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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 2007 - 2018 Intel Corporation. */
   3
   4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   5
   6 #include <linux/module.h>
   7 #include <linux/types.h>
   8 #include <linux/init.h>
   9 #include <linux/bitops.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/ipv6.h>
  14 #include <linux/slab.h>
  15 #include <net/checksum.h>
  16 #include <net/ip6_checksum.h>
  17 #include <net/pkt_sched.h>
  18 #include <net/pkt_cls.h>
  19 #include <linux/net_tstamp.h>
  20 #include <linux/mii.h>
  21 #include <linux/ethtool.h>
  22 #include <linux/if.h>
  23 #include <linux/if_vlan.h>
  24 #include <linux/pci.h>
  25 #include <linux/delay.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/ip.h>
  28 #include <linux/tcp.h>
  29 #include <linux/sctp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/aer.h>
  32 #include <linux/prefetch.h>
  33 #include <linux/pm_runtime.h>
  34 #include <linux/etherdevice.h>
  35 #ifdef CONFIG_IGB_DCA
  36 #include <linux/dca.h>
  37 #endif
  38 #include <linux/i2c.h>
  39 #include "igb.h"
  40
  41 #define MAJ 5
  42 #define MIN 6
  43 #define BUILD 0
  44 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
  45 __stringify(BUILD) "-k"
  46
  47 enum queue_mode {
  48         QUEUE_MODE_STRICT_PRIORITY,
  49         QUEUE_MODE_STREAM_RESERVATION,
  50 };
  51
  52 enum tx_queue_prio {
  53         TX_QUEUE_PRIO_HIGH,
  54         TX_QUEUE_PRIO_LOW,
  55 };
  56
  57 char igb_driver_name[] = "igb";
  58 char igb_driver_version[] = DRV_VERSION;
  59 static const char igb_driver_string[] =
  60                                 "Intel(R) Gigabit Ethernet Network Driver";
  61 static const char igb_copyright[] =
  62                                 "Copyright (c) 2007-2014 Intel Corporation.";
  63
  64 static const struct e1000_info *igb_info_tbl[] = {
  65         [board_82575] = &e1000_82575_info,
  66 };
  67
  68 static const struct pci_device_id igb_pci_tbl[] = {
  69         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
  70         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
  71         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
  72         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
  73         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
  74         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
  75         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
  76         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
  77         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
  78         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
  79         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
  80         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
  81         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
  82         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
  83         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
  84         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
  85         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
  86         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
  87         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
  88         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
  89         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
  90         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
  91         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
  92         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
  93         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
  94         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
  95         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
  96         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
  97         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
  98         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
  99         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
 100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
 101         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
 102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
 103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
 104         /* required last entry */
 105         {0, }
 106 };
 107
 108 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 109
 110 static int igb_setup_all_tx_resources(struct igb_adapter *);
 111 static int igb_setup_all_rx_resources(struct igb_adapter *);
 112 static void igb_free_all_tx_resources(struct igb_adapter *);
 113 static void igb_free_all_rx_resources(struct igb_adapter *);
 114 static void igb_setup_mrqc(struct igb_adapter *);
 115 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 116 static void igb_remove(struct pci_dev *pdev);
 117 static int igb_sw_init(struct igb_adapter *);
 118 int igb_open(struct net_device *);
 119 int igb_close(struct net_device *);
 120 static void igb_configure(struct igb_adapter *);
 121 static void igb_configure_tx(struct igb_adapter *);
 122 static void igb_configure_rx(struct igb_adapter *);
 123 static void igb_clean_all_tx_rings(struct igb_adapter *);
 124 static void igb_clean_all_rx_rings(struct igb_adapter *);
 125 static void igb_clean_tx_ring(struct igb_ring *);
 126 static void igb_clean_rx_ring(struct igb_ring *);
 127 static void igb_set_rx_mode(struct net_device *);
 128 static void igb_update_phy_info(struct timer_list *);
 129 static void igb_watchdog(struct timer_list *);
 130 static void igb_watchdog_task(struct work_struct *);
 131 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 132 static void igb_get_stats64(struct net_device *dev,
 133                             struct rtnl_link_stats64 *stats);
 134 static int igb_change_mtu(struct net_device *, int);
 135 static int igb_set_mac(struct net_device *, void *);
 136 static void igb_set_uta(struct igb_adapter *adapter, bool set);
 137 static irqreturn_t igb_intr(int irq, void *);
 138 static irqreturn_t igb_intr_msi(int irq, void *);
 139 static irqreturn_t igb_msix_other(int irq, void *);
 140 static irqreturn_t igb_msix_ring(int irq, void *);
 141 #ifdef CONFIG_IGB_DCA
 142 static void igb_update_dca(struct igb_q_vector *);
 143 static void igb_setup_dca(struct igb_adapter *);
 144 #endif /* CONFIG_IGB_DCA */
 145 static int igb_poll(struct napi_struct *, int);
 146 static bool igb_clean_tx_irq(struct igb_q_vector *, int);
 147 static int igb_clean_rx_irq(struct igb_q_vector *, int);
 148 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 149 static void igb_tx_timeout(struct net_device *, unsigned int txqueue);
 150 static void igb_reset_task(struct work_struct *);
 151 static void igb_vlan_mode(struct net_device *netdev,
 152                           netdev_features_t features);
 153 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
 154 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
 155 static void igb_restore_vlan(struct igb_adapter *);
 156 static void igb_rar_set_index(struct igb_adapter *, u32);
 157 static void igb_ping_all_vfs(struct igb_adapter *);
 158 static void igb_msg_task(struct igb_adapter *);
 159 static void igb_vmm_control(struct igb_adapter *);
 160 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 161 static void igb_flush_mac_table(struct igb_adapter *);
 162 static int igb_available_rars(struct igb_adapter *, u8);
 163 static void igb_set_default_mac_filter(struct igb_adapter *);
 164 static int igb_uc_sync(struct net_device *, const unsigned char *);
 165 static int igb_uc_unsync(struct net_device *, const unsigned char *);
 166 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 167 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 168 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 169                                int vf, u16 vlan, u8 qos, __be16 vlan_proto);
 170 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
 171 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 172                                    bool setting);
 173 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
 174                                 bool setting);
 175 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 176                                  struct ifla_vf_info *ivi);
 177 static void igb_check_vf_rate_limit(struct igb_adapter *);
 178 static void igb_nfc_filter_exit(struct igb_adapter *adapter);
 179 static void igb_nfc_filter_restore(struct igb_adapter *adapter);
 180
 181 #ifdef CONFIG_PCI_IOV
 182 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 183 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
 184 static int igb_disable_sriov(struct pci_dev *dev);
 185 static int igb_pci_disable_sriov(struct pci_dev *dev);
 186 #endif
 187
 188 static int igb_suspend(struct device *);
 189 static int igb_resume(struct device *);
 190 static int igb_runtime_suspend(struct device *dev);
 191 static int igb_runtime_resume(struct device *dev);
 192 static int igb_runtime_idle(struct device *dev);
 193 static const struct dev_pm_ops igb_pm_ops = {
 194         SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
 195         SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
 196                         igb_runtime_idle)
 197 };
 198 static void igb_shutdown(struct pci_dev *);
 199 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 200 #ifdef CONFIG_IGB_DCA
 201 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 202 static struct notifier_block dca_notifier = {
 203         .notifier_call  = igb_notify_dca,
 204         .next           = NULL,
 205         .priority       = 0
 206 };
 207 #endif
 208 #ifdef CONFIG_PCI_IOV
 209 static unsigned int max_vfs;
 210 module_param(max_vfs, uint, 0);
 211 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
 212 #endif /* CONFIG_PCI_IOV */
 213
 214 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 215                      pci_channel_state_t);
 216 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 217 static void igb_io_resume(struct pci_dev *);
 218
 219 static const struct pci_error_handlers igb_err_handler = {
 220         .error_detected = igb_io_error_detected,
 221         .slot_reset = igb_io_slot_reset,
 222         .resume = igb_io_resume,
 223 };
 224
 225 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 226
 227 static struct pci_driver igb_driver = {
 228         .name     = igb_driver_name,
 229         .id_table = igb_pci_tbl,
 230         .probe    = igb_probe,
 231         .remove   = igb_remove,
 232 #ifdef CONFIG_PM
 233         .driver.pm = &igb_pm_ops,
 234 #endif
 235         .shutdown = igb_shutdown,
 236         .sriov_configure = igb_pci_sriov_configure,
 237         .err_handler = &igb_err_handler
 238 };
 239
 240 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 241 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 242 MODULE_LICENSE("GPL v2");
 243 MODULE_VERSION(DRV_VERSION);
 244
 245 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 246 static int debug = -1;
 247 module_param(debug, int, 0);
 248 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 249
 250 struct igb_reg_info {
 251         u32 ofs;
 252         char *name;
 253 };
 254
 255 static const struct igb_reg_info igb_reg_info_tbl[] = {
 256
 257         /* General Registers */
 258         {E1000_CTRL, "CTRL"},
 259         {E1000_STATUS, "STATUS"},
 260         {E1000_CTRL_EXT, "CTRL_EXT"},
 261
 262         /* Interrupt Registers */
 263         {E1000_ICR, "ICR"},
 264
 265         /* RX Registers */
 266         {E1000_RCTL, "RCTL"},
 267         {E1000_RDLEN(0), "RDLEN"},
 268         {E1000_RDH(0), "RDH"},
 269         {E1000_RDT(0), "RDT"},
 270         {E1000_RXDCTL(0), "RXDCTL"},
 271         {E1000_RDBAL(0), "RDBAL"},
 272         {E1000_RDBAH(0), "RDBAH"},
 273
 274         /* TX Registers */
 275         {E1000_TCTL, "TCTL"},
 276         {E1000_TDBAL(0), "TDBAL"},
 277         {E1000_TDBAH(0), "TDBAH"},
 278         {E1000_TDLEN(0), "TDLEN"},
 279         {E1000_TDH(0), "TDH"},
 280         {E1000_TDT(0), "TDT"},
 281         {E1000_TXDCTL(0), "TXDCTL"},
 282         {E1000_TDFH, "TDFH"},
 283         {E1000_TDFT, "TDFT"},
 284         {E1000_TDFHS, "TDFHS"},
 285         {E1000_TDFPC, "TDFPC"},
 286
 287         /* List Terminator */
 288         {}
 289 };
 290
 291 /* igb_regdump - register printout routine */
 292 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
 293 {
 294         int n = 0;
 295         char rname[16];
 296         u32 regs[8];
 297
 298         switch (reginfo->ofs) {
 299         case E1000_RDLEN(0):
 300                 for (n = 0; n < 4; n++)
 301                         regs[n] = rd32(E1000_RDLEN(n));
 302                 break;
 303         case E1000_RDH(0):
 304                 for (n = 0; n < 4; n++)
 305                         regs[n] = rd32(E1000_RDH(n));
 306                 break;
 307         case E1000_RDT(0):
 308                 for (n = 0; n < 4; n++)
 309                         regs[n] = rd32(E1000_RDT(n));
 310                 break;
 311         case E1000_RXDCTL(0):
 312                 for (n = 0; n < 4; n++)
 313                         regs[n] = rd32(E1000_RXDCTL(n));
 314                 break;
 315         case E1000_RDBAL(0):
 316                 for (n = 0; n < 4; n++)
 317                         regs[n] = rd32(E1000_RDBAL(n));
 318                 break;
 319         case E1000_RDBAH(0):
 320                 for (n = 0; n < 4; n++)
 321                         regs[n] = rd32(E1000_RDBAH(n));
 322                 break;
 323         case E1000_TDBAL(0):
 324                 for (n = 0; n < 4; n++)
 325                         regs[n] = rd32(E1000_RDBAL(n));
 326                 break;
 327         case E1000_TDBAH(0):
 328                 for (n = 0; n < 4; n++)
 329                         regs[n] = rd32(E1000_TDBAH(n));
 330                 break;
 331         case E1000_TDLEN(0):
 332                 for (n = 0; n < 4; n++)
 333                         regs[n] = rd32(E1000_TDLEN(n));
 334                 break;
 335         case E1000_TDH(0):
 336                 for (n = 0; n < 4; n++)
 337                         regs[n] = rd32(E1000_TDH(n));
 338                 break;
 339         case E1000_TDT(0):
 340                 for (n = 0; n < 4; n++)
 341                         regs[n] = rd32(E1000_TDT(n));
 342                 break;
 343         case E1000_TXDCTL(0):
 344                 for (n = 0; n < 4; n++)
 345                         regs[n] = rd32(E1000_TXDCTL(n));
 346                 break;
 347         default:
 348                 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
 349                 return;
 350         }
 351
 352         snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
 353         pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
 354                 regs[2], regs[3]);
 355 }
 356
 357 /* igb_dump - Print registers, Tx-rings and Rx-rings */
 358 static void igb_dump(struct igb_adapter *adapter)
 359 {
 360         struct net_device *netdev = adapter->netdev;
 361         struct e1000_hw *hw = &adapter->hw;
 362         struct igb_reg_info *reginfo;
 363         struct igb_ring *tx_ring;
 364         union e1000_adv_tx_desc *tx_desc;
 365         struct my_u0 { u64 a; u64 b; } *u0;
 366         struct igb_ring *rx_ring;
 367         union e1000_adv_rx_desc *rx_desc;
 368         u32 staterr;
 369         u16 i, n;
 370
 371         if (!netif_msg_hw(adapter))
 372                 return;
 373
 374         /* Print netdevice Info */
 375         if (netdev) {
 376                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 377                 pr_info("Device Name     state            trans_start\n");
 378                 pr_info("%-15s %016lX %016lX\n", netdev->name,
 379                         netdev->state, dev_trans_start(netdev));
 380         }
 381
 382         /* Print Registers */
 383         dev_info(&adapter->pdev->dev, "Register Dump\n");
 384         pr_info(" Register Name   Value\n");
 385         for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
 386              reginfo->name; reginfo++) {
 387                 igb_regdump(hw, reginfo);
 388         }
 389
 390         /* Print TX Ring Summary */
 391         if (!netdev || !netif_running(netdev))
 392                 goto exit;
 393
 394         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 395         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 396         for (n = 0; n < adapter->num_tx_queues; n++) {
 397                 struct igb_tx_buffer *buffer_info;
 398                 tx_ring = adapter->tx_ring[n];
 399                 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 400                 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
 401                         n, tx_ring->next_to_use, tx_ring->next_to_clean,
 402                         (u64)dma_unmap_addr(buffer_info, dma),
 403                         dma_unmap_len(buffer_info, len),
 404                         buffer_info->next_to_watch,
 405                         (u64)buffer_info->time_stamp);
 406         }
 407
 408         /* Print TX Rings */
 409         if (!netif_msg_tx_done(adapter))
 410                 goto rx_ring_summary;
 411
 412         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 413
 414         /* Transmit Descriptor Formats
 415          *
 416          * Advanced Transmit Descriptor
 417          *   +--------------------------------------------------------------+
 418          * 0 |         Buffer Address [63:0]                                |
 419          *   +--------------------------------------------------------------+
 420          * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
 421          *   +--------------------------------------------------------------+
 422          *   63      46 45    40 39 38 36 35 32 31   24             15       0
 423          */
 424
 425         for (n = 0; n < adapter->num_tx_queues; n++) {
 426                 tx_ring = adapter->tx_ring[n];
 427                 pr_info("------------------------------------\n");
 428                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 429                 pr_info("------------------------------------\n");
 430                 pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] [bi->dma       ] leng  ntw timestamp        bi->skb\n");
 431
 432                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 433                         const char *next_desc;
 434                         struct igb_tx_buffer *buffer_info;
 435                         tx_desc = IGB_TX_DESC(tx_ring, i);
 436                         buffer_info = &tx_ring->tx_buffer_info[i];
 437                         u0 = (struct my_u0 *)tx_desc;
 438                         if (i == tx_ring->next_to_use &&
 439                             i == tx_ring->next_to_clean)
 440                                 next_desc = " NTC/U";
 441                         else if (i == tx_ring->next_to_use)
 442                                 next_desc = " NTU";
 443                         else if (i == tx_ring->next_to_clean)
 444                                 next_desc = " NTC";
