe_mac_82598EB:
5298                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5299                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
5300
5301                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5302                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5303
5304                 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5305                 adapter->ring_feature[RING_F_FDIR].limit = 0;
5306                 adapter->atr_sample_rate = 0;
5307                 adapter->fdir_pballoc = 0;
5308 #ifdef IXGBE_FCOE
5309                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5310                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5311 #ifdef CONFIG_IXGBE_DCB
5312                 adapter->fcoe.up = 0;
5313 #endif /* IXGBE_DCB */
5314 #endif /* IXGBE_FCOE */
5315                 break;
5316         case ixgbe_mac_82599EB:
5317                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5318                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5319                 break;
5320         case ixgbe_mac_X540:
5321                 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
5322                 if (fwsm & IXGBE_FWSM_TS_ENABLED)
5323                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5324                 break;
5325         case ixgbe_mac_X550EM_x:
5326         case ixgbe_mac_X550:
5327 #ifdef CONFIG_IXGBE_DCA
5328                 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5329 #endif
5330 #ifdef CONFIG_IXGBE_VXLAN
5331                 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
5332 #endif
5333                 break;
5334         default:
5335                 break;
5336         }
5337
5338 #ifdef IXGBE_FCOE
5339         /* FCoE support exists, always init the FCoE lock */
5340         spin_lock_init(&adapter->fcoe.lock);
5341
5342 #endif
5343         /* n-tuple support exists, always init our spinlock */
5344         spin_lock_init(&adapter->fdir_perfect_lock);
5345
5346 #ifdef CONFIG_IXGBE_DCB
5347         switch (hw->mac.type) {
5348         case ixgbe_mac_X540:
5349         case ixgbe_mac_X550:
5350         case ixgbe_mac_X550EM_x:
5351                 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5352                 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5353                 break;
5354         default:
5355                 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5356                 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5357                 break;
5358         }
5359
5360         /* Configure DCB traffic classes */
5361         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5362                 tc = &adapter->dcb_cfg.tc_config[j];
5363                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5364                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5365                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5366                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5367                 tc->dcb_pfc = pfc_disabled;
5368         }
5369
5370         /* Initialize default user to priority mapping, UPx->TC0 */
5371         tc = &adapter->dcb_cfg.tc_config[0];
5372         tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5373         tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5374
5375         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5376         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5377         adapter->dcb_cfg.pfc_mode_enable = false;
5378         adapter->dcb_set_bitmap = 0x00;
5379         adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5380         memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5381                sizeof(adapter->temp_dcb_cfg));
5382
5383 #endif
5384
5385         /* default flow control settings */
5386         hw->fc.requested_mode = ixgbe_fc_full;
5387         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
5388         ixgbe_pbthresh_setup(adapter);
5389         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5390         hw->fc.send_xon = true;
5391         hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5392
5393 #ifdef CONFIG_PCI_IOV
5394         if (max_vfs > 0)
5395                 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5396
5397         /* assign number of SR-IOV VFs */
5398         if (hw->mac.type != ixgbe_mac_82598EB) {
5399                 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5400                         adapter->num_vfs = 0;
5401                         e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5402                 } else {
5403                         adapter->num_vfs = max_vfs;
5404                 }
5405         }
5406 #endif /* CONFIG_PCI_IOV */
5407
5408         /* enable itr by default in dynamic mode */
5409         adapter->rx_itr_setting = 1;
5410         adapter->tx_itr_setting = 1;
5411
5412         /* set default ring sizes */
5413         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5414         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5415
5416         /* set default work limits */
5417         adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5418
5419         /* initialize eeprom parameters */
5420         if (ixgbe_init_eeprom_params_generic(hw)) {
5421                 e_dev_err("EEPROM initialization failed\n");
5422                 return -EIO;
5423         }
5424
5425         /* PF holds first pool slot */
5426         set_bit(0, &adapter->fwd_bitmask);
5427         set_bit(__IXGBE_DOWN, &adapter->state);
5428
5429         return 0;
5430 }
5431
5432 /**
5433  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5434  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5435  *
5436  * Return 0 on success, negative on failure
5437  **/
5438 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5439 {
5440         struct device *dev = tx_ring->dev;
5441         int orig_node = dev_to_node(dev);
5442         int ring_node = -1;
5443         int size;
5444
5445         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5446
5447         if (tx_ring->q_vector)
5448                 ring_node = tx_ring->q_vector->numa_node;
5449
5450         tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5451         if (!tx_ring->tx_buffer_info)
5452                 tx_ring->tx_buffer_info = vzalloc(size);
5453         if (!tx_ring->tx_buffer_info)
5454                 goto err;
5455
5456         u64_stats_init(&tx_ring->syncp);
5457
5458         /* round up to nearest 4K */
5459         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5460         tx_ring->size = ALIGN(tx_ring->size, 4096);
5461
5462         set_dev_node(dev, ring_node);
5463         tx_ring->desc = dma_alloc_coherent(dev,
5464                                            tx_ring->size,
5465                                            &tx_ring->dma,
5466                                            GFP_KERNEL);
5467         set_dev_node(dev, orig_node);
5468         if (!tx_ring->desc)
5469                 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5470                                                    &tx_ring->dma, GFP_KERNEL);
5471         if (!tx_ring->desc)
5472                 goto err;
5473
5474         tx_ring->next_to_use = 0;
5475         tx_ring->next_to_clean = 0;
5476         return 0;
5477
5478 err:
5479         vfree(tx_ring->tx_buffer_info);
5480         tx_ring->tx_buffer_info = NULL;
5481         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5482         return -ENOMEM;
5483 }
5484
5485 /**
5486  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5487  * @adapter: board private structure
5488  *
5489  * If this function returns with an error, then it's possible one or
5490  * more of the rings is populated (while the rest are not).  It is the
5491  * callers duty to clean those orphaned rings.
5492  *
5493  * Return 0 on success, negative on failure
5494  **/
5495 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5496 {
5497         int i, err = 0;
5498
5499         for (i = 0; i < adapter->num_tx_queues; i++) {
5500                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5501                 if (!err)
5502                         continue;
5503
5504                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5505                 goto err_setup_tx;
5506         }
5507
5508         return 0;
5509 err_setup_tx:
5510         /* rewind the index freeing the rings as we go */
5511         while (i--)
5512                 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5513         return err;
5514 }
5515
5516 /**
5517  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5518  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5519  *
5520  * Returns 0 on success, negative on failure
5521  **/
5522 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5523 {
5524         struct device *dev = rx_ring->dev;
5525         int orig_node = dev_to_node(dev);
5526         int ring_node = -1;
5527         int size;
5528
5529         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5530
5531         if (rx_ring->q_vector)
5532                 ring_node = rx_ring->q_vector->numa_node;
5533
5534         rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5535         if (!rx_ring->rx_buffer_info)
5536                 rx_ring->rx_buffer_info = vzalloc(size);
5537         if (!rx_ring->rx_buffer_info)
5538                 goto err;
5539
5540         u64_stats_init(&rx_ring->syncp);
5541
5542         /* Round up to nearest 4K */
5543         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5544         rx_ring->size = ALIGN(rx_ring->size, 4096);
5545
5546         set_dev_node(dev, ring_node);
5547         rx_ring->desc = dma_alloc_coherent(dev,
5548                                            rx_ring->size,
5549                                            &rx_ring->dma,
5550                                            GFP_KERNEL);
5551         set_dev_node(dev, orig_node);
5552         if (!rx_ring->desc)
5553                 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5554                                                    &rx_ring->dma, GFP_KERNEL);
5555         if (!rx_ring->desc)
5556                 goto err;
5557
5558         rx_ring->next_to_clean = 0;
5559         rx_ring->next_to_use = 0;
5560
5561         return 0;
5562 err:
5563         vfree(rx_ring->rx_buffer_info);
5564         rx_ring->rx_buffer_info = NULL;
5565         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5566         return -ENOMEM;
5567 }
5568
5569 /**
5570  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5571  * @adapter: board private structure
5572  *
5573  * If this function returns with an error, then it's possible one or
5574  * more of the rings is populated (while the rest are not).  It is the
5575  * callers duty to clean those orphaned rings.
5576  *
5577  * Return 0 on success, negative on failure
5578  **/
5579 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5580 {
5581         int i, err = 0;
5582
5583         for (i = 0; i < adapter->num_rx_queues; i++) {
5584                 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5585                 if (!err)
5586                         continue;
5587
5588                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5589                 goto err_setup_rx;
5590         }
5591
5592 #ifdef IXGBE_FCOE
5593         err = ixgbe_setup_fcoe_ddp_resources(adapter);
5594         if (!err)
5595 #endif
5596                 return 0;
5597 err_setup_rx:
5598         /* rewind the index freeing the rings as we go */
5599         while (i--)
5600                 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5601         return err;
5602 }
5603
5604 /**
5605  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5606  * @tx_ring: Tx descriptor ring for a specific queue
5607  *
5608  * Free all transmit software resources
5609  **/
5610 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5611 {
5612         ixgbe_clean_tx_ring(tx_ring);
5613
5614         vfree(tx_ring->tx_buffer_info);
5615         tx_ring->tx_buffer_info = NULL;
5616
5617         /* if not set, then don't free */
5618         if (!tx_ring->desc)
5619                 return;
5620
5621         dma_free_coherent(tx_ring->dev, tx_ring->size,
5622                           tx_ring->desc, tx_ring->dma);
5623
5624         tx_ring->desc = NULL;
5625 }
5626
5627 /**
5628  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5629  * @adapter: board private structure
5630  *
5631  * Free all transmit software resources
5632  **/
5633 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5634 {
5635         int i;
5636
5637         for (i = 0; i < adapter->num_tx_queues; i++)
5638                 if (adapter->tx_ring[i]->desc)
5639                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
5640 }
5641
5642 /**
5643  * ixgbe_free_rx_resources - Free Rx Resources
5644  * @rx_ring: ring to clean the resources from
5645  *
5646  * Free all receive software resources
5647  **/
5648 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5649 {
5650         ixgbe_clean_rx_ring(rx_ring);
5651
5652         vfree(rx_ring->rx_buffer_info);
5653         rx_ring->rx_buffer_info = NULL;
5654
5655         /* if not set, then don't free */
5656         if (!rx_ring->desc)
5657                 return;
5658
5659         dma_free_coherent(rx_ring->dev, rx_ring->size,
5660                           rx_ring->desc, rx_ring->dma);
5661
5662         rx_ring->desc = NULL;
5663 }
5664
5665 /**
5666  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5667  * @adapter: board private structure
5668  *
5669  * Free all receive software resources
5670  **/
5671 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5672 {
5673         int i;
5674
5675 #ifdef IXGBE_FCOE
5676         ixgbe_free_fcoe_ddp_resources(adapter);
5677
5678 #endif
5679         for (i = 0; i < adapter->num_rx_queues; i++)
5680                 if (adapter->rx_ring[i]->desc)
5681                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
5682 }
5683
5684 /**
5685  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5686  * @netdev: network interface device structure
5687  * @new_mtu: new value for maximum frame size
5688  *
5689  * Returns 0 on success, negative on failure
5690  **/
5691 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5692 {
5693         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5694         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5695
5696         /* MTU < 68 is an error and causes problems on some kernels */
5697         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5698                 return -EINVAL;
5699
5700         /*
5701          * For 82599EB we cannot allow legacy VFs to enable their receive
5702          * paths when MTU greater than 1500 is configured.  So display a
5703          * warning that legacy VFs will be disabled.
5704          */
5705         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5706             (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5707             (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5708                 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5709
5710         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5711
5712         /* must set new MTU before calling down or up */
5713         netdev->mtu = new_mtu;
5714
5715         if (netif_running(netdev))
5716                 ixgbe_reinit_locked(adapter);
5717
5718         return 0;
5719 }
5720
5721 /**
5722  * ixgbe_open - Called when a network interface is made active
5723  * @netdev: network interface device structure
5724  *
5725  * Returns 0 on success, negative value on failure
5726  *
5727  * The open entry point is called when a network interface is made
5728  * active by the system (IFF_UP).  At this point all resources needed
5729  * for transmit and receive operations are allocated, the interrupt
5730  * handler is registered with the OS, the watchdog timer is started,
5731  * and the stack is notified that the interface is ready.
5732  **/
5733 static int ixgbe_open(struct net_device *netdev)
5734 {
5735         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5736         struct ixgbe_hw *hw = &adapter->hw;
5737         int err, queues;
5738
5739         /* disallow open during test */
5740         if (test_bit(__IXGBE_TESTING, &adapter->state))
5741                 return -EBUSY;
5742
5743         netif_carrier_off(netdev);
5744
5745         /* allocate transmit descriptors */
5746         err = ixgbe_setup_all_tx_resources(adapter);
5747         if (err)
5748                 goto err_setup_tx;
5749
5750         /* allocate receive descriptors */
5751         err = ixgbe_setup_all_rx_resources(adapter);
5752         if (err)
5753                 goto err_setup_rx;
5754
5755         ixgbe_configure(adapter);
5756
5757         err = ixgbe_request_irq(adapter);
5758         if (err)
5759                 goto err_req_irq;
5760
5761         /* Notify the stack of the actual queue counts. */
5762         if (adapter->num_rx_pools > 1)
5763                 queues = adapter->num_rx_queues_per_pool;
5764         else
5765                 queues = adapter->num_tx_queues;
5766
5767         err = netif_set_real_num_tx_queues(netdev, queues);
5768         if (err)
5769                 goto err_set_queues;
5770
5771         if (adapter->num_rx_pools > 1 &&
5772             adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
5773                 queues = IXGBE_MAX_L2A_QUEUES;
5774         else
5775                 queues = adapter->num_rx_queues;
5776         err = netif_set_real_num_rx_queues(netdev, queues);
5777         if (err)
5778                 goto err_set_queues;
5779
5780         ixgbe_ptp_init(adapter);
5781
5782         ixgbe_up_complete(adapter);
5783
5784         ixgbe_clear_vxlan_port(adapter);
5785 #ifdef CONFIG_IXGBE_VXLAN
5786         vxlan_get_rx_port(netdev);
5787 #endif
5788
5789         return 0;
5790
5791 err_set_queues:
5792         ixgbe_free_irq(adapter);
5793 err_req_irq:
5794         ixgbe_free_all_rx_resources(adapter);
5795         if (hw->phy.ops.set_phy_power && !adapter->wol)
5796                 hw->phy.ops.set_phy_power(&adapter->hw, false);
5797 err_setup_rx:
5798         ixgbe_free_all_tx_resources(adapter);
5799 err_setup_tx:
5800         ixgbe_reset(adapter);
5801
5802         return err;
5803 }
5804
5805 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
5806 {
5807         ixgbe_ptp_suspend(adapter);
5808
5809         if (adapter->hw.phy.ops.enter_lplu) {
5810                 adapter->hw.phy.reset_disable = true;
5811                 ixgbe_down(adapter);
5812                 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
5813                 adapter->hw.phy.reset_disable = false;
5814         } else {
5815                 ixgbe_down(adapter);
5816         }
5817
5818         ixgbe_free_irq(adapter);
5819
5820         ixgbe_free_all_tx_resources(adapter);
5821         ixgbe_free_all_rx_resources(adapter);
5822 }
5823
5824 /**
5825  * ixgbe_close - Disables a network interface
5826  * @netdev: network interface device structure
5827  *
5828  * Returns 0, this is not allowed to fail
5829  *
5830  * The close entry point is called when an interface is de-activated
5831  * by the OS.  The hardware is still under the drivers control, but
5832  * needs to be disabled.  A global MAC reset is issued to stop the
5833  * hardware, and all transmit and receive resources are freed.
