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 <[email protected]>
  24   e1000-devel Mailing List <[email protected]>
  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 #include <net/pkt_cls.h>
  55 #include <net/tc_act/tc_gact.h>
  56
  57 #ifdef CONFIG_OF
  58 #include <linux/of_net.h>
  59 #endif
  60
  61 #ifdef CONFIG_SPARC
  62 #include <asm/idprom.h>
  63 #include <asm/prom.h>
  64 #endif
  65
  66 #include "ixgbe.h"
  67 #include "ixgbe_common.h"
  68 #include "ixgbe_dcb_82599.h"
  69 #include "ixgbe_sriov.h"
  70 #include "ixgbe_model.h"
  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 "4.2.1-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         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X550EM_X_SFP), board_X550EM_x},
 141         /* required last entry */
 142         {0, }
 143 };
 144 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
 145
 146 #ifdef CONFIG_IXGBE_DCA
 147 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
 148                             void *p);
 149 static struct notifier_block dca_notifier = {
 150         .notifier_call = ixgbe_notify_dca,
 151         .next          = NULL,
 152         .priority      = 0
 153 };
 154 #endif
 155
 156 #ifdef CONFIG_PCI_IOV
 157 static unsigned int max_vfs;
 158 module_param(max_vfs, uint, 0);
 159 MODULE_PARM_DESC(max_vfs,
 160                  "Maximum number of virtual functions to allocate per physical function - default is zero and maximum value is 63. (Deprecated)");
 161 #endif /* CONFIG_PCI_IOV */
 162
 163 static unsigned int allow_unsupported_sfp;
 164 module_param(allow_unsupported_sfp, uint, 0);
 165 MODULE_PARM_DESC(allow_unsupported_sfp,
 166                  "Allow unsupported and untested SFP+ modules on 82599-based adapters");
 167
 168 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 169 static int debug = -1;
 170 module_param(debug, int, 0);
 171 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 172
 173 MODULE_AUTHOR("Intel Corporation, <[email protected]>");
 174 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
 175 MODULE_LICENSE("GPL");
 176 MODULE_VERSION(DRV_VERSION);
 177
 178 static struct workqueue_struct *ixgbe_wq;
 179
 180 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev);
 181
 182 static int ixgbe_read_pci_cfg_word_parent(struct ixgbe_adapter *adapter,
 183                                           u32 reg, u16 *value)
 184 {
 185         struct pci_dev *parent_dev;
 186         struct pci_bus *parent_bus;
 187
 188         parent_bus = adapter->pdev->bus->parent;
 189         if (!parent_bus)
 190                 return -1;
 191
 192         parent_dev = parent_bus->self;
 193         if (!parent_dev)
 194                 return -1;
 195
 196         if (!pci_is_pcie(parent_dev))
 197                 return -1;
 198
 199         pcie_capability_read_word(parent_dev, reg, value);
 200         if (*value == IXGBE_FAILED_READ_CFG_WORD &&
 201             ixgbe_check_cfg_remove(&adapter->hw, parent_dev))
 202                 return -1;
 203         return 0;
 204 }
 205
 206 static s32 ixgbe_get_parent_bus_info(struct ixgbe_adapter *adapter)
 207 {
 208         struct ixgbe_hw *hw = &adapter->hw;
 209         u16 link_status = 0;
 210         int err;
 211
 212         hw->bus.type = ixgbe_bus_type_pci_express;
 213
 214         /* Get the negotiated link width and speed from PCI config space of the
 215          * parent, as this device is behind a switch
 216          */
 217         err = ixgbe_read_pci_cfg_word_parent(adapter, 18, &link_status);
 218
 219         /* assume caller will handle error case */
 220         if (err)
 221                 return err;
 222
 223         hw->bus.width = ixgbe_convert_bus_width(link_status);
 224         hw->bus.speed = ixgbe_convert_bus_speed(link_status);
 225
 226         return 0;
 227 }
 228
 229 /**
 230  * ixgbe_check_from_parent - Determine whether PCIe info should come from parent
 231  * @hw: hw specific details
 232  *
 233  * This function is used by probe to determine whether a device's PCI-Express
 234  * bandwidth details should be gathered from the parent bus instead of from the
 235  * device. Used to ensure that various locations all have the correct device ID
 236  * checks.
 237  */
 238 static inline bool ixgbe_pcie_from_parent(struct ixgbe_hw *hw)
 239 {
 240         switch (hw->device_id) {
 241         case IXGBE_DEV_ID_82599_SFP_SF_QP:
 242         case IXGBE_DEV_ID_82599_QSFP_SF_QP:
 243                 return true;
 244         default:
 245                 return false;
 246         }
 247 }
 248
 249 static void ixgbe_check_minimum_link(struct ixgbe_adapter *adapter,
 250                                      int expected_gts)
 251 {
 252         struct ixgbe_hw *hw = &adapter->hw;
 253         int max_gts = 0;
 254         enum pci_bus_speed speed = PCI_SPEED_UNKNOWN;
 255         enum pcie_link_width width = PCIE_LNK_WIDTH_UNKNOWN;
 256         struct pci_dev *pdev;
 257
 258         /* Some devices are not connected over PCIe and thus do not negotiate
 259          * speed. These devices do not have valid bus info, and thus any report
 260          * we generate may not be correct.
 261          */
 262         if (hw->bus.type == ixgbe_bus_type_internal)
 263                 return;
 264
 265         /* determine whether to use the parent device */
 266         if (ixgbe_pcie_from_parent(&adapter->hw))
 267                 pdev = adapter->pdev->bus->parent->self;
 268         else
 269                 pdev = adapter->pdev;
 270
 271         if (pcie_get_minimum_link(pdev, &speed, &width) ||
 272             speed == PCI_SPEED_UNKNOWN || width == PCIE_LNK_WIDTH_UNKNOWN) {
 273                 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
 274                 return;
 275         }
 276
 277         switch (speed) {
 278         case PCIE_SPEED_2_5GT:
 279                 /* 8b/10b encoding reduces max throughput by 20% */
 280                 max_gts = 2 * width;
 281                 break;
 282         case PCIE_SPEED_5_0GT:
 283                 /* 8b/10b encoding reduces max throughput by 20% */
 284                 max_gts = 4 * width;
 285                 break;
 286         case PCIE_SPEED_8_0GT:
 287                 /* 128b/130b encoding reduces throughput by less than 2% */
 288                 max_gts = 8 * width;
 289                 break;
 290         default:
 291                 e_dev_warn("Unable to determine PCI Express bandwidth.\n");
 292                 return;
 293         }
 294
 295         e_dev_info("PCI Express bandwidth of %dGT/s available\n",
 296                    max_gts);
 297         e_dev_info("(Speed:%s, Width: x%d, Encoding Loss:%s)\n",
 298                    (speed == PCIE_SPEED_8_0GT ? "8.0GT/s" :
 299                     speed == PCIE_SPEED_5_0GT ? "5.0GT/s" :
 300                     speed == PCIE_SPEED_2_5GT ? "2.5GT/s" :
 301                     "Unknown"),
 302                    width,
 303                    (speed == PCIE_SPEED_2_5GT ? "20%" :
 304                     speed == PCIE_SPEED_5_0GT ? "20%" :
 305                     speed == PCIE_SPEED_8_0GT ? "<2%" :
 306                     "Unknown"));
 307
 308         if (max_gts < expected_gts) {
 309                 e_dev_warn("This is not sufficient for optimal performance of this card.\n");
 310                 e_dev_warn("For optimal performance, at least %dGT/s of bandwidth is required.\n",
 311                         expected_gts);
 312                 e_dev_warn("A slot with more lanes and/or higher speed is suggested.\n");
 313         }
 314 }
 315
 316 static void ixgbe_service_event_schedule(struct ixgbe_adapter *adapter)
 317 {
 318         if (!test_bit(__IXGBE_DOWN, &adapter->state) &&
 319             !test_bit(__IXGBE_REMOVING, &adapter->state) &&
 320             !test_and_set_bit(__IXGBE_SERVICE_SCHED, &adapter->state))
 321                 queue_work(ixgbe_wq, &adapter->service_task);
 322 }
 323
 324 static void ixgbe_remove_adapter(struct ixgbe_hw *hw)
 325 {
 326         struct ixgbe_adapter *adapter = hw->back;
 327
 328         if (!hw->hw_addr)
 329                 return;
 330         hw->hw_addr = NULL;
 331         e_dev_err("Adapter removed\n");
 332         if (test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
 333                 ixgbe_service_event_schedule(adapter);
 334 }
 335
 336 static void ixgbe_check_remove(struct ixgbe_hw *hw, u32 reg)
 337 {
 338         u32 value;
 339
 340         /* The following check not only optimizes a bit by not
 341          * performing a read on the status register when the
 342          * register just read was a status register read that
 343          * returned IXGBE_FAILED_READ_REG. It also blocks any
 344          * potential recursion.
 345          */
 346         if (reg == IXGBE_STATUS) {
 347                 ixgbe_remove_adapter(hw);
 348                 return;
 349         }
 350         value = ixgbe_read_reg(hw, IXGBE_STATUS);
 351         if (value == IXGBE_FAILED_READ_REG)
 352                 ixgbe_remove_adapter(hw);
 353 }
 354
 355 /**
 356  * ixgbe_read_reg - Read from device register
 357  * @hw: hw specific details
 358  * @reg: offset of register to read
 359  *
 360  * Returns : value read or IXGBE_FAILED_READ_REG if removed
 361  *
 362  * This function is used to read device registers. It checks for device
 363  * removal by confirming any read that returns all ones by checking the
 364  * status register value for all ones. This function avoids reading from
 365  * the hardware if a removal was previously detected in which case it
 366  * returns IXGBE_FAILED_READ_REG (all ones).
 367  */
 368 u32 ixgbe_read_reg(struct ixgbe_hw *hw, u32 reg)
 369 {
 370         u8 __iomem *reg_addr = ACCESS_ONCE(hw->hw_addr);
 371         u32 value;
 372
 373         if (ixgbe_removed(reg_addr))
 374                 return IXGBE_FAILED_READ_REG;
 375         value = readl(reg_addr + reg);
 376         if (unlikely(value == IXGBE_FAILED_READ_REG))
 377                 ixgbe_check_remove(hw, reg);
 378         return value;
 379 }
 380
 381 static bool ixgbe_check_cfg_remove(struct ixgbe_hw *hw, struct pci_dev *pdev)
 382 {
 383         u16 value;
 384
 385         pci_read_config_word(pdev, PCI_VENDOR_ID, &value);
 386         if (value == IXGBE_FAILED_READ_CFG_WORD) {
 387                 ixgbe_remove_adapter(hw);
 388                 return true;
 389         }
 390         return false;
 391 }
 392
 393 u16 ixgbe_read_pci_cfg_word(struct ixgbe_hw *hw, u32 reg)
 394 {
 395         struct ixgbe_adapter *adapter = hw->back;
 396         u16 value;
 397
 398         if (ixgbe_removed(hw->hw_addr))
 399                 return IXGBE_FAILED_READ_CFG_WORD;
 400         pci_read_config_word(adapter->pdev, reg, &value);
 401         if (value == IXGBE_FAILED_READ_CFG_WORD &&
 402             ixgbe_check_cfg_remove(hw, adapter->pdev))
 403                 return IXGBE_FAILED_READ_CFG_WORD;
 404         return value;
 405 }
 406
 407 #ifdef CONFIG_PCI_IOV
 408 static u32 ixgbe_read_pci_cfg_dword(struct ixgbe_hw *hw, u32 reg)
 409 {
 410         struct ixgbe_adapter *adapter = hw->back;
 411         u32 value;
 412
 413         if (ixgbe_removed(hw->hw_addr))
 414                 return IXGBE_FAILED_READ_CFG_DWORD;
 415         pci_read_config_dword(adapter->pdev, reg, &value);
 416         if (value == IXGBE_FAILED_READ_CFG_DWORD &&
 417             ixgbe_check_cfg_remove(hw, adapter->pdev))
 418                 return IXGBE_FAILED_READ_CFG_DWORD;
 419         return value;
 420 }
 421 #endif /* CONFIG_PCI_IOV */
 422
 423 void ixgbe_write_pci_cfg_word(struct ixgbe_hw *hw, u32 reg, u16 value)
 424 {
 425         struct ixgbe_adapter *adapter = hw->back;
 426
 427         if (ixgbe_removed(hw->hw_addr))
 428                 return;
 429         pci_write_config_word(adapter->pdev, reg, value);
 430 }
 431
 432 static void ixgbe_service_event_complete(struct ixgbe_adapter *adapter)
 433 {
 434         BUG_ON(!test_bit(__IXGBE_SERVICE_SCHED, &adapter->state));
 435
 436         /* flush memory to make sure state is correct before next watchdog */
 437         smp_mb__before_atomic();
 438         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
 439 }
 440
 441 struct ixgbe_reg_info {
 442         u32 ofs;
 443         char *name;
 444 };
 445
 446 static const struct ixgbe_reg_info ixgbe_reg_info_tbl[] = {
 447
 448         /* General Registers */
 449         {IXGBE_CTRL, "CTRL"},
 450         {IXGBE_STATUS, "STATUS"},
 451         {IXGBE_CTRL_EXT, "CTRL_EXT"},
 452
 453         /* Interrupt Registers */
 454         {IXGBE_EICR, "EICR"},
 455
 456         /* RX Registers */
 457         {IXGBE_SRRCTL(0), "SRRCTL"},
 458         {IXGBE_DCA_RXCTRL(0), "DRXCTL"},
 459         {IXGBE_RDLEN(0), "RDLEN"},
 460         {IXGBE_RDH(0), "RDH"},
 461         {IXGBE_RDT(0), "RDT"},
 462         {IXGBE_RXDCTL(0), "RXDCTL"},
 463         {IXGBE_RDBAL(0), "RDBAL"},
 464         {IXGBE_RDBAH(0), "RDBAH"},
 465
 466         /* TX Registers */
 467         {IXGBE_TDBAL(0), "TDBAL"},
 468         {IXGBE_TDBAH(0), "TDBAH"},
 469         {IXGBE_TDLEN(0), "TDLEN"},
 470         {IXGBE_TDH(0), "TDH"},
 471         {IXGBE_TDT(0), "TDT"},
 472         {IXGBE_TXDCTL(0), "TXDCTL"},
 473
 474         /* List Terminator */
 475         { .name = NULL }
 476 };
 477
 478
 479 /*
 480  * ixgbe_regdump - register printout routine
 481  */
 482 static void ixgbe_regdump(struct ixgbe_hw *hw, struct ixgbe_reg_info *reginfo)
 483 {
 484         int i = 0, j = 0;
 485         char rname[16];
 486         u32 regs[64];
 487
 488         switch (reginfo->ofs) {
 489         case IXGBE_SRRCTL(0):
 490                 for (i = 0; i < 64; i++)
 491                         regs[i] = IXGBE_READ_REG(hw, IXGBE_SRRCTL(i));
 492                 break;
 493         case IXGBE_DCA_RXCTRL(0):
 494                 for (i = 0; i < 64; i++)
 495                         regs[i] = IXGBE_READ_REG(hw, IXGBE_DCA_RXCTRL(i));
 496                 break;
 497         case IXGBE_RDLEN(0):
 498                 for (i = 0; i < 64; i++)
 499                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDLEN(i));
 500                 break;
 501         case IXGBE_RDH(0):
 502                 for (i = 0; i < 64; i++)
 503                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDH(i));
 504                 break;
 505         case IXGBE_RDT(0):
 506                 for (i = 0; i < 64; i++)
 507                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDT(i));
 508                 break;
 509         case IXGBE_RXDCTL(0):
 510                 for (i = 0; i < 64; i++)
 511                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RXDCTL(i));
 512                 break;
 513         case IXGBE_RDBAL(0):
 514                 for (i = 0; i < 64; i++)
 515                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAL(i));
 516                 break;
 517         case IXGBE_RDBAH(0):
 518                 for (i = 0; i < 64; i++)
 519                         regs[i] = IXGBE_READ_REG(hw, IXGBE_RDBAH(i));
 520                 break;
 521         case IXGBE_TDBAL(0):
 522                 for (i = 0; i < 64; i++)
 523                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAL(i));
 524                 break;
 525         case IXGBE_TDBAH(0):
 526                 for (i = 0; i < 64; i++)
 527                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDBAH(i));
 528                 break;
 529         case IXGBE_TDLEN(0):
 530                 for (i = 0; i < 64; i++)
 531                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDLEN(i));
 532                 break;
 533         case IXGBE_TDH(0):
 534                 for (i = 0; i < 64; i++)
 535                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDH(i));
 536                 break;
 537         case IXGBE_TDT(0):
 538                 for (i = 0; i < 64; i++)
 539                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TDT(i));
 540                 break;
 541         case IXGBE_TXDCTL(0):
 542                 for (i = 0; i < 64; i++)
 543                         regs[i] = IXGBE_READ_REG(hw, IXGBE_TXDCTL(i));
 544                 break;
 545         default:
 546                 pr_info("%-15s %08x\n", reginfo->name,
 547                         IXGBE_READ_REG(hw, reginfo->ofs));
 548                 return;
 549         }
 550
 551         for (i = 0; i < 8; i++) {
 552                 snprintf(rname, 16, "%s[%d-%d]", reginfo->name, i*8, i*8+7);
 553                 pr_err("%-15s", rname);
 554                 for (j = 0; j < 8; j++)
 555                         pr_cont(" %08x", regs[i*8+j]);
 556                 pr_cont("\n");
 557         }
 558
 559 }
 560
 561 /*
 562  * ixgbe_dump - Print registers, tx-rings and rx-rings
 563  */
 564 static void ixgbe_dump(struct ixgbe_adapter *adapter)
 565 {
 566         struct net_device *netdev = adapter->netdev;
 567         struct ixgbe_hw *hw = &adapter->hw;
 568         struct ixgbe_reg_info *reginfo;
 569         int n = 0;
 570         struct ixgbe_ring *tx_ring;
 571         struct ixgbe_tx_buffer *tx_buffer;
 572         union ixgbe_adv_tx_desc *tx_desc;
 573         struct my_u0 { u64 a; u64 b; } *u0;
 574         struct ixgbe_ring *rx_ring;
 575         union ixgbe_adv_rx_desc *rx_desc;
 576         struct ixgbe_rx_buffer *rx_buffer_info;
 577         u32 staterr;
 578         int i = 0;
 579
 580         if (!netif_msg_hw(adapter))
 581                 return;
 582
 583         /* Print netdevice Info */
 584         if (netdev) {
 585                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 586                 pr_info("Device Name     state            "
 587                         "trans_start      last_rx\n");
 588                 pr_info("%-15s %016lX %016lX %016lX\n",
 589                         netdev->name,
 590                         netdev->state,
 591                         netdev->trans_start,
 592                         netdev->last_rx);
 593         }
 594
 595         /* Print Registers */
 596         dev_info(&adapter->pdev->dev, "Register Dump\n");
 597         pr_info(" Register Name   Value\n");
 598         for (reginfo = (struct ixgbe_reg_info *)ixgbe_reg_info_tbl;
 599              reginfo->name; reginfo++) {
 600                 ixgbe_regdump(hw, reginfo);
 601         }
 602
 603         /* Print TX Ring Summary */
 604         if (!netdev || !netif_running(netdev))
 605                 return;
 606
 607         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 608         pr_info(" %s     %s              %s        %s\n",
 609                 "Queue [NTU] [NTC] [bi(ntc)->dma  ]",
 610                 "leng", "ntw", "timestamp");
 611         for (n = 0; n < adapter->num_tx_queues; n++) {
 612                 tx_ring = adapter->tx_ring[n];
 613                 tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 614                 pr_info(" %5d %5X %5X %016llX %08X %p %016llX\n",
 615                            n, tx_ring->next_to_use, tx_ring->next_to_clean,
 616                            (u64)dma_unmap_addr(tx_buffer, dma),
 617                            dma_unmap_len(tx_buffer, len),
 618                            tx_buffer->next_to_watch,
 619                            (u64)tx_buffer->time_stamp);
 620         }
 621
 622         /* Print TX Rings */
 623         if (!netif_msg_tx_done(adapter))
 624                 goto rx_ring_summary;
 625
 626         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 627
 628         /* Transmit Descriptor Formats
 629          *
 630          * 82598 Advanced Transmit Descriptor
 631          *   +--------------------------------------------------------------+
 632          * 0 |         Buffer Address [63:0]                                |
 633          *   +--------------------------------------------------------------+
 634          * 8 |  PAYLEN  | POPTS  | IDX | STA | DCMD  |DTYP |  RSV |  DTALEN |
 635          *   +--------------------------------------------------------------+
 636          *   63       46 45    40 39 36 35 32 31   24 23 20 19              0
 637          *
 638          * 82598 Advanced Transmit Descriptor (Write-Back Format)
 639          *   +--------------------------------------------------------------+
 640          * 0 |                          RSV [63:0]                          |
 641          *   +--------------------------------------------------------------+
 642          * 8 |            RSV           |  STA  |          NXTSEQ           |
 643          *   +--------------------------------------------------------------+
 644          *   63                       36 35   32 31                         0
 645          *
 646          * 82599+ Advanced Transmit Descriptor
 647          *   +--------------------------------------------------------------+
 648          * 0 |         Buffer Address [63:0]                                |
 649          *   +--------------------------------------------------------------+
 650          * 8 |PAYLEN  |POPTS|CC|IDX  |STA  |DCMD  |DTYP |MAC  |RSV  |DTALEN |
 651          *   +--------------------------------------------------------------+
 652          *   63     46 45 40 39 38 36 35 32 31  24 23 20 19 18 17 16 15     0
 653          *
 654          * 82599+ Advanced Transmit Descriptor (Write-Back Format)
 655          *   +--------------------------------------------------------------+
 656          * 0 |                          RSV [63:0]                          |
 657          *   +--------------------------------------------------------------+
 658          * 8 |            RSV           |  STA  |           RSV             |
 659          *   +--------------------------------------------------------------+
 660          *   63                       36 35   32 31                         0
 661          */
 662
 663         for (n = 0; n < adapter->num_tx_queues; n++) {
 664                 tx_ring = adapter->tx_ring[n];
 665                 pr_info("------------------------------------\n");
 666                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 667                 pr_info("------------------------------------\n");
 668                 pr_info("%s%s    %s              %s        %s          %s\n",
 669                         "T [desc]     [address 63:0  ] ",
 670                         "[PlPOIdStDDt Ln] [bi->dma       ] ",
 671                         "leng", "ntw", "timestamp", "bi->skb");
 672
 673                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 674                         tx_desc = IXGBE_TX_DESC(tx_ring, i);
 675                         tx_buffer = &tx_ring->tx_buffer_info[i];
 676                         u0 = (struct my_u0 *)tx_desc;
 677                         if (dma_unmap_len(tx_buffer, len) > 0) {
 678                                 pr_info("T [0x%03X]    %016llX %016llX %016llX %08X %p %016llX %p",
 679                                         i,
 680                                         le64_to_cpu(u0->a),
 681                                         le64_to_cpu(u0->b),
 682                                         (u64)dma_unmap_addr(tx_buffer, dma),
 683                                         dma_unmap_len(tx_buffer, len),
 684                                         tx_buffer->next_to_watch,
 685                                         (u64)tx_buffer->time_stamp,
 686                                         tx_buffer->skb);
 687                                 if (i == tx_ring->next_to_use &&
 688                                         i == tx_ring->next_to_clean)
 689                                         pr_cont(" NTC/U\n");
 690                                 else if (i == tx_ring->next_to_use)
 691                                         pr_cont(" NTU\n");
 692                                 else if (i == tx_ring->next_to_clean)
 693                                         pr_cont(" NTC\n");
 694                                 else
 695                                         pr_cont("\n");
 696
 697                                 if (netif_msg_pktdata(adapter) &&
 698                                     tx_buffer->skb)
 699                                         print_hex_dump(KERN_INFO, "",
 700                                                 DUMP_PREFIX_ADDRESS, 16, 1,
 701                                                 tx_buffer->skb->data,
 702                                                 dma_unmap_len(tx_buffer, len),
 703                                                 true);
 704                         }
 705                 }
 706         }
 707
 708         /* Print RX Rings Summary */
 709 rx_ring_summary:
 710         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 711         pr_info("Queue [NTU] [NTC]\n");
 712         for (n = 0; n < adapter->num_rx_queues; n++) {
 713                 rx_ring = adapter->rx_ring[n];
 714                 pr_info("%5d %5X %5X\n",
 715                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 716         }
 717
 718         /* Print RX Rings */
 719         if (!netif_msg_rx_status(adapter))
 720                 return;
 721
 722         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 723
 724         /* Receive Descriptor Formats
 725          *
 726          * 82598 Advanced Receive Descriptor (Read) Format
 727          *    63                                           1        0
 728          *    +-----------------------------------------------------+
 729          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 730          *    +----------------------------------------------+------+
 731          *  8 |       Header Buffer Address [63:1]           |  DD  |
 732          *    +-----------------------------------------------------+
 733          *
 734          *
 735          * 82598 Advanced Receive Descriptor (Write-Back) Format
 736          *
 737          *   63       48 47    32 31  30      21 20 16 15   4 3     0
 738          *   +------------------------------------------------------+
 739          * 0 |       RSS Hash /  |SPH| HDR_LEN  | RSV |Packet|  RSS |
 740          *   | Packet   | IP     |   |          |     | Type | Type |
 741          *   | Checksum | Ident  |   |          |     |      |      |
 742          *   +------------------------------------------------------+
 743          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 744          *   +------------------------------------------------------+
 745          *   63       48 47    32 31            20 19               0
 746          *
 747          * 82599+ Advanced Receive Descriptor (Read) Format
 748          *    63                                           1        0
 749          *    +-----------------------------------------------------+
 750          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 751          *    +----------------------------------------------+------+
 752          *  8 |       Header Buffer Address [63:1]           |  DD  |
 753          *    +-----------------------------------------------------+
 754          *
 755          *
 756          * 82599+ Advanced Receive Descriptor (Write-Back) Format
 757          *
 758          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 759          *   +------------------------------------------------------+
 760          * 0 |RSS / Frag Checksum|SPH| HDR_LEN  |RSC- |Packet|  RSS |
 761          *   |/ RTT / PCoE_PARAM |   |          | CNT | Type | Type |
 762          *   |/ Flow Dir Flt ID  |   |          |     |      |      |
 763          *   +------------------------------------------------------+
 764          * 8 | VLAN Tag | Length |Extended Error| Xtnd Status/NEXTP |
 765          *   +------------------------------------------------------+
 766          *   63       48 47    32 31          20 19                 0
 767          */
 768
 769         for (n = 0; n < adapter->num_rx_queues; n++) {
 770                 rx_ring = adapter->rx_ring[n];
 771                 pr_info("------------------------------------\n");
 772                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 773                 pr_info("------------------------------------\n");
 774                 pr_info("%s%s%s",
 775                         "R  [desc]      [ PktBuf     A0] ",
 776                         "[  HeadBuf   DD] [bi->dma       ] [bi->skb       ] ",
 777                         "<-- Adv Rx Read format\n");
 778                 pr_info("%s%s%s",
 779                         "RWB[desc]      [PcsmIpSHl PtRs] ",
 780                         "[vl er S cks ln] ---------------- [bi->skb       ] ",
 781                         "<-- Adv Rx Write-Back format\n");
 782
 783                 for (i = 0; i < rx_ring->count; i++) {
 784                         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 785                         rx_desc = IXGBE_RX_DESC(rx_ring, i);
 786                         u0 = (struct my_u0 *)rx_desc;
 787                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 788                         if (staterr & IXGBE_RXD_STAT_DD) {
 789                                 /* Descriptor Done */
 790                                 pr_info("RWB[0x%03X]     %016llX "
 791                                         "%016llX ---------------- %p", i,
 792                                         le64_to_cpu(u0->a),
 793                                         le64_to_cpu(u0->b),
 794                                         rx_buffer_info->skb);
 795                         } else {
 796                                 pr_info("R  [0x%03X]     %016llX "
 797                                         "%016llX %016llX %p", i,
 798                                         le64_to_cpu(u0->a),
 799                                         le64_to_cpu(u0->b),
 800                                         (u64)rx_buffer_info->dma,
 801                                         rx_buffer_info->skb);
 802
 803                                 if (netif_msg_pktdata(adapter) &&
 804                                     rx_buffer_info->dma) {
 805                                         print_hex_dump(KERN_INFO, "",
 806                                            DUMP_PREFIX_ADDRESS, 16, 1,
 807                                            page_address(rx_buffer_info->page) +
 808                                                     rx_buffer_info->page_offset,
 809                                            ixgbe_rx_bufsz(rx_ring), true);
 810                                 }
 811                         }
 812
 813                         if (i == rx_ring->next_to_use)
 814                                 pr_cont(" NTU\n");
 815                         else if (i == rx_ring->next_to_clean)
 816                                 pr_cont(" NTC\n");
 817                         else
 818                                 pr_cont("\n");
 819
 820                 }
 821         }
 822 }
 823
 824 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 825 {
 826         u32 ctrl_ext;
 827
 828         /* Let firmware take over control of h/w */
 829         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 830         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 831                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 832 }
 833
 834 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 835 {
 836         u32 ctrl_ext;
 837
 838         /* Let firmware know the driver has taken over */
 839         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 840         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 841                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 842 }
 843
 844 /**
 845  * ixgbe_set_ivar - set the IVAR registers, mapping interrupt causes to vectors
 846  * @adapter: pointer to adapter struct
 847  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 848  * @queue: queue to map the corresponding interrupt to
 849  * @msix_vector: the vector to map to the corresponding queue
 850  *
 851  */
 852 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, s8 direction,
 853                            u8 queue, u8 msix_vector)
 854 {
 855         u32 ivar, index;
 856         struct ixgbe_hw *hw = &adapter->hw;
 857         switch (hw->mac.type) {
 858         case ixgbe_mac_82598EB:
 859                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 860                 if (direction == -1)
 861                         direction = 0;
 862                 index = (((direction * 64) + queue) >> 2) & 0x1F;
 863                 ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(index));
 864                 ivar &= ~(0xFF << (8 * (queue & 0x3)));
 865                 ivar |= (msix_vector << (8 * (queue & 0x3)));
 866                 IXGBE_WRITE_REG(hw, IXGBE_IVAR(index), ivar);
 867                 break;
 868         case ixgbe_mac_82599EB:
 869         case ixgbe_mac_X540:
 870         case ixgbe_mac_X550:
 871         case ixgbe_mac_X550EM_x:
 872                 if (direction == -1) {
 873                         /* other causes */
 874                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 875                         index = ((queue & 1) * 8);
 876                         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR_MISC);
 877                         ivar &= ~(0xFF << index);
 878                         ivar |= (msix_vector << index);
 879                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR_MISC, ivar);
 880                         break;
 881                 } else {
 882                         /* tx or rx causes */
 883                         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 884                         index = ((16 * (queue & 1)) + (8 * direction));
 885                         ivar = IXGBE_READ_REG(hw, IXGBE_IVAR(queue >> 1));
 886                         ivar &= ~(0xFF << index);
 887                         ivar |= (msix_vector << index);
 888                         IXGBE_WRITE_REG(hw, IXGBE_IVAR(queue >> 1), ivar);
 889                         break;
 890                 }
 891         default:
 892                 break;
 893         }
 894 }
 895
 896 static inline void ixgbe_irq_rearm_queues(struct ixgbe_adapter *adapter,
 897                                           u64 qmask)
 898 {
 899         u32 mask;
 900
 901         switch (adapter->hw.mac.type) {
 902         case ixgbe_mac_82598EB:
 903                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
 904                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, mask);
 905                 break;
 906         case ixgbe_mac_82599EB:
 907         case ixgbe_mac_X540:
 908         case ixgbe_mac_X550:
 909         case ixgbe_mac_X550EM_x:
 910                 mask = (qmask & 0xFFFFFFFF);
 911                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(0), mask);
 912                 mask = (qmask >> 32);
 913                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS_EX(1), mask);
 914                 break;
 915         default:
 916                 break;
 917         }
 918 }
 919
 920 void ixgbe_unmap_and_free_tx_resource(struct ixgbe_ring *ring,
 921                                       struct ixgbe_tx_buffer *tx_buffer)
 922 {
 923         if (tx_buffer->skb) {
 924                 dev_kfree_skb_any(tx_buffer->skb);
 925                 if (dma_unmap_len(tx_buffer, len))
 926                         dma_unmap_single(ring->dev,
 927                                          dma_unmap_addr(tx_buffer, dma),
 928                                          dma_unmap_len(tx_buffer, len),
 929                                          DMA_TO_DEVICE);
 930         } else if (dma_unmap_len(tx_buffer, len)) {
 931                 dma_unmap_page(ring->dev,
 932                                dma_unmap_addr(tx_buffer, dma),
 933                                dma_unmap_len(tx_buffer, len),
 934                                DMA_TO_DEVICE);
 935         }
 936         tx_buffer->next_to_watch = NULL;
 937         tx_buffer->skb = NULL;
 938         dma_unmap_len_set(tx_buffer, len, 0);
 939         /* tx_buffer must be completely set up in the transmit path */
 940 }
 941
 942 static void ixgbe_update_xoff_rx_lfc(struct ixgbe_adapter *adapter)
 943 {
 944         struct ixgbe_hw *hw = &adapter->hw;
 945         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 946         int i;
 947         u32 data;
 948
 949         if ((hw->fc.current_mode != ixgbe_fc_full) &&
 950             (hw->fc.current_mode != ixgbe_fc_rx_pause))
 951                 return;
 952
 953         switch (hw->mac.type) {
 954         case ixgbe_mac_82598EB:
 955                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
 956                 break;
 957         default:
 958                 data = IXGBE_READ_REG(hw, IXGBE_LXOFFRXCNT);
 959         }
 960         hwstats->lxoffrxc += data;
 961
 962         /* refill credits (no tx hang) if we received xoff */
 963         if (!data)
 964                 return;
 965
 966         for (i = 0; i < adapter->num_tx_queues; i++)
 967                 clear_bit(__IXGBE_HANG_CHECK_ARMED,
 968                           &adapter->tx_ring[i]->state);
 969 }
 970
 971 static void ixgbe_update_xoff_received(struct ixgbe_adapter *adapter)
 972 {
 973         struct ixgbe_hw *hw = &adapter->hw;
 974         struct ixgbe_hw_stats *hwstats = &adapter->stats;
 975         u32 xoff[8] = {0};
 976         u8 tc;
 977         int i;
 978         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
 979
 980         if (adapter->ixgbe_ieee_pfc)
 981                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
 982
 983         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED) || !pfc_en) {
 984                 ixgbe_update_xoff_rx_lfc(adapter);
 985                 return;
 986         }
 987
 988         /* update stats for each tc, only valid with PFC enabled */
 989         for (i = 0; i < MAX_TX_PACKET_BUFFERS; i++) {
 990                 u32 pxoffrxc;
 991
 992                 switch (hw->mac.type) {
 993                 case ixgbe_mac_82598EB:
 994                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXC(i));
 995                         break;
 996                 default:
 997                         pxoffrxc = IXGBE_READ_REG(hw, IXGBE_PXOFFRXCNT(i));
 998                 }
 999                 hwstats->pxoffrxc[i] += pxoffrxc;
1000                 /* Get the TC for given UP */
1001                 tc = netdev_get_prio_tc_map(adapter->netdev, i);
1002                 xoff[tc] += pxoffrxc;
1003         }
1004
1005         /* disarm tx queues that have received xoff frames */
1006         for (i = 0; i < adapter->num_tx_queues; i++) {
1007                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
1008
1009                 tc = tx_ring->dcb_tc;
1010                 if (xoff[tc])
1011                         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1012         }
1013 }
1014
1015 static u64 ixgbe_get_tx_completed(struct ixgbe_ring *ring)
1016 {
1017         return ring->stats.packets;
1018 }
1019
1020 static u64 ixgbe_get_tx_pending(struct ixgbe_ring *ring)
1021 {
1022         struct ixgbe_adapter *adapter;
1023         struct ixgbe_hw *hw;
1024         u32 head, tail;
1025
1026         if (ring->l2_accel_priv)
1027                 adapter = ring->l2_accel_priv->real_adapter;
1028         else
1029                 adapter = netdev_priv(ring->netdev);
1030
1031         hw = &adapter->hw;
1032         head = IXGBE_READ_REG(hw, IXGBE_TDH(ring->reg_idx));
1033         tail = IXGBE_READ_REG(hw, IXGBE_TDT(ring->reg_idx));
1034
1035         if (head != tail)
1036                 return (head < tail) ?
1037                         tail - head : (tail + ring->count - head);
1038
1039         return 0;
1040 }
1041
1042 static inline bool ixgbe_check_tx_hang(struct ixgbe_ring *tx_ring)
1043 {
1044         u32 tx_done = ixgbe_get_tx_completed(tx_ring);
1045         u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1046         u32 tx_pending = ixgbe_get_tx_pending(tx_ring);
1047
1048         clear_check_for_tx_hang(tx_ring);
1049
1050         /*
1051          * Check for a hung queue, but be thorough. This verifies
1052          * that a transmit has been completed since the previous
1053          * check AND there is at least one packet pending. The
1054          * ARMED bit is set to indicate a potential hang. The
1055          * bit is cleared if a pause frame is received to remove
1056          * false hang detection due to PFC or 802.3x frames. By
1057          * requiring this to fail twice we avoid races with
1058          * pfc clearing the ARMED bit and conditions where we
1059          * run the check_tx_hang logic with a transmit completion
1060          * pending but without time to complete it yet.
1061          */
1062         if (tx_done_old == tx_done && tx_pending)
1063                 /* make sure it is true for two checks in a row */
1064                 return test_and_set_bit(__IXGBE_HANG_CHECK_ARMED,
1065                                         &tx_ring->state);
1066         /* update completed stats and continue */
1067         tx_ring->tx_stats.tx_done_old = tx_done;
1068         /* reset the countdown */
1069         clear_bit(__IXGBE_HANG_CHECK_ARMED, &tx_ring->state);
1070
1071         return false;
1072 }
1073
1074 /**
1075  * ixgbe_tx_timeout_reset - initiate reset due to Tx timeout
1076  * @adapter: driver private struct
1077  **/
1078 static void ixgbe_tx_timeout_reset(struct ixgbe_adapter *adapter)
1079 {
1080
1081         /* Do the reset outside of interrupt context */
1082         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
1083                 adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
1084                 e_warn(drv, "initiating reset due to tx timeout\n");
1085                 ixgbe_service_event_schedule(adapter);
1086         }
1087 }
1088
1089 /**
1090  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
1091  * @q_vector: structure containing interrupt and ring information
1092  * @tx_ring: tx ring to clean
1093  **/
1094 static bool ixgbe_clean_tx_irq(struct ixgbe_q_vector *q_vector,
1095                                struct ixgbe_ring *tx_ring, int napi_budget)
1096 {
1097         struct ixgbe_adapter *adapter = q_vector->adapter;
1098         struct ixgbe_tx_buffer *tx_buffer;
1099         union ixgbe_adv_tx_desc *tx_desc;
1100         unsigned int total_bytes = 0, total_packets = 0;
1101         unsigned int budget = q_vector->tx.work_limit;
1102         unsigned int i = tx_ring->next_to_clean;
1103
1104         if (test_bit(__IXGBE_DOWN, &adapter->state))
1105                 return true;
1106
1107         tx_buffer = &tx_ring->tx_buffer_info[i];
1108         tx_desc = IXGBE_TX_DESC(tx_ring, i);
1109         i -= tx_ring->count;
1110
1111         do {
1112                 union ixgbe_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
1113
1114                 /* if next_to_watch is not set then there is no work pending */
1115                 if (!eop_desc)
1116                         break;
1117
1118                 /* prevent any other reads prior to eop_desc */
1119                 read_barrier_depends();
1120
1121                 /* if DD is not set pending work has not been completed */
1122                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
1123                         break;
1124
1125                 /* clear next_to_watch to prevent false hangs */
1126                 tx_buffer->next_to_watch = NULL;
1127
1128                 /* update the statistics for this packet */
1129                 total_bytes += tx_buffer->bytecount;
1130                 total_packets += tx_buffer->gso_segs;
1131
1132                 /* free the skb */
1133                 napi_consume_skb(tx_buffer->skb, napi_budget);
1134
1135                 /* unmap skb header data */
1136                 dma_unmap_single(tx_ring->dev,
1137                                  dma_unmap_addr(tx_buffer, dma),
1138                                  dma_unmap_len(tx_buffer, len),
1139                                  DMA_TO_DEVICE);
1140
1141                 /* clear tx_buffer data */
1142                 tx_buffer->skb = NULL;
1143                 dma_unmap_len_set(tx_buffer, len, 0);
1144
1145                 /* unmap remaining buffers */
1146                 while (tx_desc != eop_desc) {
1147                         tx_buffer++;
1148                         tx_desc++;
1149                         i++;
1150                         if (unlikely(!i)) {
1151                                 i -= tx_ring->count;
1152                                 tx_buffer = tx_ring->tx_buffer_info;
1153                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1154                         }
1155
1156                         /* unmap any remaining paged data */
1157                         if (dma_unmap_len(tx_buffer, len)) {
1158                                 dma_unmap_page(tx_ring->dev,
1159                                                dma_unmap_addr(tx_buffer, dma),
1160                                                dma_unmap_len(tx_buffer, len),
1161                                                DMA_TO_DEVICE);
1162                                 dma_unmap_len_set(tx_buffer, len, 0);
1163                         }
1164                 }
1165
1166                 /* move us one more past the eop_desc for start of next pkt */
1167                 tx_buffer++;
1168                 tx_desc++;
1169                 i++;
1170                 if (unlikely(!i)) {
1171                         i -= tx_ring->count;
1172                         tx_buffer = tx_ring->tx_buffer_info;
1173                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
1174                 }
1175
1176                 /* issue prefetch for next Tx descriptor */
1177                 prefetch(tx_desc);
1178
1179                 /* update budget accounting */
1180                 budget--;
1181         } while (likely(budget));
1182
1183         i += tx_ring->count;
1184         tx_ring->next_to_clean = i;
1185         u64_stats_update_begin(&tx_ring->syncp);
1186         tx_ring->stats.bytes += total_bytes;
1187         tx_ring->stats.packets += total_packets;
1188         u64_stats_update_end(&tx_ring->syncp);
1189         q_vector->tx.total_bytes += total_bytes;
1190         q_vector->tx.total_packets += total_packets;
1191
1192         if (check_for_tx_hang(tx_ring) && ixgbe_check_tx_hang(tx_ring)) {
1193                 /* schedule immediate reset if we believe we hung */
1194                 struct ixgbe_hw *hw = &adapter->hw;
1195                 e_err(drv, "Detected Tx Unit Hang\n"
1196                         "  Tx Queue             <%d>\n"
1197                         "  TDH, TDT             <%x>, <%x>\n"
1198                         "  next_to_use          <%x>\n"
1199                         "  next_to_clean        <%x>\n"
1200                         "tx_buffer_info[next_to_clean]\n"
1201                         "  time_stamp           <%lx>\n"
1202                         "  jiffies              <%lx>\n",
1203                         tx_ring->queue_index,
1204                         IXGBE_READ_REG(hw, IXGBE_TDH(tx_ring->reg_idx)),
1205                         IXGBE_READ_REG(hw, IXGBE_TDT(tx_ring->reg_idx)),
1206                         tx_ring->next_to_use, i,
1207                         tx_ring->tx_buffer_info[i].time_stamp, jiffies);
1208
1209                 netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
1210
1211                 e_info(probe,
1212                        "tx hang %d detected on queue %d, resetting adapter\n",
1213                         adapter->tx_timeout_count + 1, tx_ring->queue_index);
1214
1215                 /* schedule immediate reset if we believe we hung */
1216                 ixgbe_tx_timeout_reset(adapter);
1217
1218                 /* the adapter is about to reset, no point in enabling stuff */
1219                 return true;
1220         }
1221
1222         netdev_tx_completed_queue(txring_txq(tx_ring),
1223                                   total_packets, total_bytes);
1224
1225 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
1226         if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1227                      (ixgbe_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1228                 /* Make sure that anybody stopping the queue after this
1229                  * sees the new next_to_clean.
1230                  */
1231                 smp_mb();
1232                 if (__netif_subqueue_stopped(tx_ring->netdev,
1233                                              tx_ring->queue_index)
1234                     && !test_bit(__IXGBE_DOWN, &adapter->state)) {
1235                         netif_wake_subqueue(tx_ring->netdev,
1236                                             tx_ring->queue_index);
1237                         ++tx_ring->tx_stats.restart_queue;
1238                 }
1239         }
1240
1241         return !!budget;
1242 }
1243
1244 #ifdef CONFIG_IXGBE_DCA
1245 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
1246                                 struct ixgbe_ring *tx_ring,
1247                                 int cpu)
1248 {
1249         struct ixgbe_hw *hw = &adapter->hw;
1250         u32 txctrl = 0;
1251         u16 reg_offset;
1252
1253         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1254                 txctrl = dca3_get_tag(tx_ring->dev, cpu);
1255
1256         switch (hw->mac.type) {
1257         case ixgbe_mac_82598EB:
1258                 reg_offset = IXGBE_DCA_TXCTRL(tx_ring->reg_idx);
1259                 break;
1260         case ixgbe_mac_82599EB:
1261         case ixgbe_mac_X540:
1262                 reg_offset = IXGBE_DCA_TXCTRL_82599(tx_ring->reg_idx);
1263                 txctrl <<= IXGBE_DCA_TXCTRL_CPUID_SHIFT_82599;
1264                 break;
1265         default:
1266                 /* for unknown hardware do not write register */
1267                 return;
1268         }
1269
1270         /*
1271          * We can enable relaxed ordering for reads, but not writes when
1272          * DCA is enabled.  This is due to a known issue in some chipsets
1273          * which will cause the DCA tag to be cleared.
1274          */
1275         txctrl |= IXGBE_DCA_TXCTRL_DESC_RRO_EN |
1276                   IXGBE_DCA_TXCTRL_DATA_RRO_EN |
1277                   IXGBE_DCA_TXCTRL_DESC_DCA_EN;
1278
1279         IXGBE_WRITE_REG(hw, reg_offset, txctrl);
1280 }
1281
1282 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
1283                                 struct ixgbe_ring *rx_ring,
1284                                 int cpu)
1285 {
1286         struct ixgbe_hw *hw = &adapter->hw;
1287         u32 rxctrl = 0;
1288         u8 reg_idx = rx_ring->reg_idx;
1289
1290         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1291                 rxctrl = dca3_get_tag(rx_ring->dev, cpu);
1292
1293         switch (hw->mac.type) {
1294         case ixgbe_mac_82599EB:
1295         case ixgbe_mac_X540:
1296                 rxctrl <<= IXGBE_DCA_RXCTRL_CPUID_SHIFT_82599;
1297                 break;
1298         default:
1299                 break;
1300         }
1301
1302         /*
1303          * We can enable relaxed ordering for reads, but not writes when
1304          * DCA is enabled.  This is due to a known issue in some chipsets
1305          * which will cause the DCA tag to be cleared.
