drivers/net/ethernet/intel/igb/igb_main.c

   1 /*******************************************************************************
   2
   3   Intel(R) Gigabit Ethernet Linux driver
   4   Copyright(c) 2007-2013 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   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30 #include <linux/module.h>
  31 #include <linux/types.h>
  32 #include <linux/init.h>
  33 #include <linux/bitops.h>
  34 #include <linux/vmalloc.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/netdevice.h>
  37 #include <linux/ipv6.h>
  38 #include <linux/slab.h>
  39 #include <net/checksum.h>
  40 #include <net/ip6_checksum.h>
  41 #include <linux/net_tstamp.h>
  42 #include <linux/mii.h>
  43 #include <linux/ethtool.h>
  44 #include <linux/if.h>
  45 #include <linux/if_vlan.h>
  46 #include <linux/pci.h>
  47 #include <linux/pci-aspm.h>
  48 #include <linux/delay.h>
  49 #include <linux/interrupt.h>
  50 #include <linux/ip.h>
  51 #include <linux/tcp.h>
  52 #include <linux/sctp.h>
  53 #include <linux/if_ether.h>
  54 #include <linux/aer.h>
  55 #include <linux/prefetch.h>
  56 #include <linux/pm_runtime.h>
  57 #ifdef CONFIG_IGB_DCA
  58 #include <linux/dca.h>
  59 #endif
  60 #include <linux/i2c.h>
  61 #include "igb.h"
  62
  63 #define MAJ 4
  64 #define MIN 1
  65 #define BUILD 2
  66 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
  67 __stringify(BUILD) "-k"
  68 char igb_driver_name[] = "igb";
  69 char igb_driver_version[] = DRV_VERSION;
  70 static const char igb_driver_string[] =
  71                                 "Intel(R) Gigabit Ethernet Network Driver";
  72 static const char igb_copyright[] =
  73                                 "Copyright (c) 2007-2013 Intel Corporation.";
  74
  75 static const struct e1000_info *igb_info_tbl[] = {
  76         [board_82575] = &e1000_82575_info,
  77 };
  78
  79 static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
  80         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
  81         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
  82         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
  83         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
  84         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
  85         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
  86         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
  87         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
  88         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
  89         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
  90         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
  91         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
  92         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
  93         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
  94         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
  95         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
  96         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
  97         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
  98         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
  99         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
 100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
 101         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
 102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
 103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
 104         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
 105         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
 106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
 107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
 108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
 109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
 110         /* required last entry */
 111         {0, }
 112 };
 113
 114 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 115
 116 void igb_reset(struct igb_adapter *);
 117 static int igb_setup_all_tx_resources(struct igb_adapter *);
 118 static int igb_setup_all_rx_resources(struct igb_adapter *);
 119 static void igb_free_all_tx_resources(struct igb_adapter *);
 120 static void igb_free_all_rx_resources(struct igb_adapter *);
 121 static void igb_setup_mrqc(struct igb_adapter *);
 122 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 123 static void igb_remove(struct pci_dev *pdev);
 124 static int igb_sw_init(struct igb_adapter *);
 125 static int igb_open(struct net_device *);
 126 static int igb_close(struct net_device *);
 127 static void igb_configure(struct igb_adapter *);
 128 static void igb_configure_tx(struct igb_adapter *);
 129 static void igb_configure_rx(struct igb_adapter *);
 130 static void igb_clean_all_tx_rings(struct igb_adapter *);
 131 static void igb_clean_all_rx_rings(struct igb_adapter *);
 132 static void igb_clean_tx_ring(struct igb_ring *);
 133 static void igb_clean_rx_ring(struct igb_ring *);
 134 static void igb_set_rx_mode(struct net_device *);
 135 static void igb_update_phy_info(unsigned long);
 136 static void igb_watchdog(unsigned long);
 137 static void igb_watchdog_task(struct work_struct *);
 138 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 139 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
 140                                                  struct rtnl_link_stats64 *stats);
 141 static int igb_change_mtu(struct net_device *, int);
 142 static int igb_set_mac(struct net_device *, void *);
 143 static void igb_set_uta(struct igb_adapter *adapter);
 144 static irqreturn_t igb_intr(int irq, void *);
 145 static irqreturn_t igb_intr_msi(int irq, void *);
 146 static irqreturn_t igb_msix_other(int irq, void *);
 147 static irqreturn_t igb_msix_ring(int irq, void *);
 148 #ifdef CONFIG_IGB_DCA
 149 static void igb_update_dca(struct igb_q_vector *);
 150 static void igb_setup_dca(struct igb_adapter *);
 151 #endif /* CONFIG_IGB_DCA */
 152 static int igb_poll(struct napi_struct *, int);
 153 static bool igb_clean_tx_irq(struct igb_q_vector *);
 154 static bool igb_clean_rx_irq(struct igb_q_vector *, int);
 155 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 156 static void igb_tx_timeout(struct net_device *);
 157 static void igb_reset_task(struct work_struct *);
 158 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
 159 static int igb_vlan_rx_add_vid(struct net_device *, u16);
 160 static int igb_vlan_rx_kill_vid(struct net_device *, u16);
 161 static void igb_restore_vlan(struct igb_adapter *);
 162 static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
 163 static void igb_ping_all_vfs(struct igb_adapter *);
 164 static void igb_msg_task(struct igb_adapter *);
 165 static void igb_vmm_control(struct igb_adapter *);
 166 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 167 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 168 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 169 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 170                                int vf, u16 vlan, u8 qos);
 171 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
 172 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 173                                  struct ifla_vf_info *ivi);
 174 static void igb_check_vf_rate_limit(struct igb_adapter *);
 175
 176 #ifdef CONFIG_PCI_IOV
 177 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 178 static bool igb_vfs_are_assigned(struct igb_adapter *adapter);
 179 #endif
 180
 181 #ifdef CONFIG_PM
 182 #ifdef CONFIG_PM_SLEEP
 183 static int igb_suspend(struct device *);
 184 #endif
 185 static int igb_resume(struct device *);
 186 #ifdef CONFIG_PM_RUNTIME
 187 static int igb_runtime_suspend(struct device *dev);
 188 static int igb_runtime_resume(struct device *dev);
 189 static int igb_runtime_idle(struct device *dev);
 190 #endif
 191 static const struct dev_pm_ops igb_pm_ops = {
 192         SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
 193         SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
 194                         igb_runtime_idle)
 195 };
 196 #endif
 197 static void igb_shutdown(struct pci_dev *);
 198 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 199 #ifdef CONFIG_IGB_DCA
 200 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 201 static struct notifier_block dca_notifier = {
 202         .notifier_call  = igb_notify_dca,
 203         .next           = NULL,
 204         .priority       = 0
 205 };
 206 #endif
 207 #ifdef CONFIG_NET_POLL_CONTROLLER
 208 /* for netdump / net console */
 209 static void igb_netpoll(struct net_device *);
 210 #endif
 211 #ifdef CONFIG_PCI_IOV
 212 static unsigned int max_vfs = 0;
 213 module_param(max_vfs, uint, 0);
 214 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
 215                  "per physical function");
 216 #endif /* CONFIG_PCI_IOV */
 217
 218 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 219                      pci_channel_state_t);
 220 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 221 static void igb_io_resume(struct pci_dev *);
 222
 223 static const struct pci_error_handlers igb_err_handler = {
 224         .error_detected = igb_io_error_detected,
 225         .slot_reset = igb_io_slot_reset,
 226         .resume = igb_io_resume,
 227 };
 228
 229 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 230
 231 static struct pci_driver igb_driver = {
 232         .name     = igb_driver_name,
 233         .id_table = igb_pci_tbl,
 234         .probe    = igb_probe,
 235         .remove   = igb_remove,
 236 #ifdef CONFIG_PM
 237         .driver.pm = &igb_pm_ops,
 238 #endif
 239         .shutdown = igb_shutdown,
 240         .sriov_configure = igb_pci_sriov_configure,
 241         .err_handler = &igb_err_handler
 242 };
 243
 244 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 245 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 246 MODULE_LICENSE("GPL");
 247 MODULE_VERSION(DRV_VERSION);
 248
 249 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 250 static int debug = -1;
 251 module_param(debug, int, 0);
 252 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 253
 254 struct igb_reg_info {
 255         u32 ofs;
 256         char *name;
 257 };
 258
 259 static const struct igb_reg_info igb_reg_info_tbl[] = {
 260
 261         /* General Registers */
 262         {E1000_CTRL, "CTRL"},
 263         {E1000_STATUS, "STATUS"},
 264         {E1000_CTRL_EXT, "CTRL_EXT"},
 265
 266         /* Interrupt Registers */
 267         {E1000_ICR, "ICR"},
 268
 269         /* RX Registers */
 270         {E1000_RCTL, "RCTL"},
 271         {E1000_RDLEN(0), "RDLEN"},
 272         {E1000_RDH(0), "RDH"},
 273         {E1000_RDT(0), "RDT"},
 274         {E1000_RXDCTL(0), "RXDCTL"},
 275         {E1000_RDBAL(0), "RDBAL"},
 276         {E1000_RDBAH(0), "RDBAH"},
 277
 278         /* TX Registers */
 279         {E1000_TCTL, "TCTL"},
 280         {E1000_TDBAL(0), "TDBAL"},
 281         {E1000_TDBAH(0), "TDBAH"},
 282         {E1000_TDLEN(0), "TDLEN"},
 283         {E1000_TDH(0), "TDH"},
 284         {E1000_TDT(0), "TDT"},
 285         {E1000_TXDCTL(0), "TXDCTL"},
 286         {E1000_TDFH, "TDFH"},
 287         {E1000_TDFT, "TDFT"},
 288         {E1000_TDFHS, "TDFHS"},
 289         {E1000_TDFPC, "TDFPC"},
 290
 291         /* List Terminator */
 292         {}
 293 };
 294
 295 /*
 296  * igb_regdump - register printout routine
 297  */
 298 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
 299 {
 300         int n = 0;
 301         char rname[16];
 302         u32 regs[8];
 303
 304         switch (reginfo->ofs) {
 305         case E1000_RDLEN(0):
 306                 for (n = 0; n < 4; n++)
 307                         regs[n] = rd32(E1000_RDLEN(n));
 308                 break;
 309         case E1000_RDH(0):
 310                 for (n = 0; n < 4; n++)
 311                         regs[n] = rd32(E1000_RDH(n));
 312                 break;
 313         case E1000_RDT(0):
 314                 for (n = 0; n < 4; n++)
 315                         regs[n] = rd32(E1000_RDT(n));
 316                 break;
 317         case E1000_RXDCTL(0):
 318                 for (n = 0; n < 4; n++)
 319                         regs[n] = rd32(E1000_RXDCTL(n));
 320                 break;
 321         case E1000_RDBAL(0):
 322                 for (n = 0; n < 4; n++)
 323                         regs[n] = rd32(E1000_RDBAL(n));
 324                 break;
 325         case E1000_RDBAH(0):
 326                 for (n = 0; n < 4; n++)
 327                         regs[n] = rd32(E1000_RDBAH(n));
 328                 break;
 329         case E1000_TDBAL(0):
 330                 for (n = 0; n < 4; n++)
 331                         regs[n] = rd32(E1000_RDBAL(n));
 332                 break;
 333         case E1000_TDBAH(0):
 334                 for (n = 0; n < 4; n++)
 335                         regs[n] = rd32(E1000_TDBAH(n));
 336                 break;
 337         case E1000_TDLEN(0):
 338                 for (n = 0; n < 4; n++)
 339                         regs[n] = rd32(E1000_TDLEN(n));
 340                 break;
 341         case E1000_TDH(0):
 342                 for (n = 0; n < 4; n++)
 343                         regs[n] = rd32(E1000_TDH(n));
 344                 break;
 345         case E1000_TDT(0):
 346                 for (n = 0; n < 4; n++)
 347                         regs[n] = rd32(E1000_TDT(n));
 348                 break;
 349         case E1000_TXDCTL(0):
 350                 for (n = 0; n < 4; n++)
 351                         regs[n] = rd32(E1000_TXDCTL(n));
 352                 break;
 353         default:
 354                 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
 355                 return;
 356         }
 357
 358         snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
 359         pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
 360                 regs[2], regs[3]);
 361 }
 362
 363 /*
 364  * igb_dump - Print registers, tx-rings and rx-rings
 365  */
 366 static void igb_dump(struct igb_adapter *adapter)
 367 {
 368         struct net_device *netdev = adapter->netdev;
 369         struct e1000_hw *hw = &adapter->hw;
 370         struct igb_reg_info *reginfo;
 371         struct igb_ring *tx_ring;
 372         union e1000_adv_tx_desc *tx_desc;
 373         struct my_u0 { u64 a; u64 b; } *u0;
 374         struct igb_ring *rx_ring;
 375         union e1000_adv_rx_desc *rx_desc;
 376         u32 staterr;
 377         u16 i, n;
 378
 379         if (!netif_msg_hw(adapter))
 380                 return;
 381
 382         /* Print netdevice Info */
 383         if (netdev) {
 384                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 385                 pr_info("Device Name     state            trans_start      "
 386                         "last_rx\n");
 387                 pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
 388                         netdev->state, netdev->trans_start, netdev->last_rx);
 389         }
 390
 391         /* Print Registers */
 392         dev_info(&adapter->pdev->dev, "Register Dump\n");
 393         pr_info(" Register Name   Value\n");
 394         for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
 395              reginfo->name; reginfo++) {
 396                 igb_regdump(hw, reginfo);
 397         }
 398
 399         /* Print TX Ring Summary */
 400         if (!netdev || !netif_running(netdev))
 401                 goto exit;
 402
 403         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 404         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 405         for (n = 0; n < adapter->num_tx_queues; n++) {
 406                 struct igb_tx_buffer *buffer_info;
 407                 tx_ring = adapter->tx_ring[n];
 408                 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 409                 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
 410                         n, tx_ring->next_to_use, tx_ring->next_to_clean,
 411                         (u64)dma_unmap_addr(buffer_info, dma),
 412                         dma_unmap_len(buffer_info, len),
 413                         buffer_info->next_to_watch,
 414                         (u64)buffer_info->time_stamp);
 415         }
 416
 417         /* Print TX Rings */
 418         if (!netif_msg_tx_done(adapter))
 419                 goto rx_ring_summary;
 420
 421         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 422
 423         /* Transmit Descriptor Formats
 424          *
 425          * Advanced Transmit Descriptor
 426          *   +--------------------------------------------------------------+
 427          * 0 |         Buffer Address [63:0]                                |
 428          *   +--------------------------------------------------------------+
 429          * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
 430          *   +--------------------------------------------------------------+
 431          *   63      46 45    40 39 38 36 35 32 31   24             15       0
 432          */
 433
 434         for (n = 0; n < adapter->num_tx_queues; n++) {
 435                 tx_ring = adapter->tx_ring[n];
 436                 pr_info("------------------------------------\n");
 437                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 438                 pr_info("------------------------------------\n");
 439                 pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] "
 440                         "[bi->dma       ] leng  ntw timestamp        "
 441                         "bi->skb\n");
 442
 443                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 444                         const char *next_desc;
 445                         struct igb_tx_buffer *buffer_info;
 446                         tx_desc = IGB_TX_DESC(tx_ring, i);
 447                         buffer_info = &tx_ring->tx_buffer_info[i];
 448                         u0 = (struct my_u0 *)tx_desc;
 449                         if (i == tx_ring->next_to_use &&
 450                             i == tx_ring->next_to_clean)
 451                                 next_desc = " NTC/U";
 452                         else if (i == tx_ring->next_to_use)
 453                                 next_desc = " NTU";
 454                         else if (i == tx_ring->next_to_clean)
 455                                 next_desc = " NTC";
 456                         else
 457                                 next_desc = "";
 458
 459                         pr_info("T [0x%03X]    %016llX %016llX %016llX"
 460                                 " %04X  %p %016llX %p%s\n", i,
 461                                 le64_to_cpu(u0->a),
 462                                 le64_to_cpu(u0->b),
 463                                 (u64)dma_unmap_addr(buffer_info, dma),
 464                                 dma_unmap_len(buffer_info, len),
 465                                 buffer_info->next_to_watch,
 466                                 (u64)buffer_info->time_stamp,
 467                                 buffer_info->skb, next_desc);
 468
 469                         if (netif_msg_pktdata(adapter) && buffer_info->skb)
 470                                 print_hex_dump(KERN_INFO, "",
 471                                         DUMP_PREFIX_ADDRESS,
 472                                         16, 1, buffer_info->skb->data,
 473                                         dma_unmap_len(buffer_info, len),
 474                                         true);
 475                 }
 476         }
 477
 478         /* Print RX Rings Summary */
 479 rx_ring_summary:
 480         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 481         pr_info("Queue [NTU] [NTC]\n");
 482         for (n = 0; n < adapter->num_rx_queues; n++) {
 483                 rx_ring = adapter->rx_ring[n];
 484                 pr_info(" %5d %5X %5X\n",
 485                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 486         }
 487
 488         /* Print RX Rings */
 489         if (!netif_msg_rx_status(adapter))
 490                 goto exit;
 491
 492         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 493
 494         /* Advanced Receive Descriptor (Read) Format
 495          *    63                                           1        0
 496          *    +-----------------------------------------------------+
 497          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 498          *    +----------------------------------------------+------+
 499          *  8 |       Header Buffer Address [63:1]           |  DD  |
 500          *    +-----------------------------------------------------+
 501          *
 502          *
 503          * Advanced Receive Descriptor (Write-Back) Format
 504          *
 505          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 506          *   +------------------------------------------------------+
 507          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 508          *   | Checksum   Ident  |   |           |    | Type | Type |
 509          *   +------------------------------------------------------+
 510          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 511          *   +------------------------------------------------------+
 512          *   63       48 47    32 31            20 19               0
 513          */
 514
 515         for (n = 0; n < adapter->num_rx_queues; n++) {
 516                 rx_ring = adapter->rx_ring[n];
 517                 pr_info("------------------------------------\n");
 518                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 519                 pr_info("------------------------------------\n");
 520                 pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] "
 521                         "[bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
 522                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] -----"
 523                         "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
 524
 525                 for (i = 0; i < rx_ring->count; i++) {
 526                         const char *next_desc;
 527                         struct igb_rx_buffer *buffer_info;
 528                         buffer_info = &rx_ring->rx_buffer_info[i];
 529                         rx_desc = IGB_RX_DESC(rx_ring, i);
 530                         u0 = (struct my_u0 *)rx_desc;
 531                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 532
 533                         if (i == rx_ring->next_to_use)
 534                                 next_desc = " NTU";
 535                         else if (i == rx_ring->next_to_clean)
 536                                 next_desc = " NTC";
 537                         else
 538                                 next_desc = "";
 539
 540                         if (staterr & E1000_RXD_STAT_DD) {
 541                                 /* Descriptor Done */
 542                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
 543                                         "RWB", i,
 544                                         le64_to_cpu(u0->a),
 545                                         le64_to_cpu(u0->b),
 546                                         next_desc);
 547                         } else {
 548                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
 549                                         "R  ", i,
 550                                         le64_to_cpu(u0->a),
 551                                         le64_to_cpu(u0->b),
 552                                         (u64)buffer_info->dma,
 553                                         next_desc);
 554
 555                                 if (netif_msg_pktdata(adapter) &&
 556                                     buffer_info->dma && buffer_info->page) {
 557                                         print_hex_dump(KERN_INFO, "",
 558                                           DUMP_PREFIX_ADDRESS,
 559                                           16, 1,
 560                                           page_address(buffer_info->page) +
 561                                                       buffer_info->page_offset,
 562                                           IGB_RX_BUFSZ, true);
 563                                 }
 564                         }
 565                 }
 566         }
 567
 568 exit:
 569         return;
 570 }
 571
 572 /*  igb_get_i2c_data - Reads the I2C SDA data bit
 573  *  @hw: pointer to hardware structure
 574  *  @i2cctl: Current value of I2CCTL register
 575  *
 576  *  Returns the I2C data bit value
 577  */
 578 static int igb_get_i2c_data(void *data)
 579 {
 580         struct igb_adapter *adapter = (struct igb_adapter *)data;
 581         struct e1000_hw *hw = &adapter->hw;
 582         s32 i2cctl = rd32(E1000_I2CPARAMS);
 583
 584         return ((i2cctl & E1000_I2C_DATA_IN) != 0);
 585 }
 586
 587 /* igb_set_i2c_data - Sets the I2C data bit
 588  *  @data: pointer to hardware structure
 589  *  @state: I2C data value (0 or 1) to set
 590  *
 591  *  Sets the I2C data bit
 592  */
 593 static void igb_set_i2c_data(void *data, int state)
 594 {
 595         struct igb_adapter *adapter = (struct igb_adapter *)data;
 596         struct e1000_hw *hw = &adapter->hw;
 597         s32 i2cctl = rd32(E1000_I2CPARAMS);
 598
 599         if (state)
 600                 i2cctl |= E1000_I2C_DATA_OUT;
 601         else
 602                 i2cctl &= ~E1000_I2C_DATA_OUT;
 603
 604         i2cctl &= ~E1000_I2C_DATA_OE_N;
 605         i2cctl |= E1000_I2C_CLK_OE_N;
 606         wr32(E1000_I2CPARAMS, i2cctl);
 607         wrfl();
 608
 609 }
 610
 611 /* igb_set_i2c_clk - Sets the I2C SCL clock
 612  *  @data: pointer to hardware structure
 613  *  @state: state to set clock
 614  *
 615  *  Sets the I2C clock line to state
 616  */
 617 static void igb_set_i2c_clk(void *data, int state)
 618 {
 619         struct igb_adapter *adapter = (struct igb_adapter *)data;
 620         struct e1000_hw *hw = &adapter->hw;
 621         s32 i2cctl = rd32(E1000_I2CPARAMS);
 622
 623         if (state) {
 624                 i2cctl |= E1000_I2C_CLK_OUT;
 625                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 626         } else {
 627                 i2cctl &= ~E1000_I2C_CLK_OUT;
 628                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 629         }
 630         wr32(E1000_I2CPARAMS, i2cctl);
 631         wrfl();
 632 }
 633
 634 /* igb_get_i2c_clk - Gets the I2C SCL clock state
 635  *  @data: pointer to hardware structure
 636  *
 637  *  Gets the I2C clock state
 638  */
 639 static int igb_get_i2c_clk(void *data)
 640 {
 641         struct igb_adapter *adapter = (struct igb_adapter *)data;
 642         struct e1000_hw *hw = &adapter->hw;
 643         s32 i2cctl = rd32(E1000_I2CPARAMS);
 644
 645         return ((i2cctl & E1000_I2C_CLK_IN) != 0);
 646 }
 647
 648 static const struct i2c_algo_bit_data igb_i2c_algo = {
 649         .setsda         = igb_set_i2c_data,
 650         .setscl         = igb_set_i2c_clk,
 651         .getsda         = igb_get_i2c_data,
 652         .getscl         = igb_get_i2c_clk,
 653         .udelay         = 5,
 654         .timeout        = 20,
 655 };
 656
 657 /**
 658  * igb_get_hw_dev - return device
 659  * used by hardware layer to print debugging information
 660  **/
 661 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
 662 {
 663         struct igb_adapter *adapter = hw->back;
 664         return adapter->netdev;
 665 }
 666
 667 /**
 668  * igb_init_module - Driver Registration Routine
 669  *
 670  * igb_init_module is the first routine called when the driver is
 671  * loaded. All it does is register with the PCI subsystem.
 672  **/
 673 static int __init igb_init_module(void)
 674 {
 675         int ret;
 676         pr_info("%s - version %s\n",
 677                igb_driver_string, igb_driver_version);
 678
 679         pr_info("%s\n", igb_copyright);
 680
 681 #ifdef CONFIG_IGB_DCA
 682         dca_register_notify(&dca_notifier);
 683 #endif
 684         ret = pci_register_driver(&igb_driver);
 685         return ret;
 686 }
 687
 688 module_init(igb_init_module);
 689
 690 /**
 691  * igb_exit_module - Driver Exit Cleanup Routine
 692  *
 693  * igb_exit_module is called just before the driver is removed
 694  * from memory.
 695  **/
 696 static void __exit igb_exit_module(void)
 697 {
 698 #ifdef CONFIG_IGB_DCA
 699         dca_unregister_notify(&dca_notifier);
 700 #endif
 701         pci_unregister_driver(&igb_driver);
 702 }
 703
 704 module_exit(igb_exit_module);
 705
 706 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 707 /**
 708  * igb_cache_ring_register - Descriptor ring to register mapping
 709  * @adapter: board private structure to initialize
 710  *
 711  * Once we know the feature-set enabled for the device, we'll cache
 712  * the register offset the descriptor ring is assigned to.
 713  **/
 714 static void igb_cache_ring_register(struct igb_adapter *adapter)
 715 {
 716         int i = 0, j = 0;
 717         u32 rbase_offset = adapter->vfs_allocated_count;
 718
 719         switch (adapter->hw.mac.type) {
 720         case e1000_82576:
 721                 /* The queues are allocated for virtualization such that VF 0
 722                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 723                  * In order to avoid collision we start at the first free queue
 724                  * and continue consuming queues in the same sequence
 725                  */
 726                 if (adapter->vfs_allocated_count) {
 727                         for (; i < adapter->rss_queues; i++)
 728                                 adapter->rx_ring[i]->reg_idx = rbase_offset +
 729                                                                Q_IDX_82576(i);
 730                 }
 731         case e1000_82575:
 732         case e1000_82580:
 733         case e1000_i350:
 734         case e1000_i210:
 735         case e1000_i211:
 736         default:
 737                 for (; i < adapter->num_rx_queues; i++)
 738                         adapter->rx_ring[i]->reg_idx = rbase_offset + i;
 739                 for (; j < adapter->num_tx_queues; j++)
 740                         adapter->tx_ring[j]->reg_idx = rbase_offset + j;
 741                 break;
 742         }
 743 }
 744
 745 /**
 746  *  igb_write_ivar - configure ivar for given MSI-X vector
 747  *  @hw: pointer to the HW structure
 748  *  @msix_vector: vector number we are allocating to a given ring
 749  *  @index: row index of IVAR register to write within IVAR table
 750  *  @offset: column offset of in IVAR, should be multiple of 8
 751  *
 752  *  This function is intended to handle the writing of the IVAR register
 753  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 754  *  each containing an cause allocation for an Rx and Tx ring, and a
 755  *  variable number of rows depending on the number of queues supported.
 756  **/
 757 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 758                            int index, int offset)
 759 {
 760         u32 ivar = array_rd32(E1000_IVAR0, index);
 761
 762         /* clear any bits that are currently set */
 763         ivar &= ~((u32)0xFF << offset);
 764
 765         /* write vector and valid bit */
 766         ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 767
 768         array_wr32(E1000_IVAR0, index, ivar);
 769 }
 770
 771 #define IGB_N0_QUEUE -1
 772 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 773 {
 774         struct igb_adapter *adapter = q_vector->adapter;
 775         struct e1000_hw *hw = &adapter->hw;
 776         int rx_queue = IGB_N0_QUEUE;
 777         int tx_queue = IGB_N0_QUEUE;
 778         u32 msixbm = 0;
 779
 780         if (q_vector->rx.ring)
 781                 rx_queue = q_vector->rx.ring->reg_idx;
 782         if (q_vector->tx.ring)
 783                 tx_queue = q_vector->tx.ring->reg_idx;
 784
 785         switch (hw->mac.type) {
 786         case e1000_82575:
 787                 /* The 82575 assigns vectors using a bitmask, which matches the
 788                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
 789                    or more queues to a vector, we write the appropriate bits
 790                    into the MSIXBM register for that vector. */
 791                 if (rx_queue > IGB_N0_QUEUE)
 792                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 793                 if (tx_queue > IGB_N0_QUEUE)
 794                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 795                 if (!adapter->msix_entries && msix_vector == 0)
 796                         msixbm |= E1000_EIMS_OTHER;
 797                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 798                 q_vector->eims_value = msixbm;
 799                 break;
 800         case e1000_82576:
 801                 /*
 802                  * 82576 uses a table that essentially consists of 2 columns
 803                  * with 8 rows.  The ordering is column-major so we use the
 804                  * lower 3 bits as the row index, and the 4th bit as the
 805                  * column offset.
 806                  */
 807                 if (rx_queue > IGB_N0_QUEUE)
 808                         igb_write_ivar(hw, msix_vector,
 809                                        rx_queue & 0x7,
 810                                        (rx_queue & 0x8) << 1);
 811                 if (tx_queue > IGB_N0_QUEUE)
 812                         igb_write_ivar(hw, msix_vector,
 813                                        tx_queue & 0x7,
 814                                        ((tx_queue & 0x8) << 1) + 8);
 815                 q_vector->eims_value = 1 << msix_vector;
 816                 break;
 817         case e1000_82580:
 818         case e1000_i350:
 819         case e1000_i210:
 820         case e1000_i211:
 821                 /*
 822                  * On 82580 and newer adapters the scheme is similar to 82576
 823                  * however instead of ordering column-major we have things
 824                  * ordered row-major.  So we traverse the table by using
 825                  * bit 0 as the column offset, and the remaining bits as the
 826                  * row index.
 827                  */
 828                 if (rx_queue > IGB_N0_QUEUE)
 829                         igb_write_ivar(hw, msix_vector,
 830                                        rx_queue >> 1,
 831                                        (rx_queue & 0x1) << 4);
 832                 if (tx_queue > IGB_N0_QUEUE)
 833                         igb_write_ivar(hw, msix_vector,
 834                                        tx_queue >> 1,
 835                                        ((tx_queue & 0x1) << 4) + 8);
 836                 q_vector->eims_value = 1 << msix_vector;
 837                 break;
 838         default:
 839                 BUG();
 840                 break;
 841         }
 842
 843         /* add q_vector eims value to global eims_enable_mask */
 844         adapter->eims_enable_mask |= q_vector->eims_value;
 845
 846         /* configure q_vector to set itr on first interrupt */
 847         q_vector->set_itr = 1;
 848 }
 849
 850 /**
 851  * igb_configure_msix - Configure MSI-X hardware
 852  *
 853  * igb_configure_msix sets up the hardware to properly
 854  * generate MSI-X interrupts.
 855  **/
 856 static void igb_configure_msix(struct igb_adapter *adapter)
 857 {
 858         u32 tmp;
 859         int i, vector = 0;
 860         struct e1000_hw *hw = &adapter->hw;
 861
 862         adapter->eims_enable_mask = 0;
 863
 864         /* set vector for other causes, i.e. link changes */
 865         switch (hw->mac.type) {
 866         case e1000_82575:
 867                 tmp = rd32(E1000_CTRL_EXT);
 868                 /* enable MSI-X PBA support*/
 869                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 870
 871                 /* Auto-Mask interrupts upon ICR read. */
 872                 tmp |= E1000_CTRL_EXT_EIAME;
 873                 tmp |= E1000_CTRL_EXT_IRCA;
 874
 875                 wr32(E1000_CTRL_EXT, tmp);
 876
 877                 /* enable msix_other interrupt */
 878                 array_wr32(E1000_MSIXBM(0), vector++,
 879                                       E1000_EIMS_OTHER);
 880                 adapter->eims_other = E1000_EIMS_OTHER;
 881
 882                 break;
 883
 884         case e1000_82576:
 885         case e1000_82580:
 886         case e1000_i350:
 887         case e1000_i210:
 888         case e1000_i211:
 889                 /* Turn on MSI-X capability first, or our settings
 890                  * won't stick.  And it will take days to debug. */
 891                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 892                                 E1000_GPIE_PBA | E1000_GPIE_EIAME |
 893                                 E1000_GPIE_NSICR);
 894
 895                 /* enable msix_other interrupt */
 896                 adapter->eims_other = 1 << vector;
 897                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 898
 899                 wr32(E1000_IVAR_MISC, tmp);
 900                 break;
 901         default:
 902                 /* do nothing, since nothing else supports MSI-X */
 903                 break;
 904         } /* switch (hw->mac.type) */
 905
 906         adapter->eims_enable_mask |= adapter->eims_other;
 907
 908         for (i = 0; i < adapter->num_q_vectors; i++)
 909                 igb_assign_vector(adapter->q_vector[i], vector++);
 910
 911         wrfl();
 912 }
 913
 914 /**
 915  * igb_request_msix - Initialize MSI-X interrupts
 916  *
 917  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
 918  * kernel.