5834  **/
5835 static int ixgbe_close(struct net_device *netdev)
5836 {
5837         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5838
5839         ixgbe_ptp_stop(adapter);
5840
5841         ixgbe_close_suspend(adapter);
5842
5843         ixgbe_fdir_filter_exit(adapter);
5844
5845         ixgbe_release_hw_control(adapter);
5846
5847         return 0;
5848 }
5849
5850 #ifdef CONFIG_PM
5851 static int ixgbe_resume(struct pci_dev *pdev)
5852 {
5853         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5854         struct net_device *netdev = adapter->netdev;
5855         u32 err;
5856
5857         adapter->hw.hw_addr = adapter->io_addr;
5858         pci_set_power_state(pdev, PCI_D0);
5859         pci_restore_state(pdev);
5860         /*
5861          * pci_restore_state clears dev->state_saved so call
5862          * pci_save_state to restore it.
5863          */
5864         pci_save_state(pdev);
5865
5866         err = pci_enable_device_mem(pdev);
5867         if (err) {
5868                 e_dev_err("Cannot enable PCI device from suspend\n");
5869                 return err;
5870         }
5871         smp_mb__before_atomic();
5872         clear_bit(__IXGBE_DISABLED, &adapter->state);
5873         pci_set_master(pdev);
5874
5875         pci_wake_from_d3(pdev, false);
5876
5877         ixgbe_reset(adapter);
5878
5879         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
5880
5881         rtnl_lock();
5882         err = ixgbe_init_interrupt_scheme(adapter);
5883         if (!err && netif_running(netdev))
5884                 err = ixgbe_open(netdev);
5885
5886         rtnl_unlock();
5887
5888         if (err)
5889                 return err;
5890
5891         netif_device_attach(netdev);
5892
5893         return 0;
5894 }
5895 #endif /* CONFIG_PM */
5896
5897 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
5898 {
5899         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
5900         struct net_device *netdev = adapter->netdev;
5901         struct ixgbe_hw *hw = &adapter->hw;
5902         u32 ctrl, fctrl;
5903         u32 wufc = adapter->wol;
5904 #ifdef CONFIG_PM
5905         int retval = 0;
5906 #endif
5907
5908         netif_device_detach(netdev);
5909
5910         rtnl_lock();
5911         if (netif_running(netdev))
5912                 ixgbe_close_suspend(adapter);
5913         rtnl_unlock();
5914
5915         ixgbe_clear_interrupt_scheme(adapter);
5916
5917 #ifdef CONFIG_PM
5918         retval = pci_save_state(pdev);
5919         if (retval)
5920                 return retval;
5921
5922 #endif
5923         if (hw->mac.ops.stop_link_on_d3)
5924                 hw->mac.ops.stop_link_on_d3(hw);
5925
5926         if (wufc) {
5927                 ixgbe_set_rx_mode(netdev);
5928
5929                 /* enable the optics for 82599 SFP+ fiber as we can WoL */
5930                 if (hw->mac.ops.enable_tx_laser)
5931                         hw->mac.ops.enable_tx_laser(hw);
5932
5933                 /* turn on all-multi mode if wake on multicast is enabled */
5934                 if (wufc & IXGBE_WUFC_MC) {
5935                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
5936                         fctrl |= IXGBE_FCTRL_MPE;
5937                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
5938                 }
5939
5940                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
5941                 ctrl |= IXGBE_CTRL_GIO_DIS;
5942                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
5943
5944                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
5945         } else {
5946                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
5947                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
5948         }
5949
5950         switch (hw->mac.type) {
5951         case ixgbe_mac_82598EB:
5952                 pci_wake_from_d3(pdev, false);
5953                 break;
5954         case ixgbe_mac_82599EB:
5955         case ixgbe_mac_X540:
5956         case ixgbe_mac_X550:
5957         case ixgbe_mac_X550EM_x:
5958                 pci_wake_from_d3(pdev, !!wufc);
5959                 break;
5960         default:
5961                 break;
5962         }
5963
5964         *enable_wake = !!wufc;
5965         if (hw->phy.ops.set_phy_power && !*enable_wake)
5966                 hw->phy.ops.set_phy_power(hw, false);
5967
5968         ixgbe_release_hw_control(adapter);
5969
5970         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
5971                 pci_disable_device(pdev);
5972
5973         return 0;
5974 }
5975
5976 #ifdef CONFIG_PM
5977 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
5978 {
5979         int retval;
5980         bool wake;
5981
5982         retval = __ixgbe_shutdown(pdev, &wake);
5983         if (retval)
5984                 return retval;
5985
5986         if (wake) {
5987                 pci_prepare_to_sleep(pdev);
5988         } else {
5989                 pci_wake_from_d3(pdev, false);
5990                 pci_set_power_state(pdev, PCI_D3hot);
5991         }
5992
5993         return 0;
5994 }
5995 #endif /* CONFIG_PM */
5996
5997 static void ixgbe_shutdown(struct pci_dev *pdev)
5998 {
5999         bool wake;
6000
6001         __ixgbe_shutdown(pdev, &wake);
6002
6003         if (system_state == SYSTEM_POWER_OFF) {
6004                 pci_wake_from_d3(pdev, wake);
6005                 pci_set_power_state(pdev, PCI_D3hot);
6006         }
6007 }
6008
6009 /**
6010  * ixgbe_update_stats - Update the board statistics counters.
6011  * @adapter: board private structure
6012  **/
6013 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
6014 {
6015         struct net_device *netdev = adapter->netdev;
6016         struct ixgbe_hw *hw = &adapter->hw;
6017         struct ixgbe_hw_stats *hwstats = &adapter->stats;
6018         u64 total_mpc = 0;
6019         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
6020         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
6021         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
6022         u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
6023
6024         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6025             test_bit(__IXGBE_RESETTING, &adapter->state))
6026                 return;
6027
6028         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
6029                 u64 rsc_count = 0;
6030                 u64 rsc_flush = 0;
6031                 for (i = 0; i < adapter->num_rx_queues; i++) {
6032                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
6033                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
6034                 }
6035                 adapter->rsc_total_count = rsc_count;
6036                 adapter->rsc_total_flush = rsc_flush;
6037         }
6038
6039         for (i = 0; i < adapter->num_rx_queues; i++) {
6040                 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
6041                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
6042                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
6043                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
6044                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
6045                 bytes += rx_ring->stats.bytes;
6046                 packets += rx_ring->stats.packets;
6047         }
6048         adapter->non_eop_descs = non_eop_descs;
6049         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
6050         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
6051         adapter->hw_csum_rx_error = hw_csum_rx_error;
6052         netdev->stats.rx_bytes = bytes;
6053         netdev->stats.rx_packets = packets;
6054
6055         bytes = 0;
6056         packets = 0;
6057         /* gather some stats to the adapter struct that are per queue */
6058         for (i = 0; i < adapter->num_tx_queues; i++) {
6059                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6060                 restart_queue += tx_ring->tx_stats.restart_queue;
6061                 tx_busy += tx_ring->tx_stats.tx_busy;
6062                 bytes += tx_ring->stats.bytes;
6063                 packets += tx_ring->stats.packets;
6064         }
6065         adapter->restart_queue = restart_queue;
6066         adapter->tx_busy = tx_busy;
6067         netdev->stats.tx_bytes = bytes;
6068         netdev->stats.tx_packets = packets;
6069
6070         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
6071
6072         /* 8 register reads */
6073         for (i = 0; i < 8; i++) {
6074                 /* for packet buffers not used, the register should read 0 */
6075                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6076                 missed_rx += mpc;
6077                 hwstats->mpc[i] += mpc;
6078                 total_mpc += hwstats->mpc[i];
6079                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6080                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6081                 switch (hw->mac.type) {
6082                 case ixgbe_mac_82598EB:
6083                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6084                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6085                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6086                         hwstats->pxonrxc[i] +=
6087                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6088                         break;
6089                 case ixgbe_mac_82599EB:
6090                 case ixgbe_mac_X540:
6091                 case ixgbe_mac_X550:
6092                 case ixgbe_mac_X550EM_x:
6093                         hwstats->pxonrxc[i] +=
6094                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6095                         break;
6096                 default:
6097                         break;
6098                 }
6099         }
6100
6101         /*16 register reads */
6102         for (i = 0; i < 16; i++) {
6103                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
6104                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
6105                 if ((hw->mac.type == ixgbe_mac_82599EB) ||
6106                     (hw->mac.type == ixgbe_mac_X540) ||
6107                     (hw->mac.type == ixgbe_mac_X550) ||
6108                     (hw->mac.type == ixgbe_mac_X550EM_x)) {
6109                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
6110                         IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
6111                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
6112                         IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
6113                 }
6114         }
6115
6116         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
6117         /* work around hardware counting issue */
6118         hwstats->gprc -= missed_rx;
6119
6120         ixgbe_update_xoff_received(adapter);
6121
6122         /* 82598 hardware only has a 32 bit counter in the high register */
6123         switch (hw->mac.type) {
6124         case ixgbe_mac_82598EB:
6125                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
6126                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
6127                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
6128                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
6129                 break;
6130         case ixgbe_mac_X540:
6131         case ixgbe_mac_X550:
6132         case ixgbe_mac_X550EM_x:
6133                 /* OS2BMC stats are X540 and later */
6134                 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
6135                 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
6136                 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
6137                 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
6138         case ixgbe_mac_82599EB:
6139                 for (i = 0; i < 16; i++)
6140                         adapter->hw_rx_no_dma_resources +=
6141                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
6142                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
6143                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
6144                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
6145                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
6146                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
6147                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
6148                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
6149                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
6150                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
6151 #ifdef IXGBE_FCOE
6152                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
6153                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
6154                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
6155                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
6156                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
6157                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
6158                 /* Add up per cpu counters for total ddp aloc fail */
6159                 if (adapter->fcoe.ddp_pool) {
6160                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
6161                         struct ixgbe_fcoe_ddp_pool *ddp_pool;
6162                         unsigned int cpu;
6163                         u64 noddp = 0, noddp_ext_buff = 0;
6164                         for_each_possible_cpu(cpu) {
6165                                 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
6166                                 noddp += ddp_pool->noddp;
6167                                 noddp_ext_buff += ddp_pool->noddp_ext_buff;
6168                         }
6169                         hwstats->fcoe_noddp = noddp;
6170                         hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6171                 }
6172 #endif /* IXGBE_FCOE */
6173                 break;
6174         default:
6175                 break;
6176         }
6177         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6178         hwstats->bprc += bprc;
6179         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6180         if (hw->mac.type == ixgbe_mac_82598EB)
6181                 hwstats->mprc -= bprc;
6182         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6183         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6184         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6185         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6186         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6187         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6188         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6189         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6190         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6191         hwstats->lxontxc += lxon;
6192         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6193         hwstats->lxofftxc += lxoff;
6194         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6195         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6196         /*
6197          * 82598 errata - tx of flow control packets is included in tx counters
6198          */
6199         xon_off_tot = lxon + lxoff;
6200         hwstats->gptc -= xon_off_tot;
6201         hwstats->mptc -= xon_off_tot;
6202         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6203         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6204         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6205         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6206         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6207         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6208         hwstats->ptc64 -= xon_off_tot;
6209         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6210         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6211         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6212         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6213         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6214         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6215
6216         /* Fill out the OS statistics structure */
6217         netdev->stats.multicast = hwstats->mprc;
6218
6219         /* Rx Errors */
6220         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6221         netdev->stats.rx_dropped = 0;
6222         netdev->stats.rx_length_errors = hwstats->rlec;
6223         netdev->stats.rx_crc_errors = hwstats->crcerrs;
6224         netdev->stats.rx_missed_errors = total_mpc;
6225 }
6226
6227 /**
6228  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6229  * @adapter: pointer to the device adapter structure
6230  **/
6231 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6232 {
6233         struct ixgbe_hw *hw = &adapter->hw;
6234         int i;
6235
6236         if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6237                 return;
6238
6239         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6240
6241         /* if interface is down do nothing */
6242         if (test_bit(__IXGBE_DOWN, &adapter->state))
6243                 return;
6244
6245         /* do nothing if we are not using signature filters */
6246         if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6247                 return;
6248
6249         adapter->fdir_overflow++;
6250
6251         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6252                 for (i = 0; i < adapter->num_tx_queues; i++)
6253                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6254                                 &(adapter->tx_ring[i]->state));
6255                 /* re-enable flow director interrupts */
6256                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6257         } else {
6258                 e_err(probe, "failed to finish FDIR re-initialization, "
6259                       "ignored adding FDIR ATR filters\n");
6260         }
6261 }
6262
6263 /**
6264  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6265  * @adapter: pointer to the device adapter structure
6266  *
6267  * This function serves two purposes.  First it strobes the interrupt lines
6268  * in order to make certain interrupts are occurring.  Secondly it sets the
6269  * bits needed to check for TX hangs.  As a result we should immediately
6270  * determine if a hang has occurred.