 445                         else
 446                                 next_desc = "";
 447
 448                         pr_info("T [0x%03X]    %016llX %016llX %016llX %04X  %p %016llX %p%s\n",
 449                                 i, le64_to_cpu(u0->a),
 450                                 le64_to_cpu(u0->b),
 451                                 (u64)dma_unmap_addr(buffer_info, dma),
 452                                 dma_unmap_len(buffer_info, len),
 453                                 buffer_info->next_to_watch,
 454                                 (u64)buffer_info->time_stamp,
 455                                 buffer_info->skb, next_desc);
 456
 457                         if (netif_msg_pktdata(adapter) && buffer_info->skb)
 458                                 print_hex_dump(KERN_INFO, "",
 459                                         DUMP_PREFIX_ADDRESS,
 460                                         16, 1, buffer_info->skb->data,
 461                                         dma_unmap_len(buffer_info, len),
 462                                         true);
 463                 }
 464         }
 465
 466         /* Print RX Rings Summary */
 467 rx_ring_summary:
 468         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 469         pr_info("Queue [NTU] [NTC]\n");
 470         for (n = 0; n < adapter->num_rx_queues; n++) {
 471                 rx_ring = adapter->rx_ring[n];
 472                 pr_info(" %5d %5X %5X\n",
 473                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 474         }
 475
 476         /* Print RX Rings */
 477         if (!netif_msg_rx_status(adapter))
 478                 goto exit;
 479
 480         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 481
 482         /* Advanced Receive Descriptor (Read) Format
 483          *    63                                           1        0
 484          *    +-----------------------------------------------------+
 485          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 486          *    +----------------------------------------------+------+
 487          *  8 |       Header Buffer Address [63:1]           |  DD  |
 488          *    +-----------------------------------------------------+
 489          *
 490          *
 491          * Advanced Receive Descriptor (Write-Back) Format
 492          *
 493          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 494          *   +------------------------------------------------------+
 495          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 496          *   | Checksum   Ident  |   |           |    | Type | Type |
 497          *   +------------------------------------------------------+
 498          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 499          *   +------------------------------------------------------+
 500          *   63       48 47    32 31            20 19               0
 501          */
 502
 503         for (n = 0; n < adapter->num_rx_queues; n++) {
 504                 rx_ring = adapter->rx_ring[n];
 505                 pr_info("------------------------------------\n");
 506                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 507                 pr_info("------------------------------------\n");
 508                 pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] [bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
 509                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
 510
 511                 for (i = 0; i < rx_ring->count; i++) {
 512                         const char *next_desc;
 513                         struct igb_rx_buffer *buffer_info;
 514                         buffer_info = &rx_ring->rx_buffer_info[i];
 515                         rx_desc = IGB_RX_DESC(rx_ring, i);
 516                         u0 = (struct my_u0 *)rx_desc;
 517                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 518
 519                         if (i == rx_ring->next_to_use)
 520                                 next_desc = " NTU";
 521                         else if (i == rx_ring->next_to_clean)
 522                                 next_desc = " NTC";
 523                         else
 524                                 next_desc = "";
 525
 526                         if (staterr & E1000_RXD_STAT_DD) {
 527                                 /* Descriptor Done */
 528                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
 529                                         "RWB", i,
 530                                         le64_to_cpu(u0->a),
 531                                         le64_to_cpu(u0->b),
 532                                         next_desc);
 533                         } else {
 534                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
 535                                         "R  ", i,
 536                                         le64_to_cpu(u0->a),
 537                                         le64_to_cpu(u0->b),
 538                                         (u64)buffer_info->dma,
 539                                         next_desc);
 540
 541                                 if (netif_msg_pktdata(adapter) &&
 542                                     buffer_info->dma && buffer_info->page) {
 543                                         print_hex_dump(KERN_INFO, "",
 544                                           DUMP_PREFIX_ADDRESS,
 545                                           16, 1,
 546                                           page_address(buffer_info->page) +
 547                                                       buffer_info->page_offset,
 548                                           igb_rx_bufsz(rx_ring), true);
 549                                 }
 550                         }
 551                 }
 552         }
 553
 554 exit:
 555         return;
 556 }
 557
 558 /**
 559  *  igb_get_i2c_data - Reads the I2C SDA data bit
 560  *  @hw: pointer to hardware structure
 561  *  @i2cctl: Current value of I2CCTL register
 562  *
 563  *  Returns the I2C data bit value
 564  **/
 565 static int igb_get_i2c_data(void *data)
 566 {
 567         struct igb_adapter *adapter = (struct igb_adapter *)data;
 568         struct e1000_hw *hw = &adapter->hw;
 569         s32 i2cctl = rd32(E1000_I2CPARAMS);
 570
 571         return !!(i2cctl & E1000_I2C_DATA_IN);
 572 }
 573
 574 /**
 575  *  igb_set_i2c_data - Sets the I2C data bit
 576  *  @data: pointer to hardware structure
 577  *  @state: I2C data value (0 or 1) to set
 578  *
 579  *  Sets the I2C data bit
 580  **/
 581 static void igb_set_i2c_data(void *data, int state)
 582 {
 583         struct igb_adapter *adapter = (struct igb_adapter *)data;
 584         struct e1000_hw *hw = &adapter->hw;
 585         s32 i2cctl = rd32(E1000_I2CPARAMS);
 586
 587         if (state)
 588                 i2cctl |= E1000_I2C_DATA_OUT;
 589         else
 590                 i2cctl &= ~E1000_I2C_DATA_OUT;
 591
 592         i2cctl &= ~E1000_I2C_DATA_OE_N;
 593         i2cctl |= E1000_I2C_CLK_OE_N;
 594         wr32(E1000_I2CPARAMS, i2cctl);
 595         wrfl();
 596
 597 }
 598
 599 /**
 600  *  igb_set_i2c_clk - Sets the I2C SCL clock
 601  *  @data: pointer to hardware structure
 602  *  @state: state to set clock
 603  *
 604  *  Sets the I2C clock line to state
 605  **/
 606 static void igb_set_i2c_clk(void *data, int state)
 607 {
 608         struct igb_adapter *adapter = (struct igb_adapter *)data;
 609         struct e1000_hw *hw = &adapter->hw;
 610         s32 i2cctl = rd32(E1000_I2CPARAMS);
 611
 612         if (state) {
 613                 i2cctl |= E1000_I2C_CLK_OUT;
 614                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 615         } else {
 616                 i2cctl &= ~E1000_I2C_CLK_OUT;
 617                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 618         }
 619         wr32(E1000_I2CPARAMS, i2cctl);
 620         wrfl();
 621 }
 622
 623 /**
 624  *  igb_get_i2c_clk - Gets the I2C SCL clock state
 625  *  @data: pointer to hardware structure
 626  *
 627  *  Gets the I2C clock state
 628  **/
 629 static int igb_get_i2c_clk(void *data)
 630 {
 631         struct igb_adapter *adapter = (struct igb_adapter *)data;
 632         struct e1000_hw *hw = &adapter->hw;
 633         s32 i2cctl = rd32(E1000_I2CPARAMS);
 634
 635         return !!(i2cctl & E1000_I2C_CLK_IN);
 636 }
 637
 638 static const struct i2c_algo_bit_data igb_i2c_algo = {
 639         .setsda         = igb_set_i2c_data,
 640         .setscl         = igb_set_i2c_clk,
 641         .getsda         = igb_get_i2c_data,
 642         .getscl         = igb_get_i2c_clk,
 643         .udelay         = 5,
 644         .timeout        = 20,
 645 };
 646
 647 /**
 648  *  igb_get_hw_dev - return device
 649  *  @hw: pointer to hardware structure
 650  *
 651  *  used by hardware layer to print debugging information
 652  **/
 653 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
 654 {
 655         struct igb_adapter *adapter = hw->back;
 656         return adapter->netdev;
 657 }
 658
 659 /**
 660  *  igb_init_module - Driver Registration Routine
 661  *
 662  *  igb_init_module is the first routine called when the driver is
 663  *  loaded. All it does is register with the PCI subsystem.
 664  **/
 665 static int __init igb_init_module(void)
 666 {
 667         int ret;
 668
 669         pr_info("%s - version %s\n",
 670                igb_driver_string, igb_driver_version);
 671         pr_info("%s\n", igb_copyright);
 672
 673 #ifdef CONFIG_IGB_DCA
 674         dca_register_notify(&dca_notifier);
 675 #endif
 676         ret = pci_register_driver(&igb_driver);
 677         return ret;
 678 }
 679
 680 module_init(igb_init_module);
 681
 682 /**
 683  *  igb_exit_module - Driver Exit Cleanup Routine
 684  *
 685  *  igb_exit_module is called just before the driver is removed
 686  *  from memory.
 687  **/
 688 static void __exit igb_exit_module(void)
 689 {
 690 #ifdef CONFIG_IGB_DCA
 691         dca_unregister_notify(&dca_notifier);
 692 #endif
 693         pci_unregister_driver(&igb_driver);
 694 }
 695
 696 module_exit(igb_exit_module);
 697
 698 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 699 /**
 700  *  igb_cache_ring_register - Descriptor ring to register mapping
 701  *  @adapter: board private structure to initialize
 702  *
 703  *  Once we know the feature-set enabled for the device, we'll cache
 704  *  the register offset the descriptor ring is assigned to.
 705  **/
 706 static void igb_cache_ring_register(struct igb_adapter *adapter)
 707 {
 708         int i = 0, j = 0;
 709         u32 rbase_offset = adapter->vfs_allocated_count;
 710
 711         switch (adapter->hw.mac.type) {
 712         case e1000_82576:
 713                 /* The queues are allocated for virtualization such that VF 0
 714                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 715                  * In order to avoid collision we start at the first free queue
 716                  * and continue consuming queues in the same sequence
 717                  */
 718                 if (adapter->vfs_allocated_count) {
 719                         for (; i < adapter->rss_queues; i++)
 720                                 adapter->rx_ring[i]->reg_idx = rbase_offset +
 721                                                                Q_IDX_82576(i);
 722                 }
 723                 /* Fall through */
 724         case e1000_82575:
 725         case e1000_82580:
 726         case e1000_i350:
 727         case e1000_i354:
 728         case e1000_i210:
 729         case e1000_i211:
 730                 /* Fall through */
 731         default:
 732                 for (; i < adapter->num_rx_queues; i++)
 733                         adapter->rx_ring[i]->reg_idx = rbase_offset + i;
 734                 for (; j < adapter->num_tx_queues; j++)
 735                         adapter->tx_ring[j]->reg_idx = rbase_offset + j;
 736                 break;
 737         }
 738 }
 739
 740 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
 741 {
 742         struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
 743         u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
 744         u32 value = 0;
 745
 746         if (E1000_REMOVED(hw_addr))
 747                 return ~value;
 748
 749         value = readl(&hw_addr[reg]);
 750
 751         /* reads should not return all F's */
 752         if (!(~value) && (!reg || !(~readl(hw_addr)))) {
 753                 struct net_device *netdev = igb->netdev;
 754                 hw->hw_addr = NULL;
 755                 netdev_err(netdev, "PCIe link lost\n");
 756                 WARN(pci_device_is_present(igb->pdev),
 757                      "igb: Failed to read reg 0x%x!\n", reg);
 758         }
 759
 760         return value;
 761 }
 762
 763 /**
 764  *  igb_write_ivar - configure ivar for given MSI-X vector
 765  *  @hw: pointer to the HW structure
 766  *  @msix_vector: vector number we are allocating to a given ring
 767  *  @index: row index of IVAR register to write within IVAR table
 768  *  @offset: column offset of in IVAR, should be multiple of 8
 769  *
 770  *  This function is intended to handle the writing of the IVAR register
 771  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 772  *  each containing an cause allocation for an Rx and Tx ring, and a
 773  *  variable number of rows depending on the number of queues supported.
 774  **/
 775 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 776                            int index, int offset)
 777 {
 778         u32 ivar = array_rd32(E1000_IVAR0, index);
 779
 780         /* clear any bits that are currently set */
 781         ivar &= ~((u32)0xFF << offset);
 782
 783         /* write vector and valid bit */
 784         ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 785
 786         array_wr32(E1000_IVAR0, index, ivar);
 787 }
 788
 789 #define IGB_N0_QUEUE -1
 790 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 791 {
 792         struct igb_adapter *adapter = q_vector->adapter;
 793         struct e1000_hw *hw = &adapter->hw;
 794         int rx_queue = IGB_N0_QUEUE;
 795         int tx_queue = IGB_N0_QUEUE;
 796         u32 msixbm = 0;
 797
 798         if (q_vector->rx.ring)
 799                 rx_queue = q_vector->rx.ring->reg_idx;
 800         if (q_vector->tx.ring)
 801                 tx_queue = q_vector->tx.ring->reg_idx;
 802
 803         switch (hw->mac.type) {
 804         case e1000_82575:
 805                 /* The 82575 assigns vectors using a bitmask, which matches the
 806                  * bitmask for the EICR/EIMS/EIMC registers.  To assign one
 807                  * or more queues to a vector, we write the appropriate bits
 808                  * into the MSIXBM register for that vector.
 809                  */
 810                 if (rx_queue > IGB_N0_QUEUE)
 811                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 812                 if (tx_queue > IGB_N0_QUEUE)
 813                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 814                 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
 815                         msixbm |= E1000_EIMS_OTHER;
 816                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 817                 q_vector->eims_value = msixbm;
 818                 break;
 819         case e1000_82576:
 820                 /* 82576 uses a table that essentially consists of 2 columns
 821                  * with 8 rows.  The ordering is column-major so we use the
 822                  * lower 3 bits as the row index, and the 4th bit as the
 823                  * column offset.
 824                  */
 825                 if (rx_queue > IGB_N0_QUEUE)
 826                         igb_write_ivar(hw, msix_vector,
 827                                        rx_queue & 0x7,
 828                                        (rx_queue & 0x8) << 1);
 829                 if (tx_queue > IGB_N0_QUEUE)
 830                         igb_write_ivar(hw, msix_vector,
 831                                        tx_queue & 0x7,
 832                                        ((tx_queue & 0x8) << 1) + 8);
 833                 q_vector->eims_value = BIT(msix_vector);
 834                 break;
 835         case e1000_82580:
 836         case e1000_i350:
 837         case e1000_i354:
 838         case e1000_i210:
 839         case e1000_i211:
 840                 /* On 82580 and newer adapters the scheme is similar to 82576
 841                  * however instead of ordering column-major we have things
 842                  * ordered row-major.  So we traverse the table by using
 843                  * bit 0 as the column offset, and the remaining bits as the
 844                  * row index.
 845                  */
 846                 if (rx_queue > IGB_N0_QUEUE)
 847                         igb_write_ivar(hw, msix_vector,
 848                                        rx_queue >> 1,
 849                                        (rx_queue & 0x1) << 4);
 850                 if (tx_queue > IGB_N0_QUEUE)
 851                         igb_write_ivar(hw, msix_vector,
 852                                        tx_queue >> 1,
 853                                        ((tx_queue & 0x1) << 4) + 8);
 854                 q_vector->eims_value = BIT(msix_vector);
 855                 break;
 856         default:
 857                 BUG();
 858                 break;
 859         }
 860
 861         /* add q_vector eims value to global eims_enable_mask */
 862         adapter->eims_enable_mask |= q_vector->eims_value;
 863
 864         /* configure q_vector to set itr on first interrupt */
 865         q_vector->set_itr = 1;
 866 }
 867
 868 /**
 869  *  igb_configure_msix - Configure MSI-X hardware
 870  *  @adapter: board private structure to initialize
 871  *
 872  *  igb_configure_msix sets up the hardware to properly
 873  *  generate MSI-X interrupts.
 874  **/
 875 static void igb_configure_msix(struct igb_adapter *adapter)
 876 {
 877         u32 tmp;
 878         int i, vector = 0;
 879         struct e1000_hw *hw = &adapter->hw;
 880
 881         adapter->eims_enable_mask = 0;
 882
 883         /* set vector for other causes, i.e. link changes */
 884         switch (hw->mac.type) {
 885         case e1000_82575:
 886                 tmp = rd32(E1000_CTRL_EXT);
 887                 /* enable MSI-X PBA support*/
 888                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 889
 890                 /* Auto-Mask interrupts upon ICR read. */
 891                 tmp |= E1000_CTRL_EXT_EIAME;
 892                 tmp |= E1000_CTRL_EXT_IRCA;
 893
 894                 wr32(E1000_CTRL_EXT, tmp);
 895
 896                 /* enable msix_other interrupt */
 897                 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
 898                 adapter->eims_other = E1000_EIMS_OTHER;
 899
 900                 break;
 901
 902         case e1000_82576:
 903         case e1000_82580:
 904         case e1000_i350:
 905         case e1000_i354:
 906         case e1000_i210:
 907         case e1000_i211:
 908                 /* Turn on MSI-X capability first, or our settings
 909                  * won't stick.  And it will take days to debug.