1306          */
1307         rxctrl |= IXGBE_DCA_RXCTRL_DESC_RRO_EN |
1308                   IXGBE_DCA_RXCTRL_DATA_DCA_EN |
1309                   IXGBE_DCA_RXCTRL_DESC_DCA_EN;
1310
1311         IXGBE_WRITE_REG(hw, IXGBE_DCA_RXCTRL(reg_idx), rxctrl);
1312 }
1313
1314 static void ixgbe_update_dca(struct ixgbe_q_vector *q_vector)
1315 {
1316         struct ixgbe_adapter *adapter = q_vector->adapter;
1317         struct ixgbe_ring *ring;
1318         int cpu = get_cpu();
1319
1320         if (q_vector->cpu == cpu)
1321                 goto out_no_update;
1322
1323         ixgbe_for_each_ring(ring, q_vector->tx)
1324                 ixgbe_update_tx_dca(adapter, ring, cpu);
1325
1326         ixgbe_for_each_ring(ring, q_vector->rx)
1327                 ixgbe_update_rx_dca(adapter, ring, cpu);
1328
1329         q_vector->cpu = cpu;
1330 out_no_update:
1331         put_cpu();
1332 }
1333
1334 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
1335 {
1336         int i;
1337
1338         /* always use CB2 mode, difference is masked in the CB driver */
1339         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1340                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1341                                 IXGBE_DCA_CTRL_DCA_MODE_CB2);
1342         else
1343                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1344                                 IXGBE_DCA_CTRL_DCA_DISABLE);
1345
1346         for (i = 0; i < adapter->num_q_vectors; i++) {
1347                 adapter->q_vector[i]->cpu = -1;
1348                 ixgbe_update_dca(adapter->q_vector[i]);
1349         }
1350 }
1351
1352 static int __ixgbe_notify_dca(struct device *dev, void *data)
1353 {
1354         struct ixgbe_adapter *adapter = dev_get_drvdata(dev);
1355         unsigned long event = *(unsigned long *)data;
1356
1357         if (!(adapter->flags & IXGBE_FLAG_DCA_CAPABLE))
1358                 return 0;
1359
1360         switch (event) {
1361         case DCA_PROVIDER_ADD:
1362                 /* if we're already enabled, don't do it again */
1363                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1364                         break;
1365                 if (dca_add_requester(dev) == 0) {
1366                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
1367                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1368                                         IXGBE_DCA_CTRL_DCA_MODE_CB2);
1369                         break;
1370                 }
1371                 /* Fall Through since DCA is disabled. */
1372         case DCA_PROVIDER_REMOVE:
1373                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
1374                         dca_remove_requester(dev);
1375                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
1376                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
1377                                         IXGBE_DCA_CTRL_DCA_DISABLE);
1378                 }
1379                 break;
1380         }
1381
1382         return 0;
1383 }
1384
1385 #endif /* CONFIG_IXGBE_DCA */
1386
1387 #define IXGBE_RSS_L4_TYPES_MASK \
1388         ((1ul << IXGBE_RXDADV_RSSTYPE_IPV4_TCP) | \
1389          (1ul << IXGBE_RXDADV_RSSTYPE_IPV4_UDP) | \
1390          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_TCP) | \
1391          (1ul << IXGBE_RXDADV_RSSTYPE_IPV6_UDP))
1392
1393 static inline void ixgbe_rx_hash(struct ixgbe_ring *ring,
1394                                  union ixgbe_adv_rx_desc *rx_desc,
1395                                  struct sk_buff *skb)
1396 {
1397         u16 rss_type;
1398
1399         if (!(ring->netdev->features & NETIF_F_RXHASH))
1400                 return;
1401
1402         rss_type = le16_to_cpu(rx_desc->wb.lower.lo_dword.hs_rss.pkt_info) &
1403                    IXGBE_RXDADV_RSSTYPE_MASK;
1404
1405         if (!rss_type)
1406                 return;
1407
1408         skb_set_hash(skb, le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
1409                      (IXGBE_RSS_L4_TYPES_MASK & (1ul << rss_type)) ?
1410                      PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
1411 }
1412
1413 #ifdef IXGBE_FCOE
1414 /**
1415  * ixgbe_rx_is_fcoe - check the rx desc for incoming pkt type
1416  * @ring: structure containing ring specific data
1417  * @rx_desc: advanced rx descriptor
1418  *
1419  * Returns : true if it is FCoE pkt
1420  */
1421 static inline bool ixgbe_rx_is_fcoe(struct ixgbe_ring *ring,
1422                                     union ixgbe_adv_rx_desc *rx_desc)
1423 {
1424         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1425
1426         return test_bit(__IXGBE_RX_FCOE, &ring->state) &&
1427                ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_ETQF_MASK)) ==
1428                 (cpu_to_le16(IXGBE_ETQF_FILTER_FCOE <<
1429                              IXGBE_RXDADV_PKTTYPE_ETQF_SHIFT)));
1430 }
1431
1432 #endif /* IXGBE_FCOE */
1433 /**
1434  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
1435  * @ring: structure containing ring specific data
1436  * @rx_desc: current Rx descriptor being processed
1437  * @skb: skb currently being received and modified
1438  **/
1439 static inline void ixgbe_rx_checksum(struct ixgbe_ring *ring,
1440                                      union ixgbe_adv_rx_desc *rx_desc,
1441                                      struct sk_buff *skb)
1442 {
1443         __le16 pkt_info = rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
1444         __le16 hdr_info = rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
1445         bool encap_pkt = false;
1446
1447         skb_checksum_none_assert(skb);
1448
1449         /* Rx csum disabled */
1450         if (!(ring->netdev->features & NETIF_F_RXCSUM))
1451                 return;
1452
1453         if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_VXLAN)) &&
1454             (hdr_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_TUNNEL >> 16))) {
1455                 encap_pkt = true;
1456                 skb->encapsulation = 1;
1457         }
1458
1459         /* if IP and error */
1460         if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_IPCS) &&
1461             ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_IPE)) {
1462                 ring->rx_stats.csum_err++;
1463                 return;
1464         }
1465
1466         if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_L4CS))
1467                 return;
1468
1469         if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_TCPE)) {
1470                 /*
1471                  * 82599 errata, UDP frames with a 0 checksum can be marked as
1472                  * checksum errors.
1473                  */
1474                 if ((pkt_info & cpu_to_le16(IXGBE_RXDADV_PKTTYPE_UDP)) &&
1475                     test_bit(__IXGBE_RX_CSUM_UDP_ZERO_ERR, &ring->state))
1476                         return;
1477
1478                 ring->rx_stats.csum_err++;
1479                 return;
1480         }
1481
1482         /* It must be a TCP or UDP packet with a valid checksum */
1483         skb->ip_summed = CHECKSUM_UNNECESSARY;
1484         if (encap_pkt) {
1485                 if (!ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_OUTERIPCS))
1486                         return;
1487
1488                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXDADV_ERR_OUTERIPER)) {
1489                         skb->ip_summed = CHECKSUM_NONE;
1490                         return;
1491                 }
1492                 /* If we checked the outer header let the stack know */
1493                 skb->csum_level = 1;
1494         }
1495 }
1496
1497 static bool ixgbe_alloc_mapped_page(struct ixgbe_ring *rx_ring,
1498                                     struct ixgbe_rx_buffer *bi)
1499 {
1500         struct page *page = bi->page;
1501         dma_addr_t dma;
1502
1503         /* since we are recycling buffers we should seldom need to alloc */
1504         if (likely(page))
1505                 return true;
1506
1507         /* alloc new page for storage */
1508         page = dev_alloc_pages(ixgbe_rx_pg_order(rx_ring));
1509         if (unlikely(!page)) {
1510                 rx_ring->rx_stats.alloc_rx_page_failed++;
1511                 return false;
1512         }
1513
1514         /* map page for use */
1515         dma = dma_map_page(rx_ring->dev, page, 0,
1516                            ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1517
1518         /*
1519          * if mapping failed free memory back to system since
1520          * there isn't much point in holding memory we can't use
1521          */
1522         if (dma_mapping_error(rx_ring->dev, dma)) {
1523                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1524
1525                 rx_ring->rx_stats.alloc_rx_page_failed++;
1526                 return false;
1527         }
1528
1529         bi->dma = dma;
1530         bi->page = page;
1531         bi->page_offset = 0;
1532
1533         return true;
1534 }
1535
1536 /**
1537  * ixgbe_alloc_rx_buffers - Replace used receive buffers
1538  * @rx_ring: ring to place buffers on
1539  * @cleaned_count: number of buffers to replace
1540  **/
1541 void ixgbe_alloc_rx_buffers(struct ixgbe_ring *rx_ring, u16 cleaned_count)
1542 {
1543         union ixgbe_adv_rx_desc *rx_desc;
1544         struct ixgbe_rx_buffer *bi;
1545         u16 i = rx_ring->next_to_use;
1546
1547         /* nothing to do */
1548         if (!cleaned_count)
1549                 return;
1550
1551         rx_desc = IXGBE_RX_DESC(rx_ring, i);
1552         bi = &rx_ring->rx_buffer_info[i];
1553         i -= rx_ring->count;
1554
1555         do {
1556                 if (!ixgbe_alloc_mapped_page(rx_ring, bi))
1557                         break;
1558
1559                 /*
1560                  * Refresh the desc even if buffer_addrs didn't change
1561                  * because each write-back erases this info.
1562                  */
1563                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
1564
1565                 rx_desc++;
1566                 bi++;
1567                 i++;
1568                 if (unlikely(!i)) {
1569                         rx_desc = IXGBE_RX_DESC(rx_ring, 0);
1570                         bi = rx_ring->rx_buffer_info;
1571                         i -= rx_ring->count;
1572                 }
1573
1574                 /* clear the status bits for the next_to_use descriptor */
1575                 rx_desc->wb.upper.status_error = 0;
1576
1577                 cleaned_count--;
1578         } while (cleaned_count);
1579
1580         i += rx_ring->count;
1581
1582         if (rx_ring->next_to_use != i) {
1583                 rx_ring->next_to_use = i;
1584
1585                 /* update next to alloc since we have filled the ring */
1586                 rx_ring->next_to_alloc = i;
1587
1588                 /* Force memory writes to complete before letting h/w
1589                  * know there are new descriptors to fetch.  (Only
1590                  * applicable for weak-ordered memory model archs,
1591                  * such as IA-64).
1592                  */
1593                 wmb();
1594                 writel(i, rx_ring->tail);
1595         }
1596 }
1597
1598 static void ixgbe_set_rsc_gso_size(struct ixgbe_ring *ring,
1599                                    struct sk_buff *skb)
1600 {
1601         u16 hdr_len = skb_headlen(skb);
1602
1603         /* set gso_size to avoid messing up TCP MSS */
1604         skb_shinfo(skb)->gso_size = DIV_ROUND_UP((skb->len - hdr_len),
1605                                                  IXGBE_CB(skb)->append_cnt);
1606         skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1607 }
1608
1609 static void ixgbe_update_rsc_stats(struct ixgbe_ring *rx_ring,
1610                                    struct sk_buff *skb)
1611 {
1612         /* if append_cnt is 0 then frame is not RSC */
1613         if (!IXGBE_CB(skb)->append_cnt)
1614                 return;
1615
1616         rx_ring->rx_stats.rsc_count += IXGBE_CB(skb)->append_cnt;
1617         rx_ring->rx_stats.rsc_flush++;
1618
1619         ixgbe_set_rsc_gso_size(rx_ring, skb);
1620
1621         /* gso_size is computed using append_cnt so always clear it last */
1622         IXGBE_CB(skb)->append_cnt = 0;
1623 }
1624
1625 /**
1626  * ixgbe_process_skb_fields - Populate skb header fields from Rx descriptor
1627  * @rx_ring: rx descriptor ring packet is being transacted on
1628  * @rx_desc: pointer to the EOP Rx descriptor
1629  * @skb: pointer to current skb being populated
1630  *
1631  * This function checks the ring, descriptor, and packet information in
1632  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
1633  * other fields within the skb.
1634  **/
1635 static void ixgbe_process_skb_fields(struct ixgbe_ring *rx_ring,
1636                                      union ixgbe_adv_rx_desc *rx_desc,
1637                                      struct sk_buff *skb)
1638 {
1639         struct net_device *dev = rx_ring->netdev;
1640         u32 flags = rx_ring->q_vector->adapter->flags;
1641
1642         ixgbe_update_rsc_stats(rx_ring, skb);
1643
1644         ixgbe_rx_hash(rx_ring, rx_desc, skb);
1645
1646         ixgbe_rx_checksum(rx_ring, rx_desc, skb);
1647
1648         if (unlikely(flags & IXGBE_FLAG_RX_HWTSTAMP_ENABLED))
1649                 ixgbe_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
1650
1651         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
1652             ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_VP)) {
1653                 u16 vid = le16_to_cpu(rx_desc->wb.upper.vlan);
1654                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
1655         }
1656
1657         skb_record_rx_queue(skb, rx_ring->queue_index);
1658
1659         skb->protocol = eth_type_trans(skb, dev);
1660 }
1661
1662 static void ixgbe_rx_skb(struct ixgbe_q_vector *q_vector,
1663                          struct sk_buff *skb)
1664 {
1665         skb_mark_napi_id(skb, &q_vector->napi);
1666         if (ixgbe_qv_busy_polling(q_vector))
1667                 netif_receive_skb(skb);
1668         else
1669                 napi_gro_receive(&q_vector->napi, skb);
1670 }
1671
1672 /**
1673  * ixgbe_is_non_eop - process handling of non-EOP buffers
1674  * @rx_ring: Rx ring being processed
1675  * @rx_desc: Rx descriptor for current buffer
1676  * @skb: Current socket buffer containing buffer in progress
1677  *
1678  * This function updates next to clean.  If the buffer is an EOP buffer
1679  * this function exits returning false, otherwise it will place the
1680  * sk_buff in the next buffer to be chained and return true indicating
1681  * that this is in fact a non-EOP buffer.
1682  **/
1683 static bool ixgbe_is_non_eop(struct ixgbe_ring *rx_ring,
1684                              union ixgbe_adv_rx_desc *rx_desc,
1685                              struct sk_buff *skb)
1686 {
1687         u32 ntc = rx_ring->next_to_clean + 1;
1688
1689         /* fetch, update, and store next to clean */
1690         ntc = (ntc < rx_ring->count) ? ntc : 0;
1691         rx_ring->next_to_clean = ntc;
1692
1693         prefetch(IXGBE_RX_DESC(rx_ring, ntc));
1694
1695         /* update RSC append count if present */
1696         if (ring_is_rsc_enabled(rx_ring)) {
1697                 __le32 rsc_enabled = rx_desc->wb.lower.lo_dword.data &
1698                                      cpu_to_le32(IXGBE_RXDADV_RSCCNT_MASK);
1699
1700                 if (unlikely(rsc_enabled)) {
1701                         u32 rsc_cnt = le32_to_cpu(rsc_enabled);
1702
1703                         rsc_cnt >>= IXGBE_RXDADV_RSCCNT_SHIFT;
1704                         IXGBE_CB(skb)->append_cnt += rsc_cnt - 1;
1705
1706                         /* update ntc based on RSC value */
1707                         ntc = le32_to_cpu(rx_desc->wb.upper.status_error);
1708                         ntc &= IXGBE_RXDADV_NEXTP_MASK;
1709                         ntc >>= IXGBE_RXDADV_NEXTP_SHIFT;
1710                 }
1711         }
1712
1713         /* if we are the last buffer then there is nothing else to do */
1714         if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1715                 return false;
1716
1717         /* place skb in next buffer to be received */
1718         rx_ring->rx_buffer_info[ntc].skb = skb;
1719         rx_ring->rx_stats.non_eop_descs++;
1720
1721         return true;
1722 }
1723
1724 /**
1725  * ixgbe_pull_tail - ixgbe specific version of skb_pull_tail
1726  * @rx_ring: rx descriptor ring packet is being transacted on
1727  * @skb: pointer to current skb being adjusted
1728  *
1729  * This function is an ixgbe specific version of __pskb_pull_tail.  The
1730  * main difference between this version and the original function is that
1731  * this function can make several assumptions about the state of things
1732  * that allow for significant optimizations versus the standard function.
1733  * As a result we can do things like drop a frag and maintain an accurate
1734  * truesize for the skb.
1735  */
1736 static void ixgbe_pull_tail(struct ixgbe_ring *rx_ring,
1737                             struct sk_buff *skb)
1738 {
1739         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1740         unsigned char *va;
1741         unsigned int pull_len;
1742
1743         /*
1744          * it is valid to use page_address instead of kmap since we are
1745          * working with pages allocated out of the lomem pool per
1746          * alloc_page(GFP_ATOMIC)
1747          */
1748         va = skb_frag_address(frag);
1749
1750         /*
1751          * we need the header to contain the greater of either ETH_HLEN or
1752          * 60 bytes if the skb->len is less than 60 for skb_pad.
1753          */
1754         pull_len = eth_get_headlen(va, IXGBE_RX_HDR_SIZE);
1755
1756         /* align pull length to size of long to optimize memcpy performance */
1757         skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
1758
1759         /* update all of the pointers */
1760         skb_frag_size_sub(frag, pull_len);
1761         frag->page_offset += pull_len;
1762         skb->data_len -= pull_len;
1763         skb->tail += pull_len;
1764 }
1765
1766 /**
1767  * ixgbe_dma_sync_frag - perform DMA sync for first frag of SKB
1768  * @rx_ring: rx descriptor ring packet is being transacted on
1769  * @skb: pointer to current skb being updated
1770  *
1771  * This function provides a basic DMA sync up for the first fragment of an
1772  * skb.  The reason for doing this is that the first fragment cannot be
1773  * unmapped until we have reached the end of packet descriptor for a buffer
1774  * chain.
1775  */
1776 static void ixgbe_dma_sync_frag(struct ixgbe_ring *rx_ring,
1777                                 struct sk_buff *skb)
1778 {
1779         /* if the page was released unmap it, else just sync our portion */
1780         if (unlikely(IXGBE_CB(skb)->page_released)) {
1781                 dma_unmap_page(rx_ring->dev, IXGBE_CB(skb)->dma,
1782                                ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
1783                 IXGBE_CB(skb)->page_released = false;
1784         } else {
1785                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
1786
1787                 dma_sync_single_range_for_cpu(rx_ring->dev,
1788                                               IXGBE_CB(skb)->dma,
1789                                               frag->page_offset,
1790                                               ixgbe_rx_bufsz(rx_ring),
1791                                               DMA_FROM_DEVICE);
1792         }
1793         IXGBE_CB(skb)->dma = 0;
1794 }
1795
1796 /**
1797  * ixgbe_cleanup_headers - Correct corrupted or empty headers
1798  * @rx_ring: rx descriptor ring packet is being transacted on
1799  * @rx_desc: pointer to the EOP Rx descriptor
1800  * @skb: pointer to current skb being fixed
1801  *
1802  * Check for corrupted packet headers caused by senders on the local L2
1803  * embedded NIC switch not setting up their Tx Descriptors right.  These
1804  * should be very rare.
1805  *
1806  * Also address the case where we are pulling data in on pages only
1807  * and as such no data is present in the skb header.
1808  *
1809  * In addition if skb is not at least 60 bytes we need to pad it so that
1810  * it is large enough to qualify as a valid Ethernet frame.
1811  *
1812  * Returns true if an error was encountered and skb was freed.
1813  **/
1814 static bool ixgbe_cleanup_headers(struct ixgbe_ring *rx_ring,
1815                                   union ixgbe_adv_rx_desc *rx_desc,
1816                                   struct sk_buff *skb)
1817 {
1818         struct net_device *netdev = rx_ring->netdev;
1819
1820         /* verify that the packet does not have any known errors */
1821         if (unlikely(ixgbe_test_staterr(rx_desc,
1822                                         IXGBE_RXDADV_ERR_FRAME_ERR_MASK) &&
1823             !(netdev->features & NETIF_F_RXALL))) {
1824                 dev_kfree_skb_any(skb);
1825                 return true;
1826         }
1827
1828         /* place header in linear portion of buffer */
1829         if (skb_is_nonlinear(skb))
1830                 ixgbe_pull_tail(rx_ring, skb);
1831
1832 #ifdef IXGBE_FCOE
1833         /* do not attempt to pad FCoE Frames as this will disrupt DDP */
1834         if (ixgbe_rx_is_fcoe(rx_ring, rx_desc))
1835                 return false;
1836
1837 #endif
1838         /* if eth_skb_pad returns an error the skb was freed */
1839         if (eth_skb_pad(skb))
1840                 return true;
1841
1842         return false;
1843 }
1844
1845 /**
1846  * ixgbe_reuse_rx_page - page flip buffer and store it back on the ring
1847  * @rx_ring: rx descriptor ring to store buffers on
1848  * @old_buff: donor buffer to have page reused
1849  *
1850  * Synchronizes page for reuse by the adapter
1851  **/
1852 static void ixgbe_reuse_rx_page(struct ixgbe_ring *rx_ring,
1853                                 struct ixgbe_rx_buffer *old_buff)
1854 {
1855         struct ixgbe_rx_buffer *new_buff;
1856         u16 nta = rx_ring->next_to_alloc;
1857
1858         new_buff = &rx_ring->rx_buffer_info[nta];
1859
1860         /* update, and store next to alloc */
1861         nta++;
1862         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
1863
1864         /* transfer page from old buffer to new buffer */
1865         *new_buff = *old_buff;
1866
1867         /* sync the buffer for use by the device */
1868         dma_sync_single_range_for_device(rx_ring->dev, new_buff->dma,
1869                                          new_buff->page_offset,
1870                                          ixgbe_rx_bufsz(rx_ring),
1871                                          DMA_FROM_DEVICE);
1872 }
1873
1874 static inline bool ixgbe_page_is_reserved(struct page *page)
1875 {
1876         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
1877 }
1878
1879 /**
1880  * ixgbe_add_rx_frag - Add contents of Rx buffer to sk_buff
1881  * @rx_ring: rx descriptor ring to transact packets on
1882  * @rx_buffer: buffer containing page to add
1883  * @rx_desc: descriptor containing length of buffer written by hardware
1884  * @skb: sk_buff to place the data into
1885  *
1886  * This function will add the data contained in rx_buffer->page to the skb.
1887  * This is done either through a direct copy if the data in the buffer is
1888  * less than the skb header size, otherwise it will just attach the page as
1889  * a frag to the skb.
1890  *
1891  * The function will then update the page offset if necessary and return
1892  * true if the buffer can be reused by the adapter.
1893  **/
1894 static bool ixgbe_add_rx_frag(struct ixgbe_ring *rx_ring,
1895                               struct ixgbe_rx_buffer *rx_buffer,
1896                               union ixgbe_adv_rx_desc *rx_desc,
1897                               struct sk_buff *skb)
1898 {
1899         struct page *page = rx_buffer->page;
1900         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
1901 #if (PAGE_SIZE < 8192)
1902         unsigned int truesize = ixgbe_rx_bufsz(rx_ring);
1903 #else
1904         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
1905         unsigned int last_offset = ixgbe_rx_pg_size(rx_ring) -
1906                                    ixgbe_rx_bufsz(rx_ring);
1907 #endif
1908
1909         if ((size <= IXGBE_RX_HDR_SIZE) && !skb_is_nonlinear(skb)) {
1910                 unsigned char *va = page_address(page) + rx_buffer->page_offset;
1911
1912                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
1913
1914                 /* page is not reserved, we can reuse buffer as-is */
1915                 if (likely(!ixgbe_page_is_reserved(page)))
1916                         return true;
1917
1918                 /* this page cannot be reused so discard it */
1919                 __free_pages(page, ixgbe_rx_pg_order(rx_ring));
1920                 return false;
1921         }
1922
1923         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
1924                         rx_buffer->page_offset, size, truesize);
1925
1926         /* avoid re-using remote pages */
1927         if (unlikely(ixgbe_page_is_reserved(page)))
1928                 return false;
1929
1930 #if (PAGE_SIZE < 8192)
1931         /* if we are only owner of page we can reuse it */
1932         if (unlikely(page_count(page) != 1))
1933                 return false;
1934
1935         /* flip page offset to other buffer */
1936         rx_buffer->page_offset ^= truesize;
1937 #else
1938         /* move offset up to the next cache line */
1939         rx_buffer->page_offset += truesize;
1940
1941         if (rx_buffer->page_offset > last_offset)
1942                 return false;
1943 #endif
1944
1945         /* Even if we own the page, we are not allowed to use atomic_set()
1946          * This would break get_page_unless_zero() users.
1947          */
1948         atomic_inc(&page->_count);
1949
1950         return true;
1951 }
1952
1953 static struct sk_buff *ixgbe_fetch_rx_buffer(struct ixgbe_ring *rx_ring,
1954                                              union ixgbe_adv_rx_desc *rx_desc)
1955 {
1956         struct ixgbe_rx_buffer *rx_buffer;
1957         struct sk_buff *skb;
1958         struct page *page;
1959
1960         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
1961         page = rx_buffer->page;
1962         prefetchw(page);
1963
1964         skb = rx_buffer->skb;
1965
1966         if (likely(!skb)) {
1967                 void *page_addr = page_address(page) +
1968                                   rx_buffer->page_offset;
1969
1970                 /* prefetch first cache line of first page */
1971                 prefetch(page_addr);
1972 #if L1_CACHE_BYTES < 128
1973                 prefetch(page_addr + L1_CACHE_BYTES);
1974 #endif
1975
1976                 /* allocate a skb to store the frags */
1977                 skb = napi_alloc_skb(&rx_ring->q_vector->napi,
1978                                      IXGBE_RX_HDR_SIZE);
1979                 if (unlikely(!skb)) {
1980                         rx_ring->rx_stats.alloc_rx_buff_failed++;
1981                         return NULL;
1982                 }
1983
1984                 /*
1985                  * we will be copying header into skb->data in
1986                  * pskb_may_pull so it is in our interest to prefetch
1987                  * it now to avoid a possible cache miss
1988                  */
1989                 prefetchw(skb->data);
1990
1991                 /*
1992                  * Delay unmapping of the first packet. It carries the
1993                  * header information, HW may still access the header
1994                  * after the writeback.  Only unmap it when EOP is
1995                  * reached
1996                  */
1997                 if (likely(ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP)))
1998                         goto dma_sync;
1999
2000                 IXGBE_CB(skb)->dma = rx_buffer->dma;
2001         } else {
2002                 if (ixgbe_test_staterr(rx_desc, IXGBE_RXD_STAT_EOP))
2003                         ixgbe_dma_sync_frag(rx_ring, skb);
2004
2005 dma_sync:
2006                 /* we are reusing so sync this buffer for CPU use */
2007                 dma_sync_single_range_for_cpu(rx_ring->dev,
2008                                               rx_buffer->dma,
2009                                               rx_buffer->page_offset,
2010                                               ixgbe_rx_bufsz(rx_ring),
2011                                               DMA_FROM_DEVICE);
2012
2013                 rx_buffer->skb = NULL;
2014         }
2015
2016         /* pull page into skb */
2017         if (ixgbe_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
2018                 /* hand second half of page back to the ring */
2019                 ixgbe_reuse_rx_page(rx_ring, rx_buffer);
2020         } else if (IXGBE_CB(skb)->dma == rx_buffer->dma) {
2021                 /* the page has been released from the ring */
2022                 IXGBE_CB(skb)->page_released = true;
2023         } else {
2024                 /* we are not reusing the buffer so unmap it */
2025                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
2026                                ixgbe_rx_pg_size(rx_ring),
2027                                DMA_FROM_DEVICE);
2028         }
2029
2030         /* clear contents of buffer_info */
2031         rx_buffer->page = NULL;
2032
2033         return skb;
2034 }
2035
2036 /**
2037  * ixgbe_clean_rx_irq - Clean completed descriptors from Rx ring - bounce buf
2038  * @q_vector: structure containing interrupt and ring information
2039  * @rx_ring: rx descriptor ring to transact packets on
2040  * @budget: Total limit on number of packets to process
2041  *
2042  * This function provides a "bounce buffer" approach to Rx interrupt
2043  * processing.  The advantage to this is that on systems that have
2044  * expensive overhead for IOMMU access this provides a means of avoiding
2045  * it by maintaining the mapping of the page to the syste.
2046  *
2047  * Returns amount of work completed
2048  **/
2049 static int ixgbe_clean_rx_irq(struct ixgbe_q_vector *q_vector,
2050                                struct ixgbe_ring *rx_ring,
2051                                const int budget)
2052 {
2053         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
2054 #ifdef IXGBE_FCOE
2055         struct ixgbe_adapter *adapter = q_vector->adapter;
2056         int ddp_bytes;
2057         unsigned int mss = 0;
2058 #endif /* IXGBE_FCOE */
2059         u16 cleaned_count = ixgbe_desc_unused(rx_ring);
2060
2061         while (likely(total_rx_packets < budget)) {
2062                 union ixgbe_adv_rx_desc *rx_desc;
2063                 struct sk_buff *skb;
2064
2065                 /* return some buffers to hardware, one at a time is too slow */
2066                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
2067                         ixgbe_alloc_rx_buffers(rx_ring, cleaned_count);
2068                         cleaned_count = 0;
2069                 }
2070
2071                 rx_desc = IXGBE_RX_DESC(rx_ring, rx_ring->next_to_clean);
2072
2073                 if (!rx_desc->wb.upper.status_error)
2074                         break;
2075
2076                 /* This memory barrier is needed to keep us from reading
2077                  * any other fields out of the rx_desc until we know the
2078                  * descriptor has been written back
2079                  */
2080                 dma_rmb();
2081
2082                 /* retrieve a buffer from the ring */
2083                 skb = ixgbe_fetch_rx_buffer(rx_ring, rx_desc);
2084
2085                 /* exit if we failed to retrieve a buffer */
2086                 if (!skb)
2087                         break;
2088
2089                 cleaned_count++;
2090
2091                 /* place incomplete frames back on ring for completion */
2092                 if (ixgbe_is_non_eop(rx_ring, rx_desc, skb))
2093                         continue;
2094
2095                 /* verify the packet layout is correct */
2096                 if (ixgbe_cleanup_headers(rx_ring, rx_desc, skb))
2097                         continue;
2098
2099                 /* probably a little skewed due to removing CRC */
2100                 total_rx_bytes += skb->len;
2101
2102                 /* populate checksum, timestamp, VLAN, and protocol */
2103                 ixgbe_process_skb_fields(rx_ring, rx_desc, skb);
2104
2105 #ifdef IXGBE_FCOE
2106                 /* if ddp, not passing to ULD unless for FCP_RSP or error */
2107                 if (ixgbe_rx_is_fcoe(rx_ring, rx_desc)) {
2108                         ddp_bytes = ixgbe_fcoe_ddp(adapter, rx_desc, skb);
2109                         /* include DDPed FCoE data */
2110                         if (ddp_bytes > 0) {
2111                                 if (!mss) {
2112                                         mss = rx_ring->netdev->mtu -
2113                                                 sizeof(struct fcoe_hdr) -
2114                                                 sizeof(struct fc_frame_header) -
2115                                                 sizeof(struct fcoe_crc_eof);
2116                                         if (mss > 512)
2117                                                 mss &= ~511;
2118                                 }
2119                                 total_rx_bytes += ddp_bytes;
2120                                 total_rx_packets += DIV_ROUND_UP(ddp_bytes,
2121                                                                  mss);
2122                         }
2123                         if (!ddp_bytes) {
2124                                 dev_kfree_skb_any(skb);
2125                                 continue;
2126                         }
2127                 }
2128
2129 #endif /* IXGBE_FCOE */
2130                 ixgbe_rx_skb(q_vector, skb);
2131
2132                 /* update budget accounting */
2133                 total_rx_packets++;
2134         }
2135
2136         u64_stats_update_begin(&rx_ring->syncp);
2137         rx_ring->stats.packets += total_rx_packets;
2138         rx_ring->stats.bytes += total_rx_bytes;
2139         u64_stats_update_end(&rx_ring->syncp);
2140         q_vector->rx.total_packets += total_rx_packets;
2141         q_vector->rx.total_bytes += total_rx_bytes;
2142
2143         return total_rx_packets;
2144 }
2145
2146 #ifdef CONFIG_NET_RX_BUSY_POLL
2147 /* must be called with local_bh_disable()d */
2148 static int ixgbe_low_latency_recv(struct napi_struct *napi)
2149 {
2150         struct ixgbe_q_vector *q_vector =
2151                         container_of(napi, struct ixgbe_q_vector, napi);
2152         struct ixgbe_adapter *adapter = q_vector->adapter;
2153         struct ixgbe_ring  *ring;
2154         int found = 0;
2155
2156         if (test_bit(__IXGBE_DOWN, &adapter->state))
2157                 return LL_FLUSH_FAILED;
2158
2159         if (!ixgbe_qv_lock_poll(q_vector))
2160                 return LL_FLUSH_BUSY;
2161
2162         ixgbe_for_each_ring(ring, q_vector->rx) {
2163                 found = ixgbe_clean_rx_irq(q_vector, ring, 4);
2164 #ifdef BP_EXTENDED_STATS
2165                 if (found)
2166                         ring->stats.cleaned += found;
2167                 else
2168                         ring->stats.misses++;
2169 #endif
2170                 if (found)
2171                         break;
2172         }
2173
2174         ixgbe_qv_unlock_poll(q_vector);
2175
2176         return found;
2177 }
2178 #endif  /* CONFIG_NET_RX_BUSY_POLL */
2179
2180 /**
2181  * ixgbe_configure_msix - Configure MSI-X hardware
2182  * @adapter: board private structure
2183  *
2184  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
2185  * interrupts.
2186  **/
2187 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
2188 {
2189         struct ixgbe_q_vector *q_vector;
2190         int v_idx;
2191         u32 mask;
2192
2193         /* Populate MSIX to EITR Select */
2194         if (adapter->num_vfs > 32) {
2195                 u32 eitrsel = (1 << (adapter->num_vfs - 32)) - 1;
2196                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, eitrsel);
2197         }
2198
2199         /*
2200          * Populate the IVAR table and set the ITR values to the
2201          * corresponding register.
2202          */
2203         for (v_idx = 0; v_idx < adapter->num_q_vectors; v_idx++) {
2204                 struct ixgbe_ring *ring;
2205                 q_vector = adapter->q_vector[v_idx];
2206
2207                 ixgbe_for_each_ring(ring, q_vector->rx)
2208                         ixgbe_set_ivar(adapter, 0, ring->reg_idx, v_idx);
2209
2210                 ixgbe_for_each_ring(ring, q_vector->tx)
2211                         ixgbe_set_ivar(adapter, 1, ring->reg_idx, v_idx);
2212
2213                 ixgbe_write_eitr(q_vector);
2214         }
2215
2216         switch (adapter->hw.mac.type) {
2217         case ixgbe_mac_82598EB:
2218                 ixgbe_set_ivar(adapter, -1, IXGBE_IVAR_OTHER_CAUSES_INDEX,
2219                                v_idx);
2220                 break;
2221         case ixgbe_mac_82599EB:
2222         case ixgbe_mac_X540:
2223         case ixgbe_mac_X550:
2224         case ixgbe_mac_X550EM_x:
2225                 ixgbe_set_ivar(adapter, -1, 1, v_idx);
2226                 break;
2227         default:
2228                 break;
2229         }
2230         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
2231
2232         /* set up to autoclear timer, and the vectors */
2233         mask = IXGBE_EIMS_ENABLE_MASK;
2234         mask &= ~(IXGBE_EIMS_OTHER |
2235                   IXGBE_EIMS_MAILBOX |
2236                   IXGBE_EIMS_LSC);
2237
2238         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
2239 }
2240
2241 enum latency_range {
2242         lowest_latency = 0,
2243         low_latency = 1,
2244         bulk_latency = 2,
2245         latency_invalid = 255
2246 };
2247
2248 /**
2249  * ixgbe_update_itr - update the dynamic ITR value based on statistics
2250  * @q_vector: structure containing interrupt and ring information
2251  * @ring_container: structure containing ring performance data
2252  *
2253  *      Stores a new ITR value based on packets and byte
2254  *      counts during the last interrupt.  The advantage of per interrupt
2255  *      computation is faster updates and more accurate ITR for the current
2256  *      traffic pattern.  Constants in this function were computed
2257  *      based on theoretical maximum wire speed and thresholds were set based
2258  *      on testing data as well as attempting to minimize response time
2259  *      while increasing bulk throughput.
2260  *      this functionality is controlled by the InterruptThrottleRate module
2261  *      parameter (see ixgbe_param.c)
2262  **/
2263 static void ixgbe_update_itr(struct ixgbe_q_vector *q_vector,
2264                              struct ixgbe_ring_container *ring_container)
2265 {
2266         int bytes = ring_container->total_bytes;
2267         int packets = ring_container->total_packets;
2268         u32 timepassed_us;
2269         u64 bytes_perint;
2270         u8 itr_setting = ring_container->itr;
2271
2272         if (packets == 0)
2273                 return;
2274
2275         /* simple throttlerate management
2276          *   0-10MB/s   lowest (100000 ints/s)
2277          *  10-20MB/s   low    (20000 ints/s)
2278          *  20-1249MB/s bulk   (12000 ints/s)
2279          */
2280         /* what was last interrupt timeslice? */
2281         timepassed_us = q_vector->itr >> 2;
2282         if (timepassed_us == 0)
2283                 return;
2284
2285         bytes_perint = bytes / timepassed_us; /* bytes/usec */
2286
2287         switch (itr_setting) {
2288         case lowest_latency:
2289                 if (bytes_perint > 10)
2290                         itr_setting = low_latency;
2291                 break;
2292         case low_latency:
2293                 if (bytes_perint > 20)
2294                         itr_setting = bulk_latency;
2295                 else if (bytes_perint <= 10)
2296                         itr_setting = lowest_latency;
2297                 break;
2298         case bulk_latency:
2299                 if (bytes_perint <= 20)
2300                         itr_setting = low_latency;
2301                 break;
2302         }
2303
2304         /* clear work counters since we have the values we need */
2305         ring_container->total_bytes = 0;
2306         ring_container->total_packets = 0;
2307
2308         /* write updated itr to ring container */
2309         ring_container->itr = itr_setting;
2310 }
2311
2312 /**
2313  * ixgbe_write_eitr - write EITR register in hardware specific way
2314  * @q_vector: structure containing interrupt and ring information
2315  *
2316  * This function is made to be called by ethtool and by the driver
2317  * when it needs to update EITR registers at runtime.  Hardware
2318  * specific quirks/differences are taken care of here.
2319  */
2320 void ixgbe_write_eitr(struct ixgbe_q_vector *q_vector)
2321 {
2322         struct ixgbe_adapter *adapter = q_vector->adapter;
2323         struct ixgbe_hw *hw = &adapter->hw;
2324         int v_idx = q_vector->v_idx;
2325         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
2326
2327         switch (adapter->hw.mac.type) {
2328         case ixgbe_mac_82598EB:
2329                 /* must write high and low 16 bits to reset counter */
2330                 itr_reg |= (itr_reg << 16);
2331                 break;
2332         case ixgbe_mac_82599EB:
2333         case ixgbe_mac_X540:
2334         case ixgbe_mac_X550:
2335         case ixgbe_mac_X550EM_x:
2336                 /*
2337                  * set the WDIS bit to not clear the timer bits and cause an
2338                  * immediate assertion of the interrupt
2339                  */
2340                 itr_reg |= IXGBE_EITR_CNT_WDIS;
2341                 break;
2342         default:
2343                 break;
2344         }
2345         IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg);
2346 }
2347
2348 static void ixgbe_set_itr(struct ixgbe_q_vector *q_vector)
2349 {
2350         u32 new_itr = q_vector->itr;
2351         u8 current_itr;
2352
2353         ixgbe_update_itr(q_vector, &q_vector->tx);
2354         ixgbe_update_itr(q_vector, &q_vector->rx);
2355
2356         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
2357
2358         switch (current_itr) {
2359         /* counts and packets in update_itr are dependent on these numbers */
2360         case lowest_latency:
2361                 new_itr = IXGBE_100K_ITR;
2362                 break;
2363         case low_latency:
2364                 new_itr = IXGBE_20K_ITR;
2365                 break;
2366         case bulk_latency:
2367                 new_itr = IXGBE_12K_ITR;
2368                 break;
2369         default:
2370                 break;
2371         }
2372
2373         if (new_itr != q_vector->itr) {
2374                 /* do an exponential smoothing */
2375                 new_itr = (10 * new_itr * q_vector->itr) /
2376                           ((9 * new_itr) + q_vector->itr);
2377
2378                 /* save the algorithm value here */
2379                 q_vector->itr = new_itr;
2380
2381                 ixgbe_write_eitr(q_vector);
2382         }
2383 }
2384
2385 /**
2386  * ixgbe_check_overtemp_subtask - check for over temperature
2387  * @adapter: pointer to adapter
2388  **/
2389 static void ixgbe_check_overtemp_subtask(struct ixgbe_adapter *adapter)
2390 {
2391         struct ixgbe_hw *hw = &adapter->hw;
2392         u32 eicr = adapter->interrupt_event;
2393
2394         if (test_bit(__IXGBE_DOWN, &adapter->state))
2395                 return;
2396
2397         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) &&
2398             !(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_EVENT))
2399                 return;
2400
2401         adapter->flags2 &= ~IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2402
2403         switch (hw->device_id) {
2404         case IXGBE_DEV_ID_82599_T3_LOM:
2405                 /*
2406                  * Since the warning interrupt is for both ports
2407                  * we don't have to check if:
2408                  *  - This interrupt wasn't for our port.
2409                  *  - We may have missed the interrupt so always have to
2410                  *    check if we  got a LSC
2411                  */
2412                 if (!(eicr & IXGBE_EICR_GPI_SDP0_8259X) &&
2413                     !(eicr & IXGBE_EICR_LSC))
2414                         return;
2415
2416                 if (!(eicr & IXGBE_EICR_LSC) && hw->mac.ops.check_link) {
2417                         u32 speed;
2418                         bool link_up = false;
2419
2420                         hw->mac.ops.check_link(hw, &speed, &link_up, false);
2421
2422                         if (link_up)
2423                                 return;
2424                 }
2425
2426                 /* Check if this is not due to overtemp */
2427                 if (hw->phy.ops.check_overtemp(hw) != IXGBE_ERR_OVERTEMP)
2428                         return;
2429
2430                 break;
2431         default:
2432                 if (adapter->hw.mac.type >= ixgbe_mac_X540)
2433                         return;
2434                 if (!(eicr & IXGBE_EICR_GPI_SDP0(hw)))
2435                         return;
2436                 break;
2437         }
2438         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2439
2440         adapter->interrupt_event = 0;
2441 }
2442
2443 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
2444 {
2445         struct ixgbe_hw *hw = &adapter->hw;
2446
2447         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
2448             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2449                 e_crit(probe, "Fan has stopped, replace the adapter\n");
2450                 /* write to clear the interrupt */
2451                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2452         }
2453 }
2454
2455 static void ixgbe_check_overtemp_event(struct ixgbe_adapter *adapter, u32 eicr)
2456 {
2457         struct ixgbe_hw *hw = &adapter->hw;
2458
2459         if (!(adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE))
2460                 return;
2461
2462         switch (adapter->hw.mac.type) {
2463         case ixgbe_mac_82599EB:
2464                 /*
2465                  * Need to check link state so complete overtemp check
2466                  * on service task
2467                  */
2468                 if (((eicr & IXGBE_EICR_GPI_SDP0(hw)) ||
2469                      (eicr & IXGBE_EICR_LSC)) &&
2470                     (!test_bit(__IXGBE_DOWN, &adapter->state))) {
2471                         adapter->interrupt_event = eicr;
2472                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_EVENT;
2473                         ixgbe_service_event_schedule(adapter);
2474                         return;
2475                 }
2476                 return;
2477         case ixgbe_mac_X540:
2478                 if (!(eicr & IXGBE_EICR_TS))
2479                         return;
2480                 break;
2481         default:
2482                 return;
2483         }
2484
2485         e_crit(drv, "%s\n", ixgbe_overheat_msg);
2486 }
2487
2488 static inline bool ixgbe_is_sfp(struct ixgbe_hw *hw)
2489 {
2490         switch (hw->mac.type) {
2491         case ixgbe_mac_82598EB:
2492                 if (hw->phy.type == ixgbe_phy_nl)
2493                         return true;
2494                 return false;
2495         case ixgbe_mac_82599EB:
2496         case ixgbe_mac_X550EM_x:
2497                 switch (hw->mac.ops.get_media_type(hw)) {
2498                 case ixgbe_media_type_fiber:
2499                 case ixgbe_media_type_fiber_qsfp:
2500                         return true;
2501                 default:
2502                         return false;
2503                 }
2504         default:
2505                 return false;
2506         }
2507 }
2508
2509 static void ixgbe_check_sfp_event(struct ixgbe_adapter *adapter, u32 eicr)
2510 {
2511         struct ixgbe_hw *hw = &adapter->hw;
2512         u32 eicr_mask = IXGBE_EICR_GPI_SDP2(hw);
2513
2514         if (!ixgbe_is_sfp(hw))
2515                 return;
2516
2517         /* Later MAC's use different SDP */
2518         if (hw->mac.type >= ixgbe_mac_X540)
2519                 eicr_mask = IXGBE_EICR_GPI_SDP0_X540;
2520
2521         if (eicr & eicr_mask) {
2522                 /* Clear the interrupt */
2523                 IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr_mask);
2524                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2525                         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
2526                         adapter->sfp_poll_time = 0;
2527                         ixgbe_service_event_schedule(adapter);
2528                 }
2529         }
2530
2531         if (adapter->hw.mac.type == ixgbe_mac_82599EB &&
2532             (eicr & IXGBE_EICR_GPI_SDP1(hw))) {
2533                 /* Clear the interrupt */
2534                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1(hw));
2535                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2536                         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
2537                         ixgbe_service_event_schedule(adapter);
2538                 }
2539         }
2540 }
2541
2542 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
2543 {
2544         struct ixgbe_hw *hw = &adapter->hw;
2545
2546         adapter->lsc_int++;
2547         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2548         adapter->link_check_timeout = jiffies;
2549         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
2550                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
2551                 IXGBE_WRITE_FLUSH(hw);
2552                 ixgbe_service_event_schedule(adapter);
2553         }
2554 }
2555
2556 static inline void ixgbe_irq_enable_queues(struct ixgbe_adapter *adapter,
2557                                            u64 qmask)
2558 {
2559         u32 mask;
2560         struct ixgbe_hw *hw = &adapter->hw;
2561
2562         switch (hw->mac.type) {
2563         case ixgbe_mac_82598EB:
2564                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2565                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, mask);
2566                 break;
2567         case ixgbe_mac_82599EB:
2568         case ixgbe_mac_X540:
2569         case ixgbe_mac_X550:
2570         case ixgbe_mac_X550EM_x:
2571                 mask = (qmask & 0xFFFFFFFF);
2572                 if (mask)
2573                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(0), mask);
2574                 mask = (qmask >> 32);
2575                 if (mask)
2576                         IXGBE_WRITE_REG(hw, IXGBE_EIMS_EX(1), mask);
2577                 break;
2578         default:
2579                 break;
2580         }
2581         /* skip the flush */
2582 }
2583
2584 static inline void ixgbe_irq_disable_queues(struct ixgbe_adapter *adapter,
2585                                             u64 qmask)
2586 {
2587         u32 mask;
2588         struct ixgbe_hw *hw = &adapter->hw;
2589
2590         switch (hw->mac.type) {
2591         case ixgbe_mac_82598EB:
2592                 mask = (IXGBE_EIMS_RTX_QUEUE & qmask);
2593                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, mask);
2594                 break;
2595         case ixgbe_mac_82599EB:
2596         case ixgbe_mac_X540:
2597         case ixgbe_mac_X550:
2598         case ixgbe_mac_X550EM_x:
2599                 mask = (qmask & 0xFFFFFFFF);
2600                 if (mask)
2601                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(0), mask);
2602                 mask = (qmask >> 32);
2603                 if (mask)
2604                         IXGBE_WRITE_REG(hw, IXGBE_EIMC_EX(1), mask);
2605                 break;
2606         default:
2607                 break;
2608         }
2609         /* skip the flush */
2610 }
2611
2612 /**
2613  * ixgbe_irq_enable - Enable default interrupt generation settings
2614  * @adapter: board private structure
2615  **/
2616 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter, bool queues,
2617                                     bool flush)
2618 {
2619         struct ixgbe_hw *hw = &adapter->hw;
2620         u32 mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
2621
2622         /* don't reenable LSC while waiting for link */
2623         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
2624                 mask &= ~IXGBE_EIMS_LSC;
2625
2626         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE)
2627                 switch (adapter->hw.mac.type) {
2628                 case ixgbe_mac_82599EB:
2629                         mask |= IXGBE_EIMS_GPI_SDP0(hw);
2630                         break;
2631                 case ixgbe_mac_X540:
2632                 case ixgbe_mac_X550:
2633                 case ixgbe_mac_X550EM_x:
2634                         mask |= IXGBE_EIMS_TS;
2635                         break;
2636                 default:
2637                         break;
2638                 }
2639         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
2640                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2641         switch (adapter->hw.mac.type) {
2642         case ixgbe_mac_82599EB:
2643                 mask |= IXGBE_EIMS_GPI_SDP1(hw);
2644                 mask |= IXGBE_EIMS_GPI_SDP2(hw);
2645                 /* fall through */
2646         case ixgbe_mac_X540:
2647         case ixgbe_mac_X550:
2648         case ixgbe_mac_X550EM_x:
2649                 if (adapter->hw.device_id == IXGBE_DEV_ID_X550EM_X_SFP)
2650                         mask |= IXGBE_EIMS_GPI_SDP0(&adapter->hw);
2651                 if (adapter->hw.phy.type == ixgbe_phy_x550em_ext_t)
2652                         mask |= IXGBE_EICR_GPI_SDP0_X540;
2653                 mask |= IXGBE_EIMS_ECC;
2654                 mask |= IXGBE_EIMS_MAILBOX;
2655                 break;
2656         default:
2657                 break;
2658         }
2659
2660         if ((adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) &&
2661             !(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
2662                 mask |= IXGBE_EIMS_FLOW_DIR;
2663
2664         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
2665         if (queues)
2666                 ixgbe_irq_enable_queues(adapter, ~0);
2667         if (flush)
2668                 IXGBE_WRITE_FLUSH(&adapter->hw);
2669 }
2670
2671 static irqreturn_t ixgbe_msix_other(int irq, void *data)
2672 {
2673         struct ixgbe_adapter *adapter = data;
2674         struct ixgbe_hw *hw = &adapter->hw;
2675         u32 eicr;
2676
2677         /*
2678          * Workaround for Silicon errata.  Use clear-by-write instead
2679          * of clear-by-read.  Reading with EICS will return the
2680          * interrupt causes without clearing, which later be done
2681          * with the write to EICR.