6271  */
6272 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6273 {
6274         struct ixgbe_hw *hw = &adapter->hw;
6275         u64 eics = 0;
6276         int i;
6277
6278         /* If we're down, removing or resetting, just bail */
6279         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6280             test_bit(__IXGBE_REMOVING, &adapter->state) ||
6281             test_bit(__IXGBE_RESETTING, &adapter->state))
6282                 return;
6283
6284         /* Force detection of hung controller */
6285         if (netif_carrier_ok(adapter->netdev)) {
6286                 for (i = 0; i < adapter->num_tx_queues; i++)
6287                         set_check_for_tx_hang(adapter->tx_ring[i]);
6288         }
6289
6290         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6291                 /*
6292                  * for legacy and MSI interrupts don't set any bits
6293                  * that are enabled for EIAM, because this operation
6294                  * would set *both* EIMS and EICS for any bit in EIAM
6295                  */
6296                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6297                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6298         } else {
6299                 /* get one bit for every active tx/rx interrupt vector */
6300                 for (i = 0; i < adapter->num_q_vectors; i++) {
6301                         struct ixgbe_q_vector *qv = adapter->q_vector[i];
6302                         if (qv->rx.ring || qv->tx.ring)
6303                                 eics |= ((u64)1 << i);
6304                 }
6305         }
6306
6307         /* Cause software interrupt to ensure rings are cleaned */
6308         ixgbe_irq_rearm_queues(adapter, eics);
6309 }
6310
6311 /**
6312  * ixgbe_watchdog_update_link - update the link status
6313  * @adapter: pointer to the device adapter structure
6314  * @link_speed: pointer to a u32 to store the link_speed
6315  **/
6316 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6317 {
6318         struct ixgbe_hw *hw = &adapter->hw;
6319         u32 link_speed = adapter->link_speed;
6320         bool link_up = adapter->link_up;
6321         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6322
6323         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6324                 return;
6325
6326         if (hw->mac.ops.check_link) {
6327                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6328         } else {
6329                 /* always assume link is up, if no check link function */
6330                 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6331                 link_up = true;
6332         }
6333
6334         if (adapter->ixgbe_ieee_pfc)
6335                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6336
6337         if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6338                 hw->mac.ops.fc_enable(hw);
6339                 ixgbe_set_rx_drop_en(adapter);
6340         }
6341
6342         if (link_up ||
6343             time_after(jiffies, (adapter->link_check_timeout +
6344                                  IXGBE_TRY_LINK_TIMEOUT))) {
6345                 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6346                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6347                 IXGBE_WRITE_FLUSH(hw);
6348         }
6349
6350         adapter->link_up = link_up;
6351         adapter->link_speed = link_speed;
6352 }
6353
6354 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6355 {
6356 #ifdef CONFIG_IXGBE_DCB
6357         struct net_device *netdev = adapter->netdev;
6358         struct dcb_app app = {
6359                               .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6360                               .protocol = 0,
6361                              };
6362         u8 up = 0;
6363
6364         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6365                 up = dcb_ieee_getapp_mask(netdev, &app);
6366
6367         adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6368 #endif
6369 }
6370
6371 /**
6372  * ixgbe_watchdog_link_is_up - update netif_carrier status and
6373  *                             print link up message
6374  * @adapter: pointer to the device adapter structure
6375  **/
6376 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6377 {
6378         struct net_device *netdev = adapter->netdev;
6379         struct ixgbe_hw *hw = &adapter->hw;
6380         struct net_device *upper;
6381         struct list_head *iter;
6382         u32 link_speed = adapter->link_speed;
6383         const char *speed_str;
6384         bool flow_rx, flow_tx;
6385
6386         /* only continue if link was previously down */
6387         if (netif_carrier_ok(netdev))
6388                 return;
6389
6390         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6391
6392         switch (hw->mac.type) {
6393         case ixgbe_mac_82598EB: {
6394                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6395                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6396                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6397                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6398         }
6399                 break;
6400         case ixgbe_mac_X540:
6401         case ixgbe_mac_X550:
6402         case ixgbe_mac_X550EM_x:
6403         case ixgbe_mac_82599EB: {
6404                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6405                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6406                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6407                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6408         }
6409                 break;
6410         default:
6411                 flow_tx = false;
6412                 flow_rx = false;
6413                 break;
6414         }
6415
6416         adapter->last_rx_ptp_check = jiffies;
6417
6418         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6419                 ixgbe_ptp_start_cyclecounter(adapter);
6420
6421         switch (link_speed) {
6422         case IXGBE_LINK_SPEED_10GB_FULL:
6423                 speed_str = "10 Gbps";
6424                 break;
6425         case IXGBE_LINK_SPEED_2_5GB_FULL:
6426                 speed_str = "2.5 Gbps";
6427                 break;
6428         case IXGBE_LINK_SPEED_1GB_FULL:
6429                 speed_str = "1 Gbps";
6430                 break;
6431         case IXGBE_LINK_SPEED_100_FULL:
6432                 speed_str = "100 Mbps";
6433                 break;
6434         default:
6435                 speed_str = "unknown speed";
6436                 break;
6437         }
6438         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
6439                ((flow_rx && flow_tx) ? "RX/TX" :
6440                (flow_rx ? "RX" :
6441                (flow_tx ? "TX" : "None"))));
6442
6443         netif_carrier_on(netdev);
6444         ixgbe_check_vf_rate_limit(adapter);
6445
6446         /* enable transmits */
6447         netif_tx_wake_all_queues(adapter->netdev);
6448
6449         /* enable any upper devices */
6450         rtnl_lock();
6451         netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6452                 if (netif_is_macvlan(upper)) {
6453                         struct macvlan_dev *vlan = netdev_priv(upper);
6454
6455                         if (vlan->fwd_priv)
6456                                 netif_tx_wake_all_queues(upper);
6457                 }
6458         }
6459         rtnl_unlock();
6460
6461         /* update the default user priority for VFs */
6462         ixgbe_update_default_up(adapter);
6463
6464         /* ping all the active vfs to let them know link has changed */
6465         ixgbe_ping_all_vfs(adapter);
6466 }
6467
6468 /**
6469  * ixgbe_watchdog_link_is_down - update netif_carrier status and
6470  *                               print link down message
6471  * @adapter: pointer to the adapter structure
6472  **/
6473 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6474 {
6475         struct net_device *netdev = adapter->netdev;
6476         struct ixgbe_hw *hw = &adapter->hw;
6477
6478         adapter->link_up = false;
6479         adapter->link_speed = 0;
6480
6481         /* only continue if link was up previously */
6482         if (!netif_carrier_ok(netdev))
6483                 return;
6484
6485         /* poll for SFP+ cable when link is down */
6486         if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6487                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6488
6489         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6490                 ixgbe_ptp_start_cyclecounter(adapter);
6491
6492         e_info(drv, "NIC Link is Down\n");
6493         netif_carrier_off(netdev);
6494
6495         /* ping all the active vfs to let them know link has changed */
6496         ixgbe_ping_all_vfs(adapter);
6497 }
6498
6499 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6500 {
6501         int i;
6502
6503         for (i = 0; i < adapter->num_tx_queues; i++) {
6504                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6505
6506                 if (tx_ring->next_to_use != tx_ring->next_to_clean)
6507                         return true;
6508         }
6509
6510         return false;
6511 }
6512
6513 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6514 {
6515         struct ixgbe_hw *hw = &adapter->hw;
6516         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6517         u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6518
6519         int i, j;
6520
6521         if (!adapter->num_vfs)
6522                 return false;
6523
6524         /* resetting the PF is only needed for MAC before X550 */
6525         if (hw->mac.type >= ixgbe_mac_X550)
6526                 return false;
6527
6528         for (i = 0; i < adapter->num_vfs; i++) {
6529                 for (j = 0; j < q_per_pool; j++) {
6530                         u32 h, t;
6531
6532                         h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6533                         t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6534
6535                         if (h != t)
6536                                 return true;
6537                 }
6538         }
6539
6540         return false;
6541 }
6542
6543 /**
6544  * ixgbe_watchdog_flush_tx - flush queues on link down
6545  * @adapter: pointer to the device adapter structure
6546  **/
6547 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6548 {
6549         if (!netif_carrier_ok(adapter->netdev)) {
6550                 if (ixgbe_ring_tx_pending(adapter) ||
6551                     ixgbe_vf_tx_pending(adapter)) {
6552                         /* We've lost link, so the controller stops DMA,
6553                          * but we've got queued Tx work that's never going
6554                          * to get done, so reset controller to flush Tx.
6555                          * (Do the reset outside of interrupt context).
6556                          */
6557                         e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6558                         adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6559                 }
6560         }
6561 }
6562
6563 #ifdef CONFIG_PCI_IOV
6564 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6565                                       struct pci_dev *vfdev)
6566 {
6567         if (!pci_wait_for_pending_transaction(vfdev))
6568                 e_dev_warn("Issuing VFLR with pending transactions\n");
6569
6570         e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6571         pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6572
6573         msleep(100);
6574 }
6575
6576 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6577 {
6578         struct ixgbe_hw *hw = &adapter->hw;
6579         struct pci_dev *pdev = adapter->pdev;
6580         struct pci_dev *vfdev;
6581         u32 gpc;
6582         int pos;
6583         unsigned short vf_id;
6584
6585         if (!(netif_carrier_ok(adapter->netdev)))
6586                 return;
6587
6588         gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6589         if (gpc) /* If incrementing then no need for the check below */
6590                 return;
6591         /* Check to see if a bad DMA write target from an errant or
6592          * malicious VF has caused a PCIe error.  If so then we can
6593          * issue a VFLR to the offending VF(s) and then resume without
6594          * requesting a full slot reset.
6595          */
6596
6597         if (!pdev)
6598                 return;
6599
6600         pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_SRIOV);
6601         if (!pos)
6602                 return;
6603
6604         /* get the device ID for the VF */
6605         pci_read_config_word(pdev, pos + PCI_SRIOV_VF_DID, &vf_id);
6606
6607         /* check status reg for all VFs owned by this PF */
6608         vfdev = pci_get_device(pdev->vendor, vf_id, NULL);
6609         while (vfdev) {
6610                 if (vfdev->is_virtfn && (vfdev->physfn == pdev)) {
6611                         u16 status_reg;
6612
6613                         pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6614                         if (status_reg & PCI_STATUS_REC_MASTER_ABORT)
6615                                 /* issue VFLR */
6616                                 ixgbe_issue_vf_flr(adapter, vfdev);
6617                 }
6618
6619                 vfdev = pci_get_device(pdev->vendor, vf_id, vfdev);
6620         }
6621 }
6622
6623 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6624 {
6625         u32 ssvpc;
6626
6627         /* Do not perform spoof check for 82598 or if not in IOV mode */
6628         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6629             adapter->num_vfs == 0)
6630                 return;
6631
6632         ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6633
6634         /*
6635          * ssvpc register is cleared on read, if zero then no
6636          * spoofed packets in the last interval.
6637          */
6638         if (!ssvpc)
6639                 return;
6640
6641         e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6642 }
6643 #else
6644 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
6645 {
6646 }
6647
6648 static void
6649 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
6650 {
6651 }
6652 #endif /* CONFIG_PCI_IOV */
6653
6654
6655 /**
6656  * ixgbe_watchdog_subtask - check and bring link up
6657  * @adapter: pointer to the device adapter structure
6658  **/
6659 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6660 {
6661         /* if interface is down, removing or resetting, do nothing */
6662         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6663             test_bit(__IXGBE_REMOVING, &adapter->state) ||
6664             test_bit(__IXGBE_RESETTING, &adapter->state))
6665                 return;
6666
6667         ixgbe_watchdog_update_link(adapter);
6668
6669         if (adapter->link_up)
6670                 ixgbe_watchdog_link_is_up(adapter);
6671         else
6672                 ixgbe_watchdog_link_is_down(adapter);
6673
6674         ixgbe_check_for_bad_vf(adapter);
6675         ixgbe_spoof_check(adapter);
6676         ixgbe_update_stats(adapter);
6677
6678         ixgbe_watchdog_flush_tx(adapter);
6679 }
6680
6681 /**
6682  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6683  * @adapter: the ixgbe adapter structure
6684  **/
6685 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6686 {
6687         struct ixgbe_hw *hw = &adapter->hw;
6688         s32 err;
6689
6690         /* not searching for SFP so there is nothing to do here */
6691         if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6692             !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6693                 return;
6694
6695         /* someone else is in init, wait until next service event */
6696         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6697                 return;
6698
6699         err = hw->phy.ops.identify_sfp(hw);
6700         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6701                 goto sfp_out;
6702
6703         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6704                 /* If no cable is present, then we need to reset
6705                  * the next time we find a good cable. */
6706                 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6707         }
6708
6709         /* exit on error */
6710         if (err)
6711                 goto sfp_out;
6712
6713         /* exit if reset not needed */
6714         if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6715                 goto sfp_out;
6716
6717         adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6718
6719         /*
6720          * A module may be identified correctly, but the EEPROM may not have
6721          * support for that module.  setup_sfp() will fail in that case, so
6722          * we should not allow that module to load.
6723          */
6724         if (hw->mac.type == ixgbe_mac_82598EB)
6725                 err = hw->phy.ops.reset(hw);
6726         else
6727                 err = hw->mac.ops.setup_sfp(hw);
6728
6729         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6730                 goto sfp_out;
6731
6732         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6733         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6734
6735 sfp_out:
6736         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6737
6738         if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6739             (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6740                 e_dev_err("failed to initialize because an unsupported "
6741                           "SFP+ module type was detected.\n");
6742                 e_dev_err("Reload the driver after installing a "
6743                           "supported module.\n");
6744                 unregister_netdev(adapter->netdev);
6745         }
6746 }
6747
6748 /**
6749  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
6750  * @adapter: the ixgbe adapter structure
6751  **/
6752 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
6753 {
6754         struct ixgbe_hw *hw = &adapter->hw;
6755         u32 speed;
6756         bool autoneg = false;
6757
6758         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
6759                 return;
6760
6761         /* someone else is in init, wait until next service event */
6762         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6763                 return;
6764
6765         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
6766
6767         speed = hw->phy.autoneg_advertised;
6768         if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
6769                 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
6770
6771                 /* setup the highest link when no autoneg */
6772                 if (!autoneg) {
6773                         if (speed & IXGBE_LINK_SPEED_10GB_FULL)
6774                                 speed = IXGBE_LINK_SPEED_10GB_FULL;
6775                 }
6776         }
6777
6778         if (hw->mac.ops.setup_link)
6779                 hw->mac.ops.setup_link(hw, speed, true);
6780
6781         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
6782         adapter->link_check_timeout = jiffies;
6783         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6784 }
6785
6786 /**
6787  * ixgbe_service_timer - Timer Call-back
6788  * @data: pointer to adapter cast into an unsigned long
6789  **/
6790 static void ixgbe_service_timer(unsigned long data)
6791 {
6792         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
6793         unsigned long next_event_offset;
6794
6795         /* poll faster when waiting for link */
6796         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
6797                 next_event_offset = HZ / 10;
6798         else
6799                 next_event_offset = HZ * 2;
6800
6801         /* Reset the timer */
6802         mod_timer(&adapter->service_timer, next_event_offset + jiffies);
6803
6804         ixgbe_service_event_schedule(adapter);
6805 }
6806
6807 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
6808 {
6809         struct ixgbe_hw *hw = &adapter->hw;
6810         u32 status;
6811
6812         if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
6813                 return;
6814
6815         adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
6816
6817         if (!hw->phy.ops.handle_lasi)
6818                 return;
6819
6820         status = hw->phy.ops.handle_lasi(&adapter->hw);
6821         if (status != IXGBE_ERR_OVERTEMP)
6822                 return;
6823
6824         e_crit(drv, "%s\n", ixgbe_overheat_msg);
6825 }
6826
6827 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
6828 {
6829         if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
6830                 return;
6831
6832         adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
6833
6834         /* If we're already down, removing or resetting, just bail */
6835         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6836             test_bit(__IXGBE_REMOVING, &adapter->state) ||
6837             test_bit(__IXGBE_RESETTING, &adapter->state))
6838                 return;
6839
6840         ixgbe_dump(adapter);
6841         netdev_err(adapter->netdev, "Reset adapter\n");