 910                  */
 911                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 912                      E1000_GPIE_PBA | E1000_GPIE_EIAME |
 913                      E1000_GPIE_NSICR);
 914
 915                 /* enable msix_other interrupt */
 916                 adapter->eims_other = BIT(vector);
 917                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 918
 919                 wr32(E1000_IVAR_MISC, tmp);
 920                 break;
 921         default:
 922                 /* do nothing, since nothing else supports MSI-X */
 923                 break;
 924         } /* switch (hw->mac.type) */
 925
 926         adapter->eims_enable_mask |= adapter->eims_other;
 927
 928         for (i = 0; i < adapter->num_q_vectors; i++)
 929                 igb_assign_vector(adapter->q_vector[i], vector++);
 930
 931         wrfl();
 932 }
 933
 934 /**
 935  *  igb_request_msix - Initialize MSI-X interrupts
 936  *  @adapter: board private structure to initialize
 937  *
 938  *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
 939  *  kernel.
 940  **/
 941 static int igb_request_msix(struct igb_adapter *adapter)
 942 {
 943         struct net_device *netdev = adapter->netdev;
 944         int i, err = 0, vector = 0, free_vector = 0;
 945
 946         err = request_irq(adapter->msix_entries[vector].vector,
 947                           igb_msix_other, 0, netdev->name, adapter);
 948         if (err)
 949                 goto err_out;
 950
 951         for (i = 0; i < adapter->num_q_vectors; i++) {
 952                 struct igb_q_vector *q_vector = adapter->q_vector[i];
 953
 954                 vector++;
 955
 956                 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
 957
 958                 if (q_vector->rx.ring && q_vector->tx.ring)
 959                         sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 960                                 q_vector->rx.ring->queue_index);
 961                 else if (q_vector->tx.ring)
 962                         sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 963                                 q_vector->tx.ring->queue_index);
 964                 else if (q_vector->rx.ring)
 965                         sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 966                                 q_vector->rx.ring->queue_index);
 967                 else
 968                         sprintf(q_vector->name, "%s-unused", netdev->name);
 969
 970                 err = request_irq(adapter->msix_entries[vector].vector,
 971                                   igb_msix_ring, 0, q_vector->name,
 972                                   q_vector);
 973                 if (err)
 974                         goto err_free;
 975         }
 976
 977         igb_configure_msix(adapter);
 978         return 0;
 979
 980 err_free:
 981         /* free already assigned IRQs */
 982         free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 983
 984         vector--;
 985         for (i = 0; i < vector; i++) {
 986                 free_irq(adapter->msix_entries[free_vector++].vector,
 987                          adapter->q_vector[i]);
 988         }
 989 err_out:
 990         return err;
 991 }
 992
 993 /**
 994  *  igb_free_q_vector - Free memory allocated for specific interrupt vector
 995  *  @adapter: board private structure to initialize
 996  *  @v_idx: Index of vector to be freed
 997  *
 998  *  This function frees the memory allocated to the q_vector.
 999  **/
1000 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1001 {
1002         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1003
1004         adapter->q_vector[v_idx] = NULL;
1005
1006         /* igb_get_stats64() might access the rings on this vector,
1007          * we must wait a grace period before freeing it.
1008          */
1009         if (q_vector)
1010                 kfree_rcu(q_vector, rcu);
1011 }
1012
1013 /**
1014  *  igb_reset_q_vector - Reset config for interrupt vector
1015  *  @adapter: board private structure to initialize
1016  *  @v_idx: Index of vector to be reset
1017  *
1018  *  If NAPI is enabled it will delete any references to the
1019  *  NAPI struct. This is preparation for igb_free_q_vector.
1020  **/
1021 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1022 {
1023         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1024
1025         /* Coming from igb_set_interrupt_capability, the vectors are not yet
1026          * allocated. So, q_vector is NULL so we should stop here.
1027          */
1028         if (!q_vector)
1029                 return;
1030
1031         if (q_vector->tx.ring)
1032                 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1033
1034         if (q_vector->rx.ring)
1035                 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1036
1037         netif_napi_del(&q_vector->napi);
1038
1039 }
1040
1041 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1042 {
1043         int v_idx = adapter->num_q_vectors;
1044
1045         if (adapter->flags & IGB_FLAG_HAS_MSIX)
1046                 pci_disable_msix(adapter->pdev);
1047         else if (adapter->flags & IGB_FLAG_HAS_MSI)
1048                 pci_disable_msi(adapter->pdev);
1049
1050         while (v_idx--)
1051                 igb_reset_q_vector(adapter, v_idx);
1052 }
1053
1054 /**
1055  *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1056  *  @adapter: board private structure to initialize
1057  *
1058  *  This function frees the memory allocated to the q_vectors.  In addition if
1059  *  NAPI is enabled it will delete any references to the NAPI struct prior
1060  *  to freeing the q_vector.
1061  **/
1062 static void igb_free_q_vectors(struct igb_adapter *adapter)
1063 {
1064         int v_idx = adapter->num_q_vectors;
1065
1066         adapter->num_tx_queues = 0;
1067         adapter->num_rx_queues = 0;
1068         adapter->num_q_vectors = 0;
1069
1070         while (v_idx--) {
1071                 igb_reset_q_vector(adapter, v_idx);
1072                 igb_free_q_vector(adapter, v_idx);
1073         }
1074 }
1075
1076 /**
1077  *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1078  *  @adapter: board private structure to initialize
1079  *
1080  *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1081  *  MSI-X interrupts allocated.
1082  */
1083 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1084 {
1085         igb_free_q_vectors(adapter);
1086         igb_reset_interrupt_capability(adapter);
1087 }
1088
1089 /**
1090  *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1091  *  @adapter: board private structure to initialize
1092  *  @msix: boolean value of MSIX capability
1093  *
1094  *  Attempt to configure interrupts using the best available
1095  *  capabilities of the hardware and kernel.
1096  **/
1097 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1098 {
1099         int err;
1100         int numvecs, i;
1101
1102         if (!msix)
1103                 goto msi_only;
1104         adapter->flags |= IGB_FLAG_HAS_MSIX;
1105
1106         /* Number of supported queues. */
1107         adapter->num_rx_queues = adapter->rss_queues;
1108         if (adapter->vfs_allocated_count)
1109                 adapter->num_tx_queues = 1;
1110         else
1111                 adapter->num_tx_queues = adapter->rss_queues;
1112
1113         /* start with one vector for every Rx queue */
1114         numvecs = adapter->num_rx_queues;
1115
1116         /* if Tx handler is separate add 1 for every Tx queue */
1117         if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1118                 numvecs += adapter->num_tx_queues;
1119
1120         /* store the number of vectors reserved for queues */
1121         adapter->num_q_vectors = numvecs;
1122
1123         /* add 1 vector for link status interrupts */
1124         numvecs++;
1125         for (i = 0; i < numvecs; i++)
1126                 adapter->msix_entries[i].entry = i;
1127
1128         err = pci_enable_msix_range(adapter->pdev,
1129                                     adapter->msix_entries,
1130                                     numvecs,
1131                                     numvecs);
1132         if (err > 0)
1133                 return;
1134
1135         igb_reset_interrupt_capability(adapter);
1136
1137         /* If we can't do MSI-X, try MSI */
1138 msi_only:
1139         adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1140 #ifdef CONFIG_PCI_IOV
1141         /* disable SR-IOV for non MSI-X configurations */
1142         if (adapter->vf_data) {
1143                 struct e1000_hw *hw = &adapter->hw;
1144                 /* disable iov and allow time for transactions to clear */
1145                 pci_disable_sriov(adapter->pdev);
1146                 msleep(500);
1147
1148                 kfree(adapter->vf_mac_list);
1149                 adapter->vf_mac_list = NULL;
1150                 kfree(adapter->vf_data);
1151                 adapter->vf_data = NULL;
1152                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1153                 wrfl();
1154                 msleep(100);
1155                 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1156         }
1157 #endif
1158         adapter->vfs_allocated_count = 0;
1159         adapter->rss_queues = 1;
1160         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1161         adapter->num_rx_queues = 1;
1162         adapter->num_tx_queues = 1;
1163         adapter->num_q_vectors = 1;
1164         if (!pci_enable_msi(adapter->pdev))
1165                 adapter->flags |= IGB_FLAG_HAS_MSI;
1166 }
1167
1168 static void igb_add_ring(struct igb_ring *ring,
1169                          struct igb_ring_container *head)
1170 {
1171         head->ring = ring;
1172         head->count++;
1173 }
1174
1175 /**
1176  *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1177  *  @adapter: board private structure to initialize
1178  *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1179  *  @v_idx: index of vector in adapter struct
1180  *  @txr_count: total number of Tx rings to allocate
1181  *  @txr_idx: index of first Tx ring to allocate
1182  *  @rxr_count: total number of Rx rings to allocate
1183  *  @rxr_idx: index of first Rx ring to allocate
1184  *
1185  *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1186  **/
1187 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1188                               int v_count, int v_idx,
1189                               int txr_count, int txr_idx,
1190                               int rxr_count, int rxr_idx)
1191 {
1192         struct igb_q_vector *q_vector;
1193         struct igb_ring *ring;
1194         int ring_count;
1195         size_t size;
1196
1197         /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1198         if (txr_count > 1 || rxr_count > 1)
1199                 return -ENOMEM;
1200
1201         ring_count = txr_count + rxr_count;
1202         size = struct_size(q_vector, ring, ring_count);
1203
1204         /* allocate q_vector and rings */
1205         q_vector = adapter->q_vector[v_idx];
1206         if (!q_vector) {
1207                 q_vector = kzalloc(size, GFP_KERNEL);
1208         } else if (size > ksize(q_vector)) {
1209                 kfree_rcu(q_vector, rcu);
1210                 q_vector = kzalloc(size, GFP_KERNEL);
1211         } else {
1212                 memset(q_vector, 0, size);
1213         }
1214         if (!q_vector)
1215                 return -ENOMEM;
1216
1217         /* initialize NAPI */
1218         netif_napi_add(adapter->netdev, &q_vector->napi,
1219                        igb_poll, 64);
1220
1221         /* tie q_vector and adapter together */
1222         adapter->q_vector[v_idx] = q_vector;
1223         q_vector->adapter = adapter;
1224
1225         /* initialize work limits */
1226         q_vector->tx.work_limit = adapter->tx_work_limit;
1227
1228         /* initialize ITR configuration */
1229         q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1230         q_vector->itr_val = IGB_START_ITR;
1231
1232         /* initialize pointer to rings */
1233         ring = q_vector->ring;
1234
1235         /* intialize ITR */
1236         if (rxr_count) {
1237                 /* rx or rx/tx vector */
1238                 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1239                         q_vector->itr_val = adapter->rx_itr_setting;
1240         } else {
1241                 /* tx only vector */
1242                 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1243                         q_vector->itr_val = adapter->tx_itr_setting;
1244         }
1245
1246         if (txr_count) {
1247                 /* assign generic ring traits */
1248                 ring->dev = &adapter->pdev->dev;
1249                 ring->netdev = adapter->netdev;
1250
1251                 /* configure backlink on ring */
1252                 ring->q_vector = q_vector;
1253
1254                 /* update q_vector Tx values */
1255                 igb_add_ring(ring, &q_vector->tx);
1256
1257                 /* For 82575, context index must be unique per ring. */
1258                 if (adapter->hw.mac.type == e1000_82575)
1259                         set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1260
1261                 /* apply Tx specific ring traits */
1262                 ring->count = adapter->tx_ring_count;
1263                 ring->queue_index = txr_idx;
1264
1265                 ring->cbs_enable = false;
1266                 ring->idleslope = 0;
1267                 ring->sendslope = 0;
1268                 ring->hicredit = 0;
1269                 ring->locredit = 0;
1270
1271                 u64_stats_init(&ring->tx_syncp);
1272                 u64_stats_init(&ring->tx_syncp2);
1273
1274                 /* assign ring to adapter */
1275                 adapter->tx_ring[txr_idx] = ring;
1276
1277                 /* push pointer to next ring */
1278                 ring++;
1279         }
1280
1281         if (rxr_count) {
1282                 /* assign generic ring traits */
1283                 ring->dev = &adapter->pdev->dev;
1284                 ring->netdev = adapter->netdev;
1285
1286                 /* configure backlink on ring */
1287                 ring->q_vector = q_vector;
1288
1289                 /* update q_vector Rx values */
1290                 igb_add_ring(ring, &q_vector->rx);
1291
1292                 /* set flag indicating ring supports SCTP checksum offload */
1293                 if (adapter->hw.mac.type >= e1000_82576)
1294                         set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1295
1296                 /* On i350, i354, i210, and i211, loopback VLAN packets
1297                  * have the tag byte-swapped.
1298                  */
1299                 if (adapter->hw.mac.type >= e1000_i350)
1300                         set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1301
1302                 /* apply Rx specific ring traits */
1303                 ring->count = adapter->rx_ring_count;
1304                 ring->queue_index = rxr_idx;
1305
1306                 u64_stats_init(&ring->rx_syncp);
1307
1308                 /* assign ring to adapter */
1309                 adapter->rx_ring[rxr_idx] = ring;
1310         }
1311
1312         return 0;
1313 }
1314
1315
1316 /**
1317  *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1318  *  @adapter: board private structure to initialize
1319  *
1320  *  We allocate one q_vector per queue interrupt.  If allocation fails we
1321  *  return -ENOMEM.
1322  **/
1323 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1324 {
1325         int q_vectors = adapter->num_q_vectors;
1326         int rxr_remaining = adapter->num_rx_queues;
1327         int txr_remaining = adapter->num_tx_queues;
1328         int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1329         int err;
1330
1331         if (q_vectors >= (rxr_remaining + txr_remaining)) {
1332                 for (; rxr_remaining; v_idx++) {
1333                         err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1334                                                  0, 0, 1, rxr_idx);
1335
1336                         if (err)
1337                                 goto err_out;
1338
1339                         /* update counts and index */
1340                         rxr_remaining--;
1341                         rxr_idx++;
1342                 }
1343         }
1344
1345         for (; v_idx < q_vectors; v_idx++) {
1346                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1347                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1348
1349                 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1350                                          tqpv, txr_idx, rqpv, rxr_idx);
1351
1352                 if (err)
1353                         goto err_out;
1354
1355                 /* update counts and index */
1356                 rxr_remaining -= rqpv;
1357                 txr_remaining -= tqpv;
1358                 rxr_idx++;
1359                 txr_idx++;
1360         }
1361
1362         return 0;
1363
1364 err_out:
1365         adapter->num_tx_queues = 0;
1366         adapter->num_rx_queues = 0;
1367         adapter->num_q_vectors = 0;
1368
1369         while (v_idx--)
1370                 igb_free_q_vector(adapter, v_idx);
1371
1372         return -ENOMEM;
1373 }
1374
1375 /**
1376  *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1377  *  @adapter: board private structure to initialize
1378  *  @msix: boolean value of MSIX capability
1379  *
1380  *  This function initializes the interrupts and allocates all of the queues.
1381  **/
1382 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1383 {
1384         struct pci_dev *pdev = adapter->pdev;
1385         int err;
1386
1387         igb_set_interrupt_capability(adapter, msix);
1388
1389         err = igb_alloc_q_vectors(adapter);
1390         if (err) {
1391                 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1392                 goto err_alloc_q_vectors;
1393         }
1394
1395         igb_cache_ring_register(adapter);
1396
1397         return 0;
1398
1399 err_alloc_q_vectors:
1400         igb_reset_interrupt_capability(adapter);
1401         return err;
1402 }
1403
1404 /**
1405  *  igb_request_irq - initialize interrupts
1406  *  @adapter: board private structure to initialize
1407  *
1408  *  Attempts to configure interrupts using the best available
1409  *  capabilities of the hardware and kernel.