2682          */
2683         eicr = IXGBE_READ_REG(hw, IXGBE_EICS);
2684
2685         /* The lower 16bits of the EICR register are for the queue interrupts
2686          * which should be masked here in order to not accidentally clear them if
2687          * the bits are high when ixgbe_msix_other is called. There is a race
2688          * condition otherwise which results in possible performance loss
2689          * especially if the ixgbe_msix_other interrupt is triggering
2690          * consistently (as it would when PPS is turned on for the X540 device)
2691          */
2692         eicr &= 0xFFFF0000;
2693
2694         IXGBE_WRITE_REG(hw, IXGBE_EICR, eicr);
2695
2696         if (eicr & IXGBE_EICR_LSC)
2697                 ixgbe_check_lsc(adapter);
2698
2699         if (eicr & IXGBE_EICR_MAILBOX)
2700                 ixgbe_msg_task(adapter);
2701
2702         switch (hw->mac.type) {
2703         case ixgbe_mac_82599EB:
2704         case ixgbe_mac_X540:
2705         case ixgbe_mac_X550:
2706         case ixgbe_mac_X550EM_x:
2707                 if (hw->phy.type == ixgbe_phy_x550em_ext_t &&
2708                     (eicr & IXGBE_EICR_GPI_SDP0_X540)) {
2709                         adapter->flags2 |= IXGBE_FLAG2_PHY_INTERRUPT;
2710                         ixgbe_service_event_schedule(adapter);
2711                         IXGBE_WRITE_REG(hw, IXGBE_EICR,
2712                                         IXGBE_EICR_GPI_SDP0_X540);
2713                 }
2714                 if (eicr & IXGBE_EICR_ECC) {
2715                         e_info(link, "Received ECC Err, initiating reset\n");
2716                         adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2717                         ixgbe_service_event_schedule(adapter);
2718                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2719                 }
2720                 /* Handle Flow Director Full threshold interrupt */
2721                 if (eicr & IXGBE_EICR_FLOW_DIR) {
2722                         int reinit_count = 0;
2723                         int i;
2724                         for (i = 0; i < adapter->num_tx_queues; i++) {
2725                                 struct ixgbe_ring *ring = adapter->tx_ring[i];
2726                                 if (test_and_clear_bit(__IXGBE_TX_FDIR_INIT_DONE,
2727                                                        &ring->state))
2728                                         reinit_count++;
2729                         }
2730                         if (reinit_count) {
2731                                 /* no more flow director interrupts until after init */
2732                                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_FLOW_DIR);
2733                                 adapter->flags2 |= IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
2734                                 ixgbe_service_event_schedule(adapter);
2735                         }
2736                 }
2737                 ixgbe_check_sfp_event(adapter, eicr);
2738                 ixgbe_check_overtemp_event(adapter, eicr);
2739                 break;
2740         default:
2741                 break;
2742         }
2743
2744         ixgbe_check_fan_failure(adapter, eicr);
2745
2746         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2747                 ixgbe_ptp_check_pps_event(adapter);
2748
2749         /* re-enable the original interrupt state, no lsc, no queues */
2750         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2751                 ixgbe_irq_enable(adapter, false, false);
2752
2753         return IRQ_HANDLED;
2754 }
2755
2756 static irqreturn_t ixgbe_msix_clean_rings(int irq, void *data)
2757 {
2758         struct ixgbe_q_vector *q_vector = data;
2759
2760         /* EIAM disabled interrupts (on this vector) for us */
2761
2762         if (q_vector->rx.ring || q_vector->tx.ring)
2763                 napi_schedule_irqoff(&q_vector->napi);
2764
2765         return IRQ_HANDLED;
2766 }
2767
2768 /**
2769  * ixgbe_poll - NAPI Rx polling callback
2770  * @napi: structure for representing this polling device
2771  * @budget: how many packets driver is allowed to clean
2772  *
2773  * This function is used for legacy and MSI, NAPI mode
2774  **/
2775 int ixgbe_poll(struct napi_struct *napi, int budget)
2776 {
2777         struct ixgbe_q_vector *q_vector =
2778                                 container_of(napi, struct ixgbe_q_vector, napi);
2779         struct ixgbe_adapter *adapter = q_vector->adapter;
2780         struct ixgbe_ring *ring;
2781         int per_ring_budget, work_done = 0;
2782         bool clean_complete = true;
2783
2784 #ifdef CONFIG_IXGBE_DCA
2785         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
2786                 ixgbe_update_dca(q_vector);
2787 #endif
2788
2789         ixgbe_for_each_ring(ring, q_vector->tx)
2790                 clean_complete &= !!ixgbe_clean_tx_irq(q_vector, ring, budget);
2791
2792         /* Exit if we are called by netpoll or busy polling is active */
2793         if ((budget <= 0) || !ixgbe_qv_lock_napi(q_vector))
2794                 return budget;
2795
2796         /* attempt to distribute budget to each queue fairly, but don't allow
2797          * the budget to go below 1 because we'll exit polling */
2798         if (q_vector->rx.count > 1)
2799                 per_ring_budget = max(budget/q_vector->rx.count, 1);
2800         else
2801                 per_ring_budget = budget;
2802
2803         ixgbe_for_each_ring(ring, q_vector->rx) {
2804                 int cleaned = ixgbe_clean_rx_irq(q_vector, ring,
2805                                                  per_ring_budget);
2806
2807                 work_done += cleaned;
2808                 clean_complete &= (cleaned < per_ring_budget);
2809         }
2810
2811         ixgbe_qv_unlock_napi(q_vector);
2812         /* If all work not completed, return budget and keep polling */
2813         if (!clean_complete)
2814                 return budget;
2815
2816         /* all work done, exit the polling mode */
2817         napi_complete_done(napi, work_done);
2818         if (adapter->rx_itr_setting & 1)
2819                 ixgbe_set_itr(q_vector);
2820         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2821                 ixgbe_irq_enable_queues(adapter, ((u64)1 << q_vector->v_idx));
2822
2823         return 0;
2824 }
2825
2826 /**
2827  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
2828  * @adapter: board private structure
2829  *
2830  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
2831  * interrupts from the kernel.
2832  **/
2833 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
2834 {
2835         struct net_device *netdev = adapter->netdev;
2836         int vector, err;
2837         int ri = 0, ti = 0;
2838
2839         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
2840                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
2841                 struct msix_entry *entry = &adapter->msix_entries[vector];
2842
2843                 if (q_vector->tx.ring && q_vector->rx.ring) {
2844                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2845                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
2846                         ti++;
2847                 } else if (q_vector->rx.ring) {
2848                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2849                                  "%s-%s-%d", netdev->name, "rx", ri++);
2850                 } else if (q_vector->tx.ring) {
2851                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
2852                                  "%s-%s-%d", netdev->name, "tx", ti++);
2853                 } else {
2854                         /* skip this unused q_vector */
2855                         continue;
2856                 }
2857                 err = request_irq(entry->vector, &ixgbe_msix_clean_rings, 0,
2858                                   q_vector->name, q_vector);
2859                 if (err) {
2860                         e_err(probe, "request_irq failed for MSIX interrupt "
2861                               "Error: %d\n", err);
2862                         goto free_queue_irqs;
2863                 }
2864                 /* If Flow Director is enabled, set interrupt affinity */
2865                 if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
2866                         /* assign the mask for this irq */
2867                         irq_set_affinity_hint(entry->vector,
2868                                               &q_vector->affinity_mask);
2869                 }
2870         }
2871
2872         err = request_irq(adapter->msix_entries[vector].vector,
2873                           ixgbe_msix_other, 0, netdev->name, adapter);
2874         if (err) {
2875                 e_err(probe, "request_irq for msix_other failed: %d\n", err);
2876                 goto free_queue_irqs;
2877         }
2878
2879         return 0;
2880
2881 free_queue_irqs:
2882         while (vector) {
2883                 vector--;
2884                 irq_set_affinity_hint(adapter->msix_entries[vector].vector,
2885                                       NULL);
2886                 free_irq(adapter->msix_entries[vector].vector,
2887                          adapter->q_vector[vector]);
2888         }
2889         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2890         pci_disable_msix(adapter->pdev);
2891         kfree(adapter->msix_entries);
2892         adapter->msix_entries = NULL;
2893         return err;
2894 }
2895
2896 /**
2897  * ixgbe_intr - legacy mode Interrupt Handler
2898  * @irq: interrupt number
2899  * @data: pointer to a network interface device structure
2900  **/
2901 static irqreturn_t ixgbe_intr(int irq, void *data)
2902 {
2903         struct ixgbe_adapter *adapter = data;
2904         struct ixgbe_hw *hw = &adapter->hw;
2905         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
2906         u32 eicr;
2907
2908         /*
2909          * Workaround for silicon errata #26 on 82598.  Mask the interrupt
2910          * before the read of EICR.
2911          */
2912         IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_IRQ_CLEAR_MASK);
2913
2914         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
2915          * therefore no explicit interrupt disable is necessary */
2916         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
2917         if (!eicr) {
2918                 /*
2919                  * shared interrupt alert!
2920                  * make sure interrupts are enabled because the read will
2921                  * have disabled interrupts due to EIAM
2922                  * finish the workaround of silicon errata on 82598.  Unmask
2923                  * the interrupt that we masked before the EICR read.
2924                  */
2925                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2926                         ixgbe_irq_enable(adapter, true, true);
2927                 return IRQ_NONE;        /* Not our interrupt */
2928         }
2929
2930         if (eicr & IXGBE_EICR_LSC)
2931                 ixgbe_check_lsc(adapter);
2932
2933         switch (hw->mac.type) {
2934         case ixgbe_mac_82599EB:
2935                 ixgbe_check_sfp_event(adapter, eicr);
2936                 /* Fall through */
2937         case ixgbe_mac_X540:
2938         case ixgbe_mac_X550:
2939         case ixgbe_mac_X550EM_x:
2940                 if (eicr & IXGBE_EICR_ECC) {
2941                         e_info(link, "Received ECC Err, initiating reset\n");
2942                         adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
2943                         ixgbe_service_event_schedule(adapter);
2944                         IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_ECC);
2945                 }
2946                 ixgbe_check_overtemp_event(adapter, eicr);
2947                 break;
2948         default:
2949                 break;
2950         }
2951
2952         ixgbe_check_fan_failure(adapter, eicr);
2953         if (unlikely(eicr & IXGBE_EICR_TIMESYNC))
2954                 ixgbe_ptp_check_pps_event(adapter);
2955
2956         /* would disable interrupts here but EIAM disabled it */
2957         napi_schedule_irqoff(&q_vector->napi);
2958
2959         /*
2960          * re-enable link(maybe) and non-queue interrupts, no flush.
2961          * ixgbe_poll will re-enable the queue interrupts
2962          */
2963         if (!test_bit(__IXGBE_DOWN, &adapter->state))
2964                 ixgbe_irq_enable(adapter, false, false);
2965
2966         return IRQ_HANDLED;
2967 }
2968
2969 /**
2970  * ixgbe_request_irq - initialize interrupts
2971  * @adapter: board private structure
2972  *
2973  * Attempts to configure interrupts using the best available
2974  * capabilities of the hardware and kernel.
2975  **/
2976 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
2977 {
2978         struct net_device *netdev = adapter->netdev;
2979         int err;
2980
2981         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2982                 err = ixgbe_request_msix_irqs(adapter);
2983         else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED)
2984                 err = request_irq(adapter->pdev->irq, ixgbe_intr, 0,
2985                                   netdev->name, adapter);
2986         else
2987                 err = request_irq(adapter->pdev->irq, ixgbe_intr, IRQF_SHARED,
2988                                   netdev->name, adapter);
2989
2990         if (err)
2991                 e_err(probe, "request_irq failed, Error %d\n", err);
2992
2993         return err;
2994 }
2995
2996 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
2997 {
2998         int vector;
2999
3000         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
3001                 free_irq(adapter->pdev->irq, adapter);
3002                 return;
3003         }
3004
3005         for (vector = 0; vector < adapter->num_q_vectors; vector++) {
3006                 struct ixgbe_q_vector *q_vector = adapter->q_vector[vector];
3007                 struct msix_entry *entry = &adapter->msix_entries[vector];
3008
3009                 /* free only the irqs that were actually requested */
3010                 if (!q_vector->rx.ring && !q_vector->tx.ring)
3011                         continue;
3012
3013                 /* clear the affinity_mask in the IRQ descriptor */
3014                 irq_set_affinity_hint(entry->vector, NULL);
3015
3016                 free_irq(entry->vector, q_vector);
3017         }
3018
3019         free_irq(adapter->msix_entries[vector++].vector, adapter);
3020 }
3021
3022 /**
3023  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
3024  * @adapter: board private structure
3025  **/
3026 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
3027 {
3028         switch (adapter->hw.mac.type) {
3029         case ixgbe_mac_82598EB:
3030                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
3031                 break;
3032         case ixgbe_mac_82599EB:
3033         case ixgbe_mac_X540:
3034         case ixgbe_mac_X550:
3035         case ixgbe_mac_X550EM_x:
3036                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, 0xFFFF0000);
3037                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(0), ~0);
3038                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC_EX(1), ~0);
3039                 break;
3040         default:
3041                 break;
3042         }
3043         IXGBE_WRITE_FLUSH(&adapter->hw);
3044         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3045                 int vector;
3046
3047                 for (vector = 0; vector < adapter->num_q_vectors; vector++)
3048                         synchronize_irq(adapter->msix_entries[vector].vector);
3049
3050                 synchronize_irq(adapter->msix_entries[vector++].vector);
3051         } else {
3052                 synchronize_irq(adapter->pdev->irq);
3053         }
3054 }
3055
3056 /**
3057  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
3058  *
3059  **/
3060 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
3061 {
3062         struct ixgbe_q_vector *q_vector = adapter->q_vector[0];
3063
3064         ixgbe_write_eitr(q_vector);
3065
3066         ixgbe_set_ivar(adapter, 0, 0, 0);
3067         ixgbe_set_ivar(adapter, 1, 0, 0);
3068
3069         e_info(hw, "Legacy interrupt IVAR setup done\n");
3070 }
3071
3072 /**
3073  * ixgbe_configure_tx_ring - Configure 8259x Tx ring after Reset
3074  * @adapter: board private structure
3075  * @ring: structure containing ring specific data
3076  *
3077  * Configure the Tx descriptor ring after a reset.
3078  **/
3079 void ixgbe_configure_tx_ring(struct ixgbe_adapter *adapter,
3080                              struct ixgbe_ring *ring)
3081 {
3082         struct ixgbe_hw *hw = &adapter->hw;
3083         u64 tdba = ring->dma;
3084         int wait_loop = 10;
3085         u32 txdctl = IXGBE_TXDCTL_ENABLE;
3086         u8 reg_idx = ring->reg_idx;
3087
3088         /* disable queue to avoid issues while updating state */
3089         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), 0);
3090         IXGBE_WRITE_FLUSH(hw);
3091
3092         IXGBE_WRITE_REG(hw, IXGBE_TDBAL(reg_idx),
3093                         (tdba & DMA_BIT_MASK(32)));
3094         IXGBE_WRITE_REG(hw, IXGBE_TDBAH(reg_idx), (tdba >> 32));
3095         IXGBE_WRITE_REG(hw, IXGBE_TDLEN(reg_idx),
3096                         ring->count * sizeof(union ixgbe_adv_tx_desc));
3097         IXGBE_WRITE_REG(hw, IXGBE_TDH(reg_idx), 0);
3098         IXGBE_WRITE_REG(hw, IXGBE_TDT(reg_idx), 0);
3099         ring->tail = adapter->io_addr + IXGBE_TDT(reg_idx);
3100
3101         /*
3102          * set WTHRESH to encourage burst writeback, it should not be set
3103          * higher than 1 when:
3104          * - ITR is 0 as it could cause false TX hangs
3105          * - ITR is set to > 100k int/sec and BQL is enabled
3106          *
3107          * In order to avoid issues WTHRESH + PTHRESH should always be equal
3108          * to or less than the number of on chip descriptors, which is
3109          * currently 40.
3110          */
3111         if (!ring->q_vector || (ring->q_vector->itr < IXGBE_100K_ITR))
3112                 txdctl |= (1 << 16);    /* WTHRESH = 1 */
3113         else
3114                 txdctl |= (8 << 16);    /* WTHRESH = 8 */
3115
3116         /*
3117          * Setting PTHRESH to 32 both improves performance
3118          * and avoids a TX hang with DFP enabled
3119          */
3120         txdctl |= (1 << 8) |    /* HTHRESH = 1 */
3121                    32;          /* PTHRESH = 32 */
3122
3123         /* reinitialize flowdirector state */
3124         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
3125                 ring->atr_sample_rate = adapter->atr_sample_rate;
3126                 ring->atr_count = 0;
3127                 set_bit(__IXGBE_TX_FDIR_INIT_DONE, &ring->state);
3128         } else {
3129                 ring->atr_sample_rate = 0;
3130         }
3131
3132         /* initialize XPS */
3133         if (!test_and_set_bit(__IXGBE_TX_XPS_INIT_DONE, &ring->state)) {
3134                 struct ixgbe_q_vector *q_vector = ring->q_vector;
3135
3136                 if (q_vector)
3137                         netif_set_xps_queue(ring->netdev,
3138                                             &q_vector->affinity_mask,
3139                                             ring->queue_index);
3140         }
3141
3142         clear_bit(__IXGBE_HANG_CHECK_ARMED, &ring->state);
3143
3144         /* enable queue */
3145         IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), txdctl);
3146
3147         /* TXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3148         if (hw->mac.type == ixgbe_mac_82598EB &&
3149             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3150                 return;
3151
3152         /* poll to verify queue is enabled */
3153         do {
3154                 usleep_range(1000, 2000);
3155                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(reg_idx));
3156         } while (--wait_loop && !(txdctl & IXGBE_TXDCTL_ENABLE));
3157         if (!wait_loop)
3158                 e_err(drv, "Could not enable Tx Queue %d\n", reg_idx);
3159 }
3160
3161 static void ixgbe_setup_mtqc(struct ixgbe_adapter *adapter)
3162 {
3163         struct ixgbe_hw *hw = &adapter->hw;
3164         u32 rttdcs, mtqc;
3165         u8 tcs = netdev_get_num_tc(adapter->netdev);
3166
3167         if (hw->mac.type == ixgbe_mac_82598EB)
3168                 return;
3169
3170         /* disable the arbiter while setting MTQC */
3171         rttdcs = IXGBE_READ_REG(hw, IXGBE_RTTDCS);
3172         rttdcs |= IXGBE_RTTDCS_ARBDIS;
3173         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3174
3175         /* set transmit pool layout */
3176         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3177                 mtqc = IXGBE_MTQC_VT_ENA;
3178                 if (tcs > 4)
3179                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3180                 else if (tcs > 1)
3181                         mtqc |= IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3182                 else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3183                         mtqc |= IXGBE_MTQC_32VF;
3184                 else
3185                         mtqc |= IXGBE_MTQC_64VF;
3186         } else {
3187                 if (tcs > 4)
3188                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_8TC_8TQ;
3189                 else if (tcs > 1)
3190                         mtqc = IXGBE_MTQC_RT_ENA | IXGBE_MTQC_4TC_4TQ;
3191                 else
3192                         mtqc = IXGBE_MTQC_64Q_1PB;
3193         }
3194
3195         IXGBE_WRITE_REG(hw, IXGBE_MTQC, mtqc);
3196
3197         /* Enable Security TX Buffer IFG for multiple pb */
3198         if (tcs) {
3199                 u32 sectx = IXGBE_READ_REG(hw, IXGBE_SECTXMINIFG);
3200                 sectx |= IXGBE_SECTX_DCB;
3201                 IXGBE_WRITE_REG(hw, IXGBE_SECTXMINIFG, sectx);
3202         }
3203
3204         /* re-enable the arbiter */
3205         rttdcs &= ~IXGBE_RTTDCS_ARBDIS;
3206         IXGBE_WRITE_REG(hw, IXGBE_RTTDCS, rttdcs);
3207 }
3208
3209 /**
3210  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
3211  * @adapter: board private structure
3212  *
3213  * Configure the Tx unit of the MAC after a reset.
3214  **/
3215 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
3216 {
3217         struct ixgbe_hw *hw = &adapter->hw;
3218         u32 dmatxctl;
3219         u32 i;
3220
3221         ixgbe_setup_mtqc(adapter);
3222
3223         if (hw->mac.type != ixgbe_mac_82598EB) {
3224                 /* DMATXCTL.EN must be before Tx queues are enabled */
3225                 dmatxctl = IXGBE_READ_REG(hw, IXGBE_DMATXCTL);
3226                 dmatxctl |= IXGBE_DMATXCTL_TE;
3227                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL, dmatxctl);
3228         }
3229
3230         /* Setup the HW Tx Head and Tail descriptor pointers */
3231         for (i = 0; i < adapter->num_tx_queues; i++)
3232                 ixgbe_configure_tx_ring(adapter, adapter->tx_ring[i]);
3233 }
3234
3235 static void ixgbe_enable_rx_drop(struct ixgbe_adapter *adapter,
3236                                  struct ixgbe_ring *ring)
3237 {
3238         struct ixgbe_hw *hw = &adapter->hw;
3239         u8 reg_idx = ring->reg_idx;
3240         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3241
3242         srrctl |= IXGBE_SRRCTL_DROP_EN;
3243
3244         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3245 }
3246
3247 static void ixgbe_disable_rx_drop(struct ixgbe_adapter *adapter,
3248                                   struct ixgbe_ring *ring)
3249 {
3250         struct ixgbe_hw *hw = &adapter->hw;
3251         u8 reg_idx = ring->reg_idx;
3252         u32 srrctl = IXGBE_READ_REG(hw, IXGBE_SRRCTL(reg_idx));
3253
3254         srrctl &= ~IXGBE_SRRCTL_DROP_EN;
3255
3256         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3257 }
3258
3259 #ifdef CONFIG_IXGBE_DCB
3260 void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3261 #else
3262 static void ixgbe_set_rx_drop_en(struct ixgbe_adapter *adapter)
3263 #endif
3264 {
3265         int i;
3266         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
3267
3268         if (adapter->ixgbe_ieee_pfc)
3269                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
3270
3271         /*
3272          * We should set the drop enable bit if:
3273          *  SR-IOV is enabled
3274          *   or
3275          *  Number of Rx queues > 1 and flow control is disabled
3276          *
3277          *  This allows us to avoid head of line blocking for security
3278          *  and performance reasons.
3279          */
3280         if (adapter->num_vfs || (adapter->num_rx_queues > 1 &&
3281             !(adapter->hw.fc.current_mode & ixgbe_fc_tx_pause) && !pfc_en)) {
3282                 for (i = 0; i < adapter->num_rx_queues; i++)
3283                         ixgbe_enable_rx_drop(adapter, adapter->rx_ring[i]);
3284         } else {
3285                 for (i = 0; i < adapter->num_rx_queues; i++)
3286                         ixgbe_disable_rx_drop(adapter, adapter->rx_ring[i]);
3287         }
3288 }
3289
3290 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
3291
3292 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter,
3293                                    struct ixgbe_ring *rx_ring)
3294 {
3295         struct ixgbe_hw *hw = &adapter->hw;
3296         u32 srrctl;
3297         u8 reg_idx = rx_ring->reg_idx;
3298
3299         if (hw->mac.type == ixgbe_mac_82598EB) {
3300                 u16 mask = adapter->ring_feature[RING_F_RSS].mask;
3301
3302                 /*
3303                  * if VMDq is not active we must program one srrctl register
3304                  * per RSS queue since we have enabled RDRXCTL.MVMEN
3305                  */
3306                 reg_idx &= mask;
3307         }
3308
3309         /* configure header buffer length, needed for RSC */
3310         srrctl = IXGBE_RX_HDR_SIZE << IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT;
3311
3312         /* configure the packet buffer length */
3313         srrctl |= ixgbe_rx_bufsz(rx_ring) >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
3314
3315         /* configure descriptor type */
3316         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
3317
3318         IXGBE_WRITE_REG(hw, IXGBE_SRRCTL(reg_idx), srrctl);
3319 }
3320
3321 /**
3322  * ixgbe_rss_indir_tbl_entries - Return RSS indirection table entries
3323  * @adapter: device handle
3324  *
3325  *  - 82598/82599/X540:     128
3326  *  - X550(non-SRIOV mode): 512
3327  *  - X550(SRIOV mode):     64
3328  */
3329 u32 ixgbe_rss_indir_tbl_entries(struct ixgbe_adapter *adapter)
3330 {
3331         if (adapter->hw.mac.type < ixgbe_mac_X550)
3332                 return 128;
3333         else if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
3334                 return 64;
3335         else
3336                 return 512;
3337 }
3338
3339 /**
3340  * ixgbe_store_reta - Write the RETA table to HW
3341  * @adapter: device handle
3342  *
3343  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3344  */
3345 void ixgbe_store_reta(struct ixgbe_adapter *adapter)
3346 {
3347         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3348         struct ixgbe_hw *hw = &adapter->hw;
3349         u32 reta = 0;
3350         u32 indices_multi;
3351         u8 *indir_tbl = adapter->rss_indir_tbl;
3352
3353         /* Fill out the redirection table as follows:
3354          *  - 82598:      8 bit wide entries containing pair of 4 bit RSS
3355          *    indices.
3356          *  - 82599/X540: 8 bit wide entries containing 4 bit RSS index
3357          *  - X550:       8 bit wide entries containing 6 bit RSS index
3358          */
3359         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
3360                 indices_multi = 0x11;
3361         else
3362                 indices_multi = 0x1;
3363
3364         /* Write redirection table to HW */
3365         for (i = 0; i < reta_entries; i++) {
3366                 reta |= indices_multi * indir_tbl[i] << (i & 0x3) * 8;
3367                 if ((i & 3) == 3) {
3368                         if (i < 128)
3369                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
3370                         else
3371                                 IXGBE_WRITE_REG(hw, IXGBE_ERETA((i >> 2) - 32),
3372                                                 reta);
3373                         reta = 0;
3374                 }
3375         }
3376 }
3377
3378 /**
3379  * ixgbe_store_vfreta - Write the RETA table to HW (x550 devices in SRIOV mode)
3380  * @adapter: device handle
3381  *
3382  * Write the RSS redirection table stored in adapter.rss_indir_tbl[] to HW.
3383  */
3384 static void ixgbe_store_vfreta(struct ixgbe_adapter *adapter)
3385 {
3386         u32 i, reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3387         struct ixgbe_hw *hw = &adapter->hw;
3388         u32 vfreta = 0;
3389         unsigned int pf_pool = adapter->num_vfs;
3390
3391         /* Write redirection table to HW */
3392         for (i = 0; i < reta_entries; i++) {
3393                 vfreta |= (u32)adapter->rss_indir_tbl[i] << (i & 0x3) * 8;
3394                 if ((i & 3) == 3) {
3395                         IXGBE_WRITE_REG(hw, IXGBE_PFVFRETA(i >> 2, pf_pool),
3396                                         vfreta);
3397                         vfreta = 0;
3398                 }
3399         }
3400 }
3401
3402 static void ixgbe_setup_reta(struct ixgbe_adapter *adapter)
3403 {
3404         struct ixgbe_hw *hw = &adapter->hw;
3405         u32 i, j;
3406         u32 reta_entries = ixgbe_rss_indir_tbl_entries(adapter);
3407         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3408
3409         /* Program table for at least 2 queues w/ SR-IOV so that VFs can
3410          * make full use of any rings they may have.  We will use the
3411          * PSRTYPE register to control how many rings we use within the PF.
3412          */
3413         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) && (rss_i < 2))
3414                 rss_i = 2;
3415
3416         /* Fill out hash function seeds */
3417         for (i = 0; i < 10; i++)
3418                 IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), adapter->rss_key[i]);
3419
3420         /* Fill out redirection table */
3421         memset(adapter->rss_indir_tbl, 0, sizeof(adapter->rss_indir_tbl));
3422
3423         for (i = 0, j = 0; i < reta_entries; i++, j++) {
3424                 if (j == rss_i)
3425                         j = 0;
3426
3427                 adapter->rss_indir_tbl[i] = j;
3428         }
3429
3430         ixgbe_store_reta(adapter);
3431 }
3432
3433 static void ixgbe_setup_vfreta(struct ixgbe_adapter *adapter)
3434 {
3435         struct ixgbe_hw *hw = &adapter->hw;
3436         u16 rss_i = adapter->ring_feature[RING_F_RSS].indices;
3437         unsigned int pf_pool = adapter->num_vfs;
3438         int i, j;
3439
3440         /* Fill out hash function seeds */
3441         for (i = 0; i < 10; i++)
3442                 IXGBE_WRITE_REG(hw, IXGBE_PFVFRSSRK(i, pf_pool),
3443                                 adapter->rss_key[i]);
3444
3445         /* Fill out the redirection table */
3446         for (i = 0, j = 0; i < 64; i++, j++) {
3447                 if (j == rss_i)
3448                         j = 0;
3449
3450                 adapter->rss_indir_tbl[i] = j;
3451         }
3452
3453         ixgbe_store_vfreta(adapter);
3454 }
3455
3456 static void ixgbe_setup_mrqc(struct ixgbe_adapter *adapter)
3457 {
3458         struct ixgbe_hw *hw = &adapter->hw;
3459         u32 mrqc = 0, rss_field = 0, vfmrqc = 0;
3460         u32 rxcsum;
3461
3462         /* Disable indicating checksum in descriptor, enables RSS hash */
3463         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
3464         rxcsum |= IXGBE_RXCSUM_PCSD;
3465         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
3466
3467         if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
3468                 if (adapter->ring_feature[RING_F_RSS].mask)
3469                         mrqc = IXGBE_MRQC_RSSEN;
3470         } else {
3471                 u8 tcs = netdev_get_num_tc(adapter->netdev);
3472
3473                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
3474                         if (tcs > 4)
3475                                 mrqc = IXGBE_MRQC_VMDQRT8TCEN;  /* 8 TCs */
3476                         else if (tcs > 1)
3477                                 mrqc = IXGBE_MRQC_VMDQRT4TCEN;  /* 4 TCs */
3478                         else if (adapter->ring_feature[RING_F_RSS].indices == 4)
3479                                 mrqc = IXGBE_MRQC_VMDQRSS32EN;
3480                         else
3481                                 mrqc = IXGBE_MRQC_VMDQRSS64EN;
3482                 } else {
3483                         if (tcs > 4)
3484                                 mrqc = IXGBE_MRQC_RTRSS8TCEN;
3485                         else if (tcs > 1)
3486                                 mrqc = IXGBE_MRQC_RTRSS4TCEN;
3487                         else
3488                                 mrqc = IXGBE_MRQC_RSSEN;
3489                 }
3490         }
3491
3492         /* Perform hash on these packet types */
3493         rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4 |
3494                      IXGBE_MRQC_RSS_FIELD_IPV4_TCP |
3495                      IXGBE_MRQC_RSS_FIELD_IPV6 |
3496                      IXGBE_MRQC_RSS_FIELD_IPV6_TCP;
3497
3498         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV4_UDP)
3499                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV4_UDP;
3500         if (adapter->flags2 & IXGBE_FLAG2_RSS_FIELD_IPV6_UDP)
3501                 rss_field |= IXGBE_MRQC_RSS_FIELD_IPV6_UDP;
3502
3503         netdev_rss_key_fill(adapter->rss_key, sizeof(adapter->rss_key));
3504         if ((hw->mac.type >= ixgbe_mac_X550) &&
3505             (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)) {
3506                 unsigned int pf_pool = adapter->num_vfs;
3507
3508                 /* Enable VF RSS mode */
3509                 mrqc |= IXGBE_MRQC_MULTIPLE_RSS;
3510                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3511
3512                 /* Setup RSS through the VF registers */
3513                 ixgbe_setup_vfreta(adapter);
3514                 vfmrqc = IXGBE_MRQC_RSSEN;
3515                 vfmrqc |= rss_field;
3516                 IXGBE_WRITE_REG(hw, IXGBE_PFVFMRQC(pf_pool), vfmrqc);
3517         } else {
3518                 ixgbe_setup_reta(adapter);
3519                 mrqc |= rss_field;
3520                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
3521         }
3522 }
3523
3524 /**
3525  * ixgbe_configure_rscctl - enable RSC for the indicated ring
3526  * @adapter:    address of board private structure
3527  * @index:      index of ring to set
3528  **/
3529 static void ixgbe_configure_rscctl(struct ixgbe_adapter *adapter,
3530                                    struct ixgbe_ring *ring)
3531 {
3532         struct ixgbe_hw *hw = &adapter->hw;
3533         u32 rscctrl;
3534         u8 reg_idx = ring->reg_idx;
3535
3536         if (!ring_is_rsc_enabled(ring))
3537                 return;
3538
3539         rscctrl = IXGBE_READ_REG(hw, IXGBE_RSCCTL(reg_idx));
3540         rscctrl |= IXGBE_RSCCTL_RSCEN;
3541         /*
3542          * we must limit the number of descriptors so that the
3543          * total size of max desc * buf_len is not greater
3544          * than 65536
3545          */
3546         rscctrl |= IXGBE_RSCCTL_MAXDESC_16;
3547         IXGBE_WRITE_REG(hw, IXGBE_RSCCTL(reg_idx), rscctrl);
3548 }
3549
3550 #define IXGBE_MAX_RX_DESC_POLL 10
3551 static void ixgbe_rx_desc_queue_enable(struct ixgbe_adapter *adapter,
3552                                        struct ixgbe_ring *ring)
3553 {
3554         struct ixgbe_hw *hw = &adapter->hw;
3555         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3556         u32 rxdctl;
3557         u8 reg_idx = ring->reg_idx;
3558
3559         if (ixgbe_removed(hw->hw_addr))
3560                 return;
3561         /* RXDCTL.EN will return 0 on 82598 if link is down, so skip it */
3562         if (hw->mac.type == ixgbe_mac_82598EB &&
3563             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3564                 return;
3565
3566         do {
3567                 usleep_range(1000, 2000);
3568                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3569         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
3570
3571         if (!wait_loop) {
3572                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not set within "
3573                       "the polling period\n", reg_idx);
3574         }
3575 }
3576
3577 void ixgbe_disable_rx_queue(struct ixgbe_adapter *adapter,
3578                             struct ixgbe_ring *ring)
3579 {
3580         struct ixgbe_hw *hw = &adapter->hw;
3581         int wait_loop = IXGBE_MAX_RX_DESC_POLL;
3582         u32 rxdctl;
3583         u8 reg_idx = ring->reg_idx;
3584
3585         if (ixgbe_removed(hw->hw_addr))
3586                 return;
3587         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3588         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
3589
3590         /* write value back with RXDCTL.ENABLE bit cleared */
3591         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3592
3593         if (hw->mac.type == ixgbe_mac_82598EB &&
3594             !(IXGBE_READ_REG(hw, IXGBE_LINKS) & IXGBE_LINKS_UP))
3595                 return;
3596
3597         /* the hardware may take up to 100us to really disable the rx queue */
3598         do {
3599                 udelay(10);
3600                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3601         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
3602
3603         if (!wait_loop) {
3604                 e_err(drv, "RXDCTL.ENABLE on Rx queue %d not cleared within "
3605                       "the polling period\n", reg_idx);
3606         }
3607 }
3608
3609 void ixgbe_configure_rx_ring(struct ixgbe_adapter *adapter,
3610                              struct ixgbe_ring *ring)
3611 {
3612         struct ixgbe_hw *hw = &adapter->hw;
3613         u64 rdba = ring->dma;
3614         u32 rxdctl;
3615         u8 reg_idx = ring->reg_idx;
3616
3617         /* disable queue to avoid issues while updating state */
3618         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(reg_idx));
3619         ixgbe_disable_rx_queue(adapter, ring);
3620
3621         IXGBE_WRITE_REG(hw, IXGBE_RDBAL(reg_idx), (rdba & DMA_BIT_MASK(32)));
3622         IXGBE_WRITE_REG(hw, IXGBE_RDBAH(reg_idx), (rdba >> 32));
3623         IXGBE_WRITE_REG(hw, IXGBE_RDLEN(reg_idx),
3624                         ring->count * sizeof(union ixgbe_adv_rx_desc));
3625         /* Force flushing of IXGBE_RDLEN to prevent MDD */
3626         IXGBE_WRITE_FLUSH(hw);
3627
3628         IXGBE_WRITE_REG(hw, IXGBE_RDH(reg_idx), 0);
3629         IXGBE_WRITE_REG(hw, IXGBE_RDT(reg_idx), 0);
3630         ring->tail = adapter->io_addr + IXGBE_RDT(reg_idx);
3631
3632         ixgbe_configure_srrctl(adapter, ring);
3633         ixgbe_configure_rscctl(adapter, ring);
3634
3635         if (hw->mac.type == ixgbe_mac_82598EB) {
3636                 /*
3637                  * enable cache line friendly hardware writes:
3638                  * PTHRESH=32 descriptors (half the internal cache),
3639                  * this also removes ugly rx_no_buffer_count increment
3640                  * HTHRESH=4 descriptors (to minimize latency on fetch)
3641                  * WTHRESH=8 burst writeback up to two cache lines
3642                  */
3643                 rxdctl &= ~0x3FFFFF;
3644                 rxdctl |=  0x080420;
3645         }
3646
3647         /* enable receive descriptor ring */
3648         rxdctl |= IXGBE_RXDCTL_ENABLE;
3649         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(reg_idx), rxdctl);
3650
3651         ixgbe_rx_desc_queue_enable(adapter, ring);
3652         ixgbe_alloc_rx_buffers(ring, ixgbe_desc_unused(ring));
3653 }
3654
3655 static void ixgbe_setup_psrtype(struct ixgbe_adapter *adapter)
3656 {
3657         struct ixgbe_hw *hw = &adapter->hw;
3658         int rss_i = adapter->ring_feature[RING_F_RSS].indices;
3659         u16 pool;
3660
3661         /* PSRTYPE must be initialized in non 82598 adapters */
3662         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
3663                       IXGBE_PSRTYPE_UDPHDR |
3664                       IXGBE_PSRTYPE_IPV4HDR |
3665                       IXGBE_PSRTYPE_L2HDR |
3666                       IXGBE_PSRTYPE_IPV6HDR;
3667
3668         if (hw->mac.type == ixgbe_mac_82598EB)
3669                 return;
3670
3671         if (rss_i > 3)
3672                 psrtype |= 2 << 29;
3673         else if (rss_i > 1)
3674                 psrtype |= 1 << 29;
3675
3676         for_each_set_bit(pool, &adapter->fwd_bitmask, 32)
3677                 IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
3678 }
3679
3680 static void ixgbe_configure_virtualization(struct ixgbe_adapter *adapter)
3681 {
3682         struct ixgbe_hw *hw = &adapter->hw;
3683         u32 reg_offset, vf_shift;
3684         u32 gcr_ext, vmdctl;
3685         int i;
3686
3687         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
3688                 return;
3689
3690         vmdctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
3691         vmdctl |= IXGBE_VMD_CTL_VMDQ_EN;
3692         vmdctl &= ~IXGBE_VT_CTL_POOL_MASK;
3693         vmdctl |= VMDQ_P(0) << IXGBE_VT_CTL_POOL_SHIFT;
3694         vmdctl |= IXGBE_VT_CTL_REPLEN;
3695         IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vmdctl);
3696
3697         vf_shift = VMDQ_P(0) % 32;
3698         reg_offset = (VMDQ_P(0) >= 32) ? 1 : 0;
3699
3700         /* Enable only the PF's pool for Tx/Rx */
3701         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), (~0) << vf_shift);
3702         IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset ^ 1), reg_offset - 1);
3703         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), (~0) << vf_shift);
3704         IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset ^ 1), reg_offset - 1);
3705         if (adapter->bridge_mode == BRIDGE_MODE_VEB)
3706                 IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
3707
3708         /* Map PF MAC address in RAR Entry 0 to first pool following VFs */
3709         hw->mac.ops.set_vmdq(hw, 0, VMDQ_P(0));
3710
3711         /* clear VLAN promisc flag so VFTA will be updated if necessary */
3712         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
3713
3714         /*
3715          * Set up VF register offsets for selected VT Mode,
3716          * i.e. 32 or 64 VFs for SR-IOV
3717          */
3718         switch (adapter->ring_feature[RING_F_VMDQ].mask) {
3719         case IXGBE_82599_VMDQ_8Q_MASK:
3720                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_16;
3721                 break;
3722         case IXGBE_82599_VMDQ_4Q_MASK:
3723                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_32;
3724                 break;
3725         default:
3726                 gcr_ext = IXGBE_GCR_EXT_VT_MODE_64;
3727                 break;
3728         }
3729
3730         IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
3731
3732
3733         /* Enable MAC Anti-Spoofing */
3734         hw->mac.ops.set_mac_anti_spoofing(hw, (adapter->num_vfs != 0),
3735                                           adapter->num_vfs);
3736
3737         /* Ensure LLDP and FC is set for Ethertype Antispoofing if we will be
3738          * calling set_ethertype_anti_spoofing for each VF in loop below
3739          */
3740         if (hw->mac.ops.set_ethertype_anti_spoofing) {
3741                 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_LLDP),
3742                                 (IXGBE_ETQF_FILTER_EN    |
3743                                  IXGBE_ETQF_TX_ANTISPOOF |
3744                                  IXGBE_ETH_P_LLDP));
3745
3746                 IXGBE_WRITE_REG(hw, IXGBE_ETQF(IXGBE_ETQF_FILTER_FC),
3747                                 (IXGBE_ETQF_FILTER_EN |
3748                                  IXGBE_ETQF_TX_ANTISPOOF |
3749                                  ETH_P_PAUSE));
3750         }
3751
3752         /* For VFs that have spoof checking turned off */
3753         for (i = 0; i < adapter->num_vfs; i++) {
3754                 if (!adapter->vfinfo[i].spoofchk_enabled)
3755                         ixgbe_ndo_set_vf_spoofchk(adapter->netdev, i, false);
3756
3757                 /* enable ethertype anti spoofing if hw supports it */
3758                 if (hw->mac.ops.set_ethertype_anti_spoofing)
3759                         hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
3760
3761                 /* Enable/Disable RSS query feature  */
3762                 ixgbe_ndo_set_vf_rss_query_en(adapter->netdev, i,
3763                                           adapter->vfinfo[i].rss_query_enabled);
3764         }
3765 }
3766
3767 static void ixgbe_set_rx_buffer_len(struct ixgbe_adapter *adapter)
3768 {
3769         struct ixgbe_hw *hw = &adapter->hw;
3770         struct net_device *netdev = adapter->netdev;
3771         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
3772         struct ixgbe_ring *rx_ring;
3773         int i;
3774         u32 mhadd, hlreg0;
3775
3776 #ifdef IXGBE_FCOE
3777         /* adjust max frame to be able to do baby jumbo for FCoE */
3778         if ((adapter->flags & IXGBE_FLAG_FCOE_ENABLED) &&
3779             (max_frame < IXGBE_FCOE_JUMBO_FRAME_SIZE))
3780                 max_frame = IXGBE_FCOE_JUMBO_FRAME_SIZE;
3781
3782 #endif /* IXGBE_FCOE */
3783
3784         /* adjust max frame to be at least the size of a standard frame */
3785         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
3786                 max_frame = (ETH_FRAME_LEN + ETH_FCS_LEN);
3787
3788         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
3789         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
3790                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
3791                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
3792
3793                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
3794         }
3795
3796         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
3797         /* set jumbo enable since MHADD.MFS is keeping size locked at max_frame */
3798         hlreg0 |= IXGBE_HLREG0_JUMBOEN;
3799         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
3800
3801         /*
3802          * Setup the HW Rx Head and Tail Descriptor Pointers and
3803          * the Base and Length of the Rx Descriptor Ring
3804          */
3805         for (i = 0; i < adapter->num_rx_queues; i++) {
3806                 rx_ring = adapter->rx_ring[i];
3807                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
3808                         set_ring_rsc_enabled(rx_ring);
3809                 else
3810                         clear_ring_rsc_enabled(rx_ring);
3811         }
3812 }
3813
3814 static void ixgbe_setup_rdrxctl(struct ixgbe_adapter *adapter)
3815 {
3816         struct ixgbe_hw *hw = &adapter->hw;
3817         u32 rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
3818
3819         switch (hw->mac.type) {
3820         case ixgbe_mac_82598EB:
3821                 /*
3822                  * For VMDq support of different descriptor types or
3823                  * buffer sizes through the use of multiple SRRCTL
3824                  * registers, RDRXCTL.MVMEN must be set to 1
3825                  *
3826                  * also, the manual doesn't mention it clearly but DCA hints
3827                  * will only use queue 0's tags unless this bit is set.  Side
3828                  * effects of setting this bit are only that SRRCTL must be
3829                  * fully programmed [0..15]
3830                  */
3831                 rdrxctl |= IXGBE_RDRXCTL_MVMEN;
3832                 break;
3833         case ixgbe_mac_X550:
3834         case ixgbe_mac_X550EM_x:
3835                 if (adapter->num_vfs)
3836                         rdrxctl |= IXGBE_RDRXCTL_PSP;
3837                 /* fall through for older HW */
3838         case ixgbe_mac_82599EB:
3839         case ixgbe_mac_X540:
3840                 /* Disable RSC for ACK packets */
3841                 IXGBE_WRITE_REG(hw, IXGBE_RSCDBU,
3842                    (IXGBE_RSCDBU_RSCACKDIS | IXGBE_READ_REG(hw, IXGBE_RSCDBU)));
3843                 rdrxctl &= ~IXGBE_RDRXCTL_RSCFRSTSIZE;
3844                 /* hardware requires some bits to be set by default */
3845                 rdrxctl |= (IXGBE_RDRXCTL_RSCACKC | IXGBE_RDRXCTL_FCOE_WRFIX);
3846                 rdrxctl |= IXGBE_RDRXCTL_CRCSTRIP;
3847                 break;
3848         default:
3849                 /* We should do nothing since we don't know this hardware */
3850                 return;
3851         }
3852
3853         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
3854 }
3855
3856 /**
3857  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
3858  * @adapter: board private structure
3859  *
3860  * Configure the Rx unit of the MAC after a reset.