6842         adapter->tx_timeout_count++;
6843
6844         rtnl_lock();
6845         ixgbe_reinit_locked(adapter);
6846         rtnl_unlock();
6847 }
6848
6849 /**
6850  * ixgbe_service_task - manages and runs subtasks
6851  * @work: pointer to work_struct containing our data
6852  **/
6853 static void ixgbe_service_task(struct work_struct *work)
6854 {
6855         struct ixgbe_adapter *adapter = container_of(work,
6856                                                      struct ixgbe_adapter,
6857                                                      service_task);
6858         if (ixgbe_removed(adapter->hw.hw_addr)) {
6859                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
6860                         rtnl_lock();
6861                         ixgbe_down(adapter);
6862                         rtnl_unlock();
6863                 }
6864                 ixgbe_service_event_complete(adapter);
6865                 return;
6866         }
6867 #ifdef CONFIG_IXGBE_VXLAN
6868         if (adapter->flags2 & IXGBE_FLAG2_VXLAN_REREG_NEEDED) {
6869                 adapter->flags2 &= ~IXGBE_FLAG2_VXLAN_REREG_NEEDED;
6870                 vxlan_get_rx_port(adapter->netdev);
6871         }
6872 #endif /* CONFIG_IXGBE_VXLAN */
6873         ixgbe_reset_subtask(adapter);
6874         ixgbe_phy_interrupt_subtask(adapter);
6875         ixgbe_sfp_detection_subtask(adapter);
6876         ixgbe_sfp_link_config_subtask(adapter);
6877         ixgbe_check_overtemp_subtask(adapter);
6878         ixgbe_watchdog_subtask(adapter);
6879         ixgbe_fdir_reinit_subtask(adapter);
6880         ixgbe_check_hang_subtask(adapter);
6881
6882         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
6883                 ixgbe_ptp_overflow_check(adapter);
6884                 ixgbe_ptp_rx_hang(adapter);
6885         }
6886
6887         ixgbe_service_event_complete(adapter);
6888 }
6889
6890 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
6891                      struct ixgbe_tx_buffer *first,
6892                      u8 *hdr_len)
6893 {
6894         struct sk_buff *skb = first->skb;
6895         u32 vlan_macip_lens, type_tucmd;
6896         u32 mss_l4len_idx, l4len;
6897         int err;
6898
6899         if (skb->ip_summed != CHECKSUM_PARTIAL)
6900                 return 0;
6901
6902         if (!skb_is_gso(skb))
6903                 return 0;
6904
6905         err = skb_cow_head(skb, 0);
6906         if (err < 0)
6907                 return err;
6908
6909         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
6910         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
6911
6912         if (first->protocol == htons(ETH_P_IP)) {
6913                 struct iphdr *iph = ip_hdr(skb);
6914                 iph->tot_len = 0;
6915                 iph->check = 0;
6916                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
6917                                                          iph->daddr, 0,
6918                                                          IPPROTO_TCP,
6919                                                          0);
6920                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
6921                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6922                                    IXGBE_TX_FLAGS_CSUM |
6923                                    IXGBE_TX_FLAGS_IPV4;
6924         } else if (skb_is_gso_v6(skb)) {
6925                 ipv6_hdr(skb)->payload_len = 0;
6926                 tcp_hdr(skb)->check =
6927                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
6928                                      &ipv6_hdr(skb)->daddr,
6929                                      0, IPPROTO_TCP, 0);
6930                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
6931                                    IXGBE_TX_FLAGS_CSUM;
6932         }
6933
6934         /* compute header lengths */
6935         l4len = tcp_hdrlen(skb);
6936         *hdr_len = skb_transport_offset(skb) + l4len;
6937
6938         /* update gso size and bytecount with header size */
6939         first->gso_segs = skb_shinfo(skb)->gso_segs;
6940         first->bytecount += (first->gso_segs - 1) * *hdr_len;
6941
6942         /* mss_l4len_id: use 0 as index for TSO */
6943         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
6944         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
6945
6946         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
6947         vlan_macip_lens = skb_network_header_len(skb);
6948         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
6949         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
6950
6951         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
6952                           mss_l4len_idx);
6953
6954         return 1;
6955 }
6956
6957 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
6958                           struct ixgbe_tx_buffer *first)
6959 {
6960         struct sk_buff *skb = first->skb;
6961         u32 vlan_macip_lens = 0;
6962         u32 mss_l4len_idx = 0;
6963         u32 type_tucmd = 0;
6964
6965         if (skb->ip_summed != CHECKSUM_PARTIAL) {
6966                 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
6967                     !(first->tx_flags & IXGBE_TX_FLAGS_CC))
6968                         return;
6969                 vlan_macip_lens = skb_network_offset(skb) <<
6970                                   IXGBE_ADVTXD_MACLEN_SHIFT;
6971         } else {
6972                 u8 l4_hdr = 0;
6973                 union {
6974                         struct iphdr *ipv4;
6975                         struct ipv6hdr *ipv6;
6976                         u8 *raw;
6977                 } network_hdr;
6978                 union {
6979                         struct tcphdr *tcphdr;
6980                         u8 *raw;
6981                 } transport_hdr;
6982
6983                 if (skb->encapsulation) {
6984                         network_hdr.raw = skb_inner_network_header(skb);
6985                         transport_hdr.raw = skb_inner_transport_header(skb);
6986                         vlan_macip_lens = skb_inner_network_offset(skb) <<
6987                                           IXGBE_ADVTXD_MACLEN_SHIFT;
6988                 } else {
6989                         network_hdr.raw = skb_network_header(skb);
6990                         transport_hdr.raw = skb_transport_header(skb);
6991                         vlan_macip_lens = skb_network_offset(skb) <<
6992                                           IXGBE_ADVTXD_MACLEN_SHIFT;
6993                 }
6994
6995                 /* use first 4 bits to determine IP version */
6996                 switch (network_hdr.ipv4->version) {
6997                 case IPVERSION:
6998                         vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
6999                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7000                         l4_hdr = network_hdr.ipv4->protocol;
7001                         break;
7002                 case 6:
7003                         vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7004                         l4_hdr = network_hdr.ipv6->nexthdr;
7005                         break;
7006                 default:
7007                         if (unlikely(net_ratelimit())) {
7008                                 dev_warn(tx_ring->dev,
7009                                          "partial checksum but version=%d\n",
7010                                          network_hdr.ipv4->version);
7011                         }
7012                 }
7013
7014                 switch (l4_hdr) {
7015                 case IPPROTO_TCP:
7016                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
7017                         mss_l4len_idx = (transport_hdr.tcphdr->doff * 4) <<
7018                                         IXGBE_ADVTXD_L4LEN_SHIFT;
7019                         break;
7020                 case IPPROTO_SCTP:
7021                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
7022                         mss_l4len_idx = sizeof(struct sctphdr) <<
7023                                         IXGBE_ADVTXD_L4LEN_SHIFT;
7024                         break;
7025                 case IPPROTO_UDP:
7026                         mss_l4len_idx = sizeof(struct udphdr) <<
7027                                         IXGBE_ADVTXD_L4LEN_SHIFT;
7028                         break;
7029                 default:
7030                         if (unlikely(net_ratelimit())) {
7031                                 dev_warn(tx_ring->dev,
7032                                  "partial checksum but l4 proto=%x!\n",
7033                                  l4_hdr);
7034                         }
7035                         break;
7036                 }
7037
7038                 /* update TX checksum flag */
7039                 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
7040         }
7041
7042         /* vlan_macip_lens: MACLEN, VLAN tag */
7043         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7044
7045         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
7046                           type_tucmd, mss_l4len_idx);
7047 }
7048
7049 #define IXGBE_SET_FLAG(_input, _flag, _result) \
7050         ((_flag <= _result) ? \
7051          ((u32)(_input & _flag) * (_result / _flag)) : \
7052          ((u32)(_input & _flag) / (_flag / _result)))
7053
7054 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
7055 {
7056         /* set type for advanced descriptor with frame checksum insertion */
7057         u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
7058                        IXGBE_ADVTXD_DCMD_DEXT |
7059                        IXGBE_ADVTXD_DCMD_IFCS;
7060
7061         /* set HW vlan bit if vlan is present */
7062         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
7063                                    IXGBE_ADVTXD_DCMD_VLE);
7064
7065         /* set segmentation enable bits for TSO/FSO */
7066         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
7067                                    IXGBE_ADVTXD_DCMD_TSE);
7068
7069         /* set timestamp bit if present */
7070         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
7071                                    IXGBE_ADVTXD_MAC_TSTAMP);
7072
7073         /* insert frame checksum */
7074         cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
7075
7076         return cmd_type;
7077 }
7078
7079 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
7080                                    u32 tx_flags, unsigned int paylen)
7081 {
7082         u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
7083
7084         /* enable L4 checksum for TSO and TX checksum offload */
7085         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7086                                         IXGBE_TX_FLAGS_CSUM,
7087                                         IXGBE_ADVTXD_POPTS_TXSM);
7088
7089         /* enble IPv4 checksum for TSO */
7090         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7091                                         IXGBE_TX_FLAGS_IPV4,
7092                                         IXGBE_ADVTXD_POPTS_IXSM);
7093
7094         /*
7095          * Check Context must be set if Tx switch is enabled, which it
7096          * always is for case where virtual functions are running
7097          */
7098         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7099                                         IXGBE_TX_FLAGS_CC,
7100                                         IXGBE_ADVTXD_CC);
7101
7102         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
7103 }
7104
7105 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7106 {
7107         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7108
7109         /* Herbert's original patch had:
7110          *  smp_mb__after_netif_stop_queue();
7111          * but since that doesn't exist yet, just open code it.
7112          */
7113         smp_mb();
7114
7115         /* We need to check again in a case another CPU has just
7116          * made room available.
7117          */
7118         if (likely(ixgbe_desc_unused(tx_ring) < size))
7119                 return -EBUSY;
7120
7121         /* A reprieve! - use start_queue because it doesn't call schedule */
7122         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
7123         ++tx_ring->tx_stats.restart_queue;
7124         return 0;
7125 }
7126
7127 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7128 {
7129         if (likely(ixgbe_desc_unused(tx_ring) >= size))
7130                 return 0;
7131
7132         return __ixgbe_maybe_stop_tx(tx_ring, size);
7133 }
7134
7135 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
7136                        IXGBE_TXD_CMD_RS)
7137
7138 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
7139                          struct ixgbe_tx_buffer *first,
7140                          const u8 hdr_len)
7141 {
7142         struct sk_buff *skb = first->skb;
7143         struct ixgbe_tx_buffer *tx_buffer;
7144         union ixgbe_adv_tx_desc *tx_desc;
7145         struct skb_frag_struct *frag;
7146         dma_addr_t dma;
7147         unsigned int data_len, size;
7148         u32 tx_flags = first->tx_flags;
7149         u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
7150         u16 i = tx_ring->next_to_use;
7151
7152         tx_desc = IXGBE_TX_DESC(tx_ring, i);
7153
7154         ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
7155
7156         size = skb_headlen(skb);
7157         data_len = skb->data_len;
7158
7159 #ifdef IXGBE_FCOE
7160         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
7161                 if (data_len < sizeof(struct fcoe_crc_eof)) {
7162                         size -= sizeof(struct fcoe_crc_eof) - data_len;
7163                         data_len = 0;
7164                 } else {
7165                         data_len -= sizeof(struct fcoe_crc_eof);
7166                 }
7167         }
7168
7169 #endif
7170         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
7171
7172         tx_buffer = first;
7173
7174         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
7175                 if (dma_mapping_error(tx_ring->dev, dma))
7176                         goto dma_error;
7177
7178                 /* record length, and DMA address */
7179                 dma_unmap_len_set(tx_buffer, len, size);
7180                 dma_unmap_addr_set(tx_buffer, dma, dma);
7181
7182                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7183
7184                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
7185                         tx_desc->read.cmd_type_len =
7186                                 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
7187
7188                         i++;
7189                         tx_desc++;
7190                         if (i == tx_ring->count) {
7191                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7192                                 i = 0;
7193                         }
7194                         tx_desc->read.olinfo_status = 0;
7195
7196                         dma += IXGBE_MAX_DATA_PER_TXD;
7197                         size -= IXGBE_MAX_DATA_PER_TXD;
7198
7199                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
7200                 }
7201
7202                 if (likely(!data_len))
7203                         break;
7204
7205                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
7206
7207                 i++;
7208                 tx_desc++;
7209                 if (i == tx_ring->count) {
7210                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7211                         i = 0;
7212                 }
7213                 tx_desc->read.olinfo_status = 0;
7214
7215 #ifdef IXGBE_FCOE
7216                 size = min_t(unsigned int, data_len, skb_frag_size(frag));
7217 #else
7218                 size = skb_frag_size(frag);
7219 #endif
7220                 data_len -= size;
7221
7222                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
7223                                        DMA_TO_DEVICE);
7224
7225                 tx_buffer = &tx_ring->tx_buffer_info[i];
7226         }
7227
7228         /* write last descriptor with RS and EOP bits */
7229         cmd_type |= size | IXGBE_TXD_CMD;
7230         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7231
7232         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7233
7234         /* set the timestamp */
7235         first->time_stamp = jiffies;
7236
7237         /*
7238          * Force memory writes to complete before letting h/w know there
7239          * are new descriptors to fetch.  (Only applicable for weak-ordered
7240          * memory model archs, such as IA-64).
7241          *
7242          * We also need this memory barrier to make certain all of the
7243          * status bits have been updated before next_to_watch is written.
7244          */
7245         wmb();
7246
7247         /* set next_to_watch value indicating a packet is present */
7248         first->next_to_watch = tx_desc;
7249
7250         i++;
7251         if (i == tx_ring->count)
7252                 i = 0;
7253
7254         tx_ring->next_to_use = i;
7255
7256         ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7257
7258         if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7259                 writel(i, tx_ring->tail);
7260
7261                 /* we need this if more than one processor can write to our tail
7262                  * at a time, it synchronizes IO on IA64/Altix systems
7263                  */
7264                 mmiowb();
7265         }
7266
7267         return;
7268 dma_error:
7269         dev_err(tx_ring->dev, "TX DMA map failed\n");
7270
7271         /* clear dma mappings for failed tx_buffer_info map */
7272         for (;;) {
7273                 tx_buffer = &tx_ring->tx_buffer_info[i];
7274                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7275                 if (tx_buffer == first)
7276                         break;
7277                 if (i == 0)
7278                         i = tx_ring->count;
7279                 i--;
7280         }
7281
7282         tx_ring->next_to_use = i;
7283 }
7284
7285 static void ixgbe_atr(struct ixgbe_ring *ring,
7286                       struct ixgbe_tx_buffer *first)
7287 {
7288         struct ixgbe_q_vector *q_vector = ring->q_vector;
7289         union ixgbe_atr_hash_dword input = { .dword = 0 };
7290         union ixgbe_atr_hash_dword common = { .dword = 0 };
7291         union {
7292                 unsigned char *network;
7293                 struct iphdr *ipv4;
7294                 struct ipv6hdr *ipv6;
7295         } hdr;
7296         struct tcphdr *th;
7297         struct sk_buff *skb;
7298 #ifdef CONFIG_IXGBE_VXLAN
7299         u8 encap = false;
7300 #endif /* CONFIG_IXGBE_VXLAN */
7301         __be16 vlan_id;
7302
7303         /* if ring doesn't have a interrupt vector, cannot perform ATR */
7304         if (!q_vector)
7305                 return;
7306
7307         /* do nothing if sampling is disabled */
7308         if (!ring->atr_sample_rate)
7309                 return;
7310
7311         ring->atr_count++;
7312
7313         /* snag network header to get L4 type and address */
7314         skb = first->skb;
7315         hdr.network = skb_network_header(skb);
7316         if (skb->encapsulation) {
7317 #ifdef CONFIG_IXGBE_VXLAN
7318                 struct ixgbe_adapter *adapter = q_vector->adapter;
7319
7320                 if (!adapter->vxlan_port)
7321                         return;
7322                 if (first->protocol != htons(ETH_P_IP) ||
7323                     hdr.ipv4->version != IPVERSION ||
7324                     hdr.ipv4->protocol != IPPROTO_UDP) {
7325                         return;
7326                 }
7327                 if (ntohs(udp_hdr(skb)->dest) != adapter->vxlan_port)
7328                         return;
7329                 encap = true;
7330                 hdr.network = skb_inner_network_header(skb);
7331                 th = inner_tcp_hdr(skb);
7332 #else
7333                 return;
7334 #endif /* CONFIG_IXGBE_VXLAN */
7335         } else {
7336                 /* Currently only IPv4/IPv6 with TCP is supported */
7337                 if ((first->protocol != htons(ETH_P_IPV6) ||
7338                      hdr.ipv6->nexthdr != IPPROTO_TCP) &&
7339                     (first->protocol != htons(ETH_P_IP) ||
7340                      hdr.ipv4->protocol != IPPROTO_TCP))
7341                         return;
7342                 th = tcp_hdr(skb);
7343         }
7344
7345         /* skip this packet since it is invalid or the socket is closing */
7346         if (!th || th->fin)
7347                 return;
7348
7349         /* sample on all syn packets or once every atr sample count */
7350         if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7351                 return;
7352
7353         /* reset sample count */
7354         ring->atr_count = 0;
7355
7356         vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7357
7358         /*
7359          * src and dst are inverted, think how the receiver sees them
7360          *
7361          * The input is broken into two sections, a non-compressed section
7362          * containing vm_pool, vlan_id, and flow_type.  The rest of the data
7363          * is XORed together and stored in the compressed dword.