1410  **/
1411 static int igb_request_irq(struct igb_adapter *adapter)
1412 {
1413         struct net_device *netdev = adapter->netdev;
1414         struct pci_dev *pdev = adapter->pdev;
1415         int err = 0;
1416
1417         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1418                 err = igb_request_msix(adapter);
1419                 if (!err)
1420                         goto request_done;
1421                 /* fall back to MSI */
1422                 igb_free_all_tx_resources(adapter);
1423                 igb_free_all_rx_resources(adapter);
1424
1425                 igb_clear_interrupt_scheme(adapter);
1426                 err = igb_init_interrupt_scheme(adapter, false);
1427                 if (err)
1428                         goto request_done;
1429
1430                 igb_setup_all_tx_resources(adapter);
1431                 igb_setup_all_rx_resources(adapter);
1432                 igb_configure(adapter);
1433         }
1434
1435         igb_assign_vector(adapter->q_vector[0], 0);
1436
1437         if (adapter->flags & IGB_FLAG_HAS_MSI) {
1438                 err = request_irq(pdev->irq, igb_intr_msi, 0,
1439                                   netdev->name, adapter);
1440                 if (!err)
1441                         goto request_done;
1442
1443                 /* fall back to legacy interrupts */
1444                 igb_reset_interrupt_capability(adapter);
1445                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1446         }
1447
1448         err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1449                           netdev->name, adapter);
1450
1451         if (err)
1452                 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1453                         err);
1454
1455 request_done:
1456         return err;
1457 }
1458
1459 static void igb_free_irq(struct igb_adapter *adapter)
1460 {
1461         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1462                 int vector = 0, i;
1463
1464                 free_irq(adapter->msix_entries[vector++].vector, adapter);
1465
1466                 for (i = 0; i < adapter->num_q_vectors; i++)
1467                         free_irq(adapter->msix_entries[vector++].vector,
1468                                  adapter->q_vector[i]);
1469         } else {
1470                 free_irq(adapter->pdev->irq, adapter);
1471         }
1472 }
1473
1474 /**
1475  *  igb_irq_disable - Mask off interrupt generation on the NIC
1476  *  @adapter: board private structure
1477  **/
1478 static void igb_irq_disable(struct igb_adapter *adapter)
1479 {
1480         struct e1000_hw *hw = &adapter->hw;
1481
1482         /* we need to be careful when disabling interrupts.  The VFs are also
1483          * mapped into these registers and so clearing the bits can cause
1484          * issues on the VF drivers so we only need to clear what we set
1485          */
1486         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1487                 u32 regval = rd32(E1000_EIAM);
1488
1489                 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1490                 wr32(E1000_EIMC, adapter->eims_enable_mask);
1491                 regval = rd32(E1000_EIAC);
1492                 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1493         }
1494
1495         wr32(E1000_IAM, 0);
1496         wr32(E1000_IMC, ~0);
1497         wrfl();
1498         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1499                 int i;
1500
1501                 for (i = 0; i < adapter->num_q_vectors; i++)
1502                         synchronize_irq(adapter->msix_entries[i].vector);
1503         } else {
1504                 synchronize_irq(adapter->pdev->irq);
1505         }
1506 }
1507
1508 /**
1509  *  igb_irq_enable - Enable default interrupt generation settings
1510  *  @adapter: board private structure
1511  **/
1512 static void igb_irq_enable(struct igb_adapter *adapter)
1513 {
1514         struct e1000_hw *hw = &adapter->hw;
1515
1516         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1517                 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1518                 u32 regval = rd32(E1000_EIAC);
1519
1520                 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1521                 regval = rd32(E1000_EIAM);
1522                 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1523                 wr32(E1000_EIMS, adapter->eims_enable_mask);
1524                 if (adapter->vfs_allocated_count) {
1525                         wr32(E1000_MBVFIMR, 0xFF);
1526                         ims |= E1000_IMS_VMMB;
1527                 }
1528                 wr32(E1000_IMS, ims);
1529         } else {
1530                 wr32(E1000_IMS, IMS_ENABLE_MASK |
1531                                 E1000_IMS_DRSTA);
1532                 wr32(E1000_IAM, IMS_ENABLE_MASK |
1533                                 E1000_IMS_DRSTA);
1534         }
1535 }
1536
1537 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1538 {
1539         struct e1000_hw *hw = &adapter->hw;
1540         u16 pf_id = adapter->vfs_allocated_count;
1541         u16 vid = adapter->hw.mng_cookie.vlan_id;
1542         u16 old_vid = adapter->mng_vlan_id;
1543
1544         if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1545                 /* add VID to filter table */
1546                 igb_vfta_set(hw, vid, pf_id, true, true);
1547                 adapter->mng_vlan_id = vid;
1548         } else {
1549                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1550         }
1551
1552         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1553             (vid != old_vid) &&
1554             !test_bit(old_vid, adapter->active_vlans)) {
1555                 /* remove VID from filter table */
1556                 igb_vfta_set(hw, vid, pf_id, false, true);
1557         }
1558 }
1559
1560 /**
1561  *  igb_release_hw_control - release control of the h/w to f/w
1562  *  @adapter: address of board private structure
1563  *
1564  *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1565  *  For ASF and Pass Through versions of f/w this means that the
1566  *  driver is no longer loaded.
1567  **/
1568 static void igb_release_hw_control(struct igb_adapter *adapter)
1569 {
1570         struct e1000_hw *hw = &adapter->hw;
1571         u32 ctrl_ext;
1572
1573         /* Let firmware take over control of h/w */
1574         ctrl_ext = rd32(E1000_CTRL_EXT);
1575         wr32(E1000_CTRL_EXT,
1576                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1577 }
1578
1579 /**
1580  *  igb_get_hw_control - get control of the h/w from f/w
1581  *  @adapter: address of board private structure
1582  *
1583  *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1584  *  For ASF and Pass Through versions of f/w this means that
1585  *  the driver is loaded.
1586  **/
1587 static void igb_get_hw_control(struct igb_adapter *adapter)
1588 {
1589         struct e1000_hw *hw = &adapter->hw;
1590         u32 ctrl_ext;
1591
1592         /* Let firmware know the driver has taken over */
1593         ctrl_ext = rd32(E1000_CTRL_EXT);
1594         wr32(E1000_CTRL_EXT,
1595                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1596 }
1597
1598 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1599 {
1600         struct net_device *netdev = adapter->netdev;
1601         struct e1000_hw *hw = &adapter->hw;
1602
1603         WARN_ON(hw->mac.type != e1000_i210);
1604
1605         if (enable)
1606                 adapter->flags |= IGB_FLAG_FQTSS;
1607         else
1608                 adapter->flags &= ~IGB_FLAG_FQTSS;
1609
1610         if (netif_running(netdev))
1611                 schedule_work(&adapter->reset_task);
1612 }
1613
1614 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1615 {
1616         return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1617 }
1618
1619 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1620                                    enum tx_queue_prio prio)
1621 {
1622         u32 val;
1623
1624         WARN_ON(hw->mac.type != e1000_i210);
1625         WARN_ON(queue < 0 || queue > 4);
1626
1627         val = rd32(E1000_I210_TXDCTL(queue));
1628
1629         if (prio == TX_QUEUE_PRIO_HIGH)
1630                 val |= E1000_TXDCTL_PRIORITY;
1631         else
1632                 val &= ~E1000_TXDCTL_PRIORITY;
1633
1634         wr32(E1000_I210_TXDCTL(queue), val);
1635 }
1636
1637 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1638 {
1639         u32 val;
1640
1641         WARN_ON(hw->mac.type != e1000_i210);
1642         WARN_ON(queue < 0 || queue > 1);
1643
1644         val = rd32(E1000_I210_TQAVCC(queue));
1645
1646         if (mode == QUEUE_MODE_STREAM_RESERVATION)
1647                 val |= E1000_TQAVCC_QUEUEMODE;
1648         else
1649                 val &= ~E1000_TQAVCC_QUEUEMODE;
1650
1651         wr32(E1000_I210_TQAVCC(queue), val);
1652 }
1653
1654 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1655 {
1656         int i;
1657
1658         for (i = 0; i < adapter->num_tx_queues; i++) {
1659                 if (adapter->tx_ring[i]->cbs_enable)
1660                         return true;
1661         }
1662
1663         return false;
1664 }
1665
1666 static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1667 {
1668         int i;
1669
1670         for (i = 0; i < adapter->num_tx_queues; i++) {
1671                 if (adapter->tx_ring[i]->launchtime_enable)
1672                         return true;
1673         }
1674
1675         return false;
1676 }
1677
1678 /**
1679  *  igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1680  *  @adapter: pointer to adapter struct
1681  *  @queue: queue number
1682  *
1683  *  Configure CBS and Launchtime for a given hardware queue.
1684  *  Parameters are retrieved from the correct Tx ring, so
1685  *  igb_save_cbs_params() and igb_save_txtime_params() should be used
1686  *  for setting those correctly prior to this function being called.
1687  **/
1688 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1689 {
1690         struct igb_ring *ring = adapter->tx_ring[queue];
1691         struct net_device *netdev = adapter->netdev;
1692         struct e1000_hw *hw = &adapter->hw;
1693         u32 tqavcc, tqavctrl;
1694         u16 value;
1695
1696         WARN_ON(hw->mac.type != e1000_i210);
1697         WARN_ON(queue < 0 || queue > 1);
1698
1699         /* If any of the Qav features is enabled, configure queues as SR and
1700          * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1701          * as SP.
1702          */
1703         if (ring->cbs_enable || ring->launchtime_enable) {
1704                 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1705                 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1706         } else {
1707                 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1708                 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1709         }
1710
1711         /* If CBS is enabled, set DataTranARB and config its parameters. */
1712         if (ring->cbs_enable || queue == 0) {
1713                 /* i210 does not allow the queue 0 to be in the Strict
1714                  * Priority mode while the Qav mode is enabled, so,
1715                  * instead of disabling strict priority mode, we give
1716                  * queue 0 the maximum of credits possible.
1717                  *
1718                  * See section 8.12.19 of the i210 datasheet, "Note:
1719                  * Queue0 QueueMode must be set to 1b when
1720                  * TransmitMode is set to Qav."
1721                  */
1722                 if (queue == 0 && !ring->cbs_enable) {
1723                         /* max "linkspeed" idleslope in kbps */
1724                         ring->idleslope = 1000000;
1725                         ring->hicredit = ETH_FRAME_LEN;
1726                 }
1727
1728                 /* Always set data transfer arbitration to credit-based
1729                  * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1730                  * the queues.
1731                  */
1732                 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1733                 tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1734                 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1735
1736                 /* According to i210 datasheet section 7.2.7.7, we should set
1737                  * the 'idleSlope' field from TQAVCC register following the
1738                  * equation:
1739                  *
1740                  * For 100 Mbps link speed:
1741                  *
1742                  *     value = BW * 0x7735 * 0.2                          (E1)
1743                  *
1744                  * For 1000Mbps link speed:
1745                  *
1746                  *     value = BW * 0x7735 * 2                            (E2)
1747                  *
1748                  * E1 and E2 can be merged into one equation as shown below.
1749                  * Note that 'link-speed' is in Mbps.
1750                  *
1751                  *     value = BW * 0x7735 * 2 * link-speed
1752                  *                           --------------               (E3)
1753                  *                                1000
1754                  *
1755                  * 'BW' is the percentage bandwidth out of full link speed
1756                  * which can be found with the following equation. Note that
1757                  * idleSlope here is the parameter from this function which
1758                  * is in kbps.
1759                  *
1760                  *     BW =     idleSlope
1761                  *          -----------------                             (E4)
1762                  *          link-speed * 1000
1763                  *
1764                  * That said, we can come up with a generic equation to
1765                  * calculate the value we should set it TQAVCC register by
1766                  * replacing 'BW' in E3 by E4. The resulting equation is:
1767                  *
1768                  * value =     idleSlope     * 0x7735 * 2 * link-speed
1769                  *         -----------------            --------------    (E5)
1770                  *         link-speed * 1000                 1000
1771                  *
1772                  * 'link-speed' is present in both sides of the fraction so
1773                  * it is canceled out. The final equation is the following:
1774                  *
1775                  *     value = idleSlope * 61034
1776                  *             -----------------                          (E6)
1777                  *                  1000000
1778                  *
1779                  * NOTE: For i210, given the above, we can see that idleslope
1780                  *       is represented in 16.38431 kbps units by the value at
1781                  *       the TQAVCC register (1Gbps / 61034), which reduces
1782                  *       the granularity for idleslope increments.
1783                  *       For instance, if you want to configure a 2576kbps
1784                  *       idleslope, the value to be written on the register
1785                  *       would have to be 157.23. If rounded down, you end
1786                  *       up with less bandwidth available than originally
1787                  *       required (~2572 kbps). If rounded up, you end up
1788                  *       with a higher bandwidth (~2589 kbps). Below the
1789                  *       approach we take is to always round up the
1790                  *       calculated value, so the resulting bandwidth might
1791                  *       be slightly higher for some configurations.
1792                  */
1793                 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1794
1795                 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1796                 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1797                 tqavcc |= value;
1798                 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1799
1800                 wr32(E1000_I210_TQAVHC(queue),
1801                      0x80000000 + ring->hicredit * 0x7735);
1802         } else {
1803
1804                 /* Set idleSlope to zero. */
1805                 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1806                 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1807                 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1808
1809                 /* Set hiCredit to zero. */
1810                 wr32(E1000_I210_TQAVHC(queue), 0);
1811
1812                 /* If CBS is not enabled for any queues anymore, then return to
1813                  * the default state of Data Transmission Arbitration on
1814                  * TQAVCTRL.
1815                  */
1816                 if (!is_any_cbs_enabled(adapter)) {
1817                         tqavctrl = rd32(E1000_I210_TQAVCTRL);
1818                         tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1819                         wr32(E1000_I210_TQAVCTRL, tqavctrl);
1820                 }
1821         }
1822
1823         /* If LaunchTime is enabled, set DataTranTIM. */
1824         if (ring->launchtime_enable) {
1825                 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1826                  * for any of the SR queues, and configure fetchtime delta.
1827                  * XXX NOTE:
1828                  *     - LaunchTime will be enabled for all SR queues.
1829                  *     - A fixed offset can be added relative to the launch
1830                  *       time of all packets if configured at reg LAUNCH_OS0.
1831                  *       We are keeping it as 0 for now (default value).
1832                  */
1833                 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1834                 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1835                        E1000_TQAVCTRL_FETCHTIME_DELTA;
1836                 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1837         } else {
1838                 /* If Launchtime is not enabled for any SR queues anymore,
1839                  * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1840                  * effectively disabling Launchtime.
1841                  */
1842                 if (!is_any_txtime_enabled(adapter)) {
1843                         tqavctrl = rd32(E1000_I210_TQAVCTRL);
1844                         tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1845                         tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1846                         wr32(E1000_I210_TQAVCTRL, tqavctrl);
1847                 }
1848         }
1849
1850         /* XXX: In i210 controller the sendSlope and loCredit parameters from
1851          * CBS are not configurable by software so we don't do any 'controller
1852          * configuration' in respect to these parameters.
1853          */
1854
1855         netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1856                    ring->cbs_enable ? "enabled" : "disabled",
1857                    ring->launchtime_enable ? "enabled" : "disabled",
1858                    queue,
1859                    ring->idleslope, ring->sendslope,
1860                    ring->hicredit, ring->locredit);
1861 }
1862
1863 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1864                                   bool enable)
1865 {
1866         struct igb_ring *ring;
1867
1868         if (queue < 0 || queue > adapter->num_tx_queues)
1869                 return -EINVAL;
1870
1871         ring = adapter->tx_ring[queue];
1872         ring->launchtime_enable = enable;
1873
1874         return 0;
1875 }
1876
1877 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1878                                bool enable, int idleslope, int sendslope,
1879                                int hicredit, int locredit)
1880 {
1881         struct igb_ring *ring;
1882
1883         if (queue < 0 || queue > adapter->num_tx_queues)
1884                 return -EINVAL;
1885
1886         ring = adapter->tx_ring[queue];
1887
1888         ring->cbs_enable = enable;
1889         ring->idleslope = idleslope;
1890         ring->sendslope = sendslope;
1891         ring->hicredit = hicredit;
1892         ring->locredit = locredit;
1893
1894         return 0;
1895 }
1896
1897 /**
1898  *  igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1899  *  @adapter: pointer to adapter struct
1900  *
1901  *  Configure TQAVCTRL register switching the controller's Tx mode
1902  *  if FQTSS mode is enabled or disabled. Additionally, will issue
1903  *  a call to igb_config_tx_modes() per queue so any previously saved
1904  *  Tx parameters are applied.
1905  **/
1906 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1907 {
1908         struct net_device *netdev = adapter->netdev;
1909         struct e1000_hw *hw = &adapter->hw;
1910         u32 val;
1911
1912         /* Only i210 controller supports changing the transmission mode. */
1913         if (hw->mac.type != e1000_i210)
1914                 return;
1915
1916         if (is_fqtss_enabled(adapter)) {
1917                 int i, max_queue;
1918
1919                 /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1920                  * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1921                  * so SP queues wait for SR ones.
1922                  */
1923                 val = rd32(E1000_I210_TQAVCTRL);
1924                 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1925                 val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1926                 wr32(E1000_I210_TQAVCTRL, val);
1927
1928                 /* Configure Tx and Rx packet buffers sizes as described in
1929                  * i210 datasheet section 7.2.7.7.