3861  **/
3862 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
3863 {
3864         struct ixgbe_hw *hw = &adapter->hw;
3865         int i;
3866         u32 rxctrl, rfctl;
3867
3868         /* disable receives while setting up the descriptors */
3869         hw->mac.ops.disable_rx(hw);
3870
3871         ixgbe_setup_psrtype(adapter);
3872         ixgbe_setup_rdrxctl(adapter);
3873
3874         /* RSC Setup */
3875         rfctl = IXGBE_READ_REG(hw, IXGBE_RFCTL);
3876         rfctl &= ~IXGBE_RFCTL_RSC_DIS;
3877         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED))
3878                 rfctl |= IXGBE_RFCTL_RSC_DIS;
3879         IXGBE_WRITE_REG(hw, IXGBE_RFCTL, rfctl);
3880
3881         /* Program registers for the distribution of queues */
3882         ixgbe_setup_mrqc(adapter);
3883
3884         /* set_rx_buffer_len must be called before ring initialization */
3885         ixgbe_set_rx_buffer_len(adapter);
3886
3887         /*
3888          * Setup the HW Rx Head and Tail Descriptor Pointers and
3889          * the Base and Length of the Rx Descriptor Ring
3890          */
3891         for (i = 0; i < adapter->num_rx_queues; i++)
3892                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[i]);
3893
3894         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
3895         /* disable drop enable for 82598 parts */
3896         if (hw->mac.type == ixgbe_mac_82598EB)
3897                 rxctrl |= IXGBE_RXCTRL_DMBYPS;
3898
3899         /* enable all receives */
3900         rxctrl |= IXGBE_RXCTRL_RXEN;
3901         hw->mac.ops.enable_rx_dma(hw, rxctrl);
3902 }
3903
3904 static int ixgbe_vlan_rx_add_vid(struct net_device *netdev,
3905                                  __be16 proto, u16 vid)
3906 {
3907         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3908         struct ixgbe_hw *hw = &adapter->hw;
3909
3910         /* add VID to filter table */
3911         hw->mac.ops.set_vfta(&adapter->hw, vid, VMDQ_P(0), true, true);
3912         set_bit(vid, adapter->active_vlans);
3913
3914         return 0;
3915 }
3916
3917 static int ixgbe_find_vlvf_entry(struct ixgbe_hw *hw, u32 vlan)
3918 {
3919         u32 vlvf;
3920         int idx;
3921
3922         /* short cut the special case */
3923         if (vlan == 0)
3924                 return 0;
3925
3926         /* Search for the vlan id in the VLVF entries */
3927         for (idx = IXGBE_VLVF_ENTRIES; --idx;) {
3928                 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(idx));
3929                 if ((vlvf & VLAN_VID_MASK) == vlan)
3930                         break;
3931         }
3932
3933         return idx;
3934 }
3935
3936 void ixgbe_update_pf_promisc_vlvf(struct ixgbe_adapter *adapter, u32 vid)
3937 {
3938         struct ixgbe_hw *hw = &adapter->hw;
3939         u32 bits, word;
3940         int idx;
3941
3942         idx = ixgbe_find_vlvf_entry(hw, vid);
3943         if (!idx)
3944                 return;
3945
3946         /* See if any other pools are set for this VLAN filter
3947          * entry other than the PF.
3948          */
3949         word = idx * 2 + (VMDQ_P(0) / 32);
3950         bits = ~(1 << (VMDQ_P(0)) % 32);
3951         bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
3952
3953         /* Disable the filter so this falls into the default pool. */
3954         if (!bits && !IXGBE_READ_REG(hw, IXGBE_VLVFB(word ^ 1))) {
3955                 if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
3956                         IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), 0);
3957                 IXGBE_WRITE_REG(hw, IXGBE_VLVF(idx), 0);
3958         }
3959 }
3960
3961 static int ixgbe_vlan_rx_kill_vid(struct net_device *netdev,
3962                                   __be16 proto, u16 vid)
3963 {
3964         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3965         struct ixgbe_hw *hw = &adapter->hw;
3966
3967         /* remove VID from filter table */
3968         if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
3969                 ixgbe_update_pf_promisc_vlvf(adapter, vid);
3970         else
3971                 hw->mac.ops.set_vfta(hw, vid, VMDQ_P(0), false, true);
3972
3973         clear_bit(vid, adapter->active_vlans);
3974
3975         return 0;
3976 }
3977
3978 /**
3979  * ixgbe_vlan_strip_disable - helper to disable hw vlan stripping
3980  * @adapter: driver data
3981  */
3982 static void ixgbe_vlan_strip_disable(struct ixgbe_adapter *adapter)
3983 {
3984         struct ixgbe_hw *hw = &adapter->hw;
3985         u32 vlnctrl;
3986         int i, j;
3987
3988         switch (hw->mac.type) {
3989         case ixgbe_mac_82598EB:
3990                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
3991                 vlnctrl &= ~IXGBE_VLNCTRL_VME;
3992                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
3993                 break;
3994         case ixgbe_mac_82599EB:
3995         case ixgbe_mac_X540:
3996         case ixgbe_mac_X550:
3997         case ixgbe_mac_X550EM_x:
3998                 for (i = 0; i < adapter->num_rx_queues; i++) {
3999                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4000
4001                         if (ring->l2_accel_priv)
4002                                 continue;
4003                         j = ring->reg_idx;
4004                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4005                         vlnctrl &= ~IXGBE_RXDCTL_VME;
4006                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4007                 }
4008                 break;
4009         default:
4010                 break;
4011         }
4012 }
4013
4014 /**
4015  * ixgbe_vlan_strip_enable - helper to enable hw vlan stripping
4016  * @adapter: driver data
4017  */
4018 static void ixgbe_vlan_strip_enable(struct ixgbe_adapter *adapter)
4019 {
4020         struct ixgbe_hw *hw = &adapter->hw;
4021         u32 vlnctrl;
4022         int i, j;
4023
4024         switch (hw->mac.type) {
4025         case ixgbe_mac_82598EB:
4026                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4027                 vlnctrl |= IXGBE_VLNCTRL_VME;
4028                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4029                 break;
4030         case ixgbe_mac_82599EB:
4031         case ixgbe_mac_X540:
4032         case ixgbe_mac_X550:
4033         case ixgbe_mac_X550EM_x:
4034                 for (i = 0; i < adapter->num_rx_queues; i++) {
4035                         struct ixgbe_ring *ring = adapter->rx_ring[i];
4036
4037                         if (ring->l2_accel_priv)
4038                                 continue;
4039                         j = ring->reg_idx;
4040                         vlnctrl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
4041                         vlnctrl |= IXGBE_RXDCTL_VME;
4042                         IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), vlnctrl);
4043                 }
4044                 break;
4045         default:
4046                 break;
4047         }
4048 }
4049
4050 static void ixgbe_vlan_promisc_enable(struct ixgbe_adapter *adapter)
4051 {
4052         struct ixgbe_hw *hw = &adapter->hw;
4053         u32 vlnctrl, i;
4054
4055         switch (hw->mac.type) {
4056         case ixgbe_mac_82599EB:
4057         case ixgbe_mac_X540:
4058         case ixgbe_mac_X550:
4059         case ixgbe_mac_X550EM_x:
4060         default:
4061                 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED)
4062                         break;
4063                 /* fall through */
4064         case ixgbe_mac_82598EB:
4065                 /* legacy case, we can just disable VLAN filtering */
4066                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4067                 vlnctrl &= ~(IXGBE_VLNCTRL_VFE | IXGBE_VLNCTRL_CFIEN);
4068                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4069                 return;
4070         }
4071
4072         /* We are already in VLAN promisc, nothing to do */
4073         if (adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC)
4074                 return;
4075
4076         /* Set flag so we don't redo unnecessary work */
4077         adapter->flags2 |= IXGBE_FLAG2_VLAN_PROMISC;
4078
4079         /* Add PF to all active pools */
4080         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4081                 u32 reg_offset = IXGBE_VLVFB(i * 2 + VMDQ_P(0) / 32);
4082                 u32 vlvfb = IXGBE_READ_REG(hw, reg_offset);
4083
4084                 vlvfb |= 1 << (VMDQ_P(0) % 32);
4085                 IXGBE_WRITE_REG(hw, reg_offset, vlvfb);
4086         }
4087
4088         /* Set all bits in the VLAN filter table array */
4089         for (i = hw->mac.vft_size; i--;)
4090                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), ~0U);
4091 }
4092
4093 #define VFTA_BLOCK_SIZE 8
4094 static void ixgbe_scrub_vfta(struct ixgbe_adapter *adapter, u32 vfta_offset)
4095 {
4096         struct ixgbe_hw *hw = &adapter->hw;
4097         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4098         u32 vid_start = vfta_offset * 32;
4099         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4100         u32 i, vid, word, bits;
4101
4102         for (i = IXGBE_VLVF_ENTRIES; --i;) {
4103                 u32 vlvf = IXGBE_READ_REG(hw, IXGBE_VLVF(i));
4104
4105                 /* pull VLAN ID from VLVF */
4106                 vid = vlvf & VLAN_VID_MASK;
4107
4108                 /* only concern outselves with a certain range */
4109                 if (vid < vid_start || vid >= vid_end)
4110                         continue;
4111
4112                 if (vlvf) {
4113                         /* record VLAN ID in VFTA */
4114                         vfta[(vid - vid_start) / 32] |= 1 << (vid % 32);
4115
4116                         /* if PF is part of this then continue */
4117                         if (test_bit(vid, adapter->active_vlans))
4118                                 continue;
4119                 }
4120
4121                 /* remove PF from the pool */
4122                 word = i * 2 + VMDQ_P(0) / 32;
4123                 bits = ~(1 << (VMDQ_P(0) % 32));
4124                 bits &= IXGBE_READ_REG(hw, IXGBE_VLVFB(word));
4125                 IXGBE_WRITE_REG(hw, IXGBE_VLVFB(word), bits);
4126         }
4127
4128         /* extract values from active_vlans and write back to VFTA */
4129         for (i = VFTA_BLOCK_SIZE; i--;) {
4130                 vid = (vfta_offset + i) * 32;
4131                 word = vid / BITS_PER_LONG;
4132                 bits = vid % BITS_PER_LONG;
4133
4134                 vfta[i] |= adapter->active_vlans[word] >> bits;
4135
4136                 IXGBE_WRITE_REG(hw, IXGBE_VFTA(vfta_offset + i), vfta[i]);
4137         }
4138 }
4139
4140 static void ixgbe_vlan_promisc_disable(struct ixgbe_adapter *adapter)
4141 {
4142         struct ixgbe_hw *hw = &adapter->hw;
4143         u32 vlnctrl, i;
4144
4145         switch (hw->mac.type) {
4146         case ixgbe_mac_82599EB:
4147         case ixgbe_mac_X540:
4148         case ixgbe_mac_X550:
4149         case ixgbe_mac_X550EM_x:
4150         default:
4151                 if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED)
4152                         break;
4153                 /* fall through */
4154         case ixgbe_mac_82598EB:
4155                 vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
4156                 vlnctrl &= ~IXGBE_VLNCTRL_CFIEN;
4157                 vlnctrl |= IXGBE_VLNCTRL_VFE;
4158                 IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
4159                 return;
4160         }
4161
4162         /* We are not in VLAN promisc, nothing to do */
4163         if (!(adapter->flags2 & IXGBE_FLAG2_VLAN_PROMISC))
4164                 return;
4165
4166         /* Set flag so we don't redo unnecessary work */
4167         adapter->flags2 &= ~IXGBE_FLAG2_VLAN_PROMISC;
4168
4169         for (i = 0; i < hw->mac.vft_size; i += VFTA_BLOCK_SIZE)
4170                 ixgbe_scrub_vfta(adapter, i);
4171 }
4172
4173 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
4174 {
4175         u16 vid;
4176
4177         ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
4178
4179         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4180                 ixgbe_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
4181 }
4182
4183 /**
4184  * ixgbe_write_mc_addr_list - write multicast addresses to MTA
4185  * @netdev: network interface device structure
4186  *
4187  * Writes multicast address list to the MTA hash table.
4188  * Returns: -ENOMEM on failure
4189  *                0 on no addresses written
4190  *                X on writing X addresses to MTA
4191  **/
4192 static int ixgbe_write_mc_addr_list(struct net_device *netdev)
4193 {
4194         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4195         struct ixgbe_hw *hw = &adapter->hw;
4196
4197         if (!netif_running(netdev))
4198                 return 0;
4199
4200         if (hw->mac.ops.update_mc_addr_list)
4201                 hw->mac.ops.update_mc_addr_list(hw, netdev);
4202         else
4203                 return -ENOMEM;
4204
4205 #ifdef CONFIG_PCI_IOV
4206         ixgbe_restore_vf_multicasts(adapter);
4207 #endif
4208
4209         return netdev_mc_count(netdev);
4210 }
4211
4212 #ifdef CONFIG_PCI_IOV
4213 void ixgbe_full_sync_mac_table(struct ixgbe_adapter *adapter)
4214 {
4215         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4216         struct ixgbe_hw *hw = &adapter->hw;
4217         int i;
4218
4219         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4220                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4221
4222                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4223                         hw->mac.ops.set_rar(hw, i,
4224                                             mac_table->addr,
4225                                             mac_table->pool,
4226                                             IXGBE_RAH_AV);
4227                 else
4228                         hw->mac.ops.clear_rar(hw, i);
4229         }
4230 }
4231
4232 #endif
4233 static void ixgbe_sync_mac_table(struct ixgbe_adapter *adapter)
4234 {
4235         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4236         struct ixgbe_hw *hw = &adapter->hw;
4237         int i;
4238
4239         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4240                 if (!(mac_table->state & IXGBE_MAC_STATE_MODIFIED))
4241                         continue;
4242
4243                 mac_table->state &= ~IXGBE_MAC_STATE_MODIFIED;
4244
4245                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4246                         hw->mac.ops.set_rar(hw, i,
4247                                             mac_table->addr,
4248                                             mac_table->pool,
4249                                             IXGBE_RAH_AV);
4250                 else
4251                         hw->mac.ops.clear_rar(hw, i);
4252         }
4253 }
4254
4255 static void ixgbe_flush_sw_mac_table(struct ixgbe_adapter *adapter)
4256 {
4257         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4258         struct ixgbe_hw *hw = &adapter->hw;
4259         int i;
4260
4261         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4262                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4263                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4264         }
4265
4266         ixgbe_sync_mac_table(adapter);
4267 }
4268
4269 static int ixgbe_available_rars(struct ixgbe_adapter *adapter, u16 pool)
4270 {
4271         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4272         struct ixgbe_hw *hw = &adapter->hw;
4273         int i, count = 0;
4274
4275         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4276                 /* do not count default RAR as available */
4277                 if (mac_table->state & IXGBE_MAC_STATE_DEFAULT)
4278                         continue;
4279
4280                 /* only count unused and addresses that belong to us */
4281                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE) {
4282                         if (mac_table->pool != pool)
4283                                 continue;
4284                 }
4285
4286                 count++;
4287         }
4288
4289         return count;
4290 }
4291
4292 /* this function destroys the first RAR entry */
4293 static void ixgbe_mac_set_default_filter(struct ixgbe_adapter *adapter)
4294 {
4295         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4296         struct ixgbe_hw *hw = &adapter->hw;
4297
4298         memcpy(&mac_table->addr, hw->mac.addr, ETH_ALEN);
4299         mac_table->pool = VMDQ_P(0);
4300
4301         mac_table->state = IXGBE_MAC_STATE_DEFAULT | IXGBE_MAC_STATE_IN_USE;
4302
4303         hw->mac.ops.set_rar(hw, 0, mac_table->addr, mac_table->pool,
4304                             IXGBE_RAH_AV);
4305 }
4306
4307 int ixgbe_add_mac_filter(struct ixgbe_adapter *adapter,
4308                          const u8 *addr, u16 pool)
4309 {
4310         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4311         struct ixgbe_hw *hw = &adapter->hw;
4312         int i;
4313
4314         if (is_zero_ether_addr(addr))
4315                 return -EINVAL;
4316
4317         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4318                 if (mac_table->state & IXGBE_MAC_STATE_IN_USE)
4319                         continue;
4320
4321                 ether_addr_copy(mac_table->addr, addr);
4322                 mac_table->pool = pool;
4323
4324                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED |
4325                                     IXGBE_MAC_STATE_IN_USE;
4326
4327                 ixgbe_sync_mac_table(adapter);
4328
4329                 return i;
4330         }
4331
4332         return -ENOMEM;
4333 }
4334
4335 int ixgbe_del_mac_filter(struct ixgbe_adapter *adapter,
4336                          const u8 *addr, u16 pool)
4337 {
4338         struct ixgbe_mac_addr *mac_table = &adapter->mac_table[0];
4339         struct ixgbe_hw *hw = &adapter->hw;
4340         int i;
4341
4342         if (is_zero_ether_addr(addr))
4343                 return -EINVAL;
4344
4345         /* search table for addr, if found clear IN_USE flag and sync */
4346         for (i = 0; i < hw->mac.num_rar_entries; i++, mac_table++) {
4347                 /* we can only delete an entry if it is in use */
4348                 if (!(mac_table->state & IXGBE_MAC_STATE_IN_USE))
4349                         continue;
4350                 /* we only care about entries that belong to the given pool */
4351                 if (mac_table->pool != pool)
4352                         continue;
4353                 /* we only care about a specific MAC address */
4354                 if (!ether_addr_equal(addr, mac_table->addr))
4355                         continue;
4356
4357                 mac_table->state |= IXGBE_MAC_STATE_MODIFIED;
4358                 mac_table->state &= ~IXGBE_MAC_STATE_IN_USE;
4359
4360                 ixgbe_sync_mac_table(adapter);
4361
4362                 return 0;
4363         }
4364
4365         return -ENOMEM;
4366 }
4367 /**
4368  * ixgbe_write_uc_addr_list - write unicast addresses to RAR table
4369  * @netdev: network interface device structure
4370  *
4371  * Writes unicast address list to the RAR table.
4372  * Returns: -ENOMEM on failure/insufficient address space
4373  *                0 on no addresses written
4374  *                X on writing X addresses to the RAR table
4375  **/
4376 static int ixgbe_write_uc_addr_list(struct net_device *netdev, int vfn)
4377 {
4378         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4379         int count = 0;
4380
4381         /* return ENOMEM indicating insufficient memory for addresses */
4382         if (netdev_uc_count(netdev) > ixgbe_available_rars(adapter, vfn))
4383                 return -ENOMEM;
4384
4385         if (!netdev_uc_empty(netdev)) {
4386                 struct netdev_hw_addr *ha;
4387                 netdev_for_each_uc_addr(ha, netdev) {
4388                         ixgbe_del_mac_filter(adapter, ha->addr, vfn);
4389                         ixgbe_add_mac_filter(adapter, ha->addr, vfn);
4390                         count++;
4391                 }
4392         }
4393         return count;
4394 }
4395
4396 static int ixgbe_uc_sync(struct net_device *netdev, const unsigned char *addr)
4397 {
4398         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4399         int ret;
4400
4401         ret = ixgbe_add_mac_filter(adapter, addr, VMDQ_P(0));
4402
4403         return min_t(int, ret, 0);
4404 }
4405
4406 static int ixgbe_uc_unsync(struct net_device *netdev, const unsigned char *addr)
4407 {
4408         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4409
4410         ixgbe_del_mac_filter(adapter, addr, VMDQ_P(0));
4411
4412         return 0;
4413 }
4414
4415 /**
4416  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
4417  * @netdev: network interface device structure
4418  *
4419  * The set_rx_method entry point is called whenever the unicast/multicast
4420  * address list or the network interface flags are updated.  This routine is
4421  * responsible for configuring the hardware for proper unicast, multicast and
4422  * promiscuous mode.
4423  **/
4424 void ixgbe_set_rx_mode(struct net_device *netdev)
4425 {
4426         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4427         struct ixgbe_hw *hw = &adapter->hw;
4428         u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
4429         int count;
4430
4431         /* Check for Promiscuous and All Multicast modes */
4432         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
4433
4434         /* set all bits that we expect to always be set */
4435         fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
4436         fctrl |= IXGBE_FCTRL_BAM;
4437         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
4438         fctrl |= IXGBE_FCTRL_PMCF;
4439
4440         /* clear the bits we are changing the status of */
4441         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4442         if (netdev->flags & IFF_PROMISC) {
4443                 hw->addr_ctrl.user_set_promisc = true;
4444                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
4445                 vmolr |= IXGBE_VMOLR_MPE;
4446                 ixgbe_vlan_promisc_enable(adapter);
4447         } else {
4448                 if (netdev->flags & IFF_ALLMULTI) {
4449                         fctrl |= IXGBE_FCTRL_MPE;
4450                         vmolr |= IXGBE_VMOLR_MPE;
4451                 }
4452                 hw->addr_ctrl.user_set_promisc = false;
4453                 ixgbe_vlan_promisc_disable(adapter);
4454         }
4455
4456         /*
4457          * Write addresses to available RAR registers, if there is not
4458          * sufficient space to store all the addresses then enable
4459          * unicast promiscuous mode
4460          */
4461         if (__dev_uc_sync(netdev, ixgbe_uc_sync, ixgbe_uc_unsync)) {
4462                 fctrl |= IXGBE_FCTRL_UPE;
4463                 vmolr |= IXGBE_VMOLR_ROPE;
4464         }
4465
4466         /* Write addresses to the MTA, if the attempt fails
4467          * then we should just turn on promiscuous mode so
4468          * that we can at least receive multicast traffic
4469          */
4470         count = ixgbe_write_mc_addr_list(netdev);
4471         if (count < 0) {
4472                 fctrl |= IXGBE_FCTRL_MPE;
4473                 vmolr |= IXGBE_VMOLR_MPE;
4474         } else if (count) {
4475                 vmolr |= IXGBE_VMOLR_ROMPE;
4476         }
4477
4478         if (hw->mac.type != ixgbe_mac_82598EB) {
4479                 vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(VMDQ_P(0))) &
4480                          ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
4481                            IXGBE_VMOLR_ROPE);
4482                 IXGBE_WRITE_REG(hw, IXGBE_VMOLR(VMDQ_P(0)), vmolr);
4483         }
4484
4485         /* This is useful for sniffing bad packets. */
4486         if (adapter->netdev->features & NETIF_F_RXALL) {
4487                 /* UPE and MPE will be handled by normal PROMISC logic
4488                  * in e1000e_set_rx_mode */
4489                 fctrl |= (IXGBE_FCTRL_SBP | /* Receive bad packets */
4490                           IXGBE_FCTRL_BAM | /* RX All Bcast Pkts */
4491                           IXGBE_FCTRL_PMCF); /* RX All MAC Ctrl Pkts */
4492
4493                 fctrl &= ~(IXGBE_FCTRL_DPF);
4494                 /* NOTE:  VLAN filtering is disabled by setting PROMISC */
4495         }
4496
4497         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
4498
4499         if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
4500                 ixgbe_vlan_strip_enable(adapter);
4501         else
4502                 ixgbe_vlan_strip_disable(adapter);
4503 }
4504
4505 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
4506 {
4507         int q_idx;
4508
4509         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4510                 ixgbe_qv_init_lock(adapter->q_vector[q_idx]);
4511                 napi_enable(&adapter->q_vector[q_idx]->napi);
4512         }
4513 }
4514
4515 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
4516 {
4517         int q_idx;
4518
4519         for (q_idx = 0; q_idx < adapter->num_q_vectors; q_idx++) {
4520                 napi_disable(&adapter->q_vector[q_idx]->napi);
4521                 while (!ixgbe_qv_disable(adapter->q_vector[q_idx])) {
4522                         pr_info("QV %d locked\n", q_idx);
4523                         usleep_range(1000, 20000);
4524                 }
4525         }
4526 }
4527
4528 static void ixgbe_clear_vxlan_port(struct ixgbe_adapter *adapter)
4529 {
4530         switch (adapter->hw.mac.type) {
4531         case ixgbe_mac_X550:
4532         case ixgbe_mac_X550EM_x:
4533                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VXLANCTRL, 0);
4534 #ifdef CONFIG_IXGBE_VXLAN
4535                 adapter->vxlan_port = 0;
4536 #endif
4537                 break;
4538         default:
4539                 break;
4540         }
4541 }
4542
4543 #ifdef CONFIG_IXGBE_DCB
4544 /**
4545  * ixgbe_configure_dcb - Configure DCB hardware
4546  * @adapter: ixgbe adapter struct
4547  *
4548  * This is called by the driver on open to configure the DCB hardware.
4549  * This is also called by the gennetlink interface when reconfiguring
4550  * the DCB state.
4551  */
4552 static void ixgbe_configure_dcb(struct ixgbe_adapter *adapter)
4553 {
4554         struct ixgbe_hw *hw = &adapter->hw;
4555         int max_frame = adapter->netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
4556
4557         if (!(adapter->flags & IXGBE_FLAG_DCB_ENABLED)) {
4558                 if (hw->mac.type == ixgbe_mac_82598EB)
4559                         netif_set_gso_max_size(adapter->netdev, 65536);
4560                 return;
4561         }
4562
4563         if (hw->mac.type == ixgbe_mac_82598EB)
4564                 netif_set_gso_max_size(adapter->netdev, 32768);
4565
4566 #ifdef IXGBE_FCOE
4567         if (adapter->netdev->features & NETIF_F_FCOE_MTU)
4568                 max_frame = max(max_frame, IXGBE_FCOE_JUMBO_FRAME_SIZE);
4569 #endif
4570
4571         /* reconfigure the hardware */
4572         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE) {
4573                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4574                                                 DCB_TX_CONFIG);
4575                 ixgbe_dcb_calculate_tc_credits(hw, &adapter->dcb_cfg, max_frame,
4576                                                 DCB_RX_CONFIG);
4577                 ixgbe_dcb_hw_config(hw, &adapter->dcb_cfg);
4578         } else if (adapter->ixgbe_ieee_ets && adapter->ixgbe_ieee_pfc) {
4579                 ixgbe_dcb_hw_ets(&adapter->hw,
4580                                  adapter->ixgbe_ieee_ets,
4581                                  max_frame);
4582                 ixgbe_dcb_hw_pfc_config(&adapter->hw,
4583                                         adapter->ixgbe_ieee_pfc->pfc_en,
4584                                         adapter->ixgbe_ieee_ets->prio_tc);
4585         }
4586
4587         /* Enable RSS Hash per TC */
4588         if (hw->mac.type != ixgbe_mac_82598EB) {
4589                 u32 msb = 0;
4590                 u16 rss_i = adapter->ring_feature[RING_F_RSS].indices - 1;
4591
4592                 while (rss_i) {
4593                         msb++;
4594                         rss_i >>= 1;
4595                 }
4596
4597                 /* write msb to all 8 TCs in one write */
4598                 IXGBE_WRITE_REG(hw, IXGBE_RQTC, msb * 0x11111111);
4599         }
4600 }
4601 #endif
4602
4603 /* Additional bittime to account for IXGBE framing */
4604 #define IXGBE_ETH_FRAMING 20
4605
4606 /**
4607  * ixgbe_hpbthresh - calculate high water mark for flow control
4608  *
4609  * @adapter: board private structure to calculate for
4610  * @pb: packet buffer to calculate
4611  */
4612 static int ixgbe_hpbthresh(struct ixgbe_adapter *adapter, int pb)
4613 {
4614         struct ixgbe_hw *hw = &adapter->hw;
4615         struct net_device *dev = adapter->netdev;
4616         int link, tc, kb, marker;
4617         u32 dv_id, rx_pba;
4618
4619         /* Calculate max LAN frame size */
4620         tc = link = dev->mtu + ETH_HLEN + ETH_FCS_LEN + IXGBE_ETH_FRAMING;
4621
4622 #ifdef IXGBE_FCOE
4623         /* FCoE traffic class uses FCOE jumbo frames */
4624         if ((dev->features & NETIF_F_FCOE_MTU) &&
4625             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4626             (pb == ixgbe_fcoe_get_tc(adapter)))
4627                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4628 #endif
4629
4630         /* Calculate delay value for device */
4631         switch (hw->mac.type) {
4632         case ixgbe_mac_X540:
4633         case ixgbe_mac_X550:
4634         case ixgbe_mac_X550EM_x:
4635                 dv_id = IXGBE_DV_X540(link, tc);
4636                 break;
4637         default:
4638                 dv_id = IXGBE_DV(link, tc);
4639                 break;
4640         }
4641
4642         /* Loopback switch introduces additional latency */
4643         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
4644                 dv_id += IXGBE_B2BT(tc);
4645
4646         /* Delay value is calculated in bit times convert to KB */
4647         kb = IXGBE_BT2KB(dv_id);
4648         rx_pba = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(pb)) >> 10;
4649
4650         marker = rx_pba - kb;
4651
4652         /* It is possible that the packet buffer is not large enough
4653          * to provide required headroom. In this case throw an error
4654          * to user and a do the best we can.
4655          */
4656         if (marker < 0) {
4657                 e_warn(drv, "Packet Buffer(%i) can not provide enough"
4658                             "headroom to support flow control."
4659                             "Decrease MTU or number of traffic classes\n", pb);
4660                 marker = tc + 1;
4661         }
4662
4663         return marker;
4664 }
4665
4666 /**
4667  * ixgbe_lpbthresh - calculate low water mark for for flow control
4668  *
4669  * @adapter: board private structure to calculate for
4670  * @pb: packet buffer to calculate
4671  */
4672 static int ixgbe_lpbthresh(struct ixgbe_adapter *adapter, int pb)
4673 {
4674         struct ixgbe_hw *hw = &adapter->hw;
4675         struct net_device *dev = adapter->netdev;
4676         int tc;
4677         u32 dv_id;
4678
4679         /* Calculate max LAN frame size */
4680         tc = dev->mtu + ETH_HLEN + ETH_FCS_LEN;
4681
4682 #ifdef IXGBE_FCOE
4683         /* FCoE traffic class uses FCOE jumbo frames */
4684         if ((dev->features & NETIF_F_FCOE_MTU) &&
4685             (tc < IXGBE_FCOE_JUMBO_FRAME_SIZE) &&
4686             (pb == netdev_get_prio_tc_map(dev, adapter->fcoe.up)))
4687                 tc = IXGBE_FCOE_JUMBO_FRAME_SIZE;
4688 #endif
4689
4690         /* Calculate delay value for device */
4691         switch (hw->mac.type) {
4692         case ixgbe_mac_X540:
4693         case ixgbe_mac_X550:
4694         case ixgbe_mac_X550EM_x:
4695                 dv_id = IXGBE_LOW_DV_X540(tc);
4696                 break;
4697         default:
4698                 dv_id = IXGBE_LOW_DV(tc);
4699                 break;
4700         }
4701
4702         /* Delay value is calculated in bit times convert to KB */
4703         return IXGBE_BT2KB(dv_id);
4704 }
4705
4706 /*
4707  * ixgbe_pbthresh_setup - calculate and setup high low water marks
4708  */
4709 static void ixgbe_pbthresh_setup(struct ixgbe_adapter *adapter)
4710 {
4711         struct ixgbe_hw *hw = &adapter->hw;
4712         int num_tc = netdev_get_num_tc(adapter->netdev);
4713         int i;
4714
4715         if (!num_tc)
4716                 num_tc = 1;
4717
4718         for (i = 0; i < num_tc; i++) {
4719                 hw->fc.high_water[i] = ixgbe_hpbthresh(adapter, i);
4720                 hw->fc.low_water[i] = ixgbe_lpbthresh(adapter, i);
4721
4722                 /* Low water marks must not be larger than high water marks */
4723                 if (hw->fc.low_water[i] > hw->fc.high_water[i])
4724                         hw->fc.low_water[i] = 0;
4725         }
4726
4727         for (; i < MAX_TRAFFIC_CLASS; i++)
4728                 hw->fc.high_water[i] = 0;
4729 }
4730
4731 static void ixgbe_configure_pb(struct ixgbe_adapter *adapter)
4732 {
4733         struct ixgbe_hw *hw = &adapter->hw;
4734         int hdrm;
4735         u8 tc = netdev_get_num_tc(adapter->netdev);
4736
4737         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE ||
4738             adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
4739                 hdrm = 32 << adapter->fdir_pballoc;
4740         else
4741                 hdrm = 0;
4742
4743         hw->mac.ops.set_rxpba(hw, tc, hdrm, PBA_STRATEGY_EQUAL);
4744         ixgbe_pbthresh_setup(adapter);
4745 }
4746
4747 static void ixgbe_fdir_filter_restore(struct ixgbe_adapter *adapter)
4748 {
4749         struct ixgbe_hw *hw = &adapter->hw;
4750         struct hlist_node *node2;
4751         struct ixgbe_fdir_filter *filter;
4752
4753         spin_lock(&adapter->fdir_perfect_lock);
4754
4755         if (!hlist_empty(&adapter->fdir_filter_list))
4756                 ixgbe_fdir_set_input_mask_82599(hw, &adapter->fdir_mask);
4757
4758         hlist_for_each_entry_safe(filter, node2,
4759                                   &adapter->fdir_filter_list, fdir_node) {
4760                 ixgbe_fdir_write_perfect_filter_82599(hw,
4761                                 &filter->filter,
4762                                 filter->sw_idx,
4763                                 (filter->action == IXGBE_FDIR_DROP_QUEUE) ?
4764                                 IXGBE_FDIR_DROP_QUEUE :
4765                                 adapter->rx_ring[filter->action]->reg_idx);
4766         }
4767
4768         spin_unlock(&adapter->fdir_perfect_lock);
4769 }
4770
4771 static void ixgbe_macvlan_set_rx_mode(struct net_device *dev, unsigned int pool,
4772                                       struct ixgbe_adapter *adapter)
4773 {
4774         struct ixgbe_hw *hw = &adapter->hw;
4775         u32 vmolr;
4776
4777         /* No unicast promiscuous support for VMDQ devices. */
4778         vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(pool));
4779         vmolr |= (IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
4780
4781         /* clear the affected bit */
4782         vmolr &= ~IXGBE_VMOLR_MPE;
4783
4784         if (dev->flags & IFF_ALLMULTI) {
4785                 vmolr |= IXGBE_VMOLR_MPE;
4786         } else {
4787                 vmolr |= IXGBE_VMOLR_ROMPE;
4788                 hw->mac.ops.update_mc_addr_list(hw, dev);
4789         }
4790         ixgbe_write_uc_addr_list(adapter->netdev, pool);
4791         IXGBE_WRITE_REG(hw, IXGBE_VMOLR(pool), vmolr);
4792 }
4793
4794 static void ixgbe_fwd_psrtype(struct ixgbe_fwd_adapter *vadapter)
4795 {
4796         struct ixgbe_adapter *adapter = vadapter->real_adapter;
4797         int rss_i = adapter->num_rx_queues_per_pool;
4798         struct ixgbe_hw *hw = &adapter->hw;
4799         u16 pool = vadapter->pool;
4800         u32 psrtype = IXGBE_PSRTYPE_TCPHDR |
4801                       IXGBE_PSRTYPE_UDPHDR |
4802                       IXGBE_PSRTYPE_IPV4HDR |
4803                       IXGBE_PSRTYPE_L2HDR |
4804                       IXGBE_PSRTYPE_IPV6HDR;
4805
4806         if (hw->mac.type == ixgbe_mac_82598EB)
4807                 return;
4808
4809         if (rss_i > 3)
4810                 psrtype |= 2 << 29;
4811         else if (rss_i > 1)
4812                 psrtype |= 1 << 29;
4813
4814         IXGBE_WRITE_REG(hw, IXGBE_PSRTYPE(VMDQ_P(pool)), psrtype);
4815 }
4816
4817 /**
4818  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
4819  * @rx_ring: ring to free buffers from
4820  **/
4821 static void ixgbe_clean_rx_ring(struct ixgbe_ring *rx_ring)
4822 {
4823         struct device *dev = rx_ring->dev;
4824         unsigned long size;
4825         u16 i;
4826
4827         /* ring already cleared, nothing to do */
4828         if (!rx_ring->rx_buffer_info)
4829                 return;
4830
4831         /* Free all the Rx ring sk_buffs */
4832         for (i = 0; i < rx_ring->count; i++) {
4833                 struct ixgbe_rx_buffer *rx_buffer = &rx_ring->rx_buffer_info[i];
4834
4835                 if (rx_buffer->skb) {
4836                         struct sk_buff *skb = rx_buffer->skb;
4837                         if (IXGBE_CB(skb)->page_released)
4838                                 dma_unmap_page(dev,
4839                                                IXGBE_CB(skb)->dma,
4840                                                ixgbe_rx_bufsz(rx_ring),
4841                                                DMA_FROM_DEVICE);
4842                         dev_kfree_skb(skb);
4843                         rx_buffer->skb = NULL;
4844                 }
4845
4846                 if (!rx_buffer->page)
4847                         continue;
4848
4849                 dma_unmap_page(dev, rx_buffer->dma,
4850                                ixgbe_rx_pg_size(rx_ring), DMA_FROM_DEVICE);
4851                 __free_pages(rx_buffer->page, ixgbe_rx_pg_order(rx_ring));
4852
4853                 rx_buffer->page = NULL;
4854         }
4855
4856         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
4857         memset(rx_ring->rx_buffer_info, 0, size);
4858
4859         /* Zero out the descriptor ring */
4860         memset(rx_ring->desc, 0, rx_ring->size);
4861
4862         rx_ring->next_to_alloc = 0;
4863         rx_ring->next_to_clean = 0;
4864         rx_ring->next_to_use = 0;
4865 }
4866
4867 static void ixgbe_disable_fwd_ring(struct ixgbe_fwd_adapter *vadapter,
4868                                    struct ixgbe_ring *rx_ring)
4869 {
4870         struct ixgbe_adapter *adapter = vadapter->real_adapter;
4871         int index = rx_ring->queue_index + vadapter->rx_base_queue;
4872
4873         /* shutdown specific queue receive and wait for dma to settle */
4874         ixgbe_disable_rx_queue(adapter, rx_ring);
4875         usleep_range(10000, 20000);
4876         ixgbe_irq_disable_queues(adapter, ((u64)1 << index));
4877         ixgbe_clean_rx_ring(rx_ring);
4878         rx_ring->l2_accel_priv = NULL;
4879 }
4880
4881 static int ixgbe_fwd_ring_down(struct net_device *vdev,
4882                                struct ixgbe_fwd_adapter *accel)
4883 {
4884         struct ixgbe_adapter *adapter = accel->real_adapter;
4885         unsigned int rxbase = accel->rx_base_queue;
4886         unsigned int txbase = accel->tx_base_queue;
4887         int i;
4888
4889         netif_tx_stop_all_queues(vdev);
4890
4891         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4892                 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4893                 adapter->rx_ring[rxbase + i]->netdev = adapter->netdev;
4894         }
4895
4896         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4897                 adapter->tx_ring[txbase + i]->l2_accel_priv = NULL;
4898                 adapter->tx_ring[txbase + i]->netdev = adapter->netdev;
4899         }
4900
4901
4902         return 0;
4903 }
4904
4905 static int ixgbe_fwd_ring_up(struct net_device *vdev,
4906                              struct ixgbe_fwd_adapter *accel)
4907 {
4908         struct ixgbe_adapter *adapter = accel->real_adapter;
4909         unsigned int rxbase, txbase, queues;
4910         int i, baseq, err = 0;
4911
4912         if (!test_bit(accel->pool, &adapter->fwd_bitmask))
4913                 return 0;
4914
4915         baseq = accel->pool * adapter->num_rx_queues_per_pool;
4916         netdev_dbg(vdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
4917                    accel->pool, adapter->num_rx_pools,
4918                    baseq, baseq + adapter->num_rx_queues_per_pool,
4919                    adapter->fwd_bitmask);
4920
4921         accel->netdev = vdev;
4922         accel->rx_base_queue = rxbase = baseq;
4923         accel->tx_base_queue = txbase = baseq;
4924
4925         for (i = 0; i < adapter->num_rx_queues_per_pool; i++)
4926                 ixgbe_disable_fwd_ring(accel, adapter->rx_ring[rxbase + i]);
4927
4928         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4929                 adapter->rx_ring[rxbase + i]->netdev = vdev;
4930                 adapter->rx_ring[rxbase + i]->l2_accel_priv = accel;
4931                 ixgbe_configure_rx_ring(adapter, adapter->rx_ring[rxbase + i]);
4932         }
4933
4934         for (i = 0; i < adapter->num_rx_queues_per_pool; i++) {
4935                 adapter->tx_ring[txbase + i]->netdev = vdev;
4936                 adapter->tx_ring[txbase + i]->l2_accel_priv = accel;
4937         }
4938
4939         queues = min_t(unsigned int,
4940                        adapter->num_rx_queues_per_pool, vdev->num_tx_queues);
4941         err = netif_set_real_num_tx_queues(vdev, queues);
4942         if (err)
4943                 goto fwd_queue_err;
4944
4945         err = netif_set_real_num_rx_queues(vdev, queues);
4946         if (err)
4947                 goto fwd_queue_err;
4948
4949         if (is_valid_ether_addr(vdev->dev_addr))
4950                 ixgbe_add_mac_filter(adapter, vdev->dev_addr, accel->pool);
4951
4952         ixgbe_fwd_psrtype(accel);
4953         ixgbe_macvlan_set_rx_mode(vdev, accel->pool, adapter);
4954         return err;
4955 fwd_queue_err:
4956         ixgbe_fwd_ring_down(vdev, accel);
4957         return err;
4958 }
4959
4960 static void ixgbe_configure_dfwd(struct ixgbe_adapter *adapter)
4961 {
4962         struct net_device *upper;
4963         struct list_head *iter;
4964         int err;
4965
4966         netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
4967                 if (netif_is_macvlan(upper)) {
4968                         struct macvlan_dev *dfwd = netdev_priv(upper);
4969                         struct ixgbe_fwd_adapter *vadapter = dfwd->fwd_priv;
4970
4971                         if (dfwd->fwd_priv) {
4972                                 err = ixgbe_fwd_ring_up(upper, vadapter);
4973                                 if (err)
4974                                         continue;
4975                         }
4976                 }
4977         }
4978 }
4979
4980 static void ixgbe_configure(struct ixgbe_adapter *adapter)
4981 {
4982         struct ixgbe_hw *hw = &adapter->hw;
4983
4984         ixgbe_configure_pb(adapter);
4985 #ifdef CONFIG_IXGBE_DCB
4986         ixgbe_configure_dcb(adapter);
4987 #endif
4988         /*
4989          * We must restore virtualization before VLANs or else
4990          * the VLVF registers will not be populated
4991          */
4992         ixgbe_configure_virtualization(adapter);
4993
4994         ixgbe_set_rx_mode(adapter->netdev);
4995         ixgbe_restore_vlan(adapter);
4996
4997         switch (hw->mac.type) {
4998         case ixgbe_mac_82599EB:
4999         case ixgbe_mac_X540:
5000                 hw->mac.ops.disable_rx_buff(hw);
5001                 break;
5002         default:
5003                 break;
5004         }
5005
5006         if (adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE) {
5007                 ixgbe_init_fdir_signature_82599(&adapter->hw,
5008                                                 adapter->fdir_pballoc);
5009         } else if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE) {
5010                 ixgbe_init_fdir_perfect_82599(&adapter->hw,
5011                                               adapter->fdir_pballoc);
5012                 ixgbe_fdir_filter_restore(adapter);
5013         }
5014
5015         switch (hw->mac.type) {
5016         case ixgbe_mac_82599EB:
5017         case ixgbe_mac_X540:
5018                 hw->mac.ops.enable_rx_buff(hw);
5019                 break;
5020         default:
5021                 break;
5022         }
5023
5024 #ifdef CONFIG_IXGBE_DCA
5025         /* configure DCA */
5026         if (adapter->flags & IXGBE_FLAG_DCA_CAPABLE)
5027                 ixgbe_setup_dca(adapter);
5028 #endif /* CONFIG_IXGBE_DCA */
5029
5030 #ifdef IXGBE_FCOE
5031         /* configure FCoE L2 filters, redirection table, and Rx control */
5032         ixgbe_configure_fcoe(adapter);
5033
5034 #endif /* IXGBE_FCOE */
5035         ixgbe_configure_tx(adapter);
5036         ixgbe_configure_rx(adapter);
5037         ixgbe_configure_dfwd(adapter);
5038 }
5039
5040 /**
5041  * ixgbe_sfp_link_config - set up SFP+ link
5042  * @adapter: pointer to private adapter struct
5043  **/
5044 static void ixgbe_sfp_link_config(struct ixgbe_adapter *adapter)
5045 {
5046         /*
5047          * We are assuming the worst case scenario here, and that
5048          * is that an SFP was inserted/removed after the reset
5049          * but before SFP detection was enabled.  As such the best
5050          * solution is to just start searching as soon as we start
5051          */
5052         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
5053                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
5054
5055         adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
5056         adapter->sfp_poll_time = 0;
5057 }
5058
5059 /**
5060  * ixgbe_non_sfp_link_config - set up non-SFP+ link
5061  * @hw: pointer to private hardware struct
5062  *
5063  * Returns 0 on success, negative on failure
5064  **/
5065 static int ixgbe_non_sfp_link_config(struct ixgbe_hw *hw)
5066 {
5067         u32 speed;
5068         bool autoneg, link_up = false;
5069         int ret = IXGBE_ERR_LINK_SETUP;
5070
5071         if (hw->mac.ops.check_link)
5072                 ret = hw->mac.ops.check_link(hw, &speed, &link_up, false);
5073
5074         if (ret)
5075                 return ret;
5076
5077         speed = hw->phy.autoneg_advertised;
5078         if ((!speed) && (hw->mac.ops.get_link_capabilities))
5079                 ret = hw->mac.ops.get_link_capabilities(hw, &speed,
5080                                                         &autoneg);
5081         if (ret)
5082                 return ret;
5083
5084         if (hw->mac.ops.setup_link)
5085                 ret = hw->mac.ops.setup_link(hw, speed, link_up);
5086
5087         return ret;
5088 }
5089
5090 static void ixgbe_setup_gpie(struct ixgbe_adapter *adapter)
5091 {
5092         struct ixgbe_hw *hw = &adapter->hw;
5093         u32 gpie = 0;
5094
5095         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
5096                 gpie = IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT |
5097                        IXGBE_GPIE_OCD;
5098                 gpie |= IXGBE_GPIE_EIAME;
5099                 /*
5100                  * use EIAM to auto-mask when MSI-X interrupt is asserted
5101                  * this saves a register write for every interrupt
5102                  */
5103                 switch (hw->mac.type) {
5104                 case ixgbe_mac_82598EB:
5105                         IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5106                         break;
5107                 case ixgbe_mac_82599EB:
5108                 case ixgbe_mac_X540:
5109                 case ixgbe_mac_X550:
5110                 case ixgbe_mac_X550EM_x:
5111                 default:
5112                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(0), 0xFFFFFFFF);
5113                         IXGBE_WRITE_REG(hw, IXGBE_EIAM_EX(1), 0xFFFFFFFF);
5114                         break;
5115                 }
5116         } else {
5117                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
5118                  * specifically only auto mask tx and rx interrupts */
5119                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
5120         }
5121
5122         /* XXX: to interrupt immediately for EICS writes, enable this */
5123         /* gpie |= IXGBE_GPIE_EIMEN; */
5124
5125         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
5126                 gpie &= ~IXGBE_GPIE_VTMODE_MASK;
5127
5128                 switch (adapter->ring_feature[RING_F_VMDQ].mask) {
5129                 case IXGBE_82599_VMDQ_8Q_MASK:
5130                         gpie |= IXGBE_GPIE_VTMODE_16;
5131                         break;
5132                 case IXGBE_82599_VMDQ_4Q_MASK:
5133                         gpie |= IXGBE_GPIE_VTMODE_32;
5134                         break;
5135                 default:
5136                         gpie |= IXGBE_GPIE_VTMODE_64;
5137                         break;
5138                 }
5139         }
5140
5141         /* Enable Thermal over heat sensor interrupt */
5142         if (adapter->flags2 & IXGBE_FLAG2_TEMP_SENSOR_CAPABLE) {
5143                 switch (adapter->hw.mac.type) {
5144                 case ixgbe_mac_82599EB:
5145                         gpie |= IXGBE_SDP0_GPIEN_8259X;
5146                         break;
5147                 default:
5148                         break;
5149                 }
5150         }
5151
5152         /* Enable fan failure interrupt */
5153         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
5154                 gpie |= IXGBE_SDP1_GPIEN(hw);
5155
5156         switch (hw->mac.type) {
5157         case ixgbe_mac_82599EB:
5158                 gpie |= IXGBE_SDP1_GPIEN_8259X | IXGBE_SDP2_GPIEN_8259X;
5159                 break;
5160         case ixgbe_mac_X550EM_x:
5161                 gpie |= IXGBE_SDP0_GPIEN_X540;
5162                 break;
5163         default:
5164                 break;
5165         }
5166
5167         IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
5168 }
5169
5170 static void ixgbe_up_complete(struct ixgbe_adapter *adapter)
5171 {
5172         struct ixgbe_hw *hw = &adapter->hw;
5173         int err;
5174         u32 ctrl_ext;
5175
5176         ixgbe_get_hw_control(adapter);
5177         ixgbe_setup_gpie(adapter);
5178
5179         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
5180                 ixgbe_configure_msix(adapter);
5181         else
5182                 ixgbe_configure_msi_and_legacy(adapter);
5183
5184         /* enable the optics for 82599 SFP+ fiber */
5185         if (hw->mac.ops.enable_tx_laser)
5186                 hw->mac.ops.enable_tx_laser(hw);
5187
5188         if (hw->phy.ops.set_phy_power)
5189                 hw->phy.ops.set_phy_power(hw, true);
5190
5191         smp_mb__before_atomic();
5192         clear_bit(__IXGBE_DOWN, &adapter->state);
5193         ixgbe_napi_enable_all(adapter);
5194
5195         if (ixgbe_is_sfp(hw)) {
5196                 ixgbe_sfp_link_config(adapter);
5197         } else {
5198                 err = ixgbe_non_sfp_link_config(hw);
5199                 if (err)
5200                         e_err(probe, "link_config FAILED %d\n", err);
5201         }
5202
5203         /* clear any pending interrupts, may auto mask */
5204         IXGBE_READ_REG(hw, IXGBE_EICR);
5205         ixgbe_irq_enable(adapter, true, true);
5206
5207         /*
5208          * If this adapter has a fan, check to see if we had a failure
5209          * before we enabled the interrupt.