 919  **/
 920 static int igb_request_msix(struct igb_adapter *adapter)
 921 {
 922         struct net_device *netdev = adapter->netdev;
 923         struct e1000_hw *hw = &adapter->hw;
 924         int i, err = 0, vector = 0, free_vector = 0;
 925
 926         err = request_irq(adapter->msix_entries[vector].vector,
 927                           igb_msix_other, 0, netdev->name, adapter);
 928         if (err)
 929                 goto err_out;
 930
 931         for (i = 0; i < adapter->num_q_vectors; i++) {
 932                 struct igb_q_vector *q_vector = adapter->q_vector[i];
 933
 934                 vector++;
 935
 936                 q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
 937
 938                 if (q_vector->rx.ring && q_vector->tx.ring)
 939                         sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 940                                 q_vector->rx.ring->queue_index);
 941                 else if (q_vector->tx.ring)
 942                         sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 943                                 q_vector->tx.ring->queue_index);
 944                 else if (q_vector->rx.ring)
 945                         sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 946                                 q_vector->rx.ring->queue_index);
 947                 else
 948                         sprintf(q_vector->name, "%s-unused", netdev->name);
 949
 950                 err = request_irq(adapter->msix_entries[vector].vector,
 951                                   igb_msix_ring, 0, q_vector->name,
 952                                   q_vector);
 953                 if (err)
 954                         goto err_free;
 955         }
 956
 957         igb_configure_msix(adapter);
 958         return 0;
 959
 960 err_free:
 961         /* free already assigned IRQs */
 962         free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 963
 964         vector--;
 965         for (i = 0; i < vector; i++) {
 966                 free_irq(adapter->msix_entries[free_vector++].vector,
 967                          adapter->q_vector[i]);
 968         }
 969 err_out:
 970         return err;
 971 }
 972
 973 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 974 {
 975         if (adapter->msix_entries) {
 976                 pci_disable_msix(adapter->pdev);
 977                 kfree(adapter->msix_entries);
 978                 adapter->msix_entries = NULL;
 979         } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
 980                 pci_disable_msi(adapter->pdev);
 981         }
 982 }
 983
 984 /**
 985  * igb_free_q_vector - Free memory allocated for specific interrupt vector
 986  * @adapter: board private structure to initialize
 987  * @v_idx: Index of vector to be freed
 988  *
 989  * This function frees the memory allocated to the q_vector.  In addition if
 990  * NAPI is enabled it will delete any references to the NAPI struct prior
 991  * to freeing the q_vector.
 992  **/
 993 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
 994 {
 995         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
 996
 997         if (q_vector->tx.ring)
 998                 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
 999
1000         if (q_vector->rx.ring)
1001                 adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1002
1003         adapter->q_vector[v_idx] = NULL;
1004         netif_napi_del(&q_vector->napi);
1005
1006         /*
1007          * ixgbe_get_stats64() might access the rings on this vector,
1008          * we must wait a grace period before freeing it.
1009          */
1010         kfree_rcu(q_vector, rcu);
1011 }
1012
1013 /**
1014  * igb_free_q_vectors - Free memory allocated for interrupt vectors
1015  * @adapter: board private structure to initialize
1016  *
1017  * This function frees the memory allocated to the q_vectors.  In addition if
1018  * NAPI is enabled it will delete any references to the NAPI struct prior
1019  * to freeing the q_vector.
1020  **/
1021 static void igb_free_q_vectors(struct igb_adapter *adapter)
1022 {
1023         int v_idx = adapter->num_q_vectors;
1024
1025         adapter->num_tx_queues = 0;
1026         adapter->num_rx_queues = 0;
1027         adapter->num_q_vectors = 0;
1028
1029         while (v_idx--)
1030                 igb_free_q_vector(adapter, v_idx);
1031 }
1032
1033 /**
1034  * igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1035  *
1036  * This function resets the device so that it has 0 rx queues, tx queues, and
1037  * MSI-X interrupts allocated.
1038  */
1039 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1040 {
1041         igb_free_q_vectors(adapter);
1042         igb_reset_interrupt_capability(adapter);
1043 }
1044
1045 /**
1046  * igb_set_interrupt_capability - set MSI or MSI-X if supported
1047  *
1048  * Attempt to configure interrupts using the best available
1049  * capabilities of the hardware and kernel.
1050  **/
1051 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1052 {
1053         int err;
1054         int numvecs, i;
1055
1056         if (!msix)
1057                 goto msi_only;
1058
1059         /* Number of supported queues. */
1060         adapter->num_rx_queues = adapter->rss_queues;
1061         if (adapter->vfs_allocated_count)
1062                 adapter->num_tx_queues = 1;
1063         else
1064                 adapter->num_tx_queues = adapter->rss_queues;
1065
1066         /* start with one vector for every rx queue */
1067         numvecs = adapter->num_rx_queues;
1068
1069         /* if tx handler is separate add 1 for every tx queue */
1070         if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1071                 numvecs += adapter->num_tx_queues;
1072
1073         /* store the number of vectors reserved for queues */
1074         adapter->num_q_vectors = numvecs;
1075
1076         /* add 1 vector for link status interrupts */
1077         numvecs++;
1078         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1079                                         GFP_KERNEL);
1080
1081         if (!adapter->msix_entries)
1082                 goto msi_only;
1083
1084         for (i = 0; i < numvecs; i++)
1085                 adapter->msix_entries[i].entry = i;
1086
1087         err = pci_enable_msix(adapter->pdev,
1088                               adapter->msix_entries,
1089                               numvecs);
1090         if (err == 0)
1091                 return;
1092
1093         igb_reset_interrupt_capability(adapter);
1094
1095         /* If we can't do MSI-X, try MSI */
1096 msi_only:
1097 #ifdef CONFIG_PCI_IOV
1098         /* disable SR-IOV for non MSI-X configurations */
1099         if (adapter->vf_data) {
1100                 struct e1000_hw *hw = &adapter->hw;
1101                 /* disable iov and allow time for transactions to clear */
1102                 pci_disable_sriov(adapter->pdev);
1103                 msleep(500);
1104
1105                 kfree(adapter->vf_data);
1106                 adapter->vf_data = NULL;
1107                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1108                 wrfl();
1109                 msleep(100);
1110                 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1111         }
1112 #endif
1113         adapter->vfs_allocated_count = 0;
1114         adapter->rss_queues = 1;
1115         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1116         adapter->num_rx_queues = 1;
1117         adapter->num_tx_queues = 1;
1118         adapter->num_q_vectors = 1;
1119         if (!pci_enable_msi(adapter->pdev))
1120                 adapter->flags |= IGB_FLAG_HAS_MSI;
1121 }
1122
1123 static void igb_add_ring(struct igb_ring *ring,
1124                          struct igb_ring_container *head)
1125 {
1126         head->ring = ring;
1127         head->count++;
1128 }
1129
1130 /**
1131  * igb_alloc_q_vector - Allocate memory for a single interrupt vector
1132  * @adapter: board private structure to initialize
1133  * @v_count: q_vectors allocated on adapter, used for ring interleaving
1134  * @v_idx: index of vector in adapter struct
1135  * @txr_count: total number of Tx rings to allocate
1136  * @txr_idx: index of first Tx ring to allocate
1137  * @rxr_count: total number of Rx rings to allocate
1138  * @rxr_idx: index of first Rx ring to allocate
1139  *
1140  * We allocate one q_vector.  If allocation fails we return -ENOMEM.
1141  **/
1142 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1143                               int v_count, int v_idx,
1144                               int txr_count, int txr_idx,
1145                               int rxr_count, int rxr_idx)
1146 {
1147         struct igb_q_vector *q_vector;
1148         struct igb_ring *ring;
1149         int ring_count, size;
1150
1151         /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1152         if (txr_count > 1 || rxr_count > 1)
1153                 return -ENOMEM;
1154
1155         ring_count = txr_count + rxr_count;
1156         size = sizeof(struct igb_q_vector) +
1157                (sizeof(struct igb_ring) * ring_count);
1158
1159         /* allocate q_vector and rings */
1160         q_vector = kzalloc(size, GFP_KERNEL);
1161         if (!q_vector)
1162                 return -ENOMEM;
1163
1164         /* initialize NAPI */
1165         netif_napi_add(adapter->netdev, &q_vector->napi,
1166                        igb_poll, 64);
1167
1168         /* tie q_vector and adapter together */
1169         adapter->q_vector[v_idx] = q_vector;
1170         q_vector->adapter = adapter;
1171
1172         /* initialize work limits */
1173         q_vector->tx.work_limit = adapter->tx_work_limit;
1174
1175         /* initialize ITR configuration */
1176         q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1177         q_vector->itr_val = IGB_START_ITR;
1178
1179         /* initialize pointer to rings */
1180         ring = q_vector->ring;
1181
1182         if (txr_count) {
1183                 /* assign generic ring traits */
1184                 ring->dev = &adapter->pdev->dev;
1185                 ring->netdev = adapter->netdev;
1186
1187                 /* configure backlink on ring */
1188                 ring->q_vector = q_vector;
1189
1190                 /* update q_vector Tx values */
1191                 igb_add_ring(ring, &q_vector->tx);
1192
1193                 /* For 82575, context index must be unique per ring. */
1194                 if (adapter->hw.mac.type == e1000_82575)
1195                         set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1196
1197                 /* apply Tx specific ring traits */
1198                 ring->count = adapter->tx_ring_count;
1199                 ring->queue_index = txr_idx;
1200
1201                 /* assign ring to adapter */
1202                 adapter->tx_ring[txr_idx] = ring;
1203
1204                 /* push pointer to next ring */
1205                 ring++;
1206         }
1207
1208         if (rxr_count) {
1209                 /* assign generic ring traits */
1210                 ring->dev = &adapter->pdev->dev;
1211                 ring->netdev = adapter->netdev;
1212
1213                 /* configure backlink on ring */
1214                 ring->q_vector = q_vector;
1215
1216                 /* update q_vector Rx values */
1217                 igb_add_ring(ring, &q_vector->rx);
1218
1219                 /* set flag indicating ring supports SCTP checksum offload */
1220                 if (adapter->hw.mac.type >= e1000_82576)
1221                         set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1222
1223                 /*
1224                  * On i350, i210, and i211, loopback VLAN packets
1225                  * have the tag byte-swapped.
1226                  * */
1227                 if (adapter->hw.mac.type >= e1000_i350)
1228                         set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1229
1230                 /* apply Rx specific ring traits */
1231                 ring->count = adapter->rx_ring_count;
1232                 ring->queue_index = rxr_idx;
1233
1234                 /* assign ring to adapter */
1235                 adapter->rx_ring[rxr_idx] = ring;
1236         }
1237
1238         return 0;
1239 }
1240
1241
1242 /**
1243  * igb_alloc_q_vectors - Allocate memory for interrupt vectors
1244  * @adapter: board private structure to initialize
1245  *
1246  * We allocate one q_vector per queue interrupt.  If allocation fails we
1247  * return -ENOMEM.
1248  **/
1249 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1250 {
1251         int q_vectors = adapter->num_q_vectors;
1252         int rxr_remaining = adapter->num_rx_queues;
1253         int txr_remaining = adapter->num_tx_queues;
1254         int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1255         int err;
1256
1257         if (q_vectors >= (rxr_remaining + txr_remaining)) {
1258                 for (; rxr_remaining; v_idx++) {
1259                         err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1260                                                  0, 0, 1, rxr_idx);
1261
1262                         if (err)
1263                                 goto err_out;
1264
1265                         /* update counts and index */
1266                         rxr_remaining--;
1267                         rxr_idx++;
1268                 }
1269         }
1270
1271         for (; v_idx < q_vectors; v_idx++) {
1272                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1273                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1274                 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1275                                          tqpv, txr_idx, rqpv, rxr_idx);
1276
1277                 if (err)
1278                         goto err_out;
1279
1280                 /* update counts and index */
1281                 rxr_remaining -= rqpv;
1282                 txr_remaining -= tqpv;
1283                 rxr_idx++;
1284                 txr_idx++;
1285         }
1286
1287         return 0;
1288
1289 err_out:
1290         adapter->num_tx_queues = 0;
1291         adapter->num_rx_queues = 0;
1292         adapter->num_q_vectors = 0;
1293
1294         while (v_idx--)
1295                 igb_free_q_vector(adapter, v_idx);
1296
1297         return -ENOMEM;
1298 }
1299
1300 /**
1301  * igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1302  *
1303  * This function initializes the interrupts and allocates all of the queues.
1304  **/
1305 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1306 {
1307         struct pci_dev *pdev = adapter->pdev;
1308         int err;
1309
1310         igb_set_interrupt_capability(adapter, msix);
1311
1312         err = igb_alloc_q_vectors(adapter);
1313         if (err) {
1314                 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1315                 goto err_alloc_q_vectors;
1316         }
1317
1318         igb_cache_ring_register(adapter);
1319
1320         return 0;
1321
1322 err_alloc_q_vectors:
1323         igb_reset_interrupt_capability(adapter);
1324         return err;
1325 }
1326
1327 /**
1328  * igb_request_irq - initialize interrupts
1329  *
1330  * Attempts to configure interrupts using the best available
1331  * capabilities of the hardware and kernel.
1332  **/
1333 static int igb_request_irq(struct igb_adapter *adapter)
1334 {
1335         struct net_device *netdev = adapter->netdev;
1336         struct pci_dev *pdev = adapter->pdev;
1337         int err = 0;
1338
1339         if (adapter->msix_entries) {
1340                 err = igb_request_msix(adapter);
1341                 if (!err)
1342                         goto request_done;
1343                 /* fall back to MSI */
1344                 igb_free_all_tx_resources(adapter);
1345                 igb_free_all_rx_resources(adapter);
1346
1347                 igb_clear_interrupt_scheme(adapter);
1348                 err = igb_init_interrupt_scheme(adapter, false);
1349                 if (err)
1350                         goto request_done;
1351
1352                 igb_setup_all_tx_resources(adapter);
1353                 igb_setup_all_rx_resources(adapter);
1354                 igb_configure(adapter);
1355         }
1356
1357         igb_assign_vector(adapter->q_vector[0], 0);
1358
1359         if (adapter->flags & IGB_FLAG_HAS_MSI) {
1360                 err = request_irq(pdev->irq, igb_intr_msi, 0,
1361                                   netdev->name, adapter);
1362                 if (!err)
1363                         goto request_done;
1364
1365                 /* fall back to legacy interrupts */
1366                 igb_reset_interrupt_capability(adapter);
1367                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1368         }
1369
1370         err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1371                           netdev->name, adapter);
1372
1373         if (err)
1374                 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1375                         err);
1376
1377 request_done:
1378         return err;
1379 }
1380
1381 static void igb_free_irq(struct igb_adapter *adapter)
1382 {
1383         if (adapter->msix_entries) {
1384                 int vector = 0, i;
1385
1386                 free_irq(adapter->msix_entries[vector++].vector, adapter);
1387
1388                 for (i = 0; i < adapter->num_q_vectors; i++)
1389                         free_irq(adapter->msix_entries[vector++].vector,
1390                                  adapter->q_vector[i]);
1391         } else {
1392                 free_irq(adapter->pdev->irq, adapter);
1393         }
1394 }
1395
1396 /**
1397  * igb_irq_disable - Mask off interrupt generation on the NIC
1398  * @adapter: board private structure
1399  **/
1400 static void igb_irq_disable(struct igb_adapter *adapter)
1401 {
1402         struct e1000_hw *hw = &adapter->hw;
1403
1404         /*
1405          * we need to be careful when disabling interrupts.  The VFs are also
1406          * mapped into these registers and so clearing the bits can cause
1407          * issues on the VF drivers so we only need to clear what we set
1408          */
1409         if (adapter->msix_entries) {
1410                 u32 regval = rd32(E1000_EIAM);
1411                 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1412                 wr32(E1000_EIMC, adapter->eims_enable_mask);
1413                 regval = rd32(E1000_EIAC);
1414                 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1415         }
1416
1417         wr32(E1000_IAM, 0);
1418         wr32(E1000_IMC, ~0);
1419         wrfl();
1420         if (adapter->msix_entries) {
1421                 int i;
1422                 for (i = 0; i < adapter->num_q_vectors; i++)
1423                         synchronize_irq(adapter->msix_entries[i].vector);
1424         } else {
1425                 synchronize_irq(adapter->pdev->irq);
1426         }
1427 }
1428
1429 /**
1430  * igb_irq_enable - Enable default interrupt generation settings
1431  * @adapter: board private structure
1432  **/
1433 static void igb_irq_enable(struct igb_adapter *adapter)
1434 {
1435         struct e1000_hw *hw = &adapter->hw;
1436
1437         if (adapter->msix_entries) {
1438                 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1439                 u32 regval = rd32(E1000_EIAC);
1440                 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1441                 regval = rd32(E1000_EIAM);
1442                 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1443                 wr32(E1000_EIMS, adapter->eims_enable_mask);
1444                 if (adapter->vfs_allocated_count) {
1445                         wr32(E1000_MBVFIMR, 0xFF);
1446                         ims |= E1000_IMS_VMMB;
1447                 }
1448                 wr32(E1000_IMS, ims);
1449         } else {
1450                 wr32(E1000_IMS, IMS_ENABLE_MASK |
1451                                 E1000_IMS_DRSTA);
1452                 wr32(E1000_IAM, IMS_ENABLE_MASK |
1453                                 E1000_IMS_DRSTA);
1454         }
1455 }
1456
1457 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1458 {
1459         struct e1000_hw *hw = &adapter->hw;
1460         u16 vid = adapter->hw.mng_cookie.vlan_id;
1461         u16 old_vid = adapter->mng_vlan_id;
1462
1463         if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1464                 /* add VID to filter table */
1465                 igb_vfta_set(hw, vid, true);
1466                 adapter->mng_vlan_id = vid;
1467         } else {
1468                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1469         }
1470
1471         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1472             (vid != old_vid) &&
1473             !test_bit(old_vid, adapter->active_vlans)) {
1474                 /* remove VID from filter table */
1475                 igb_vfta_set(hw, old_vid, false);
1476         }
1477 }
1478
1479 /**
1480  * igb_release_hw_control - release control of the h/w to f/w
1481  * @adapter: address of board private structure
1482  *
1483  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1484  * For ASF and Pass Through versions of f/w this means that the
1485  * driver is no longer loaded.
1486  *
1487  **/
1488 static void igb_release_hw_control(struct igb_adapter *adapter)
1489 {
1490         struct e1000_hw *hw = &adapter->hw;
1491         u32 ctrl_ext;
1492
1493         /* Let firmware take over control of h/w */
1494         ctrl_ext = rd32(E1000_CTRL_EXT);
1495         wr32(E1000_CTRL_EXT,
1496                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1497 }
1498
1499 /**
1500  * igb_get_hw_control - get control of the h/w from f/w
1501  * @adapter: address of board private structure
1502  *
1503  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1504  * For ASF and Pass Through versions of f/w this means that
1505  * the driver is loaded.
1506  *
1507  **/
1508 static void igb_get_hw_control(struct igb_adapter *adapter)
1509 {
1510         struct e1000_hw *hw = &adapter->hw;
1511         u32 ctrl_ext;
1512
1513         /* Let firmware know the driver has taken over */
1514         ctrl_ext = rd32(E1000_CTRL_EXT);
1515         wr32(E1000_CTRL_EXT,
1516                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1517 }
1518
1519 /**
1520  * igb_configure - configure the hardware for RX and TX
1521  * @adapter: private board structure
1522  **/
1523 static void igb_configure(struct igb_adapter *adapter)
1524 {
1525         struct net_device *netdev = adapter->netdev;
1526         int i;
1527
1528         igb_get_hw_control(adapter);
1529         igb_set_rx_mode(netdev);
1530
1531         igb_restore_vlan(adapter);
1532
1533         igb_setup_tctl(adapter);
1534         igb_setup_mrqc(adapter);
1535         igb_setup_rctl(adapter);
1536
1537         igb_configure_tx(adapter);
1538         igb_configure_rx(adapter);
1539
1540         igb_rx_fifo_flush_82575(&adapter->hw);
1541
1542         /* call igb_desc_unused which always leaves
1543          * at least 1 descriptor unused to make sure
1544          * next_to_use != next_to_clean */
1545         for (i = 0; i < adapter->num_rx_queues; i++) {
1546                 struct igb_ring *ring = adapter->rx_ring[i];
1547                 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1548         }
1549 }
1550
1551 /**
1552  * igb_power_up_link - Power up the phy/serdes link
1553  * @adapter: address of board private structure
1554  **/
1555 void igb_power_up_link(struct igb_adapter *adapter)
1556 {
1557         igb_reset_phy(&adapter->hw);
1558
1559         if (adapter->hw.phy.media_type == e1000_media_type_copper)
1560                 igb_power_up_phy_copper(&adapter->hw);
1561         else
1562                 igb_power_up_serdes_link_82575(&adapter->hw);
1563 }
1564
1565 /**
1566  * igb_power_down_link - Power down the phy/serdes link
1567  * @adapter: address of board private structure
1568  */
1569 static void igb_power_down_link(struct igb_adapter *adapter)
1570 {
1571         if (adapter->hw.phy.media_type == e1000_media_type_copper)
1572                 igb_power_down_phy_copper_82575(&adapter->hw);
1573         else
1574                 igb_shutdown_serdes_link_82575(&adapter->hw);
1575 }
1576
1577 /**
1578  * igb_up - Open the interface and prepare it to handle traffic
1579  * @adapter: board private structure
1580  **/
1581 int igb_up(struct igb_adapter *adapter)
1582 {
1583         struct e1000_hw *hw = &adapter->hw;
1584         int i;
1585
1586         /* hardware has been reset, we need to reload some things */
1587         igb_configure(adapter);
1588
1589         clear_bit(__IGB_DOWN, &adapter->state);
1590
1591         for (i = 0; i < adapter->num_q_vectors; i++)
1592                 napi_enable(&(adapter->q_vector[i]->napi));
1593
1594         if (adapter->msix_entries)
1595                 igb_configure_msix(adapter);
1596         else
1597                 igb_assign_vector(adapter->q_vector[0], 0);
1598
1599         /* Clear any pending interrupts. */
1600         rd32(E1000_ICR);
1601         igb_irq_enable(adapter);
1602
1603         /* notify VFs that reset has been completed */
1604         if (adapter->vfs_allocated_count) {
1605                 u32 reg_data = rd32(E1000_CTRL_EXT);
1606                 reg_data |= E1000_CTRL_EXT_PFRSTD;
1607                 wr32(E1000_CTRL_EXT, reg_data);
1608         }
1609
1610         netif_tx_start_all_queues(adapter->netdev);
1611
1612         /* start the watchdog. */
1613         hw->mac.get_link_status = 1;
1614         schedule_work(&adapter->watchdog_task);
1615
1616         return 0;
1617 }
1618
1619 void igb_down(struct igb_adapter *adapter)
1620 {
1621         struct net_device *netdev = adapter->netdev;
1622         struct e1000_hw *hw = &adapter->hw;
1623         u32 tctl, rctl;
1624         int i;
1625
1626         /* signal that we're down so the interrupt handler does not
1627          * reschedule our watchdog timer */
1628         set_bit(__IGB_DOWN, &adapter->state);
1629
1630         /* disable receives in the hardware */
1631         rctl = rd32(E1000_RCTL);
1632         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1633         /* flush and sleep below */
1634
1635         netif_tx_stop_all_queues(netdev);
1636
1637         /* disable transmits in the hardware */
1638         tctl = rd32(E1000_TCTL);
1639         tctl &= ~E1000_TCTL_EN;
1640         wr32(E1000_TCTL, tctl);
1641         /* flush both disables and wait for them to finish */
1642         wrfl();
1643         msleep(10);
1644
1645         for (i = 0; i < adapter->num_q_vectors; i++)
1646                 napi_disable(&(adapter->q_vector[i]->napi));
1647
1648         igb_irq_disable(adapter);
1649
1650         del_timer_sync(&adapter->watchdog_timer);
1651         del_timer_sync(&adapter->phy_info_timer);
1652
1653         netif_carrier_off(netdev);
1654
1655         /* record the stats before reset*/
1656         spin_lock(&adapter->stats64_lock);
1657         igb_update_stats(adapter, &adapter->stats64);
1658         spin_unlock(&adapter->stats64_lock);
1659
1660         adapter->link_speed = 0;
1661         adapter->link_duplex = 0;
1662
1663         if (!pci_channel_offline(adapter->pdev))
1664                 igb_reset(adapter);
1665         igb_clean_all_tx_rings(adapter);
1666         igb_clean_all_rx_rings(adapter);
1667 #ifdef CONFIG_IGB_DCA
1668
1669         /* since we reset the hardware DCA settings were cleared */
1670         igb_setup_dca(adapter);
1671 #endif
1672 }
1673
1674 void igb_reinit_locked(struct igb_adapter *adapter)
1675 {
1676         WARN_ON(in_interrupt());
1677         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1678                 msleep(1);
1679         igb_down(adapter);
1680         igb_up(adapter);
1681         clear_bit(__IGB_RESETTING, &adapter->state);
1682 }
1683
1684 void igb_reset(struct igb_adapter *adapter)
1685 {
1686         struct pci_dev *pdev = adapter->pdev;
1687         struct e1000_hw *hw = &adapter->hw;
1688         struct e1000_mac_info *mac = &hw->mac;
1689         struct e1000_fc_info *fc = &hw->fc;
1690         u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1691
1692         /* Repartition Pba for greater than 9k mtu
1693          * To take effect CTRL.RST is required.
1694          */
1695         switch (mac->type) {
1696         case e1000_i350:
1697         case e1000_82580:
1698                 pba = rd32(E1000_RXPBS);
1699                 pba = igb_rxpbs_adjust_82580(pba);
1700                 break;
1701         case e1000_82576:
1702                 pba = rd32(E1000_RXPBS);
1703                 pba &= E1000_RXPBS_SIZE_MASK_82576;
1704                 break;
1705         case e1000_82575:
1706         case e1000_i210:
1707         case e1000_i211:
1708         default:
1709                 pba = E1000_PBA_34K;
1710                 break;
1711         }
1712
1713         if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1714             (mac->type < e1000_82576)) {
1715                 /* adjust PBA for jumbo frames */
1716                 wr32(E1000_PBA, pba);
1717
1718                 /* To maintain wire speed transmits, the Tx FIFO should be
1719                  * large enough to accommodate two full transmit packets,
1720                  * rounded up to the next 1KB and expressed in KB.  Likewise,
1721                  * the Rx FIFO should be large enough to accommodate at least
1722                  * one full receive packet and is similarly rounded up and
1723                  * expressed in KB. */
1724                 pba = rd32(E1000_PBA);
1725                 /* upper 16 bits has Tx packet buffer allocation size in KB */
1726                 tx_space = pba >> 16;
1727                 /* lower 16 bits has Rx packet buffer allocation size in KB */
1728                 pba &= 0xffff;
1729                 /* the tx fifo also stores 16 bytes of information about the tx
1730                  * but don't include ethernet FCS because hardware appends it */
1731                 min_tx_space = (adapter->max_frame_size +
1732                                 sizeof(union e1000_adv_tx_desc) -
1733                                 ETH_FCS_LEN) * 2;
1734                 min_tx_space = ALIGN(min_tx_space, 1024);
1735                 min_tx_space >>= 10;
1736                 /* software strips receive CRC, so leave room for it */
1737                 min_rx_space = adapter->max_frame_size;
1738                 min_rx_space = ALIGN(min_rx_space, 1024);
1739                 min_rx_space >>= 10;
1740
1741                 /* If current Tx allocation is less than the min Tx FIFO size,
1742                  * and the min Tx FIFO size is less than the current Rx FIFO
1743                  * allocation, take space away from current Rx allocation */
1744                 if (tx_space < min_tx_space &&
1745                     ((min_tx_space - tx_space) < pba)) {
1746                         pba = pba - (min_tx_space - tx_space);
1747
1748                         /* if short on rx space, rx wins and must trump tx
1749                          * adjustment */
1750                         if (pba < min_rx_space)
1751                                 pba = min_rx_space;
1752                 }
1753                 wr32(E1000_PBA, pba);
1754         }
1755
1756         /* flow control settings */
1757         /* The high water mark must be low enough to fit one full frame
1758          * (or the size used for early receive) above it in the Rx FIFO.
1759          * Set it to the lower of:
1760          * - 90% of the Rx FIFO size, or
1761          * - the full Rx FIFO size minus one full frame */
1762         hwm = min(((pba << 10) * 9 / 10),
1763                         ((pba << 10) - 2 * adapter->max_frame_size));
1764
1765         fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
1766         fc->low_water = fc->high_water - 16;
1767         fc->pause_time = 0xFFFF;
1768         fc->send_xon = 1;
1769         fc->current_mode = fc->requested_mode;
1770
1771         /* disable receive for all VFs and wait one second */
1772         if (adapter->vfs_allocated_count) {
1773                 int i;
1774                 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1775                         adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1776
1777                 /* ping all the active vfs to let them know we are going down */
1778                 igb_ping_all_vfs(adapter);
1779
1780                 /* disable transmits and receives */
1781                 wr32(E1000_VFRE, 0);
1782                 wr32(E1000_VFTE, 0);
1783         }
1784
1785         /* Allow time for pending master requests to run */
1786         hw->mac.ops.reset_hw(hw);
1787         wr32(E1000_WUC, 0);
1788
1789         if (hw->mac.ops.init_hw(hw))
1790                 dev_err(&pdev->dev, "Hardware Error\n");
1791
1792         /*
1793          * Flow control settings reset on hardware reset, so guarantee flow
1794          * control is off when forcing speed.
1795          */
1796         if (!hw->mac.autoneg)
1797                 igb_force_mac_fc(hw);
1798
1799         igb_init_dmac(adapter, pba);
1800 #ifdef CONFIG_IGB_HWMON
1801         /* Re-initialize the thermal sensor on i350 devices. */
1802         if (!test_bit(__IGB_DOWN, &adapter->state)) {
1803                 if (mac->type == e1000_i350 && hw->bus.func == 0) {
1804                         /* If present, re-initialize the external thermal sensor
1805                          * interface.
1806                          */
1807                         if (adapter->ets)
1808                                 mac->ops.init_thermal_sensor_thresh(hw);
1809                 }
1810         }
1811 #endif
1812         if (!netif_running(adapter->netdev))
1813                 igb_power_down_link(adapter);
1814
1815         igb_update_mng_vlan(adapter);
1816
1817         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1818         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1819
1820         /* Re-enable PTP, where applicable. */
1821         igb_ptp_reset(adapter);
1822
1823         igb_get_phy_info(hw);
1824 }
1825
1826 static netdev_features_t igb_fix_features(struct net_device *netdev,
1827         netdev_features_t features)
1828 {
1829         /*
1830          * Since there is no support for separate rx/tx vlan accel
1831          * enable/disable make sure tx flag is always in same state as rx.
1832          */
1833         if (features & NETIF_F_HW_VLAN_RX)
1834                 features |= NETIF_F_HW_VLAN_TX;
1835         else
1836                 features &= ~NETIF_F_HW_VLAN_TX;
1837
1838         return features;
1839 }
1840
1841 static int igb_set_features(struct net_device *netdev,
1842         netdev_features_t features)
1843 {
1844         netdev_features_t changed = netdev->features ^ features;
1845         struct igb_adapter *adapter = netdev_priv(netdev);
1846
1847         if (changed & NETIF_F_HW_VLAN_RX)
1848                 igb_vlan_mode(netdev, features);
1849
1850         if (!(changed & NETIF_F_RXALL))
1851                 return 0;
1852
1853         netdev->features = features;
1854
1855         if (netif_running(netdev))
1856                 igb_reinit_locked(adapter);
1857         else
1858                 igb_reset(adapter);
1859
1860         return 0;
1861 }
1862
1863 static const struct net_device_ops igb_netdev_ops = {
1864         .ndo_open               = igb_open,
1865         .ndo_stop               = igb_close,
1866         .ndo_start_xmit         = igb_xmit_frame,
1867         .ndo_get_stats64        = igb_get_stats64,
1868         .ndo_set_rx_mode        = igb_set_rx_mode,
1869         .ndo_set_mac_address    = igb_set_mac,
1870         .ndo_change_mtu         = igb_change_mtu,
1871         .ndo_do_ioctl           = igb_ioctl,
1872         .ndo_tx_timeout         = igb_tx_timeout,
1873         .ndo_validate_addr      = eth_validate_addr,
1874         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
1875         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
1876         .ndo_set_vf_mac         = igb_ndo_set_vf_mac,
1877         .ndo_set_vf_vlan        = igb_ndo_set_vf_vlan,
1878         .ndo_set_vf_tx_rate     = igb_ndo_set_vf_bw,
1879         .ndo_get_vf_config      = igb_ndo_get_vf_config,
1880 #ifdef CONFIG_NET_POLL_CONTROLLER
1881         .ndo_poll_controller    = igb_netpoll,
1882 #endif
1883         .ndo_fix_features       = igb_fix_features,
1884         .ndo_set_features       = igb_set_features,
1885 };
1886
1887 /**
1888  * igb_set_fw_version - Configure version string for ethtool
1889  * @adapter: adapter struct
1890  *
1891  **/
1892 void igb_set_fw_version(struct igb_adapter *adapter)
1893 {
1894         struct e1000_hw *hw = &adapter->hw;
1895         struct e1000_fw_version fw;
1896
1897         igb_get_fw_version(hw, &fw);
1898
1899         switch (hw->mac.type) {
1900         case e1000_i211:
1901                 snprintf(adapter->fw_version, sizeof(adapter->fw_version),
1902                          "%2d.%2d-%d",
1903                          fw.invm_major, fw.invm_minor, fw.invm_img_type);
1904                 break;
1905
1906         default:
1907                 /* if option is rom valid, display its version too */
1908                 if (fw.or_valid) {
1909                         snprintf(adapter->fw_version,
1910                                  sizeof(adapter->fw_version),
1911                                  "%d.%d, 0x%08x, %d.%d.%d",
1912                                  fw.eep_major, fw.eep_minor, fw.etrack_id,
1913                                  fw.or_major, fw.or_build, fw.or_patch);
1914                 /* no option rom */
1915                 } else {
1916                         snprintf(adapter->fw_version,
1917                                  sizeof(adapter->fw_version),
1918                                  "%d.%d, 0x%08x",
1919                                  fw.eep_major, fw.eep_minor, fw.etrack_id);
1920                 }
1921                 break;
1922         }
1923         return;
1924 }
1925
1926 /*  igb_init_i2c - Init I2C interface
1927  *  @adapter: pointer to adapter structure
1928  *
1929  */
1930 static s32 igb_init_i2c(struct igb_adapter *adapter)
1931 {
1932         s32 status = E1000_SUCCESS;
1933
1934         /* I2C interface supported on i350 devices */
1935         if (adapter->hw.mac.type != e1000_i350)
1936                 return E1000_SUCCESS;
1937
1938         /* Initialize the i2c bus which is controlled by the registers.