1930                  */
1931                 val = rd32(E1000_TXPBS);
1932                 val &= ~I210_TXPBSIZE_MASK;
1933                 val |= I210_TXPBSIZE_PB0_8KB | I210_TXPBSIZE_PB1_8KB |
1934                         I210_TXPBSIZE_PB2_4KB | I210_TXPBSIZE_PB3_4KB;
1935                 wr32(E1000_TXPBS, val);
1936
1937                 val = rd32(E1000_RXPBS);
1938                 val &= ~I210_RXPBSIZE_MASK;
1939                 val |= I210_RXPBSIZE_PB_30KB;
1940                 wr32(E1000_RXPBS, val);
1941
1942                 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1943                  * register should not exceed the buffer size programmed in
1944                  * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1945                  * so according to the datasheet we should set MAX_TPKT_SIZE to
1946                  * 4kB / 64.
1947                  *
1948                  * However, when we do so, no frame from queue 2 and 3 are
1949                  * transmitted.  It seems the MAX_TPKT_SIZE should not be great
1950                  * or _equal_ to the buffer size programmed in TXPBS. For this
1951                  * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1952                  */
1953                 val = (4096 - 1) / 64;
1954                 wr32(E1000_I210_DTXMXPKTSZ, val);
1955
1956                 /* Since FQTSS mode is enabled, apply any CBS configuration
1957                  * previously set. If no previous CBS configuration has been
1958                  * done, then the initial configuration is applied, which means
1959                  * CBS is disabled.
1960                  */
1961                 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1962                             adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1963
1964                 for (i = 0; i < max_queue; i++) {
1965                         igb_config_tx_modes(adapter, i);
1966                 }
1967         } else {
1968                 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1969                 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1970                 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1971
1972                 val = rd32(E1000_I210_TQAVCTRL);
1973                 /* According to Section 8.12.21, the other flags we've set when
1974                  * enabling FQTSS are not relevant when disabling FQTSS so we
1975                  * don't set they here.
1976                  */
1977                 val &= ~E1000_TQAVCTRL_XMIT_MODE;
1978                 wr32(E1000_I210_TQAVCTRL, val);
1979         }
1980
1981         netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1982                    "enabled" : "disabled");
1983 }
1984
1985 /**
1986  *  igb_configure - configure the hardware for RX and TX
1987  *  @adapter: private board structure
1988  **/
1989 static void igb_configure(struct igb_adapter *adapter)
1990 {
1991         struct net_device *netdev = adapter->netdev;
1992         int i;
1993
1994         igb_get_hw_control(adapter);
1995         igb_set_rx_mode(netdev);
1996         igb_setup_tx_mode(adapter);
1997
1998         igb_restore_vlan(adapter);
1999
2000         igb_setup_tctl(adapter);
2001         igb_setup_mrqc(adapter);
2002         igb_setup_rctl(adapter);
2003
2004         igb_nfc_filter_restore(adapter);
2005         igb_configure_tx(adapter);
2006         igb_configure_rx(adapter);
2007
2008         igb_rx_fifo_flush_82575(&adapter->hw);
2009
2010         /* call igb_desc_unused which always leaves
2011          * at least 1 descriptor unused to make sure
2012          * next_to_use != next_to_clean
2013          */
2014         for (i = 0; i < adapter->num_rx_queues; i++) {
2015                 struct igb_ring *ring = adapter->rx_ring[i];
2016                 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
2017         }
2018 }
2019
2020 /**
2021  *  igb_power_up_link - Power up the phy/serdes link
2022  *  @adapter: address of board private structure
2023  **/
2024 void igb_power_up_link(struct igb_adapter *adapter)
2025 {
2026         igb_reset_phy(&adapter->hw);
2027
2028         if (adapter->hw.phy.media_type == e1000_media_type_copper)
2029                 igb_power_up_phy_copper(&adapter->hw);
2030         else
2031                 igb_power_up_serdes_link_82575(&adapter->hw);
2032
2033         igb_setup_link(&adapter->hw);
2034 }
2035
2036 /**
2037  *  igb_power_down_link - Power down the phy/serdes link
2038  *  @adapter: address of board private structure
2039  */
2040 static void igb_power_down_link(struct igb_adapter *adapter)
2041 {
2042         if (adapter->hw.phy.media_type == e1000_media_type_copper)
2043                 igb_power_down_phy_copper_82575(&adapter->hw);
2044         else
2045                 igb_shutdown_serdes_link_82575(&adapter->hw);
2046 }
2047
2048 /**
2049  * Detect and switch function for Media Auto Sense
2050  * @adapter: address of the board private structure
2051  **/
2052 static void igb_check_swap_media(struct igb_adapter *adapter)
2053 {
2054         struct e1000_hw *hw = &adapter->hw;
2055         u32 ctrl_ext, connsw;
2056         bool swap_now = false;
2057
2058         ctrl_ext = rd32(E1000_CTRL_EXT);
2059         connsw = rd32(E1000_CONNSW);
2060
2061         /* need to live swap if current media is copper and we have fiber/serdes
2062          * to go to.
2063          */
2064
2065         if ((hw->phy.media_type == e1000_media_type_copper) &&
2066             (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2067                 swap_now = true;
2068         } else if ((hw->phy.media_type != e1000_media_type_copper) &&
2069                    !(connsw & E1000_CONNSW_SERDESD)) {
2070                 /* copper signal takes time to appear */
2071                 if (adapter->copper_tries < 4) {
2072                         adapter->copper_tries++;
2073                         connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2074                         wr32(E1000_CONNSW, connsw);
2075                         return;
2076                 } else {
2077                         adapter->copper_tries = 0;
2078                         if ((connsw & E1000_CONNSW_PHYSD) &&
2079                             (!(connsw & E1000_CONNSW_PHY_PDN))) {
2080                                 swap_now = true;
2081                                 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2082                                 wr32(E1000_CONNSW, connsw);
2083                         }
2084                 }
2085         }
2086
2087         if (!swap_now)
2088                 return;
2089
2090         switch (hw->phy.media_type) {
2091         case e1000_media_type_copper:
2092                 netdev_info(adapter->netdev,
2093                         "MAS: changing media to fiber/serdes\n");
2094                 ctrl_ext |=
2095                         E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2096                 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2097                 adapter->copper_tries = 0;
2098                 break;
2099         case e1000_media_type_internal_serdes:
2100         case e1000_media_type_fiber:
2101                 netdev_info(adapter->netdev,
2102                         "MAS: changing media to copper\n");
2103                 ctrl_ext &=
2104                         ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2105                 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2106                 break;
2107         default:
2108                 /* shouldn't get here during regular operation */
2109                 netdev_err(adapter->netdev,
2110                         "AMS: Invalid media type found, returning\n");
2111                 break;
2112         }
2113         wr32(E1000_CTRL_EXT, ctrl_ext);
2114 }
2115
2116 /**
2117  *  igb_up - Open the interface and prepare it to handle traffic
2118  *  @adapter: board private structure
2119  **/
2120 int igb_up(struct igb_adapter *adapter)
2121 {
2122         struct e1000_hw *hw = &adapter->hw;
2123         int i;
2124
2125         /* hardware has been reset, we need to reload some things */
2126         igb_configure(adapter);
2127
2128         clear_bit(__IGB_DOWN, &adapter->state);
2129
2130         for (i = 0; i < adapter->num_q_vectors; i++)
2131                 napi_enable(&(adapter->q_vector[i]->napi));
2132
2133         if (adapter->flags & IGB_FLAG_HAS_MSIX)
2134                 igb_configure_msix(adapter);
2135         else
2136                 igb_assign_vector(adapter->q_vector[0], 0);
2137
2138         /* Clear any pending interrupts. */
2139         rd32(E1000_TSICR);
2140         rd32(E1000_ICR);
2141         igb_irq_enable(adapter);
2142
2143         /* notify VFs that reset has been completed */
2144         if (adapter->vfs_allocated_count) {
2145                 u32 reg_data = rd32(E1000_CTRL_EXT);
2146
2147                 reg_data |= E1000_CTRL_EXT_PFRSTD;
2148                 wr32(E1000_CTRL_EXT, reg_data);
2149         }
2150
2151         netif_tx_start_all_queues(adapter->netdev);
2152
2153         /* start the watchdog. */
2154         hw->mac.get_link_status = 1;
2155         schedule_work(&adapter->watchdog_task);
2156
2157         if ((adapter->flags & IGB_FLAG_EEE) &&
2158             (!hw->dev_spec._82575.eee_disable))
2159                 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2160
2161         return 0;
2162 }
2163
2164 void igb_down(struct igb_adapter *adapter)
2165 {
2166         struct net_device *netdev = adapter->netdev;
2167         struct e1000_hw *hw = &adapter->hw;
2168         u32 tctl, rctl;
2169         int i;
2170
2171         /* signal that we're down so the interrupt handler does not
2172          * reschedule our watchdog timer
2173          */
2174         set_bit(__IGB_DOWN, &adapter->state);
2175
2176         /* disable receives in the hardware */
2177         rctl = rd32(E1000_RCTL);
2178         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2179         /* flush and sleep below */
2180
2181         igb_nfc_filter_exit(adapter);
2182
2183         netif_carrier_off(netdev);
2184         netif_tx_stop_all_queues(netdev);
2185
2186         /* disable transmits in the hardware */
2187         tctl = rd32(E1000_TCTL);
2188         tctl &= ~E1000_TCTL_EN;
2189         wr32(E1000_TCTL, tctl);
2190         /* flush both disables and wait for them to finish */
2191         wrfl();
2192         usleep_range(10000, 11000);
2193
2194         igb_irq_disable(adapter);
2195
2196         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2197
2198         for (i = 0; i < adapter->num_q_vectors; i++) {
2199                 if (adapter->q_vector[i]) {
2200                         napi_synchronize(&adapter->q_vector[i]->napi);
2201                         napi_disable(&adapter->q_vector[i]->napi);
2202                 }
2203         }
2204
2205         del_timer_sync(&adapter->watchdog_timer);
2206         del_timer_sync(&adapter->phy_info_timer);
2207
2208         /* record the stats before reset*/
2209         spin_lock(&adapter->stats64_lock);
2210         igb_update_stats(adapter);
2211         spin_unlock(&adapter->stats64_lock);
2212
2213         adapter->link_speed = 0;
2214         adapter->link_duplex = 0;
2215
2216         if (!pci_channel_offline(adapter->pdev))
2217                 igb_reset(adapter);
2218
2219         /* clear VLAN promisc flag so VFTA will be updated if necessary */
2220         adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2221
2222         igb_clean_all_tx_rings(adapter);
2223         igb_clean_all_rx_rings(adapter);
2224 #ifdef CONFIG_IGB_DCA
2225
2226         /* since we reset the hardware DCA settings were cleared */
2227         igb_setup_dca(adapter);
2228 #endif
2229 }
2230
2231 void igb_reinit_locked(struct igb_adapter *adapter)
2232 {
2233         WARN_ON(in_interrupt());
2234         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2235                 usleep_range(1000, 2000);
2236         igb_down(adapter);
2237         igb_up(adapter);
2238         clear_bit(__IGB_RESETTING, &adapter->state);
2239 }
2240
2241 /** igb_enable_mas - Media Autosense re-enable after swap
2242  *
2243  * @adapter: adapter struct
2244  **/
2245 static void igb_enable_mas(struct igb_adapter *adapter)
2246 {
2247         struct e1000_hw *hw = &adapter->hw;
2248         u32 connsw = rd32(E1000_CONNSW);
2249
2250         /* configure for SerDes media detect */
2251         if ((hw->phy.media_type == e1000_media_type_copper) &&
2252             (!(connsw & E1000_CONNSW_SERDESD))) {
2253                 connsw |= E1000_CONNSW_ENRGSRC;
2254                 connsw |= E1000_CONNSW_AUTOSENSE_EN;
2255                 wr32(E1000_CONNSW, connsw);
2256                 wrfl();
2257         }
2258 }
2259
2260 void igb_reset(struct igb_adapter *adapter)
2261 {
2262         struct pci_dev *pdev = adapter->pdev;
2263         struct e1000_hw *hw = &adapter->hw;
2264         struct e1000_mac_info *mac = &hw->mac;
2265         struct e1000_fc_info *fc = &hw->fc;
2266         u32 pba, hwm;
2267
2268         /* Repartition Pba for greater than 9k mtu
2269          * To take effect CTRL.RST is required.
2270          */
2271         switch (mac->type) {
2272         case e1000_i350:
2273         case e1000_i354:
2274         case e1000_82580:
2275                 pba = rd32(E1000_RXPBS);
2276                 pba = igb_rxpbs_adjust_82580(pba);
2277                 break;
2278         case e1000_82576:
2279                 pba = rd32(E1000_RXPBS);
2280                 pba &= E1000_RXPBS_SIZE_MASK_82576;
2281                 break;
2282         case e1000_82575:
2283         case e1000_i210:
2284         case e1000_i211:
2285         default:
2286                 pba = E1000_PBA_34K;
2287                 break;
2288         }
2289
2290         if (mac->type == e1000_82575) {
2291                 u32 min_rx_space, min_tx_space, needed_tx_space;
2292
2293                 /* write Rx PBA so that hardware can report correct Tx PBA */
2294                 wr32(E1000_PBA, pba);
2295
2296                 /* To maintain wire speed transmits, the Tx FIFO should be
2297                  * large enough to accommodate two full transmit packets,
2298                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2299                  * the Rx FIFO should be large enough to accommodate at least
2300                  * one full receive packet and is similarly rounded up and
2301                  * expressed in KB.
2302                  */
2303                 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2304
2305                 /* The Tx FIFO also stores 16 bytes of information about the Tx
2306                  * but don't include Ethernet FCS because hardware appends it.
2307                  * We only need to round down to the nearest 512 byte block
2308                  * count since the value we care about is 2 frames, not 1.
2309                  */
2310                 min_tx_space = adapter->max_frame_size;
2311                 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2312                 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2313
2314                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2315                 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2316
2317                 /* If current Tx allocation is less than the min Tx FIFO size,
2318                  * and the min Tx FIFO size is less than the current Rx FIFO
2319                  * allocation, take space away from current Rx allocation.
2320                  */
2321                 if (needed_tx_space < pba) {
2322                         pba -= needed_tx_space;
2323
2324                         /* if short on Rx space, Rx wins and must trump Tx
2325                          * adjustment
2326                          */
2327                         if (pba < min_rx_space)
2328                                 pba = min_rx_space;
2329                 }
2330
2331                 /* adjust PBA for jumbo frames */
2332                 wr32(E1000_PBA, pba);
2333         }
2334
2335         /* flow control settings
2336          * The high water mark must be low enough to fit one full frame
2337          * after transmitting the pause frame.  As such we must have enough
2338          * space to allow for us to complete our current transmit and then
2339          * receive the frame that is in progress from the link partner.
2340          * Set it to:
2341          * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2342          */
2343         hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2344
2345         fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
2346         fc->low_water = fc->high_water - 16;
2347         fc->pause_time = 0xFFFF;
2348         fc->send_xon = 1;
2349         fc->current_mode = fc->requested_mode;
2350
2351         /* disable receive for all VFs and wait one second */
2352         if (adapter->vfs_allocated_count) {
2353                 int i;
2354
2355                 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2356                         adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2357
2358                 /* ping all the active vfs to let them know we are going down */
2359                 igb_ping_all_vfs(adapter);
2360
2361                 /* disable transmits and receives */
2362                 wr32(E1000_VFRE, 0);
2363                 wr32(E1000_VFTE, 0);
2364         }
2365
2366         /* Allow time for pending master requests to run */
2367         hw->mac.ops.reset_hw(hw);
2368         wr32(E1000_WUC, 0);
2369
2370         if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2371                 /* need to resetup here after media swap */
2372                 adapter->ei.get_invariants(hw);
2373                 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2374         }
2375         if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
2376             (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2377                 igb_enable_mas(adapter);
2378         }
2379         if (hw->mac.ops.init_hw(hw))
2380                 dev_err(&pdev->dev, "Hardware Error\n");
2381
2382         /* RAR registers were cleared during init_hw, clear mac table */
2383         igb_flush_mac_table(adapter);
2384         __dev_uc_unsync(adapter->netdev, NULL);
2385
2386         /* Recover default RAR entry */
2387         igb_set_default_mac_filter(adapter);
2388
2389         /* Flow control settings reset on hardware reset, so guarantee flow
2390          * control is off when forcing speed.