5210          */
5211         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
5212                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
5213                 if (esdp & IXGBE_ESDP_SDP1)
5214                         e_crit(drv, "Fan has stopped, replace the adapter\n");
5215         }
5216
5217         /* bring the link up in the watchdog, this could race with our first
5218          * link up interrupt but shouldn't be a problem */
5219         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
5220         adapter->link_check_timeout = jiffies;
5221         mod_timer(&adapter->service_timer, jiffies);
5222
5223         /* Set PF Reset Done bit so PF/VF Mail Ops can work */
5224         ctrl_ext = IXGBE_READ_REG(hw, IXGBE_CTRL_EXT);
5225         ctrl_ext |= IXGBE_CTRL_EXT_PFRSTD;
5226         IXGBE_WRITE_REG(hw, IXGBE_CTRL_EXT, ctrl_ext);
5227 }
5228
5229 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
5230 {
5231         WARN_ON(in_interrupt());
5232         /* put off any impending NetWatchDogTimeout */
5233         adapter->netdev->trans_start = jiffies;
5234
5235         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
5236                 usleep_range(1000, 2000);
5237         ixgbe_down(adapter);
5238         /*
5239          * If SR-IOV enabled then wait a bit before bringing the adapter
5240          * back up to give the VFs time to respond to the reset.  The
5241          * two second wait is based upon the watchdog timer cycle in
5242          * the VF driver.
5243          */
5244         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
5245                 msleep(2000);
5246         ixgbe_up(adapter);
5247         clear_bit(__IXGBE_RESETTING, &adapter->state);
5248 }
5249
5250 void ixgbe_up(struct ixgbe_adapter *adapter)
5251 {
5252         /* hardware has been reset, we need to reload some things */
5253         ixgbe_configure(adapter);
5254
5255         ixgbe_up_complete(adapter);
5256 }
5257
5258 void ixgbe_reset(struct ixgbe_adapter *adapter)
5259 {
5260         struct ixgbe_hw *hw = &adapter->hw;
5261         struct net_device *netdev = adapter->netdev;
5262         int err;
5263
5264         if (ixgbe_removed(hw->hw_addr))
5265                 return;
5266         /* lock SFP init bit to prevent race conditions with the watchdog */
5267         while (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
5268                 usleep_range(1000, 2000);
5269
5270         /* clear all SFP and link config related flags while holding SFP_INIT */
5271         adapter->flags2 &= ~(IXGBE_FLAG2_SEARCH_FOR_SFP |
5272                              IXGBE_FLAG2_SFP_NEEDS_RESET);
5273         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
5274
5275         err = hw->mac.ops.init_hw(hw);
5276         switch (err) {
5277         case 0:
5278         case IXGBE_ERR_SFP_NOT_PRESENT:
5279         case IXGBE_ERR_SFP_NOT_SUPPORTED:
5280                 break;
5281         case IXGBE_ERR_MASTER_REQUESTS_PENDING:
5282                 e_dev_err("master disable timed out\n");
5283                 break;
5284         case IXGBE_ERR_EEPROM_VERSION:
5285                 /* We are running on a pre-production device, log a warning */
5286                 e_dev_warn("This device is a pre-production adapter/LOM. "
5287                            "Please be aware there may be issues associated with "
5288                            "your hardware.  If you are experiencing problems "
5289                            "please contact your Intel or hardware "
5290                            "representative who provided you with this "
5291                            "hardware.\n");
5292                 break;
5293         default:
5294                 e_dev_err("Hardware Error: %d\n", err);
5295         }
5296
5297         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
5298
5299         /* flush entries out of MAC table */
5300         ixgbe_flush_sw_mac_table(adapter);
5301         __dev_uc_unsync(netdev, NULL);
5302
5303         /* do not flush user set addresses */
5304         ixgbe_mac_set_default_filter(adapter);
5305
5306         /* update SAN MAC vmdq pool selection */
5307         if (hw->mac.san_mac_rar_index)
5308                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
5309
5310         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
5311                 ixgbe_ptp_reset(adapter);
5312
5313         if (hw->phy.ops.set_phy_power) {
5314                 if (!netif_running(adapter->netdev) && !adapter->wol)
5315                         hw->phy.ops.set_phy_power(hw, false);
5316                 else
5317                         hw->phy.ops.set_phy_power(hw, true);
5318         }
5319 }
5320
5321 /**
5322  * ixgbe_clean_tx_ring - Free Tx Buffers
5323  * @tx_ring: ring to be cleaned
5324  **/
5325 static void ixgbe_clean_tx_ring(struct ixgbe_ring *tx_ring)
5326 {
5327         struct ixgbe_tx_buffer *tx_buffer_info;
5328         unsigned long size;
5329         u16 i;
5330
5331         /* ring already cleared, nothing to do */
5332         if (!tx_ring->tx_buffer_info)
5333                 return;
5334
5335         /* Free all the Tx ring sk_buffs */
5336         for (i = 0; i < tx_ring->count; i++) {
5337                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
5338                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
5339         }
5340
5341         netdev_tx_reset_queue(txring_txq(tx_ring));
5342
5343         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5344         memset(tx_ring->tx_buffer_info, 0, size);
5345
5346         /* Zero out the descriptor ring */
5347         memset(tx_ring->desc, 0, tx_ring->size);
5348
5349         tx_ring->next_to_use = 0;
5350         tx_ring->next_to_clean = 0;
5351 }
5352
5353 /**
5354  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
5355  * @adapter: board private structure
5356  **/
5357 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
5358 {
5359         int i;
5360
5361         for (i = 0; i < adapter->num_rx_queues; i++)
5362                 ixgbe_clean_rx_ring(adapter->rx_ring[i]);
5363 }
5364
5365 /**
5366  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
5367  * @adapter: board private structure
5368  **/
5369 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
5370 {
5371         int i;
5372
5373         for (i = 0; i < adapter->num_tx_queues; i++)
5374                 ixgbe_clean_tx_ring(adapter->tx_ring[i]);
5375 }
5376
5377 static void ixgbe_fdir_filter_exit(struct ixgbe_adapter *adapter)
5378 {
5379         struct hlist_node *node2;
5380         struct ixgbe_fdir_filter *filter;
5381
5382         spin_lock(&adapter->fdir_perfect_lock);
5383
5384         hlist_for_each_entry_safe(filter, node2,
5385                                   &adapter->fdir_filter_list, fdir_node) {
5386                 hlist_del(&filter->fdir_node);
5387                 kfree(filter);
5388         }
5389         adapter->fdir_filter_count = 0;
5390
5391         spin_unlock(&adapter->fdir_perfect_lock);
5392 }
5393
5394 void ixgbe_down(struct ixgbe_adapter *adapter)
5395 {
5396         struct net_device *netdev = adapter->netdev;
5397         struct ixgbe_hw *hw = &adapter->hw;
5398         struct net_device *upper;
5399         struct list_head *iter;
5400         int i;
5401
5402         /* signal that we are down to the interrupt handler */
5403         if (test_and_set_bit(__IXGBE_DOWN, &adapter->state))
5404                 return; /* do nothing if already down */
5405
5406         /* disable receives */
5407         hw->mac.ops.disable_rx(hw);
5408
5409         /* disable all enabled rx queues */
5410         for (i = 0; i < adapter->num_rx_queues; i++)
5411                 /* this call also flushes the previous write */
5412                 ixgbe_disable_rx_queue(adapter, adapter->rx_ring[i]);
5413
5414         usleep_range(10000, 20000);
5415
5416         netif_tx_stop_all_queues(netdev);
5417
5418         /* call carrier off first to avoid false dev_watchdog timeouts */
5419         netif_carrier_off(netdev);
5420         netif_tx_disable(netdev);
5421
5422         /* disable any upper devices */
5423         netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
5424                 if (netif_is_macvlan(upper)) {
5425                         struct macvlan_dev *vlan = netdev_priv(upper);
5426
5427                         if (vlan->fwd_priv) {
5428                                 netif_tx_stop_all_queues(upper);
5429                                 netif_carrier_off(upper);
5430                                 netif_tx_disable(upper);
5431                         }
5432                 }
5433         }
5434
5435         ixgbe_irq_disable(adapter);
5436
5437         ixgbe_napi_disable_all(adapter);
5438
5439         adapter->flags2 &= ~(IXGBE_FLAG2_FDIR_REQUIRES_REINIT |
5440                              IXGBE_FLAG2_RESET_REQUESTED);
5441         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
5442
5443         del_timer_sync(&adapter->service_timer);
5444
5445         if (adapter->num_vfs) {
5446                 /* Clear EITR Select mapping */
5447                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITRSEL, 0);
5448
5449                 /* Mark all the VFs as inactive */
5450                 for (i = 0 ; i < adapter->num_vfs; i++)
5451                         adapter->vfinfo[i].clear_to_send = false;
5452
5453                 /* ping all the active vfs to let them know we are going down */
5454                 ixgbe_ping_all_vfs(adapter);
5455
5456                 /* Disable all VFTE/VFRE TX/RX */
5457                 ixgbe_disable_tx_rx(adapter);
5458         }
5459
5460         /* disable transmits in the hardware now that interrupts are off */
5461         for (i = 0; i < adapter->num_tx_queues; i++) {
5462                 u8 reg_idx = adapter->tx_ring[i]->reg_idx;
5463                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(reg_idx), IXGBE_TXDCTL_SWFLSH);
5464         }
5465
5466         /* Disable the Tx DMA engine on 82599 and later MAC */
5467         switch (hw->mac.type) {
5468         case ixgbe_mac_82599EB:
5469         case ixgbe_mac_X540:
5470         case ixgbe_mac_X550:
5471         case ixgbe_mac_X550EM_x:
5472                 IXGBE_WRITE_REG(hw, IXGBE_DMATXCTL,
5473                                 (IXGBE_READ_REG(hw, IXGBE_DMATXCTL) &
5474                                  ~IXGBE_DMATXCTL_TE));
5475                 break;
5476         default:
5477                 break;
5478         }
5479
5480         if (!pci_channel_offline(adapter->pdev))
5481                 ixgbe_reset(adapter);
5482
5483         /* power down the optics for 82599 SFP+ fiber */
5484         if (hw->mac.ops.disable_tx_laser)
5485                 hw->mac.ops.disable_tx_laser(hw);
5486
5487         ixgbe_clean_all_tx_rings(adapter);
5488         ixgbe_clean_all_rx_rings(adapter);
5489 }
5490
5491 /**
5492  * ixgbe_tx_timeout - Respond to a Tx Hang
5493  * @netdev: network interface device structure
5494  **/
5495 static void ixgbe_tx_timeout(struct net_device *netdev)
5496 {
5497         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5498
5499         /* Do the reset outside of interrupt context */
5500         ixgbe_tx_timeout_reset(adapter);
5501 }
5502
5503 /**
5504  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
5505  * @adapter: board private structure to initialize
5506  *
5507  * ixgbe_sw_init initializes the Adapter private data structure.
5508  * Fields are initialized based on PCI device information and
5509  * OS network device settings (MTU size).
5510  **/
5511 static int ixgbe_sw_init(struct ixgbe_adapter *adapter)
5512 {
5513         struct ixgbe_hw *hw = &adapter->hw;
5514         struct pci_dev *pdev = adapter->pdev;
5515         unsigned int rss, fdir;
5516         u32 fwsm;
5517 #ifdef CONFIG_IXGBE_DCB
5518         int j;
5519         struct tc_configuration *tc;
5520 #endif
5521
5522         /* PCI config space info */
5523
5524         hw->vendor_id = pdev->vendor;
5525         hw->device_id = pdev->device;
5526         hw->revision_id = pdev->revision;
5527         hw->subsystem_vendor_id = pdev->subsystem_vendor;
5528         hw->subsystem_device_id = pdev->subsystem_device;
5529
5530         /* Set common capability flags and settings */
5531         rss = min_t(int, ixgbe_max_rss_indices(adapter), num_online_cpus());
5532         adapter->ring_feature[RING_F_RSS].limit = rss;
5533         adapter->flags2 |= IXGBE_FLAG2_RSC_CAPABLE;
5534         adapter->max_q_vectors = MAX_Q_VECTORS_82599;
5535         adapter->atr_sample_rate = 20;
5536         fdir = min_t(int, IXGBE_MAX_FDIR_INDICES, num_online_cpus());
5537         adapter->ring_feature[RING_F_FDIR].limit = fdir;
5538         adapter->fdir_pballoc = IXGBE_FDIR_PBALLOC_64K;
5539 #ifdef CONFIG_IXGBE_DCA
5540         adapter->flags |= IXGBE_FLAG_DCA_CAPABLE;
5541 #endif
5542 #ifdef IXGBE_FCOE
5543         adapter->flags |= IXGBE_FLAG_FCOE_CAPABLE;
5544         adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5545 #ifdef CONFIG_IXGBE_DCB
5546         /* Default traffic class to use for FCoE */
5547         adapter->fcoe.up = IXGBE_FCOE_DEFTC;
5548 #endif /* CONFIG_IXGBE_DCB */
5549 #endif /* IXGBE_FCOE */
5550
5551         /* initialize static ixgbe jump table entries */
5552         adapter->jump_tables[0] = ixgbe_ipv4_fields;
5553
5554         adapter->mac_table = kzalloc(sizeof(struct ixgbe_mac_addr) *
5555                                      hw->mac.num_rar_entries,
5556                                      GFP_ATOMIC);
5557         if (!adapter->mac_table)
5558                 return -ENOMEM;
5559
5560         /* Set MAC specific capability flags and exceptions */
5561         switch (hw->mac.type) {
5562         case ixgbe_mac_82598EB:
5563                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_CAPABLE;
5564
5565                 if (hw->device_id == IXGBE_DEV_ID_82598AT)
5566                         adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
5567
5568                 adapter->max_q_vectors = MAX_Q_VECTORS_82598;
5569                 adapter->ring_feature[RING_F_FDIR].limit = 0;
5570                 adapter->atr_sample_rate = 0;
5571                 adapter->fdir_pballoc = 0;
5572 #ifdef IXGBE_FCOE
5573                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
5574                 adapter->flags &= ~IXGBE_FLAG_FCOE_ENABLED;
5575 #ifdef CONFIG_IXGBE_DCB
5576                 adapter->fcoe.up = 0;
5577 #endif /* IXGBE_DCB */
5578 #endif /* IXGBE_FCOE */
5579                 break;
5580         case ixgbe_mac_82599EB:
5581                 if (hw->device_id == IXGBE_DEV_ID_82599_T3_LOM)
5582                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5583                 break;
5584         case ixgbe_mac_X540:
5585                 fwsm = IXGBE_READ_REG(hw, IXGBE_FWSM(hw));
5586                 if (fwsm & IXGBE_FWSM_TS_ENABLED)
5587                         adapter->flags2 |= IXGBE_FLAG2_TEMP_SENSOR_CAPABLE;
5588                 break;
5589         case ixgbe_mac_X550EM_x:
5590         case ixgbe_mac_X550:
5591 #ifdef CONFIG_IXGBE_DCA
5592                 adapter->flags &= ~IXGBE_FLAG_DCA_CAPABLE;
5593 #endif
5594 #ifdef CONFIG_IXGBE_VXLAN
5595                 adapter->flags |= IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE;
5596 #endif
5597                 break;
5598         default:
5599                 break;
5600         }
5601
5602 #ifdef IXGBE_FCOE
5603         /* FCoE support exists, always init the FCoE lock */
5604         spin_lock_init(&adapter->fcoe.lock);
5605
5606 #endif
5607         /* n-tuple support exists, always init our spinlock */
5608         spin_lock_init(&adapter->fdir_perfect_lock);
5609
5610 #ifdef CONFIG_IXGBE_DCB
5611         switch (hw->mac.type) {
5612         case ixgbe_mac_X540:
5613         case ixgbe_mac_X550:
5614         case ixgbe_mac_X550EM_x:
5615                 adapter->dcb_cfg.num_tcs.pg_tcs = X540_TRAFFIC_CLASS;
5616                 adapter->dcb_cfg.num_tcs.pfc_tcs = X540_TRAFFIC_CLASS;
5617                 break;
5618         default:
5619                 adapter->dcb_cfg.num_tcs.pg_tcs = MAX_TRAFFIC_CLASS;
5620                 adapter->dcb_cfg.num_tcs.pfc_tcs = MAX_TRAFFIC_CLASS;
5621                 break;
5622         }
5623
5624         /* Configure DCB traffic classes */
5625         for (j = 0; j < MAX_TRAFFIC_CLASS; j++) {
5626                 tc = &adapter->dcb_cfg.tc_config[j];
5627                 tc->path[DCB_TX_CONFIG].bwg_id = 0;
5628                 tc->path[DCB_TX_CONFIG].bwg_percent = 12 + (j & 1);
5629                 tc->path[DCB_RX_CONFIG].bwg_id = 0;
5630                 tc->path[DCB_RX_CONFIG].bwg_percent = 12 + (j & 1);
5631                 tc->dcb_pfc = pfc_disabled;
5632         }
5633
5634         /* Initialize default user to priority mapping, UPx->TC0 */
5635         tc = &adapter->dcb_cfg.tc_config[0];
5636         tc->path[DCB_TX_CONFIG].up_to_tc_bitmap = 0xFF;
5637         tc->path[DCB_RX_CONFIG].up_to_tc_bitmap = 0xFF;
5638
5639         adapter->dcb_cfg.bw_percentage[DCB_TX_CONFIG][0] = 100;
5640         adapter->dcb_cfg.bw_percentage[DCB_RX_CONFIG][0] = 100;
5641         adapter->dcb_cfg.pfc_mode_enable = false;
5642         adapter->dcb_set_bitmap = 0x00;
5643         adapter->dcbx_cap = DCB_CAP_DCBX_HOST | DCB_CAP_DCBX_VER_CEE;
5644         memcpy(&adapter->temp_dcb_cfg, &adapter->dcb_cfg,
5645                sizeof(adapter->temp_dcb_cfg));
5646
5647 #endif
5648
5649         /* default flow control settings */
5650         hw->fc.requested_mode = ixgbe_fc_full;
5651         hw->fc.current_mode = ixgbe_fc_full;    /* init for ethtool output */
5652         ixgbe_pbthresh_setup(adapter);
5653         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
5654         hw->fc.send_xon = true;
5655         hw->fc.disable_fc_autoneg = ixgbe_device_supports_autoneg_fc(hw);
5656
5657 #ifdef CONFIG_PCI_IOV
5658         if (max_vfs > 0)
5659                 e_dev_warn("Enabling SR-IOV VFs using the max_vfs module parameter is deprecated - please use the pci sysfs interface instead.\n");
5660
5661         /* assign number of SR-IOV VFs */
5662         if (hw->mac.type != ixgbe_mac_82598EB) {
5663                 if (max_vfs > IXGBE_MAX_VFS_DRV_LIMIT) {
5664                         adapter->num_vfs = 0;
5665                         e_dev_warn("max_vfs parameter out of range. Not assigning any SR-IOV VFs\n");
5666                 } else {
5667                         adapter->num_vfs = max_vfs;
5668                 }
5669         }
5670 #endif /* CONFIG_PCI_IOV */
5671
5672         /* enable itr by default in dynamic mode */
5673         adapter->rx_itr_setting = 1;
5674         adapter->tx_itr_setting = 1;
5675
5676         /* set default ring sizes */
5677         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
5678         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
5679
5680         /* set default work limits */
5681         adapter->tx_work_limit = IXGBE_DEFAULT_TX_WORK;
5682
5683         /* initialize eeprom parameters */
5684         if (ixgbe_init_eeprom_params_generic(hw)) {
5685                 e_dev_err("EEPROM initialization failed\n");
5686                 return -EIO;
5687         }
5688
5689         /* PF holds first pool slot */
5690         set_bit(0, &adapter->fwd_bitmask);
5691         set_bit(__IXGBE_DOWN, &adapter->state);
5692
5693         return 0;
5694 }
5695
5696 /**
5697  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
5698  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
5699  *
5700  * Return 0 on success, negative on failure
5701  **/
5702 int ixgbe_setup_tx_resources(struct ixgbe_ring *tx_ring)
5703 {
5704         struct device *dev = tx_ring->dev;
5705         int orig_node = dev_to_node(dev);
5706         int ring_node = -1;
5707         int size;
5708
5709         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
5710
5711         if (tx_ring->q_vector)
5712                 ring_node = tx_ring->q_vector->numa_node;
5713
5714         tx_ring->tx_buffer_info = vzalloc_node(size, ring_node);
5715         if (!tx_ring->tx_buffer_info)
5716                 tx_ring->tx_buffer_info = vzalloc(size);
5717         if (!tx_ring->tx_buffer_info)
5718                 goto err;
5719
5720         u64_stats_init(&tx_ring->syncp);
5721
5722         /* round up to nearest 4K */
5723         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
5724         tx_ring->size = ALIGN(tx_ring->size, 4096);
5725
5726         set_dev_node(dev, ring_node);
5727         tx_ring->desc = dma_alloc_coherent(dev,
5728                                            tx_ring->size,
5729                                            &tx_ring->dma,
5730                                            GFP_KERNEL);
5731         set_dev_node(dev, orig_node);
5732         if (!tx_ring->desc)
5733                 tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
5734                                                    &tx_ring->dma, GFP_KERNEL);
5735         if (!tx_ring->desc)
5736                 goto err;
5737
5738         tx_ring->next_to_use = 0;
5739         tx_ring->next_to_clean = 0;
5740         return 0;
5741
5742 err:
5743         vfree(tx_ring->tx_buffer_info);
5744         tx_ring->tx_buffer_info = NULL;
5745         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
5746         return -ENOMEM;
5747 }
5748
5749 /**
5750  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
5751  * @adapter: board private structure
5752  *
5753  * If this function returns with an error, then it's possible one or
5754  * more of the rings is populated (while the rest are not).  It is the
5755  * callers duty to clean those orphaned rings.
5756  *
5757  * Return 0 on success, negative on failure
5758  **/
5759 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
5760 {
5761         int i, err = 0;
5762
5763         for (i = 0; i < adapter->num_tx_queues; i++) {
5764                 err = ixgbe_setup_tx_resources(adapter->tx_ring[i]);
5765                 if (!err)
5766                         continue;
5767
5768                 e_err(probe, "Allocation for Tx Queue %u failed\n", i);
5769                 goto err_setup_tx;
5770         }
5771
5772         return 0;
5773 err_setup_tx:
5774         /* rewind the index freeing the rings as we go */
5775         while (i--)
5776                 ixgbe_free_tx_resources(adapter->tx_ring[i]);
5777         return err;
5778 }
5779
5780 /**
5781  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
5782  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
5783  *
5784  * Returns 0 on success, negative on failure
5785  **/
5786 int ixgbe_setup_rx_resources(struct ixgbe_ring *rx_ring)
5787 {
5788         struct device *dev = rx_ring->dev;
5789         int orig_node = dev_to_node(dev);
5790         int ring_node = -1;
5791         int size;
5792
5793         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
5794
5795         if (rx_ring->q_vector)
5796                 ring_node = rx_ring->q_vector->numa_node;
5797
5798         rx_ring->rx_buffer_info = vzalloc_node(size, ring_node);
5799         if (!rx_ring->rx_buffer_info)
5800                 rx_ring->rx_buffer_info = vzalloc(size);
5801         if (!rx_ring->rx_buffer_info)
5802                 goto err;
5803
5804         u64_stats_init(&rx_ring->syncp);
5805
5806         /* Round up to nearest 4K */
5807         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
5808         rx_ring->size = ALIGN(rx_ring->size, 4096);
5809
5810         set_dev_node(dev, ring_node);
5811         rx_ring->desc = dma_alloc_coherent(dev,
5812                                            rx_ring->size,
5813                                            &rx_ring->dma,
5814                                            GFP_KERNEL);
5815         set_dev_node(dev, orig_node);
5816         if (!rx_ring->desc)
5817                 rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
5818                                                    &rx_ring->dma, GFP_KERNEL);
5819         if (!rx_ring->desc)
5820                 goto err;
5821
5822         rx_ring->next_to_clean = 0;
5823         rx_ring->next_to_use = 0;
5824
5825         return 0;
5826 err:
5827         vfree(rx_ring->rx_buffer_info);
5828         rx_ring->rx_buffer_info = NULL;
5829         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
5830         return -ENOMEM;
5831 }
5832
5833 /**
5834  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
5835  * @adapter: board private structure
5836  *
5837  * If this function returns with an error, then it's possible one or
5838  * more of the rings is populated (while the rest are not).  It is the
5839  * callers duty to clean those orphaned rings.
5840  *
5841  * Return 0 on success, negative on failure
5842  **/
5843 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
5844 {
5845         int i, err = 0;
5846
5847         for (i = 0; i < adapter->num_rx_queues; i++) {
5848                 err = ixgbe_setup_rx_resources(adapter->rx_ring[i]);
5849                 if (!err)
5850                         continue;
5851
5852                 e_err(probe, "Allocation for Rx Queue %u failed\n", i);
5853                 goto err_setup_rx;
5854         }
5855
5856 #ifdef IXGBE_FCOE
5857         err = ixgbe_setup_fcoe_ddp_resources(adapter);
5858         if (!err)
5859 #endif
5860                 return 0;
5861 err_setup_rx:
5862         /* rewind the index freeing the rings as we go */
5863         while (i--)
5864                 ixgbe_free_rx_resources(adapter->rx_ring[i]);
5865         return err;
5866 }
5867
5868 /**
5869  * ixgbe_free_tx_resources - Free Tx Resources per Queue
5870  * @tx_ring: Tx descriptor ring for a specific queue
5871  *
5872  * Free all transmit software resources
5873  **/
5874 void ixgbe_free_tx_resources(struct ixgbe_ring *tx_ring)
5875 {
5876         ixgbe_clean_tx_ring(tx_ring);
5877
5878         vfree(tx_ring->tx_buffer_info);
5879         tx_ring->tx_buffer_info = NULL;
5880
5881         /* if not set, then don't free */
5882         if (!tx_ring->desc)
5883                 return;
5884
5885         dma_free_coherent(tx_ring->dev, tx_ring->size,
5886                           tx_ring->desc, tx_ring->dma);
5887
5888         tx_ring->desc = NULL;
5889 }
5890
5891 /**
5892  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
5893  * @adapter: board private structure
5894  *
5895  * Free all transmit software resources
5896  **/
5897 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
5898 {
5899         int i;
5900
5901         for (i = 0; i < adapter->num_tx_queues; i++)
5902                 if (adapter->tx_ring[i]->desc)
5903                         ixgbe_free_tx_resources(adapter->tx_ring[i]);
5904 }
5905
5906 /**
5907  * ixgbe_free_rx_resources - Free Rx Resources
5908  * @rx_ring: ring to clean the resources from
5909  *
5910  * Free all receive software resources
5911  **/
5912 void ixgbe_free_rx_resources(struct ixgbe_ring *rx_ring)
5913 {
5914         ixgbe_clean_rx_ring(rx_ring);
5915
5916         vfree(rx_ring->rx_buffer_info);
5917         rx_ring->rx_buffer_info = NULL;
5918
5919         /* if not set, then don't free */
5920         if (!rx_ring->desc)
5921                 return;
5922
5923         dma_free_coherent(rx_ring->dev, rx_ring->size,
5924                           rx_ring->desc, rx_ring->dma);
5925
5926         rx_ring->desc = NULL;
5927 }
5928
5929 /**
5930  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
5931  * @adapter: board private structure
5932  *
5933  * Free all receive software resources
5934  **/
5935 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
5936 {
5937         int i;
5938
5939 #ifdef IXGBE_FCOE
5940         ixgbe_free_fcoe_ddp_resources(adapter);
5941
5942 #endif
5943         for (i = 0; i < adapter->num_rx_queues; i++)
5944                 if (adapter->rx_ring[i]->desc)
5945                         ixgbe_free_rx_resources(adapter->rx_ring[i]);
5946 }
5947
5948 /**
5949  * ixgbe_change_mtu - Change the Maximum Transfer Unit
5950  * @netdev: network interface device structure
5951  * @new_mtu: new value for maximum frame size
5952  *
5953  * Returns 0 on success, negative on failure
5954  **/
5955 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
5956 {
5957         struct ixgbe_adapter *adapter = netdev_priv(netdev);
5958         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
5959
5960         /* MTU < 68 is an error and causes problems on some kernels */
5961         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
5962                 return -EINVAL;
5963
5964         /*
5965          * For 82599EB we cannot allow legacy VFs to enable their receive
5966          * paths when MTU greater than 1500 is configured.  So display a
5967          * warning that legacy VFs will be disabled.
5968          */
5969         if ((adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) &&
5970             (adapter->hw.mac.type == ixgbe_mac_82599EB) &&
5971             (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)))
5972                 e_warn(probe, "Setting MTU > 1500 will disable legacy VFs\n");
5973
5974         e_info(probe, "changing MTU from %d to %d\n", netdev->mtu, new_mtu);
5975
5976         /* must set new MTU before calling down or up */
5977         netdev->mtu = new_mtu;
5978
5979         if (netif_running(netdev))
5980                 ixgbe_reinit_locked(adapter);
5981
5982         return 0;
5983 }
5984
5985 /**
5986  * ixgbe_open - Called when a network interface is made active
5987  * @netdev: network interface device structure
5988  *
5989  * Returns 0 on success, negative value on failure
5990  *
5991  * The open entry point is called when a network interface is made
5992  * active by the system (IFF_UP).  At this point all resources needed
5993  * for transmit and receive operations are allocated, the interrupt
5994  * handler is registered with the OS, the watchdog timer is started,
5995  * and the stack is notified that the interface is ready.
5996  **/
5997 static int ixgbe_open(struct net_device *netdev)
5998 {
5999         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6000         struct ixgbe_hw *hw = &adapter->hw;
6001         int err, queues;
6002
6003         /* disallow open during test */
6004         if (test_bit(__IXGBE_TESTING, &adapter->state))
6005                 return -EBUSY;
6006
6007         netif_carrier_off(netdev);
6008
6009         /* allocate transmit descriptors */
6010         err = ixgbe_setup_all_tx_resources(adapter);
6011         if (err)
6012                 goto err_setup_tx;
6013
6014         /* allocate receive descriptors */
6015         err = ixgbe_setup_all_rx_resources(adapter);
6016         if (err)
6017                 goto err_setup_rx;
6018
6019         ixgbe_configure(adapter);
6020
6021         err = ixgbe_request_irq(adapter);
6022         if (err)
6023                 goto err_req_irq;
6024
6025         /* Notify the stack of the actual queue counts. */
6026         if (adapter->num_rx_pools > 1)
6027                 queues = adapter->num_rx_queues_per_pool;
6028         else
6029                 queues = adapter->num_tx_queues;
6030
6031         err = netif_set_real_num_tx_queues(netdev, queues);
6032         if (err)
6033                 goto err_set_queues;
6034
6035         if (adapter->num_rx_pools > 1 &&
6036             adapter->num_rx_queues > IXGBE_MAX_L2A_QUEUES)
6037                 queues = IXGBE_MAX_L2A_QUEUES;
6038         else
6039                 queues = adapter->num_rx_queues;
6040         err = netif_set_real_num_rx_queues(netdev, queues);
6041         if (err)
6042                 goto err_set_queues;
6043
6044         ixgbe_ptp_init(adapter);
6045
6046         ixgbe_up_complete(adapter);
6047
6048         ixgbe_clear_vxlan_port(adapter);
6049 #ifdef CONFIG_IXGBE_VXLAN
6050         vxlan_get_rx_port(netdev);
6051 #endif
6052
6053         return 0;
6054
6055 err_set_queues:
6056         ixgbe_free_irq(adapter);
6057 err_req_irq:
6058         ixgbe_free_all_rx_resources(adapter);
6059         if (hw->phy.ops.set_phy_power && !adapter->wol)
6060                 hw->phy.ops.set_phy_power(&adapter->hw, false);
6061 err_setup_rx:
6062         ixgbe_free_all_tx_resources(adapter);
6063 err_setup_tx:
6064         ixgbe_reset(adapter);
6065
6066         return err;
6067 }
6068
6069 static void ixgbe_close_suspend(struct ixgbe_adapter *adapter)
6070 {
6071         ixgbe_ptp_suspend(adapter);
6072
6073         if (adapter->hw.phy.ops.enter_lplu) {
6074                 adapter->hw.phy.reset_disable = true;
6075                 ixgbe_down(adapter);
6076                 adapter->hw.phy.ops.enter_lplu(&adapter->hw);
6077                 adapter->hw.phy.reset_disable = false;
6078         } else {
6079                 ixgbe_down(adapter);
6080         }
6081
6082         ixgbe_free_irq(adapter);
6083
6084         ixgbe_free_all_tx_resources(adapter);
6085         ixgbe_free_all_rx_resources(adapter);
6086 }
6087
6088 /**
6089  * ixgbe_close - Disables a network interface
6090  * @netdev: network interface device structure
6091  *
6092  * Returns 0, this is not allowed to fail
6093  *
6094  * The close entry point is called when an interface is de-activated
6095  * by the OS.  The hardware is still under the drivers control, but
6096  * needs to be disabled.  A global MAC reset is issued to stop the
6097  * hardware, and all transmit and receive resources are freed.
6098  **/
6099 static int ixgbe_close(struct net_device *netdev)
6100 {
6101         struct ixgbe_adapter *adapter = netdev_priv(netdev);
6102
6103         ixgbe_ptp_stop(adapter);
6104
6105         ixgbe_close_suspend(adapter);
6106
6107         ixgbe_fdir_filter_exit(adapter);
6108
6109         ixgbe_release_hw_control(adapter);
6110
6111         return 0;
6112 }
6113
6114 #ifdef CONFIG_PM
6115 static int ixgbe_resume(struct pci_dev *pdev)
6116 {
6117         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6118         struct net_device *netdev = adapter->netdev;
6119         u32 err;
6120
6121         adapter->hw.hw_addr = adapter->io_addr;
6122         pci_set_power_state(pdev, PCI_D0);
6123         pci_restore_state(pdev);
6124         /*
6125          * pci_restore_state clears dev->state_saved so call
6126          * pci_save_state to restore it.
6127          */
6128         pci_save_state(pdev);
6129
6130         err = pci_enable_device_mem(pdev);
6131         if (err) {
6132                 e_dev_err("Cannot enable PCI device from suspend\n");
6133                 return err;
6134         }
6135         smp_mb__before_atomic();
6136         clear_bit(__IXGBE_DISABLED, &adapter->state);
6137         pci_set_master(pdev);
6138
6139         pci_wake_from_d3(pdev, false);
6140
6141         ixgbe_reset(adapter);
6142
6143         IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
6144
6145         rtnl_lock();
6146         err = ixgbe_init_interrupt_scheme(adapter);
6147         if (!err && netif_running(netdev))
6148                 err = ixgbe_open(netdev);
6149
6150         rtnl_unlock();
6151
6152         if (err)
6153                 return err;
6154
6155         netif_device_attach(netdev);
6156
6157         return 0;
6158 }
6159 #endif /* CONFIG_PM */
6160
6161 static int __ixgbe_shutdown(struct pci_dev *pdev, bool *enable_wake)
6162 {
6163         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
6164         struct net_device *netdev = adapter->netdev;
6165         struct ixgbe_hw *hw = &adapter->hw;
6166         u32 ctrl, fctrl;
6167         u32 wufc = adapter->wol;
6168 #ifdef CONFIG_PM
6169         int retval = 0;
6170 #endif
6171
6172         netif_device_detach(netdev);
6173
6174         rtnl_lock();
6175         if (netif_running(netdev))
6176                 ixgbe_close_suspend(adapter);
6177         rtnl_unlock();
6178
6179         ixgbe_clear_interrupt_scheme(adapter);
6180
6181 #ifdef CONFIG_PM
6182         retval = pci_save_state(pdev);
6183         if (retval)
6184                 return retval;
6185
6186 #endif
6187         if (hw->mac.ops.stop_link_on_d3)
6188                 hw->mac.ops.stop_link_on_d3(hw);
6189
6190         if (wufc) {
6191                 ixgbe_set_rx_mode(netdev);
6192
6193                 /* enable the optics for 82599 SFP+ fiber as we can WoL */
6194                 if (hw->mac.ops.enable_tx_laser)
6195                         hw->mac.ops.enable_tx_laser(hw);
6196
6197                 /* turn on all-multi mode if wake on multicast is enabled */
6198                 if (wufc & IXGBE_WUFC_MC) {
6199                         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6200                         fctrl |= IXGBE_FCTRL_MPE;
6201                         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
6202                 }
6203
6204                 ctrl = IXGBE_READ_REG(hw, IXGBE_CTRL);
6205                 ctrl |= IXGBE_CTRL_GIO_DIS;
6206                 IXGBE_WRITE_REG(hw, IXGBE_CTRL, ctrl);
6207
6208                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, wufc);
6209         } else {
6210                 IXGBE_WRITE_REG(hw, IXGBE_WUC, 0);
6211                 IXGBE_WRITE_REG(hw, IXGBE_WUFC, 0);
6212         }
6213
6214         switch (hw->mac.type) {
6215         case ixgbe_mac_82598EB:
6216                 pci_wake_from_d3(pdev, false);
6217                 break;
6218         case ixgbe_mac_82599EB:
6219         case ixgbe_mac_X540:
6220         case ixgbe_mac_X550:
6221         case ixgbe_mac_X550EM_x:
6222                 pci_wake_from_d3(pdev, !!wufc);
6223                 break;
6224         default:
6225                 break;
6226         }
6227
6228         *enable_wake = !!wufc;
6229         if (hw->phy.ops.set_phy_power && !*enable_wake)
6230                 hw->phy.ops.set_phy_power(hw, false);
6231
6232         ixgbe_release_hw_control(adapter);
6233
6234         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
6235                 pci_disable_device(pdev);
6236
6237         return 0;
6238 }
6239
6240 #ifdef CONFIG_PM
6241 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
6242 {
6243         int retval;
6244         bool wake;
6245
6246         retval = __ixgbe_shutdown(pdev, &wake);
6247         if (retval)
6248                 return retval;
6249
6250         if (wake) {
6251                 pci_prepare_to_sleep(pdev);
6252         } else {
6253                 pci_wake_from_d3(pdev, false);
6254                 pci_set_power_state(pdev, PCI_D3hot);
6255         }
6256
6257         return 0;
6258 }
6259 #endif /* CONFIG_PM */
6260
6261 static void ixgbe_shutdown(struct pci_dev *pdev)
6262 {
6263         bool wake;
6264
6265         __ixgbe_shutdown(pdev, &wake);
6266
6267         if (system_state == SYSTEM_POWER_OFF) {
6268                 pci_wake_from_d3(pdev, wake);
6269                 pci_set_power_state(pdev, PCI_D3hot);
6270         }
6271 }
6272
6273 /**
6274  * ixgbe_update_stats - Update the board statistics counters.