1939          * This bus will use the i2c_algo_bit structue that implements
1940          * the protocol through toggling of the 4 bits in the register.
1941          */
1942         adapter->i2c_adap.owner = THIS_MODULE;
1943         adapter->i2c_algo = igb_i2c_algo;
1944         adapter->i2c_algo.data = adapter;
1945         adapter->i2c_adap.algo_data = &adapter->i2c_algo;
1946         adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
1947         strlcpy(adapter->i2c_adap.name, "igb BB",
1948                 sizeof(adapter->i2c_adap.name));
1949         status = i2c_bit_add_bus(&adapter->i2c_adap);
1950         return status;
1951 }
1952
1953 /**
1954  * igb_probe - Device Initialization Routine
1955  * @pdev: PCI device information struct
1956  * @ent: entry in igb_pci_tbl
1957  *
1958  * Returns 0 on success, negative on failure
1959  *
1960  * igb_probe initializes an adapter identified by a pci_dev structure.
1961  * The OS initialization, configuring of the adapter private structure,
1962  * and a hardware reset occur.
1963  **/
1964 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
1965 {
1966         struct net_device *netdev;
1967         struct igb_adapter *adapter;
1968         struct e1000_hw *hw;
1969         u16 eeprom_data = 0;
1970         s32 ret_val;
1971         static int global_quad_port_a; /* global quad port a indication */
1972         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1973         unsigned long mmio_start, mmio_len;
1974         int err, pci_using_dac;
1975         u8 part_str[E1000_PBANUM_LENGTH];
1976
1977         /* Catch broken hardware that put the wrong VF device ID in
1978          * the PCIe SR-IOV capability.
1979          */
1980         if (pdev->is_virtfn) {
1981                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
1982                         pci_name(pdev), pdev->vendor, pdev->device);
1983                 return -EINVAL;
1984         }
1985
1986         err = pci_enable_device_mem(pdev);
1987         if (err)
1988                 return err;
1989
1990         pci_using_dac = 0;
1991         err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
1992         if (!err) {
1993                 err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
1994                 if (!err)
1995                         pci_using_dac = 1;
1996         } else {
1997                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1998                 if (err) {
1999                         err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(32));
2000                         if (err) {
2001                                 dev_err(&pdev->dev, "No usable DMA "
2002                                         "configuration, aborting\n");
2003                                 goto err_dma;
2004                         }
2005                 }
2006         }
2007
2008         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2009                                            IORESOURCE_MEM),
2010                                            igb_driver_name);
2011         if (err)
2012                 goto err_pci_reg;
2013
2014         pci_enable_pcie_error_reporting(pdev);
2015
2016         pci_set_master(pdev);
2017         pci_save_state(pdev);
2018
2019         err = -ENOMEM;
2020         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2021                                    IGB_MAX_TX_QUEUES);
2022         if (!netdev)
2023                 goto err_alloc_etherdev;
2024
2025         SET_NETDEV_DEV(netdev, &pdev->dev);
2026
2027         pci_set_drvdata(pdev, netdev);
2028         adapter = netdev_priv(netdev);
2029         adapter->netdev = netdev;
2030         adapter->pdev = pdev;
2031         hw = &adapter->hw;
2032         hw->back = adapter;
2033         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2034
2035         mmio_start = pci_resource_start(pdev, 0);
2036         mmio_len = pci_resource_len(pdev, 0);
2037
2038         err = -EIO;
2039         hw->hw_addr = ioremap(mmio_start, mmio_len);
2040         if (!hw->hw_addr)
2041                 goto err_ioremap;
2042
2043         netdev->netdev_ops = &igb_netdev_ops;
2044         igb_set_ethtool_ops(netdev);
2045         netdev->watchdog_timeo = 5 * HZ;
2046
2047         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2048
2049         netdev->mem_start = mmio_start;
2050         netdev->mem_end = mmio_start + mmio_len;
2051
2052         /* PCI config space info */
2053         hw->vendor_id = pdev->vendor;
2054         hw->device_id = pdev->device;
2055         hw->revision_id = pdev->revision;
2056         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2057         hw->subsystem_device_id = pdev->subsystem_device;
2058
2059         /* Copy the default MAC, PHY and NVM function pointers */
2060         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2061         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2062         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2063         /* Initialize skew-specific constants */
2064         err = ei->get_invariants(hw);
2065         if (err)
2066                 goto err_sw_init;
2067
2068         /* setup the private structure */
2069         err = igb_sw_init(adapter);
2070         if (err)
2071                 goto err_sw_init;
2072
2073         igb_get_bus_info_pcie(hw);
2074
2075         hw->phy.autoneg_wait_to_complete = false;
2076
2077         /* Copper options */
2078         if (hw->phy.media_type == e1000_media_type_copper) {
2079                 hw->phy.mdix = AUTO_ALL_MODES;
2080                 hw->phy.disable_polarity_correction = false;
2081                 hw->phy.ms_type = e1000_ms_hw_default;
2082         }
2083
2084         if (igb_check_reset_block(hw))
2085                 dev_info(&pdev->dev,
2086                         "PHY reset is blocked due to SOL/IDER session.\n");
2087
2088         /*
2089          * features is initialized to 0 in allocation, it might have bits
2090          * set by igb_sw_init so we should use an or instead of an
2091          * assignment.
2092          */
2093         netdev->features |= NETIF_F_SG |
2094                             NETIF_F_IP_CSUM |
2095                             NETIF_F_IPV6_CSUM |
2096                             NETIF_F_TSO |
2097                             NETIF_F_TSO6 |
2098                             NETIF_F_RXHASH |
2099                             NETIF_F_RXCSUM |
2100                             NETIF_F_HW_VLAN_RX |
2101                             NETIF_F_HW_VLAN_TX;
2102
2103         /* copy netdev features into list of user selectable features */
2104         netdev->hw_features |= netdev->features;
2105         netdev->hw_features |= NETIF_F_RXALL;
2106
2107         /* set this bit last since it cannot be part of hw_features */
2108         netdev->features |= NETIF_F_HW_VLAN_FILTER;
2109
2110         netdev->vlan_features |= NETIF_F_TSO |
2111                                  NETIF_F_TSO6 |
2112                                  NETIF_F_IP_CSUM |
2113                                  NETIF_F_IPV6_CSUM |
2114                                  NETIF_F_SG;
2115
2116         netdev->priv_flags |= IFF_SUPP_NOFCS;
2117
2118         if (pci_using_dac) {
2119                 netdev->features |= NETIF_F_HIGHDMA;
2120                 netdev->vlan_features |= NETIF_F_HIGHDMA;
2121         }
2122
2123         if (hw->mac.type >= e1000_82576) {
2124                 netdev->hw_features |= NETIF_F_SCTP_CSUM;
2125                 netdev->features |= NETIF_F_SCTP_CSUM;
2126         }
2127
2128         netdev->priv_flags |= IFF_UNICAST_FLT;
2129
2130         adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2131
2132         /* before reading the NVM, reset the controller to put the device in a
2133          * known good starting state */
2134         hw->mac.ops.reset_hw(hw);
2135
2136         /*
2137          * make sure the NVM is good , i211 parts have special NVM that
2138          * doesn't contain a checksum
2139          */
2140         if (hw->mac.type != e1000_i211) {
2141                 if (hw->nvm.ops.validate(hw) < 0) {
2142                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2143                         err = -EIO;
2144                         goto err_eeprom;
2145                 }
2146         }
2147
2148         /* copy the MAC address out of the NVM */
2149         if (hw->mac.ops.read_mac_addr(hw))
2150                 dev_err(&pdev->dev, "NVM Read Error\n");
2151
2152         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2153
2154         if (!is_valid_ether_addr(netdev->dev_addr)) {
2155                 dev_err(&pdev->dev, "Invalid MAC Address\n");
2156                 err = -EIO;
2157                 goto err_eeprom;
2158         }
2159
2160         /* get firmware version for ethtool -i */
2161         igb_set_fw_version(adapter);
2162
2163         setup_timer(&adapter->watchdog_timer, igb_watchdog,
2164                     (unsigned long) adapter);
2165         setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2166                     (unsigned long) adapter);
2167
2168         INIT_WORK(&adapter->reset_task, igb_reset_task);
2169         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2170
2171         /* Initialize link properties that are user-changeable */
2172         adapter->fc_autoneg = true;
2173         hw->mac.autoneg = true;
2174         hw->phy.autoneg_advertised = 0x2f;
2175
2176         hw->fc.requested_mode = e1000_fc_default;
2177         hw->fc.current_mode = e1000_fc_default;
2178
2179         igb_validate_mdi_setting(hw);
2180
2181         /* By default, support wake on port A */
2182         if (hw->bus.func == 0)
2183                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2184
2185         /* Check the NVM for wake support on non-port A ports */
2186         if (hw->mac.type >= e1000_82580)
2187                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2188                                  NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2189                                  &eeprom_data);
2190         else if (hw->bus.func == 1)
2191                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2192
2193         if (eeprom_data & IGB_EEPROM_APME)
2194                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2195
2196         /* now that we have the eeprom settings, apply the special cases where
2197          * the eeprom may be wrong or the board simply won't support wake on
2198          * lan on a particular port */
2199         switch (pdev->device) {
2200         case E1000_DEV_ID_82575GB_QUAD_COPPER:
2201                 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2202                 break;
2203         case E1000_DEV_ID_82575EB_FIBER_SERDES:
2204         case E1000_DEV_ID_82576_FIBER:
2205         case E1000_DEV_ID_82576_SERDES:
2206                 /* Wake events only supported on port A for dual fiber
2207                  * regardless of eeprom setting */
2208                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2209                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2210                 break;
2211         case E1000_DEV_ID_82576_QUAD_COPPER:
2212         case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2213                 /* if quad port adapter, disable WoL on all but port A */
2214                 if (global_quad_port_a != 0)
2215                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2216                 else
2217                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2218                 /* Reset for multiple quad port adapters */
2219                 if (++global_quad_port_a == 4)
2220                         global_quad_port_a = 0;
2221                 break;
2222         default:
2223                 /* If the device can't wake, don't set software support */
2224                 if (!device_can_wakeup(&adapter->pdev->dev))
2225                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2226         }
2227
2228         /* initialize the wol settings based on the eeprom settings */
2229         if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2230                 adapter->wol |= E1000_WUFC_MAG;
2231
2232         /* Some vendors want WoL disabled by default, but still supported */
2233         if ((hw->mac.type == e1000_i350) &&
2234             (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2235                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2236                 adapter->wol = 0;
2237         }
2238
2239         device_set_wakeup_enable(&adapter->pdev->dev,
2240                                  adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2241
2242         /* reset the hardware with the new settings */
2243         igb_reset(adapter);
2244
2245         /* Init the I2C interface */
2246         err = igb_init_i2c(adapter);
2247         if (err) {
2248                 dev_err(&pdev->dev, "failed to init i2c interface\n");
2249                 goto err_eeprom;
2250         }
2251
2252         /* let the f/w know that the h/w is now under the control of the
2253          * driver. */
2254         igb_get_hw_control(adapter);
2255
2256         strcpy(netdev->name, "eth%d");
2257         err = register_netdev(netdev);
2258         if (err)
2259                 goto err_register;
2260
2261         /* carrier off reporting is important to ethtool even BEFORE open */
2262         netif_carrier_off(netdev);
2263
2264 #ifdef CONFIG_IGB_DCA
2265         if (dca_add_requester(&pdev->dev) == 0) {
2266                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2267                 dev_info(&pdev->dev, "DCA enabled\n");
2268                 igb_setup_dca(adapter);
2269         }
2270
2271 #endif
2272 #ifdef CONFIG_IGB_HWMON
2273         /* Initialize the thermal sensor on i350 devices. */
2274         if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2275                 u16 ets_word;
2276
2277                 /*
2278                  * Read the NVM to determine if this i350 device supports an
2279                  * external thermal sensor.
2280                  */
2281                 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2282                 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2283                         adapter->ets = true;
2284                 else
2285                         adapter->ets = false;
2286                 if (igb_sysfs_init(adapter))
2287                         dev_err(&pdev->dev,
2288                                 "failed to allocate sysfs resources\n");
2289         } else {
2290                 adapter->ets = false;
2291         }
2292 #endif
2293         /* do hw tstamp init after resetting */
2294         igb_ptp_init(adapter);
2295
2296         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2297         /* print bus type/speed/width info */
2298         dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2299                  netdev->name,
2300                  ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2301                   (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2302                                                             "unknown"),
2303                  ((hw->bus.width == e1000_bus_width_pcie_x4) ? "Width x4" :
2304                   (hw->bus.width == e1000_bus_width_pcie_x2) ? "Width x2" :
2305                   (hw->bus.width == e1000_bus_width_pcie_x1) ? "Width x1" :
2306                    "unknown"),
2307                  netdev->dev_addr);
2308
2309         ret_val = igb_read_part_string(hw, part_str, E1000_PBANUM_LENGTH);
2310         if (ret_val)
2311                 strcpy(part_str, "Unknown");
2312         dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2313         dev_info(&pdev->dev,
2314                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2315                 adapter->msix_entries ? "MSI-X" :
2316                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2317                 adapter->num_rx_queues, adapter->num_tx_queues);
2318         switch (hw->mac.type) {
2319         case e1000_i350:
2320         case e1000_i210:
2321         case e1000_i211:
2322                 igb_set_eee_i350(hw);
2323                 break;
2324         default:
2325                 break;
2326         }
2327
2328         pm_runtime_put_noidle(&pdev->dev);
2329         return 0;
2330
2331 err_register:
2332         igb_release_hw_control(adapter);
2333         memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2334 err_eeprom:
2335         if (!igb_check_reset_block(hw))
2336                 igb_reset_phy(hw);
2337
2338         if (hw->flash_address)
2339                 iounmap(hw->flash_address);
2340 err_sw_init:
2341         igb_clear_interrupt_scheme(adapter);
2342         iounmap(hw->hw_addr);
2343 err_ioremap:
2344         free_netdev(netdev);
2345 err_alloc_etherdev:
2346         pci_release_selected_regions(pdev,
2347                                      pci_select_bars(pdev, IORESOURCE_MEM));
2348 err_pci_reg:
2349 err_dma:
2350         pci_disable_device(pdev);
2351         return err;
2352 }
2353
2354 #ifdef CONFIG_PCI_IOV
2355 static int  igb_disable_sriov(struct pci_dev *pdev)
2356 {
2357         struct net_device *netdev = pci_get_drvdata(pdev);
2358         struct igb_adapter *adapter = netdev_priv(netdev);
2359         struct e1000_hw *hw = &adapter->hw;
2360
2361         /* reclaim resources allocated to VFs */
2362         if (adapter->vf_data) {
2363                 /* disable iov and allow time for transactions to clear */
2364                 if (igb_vfs_are_assigned(adapter)) {
2365                         dev_warn(&pdev->dev,
2366                                  "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2367                         return -EPERM;
2368                 } else {
2369                         pci_disable_sriov(pdev);
2370                         msleep(500);
2371                 }
2372
2373                 kfree(adapter->vf_data);
2374                 adapter->vf_data = NULL;
2375                 adapter->vfs_allocated_count = 0;
2376                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2377                 wrfl();
2378                 msleep(100);
2379                 dev_info(&pdev->dev, "IOV Disabled\n");
2380
2381                 /* Re-enable DMA Coalescing flag since IOV is turned off */
2382                 adapter->flags |= IGB_FLAG_DMAC;
2383         }
2384
2385         return 0;
2386 }
2387
2388 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2389 {
2390         struct net_device *netdev = pci_get_drvdata(pdev);
2391         struct igb_adapter *adapter = netdev_priv(netdev);
2392         int old_vfs = pci_num_vf(pdev);
2393         int err = 0;
2394         int i;
2395
2396         if (!num_vfs)
2397                 goto out;
2398         else if (old_vfs && old_vfs == num_vfs)
2399                 goto out;
2400         else if (old_vfs && old_vfs != num_vfs)
2401                 err = igb_disable_sriov(pdev);
2402
2403         if (err)
2404                 goto out;
2405
2406         if (num_vfs > 7) {
2407                 err = -EPERM;
2408                 goto out;
2409         }
2410
2411         adapter->vfs_allocated_count = num_vfs;
2412
2413         adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2414                                 sizeof(struct vf_data_storage), GFP_KERNEL);
2415
2416         /* if allocation failed then we do not support SR-IOV */
2417         if (!adapter->vf_data) {
2418                 adapter->vfs_allocated_count = 0;
2419                 dev_err(&pdev->dev,
2420                         "Unable to allocate memory for VF Data Storage\n");
2421                 err = -ENOMEM;
2422                 goto out;
2423         }
2424
2425         err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2426         if (err)
2427                 goto err_out;
2428
2429         dev_info(&pdev->dev, "%d VFs allocated\n",
2430                  adapter->vfs_allocated_count);
2431         for (i = 0; i < adapter->vfs_allocated_count; i++)
2432                 igb_vf_configure(adapter, i);
2433
2434         /* DMA Coalescing is not supported in IOV mode. */
2435         adapter->flags &= ~IGB_FLAG_DMAC;
2436         goto out;
2437
2438 err_out:
2439         kfree(adapter->vf_data);
2440         adapter->vf_data = NULL;
2441         adapter->vfs_allocated_count = 0;
2442 out:
2443         return err;
2444 }
2445
2446 #endif
2447 /*
2448  *  igb_remove_i2c - Cleanup  I2C interface
2449  *  @adapter: pointer to adapter structure
2450  *
2451  */
2452 static void igb_remove_i2c(struct igb_adapter *adapter)
2453 {
2454
2455         /* free the adapter bus structure */
2456         i2c_del_adapter(&adapter->i2c_adap);
2457 }
2458
2459 /**
2460  * igb_remove - Device Removal Routine
2461  * @pdev: PCI device information struct
2462  *
2463  * igb_remove is called by the PCI subsystem to alert the driver
2464  * that it should release a PCI device.  The could be caused by a
2465  * Hot-Plug event, or because the driver is going to be removed from
2466  * memory.
2467  **/
2468 static void igb_remove(struct pci_dev *pdev)
2469 {
2470         struct net_device *netdev = pci_get_drvdata(pdev);
2471         struct igb_adapter *adapter = netdev_priv(netdev);
2472         struct e1000_hw *hw = &adapter->hw;
2473
2474         pm_runtime_get_noresume(&pdev->dev);
2475 #ifdef CONFIG_IGB_HWMON
2476         igb_sysfs_exit(adapter);
2477 #endif
2478         igb_remove_i2c(adapter);
2479         igb_ptp_stop(adapter);
2480         /*
2481          * The watchdog timer may be rescheduled, so explicitly
2482          * disable watchdog from being rescheduled.
2483          */
2484         set_bit(__IGB_DOWN, &adapter->state);
2485         del_timer_sync(&adapter->watchdog_timer);
2486         del_timer_sync(&adapter->phy_info_timer);
2487
2488         cancel_work_sync(&adapter->reset_task);
2489         cancel_work_sync(&adapter->watchdog_task);
2490
2491 #ifdef CONFIG_IGB_DCA
2492         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2493                 dev_info(&pdev->dev, "DCA disabled\n");
2494                 dca_remove_requester(&pdev->dev);
2495                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2496                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2497         }
2498 #endif
2499
2500         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
2501          * would have already happened in close and is redundant. */
2502         igb_release_hw_control(adapter);
2503
2504         unregister_netdev(netdev);
2505
2506         igb_clear_interrupt_scheme(adapter);
2507
2508 #ifdef CONFIG_PCI_IOV
2509         igb_disable_sriov(pdev);
2510 #endif
2511
2512         iounmap(hw->hw_addr);
2513         if (hw->flash_address)
2514                 iounmap(hw->flash_address);
2515         pci_release_selected_regions(pdev,
2516                                      pci_select_bars(pdev, IORESOURCE_MEM));
2517
2518         kfree(adapter->shadow_vfta);
2519         free_netdev(netdev);
2520
2521         pci_disable_pcie_error_reporting(pdev);
2522
2523         pci_disable_device(pdev);
2524 }
2525
2526 /**
2527  * igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2528  * @adapter: board private structure to initialize
2529  *
2530  * This function initializes the vf specific data storage and then attempts to
2531  * allocate the VFs.  The reason for ordering it this way is because it is much
2532  * mor expensive time wise to disable SR-IOV than it is to allocate and free
2533  * the memory for the VFs.
2534  **/
2535 static void igb_probe_vfs(struct igb_adapter *adapter)
2536 {
2537 #ifdef CONFIG_PCI_IOV
2538         struct pci_dev *pdev = adapter->pdev;
2539         struct e1000_hw *hw = &adapter->hw;
2540
2541         /* Virtualization features not supported on i210 family. */
2542         if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2543                 return;
2544
2545         pci_sriov_set_totalvfs(pdev, 7);
2546         igb_enable_sriov(pdev, max_vfs);
2547
2548 #endif /* CONFIG_PCI_IOV */
2549 }
2550
2551 static void igb_init_queue_configuration(struct igb_adapter *adapter)
2552 {
2553         struct e1000_hw *hw = &adapter->hw;
2554         u32 max_rss_queues;
2555
2556         /* Determine the maximum number of RSS queues supported. */
2557         switch (hw->mac.type) {
2558         case e1000_i211:
2559                 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2560                 break;
2561         case e1000_82575:
2562         case e1000_i210:
2563                 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2564                 break;
2565         case e1000_i350:
2566                 /* I350 cannot do RSS and SR-IOV at the same time */
2567                 if (!!adapter->vfs_allocated_count) {
2568                         max_rss_queues = 1;
2569                         break;
2570                 }
2571                 /* fall through */
2572         case e1000_82576:
2573                 if (!!adapter->vfs_allocated_count) {
2574                         max_rss_queues = 2;
2575                         break;
2576                 }
2577                 /* fall through */
2578         case e1000_82580:
2579         default:
2580                 max_rss_queues = IGB_MAX_RX_QUEUES;
2581                 break;
2582         }
2583
2584         adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2585
2586         /* Determine if we need to pair queues. */
2587         switch (hw->mac.type) {
2588         case e1000_82575:
2589         case e1000_i211:
2590                 /* Device supports enough interrupts without queue pairing. */
2591                 break;
2592         case e1000_82576:
2593                 /*
2594                  * If VFs are going to be allocated with RSS queues then we
2595                  * should pair the queues in order to conserve interrupts due
2596                  * to limited supply.
2597                  */
2598                 if ((adapter->rss_queues > 1) &&
2599                     (adapter->vfs_allocated_count > 6))
2600                         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2601                 /* fall through */
2602         case e1000_82580:
2603         case e1000_i350:
2604         case e1000_i210:
2605         default:
2606                 /*
2607                  * If rss_queues > half of max_rss_queues, pair the queues in
2608                  * order to conserve interrupts due to limited supply.
2609                  */
2610                 if (adapter->rss_queues > (max_rss_queues / 2))
2611                         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2612                 break;
2613         }
2614 }
2615
2616 /**
2617  * igb_sw_init - Initialize general software structures (struct igb_adapter)
2618  * @adapter: board private structure to initialize
2619  *
2620  * igb_sw_init initializes the Adapter private data structure.
2621  * Fields are initialized based on PCI device information and
2622  * OS network device settings (MTU size).
2623  **/
2624 static int igb_sw_init(struct igb_adapter *adapter)
2625 {
2626         struct e1000_hw *hw = &adapter->hw;
2627         struct net_device *netdev = adapter->netdev;
2628         struct pci_dev *pdev = adapter->pdev;
2629
2630         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2631
2632         /* set default ring sizes */
2633         adapter->tx_ring_count = IGB_DEFAULT_TXD;
2634         adapter->rx_ring_count = IGB_DEFAULT_RXD;
2635
2636         /* set default ITR values */
2637         adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2638         adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2639
2640         /* set default work limits */
2641         adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2642
2643         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2644                                   VLAN_HLEN;
2645         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2646
2647         spin_lock_init(&adapter->stats64_lock);
2648 #ifdef CONFIG_PCI_IOV
2649         switch (hw->mac.type) {
2650         case e1000_82576:
2651         case e1000_i350:
2652                 if (max_vfs > 7) {
2653                         dev_warn(&pdev->dev,
2654                                  "Maximum of 7 VFs per PF, using max\n");
2655                         max_vfs = adapter->vfs_allocated_count = 7;
2656                 } else
2657                         adapter->vfs_allocated_count = max_vfs;
2658                 if (adapter->vfs_allocated_count)
2659                         dev_warn(&pdev->dev,
2660                                  "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2661                 break;
2662         default:
2663                 break;
2664         }
2665 #endif /* CONFIG_PCI_IOV */
2666
2667         igb_init_queue_configuration(adapter);
2668
2669         /* Setup and initialize a copy of the hw vlan table array */
2670         adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2671                                        GFP_ATOMIC);
2672
2673         /* This call may decrease the number of queues */
2674         if (igb_init_interrupt_scheme(adapter, true)) {
2675                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2676                 return -ENOMEM;
2677         }
2678
2679         igb_probe_vfs(adapter);
2680
2681         /* Explicitly disable IRQ since the NIC can be in any state. */
2682         igb_irq_disable(adapter);
2683
2684         if (hw->mac.type >= e1000_i350)
2685                 adapter->flags &= ~IGB_FLAG_DMAC;
2686
2687         set_bit(__IGB_DOWN, &adapter->state);
2688         return 0;
2689 }
2690
2691 /**
2692  * igb_open - Called when a network interface is made active
2693  * @netdev: network interface device structure
2694  *
2695  * Returns 0 on success, negative value on failure
2696  *
2697  * The open entry point is called when a network interface is made
2698  * active by the system (IFF_UP).  At this point all resources needed
2699  * for transmit and receive operations are allocated, the interrupt
2700  * handler is registered with the OS, the watchdog timer is started,
2701  * and the stack is notified that the interface is ready.
2702  **/
2703 static int __igb_open(struct net_device *netdev, bool resuming)
2704 {
2705         struct igb_adapter *adapter = netdev_priv(netdev);
2706         struct e1000_hw *hw = &adapter->hw;
2707         struct pci_dev *pdev = adapter->pdev;
2708         int err;
2709         int i;
2710
2711         /* disallow open during test */
2712         if (test_bit(__IGB_TESTING, &adapter->state)) {
2713                 WARN_ON(resuming);
2714                 return -EBUSY;
2715         }
2716
2717         if (!resuming)
2718                 pm_runtime_get_sync(&pdev->dev);
2719
2720         netif_carrier_off(netdev);
2721
2722         /* allocate transmit descriptors */
2723         err = igb_setup_all_tx_resources(adapter);
2724         if (err)
2725                 goto err_setup_tx;
2726
2727         /* allocate receive descriptors */
2728         err = igb_setup_all_rx_resources(adapter);
2729         if (err)
2730                 goto err_setup_rx;
2731
2732         igb_power_up_link(adapter);
2733
2734         /* before we allocate an interrupt, we must be ready to handle it.
2735          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2736          * as soon as we call pci_request_irq, so we have to setup our
2737          * clean_rx handler before we do so.  */
2738         igb_configure(adapter);
2739
2740         err = igb_request_irq(adapter);
2741         if (err)
2742                 goto err_req_irq;
2743
2744         /* Notify the stack of the actual queue counts. */
2745         err = netif_set_real_num_tx_queues(adapter->netdev,
2746                                            adapter->num_tx_queues);
2747         if (err)
2748                 goto err_set_queues;
2749
2750         err = netif_set_real_num_rx_queues(adapter->netdev,
2751                                            adapter->num_rx_queues);
2752         if (err)
2753                 goto err_set_queues;
2754
2755         /* From here on the code is the same as igb_up() */
2756         clear_bit(__IGB_DOWN, &adapter->state);
2757
2758         for (i = 0; i < adapter->num_q_vectors; i++)
2759                 napi_enable(&(adapter->q_vector[i]->napi));
2760
2761         /* Clear any pending interrupts. */
2762         rd32(E1000_ICR);
2763
2764         igb_irq_enable(adapter);
2765
2766         /* notify VFs that reset has been completed */
2767         if (adapter->vfs_allocated_count) {
2768                 u32 reg_data = rd32(E1000_CTRL_EXT);
2769                 reg_data |= E1000_CTRL_EXT_PFRSTD;
2770                 wr32(E1000_CTRL_EXT, reg_data);
2771         }
2772
2773         netif_tx_start_all_queues(netdev);
2774
2775         if (!resuming)
2776                 pm_runtime_put(&pdev->dev);
2777
2778         /* start the watchdog. */
2779         hw->mac.get_link_status = 1;
2780         schedule_work(&adapter->watchdog_task);
2781
2782         return 0;
2783
2784 err_set_queues:
2785         igb_free_irq(adapter);
2786 err_req_irq:
2787         igb_release_hw_control(adapter);
2788         igb_power_down_link(adapter);
2789         igb_free_all_rx_resources(adapter);
2790 err_setup_rx:
2791         igb_free_all_tx_resources(adapter);
2792 err_setup_tx:
2793         igb_reset(adapter);
2794         if (!resuming)
2795                 pm_runtime_put(&pdev->dev);
2796
2797         return err;
2798 }
2799
2800 static int igb_open(struct net_device *netdev)
2801 {
2802         return __igb_open(netdev, false);
2803 }
2804
2805 /**
2806  * igb_close - Disables a network interface
2807  * @netdev: network interface device structure
2808  *
2809  * Returns 0, this is not allowed to fail
2810  *
2811  * The close entry point is called when an interface is de-activated
2812  * by the OS.  The hardware is still under the driver's control, but
2813  * needs to be disabled.  A global MAC reset is issued to stop the
2814  * hardware, and all transmit and receive resources are freed.
2815  **/
2816 static int __igb_close(struct net_device *netdev, bool suspending)
2817 {
2818         struct igb_adapter *adapter = netdev_priv(netdev);
2819         struct pci_dev *pdev = adapter->pdev;
2820
2821         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2822
2823         if (!suspending)
2824                 pm_runtime_get_sync(&pdev->dev);
2825
2826         igb_down(adapter);
2827         igb_free_irq(adapter);
2828
2829         igb_free_all_tx_resources(adapter);
2830         igb_free_all_rx_resources(adapter);
2831
2832         if (!suspending)
2833                 pm_runtime_put_sync(&pdev->dev);
2834         return 0;
2835 }
2836
2837 static int igb_close(struct net_device *netdev)
2838 {
2839         return __igb_close(netdev, false);
2840 }
2841
2842 /**
2843  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
2844  * @tx_ring: tx descriptor ring (for a specific queue) to setup
2845  *
2846  * Return 0 on success, negative on failure
2847  **/
2848 int igb_setup_tx_resources(struct igb_ring *tx_ring)
2849 {
2850         struct device *dev = tx_ring->dev;
2851         int size;
2852
2853         size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2854
2855         tx_ring->tx_buffer_info = vzalloc(size);
2856         if (!tx_ring->tx_buffer_info)
2857                 goto err;
2858
2859         /* round up to nearest 4K */
2860         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2861         tx_ring->size = ALIGN(tx_ring->size, 4096);
2862
2863         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2864                                            &tx_ring->dma, GFP_KERNEL);
2865         if (!tx_ring->desc)
2866                 goto err;
2867
2868         tx_ring->next_to_use = 0;
2869         tx_ring->next_to_clean = 0;
2870
2871         return 0;
2872
2873 err:
2874         vfree(tx_ring->tx_buffer_info);
2875         tx_ring->tx_buffer_info = NULL;
2876         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2877         return -ENOMEM;
2878 }
2879
2880 /**
2881  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
2882  *                                (Descriptors) for all queues
2883  * @adapter: board private structure
2884  *
2885  * Return 0 on success, negative on failure
2886  **/
2887 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2888 {
2889         struct pci_dev *pdev = adapter->pdev;
2890         int i, err = 0;
2891
2892         for (i = 0; i < adapter->num_tx_queues; i++) {
2893                 err = igb_setup_tx_resources(adapter->tx_ring[i]);
2894                 if (err) {
2895                         dev_err(&pdev->dev,
2896                                 "Allocation for Tx Queue %u failed\n", i);
2897                         for (i--; i >= 0; i--)
2898                                 igb_free_tx_resources(adapter->tx_ring[i]);
2899                         break;
2900                 }
2901         }
2902
2903         return err;
2904 }
2905
2906 /**
2907  * igb_setup_tctl - configure the transmit control registers
2908  * @adapter: Board private structure
2909  **/
2910 void igb_setup_tctl(struct igb_adapter *adapter)
2911 {
2912         struct e1000_hw *hw = &adapter->hw;
2913         u32 tctl;
2914
2915         /* disable queue 0 which is enabled by default on 82575 and 82576 */
2916         wr32(E1000_TXDCTL(0), 0);
2917
2918         /* Program the Transmit Control Register */
2919         tctl = rd32(E1000_TCTL);
2920         tctl &= ~E1000_TCTL_CT;
2921         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
2922                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2923
2924         igb_config_collision_dist(hw);
2925
2926         /* Enable transmits */
2927         tctl |= E1000_TCTL_EN;
2928
2929         wr32(E1000_TCTL, tctl);
2930 }
2931
2932 /**
2933  * igb_configure_tx_ring - Configure transmit ring after Reset
2934  * @adapter: board private structure
2935  * @ring: tx ring to configure
2936  *
2937  * Configure a transmit ring after a reset.