7364          */
7365         input.formatted.vlan_id = vlan_id;
7366
7367         /*
7368          * since src port and flex bytes occupy the same word XOR them together
7369          * and write the value to source port portion of compressed dword
7370          */
7371         if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7372                 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7373         else
7374                 common.port.src ^= th->dest ^ first->protocol;
7375         common.port.dst ^= th->source;
7376
7377         if (first->protocol == htons(ETH_P_IP)) {
7378                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7379                 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7380         } else {
7381                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7382                 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7383                              hdr.ipv6->saddr.s6_addr32[1] ^
7384                              hdr.ipv6->saddr.s6_addr32[2] ^
7385                              hdr.ipv6->saddr.s6_addr32[3] ^
7386                              hdr.ipv6->daddr.s6_addr32[0] ^
7387                              hdr.ipv6->daddr.s6_addr32[1] ^
7388                              hdr.ipv6->daddr.s6_addr32[2] ^
7389                              hdr.ipv6->daddr.s6_addr32[3];
7390         }
7391
7392 #ifdef CONFIG_IXGBE_VXLAN
7393         if (encap)
7394                 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
7395 #endif /* CONFIG_IXGBE_VXLAN */
7396
7397         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7398         ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7399                                               input, common, ring->queue_index);
7400 }
7401
7402 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7403                               void *accel_priv, select_queue_fallback_t fallback)
7404 {
7405         struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7406 #ifdef IXGBE_FCOE
7407         struct ixgbe_adapter *adapter;
7408         struct ixgbe_ring_feature *f;
7409         int txq;
7410 #endif
7411
7412         if (fwd_adapter)
7413                 return skb->queue_mapping + fwd_adapter->tx_base_queue;
7414
7415 #ifdef IXGBE_FCOE
7416
7417         /*
7418          * only execute the code below if protocol is FCoE
7419          * or FIP and we have FCoE enabled on the adapter
7420          */
7421         switch (vlan_get_protocol(skb)) {
7422         case htons(ETH_P_FCOE):
7423         case htons(ETH_P_FIP):
7424                 adapter = netdev_priv(dev);
7425
7426                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7427                         break;
7428         default:
7429                 return fallback(dev, skb);
7430         }
7431
7432         f = &adapter->ring_feature[RING_F_FCOE];
7433
7434         txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7435                                            smp_processor_id();
7436
7437         while (txq >= f->indices)
7438                 txq -= f->indices;
7439
7440         return txq + f->offset;
7441 #else
7442         return fallback(dev, skb);
7443 #endif
7444 }
7445
7446 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7447                           struct ixgbe_adapter *adapter,
7448                           struct ixgbe_ring *tx_ring)
7449 {
7450         struct ixgbe_tx_buffer *first;
7451         int tso;
7452         u32 tx_flags = 0;
7453         unsigned short f;
7454         u16 count = TXD_USE_COUNT(skb_headlen(skb));
7455         __be16 protocol = skb->protocol;
7456         u8 hdr_len = 0;
7457
7458         /*
7459          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7460          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7461          *       + 2 desc gap to keep tail from touching head,
7462          *       + 1 desc for context descriptor,
7463          * otherwise try next time
7464          */
7465         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7466                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7467
7468         if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7469                 tx_ring->tx_stats.tx_busy++;
7470                 return NETDEV_TX_BUSY;
7471         }
7472
7473         /* record the location of the first descriptor for this packet */
7474         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7475         first->skb = skb;
7476         first->bytecount = skb->len;
7477         first->gso_segs = 1;
7478
7479         /* if we have a HW VLAN tag being added default to the HW one */
7480         if (skb_vlan_tag_present(skb)) {
7481                 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7482                 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7483         /* else if it is a SW VLAN check the next protocol and store the tag */
7484         } else if (protocol == htons(ETH_P_8021Q)) {
7485                 struct vlan_hdr *vhdr, _vhdr;
7486                 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7487                 if (!vhdr)
7488                         goto out_drop;
7489
7490                 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7491                                   IXGBE_TX_FLAGS_VLAN_SHIFT;
7492                 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7493         }
7494         protocol = vlan_get_protocol(skb);
7495
7496         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7497             adapter->ptp_clock &&
7498             !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7499                                    &adapter->state)) {
7500                 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7501                 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7502
7503                 /* schedule check for Tx timestamp */
7504                 adapter->ptp_tx_skb = skb_get(skb);
7505                 adapter->ptp_tx_start = jiffies;
7506                 schedule_work(&adapter->ptp_tx_work);
7507         }
7508
7509         skb_tx_timestamp(skb);
7510
7511 #ifdef CONFIG_PCI_IOV
7512         /*
7513          * Use the l2switch_enable flag - would be false if the DMA
7514          * Tx switch had been disabled.
7515          */
7516         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7517                 tx_flags |= IXGBE_TX_FLAGS_CC;
7518
7519 #endif
7520         /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7521         if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7522             ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7523              (skb->priority != TC_PRIO_CONTROL))) {
7524                 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7525                 tx_flags |= (skb->priority & 0x7) <<
7526                                         IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7527                 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7528                         struct vlan_ethhdr *vhdr;
7529
7530                         if (skb_cow_head(skb, 0))
7531                                 goto out_drop;
7532                         vhdr = (struct vlan_ethhdr *)skb->data;
7533                         vhdr->h_vlan_TCI = htons(tx_flags >>
7534                                                  IXGBE_TX_FLAGS_VLAN_SHIFT);
7535                 } else {
7536                         tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7537                 }
7538         }
7539
7540         /* record initial flags and protocol */
7541         first->tx_flags = tx_flags;
7542         first->protocol = protocol;
7543
7544 #ifdef IXGBE_FCOE
7545         /* setup tx offload for FCoE */
7546         if ((protocol == htons(ETH_P_FCOE)) &&
7547             (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7548                 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7549                 if (tso < 0)
7550                         goto out_drop;
7551
7552                 goto xmit_fcoe;
7553         }
7554
7555 #endif /* IXGBE_FCOE */
7556         tso = ixgbe_tso(tx_ring, first, &hdr_len);
7557         if (tso < 0)
7558                 goto out_drop;
7559         else if (!tso)
7560                 ixgbe_tx_csum(tx_ring, first);
7561
7562         /* add the ATR filter if ATR is on */
7563         if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7564                 ixgbe_atr(tx_ring, first);
7565
7566 #ifdef IXGBE_FCOE
7567 xmit_fcoe:
7568 #endif /* IXGBE_FCOE */
7569         ixgbe_tx_map(tx_ring, first, hdr_len);
7570
7571         return NETDEV_TX_OK;
7572
7573 out_drop:
7574         dev_kfree_skb_any(first->skb);
7575         first->skb = NULL;
7576
7577         return NETDEV_TX_OK;
7578 }
7579
7580 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7581                                       struct net_device *netdev,
7582                                       struct ixgbe_ring *ring)
7583 {
7584         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7585         struct ixgbe_ring *tx_ring;
7586
7587         /*
7588          * The minimum packet size for olinfo paylen is 17 so pad the skb
7589          * in order to meet this minimum size requirement.
7590          */
7591         if (skb_put_padto(skb, 17))
7592                 return NETDEV_TX_OK;
7593
7594         tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7595
7596         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7597 }
7598
7599 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7600                                     struct net_device *netdev)
7601 {
7602         return __ixgbe_xmit_frame(skb, netdev, NULL);
7603 }
7604
7605 /**
7606  * ixgbe_set_mac - Change the Ethernet Address of the NIC
7607  * @netdev: network interface device structure
7608  * @p: pointer to an address structure
7609  *
7610  * Returns 0 on success, negative on failure
7611  **/
7612 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7613 {
7614         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7615         struct ixgbe_hw *hw = &adapter->hw;
7616         struct sockaddr *addr = p;
7617         int ret;
7618
7619         if (!is_valid_ether_addr(addr->sa_data))
7620                 return -EADDRNOTAVAIL;
7621
7622         ixgbe_del_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7623         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7624         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7625
7626         ret = ixgbe_add_mac_filter(adapter, hw->mac.addr, VMDQ_P(0));
7627         return ret > 0 ? 0 : ret;
7628 }
7629
7630 static int
7631 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7632 {
7633         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7634         struct ixgbe_hw *hw = &adapter->hw;
7635         u16 value;
7636         int rc;
7637
7638         if (prtad != hw->phy.mdio.prtad)
7639                 return -EINVAL;
7640         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7641         if (!rc)
7642                 rc = value;
7643         return rc;
7644 }
7645
7646 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7647                             u16 addr, u16 value)
7648 {
7649         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7650         struct ixgbe_hw *hw = &adapter->hw;
7651
7652         if (prtad != hw->phy.mdio.prtad)
7653                 return -EINVAL;
7654         return hw->phy.ops.write_reg(hw, addr, devad, value);
7655 }
7656
7657 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7658 {
7659         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7660
7661         switch (cmd) {
7662         case SIOCSHWTSTAMP:
7663                 return ixgbe_ptp_set_ts_config(adapter, req);
7664         case SIOCGHWTSTAMP:
7665                 return ixgbe_ptp_get_ts_config(adapter, req);
7666         default:
7667                 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7668         }
7669 }
7670
7671 /**
7672  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7673  * netdev->dev_addrs
7674  * @netdev: network interface device structure
7675  *
7676  * Returns non-zero on failure
7677  **/
7678 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7679 {
7680         int err = 0;
7681         struct ixgbe_adapter *adapter = netdev_priv(dev);
7682         struct ixgbe_hw *hw = &adapter->hw;
7683
7684         if (is_valid_ether_addr(hw->mac.san_addr)) {
7685                 rtnl_lock();
7686                 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7687                 rtnl_unlock();
7688
7689                 /* update SAN MAC vmdq pool selection */
7690                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7691         }
7692         return err;
7693 }
7694
7695 /**
7696  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7697  * netdev->dev_addrs
7698  * @netdev: network interface device structure
7699  *
7700  * Returns non-zero on failure
7701  **/
7702 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7703 {
7704         int err = 0;
7705         struct ixgbe_adapter *adapter = netdev_priv(dev);
7706         struct ixgbe_mac_info *mac = &adapter->hw.mac;
7707
7708         if (is_valid_ether_addr(mac->san_addr)) {
7709                 rtnl_lock();
7710                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
7711                 rtnl_unlock();
7712         }
7713         return err;
7714 }
7715
7716 #ifdef CONFIG_NET_POLL_CONTROLLER
7717 /*
7718  * Polling 'interrupt' - used by things like netconsole to send skbs
7719  * without having to re-enable interrupts. It's not called while
7720  * the interrupt routine is executing.
7721  */
7722 static void ixgbe_netpoll(struct net_device *netdev)
7723 {
7724         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7725         int i;
7726
7727         /* if interface is down do nothing */
7728         if (test_bit(__IXGBE_DOWN, &adapter->state))
7729                 return;
7730
7731         /* loop through and schedule all active queues */
7732         for (i = 0; i < adapter->num_q_vectors; i++)
7733                 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
7734 }
7735
7736 #endif
7737 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
7738                                                    struct rtnl_link_stats64 *stats)
7739 {
7740         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7741         int i;
7742
7743         rcu_read_lock();
7744         for (i = 0; i < adapter->num_rx_queues; i++) {
7745                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
7746                 u64 bytes, packets;
7747                 unsigned int start;
7748
7749                 if (ring) {
7750                         do {
7751                                 start = u64_stats_fetch_begin_irq(&ring->syncp);
7752                                 packets = ring->stats.packets;
7753                                 bytes   = ring->stats.bytes;
7754                         } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7755                         stats->rx_packets += packets;
7756                         stats->rx_bytes   += bytes;
7757                 }
7758         }
7759
7760         for (i = 0; i < adapter->num_tx_queues; i++) {
7761                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
7762                 u64 bytes, packets;
7763                 unsigned int start;
7764
7765                 if (ring) {
7766                         do {
7767                                 start = u64_stats_fetch_begin_irq(&ring->syncp);
7768                                 packets = ring->stats.packets;
7769                                 bytes   = ring->stats.bytes;
7770                         } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
7771                         stats->tx_packets += packets;
7772                         stats->tx_bytes   += bytes;
7773                 }
7774         }
7775         rcu_read_unlock();
7776         /* following stats updated by ixgbe_watchdog_task() */
7777         stats->multicast        = netdev->stats.multicast;
7778         stats->rx_errors        = netdev->stats.rx_errors;
7779         stats->rx_length_errors = netdev->stats.rx_length_errors;
7780         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
7781         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
7782         return stats;
7783 }
7784
7785 #ifdef CONFIG_IXGBE_DCB
7786 /**
7787  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
7788  * @adapter: pointer to ixgbe_adapter
7789  * @tc: number of traffic classes currently enabled
7790  *
7791  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
7792  * 802.1Q priority maps to a packet buffer that exists.
7793  */
7794 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
7795 {
7796         struct ixgbe_hw *hw = &adapter->hw;
7797         u32 reg, rsave;
7798         int i;
7799
7800         /* 82598 have a static priority to TC mapping that can not
7801          * be changed so no validation is needed.
7802          */
7803         if (hw->mac.type == ixgbe_mac_82598EB)
7804                 return;
7805
7806         reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
7807         rsave = reg;
7808
7809         for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
7810                 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
7811
7812                 /* If up2tc is out of bounds default to zero */
7813                 if (up2tc > tc)
7814                         reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
7815         }
7816
7817         if (reg != rsave)
7818                 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
7819
7820         return;
7821 }
7822
7823 /**
7824  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
7825  * @adapter: Pointer to adapter struct
7826  *
7827  * Populate the netdev user priority to tc map
7828  */
7829 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
7830 {
7831         struct net_device *dev = adapter->netdev;
7832         struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
7833         struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
7834         u8 prio;
7835
7836         for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
7837                 u8 tc = 0;
7838
7839                 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
7840                         tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
7841                 else if (ets)
7842                         tc = ets->prio_tc[prio];
7843
7844                 netdev_set_prio_tc_map(dev, prio, tc);
7845         }
7846 }
7847
7848 #endif /* CONFIG_IXGBE_DCB */
7849 /**
7850  * ixgbe_setup_tc - configure net_device for multiple traffic classes
7851  *
7852  * @netdev: net device to configure
7853  * @tc: number of traffic classes to enable
7854  */
7855 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
7856 {
7857         struct ixgbe_adapter *adapter = netdev_priv(dev);
7858         struct ixgbe_hw *hw = &adapter->hw;
7859         bool pools;
7860
7861         /* Hardware supports up to 8 traffic classes */
7862         if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
7863                 return -EINVAL;
7864
7865         if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
7866                 return -EINVAL;
7867
7868         pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
7869         if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
7870                 return -EBUSY;
7871
7872         /* Hardware has to reinitialize queues and interrupts to
7873          * match packet buffer alignment. Unfortunately, the
7874          * hardware is not flexible enough to do this dynamically.