6275  * @adapter: board private structure
6276  **/
6277 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
6278 {
6279         struct net_device *netdev = adapter->netdev;
6280         struct ixgbe_hw *hw = &adapter->hw;
6281         struct ixgbe_hw_stats *hwstats = &adapter->stats;
6282         u64 total_mpc = 0;
6283         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
6284         u64 non_eop_descs = 0, restart_queue = 0, tx_busy = 0;
6285         u64 alloc_rx_page_failed = 0, alloc_rx_buff_failed = 0;
6286         u64 bytes = 0, packets = 0, hw_csum_rx_error = 0;
6287
6288         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6289             test_bit(__IXGBE_RESETTING, &adapter->state))
6290                 return;
6291
6292         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED) {
6293                 u64 rsc_count = 0;
6294                 u64 rsc_flush = 0;
6295                 for (i = 0; i < adapter->num_rx_queues; i++) {
6296                         rsc_count += adapter->rx_ring[i]->rx_stats.rsc_count;
6297                         rsc_flush += adapter->rx_ring[i]->rx_stats.rsc_flush;
6298                 }
6299                 adapter->rsc_total_count = rsc_count;
6300                 adapter->rsc_total_flush = rsc_flush;
6301         }
6302
6303         for (i = 0; i < adapter->num_rx_queues; i++) {
6304                 struct ixgbe_ring *rx_ring = adapter->rx_ring[i];
6305                 non_eop_descs += rx_ring->rx_stats.non_eop_descs;
6306                 alloc_rx_page_failed += rx_ring->rx_stats.alloc_rx_page_failed;
6307                 alloc_rx_buff_failed += rx_ring->rx_stats.alloc_rx_buff_failed;
6308                 hw_csum_rx_error += rx_ring->rx_stats.csum_err;
6309                 bytes += rx_ring->stats.bytes;
6310                 packets += rx_ring->stats.packets;
6311         }
6312         adapter->non_eop_descs = non_eop_descs;
6313         adapter->alloc_rx_page_failed = alloc_rx_page_failed;
6314         adapter->alloc_rx_buff_failed = alloc_rx_buff_failed;
6315         adapter->hw_csum_rx_error = hw_csum_rx_error;
6316         netdev->stats.rx_bytes = bytes;
6317         netdev->stats.rx_packets = packets;
6318
6319         bytes = 0;
6320         packets = 0;
6321         /* gather some stats to the adapter struct that are per queue */
6322         for (i = 0; i < adapter->num_tx_queues; i++) {
6323                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6324                 restart_queue += tx_ring->tx_stats.restart_queue;
6325                 tx_busy += tx_ring->tx_stats.tx_busy;
6326                 bytes += tx_ring->stats.bytes;
6327                 packets += tx_ring->stats.packets;
6328         }
6329         adapter->restart_queue = restart_queue;
6330         adapter->tx_busy = tx_busy;
6331         netdev->stats.tx_bytes = bytes;
6332         netdev->stats.tx_packets = packets;
6333
6334         hwstats->crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
6335
6336         /* 8 register reads */
6337         for (i = 0; i < 8; i++) {
6338                 /* for packet buffers not used, the register should read 0 */
6339                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
6340                 missed_rx += mpc;
6341                 hwstats->mpc[i] += mpc;
6342                 total_mpc += hwstats->mpc[i];
6343                 hwstats->pxontxc[i] += IXGBE_READ_REG(hw, IXGBE_PXONTXC(i));
6344                 hwstats->pxofftxc[i] += IXGBE_READ_REG(hw, IXGBE_PXOFFTXC(i));
6345                 switch (hw->mac.type) {
6346                 case ixgbe_mac_82598EB:
6347                         hwstats->rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
6348                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC(i));
6349                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC(i));
6350                         hwstats->pxonrxc[i] +=
6351                                 IXGBE_READ_REG(hw, IXGBE_PXONRXC(i));
6352                         break;
6353                 case ixgbe_mac_82599EB:
6354                 case ixgbe_mac_X540:
6355                 case ixgbe_mac_X550:
6356                 case ixgbe_mac_X550EM_x:
6357                         hwstats->pxonrxc[i] +=
6358                                 IXGBE_READ_REG(hw, IXGBE_PXONRXCNT(i));
6359                         break;
6360                 default:
6361                         break;
6362                 }
6363         }
6364
6365         /*16 register reads */
6366         for (i = 0; i < 16; i++) {
6367                 hwstats->qptc[i] += IXGBE_READ_REG(hw, IXGBE_QPTC(i));
6368                 hwstats->qprc[i] += IXGBE_READ_REG(hw, IXGBE_QPRC(i));
6369                 if ((hw->mac.type == ixgbe_mac_82599EB) ||
6370                     (hw->mac.type == ixgbe_mac_X540) ||
6371                     (hw->mac.type == ixgbe_mac_X550) ||
6372                     (hw->mac.type == ixgbe_mac_X550EM_x)) {
6373                         hwstats->qbtc[i] += IXGBE_READ_REG(hw, IXGBE_QBTC_L(i));
6374                         IXGBE_READ_REG(hw, IXGBE_QBTC_H(i)); /* to clear */
6375                         hwstats->qbrc[i] += IXGBE_READ_REG(hw, IXGBE_QBRC_L(i));
6376                         IXGBE_READ_REG(hw, IXGBE_QBRC_H(i)); /* to clear */
6377                 }
6378         }
6379
6380         hwstats->gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
6381         /* work around hardware counting issue */
6382         hwstats->gprc -= missed_rx;
6383
6384         ixgbe_update_xoff_received(adapter);
6385
6386         /* 82598 hardware only has a 32 bit counter in the high register */
6387         switch (hw->mac.type) {
6388         case ixgbe_mac_82598EB:
6389                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
6390                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
6391                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
6392                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORH);
6393                 break;
6394         case ixgbe_mac_X540:
6395         case ixgbe_mac_X550:
6396         case ixgbe_mac_X550EM_x:
6397                 /* OS2BMC stats are X540 and later */
6398                 hwstats->o2bgptc += IXGBE_READ_REG(hw, IXGBE_O2BGPTC);
6399                 hwstats->o2bspc += IXGBE_READ_REG(hw, IXGBE_O2BSPC);
6400                 hwstats->b2ospc += IXGBE_READ_REG(hw, IXGBE_B2OSPC);
6401                 hwstats->b2ogprc += IXGBE_READ_REG(hw, IXGBE_B2OGPRC);
6402         case ixgbe_mac_82599EB:
6403                 for (i = 0; i < 16; i++)
6404                         adapter->hw_rx_no_dma_resources +=
6405                                              IXGBE_READ_REG(hw, IXGBE_QPRDC(i));
6406                 hwstats->gorc += IXGBE_READ_REG(hw, IXGBE_GORCL);
6407                 IXGBE_READ_REG(hw, IXGBE_GORCH); /* to clear */
6408                 hwstats->gotc += IXGBE_READ_REG(hw, IXGBE_GOTCL);
6409                 IXGBE_READ_REG(hw, IXGBE_GOTCH); /* to clear */
6410                 hwstats->tor += IXGBE_READ_REG(hw, IXGBE_TORL);
6411                 IXGBE_READ_REG(hw, IXGBE_TORH); /* to clear */
6412                 hwstats->lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXCNT);
6413                 hwstats->fdirmatch += IXGBE_READ_REG(hw, IXGBE_FDIRMATCH);
6414                 hwstats->fdirmiss += IXGBE_READ_REG(hw, IXGBE_FDIRMISS);
6415 #ifdef IXGBE_FCOE
6416                 hwstats->fccrc += IXGBE_READ_REG(hw, IXGBE_FCCRC);
6417                 hwstats->fcoerpdc += IXGBE_READ_REG(hw, IXGBE_FCOERPDC);
6418                 hwstats->fcoeprc += IXGBE_READ_REG(hw, IXGBE_FCOEPRC);
6419                 hwstats->fcoeptc += IXGBE_READ_REG(hw, IXGBE_FCOEPTC);
6420                 hwstats->fcoedwrc += IXGBE_READ_REG(hw, IXGBE_FCOEDWRC);
6421                 hwstats->fcoedwtc += IXGBE_READ_REG(hw, IXGBE_FCOEDWTC);
6422                 /* Add up per cpu counters for total ddp aloc fail */
6423                 if (adapter->fcoe.ddp_pool) {
6424                         struct ixgbe_fcoe *fcoe = &adapter->fcoe;
6425                         struct ixgbe_fcoe_ddp_pool *ddp_pool;
6426                         unsigned int cpu;
6427                         u64 noddp = 0, noddp_ext_buff = 0;
6428                         for_each_possible_cpu(cpu) {
6429                                 ddp_pool = per_cpu_ptr(fcoe->ddp_pool, cpu);
6430                                 noddp += ddp_pool->noddp;
6431                                 noddp_ext_buff += ddp_pool->noddp_ext_buff;
6432                         }
6433                         hwstats->fcoe_noddp = noddp;
6434                         hwstats->fcoe_noddp_ext_buff = noddp_ext_buff;
6435                 }
6436 #endif /* IXGBE_FCOE */
6437                 break;
6438         default:
6439                 break;
6440         }
6441         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
6442         hwstats->bprc += bprc;
6443         hwstats->mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
6444         if (hw->mac.type == ixgbe_mac_82598EB)
6445                 hwstats->mprc -= bprc;
6446         hwstats->roc += IXGBE_READ_REG(hw, IXGBE_ROC);
6447         hwstats->prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
6448         hwstats->prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
6449         hwstats->prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
6450         hwstats->prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
6451         hwstats->prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
6452         hwstats->prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
6453         hwstats->rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
6454         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
6455         hwstats->lxontxc += lxon;
6456         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
6457         hwstats->lxofftxc += lxoff;
6458         hwstats->gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
6459         hwstats->mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
6460         /*
6461          * 82598 errata - tx of flow control packets is included in tx counters
6462          */
6463         xon_off_tot = lxon + lxoff;
6464         hwstats->gptc -= xon_off_tot;
6465         hwstats->mptc -= xon_off_tot;
6466         hwstats->gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
6467         hwstats->ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
6468         hwstats->rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
6469         hwstats->rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
6470         hwstats->tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
6471         hwstats->ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
6472         hwstats->ptc64 -= xon_off_tot;
6473         hwstats->ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
6474         hwstats->ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
6475         hwstats->ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
6476         hwstats->ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
6477         hwstats->ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
6478         hwstats->bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
6479
6480         /* Fill out the OS statistics structure */
6481         netdev->stats.multicast = hwstats->mprc;
6482
6483         /* Rx Errors */
6484         netdev->stats.rx_errors = hwstats->crcerrs + hwstats->rlec;
6485         netdev->stats.rx_dropped = 0;
6486         netdev->stats.rx_length_errors = hwstats->rlec;
6487         netdev->stats.rx_crc_errors = hwstats->crcerrs;
6488         netdev->stats.rx_missed_errors = total_mpc;
6489 }
6490
6491 /**
6492  * ixgbe_fdir_reinit_subtask - worker thread to reinit FDIR filter table
6493  * @adapter: pointer to the device adapter structure
6494  **/
6495 static void ixgbe_fdir_reinit_subtask(struct ixgbe_adapter *adapter)
6496 {
6497         struct ixgbe_hw *hw = &adapter->hw;
6498         int i;
6499
6500         if (!(adapter->flags2 & IXGBE_FLAG2_FDIR_REQUIRES_REINIT))
6501                 return;
6502
6503         adapter->flags2 &= ~IXGBE_FLAG2_FDIR_REQUIRES_REINIT;
6504
6505         /* if interface is down do nothing */
6506         if (test_bit(__IXGBE_DOWN, &adapter->state))
6507                 return;
6508
6509         /* do nothing if we are not using signature filters */
6510         if (!(adapter->flags & IXGBE_FLAG_FDIR_HASH_CAPABLE))
6511                 return;
6512
6513         adapter->fdir_overflow++;
6514
6515         if (ixgbe_reinit_fdir_tables_82599(hw) == 0) {
6516                 for (i = 0; i < adapter->num_tx_queues; i++)
6517                         set_bit(__IXGBE_TX_FDIR_INIT_DONE,
6518                                 &(adapter->tx_ring[i]->state));
6519                 /* re-enable flow director interrupts */
6520                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_FLOW_DIR);
6521         } else {
6522                 e_err(probe, "failed to finish FDIR re-initialization, "
6523                       "ignored adding FDIR ATR filters\n");
6524         }
6525 }
6526
6527 /**
6528  * ixgbe_check_hang_subtask - check for hung queues and dropped interrupts
6529  * @adapter: pointer to the device adapter structure
6530  *
6531  * This function serves two purposes.  First it strobes the interrupt lines
6532  * in order to make certain interrupts are occurring.  Secondly it sets the
6533  * bits needed to check for TX hangs.  As a result we should immediately
6534  * determine if a hang has occurred.
6535  */
6536 static void ixgbe_check_hang_subtask(struct ixgbe_adapter *adapter)
6537 {
6538         struct ixgbe_hw *hw = &adapter->hw;
6539         u64 eics = 0;
6540         int i;
6541
6542         /* If we're down, removing or resetting, just bail */
6543         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6544             test_bit(__IXGBE_REMOVING, &adapter->state) ||
6545             test_bit(__IXGBE_RESETTING, &adapter->state))
6546                 return;
6547
6548         /* Force detection of hung controller */
6549         if (netif_carrier_ok(adapter->netdev)) {
6550                 for (i = 0; i < adapter->num_tx_queues; i++)
6551                         set_check_for_tx_hang(adapter->tx_ring[i]);
6552         }
6553
6554         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
6555                 /*
6556                  * for legacy and MSI interrupts don't set any bits
6557                  * that are enabled for EIAM, because this operation
6558                  * would set *both* EIMS and EICS for any bit in EIAM
6559                  */
6560                 IXGBE_WRITE_REG(hw, IXGBE_EICS,
6561                         (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
6562         } else {
6563                 /* get one bit for every active tx/rx interrupt vector */
6564                 for (i = 0; i < adapter->num_q_vectors; i++) {
6565                         struct ixgbe_q_vector *qv = adapter->q_vector[i];
6566                         if (qv->rx.ring || qv->tx.ring)
6567                                 eics |= ((u64)1 << i);
6568                 }
6569         }
6570
6571         /* Cause software interrupt to ensure rings are cleaned */
6572         ixgbe_irq_rearm_queues(adapter, eics);
6573 }
6574
6575 /**
6576  * ixgbe_watchdog_update_link - update the link status
6577  * @adapter: pointer to the device adapter structure
6578  * @link_speed: pointer to a u32 to store the link_speed
6579  **/
6580 static void ixgbe_watchdog_update_link(struct ixgbe_adapter *adapter)
6581 {
6582         struct ixgbe_hw *hw = &adapter->hw;
6583         u32 link_speed = adapter->link_speed;
6584         bool link_up = adapter->link_up;
6585         bool pfc_en = adapter->dcb_cfg.pfc_mode_enable;
6586
6587         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE))
6588                 return;
6589
6590         if (hw->mac.ops.check_link) {
6591                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
6592         } else {
6593                 /* always assume link is up, if no check link function */
6594                 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
6595                 link_up = true;
6596         }
6597
6598         if (adapter->ixgbe_ieee_pfc)
6599                 pfc_en |= !!(adapter->ixgbe_ieee_pfc->pfc_en);
6600
6601         if (link_up && !((adapter->flags & IXGBE_FLAG_DCB_ENABLED) && pfc_en)) {
6602                 hw->mac.ops.fc_enable(hw);
6603                 ixgbe_set_rx_drop_en(adapter);
6604         }
6605
6606         if (link_up ||
6607             time_after(jiffies, (adapter->link_check_timeout +
6608                                  IXGBE_TRY_LINK_TIMEOUT))) {
6609                 adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
6610                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
6611                 IXGBE_WRITE_FLUSH(hw);
6612         }
6613
6614         adapter->link_up = link_up;
6615         adapter->link_speed = link_speed;
6616 }
6617
6618 static void ixgbe_update_default_up(struct ixgbe_adapter *adapter)
6619 {
6620 #ifdef CONFIG_IXGBE_DCB
6621         struct net_device *netdev = adapter->netdev;
6622         struct dcb_app app = {
6623                               .selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE,
6624                               .protocol = 0,
6625                              };
6626         u8 up = 0;
6627
6628         if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_IEEE)
6629                 up = dcb_ieee_getapp_mask(netdev, &app);
6630
6631         adapter->default_up = (up > 1) ? (ffs(up) - 1) : 0;
6632 #endif
6633 }
6634
6635 /**
6636  * ixgbe_watchdog_link_is_up - update netif_carrier status and
6637  *                             print link up message
6638  * @adapter: pointer to the device adapter structure
6639  **/
6640 static void ixgbe_watchdog_link_is_up(struct ixgbe_adapter *adapter)
6641 {
6642         struct net_device *netdev = adapter->netdev;
6643         struct ixgbe_hw *hw = &adapter->hw;
6644         struct net_device *upper;
6645         struct list_head *iter;
6646         u32 link_speed = adapter->link_speed;
6647         const char *speed_str;
6648         bool flow_rx, flow_tx;
6649
6650         /* only continue if link was previously down */
6651         if (netif_carrier_ok(netdev))
6652                 return;
6653
6654         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
6655
6656         switch (hw->mac.type) {
6657         case ixgbe_mac_82598EB: {
6658                 u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
6659                 u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
6660                 flow_rx = !!(frctl & IXGBE_FCTRL_RFCE);
6661                 flow_tx = !!(rmcs & IXGBE_RMCS_TFCE_802_3X);
6662         }
6663                 break;
6664         case ixgbe_mac_X540:
6665         case ixgbe_mac_X550:
6666         case ixgbe_mac_X550EM_x:
6667         case ixgbe_mac_82599EB: {
6668                 u32 mflcn = IXGBE_READ_REG(hw, IXGBE_MFLCN);
6669                 u32 fccfg = IXGBE_READ_REG(hw, IXGBE_FCCFG);
6670                 flow_rx = !!(mflcn & IXGBE_MFLCN_RFCE);
6671                 flow_tx = !!(fccfg & IXGBE_FCCFG_TFCE_802_3X);
6672         }
6673                 break;
6674         default:
6675                 flow_tx = false;
6676                 flow_rx = false;
6677                 break;
6678         }
6679
6680         adapter->last_rx_ptp_check = jiffies;
6681
6682         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6683                 ixgbe_ptp_start_cyclecounter(adapter);
6684
6685         switch (link_speed) {
6686         case IXGBE_LINK_SPEED_10GB_FULL:
6687                 speed_str = "10 Gbps";
6688                 break;
6689         case IXGBE_LINK_SPEED_2_5GB_FULL:
6690                 speed_str = "2.5 Gbps";
6691                 break;
6692         case IXGBE_LINK_SPEED_1GB_FULL:
6693                 speed_str = "1 Gbps";
6694                 break;
6695         case IXGBE_LINK_SPEED_100_FULL:
6696                 speed_str = "100 Mbps";
6697                 break;
6698         default:
6699                 speed_str = "unknown speed";
6700                 break;
6701         }
6702         e_info(drv, "NIC Link is Up %s, Flow Control: %s\n", speed_str,
6703                ((flow_rx && flow_tx) ? "RX/TX" :
6704                (flow_rx ? "RX" :
6705                (flow_tx ? "TX" : "None"))));
6706
6707         netif_carrier_on(netdev);
6708         ixgbe_check_vf_rate_limit(adapter);
6709
6710         /* enable transmits */
6711         netif_tx_wake_all_queues(adapter->netdev);
6712
6713         /* enable any upper devices */
6714         rtnl_lock();
6715         netdev_for_each_all_upper_dev_rcu(adapter->netdev, upper, iter) {
6716                 if (netif_is_macvlan(upper)) {
6717                         struct macvlan_dev *vlan = netdev_priv(upper);
6718
6719                         if (vlan->fwd_priv)
6720                                 netif_tx_wake_all_queues(upper);
6721                 }
6722         }
6723         rtnl_unlock();
6724
6725         /* update the default user priority for VFs */
6726         ixgbe_update_default_up(adapter);
6727
6728         /* ping all the active vfs to let them know link has changed */
6729         ixgbe_ping_all_vfs(adapter);
6730 }
6731
6732 /**
6733  * ixgbe_watchdog_link_is_down - update netif_carrier status and
6734  *                               print link down message
6735  * @adapter: pointer to the adapter structure
6736  **/
6737 static void ixgbe_watchdog_link_is_down(struct ixgbe_adapter *adapter)
6738 {
6739         struct net_device *netdev = adapter->netdev;
6740         struct ixgbe_hw *hw = &adapter->hw;
6741
6742         adapter->link_up = false;
6743         adapter->link_speed = 0;
6744
6745         /* only continue if link was up previously */
6746         if (!netif_carrier_ok(netdev))
6747                 return;
6748
6749         /* poll for SFP+ cable when link is down */
6750         if (ixgbe_is_sfp(hw) && hw->mac.type == ixgbe_mac_82598EB)
6751                 adapter->flags2 |= IXGBE_FLAG2_SEARCH_FOR_SFP;
6752
6753         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state))
6754                 ixgbe_ptp_start_cyclecounter(adapter);
6755
6756         e_info(drv, "NIC Link is Down\n");
6757         netif_carrier_off(netdev);
6758
6759         /* ping all the active vfs to let them know link has changed */
6760         ixgbe_ping_all_vfs(adapter);
6761 }
6762
6763 static bool ixgbe_ring_tx_pending(struct ixgbe_adapter *adapter)
6764 {
6765         int i;
6766
6767         for (i = 0; i < adapter->num_tx_queues; i++) {
6768                 struct ixgbe_ring *tx_ring = adapter->tx_ring[i];
6769
6770                 if (tx_ring->next_to_use != tx_ring->next_to_clean)
6771                         return true;
6772         }
6773
6774         return false;
6775 }
6776
6777 static bool ixgbe_vf_tx_pending(struct ixgbe_adapter *adapter)
6778 {
6779         struct ixgbe_hw *hw = &adapter->hw;
6780         struct ixgbe_ring_feature *vmdq = &adapter->ring_feature[RING_F_VMDQ];
6781         u32 q_per_pool = __ALIGN_MASK(1, ~vmdq->mask);
6782
6783         int i, j;
6784
6785         if (!adapter->num_vfs)
6786                 return false;
6787
6788         /* resetting the PF is only needed for MAC before X550 */
6789         if (hw->mac.type >= ixgbe_mac_X550)
6790                 return false;
6791
6792         for (i = 0; i < adapter->num_vfs; i++) {
6793                 for (j = 0; j < q_per_pool; j++) {
6794                         u32 h, t;
6795
6796                         h = IXGBE_READ_REG(hw, IXGBE_PVFTDHN(q_per_pool, i, j));
6797                         t = IXGBE_READ_REG(hw, IXGBE_PVFTDTN(q_per_pool, i, j));
6798
6799                         if (h != t)
6800                                 return true;
6801                 }
6802         }
6803
6804         return false;
6805 }
6806
6807 /**
6808  * ixgbe_watchdog_flush_tx - flush queues on link down
6809  * @adapter: pointer to the device adapter structure
6810  **/
6811 static void ixgbe_watchdog_flush_tx(struct ixgbe_adapter *adapter)
6812 {
6813         if (!netif_carrier_ok(adapter->netdev)) {
6814                 if (ixgbe_ring_tx_pending(adapter) ||
6815                     ixgbe_vf_tx_pending(adapter)) {
6816                         /* We've lost link, so the controller stops DMA,
6817                          * but we've got queued Tx work that's never going
6818                          * to get done, so reset controller to flush Tx.
6819                          * (Do the reset outside of interrupt context).
6820                          */
6821                         e_warn(drv, "initiating reset to clear Tx work after link loss\n");
6822                         adapter->flags2 |= IXGBE_FLAG2_RESET_REQUESTED;
6823                 }
6824         }
6825 }
6826
6827 #ifdef CONFIG_PCI_IOV
6828 static inline void ixgbe_issue_vf_flr(struct ixgbe_adapter *adapter,
6829                                       struct pci_dev *vfdev)
6830 {
6831         if (!pci_wait_for_pending_transaction(vfdev))
6832                 e_dev_warn("Issuing VFLR with pending transactions\n");
6833
6834         e_dev_err("Issuing VFLR for VF %s\n", pci_name(vfdev));
6835         pcie_capability_set_word(vfdev, PCI_EXP_DEVCTL, PCI_EXP_DEVCTL_BCR_FLR);
6836
6837         msleep(100);
6838 }
6839
6840 static void ixgbe_check_for_bad_vf(struct ixgbe_adapter *adapter)
6841 {
6842         struct ixgbe_hw *hw = &adapter->hw;
6843         struct pci_dev *pdev = adapter->pdev;
6844         unsigned int vf;
6845         u32 gpc;
6846
6847         if (!(netif_carrier_ok(adapter->netdev)))
6848                 return;
6849
6850         gpc = IXGBE_READ_REG(hw, IXGBE_TXDGPC);
6851         if (gpc) /* If incrementing then no need for the check below */
6852                 return;
6853         /* Check to see if a bad DMA write target from an errant or
6854          * malicious VF has caused a PCIe error.  If so then we can
6855          * issue a VFLR to the offending VF(s) and then resume without
6856          * requesting a full slot reset.
6857          */
6858
6859         if (!pdev)
6860                 return;
6861
6862         /* check status reg for all VFs owned by this PF */
6863         for (vf = 0; vf < adapter->num_vfs; ++vf) {
6864                 struct pci_dev *vfdev = adapter->vfinfo[vf].vfdev;
6865                 u16 status_reg;
6866
6867                 if (!vfdev)
6868                         continue;
6869                 pci_read_config_word(vfdev, PCI_STATUS, &status_reg);
6870                 if (status_reg != IXGBE_FAILED_READ_CFG_WORD &&
6871                     status_reg & PCI_STATUS_REC_MASTER_ABORT)
6872                         ixgbe_issue_vf_flr(adapter, vfdev);
6873         }
6874 }
6875
6876 static void ixgbe_spoof_check(struct ixgbe_adapter *adapter)
6877 {
6878         u32 ssvpc;
6879
6880         /* Do not perform spoof check for 82598 or if not in IOV mode */
6881         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
6882             adapter->num_vfs == 0)
6883                 return;
6884
6885         ssvpc = IXGBE_READ_REG(&adapter->hw, IXGBE_SSVPC);
6886
6887         /*
6888          * ssvpc register is cleared on read, if zero then no
6889          * spoofed packets in the last interval.
6890          */
6891         if (!ssvpc)
6892                 return;
6893
6894         e_warn(drv, "%u Spoofed packets detected\n", ssvpc);
6895 }
6896 #else
6897 static void ixgbe_spoof_check(struct ixgbe_adapter __always_unused *adapter)
6898 {
6899 }
6900
6901 static void
6902 ixgbe_check_for_bad_vf(struct ixgbe_adapter __always_unused *adapter)
6903 {
6904 }
6905 #endif /* CONFIG_PCI_IOV */
6906
6907
6908 /**
6909  * ixgbe_watchdog_subtask - check and bring link up
6910  * @adapter: pointer to the device adapter structure
6911  **/
6912 static void ixgbe_watchdog_subtask(struct ixgbe_adapter *adapter)
6913 {
6914         /* if interface is down, removing or resetting, do nothing */
6915         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
6916             test_bit(__IXGBE_REMOVING, &adapter->state) ||
6917             test_bit(__IXGBE_RESETTING, &adapter->state))
6918                 return;
6919
6920         ixgbe_watchdog_update_link(adapter);
6921
6922         if (adapter->link_up)
6923                 ixgbe_watchdog_link_is_up(adapter);
6924         else
6925                 ixgbe_watchdog_link_is_down(adapter);
6926
6927         ixgbe_check_for_bad_vf(adapter);
6928         ixgbe_spoof_check(adapter);
6929         ixgbe_update_stats(adapter);
6930
6931         ixgbe_watchdog_flush_tx(adapter);
6932 }
6933
6934 /**
6935  * ixgbe_sfp_detection_subtask - poll for SFP+ cable
6936  * @adapter: the ixgbe adapter structure
6937  **/
6938 static void ixgbe_sfp_detection_subtask(struct ixgbe_adapter *adapter)
6939 {
6940         struct ixgbe_hw *hw = &adapter->hw;
6941         s32 err;
6942
6943         /* not searching for SFP so there is nothing to do here */
6944         if (!(adapter->flags2 & IXGBE_FLAG2_SEARCH_FOR_SFP) &&
6945             !(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6946                 return;
6947
6948         if (adapter->sfp_poll_time &&
6949             time_after(adapter->sfp_poll_time, jiffies))
6950                 return; /* If not yet time to poll for SFP */
6951
6952         /* someone else is in init, wait until next service event */
6953         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
6954                 return;
6955
6956         adapter->sfp_poll_time = jiffies + IXGBE_SFP_POLL_JIFFIES - 1;
6957
6958         err = hw->phy.ops.identify_sfp(hw);
6959         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6960                 goto sfp_out;
6961
6962         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
6963                 /* If no cable is present, then we need to reset
6964                  * the next time we find a good cable. */
6965                 adapter->flags2 |= IXGBE_FLAG2_SFP_NEEDS_RESET;
6966         }
6967
6968         /* exit on error */
6969         if (err)
6970                 goto sfp_out;
6971
6972         /* exit if reset not needed */
6973         if (!(adapter->flags2 & IXGBE_FLAG2_SFP_NEEDS_RESET))
6974                 goto sfp_out;
6975
6976         adapter->flags2 &= ~IXGBE_FLAG2_SFP_NEEDS_RESET;
6977
6978         /*
6979          * A module may be identified correctly, but the EEPROM may not have
6980          * support for that module.  setup_sfp() will fail in that case, so
6981          * we should not allow that module to load.
6982          */
6983         if (hw->mac.type == ixgbe_mac_82598EB)
6984                 err = hw->phy.ops.reset(hw);
6985         else
6986                 err = hw->mac.ops.setup_sfp(hw);
6987
6988         if (err == IXGBE_ERR_SFP_NOT_SUPPORTED)
6989                 goto sfp_out;
6990
6991         adapter->flags |= IXGBE_FLAG_NEED_LINK_CONFIG;
6992         e_info(probe, "detected SFP+: %d\n", hw->phy.sfp_type);
6993
6994 sfp_out:
6995         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
6996
6997         if ((err == IXGBE_ERR_SFP_NOT_SUPPORTED) &&
6998             (adapter->netdev->reg_state == NETREG_REGISTERED)) {
6999                 e_dev_err("failed to initialize because an unsupported "
7000                           "SFP+ module type was detected.\n");
7001                 e_dev_err("Reload the driver after installing a "
7002                           "supported module.\n");
7003                 unregister_netdev(adapter->netdev);
7004         }
7005 }
7006
7007 /**
7008  * ixgbe_sfp_link_config_subtask - set up link SFP after module install
7009  * @adapter: the ixgbe adapter structure
7010  **/
7011 static void ixgbe_sfp_link_config_subtask(struct ixgbe_adapter *adapter)
7012 {
7013         struct ixgbe_hw *hw = &adapter->hw;
7014         u32 speed;
7015         bool autoneg = false;
7016
7017         if (!(adapter->flags & IXGBE_FLAG_NEED_LINK_CONFIG))
7018                 return;
7019
7020         /* someone else is in init, wait until next service event */
7021         if (test_and_set_bit(__IXGBE_IN_SFP_INIT, &adapter->state))
7022                 return;
7023
7024         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_CONFIG;
7025
7026         speed = hw->phy.autoneg_advertised;
7027         if ((!speed) && (hw->mac.ops.get_link_capabilities)) {
7028                 hw->mac.ops.get_link_capabilities(hw, &speed, &autoneg);
7029
7030                 /* setup the highest link when no autoneg */
7031                 if (!autoneg) {
7032                         if (speed & IXGBE_LINK_SPEED_10GB_FULL)
7033                                 speed = IXGBE_LINK_SPEED_10GB_FULL;
7034                 }
7035         }
7036
7037         if (hw->mac.ops.setup_link)
7038                 hw->mac.ops.setup_link(hw, speed, true);
7039
7040         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
7041         adapter->link_check_timeout = jiffies;
7042         clear_bit(__IXGBE_IN_SFP_INIT, &adapter->state);
7043 }
7044
7045 /**
7046  * ixgbe_service_timer - Timer Call-back
7047  * @data: pointer to adapter cast into an unsigned long
7048  **/
7049 static void ixgbe_service_timer(unsigned long data)
7050 {
7051         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
7052         unsigned long next_event_offset;
7053
7054         /* poll faster when waiting for link */
7055         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE)
7056                 next_event_offset = HZ / 10;
7057         else
7058                 next_event_offset = HZ * 2;
7059
7060         /* Reset the timer */
7061         mod_timer(&adapter->service_timer, next_event_offset + jiffies);
7062
7063         ixgbe_service_event_schedule(adapter);
7064 }
7065
7066 static void ixgbe_phy_interrupt_subtask(struct ixgbe_adapter *adapter)
7067 {
7068         struct ixgbe_hw *hw = &adapter->hw;
7069         u32 status;
7070
7071         if (!(adapter->flags2 & IXGBE_FLAG2_PHY_INTERRUPT))
7072                 return;
7073
7074         adapter->flags2 &= ~IXGBE_FLAG2_PHY_INTERRUPT;
7075
7076         if (!hw->phy.ops.handle_lasi)
7077                 return;
7078
7079         status = hw->phy.ops.handle_lasi(&adapter->hw);
7080         if (status != IXGBE_ERR_OVERTEMP)
7081                 return;
7082
7083         e_crit(drv, "%s\n", ixgbe_overheat_msg);
7084 }
7085
7086 static void ixgbe_reset_subtask(struct ixgbe_adapter *adapter)
7087 {
7088         if (!(adapter->flags2 & IXGBE_FLAG2_RESET_REQUESTED))
7089                 return;
7090
7091         adapter->flags2 &= ~IXGBE_FLAG2_RESET_REQUESTED;
7092
7093         /* If we're already down, removing or resetting, just bail */
7094         if (test_bit(__IXGBE_DOWN, &adapter->state) ||
7095             test_bit(__IXGBE_REMOVING, &adapter->state) ||
7096             test_bit(__IXGBE_RESETTING, &adapter->state))
7097                 return;
7098
7099         ixgbe_dump(adapter);
7100         netdev_err(adapter->netdev, "Reset adapter\n");
7101         adapter->tx_timeout_count++;
7102
7103         rtnl_lock();
7104         ixgbe_reinit_locked(adapter);
7105         rtnl_unlock();
7106 }
7107
7108 /**
7109  * ixgbe_service_task - manages and runs subtasks
7110  * @work: pointer to work_struct containing our data
7111  **/
7112 static void ixgbe_service_task(struct work_struct *work)
7113 {
7114         struct ixgbe_adapter *adapter = container_of(work,
7115                                                      struct ixgbe_adapter,
7116                                                      service_task);
7117         if (ixgbe_removed(adapter->hw.hw_addr)) {
7118                 if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
7119                         rtnl_lock();
7120                         ixgbe_down(adapter);
7121                         rtnl_unlock();
7122                 }
7123                 ixgbe_service_event_complete(adapter);
7124                 return;
7125         }
7126 #ifdef CONFIG_IXGBE_VXLAN
7127         if (adapter->flags2 & IXGBE_FLAG2_VXLAN_REREG_NEEDED) {
7128                 adapter->flags2 &= ~IXGBE_FLAG2_VXLAN_REREG_NEEDED;
7129                 vxlan_get_rx_port(adapter->netdev);
7130         }
7131 #endif /* CONFIG_IXGBE_VXLAN */
7132         ixgbe_reset_subtask(adapter);
7133         ixgbe_phy_interrupt_subtask(adapter);
7134         ixgbe_sfp_detection_subtask(adapter);
7135         ixgbe_sfp_link_config_subtask(adapter);
7136         ixgbe_check_overtemp_subtask(adapter);
7137         ixgbe_watchdog_subtask(adapter);
7138         ixgbe_fdir_reinit_subtask(adapter);
7139         ixgbe_check_hang_subtask(adapter);
7140
7141         if (test_bit(__IXGBE_PTP_RUNNING, &adapter->state)) {
7142                 ixgbe_ptp_overflow_check(adapter);
7143                 ixgbe_ptp_rx_hang(adapter);
7144         }
7145
7146         ixgbe_service_event_complete(adapter);
7147 }
7148
7149 static int ixgbe_tso(struct ixgbe_ring *tx_ring,
7150                      struct ixgbe_tx_buffer *first,
7151                      u8 *hdr_len)
7152 {
7153         struct sk_buff *skb = first->skb;
7154         u32 vlan_macip_lens, type_tucmd;
7155         u32 mss_l4len_idx, l4len;
7156         int err;
7157
7158         if (skb->ip_summed != CHECKSUM_PARTIAL)
7159                 return 0;
7160
7161         if (!skb_is_gso(skb))
7162                 return 0;
7163
7164         err = skb_cow_head(skb, 0);
7165         if (err < 0)
7166                 return err;
7167
7168         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
7169         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
7170
7171         if (first->protocol == htons(ETH_P_IP)) {
7172                 struct iphdr *iph = ip_hdr(skb);
7173                 iph->tot_len = 0;
7174                 iph->check = 0;
7175                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
7176                                                          iph->daddr, 0,
7177                                                          IPPROTO_TCP,
7178                                                          0);
7179                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7180                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7181                                    IXGBE_TX_FLAGS_CSUM |
7182                                    IXGBE_TX_FLAGS_IPV4;
7183         } else if (skb_is_gso_v6(skb)) {
7184                 ipv6_hdr(skb)->payload_len = 0;
7185                 tcp_hdr(skb)->check =
7186                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
7187                                      &ipv6_hdr(skb)->daddr,
7188                                      0, IPPROTO_TCP, 0);
7189                 first->tx_flags |= IXGBE_TX_FLAGS_TSO |
7190                                    IXGBE_TX_FLAGS_CSUM;
7191         }
7192
7193         /* compute header lengths */
7194         l4len = tcp_hdrlen(skb);
7195         *hdr_len = skb_transport_offset(skb) + l4len;
7196
7197         /* update gso size and bytecount with header size */
7198         first->gso_segs = skb_shinfo(skb)->gso_segs;
7199         first->bytecount += (first->gso_segs - 1) * *hdr_len;
7200
7201         /* mss_l4len_id: use 0 as index for TSO */
7202         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
7203         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
7204
7205         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
7206         vlan_macip_lens = skb_network_header_len(skb);
7207         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
7208         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7209
7210         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0, type_tucmd,
7211                           mss_l4len_idx);
7212
7213         return 1;
7214 }
7215
7216 static void ixgbe_tx_csum(struct ixgbe_ring *tx_ring,
7217                           struct ixgbe_tx_buffer *first)
7218 {
7219         struct sk_buff *skb = first->skb;
7220         u32 vlan_macip_lens = 0;
7221         u32 mss_l4len_idx = 0;
7222         u32 type_tucmd = 0;
7223
7224         if (skb->ip_summed != CHECKSUM_PARTIAL) {
7225                 if (!(first->tx_flags & IXGBE_TX_FLAGS_HW_VLAN) &&
7226                     !(first->tx_flags & IXGBE_TX_FLAGS_CC))
7227                         return;
7228                 vlan_macip_lens = skb_network_offset(skb) <<
7229                                   IXGBE_ADVTXD_MACLEN_SHIFT;
7230         } else {
7231                 u8 l4_hdr = 0;
7232                 union {
7233                         struct iphdr *ipv4;
7234                         struct ipv6hdr *ipv6;
7235                         u8 *raw;
7236                 } network_hdr;
7237                 union {
7238                         struct tcphdr *tcphdr;
7239                         u8 *raw;
7240                 } transport_hdr;
7241                 __be16 frag_off;
7242
7243                 if (skb->encapsulation) {
7244                         network_hdr.raw = skb_inner_network_header(skb);
7245                         transport_hdr.raw = skb_inner_transport_header(skb);
7246                         vlan_macip_lens = skb_inner_network_offset(skb) <<
7247                                           IXGBE_ADVTXD_MACLEN_SHIFT;
7248                 } else {
7249                         network_hdr.raw = skb_network_header(skb);
7250                         transport_hdr.raw = skb_transport_header(skb);
7251                         vlan_macip_lens = skb_network_offset(skb) <<
7252                                           IXGBE_ADVTXD_MACLEN_SHIFT;
7253                 }
7254
7255                 /* use first 4 bits to determine IP version */
7256                 switch (network_hdr.ipv4->version) {
7257                 case IPVERSION:
7258                         vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7259                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
7260                         l4_hdr = network_hdr.ipv4->protocol;
7261                         break;
7262                 case 6:
7263                         vlan_macip_lens |= transport_hdr.raw - network_hdr.raw;
7264                         l4_hdr = network_hdr.ipv6->nexthdr;
7265                         if (likely((transport_hdr.raw - network_hdr.raw) ==
7266                                    sizeof(struct ipv6hdr)))
7267                                 break;
7268                         ipv6_skip_exthdr(skb, network_hdr.raw - skb->data +
7269                                               sizeof(struct ipv6hdr),
7270                                          &l4_hdr, &frag_off);
7271                         if (unlikely(frag_off))
7272                                 l4_hdr = NEXTHDR_FRAGMENT;
7273                         break;
7274                 default:
7275                         break;
7276                 }
7277
7278                 switch (l4_hdr) {
7279                 case IPPROTO_TCP:
7280                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
7281                         mss_l4len_idx = (transport_hdr.tcphdr->doff * 4) <<
7282                                         IXGBE_ADVTXD_L4LEN_SHIFT;
7283                         break;
7284                 case IPPROTO_SCTP:
7285                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
7286                         mss_l4len_idx = sizeof(struct sctphdr) <<
7287                                         IXGBE_ADVTXD_L4LEN_SHIFT;
7288                         break;
7289                 case IPPROTO_UDP:
7290                         mss_l4len_idx = sizeof(struct udphdr) <<
7291                                         IXGBE_ADVTXD_L4LEN_SHIFT;
7292                         break;
7293                 default:
7294                         if (unlikely(net_ratelimit())) {
7295                                 dev_warn(tx_ring->dev,
7296                                          "partial checksum, version=%d, l4 proto=%x\n",
7297                                          network_hdr.ipv4->version, l4_hdr);
7298                         }
7299                         skb_checksum_help(skb);
7300                         goto no_csum;
7301                 }
7302
7303                 /* update TX checksum flag */
7304                 first->tx_flags |= IXGBE_TX_FLAGS_CSUM;
7305         }
7306
7307 no_csum:
7308         /* vlan_macip_lens: MACLEN, VLAN tag */
7309         vlan_macip_lens |= first->tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
7310
7311         ixgbe_tx_ctxtdesc(tx_ring, vlan_macip_lens, 0,
7312                           type_tucmd, mss_l4len_idx);
7313 }
7314
7315 #define IXGBE_SET_FLAG(_input, _flag, _result) \
7316         ((_flag <= _result) ? \
7317          ((u32)(_input & _flag) * (_result / _flag)) : \
7318          ((u32)(_input & _flag) / (_flag / _result)))
7319
7320 static u32 ixgbe_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
7321 {
7322         /* set type for advanced descriptor with frame checksum insertion */
7323         u32 cmd_type = IXGBE_ADVTXD_DTYP_DATA |
7324                        IXGBE_ADVTXD_DCMD_DEXT |
7325                        IXGBE_ADVTXD_DCMD_IFCS;
7326
7327         /* set HW vlan bit if vlan is present */
7328         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_HW_VLAN,
7329                                    IXGBE_ADVTXD_DCMD_VLE);
7330
7331         /* set segmentation enable bits for TSO/FSO */
7332         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSO,
7333                                    IXGBE_ADVTXD_DCMD_TSE);
7334
7335         /* set timestamp bit if present */
7336         cmd_type |= IXGBE_SET_FLAG(tx_flags, IXGBE_TX_FLAGS_TSTAMP,
7337                                    IXGBE_ADVTXD_MAC_TSTAMP);
7338
7339         /* insert frame checksum */
7340         cmd_type ^= IXGBE_SET_FLAG(skb->no_fcs, 1, IXGBE_ADVTXD_DCMD_IFCS);
7341
7342         return cmd_type;
7343 }
7344
7345 static void ixgbe_tx_olinfo_status(union ixgbe_adv_tx_desc *tx_desc,
7346                                    u32 tx_flags, unsigned int paylen)
7347 {
7348         u32 olinfo_status = paylen << IXGBE_ADVTXD_PAYLEN_SHIFT;
7349
7350         /* enable L4 checksum for TSO and TX checksum offload */
7351         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7352                                         IXGBE_TX_FLAGS_CSUM,
7353                                         IXGBE_ADVTXD_POPTS_TXSM);
7354
7355         /* enble IPv4 checksum for TSO */
7356         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7357                                         IXGBE_TX_FLAGS_IPV4,
7358                                         IXGBE_ADVTXD_POPTS_IXSM);
7359
7360         /*
7361          * Check Context must be set if Tx switch is enabled, which it
7362          * always is for case where virtual functions are running
7363          */
7364         olinfo_status |= IXGBE_SET_FLAG(tx_flags,
7365                                         IXGBE_TX_FLAGS_CC,
7366                                         IXGBE_ADVTXD_CC);
7367
7368         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
7369 }
7370
7371 static int __ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7372 {
7373         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
7374
7375         /* Herbert's original patch had:
7376          *  smp_mb__after_netif_stop_queue();
7377          * but since that doesn't exist yet, just open code it.
7378          */
7379         smp_mb();
7380
7381         /* We need to check again in a case another CPU has just
7382          * made room available.
7383          */
7384         if (likely(ixgbe_desc_unused(tx_ring) < size))
7385                 return -EBUSY;
7386
7387         /* A reprieve! - use start_queue because it doesn't call schedule */
7388         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
7389         ++tx_ring->tx_stats.restart_queue;
7390         return 0;
7391 }
7392
7393 static inline int ixgbe_maybe_stop_tx(struct ixgbe_ring *tx_ring, u16 size)
7394 {
7395         if (likely(ixgbe_desc_unused(tx_ring) >= size))
7396                 return 0;
7397
7398         return __ixgbe_maybe_stop_tx(tx_ring, size);
7399 }
7400
7401 #define IXGBE_TXD_CMD (IXGBE_TXD_CMD_EOP | \
7402                        IXGBE_TXD_CMD_RS)
7403
7404 static void ixgbe_tx_map(struct ixgbe_ring *tx_ring,
7405                          struct ixgbe_tx_buffer *first,
7406                          const u8 hdr_len)
7407 {
7408         struct sk_buff *skb = first->skb;
7409         struct ixgbe_tx_buffer *tx_buffer;
7410         union ixgbe_adv_tx_desc *tx_desc;
7411         struct skb_frag_struct *frag;
7412         dma_addr_t dma;
7413         unsigned int data_len, size;
7414         u32 tx_flags = first->tx_flags;
7415         u32 cmd_type = ixgbe_tx_cmd_type(skb, tx_flags);
7416         u16 i = tx_ring->next_to_use;
7417
7418         tx_desc = IXGBE_TX_DESC(tx_ring, i);
7419
7420         ixgbe_tx_olinfo_status(tx_desc, tx_flags, skb->len - hdr_len);
7421
7422         size = skb_headlen(skb);
7423         data_len = skb->data_len;
7424
7425 #ifdef IXGBE_FCOE
7426         if (tx_flags & IXGBE_TX_FLAGS_FCOE) {
7427                 if (data_len < sizeof(struct fcoe_crc_eof)) {
7428                         size -= sizeof(struct fcoe_crc_eof) - data_len;
7429                         data_len = 0;
7430                 } else {
7431                         data_len -= sizeof(struct fcoe_crc_eof);
7432                 }
7433         }
7434
7435 #endif
7436         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
7437
7438         tx_buffer = first;
7439
7440         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
7441                 if (dma_mapping_error(tx_ring->dev, dma))
7442                         goto dma_error;
7443
7444                 /* record length, and DMA address */
7445                 dma_unmap_len_set(tx_buffer, len, size);
7446                 dma_unmap_addr_set(tx_buffer, dma, dma);
7447
7448                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
7449
7450                 while (unlikely(size > IXGBE_MAX_DATA_PER_TXD)) {
7451                         tx_desc->read.cmd_type_len =
7452                                 cpu_to_le32(cmd_type ^ IXGBE_MAX_DATA_PER_TXD);
7453
7454                         i++;
7455                         tx_desc++;
7456                         if (i == tx_ring->count) {
7457                                 tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7458                                 i = 0;
7459                         }
7460                         tx_desc->read.olinfo_status = 0;
7461
7462                         dma += IXGBE_MAX_DATA_PER_TXD;
7463                         size -= IXGBE_MAX_DATA_PER_TXD;
7464
7465                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
7466                 }
7467
7468                 if (likely(!data_len))
7469                         break;
7470
7471                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
7472
7473                 i++;
7474                 tx_desc++;
7475                 if (i == tx_ring->count) {
7476                         tx_desc = IXGBE_TX_DESC(tx_ring, 0);
7477                         i = 0;
7478                 }
7479                 tx_desc->read.olinfo_status = 0;
7480
7481 #ifdef IXGBE_FCOE
7482                 size = min_t(unsigned int, data_len, skb_frag_size(frag));
7483 #else
7484                 size = skb_frag_size(frag);
7485 #endif
7486                 data_len -= size;
7487
7488                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
7489                                        DMA_TO_DEVICE);
7490
7491                 tx_buffer = &tx_ring->tx_buffer_info[i];
7492         }
7493
7494         /* write last descriptor with RS and EOP bits */
7495         cmd_type |= size | IXGBE_TXD_CMD;
7496         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
7497
7498         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
7499
7500         /* set the timestamp */
7501         first->time_stamp = jiffies;
7502
7503         /*
7504          * Force memory writes to complete before letting h/w know there
7505          * are new descriptors to fetch.  (Only applicable for weak-ordered
7506          * memory model archs, such as IA-64).