2938  **/
2939 void igb_configure_tx_ring(struct igb_adapter *adapter,
2940                            struct igb_ring *ring)
2941 {
2942         struct e1000_hw *hw = &adapter->hw;
2943         u32 txdctl = 0;
2944         u64 tdba = ring->dma;
2945         int reg_idx = ring->reg_idx;
2946
2947         /* disable the queue */
2948         wr32(E1000_TXDCTL(reg_idx), 0);
2949         wrfl();
2950         mdelay(10);
2951
2952         wr32(E1000_TDLEN(reg_idx),
2953                         ring->count * sizeof(union e1000_adv_tx_desc));
2954         wr32(E1000_TDBAL(reg_idx),
2955                         tdba & 0x00000000ffffffffULL);
2956         wr32(E1000_TDBAH(reg_idx), tdba >> 32);
2957
2958         ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
2959         wr32(E1000_TDH(reg_idx), 0);
2960         writel(0, ring->tail);
2961
2962         txdctl |= IGB_TX_PTHRESH;
2963         txdctl |= IGB_TX_HTHRESH << 8;
2964         txdctl |= IGB_TX_WTHRESH << 16;
2965
2966         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
2967         wr32(E1000_TXDCTL(reg_idx), txdctl);
2968 }
2969
2970 /**
2971  * igb_configure_tx - Configure transmit Unit after Reset
2972  * @adapter: board private structure
2973  *
2974  * Configure the Tx unit of the MAC after a reset.
2975  **/
2976 static void igb_configure_tx(struct igb_adapter *adapter)
2977 {
2978         int i;
2979
2980         for (i = 0; i < adapter->num_tx_queues; i++)
2981                 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
2982 }
2983
2984 /**
2985  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
2986  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2987  *
2988  * Returns 0 on success, negative on failure
2989  **/
2990 int igb_setup_rx_resources(struct igb_ring *rx_ring)
2991 {
2992         struct device *dev = rx_ring->dev;
2993         int size;
2994
2995         size = sizeof(struct igb_rx_buffer) * rx_ring->count;
2996
2997         rx_ring->rx_buffer_info = vzalloc(size);
2998         if (!rx_ring->rx_buffer_info)
2999                 goto err;
3000
3001         /* Round up to nearest 4K */
3002         rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3003         rx_ring->size = ALIGN(rx_ring->size, 4096);
3004
3005         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3006                                            &rx_ring->dma, GFP_KERNEL);
3007         if (!rx_ring->desc)
3008                 goto err;
3009
3010         rx_ring->next_to_alloc = 0;
3011         rx_ring->next_to_clean = 0;
3012         rx_ring->next_to_use = 0;
3013
3014         return 0;
3015
3016 err:
3017         vfree(rx_ring->rx_buffer_info);
3018         rx_ring->rx_buffer_info = NULL;
3019         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3020         return -ENOMEM;
3021 }
3022
3023 /**
3024  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
3025  *                                (Descriptors) for all queues
3026  * @adapter: board private structure
3027  *
3028  * Return 0 on success, negative on failure
3029  **/
3030 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3031 {
3032         struct pci_dev *pdev = adapter->pdev;
3033         int i, err = 0;
3034
3035         for (i = 0; i < adapter->num_rx_queues; i++) {
3036                 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3037                 if (err) {
3038                         dev_err(&pdev->dev,
3039                                 "Allocation for Rx Queue %u failed\n", i);
3040                         for (i--; i >= 0; i--)
3041                                 igb_free_rx_resources(adapter->rx_ring[i]);
3042                         break;
3043                 }
3044         }
3045
3046         return err;
3047 }
3048
3049 /**
3050  * igb_setup_mrqc - configure the multiple receive queue control registers
3051  * @adapter: Board private structure
3052  **/
3053 static void igb_setup_mrqc(struct igb_adapter *adapter)
3054 {
3055         struct e1000_hw *hw = &adapter->hw;
3056         u32 mrqc, rxcsum;
3057         u32 j, num_rx_queues, shift = 0;
3058         static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3059                                         0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3060                                         0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3061                                         0xFA01ACBE };
3062
3063         /* Fill out hash function seeds */
3064         for (j = 0; j < 10; j++)
3065                 wr32(E1000_RSSRK(j), rsskey[j]);
3066
3067         num_rx_queues = adapter->rss_queues;
3068
3069         switch (hw->mac.type) {
3070         case e1000_82575:
3071                 shift = 6;
3072                 break;
3073         case e1000_82576:
3074                 /* 82576 supports 2 RSS queues for SR-IOV */
3075                 if (adapter->vfs_allocated_count) {
3076                         shift = 3;
3077                         num_rx_queues = 2;
3078                 }
3079                 break;
3080         default:
3081                 break;
3082         }
3083
3084         /*
3085          * Populate the indirection table 4 entries at a time.  To do this
3086          * we are generating the results for n and n+2 and then interleaving
3087          * those with the results with n+1 and n+3.
3088          */
3089         for (j = 0; j < 32; j++) {
3090                 /* first pass generates n and n+2 */
3091                 u32 base = ((j * 0x00040004) + 0x00020000) * num_rx_queues;
3092                 u32 reta = (base & 0x07800780) >> (7 - shift);
3093
3094                 /* second pass generates n+1 and n+3 */
3095                 base += 0x00010001 * num_rx_queues;
3096                 reta |= (base & 0x07800780) << (1 + shift);
3097
3098                 wr32(E1000_RETA(j), reta);
3099         }
3100
3101         /*
3102          * Disable raw packet checksumming so that RSS hash is placed in
3103          * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3104          * offloads as they are enabled by default
3105          */
3106         rxcsum = rd32(E1000_RXCSUM);
3107         rxcsum |= E1000_RXCSUM_PCSD;
3108
3109         if (adapter->hw.mac.type >= e1000_82576)
3110                 /* Enable Receive Checksum Offload for SCTP */
3111                 rxcsum |= E1000_RXCSUM_CRCOFL;
3112
3113         /* Don't need to set TUOFL or IPOFL, they default to 1 */
3114         wr32(E1000_RXCSUM, rxcsum);
3115
3116         /* Generate RSS hash based on packet types, TCP/UDP
3117          * port numbers and/or IPv4/v6 src and dst addresses
3118          */
3119         mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3120                E1000_MRQC_RSS_FIELD_IPV4_TCP |
3121                E1000_MRQC_RSS_FIELD_IPV6 |
3122                E1000_MRQC_RSS_FIELD_IPV6_TCP |
3123                E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3124
3125         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3126                 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3127         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3128                 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3129
3130         /* If VMDq is enabled then we set the appropriate mode for that, else
3131          * we default to RSS so that an RSS hash is calculated per packet even
3132          * if we are only using one queue */
3133         if (adapter->vfs_allocated_count) {
3134                 if (hw->mac.type > e1000_82575) {
3135                         /* Set the default pool for the PF's first queue */
3136                         u32 vtctl = rd32(E1000_VT_CTL);
3137                         vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3138                                    E1000_VT_CTL_DISABLE_DEF_POOL);
3139                         vtctl |= adapter->vfs_allocated_count <<
3140                                 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3141                         wr32(E1000_VT_CTL, vtctl);
3142                 }
3143                 if (adapter->rss_queues > 1)
3144                         mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3145                 else
3146                         mrqc |= E1000_MRQC_ENABLE_VMDQ;
3147         } else {
3148                 if (hw->mac.type != e1000_i211)
3149                         mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3150         }
3151         igb_vmm_control(adapter);
3152
3153         wr32(E1000_MRQC, mrqc);
3154 }
3155
3156 /**
3157  * igb_setup_rctl - configure the receive control registers
3158  * @adapter: Board private structure
3159  **/
3160 void igb_setup_rctl(struct igb_adapter *adapter)
3161 {
3162         struct e1000_hw *hw = &adapter->hw;
3163         u32 rctl;
3164
3165         rctl = rd32(E1000_RCTL);
3166
3167         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3168         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3169
3170         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3171                 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3172
3173         /*
3174          * enable stripping of CRC. It's unlikely this will break BMC
3175          * redirection as it did with e1000. Newer features require
3176          * that the HW strips the CRC.
3177          */
3178         rctl |= E1000_RCTL_SECRC;
3179
3180         /* disable store bad packets and clear size bits. */
3181         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3182
3183         /* enable LPE to prevent packets larger than max_frame_size */
3184         rctl |= E1000_RCTL_LPE;
3185
3186         /* disable queue 0 to prevent tail write w/o re-config */
3187         wr32(E1000_RXDCTL(0), 0);
3188
3189         /* Attention!!!  For SR-IOV PF driver operations you must enable
3190          * queue drop for all VF and PF queues to prevent head of line blocking
3191          * if an un-trusted VF does not provide descriptors to hardware.
3192          */
3193         if (adapter->vfs_allocated_count) {
3194                 /* set all queue drop enable bits */
3195                 wr32(E1000_QDE, ALL_QUEUES);
3196         }
3197
3198         /* This is useful for sniffing bad packets. */
3199         if (adapter->netdev->features & NETIF_F_RXALL) {
3200                 /* UPE and MPE will be handled by normal PROMISC logic
3201                  * in e1000e_set_rx_mode */
3202                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3203                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
3204                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3205
3206                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3207                           E1000_RCTL_DPF | /* Allow filtered pause */
3208                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3209                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3210                  * and that breaks VLANs.
3211                  */
3212         }
3213
3214         wr32(E1000_RCTL, rctl);
3215 }
3216
3217 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3218                                    int vfn)
3219 {
3220         struct e1000_hw *hw = &adapter->hw;
3221         u32 vmolr;
3222
3223         /* if it isn't the PF check to see if VFs are enabled and
3224          * increase the size to support vlan tags */
3225         if (vfn < adapter->vfs_allocated_count &&
3226             adapter->vf_data[vfn].vlans_enabled)
3227                 size += VLAN_TAG_SIZE;
3228
3229         vmolr = rd32(E1000_VMOLR(vfn));
3230         vmolr &= ~E1000_VMOLR_RLPML_MASK;
3231         vmolr |= size | E1000_VMOLR_LPE;
3232         wr32(E1000_VMOLR(vfn), vmolr);
3233
3234         return 0;
3235 }
3236
3237 /**
3238  * igb_rlpml_set - set maximum receive packet size
3239  * @adapter: board private structure
3240  *
3241  * Configure maximum receivable packet size.
3242  **/
3243 static void igb_rlpml_set(struct igb_adapter *adapter)
3244 {
3245         u32 max_frame_size = adapter->max_frame_size;
3246         struct e1000_hw *hw = &adapter->hw;
3247         u16 pf_id = adapter->vfs_allocated_count;
3248
3249         if (pf_id) {
3250                 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3251                 /*
3252                  * If we're in VMDQ or SR-IOV mode, then set global RLPML
3253                  * to our max jumbo frame size, in case we need to enable
3254                  * jumbo frames on one of the rings later.
3255                  * This will not pass over-length frames into the default
3256                  * queue because it's gated by the VMOLR.RLPML.
3257                  */
3258                 max_frame_size = MAX_JUMBO_FRAME_SIZE;
3259         }
3260
3261         wr32(E1000_RLPML, max_frame_size);
3262 }
3263
3264 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3265                                  int vfn, bool aupe)
3266 {
3267         struct e1000_hw *hw = &adapter->hw;
3268         u32 vmolr;
3269
3270         /*
3271          * This register exists only on 82576 and newer so if we are older then
3272          * we should exit and do nothing
3273          */
3274         if (hw->mac.type < e1000_82576)
3275                 return;
3276
3277         vmolr = rd32(E1000_VMOLR(vfn));
3278         vmolr |= E1000_VMOLR_STRVLAN;      /* Strip vlan tags */
3279         if (aupe)
3280                 vmolr |= E1000_VMOLR_AUPE;        /* Accept untagged packets */
3281         else
3282                 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3283
3284         /* clear all bits that might not be set */
3285         vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3286
3287         if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3288                 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3289         /*
3290          * for VMDq only allow the VFs and pool 0 to accept broadcast and
3291          * multicast packets
3292          */
3293         if (vfn <= adapter->vfs_allocated_count)
3294                 vmolr |= E1000_VMOLR_BAM;          /* Accept broadcast */
3295
3296         wr32(E1000_VMOLR(vfn), vmolr);
3297 }
3298
3299 /**
3300  * igb_configure_rx_ring - Configure a receive ring after Reset
3301  * @adapter: board private structure
3302  * @ring: receive ring to be configured
3303  *
3304  * Configure the Rx unit of the MAC after a reset.
3305  **/
3306 void igb_configure_rx_ring(struct igb_adapter *adapter,
3307                            struct igb_ring *ring)
3308 {
3309         struct e1000_hw *hw = &adapter->hw;
3310         u64 rdba = ring->dma;
3311         int reg_idx = ring->reg_idx;
3312         u32 srrctl = 0, rxdctl = 0;
3313
3314         /* disable the queue */
3315         wr32(E1000_RXDCTL(reg_idx), 0);
3316
3317         /* Set DMA base address registers */
3318         wr32(E1000_RDBAL(reg_idx),
3319              rdba & 0x00000000ffffffffULL);
3320         wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3321         wr32(E1000_RDLEN(reg_idx),
3322                        ring->count * sizeof(union e1000_adv_rx_desc));
3323
3324         /* initialize head and tail */
3325         ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3326         wr32(E1000_RDH(reg_idx), 0);
3327         writel(0, ring->tail);
3328
3329         /* set descriptor configuration */
3330         srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3331         srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3332         srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3333         if (hw->mac.type >= e1000_82580)
3334                 srrctl |= E1000_SRRCTL_TIMESTAMP;
3335         /* Only set Drop Enable if we are supporting multiple queues */
3336         if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3337                 srrctl |= E1000_SRRCTL_DROP_EN;
3338
3339         wr32(E1000_SRRCTL(reg_idx), srrctl);
3340
3341         /* set filtering for VMDQ pools */
3342         igb_set_vmolr(adapter, reg_idx & 0x7, true);
3343
3344         rxdctl |= IGB_RX_PTHRESH;
3345         rxdctl |= IGB_RX_HTHRESH << 8;
3346         rxdctl |= IGB_RX_WTHRESH << 16;
3347
3348         /* enable receive descriptor fetching */
3349         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3350         wr32(E1000_RXDCTL(reg_idx), rxdctl);
3351 }
3352
3353 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
3354                                   struct igb_ring *rx_ring)
3355 {
3356 #define IGB_MAX_BUILD_SKB_SIZE \
3357         (SKB_WITH_OVERHEAD(IGB_RX_BUFSZ) - \
3358          (NET_SKB_PAD + NET_IP_ALIGN + IGB_TS_HDR_LEN))
3359
3360         /* set build_skb flag */
3361         if (adapter->max_frame_size <= IGB_MAX_BUILD_SKB_SIZE)
3362                 set_ring_build_skb_enabled(rx_ring);
3363         else
3364                 clear_ring_build_skb_enabled(rx_ring);
3365 }
3366
3367 /**
3368  * igb_configure_rx - Configure receive Unit after Reset
3369  * @adapter: board private structure
3370  *
3371  * Configure the Rx unit of the MAC after a reset.
3372  **/
3373 static void igb_configure_rx(struct igb_adapter *adapter)
3374 {
3375         int i;
3376
3377         /* set UTA to appropriate mode */
3378         igb_set_uta(adapter);
3379
3380         /* set the correct pool for the PF default MAC address in entry 0 */
3381         igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3382                          adapter->vfs_allocated_count);
3383
3384         /* Setup the HW Rx Head and Tail Descriptor Pointers and
3385          * the Base and Length of the Rx Descriptor Ring */
3386         for (i = 0; i < adapter->num_rx_queues; i++) {
3387                 struct igb_ring *rx_ring = adapter->rx_ring[i];
3388                 igb_set_rx_buffer_len(adapter, rx_ring);
3389                 igb_configure_rx_ring(adapter, rx_ring);
3390         }
3391 }
3392
3393 /**
3394  * igb_free_tx_resources - Free Tx Resources per Queue
3395  * @tx_ring: Tx descriptor ring for a specific queue
3396  *
3397  * Free all transmit software resources
3398  **/
3399 void igb_free_tx_resources(struct igb_ring *tx_ring)
3400 {
3401         igb_clean_tx_ring(tx_ring);
3402
3403         vfree(tx_ring->tx_buffer_info);
3404         tx_ring->tx_buffer_info = NULL;
3405
3406         /* if not set, then don't free */
3407         if (!tx_ring->desc)
3408                 return;
3409
3410         dma_free_coherent(tx_ring->dev, tx_ring->size,
3411                           tx_ring->desc, tx_ring->dma);
3412
3413         tx_ring->desc = NULL;
3414 }
3415
3416 /**
3417  * igb_free_all_tx_resources - Free Tx Resources for All Queues
3418  * @adapter: board private structure
3419  *
3420  * Free all transmit software resources
3421  **/
3422 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3423 {
3424         int i;
3425
3426         for (i = 0; i < adapter->num_tx_queues; i++)
3427                 igb_free_tx_resources(adapter->tx_ring[i]);
3428 }
3429
3430 void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3431                                     struct igb_tx_buffer *tx_buffer)
3432 {
3433         if (tx_buffer->skb) {
3434                 dev_kfree_skb_any(tx_buffer->skb);
3435                 if (dma_unmap_len(tx_buffer, len))
3436                         dma_unmap_single(ring->dev,
3437                                          dma_unmap_addr(tx_buffer, dma),
3438                                          dma_unmap_len(tx_buffer, len),
3439                                          DMA_TO_DEVICE);
3440         } else if (dma_unmap_len(tx_buffer, len)) {
3441                 dma_unmap_page(ring->dev,
3442                                dma_unmap_addr(tx_buffer, dma),
3443                                dma_unmap_len(tx_buffer, len),
3444                                DMA_TO_DEVICE);
3445         }
3446         tx_buffer->next_to_watch = NULL;
3447         tx_buffer->skb = NULL;
3448         dma_unmap_len_set(tx_buffer, len, 0);
3449         /* buffer_info must be completely set up in the transmit path */
3450 }
3451
3452 /**
3453  * igb_clean_tx_ring - Free Tx Buffers
3454  * @tx_ring: ring to be cleaned
3455  **/
3456 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3457 {
3458         struct igb_tx_buffer *buffer_info;
3459         unsigned long size;
3460         u16 i;
3461
3462         if (!tx_ring->tx_buffer_info)
3463                 return;
3464         /* Free all the Tx ring sk_buffs */
3465
3466         for (i = 0; i < tx_ring->count; i++) {
3467                 buffer_info = &tx_ring->tx_buffer_info[i];
3468                 igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3469         }
3470
3471         netdev_tx_reset_queue(txring_txq(tx_ring));
3472
3473         size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3474         memset(tx_ring->tx_buffer_info, 0, size);
3475
3476         /* Zero out the descriptor ring */
3477         memset(tx_ring->desc, 0, tx_ring->size);
3478
3479         tx_ring->next_to_use = 0;
3480         tx_ring->next_to_clean = 0;
3481 }
3482
3483 /**
3484  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
3485  * @adapter: board private structure
3486  **/
3487 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3488 {
3489         int i;
3490
3491         for (i = 0; i < adapter->num_tx_queues; i++)
3492                 igb_clean_tx_ring(adapter->tx_ring[i]);
3493 }
3494
3495 /**
3496  * igb_free_rx_resources - Free Rx Resources
3497  * @rx_ring: ring to clean the resources from
3498  *
3499  * Free all receive software resources
3500  **/
3501 void igb_free_rx_resources(struct igb_ring *rx_ring)
3502 {
3503         igb_clean_rx_ring(rx_ring);
3504
3505         vfree(rx_ring->rx_buffer_info);
3506         rx_ring->rx_buffer_info = NULL;
3507
3508         /* if not set, then don't free */
3509         if (!rx_ring->desc)
3510                 return;
3511
3512         dma_free_coherent(rx_ring->dev, rx_ring->size,
3513                           rx_ring->desc, rx_ring->dma);
3514
3515         rx_ring->desc = NULL;
3516 }
3517
3518 /**
3519  * igb_free_all_rx_resources - Free Rx Resources for All Queues
3520  * @adapter: board private structure
3521  *
3522  * Free all receive software resources
3523  **/
3524 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3525 {
3526         int i;
3527
3528         for (i = 0; i < adapter->num_rx_queues; i++)
3529                 igb_free_rx_resources(adapter->rx_ring[i]);
3530 }
3531
3532 /**
3533  * igb_clean_rx_ring - Free Rx Buffers per Queue
3534  * @rx_ring: ring to free buffers from
3535  **/
3536 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3537 {
3538         unsigned long size;
3539         u16 i;
3540
3541         if (rx_ring->skb)
3542                 dev_kfree_skb(rx_ring->skb);
3543         rx_ring->skb = NULL;
3544
3545         if (!rx_ring->rx_buffer_info)
3546                 return;
3547
3548         /* Free all the Rx ring sk_buffs */
3549         for (i = 0; i < rx_ring->count; i++) {
3550                 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3551
3552                 if (!buffer_info->page)
3553                         continue;
3554
3555                 dma_unmap_page(rx_ring->dev,
3556                                buffer_info->dma,
3557                                PAGE_SIZE,
3558                                DMA_FROM_DEVICE);
3559                 __free_page(buffer_info->page);
3560
3561                 buffer_info->page = NULL;
3562         }
3563
3564         size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3565         memset(rx_ring->rx_buffer_info, 0, size);
3566
3567         /* Zero out the descriptor ring */
3568         memset(rx_ring->desc, 0, rx_ring->size);
3569
3570         rx_ring->next_to_alloc = 0;
3571         rx_ring->next_to_clean = 0;
3572         rx_ring->next_to_use = 0;
3573 }
3574
3575 /**
3576  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
3577  * @adapter: board private structure
3578  **/
3579 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3580 {
3581         int i;
3582
3583         for (i = 0; i < adapter->num_rx_queues; i++)
3584                 igb_clean_rx_ring(adapter->rx_ring[i]);
3585 }
3586
3587 /**
3588  * igb_set_mac - Change the Ethernet Address of the NIC
3589  * @netdev: network interface device structure
3590  * @p: pointer to an address structure
3591  *
3592  * Returns 0 on success, negative on failure
3593  **/
3594 static int igb_set_mac(struct net_device *netdev, void *p)
3595 {
3596         struct igb_adapter *adapter = netdev_priv(netdev);
3597         struct e1000_hw *hw = &adapter->hw;
3598         struct sockaddr *addr = p;
3599
3600         if (!is_valid_ether_addr(addr->sa_data))
3601                 return -EADDRNOTAVAIL;
3602
3603         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3604         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3605
3606         /* set the correct pool for the new PF MAC address in entry 0 */
3607         igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3608                          adapter->vfs_allocated_count);
3609
3610         return 0;
3611 }
3612
3613 /**
3614  * igb_write_mc_addr_list - write multicast addresses to MTA
3615  * @netdev: network interface device structure
3616  *
3617  * Writes multicast address list to the MTA hash table.
3618  * Returns: -ENOMEM on failure
3619  *                0 on no addresses written
3620  *                X on writing X addresses to MTA
3621  **/
3622 static int igb_write_mc_addr_list(struct net_device *netdev)
3623 {
3624         struct igb_adapter *adapter = netdev_priv(netdev);
3625         struct e1000_hw *hw = &adapter->hw;
3626         struct netdev_hw_addr *ha;
3627         u8  *mta_list;
3628         int i;
3629
3630         if (netdev_mc_empty(netdev)) {
3631                 /* nothing to program, so clear mc list */
3632                 igb_update_mc_addr_list(hw, NULL, 0);
3633                 igb_restore_vf_multicasts(adapter);
3634                 return 0;
3635         }
3636
3637         mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3638         if (!mta_list)
3639                 return -ENOMEM;
3640
3641         /* The shared function expects a packed array of only addresses. */
3642         i = 0;
3643         netdev_for_each_mc_addr(ha, netdev)
3644                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3645
3646         igb_update_mc_addr_list(hw, mta_list, i);
3647         kfree(mta_list);
3648
3649         return netdev_mc_count(netdev);
3650 }
3651
3652 /**
3653  * igb_write_uc_addr_list - write unicast addresses to RAR table
3654  * @netdev: network interface device structure
3655  *
3656  * Writes unicast address list to the RAR table.
3657  * Returns: -ENOMEM on failure/insufficient address space
3658  *                0 on no addresses written
3659  *                X on writing X addresses to the RAR table
3660  **/
3661 static int igb_write_uc_addr_list(struct net_device *netdev)
3662 {
3663         struct igb_adapter *adapter = netdev_priv(netdev);
3664         struct e1000_hw *hw = &adapter->hw;
3665         unsigned int vfn = adapter->vfs_allocated_count;
3666         unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3667         int count = 0;
3668
3669         /* return ENOMEM indicating insufficient memory for addresses */
3670         if (netdev_uc_count(netdev) > rar_entries)
3671                 return -ENOMEM;
3672
3673         if (!netdev_uc_empty(netdev) && rar_entries) {
3674                 struct netdev_hw_addr *ha;
3675
3676                 netdev_for_each_uc_addr(ha, netdev) {
3677                         if (!rar_entries)
3678                                 break;
3679                         igb_rar_set_qsel(adapter, ha->addr,
3680                                          rar_entries--,
3681                                          vfn);
3682                         count++;
3683                 }
3684         }
3685         /* write the addresses in reverse order to avoid write combining */
3686         for (; rar_entries > 0 ; rar_entries--) {
3687                 wr32(E1000_RAH(rar_entries), 0);
3688                 wr32(E1000_RAL(rar_entries), 0);
3689         }
3690         wrfl();
3691
3692         return count;
3693 }
3694
3695 /**
3696  * igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3697  * @netdev: network interface device structure
3698  *
3699  * The set_rx_mode entry point is called whenever the unicast or multicast
3700  * address lists or the network interface flags are updated.  This routine is
3701  * responsible for configuring the hardware for proper unicast, multicast,
3702  * promiscuous mode, and all-multi behavior.
3703  **/
3704 static void igb_set_rx_mode(struct net_device *netdev)
3705 {
3706         struct igb_adapter *adapter = netdev_priv(netdev);
3707         struct e1000_hw *hw = &adapter->hw;
3708         unsigned int vfn = adapter->vfs_allocated_count;
3709         u32 rctl, vmolr = 0;
3710         int count;
3711
3712         /* Check for Promiscuous and All Multicast modes */
3713         rctl = rd32(E1000_RCTL);
3714
3715         /* clear the effected bits */
3716         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3717
3718         if (netdev->flags & IFF_PROMISC) {
3719                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3720                 vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3721         } else {
3722                 if (netdev->flags & IFF_ALLMULTI) {
3723                         rctl |= E1000_RCTL_MPE;
3724                         vmolr |= E1000_VMOLR_MPME;
3725                 } else {
3726                         /*
3727                          * Write addresses to the MTA, if the attempt fails
3728                          * then we should just turn on promiscuous mode so
3729                          * that we can at least receive multicast traffic
3730                          */
3731                         count = igb_write_mc_addr_list(netdev);
3732                         if (count < 0) {
3733                                 rctl |= E1000_RCTL_MPE;
3734                                 vmolr |= E1000_VMOLR_MPME;
3735                         } else if (count) {
3736                                 vmolr |= E1000_VMOLR_ROMPE;
3737                         }
3738                 }
3739                 /*
3740                  * Write addresses to available RAR registers, if there is not
3741                  * sufficient space to store all the addresses then enable
3742                  * unicast promiscuous mode
3743                  */
3744                 count = igb_write_uc_addr_list(netdev);
3745                 if (count < 0) {
3746                         rctl |= E1000_RCTL_UPE;
3747                         vmolr |= E1000_VMOLR_ROPE;
3748                 }
3749                 rctl |= E1000_RCTL_VFE;
3750         }
3751         wr32(E1000_RCTL, rctl);
3752
3753         /*
3754          * In order to support SR-IOV and eventually VMDq it is necessary to set
3755          * the VMOLR to enable the appropriate modes.  Without this workaround
3756          * we will have issues with VLAN tag stripping not being done for frames
3757          * that are only arriving because we are the default pool
3758          */
3759         if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3760                 return;
3761
3762         vmolr |= rd32(E1000_VMOLR(vfn)) &
3763                  ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3764         wr32(E1000_VMOLR(vfn), vmolr);
3765         igb_restore_vf_multicasts(adapter);
3766 }
3767
3768 static void igb_check_wvbr(struct igb_adapter *adapter)
3769 {
3770         struct e1000_hw *hw = &adapter->hw;
3771         u32 wvbr = 0;
3772
3773         switch (hw->mac.type) {
3774         case e1000_82576:
3775         case e1000_i350:
3776                 if (!(wvbr = rd32(E1000_WVBR)))
3777                         return;
3778                 break;
3779         default:
3780                 break;
3781         }
3782
3783         adapter->wvbr |= wvbr;
3784 }
3785
3786 #define IGB_STAGGERED_QUEUE_OFFSET 8
3787
3788 static void igb_spoof_check(struct igb_adapter *adapter)
3789 {
3790         int j;
3791
3792         if (!adapter->wvbr)
3793                 return;
3794
3795         for(j = 0; j < adapter->vfs_allocated_count; j++) {
3796                 if (adapter->wvbr & (1 << j) ||
3797                     adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3798                         dev_warn(&adapter->pdev->dev,
3799                                 "Spoof event(s) detected on VF %d\n", j);
3800                         adapter->wvbr &=
3801                                 ~((1 << j) |
3802                                   (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3803                 }
3804         }
3805 }
3806
3807 /* Need to wait a few seconds after link up to get diagnostic information from
3808  * the phy */
3809 static void igb_update_phy_info(unsigned long data)
3810 {
3811         struct igb_adapter *adapter = (struct igb_adapter *) data;
3812         igb_get_phy_info(&adapter->hw);
3813 }
3814
3815 /**
3816  * igb_has_link - check shared code for link and determine up/down
3817  * @adapter: pointer to driver private info
3818  **/
3819 bool igb_has_link(struct igb_adapter *adapter)
3820 {
3821         struct e1000_hw *hw = &adapter->hw;
3822         bool link_active = false;
3823         s32 ret_val = 0;
3824
3825         /* get_link_status is set on LSC (link status) interrupt or
3826          * rx sequence error interrupt.  get_link_status will stay
3827          * false until the e1000_check_for_link establishes link
3828          * for copper adapters ONLY
3829          */
3830         switch (hw->phy.media_type) {
3831         case e1000_media_type_copper:
3832                 if (hw->mac.get_link_status) {
3833                         ret_val = hw->mac.ops.check_for_link(hw);
3834                         link_active = !hw->mac.get_link_status;
3835                 } else {
3836                         link_active = true;
3837                 }
3838                 break;
3839         case e1000_media_type_internal_serdes:
3840                 ret_val = hw->mac.ops.check_for_link(hw);
3841                 link_active = hw->mac.serdes_has_link;
3842                 break;
3843         default:
3844         case e1000_media_type_unknown:
3845                 break;
3846         }
3847
3848         return link_active;
3849 }
3850
3851 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3852 {
3853         bool ret = false;
3854         u32 ctrl_ext, thstat;
3855
3856         /* check for thermal sensor event on i350 copper only */
3857         if (hw->mac.type == e1000_i350) {
3858                 thstat = rd32(E1000_THSTAT);
3859                 ctrl_ext = rd32(E1000_CTRL_EXT);
3860
3861                 if ((hw->phy.media_type == e1000_media_type_copper) &&
3862                     !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
3863                         ret = !!(thstat & event);
3864         }
3865
3866         return ret;
3867 }
3868
3869 /**
3870  * igb_watchdog - Timer Call-back
3871  * @data: pointer to adapter cast into an unsigned long
3872  **/
3873 static void igb_watchdog(unsigned long data)
3874 {
3875         struct igb_adapter *adapter = (struct igb_adapter *)data;
3876         /* Do the rest outside of interrupt context */
3877         schedule_work(&adapter->watchdog_task);
3878 }
3879
3880 static void igb_watchdog_task(struct work_struct *work)
3881 {
3882         struct igb_adapter *adapter = container_of(work,
3883                                                    struct igb_adapter,
3884                                                    watchdog_task);
3885         struct e1000_hw *hw = &adapter->hw;
3886         struct net_device *netdev = adapter->netdev;
3887         u32 link;
3888         int i;
3889
3890         link = igb_has_link(adapter);
3891         if (link) {
3892                 /* Cancel scheduled suspend requests. */
3893                 pm_runtime_resume(netdev->dev.parent);
3894
3895                 if (!netif_carrier_ok(netdev)) {
3896                         u32 ctrl;
3897                         hw->mac.ops.get_speed_and_duplex(hw,
3898                                                          &adapter->link_speed,
3899                                                          &adapter->link_duplex);
3900
3901                         ctrl = rd32(E1000_CTRL);
3902                         /* Links status message must follow this format */
3903                         printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3904                                "Duplex, Flow Control: %s\n",
3905                                netdev->name,
3906                                adapter->link_speed,
3907                                adapter->link_duplex == FULL_DUPLEX ?