2391          */
2392         if (!hw->mac.autoneg)
2393                 igb_force_mac_fc(hw);
2394
2395         igb_init_dmac(adapter, pba);
2396 #ifdef CONFIG_IGB_HWMON
2397         /* Re-initialize the thermal sensor on i350 devices. */
2398         if (!test_bit(__IGB_DOWN, &adapter->state)) {
2399                 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2400                         /* If present, re-initialize the external thermal sensor
2401                          * interface.
2402                          */
2403                         if (adapter->ets)
2404                                 mac->ops.init_thermal_sensor_thresh(hw);
2405                 }
2406         }
2407 #endif
2408         /* Re-establish EEE setting */
2409         if (hw->phy.media_type == e1000_media_type_copper) {
2410                 switch (mac->type) {
2411                 case e1000_i350:
2412                 case e1000_i210:
2413                 case e1000_i211:
2414                         igb_set_eee_i350(hw, true, true);
2415                         break;
2416                 case e1000_i354:
2417                         igb_set_eee_i354(hw, true, true);
2418                         break;
2419                 default:
2420                         break;
2421                 }
2422         }
2423         if (!netif_running(adapter->netdev))
2424                 igb_power_down_link(adapter);
2425
2426         igb_update_mng_vlan(adapter);
2427
2428         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2429         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2430
2431         /* Re-enable PTP, where applicable. */
2432         if (adapter->ptp_flags & IGB_PTP_ENABLED)
2433                 igb_ptp_reset(adapter);
2434
2435         igb_get_phy_info(hw);
2436 }
2437
2438 static netdev_features_t igb_fix_features(struct net_device *netdev,
2439         netdev_features_t features)
2440 {
2441         /* Since there is no support for separate Rx/Tx vlan accel
2442          * enable/disable make sure Tx flag is always in same state as Rx.
2443          */
2444         if (features & NETIF_F_HW_VLAN_CTAG_RX)
2445                 features |= NETIF_F_HW_VLAN_CTAG_TX;
2446         else
2447                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2448
2449         return features;
2450 }
2451
2452 static int igb_set_features(struct net_device *netdev,
2453         netdev_features_t features)
2454 {
2455         netdev_features_t changed = netdev->features ^ features;
2456         struct igb_adapter *adapter = netdev_priv(netdev);
2457
2458         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2459                 igb_vlan_mode(netdev, features);
2460
2461         if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2462                 return 0;
2463
2464         if (!(features & NETIF_F_NTUPLE)) {
2465                 struct hlist_node *node2;
2466                 struct igb_nfc_filter *rule;
2467
2468                 spin_lock(&adapter->nfc_lock);
2469                 hlist_for_each_entry_safe(rule, node2,
2470                                           &adapter->nfc_filter_list, nfc_node) {
2471                         igb_erase_filter(adapter, rule);
2472                         hlist_del(&rule->nfc_node);
2473                         kfree(rule);
2474                 }
2475                 spin_unlock(&adapter->nfc_lock);
2476                 adapter->nfc_filter_count = 0;
2477         }
2478
2479         netdev->features = features;
2480
2481         if (netif_running(netdev))
2482                 igb_reinit_locked(adapter);
2483         else
2484                 igb_reset(adapter);
2485
2486         return 1;
2487 }
2488
2489 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2490                            struct net_device *dev,
2491                            const unsigned char *addr, u16 vid,
2492                            u16 flags,
2493                            struct netlink_ext_ack *extack)
2494 {
2495         /* guarantee we can provide a unique filter for the unicast address */
2496         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2497                 struct igb_adapter *adapter = netdev_priv(dev);
2498                 int vfn = adapter->vfs_allocated_count;
2499
2500                 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2501                         return -ENOMEM;
2502         }
2503
2504         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2505 }
2506
2507 #define IGB_MAX_MAC_HDR_LEN     127
2508 #define IGB_MAX_NETWORK_HDR_LEN 511
2509
2510 static netdev_features_t
2511 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2512                    netdev_features_t features)
2513 {
2514         unsigned int network_hdr_len, mac_hdr_len;
2515
2516         /* Make certain the headers can be described by a context descriptor */
2517         mac_hdr_len = skb_network_header(skb) - skb->data;
2518         if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2519                 return features & ~(NETIF_F_HW_CSUM |
2520                                     NETIF_F_SCTP_CRC |
2521                                     NETIF_F_GSO_UDP_L4 |
2522                                     NETIF_F_HW_VLAN_CTAG_TX |
2523                                     NETIF_F_TSO |
2524                                     NETIF_F_TSO6);
2525
2526         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2527         if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
2528                 return features & ~(NETIF_F_HW_CSUM |
2529                                     NETIF_F_SCTP_CRC |
2530                                     NETIF_F_GSO_UDP_L4 |
2531                                     NETIF_F_TSO |
2532                                     NETIF_F_TSO6);
2533
2534         /* We can only support IPV4 TSO in tunnels if we can mangle the
2535          * inner IP ID field, so strip TSO if MANGLEID is not supported.
2536          */
2537         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2538                 features &= ~NETIF_F_TSO;
2539
2540         return features;
2541 }
2542
2543 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2544 {
2545         if (!is_fqtss_enabled(adapter)) {
2546                 enable_fqtss(adapter, true);
2547                 return;
2548         }
2549
2550         igb_config_tx_modes(adapter, queue);
2551
2552         if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2553                 enable_fqtss(adapter, false);
2554 }
2555
2556 static int igb_offload_cbs(struct igb_adapter *adapter,
2557                            struct tc_cbs_qopt_offload *qopt)
2558 {
2559         struct e1000_hw *hw = &adapter->hw;
2560         int err;
2561
2562         /* CBS offloading is only supported by i210 controller. */
2563         if (hw->mac.type != e1000_i210)
2564                 return -EOPNOTSUPP;
2565
2566         /* CBS offloading is only supported by queue 0 and queue 1. */
2567         if (qopt->queue < 0 || qopt->queue > 1)
2568                 return -EINVAL;
2569
2570         err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2571                                   qopt->idleslope, qopt->sendslope,
2572                                   qopt->hicredit, qopt->locredit);
2573         if (err)
2574                 return err;
2575
2576         igb_offload_apply(adapter, qopt->queue);
2577
2578         return 0;
2579 }
2580
2581 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2582 #define VLAN_PRIO_FULL_MASK (0x07)
2583
2584 static int igb_parse_cls_flower(struct igb_adapter *adapter,
2585                                 struct flow_cls_offload *f,
2586                                 int traffic_class,
2587                                 struct igb_nfc_filter *input)
2588 {
2589         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2590         struct flow_dissector *dissector = rule->match.dissector;
2591         struct netlink_ext_ack *extack = f->common.extack;
2592
2593         if (dissector->used_keys &
2594             ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
2595               BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2596               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2597               BIT(FLOW_DISSECTOR_KEY_VLAN))) {
2598                 NL_SET_ERR_MSG_MOD(extack,
2599                                    "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2600                 return -EOPNOTSUPP;
2601         }
2602
2603         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2604                 struct flow_match_eth_addrs match;
2605
2606                 flow_rule_match_eth_addrs(rule, &match);
2607                 if (!is_zero_ether_addr(match.mask->dst)) {
2608                         if (!is_broadcast_ether_addr(match.mask->dst)) {
2609                                 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2610                                 return -EINVAL;
2611                         }
2612
2613                         input->filter.match_flags |=
2614                                 IGB_FILTER_FLAG_DST_MAC_ADDR;
2615                         ether_addr_copy(input->filter.dst_addr, match.key->dst);
2616                 }
2617
2618                 if (!is_zero_ether_addr(match.mask->src)) {
2619                         if (!is_broadcast_ether_addr(match.mask->src)) {
2620                                 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2621                                 return -EINVAL;
2622                         }
2623
2624                         input->filter.match_flags |=
2625                                 IGB_FILTER_FLAG_SRC_MAC_ADDR;
2626                         ether_addr_copy(input->filter.src_addr, match.key->src);
2627                 }
2628         }
2629
2630         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2631                 struct flow_match_basic match;
2632
2633                 flow_rule_match_basic(rule, &match);
2634                 if (match.mask->n_proto) {
2635                         if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
2636                                 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2637                                 return -EINVAL;
2638                         }
2639
2640                         input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2641                         input->filter.etype = match.key->n_proto;
2642                 }
2643         }
2644
2645         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
2646                 struct flow_match_vlan match;
2647
2648                 flow_rule_match_vlan(rule, &match);
2649                 if (match.mask->vlan_priority) {
2650                         if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2651                                 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2652                                 return -EINVAL;
2653                         }
2654
2655                         input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2656                         input->filter.vlan_tci = match.key->vlan_priority;
2657                 }
2658         }
2659
2660         input->action = traffic_class;
2661         input->cookie = f->cookie;
2662
2663         return 0;
2664 }
2665
2666 static int igb_configure_clsflower(struct igb_adapter *adapter,
2667                                    struct flow_cls_offload *cls_flower)
2668 {
2669         struct netlink_ext_ack *extack = cls_flower->common.extack;
2670         struct igb_nfc_filter *filter, *f;
2671         int err, tc;
2672
2673         tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2674         if (tc < 0) {
2675                 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2676                 return -EINVAL;
2677         }
2678
2679         filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2680         if (!filter)
2681                 return -ENOMEM;
2682
2683         err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2684         if (err < 0)
2685                 goto err_parse;
2686
2687         spin_lock(&adapter->nfc_lock);
2688
2689         hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2690                 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2691                         err = -EEXIST;
2692                         NL_SET_ERR_MSG_MOD(extack,
2693                                            "This filter is already set in ethtool");
2694                         goto err_locked;
2695                 }
2696         }
2697
2698         hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2699                 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2700                         err = -EEXIST;
2701                         NL_SET_ERR_MSG_MOD(extack,
2702                                            "This filter is already set in cls_flower");
2703                         goto err_locked;
2704                 }
2705         }
2706
2707         err = igb_add_filter(adapter, filter);
2708         if (err < 0) {
2709                 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2710                 goto err_locked;
2711         }
2712
2713         hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2714
2715         spin_unlock(&adapter->nfc_lock);
2716
2717         return 0;
2718
2719 err_locked:
2720         spin_unlock(&adapter->nfc_lock);
2721
2722 err_parse:
2723         kfree(filter);
2724
2725         return err;
2726 }
2727
2728 static int igb_delete_clsflower(struct igb_adapter *adapter,
2729                                 struct flow_cls_offload *cls_flower)
2730 {
2731         struct igb_nfc_filter *filter;
2732         int err;
2733
2734         spin_lock(&adapter->nfc_lock);
2735
2736         hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2737                 if (filter->cookie == cls_flower->cookie)
2738                         break;
2739
2740         if (!filter) {
2741                 err = -ENOENT;
2742                 goto out;
2743         }
2744
2745         err = igb_erase_filter(adapter, filter);
2746         if (err < 0)
2747                 goto out;
2748
2749         hlist_del(&filter->nfc_node);
2750         kfree(filter);
2751
2752 out:
2753         spin_unlock(&adapter->nfc_lock);
2754
2755         return err;
2756 }
2757
2758 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2759                                    struct flow_cls_offload *cls_flower)
2760 {
2761         switch (cls_flower->command) {
2762         case FLOW_CLS_REPLACE:
2763                 return igb_configure_clsflower(adapter, cls_flower);
2764         case FLOW_CLS_DESTROY:
2765                 return igb_delete_clsflower(adapter, cls_flower);
2766         case FLOW_CLS_STATS:
2767                 return -EOPNOTSUPP;
2768         default:
2769                 return -EOPNOTSUPP;
2770         }
2771 }
2772
2773 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2774                                  void *cb_priv)
2775 {
2776         struct igb_adapter *adapter = cb_priv;
2777
2778         if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2779                 return -EOPNOTSUPP;
2780
2781         switch (type) {
2782         case TC_SETUP_CLSFLOWER:
2783                 return igb_setup_tc_cls_flower(adapter, type_data);
2784
2785         default:
2786                 return -EOPNOTSUPP;
2787         }
2788 }
2789
2790 static int igb_offload_txtime(struct igb_adapter *adapter,
2791                               struct tc_etf_qopt_offload *qopt)
2792 {
2793         struct e1000_hw *hw = &adapter->hw;
2794         int err;
2795
2796         /* Launchtime offloading is only supported by i210 controller. */
2797         if (hw->mac.type != e1000_i210)
2798                 return -EOPNOTSUPP;
2799
2800         /* Launchtime offloading is only supported by queues 0 and 1. */
2801         if (qopt->queue < 0 || qopt->queue > 1)
2802                 return -EINVAL;
2803
2804         err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2805         if (err)
2806                 return err;
2807
2808         igb_offload_apply(adapter, qopt->queue);
2809
2810         return 0;
2811 }
2812
2813 static LIST_HEAD(igb_block_cb_list);
2814
2815 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2816                         void *type_data)
2817 {
2818         struct igb_adapter *adapter = netdev_priv(dev);
2819
2820         switch (type) {
2821         case TC_SETUP_QDISC_CBS:
2822                 return igb_offload_cbs(adapter, type_data);
2823         case TC_SETUP_BLOCK:
2824                 return flow_block_cb_setup_simple(type_data,
2825                                                   &igb_block_cb_list,
2826                                                   igb_setup_tc_block_cb,
2827                                                   adapter, adapter, true);
2828
2829         case TC_SETUP_QDISC_ETF:
2830                 return igb_offload_txtime(adapter, type_data);
2831
2832         default:
2833                 return -EOPNOTSUPP;
2834         }
2835 }
2836
2837 static const struct net_device_ops igb_netdev_ops = {
2838         .ndo_open               = igb_open,
2839         .ndo_stop               = igb_close,
2840         .ndo_start_xmit         = igb_xmit_frame,
2841         .ndo_get_stats64        = igb_get_stats64,
2842         .ndo_set_rx_mode        = igb_set_rx_mode,
2843         .ndo_set_mac_address    = igb_set_mac,
2844         .ndo_change_mtu         = igb_change_mtu,
2845         .ndo_do_ioctl           = igb_ioctl,
2846         .ndo_tx_timeout         = igb_tx_timeout,
2847         .ndo_validate_addr      = eth_validate_addr,
2848         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
2849         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
2850         .ndo_set_vf_mac         = igb_ndo_set_vf_mac,
2851         .ndo_set_vf_vlan        = igb_ndo_set_vf_vlan,
2852         .ndo_set_vf_rate        = igb_ndo_set_vf_bw,
2853         .ndo_set_vf_spoofchk    = igb_ndo_set_vf_spoofchk,
2854         .ndo_set_vf_trust       = igb_ndo_set_vf_trust,
2855         .ndo_get_vf_config      = igb_ndo_get_vf_config,
2856         .ndo_fix_features       = igb_fix_features,
2857         .ndo_set_features       = igb_set_features,
2858         .ndo_fdb_add            = igb_ndo_fdb_add,
2859         .ndo_features_check     = igb_features_check,
2860         .ndo_setup_tc           = igb_setup_tc,
2861 };
2862
2863 /**
2864  * igb_set_fw_version - Configure version string for ethtool
2865  * @adapter: adapter struct
2866  **/
2867 void igb_set_fw_version(struct igb_adapter *adapter)
2868 {
2869         struct e1000_hw *hw = &adapter->hw;
2870         struct e1000_fw_version fw;
2871
2872         igb_get_fw_version(hw, &fw);
2873
2874         switch (hw->mac.type) {
2875         case e1000_i210:
2876         case e1000_i211:
2877                 if (!(igb_get_flash_presence_i210(hw))) {
2878                         snprintf(adapter->fw_version,
2879                                  sizeof(adapter->fw_version),
2880                                  "%2d.%2d-%d",
2881                                  fw.invm_major, fw.invm_minor,
2882                                  fw.invm_img_type);
2883                         break;
2884                 }
2885                 /* fall through */
2886         default:
2887                 /* if option is rom valid, display its version too */
2888                 if (fw.or_valid) {
2889                         snprintf(adapter->fw_version,
2890                                  sizeof(adapter->fw_version),
2891                                  "%d.%d, 0x%08x, %d.%d.%d",
2892                                  fw.eep_major, fw.eep_minor, fw.etrack_id,
2893                                  fw.or_major, fw.or_build, fw.or_patch);
2894                 /* no option rom */
2895                 } else if (fw.etrack_id != 0X0000) {
2896                         snprintf(adapter->fw_version,
2897                             sizeof(adapter->fw_version),
2898                             "%d.%d, 0x%08x",
2899                             fw.eep_major, fw.eep_minor, fw.etrack_id);
2900                 } else {
2901                 snprintf(adapter->fw_version,
2902                     sizeof(adapter->fw_version),
2903                     "%d.%d.%d",
2904                     fw.eep_major, fw.eep_minor, fw.eep_build);
2905                 }
2906                 break;
2907         }
2908 }
2909
2910 /**
2911  * igb_init_mas - init Media Autosense feature if enabled in the NVM
2912  *
2913  * @adapter: adapter struct
2914  **/
2915 static void igb_init_mas(struct igb_adapter *adapter)
2916 {
2917         struct e1000_hw *hw = &adapter->hw;
2918         u16 eeprom_data;
2919
2920         hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2921         switch (hw->bus.func) {
2922         case E1000_FUNC_0:
2923                 if (eeprom_data & IGB_MAS_ENABLE_0) {
2924                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2925                         netdev_info(adapter->netdev,
2926                                 "MAS: Enabling Media Autosense for port %d\n",
2927                                 hw->bus.func);
2928                 }
2929                 break;
2930         case E1000_FUNC_1:
2931                 if (eeprom_data & IGB_MAS_ENABLE_1) {
2932                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2933                         netdev_info(adapter->netdev,
2934                                 "MAS: Enabling Media Autosense for port %d\n",
2935                                 hw->bus.func);
2936                 }
2937                 break;
2938         case E1000_FUNC_2:
2939                 if (eeprom_data & IGB_MAS_ENABLE_2) {
2940                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2941                         netdev_info(adapter->netdev,
2942                                 "MAS: Enabling Media Autosense for port %d\n",
2943                                 hw->bus.func);
2944                 }
2945                 break;
2946         case E1000_FUNC_3:
2947                 if (eeprom_data & IGB_MAS_ENABLE_3) {
2948                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2949                         netdev_info(adapter->netdev,
2950                                 "MAS: Enabling Media Autosense for port %d\n",
2951                                 hw->bus.func);
2952                 }
2953                 break;
2954         default:
2955                 /* Shouldn't get here */
2956                 netdev_err(adapter->netdev,
2957                         "MAS: Invalid port configuration, returning\n");
2958                 break;
2959         }
2960 }
2961
2962 /**
2963  *  igb_init_i2c - Init I2C interface
2964  *  @adapter: pointer to adapter structure
2965  **/
2966 static s32 igb_init_i2c(struct igb_adapter *adapter)
2967 {
2968         s32 status = 0;
2969
2970         /* I2C interface supported on i350 devices */
2971         if (adapter->hw.mac.type != e1000_i350)
2972                 return 0;
2973
2974         /* Initialize the i2c bus which is controlled by the registers.