7507          *
7508          * We also need this memory barrier to make certain all of the
7509          * status bits have been updated before next_to_watch is written.
7510          */
7511         wmb();
7512
7513         /* set next_to_watch value indicating a packet is present */
7514         first->next_to_watch = tx_desc;
7515
7516         i++;
7517         if (i == tx_ring->count)
7518                 i = 0;
7519
7520         tx_ring->next_to_use = i;
7521
7522         ixgbe_maybe_stop_tx(tx_ring, DESC_NEEDED);
7523
7524         if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
7525                 writel(i, tx_ring->tail);
7526
7527                 /* we need this if more than one processor can write to our tail
7528                  * at a time, it synchronizes IO on IA64/Altix systems
7529                  */
7530                 mmiowb();
7531         }
7532
7533         return;
7534 dma_error:
7535         dev_err(tx_ring->dev, "TX DMA map failed\n");
7536
7537         /* clear dma mappings for failed tx_buffer_info map */
7538         for (;;) {
7539                 tx_buffer = &tx_ring->tx_buffer_info[i];
7540                 ixgbe_unmap_and_free_tx_resource(tx_ring, tx_buffer);
7541                 if (tx_buffer == first)
7542                         break;
7543                 if (i == 0)
7544                         i = tx_ring->count;
7545                 i--;
7546         }
7547
7548         tx_ring->next_to_use = i;
7549 }
7550
7551 static void ixgbe_atr(struct ixgbe_ring *ring,
7552                       struct ixgbe_tx_buffer *first)
7553 {
7554         struct ixgbe_q_vector *q_vector = ring->q_vector;
7555         union ixgbe_atr_hash_dword input = { .dword = 0 };
7556         union ixgbe_atr_hash_dword common = { .dword = 0 };
7557         union {
7558                 unsigned char *network;
7559                 struct iphdr *ipv4;
7560                 struct ipv6hdr *ipv6;
7561         } hdr;
7562         struct tcphdr *th;
7563         struct sk_buff *skb;
7564 #ifdef CONFIG_IXGBE_VXLAN
7565         u8 encap = false;
7566 #endif /* CONFIG_IXGBE_VXLAN */
7567         __be16 vlan_id;
7568
7569         /* if ring doesn't have a interrupt vector, cannot perform ATR */
7570         if (!q_vector)
7571                 return;
7572
7573         /* do nothing if sampling is disabled */
7574         if (!ring->atr_sample_rate)
7575                 return;
7576
7577         ring->atr_count++;
7578
7579         /* snag network header to get L4 type and address */
7580         skb = first->skb;
7581         hdr.network = skb_network_header(skb);
7582         if (!skb->encapsulation) {
7583                 th = tcp_hdr(skb);
7584         } else {
7585 #ifdef CONFIG_IXGBE_VXLAN
7586                 struct ixgbe_adapter *adapter = q_vector->adapter;
7587
7588                 if (!adapter->vxlan_port)
7589                         return;
7590                 if (first->protocol != htons(ETH_P_IP) ||
7591                     hdr.ipv4->version != IPVERSION ||
7592                     hdr.ipv4->protocol != IPPROTO_UDP) {
7593                         return;
7594                 }
7595                 if (ntohs(udp_hdr(skb)->dest) != adapter->vxlan_port)
7596                         return;
7597                 encap = true;
7598                 hdr.network = skb_inner_network_header(skb);
7599                 th = inner_tcp_hdr(skb);
7600 #else
7601                 return;
7602 #endif /* CONFIG_IXGBE_VXLAN */
7603         }
7604
7605         /* Currently only IPv4/IPv6 with TCP is supported */
7606         switch (hdr.ipv4->version) {
7607         case IPVERSION:
7608                 if (hdr.ipv4->protocol != IPPROTO_TCP)
7609                         return;
7610                 break;
7611         case 6:
7612                 if (likely((unsigned char *)th - hdr.network ==
7613                            sizeof(struct ipv6hdr))) {
7614                         if (hdr.ipv6->nexthdr != IPPROTO_TCP)
7615                                 return;
7616                 } else {
7617                         __be16 frag_off;
7618                         u8 l4_hdr;
7619
7620                         ipv6_skip_exthdr(skb, hdr.network - skb->data +
7621                                               sizeof(struct ipv6hdr),
7622                                          &l4_hdr, &frag_off);
7623                         if (unlikely(frag_off))
7624                                 return;
7625                         if (l4_hdr != IPPROTO_TCP)
7626                                 return;
7627                 }
7628                 break;
7629         default:
7630                 return;
7631         }
7632
7633         /* skip this packet since it is invalid or the socket is closing */
7634         if (!th || th->fin)
7635                 return;
7636
7637         /* sample on all syn packets or once every atr sample count */
7638         if (!th->syn && (ring->atr_count < ring->atr_sample_rate))
7639                 return;
7640
7641         /* reset sample count */
7642         ring->atr_count = 0;
7643
7644         vlan_id = htons(first->tx_flags >> IXGBE_TX_FLAGS_VLAN_SHIFT);
7645
7646         /*
7647          * src and dst are inverted, think how the receiver sees them
7648          *
7649          * The input is broken into two sections, a non-compressed section
7650          * containing vm_pool, vlan_id, and flow_type.  The rest of the data
7651          * is XORed together and stored in the compressed dword.
7652          */
7653         input.formatted.vlan_id = vlan_id;
7654
7655         /*
7656          * since src port and flex bytes occupy the same word XOR them together
7657          * and write the value to source port portion of compressed dword
7658          */
7659         if (first->tx_flags & (IXGBE_TX_FLAGS_SW_VLAN | IXGBE_TX_FLAGS_HW_VLAN))
7660                 common.port.src ^= th->dest ^ htons(ETH_P_8021Q);
7661         else
7662                 common.port.src ^= th->dest ^ first->protocol;
7663         common.port.dst ^= th->source;
7664
7665         switch (hdr.ipv4->version) {
7666         case IPVERSION:
7667                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV4;
7668                 common.ip ^= hdr.ipv4->saddr ^ hdr.ipv4->daddr;
7669                 break;
7670         case 6:
7671                 input.formatted.flow_type = IXGBE_ATR_FLOW_TYPE_TCPV6;
7672                 common.ip ^= hdr.ipv6->saddr.s6_addr32[0] ^
7673                              hdr.ipv6->saddr.s6_addr32[1] ^
7674                              hdr.ipv6->saddr.s6_addr32[2] ^
7675                              hdr.ipv6->saddr.s6_addr32[3] ^
7676                              hdr.ipv6->daddr.s6_addr32[0] ^
7677                              hdr.ipv6->daddr.s6_addr32[1] ^
7678                              hdr.ipv6->daddr.s6_addr32[2] ^
7679                              hdr.ipv6->daddr.s6_addr32[3];
7680                 break;
7681         default:
7682                 break;
7683         }
7684
7685 #ifdef CONFIG_IXGBE_VXLAN
7686         if (encap)
7687                 input.formatted.flow_type |= IXGBE_ATR_L4TYPE_TUNNEL_MASK;
7688 #endif /* CONFIG_IXGBE_VXLAN */
7689
7690         /* This assumes the Rx queue and Tx queue are bound to the same CPU */
7691         ixgbe_fdir_add_signature_filter_82599(&q_vector->adapter->hw,
7692                                               input, common, ring->queue_index);
7693 }
7694
7695 static u16 ixgbe_select_queue(struct net_device *dev, struct sk_buff *skb,
7696                               void *accel_priv, select_queue_fallback_t fallback)
7697 {
7698         struct ixgbe_fwd_adapter *fwd_adapter = accel_priv;
7699 #ifdef IXGBE_FCOE
7700         struct ixgbe_adapter *adapter;
7701         struct ixgbe_ring_feature *f;
7702         int txq;
7703 #endif
7704
7705         if (fwd_adapter)
7706                 return skb->queue_mapping + fwd_adapter->tx_base_queue;
7707
7708 #ifdef IXGBE_FCOE
7709
7710         /*
7711          * only execute the code below if protocol is FCoE
7712          * or FIP and we have FCoE enabled on the adapter
7713          */
7714         switch (vlan_get_protocol(skb)) {
7715         case htons(ETH_P_FCOE):
7716         case htons(ETH_P_FIP):
7717                 adapter = netdev_priv(dev);
7718
7719                 if (adapter->flags & IXGBE_FLAG_FCOE_ENABLED)
7720                         break;
7721         default:
7722                 return fallback(dev, skb);
7723         }
7724
7725         f = &adapter->ring_feature[RING_F_FCOE];
7726
7727         txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) :
7728                                            smp_processor_id();
7729
7730         while (txq >= f->indices)
7731                 txq -= f->indices;
7732
7733         return txq + f->offset;
7734 #else
7735         return fallback(dev, skb);
7736 #endif
7737 }
7738
7739 netdev_tx_t ixgbe_xmit_frame_ring(struct sk_buff *skb,
7740                           struct ixgbe_adapter *adapter,
7741                           struct ixgbe_ring *tx_ring)
7742 {
7743         struct ixgbe_tx_buffer *first;
7744         int tso;
7745         u32 tx_flags = 0;
7746         unsigned short f;
7747         u16 count = TXD_USE_COUNT(skb_headlen(skb));
7748         __be16 protocol = skb->protocol;
7749         u8 hdr_len = 0;
7750
7751         /*
7752          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
7753          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
7754          *       + 2 desc gap to keep tail from touching head,
7755          *       + 1 desc for context descriptor,
7756          * otherwise try next time
7757          */
7758         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
7759                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
7760
7761         if (ixgbe_maybe_stop_tx(tx_ring, count + 3)) {
7762                 tx_ring->tx_stats.tx_busy++;
7763                 return NETDEV_TX_BUSY;
7764         }
7765
7766         /* record the location of the first descriptor for this packet */
7767         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
7768         first->skb = skb;
7769         first->bytecount = skb->len;
7770         first->gso_segs = 1;
7771
7772         /* if we have a HW VLAN tag being added default to the HW one */
7773         if (skb_vlan_tag_present(skb)) {
7774                 tx_flags |= skb_vlan_tag_get(skb) << IXGBE_TX_FLAGS_VLAN_SHIFT;
7775                 tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7776         /* else if it is a SW VLAN check the next protocol and store the tag */
7777         } else if (protocol == htons(ETH_P_8021Q)) {
7778                 struct vlan_hdr *vhdr, _vhdr;
7779                 vhdr = skb_header_pointer(skb, ETH_HLEN, sizeof(_vhdr), &_vhdr);
7780                 if (!vhdr)
7781                         goto out_drop;
7782
7783                 tx_flags |= ntohs(vhdr->h_vlan_TCI) <<
7784                                   IXGBE_TX_FLAGS_VLAN_SHIFT;
7785                 tx_flags |= IXGBE_TX_FLAGS_SW_VLAN;
7786         }
7787         protocol = vlan_get_protocol(skb);
7788
7789         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
7790             adapter->ptp_clock &&
7791             !test_and_set_bit_lock(__IXGBE_PTP_TX_IN_PROGRESS,
7792                                    &adapter->state)) {
7793                 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
7794                 tx_flags |= IXGBE_TX_FLAGS_TSTAMP;
7795
7796                 /* schedule check for Tx timestamp */
7797                 adapter->ptp_tx_skb = skb_get(skb);
7798                 adapter->ptp_tx_start = jiffies;
7799                 schedule_work(&adapter->ptp_tx_work);
7800         }
7801
7802         skb_tx_timestamp(skb);
7803
7804 #ifdef CONFIG_PCI_IOV
7805         /*
7806          * Use the l2switch_enable flag - would be false if the DMA
7807          * Tx switch had been disabled.
7808          */
7809         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED)
7810                 tx_flags |= IXGBE_TX_FLAGS_CC;
7811
7812 #endif
7813         /* DCB maps skb priorities 0-7 onto 3 bit PCP of VLAN tag. */
7814         if ((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
7815             ((tx_flags & (IXGBE_TX_FLAGS_HW_VLAN | IXGBE_TX_FLAGS_SW_VLAN)) ||
7816              (skb->priority != TC_PRIO_CONTROL))) {
7817                 tx_flags &= ~IXGBE_TX_FLAGS_VLAN_PRIO_MASK;
7818                 tx_flags |= (skb->priority & 0x7) <<
7819                                         IXGBE_TX_FLAGS_VLAN_PRIO_SHIFT;
7820                 if (tx_flags & IXGBE_TX_FLAGS_SW_VLAN) {
7821                         struct vlan_ethhdr *vhdr;
7822
7823                         if (skb_cow_head(skb, 0))
7824                                 goto out_drop;
7825                         vhdr = (struct vlan_ethhdr *)skb->data;
7826                         vhdr->h_vlan_TCI = htons(tx_flags >>
7827                                                  IXGBE_TX_FLAGS_VLAN_SHIFT);
7828                 } else {
7829                         tx_flags |= IXGBE_TX_FLAGS_HW_VLAN;
7830                 }
7831         }
7832
7833         /* record initial flags and protocol */
7834         first->tx_flags = tx_flags;
7835         first->protocol = protocol;
7836
7837 #ifdef IXGBE_FCOE
7838         /* setup tx offload for FCoE */
7839         if ((protocol == htons(ETH_P_FCOE)) &&
7840             (tx_ring->netdev->features & (NETIF_F_FSO | NETIF_F_FCOE_CRC))) {
7841                 tso = ixgbe_fso(tx_ring, first, &hdr_len);
7842                 if (tso < 0)
7843                         goto out_drop;
7844
7845                 goto xmit_fcoe;
7846         }
7847
7848 #endif /* IXGBE_FCOE */
7849         tso = ixgbe_tso(tx_ring, first, &hdr_len);
7850         if (tso < 0)
7851                 goto out_drop;
7852         else if (!tso)
7853                 ixgbe_tx_csum(tx_ring, first);
7854
7855         /* add the ATR filter if ATR is on */
7856         if (test_bit(__IXGBE_TX_FDIR_INIT_DONE, &tx_ring->state))
7857                 ixgbe_atr(tx_ring, first);
7858
7859 #ifdef IXGBE_FCOE
7860 xmit_fcoe:
7861 #endif /* IXGBE_FCOE */
7862         ixgbe_tx_map(tx_ring, first, hdr_len);
7863
7864         return NETDEV_TX_OK;
7865
7866 out_drop:
7867         dev_kfree_skb_any(first->skb);
7868         first->skb = NULL;
7869
7870         return NETDEV_TX_OK;
7871 }
7872
7873 static netdev_tx_t __ixgbe_xmit_frame(struct sk_buff *skb,
7874                                       struct net_device *netdev,
7875                                       struct ixgbe_ring *ring)
7876 {
7877         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7878         struct ixgbe_ring *tx_ring;
7879
7880         /*
7881          * The minimum packet size for olinfo paylen is 17 so pad the skb
7882          * in order to meet this minimum size requirement.
7883          */
7884         if (skb_put_padto(skb, 17))
7885                 return NETDEV_TX_OK;
7886
7887         tx_ring = ring ? ring : adapter->tx_ring[skb->queue_mapping];
7888
7889         return ixgbe_xmit_frame_ring(skb, adapter, tx_ring);
7890 }
7891
7892 static netdev_tx_t ixgbe_xmit_frame(struct sk_buff *skb,
7893                                     struct net_device *netdev)
7894 {
7895         return __ixgbe_xmit_frame(skb, netdev, NULL);
7896 }
7897
7898 /**
7899  * ixgbe_set_mac - Change the Ethernet Address of the NIC
7900  * @netdev: network interface device structure
7901  * @p: pointer to an address structure
7902  *
7903  * Returns 0 on success, negative on failure
7904  **/
7905 static int ixgbe_set_mac(struct net_device *netdev, void *p)
7906 {
7907         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7908         struct ixgbe_hw *hw = &adapter->hw;
7909         struct sockaddr *addr = p;
7910
7911         if (!is_valid_ether_addr(addr->sa_data))
7912                 return -EADDRNOTAVAIL;
7913
7914         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
7915         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
7916
7917         ixgbe_mac_set_default_filter(adapter);
7918
7919         return 0;
7920 }
7921
7922 static int
7923 ixgbe_mdio_read(struct net_device *netdev, int prtad, int devad, u16 addr)
7924 {
7925         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7926         struct ixgbe_hw *hw = &adapter->hw;
7927         u16 value;
7928         int rc;
7929
7930         if (prtad != hw->phy.mdio.prtad)
7931                 return -EINVAL;
7932         rc = hw->phy.ops.read_reg(hw, addr, devad, &value);
7933         if (!rc)
7934                 rc = value;
7935         return rc;
7936 }
7937
7938 static int ixgbe_mdio_write(struct net_device *netdev, int prtad, int devad,
7939                             u16 addr, u16 value)
7940 {
7941         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7942         struct ixgbe_hw *hw = &adapter->hw;
7943
7944         if (prtad != hw->phy.mdio.prtad)
7945                 return -EINVAL;
7946         return hw->phy.ops.write_reg(hw, addr, devad, value);
7947 }
7948
7949 static int ixgbe_ioctl(struct net_device *netdev, struct ifreq *req, int cmd)
7950 {
7951         struct ixgbe_adapter *adapter = netdev_priv(netdev);
7952
7953         switch (cmd) {
7954         case SIOCSHWTSTAMP:
7955                 return ixgbe_ptp_set_ts_config(adapter, req);
7956         case SIOCGHWTSTAMP:
7957                 return ixgbe_ptp_get_ts_config(adapter, req);
7958         default:
7959                 return mdio_mii_ioctl(&adapter->hw.phy.mdio, if_mii(req), cmd);
7960         }
7961 }
7962
7963 /**
7964  * ixgbe_add_sanmac_netdev - Add the SAN MAC address to the corresponding
7965  * netdev->dev_addrs
7966  * @netdev: network interface device structure
7967  *
7968  * Returns non-zero on failure
7969  **/
7970 static int ixgbe_add_sanmac_netdev(struct net_device *dev)
7971 {
7972         int err = 0;
7973         struct ixgbe_adapter *adapter = netdev_priv(dev);
7974         struct ixgbe_hw *hw = &adapter->hw;
7975
7976         if (is_valid_ether_addr(hw->mac.san_addr)) {
7977                 rtnl_lock();
7978                 err = dev_addr_add(dev, hw->mac.san_addr, NETDEV_HW_ADDR_T_SAN);
7979                 rtnl_unlock();
7980
7981                 /* update SAN MAC vmdq pool selection */
7982                 hw->mac.ops.set_vmdq_san_mac(hw, VMDQ_P(0));
7983         }
7984         return err;
7985 }
7986
7987 /**
7988  * ixgbe_del_sanmac_netdev - Removes the SAN MAC address to the corresponding
7989  * netdev->dev_addrs
7990  * @netdev: network interface device structure
7991  *
7992  * Returns non-zero on failure
7993  **/
7994 static int ixgbe_del_sanmac_netdev(struct net_device *dev)
7995 {
7996         int err = 0;
7997         struct ixgbe_adapter *adapter = netdev_priv(dev);
7998         struct ixgbe_mac_info *mac = &adapter->hw.mac;
7999
8000         if (is_valid_ether_addr(mac->san_addr)) {
8001                 rtnl_lock();
8002                 err = dev_addr_del(dev, mac->san_addr, NETDEV_HW_ADDR_T_SAN);
8003                 rtnl_unlock();
8004         }
8005         return err;
8006 }
8007
8008 #ifdef CONFIG_NET_POLL_CONTROLLER
8009 /*
8010  * Polling 'interrupt' - used by things like netconsole to send skbs
8011  * without having to re-enable interrupts. It's not called while
8012  * the interrupt routine is executing.
8013  */
8014 static void ixgbe_netpoll(struct net_device *netdev)
8015 {
8016         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8017         int i;
8018
8019         /* if interface is down do nothing */
8020         if (test_bit(__IXGBE_DOWN, &adapter->state))
8021                 return;
8022
8023         /* loop through and schedule all active queues */
8024         for (i = 0; i < adapter->num_q_vectors; i++)
8025                 ixgbe_msix_clean_rings(0, adapter->q_vector[i]);
8026 }
8027
8028 #endif
8029 static struct rtnl_link_stats64 *ixgbe_get_stats64(struct net_device *netdev,
8030                                                    struct rtnl_link_stats64 *stats)
8031 {
8032         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8033         int i;
8034
8035         rcu_read_lock();
8036         for (i = 0; i < adapter->num_rx_queues; i++) {
8037                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->rx_ring[i]);
8038                 u64 bytes, packets;
8039                 unsigned int start;
8040
8041                 if (ring) {
8042                         do {
8043                                 start = u64_stats_fetch_begin_irq(&ring->syncp);
8044                                 packets = ring->stats.packets;
8045                                 bytes   = ring->stats.bytes;
8046                         } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8047                         stats->rx_packets += packets;
8048                         stats->rx_bytes   += bytes;
8049                 }
8050         }
8051
8052         for (i = 0; i < adapter->num_tx_queues; i++) {
8053                 struct ixgbe_ring *ring = ACCESS_ONCE(adapter->tx_ring[i]);
8054                 u64 bytes, packets;
8055                 unsigned int start;
8056
8057                 if (ring) {
8058                         do {
8059                                 start = u64_stats_fetch_begin_irq(&ring->syncp);
8060                                 packets = ring->stats.packets;
8061                                 bytes   = ring->stats.bytes;
8062                         } while (u64_stats_fetch_retry_irq(&ring->syncp, start));
8063                         stats->tx_packets += packets;
8064                         stats->tx_bytes   += bytes;
8065                 }
8066         }
8067         rcu_read_unlock();
8068         /* following stats updated by ixgbe_watchdog_task() */
8069         stats->multicast        = netdev->stats.multicast;
8070         stats->rx_errors        = netdev->stats.rx_errors;
8071         stats->rx_length_errors = netdev->stats.rx_length_errors;
8072         stats->rx_crc_errors    = netdev->stats.rx_crc_errors;
8073         stats->rx_missed_errors = netdev->stats.rx_missed_errors;
8074         return stats;
8075 }
8076
8077 #ifdef CONFIG_IXGBE_DCB
8078 /**
8079  * ixgbe_validate_rtr - verify 802.1Qp to Rx packet buffer mapping is valid.
8080  * @adapter: pointer to ixgbe_adapter
8081  * @tc: number of traffic classes currently enabled
8082  *
8083  * Configure a valid 802.1Qp to Rx packet buffer mapping ie confirm
8084  * 802.1Q priority maps to a packet buffer that exists.
8085  */
8086 static void ixgbe_validate_rtr(struct ixgbe_adapter *adapter, u8 tc)
8087 {
8088         struct ixgbe_hw *hw = &adapter->hw;
8089         u32 reg, rsave;
8090         int i;
8091
8092         /* 82598 have a static priority to TC mapping that can not
8093          * be changed so no validation is needed.
8094          */
8095         if (hw->mac.type == ixgbe_mac_82598EB)
8096                 return;
8097
8098         reg = IXGBE_READ_REG(hw, IXGBE_RTRUP2TC);
8099         rsave = reg;
8100
8101         for (i = 0; i < MAX_TRAFFIC_CLASS; i++) {
8102                 u8 up2tc = reg >> (i * IXGBE_RTRUP2TC_UP_SHIFT);
8103
8104                 /* If up2tc is out of bounds default to zero */
8105                 if (up2tc > tc)
8106                         reg &= ~(0x7 << IXGBE_RTRUP2TC_UP_SHIFT);
8107         }
8108
8109         if (reg != rsave)
8110                 IXGBE_WRITE_REG(hw, IXGBE_RTRUP2TC, reg);
8111
8112         return;
8113 }
8114
8115 /**
8116  * ixgbe_set_prio_tc_map - Configure netdev prio tc map
8117  * @adapter: Pointer to adapter struct
8118  *
8119  * Populate the netdev user priority to tc map
8120  */
8121 static void ixgbe_set_prio_tc_map(struct ixgbe_adapter *adapter)
8122 {
8123         struct net_device *dev = adapter->netdev;
8124         struct ixgbe_dcb_config *dcb_cfg = &adapter->dcb_cfg;
8125         struct ieee_ets *ets = adapter->ixgbe_ieee_ets;
8126         u8 prio;
8127
8128         for (prio = 0; prio < MAX_USER_PRIORITY; prio++) {
8129                 u8 tc = 0;
8130
8131                 if (adapter->dcbx_cap & DCB_CAP_DCBX_VER_CEE)
8132                         tc = ixgbe_dcb_get_tc_from_up(dcb_cfg, 0, prio);
8133                 else if (ets)
8134                         tc = ets->prio_tc[prio];
8135
8136                 netdev_set_prio_tc_map(dev, prio, tc);
8137         }
8138 }
8139
8140 #endif /* CONFIG_IXGBE_DCB */
8141 /**
8142  * ixgbe_setup_tc - configure net_device for multiple traffic classes
8143  *
8144  * @netdev: net device to configure
8145  * @tc: number of traffic classes to enable
8146  */
8147 int ixgbe_setup_tc(struct net_device *dev, u8 tc)
8148 {
8149         struct ixgbe_adapter *adapter = netdev_priv(dev);
8150         struct ixgbe_hw *hw = &adapter->hw;
8151         bool pools;
8152
8153         /* Hardware supports up to 8 traffic classes */
8154         if (tc > adapter->dcb_cfg.num_tcs.pg_tcs)
8155                 return -EINVAL;
8156
8157         if (hw->mac.type == ixgbe_mac_82598EB && tc && tc < MAX_TRAFFIC_CLASS)
8158                 return -EINVAL;
8159
8160         pools = (find_first_zero_bit(&adapter->fwd_bitmask, 32) > 1);
8161         if (tc && pools && adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS)
8162                 return -EBUSY;
8163
8164         /* Hardware has to reinitialize queues and interrupts to
8165          * match packet buffer alignment. Unfortunately, the
8166          * hardware is not flexible enough to do this dynamically.
8167          */
8168         if (netif_running(dev))
8169                 ixgbe_close(dev);
8170         else
8171                 ixgbe_reset(adapter);
8172
8173         ixgbe_clear_interrupt_scheme(adapter);
8174
8175 #ifdef CONFIG_IXGBE_DCB
8176         if (tc) {
8177                 netdev_set_num_tc(dev, tc);
8178                 ixgbe_set_prio_tc_map(adapter);
8179
8180                 adapter->flags |= IXGBE_FLAG_DCB_ENABLED;
8181
8182                 if (adapter->hw.mac.type == ixgbe_mac_82598EB) {
8183                         adapter->last_lfc_mode = adapter->hw.fc.requested_mode;
8184                         adapter->hw.fc.requested_mode = ixgbe_fc_none;
8185                 }
8186         } else {
8187                 netdev_reset_tc(dev);
8188
8189                 if (adapter->hw.mac.type == ixgbe_mac_82598EB)
8190                         adapter->hw.fc.requested_mode = adapter->last_lfc_mode;
8191
8192                 adapter->flags &= ~IXGBE_FLAG_DCB_ENABLED;
8193
8194                 adapter->temp_dcb_cfg.pfc_mode_enable = false;
8195                 adapter->dcb_cfg.pfc_mode_enable = false;
8196         }
8197
8198         ixgbe_validate_rtr(adapter, tc);
8199
8200 #endif /* CONFIG_IXGBE_DCB */
8201         ixgbe_init_interrupt_scheme(adapter);
8202
8203         if (netif_running(dev))
8204                 return ixgbe_open(dev);
8205
8206         return 0;
8207 }
8208
8209 static int ixgbe_delete_clsu32(struct ixgbe_adapter *adapter,
8210                                struct tc_cls_u32_offload *cls)
8211 {
8212         int err;
8213
8214         spin_lock(&adapter->fdir_perfect_lock);
8215         err = ixgbe_update_ethtool_fdir_entry(adapter, NULL, cls->knode.handle);
8216         spin_unlock(&adapter->fdir_perfect_lock);
8217         return err;
8218 }
8219
8220 static int ixgbe_configure_clsu32_add_hnode(struct ixgbe_adapter *adapter,
8221                                             __be16 protocol,
8222                                             struct tc_cls_u32_offload *cls)
8223 {
8224         /* This ixgbe devices do not support hash tables at the moment
8225          * so abort when given hash tables.
8226          */
8227         if (cls->hnode.divisor > 0)
8228                 return -EINVAL;
8229
8230         set_bit(TC_U32_USERHTID(cls->hnode.handle), &adapter->tables);
8231         return 0;
8232 }
8233
8234 static int ixgbe_configure_clsu32_del_hnode(struct ixgbe_adapter *adapter,
8235                                             struct tc_cls_u32_offload *cls)
8236 {
8237         clear_bit(TC_U32_USERHTID(cls->hnode.handle), &adapter->tables);
8238         return 0;
8239 }
8240
8241 static int ixgbe_configure_clsu32(struct ixgbe_adapter *adapter,
8242                                   __be16 protocol,
8243                                   struct tc_cls_u32_offload *cls)
8244 {
8245         u32 loc = cls->knode.handle & 0xfffff;
8246         struct ixgbe_hw *hw = &adapter->hw;
8247         struct ixgbe_mat_field *field_ptr;
8248         struct ixgbe_fdir_filter *input;
8249         union ixgbe_atr_input mask;
8250 #ifdef CONFIG_NET_CLS_ACT
8251         const struct tc_action *a;
8252 #endif
8253         int i, err = 0;
8254         u8 queue;
8255         u32 handle;
8256
8257         memset(&mask, 0, sizeof(union ixgbe_atr_input));
8258         handle = cls->knode.handle;
8259
8260         /* At the moment cls_u32 jumps to transport layer and skips past
8261          * L2 headers. The canonical method to match L2 frames is to use
8262          * negative values. However this is error prone at best but really
8263          * just broken because there is no way to "know" what sort of hdr
8264          * is in front of the transport layer. Fix cls_u32 to support L2
8265          * headers when needed.
8266          */
8267         if (protocol != htons(ETH_P_IP))
8268                 return -EINVAL;
8269
8270         if (cls->knode.link_handle ||
8271             cls->knode.link_handle >= IXGBE_MAX_LINK_HANDLE) {
8272                 struct ixgbe_nexthdr *nexthdr = ixgbe_ipv4_jumps;
8273                 u32 uhtid = TC_U32_USERHTID(cls->knode.link_handle);
8274
8275                 if (!test_bit(uhtid, &adapter->tables))
8276                         return -EINVAL;
8277
8278                 for (i = 0; nexthdr[i].jump; i++) {
8279                         if (nexthdr->o != cls->knode.sel->offoff ||
8280                             nexthdr->s != cls->knode.sel->offshift ||
8281                             nexthdr->m != cls->knode.sel->offmask ||
8282                             /* do not support multiple key jumps its just mad */
8283                             cls->knode.sel->nkeys > 1)
8284                                 return -EINVAL;
8285
8286                         if (nexthdr->off != cls->knode.sel->keys[0].off ||
8287                             nexthdr->val != cls->knode.sel->keys[0].val ||
8288                             nexthdr->mask != cls->knode.sel->keys[0].mask)
8289                                 return -EINVAL;
8290
8291                         if (uhtid >= IXGBE_MAX_LINK_HANDLE)
8292                                 return -EINVAL;
8293
8294                         adapter->jump_tables[uhtid] = nexthdr->jump;
8295                 }
8296                 return 0;
8297         }
8298
8299         if (loc >= ((1024 << adapter->fdir_pballoc) - 2)) {
8300                 e_err(drv, "Location out of range\n");
8301                 return -EINVAL;
8302         }
8303
8304         /* cls u32 is a graph starting at root node 0x800. The driver tracks
8305          * links and also the fields used to advance the parser across each
8306          * link (e.g. nexthdr/eat parameters from 'tc'). This way we can map
8307          * the u32 graph onto the hardware parse graph denoted in ixgbe_model.h
8308          * To add support for new nodes update ixgbe_model.h parse structures
8309          * this function _should_ be generic try not to hardcode values here.
8310          */
8311         if (TC_U32_USERHTID(handle) == 0x800) {
8312                 field_ptr = adapter->jump_tables[0];
8313         } else {
8314                 if (TC_U32_USERHTID(handle) >= ARRAY_SIZE(adapter->jump_tables))
8315                         return -EINVAL;
8316
8317                 field_ptr = adapter->jump_tables[TC_U32_USERHTID(handle)];
8318         }
8319
8320         if (!field_ptr)
8321                 return -EINVAL;
8322
8323         input = kzalloc(sizeof(*input), GFP_KERNEL);
8324         if (!input)
8325                 return -ENOMEM;
8326
8327         for (i = 0; i < cls->knode.sel->nkeys; i++) {
8328                 int off = cls->knode.sel->keys[i].off;
8329                 __be32 val = cls->knode.sel->keys[i].val;
8330                 __be32 m = cls->knode.sel->keys[i].mask;
8331                 bool found_entry = false;
8332                 int j;
8333
8334                 for (j = 0; field_ptr[j].val; j++) {
8335                         if (field_ptr[j].off == off &&
8336                             field_ptr[j].mask == m) {
8337                                 field_ptr[j].val(input, &mask, val, m);
8338                                 input->filter.formatted.flow_type |=
8339                                         field_ptr[j].type;
8340                                 found_entry = true;
8341                                 break;
8342                         }
8343                 }
8344
8345                 if (!found_entry)
8346                         goto err_out;
8347         }
8348
8349         mask.formatted.flow_type = IXGBE_ATR_L4TYPE_IPV6_MASK |
8350                                    IXGBE_ATR_L4TYPE_MASK;
8351
8352         if (input->filter.formatted.flow_type == IXGBE_ATR_FLOW_TYPE_IPV4)
8353                 mask.formatted.flow_type &= IXGBE_ATR_L4TYPE_IPV6_MASK;
8354
8355 #ifdef CONFIG_NET_CLS_ACT
8356         if (list_empty(&cls->knode.exts->actions))
8357                 goto err_out;
8358
8359         list_for_each_entry(a, &cls->knode.exts->actions, list) {
8360                 if (!is_tcf_gact_shot(a))
8361                         goto err_out;
8362         }
8363 #endif
8364
8365         input->action = IXGBE_FDIR_DROP_QUEUE;
8366         queue = IXGBE_FDIR_DROP_QUEUE;
8367         input->sw_idx = loc;
8368
8369         spin_lock(&adapter->fdir_perfect_lock);
8370
8371         if (hlist_empty(&adapter->fdir_filter_list)) {
8372                 memcpy(&adapter->fdir_mask, &mask, sizeof(mask));
8373                 err = ixgbe_fdir_set_input_mask_82599(hw, &mask);
8374                 if (err)
8375                         goto err_out_w_lock;
8376         } else if (memcmp(&adapter->fdir_mask, &mask, sizeof(mask))) {
8377                 err = -EINVAL;
8378                 goto err_out_w_lock;
8379         }
8380
8381         ixgbe_atr_compute_perfect_hash_82599(&input->filter, &mask);
8382         err = ixgbe_fdir_write_perfect_filter_82599(hw, &input->filter,
8383                                                     input->sw_idx, queue);
8384         if (!err)
8385                 ixgbe_update_ethtool_fdir_entry(adapter, input, input->sw_idx);
8386         spin_unlock(&adapter->fdir_perfect_lock);
8387
8388         return err;
8389 err_out_w_lock:
8390         spin_unlock(&adapter->fdir_perfect_lock);
8391 err_out:
8392         kfree(input);
8393         return -EINVAL;
8394 }
8395
8396 int __ixgbe_setup_tc(struct net_device *dev, u32 handle, __be16 proto,
8397                      struct tc_to_netdev *tc)
8398 {
8399         struct ixgbe_adapter *adapter = netdev_priv(dev);
8400
8401         if (TC_H_MAJ(handle) == TC_H_MAJ(TC_H_INGRESS) &&
8402             tc->type == TC_SETUP_CLSU32) {
8403                 switch (tc->cls_u32->command) {
8404                 case TC_CLSU32_NEW_KNODE:
8405                 case TC_CLSU32_REPLACE_KNODE:
8406                         return ixgbe_configure_clsu32(adapter,
8407                                                       proto, tc->cls_u32);
8408                 case TC_CLSU32_DELETE_KNODE:
8409                         return ixgbe_delete_clsu32(adapter, tc->cls_u32);
8410                 case TC_CLSU32_NEW_HNODE:
8411                 case TC_CLSU32_REPLACE_HNODE:
8412                         return ixgbe_configure_clsu32_add_hnode(adapter, proto,
8413                                                                 tc->cls_u32);
8414                 case TC_CLSU32_DELETE_HNODE:
8415                         return ixgbe_configure_clsu32_del_hnode(adapter,
8416                                                                 tc->cls_u32);
8417                 default:
8418                         return -EINVAL;
8419                 }
8420         }
8421
8422         if (tc->type != TC_SETUP_MQPRIO)
8423                 return -EINVAL;
8424
8425         return ixgbe_setup_tc(dev, tc->tc);
8426 }
8427
8428 #ifdef CONFIG_PCI_IOV
8429 void ixgbe_sriov_reinit(struct ixgbe_adapter *adapter)
8430 {
8431         struct net_device *netdev = adapter->netdev;
8432
8433         rtnl_lock();
8434         ixgbe_setup_tc(netdev, netdev_get_num_tc(netdev));
8435         rtnl_unlock();
8436 }
8437
8438 #endif
8439 void ixgbe_do_reset(struct net_device *netdev)
8440 {
8441         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8442
8443         if (netif_running(netdev))
8444                 ixgbe_reinit_locked(adapter);
8445         else
8446                 ixgbe_reset(adapter);
8447 }
8448
8449 static netdev_features_t ixgbe_fix_features(struct net_device *netdev,
8450                                             netdev_features_t features)
8451 {
8452         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8453
8454         /* If Rx checksum is disabled, then RSC/LRO should also be disabled */
8455         if (!(features & NETIF_F_RXCSUM))
8456                 features &= ~NETIF_F_LRO;
8457
8458         /* Turn off LRO if not RSC capable */
8459         if (!(adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE))
8460                 features &= ~NETIF_F_LRO;
8461
8462         return features;
8463 }
8464
8465 static int ixgbe_set_features(struct net_device *netdev,
8466                               netdev_features_t features)
8467 {
8468         struct ixgbe_adapter *adapter = netdev_priv(netdev);
8469         netdev_features_t changed = netdev->features ^ features;
8470         bool need_reset = false;
8471
8472         /* Make sure RSC matches LRO, reset if change */
8473         if (!(features & NETIF_F_LRO)) {
8474                 if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
8475                         need_reset = true;
8476                 adapter->flags2 &= ~IXGBE_FLAG2_RSC_ENABLED;
8477         } else if ((adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE) &&
8478                    !(adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)) {
8479                 if (adapter->rx_itr_setting == 1 ||
8480                     adapter->rx_itr_setting > IXGBE_MIN_RSC_ITR) {
8481                         adapter->flags2 |= IXGBE_FLAG2_RSC_ENABLED;
8482                         need_reset = true;
8483                 } else if ((changed ^ features) & NETIF_F_LRO) {
8484                         e_info(probe, "rx-usecs set too low, "
8485                                "disabling RSC\n");
8486                 }
8487         }
8488
8489         /*
8490          * Check if Flow Director n-tuple support or hw_tc support was
8491          * enabled or disabled.  If the state changed, we need to reset.
8492          */
8493         if ((features & NETIF_F_NTUPLE) || (features & NETIF_F_HW_TC)) {
8494                 /* turn off ATR, enable perfect filters and reset */
8495                 if (!(adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE))
8496                         need_reset = true;
8497
8498                 adapter->flags &= ~IXGBE_FLAG_FDIR_HASH_CAPABLE;
8499                 adapter->flags |= IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8500         } else {
8501                 /* turn off perfect filters, enable ATR and reset */
8502                 if (adapter->flags & IXGBE_FLAG_FDIR_PERFECT_CAPABLE)
8503                         need_reset = true;
8504
8505                 adapter->flags &= ~IXGBE_FLAG_FDIR_PERFECT_CAPABLE;
8506
8507                 /* We cannot enable ATR if SR-IOV is enabled */
8508                 if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED ||
8509                     /* We cannot enable ATR if we have 2 or more tcs */
8510                     (netdev_get_num_tc(netdev) > 1) ||
8511                     /* We cannot enable ATR if RSS is disabled */
8512                     (adapter->ring_feature[RING_F_RSS].limit <= 1) ||
8513                     /* A sample rate of 0 indicates ATR disabled */
8514                     (!adapter->atr_sample_rate))
8515                         ; /* do nothing not supported */
8516                 else /* otherwise supported and set the flag */
8517                         adapter->flags |= IXGBE_FLAG_FDIR_HASH_CAPABLE;
8518         }
8519
8520         if (features & NETIF_F_HW_VLAN_CTAG_RX)
8521                 ixgbe_vlan_strip_enable(adapter);
8522         else
8523                 ixgbe_vlan_strip_disable(adapter);
8524
8525         if (changed & NETIF_F_RXALL)
8526                 need_reset = true;
8527
8528         netdev->features = features;
8529
8530 #ifdef CONFIG_IXGBE_VXLAN
8531         if ((adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE)) {
8532                 if (features & NETIF_F_RXCSUM)
8533                         adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8534                 else
8535                         ixgbe_clear_vxlan_port(adapter);
8536         }
8537 #endif /* CONFIG_IXGBE_VXLAN */
8538
8539         if (need_reset)
8540                 ixgbe_do_reset(netdev);
8541
8542         return 0;
8543 }
8544
8545 #ifdef CONFIG_IXGBE_VXLAN
8546 /**
8547  * ixgbe_add_vxlan_port - Get notifications about VXLAN ports that come up
8548  * @dev: The port's netdev
8549  * @sa_family: Socket Family that VXLAN is notifiying us about
8550  * @port: New UDP port number that VXLAN started listening to
8551  **/
8552 static void ixgbe_add_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8553                                  __be16 port)
8554 {
8555         struct ixgbe_adapter *adapter = netdev_priv(dev);
8556         struct ixgbe_hw *hw = &adapter->hw;
8557         u16 new_port = ntohs(port);
8558
8559         if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8560                 return;
8561
8562         if (sa_family == AF_INET6)
8563                 return;
8564
8565         if (adapter->vxlan_port == new_port)
8566                 return;
8567
8568         if (adapter->vxlan_port) {
8569                 netdev_info(dev,
8570                             "Hit Max num of VXLAN ports, not adding port %d\n",
8571                             new_port);
8572                 return;
8573         }
8574
8575         adapter->vxlan_port = new_port;
8576         IXGBE_WRITE_REG(hw, IXGBE_VXLANCTRL, new_port);
8577 }
8578
8579 /**
8580  * ixgbe_del_vxlan_port - Get notifications about VXLAN ports that go away
8581  * @dev: The port's netdev
8582  * @sa_family: Socket Family that VXLAN is notifying us about
8583  * @port: UDP port number that VXLAN stopped listening to
8584  **/
8585 static void ixgbe_del_vxlan_port(struct net_device *dev, sa_family_t sa_family,
8586                                  __be16 port)
8587 {
8588         struct ixgbe_adapter *adapter = netdev_priv(dev);
8589         u16 new_port = ntohs(port);
8590
8591         if (!(adapter->flags & IXGBE_FLAG_VXLAN_OFFLOAD_CAPABLE))
8592                 return;
8593
8594         if (sa_family == AF_INET6)
8595                 return;
8596
8597         if (adapter->vxlan_port != new_port) {
8598                 netdev_info(dev, "Port %d was not found, not deleting\n",
8599                             new_port);
8600                 return;
8601         }
8602
8603         ixgbe_clear_vxlan_port(adapter);
8604         adapter->flags2 |= IXGBE_FLAG2_VXLAN_REREG_NEEDED;
8605 }
8606 #endif /* CONFIG_IXGBE_VXLAN */
8607
8608 static int ixgbe_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
8609                              struct net_device *dev,
8610                              const unsigned char *addr, u16 vid,
8611                              u16 flags)
8612 {
8613         /* guarantee we can provide a unique filter for the unicast address */
8614         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
8615                 struct ixgbe_adapter *adapter = netdev_priv(dev);
8616                 u16 pool = VMDQ_P(0);
8617
8618                 if (netdev_uc_count(dev) >= ixgbe_available_rars(adapter, pool))
8619                         return -ENOMEM;
8620         }
8621
8622         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
8623 }
8624
8625 /**
8626  * ixgbe_configure_bridge_mode - set various bridge modes
8627  * @adapter - the private structure
8628  * @mode - requested bridge mode
8629  *
8630  * Configure some settings require for various bridge modes.
8631  **/
8632 static int ixgbe_configure_bridge_mode(struct ixgbe_adapter *adapter,
8633                                        __u16 mode)
8634 {
8635         struct ixgbe_hw *hw = &adapter->hw;
8636         unsigned int p, num_pools;
8637         u32 vmdctl;
8638
8639         switch (mode) {
8640         case BRIDGE_MODE_VEPA:
8641                 /* disable Tx loopback, rely on switch hairpin mode */
8642                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC, 0);
8643
8644                 /* must enable Rx switching replication to allow multicast
8645                  * packet reception on all VFs, and to enable source address
8646                  * pruning.
8647                  */
8648                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8649                 vmdctl |= IXGBE_VT_CTL_REPLEN;
8650                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8651
8652                 /* enable Rx source address pruning. Note, this requires
8653                  * replication to be enabled or else it does nothing.
8654                  */
8655                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8656                 for (p = 0; p < num_pools; p++) {
8657                         if (hw->mac.ops.set_source_address_pruning)
8658                                 hw->mac.ops.set_source_address_pruning(hw,
8659                                                                        true,
8660                                                                        p);
8661                 }
8662                 break;
8663         case BRIDGE_MODE_VEB:
8664                 /* enable Tx loopback for internal VF/PF communication */
8665                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_PFDTXGSWC,
8666                                 IXGBE_PFDTXGSWC_VT_LBEN);
8667
8668                 /* disable Rx switching replication unless we have SR-IOV
8669                  * virtual functions
8670                  */
8671                 vmdctl = IXGBE_READ_REG(hw, IXGBE_VMD_CTL);
8672                 if (!adapter->num_vfs)
8673                         vmdctl &= ~IXGBE_VT_CTL_REPLEN;
8674                 IXGBE_WRITE_REG(hw, IXGBE_VMD_CTL, vmdctl);
8675
8676                 /* disable Rx source address pruning, since we don't expect to
8677                  * be receiving external loopback of our transmitted frames.