3908                                "Full" : "Half",
3909                                (ctrl & E1000_CTRL_TFCE) &&
3910                                (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3911                                (ctrl & E1000_CTRL_RFCE) ?  "RX" :
3912                                (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
3913
3914                         /* check for thermal sensor event */
3915                         if (igb_thermal_sensor_event(hw,
3916                             E1000_THSTAT_LINK_THROTTLE)) {
3917                                 netdev_info(netdev, "The network adapter link "
3918                                             "speed was downshifted because it "
3919                                             "overheated\n");
3920                         }
3921
3922                         /* adjust timeout factor according to speed/duplex */
3923                         adapter->tx_timeout_factor = 1;
3924                         switch (adapter->link_speed) {
3925                         case SPEED_10:
3926                                 adapter->tx_timeout_factor = 14;
3927                                 break;
3928                         case SPEED_100:
3929                                 /* maybe add some timeout factor ? */
3930                                 break;
3931                         }
3932
3933                         netif_carrier_on(netdev);
3934
3935                         igb_ping_all_vfs(adapter);
3936                         igb_check_vf_rate_limit(adapter);
3937
3938                         /* link state has changed, schedule phy info update */
3939                         if (!test_bit(__IGB_DOWN, &adapter->state))
3940                                 mod_timer(&adapter->phy_info_timer,
3941                                           round_jiffies(jiffies + 2 * HZ));
3942                 }
3943         } else {
3944                 if (netif_carrier_ok(netdev)) {
3945                         adapter->link_speed = 0;
3946                         adapter->link_duplex = 0;
3947
3948                         /* check for thermal sensor event */
3949                         if (igb_thermal_sensor_event(hw,
3950                             E1000_THSTAT_PWR_DOWN)) {
3951                                 netdev_err(netdev, "The network adapter was "
3952                                            "stopped because it overheated\n");
3953                         }
3954
3955                         /* Links status message must follow this format */
3956                         printk(KERN_INFO "igb: %s NIC Link is Down\n",
3957                                netdev->name);
3958                         netif_carrier_off(netdev);
3959
3960                         igb_ping_all_vfs(adapter);
3961
3962                         /* link state has changed, schedule phy info update */
3963                         if (!test_bit(__IGB_DOWN, &adapter->state))
3964                                 mod_timer(&adapter->phy_info_timer,
3965                                           round_jiffies(jiffies + 2 * HZ));
3966
3967                         pm_schedule_suspend(netdev->dev.parent,
3968                                             MSEC_PER_SEC * 5);
3969                 }
3970         }
3971
3972         spin_lock(&adapter->stats64_lock);
3973         igb_update_stats(adapter, &adapter->stats64);
3974         spin_unlock(&adapter->stats64_lock);
3975
3976         for (i = 0; i < adapter->num_tx_queues; i++) {
3977                 struct igb_ring *tx_ring = adapter->tx_ring[i];
3978                 if (!netif_carrier_ok(netdev)) {
3979                         /* We've lost link, so the controller stops DMA,
3980                          * but we've got queued Tx work that's never going
3981                          * to get done, so reset controller to flush Tx.
3982                          * (Do the reset outside of interrupt context). */
3983                         if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
3984                                 adapter->tx_timeout_count++;
3985                                 schedule_work(&adapter->reset_task);
3986                                 /* return immediately since reset is imminent */
3987                                 return;
3988                         }
3989                 }
3990
3991                 /* Force detection of hung controller every watchdog period */
3992                 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
3993         }
3994
3995         /* Cause software interrupt to ensure rx ring is cleaned */
3996         if (adapter->msix_entries) {
3997                 u32 eics = 0;
3998                 for (i = 0; i < adapter->num_q_vectors; i++)
3999                         eics |= adapter->q_vector[i]->eims_value;
4000                 wr32(E1000_EICS, eics);
4001         } else {
4002                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4003         }
4004
4005         igb_spoof_check(adapter);
4006         igb_ptp_rx_hang(adapter);
4007
4008         /* Reset the timer */
4009         if (!test_bit(__IGB_DOWN, &adapter->state))
4010                 mod_timer(&adapter->watchdog_timer,
4011                           round_jiffies(jiffies + 2 * HZ));
4012 }
4013
4014 enum latency_range {
4015         lowest_latency = 0,
4016         low_latency = 1,
4017         bulk_latency = 2,
4018         latency_invalid = 255
4019 };
4020
4021 /**
4022  * igb_update_ring_itr - update the dynamic ITR value based on packet size
4023  *
4024  *      Stores a new ITR value based on strictly on packet size.  This
4025  *      algorithm is less sophisticated than that used in igb_update_itr,
4026  *      due to the difficulty of synchronizing statistics across multiple
4027  *      receive rings.  The divisors and thresholds used by this function
4028  *      were determined based on theoretical maximum wire speed and testing
4029  *      data, in order to minimize response time while increasing bulk
4030  *      throughput.
4031  *      This functionality is controlled by the InterruptThrottleRate module
4032  *      parameter (see igb_param.c)
4033  *      NOTE:  This function is called only when operating in a multiqueue
4034  *             receive environment.
4035  * @q_vector: pointer to q_vector
4036  **/
4037 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4038 {
4039         int new_val = q_vector->itr_val;
4040         int avg_wire_size = 0;
4041         struct igb_adapter *adapter = q_vector->adapter;
4042         unsigned int packets;
4043
4044         /* For non-gigabit speeds, just fix the interrupt rate at 4000
4045          * ints/sec - ITR timer value of 120 ticks.
4046          */
4047         if (adapter->link_speed != SPEED_1000) {
4048                 new_val = IGB_4K_ITR;
4049                 goto set_itr_val;
4050         }
4051
4052         packets = q_vector->rx.total_packets;
4053         if (packets)
4054                 avg_wire_size = q_vector->rx.total_bytes / packets;
4055
4056         packets = q_vector->tx.total_packets;
4057         if (packets)
4058                 avg_wire_size = max_t(u32, avg_wire_size,
4059                                       q_vector->tx.total_bytes / packets);
4060
4061         /* if avg_wire_size isn't set no work was done */
4062         if (!avg_wire_size)
4063                 goto clear_counts;
4064
4065         /* Add 24 bytes to size to account for CRC, preamble, and gap */
4066         avg_wire_size += 24;
4067
4068         /* Don't starve jumbo frames */
4069         avg_wire_size = min(avg_wire_size, 3000);
4070
4071         /* Give a little boost to mid-size frames */
4072         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4073                 new_val = avg_wire_size / 3;
4074         else
4075                 new_val = avg_wire_size / 2;
4076
4077         /* conservative mode (itr 3) eliminates the lowest_latency setting */
4078         if (new_val < IGB_20K_ITR &&
4079             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4080              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4081                 new_val = IGB_20K_ITR;
4082
4083 set_itr_val:
4084         if (new_val != q_vector->itr_val) {
4085                 q_vector->itr_val = new_val;
4086                 q_vector->set_itr = 1;
4087         }
4088 clear_counts:
4089         q_vector->rx.total_bytes = 0;
4090         q_vector->rx.total_packets = 0;
4091         q_vector->tx.total_bytes = 0;
4092         q_vector->tx.total_packets = 0;
4093 }
4094
4095 /**
4096  * igb_update_itr - update the dynamic ITR value based on statistics
4097  *      Stores a new ITR value based on packets and byte
4098  *      counts during the last interrupt.  The advantage of per interrupt
4099  *      computation is faster updates and more accurate ITR for the current
4100  *      traffic pattern.  Constants in this function were computed
4101  *      based on theoretical maximum wire speed and thresholds were set based
4102  *      on testing data as well as attempting to minimize response time
4103  *      while increasing bulk throughput.
4104  *      this functionality is controlled by the InterruptThrottleRate module
4105  *      parameter (see igb_param.c)
4106  *      NOTE:  These calculations are only valid when operating in a single-
4107  *             queue environment.
4108  * @q_vector: pointer to q_vector
4109  * @ring_container: ring info to update the itr for
4110  **/
4111 static void igb_update_itr(struct igb_q_vector *q_vector,
4112                            struct igb_ring_container *ring_container)
4113 {
4114         unsigned int packets = ring_container->total_packets;
4115         unsigned int bytes = ring_container->total_bytes;
4116         u8 itrval = ring_container->itr;
4117
4118         /* no packets, exit with status unchanged */
4119         if (packets == 0)
4120                 return;
4121
4122         switch (itrval) {
4123         case lowest_latency:
4124                 /* handle TSO and jumbo frames */
4125                 if (bytes/packets > 8000)
4126                         itrval = bulk_latency;
4127                 else if ((packets < 5) && (bytes > 512))
4128                         itrval = low_latency;
4129                 break;
4130         case low_latency:  /* 50 usec aka 20000 ints/s */
4131                 if (bytes > 10000) {
4132                         /* this if handles the TSO accounting */
4133                         if (bytes/packets > 8000) {
4134                                 itrval = bulk_latency;
4135                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
4136                                 itrval = bulk_latency;
4137                         } else if ((packets > 35)) {
4138                                 itrval = lowest_latency;
4139                         }
4140                 } else if (bytes/packets > 2000) {
4141                         itrval = bulk_latency;
4142                 } else if (packets <= 2 && bytes < 512) {
4143                         itrval = lowest_latency;
4144                 }
4145                 break;
4146         case bulk_latency: /* 250 usec aka 4000 ints/s */
4147                 if (bytes > 25000) {
4148                         if (packets > 35)
4149                                 itrval = low_latency;
4150                 } else if (bytes < 1500) {
4151                         itrval = low_latency;
4152                 }
4153                 break;
4154         }
4155
4156         /* clear work counters since we have the values we need */
4157         ring_container->total_bytes = 0;
4158         ring_container->total_packets = 0;
4159
4160         /* write updated itr to ring container */
4161         ring_container->itr = itrval;
4162 }
4163
4164 static void igb_set_itr(struct igb_q_vector *q_vector)
4165 {
4166         struct igb_adapter *adapter = q_vector->adapter;
4167         u32 new_itr = q_vector->itr_val;
4168         u8 current_itr = 0;
4169
4170         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4171         if (adapter->link_speed != SPEED_1000) {
4172                 current_itr = 0;
4173                 new_itr = IGB_4K_ITR;
4174                 goto set_itr_now;
4175         }
4176
4177         igb_update_itr(q_vector, &q_vector->tx);
4178         igb_update_itr(q_vector, &q_vector->rx);
4179
4180         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4181
4182         /* conservative mode (itr 3) eliminates the lowest_latency setting */
4183         if (current_itr == lowest_latency &&
4184             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4185              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4186                 current_itr = low_latency;
4187
4188         switch (current_itr) {
4189         /* counts and packets in update_itr are dependent on these numbers */
4190         case lowest_latency:
4191                 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4192                 break;
4193         case low_latency:
4194                 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4195                 break;
4196         case bulk_latency:
4197                 new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
4198                 break;
4199         default:
4200                 break;
4201         }
4202
4203 set_itr_now:
4204         if (new_itr != q_vector->itr_val) {
4205                 /* this attempts to bias the interrupt rate towards Bulk
4206                  * by adding intermediate steps when interrupt rate is
4207                  * increasing */
4208                 new_itr = new_itr > q_vector->itr_val ?
4209                              max((new_itr * q_vector->itr_val) /
4210                                  (new_itr + (q_vector->itr_val >> 2)),
4211                                  new_itr) :
4212                              new_itr;
4213                 /* Don't write the value here; it resets the adapter's
4214                  * internal timer, and causes us to delay far longer than
4215                  * we should between interrupts.  Instead, we write the ITR
4216                  * value at the beginning of the next interrupt so the timing
4217                  * ends up being correct.
4218                  */
4219                 q_vector->itr_val = new_itr;
4220                 q_vector->set_itr = 1;
4221         }
4222 }
4223
4224 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4225                             u32 type_tucmd, u32 mss_l4len_idx)
4226 {
4227         struct e1000_adv_tx_context_desc *context_desc;
4228         u16 i = tx_ring->next_to_use;
4229
4230         context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4231
4232         i++;
4233         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4234
4235         /* set bits to identify this as an advanced context descriptor */
4236         type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4237
4238         /* For 82575, context index must be unique per ring. */
4239         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4240                 mss_l4len_idx |= tx_ring->reg_idx << 4;
4241
4242         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
4243         context_desc->seqnum_seed       = 0;
4244         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
4245         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
4246 }
4247
4248 static int igb_tso(struct igb_ring *tx_ring,
4249                    struct igb_tx_buffer *first,
4250                    u8 *hdr_len)
4251 {
4252         struct sk_buff *skb = first->skb;
4253         u32 vlan_macip_lens, type_tucmd;
4254         u32 mss_l4len_idx, l4len;
4255
4256         if (skb->ip_summed != CHECKSUM_PARTIAL)
4257                 return 0;
4258
4259         if (!skb_is_gso(skb))
4260                 return 0;
4261
4262         if (skb_header_cloned(skb)) {
4263                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4264                 if (err)
4265                         return err;
4266         }
4267
4268         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4269         type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4270
4271         if (first->protocol == __constant_htons(ETH_P_IP)) {
4272                 struct iphdr *iph = ip_hdr(skb);
4273                 iph->tot_len = 0;
4274                 iph->check = 0;
4275                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4276                                                          iph->daddr, 0,
4277                                                          IPPROTO_TCP,
4278                                                          0);
4279                 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4280                 first->tx_flags |= IGB_TX_FLAGS_TSO |
4281                                    IGB_TX_FLAGS_CSUM |
4282                                    IGB_TX_FLAGS_IPV4;
4283         } else if (skb_is_gso_v6(skb)) {
4284                 ipv6_hdr(skb)->payload_len = 0;
4285                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4286                                                        &ipv6_hdr(skb)->daddr,
4287                                                        0, IPPROTO_TCP, 0);
4288                 first->tx_flags |= IGB_TX_FLAGS_TSO |
4289                                    IGB_TX_FLAGS_CSUM;
4290         }
4291
4292         /* compute header lengths */
4293         l4len = tcp_hdrlen(skb);
4294         *hdr_len = skb_transport_offset(skb) + l4len;
4295
4296         /* update gso size and bytecount with header size */
4297         first->gso_segs = skb_shinfo(skb)->gso_segs;
4298         first->bytecount += (first->gso_segs - 1) * *hdr_len;
4299
4300         /* MSS L4LEN IDX */
4301         mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4302         mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4303
4304         /* VLAN MACLEN IPLEN */
4305         vlan_macip_lens = skb_network_header_len(skb);
4306         vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4307         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4308
4309         igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4310
4311         return 1;
4312 }
4313
4314 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4315 {
4316         struct sk_buff *skb = first->skb;
4317         u32 vlan_macip_lens = 0;
4318         u32 mss_l4len_idx = 0;
4319         u32 type_tucmd = 0;
4320
4321         if (skb->ip_summed != CHECKSUM_PARTIAL) {
4322                 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4323                         return;
4324         } else {
4325                 u8 l4_hdr = 0;
4326                 switch (first->protocol) {
4327                 case __constant_htons(ETH_P_IP):
4328                         vlan_macip_lens |= skb_network_header_len(skb);
4329                         type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4330                         l4_hdr = ip_hdr(skb)->protocol;
4331                         break;
4332                 case __constant_htons(ETH_P_IPV6):
4333                         vlan_macip_lens |= skb_network_header_len(skb);
4334                         l4_hdr = ipv6_hdr(skb)->nexthdr;
4335                         break;
4336                 default:
4337                         if (unlikely(net_ratelimit())) {
4338                                 dev_warn(tx_ring->dev,
4339                                  "partial checksum but proto=%x!\n",
4340                                  first->protocol);
4341                         }
4342                         break;
4343                 }
4344
4345                 switch (l4_hdr) {
4346                 case IPPROTO_TCP:
4347                         type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4348                         mss_l4len_idx = tcp_hdrlen(skb) <<
4349                                         E1000_ADVTXD_L4LEN_SHIFT;
4350                         break;
4351                 case IPPROTO_SCTP:
4352                         type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4353                         mss_l4len_idx = sizeof(struct sctphdr) <<
4354                                         E1000_ADVTXD_L4LEN_SHIFT;
4355                         break;
4356                 case IPPROTO_UDP:
4357                         mss_l4len_idx = sizeof(struct udphdr) <<
4358                                         E1000_ADVTXD_L4LEN_SHIFT;
4359                         break;
4360                 default:
4361                         if (unlikely(net_ratelimit())) {
4362                                 dev_warn(tx_ring->dev,
4363                                  "partial checksum but l4 proto=%x!\n",
4364                                  l4_hdr);
4365                         }
4366                         break;
4367                 }
4368
4369                 /* update TX checksum flag */
4370                 first->tx_flags |= IGB_TX_FLAGS_CSUM;
4371         }
4372
4373         vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4374         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4375
4376         igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4377 }
4378
4379 #define IGB_SET_FLAG(_input, _flag, _result) \
4380         ((_flag <= _result) ? \
4381          ((u32)(_input & _flag) * (_result / _flag)) : \
4382          ((u32)(_input & _flag) / (_flag / _result)))
4383
4384 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4385 {
4386         /* set type for advanced descriptor with frame checksum insertion */
4387         u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4388                        E1000_ADVTXD_DCMD_DEXT |
4389                        E1000_ADVTXD_DCMD_IFCS;
4390
4391         /* set HW vlan bit if vlan is present */
4392         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4393                                  (E1000_ADVTXD_DCMD_VLE));
4394
4395         /* set segmentation bits for TSO */
4396         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4397                                  (E1000_ADVTXD_DCMD_TSE));
4398
4399         /* set timestamp bit if present */
4400         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4401                                  (E1000_ADVTXD_MAC_TSTAMP));
4402
4403         /* insert frame checksum */
4404         cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4405
4406         return cmd_type;
4407 }
4408
4409 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4410                                  union e1000_adv_tx_desc *tx_desc,
4411                                  u32 tx_flags, unsigned int paylen)
4412 {
4413         u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4414
4415         /* 82575 requires a unique index per ring */
4416         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4417                 olinfo_status |= tx_ring->reg_idx << 4;
4418
4419         /* insert L4 checksum */
4420         olinfo_status |= IGB_SET_FLAG(tx_flags,
4421                                       IGB_TX_FLAGS_CSUM,
4422                                       (E1000_TXD_POPTS_TXSM << 8));
4423
4424         /* insert IPv4 checksum */
4425         olinfo_status |= IGB_SET_FLAG(tx_flags,
4426                                       IGB_TX_FLAGS_IPV4,
4427                                       (E1000_TXD_POPTS_IXSM << 8));
4428
4429         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4430 }
4431
4432 static void igb_tx_map(struct igb_ring *tx_ring,
4433                        struct igb_tx_buffer *first,
4434                        const u8 hdr_len)
4435 {
4436         struct sk_buff *skb = first->skb;
4437         struct igb_tx_buffer *tx_buffer;
4438         union e1000_adv_tx_desc *tx_desc;
4439         struct skb_frag_struct *frag;
4440         dma_addr_t dma;
4441         unsigned int data_len, size;
4442         u32 tx_flags = first->tx_flags;
4443         u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4444         u16 i = tx_ring->next_to_use;
4445
4446         tx_desc = IGB_TX_DESC(tx_ring, i);
4447
4448         igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4449
4450         size = skb_headlen(skb);
4451         data_len = skb->data_len;
4452
4453         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4454
4455         tx_buffer = first;
4456
4457         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4458                 if (dma_mapping_error(tx_ring->dev, dma))
4459                         goto dma_error;
4460
4461                 /* record length, and DMA address */
4462                 dma_unmap_len_set(tx_buffer, len, size);
4463                 dma_unmap_addr_set(tx_buffer, dma, dma);
4464
4465                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
4466
4467                 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4468                         tx_desc->read.cmd_type_len =
4469                                 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4470
4471                         i++;
4472                         tx_desc++;
4473                         if (i == tx_ring->count) {
4474                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
4475                                 i = 0;
4476                         }
4477                         tx_desc->read.olinfo_status = 0;
4478
4479                         dma += IGB_MAX_DATA_PER_TXD;
4480                         size -= IGB_MAX_DATA_PER_TXD;
4481
4482                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
4483                 }
4484
4485                 if (likely(!data_len))
4486                         break;
4487
4488                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4489
4490                 i++;
4491                 tx_desc++;
4492                 if (i == tx_ring->count) {
4493                         tx_desc = IGB_TX_DESC(tx_ring, 0);
4494                         i = 0;
4495                 }
4496                 tx_desc->read.olinfo_status = 0;
4497
4498                 size = skb_frag_size(frag);
4499                 data_len -= size;
4500
4501                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4502                                        size, DMA_TO_DEVICE);
4503
4504                 tx_buffer = &tx_ring->tx_buffer_info[i];
4505         }
4506
4507         /* write last descriptor with RS and EOP bits */
4508         cmd_type |= size | IGB_TXD_DCMD;
4509         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4510
4511         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4512
4513         /* set the timestamp */
4514         first->time_stamp = jiffies;
4515
4516         /*
4517          * Force memory writes to complete before letting h/w know there
4518          * are new descriptors to fetch.  (Only applicable for weak-ordered
4519          * memory model archs, such as IA-64).
4520          *
4521          * We also need this memory barrier to make certain all of the
4522          * status bits have been updated before next_to_watch is written.
4523          */
4524         wmb();
4525
4526         /* set next_to_watch value indicating a packet is present */
4527         first->next_to_watch = tx_desc;
4528
4529         i++;
4530         if (i == tx_ring->count)
4531                 i = 0;
4532
4533         tx_ring->next_to_use = i;
4534
4535         writel(i, tx_ring->tail);
4536
4537         /* we need this if more than one processor can write to our tail
4538          * at a time, it syncronizes IO on IA64/Altix systems */
4539         mmiowb();
4540
4541         return;
4542
4543 dma_error:
4544         dev_err(tx_ring->dev, "TX DMA map failed\n");
4545
4546         /* clear dma mappings for failed tx_buffer_info map */
4547         for (;;) {
4548                 tx_buffer = &tx_ring->tx_buffer_info[i];
4549                 igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4550                 if (tx_buffer == first)
4551                         break;
4552                 if (i == 0)
4553                         i = tx_ring->count;
4554                 i--;
4555         }
4556
4557         tx_ring->next_to_use = i;
4558 }
4559
4560 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4561 {
4562         struct net_device *netdev = tx_ring->netdev;
4563
4564         netif_stop_subqueue(netdev, tx_ring->queue_index);
4565
4566         /* Herbert's original patch had:
4567          *  smp_mb__after_netif_stop_queue();
4568          * but since that doesn't exist yet, just open code it. */
4569         smp_mb();
4570
4571         /* We need to check again in a case another CPU has just
4572          * made room available. */
4573         if (igb_desc_unused(tx_ring) < size)
4574                 return -EBUSY;
4575
4576         /* A reprieve! */
4577         netif_wake_subqueue(netdev, tx_ring->queue_index);
4578
4579         u64_stats_update_begin(&tx_ring->tx_syncp2);
4580         tx_ring->tx_stats.restart_queue2++;
4581         u64_stats_update_end(&tx_ring->tx_syncp2);
4582
4583         return 0;
4584 }
4585
4586 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4587 {
4588         if (igb_desc_unused(tx_ring) >= size)
4589                 return 0;
4590         return __igb_maybe_stop_tx(tx_ring, size);
4591 }
4592
4593 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4594                                 struct igb_ring *tx_ring)
4595 {
4596         struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4597         struct igb_tx_buffer *first;
4598         int tso;
4599         u32 tx_flags = 0;
4600         u16 count = TXD_USE_COUNT(skb_headlen(skb));
4601         __be16 protocol = vlan_get_protocol(skb);
4602         u8 hdr_len = 0;
4603
4604         /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4605          *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4606          *       + 2 desc gap to keep tail from touching head,
4607          *       + 1 desc for context descriptor,
4608          * otherwise try next time
4609          */
4610         if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4611                 unsigned short f;
4612                 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4613                         count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4614         } else {
4615                 count += skb_shinfo(skb)->nr_frags;
4616         }
4617
4618         if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4619                 /* this is a hard error */
4620                 return NETDEV_TX_BUSY;
4621         }
4622
4623         /* record the location of the first descriptor for this packet */
4624         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4625         first->skb = skb;
4626         first->bytecount = skb->len;
4627         first->gso_segs = 1;
4628
4629         skb_tx_timestamp(skb);
4630
4631         if (unlikely((skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP) &&
4632                      !(adapter->ptp_tx_skb))) {
4633                 skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4634                 tx_flags |= IGB_TX_FLAGS_TSTAMP;
4635
4636                 adapter->ptp_tx_skb = skb_get(skb);
4637                 adapter->ptp_tx_start = jiffies;
4638                 if (adapter->hw.mac.type == e1000_82576)
4639                         schedule_work(&adapter->ptp_tx_work);
4640         }
4641
4642         if (vlan_tx_tag_present(skb)) {
4643                 tx_flags |= IGB_TX_FLAGS_VLAN;
4644                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4645         }
4646
4647         /* record initial flags and protocol */
4648         first->tx_flags = tx_flags;
4649         first->protocol = protocol;
4650
4651         tso = igb_tso(tx_ring, first, &hdr_len);
4652         if (tso < 0)
4653                 goto out_drop;
4654         else if (!tso)
4655                 igb_tx_csum(tx_ring, first);
4656
4657         igb_tx_map(tx_ring, first, hdr_len);
4658
4659         /* Make sure there is space in the ring for the next send. */
4660         igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4661
4662         return NETDEV_TX_OK;
4663
4664 out_drop:
4665         igb_unmap_and_free_tx_resource(tx_ring, first);
4666
4667         return NETDEV_TX_OK;
4668 }
4669
4670 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4671                                                     struct sk_buff *skb)
4672 {
4673         unsigned int r_idx = skb->queue_mapping;
4674
4675         if (r_idx >= adapter->num_tx_queues)
4676                 r_idx = r_idx % adapter->num_tx_queues;
4677
4678         return adapter->tx_ring[r_idx];
4679 }
4680
4681 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4682                                   struct net_device *netdev)
4683 {
4684         struct igb_adapter *adapter = netdev_priv(netdev);
4685
4686         if (test_bit(__IGB_DOWN, &adapter->state)) {
4687                 dev_kfree_skb_any(skb);
4688                 return NETDEV_TX_OK;
4689         }
4690
4691         if (skb->len <= 0) {
4692                 dev_kfree_skb_any(skb);
4693                 return NETDEV_TX_OK;
4694         }
4695
4696         /*
4697          * The minimum packet size with TCTL.PSP set is 17 so pad the skb
4698          * in order to meet this minimum size requirement.
4699          */
4700         if (unlikely(skb->len < 17)) {
4701                 if (skb_pad(skb, 17 - skb->len))
4702                         return NETDEV_TX_OK;
4703                 skb->len = 17;
4704                 skb_set_tail_pointer(skb, 17);
4705         }
4706
4707         return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4708 }
4709
4710 /**
4711  * igb_tx_timeout - Respond to a Tx Hang
4712  * @netdev: network interface device structure
4713  **/
4714 static void igb_tx_timeout(struct net_device *netdev)
4715 {
4716         struct igb_adapter *adapter = netdev_priv(netdev);
4717         struct e1000_hw *hw = &adapter->hw;
4718
4719         /* Do the reset outside of interrupt context */
4720         adapter->tx_timeout_count++;
4721
4722         if (hw->mac.type >= e1000_82580)
4723                 hw->dev_spec._82575.global_device_reset = true;
4724
4725         schedule_work(&adapter->reset_task);
4726         wr32(E1000_EICS,
4727              (adapter->eims_enable_mask & ~adapter->eims_other));
4728 }
4729
4730 static void igb_reset_task(struct work_struct *work)
4731 {
4732         struct igb_adapter *adapter;
4733         adapter = container_of(work, struct igb_adapter, reset_task);
4734
4735         igb_dump(adapter);
4736         netdev_err(adapter->netdev, "Reset adapter\n");
4737         igb_reinit_locked(adapter);
4738 }
4739
4740 /**
4741  * igb_get_stats64 - Get System Network Statistics
4742  * @netdev: network interface device structure
4743  * @stats: rtnl_link_stats64 pointer
4744  *
4745  **/
4746 static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4747                                                  struct rtnl_link_stats64 *stats)
4748 {
4749         struct igb_adapter *adapter = netdev_priv(netdev);
4750
4751         spin_lock(&adapter->stats64_lock);
4752         igb_update_stats(adapter, &adapter->stats64);
4753         memcpy(stats, &adapter->stats64, sizeof(*stats));
4754         spin_unlock(&adapter->stats64_lock);
4755
4756         return stats;
4757 }
4758
4759 /**
4760  * igb_change_mtu - Change the Maximum Transfer Unit
4761  * @netdev: network interface device structure
4762  * @new_mtu: new value for maximum frame size
4763  *
4764  * Returns 0 on success, negative on failure
4765  **/
4766 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4767 {
4768         struct igb_adapter *adapter = netdev_priv(netdev);
4769         struct pci_dev *pdev = adapter->pdev;
4770         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4771
4772         if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4773                 dev_err(&pdev->dev, "Invalid MTU setting\n");
4774                 return -EINVAL;
4775         }
4776
4777 #define MAX_STD_JUMBO_FRAME_SIZE 9238
4778         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4779                 dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4780                 return -EINVAL;
4781         }
4782
4783         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4784                 msleep(1);
4785
4786         /* igb_down has a dependency on max_frame_size */
4787         adapter->max_frame_size = max_frame;
4788
4789         if (netif_running(netdev))
4790                 igb_down(adapter);
4791
4792         dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4793                  netdev->mtu, new_mtu);
4794         netdev->mtu = new_mtu;
4795
4796         if (netif_running(netdev))
4797                 igb_up(adapter);
4798         else
4799                 igb_reset(adapter);
4800
4801         clear_bit(__IGB_RESETTING, &adapter->state);
4802
4803         return 0;
4804 }
4805
4806 /**
4807  * igb_update_stats - Update the board statistics counters
4808  * @adapter: board private structure
4809  **/
4810
4811 void igb_update_stats(struct igb_adapter *adapter,
4812                       struct rtnl_link_stats64 *net_stats)
4813 {
4814         struct e1000_hw *hw = &adapter->hw;
4815         struct pci_dev *pdev = adapter->pdev;
4816         u32 reg, mpc;
4817         u16 phy_tmp;
4818         int i;
4819         u64 bytes, packets;
4820         unsigned int start;
4821         u64 _bytes, _packets;
4822
4823 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4824
4825         /*
4826          * Prevent stats update while adapter is being reset, or if the pci
4827          * connection is down.