7875          */
7876         if (netif_running(dev))
7877                 ixgbe_close(dev);
7878         ixgbe_clear_interrupt_scheme(adapter);
7879
7880 #ifdef CONFIG_IXGBE_DCB
7881         if (tc) {
7882                 netdev_set_num_tc(dev, tc);
7883                 ixgbe_set_prio_tc_map(adapter);
7884
7885                 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
7886
7887                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
7888                         adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
7889                         adapter->hw.fc.requested_mode = ixgbe_fc_none;
7890                 }
7891         } else {
7892                 netdev_reset_tc(dev);
7893
7894                 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
7895                         adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
7896
7897                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
7898
7899                 adapter->temp_dcb_cfg.pfc_mode_enable = false;
7900                 adapter->dcb_cfg.pfc_mode_enable = false;
7901         }
7902
7903         ixgbe_validate_rtr(adapter, tc);
7904
7905 #endif /* CONFIG_IXGBE_DCB */
7906         ixgbe_init_interrupt_scheme(adapter);
7907
7908         if (netif_running(dev))
7909                 return ixgbe_open(dev);
7910
7911         return 0;
7912 }
7913
7914 #ifdef CONFIG_PCI_IOV
7915 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
7916 {
7917         struct net_device *netdev = adapter->netdev;
7918
7919         rtnl_lock();
7920         ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
7921         rtnl_unlock();
7922 }
7923
7924 #endif
7925 void ixgbe_do_reset(struct net_device *netdev)
7926 {
7927         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7928
7929         if (netif_running(netdev))
7930                 ixgbe_reinit_locked(adapter);
7931         else
7932                 ixgbe_reset(adapter);
7933 }
7934
7935 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
7936                                             netdev_features_t features)
7937 {
7938         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7939
7940         /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
7941         if (!(features & NETIF_F_RXCSUM))
7942                 features &= ~NETIF_F_LRO;
7943
7944         /* Turn off LRO if not RSC capable */
7945         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
7946                 features &= ~NETIF_F_LRO;
7947
7948         return features;
7949 }
7950
7951 static int ixgbe_set_features(struct net_device *netdev,
7952                               netdev_features_t features)
7953 {
7954         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7955         netdev_features_t changed = netdev->features ^ features;
7956         bool need_reset = false;
7957
7958         /* Make sure RSC matches LRO, reset if change */
7959         if (!(features & NETIF_F_LRO)) {
7960                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
7961                         need_reset = true;
7962                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
7963         } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
7964                    !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
7965                 if (adapter->rx_itr_setting == 1 ||
7966                     adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
7967                         adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
7968                         need_reset = true;
7969                 } else if ((changed ^ features) & NETIF_F_LRO) {
7970                         e_info(probe, "rx-usecs set too low, "
7971                                "disabling RSC\n");
7972                 }
7973         }
7974
7975         /*
7976          * Check if Flow Director n-tuple support was enabled or disabled.  If
7977          * the state changed, we need to reset.
7978          */
7979         switch (features & NETIF_F_NTUPLE) {
7980         case NETIF_F_NTUPLE:
7981                 /* turn off ATR, enable perfect filters and reset */
7982                 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
7983                         need_reset = true;
7984
7985                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
7986                 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7987                 break;
7988         default:
7989                 /* turn off perfect filters, enable ATR and reset */
7990                 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
7991                         need_reset = true;
7992
7993                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
7994
7995                 /* We cannot enable ATR if SR-IOV is enabled */
7996                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7997                         break;
7998
7999                 /* We cannot enable ATR if we have 2 or more traffic classes */
8000                 if (netdev_get_num_tc(netdev) > 1)
8001                         break;
8002
8003                 /* We cannot enable ATR if RSS is disabled */
8004                 if (adapter->ring_feature[RING_F_RSS].limit <= 1)
8005                         break;
8006
8007                 /* A sample rate of 0 indicates ATR disabled */
8008                 if (!adapter->atr_sample_rate)
8009                         break;
8010
8011                 adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
8012                 break;
8013         }
8014
8015         if (features & NETIF_F_HW_VLAN_CTAG_RX)
8016                 ixgbe_vlan_strip_enable(adapter);
8017         else
8018                 ixgbe_vlan_strip_disable(adapter);
8019
8020         if (changed & NETIF_F_RXALL)
8021                 need_reset = true;
8022
8023         netdev->features = features;
8024
8025 #ifdef CONFIG_IXGBE_VXLAN
8026         if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
8027                 if (features & NETIF_F_RXCSUM)
8028                         adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8029                 else
8030                         ixgbe_clear_vxlan_port(adapter);
8031         }
8032 #endif /* CONFIG_IXGBE_VXLAN */
8033
8034         if (need_reset)
8035                 ixgbe_do_reset(netdev);
8036
8037         return 0;
8038 }
8039
8040 #ifdef CONFIG_IXGBE_VXLAN
8041 /**
8042  * ixgbe_add_vxlan_port - Get notifications about VXLAN ports that come up
8043  * @dev: The port's netdev
8044  * @sa_family: Socket Family that VXLAN is notifiying us about
8045  * @port: New UDP port number that VXLAN started listening to
8046  **/
8047 static void ixgbe_add_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8048                                  __be16 port)
8049 {
8050         struct ixgbe_adapter *adapter = netdev_priv(dev);
8051         struct ixgbe_hw *hw = &adapter->hw;
8052         u16 new_port = ntohs(port);
8053
8054         if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8055                 return;
8056
8057         if (sa_family == AF_INET6)
8058                 return;
8059
8060         if (adapter->vxlan_port == new_port)
8061                 return;
8062
8063         if (adapter->vxlan_port) {
8064                 netdev_info(dev,
8065                             "Hit Max num of VXLAN ports, not adding port %d\n",
8066                             new_port);
8067                 return;
8068         }
8069
8070         adapter->vxlan_port = new_port;
8071         IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, new_port);
8072 }
8073
8074 /**
8075  * ixgbe_del_vxlan_port - Get notifications about VXLAN ports that go away
8076  * @dev: The port's netdev
8077  * @sa_family: Socket Family that VXLAN is notifying us about
8078  * @port: UDP port number that VXLAN stopped listening to
8079  **/
8080 static void ixgbe_del_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8081                                  __be16 port)
8082 {
8083         struct ixgbe_adapter *adapter = netdev_priv(dev);
8084         u16 new_port = ntohs(port);
8085
8086         if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8087                 return;
8088
8089         if (sa_family == AF_INET6)
8090                 return;
8091
8092         if (adapter->vxlan_port != new_port) {
8093                 netdev_info(dev, "Port %d was not found, not deleting\n",
8094                             new_port);
8095                 return;
8096         }
8097
8098         ixgbe_clear_vxlan_port(adapter);
8099         adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8100 }
8101 #endif /* CONFIG_IXGBE_VXLAN */
8102
8103 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
8104                              struct net_device *dev,
8105                              const unsigned char *addr, u16 vid,
8106                              u16 flags)
8107 {
8108         /* guarantee we can provide a unique filter for the unicast address */
8109         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
8110                 if (IXGBE_MAX_PF_MACVLANS <= netdev_uc_count(dev))
8111                         return -ENOMEM;
8112         }
8113
8114         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
8115 }
8116
8117 /**
8118  * ixgbe_configure_bridge_mode - set various bridge modes
8119  * @adapter - the private structure
8120  * @mode - requested bridge mode
8121  *
8122  * Configure some settings require for various bridge modes.
8123  **/
8124 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
8125                                        __u16 mode)
8126 {
8127         struct ixgbe_hw *hw = &adapter->hw;
8128         unsigned int p, num_pools;
8129         u32 vmdctl;
8130
8131         switch (mode) {
8132         case BRIDGE_MODE_VEPA:
8133                 /* disable Tx loopback, rely on switch hairpin mode */
8134                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
8135
8136                 /* must enable Rx switching replication to allow multicast
8137                  * packet reception on all VFs, and to enable source address
8138                  * pruning.
8139                  */
8140                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8141                 vmdctl |= IXGBE_VT_CTL_REPLEN;
8142                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8143
8144                 /* enable Rx source address pruning. Note, this requires
8145                  * replication to be enabled or else it does nothing.
8146                  */
8147                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8148                 for (p = 0; p < num_pools; p++) {
8149                         if (hw->mac.ops.set_source_address_pruning)
8150                                 hw->mac.ops.set_source_address_pruning(hw,
8151                                                                        true,
8152                                                                        p);
8153                 }
8154                 break;
8155         case BRIDGE_MODE_VEB:
8156                 /* enable Tx loopback for internal VF/PF communication */
8157                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
8158                                 IXGBE_PFDTXGSWC_VT_LBEN);
8159
8160                 /* disable Rx switching replication unless we have SR-IOV
8161                  * virtual functions
8162                  */
8163                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8164                 if (!adapter->num_vfs)
8165                         vmdctl &= ~IXGBE_VT_CTL_REPLEN;
8166                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8167
8168                 /* disable Rx source address pruning, since we don't expect to
8169                  * be receiving external loopback of our transmitted frames.
8170                  */
8171                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8172                 for (p = 0; p < num_pools; p++) {
8173                         if (hw->mac.ops.set_source_address_pruning)
8174                                 hw->mac.ops.set_source_address_pruning(hw,
8175                                                                        false,
8176                                                                        p);
8177                 }
8178                 break;
8179         default:
8180                 return -EINVAL;
8181         }
8182
8183         adapter->bridge_mode = mode;
8184
8185         e_info(drv, "enabling bridge mode: %s\n",
8186                mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
8187
8188         return 0;
8189 }
8190
8191 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
8192                                     struct nlmsghdr *nlh, u16 flags)
8193 {
8194         struct ixgbe_adapter *adapter = netdev_priv(dev);
8195         struct nlattr *attr, *br_spec;
8196         int rem;
8197
8198         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8199                 return -EOPNOTSUPP;
8200
8201         br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
8202         if (!br_spec)
8203                 return -EINVAL;
8204
8205         nla_for_each_nested(attr, br_spec, rem) {
8206                 int status;
8207                 __u16 mode;
8208
8209                 if (nla_type(attr) != IFLA_BRIDGE_MODE)
8210                         continue;
8211
8212                 if (nla_len(attr) < sizeof(mode))
8213                         return -EINVAL;
8214
8215                 mode = nla_get_u16(attr);
8216                 status = ixgbe_configure_bridge_mode(adapter, mode);
8217                 if (status)
8218                         return status;
8219
8220                 break;
8221         }
8222
8223         return 0;
8224 }
8225
8226 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
8227                                     struct net_device *dev,
8228                                     u32 filter_mask, int nlflags)
8229 {
8230         struct ixgbe_adapter *adapter = netdev_priv(dev);
8231
8232         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8233                 return 0;
8234
8235         return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
8236                                        adapter->bridge_mode, 0, 0, nlflags,
8237                                        filter_mask, NULL);
8238 }
8239
8240 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
8241 {
8242         struct ixgbe_fwd_adapter *fwd_adapter = NULL;
8243         struct ixgbe_adapter *adapter = netdev_priv(pdev);
8244         int used_pools = adapter->num_vfs + adapter->num_rx_pools;
8245         unsigned int limit;
8246         int pool, err;
8247
8248         /* Hardware has a limited number of available pools. Each VF, and the
8249          * PF require a pool. Check to ensure we don't attempt to use more
8250          * then the available number of pools.
8251          */
8252         if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
8253                 return ERR_PTR(-EINVAL);
8254
8255 #ifdef CONFIG_RPS
8256         if (vdev->num_rx_queues != vdev->num_tx_queues) {
8257                 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
8258                             vdev->name);
8259                 return ERR_PTR(-EINVAL);
8260         }
8261 #endif
8262         /* Check for hardware restriction on number of rx/tx queues */
8263         if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
8264             vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
8265                 netdev_info(pdev,
8266                             "%s: Supports RX/TX Queue counts 1,2, and 4\n",
8267                             pdev->name);
8268                 return ERR_PTR(-EINVAL);
8269         }
8270
8271         if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8272               adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
8273             (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
8274                 return ERR_PTR(-EBUSY);
8275
8276         fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL);
8277         if (!fwd_adapter)
8278                 return ERR_PTR(-ENOMEM);
8279
8280         pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
8281         adapter->num_rx_pools++;
8282         set_bit(pool, &adapter->fwd_bitmask);
8283         limit = find_last_bit(&adapter->fwd_bitmask, 32);
8284
8285         /* Enable VMDq flag so device will be set in VM mode */
8286         adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
8287         adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8288         adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
8289
8290         /* Force reinit of ring allocation with VMDQ enabled */
8291         err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8292         if (err)
8293                 goto fwd_add_err;
8294         fwd_adapter->pool = pool;
8295         fwd_adapter->real_adapter = adapter;
8296         err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
8297         if (err)
8298                 goto fwd_add_err;
8299         netif_tx_start_all_queues(vdev);
8300         return fwd_adapter;
8301 fwd_add_err:
8302         /* unwind counter and free adapter struct */
8303         netdev_info(pdev,
8304                     "%s: dfwd hardware acceleration failed\n", vdev->name);
8305         clear_bit(pool, &adapter->fwd_bitmask);
8306         adapter->num_rx_pools--;
8307         kfree(fwd_adapter);
8308         return ERR_PTR(err);
8309 }
8310
8311 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
8312 {
8313         struct ixgbe_fwd_adapter *fwd_adapter = priv;
8314         struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
8315         unsigned int limit;
8316
8317         clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
8318         adapter->num_rx_pools--;
8319
8320         limit = find_last_bit(&adapter->fwd_bitmask, 32);
8321         adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8322         ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
8323         ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8324         netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
8325                    fwd_adapter->pool, adapter->num_rx_pools,
8326                    fwd_adapter->rx_base_queue,
8327                    fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
8328                    adapter->fwd_bitmask);
8329         kfree(fwd_adapter);
8330 }
8331
8332 #define IXGBE_MAX_TUNNEL_HDR_LEN 80
8333 static netdev_features_t
8334 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
8335                      netdev_features_t features)
8336 {
8337         if (!skb->encapsulation)
8338                 return features;
8339
8340         if (unlikely(skb_inner_mac_header(skb) - skb_transport_header(skb) >
8341                      IXGBE_MAX_TUNNEL_HDR_LEN))
8342                 return features & ~NETIF_F_ALL_CSUM;
8343
8344         return features;
8345 }
8346
8347 static const struct net_device_ops ixgbe_netdev_ops = {
8348         .ndo_open               = ixgbe_open,
8349         .ndo_stop               = ixgbe_close,
8350         .ndo_start_xmit         = ixgbe_xmit_frame,
8351         .ndo_select_queue       = ixgbe_select_queue,
8352         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
8353         .ndo_validate_addr      = eth_validate_addr,
8354         .ndo_set_mac_address    = ixgbe_set_mac,
8355         .ndo_change_mtu         = ixgbe_change_mtu,
8356         .ndo_tx_timeout         = ixgbe_tx_timeout,
8357         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
8358         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
8359         .ndo_do_ioctl           = ixgbe_ioctl,
8360         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
8361         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
8362         .ndo_set_vf_rate        = ixgbe_ndo_set_vf_bw,
8363         .ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
8364         .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
8365         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
8366         .ndo_get_stats64        = ixgbe_get_stats64,
8367 #ifdef CONFIG_IXGBE_DCB
8368         .ndo_setup_tc           = ixgbe_setup_tc,
8369 #endif
8370 #ifdef CONFIG_NET_POLL_CONTROLLER
8371         .ndo_poll_controller    = ixgbe_netpoll,
8372 #endif
8373 #ifdef CONFIG_NET_RX_BUSY_POLL
8374         .ndo_busy_poll          = ixgbe_low_latency_recv,
8375 #endif
8376 #ifdef IXGBE_FCOE
8377         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
8378         .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
8379         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
8380         .ndo_fcoe_enable = ixgbe_fcoe_enable,
8381         .ndo_fcoe_disable = ixgbe_fcoe_disable,
8382         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
8383         .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
8384 #endif /* IXGBE_FCOE */
8385         .ndo_set_features = ixgbe_set_features,
8386         .ndo_fix_features = ixgbe_fix_features,
8387         .ndo_fdb_add            = ixgbe_ndo_fdb_add,
8388         .ndo_bridge_setlink     = ixgbe_ndo_bridge_setlink,
8389         .ndo_bridge_getlink     = ixgbe_ndo_bridge_getlink,
8390         .ndo_dfwd_add_station   = ixgbe_fwd_add,
8391         .ndo_dfwd_del_station   = ixgbe_fwd_del,
8392 #ifdef CONFIG_IXGBE_VXLAN
8393         .ndo_add_vxlan_port     = ixgbe_add_vxlan_port,
8394         .ndo_del_vxlan_port     = ixgbe_del_vxlan_port,
8395 #endif /* CONFIG_IXGBE_VXLAN */
8396         .ndo_features_check     = ixgbe_features_check,
8397 };
8398
8399 /**
8400  * ixgbe_enumerate_functions - Get the number of ports this device has
8401  * @adapter: adapter structure
8402  *
8403  * This function enumerates the phsyical functions co-located on a single slot,
8404  * in order to determine how many ports a device has. This is most useful in
8405  * determining the required GT/s of PCIe bandwidth necessary for optimal
8406  * performance.