2975          * This bus will use the i2c_algo_bit structue that implements
2976          * the protocol through toggling of the 4 bits in the register.
2977          */
2978         adapter->i2c_adap.owner = THIS_MODULE;
2979         adapter->i2c_algo = igb_i2c_algo;
2980         adapter->i2c_algo.data = adapter;
2981         adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2982         adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2983         strlcpy(adapter->i2c_adap.name, "igb BB",
2984                 sizeof(adapter->i2c_adap.name));
2985         status = i2c_bit_add_bus(&adapter->i2c_adap);
2986         return status;
2987 }
2988
2989 /**
2990  *  igb_probe - Device Initialization Routine
2991  *  @pdev: PCI device information struct
2992  *  @ent: entry in igb_pci_tbl
2993  *
2994  *  Returns 0 on success, negative on failure
2995  *
2996  *  igb_probe initializes an adapter identified by a pci_dev structure.
2997  *  The OS initialization, configuring of the adapter private structure,
2998  *  and a hardware reset occur.
2999  **/
3000 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3001 {
3002         struct net_device *netdev;
3003         struct igb_adapter *adapter;
3004         struct e1000_hw *hw;
3005         u16 eeprom_data = 0;
3006         s32 ret_val;
3007         static int global_quad_port_a; /* global quad port a indication */
3008         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3009         int err, pci_using_dac;
3010         u8 part_str[E1000_PBANUM_LENGTH];
3011
3012         /* Catch broken hardware that put the wrong VF device ID in
3013          * the PCIe SR-IOV capability.
3014          */
3015         if (pdev->is_virtfn) {
3016                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
3017                         pci_name(pdev), pdev->vendor, pdev->device);
3018                 return -EINVAL;
3019         }
3020
3021         err = pci_enable_device_mem(pdev);
3022         if (err)
3023                 return err;
3024
3025         pci_using_dac = 0;
3026         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3027         if (!err) {
3028                 pci_using_dac = 1;
3029         } else {
3030                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3031                 if (err) {
3032                         dev_err(&pdev->dev,
3033                                 "No usable DMA configuration, aborting\n");
3034                         goto err_dma;
3035                 }
3036         }
3037
3038         err = pci_request_mem_regions(pdev, igb_driver_name);
3039         if (err)
3040                 goto err_pci_reg;
3041
3042         pci_enable_pcie_error_reporting(pdev);
3043
3044         pci_set_master(pdev);
3045         pci_save_state(pdev);
3046
3047         err = -ENOMEM;
3048         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3049                                    IGB_MAX_TX_QUEUES);
3050         if (!netdev)
3051                 goto err_alloc_etherdev;
3052
3053         SET_NETDEV_DEV(netdev, &pdev->dev);
3054
3055         pci_set_drvdata(pdev, netdev);
3056         adapter = netdev_priv(netdev);
3057         adapter->netdev = netdev;
3058         adapter->pdev = pdev;
3059         hw = &adapter->hw;
3060         hw->back = adapter;
3061         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3062
3063         err = -EIO;
3064         adapter->io_addr = pci_iomap(pdev, 0, 0);
3065         if (!adapter->io_addr)
3066                 goto err_ioremap;
3067         /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3068         hw->hw_addr = adapter->io_addr;
3069
3070         netdev->netdev_ops = &igb_netdev_ops;
3071         igb_set_ethtool_ops(netdev);
3072         netdev->watchdog_timeo = 5 * HZ;
3073
3074         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3075
3076         netdev->mem_start = pci_resource_start(pdev, 0);
3077         netdev->mem_end = pci_resource_end(pdev, 0);
3078
3079         /* PCI config space info */
3080         hw->vendor_id = pdev->vendor;
3081         hw->device_id = pdev->device;
3082         hw->revision_id = pdev->revision;
3083         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3084         hw->subsystem_device_id = pdev->subsystem_device;
3085
3086         /* Copy the default MAC, PHY and NVM function pointers */
3087         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3088         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3089         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3090         /* Initialize skew-specific constants */
3091         err = ei->get_invariants(hw);
3092         if (err)
3093                 goto err_sw_init;
3094
3095         /* setup the private structure */
3096         err = igb_sw_init(adapter);
3097         if (err)
3098                 goto err_sw_init;
3099
3100         igb_get_bus_info_pcie(hw);
3101
3102         hw->phy.autoneg_wait_to_complete = false;
3103
3104         /* Copper options */
3105         if (hw->phy.media_type == e1000_media_type_copper) {
3106                 hw->phy.mdix = AUTO_ALL_MODES;
3107                 hw->phy.disable_polarity_correction = false;
3108                 hw->phy.ms_type = e1000_ms_hw_default;
3109         }
3110
3111         if (igb_check_reset_block(hw))
3112                 dev_info(&pdev->dev,
3113                         "PHY reset is blocked due to SOL/IDER session.\n");
3114
3115         /* features is initialized to 0 in allocation, it might have bits
3116          * set by igb_sw_init so we should use an or instead of an
3117          * assignment.
3118          */
3119         netdev->features |= NETIF_F_SG |
3120                             NETIF_F_TSO |
3121                             NETIF_F_TSO6 |
3122                             NETIF_F_RXHASH |
3123                             NETIF_F_RXCSUM |
3124                             NETIF_F_HW_CSUM;
3125
3126         if (hw->mac.type >= e1000_82576)
3127                 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
3128
3129         if (hw->mac.type >= e1000_i350)
3130                 netdev->features |= NETIF_F_HW_TC;
3131
3132 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3133                                   NETIF_F_GSO_GRE_CSUM | \
3134                                   NETIF_F_GSO_IPXIP4 | \
3135                                   NETIF_F_GSO_IPXIP6 | \
3136                                   NETIF_F_GSO_UDP_TUNNEL | \
3137                                   NETIF_F_GSO_UDP_TUNNEL_CSUM)
3138
3139         netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3140         netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3141
3142         /* copy netdev features into list of user selectable features */
3143         netdev->hw_features |= netdev->features |
3144                                NETIF_F_HW_VLAN_CTAG_RX |
3145                                NETIF_F_HW_VLAN_CTAG_TX |
3146                                NETIF_F_RXALL;
3147
3148         if (hw->mac.type >= e1000_i350)
3149                 netdev->hw_features |= NETIF_F_NTUPLE;
3150
3151         if (pci_using_dac)
3152                 netdev->features |= NETIF_F_HIGHDMA;
3153
3154         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3155         netdev->mpls_features |= NETIF_F_HW_CSUM;
3156         netdev->hw_enc_features |= netdev->vlan_features;
3157
3158         /* set this bit last since it cannot be part of vlan_features */
3159         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3160                             NETIF_F_HW_VLAN_CTAG_RX |
3161                             NETIF_F_HW_VLAN_CTAG_TX;
3162
3163         netdev->priv_flags |= IFF_SUPP_NOFCS;
3164
3165         netdev->priv_flags |= IFF_UNICAST_FLT;
3166
3167         /* MTU range: 68 - 9216 */
3168         netdev->min_mtu = ETH_MIN_MTU;
3169         netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3170
3171         adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3172
3173         /* before reading the NVM, reset the controller to put the device in a
3174          * known good starting state
3175          */
3176         hw->mac.ops.reset_hw(hw);
3177
3178         /* make sure the NVM is good , i211/i210 parts can have special NVM
3179          * that doesn't contain a checksum
3180          */
3181         switch (hw->mac.type) {
3182         case e1000_i210:
3183         case e1000_i211:
3184                 if (igb_get_flash_presence_i210(hw)) {
3185                         if (hw->nvm.ops.validate(hw) < 0) {
3186                                 dev_err(&pdev->dev,
3187                                         "The NVM Checksum Is Not Valid\n");
3188                                 err = -EIO;
3189                                 goto err_eeprom;
3190                         }
3191                 }
3192                 break;
3193         default:
3194                 if (hw->nvm.ops.validate(hw) < 0) {
3195                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3196                         err = -EIO;
3197                         goto err_eeprom;
3198                 }
3199                 break;
3200         }
3201
3202         if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3203                 /* copy the MAC address out of the NVM */
3204                 if (hw->mac.ops.read_mac_addr(hw))
3205                         dev_err(&pdev->dev, "NVM Read Error\n");
3206         }
3207
3208         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
3209
3210         if (!is_valid_ether_addr(netdev->dev_addr)) {
3211                 dev_err(&pdev->dev, "Invalid MAC Address\n");
3212                 err = -EIO;
3213                 goto err_eeprom;
3214         }
3215
3216         igb_set_default_mac_filter(adapter);
3217
3218         /* get firmware version for ethtool -i */
3219         igb_set_fw_version(adapter);
3220
3221         /* configure RXPBSIZE and TXPBSIZE */
3222         if (hw->mac.type == e1000_i210) {
3223                 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3224                 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3225         }
3226
3227         timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3228         timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3229
3230         INIT_WORK(&adapter->reset_task, igb_reset_task);
3231         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3232
3233         /* Initialize link properties that are user-changeable */
3234         adapter->fc_autoneg = true;
3235         hw->mac.autoneg = true;
3236         hw->phy.autoneg_advertised = 0x2f;
3237
3238         hw->fc.requested_mode = e1000_fc_default;
3239         hw->fc.current_mode = e1000_fc_default;
3240
3241         igb_validate_mdi_setting(hw);
3242
3243         /* By default, support wake on port A */
3244         if (hw->bus.func == 0)
3245                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3246
3247         /* Check the NVM for wake support on non-port A ports */
3248         if (hw->mac.type >= e1000_82580)
3249                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3250                                  NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3251                                  &eeprom_data);
3252         else if (hw->bus.func == 1)
3253                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3254
3255         if (eeprom_data & IGB_EEPROM_APME)
3256                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3257
3258         /* now that we have the eeprom settings, apply the special cases where
3259          * the eeprom may be wrong or the board simply won't support wake on
3260          * lan on a particular port
3261          */
3262         switch (pdev->device) {
3263         case E1000_DEV_ID_82575GB_QUAD_COPPER:
3264                 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3265                 break;
3266         case E1000_DEV_ID_82575EB_FIBER_SERDES:
3267         case E1000_DEV_ID_82576_FIBER:
3268         case E1000_DEV_ID_82576_SERDES:
3269                 /* Wake events only supported on port A for dual fiber
3270                  * regardless of eeprom setting
3271                  */
3272                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3273                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3274                 break;
3275         case E1000_DEV_ID_82576_QUAD_COPPER:
3276         case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3277                 /* if quad port adapter, disable WoL on all but port A */
3278                 if (global_quad_port_a != 0)
3279                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3280                 else
3281                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3282                 /* Reset for multiple quad port adapters */
3283                 if (++global_quad_port_a == 4)
3284                         global_quad_port_a = 0;
3285                 break;
3286         default:
3287                 /* If the device can't wake, don't set software support */
3288                 if (!device_can_wakeup(&adapter->pdev->dev))
3289                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3290         }
3291
3292         /* initialize the wol settings based on the eeprom settings */
3293         if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3294                 adapter->wol |= E1000_WUFC_MAG;
3295
3296         /* Some vendors want WoL disabled by default, but still supported */
3297         if ((hw->mac.type == e1000_i350) &&
3298             (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3299                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3300                 adapter->wol = 0;
3301         }
3302
3303         /* Some vendors want the ability to Use the EEPROM setting as
3304          * enable/disable only, and not for capability
3305          */
3306         if (((hw->mac.type == e1000_i350) ||
3307              (hw->mac.type == e1000_i354)) &&
3308             (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3309                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3310                 adapter->wol = 0;
3311         }
3312         if (hw->mac.type == e1000_i350) {
3313                 if (((pdev->subsystem_device == 0x5001) ||
3314                      (pdev->subsystem_device == 0x5002)) &&
3315                                 (hw->bus.func == 0)) {
3316                         adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3317                         adapter->wol = 0;
3318                 }
3319                 if (pdev->subsystem_device == 0x1F52)
3320                         adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3321         }
3322
3323         device_set_wakeup_enable(&adapter->pdev->dev,
3324                                  adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3325
3326         /* reset the hardware with the new settings */
3327         igb_reset(adapter);
3328
3329         /* Init the I2C interface */
3330         err = igb_init_i2c(adapter);
3331         if (err) {
3332                 dev_err(&pdev->dev, "failed to init i2c interface\n");
3333                 goto err_eeprom;
3334         }
3335
3336         /* let the f/w know that the h/w is now under the control of the
3337          * driver.
3338          */
3339         igb_get_hw_control(adapter);
3340
3341         strcpy(netdev->name, "eth%d");
3342         err = register_netdev(netdev);
3343         if (err)
3344                 goto err_register;
3345
3346         /* carrier off reporting is important to ethtool even BEFORE open */
3347         netif_carrier_off(netdev);
3348
3349 #ifdef CONFIG_IGB_DCA
3350         if (dca_add_requester(&pdev->dev) == 0) {
3351                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3352                 dev_info(&pdev->dev, "DCA enabled\n");
3353                 igb_setup_dca(adapter);
3354         }
3355
3356 #endif
3357 #ifdef CONFIG_IGB_HWMON
3358         /* Initialize the thermal sensor on i350 devices. */
3359         if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3360                 u16 ets_word;
3361
3362                 /* Read the NVM to determine if this i350 device supports an
3363                  * external thermal sensor.
3364                  */
3365                 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3366                 if (ets_word != 0x0000 && ets_word != 0xFFFF)
3367                         adapter->ets = true;
3368                 else
3369                         adapter->ets = false;
3370                 if (igb_sysfs_init(adapter))
3371                         dev_err(&pdev->dev,
3372                                 "failed to allocate sysfs resources\n");
3373         } else {
3374                 adapter->ets = false;
3375         }
3376 #endif
3377         /* Check if Media Autosense is enabled */
3378         adapter->ei = *ei;
3379         if (hw->dev_spec._82575.mas_capable)
3380                 igb_init_mas(adapter);
3381
3382         /* do hw tstamp init after resetting */
3383         igb_ptp_init(adapter);
3384
3385         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3386         /* print bus type/speed/width info, not applicable to i354 */
3387         if (hw->mac.type != e1000_i354) {
3388                 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3389                          netdev->name,
3390                          ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3391                           (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3392                            "unknown"),
3393                          ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3394                           "Width x4" :
3395                           (hw->bus.width == e1000_bus_width_pcie_x2) ?