4828          */
4829         if (adapter->link_speed == 0)
4830                 return;
4831         if (pci_channel_offline(pdev))
4832                 return;
4833
4834         bytes = 0;
4835         packets = 0;
4836         for (i = 0; i < adapter->num_rx_queues; i++) {
4837                 u32 rqdpc = rd32(E1000_RQDPC(i));
4838                 struct igb_ring *ring = adapter->rx_ring[i];
4839
4840                 if (rqdpc) {
4841                         ring->rx_stats.drops += rqdpc;
4842                         net_stats->rx_fifo_errors += rqdpc;
4843                 }
4844
4845                 do {
4846                         start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4847                         _bytes = ring->rx_stats.bytes;
4848                         _packets = ring->rx_stats.packets;
4849                 } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4850                 bytes += _bytes;
4851                 packets += _packets;
4852         }
4853
4854         net_stats->rx_bytes = bytes;
4855         net_stats->rx_packets = packets;
4856
4857         bytes = 0;
4858         packets = 0;
4859         for (i = 0; i < adapter->num_tx_queues; i++) {
4860                 struct igb_ring *ring = adapter->tx_ring[i];
4861                 do {
4862                         start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4863                         _bytes = ring->tx_stats.bytes;
4864                         _packets = ring->tx_stats.packets;
4865                 } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4866                 bytes += _bytes;
4867                 packets += _packets;
4868         }
4869         net_stats->tx_bytes = bytes;
4870         net_stats->tx_packets = packets;
4871
4872         /* read stats registers */
4873         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4874         adapter->stats.gprc += rd32(E1000_GPRC);
4875         adapter->stats.gorc += rd32(E1000_GORCL);
4876         rd32(E1000_GORCH); /* clear GORCL */
4877         adapter->stats.bprc += rd32(E1000_BPRC);
4878         adapter->stats.mprc += rd32(E1000_MPRC);
4879         adapter->stats.roc += rd32(E1000_ROC);
4880
4881         adapter->stats.prc64 += rd32(E1000_PRC64);
4882         adapter->stats.prc127 += rd32(E1000_PRC127);
4883         adapter->stats.prc255 += rd32(E1000_PRC255);
4884         adapter->stats.prc511 += rd32(E1000_PRC511);
4885         adapter->stats.prc1023 += rd32(E1000_PRC1023);
4886         adapter->stats.prc1522 += rd32(E1000_PRC1522);
4887         adapter->stats.symerrs += rd32(E1000_SYMERRS);
4888         adapter->stats.sec += rd32(E1000_SEC);
4889
4890         mpc = rd32(E1000_MPC);
4891         adapter->stats.mpc += mpc;
4892         net_stats->rx_fifo_errors += mpc;
4893         adapter->stats.scc += rd32(E1000_SCC);
4894         adapter->stats.ecol += rd32(E1000_ECOL);
4895         adapter->stats.mcc += rd32(E1000_MCC);
4896         adapter->stats.latecol += rd32(E1000_LATECOL);
4897         adapter->stats.dc += rd32(E1000_DC);
4898         adapter->stats.rlec += rd32(E1000_RLEC);
4899         adapter->stats.xonrxc += rd32(E1000_XONRXC);
4900         adapter->stats.xontxc += rd32(E1000_XONTXC);
4901         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4902         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4903         adapter->stats.fcruc += rd32(E1000_FCRUC);
4904         adapter->stats.gptc += rd32(E1000_GPTC);
4905         adapter->stats.gotc += rd32(E1000_GOTCL);
4906         rd32(E1000_GOTCH); /* clear GOTCL */
4907         adapter->stats.rnbc += rd32(E1000_RNBC);
4908         adapter->stats.ruc += rd32(E1000_RUC);
4909         adapter->stats.rfc += rd32(E1000_RFC);
4910         adapter->stats.rjc += rd32(E1000_RJC);
4911         adapter->stats.tor += rd32(E1000_TORH);
4912         adapter->stats.tot += rd32(E1000_TOTH);
4913         adapter->stats.tpr += rd32(E1000_TPR);
4914
4915         adapter->stats.ptc64 += rd32(E1000_PTC64);
4916         adapter->stats.ptc127 += rd32(E1000_PTC127);
4917         adapter->stats.ptc255 += rd32(E1000_PTC255);
4918         adapter->stats.ptc511 += rd32(E1000_PTC511);
4919         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
4920         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
4921
4922         adapter->stats.mptc += rd32(E1000_MPTC);
4923         adapter->stats.bptc += rd32(E1000_BPTC);
4924
4925         adapter->stats.tpt += rd32(E1000_TPT);
4926         adapter->stats.colc += rd32(E1000_COLC);
4927
4928         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
4929         /* read internal phy specific stats */
4930         reg = rd32(E1000_CTRL_EXT);
4931         if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
4932                 adapter->stats.rxerrc += rd32(E1000_RXERRC);
4933
4934                 /* this stat has invalid values on i210/i211 */
4935                 if ((hw->mac.type != e1000_i210) &&
4936                     (hw->mac.type != e1000_i211))
4937                         adapter->stats.tncrs += rd32(E1000_TNCRS);
4938         }
4939
4940         adapter->stats.tsctc += rd32(E1000_TSCTC);
4941         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
4942
4943         adapter->stats.iac += rd32(E1000_IAC);
4944         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
4945         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
4946         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
4947         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
4948         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
4949         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
4950         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
4951         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
4952
4953         /* Fill out the OS statistics structure */
4954         net_stats->multicast = adapter->stats.mprc;
4955         net_stats->collisions = adapter->stats.colc;
4956
4957         /* Rx Errors */
4958
4959         /* RLEC on some newer hardware can be incorrect so build
4960          * our own version based on RUC and ROC */
4961         net_stats->rx_errors = adapter->stats.rxerrc +
4962                 adapter->stats.crcerrs + adapter->stats.algnerrc +
4963                 adapter->stats.ruc + adapter->stats.roc +
4964                 adapter->stats.cexterr;
4965         net_stats->rx_length_errors = adapter->stats.ruc +
4966                                       adapter->stats.roc;
4967         net_stats->rx_crc_errors = adapter->stats.crcerrs;
4968         net_stats->rx_frame_errors = adapter->stats.algnerrc;
4969         net_stats->rx_missed_errors = adapter->stats.mpc;
4970
4971         /* Tx Errors */
4972         net_stats->tx_errors = adapter->stats.ecol +
4973                                adapter->stats.latecol;
4974         net_stats->tx_aborted_errors = adapter->stats.ecol;
4975         net_stats->tx_window_errors = adapter->stats.latecol;
4976         net_stats->tx_carrier_errors = adapter->stats.tncrs;
4977
4978         /* Tx Dropped needs to be maintained elsewhere */
4979
4980         /* Phy Stats */
4981         if (hw->phy.media_type == e1000_media_type_copper) {
4982                 if ((adapter->link_speed == SPEED_1000) &&
4983                    (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
4984                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
4985                         adapter->phy_stats.idle_errors += phy_tmp;
4986                 }
4987         }
4988
4989         /* Management Stats */
4990         adapter->stats.mgptc += rd32(E1000_MGTPTC);
4991         adapter->stats.mgprc += rd32(E1000_MGTPRC);
4992         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
4993
4994         /* OS2BMC Stats */
4995         reg = rd32(E1000_MANC);
4996         if (reg & E1000_MANC_EN_BMC2OS) {
4997                 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
4998                 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
4999                 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5000                 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5001         }
5002 }
5003
5004 static irqreturn_t igb_msix_other(int irq, void *data)
5005 {
5006         struct igb_adapter *adapter = data;
5007         struct e1000_hw *hw = &adapter->hw;
5008         u32 icr = rd32(E1000_ICR);
5009         /* reading ICR causes bit 31 of EICR to be cleared */
5010
5011         if (icr & E1000_ICR_DRSTA)
5012                 schedule_work(&adapter->reset_task);
5013
5014         if (icr & E1000_ICR_DOUTSYNC) {
5015                 /* HW is reporting DMA is out of sync */
5016                 adapter->stats.doosync++;
5017                 /* The DMA Out of Sync is also indication of a spoof event
5018                  * in IOV mode. Check the Wrong VM Behavior register to
5019                  * see if it is really a spoof event. */
5020                 igb_check_wvbr(adapter);
5021         }
5022
5023         /* Check for a mailbox event */
5024         if (icr & E1000_ICR_VMMB)
5025                 igb_msg_task(adapter);
5026
5027         if (icr & E1000_ICR_LSC) {
5028                 hw->mac.get_link_status = 1;
5029                 /* guard against interrupt when we're going down */
5030                 if (!test_bit(__IGB_DOWN, &adapter->state))
5031                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5032         }
5033
5034         if (icr & E1000_ICR_TS) {
5035                 u32 tsicr = rd32(E1000_TSICR);
5036
5037                 if (tsicr & E1000_TSICR_TXTS) {
5038                         /* acknowledge the interrupt */
5039                         wr32(E1000_TSICR, E1000_TSICR_TXTS);
5040                         /* retrieve hardware timestamp */
5041                         schedule_work(&adapter->ptp_tx_work);
5042                 }
5043         }
5044
5045         wr32(E1000_EIMS, adapter->eims_other);
5046
5047         return IRQ_HANDLED;
5048 }
5049
5050 static void igb_write_itr(struct igb_q_vector *q_vector)
5051 {
5052         struct igb_adapter *adapter = q_vector->adapter;
5053         u32 itr_val = q_vector->itr_val & 0x7FFC;
5054
5055         if (!q_vector->set_itr)
5056                 return;
5057
5058         if (!itr_val)
5059                 itr_val = 0x4;
5060
5061         if (adapter->hw.mac.type == e1000_82575)
5062                 itr_val |= itr_val << 16;
5063         else
5064                 itr_val |= E1000_EITR_CNT_IGNR;
5065
5066         writel(itr_val, q_vector->itr_register);
5067         q_vector->set_itr = 0;
5068 }
5069
5070 static irqreturn_t igb_msix_ring(int irq, void *data)
5071 {
5072         struct igb_q_vector *q_vector = data;
5073
5074         /* Write the ITR value calculated from the previous interrupt. */
5075         igb_write_itr(q_vector);
5076
5077         napi_schedule(&q_vector->napi);
5078
5079         return IRQ_HANDLED;
5080 }
5081
5082 #ifdef CONFIG_IGB_DCA
5083 static void igb_update_tx_dca(struct igb_adapter *adapter,
5084                               struct igb_ring *tx_ring,
5085                               int cpu)
5086 {
5087         struct e1000_hw *hw = &adapter->hw;
5088         u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5089
5090         if (hw->mac.type != e1000_82575)
5091                 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5092
5093         /*
5094          * We can enable relaxed ordering for reads, but not writes when
5095          * DCA is enabled.  This is due to a known issue in some chipsets
5096          * which will cause the DCA tag to be cleared.
5097          */
5098         txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5099                   E1000_DCA_TXCTRL_DATA_RRO_EN |
5100                   E1000_DCA_TXCTRL_DESC_DCA_EN;
5101
5102         wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5103 }
5104
5105 static void igb_update_rx_dca(struct igb_adapter *adapter,
5106                               struct igb_ring *rx_ring,
5107                               int cpu)
5108 {
5109         struct e1000_hw *hw = &adapter->hw;
5110         u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5111
5112         if (hw->mac.type != e1000_82575)
5113                 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5114
5115         /*
5116          * We can enable relaxed ordering for reads, but not writes when
5117          * DCA is enabled.  This is due to a known issue in some chipsets
5118          * which will cause the DCA tag to be cleared.
5119          */
5120         rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5121                   E1000_DCA_RXCTRL_DESC_DCA_EN;
5122
5123         wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5124 }
5125
5126 static void igb_update_dca(struct igb_q_vector *q_vector)
5127 {
5128         struct igb_adapter *adapter = q_vector->adapter;
5129         int cpu = get_cpu();
5130
5131         if (q_vector->cpu == cpu)
5132                 goto out_no_update;
5133
5134         if (q_vector->tx.ring)
5135                 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5136
5137         if (q_vector->rx.ring)
5138                 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5139
5140         q_vector->cpu = cpu;
5141 out_no_update:
5142         put_cpu();
5143 }
5144
5145 static void igb_setup_dca(struct igb_adapter *adapter)
5146 {
5147         struct e1000_hw *hw = &adapter->hw;
5148         int i;
5149
5150         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5151                 return;
5152
5153         /* Always use CB2 mode, difference is masked in the CB driver. */
5154         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5155
5156         for (i = 0; i < adapter->num_q_vectors; i++) {
5157                 adapter->q_vector[i]->cpu = -1;
5158                 igb_update_dca(adapter->q_vector[i]);
5159         }
5160 }
5161
5162 static int __igb_notify_dca(struct device *dev, void *data)
5163 {
5164         struct net_device *netdev = dev_get_drvdata(dev);
5165         struct igb_adapter *adapter = netdev_priv(netdev);
5166         struct pci_dev *pdev = adapter->pdev;
5167         struct e1000_hw *hw = &adapter->hw;
5168         unsigned long event = *(unsigned long *)data;
5169
5170         switch (event) {
5171         case DCA_PROVIDER_ADD:
5172                 /* if already enabled, don't do it again */
5173                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5174                         break;
5175                 if (dca_add_requester(dev) == 0) {
5176                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
5177                         dev_info(&pdev->dev, "DCA enabled\n");
5178                         igb_setup_dca(adapter);
5179                         break;
5180                 }
5181                 /* Fall Through since DCA is disabled. */
5182         case DCA_PROVIDER_REMOVE:
5183                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5184                         /* without this a class_device is left
5185                          * hanging around in the sysfs model */
5186                         dca_remove_requester(dev);
5187                         dev_info(&pdev->dev, "DCA disabled\n");
5188                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5189                         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5190                 }
5191                 break;
5192         }
5193
5194         return 0;
5195 }
5196
5197 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5198                           void *p)
5199 {
5200         int ret_val;
5201
5202         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5203                                          __igb_notify_dca);
5204
5205         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5206 }
5207 #endif /* CONFIG_IGB_DCA */
5208
5209 #ifdef CONFIG_PCI_IOV
5210 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5211 {
5212         unsigned char mac_addr[ETH_ALEN];
5213
5214         eth_zero_addr(mac_addr);
5215         igb_set_vf_mac(adapter, vf, mac_addr);
5216
5217         return 0;
5218 }
5219
5220 static bool igb_vfs_are_assigned(struct igb_adapter *adapter)
5221 {
5222         struct pci_dev *pdev = adapter->pdev;
5223         struct pci_dev *vfdev;
5224         int dev_id;
5225
5226         switch (adapter->hw.mac.type) {
5227         case e1000_82576:
5228                 dev_id = IGB_82576_VF_DEV_ID;
5229                 break;
5230         case e1000_i350:
5231                 dev_id = IGB_I350_VF_DEV_ID;
5232                 break;
5233         default:
5234                 return false;
5235         }
5236
5237         /* loop through all the VFs to see if we own any that are assigned */
5238         vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, NULL);
5239         while (vfdev) {
5240                 /* if we don't own it we don't care */
5241                 if (vfdev->is_virtfn && vfdev->physfn == pdev) {
5242                         /* if it is assigned we cannot release it */
5243                         if (vfdev->dev_flags & PCI_DEV_FLAGS_ASSIGNED)
5244                                 return true;
5245                 }
5246
5247                 vfdev = pci_get_device(PCI_VENDOR_ID_INTEL, dev_id, vfdev);
5248         }
5249
5250         return false;
5251 }
5252
5253 #endif
5254 static void igb_ping_all_vfs(struct igb_adapter *adapter)
5255 {
5256         struct e1000_hw *hw = &adapter->hw;
5257         u32 ping;
5258         int i;
5259
5260         for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5261                 ping = E1000_PF_CONTROL_MSG;
5262                 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5263                         ping |= E1000_VT_MSGTYPE_CTS;
5264                 igb_write_mbx(hw, &ping, 1, i);
5265         }
5266 }
5267
5268 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5269 {
5270         struct e1000_hw *hw = &adapter->hw;
5271         u32 vmolr = rd32(E1000_VMOLR(vf));
5272         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5273
5274         vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5275                             IGB_VF_FLAG_MULTI_PROMISC);
5276         vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5277
5278         if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5279                 vmolr |= E1000_VMOLR_MPME;
5280                 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5281                 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5282         } else {
5283                 /*
5284                  * if we have hashes and we are clearing a multicast promisc
5285                  * flag we need to write the hashes to the MTA as this step
5286                  * was previously skipped
5287                  */
5288                 if (vf_data->num_vf_mc_hashes > 30) {
5289                         vmolr |= E1000_VMOLR_MPME;
5290                 } else if (vf_data->num_vf_mc_hashes) {
5291                         int j;
5292                         vmolr |= E1000_VMOLR_ROMPE;
5293                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5294                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5295                 }
5296         }
5297
5298         wr32(E1000_VMOLR(vf), vmolr);
5299
5300         /* there are flags left unprocessed, likely not supported */
5301         if (*msgbuf & E1000_VT_MSGINFO_MASK)
5302                 return -EINVAL;
5303
5304         return 0;
5305
5306 }
5307
5308 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5309                                   u32 *msgbuf, u32 vf)
5310 {
5311         int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5312         u16 *hash_list = (u16 *)&msgbuf[1];
5313         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5314         int i;
5315
5316         /* salt away the number of multicast addresses assigned
5317          * to this VF for later use to restore when the PF multi cast
5318          * list changes
5319          */
5320         vf_data->num_vf_mc_hashes = n;
5321
5322         /* only up to 30 hash values supported */
5323         if (n > 30)
5324                 n = 30;
5325
5326         /* store the hashes for later use */
5327         for (i = 0; i < n; i++)
5328                 vf_data->vf_mc_hashes[i] = hash_list[i];
5329
5330         /* Flush and reset the mta with the new values */
5331         igb_set_rx_mode(adapter->netdev);
5332
5333         return 0;
5334 }
5335
5336 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5337 {
5338         struct e1000_hw *hw = &adapter->hw;
5339         struct vf_data_storage *vf_data;
5340         int i, j;
5341
5342         for (i = 0; i < adapter->vfs_allocated_count; i++) {
5343                 u32 vmolr = rd32(E1000_VMOLR(i));
5344                 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5345
5346                 vf_data = &adapter->vf_data[i];
5347
5348                 if ((vf_data->num_vf_mc_hashes > 30) ||
5349                     (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5350                         vmolr |= E1000_VMOLR_MPME;
5351                 } else if (vf_data->num_vf_mc_hashes) {
5352                         vmolr |= E1000_VMOLR_ROMPE;
5353                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5354                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5355                 }
5356                 wr32(E1000_VMOLR(i), vmolr);
5357         }
5358 }
5359
5360 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5361 {
5362         struct e1000_hw *hw = &adapter->hw;
5363         u32 pool_mask, reg, vid;
5364         int i;
5365
5366         pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5367
5368         /* Find the vlan filter for this id */
5369         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5370                 reg = rd32(E1000_VLVF(i));
5371
5372                 /* remove the vf from the pool */
5373                 reg &= ~pool_mask;
5374
5375                 /* if pool is empty then remove entry from vfta */
5376                 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5377                     (reg & E1000_VLVF_VLANID_ENABLE)) {
5378                         reg = 0;
5379                         vid = reg & E1000_VLVF_VLANID_MASK;
5380                         igb_vfta_set(hw, vid, false);
5381                 }
5382
5383                 wr32(E1000_VLVF(i), reg);
5384         }
5385
5386         adapter->vf_data[vf].vlans_enabled = 0;
5387 }
5388
5389 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5390 {
5391         struct e1000_hw *hw = &adapter->hw;
5392         u32 reg, i;
5393
5394         /* The vlvf table only exists on 82576 hardware and newer */
5395         if (hw->mac.type < e1000_82576)
5396                 return -1;
5397
5398         /* we only need to do this if VMDq is enabled */
5399         if (!adapter->vfs_allocated_count)
5400                 return -1;
5401
5402         /* Find the vlan filter for this id */
5403         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5404                 reg = rd32(E1000_VLVF(i));
5405                 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5406                     vid == (reg & E1000_VLVF_VLANID_MASK))
5407                         break;
5408         }
5409
5410         if (add) {
5411                 if (i == E1000_VLVF_ARRAY_SIZE) {
5412                         /* Did not find a matching VLAN ID entry that was
5413                          * enabled.  Search for a free filter entry, i.e.
5414                          * one without the enable bit set
5415                          */
5416                         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5417                                 reg = rd32(E1000_VLVF(i));
5418                                 if (!(reg & E1000_VLVF_VLANID_ENABLE))
5419                                         break;
5420                         }
5421                 }
5422                 if (i < E1000_VLVF_ARRAY_SIZE) {
5423                         /* Found an enabled/available entry */
5424                         reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5425
5426                         /* if !enabled we need to set this up in vfta */
5427                         if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5428                                 /* add VID to filter table */
5429                                 igb_vfta_set(hw, vid, true);
5430                                 reg |= E1000_VLVF_VLANID_ENABLE;
5431                         }
5432                         reg &= ~E1000_VLVF_VLANID_MASK;
5433                         reg |= vid;
5434                         wr32(E1000_VLVF(i), reg);
5435
5436                         /* do not modify RLPML for PF devices */
5437                         if (vf >= adapter->vfs_allocated_count)
5438                                 return 0;
5439
5440                         if (!adapter->vf_data[vf].vlans_enabled) {
5441                                 u32 size;
5442                                 reg = rd32(E1000_VMOLR(vf));
5443                                 size = reg & E1000_VMOLR_RLPML_MASK;
5444                                 size += 4;
5445                                 reg &= ~E1000_VMOLR_RLPML_MASK;
5446                                 reg |= size;
5447                                 wr32(E1000_VMOLR(vf), reg);
5448                         }
5449
5450                         adapter->vf_data[vf].vlans_enabled++;
5451                 }
5452         } else {
5453                 if (i < E1000_VLVF_ARRAY_SIZE) {
5454                         /* remove vf from the pool */
5455                         reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5456                         /* if pool is empty then remove entry from vfta */
5457                         if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5458                                 reg = 0;
5459                                 igb_vfta_set(hw, vid, false);
5460                         }
5461                         wr32(E1000_VLVF(i), reg);
5462
5463                         /* do not modify RLPML for PF devices */
5464                         if (vf >= adapter->vfs_allocated_count)
5465                                 return 0;
5466
5467                         adapter->vf_data[vf].vlans_enabled--;
5468                         if (!adapter->vf_data[vf].vlans_enabled) {
5469                                 u32 size;
5470                                 reg = rd32(E1000_VMOLR(vf));
5471                                 size = reg & E1000_VMOLR_RLPML_MASK;
5472                                 size -= 4;
5473                                 reg &= ~E1000_VMOLR_RLPML_MASK;
5474                                 reg |= size;
5475                                 wr32(E1000_VMOLR(vf), reg);
5476                         }
5477                 }
5478         }
5479         return 0;
5480 }
5481
5482 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5483 {
5484         struct e1000_hw *hw = &adapter->hw;
5485
5486         if (vid)
5487                 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5488         else
5489                 wr32(E1000_VMVIR(vf), 0);
5490 }
5491
5492 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5493                                int vf, u16 vlan, u8 qos)
5494 {
5495         int err = 0;
5496         struct igb_adapter *adapter = netdev_priv(netdev);
5497
5498         if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5499                 return -EINVAL;
5500         if (vlan || qos) {
5501                 err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5502                 if (err)
5503                         goto out;
5504                 igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5505                 igb_set_vmolr(adapter, vf, !vlan);
5506                 adapter->vf_data[vf].pf_vlan = vlan;
5507                 adapter->vf_data[vf].pf_qos = qos;
5508                 dev_info(&adapter->pdev->dev,
5509                          "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5510                 if (test_bit(__IGB_DOWN, &adapter->state)) {
5511                         dev_warn(&adapter->pdev->dev,
5512                                  "The VF VLAN has been set,"
5513                                  " but the PF device is not up.\n");
5514                         dev_warn(&adapter->pdev->dev,
5515                                  "Bring the PF device up before"
5516                                  " attempting to use the VF device.\n");
5517                 }
5518         } else {
5519                 igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5520                                    false, vf);
5521                 igb_set_vmvir(adapter, vlan, vf);
5522                 igb_set_vmolr(adapter, vf, true);
5523                 adapter->vf_data[vf].pf_vlan = 0;
5524                 adapter->vf_data[vf].pf_qos = 0;
5525        }
5526 out:
5527        return err;
5528 }
5529
5530 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5531 {
5532         int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5533         int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5534
5535         return igb_vlvf_set(adapter, vid, add, vf);
5536 }
5537
5538 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5539 {
5540         /* clear flags - except flag that indicates PF has set the MAC */
5541         adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5542         adapter->vf_data[vf].last_nack = jiffies;
5543
5544         /* reset offloads to defaults */
5545         igb_set_vmolr(adapter, vf, true);
5546
5547         /* reset vlans for device */
5548         igb_clear_vf_vfta(adapter, vf);
5549         if (adapter->vf_data[vf].pf_vlan)
5550                 igb_ndo_set_vf_vlan(adapter->netdev, vf,
5551                                     adapter->vf_data[vf].pf_vlan,
5552                                     adapter->vf_data[vf].pf_qos);
5553         else
5554                 igb_clear_vf_vfta(adapter, vf);
5555
5556         /* reset multicast table array for vf */
5557         adapter->vf_data[vf].num_vf_mc_hashes = 0;
5558
5559         /* Flush and reset the mta with the new values */
5560         igb_set_rx_mode(adapter->netdev);
5561 }
5562
5563 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5564 {
5565         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5566
5567         /* clear mac address as we were hotplug removed/added */
5568         if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5569                 eth_zero_addr(vf_mac);
5570
5571         /* process remaining reset events */
5572         igb_vf_reset(adapter, vf);
5573 }
5574
5575 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5576 {
5577         struct e1000_hw *hw = &adapter->hw;
5578         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5579         int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5580         u32 reg, msgbuf[3];
5581         u8 *addr = (u8 *)(&msgbuf[1]);
5582
5583         /* process all the same items cleared in a function level reset */
5584         igb_vf_reset(adapter, vf);
5585
5586         /* set vf mac address */
5587         igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5588
5589         /* enable transmit and receive for vf */
5590         reg = rd32(E1000_VFTE);
5591         wr32(E1000_VFTE, reg | (1 << vf));
5592         reg = rd32(E1000_VFRE);
5593         wr32(E1000_VFRE, reg | (1 << vf));
5594
5595         adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5596
5597         /* reply to reset with ack and vf mac address */
5598         msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5599         memcpy(addr, vf_mac, 6);
5600         igb_write_mbx(hw, msgbuf, 3, vf);
5601 }
5602
5603 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5604 {
5605         /*
5606          * The VF MAC Address is stored in a packed array of bytes
5607          * starting at the second 32 bit word of the msg array
5608          */
5609         unsigned char *addr = (char *)&msg[1];
5610         int err = -1;
5611
5612         if (is_valid_ether_addr(addr))
5613                 err = igb_set_vf_mac(adapter, vf, addr);
5614
5615         return err;
5616 }
5617
5618 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5619 {
5620         struct e1000_hw *hw = &adapter->hw;
5621         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5622         u32 msg = E1000_VT_MSGTYPE_NACK;
5623
5624         /* if device isn't clear to send it shouldn't be reading either */
5625         if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5626             time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5627                 igb_write_mbx(hw, &msg, 1, vf);
5628                 vf_data->last_nack = jiffies;
5629         }
5630 }
5631
5632 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5633 {
5634         struct pci_dev *pdev = adapter->pdev;
5635         u32 msgbuf[E1000_VFMAILBOX_SIZE];
5636         struct e1000_hw *hw = &adapter->hw;
5637         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5638         s32 retval;
5639
5640         retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5641
5642         if (retval) {
5643                 /* if receive failed revoke VF CTS stats and restart init */
5644                 dev_err(&pdev->dev, "Error receiving message from VF\n");
5645                 vf_data->flags &= ~IGB_VF_FLAG_CTS;
5646                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5647                         return;
5648                 goto out;
5649         }
5650
5651         /* this is a message we already processed, do nothing */
5652         if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5653                 return;
5654
5655         /*
5656          * until the vf completes a reset it should not be
5657          * allowed to start any configuration.
5658          */
5659
5660         if (msgbuf[0] == E1000_VF_RESET) {
5661                 igb_vf_reset_msg(adapter, vf);
5662                 return;
5663         }
5664
5665         if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5666                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5667                         return;
5668                 retval = -1;
5669                 goto out;
5670         }
5671
5672         switch ((msgbuf[0] & 0xFFFF)) {
5673         case E1000_VF_SET_MAC_ADDR:
5674                 retval = -EINVAL;
5675                 if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5676                         retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5677                 else
5678                         dev_warn(&pdev->dev,
5679                                  "VF %d attempted to override administratively "
5680                                  "set MAC address\nReload the VF driver to "
5681                                  "resume operations\n", vf);
5682                 break;
5683         case E1000_VF_SET_PROMISC:
5684                 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5685                 break;
5686         case E1000_VF_SET_MULTICAST:
5687                 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5688                 break;
5689         case E1000_VF_SET_LPE:
5690                 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5691                 break;
5692         case E1000_VF_SET_VLAN:
5693                 retval = -1;
5694                 if (vf_data->pf_vlan)
5695                         dev_warn(&pdev->dev,
5696                                  "VF %d attempted to override administratively "
5697                                  "set VLAN tag\nReload the VF driver to "
5698                                  "resume operations\n", vf);
5699                 else
5700                         retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5701                 break;
5702         default:
5703                 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5704                 retval = -1;
5705                 break;
5706         }
5707
5708         msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5709 out:
5710         /* notify the VF of the results of what it sent us */
5711         if (retval)
5712                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5713         else
5714                 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5715
5716         igb_write_mbx(hw, msgbuf, 1, vf);
5717 }
5718
5719 static void igb_msg_task(struct igb_adapter *adapter)
5720 {
5721         struct e1000_hw *hw = &adapter->hw;
5722         u32 vf;
5723
5724         for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5725                 /* process any reset requests */
5726                 if (!igb_check_for_rst(hw, vf))
5727                         igb_vf_reset_event(adapter, vf);
5728
5729                 /* process any messages pending */
5730                 if (!igb_check_for_msg(hw, vf))
5731                         igb_rcv_msg_from_vf(adapter, vf);
5732
5733                 /* process any acks */
5734                 if (!igb_check_for_ack(hw, vf))
5735                         igb_rcv_ack_from_vf(adapter, vf);
5736         }
5737 }
5738
5739 /**
5740  *  igb_set_uta - Set unicast filter table address
5741  *  @adapter: board private structure
5742  *
5743  *  The unicast table address is a register array of 32-bit registers.