8407  **/
8408 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
8409 {
8410         struct pci_dev *entry, *pdev = adapter->pdev;
8411         int physfns = 0;
8412
8413         /* Some cards can not use the generic count PCIe functions method,
8414          * because they are behind a parent switch, so we hardcode these with
8415          * the correct number of functions.
8416          */
8417         if (ixgbe_pcie_from_parent(&adapter->hw))
8418                 physfns = 4;
8419
8420         list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
8421                 /* don't count virtual functions */
8422                 if (entry->is_virtfn)
8423                         continue;
8424
8425                 /* When the devices on the bus don't all match our device ID,
8426                  * we can't reliably determine the correct number of
8427                  * functions. This can occur if a function has been direct
8428                  * attached to a virtual machine using VT-d, for example. In
8429                  * this case, simply return -1 to indicate this.
8430                  */
8431                 if ((entry->vendor != pdev->vendor) ||
8432                     (entry->device != pdev->device))
8433                         return -1;
8434
8435                 physfns++;
8436         }
8437
8438         return physfns;
8439 }
8440
8441 /**
8442  * ixgbe_wol_supported - Check whether device supports WoL
8443  * @hw: hw specific details
8444  * @device_id: the device ID
8445  * @subdev_id: the subsystem device ID
8446  *
8447  * This function is used by probe and ethtool to determine
8448  * which devices have WoL support
8449  *
8450  **/
8451 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
8452                         u16 subdevice_id)
8453 {
8454         struct ixgbe_hw *hw = &adapter->hw;
8455         u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
8456         int is_wol_supported = 0;
8457
8458         switch (device_id) {
8459         case IXGBE_DEV_ID_82599_SFP:
8460                 /* Only these subdevices could supports WOL */
8461                 switch (subdevice_id) {
8462                 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
8463                 case IXGBE_SUBDEV_ID_82599_560FLR:
8464                         /* only support first port */
8465                         if (hw->bus.func != 0)
8466                                 break;
8467                 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
8468                 case IXGBE_SUBDEV_ID_82599_SFP:
8469                 case IXGBE_SUBDEV_ID_82599_RNDC:
8470                 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
8471                 case IXGBE_SUBDEV_ID_82599_LOM_SFP:
8472                         is_wol_supported = 1;
8473                         break;
8474                 }
8475                 break;
8476         case IXGBE_DEV_ID_82599EN_SFP:
8477                 /* Only this subdevice supports WOL */
8478                 switch (subdevice_id) {
8479                 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
8480                         is_wol_supported = 1;
8481                         break;
8482                 }
8483                 break;
8484         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
8485                 /* All except this subdevice support WOL */
8486                 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
8487                         is_wol_supported = 1;
8488                 break;
8489         case IXGBE_DEV_ID_82599_KX4:
8490                 is_wol_supported = 1;
8491                 break;
8492         case IXGBE_DEV_ID_X540T:
8493         case IXGBE_DEV_ID_X540T1:
8494         case IXGBE_DEV_ID_X550T:
8495         case IXGBE_DEV_ID_X550EM_X_KX4:
8496         case IXGBE_DEV_ID_X550EM_X_KR:
8497         case IXGBE_DEV_ID_X550EM_X_10G_T:
8498                 /* check eeprom to see if enabled wol */
8499                 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
8500                     ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
8501                      (hw->bus.func == 0))) {
8502                         is_wol_supported = 1;
8503                 }
8504                 break;
8505         }
8506
8507         return is_wol_supported;
8508 }
8509
8510 /**
8511  * ixgbe_get_platform_mac_addr - Look up MAC address in Open Firmware / IDPROM
8512  * @adapter: Pointer to adapter struct
8513  */
8514 static void ixgbe_get_platform_mac_addr(struct ixgbe_adapter *adapter)
8515 {
8516 #ifdef CONFIG_OF
8517         struct device_node *dp = pci_device_to_OF_node(adapter->pdev);
8518         struct ixgbe_hw *hw = &adapter->hw;
8519         const unsigned char *addr;
8520
8521         addr = of_get_mac_address(dp);
8522         if (addr) {
8523                 ether_addr_copy(hw->mac.perm_addr, addr);
8524                 return;
8525         }
8526 #endif /* CONFIG_OF */
8527
8528 #ifdef CONFIG_SPARC
8529         ether_addr_copy(hw->mac.perm_addr, idprom->id_ethaddr);
8530 #endif /* CONFIG_SPARC */
8531 }
8532
8533 /**
8534  * ixgbe_probe - Device Initialization Routine
8535  * @pdev: PCI device information struct
8536  * @ent: entry in ixgbe_pci_tbl
8537  *
8538  * Returns 0 on success, negative on failure
8539  *
8540  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
8541  * The OS initialization, configuring of the adapter private structure,
8542  * and a hardware reset occur.
8543  **/
8544 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
8545 {
8546         struct net_device *netdev;
8547         struct ixgbe_adapter *adapter = NULL;
8548         struct ixgbe_hw *hw;
8549         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
8550         int i, err, pci_using_dac, expected_gts;
8551         unsigned int indices = MAX_TX_QUEUES;
8552         u8 part_str[IXGBE_PBANUM_LENGTH];
8553         bool disable_dev = false;
8554 #ifdef IXGBE_FCOE
8555         u16 device_caps;
8556 #endif
8557         u32 eec;
8558
8559         /* Catch broken hardware that put the wrong VF device ID in
8560          * the PCIe SR-IOV capability.
8561          */
8562         if (pdev->is_virtfn) {
8563                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
8564                      pci_name(pdev), pdev->vendor, pdev->device);
8565                 return -EINVAL;
8566         }
8567
8568         err = pci_enable_device_mem(pdev);
8569         if (err)
8570                 return err;
8571
8572         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
8573                 pci_using_dac = 1;
8574         } else {
8575                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
8576                 if (err) {
8577                         dev_err(&pdev->dev,
8578                                 "No usable DMA configuration, aborting\n");
8579                         goto err_dma;
8580                 }
8581                 pci_using_dac = 0;
8582         }
8583
8584         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
8585                                            IORESOURCE_MEM), ixgbe_driver_name);
8586         if (err) {
8587                 dev_err(&pdev->dev,
8588                         "pci_request_selected_regions failed 0x%x\n", err);
8589                 goto err_pci_reg;
8590         }
8591
8592         pci_enable_pcie_error_reporting(pdev);
8593
8594         pci_set_master(pdev);
8595         pci_save_state(pdev);
8596
8597         if (ii->mac == ixgbe_mac_82598EB) {
8598 #ifdef CONFIG_IXGBE_DCB
8599                 /* 8 TC w/ 4 queues per TC */
8600                 indices = 4 * MAX_TRAFFIC_CLASS;
8601 #else
8602                 indices = IXGBE_MAX_RSS_INDICES;
8603 #endif
8604         }
8605
8606         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
8607         if (!netdev) {
8608                 err = -ENOMEM;
8609                 goto err_alloc_etherdev;
8610         }
8611
8612         SET_NETDEV_DEV(netdev, &pdev->dev);
8613
8614         adapter = netdev_priv(netdev);
8615
8616         adapter->netdev = netdev;
8617         adapter->pdev = pdev;
8618         hw = &adapter->hw;
8619         hw->back = adapter;
8620         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
8621
8622         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
8623                               pci_resource_len(pdev, 0));
8624         adapter->io_addr = hw->hw_addr;
8625         if (!hw->hw_addr) {
8626                 err = -EIO;
8627                 goto err_ioremap;
8628         }
8629
8630         netdev->netdev_ops = &ixgbe_netdev_ops;
8631         ixgbe_set_ethtool_ops(netdev);
8632         netdev->watchdog_timeo = 5 * HZ;
8633         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
8634
8635         /* Setup hw api */
8636         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
8637         hw->mac.type  = ii->mac;
8638         hw->mvals     = ii->mvals;
8639
8640         /* EEPROM */
8641         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
8642         eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
8643         if (ixgbe_removed(hw->hw_addr)) {
8644                 err = -EIO;
8645                 goto err_ioremap;
8646         }
8647         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
8648         if (!(eec & (1 << 8)))
8649                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
8650
8651         /* PHY */
8652         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
8653         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
8654         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
8655         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
8656         hw->phy.mdio.mmds = 0;
8657         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
8658         hw->phy.mdio.dev = netdev;
8659         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
8660         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
8661
8662         ii->get_invariants(hw);
8663
8664         /* setup the private structure */
8665         err = ixgbe_sw_init(adapter);
8666         if (err)
8667                 goto err_sw_init;
8668
8669         /* Make it possible the adapter to be woken up via WOL */
8670         switch (adapter->hw.mac.type) {
8671         case ixgbe_mac_82599EB:
8672         case ixgbe_mac_X540:
8673         case ixgbe_mac_X550:
8674         case ixgbe_mac_X550EM_x:
8675                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
8676                 break;
8677         default:
8678                 break;
8679         }
8680
8681         /*
8682          * If there is a fan on this device and it has failed log the
8683          * failure.
8684          */
8685         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
8686                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
8687                 if (esdp & IXGBE_ESDP_SDP1)
8688                         e_crit(probe, "Fan has stopped, replace the adapter\n");
8689         }
8690
8691         if (allow_unsupported_sfp)
8692                 hw->allow_unsupported_sfp = allow_unsupported_sfp;
8693
8694         /* reset_hw fills in the perm_addr as well */
8695         hw->phy.reset_if_overtemp = true;
8696         err = hw->mac.ops.reset_hw(hw);
8697         hw->phy.reset_if_overtemp = false;
8698         if (err == IXGBE_ERR_SFP_NOT_PRESENT &&
8699             hw->mac.type == ixgbe_mac_82598EB) {
8700                 err = 0;
8701         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
8702                 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
8703                 e_dev_err("Reload the driver after installing a supported module.\n");
8704                 goto err_sw_init;
8705         } else if (err) {
8706                 e_dev_err("HW Init failed: %d\n", err);
8707                 goto err_sw_init;
8708         }
8709
8710 #ifdef CONFIG_PCI_IOV
8711         /* SR-IOV not supported on the 82598 */
8712         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8713                 goto skip_sriov;
8714         /* Mailbox */
8715         ixgbe_init_mbx_params_pf(hw);
8716         memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
8717         pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
8718         ixgbe_enable_sriov(adapter);
8719 skip_sriov:
8720
8721 #endif
8722         netdev->features = NETIF_F_SG |
8723                            NETIF_F_IP_CSUM |
8724                            NETIF_F_IPV6_CSUM |
8725                            NETIF_F_HW_VLAN_CTAG_TX |
8726                            NETIF_F_HW_VLAN_CTAG_RX |
8727                            NETIF_F_TSO |
8728                            NETIF_F_TSO6 |
8729                            NETIF_F_RXHASH |
8730                            NETIF_F_RXCSUM;
8731
8732         netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
8733
8734         switch (adapter->hw.mac.type) {
8735         case ixgbe_mac_82599EB:
8736         case ixgbe_mac_X540:
8737         case ixgbe_mac_X550:
8738         case ixgbe_mac_X550EM_x:
8739                 netdev->features |= NETIF_F_SCTP_CSUM;
8740                 netdev->hw_features |= NETIF_F_SCTP_CSUM |
8741                                        NETIF_F_NTUPLE;
8742                 break;
8743         default:
8744                 break;
8745         }
8746
8747         netdev->hw_features |= NETIF_F_RXALL;
8748         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
8749
8750         netdev->vlan_features |= NETIF_F_TSO;
8751         netdev->vlan_features |= NETIF_F_TSO6;
8752         netdev->vlan_features |= NETIF_F_IP_CSUM;
8753         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
8754         netdev->vlan_features |= NETIF_F_SG;
8755
8756         netdev->hw_enc_features |= NETIF_F_SG | NETIF_F_IP_CSUM |
8757                                    NETIF_F_IPV6_CSUM;
8758
8759         netdev->priv_flags |= IFF_UNICAST_FLT;
8760         netdev->priv_flags |= IFF_SUPP_NOFCS;
8761
8762 #ifdef CONFIG_IXGBE_VXLAN
8763         switch (adapter->hw.mac.type) {
8764         case ixgbe_mac_X550:
8765         case ixgbe_mac_X550EM_x:
8766                 netdev->hw_enc_features |= NETIF_F_RXCSUM |
8767                                            NETIF_F_IP_CSUM |
8768                                            NETIF_F_IPV6_CSUM;
8769                 break;
8770         default:
8771                 break;
8772         }
8773 #endif /* CONFIG_IXGBE_VXLAN */
8774
8775 #ifdef CONFIG_IXGBE_DCB
8776         netdev->dcbnl_ops = &dcbnl_ops;
8777 #endif
8778
8779 #ifdef IXGBE_FCOE
8780         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
8781                 unsigned int fcoe_l;
8782
8783                 if (hw->mac.ops.get_device_caps) {
8784                         hw->mac.ops.get_device_caps(hw, &device_caps);
8785                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
8786                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
8787                 }
8788
8789
8790                 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
8791                 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
8792
8793                 netdev->features |= NETIF_F_FSO |
8794                                     NETIF_F_FCOE_CRC;
8795
8796                 netdev->vlan_features |= NETIF_F_FSO |
8797                                          NETIF_F_FCOE_CRC |
8798                                          NETIF_F_FCOE_MTU;
8799         }
8800 #endif /* IXGBE_FCOE */
8801         if (pci_using_dac) {
8802                 netdev->features |= NETIF_F_HIGHDMA;
8803                 netdev->vlan_features |= NETIF_F_HIGHDMA;
8804         }
8805
8806         if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
8807                 netdev->hw_features |= NETIF_F_LRO;
8808         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8809                 netdev->features |= NETIF_F_LRO;
8810
8811         /* make sure the EEPROM is good */
8812         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
8813                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
8814                 err = -EIO;
8815                 goto err_sw_init;
8816         }
8817
8818         ixgbe_get_platform_mac_addr(adapter);
8819
8820         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
8821
8822         if (!is_valid_ether_addr(netdev->dev_addr)) {
8823                 e_dev_err("invalid MAC address\n");
8824                 err = -EIO;
8825                 goto err_sw_init;
8826         }
8827
8828         ixgbe_mac_set_default_filter(adapter, hw->mac.perm_addr);
8829
8830         setup_timer(&adapter->service_timer, &ixgbe_service_timer,
8831                     (unsigned long) adapter);
8832
8833         if (ixgbe_removed(hw->hw_addr)) {
8834                 err = -EIO;
8835                 goto err_sw_init;
8836         }
8837         INIT_WORK(&adapter->service_task, ixgbe_service_task);
8838         set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
8839         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
8840
8841         err = ixgbe_init_interrupt_scheme(adapter);
8842         if (err)
8843                 goto err_sw_init;
8844
8845         /* WOL not supported for all devices */
8846         adapter->wol = 0;
8847         hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
8848         hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
8849                                                 pdev->subsystem_device);
8850         if (hw->wol_enabled)
8851                 adapter->wol = IXGBE_WUFC_MAG;
8852
8853         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
8854
8855         /* save off EEPROM version number */
8856         hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
8857         hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
8858
8859         /* pick up the PCI bus settings for reporting later */
8860         if (ixgbe_pcie_from_parent(hw))
8861                 ixgbe_get_parent_bus_info(adapter);
8862         else
8863                  hw->mac.ops.get_bus_info(hw);
8864
8865         /* calculate the expected PCIe bandwidth required for optimal
8866          * performance. Note that some older parts will never have enough
8867          * bandwidth due to being older generation PCIe parts. We clamp these
8868          * parts to ensure no warning is displayed if it can't be fixed.