8678                  */
8679                 num_pools = adapter->num_vfs + adapter->num_rx_pools;
8680                 for (p = 0; p < num_pools; p++) {
8681                         if (hw->mac.ops.set_source_address_pruning)
8682                                 hw->mac.ops.set_source_address_pruning(hw,
8683                                                                        false,
8684                                                                        p);
8685                 }
8686                 break;
8687         default:
8688                 return -EINVAL;
8689         }
8690
8691         adapter->bridge_mode = mode;
8692
8693         e_info(drv, "enabling bridge mode: %s\n",
8694                mode == BRIDGE_MODE_VEPA ? "VEPA" : "VEB");
8695
8696         return 0;
8697 }
8698
8699 static int ixgbe_ndo_bridge_setlink(struct net_device *dev,
8700                                     struct nlmsghdr *nlh, u16 flags)
8701 {
8702         struct ixgbe_adapter *adapter = netdev_priv(dev);
8703         struct nlattr *attr, *br_spec;
8704         int rem;
8705
8706         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8707                 return -EOPNOTSUPP;
8708
8709         br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
8710         if (!br_spec)
8711                 return -EINVAL;
8712
8713         nla_for_each_nested(attr, br_spec, rem) {
8714                 int status;
8715                 __u16 mode;
8716
8717                 if (nla_type(attr) != IFLA_BRIDGE_MODE)
8718                         continue;
8719
8720                 if (nla_len(attr) < sizeof(mode))
8721                         return -EINVAL;
8722
8723                 mode = nla_get_u16(attr);
8724                 status = ixgbe_configure_bridge_mode(adapter, mode);
8725                 if (status)
8726                         return status;
8727
8728                 break;
8729         }
8730
8731         return 0;
8732 }
8733
8734 static int ixgbe_ndo_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
8735                                     struct net_device *dev,
8736                                     u32 filter_mask, int nlflags)
8737 {
8738         struct ixgbe_adapter *adapter = netdev_priv(dev);
8739
8740         if (!(adapter->flags & IXGBE_FLAG_SRIOV_ENABLED))
8741                 return 0;
8742
8743         return ndo_dflt_bridge_getlink(skb, pid, seq, dev,
8744                                        adapter->bridge_mode, 0, 0, nlflags,
8745                                        filter_mask, NULL);
8746 }
8747
8748 static void *ixgbe_fwd_add(struct net_device *pdev, struct net_device *vdev)
8749 {
8750         struct ixgbe_fwd_adapter *fwd_adapter = NULL;
8751         struct ixgbe_adapter *adapter = netdev_priv(pdev);
8752         int used_pools = adapter->num_vfs + adapter->num_rx_pools;
8753         unsigned int limit;
8754         int pool, err;
8755
8756         /* Hardware has a limited number of available pools. Each VF, and the
8757          * PF require a pool. Check to ensure we don't attempt to use more
8758          * then the available number of pools.
8759          */
8760         if (used_pools >= IXGBE_MAX_VF_FUNCTIONS)
8761                 return ERR_PTR(-EINVAL);
8762
8763 #ifdef CONFIG_RPS
8764         if (vdev->num_rx_queues != vdev->num_tx_queues) {
8765                 netdev_info(pdev, "%s: Only supports a single queue count for TX and RX\n",
8766                             vdev->name);
8767                 return ERR_PTR(-EINVAL);
8768         }
8769 #endif
8770         /* Check for hardware restriction on number of rx/tx queues */
8771         if (vdev->num_tx_queues > IXGBE_MAX_L2A_QUEUES ||
8772             vdev->num_tx_queues == IXGBE_BAD_L2A_QUEUE) {
8773                 netdev_info(pdev,
8774                             "%s: Supports RX/TX Queue counts 1,2, and 4\n",
8775                             pdev->name);
8776                 return ERR_PTR(-EINVAL);
8777         }
8778
8779         if (((adapter->flags & IXGBE_FLAG_DCB_ENABLED) &&
8780               adapter->num_rx_pools > IXGBE_MAX_DCBMACVLANS - 1) ||
8781             (adapter->num_rx_pools > IXGBE_MAX_MACVLANS))
8782                 return ERR_PTR(-EBUSY);
8783
8784         fwd_adapter = kzalloc(sizeof(*fwd_adapter), GFP_KERNEL);
8785         if (!fwd_adapter)
8786                 return ERR_PTR(-ENOMEM);
8787
8788         pool = find_first_zero_bit(&adapter->fwd_bitmask, 32);
8789         adapter->num_rx_pools++;
8790         set_bit(pool, &adapter->fwd_bitmask);
8791         limit = find_last_bit(&adapter->fwd_bitmask, 32);
8792
8793         /* Enable VMDq flag so device will be set in VM mode */
8794         adapter->flags |= IXGBE_FLAG_VMDQ_ENABLED | IXGBE_FLAG_SRIOV_ENABLED;
8795         adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8796         adapter->ring_feature[RING_F_RSS].limit = vdev->num_tx_queues;
8797
8798         /* Force reinit of ring allocation with VMDQ enabled */
8799         err = ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8800         if (err)
8801                 goto fwd_add_err;
8802         fwd_adapter->pool = pool;
8803         fwd_adapter->real_adapter = adapter;
8804         err = ixgbe_fwd_ring_up(vdev, fwd_adapter);
8805         if (err)
8806                 goto fwd_add_err;
8807         netif_tx_start_all_queues(vdev);
8808         return fwd_adapter;
8809 fwd_add_err:
8810         /* unwind counter and free adapter struct */
8811         netdev_info(pdev,
8812                     "%s: dfwd hardware acceleration failed\n", vdev->name);
8813         clear_bit(pool, &adapter->fwd_bitmask);
8814         adapter->num_rx_pools--;
8815         kfree(fwd_adapter);
8816         return ERR_PTR(err);
8817 }
8818
8819 static void ixgbe_fwd_del(struct net_device *pdev, void *priv)
8820 {
8821         struct ixgbe_fwd_adapter *fwd_adapter = priv;
8822         struct ixgbe_adapter *adapter = fwd_adapter->real_adapter;
8823         unsigned int limit;
8824
8825         clear_bit(fwd_adapter->pool, &adapter->fwd_bitmask);
8826         adapter->num_rx_pools--;
8827
8828         limit = find_last_bit(&adapter->fwd_bitmask, 32);
8829         adapter->ring_feature[RING_F_VMDQ].limit = limit + 1;
8830         ixgbe_fwd_ring_down(fwd_adapter->netdev, fwd_adapter);
8831         ixgbe_setup_tc(pdev, netdev_get_num_tc(pdev));
8832         netdev_dbg(pdev, "pool %i:%i queues %i:%i VSI bitmask %lx\n",
8833                    fwd_adapter->pool, adapter->num_rx_pools,
8834                    fwd_adapter->rx_base_queue,
8835                    fwd_adapter->rx_base_queue + adapter->num_rx_queues_per_pool,
8836                    adapter->fwd_bitmask);
8837         kfree(fwd_adapter);
8838 }
8839
8840 #define IXGBE_MAX_TUNNEL_HDR_LEN 80
8841 static netdev_features_t
8842 ixgbe_features_check(struct sk_buff *skb, struct net_device *dev,
8843                      netdev_features_t features)
8844 {
8845         if (!skb->encapsulation)
8846                 return features;
8847
8848         if (unlikely(skb_inner_mac_header(skb) - skb_transport_header(skb) >
8849                      IXGBE_MAX_TUNNEL_HDR_LEN))
8850                 return features & ~NETIF_F_CSUM_MASK;
8851
8852         return features;
8853 }
8854
8855 static const struct net_device_ops ixgbe_netdev_ops = {
8856         .ndo_open               = ixgbe_open,
8857         .ndo_stop               = ixgbe_close,
8858         .ndo_start_xmit         = ixgbe_xmit_frame,
8859         .ndo_select_queue       = ixgbe_select_queue,
8860         .ndo_set_rx_mode        = ixgbe_set_rx_mode,
8861         .ndo_validate_addr      = eth_validate_addr,
8862         .ndo_set_mac_address    = ixgbe_set_mac,
8863         .ndo_change_mtu         = ixgbe_change_mtu,
8864         .ndo_tx_timeout         = ixgbe_tx_timeout,
8865         .ndo_vlan_rx_add_vid    = ixgbe_vlan_rx_add_vid,
8866         .ndo_vlan_rx_kill_vid   = ixgbe_vlan_rx_kill_vid,
8867         .ndo_do_ioctl           = ixgbe_ioctl,
8868         .ndo_set_vf_mac         = ixgbe_ndo_set_vf_mac,
8869         .ndo_set_vf_vlan        = ixgbe_ndo_set_vf_vlan,
8870         .ndo_set_vf_rate        = ixgbe_ndo_set_vf_bw,
8871         .ndo_set_vf_spoofchk    = ixgbe_ndo_set_vf_spoofchk,
8872         .ndo_set_vf_rss_query_en = ixgbe_ndo_set_vf_rss_query_en,
8873         .ndo_set_vf_trust       = ixgbe_ndo_set_vf_trust,
8874         .ndo_get_vf_config      = ixgbe_ndo_get_vf_config,
8875         .ndo_get_stats64        = ixgbe_get_stats64,
8876         .ndo_setup_tc           = __ixgbe_setup_tc,
8877 #ifdef CONFIG_NET_POLL_CONTROLLER
8878         .ndo_poll_controller    = ixgbe_netpoll,
8879 #endif
8880 #ifdef CONFIG_NET_RX_BUSY_POLL
8881         .ndo_busy_poll          = ixgbe_low_latency_recv,
8882 #endif
8883 #ifdef IXGBE_FCOE
8884         .ndo_fcoe_ddp_setup = ixgbe_fcoe_ddp_get,
8885         .ndo_fcoe_ddp_target = ixgbe_fcoe_ddp_target,
8886         .ndo_fcoe_ddp_done = ixgbe_fcoe_ddp_put,
8887         .ndo_fcoe_enable = ixgbe_fcoe_enable,
8888         .ndo_fcoe_disable = ixgbe_fcoe_disable,
8889         .ndo_fcoe_get_wwn = ixgbe_fcoe_get_wwn,
8890         .ndo_fcoe_get_hbainfo = ixgbe_fcoe_get_hbainfo,
8891 #endif /* IXGBE_FCOE */
8892         .ndo_set_features = ixgbe_set_features,
8893         .ndo_fix_features = ixgbe_fix_features,
8894         .ndo_fdb_add            = ixgbe_ndo_fdb_add,
8895         .ndo_bridge_setlink     = ixgbe_ndo_bridge_setlink,
8896         .ndo_bridge_getlink     = ixgbe_ndo_bridge_getlink,
8897         .ndo_dfwd_add_station   = ixgbe_fwd_add,
8898         .ndo_dfwd_del_station   = ixgbe_fwd_del,
8899 #ifdef CONFIG_IXGBE_VXLAN
8900         .ndo_add_vxlan_port     = ixgbe_add_vxlan_port,
8901         .ndo_del_vxlan_port     = ixgbe_del_vxlan_port,
8902 #endif /* CONFIG_IXGBE_VXLAN */
8903         .ndo_features_check     = ixgbe_features_check,
8904 };
8905
8906 /**
8907  * ixgbe_enumerate_functions - Get the number of ports this device has
8908  * @adapter: adapter structure
8909  *
8910  * This function enumerates the phsyical functions co-located on a single slot,
8911  * in order to determine how many ports a device has. This is most useful in
8912  * determining the required GT/s of PCIe bandwidth necessary for optimal
8913  * performance.
8914  **/
8915 static inline int ixgbe_enumerate_functions(struct ixgbe_adapter *adapter)
8916 {
8917         struct pci_dev *entry, *pdev = adapter->pdev;
8918         int physfns = 0;
8919
8920         /* Some cards can not use the generic count PCIe functions method,
8921          * because they are behind a parent switch, so we hardcode these with
8922          * the correct number of functions.
8923          */
8924         if (ixgbe_pcie_from_parent(&adapter->hw))
8925                 physfns = 4;
8926
8927         list_for_each_entry(entry, &adapter->pdev->bus->devices, bus_list) {
8928                 /* don't count virtual functions */
8929                 if (entry->is_virtfn)
8930                         continue;
8931
8932                 /* When the devices on the bus don't all match our device ID,
8933                  * we can't reliably determine the correct number of
8934                  * functions. This can occur if a function has been direct
8935                  * attached to a virtual machine using VT-d, for example. In
8936                  * this case, simply return -1 to indicate this.
8937                  */
8938                 if ((entry->vendor != pdev->vendor) ||
8939                     (entry->device != pdev->device))
8940                         return -1;
8941
8942                 physfns++;
8943         }
8944
8945         return physfns;
8946 }
8947
8948 /**
8949  * ixgbe_wol_supported - Check whether device supports WoL
8950  * @hw: hw specific details
8951  * @device_id: the device ID
8952  * @subdev_id: the subsystem device ID
8953  *
8954  * This function is used by probe and ethtool to determine
8955  * which devices have WoL support
8956  *
8957  **/
8958 int ixgbe_wol_supported(struct ixgbe_adapter *adapter, u16 device_id,
8959                         u16 subdevice_id)
8960 {
8961         struct ixgbe_hw *hw = &adapter->hw;
8962         u16 wol_cap = adapter->eeprom_cap & IXGBE_DEVICE_CAPS_WOL_MASK;
8963         int is_wol_supported = 0;
8964
8965         switch (device_id) {
8966         case IXGBE_DEV_ID_82599_SFP:
8967                 /* Only these subdevices could supports WOL */
8968                 switch (subdevice_id) {
8969                 case IXGBE_SUBDEV_ID_82599_SFP_WOL0:
8970                 case IXGBE_SUBDEV_ID_82599_560FLR:
8971                         /* only support first port */
8972                         if (hw->bus.func != 0)
8973                                 break;
8974                 case IXGBE_SUBDEV_ID_82599_SP_560FLR:
8975                 case IXGBE_SUBDEV_ID_82599_SFP:
8976                 case IXGBE_SUBDEV_ID_82599_RNDC:
8977                 case IXGBE_SUBDEV_ID_82599_ECNA_DP:
8978                 case IXGBE_SUBDEV_ID_82599_LOM_SFP:
8979                         is_wol_supported = 1;
8980                         break;
8981                 }
8982                 break;
8983         case IXGBE_DEV_ID_82599EN_SFP:
8984                 /* Only this subdevice supports WOL */
8985                 switch (subdevice_id) {
8986                 case IXGBE_SUBDEV_ID_82599EN_SFP_OCP1:
8987                         is_wol_supported = 1;
8988                         break;
8989                 }
8990                 break;
8991         case IXGBE_DEV_ID_82599_COMBO_BACKPLANE:
8992                 /* All except this subdevice support WOL */
8993                 if (subdevice_id != IXGBE_SUBDEV_ID_82599_KX4_KR_MEZZ)
8994                         is_wol_supported = 1;
8995                 break;
8996         case IXGBE_DEV_ID_82599_KX4:
8997                 is_wol_supported = 1;
8998                 break;
8999         case IXGBE_DEV_ID_X540T:
9000         case IXGBE_DEV_ID_X540T1:
9001         case IXGBE_DEV_ID_X550T:
9002         case IXGBE_DEV_ID_X550EM_X_KX4:
9003         case IXGBE_DEV_ID_X550EM_X_KR:
9004         case IXGBE_DEV_ID_X550EM_X_10G_T:
9005                 /* check eeprom to see if enabled wol */
9006                 if ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0_1) ||
9007                     ((wol_cap == IXGBE_DEVICE_CAPS_WOL_PORT0) &&
9008                      (hw->bus.func == 0))) {
9009                         is_wol_supported = 1;
9010                 }
9011                 break;
9012         }
9013
9014         return is_wol_supported;
9015 }
9016
9017 /**
9018  * ixgbe_get_platform_mac_addr - Look up MAC address in Open Firmware / IDPROM
9019  * @adapter: Pointer to adapter struct
9020  */
9021 static void ixgbe_get_platform_mac_addr(struct ixgbe_adapter *adapter)
9022 {
9023 #ifdef CONFIG_OF
9024         struct device_node *dp = pci_device_to_OF_node(adapter->pdev);
9025         struct ixgbe_hw *hw = &adapter->hw;
9026         const unsigned char *addr;
9027
9028         addr = of_get_mac_address(dp);
9029         if (addr) {
9030                 ether_addr_copy(hw->mac.perm_addr, addr);
9031                 return;
9032         }
9033 #endif /* CONFIG_OF */
9034
9035 #ifdef CONFIG_SPARC
9036         ether_addr_copy(hw->mac.perm_addr, idprom->id_ethaddr);
9037 #endif /* CONFIG_SPARC */
9038 }
9039
9040 /**
9041  * ixgbe_probe - Device Initialization Routine
9042  * @pdev: PCI device information struct
9043  * @ent: entry in ixgbe_pci_tbl
9044  *
9045  * Returns 0 on success, negative on failure
9046  *
9047  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
9048  * The OS initialization, configuring of the adapter private structure,
9049  * and a hardware reset occur.
9050  **/
9051 static int ixgbe_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
9052 {
9053         struct net_device *netdev;
9054         struct ixgbe_adapter *adapter = NULL;
9055         struct ixgbe_hw *hw;
9056         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
9057         int i, err, pci_using_dac, expected_gts;
9058         unsigned int indices = MAX_TX_QUEUES;
9059         u8 part_str[IXGBE_PBANUM_LENGTH];
9060         bool disable_dev = false;
9061 #ifdef IXGBE_FCOE
9062         u16 device_caps;
9063 #endif
9064         u32 eec;
9065
9066         /* Catch broken hardware that put the wrong VF device ID in
9067          * the PCIe SR-IOV capability.
9068          */
9069         if (pdev->is_virtfn) {
9070                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
9071                      pci_name(pdev), pdev->vendor, pdev->device);
9072                 return -EINVAL;
9073         }
9074
9075         err = pci_enable_device_mem(pdev);
9076         if (err)
9077                 return err;
9078
9079         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
9080                 pci_using_dac = 1;
9081         } else {
9082                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
9083                 if (err) {
9084                         dev_err(&pdev->dev,
9085                                 "No usable DMA configuration, aborting\n");
9086                         goto err_dma;
9087                 }
9088                 pci_using_dac = 0;
9089         }
9090
9091         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
9092                                            IORESOURCE_MEM), ixgbe_driver_name);
9093         if (err) {
9094                 dev_err(&pdev->dev,
9095                         "pci_request_selected_regions failed 0x%x\n", err);
9096                 goto err_pci_reg;
9097         }
9098
9099         pci_enable_pcie_error_reporting(pdev);
9100
9101         pci_set_master(pdev);
9102         pci_save_state(pdev);
9103
9104         if (ii->mac == ixgbe_mac_82598EB) {
9105 #ifdef CONFIG_IXGBE_DCB
9106                 /* 8 TC w/ 4 queues per TC */
9107                 indices = 4 * MAX_TRAFFIC_CLASS;
9108 #else
9109                 indices = IXGBE_MAX_RSS_INDICES;
9110 #endif
9111         }
9112
9113         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), indices);
9114         if (!netdev) {
9115                 err = -ENOMEM;
9116                 goto err_alloc_etherdev;
9117         }
9118
9119         SET_NETDEV_DEV(netdev, &pdev->dev);
9120
9121         adapter = netdev_priv(netdev);
9122
9123         adapter->netdev = netdev;
9124         adapter->pdev = pdev;
9125         hw = &adapter->hw;
9126         hw->back = adapter;
9127         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
9128
9129         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
9130                               pci_resource_len(pdev, 0));
9131         adapter->io_addr = hw->hw_addr;
9132         if (!hw->hw_addr) {
9133                 err = -EIO;
9134                 goto err_ioremap;
9135         }
9136
9137         netdev->netdev_ops = &ixgbe_netdev_ops;
9138         ixgbe_set_ethtool_ops(netdev);
9139         netdev->watchdog_timeo = 5 * HZ;
9140         strlcpy(netdev->name, pci_name(pdev), sizeof(netdev->name));
9141
9142         /* Setup hw api */
9143         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
9144         hw->mac.type  = ii->mac;
9145         hw->mvals     = ii->mvals;
9146
9147         /* EEPROM */
9148         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
9149         eec = IXGBE_READ_REG(hw, IXGBE_EEC(hw));
9150         if (ixgbe_removed(hw->hw_addr)) {
9151                 err = -EIO;
9152                 goto err_ioremap;
9153         }
9154         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
9155         if (!(eec & (1 << 8)))
9156                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
9157
9158         /* PHY */
9159         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
9160         hw->phy.sfp_type = ixgbe_sfp_type_unknown;
9161         /* ixgbe_identify_phy_generic will set prtad and mmds properly */
9162         hw->phy.mdio.prtad = MDIO_PRTAD_NONE;
9163         hw->phy.mdio.mmds = 0;
9164         hw->phy.mdio.mode_support = MDIO_SUPPORTS_C45 | MDIO_EMULATE_C22;
9165         hw->phy.mdio.dev = netdev;
9166         hw->phy.mdio.mdio_read = ixgbe_mdio_read;
9167         hw->phy.mdio.mdio_write = ixgbe_mdio_write;
9168
9169         ii->get_invariants(hw);
9170
9171         /* setup the private structure */
9172         err = ixgbe_sw_init(adapter);
9173         if (err)
9174                 goto err_sw_init;
9175
9176         /* Make it possible the adapter to be woken up via WOL */
9177         switch (adapter->hw.mac.type) {
9178         case ixgbe_mac_82599EB:
9179         case ixgbe_mac_X540:
9180         case ixgbe_mac_X550:
9181         case ixgbe_mac_X550EM_x:
9182                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9183                 break;
9184         default:
9185                 break;
9186         }
9187
9188         /*
9189          * If there is a fan on this device and it has failed log the
9190          * failure.
9191          */
9192         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
9193                 u32 esdp = IXGBE_READ_REG(hw, IXGBE_ESDP);
9194                 if (esdp & IXGBE_ESDP_SDP1)
9195                         e_crit(probe, "Fan has stopped, replace the adapter\n");
9196         }
9197
9198         if (allow_unsupported_sfp)
9199                 hw->allow_unsupported_sfp = allow_unsupported_sfp;
9200
9201         /* reset_hw fills in the perm_addr as well */
9202         hw->phy.reset_if_overtemp = true;
9203         err = hw->mac.ops.reset_hw(hw);
9204         hw->phy.reset_if_overtemp = false;
9205         if (err == IXGBE_ERR_SFP_NOT_PRESENT) {
9206                 err = 0;
9207         } else if (err == IXGBE_ERR_SFP_NOT_SUPPORTED) {
9208                 e_dev_err("failed to load because an unsupported SFP+ or QSFP module type was detected.\n");
9209                 e_dev_err("Reload the driver after installing a supported module.\n");
9210                 goto err_sw_init;
9211         } else if (err) {
9212                 e_dev_err("HW Init failed: %d\n", err);
9213                 goto err_sw_init;
9214         }
9215
9216 #ifdef CONFIG_PCI_IOV
9217         /* SR-IOV not supported on the 82598 */
9218         if (adapter->hw.mac.type == ixgbe_mac_82598EB)
9219                 goto skip_sriov;
9220         /* Mailbox */
9221         ixgbe_init_mbx_params_pf(hw);
9222         memcpy(&hw->mbx.ops, ii->mbx_ops, sizeof(hw->mbx.ops));
9223         pci_sriov_set_totalvfs(pdev, IXGBE_MAX_VFS_DRV_LIMIT);
9224         ixgbe_enable_sriov(adapter);
9225 skip_sriov:
9226
9227 #endif
9228         netdev->features = NETIF_F_SG |
9229                            NETIF_F_IP_CSUM |
9230                            NETIF_F_IPV6_CSUM |
9231                            NETIF_F_HW_VLAN_CTAG_TX |
9232                            NETIF_F_HW_VLAN_CTAG_RX |
9233                            NETIF_F_TSO |
9234                            NETIF_F_TSO6 |
9235                            NETIF_F_RXHASH |
9236                            NETIF_F_RXCSUM;
9237
9238         netdev->hw_features = netdev->features | NETIF_F_HW_L2FW_DOFFLOAD;
9239
9240         switch (adapter->hw.mac.type) {
9241         case ixgbe_mac_82599EB:
9242         case ixgbe_mac_X540:
9243         case ixgbe_mac_X550:
9244         case ixgbe_mac_X550EM_x:
9245                 netdev->features |= NETIF_F_SCTP_CRC;
9246                 netdev->hw_features |= NETIF_F_SCTP_CRC |
9247                                        NETIF_F_NTUPLE |
9248                                        NETIF_F_HW_TC;
9249                 break;
9250         default:
9251                 break;
9252         }
9253
9254         netdev->hw_features |= NETIF_F_RXALL;
9255         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
9256
9257         netdev->vlan_features |= NETIF_F_TSO;
9258         netdev->vlan_features |= NETIF_F_TSO6;
9259         netdev->vlan_features |= NETIF_F_IP_CSUM;
9260         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
9261         netdev->vlan_features |= NETIF_F_SG;
9262
9263         netdev->hw_enc_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
9264
9265         netdev->priv_flags |= IFF_UNICAST_FLT;
9266         netdev->priv_flags |= IFF_SUPP_NOFCS;
9267
9268 #ifdef CONFIG_IXGBE_VXLAN
9269         switch (adapter->hw.mac.type) {
9270         case ixgbe_mac_X550:
9271         case ixgbe_mac_X550EM_x:
9272                 netdev->hw_enc_features |= NETIF_F_RXCSUM;
9273                 break;
9274         default:
9275                 break;
9276         }
9277 #endif /* CONFIG_IXGBE_VXLAN */
9278
9279 #ifdef CONFIG_IXGBE_DCB
9280         netdev->dcbnl_ops = &dcbnl_ops;
9281 #endif
9282
9283 #ifdef IXGBE_FCOE
9284         if (adapter->flags & IXGBE_FLAG_FCOE_CAPABLE) {
9285                 unsigned int fcoe_l;
9286
9287                 if (hw->mac.ops.get_device_caps) {
9288                         hw->mac.ops.get_device_caps(hw, &device_caps);
9289                         if (device_caps & IXGBE_DEVICE_CAPS_FCOE_OFFLOADS)
9290                                 adapter->flags &= ~IXGBE_FLAG_FCOE_CAPABLE;
9291                 }
9292
9293
9294                 fcoe_l = min_t(int, IXGBE_FCRETA_SIZE, num_online_cpus());
9295                 adapter->ring_feature[RING_F_FCOE].limit = fcoe_l;
9296
9297                 netdev->features |= NETIF_F_FSO |
9298                                     NETIF_F_FCOE_CRC;
9299
9300                 netdev->vlan_features |= NETIF_F_FSO |
9301                                          NETIF_F_FCOE_CRC |
9302                                          NETIF_F_FCOE_MTU;
9303         }
9304 #endif /* IXGBE_FCOE */
9305         if (pci_using_dac) {
9306                 netdev->features |= NETIF_F_HIGHDMA;
9307                 netdev->vlan_features |= NETIF_F_HIGHDMA;
9308         }
9309
9310         if (adapter->flags2 & IXGBE_FLAG2_RSC_CAPABLE)
9311                 netdev->hw_features |= NETIF_F_LRO;
9312         if (adapter->flags2 & IXGBE_FLAG2_RSC_ENABLED)
9313                 netdev->features |= NETIF_F_LRO;
9314
9315         /* make sure the EEPROM is good */
9316         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
9317                 e_dev_err("The EEPROM Checksum Is Not Valid\n");
9318                 err = -EIO;
9319                 goto err_sw_init;
9320         }
9321
9322         ixgbe_get_platform_mac_addr(adapter);
9323
9324         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
9325
9326         if (!is_valid_ether_addr(netdev->dev_addr)) {
9327                 e_dev_err("invalid MAC address\n");
9328                 err = -EIO;
9329                 goto err_sw_init;
9330         }
9331
9332         ixgbe_mac_set_default_filter(adapter);
9333
9334         setup_timer(&adapter->service_timer, &ixgbe_service_timer,
9335                     (unsigned long) adapter);
9336
9337         if (ixgbe_removed(hw->hw_addr)) {
9338                 err = -EIO;
9339                 goto err_sw_init;
9340         }
9341         INIT_WORK(&adapter->service_task, ixgbe_service_task);
9342         set_bit(__IXGBE_SERVICE_INITED, &adapter->state);
9343         clear_bit(__IXGBE_SERVICE_SCHED, &adapter->state);
9344
9345         err = ixgbe_init_interrupt_scheme(adapter);
9346         if (err)
9347                 goto err_sw_init;
9348
9349         /* WOL not supported for all devices */
9350         adapter->wol = 0;
9351         hw->eeprom.ops.read(hw, 0x2c, &adapter->eeprom_cap);
9352         hw->wol_enabled = ixgbe_wol_supported(adapter, pdev->device,
9353                                                 pdev->subsystem_device);
9354         if (hw->wol_enabled)
9355                 adapter->wol = IXGBE_WUFC_MAG;
9356
9357         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
9358
9359         /* save off EEPROM version number */
9360         hw->eeprom.ops.read(hw, 0x2e, &adapter->eeprom_verh);
9361         hw->eeprom.ops.read(hw, 0x2d, &adapter->eeprom_verl);
9362
9363         /* pick up the PCI bus settings for reporting later */
9364         if (ixgbe_pcie_from_parent(hw))
9365                 ixgbe_get_parent_bus_info(adapter);
9366         else
9367                  hw->mac.ops.get_bus_info(hw);
9368
9369         /* calculate the expected PCIe bandwidth required for optimal
9370          * performance. Note that some older parts will never have enough
9371          * bandwidth due to being older generation PCIe parts. We clamp these
9372          * parts to ensure no warning is displayed if it can't be fixed.
9373          */
9374         switch (hw->mac.type) {
9375         case ixgbe_mac_82598EB:
9376                 expected_gts = min(ixgbe_enumerate_functions(adapter) * 10, 16);
9377                 break;
9378         default:
9379                 expected_gts = ixgbe_enumerate_functions(adapter) * 10;
9380                 break;
9381         }
9382
9383         /* don't check link if we failed to enumerate functions */
9384         if (expected_gts > 0)
9385                 ixgbe_check_minimum_link(adapter, expected_gts);
9386
9387         err = ixgbe_read_pba_string_generic(hw, part_str, sizeof(part_str));
9388         if (err)
9389                 strlcpy(part_str, "Unknown", sizeof(part_str));
9390         if (ixgbe_is_sfp(hw) && hw->phy.sfp_type != ixgbe_sfp_type_not_present)
9391                 e_dev_info("MAC: %d, PHY: %d, SFP+: %d, PBA No: %s\n",
9392                            hw->mac.type, hw->phy.type, hw->phy.sfp_type,
9393                            part_str);
9394         else
9395                 e_dev_info("MAC: %d, PHY: %d, PBA No: %s\n",
9396                            hw->mac.type, hw->phy.type, part_str);
9397
9398         e_dev_info("%pM\n", netdev->dev_addr);
9399
9400         /* reset the hardware with the new settings */
9401         err = hw->mac.ops.start_hw(hw);
9402         if (err == IXGBE_ERR_EEPROM_VERSION) {
9403                 /* We are running on a pre-production device, log a warning */
9404                 e_dev_warn("This device is a pre-production adapter/LOM. "
9405                            "Please be aware there may be issues associated "
9406                            "with your hardware.  If you are experiencing "
9407                            "problems please contact your Intel or hardware "
9408                            "representative who provided you with this "
9409                            "hardware.\n");
9410         }
9411         strcpy(netdev->name, "eth%d");
9412         err = register_netdev(netdev);
9413         if (err)
9414                 goto err_register;
9415
9416         pci_set_drvdata(pdev, adapter);
9417
9418         /* power down the optics for 82599 SFP+ fiber */
9419         if (hw->mac.ops.disable_tx_laser)
9420                 hw->mac.ops.disable_tx_laser(hw);
9421
9422         /* carrier off reporting is important to ethtool even BEFORE open */
9423         netif_carrier_off(netdev);
9424
9425 #ifdef CONFIG_IXGBE_DCA
9426         if (dca_add_requester(&pdev->dev) == 0) {
9427                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
9428                 ixgbe_setup_dca(adapter);
9429         }
9430 #endif
9431         if (adapter->flags & IXGBE_FLAG_SRIOV_ENABLED) {
9432                 e_info(probe, "IOV is enabled with %d VFs\n", adapter->num_vfs);
9433                 for (i = 0; i < adapter->num_vfs; i++)
9434                         ixgbe_vf_configuration(pdev, (i | 0x10000000));
9435         }
9436
9437         /* firmware requires driver version to be 0xFFFFFFFF
9438          * since os does not support feature
9439          */
9440         if (hw->mac.ops.set_fw_drv_ver)
9441                 hw->mac.ops.set_fw_drv_ver(hw, 0xFF, 0xFF, 0xFF,
9442                                            0xFF);
9443
9444         /* add san mac addr to netdev */
9445         ixgbe_add_sanmac_netdev(netdev);
9446
9447         e_dev_info("%s\n", ixgbe_default_device_descr);
9448
9449 #ifdef CONFIG_IXGBE_HWMON
9450         if (ixgbe_sysfs_init(adapter))
9451                 e_err(probe, "failed to allocate sysfs resources\n");
9452 #endif /* CONFIG_IXGBE_HWMON */
9453
9454         ixgbe_dbg_adapter_init(adapter);
9455
9456         /* setup link for SFP devices with MNG FW, else wait for IXGBE_UP */
9457         if (ixgbe_mng_enabled(hw) && ixgbe_is_sfp(hw) && hw->mac.ops.setup_link)
9458                 hw->mac.ops.setup_link(hw,
9459                         IXGBE_LINK_SPEED_10GB_FULL | IXGBE_LINK_SPEED_1GB_FULL,
9460                         true);
9461
9462         return 0;
9463
9464 err_register:
9465         ixgbe_release_hw_control(adapter);
9466         ixgbe_clear_interrupt_scheme(adapter);
9467 err_sw_init:
9468         ixgbe_disable_sriov(adapter);
9469         adapter->flags2 &= ~IXGBE_FLAG2_SEARCH_FOR_SFP;
9470         iounmap(adapter->io_addr);
9471         kfree(adapter->mac_table);
9472 err_ioremap:
9473         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9474         free_netdev(netdev);
9475 err_alloc_etherdev:
9476         pci_release_selected_regions(pdev,
9477                                      pci_select_bars(pdev, IORESOURCE_MEM));
9478 err_pci_reg:
9479 err_dma:
9480         if (!adapter || disable_dev)
9481                 pci_disable_device(pdev);
9482         return err;
9483 }
9484
9485 /**
9486  * ixgbe_remove - Device Removal Routine
9487  * @pdev: PCI device information struct
9488  *
9489  * ixgbe_remove is called by the PCI subsystem to alert the driver
9490  * that it should release a PCI device.  The could be caused by a
9491  * Hot-Plug event, or because the driver is going to be removed from
9492  * memory.
9493  **/
9494 static void ixgbe_remove(struct pci_dev *pdev)
9495 {
9496         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9497         struct net_device *netdev;
9498         bool disable_dev;
9499
9500         /* if !adapter then we already cleaned up in probe */
9501         if (!adapter)
9502                 return;
9503
9504         netdev  = adapter->netdev;
9505         ixgbe_dbg_adapter_exit(adapter);
9506
9507         set_bit(__IXGBE_REMOVING, &adapter->state);
9508         cancel_work_sync(&adapter->service_task);
9509
9510
9511 #ifdef CONFIG_IXGBE_DCA
9512         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
9513                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
9514                 dca_remove_requester(&pdev->dev);
9515                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL,
9516                                 IXGBE_DCA_CTRL_DCA_DISABLE);
9517         }
9518
9519 #endif
9520 #ifdef CONFIG_IXGBE_HWMON
9521         ixgbe_sysfs_exit(adapter);
9522 #endif /* CONFIG_IXGBE_HWMON */
9523
9524         /* remove the added san mac */
9525         ixgbe_del_sanmac_netdev(netdev);
9526
9527 #ifdef CONFIG_PCI_IOV
9528         ixgbe_disable_sriov(adapter);
9529 #endif
9530         if (netdev->reg_state == NETREG_REGISTERED)
9531                 unregister_netdev(netdev);
9532
9533         ixgbe_clear_interrupt_scheme(adapter);
9534
9535         ixgbe_release_hw_control(adapter);
9536
9537 #ifdef CONFIG_DCB
9538         kfree(adapter->ixgbe_ieee_pfc);
9539         kfree(adapter->ixgbe_ieee_ets);
9540
9541 #endif
9542         iounmap(adapter->io_addr);
9543         pci_release_selected_regions(pdev, pci_select_bars(pdev,
9544                                      IORESOURCE_MEM));
9545
9546         e_dev_info("complete\n");
9547
9548         kfree(adapter->mac_table);
9549         disable_dev = !test_and_set_bit(__IXGBE_DISABLED, &adapter->state);
9550         free_netdev(netdev);
9551
9552         pci_disable_pcie_error_reporting(pdev);
9553
9554         if (disable_dev)
9555                 pci_disable_device(pdev);
9556 }
9557
9558 /**
9559  * ixgbe_io_error_detected - called when PCI error is detected
9560  * @pdev: Pointer to PCI device
9561  * @state: The current pci connection state
9562  *
9563  * This function is called after a PCI bus error affecting
9564  * this device has been detected.
9565  */
9566 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
9567                                                 pci_channel_state_t state)
9568 {
9569         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9570         struct net_device *netdev = adapter->netdev;
9571
9572 #ifdef CONFIG_PCI_IOV
9573         struct ixgbe_hw *hw = &adapter->hw;
9574         struct pci_dev *bdev, *vfdev;
9575         u32 dw0, dw1, dw2, dw3;
9576         int vf, pos;
9577         u16 req_id, pf_func;
9578
9579         if (adapter->hw.mac.type == ixgbe_mac_82598EB ||
9580             adapter->num_vfs == 0)
9581                 goto skip_bad_vf_detection;
9582
9583         bdev = pdev->bus->self;
9584         while (bdev && (pci_pcie_type(bdev) != PCI_EXP_TYPE_ROOT_PORT))
9585                 bdev = bdev->bus->self;
9586
9587         if (!bdev)
9588                 goto skip_bad_vf_detection;
9589
9590         pos = pci_find_ext_capability(bdev, PCI_EXT_CAP_ID_ERR);
9591         if (!pos)
9592                 goto skip_bad_vf_detection;
9593
9594         dw0 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG);
9595         dw1 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 4);
9596         dw2 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 8);
9597         dw3 = ixgbe_read_pci_cfg_dword(hw, pos + PCI_ERR_HEADER_LOG + 12);
9598         if (ixgbe_removed(hw->hw_addr))
9599                 goto skip_bad_vf_detection;
9600
9601         req_id = dw1 >> 16;
9602         /* On the 82599 if bit 7 of the requestor ID is set then it's a VF */
9603         if (!(req_id & 0x0080))
9604                 goto skip_bad_vf_detection;
9605
9606         pf_func = req_id & 0x01;
9607         if ((pf_func & 1) == (pdev->devfn & 1)) {
9608                 unsigned int device_id;
9609
9610                 vf = (req_id & 0x7F) >> 1;
9611                 e_dev_err("VF %d has caused a PCIe error\n", vf);
9612                 e_dev_err("TLP: dw0: %8.8x\tdw1: %8.8x\tdw2: "
9613                                 "%8.8x\tdw3: %8.8x\n",
9614                 dw0, dw1, dw2, dw3);
9615                 switch (adapter->hw.mac.type) {
9616                 case ixgbe_mac_82599EB:
9617                         device_id = IXGBE_82599_VF_DEVICE_ID;
9618                         break;
9619                 case ixgbe_mac_X540:
9620                         device_id = IXGBE_X540_VF_DEVICE_ID;
9621                         break;
9622                 case ixgbe_mac_X550:
9623                         device_id = IXGBE_DEV_ID_X550_VF;
9624                         break;
9625                 case ixgbe_mac_X550EM_x:
9626                         device_id = IXGBE_DEV_ID_X550EM_X_VF;
9627                         break;
9628                 default:
9629                         device_id = 0;
9630                         break;
9631                 }
9632
9633                 /* Find the pci device of the offending VF */
9634                 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, device_id, NULL);
9635                 while (vfdev) {
9636                         if (vfdev->devfn == (req_id & 0xFF))
9637                                 break;
9638                         vfdev = pci_get_device(PCI_VENDOR_ID_INTEL,
9639                                                device_id, vfdev);
9640                 }
9641                 /*
9642                  * There's a slim chance the VF could have been hot plugged,
9643                  * so if it is no longer present we don't need to issue the
9644                  * VFLR.  Just clean up the AER in that case.
9645                  */
9646                 if (vfdev) {
9647                         ixgbe_issue_vf_flr(adapter, vfdev);
9648                         /* Free device reference count */
9649                         pci_dev_put(vfdev);
9650                 }
9651
9652                 pci_cleanup_aer_uncorrect_error_status(pdev);
9653         }
9654
9655         /*
9656          * Even though the error may have occurred on the other port
9657          * we still need to increment the vf error reference count for
9658          * both ports because the I/O resume function will be called
9659          * for both of them.
9660          */
9661         adapter->vferr_refcount++;
9662
9663         return PCI_ERS_RESULT_RECOVERED;
9664
9665 skip_bad_vf_detection:
9666 #endif /* CONFIG_PCI_IOV */
9667         if (!test_bit(__IXGBE_SERVICE_INITED, &adapter->state))
9668                 return PCI_ERS_RESULT_DISCONNECT;
9669
9670         rtnl_lock();
9671         netif_device_detach(netdev);
9672
9673         if (state == pci_channel_io_perm_failure) {
9674                 rtnl_unlock();
9675                 return PCI_ERS_RESULT_DISCONNECT;
9676         }
9677
9678         if (netif_running(netdev))
9679                 ixgbe_down(adapter);
9680
9681         if (!test_and_set_bit(__IXGBE_DISABLED, &adapter->state))
9682                 pci_disable_device(pdev);
9683         rtnl_unlock();
9684
9685         /* Request a slot reset. */
9686         return PCI_ERS_RESULT_NEED_RESET;
9687 }
9688
9689 /**
9690  * ixgbe_io_slot_reset - called after the pci bus has been reset.
9691  * @pdev: Pointer to PCI device
9692  *
9693  * Restart the card from scratch, as if from a cold-boot.
9694  */
9695 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
9696 {
9697         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9698         pci_ers_result_t result;
9699         int err;
9700
9701         if (pci_enable_device_mem(pdev)) {
9702                 e_err(probe, "Cannot re-enable PCI device after reset.\n");
9703                 result = PCI_ERS_RESULT_DISCONNECT;
9704         } else {
9705                 smp_mb__before_atomic();
9706                 clear_bit(__IXGBE_DISABLED, &adapter->state);
9707                 adapter->hw.hw_addr = adapter->io_addr;
9708                 pci_set_master(pdev);
9709                 pci_restore_state(pdev);
9710                 pci_save_state(pdev);
9711
9712                 pci_wake_from_d3(pdev, false);
9713
9714                 ixgbe_reset(adapter);
9715                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_WUS, ~0);
9716                 result = PCI_ERS_RESULT_RECOVERED;
9717         }
9718
9719         err = pci_cleanup_aer_uncorrect_error_status(pdev);
9720         if (err) {
9721                 e_dev_err("pci_cleanup_aer_uncorrect_error_status "
9722                           "failed 0x%0x\n", err);
9723                 /* non-fatal, continue */
9724         }
9725
9726         return result;
9727 }
9728
9729 /**
9730  * ixgbe_io_resume - called when traffic can start flowing again.
9731  * @pdev: Pointer to PCI device
9732  *
9733  * This callback is called when the error recovery driver tells us that
9734  * its OK to resume normal operation.
9735  */
9736 static void ixgbe_io_resume(struct pci_dev *pdev)
9737 {
9738         struct ixgbe_adapter *adapter = pci_get_drvdata(pdev);
9739         struct net_device *netdev = adapter->netdev;
9740
9741 #ifdef CONFIG_PCI_IOV
9742         if (adapter->vferr_refcount) {
9743                 e_info(drv, "Resuming after VF err\n");
9744                 adapter->vferr_refcount--;
9745                 return;
9746         }
9747
9748 #endif
9749         if (netif_running(netdev))
9750                 ixgbe_up(adapter);
9751
9752         netif_device_attach(netdev);
9753 }
9754
9755 static const struct pci_error_handlers ixgbe_err_handler = {
9756         .error_detected = ixgbe_io_error_detected,
9757         .slot_reset = ixgbe_io_slot_reset,
9758         .resume = ixgbe_io_resume,
9759 };
9760
9761 static struct pci_driver ixgbe_driver = {
9762         .name     = ixgbe_driver_name,
9763         .id_table = ixgbe_pci_tbl,
9764         .probe    = ixgbe_probe,
9765         .remove   = ixgbe_remove,
9766 #ifdef CONFIG_PM
9767         .suspend  = ixgbe_suspend,
9768         .resume   = ixgbe_resume,
9769 #endif
9770         .shutdown = ixgbe_shutdown,
9771         .sriov_configure = ixgbe_pci_sriov_configure,
9772         .err_handler = &ixgbe_err_handler
9773 };
9774
9775 /**
9776  * ixgbe_init_module - Driver Registration Routine
9777  *
9778  * ixgbe_init_module is the first routine called when the driver is
9779  * loaded. All it does is register with the PCI subsystem.
9780  **/
9781 static int __init ixgbe_init_module(void)
9782 {
9783         int ret;
9784         pr_info("%s - version %s\n", ixgbe_driver_string, ixgbe_driver_version);
9785         pr_info("%s\n", ixgbe_copyright);
9786
9787         ixgbe_wq = create_singlethread_workqueue(ixgbe_driver_name);
9788         if (!ixgbe_wq) {
9789                 pr_err("%s: Failed to create workqueue\n", ixgbe_driver_name);
9790                 return -ENOMEM;
9791         }
9792
9793         ixgbe_dbg_init();
9794
9795         ret = pci_register_driver(&ixgbe_driver);
9796         if (ret) {
9797                 ixgbe_dbg_exit();
9798                 return ret;
9799         }
9800
9801 #ifdef CONFIG_IXGBE_DCA
9802         dca_register_notify(&dca_notifier);
9803 #endif
9804
9805         return 0;
9806 }
9807
9808 module_init(ixgbe_init_module);
9809
9810 /**
9811  * ixgbe_exit_module - Driver Exit Cleanup Routine
9812  *
9813  * ixgbe_exit_module is called just before the driver is removed
9814  * from memory.
9815  **/
9816 static void __exit ixgbe_exit_module(void)
9817 {
9818 #ifdef CONFIG_IXGBE_DCA
9819         dca_unregister_notify(&dca_notifier);
9820 #endif
9821         pci_unregister_driver(&ixgbe_driver);
9822
9823         ixgbe_dbg_exit();
9824         if (ixgbe_wq) {
9825                 destroy_workqueue(ixgbe_wq);
9826                 ixgbe_wq = NULL;
9827         }
9828 }
9829
9830 #ifdef CONFIG_IXGBE_DCA
9831 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
9832                             void *p)
9833 {
9834         int ret_val;
9835
9836         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
9837                                          __ixgbe_notify_dca);
9838
9839         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
9840 }
9841
9842 #endif /* CONFIG_IXGBE_DCA */
9843
9844 module_exit(ixgbe_exit_module);
9845
9846 /* ixgbe_main.c */