5744  *  The table is meant to be used in a way similar to how the MTA is used
5745  *  however due to certain limitations in the hardware it is necessary to
5746  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5747  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
5748  **/
5749 static void igb_set_uta(struct igb_adapter *adapter)
5750 {
5751         struct e1000_hw *hw = &adapter->hw;
5752         int i;
5753
5754         /* The UTA table only exists on 82576 hardware and newer */
5755         if (hw->mac.type < e1000_82576)
5756                 return;
5757
5758         /* we only need to do this if VMDq is enabled */
5759         if (!adapter->vfs_allocated_count)
5760                 return;
5761
5762         for (i = 0; i < hw->mac.uta_reg_count; i++)
5763                 array_wr32(E1000_UTA, i, ~0);
5764 }
5765
5766 /**
5767  * igb_intr_msi - Interrupt Handler
5768  * @irq: interrupt number
5769  * @data: pointer to a network interface device structure
5770  **/
5771 static irqreturn_t igb_intr_msi(int irq, void *data)
5772 {
5773         struct igb_adapter *adapter = data;
5774         struct igb_q_vector *q_vector = adapter->q_vector[0];
5775         struct e1000_hw *hw = &adapter->hw;
5776         /* read ICR disables interrupts using IAM */
5777         u32 icr = rd32(E1000_ICR);
5778
5779         igb_write_itr(q_vector);
5780
5781         if (icr & E1000_ICR_DRSTA)
5782                 schedule_work(&adapter->reset_task);
5783
5784         if (icr & E1000_ICR_DOUTSYNC) {
5785                 /* HW is reporting DMA is out of sync */
5786                 adapter->stats.doosync++;
5787         }
5788
5789         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5790                 hw->mac.get_link_status = 1;
5791                 if (!test_bit(__IGB_DOWN, &adapter->state))
5792                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5793         }
5794
5795         if (icr & E1000_ICR_TS) {
5796                 u32 tsicr = rd32(E1000_TSICR);
5797
5798                 if (tsicr & E1000_TSICR_TXTS) {
5799                         /* acknowledge the interrupt */
5800                         wr32(E1000_TSICR, E1000_TSICR_TXTS);
5801                         /* retrieve hardware timestamp */
5802                         schedule_work(&adapter->ptp_tx_work);
5803                 }
5804         }
5805
5806         napi_schedule(&q_vector->napi);
5807
5808         return IRQ_HANDLED;
5809 }
5810
5811 /**
5812  * igb_intr - Legacy Interrupt Handler
5813  * @irq: interrupt number
5814  * @data: pointer to a network interface device structure
5815  **/
5816 static irqreturn_t igb_intr(int irq, void *data)
5817 {
5818         struct igb_adapter *adapter = data;
5819         struct igb_q_vector *q_vector = adapter->q_vector[0];
5820         struct e1000_hw *hw = &adapter->hw;
5821         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
5822          * need for the IMC write */
5823         u32 icr = rd32(E1000_ICR);
5824
5825         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5826          * not set, then the adapter didn't send an interrupt */
5827         if (!(icr & E1000_ICR_INT_ASSERTED))
5828                 return IRQ_NONE;
5829
5830         igb_write_itr(q_vector);
5831
5832         if (icr & E1000_ICR_DRSTA)
5833                 schedule_work(&adapter->reset_task);
5834
5835         if (icr & E1000_ICR_DOUTSYNC) {
5836                 /* HW is reporting DMA is out of sync */
5837                 adapter->stats.doosync++;
5838         }
5839
5840         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5841                 hw->mac.get_link_status = 1;
5842                 /* guard against interrupt when we're going down */
5843                 if (!test_bit(__IGB_DOWN, &adapter->state))
5844                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5845         }
5846
5847         if (icr & E1000_ICR_TS) {
5848                 u32 tsicr = rd32(E1000_TSICR);
5849
5850                 if (tsicr & E1000_TSICR_TXTS) {
5851                         /* acknowledge the interrupt */
5852                         wr32(E1000_TSICR, E1000_TSICR_TXTS);
5853                         /* retrieve hardware timestamp */
5854                         schedule_work(&adapter->ptp_tx_work);
5855                 }
5856         }
5857
5858         napi_schedule(&q_vector->napi);
5859
5860         return IRQ_HANDLED;
5861 }
5862
5863 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5864 {
5865         struct igb_adapter *adapter = q_vector->adapter;
5866         struct e1000_hw *hw = &adapter->hw;
5867
5868         if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5869             (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5870                 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5871                         igb_set_itr(q_vector);
5872                 else
5873                         igb_update_ring_itr(q_vector);
5874         }
5875
5876         if (!test_bit(__IGB_DOWN, &adapter->state)) {
5877                 if (adapter->msix_entries)
5878                         wr32(E1000_EIMS, q_vector->eims_value);
5879                 else
5880                         igb_irq_enable(adapter);
5881         }
5882 }
5883
5884 /**
5885  * igb_poll - NAPI Rx polling callback
5886  * @napi: napi polling structure
5887  * @budget: count of how many packets we should handle
5888  **/
5889 static int igb_poll(struct napi_struct *napi, int budget)
5890 {
5891         struct igb_q_vector *q_vector = container_of(napi,
5892                                                      struct igb_q_vector,
5893                                                      napi);
5894         bool clean_complete = true;
5895
5896 #ifdef CONFIG_IGB_DCA
5897         if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
5898                 igb_update_dca(q_vector);
5899 #endif
5900         if (q_vector->tx.ring)
5901                 clean_complete = igb_clean_tx_irq(q_vector);
5902
5903         if (q_vector->rx.ring)
5904                 clean_complete &= igb_clean_rx_irq(q_vector, budget);
5905
5906         /* If all work not completed, return budget and keep polling */
5907         if (!clean_complete)
5908                 return budget;
5909
5910         /* If not enough Rx work done, exit the polling mode */
5911         napi_complete(napi);
5912         igb_ring_irq_enable(q_vector);
5913
5914         return 0;
5915 }
5916
5917 /**
5918  * igb_clean_tx_irq - Reclaim resources after transmit completes
5919  * @q_vector: pointer to q_vector containing needed info
5920  *
5921  * returns true if ring is completely cleaned
5922  **/
5923 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
5924 {
5925         struct igb_adapter *adapter = q_vector->adapter;
5926         struct igb_ring *tx_ring = q_vector->tx.ring;
5927         struct igb_tx_buffer *tx_buffer;
5928         union e1000_adv_tx_desc *tx_desc;
5929         unsigned int total_bytes = 0, total_packets = 0;
5930         unsigned int budget = q_vector->tx.work_limit;
5931         unsigned int i = tx_ring->next_to_clean;
5932
5933         if (test_bit(__IGB_DOWN, &adapter->state))
5934                 return true;
5935
5936         tx_buffer = &tx_ring->tx_buffer_info[i];
5937         tx_desc = IGB_TX_DESC(tx_ring, i);
5938         i -= tx_ring->count;
5939
5940         do {
5941                 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
5942
5943                 /* if next_to_watch is not set then there is no work pending */
5944                 if (!eop_desc)
5945                         break;
5946
5947                 /* prevent any other reads prior to eop_desc */
5948                 read_barrier_depends();
5949
5950                 /* if DD is not set pending work has not been completed */
5951                 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
5952                         break;
5953
5954                 /* clear next_to_watch to prevent false hangs */
5955                 tx_buffer->next_to_watch = NULL;
5956
5957                 /* update the statistics for this packet */
5958                 total_bytes += tx_buffer->bytecount;
5959                 total_packets += tx_buffer->gso_segs;
5960
5961                 /* free the skb */
5962                 dev_kfree_skb_any(tx_buffer->skb);
5963
5964                 /* unmap skb header data */
5965                 dma_unmap_single(tx_ring->dev,
5966                                  dma_unmap_addr(tx_buffer, dma),
5967                                  dma_unmap_len(tx_buffer, len),
5968                                  DMA_TO_DEVICE);
5969
5970                 /* clear tx_buffer data */
5971                 tx_buffer->skb = NULL;
5972                 dma_unmap_len_set(tx_buffer, len, 0);
5973
5974                 /* clear last DMA location and unmap remaining buffers */
5975                 while (tx_desc != eop_desc) {
5976                         tx_buffer++;
5977                         tx_desc++;
5978                         i++;
5979                         if (unlikely(!i)) {
5980                                 i -= tx_ring->count;
5981                                 tx_buffer = tx_ring->tx_buffer_info;
5982                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
5983                         }
5984
5985                         /* unmap any remaining paged data */
5986                         if (dma_unmap_len(tx_buffer, len)) {
5987                                 dma_unmap_page(tx_ring->dev,
5988                                                dma_unmap_addr(tx_buffer, dma),
5989                                                dma_unmap_len(tx_buffer, len),
5990                                                DMA_TO_DEVICE);
5991                                 dma_unmap_len_set(tx_buffer, len, 0);
5992                         }
5993                 }
5994
5995                 /* move us one more past the eop_desc for start of next pkt */
5996                 tx_buffer++;
5997                 tx_desc++;
5998                 i++;
5999                 if (unlikely(!i)) {
6000                         i -= tx_ring->count;
6001                         tx_buffer = tx_ring->tx_buffer_info;
6002                         tx_desc = IGB_TX_DESC(tx_ring, 0);
6003                 }
6004
6005                 /* issue prefetch for next Tx descriptor */
6006                 prefetch(tx_desc);
6007
6008                 /* update budget accounting */
6009                 budget--;
6010         } while (likely(budget));
6011
6012         netdev_tx_completed_queue(txring_txq(tx_ring),
6013                                   total_packets, total_bytes);
6014         i += tx_ring->count;
6015         tx_ring->next_to_clean = i;
6016         u64_stats_update_begin(&tx_ring->tx_syncp);
6017         tx_ring->tx_stats.bytes += total_bytes;
6018         tx_ring->tx_stats.packets += total_packets;
6019         u64_stats_update_end(&tx_ring->tx_syncp);
6020         q_vector->tx.total_bytes += total_bytes;
6021         q_vector->tx.total_packets += total_packets;
6022
6023         if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6024                 struct e1000_hw *hw = &adapter->hw;
6025
6026                 /* Detect a transmit hang in hardware, this serializes the
6027                  * check with the clearing of time_stamp and movement of i */
6028                 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6029                 if (tx_buffer->next_to_watch &&
6030                     time_after(jiffies, tx_buffer->time_stamp +
6031                                (adapter->tx_timeout_factor * HZ)) &&
6032                     !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6033
6034                         /* detected Tx unit hang */
6035                         dev_err(tx_ring->dev,
6036                                 "Detected Tx Unit Hang\n"
6037                                 "  Tx Queue             <%d>\n"
6038                                 "  TDH                  <%x>\n"
6039                                 "  TDT                  <%x>\n"
6040                                 "  next_to_use          <%x>\n"
6041                                 "  next_to_clean        <%x>\n"
6042                                 "buffer_info[next_to_clean]\n"
6043                                 "  time_stamp           <%lx>\n"
6044                                 "  next_to_watch        <%p>\n"
6045                                 "  jiffies              <%lx>\n"
6046                                 "  desc.status          <%x>\n",
6047                                 tx_ring->queue_index,
6048                                 rd32(E1000_TDH(tx_ring->reg_idx)),
6049                                 readl(tx_ring->tail),
6050                                 tx_ring->next_to_use,
6051                                 tx_ring->next_to_clean,
6052                                 tx_buffer->time_stamp,
6053                                 tx_buffer->next_to_watch,
6054                                 jiffies,
6055                                 tx_buffer->next_to_watch->wb.status);
6056                         netif_stop_subqueue(tx_ring->netdev,
6057                                             tx_ring->queue_index);
6058
6059                         /* we are about to reset, no point in enabling stuff */
6060                         return true;
6061                 }
6062         }
6063
6064 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6065         if (unlikely(total_packets &&
6066                      netif_carrier_ok(tx_ring->netdev) &&
6067                      igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6068                 /* Make sure that anybody stopping the queue after this
6069                  * sees the new next_to_clean.
6070                  */
6071                 smp_mb();
6072                 if (__netif_subqueue_stopped(tx_ring->netdev,
6073                                              tx_ring->queue_index) &&
6074                     !(test_bit(__IGB_DOWN, &adapter->state))) {
6075                         netif_wake_subqueue(tx_ring->netdev,
6076                                             tx_ring->queue_index);
6077
6078                         u64_stats_update_begin(&tx_ring->tx_syncp);
6079                         tx_ring->tx_stats.restart_queue++;
6080                         u64_stats_update_end(&tx_ring->tx_syncp);
6081                 }
6082         }
6083
6084         return !!budget;
6085 }
6086
6087 /**
6088  * igb_reuse_rx_page - page flip buffer and store it back on the ring
6089  * @rx_ring: rx descriptor ring to store buffers on
6090  * @old_buff: donor buffer to have page reused
6091  *
6092  * Synchronizes page for reuse by the adapter
6093  **/
6094 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6095                               struct igb_rx_buffer *old_buff)
6096 {
6097         struct igb_rx_buffer *new_buff;
6098         u16 nta = rx_ring->next_to_alloc;
6099
6100         new_buff = &rx_ring->rx_buffer_info[nta];
6101
6102         /* update, and store next to alloc */
6103         nta++;
6104         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6105
6106         /* transfer page from old buffer to new buffer */
6107         memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6108
6109         /* sync the buffer for use by the device */
6110         dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6111                                          old_buff->page_offset,
6112                                          IGB_RX_BUFSZ,
6113                                          DMA_FROM_DEVICE);
6114 }
6115
6116 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6117                                   struct page *page,
6118                                   unsigned int truesize)
6119 {
6120         /* avoid re-using remote pages */
6121         if (unlikely(page_to_nid(page) != numa_node_id()))
6122                 return false;
6123
6124 #if (PAGE_SIZE < 8192)
6125         /* if we are only owner of page we can reuse it */
6126         if (unlikely(page_count(page) != 1))
6127                 return false;
6128
6129         /* flip page offset to other buffer */
6130         rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6131
6132         /* since we are the only owner of the page and we need to
6133          * increment it, just set the value to 2 in order to avoid
6134          * an unnecessary locked operation
6135          */
6136         atomic_set(&page->_count, 2);
6137 #else
6138         /* move offset up to the next cache line */
6139         rx_buffer->page_offset += truesize;
6140
6141         if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6142                 return false;
6143
6144         /* bump ref count on page before it is given to the stack */
6145         get_page(page);
6146 #endif
6147
6148         return true;
6149 }
6150
6151 /**
6152  * igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6153  * @rx_ring: rx descriptor ring to transact packets on
6154  * @rx_buffer: buffer containing page to add
6155  * @rx_desc: descriptor containing length of buffer written by hardware
6156  * @skb: sk_buff to place the data into
6157  *
6158  * This function will add the data contained in rx_buffer->page to the skb.
6159  * This is done either through a direct copy if the data in the buffer is
6160  * less than the skb header size, otherwise it will just attach the page as
6161  * a frag to the skb.
6162  *
6163  * The function will then update the page offset if necessary and return
6164  * true if the buffer can be reused by the adapter.
6165  **/
6166 static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6167                             struct igb_rx_buffer *rx_buffer,
6168                             union e1000_adv_rx_desc *rx_desc,
6169                             struct sk_buff *skb)
6170 {
6171         struct page *page = rx_buffer->page;
6172         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6173 #if (PAGE_SIZE < 8192)
6174         unsigned int truesize = IGB_RX_BUFSZ;
6175 #else
6176         unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6177 #endif
6178
6179         if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6180                 unsigned char *va = page_address(page) + rx_buffer->page_offset;
6181
6182                 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6183                         igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6184                         va += IGB_TS_HDR_LEN;
6185                         size -= IGB_TS_HDR_LEN;
6186                 }
6187
6188                 memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6189
6190                 /* we can reuse buffer as-is, just make sure it is local */
6191                 if (likely(page_to_nid(page) == numa_node_id()))
6192                         return true;
6193
6194                 /* this page cannot be reused so discard it */
6195                 put_page(page);
6196                 return false;
6197         }
6198
6199         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6200                         rx_buffer->page_offset, size, truesize);
6201
6202         return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6203 }
6204
6205 static struct sk_buff *igb_build_rx_buffer(struct igb_ring *rx_ring,
6206                                            union e1000_adv_rx_desc *rx_desc)
6207 {
6208         struct igb_rx_buffer *rx_buffer;
6209         struct sk_buff *skb;
6210         struct page *page;
6211         void *page_addr;
6212         unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6213 #if (PAGE_SIZE < 8192)
6214         unsigned int truesize = IGB_RX_BUFSZ;
6215 #else
6216         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
6217                                 SKB_DATA_ALIGN(NET_SKB_PAD +
6218                                                NET_IP_ALIGN +
6219                                                size);
6220 #endif
6221
6222         /* If we spanned a buffer we have a huge mess so test for it */
6223         BUG_ON(unlikely(!igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)));
6224
6225         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6226         page = rx_buffer->page;
6227         prefetchw(page);
6228
6229         page_addr = page_address(page) + rx_buffer->page_offset;
6230
6231         /* prefetch first cache line of first page */
6232         prefetch(page_addr + NET_SKB_PAD + NET_IP_ALIGN);
6233 #if L1_CACHE_BYTES < 128
6234         prefetch(page_addr + L1_CACHE_BYTES + NET_SKB_PAD + NET_IP_ALIGN);
6235 #endif
6236
6237         /* build an skb to around the page buffer */
6238         skb = build_skb(page_addr, truesize);
6239         if (unlikely(!skb)) {
6240                 rx_ring->rx_stats.alloc_failed++;
6241                 return NULL;
6242         }
6243
6244         /* we are reusing so sync this buffer for CPU use */
6245         dma_sync_single_range_for_cpu(rx_ring->dev,
6246                                       rx_buffer->dma,
6247                                       rx_buffer->page_offset,
6248                                       IGB_RX_BUFSZ,
6249                                       DMA_FROM_DEVICE);
6250
6251         /* update pointers within the skb to store the data */
6252         skb_reserve(skb, NET_IP_ALIGN + NET_SKB_PAD);
6253         __skb_put(skb, size);
6254
6255         /* pull timestamp out of packet data */
6256         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6257                 igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
6258                 __skb_pull(skb, IGB_TS_HDR_LEN);
6259         }
6260
6261         if (igb_can_reuse_rx_page(rx_buffer, page, truesize)) {
6262                 /* hand second half of page back to the ring */
6263                 igb_reuse_rx_page(rx_ring, rx_buffer);
6264         } else {
6265                 /* we are not reusing the buffer so unmap it */
6266                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6267                                PAGE_SIZE, DMA_FROM_DEVICE);
6268         }
6269
6270         /* clear contents of buffer_info */
6271         rx_buffer->dma = 0;
6272         rx_buffer->page = NULL;
6273
6274         return skb;
6275 }
6276
6277 static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6278                                            union e1000_adv_rx_desc *rx_desc,
6279                                            struct sk_buff *skb)
6280 {
6281         struct igb_rx_buffer *rx_buffer;
6282         struct page *page;
6283
6284         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6285
6286         page = rx_buffer->page;
6287         prefetchw(page);
6288
6289         if (likely(!skb)) {
6290                 void *page_addr = page_address(page) +
6291                                   rx_buffer->page_offset;
6292
6293                 /* prefetch first cache line of first page */
6294                 prefetch(page_addr);
6295 #if L1_CACHE_BYTES < 128
6296                 prefetch(page_addr + L1_CACHE_BYTES);
6297 #endif
6298
6299                 /* allocate a skb to store the frags */
6300                 skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6301                                                 IGB_RX_HDR_LEN);
6302                 if (unlikely(!skb)) {
6303                         rx_ring->rx_stats.alloc_failed++;
6304                         return NULL;
6305                 }
6306
6307                 /*
6308                  * we will be copying header into skb->data in
6309                  * pskb_may_pull so it is in our interest to prefetch
6310                  * it now to avoid a possible cache miss
6311                  */
6312                 prefetchw(skb->data);
6313         }
6314
6315         /* we are reusing so sync this buffer for CPU use */
6316         dma_sync_single_range_for_cpu(rx_ring->dev,
6317                                       rx_buffer->dma,
6318                                       rx_buffer->page_offset,
6319                                       IGB_RX_BUFSZ,
6320                                       DMA_FROM_DEVICE);
6321
6322         /* pull page into skb */
6323         if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6324                 /* hand second half of page back to the ring */
6325                 igb_reuse_rx_page(rx_ring, rx_buffer);
6326         } else {
6327                 /* we are not reusing the buffer so unmap it */
6328                 dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6329                                PAGE_SIZE, DMA_FROM_DEVICE);
6330         }
6331
6332         /* clear contents of rx_buffer */
6333         rx_buffer->page = NULL;
6334
6335         return skb;
6336 }
6337
6338 static inline void igb_rx_checksum(struct igb_ring *ring,
6339                                    union e1000_adv_rx_desc *rx_desc,
6340                                    struct sk_buff *skb)
6341 {
6342         skb_checksum_none_assert(skb);
6343
6344         /* Ignore Checksum bit is set */
6345         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6346                 return;
6347
6348         /* Rx checksum disabled via ethtool */
6349         if (!(ring->netdev->features & NETIF_F_RXCSUM))
6350                 return;
6351
6352         /* TCP/UDP checksum error bit is set */
6353         if (igb_test_staterr(rx_desc,
6354                              E1000_RXDEXT_STATERR_TCPE |
6355                              E1000_RXDEXT_STATERR_IPE)) {
6356                 /*
6357                  * work around errata with sctp packets where the TCPE aka
6358                  * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6359                  * packets, (aka let the stack check the crc32c)
6360                  */
6361                 if (!((skb->len == 60) &&
6362                       test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6363                         u64_stats_update_begin(&ring->rx_syncp);
6364                         ring->rx_stats.csum_err++;
6365                         u64_stats_update_end(&ring->rx_syncp);
6366                 }
6367                 /* let the stack verify checksum errors */
6368                 return;
6369         }
6370         /* It must be a TCP or UDP packet with a valid checksum */
6371         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6372                                       E1000_RXD_STAT_UDPCS))
6373                 skb->ip_summed = CHECKSUM_UNNECESSARY;
6374
6375         dev_dbg(ring->dev, "cksum success: bits %08X\n",
6376                 le32_to_cpu(rx_desc->wb.upper.status_error));
6377 }
6378
6379 static inline void igb_rx_hash(struct igb_ring *ring,
6380                                union e1000_adv_rx_desc *rx_desc,
6381                                struct sk_buff *skb)
6382 {
6383         if (ring->netdev->features & NETIF_F_RXHASH)
6384                 skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6385 }
6386
6387 /**
6388  * igb_is_non_eop - process handling of non-EOP buffers
6389  * @rx_ring: Rx ring being processed
6390  * @rx_desc: Rx descriptor for current buffer
6391  * @skb: current socket buffer containing buffer in progress
6392  *
6393  * This function updates next to clean.  If the buffer is an EOP buffer
6394  * this function exits returning false, otherwise it will place the
6395  * sk_buff in the next buffer to be chained and return true indicating
6396  * that this is in fact a non-EOP buffer.
6397  **/
6398 static bool igb_is_non_eop(struct igb_ring *rx_ring,
6399                            union e1000_adv_rx_desc *rx_desc)
6400 {
6401         u32 ntc = rx_ring->next_to_clean + 1;
6402
6403         /* fetch, update, and store next to clean */
6404         ntc = (ntc < rx_ring->count) ? ntc : 0;
6405         rx_ring->next_to_clean = ntc;
6406
6407         prefetch(IGB_RX_DESC(rx_ring, ntc));
6408
6409         if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6410                 return false;
6411
6412         return true;
6413 }
6414
6415 /**
6416  * igb_get_headlen - determine size of header for LRO/GRO
6417  * @data: pointer to the start of the headers
6418  * @max_len: total length of section to find headers in
6419  *
6420  * This function is meant to determine the length of headers that will
6421  * be recognized by hardware for LRO, and GRO offloads.  The main
6422  * motivation of doing this is to only perform one pull for IPv4 TCP
6423  * packets so that we can do basic things like calculating the gso_size
6424  * based on the average data per packet.
6425  **/
6426 static unsigned int igb_get_headlen(unsigned char *data,
6427                                     unsigned int max_len)
6428 {
6429         union {
6430                 unsigned char *network;
6431                 /* l2 headers */
6432                 struct ethhdr *eth;
6433                 struct vlan_hdr *vlan;
6434                 /* l3 headers */
6435                 struct iphdr *ipv4;
6436                 struct ipv6hdr *ipv6;
6437         } hdr;
6438         __be16 protocol;
6439         u8 nexthdr = 0; /* default to not TCP */
6440         u8 hlen;
6441
6442         /* this should never happen, but better safe than sorry */
6443         if (max_len < ETH_HLEN)
6444                 return max_len;
6445
6446         /* initialize network frame pointer */
6447         hdr.network = data;
6448
6449         /* set first protocol and move network header forward */
6450         protocol = hdr.eth->h_proto;
6451         hdr.network += ETH_HLEN;
6452
6453         /* handle any vlan tag if present */
6454         if (protocol == __constant_htons(ETH_P_8021Q)) {
6455                 if ((hdr.network - data) > (max_len - VLAN_HLEN))
6456                         return max_len;
6457
6458                 protocol = hdr.vlan->h_vlan_encapsulated_proto;
6459                 hdr.network += VLAN_HLEN;
6460         }
6461
6462         /* handle L3 protocols */
6463         if (protocol == __constant_htons(ETH_P_IP)) {
6464                 if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6465                         return max_len;
6466
6467                 /* access ihl as a u8 to avoid unaligned access on ia64 */
6468                 hlen = (hdr.network[0] & 0x0F) << 2;
6469
6470                 /* verify hlen meets minimum size requirements */
6471                 if (hlen < sizeof(struct iphdr))
6472                         return hdr.network - data;
6473
6474                 /* record next protocol if header is present */
6475                 if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6476                         nexthdr = hdr.ipv4->protocol;
6477         } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6478                 if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6479                         return max_len;
6480
6481                 /* record next protocol */
6482                 nexthdr = hdr.ipv6->nexthdr;
6483                 hlen = sizeof(struct ipv6hdr);
6484         } else {
6485                 return hdr.network - data;
6486         }
6487
6488         /* relocate pointer to start of L4 header */
6489         hdr.network += hlen;
6490
6491         /* finally sort out TCP */
6492         if (nexthdr == IPPROTO_TCP) {
6493                 if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6494                         return max_len;
6495
6496                 /* access doff as a u8 to avoid unaligned access on ia64 */
6497                 hlen = (hdr.network[12] & 0xF0) >> 2;
6498
6499                 /* verify hlen meets minimum size requirements */
6500                 if (hlen < sizeof(struct tcphdr))
6501                         return hdr.network - data;
6502
6503                 hdr.network += hlen;
6504         } else if (nexthdr == IPPROTO_UDP) {
6505                 if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6506                         return max_len;
6507
6508                 hdr.network += sizeof(struct udphdr);
6509         }
6510
6511         /*
6512          * If everything has gone correctly hdr.network should be the
6513          * data section of the packet and will be the end of the header.
6514          * If not then it probably represents the end of the last recognized
6515          * header.
6516          */
6517         if ((hdr.network - data) < max_len)
6518                 return hdr.network - data;
6519         else
6520                 return max_len;
6521 }
6522
6523 /**
6524  * igb_pull_tail - igb specific version of skb_pull_tail
6525  * @rx_ring: rx descriptor ring packet is being transacted on
6526  * @rx_desc: pointer to the EOP Rx descriptor
6527  * @skb: pointer to current skb being adjusted
6528  *
6529  * This function is an igb specific version of __pskb_pull_tail.  The
6530  * main difference between this version and the original function is that
6531  * this function can make several assumptions about the state of things
6532  * that allow for significant optimizations versus the standard function.
6533  * As a result we can do things like drop a frag and maintain an accurate
6534  * truesize for the skb.
6535  */
6536 static void igb_pull_tail(struct igb_ring *rx_ring,
6537                           union e1000_adv_rx_desc *rx_desc,
6538                           struct sk_buff *skb)
6539 {
6540         struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6541         unsigned char *va;
6542         unsigned int pull_len;
6543
6544         /*
6545          * it is valid to use page_address instead of kmap since we are
6546          * working with pages allocated out of the lomem pool per
6547          * alloc_page(GFP_ATOMIC)
6548          */
6549         va = skb_frag_address(frag);
6550
6551         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6552                 /* retrieve timestamp from buffer */
6553                 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6554
6555                 /* update pointers to remove timestamp header */
6556                 skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6557                 frag->page_offset += IGB_TS_HDR_LEN;
6558                 skb->data_len -= IGB_TS_HDR_LEN;
6559                 skb->len -= IGB_TS_HDR_LEN;
6560
6561                 /* move va to start of packet data */
6562                 va += IGB_TS_HDR_LEN;
6563         }
6564
6565         /*
6566          * we need the header to contain the greater of either ETH_HLEN or
6567          * 60 bytes if the skb->len is less than 60 for skb_pad.
6568          */
6569         pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6570
6571         /* align pull length to size of long to optimize memcpy performance */
6572         skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6573
6574         /* update all of the pointers */
6575         skb_frag_size_sub(frag, pull_len);
6576         frag->page_offset += pull_len;
6577         skb->data_len -= pull_len;
6578         skb->tail += pull_len;
6579 }
6580
6581 /**
6582  * igb_cleanup_headers - Correct corrupted or empty headers
6583  * @rx_ring: rx descriptor ring packet is being transacted on
6584  * @rx_desc: pointer to the EOP Rx descriptor
6585  * @skb: pointer to current skb being fixed
6586  *
6587  * Address the case where we are pulling data in on pages only
6588  * and as such no data is present in the skb header.
6589  *
6590  * In addition if skb is not at least 60 bytes we need to pad it so that
6591  * it is large enough to qualify as a valid Ethernet frame.
6592  *
6593  * Returns true if an error was encountered and skb was freed.
6594  **/
6595 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6596                                 union e1000_adv_rx_desc *rx_desc,
6597                                 struct sk_buff *skb)
6598 {
6599
6600         if (unlikely((igb_test_staterr(rx_desc,
6601                                        E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6602                 struct net_device *netdev = rx_ring->netdev;
6603                 if (!(netdev->features & NETIF_F_RXALL)) {
6604                         dev_kfree_skb_any(skb);
6605                         return true;
6606                 }
6607         }
6608
6609         /* place header in linear portion of buffer */
6610         if (skb_is_nonlinear(skb))
6611                 igb_pull_tail(rx_ring, rx_desc, skb);
6612
6613         /* if skb_pad returns an error the skb was freed */
6614         if (unlikely(skb->len < 60)) {
6615                 int pad_len = 60 - skb->len;
6616
6617                 if (skb_pad(skb, pad_len))
6618                         return true;
6619                 __skb_put(skb, pad_len);
6620         }
6621
6622         return false;
6623 }
6624
6625 /**
6626  * igb_process_skb_fields - Populate skb header fields from Rx descriptor
6627  * @rx_ring: rx descriptor ring packet is being transacted on
6628  * @rx_desc: pointer to the EOP Rx descriptor
6629  * @skb: pointer to current skb being populated
6630  *
6631  * This function checks the ring, descriptor, and packet information in
6632  * order to populate the hash, checksum, VLAN, timestamp, protocol, and
6633  * other fields within the skb.
6634  **/
6635 static void igb_process_skb_fields(struct igb_ring *rx_ring,
6636                                    union e1000_adv_rx_desc *rx_desc,
6637                                    struct sk_buff *skb)
6638 {
6639         struct net_device *dev = rx_ring->netdev;
6640
6641         igb_rx_hash(rx_ring, rx_desc, skb);
6642
6643         igb_rx_checksum(rx_ring, rx_desc, skb);
6644
6645         igb_ptp_rx_hwtstamp(rx_ring->q_vector, rx_desc, skb);
6646
6647         if ((dev->features & NETIF_F_HW_VLAN_RX) &&
6648             igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6649                 u16 vid;
6650                 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6651                     test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6652                         vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6653                 else
6654                         vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6655
6656                 __vlan_hwaccel_put_tag(skb, vid);
6657         }
6658
6659         skb_record_rx_queue(skb, rx_ring->queue_index);
6660
6661         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6662 }
6663
6664 static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6665 {
6666         struct igb_ring *rx_ring = q_vector->rx.ring;
6667         struct sk_buff *skb = rx_ring->skb;
6668         unsigned int total_bytes = 0, total_packets = 0;
6669         u16 cleaned_count = igb_desc_unused(rx_ring);
6670
6671         do {
6672                 union e1000_adv_rx_desc *rx_desc;
6673
6674                 /* return some buffers to hardware, one at a time is too slow */
6675                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6676                         igb_alloc_rx_buffers(rx_ring, cleaned_count);
6677                         cleaned_count = 0;
6678                 }
6679
6680                 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6681
6682                 if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6683                         break;
6684
6685                 /* This memory barrier is needed to keep us from reading
6686                  * any other fields out of the rx_desc until we know the
6687                  * RXD_STAT_DD bit is set
6688                  */
6689                 rmb();
6690
6691                 /* retrieve a buffer from the ring */
6692                 if (ring_uses_build_skb(rx_ring))
6693                         skb = igb_build_rx_buffer(rx_ring, rx_desc);
6694                 else
6695                         skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6696
6697                 /* exit if we failed to retrieve a buffer */
6698                 if (!skb)
6699                         break;
6700
6701                 cleaned_count++;
6702
6703                 /* fetch next buffer in frame if non-eop */
6704                 if (igb_is_non_eop(rx_ring, rx_desc))
6705                         continue;
6706
6707                 /* verify the packet layout is correct */
6708                 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6709                         skb = NULL;
6710                         continue;
6711                 }
6712
6713                 /* probably a little skewed due to removing CRC */
6714                 total_bytes += skb->len;
6715
6716                 /* populate checksum, timestamp, VLAN, and protocol */
6717                 igb_process_skb_fields(rx_ring, rx_desc, skb);
6718
6719                 napi_gro_receive(&q_vector->napi, skb);
6720
6721                 /* reset skb pointer */
6722                 skb = NULL;
6723
6724                 /* update budget accounting */
6725                 total_packets++;
6726         } while (likely(total_packets < budget));
6727
6728         /* place incomplete frames back on ring for completion */
6729         rx_ring->skb = skb;
6730
6731         u64_stats_update_begin(&rx_ring->rx_syncp);
6732         rx_ring->rx_stats.packets += total_packets;
6733         rx_ring->rx_stats.bytes += total_bytes;
6734         u64_stats_update_end(&rx_ring->rx_syncp);
6735         q_vector->rx.total_packets += total_packets;
6736         q_vector->rx.total_bytes += total_bytes;
6737
6738         if (cleaned_count)
6739                 igb_alloc_rx_buffers(rx_ring, cleaned_count);
6740
6741         return (total_packets < budget);
6742 }
6743
6744 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6745                                   struct igb_rx_buffer *bi)
6746 {
6747         struct page *page = bi->page;
6748         dma_addr_t dma;
6749
6750         /* since we are recycling buffers we should seldom need to alloc */
6751         if (likely(page))
6752                 return true;
6753
6754         /* alloc new page for storage */
6755         page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6756         if (unlikely(!page)) {
6757                 rx_ring->rx_stats.alloc_failed++;
6758                 return false;
6759         }
6760
6761         /* map page for use */
6762         dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6763
6764         /*
6765          * if mapping failed free memory back to system since
6766          * there isn't much point in holding memory we can't use
6767          */
6768         if (dma_mapping_error(rx_ring->dev, dma)) {
6769                 __free_page(page);
6770
6771                 rx_ring->rx_stats.alloc_failed++;
6772                 return false;
6773         }
6774
6775         bi->dma = dma;
6776         bi->page = page;
6777         bi->page_offset = 0;
6778
6779         return true;
6780 }
6781
6782 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
6783 {
6784         if (ring_uses_build_skb(rx_ring))
6785                 return NET_SKB_PAD + NET_IP_ALIGN;
6786         else
6787                 return 0;
6788 }
6789
6790 /**
6791  * igb_alloc_rx_buffers - Replace used receive buffers; packet split
6792  * @adapter: address of board private structure
6793  **/
6794 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6795 {
6796         union e1000_adv_rx_desc *rx_desc;
6797         struct igb_rx_buffer *bi;
6798         u16 i = rx_ring->next_to_use;
6799
6800         /* nothing to do */
6801         if (!cleaned_count)
6802                 return;
6803
6804         rx_desc = IGB_RX_DESC(rx_ring, i);
6805         bi = &rx_ring->rx_buffer_info[i];
6806         i -= rx_ring->count;
6807
6808         do {
6809                 if (!igb_alloc_mapped_page(rx_ring, bi))
6810                         break;
6811
6812                 /*
6813                  * Refresh the desc even if buffer_addrs didn't change
6814                  * because each write-back erases this info.