8869          */
8870         switch (hw->mac.type) {
8871         case ixgbe_mac_82598EB:
8872                 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
8873                 break;
8874         default:
8875                 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
8876                 break;
8877         }
8878
8879         /* don't check link if we failed to enumerate functions */
8880         if (expected_gts > 0)
8881                 ixgbe_check_minimum_link(adapter, expected_gts);
8882
8883         err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
8884         if (err)
8885                 strlcpy(part_str, "Unknown", sizeof(part_str));
8886         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
8887                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
8888                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
8889                            part_str);
8890         else
8891                 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
8892                            hw->mac.type, hw->phy.type, part_str);
8893
8894         e_dev_info("%pM\n", netdev->dev_addr);
8895
8896         /* reset the hardware with the new settings */
8897         err = hw->mac.ops.start_hw(hw);
8898         if (err == IXGBE_ERR_EEPROM_VERSION) {
8899                 /* We are running on a pre-production device, log a warning */
8900                 e_dev_warn("This device is a pre-production adapter/LOM. "
8901                            "Please be aware there may be issues associated "
8902                            "with your hardware.  If you are experiencing "
8903                            "problems please contact your Intel or hardware "
8904                            "representative who provided you with this "
8905                            "hardware.\n");
8906         }
8907         strcpy(netdev->name, "eth%d");
8908         err = register_netdev(netdev);
8909         if (err)
8910                 goto err_register;
8911
8912         pci_set_drvdata(pdev, adapter);
8913
8914         /* power down the optics for 82599 SFP+ fiber */
8915         if (hw->mac.ops.disable_tx_laser)
8916                 hw->mac.ops.disable_tx_laser(hw);
8917
8918         /* carrier off reporting is important to ethtool even BEFORE open */
8919         netif_carrier_off(netdev);
8920
8921 #ifdef CONFIG_IXGBE_DCA
8922         if (dca_add_requester(&pdev->dev) == 0) {
8923                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
8924                 ixgbe_setup_dca(adapter);
8925         }
8926 #endif
8927         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
8928                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
8929                 for (i = 0; i < adapter->num_vfs; i++)
8930                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
8931         }
8932
8933         /* firmware requires driver version to be 0xFFFFFFFF
8934          * since os does not support feature
8935          */
8936         if (hw->mac.ops.set_fw_drv_ver)
8937                 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
8938                                            0xFF);
8939
8940         /* add san mac addr to netdev */
8941         ixgbe_add_sanmac_netdev(netdev);
8942
8943         e_dev_info("%s\n", ixgbe_default_device_descr);
8944
8945 #ifdef CONFIG_IXGBE_HWMON
8946         if (ixgbe_sysfs_init(adapter))
8947                 e_err(probe, "failed to allocate sysfs resources\n");
8948 #endif /* CONFIG_IXGBE_HWMON */
8949
8950         ixgbe_dbg_adapter_init(adapter);
8951
8952         /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
8953         if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
8954                 hw->mac.ops.setup_link(hw,
8955                         IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
8956                         true);
8957
8958         return 0;
8959
8960 err_register:
8961         ixgbe_release_hw_control(adapter);
8962         ixgbe_clear_interrupt_scheme(adapter);
8963 err_sw_init:
8964         ixgbe_disable_sriov(adapter);
8965         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
8966         iounmap(adapter->io_addr);
8967         kfree(adapter->mac_table);
8968 err_ioremap:
8969         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
8970         free_netdev(netdev);
8971 err_alloc_etherdev:
8972         pci_release_selected_regions(pdev,
8973                                      pci_select_bars(pdev, IORESOURCE_MEM));
8974 err_pci_reg:
8975 err_dma:
8976         if (!adapter || disable_dev)
8977                 pci_disable_device(pdev);
8978         return err;
8979 }
8980
8981 /**
8982  * ixgbe_remove - Device Removal Routine
8983  * @pdev: PCI device information struct
8984  *
8985  * ixgbe_remove is called by the PCI subsystem to alert the driver
8986  * that it should release a PCI device.  The could be caused by a
8987  * Hot-Plug event, or because the driver is going to be removed from
8988  * memory.
8989  **/
8990 static void ixgbe_remove(struct pci_dev *pdev)
8991 {
8992         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
8993         struct net_device *netdev;
8994         bool disable_dev;
8995
8996         /* if !adapter then we already cleaned up in probe */
8997         if (!adapter)
8998                 return;
8999
9000         netdev  = adapter->netdev;
9001         ixgbe_dbg_adapter_exit(adapter);
9002
9003         set_bit(__IXGBE_REMOVING, &adapter->state);
9004         cancel_work_sync(&adapter->service_task);
9005
9006
9007 #ifdef CONFIG_IXGBE_DCA
9008         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
9009                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
9010                 dca_remove_requester(&pdev->dev);
9011                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
9012         }
9013
9014 #endif
9015 #ifdef CONFIG_IXGBE_HWMON
9016         ixgbe_sysfs_exit(adapter);
9017 #endif /* CONFIG_IXGBE_HWMON */
9018
9019         /* remove the added san mac */
9020         ixgbe_del_sanmac_netdev(netdev);
9021
9022         if (netdev->reg_state == NETREG_REGISTERED)
9023                 unregister_netdev(netdev);
9024
9025 #ifdef CONFIG_PCI_IOV
9026         ixgbe_disable_sriov(adapter);
9027 #endif
9028         ixgbe_clear_interrupt_scheme(adapter);
9029
9030         ixgbe_release_hw_control(adapter);
9031
9032 #ifdef CONFIG_DCB
9033         kfree(adapter->ixgbe_ieee_pfc);
9034         kfree(adapter->ixgbe_ieee_ets);
9035
9036 #endif
9037         iounmap(adapter->io_addr);
9038         pci_release_selected_regions(pdev, pci_select_bars(pdev,
9039                                      IORESOURCE_MEM));
9040
9041         e_dev_info("complete\n");
9042
9043         kfree(adapter->mac_table);
9044         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9045         free_netdev(netdev);
9046
9047         pci_disable_pcie_error_reporting(pdev);
9048
9049         if (disable_dev)
9050                 pci_disable_device(pdev);
9051 }
9052
9053 /**
9054  * ixgbe_io_error_detected - called when PCI error is detected
9055  * @pdev: Pointer to PCI device
9056  * @state: The current pci connection state
9057  *
9058  * This function is called after a PCI bus error affecting
9059  * this device has been detected.
9060  */
9061 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
9062                                                 pci_channel_state_t state)
9063 {
9064         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9065         struct net_device *netdev = adapter->netdev;
9066
9067 #ifdef CONFIG_PCI_IOV
9068         struct ixgbe_hw *hw = &adapter->hw;
9069         struct pci_dev *bdev, *vfdev;
9070         u32 dw0, dw1, dw2, dw3;
9071         int vf, pos;
9072         u16 req_id, pf_func;
9073
9074         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
9075             adapter->num_vfs == 0)
9076                 goto skip_bad_vf_detection;
9077
9078         bdev = pdev->bus->self;
9079         while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
9080                 bdev = bdev->bus->self;
9081
9082         if (!bdev)
9083                 goto skip_bad_vf_detection;
9084
9085         pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
9086         if (!pos)
9087                 goto skip_bad_vf_detection;
9088
9089         dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
9090         dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
9091         dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
9092         dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
9093         if (ixgbe_removed(hw->hw_addr))
9094                 goto skip_bad_vf_detection;
9095
9096         req_id = dw1 >> 16;
9097         /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
9098         if (!(req_id & 0x0080))
9099                 goto skip_bad_vf_detection;
9100
9101         pf_func = req_id & 0x01;
9102         if ((pf_func & 1) == (pdev->devfn & 1)) {
9103                 unsigned int device_id;
9104
9105                 vf = (req_id & 0x7F) >> 1;
9106                 e_dev_err("VF %d has caused a PCIe error\n", vf);
9107                 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
9108                                 "%8.8x\tdw3: %8.8x\n",
9109                 dw0, dw1, dw2, dw3);
9110                 switch (adapter->hw.mac.type) {
9111                 case ixgbe_mac_82599EB:
9112                         device_id = IXGBE_82599_VF_DEVICE_ID;
9113                         break;
9114                 case ixgbe_mac_X540:
9115                         device_id = IXGBE_X540_VF_DEVICE_ID;
9116                         break;
9117                 case ixgbe_mac_X550:
9118                         device_id = IXGBE_DEV_ID_X550_VF;
9119                         break;
9120                 case ixgbe_mac_X550EM_x:
9121                         device_id = IXGBE_DEV_ID_X550EM_X_VF;
9122                         break;
9123                 default:
9124                         device_id = 0;
9125                         break;
9126                 }
9127
9128                 /* Find the pci device of the offending VF */
9129                 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
9130                 while (vfdev) {
9131                         if (vfdev->devfn == (req_id & 0xFF))
9132                                 break;
9133                         vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
9134                                                device_id, vfdev);
9135                 }
9136                 /*
9137                  * There's a slim chance the VF could have been hot plugged,
9138                  * so if it is no longer present we don't need to issue the
9139                  * VFLR.  Just clean up the AER in that case.
9140                  */
9141                 if (vfdev) {
9142                         ixgbe_issue_vf_flr(adapter, vfdev);
9143                         /* Free device reference count */
9144                         pci_dev_put(vfdev);
9145                 }
9146
9147                 pci_cleanup_aer_uncorrect_error_status(pdev);
9148         }
9149
9150         /*
9151          * Even though the error may have occurred on the other port
9152          * we still need to increment the vf error reference count for
9153          * both ports because the I/O resume function will be called
9154          * for both of them.
9155          */
9156         adapter->vferr_refcount++;
9157
9158         return PCI_ERS_RESULT_RECOVERED;
9159
9160 skip_bad_vf_detection:
9161 #endif /* CONFIG_PCI_IOV */
9162         if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
9163                 return PCI_ERS_RESULT_DISCONNECT;
9164
9165         rtnl_lock();
9166         netif_device_detach(netdev);
9167
9168         if (state == pci_channel_io_perm_failure) {
9169                 rtnl_unlock();
9170                 return PCI_ERS_RESULT_DISCONNECT;
9171         }
9172
9173         if (netif_running(netdev))
9174                 ixgbe_down(adapter);
9175
9176         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
9177                 pci_disable_device(pdev);
9178         rtnl_unlock();
9179
9180         /* Request a slot reset. */
9181         return PCI_ERS_RESULT_NEED_RESET;
9182 }
9183
9184 /**
9185  * ixgbe_io_slot_reset - called after the pci bus has been reset.
9186  * @pdev: Pointer to PCI device
9187  *
9188  * Restart the card from scratch, as if from a cold-boot.
9189  */
9190 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
9191 {
9192         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9193         pci_ers_result_t result;
9194         int err;
9195
9196         if (pci_enable_device_mem(pdev)) {
9197                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
9198                 result = PCI_ERS_RESULT_DISCONNECT;
9199         } else {
9200                 smp_mb__before_atomic();
9201                 clear_bit(__IXGBE_DISABLED, &adapter->state);
9202                 adapter->hw.hw_addr = adapter->io_addr;
9203                 pci_set_master(pdev);
9204                 pci_restore_state(pdev);
9205                 pci_save_state(pdev);
9206
9207                 pci_wake_from_d3(pdev, false);
9208
9209                 ixgbe_reset(adapter);
9210                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9211                 result = PCI_ERS_RESULT_RECOVERED;
9212         }
9213
9214         err = pci_cleanup_aer_uncorrect_error_status(pdev);
9215         if (err) {
9216                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
9217                           "failed 0x%0x\n", err);
9218                 /* non-fatal, continue */
9219         }
9220
9221         return result;
9222 }
9223
9224 /**
9225  * ixgbe_io_resume - called when traffic can start flowing again.
9226  * @pdev: Pointer to PCI device
9227  *
9228  * This callback is called when the error recovery driver tells us that
9229  * its OK to resume normal operation.
9230  */
9231 static void ixgbe_io_resume(struct pci_dev *pdev)
9232 {
9233         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9234         struct net_device *netdev = adapter->netdev;
9235
9236 #ifdef CONFIG_PCI_IOV
9237         if (adapter->vferr_refcount) {
9238                 e_info(drv, "Resuming after VF err\n");
9239                 adapter->vferr_refcount--;
9240                 return;
9241         }
9242
9243 #endif
9244         if (netif_running(netdev))
9245                 ixgbe_up(adapter);
9246
9247         netif_device_attach(netdev);
9248 }
9249
9250 static const struct pci_error_handlers ixgbe_err_handler = {
9251         .error_detected = ixgbe_io_error_detected,
9252         .slot_reset = ixgbe_io_slot_reset,
9253         .resume = ixgbe_io_resume,
9254 };
9255
9256 static struct pci_driver ixgbe_driver = {
9257         .name     = ixgbe_driver_name,
9258         .id_table = ixgbe_pci_tbl,
9259         .probe    = ixgbe_probe,
9260         .remove   = ixgbe_remove,
9261 #ifdef CONFIG_PM
9262         .suspend  = ixgbe_suspend,
9263         .resume   = ixgbe_resume,
9264 #endif
9265         .shutdown = ixgbe_shutdown,
9266         .sriov_configure = ixgbe_pci_sriov_configure,
9267         .err_handler = &ixgbe_err_handler
9268 };
9269
9270 /**
9271  * ixgbe_init_module - Driver Registration Routine
9272  *
9273  * ixgbe_init_module is the first routine called when the driver is
9274  * loaded. All it does is register with the PCI subsystem.
9275  **/
9276 static int __init ixgbe_init_module(void)
9277 {
9278         int ret;
9279         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
9280         pr_info("%s\n", ixgbe_copyright);
9281
9282         ixgbe_dbg_init();
9283
9284         ret = pci_register_driver(&ixgbe_driver);
9285         if (ret) {
9286                 ixgbe_dbg_exit();
9287                 return ret;
9288         }
9289
9290 #ifdef CONFIG_IXGBE_DCA
9291         dca_register_notify(&dca_notifier);
9292 #endif
9293
9294         return 0;
9295 }
9296
9297 module_init(ixgbe_init_module);
9298
9299 /**
9300  * ixgbe_exit_module - Driver Exit Cleanup Routine
9301  *
9302  * ixgbe_exit_module is called just before the driver is removed
9303  * from memory.
9304  **/
9305 static void __exit ixgbe_exit_module(void)
9306 {
9307 #ifdef CONFIG_IXGBE_DCA
9308         dca_unregister_notify(&dca_notifier);
9309 #endif
9310         pci_unregister_driver(&ixgbe_driver);
9311
9312         ixgbe_dbg_exit();
9313 }
9314
9315 #ifdef CONFIG_IXGBE_DCA
9316 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
9317                             void *p)
9318 {
9319         int ret_val;
9320
9321         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
9322                                          __ixgbe_notify_dca);
9323
9324         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
9325 }
9326
9327 #endif /* CONFIG_IXGBE_DCA */
9328
9329 module_exit(ixgbe_exit_module);
9330
9331 /* ixgbe_main.c */