6815                  */
6816                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma +
6817                                                      bi->page_offset +
6818                                                      igb_rx_offset(rx_ring));
6819
6820                 rx_desc++;
6821                 bi++;
6822                 i++;
6823                 if (unlikely(!i)) {
6824                         rx_desc = IGB_RX_DESC(rx_ring, 0);
6825                         bi = rx_ring->rx_buffer_info;
6826                         i -= rx_ring->count;
6827                 }
6828
6829                 /* clear the hdr_addr for the next_to_use descriptor */
6830                 rx_desc->read.hdr_addr = 0;
6831
6832                 cleaned_count--;
6833         } while (cleaned_count);
6834
6835         i += rx_ring->count;
6836
6837         if (rx_ring->next_to_use != i) {
6838                 /* record the next descriptor to use */
6839                 rx_ring->next_to_use = i;
6840
6841                 /* update next to alloc since we have filled the ring */
6842                 rx_ring->next_to_alloc = i;
6843
6844                 /*
6845                  * Force memory writes to complete before letting h/w
6846                  * know there are new descriptors to fetch.  (Only
6847                  * applicable for weak-ordered memory model archs,
6848                  * such as IA-64).
6849                  */
6850                 wmb();
6851                 writel(i, rx_ring->tail);
6852         }
6853 }
6854
6855 /**
6856  * igb_mii_ioctl -
6857  * @netdev:
6858  * @ifreq:
6859  * @cmd:
6860  **/
6861 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6862 {
6863         struct igb_adapter *adapter = netdev_priv(netdev);
6864         struct mii_ioctl_data *data = if_mii(ifr);
6865
6866         if (adapter->hw.phy.media_type != e1000_media_type_copper)
6867                 return -EOPNOTSUPP;
6868
6869         switch (cmd) {
6870         case SIOCGMIIPHY:
6871                 data->phy_id = adapter->hw.phy.addr;
6872                 break;
6873         case SIOCGMIIREG:
6874                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6875                                      &data->val_out))
6876                         return -EIO;
6877                 break;
6878         case SIOCSMIIREG:
6879         default:
6880                 return -EOPNOTSUPP;
6881         }
6882         return 0;
6883 }
6884
6885 /**
6886  * igb_ioctl -
6887  * @netdev:
6888  * @ifreq:
6889  * @cmd:
6890  **/
6891 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6892 {
6893         switch (cmd) {
6894         case SIOCGMIIPHY:
6895         case SIOCGMIIREG:
6896         case SIOCSMIIREG:
6897                 return igb_mii_ioctl(netdev, ifr, cmd);
6898         case SIOCSHWTSTAMP:
6899                 return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6900         default:
6901                 return -EOPNOTSUPP;
6902         }
6903 }
6904
6905 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6906 {
6907         struct igb_adapter *adapter = hw->back;
6908
6909         if (pcie_capability_read_word(adapter->pdev, reg, value))
6910                 return -E1000_ERR_CONFIG;
6911
6912         return 0;
6913 }
6914
6915 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6916 {
6917         struct igb_adapter *adapter = hw->back;
6918
6919         if (pcie_capability_write_word(adapter->pdev, reg, *value))
6920                 return -E1000_ERR_CONFIG;
6921
6922         return 0;
6923 }
6924
6925 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6926 {
6927         struct igb_adapter *adapter = netdev_priv(netdev);
6928         struct e1000_hw *hw = &adapter->hw;
6929         u32 ctrl, rctl;
6930         bool enable = !!(features & NETIF_F_HW_VLAN_RX);
6931
6932         if (enable) {
6933                 /* enable VLAN tag insert/strip */
6934                 ctrl = rd32(E1000_CTRL);
6935                 ctrl |= E1000_CTRL_VME;
6936                 wr32(E1000_CTRL, ctrl);
6937
6938                 /* Disable CFI check */
6939                 rctl = rd32(E1000_RCTL);
6940                 rctl &= ~E1000_RCTL_CFIEN;
6941                 wr32(E1000_RCTL, rctl);
6942         } else {
6943                 /* disable VLAN tag insert/strip */
6944                 ctrl = rd32(E1000_CTRL);
6945                 ctrl &= ~E1000_CTRL_VME;
6946                 wr32(E1000_CTRL, ctrl);
6947         }
6948
6949         igb_rlpml_set(adapter);
6950 }
6951
6952 static int igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
6953 {
6954         struct igb_adapter *adapter = netdev_priv(netdev);
6955         struct e1000_hw *hw = &adapter->hw;
6956         int pf_id = adapter->vfs_allocated_count;
6957
6958         /* attempt to add filter to vlvf array */
6959         igb_vlvf_set(adapter, vid, true, pf_id);
6960
6961         /* add the filter since PF can receive vlans w/o entry in vlvf */
6962         igb_vfta_set(hw, vid, true);
6963
6964         set_bit(vid, adapter->active_vlans);
6965
6966         return 0;
6967 }
6968
6969 static int igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
6970 {
6971         struct igb_adapter *adapter = netdev_priv(netdev);
6972         struct e1000_hw *hw = &adapter->hw;
6973         int pf_id = adapter->vfs_allocated_count;
6974         s32 err;
6975
6976         /* remove vlan from VLVF table array */
6977         err = igb_vlvf_set(adapter, vid, false, pf_id);
6978
6979         /* if vid was not present in VLVF just remove it from table */
6980         if (err)
6981                 igb_vfta_set(hw, vid, false);
6982
6983         clear_bit(vid, adapter->active_vlans);
6984
6985         return 0;
6986 }
6987
6988 static void igb_restore_vlan(struct igb_adapter *adapter)
6989 {
6990         u16 vid;
6991
6992         igb_vlan_mode(adapter->netdev, adapter->netdev->features);
6993
6994         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
6995                 igb_vlan_rx_add_vid(adapter->netdev, vid);
6996 }
6997
6998 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
6999 {
7000         struct pci_dev *pdev = adapter->pdev;
7001         struct e1000_mac_info *mac = &adapter->hw.mac;
7002
7003         mac->autoneg = 0;
7004
7005         /* Make sure dplx is at most 1 bit and lsb of speed is not set
7006          * for the switch() below to work */
7007         if ((spd & 1) || (dplx & ~1))
7008                 goto err_inval;
7009
7010         /* Fiber NIC's only allow 1000 gbps Full duplex
7011          * and 100Mbps Full duplex for 100baseFx sfp
7012          */
7013         if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7014                 switch (spd + dplx) {
7015                 case SPEED_10 + DUPLEX_HALF:
7016                 case SPEED_10 + DUPLEX_FULL:
7017                 case SPEED_100 + DUPLEX_HALF:
7018                         goto err_inval;
7019                 default:
7020                         break;
7021                 }
7022         }
7023
7024         switch (spd + dplx) {
7025         case SPEED_10 + DUPLEX_HALF:
7026                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7027                 break;
7028         case SPEED_10 + DUPLEX_FULL:
7029                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7030                 break;
7031         case SPEED_100 + DUPLEX_HALF:
7032                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7033                 break;
7034         case SPEED_100 + DUPLEX_FULL:
7035                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7036                 break;
7037         case SPEED_1000 + DUPLEX_FULL:
7038                 mac->autoneg = 1;
7039                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7040                 break;
7041         case SPEED_1000 + DUPLEX_HALF: /* not supported */
7042         default:
7043                 goto err_inval;
7044         }
7045
7046         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7047         adapter->hw.phy.mdix = AUTO_ALL_MODES;
7048
7049         return 0;
7050
7051 err_inval:
7052         dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7053         return -EINVAL;
7054 }
7055
7056 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7057                           bool runtime)
7058 {
7059         struct net_device *netdev = pci_get_drvdata(pdev);
7060         struct igb_adapter *adapter = netdev_priv(netdev);
7061         struct e1000_hw *hw = &adapter->hw;
7062         u32 ctrl, rctl, status;
7063         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7064 #ifdef CONFIG_PM
7065         int retval = 0;
7066 #endif
7067
7068         netif_device_detach(netdev);
7069
7070         if (netif_running(netdev))
7071                 __igb_close(netdev, true);
7072
7073         igb_clear_interrupt_scheme(adapter);
7074
7075 #ifdef CONFIG_PM
7076         retval = pci_save_state(pdev);
7077         if (retval)
7078                 return retval;
7079 #endif
7080
7081         status = rd32(E1000_STATUS);
7082         if (status & E1000_STATUS_LU)
7083                 wufc &= ~E1000_WUFC_LNKC;
7084
7085         if (wufc) {
7086                 igb_setup_rctl(adapter);
7087                 igb_set_rx_mode(netdev);
7088
7089                 /* turn on all-multi mode if wake on multicast is enabled */
7090                 if (wufc & E1000_WUFC_MC) {
7091                         rctl = rd32(E1000_RCTL);
7092                         rctl |= E1000_RCTL_MPE;
7093                         wr32(E1000_RCTL, rctl);
7094                 }
7095
7096                 ctrl = rd32(E1000_CTRL);
7097                 /* advertise wake from D3Cold */
7098                 #define E1000_CTRL_ADVD3WUC 0x00100000
7099                 /* phy power management enable */
7100                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7101                 ctrl |= E1000_CTRL_ADVD3WUC;
7102                 wr32(E1000_CTRL, ctrl);
7103
7104                 /* Allow time for pending master requests to run */
7105                 igb_disable_pcie_master(hw);
7106
7107                 wr32(E1000_WUC, E1000_WUC_PME_EN);
7108                 wr32(E1000_WUFC, wufc);
7109         } else {
7110                 wr32(E1000_WUC, 0);
7111                 wr32(E1000_WUFC, 0);
7112         }
7113
7114         *enable_wake = wufc || adapter->en_mng_pt;
7115         if (!*enable_wake)
7116                 igb_power_down_link(adapter);
7117         else
7118                 igb_power_up_link(adapter);
7119
7120         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
7121          * would have already happened in close and is redundant. */
7122         igb_release_hw_control(adapter);
7123
7124         pci_disable_device(pdev);
7125
7126         return 0;
7127 }
7128
7129 #ifdef CONFIG_PM
7130 #ifdef CONFIG_PM_SLEEP
7131 static int igb_suspend(struct device *dev)
7132 {
7133         int retval;
7134         bool wake;
7135         struct pci_dev *pdev = to_pci_dev(dev);
7136
7137         retval = __igb_shutdown(pdev, &wake, 0);
7138         if (retval)
7139                 return retval;
7140
7141         if (wake) {
7142                 pci_prepare_to_sleep(pdev);
7143         } else {
7144                 pci_wake_from_d3(pdev, false);
7145                 pci_set_power_state(pdev, PCI_D3hot);
7146         }
7147
7148         return 0;
7149 }
7150 #endif /* CONFIG_PM_SLEEP */
7151
7152 static int igb_resume(struct device *dev)
7153 {
7154         struct pci_dev *pdev = to_pci_dev(dev);
7155         struct net_device *netdev = pci_get_drvdata(pdev);
7156         struct igb_adapter *adapter = netdev_priv(netdev);
7157         struct e1000_hw *hw = &adapter->hw;
7158         u32 err;
7159
7160         pci_set_power_state(pdev, PCI_D0);
7161         pci_restore_state(pdev);
7162         pci_save_state(pdev);
7163
7164         err = pci_enable_device_mem(pdev);
7165         if (err) {
7166                 dev_err(&pdev->dev,
7167                         "igb: Cannot enable PCI device from suspend\n");
7168                 return err;
7169         }
7170         pci_set_master(pdev);
7171
7172         pci_enable_wake(pdev, PCI_D3hot, 0);
7173         pci_enable_wake(pdev, PCI_D3cold, 0);
7174
7175         if (igb_init_interrupt_scheme(adapter, true)) {
7176                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7177                 return -ENOMEM;
7178         }
7179
7180         igb_reset(adapter);
7181
7182         /* let the f/w know that the h/w is now under the control of the
7183          * driver. */
7184         igb_get_hw_control(adapter);
7185
7186         wr32(E1000_WUS, ~0);
7187
7188         if (netdev->flags & IFF_UP) {
7189                 rtnl_lock();
7190                 err = __igb_open(netdev, true);
7191                 rtnl_unlock();
7192                 if (err)
7193                         return err;
7194         }
7195
7196         netif_device_attach(netdev);
7197         return 0;
7198 }
7199
7200 #ifdef CONFIG_PM_RUNTIME
7201 static int igb_runtime_idle(struct device *dev)
7202 {
7203         struct pci_dev *pdev = to_pci_dev(dev);
7204         struct net_device *netdev = pci_get_drvdata(pdev);
7205         struct igb_adapter *adapter = netdev_priv(netdev);
7206
7207         if (!igb_has_link(adapter))
7208                 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7209
7210         return -EBUSY;
7211 }
7212
7213 static int igb_runtime_suspend(struct device *dev)
7214 {
7215         struct pci_dev *pdev = to_pci_dev(dev);
7216         int retval;
7217         bool wake;
7218
7219         retval = __igb_shutdown(pdev, &wake, 1);
7220         if (retval)
7221                 return retval;
7222
7223         if (wake) {
7224                 pci_prepare_to_sleep(pdev);
7225         } else {
7226                 pci_wake_from_d3(pdev, false);
7227                 pci_set_power_state(pdev, PCI_D3hot);
7228         }
7229
7230         return 0;
7231 }
7232
7233 static int igb_runtime_resume(struct device *dev)
7234 {
7235         return igb_resume(dev);
7236 }
7237 #endif /* CONFIG_PM_RUNTIME */
7238 #endif
7239
7240 static void igb_shutdown(struct pci_dev *pdev)
7241 {
7242         bool wake;
7243
7244         __igb_shutdown(pdev, &wake, 0);
7245
7246         if (system_state == SYSTEM_POWER_OFF) {
7247                 pci_wake_from_d3(pdev, wake);
7248                 pci_set_power_state(pdev, PCI_D3hot);
7249         }
7250 }
7251
7252 #ifdef CONFIG_PCI_IOV
7253 static int igb_sriov_reinit(struct pci_dev *dev)
7254 {
7255         struct net_device *netdev = pci_get_drvdata(dev);
7256         struct igb_adapter *adapter = netdev_priv(netdev);
7257         struct pci_dev *pdev = adapter->pdev;
7258
7259         rtnl_lock();
7260
7261         if (netif_running(netdev))
7262                 igb_close(netdev);
7263
7264         igb_clear_interrupt_scheme(adapter);
7265
7266         igb_init_queue_configuration(adapter);
7267
7268         if (igb_init_interrupt_scheme(adapter, true)) {
7269                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7270                 return -ENOMEM;
7271         }
7272
7273         if (netif_running(netdev))
7274                 igb_open(netdev);
7275
7276         rtnl_unlock();
7277
7278         return 0;
7279 }
7280
7281 static int igb_pci_disable_sriov(struct pci_dev *dev)
7282 {
7283         int err = igb_disable_sriov(dev);
7284
7285         if (!err)
7286                 err = igb_sriov_reinit(dev);
7287
7288         return err;
7289 }
7290
7291 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7292 {
7293         int err = igb_enable_sriov(dev, num_vfs);
7294
7295         if (err)
7296                 goto out;
7297
7298         err = igb_sriov_reinit(dev);
7299         if (!err)
7300                 return num_vfs;
7301
7302 out:
7303         return err;
7304 }
7305
7306 #endif
7307 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7308 {
7309 #ifdef CONFIG_PCI_IOV
7310         if (num_vfs == 0)
7311                 return igb_pci_disable_sriov(dev);
7312         else
7313                 return igb_pci_enable_sriov(dev, num_vfs);
7314 #endif
7315         return 0;
7316 }
7317
7318 #ifdef CONFIG_NET_POLL_CONTROLLER
7319 /*
7320  * Polling 'interrupt' - used by things like netconsole to send skbs
7321  * without having to re-enable interrupts. It's not called while
7322  * the interrupt routine is executing.
7323  */
7324 static void igb_netpoll(struct net_device *netdev)
7325 {
7326         struct igb_adapter *adapter = netdev_priv(netdev);
7327         struct e1000_hw *hw = &adapter->hw;
7328         struct igb_q_vector *q_vector;
7329         int i;
7330
7331         for (i = 0; i < adapter->num_q_vectors; i++) {
7332                 q_vector = adapter->q_vector[i];
7333                 if (adapter->msix_entries)
7334                         wr32(E1000_EIMC, q_vector->eims_value);
7335                 else
7336                         igb_irq_disable(adapter);
7337                 napi_schedule(&q_vector->napi);
7338         }
7339 }
7340 #endif /* CONFIG_NET_POLL_CONTROLLER */
7341
7342 /**
7343  * igb_io_error_detected - called when PCI error is detected
7344  * @pdev: Pointer to PCI device
7345  * @state: The current pci connection state
7346  *
7347  * This function is called after a PCI bus error affecting
7348  * this device has been detected.
7349  */
7350 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7351                                               pci_channel_state_t state)
7352 {
7353         struct net_device *netdev = pci_get_drvdata(pdev);
7354         struct igb_adapter *adapter = netdev_priv(netdev);
7355
7356         netif_device_detach(netdev);
7357
7358         if (state == pci_channel_io_perm_failure)
7359                 return PCI_ERS_RESULT_DISCONNECT;
7360
7361         if (netif_running(netdev))
7362                 igb_down(adapter);
7363         pci_disable_device(pdev);
7364
7365         /* Request a slot slot reset. */
7366         return PCI_ERS_RESULT_NEED_RESET;
7367 }
7368
7369 /**
7370  * igb_io_slot_reset - called after the pci bus has been reset.
7371  * @pdev: Pointer to PCI device
7372  *
7373  * Restart the card from scratch, as if from a cold-boot. Implementation
7374  * resembles the first-half of the igb_resume routine.
7375  */
7376 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7377 {
7378         struct net_device *netdev = pci_get_drvdata(pdev);
7379         struct igb_adapter *adapter = netdev_priv(netdev);
7380         struct e1000_hw *hw = &adapter->hw;
7381         pci_ers_result_t result;
7382         int err;
7383
7384         if (pci_enable_device_mem(pdev)) {
7385                 dev_err(&pdev->dev,
7386                         "Cannot re-enable PCI device after reset.\n");
7387                 result = PCI_ERS_RESULT_DISCONNECT;
7388         } else {
7389                 pci_set_master(pdev);
7390                 pci_restore_state(pdev);
7391                 pci_save_state(pdev);
7392
7393                 pci_enable_wake(pdev, PCI_D3hot, 0);
7394                 pci_enable_wake(pdev, PCI_D3cold, 0);
7395
7396                 igb_reset(adapter);
7397                 wr32(E1000_WUS, ~0);
7398                 result = PCI_ERS_RESULT_RECOVERED;
7399         }
7400
7401         err = pci_cleanup_aer_uncorrect_error_status(pdev);
7402         if (err) {
7403                 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
7404                         "failed 0x%0x\n", err);
7405                 /* non-fatal, continue */
7406         }
7407
7408         return result;
7409 }
7410
7411 /**
7412  * igb_io_resume - called when traffic can start flowing again.
7413  * @pdev: Pointer to PCI device
7414  *
7415  * This callback is called when the error recovery driver tells us that
7416  * its OK to resume normal operation. Implementation resembles the
7417  * second-half of the igb_resume routine.
7418  */
7419 static void igb_io_resume(struct pci_dev *pdev)
7420 {
7421         struct net_device *netdev = pci_get_drvdata(pdev);
7422         struct igb_adapter *adapter = netdev_priv(netdev);
7423
7424         if (netif_running(netdev)) {
7425                 if (igb_up(adapter)) {
7426                         dev_err(&pdev->dev, "igb_up failed after reset\n");
7427                         return;
7428                 }
7429         }
7430
7431         netif_device_attach(netdev);
7432
7433         /* let the f/w know that the h/w is now under the control of the
7434          * driver. */
7435         igb_get_hw_control(adapter);
7436 }
7437
7438 static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7439                              u8 qsel)
7440 {
7441         u32 rar_low, rar_high;
7442         struct e1000_hw *hw = &adapter->hw;
7443
7444         /* HW expects these in little endian so we reverse the byte order
7445          * from network order (big endian) to little endian
7446          */
7447         rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7448                   ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7449         rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7450
7451         /* Indicate to hardware the Address is Valid. */
7452         rar_high |= E1000_RAH_AV;
7453
7454         if (hw->mac.type == e1000_82575)
7455                 rar_high |= E1000_RAH_POOL_1 * qsel;
7456         else
7457                 rar_high |= E1000_RAH_POOL_1 << qsel;
7458
7459         wr32(E1000_RAL(index), rar_low);
7460         wrfl();
7461         wr32(E1000_RAH(index), rar_high);
7462         wrfl();
7463 }
7464
7465 static int igb_set_vf_mac(struct igb_adapter *adapter,
7466                           int vf, unsigned char *mac_addr)
7467 {
7468         struct e1000_hw *hw = &adapter->hw;
7469         /* VF MAC addresses start at end of receive addresses and moves
7470          * torwards the first, as a result a collision should not be possible */
7471         int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7472
7473         memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7474
7475         igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7476
7477         return 0;
7478 }
7479
7480 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7481 {
7482         struct igb_adapter *adapter = netdev_priv(netdev);
7483         if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7484                 return -EINVAL;
7485         adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7486         dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7487         dev_info(&adapter->pdev->dev, "Reload the VF driver to make this"
7488                                       " change effective.");
7489         if (test_bit(__IGB_DOWN, &adapter->state)) {
7490                 dev_warn(&adapter->pdev->dev, "The VF MAC address has been set,"
7491                          " but the PF device is not up.\n");
7492                 dev_warn(&adapter->pdev->dev, "Bring the PF device up before"
7493                          " attempting to use the VF device.\n");
7494         }
7495         return igb_set_vf_mac(adapter, vf, mac);
7496 }
7497
7498 static int igb_link_mbps(int internal_link_speed)
7499 {
7500         switch (internal_link_speed) {
7501         case SPEED_100:
7502                 return 100;
7503         case SPEED_1000:
7504                 return 1000;
7505         default:
7506                 return 0;
7507         }
7508 }
7509
7510 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7511                                   int link_speed)
7512 {
7513         int rf_dec, rf_int;
7514         u32 bcnrc_val;
7515
7516         if (tx_rate != 0) {
7517                 /* Calculate the rate factor values to set */
7518                 rf_int = link_speed / tx_rate;
7519                 rf_dec = (link_speed - (rf_int * tx_rate));
7520                 rf_dec = (rf_dec * (1<<E1000_RTTBCNRC_RF_INT_SHIFT)) / tx_rate;
7521
7522                 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7523                 bcnrc_val |= ((rf_int<<E1000_RTTBCNRC_RF_INT_SHIFT) &
7524                                E1000_RTTBCNRC_RF_INT_MASK);
7525                 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7526         } else {
7527                 bcnrc_val = 0;
7528         }
7529
7530         wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7531         /*
7532          * Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7533          * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7534          */
7535         wr32(E1000_RTTBCNRM, 0x14);
7536         wr32(E1000_RTTBCNRC, bcnrc_val);
7537 }
7538
7539 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7540 {
7541         int actual_link_speed, i;
7542         bool reset_rate = false;
7543
7544         /* VF TX rate limit was not set or not supported */
7545         if ((adapter->vf_rate_link_speed == 0) ||
7546             (adapter->hw.mac.type != e1000_82576))
7547                 return;
7548
7549         actual_link_speed = igb_link_mbps(adapter->link_speed);
7550         if (actual_link_speed != adapter->vf_rate_link_speed) {
7551                 reset_rate = true;
7552                 adapter->vf_rate_link_speed = 0;
7553                 dev_info(&adapter->pdev->dev,
7554                          "Link speed has been changed. VF Transmit "
7555                          "rate is disabled\n");
7556         }
7557
7558         for (i = 0; i < adapter->vfs_allocated_count; i++) {
7559                 if (reset_rate)
7560                         adapter->vf_data[i].tx_rate = 0;
7561
7562                 igb_set_vf_rate_limit(&adapter->hw, i,
7563                                       adapter->vf_data[i].tx_rate,
7564                                       actual_link_speed);
7565         }
7566 }
7567
7568 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7569 {
7570         struct igb_adapter *adapter = netdev_priv(netdev);
7571         struct e1000_hw *hw = &adapter->hw;
7572         int actual_link_speed;
7573
7574         if (hw->mac.type != e1000_82576)
7575                 return -EOPNOTSUPP;
7576
7577         actual_link_speed = igb_link_mbps(adapter->link_speed);
7578         if ((vf >= adapter->vfs_allocated_count) ||
7579             (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7580             (tx_rate < 0) || (tx_rate > actual_link_speed))
7581                 return -EINVAL;
7582
7583         adapter->vf_rate_link_speed = actual_link_speed;
7584         adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7585         igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7586
7587         return 0;
7588 }
7589
7590 static int igb_ndo_get_vf_config(struct net_device *netdev,
7591                                  int vf, struct ifla_vf_info *ivi)
7592 {
7593         struct igb_adapter *adapter = netdev_priv(netdev);
7594         if (vf >= adapter->vfs_allocated_count)
7595                 return -EINVAL;
7596         ivi->vf = vf;
7597         memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7598         ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7599         ivi->vlan = adapter->vf_data[vf].pf_vlan;
7600         ivi->qos = adapter->vf_data[vf].pf_qos;
7601         return 0;
7602 }
7603
7604 static void igb_vmm_control(struct igb_adapter *adapter)
7605 {
7606         struct e1000_hw *hw = &adapter->hw;
7607         u32 reg;
7608
7609         switch (hw->mac.type) {
7610         case e1000_82575:
7611         case e1000_i210:
7612         case e1000_i211:
7613         default:
7614                 /* replication is not supported for 82575 */
7615                 return;
7616         case e1000_82576:
7617                 /* notify HW that the MAC is adding vlan tags */
7618                 reg = rd32(E1000_DTXCTL);
7619                 reg |= E1000_DTXCTL_VLAN_ADDED;
7620                 wr32(E1000_DTXCTL, reg);
7621         case e1000_82580:
7622                 /* enable replication vlan tag stripping */
7623                 reg = rd32(E1000_RPLOLR);
7624                 reg |= E1000_RPLOLR_STRVLAN;
7625                 wr32(E1000_RPLOLR, reg);
7626         case e1000_i350:
7627                 /* none of the above registers are supported by i350 */
7628                 break;
7629         }
7630
7631         if (adapter->vfs_allocated_count) {
7632                 igb_vmdq_set_loopback_pf(hw, true);
7633                 igb_vmdq_set_replication_pf(hw, true);
7634                 igb_vmdq_set_anti_spoofing_pf(hw, true,
7635                                                 adapter->vfs_allocated_count);
7636         } else {
7637                 igb_vmdq_set_loopback_pf(hw, false);
7638                 igb_vmdq_set_replication_pf(hw, false);
7639         }
7640 }
7641
7642 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7643 {
7644         struct e1000_hw *hw = &adapter->hw;
7645         u32 dmac_thr;
7646         u16 hwm;
7647
7648         if (hw->mac.type > e1000_82580) {
7649                 if (adapter->flags & IGB_FLAG_DMAC) {
7650                         u32 reg;
7651
7652                         /* force threshold to 0. */
7653                         wr32(E1000_DMCTXTH, 0);
7654
7655                         /*
7656                          * DMA Coalescing high water mark needs to be greater
7657                          * than the Rx threshold. Set hwm to PBA - max frame
7658                          * size in 16B units, capping it at PBA - 6KB.
7659                          */
7660                         hwm = 64 * pba - adapter->max_frame_size / 16;
7661                         if (hwm < 64 * (pba - 6))
7662                                 hwm = 64 * (pba - 6);
7663                         reg = rd32(E1000_FCRTC);
7664                         reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7665                         reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7666                                 & E1000_FCRTC_RTH_COAL_MASK);
7667                         wr32(E1000_FCRTC, reg);
7668
7669                         /*
7670                          * Set the DMA Coalescing Rx threshold to PBA - 2 * max
7671                          * frame size, capping it at PBA - 10KB.
7672                          */
7673                         dmac_thr = pba - adapter->max_frame_size / 512;
7674                         if (dmac_thr < pba - 10)
7675                                 dmac_thr = pba - 10;
7676                         reg = rd32(E1000_DMACR);
7677                         reg &= ~E1000_DMACR_DMACTHR_MASK;
7678                         reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7679                                 & E1000_DMACR_DMACTHR_MASK);
7680
7681                         /* transition to L0x or L1 if available..*/
7682                         reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7683
7684                         /* watchdog timer= +-1000 usec in 32usec intervals */
7685                         reg |= (1000 >> 5);
7686
7687                         /* Disable BMC-to-OS Watchdog Enable */
7688                         reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7689                         wr32(E1000_DMACR, reg);
7690
7691                         /*
7692                          * no lower threshold to disable
7693                          * coalescing(smart fifb)-UTRESH=0
7694                          */
7695                         wr32(E1000_DMCRTRH, 0);
7696
7697                         reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7698
7699                         wr32(E1000_DMCTLX, reg);
7700
7701                         /*
7702                          * free space in tx packet buffer to wake from
7703                          * DMA coal
7704                          */
7705                         wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7706                              (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7707
7708                         /*
7709                          * make low power state decision controlled
7710                          * by DMA coal
7711                          */
7712                         reg = rd32(E1000_PCIEMISC);
7713                         reg &= ~E1000_PCIEMISC_LX_DECISION;
7714                         wr32(E1000_PCIEMISC, reg);
7715                 } /* endif adapter->dmac is not disabled */
7716         } else if (hw->mac.type == e1000_82580) {
7717                 u32 reg = rd32(E1000_PCIEMISC);
7718                 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7719                 wr32(E1000_DMACR, 0);
7720         }
7721 }
7722
7723 /*  igb_read_i2c_byte - Reads 8 bit word over I2C
7724  *  @hw: pointer to hardware structure
7725  *  @byte_offset: byte offset to read
7726  *  @dev_addr: device address
7727  *  @data: value read
7728  *
7729  *  Performs byte read operation over I2C interface at
7730  *  a specified device address.
7731  */
7732 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7733                                 u8 dev_addr, u8 *data)
7734 {
7735         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7736         struct i2c_client *this_client = adapter->i2c_client;
7737         s32 status;
7738         u16 swfw_mask = 0;
7739
7740         if (!this_client)
7741                 return E1000_ERR_I2C;
7742
7743         swfw_mask = E1000_SWFW_PHY0_SM;
7744
7745         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7746             != E1000_SUCCESS)
7747                 return E1000_ERR_SWFW_SYNC;
7748
7749         status = i2c_smbus_read_byte_data(this_client, byte_offset);
7750         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7751
7752         if (status < 0)
7753                 return E1000_ERR_I2C;
7754         else {
7755                 *data = status;
7756                 return E1000_SUCCESS;
7757         }
7758 }
7759
7760 /*  igb_write_i2c_byte - Writes 8 bit word over I2C
7761  *  @hw: pointer to hardware structure
7762  *  @byte_offset: byte offset to write
7763  *  @dev_addr: device address
7764  *  @data: value to write
7765  *
7766  *  Performs byte write operation over I2C interface at
7767  *  a specified device address.
7768  */
7769 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7770                                  u8 dev_addr, u8 data)
7771 {
7772         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7773         struct i2c_client *this_client = adapter->i2c_client;
7774         s32 status;
7775         u16 swfw_mask = E1000_SWFW_PHY0_SM;
7776
7777         if (!this_client)
7778                 return E1000_ERR_I2C;
7779
7780         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
7781                 return E1000_ERR_SWFW_SYNC;
7782         status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
7783         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7784
7785         if (status)
7786                 return E1000_ERR_I2C;
7787         else
7788                 return E1000_SUCCESS;
7789
7790 }
7791 /* igb_main.c */