3396                           "Width x2" :
3397                           (hw->bus.width == e1000_bus_width_pcie_x1) ?
3398                           "Width x1" : "unknown"), netdev->dev_addr);
3399         }
3400
3401         if ((hw->mac.type >= e1000_i210 ||
3402              igb_get_flash_presence_i210(hw))) {
3403                 ret_val = igb_read_part_string(hw, part_str,
3404                                                E1000_PBANUM_LENGTH);
3405         } else {
3406                 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3407         }
3408
3409         if (ret_val)
3410                 strcpy(part_str, "Unknown");
3411         dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3412         dev_info(&pdev->dev,
3413                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3414                 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3415                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3416                 adapter->num_rx_queues, adapter->num_tx_queues);
3417         if (hw->phy.media_type == e1000_media_type_copper) {
3418                 switch (hw->mac.type) {
3419                 case e1000_i350:
3420                 case e1000_i210:
3421                 case e1000_i211:
3422                         /* Enable EEE for internal copper PHY devices */
3423                         err = igb_set_eee_i350(hw, true, true);
3424                         if ((!err) &&
3425                             (!hw->dev_spec._82575.eee_disable)) {
3426                                 adapter->eee_advert =
3427                                         MDIO_EEE_100TX | MDIO_EEE_1000T;
3428                                 adapter->flags |= IGB_FLAG_EEE;
3429                         }
3430                         break;
3431                 case e1000_i354:
3432                         if ((rd32(E1000_CTRL_EXT) &
3433                             E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3434                                 err = igb_set_eee_i354(hw, true, true);
3435                                 if ((!err) &&
3436                                         (!hw->dev_spec._82575.eee_disable)) {
3437                                         adapter->eee_advert =
3438                                            MDIO_EEE_100TX | MDIO_EEE_1000T;
3439                                         adapter->flags |= IGB_FLAG_EEE;
3440                                 }
3441                         }
3442                         break;
3443                 default:
3444                         break;
3445                 }
3446         }
3447
3448         dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NEVER_SKIP);
3449
3450         pm_runtime_put_noidle(&pdev->dev);
3451         return 0;
3452
3453 err_register:
3454         igb_release_hw_control(adapter);
3455         memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3456 err_eeprom:
3457         if (!igb_check_reset_block(hw))
3458                 igb_reset_phy(hw);
3459
3460         if (hw->flash_address)
3461                 iounmap(hw->flash_address);
3462 err_sw_init:
3463         kfree(adapter->mac_table);
3464         kfree(adapter->shadow_vfta);
3465         igb_clear_interrupt_scheme(adapter);
3466 #ifdef CONFIG_PCI_IOV
3467         igb_disable_sriov(pdev);
3468 #endif
3469         pci_iounmap(pdev, adapter->io_addr);
3470 err_ioremap:
3471         free_netdev(netdev);
3472 err_alloc_etherdev:
3473         pci_release_mem_regions(pdev);
3474 err_pci_reg:
3475 err_dma:
3476         pci_disable_device(pdev);
3477         return err;
3478 }
3479
3480 #ifdef CONFIG_PCI_IOV
3481 static int igb_disable_sriov(struct pci_dev *pdev)
3482 {
3483         struct net_device *netdev = pci_get_drvdata(pdev);
3484         struct igb_adapter *adapter = netdev_priv(netdev);
3485         struct e1000_hw *hw = &adapter->hw;
3486
3487         /* reclaim resources allocated to VFs */
3488         if (adapter->vf_data) {
3489                 /* disable iov and allow time for transactions to clear */
3490                 if (pci_vfs_assigned(pdev)) {
3491                         dev_warn(&pdev->dev,
3492                                  "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3493                         return -EPERM;
3494                 } else {
3495                         pci_disable_sriov(pdev);
3496                         msleep(500);
3497                 }
3498
3499                 kfree(adapter->vf_mac_list);
3500                 adapter->vf_mac_list = NULL;
3501                 kfree(adapter->vf_data);
3502                 adapter->vf_data = NULL;
3503                 adapter->vfs_allocated_count = 0;
3504                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3505                 wrfl();
3506                 msleep(100);
3507                 dev_info(&pdev->dev, "IOV Disabled\n");
3508
3509                 /* Re-enable DMA Coalescing flag since IOV is turned off */
3510                 adapter->flags |= IGB_FLAG_DMAC;
3511         }
3512
3513         return 0;
3514 }
3515
3516 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
3517 {
3518         struct net_device *netdev = pci_get_drvdata(pdev);
3519         struct igb_adapter *adapter = netdev_priv(netdev);
3520         int old_vfs = pci_num_vf(pdev);
3521         struct vf_mac_filter *mac_list;
3522         int err = 0;
3523         int num_vf_mac_filters, i;
3524
3525         if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3526                 err = -EPERM;
3527                 goto out;
3528         }
3529         if (!num_vfs)
3530                 goto out;
3531
3532         if (old_vfs) {
3533                 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3534                          old_vfs, max_vfs);
3535                 adapter->vfs_allocated_count = old_vfs;
3536         } else
3537                 adapter->vfs_allocated_count = num_vfs;
3538
3539         adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3540                                 sizeof(struct vf_data_storage), GFP_KERNEL);
3541
3542         /* if allocation failed then we do not support SR-IOV */
3543         if (!adapter->vf_data) {
3544                 adapter->vfs_allocated_count = 0;
3545                 err = -ENOMEM;
3546                 goto out;
3547         }
3548
3549         /* Due to the limited number of RAR entries calculate potential
3550          * number of MAC filters available for the VFs. Reserve entries
3551          * for PF default MAC, PF MAC filters and at least one RAR entry
3552          * for each VF for VF MAC.
3553          */
3554         num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3555                              (1 + IGB_PF_MAC_FILTERS_RESERVED +
3556                               adapter->vfs_allocated_count);
3557
3558         adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3559                                        sizeof(struct vf_mac_filter),
3560                                        GFP_KERNEL);
3561
3562         mac_list = adapter->vf_mac_list;
3563         INIT_LIST_HEAD(&adapter->vf_macs.l);
3564
3565         if (adapter->vf_mac_list) {
3566                 /* Initialize list of VF MAC filters */
3567                 for (i = 0; i < num_vf_mac_filters; i++) {
3568                         mac_list->vf = -1;
3569                         mac_list->free = true;
3570                         list_add(&mac_list->l, &adapter->vf_macs.l);
3571                         mac_list++;
3572                 }
3573         } else {
3574                 /* If we could not allocate memory for the VF MAC filters
3575                  * we can continue without this feature but warn user.
3576                  */
3577                 dev_err(&pdev->dev,
3578                         "Unable to allocate memory for VF MAC filter list\n");
3579         }
3580
3581         /* only call pci_enable_sriov() if no VFs are allocated already */
3582         if (!old_vfs) {
3583                 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3584                 if (err)
3585                         goto err_out;
3586         }
3587         dev_info(&pdev->dev, "%d VFs allocated\n",
3588                  adapter->vfs_allocated_count);
3589         for (i = 0; i < adapter->vfs_allocated_count; i++)
3590                 igb_vf_configure(adapter, i);
3591
3592         /* DMA Coalescing is not supported in IOV mode. */
3593         adapter->flags &= ~IGB_FLAG_DMAC;
3594         goto out;
3595
3596 err_out:
3597         kfree(adapter->vf_mac_list);
3598         adapter->vf_mac_list = NULL;
3599         kfree(adapter->vf_data);
3600         adapter->vf_data = NULL;
3601         adapter->vfs_allocated_count = 0;
3602 out:
3603         return err;
3604 }
3605
3606 #endif
3607 /**
3608  *  igb_remove_i2c - Cleanup  I2C interface
3609  *  @adapter: pointer to adapter structure
3610  **/
3611 static void igb_remove_i2c(struct igb_adapter *adapter)
3612 {
3613         /* free the adapter bus structure */
3614         i2c_del_adapter(&adapter->i2c_adap);
3615 }
3616
3617 /**
3618  *  igb_remove - Device Removal Routine
3619  *  @pdev: PCI device information struct
3620  *
3621  *  igb_remove is called by the PCI subsystem to alert the driver
3622  *  that it should release a PCI device.  The could be caused by a
3623  *  Hot-Plug event, or because the driver is going to be removed from
3624  *  memory.
3625  **/
3626 static void igb_remove(struct pci_dev *pdev)
3627 {
3628         struct net_device *netdev = pci_get_drvdata(pdev);
3629         struct igb_adapter *adapter = netdev_priv(netdev);
3630         struct e1000_hw *hw = &adapter->hw;
3631
3632         pm_runtime_get_noresume(&pdev->dev);
3633 #ifdef CONFIG_IGB_HWMON
3634         igb_sysfs_exit(adapter);
3635 #endif
3636         igb_remove_i2c(adapter);
3637         igb_ptp_stop(adapter);
3638         /* The watchdog timer may be rescheduled, so explicitly
3639          * disable watchdog from being rescheduled.
3640          */
3641         set_bit(__IGB_DOWN, &adapter->state);
3642         del_timer_sync(&adapter->watchdog_timer);
3643         del_timer_sync(&adapter->phy_info_timer);
3644
3645         cancel_work_sync(&adapter->reset_task);
3646         cancel_work_sync(&adapter->watchdog_task);
3647
3648 #ifdef CONFIG_IGB_DCA
3649         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3650                 dev_info(&pdev->dev, "DCA disabled\n");
3651                 dca_remove_requester(&pdev->dev);
3652                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3653                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3654         }
3655 #endif
3656
3657         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3658          * would have already happened in close and is redundant.
3659          */
3660         igb_release_hw_control(adapter);
3661
3662 #ifdef CONFIG_PCI_IOV
3663         igb_disable_sriov(pdev);
3664 #endif
3665
3666         unregister_netdev(netdev);
3667
3668         igb_clear_interrupt_scheme(adapter);
3669
3670         pci_iounmap(pdev, adapter->io_addr);
3671         if (hw->flash_address)
3672                 iounmap(hw->flash_address);
3673         pci_release_mem_regions(pdev);
3674
3675         kfree(adapter->mac_table);
3676         kfree(adapter->shadow_vfta);
3677         free_netdev(netdev);
3678
3679         pci_disable_pcie_error_reporting(pdev);
3680
3681         pci_disable_device(pdev);
3682 }
3683
3684 /**
3685  *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3686  *  @adapter: board private structure to initialize
3687  *
3688  *  This function initializes the vf specific data storage and then attempts to
3689  *  allocate the VFs.  The reason for ordering it this way is because it is much
3690  *  mor expensive time wise to disable SR-IOV than it is to allocate and free
3691  *  the memory for the VFs.
3692  **/
3693 static void igb_probe_vfs(struct igb_adapter *adapter)
3694 {
3695 #ifdef CONFIG_PCI_IOV
3696         struct pci_dev *pdev = adapter->pdev;
3697         struct e1000_hw *hw = &adapter->hw;
3698
3699         /* Virtualization features not supported on i210 family. */
3700         if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3701                 return;
3702
3703         /* Of the below we really only want the effect of getting
3704          * IGB_FLAG_HAS_MSIX set (if available), without which
3705          * igb_enable_sriov() has no effect.
3706          */
3707         igb_set_interrupt_capability(adapter, true);
3708         igb_reset_interrupt_capability(adapter);
3709
3710         pci_sriov_set_totalvfs(pdev, 7);
3711         igb_enable_sriov(pdev, max_vfs);
3712
3713 #endif /* CONFIG_PCI_IOV */
3714 }
3715
3716 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3717 {
3718         struct e1000_hw *hw = &adapter->hw;
3719         unsigned int max_rss_queues;
3720
3721         /* Determine the maximum number of RSS queues supported. */
3722         switch (hw->mac.type) {
3723         case e1000_i211:
3724                 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3725                 break;
3726         case e1000_82575:
3727         case e1000_i210:
3728                 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3729                 break;
3730         case e1000_i350:
3731                 /* I350 cannot do RSS and SR-IOV at the same time */
3732                 if (!!adapter->vfs_allocated_count) {
3733                         max_rss_queues = 1;
3734                         break;
3735                 }
3736                 /* fall through */
3737         case e1000_82576:
3738                 if (!!adapter->vfs_allocated_count) {
3739                         max_rss_queues = 2;
3740                         break;
3741                 }
3742                 /* fall through */
3743         case e1000_82580:
3744         case e1000_i354:
3745         default:
3746                 max_rss_queues = IGB_MAX_RX_QUEUES;
3747                 break;
3748         }
3749
3750         return max_rss_queues;
3751 }
3752
3753 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3754 {
3755         u32 max_rss_queues;
3756
3757         max_rss_queues = igb_get_max_rss_queues(adapter);
3758         adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3759
3760         igb_set_flag_queue_pairs(adapter, max_rss_queues);
3761 }
3762
3763 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3764                               const u32 max_rss_queues)
3765 {
3766         struct e1000_hw *hw = &adapter->hw;
3767
3768         /* Determine if we need to pair queues. */
3769         switch (hw->mac.type) {
3770         case e1000_82575:
3771         case e1000_i211:
3772                 /* Device supports enough interrupts without queue pairing. */
3773                 break;
3774         case e1000_82576:
3775         case e1000_82580:
3776         case e1000_i350:
3777         case e1000_i354:
3778         case e1000_i210:
3779         default:
3780                 /* If rss_queues > half of max_rss_queues, pair the queues in
3781                  * order to conserve interrupts due to limited supply.
3782                  */
3783                 if (adapter->rss_queues > (max_rss_queues / 2))
3784                         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3785                 else
3786                         adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3787                 break;
3788         }
3789 }
3790
3791 /**
3792  *  igb_sw_init - Initialize general software structures (struct igb_adapter)
3793  *  @adapter: board private structure to initialize
3794  *
3795  *  igb_sw_init initializes the Adapter private data structure.
3796  *  Fields are initialized based on PCI device information and
3797  *  OS network device settings (MTU size).
3798  **/
3799 static int igb_sw_init(struct igb_adapter *adapter)
3800 {
3801         struct e1000_hw *hw = &adapter->hw;
3802         struct net_device *netdev = adapter->netdev;
3803         struct pci_dev *pdev = adapter->pdev;
3804
3805         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3806
3807         /* set default ring sizes */
3808         adapter->tx_ring_count = IGB_DEFAULT_TXD;
3809         adapter->rx_ring_count = IGB_DEFAULT_RXD;
3810
3811         /* set default ITR values */
3812         adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3813         adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3814
3815         /* set default work limits */
3816         adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3817
3818         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3819                                   VLAN_HLEN;
3820         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3821
3822         spin_lock_init(&adapter->nfc_lock);
3823         spin_lock_init(&adapter->stats64_lock);
3824 #ifdef CONFIG_PCI_IOV
3825         switch (hw->mac.type) {
3826         case e1000_82576:
3827         case e1000_i350:
3828                 if (max_vfs > 7) {
3829                         dev_warn(&pdev->dev,
3830                                  "Maximum of 7 VFs per PF, using max\n");
3831                         max_vfs = adapter->vfs_allocated_count = 7;
3832                 } else
3833                         adapter->vfs_allocated_count = max_vfs;
3834                 if (adapter->vfs_allocated_count)
3835                         dev_warn(&pdev->dev,
3836                                  "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3837                 break;
3838         default:
3839                 break;
3840         }
3841 #endif /* CONFIG_PCI_IOV */
3842
3843         /* Assume MSI-X interrupts, will be checked during IRQ allocation */
3844         adapter->flags |= IGB_FLAG_HAS_MSIX;
3845
3846         adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
3847                                      sizeof(struct igb_mac_addr),
3848                                      GFP_KERNEL);
3849         if (!adapter->mac_table)
3850                 return -ENOMEM;
3851
3852         igb_probe_vfs(adapter);
3853
3854         igb_init_queue_configuration(adapter);
3855
3856         /* Setup and initialize a copy of the hw vlan table array */
3857         adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
3858                                        GFP_KERNEL);
3859         if (!adapter->shadow_vfta)
3860                 return -ENOMEM;
3861
3862         /* This call may decrease the number of queues */
3863         if (igb_init_interrupt_scheme(adapter, true)) {
3864                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3865                 return -ENOMEM;
3866         }
3867
3868         /* Explicitly disable IRQ since the NIC can be in any state. */
3869         igb_irq_disable(adapter);
3870
3871         if (hw->mac.type >= e1000_i350)
3872                 adapter->flags &= ~IGB_FLAG_DMAC;
3873
3874         set_bit(__IGB_DOWN, &adapter->state);
3875         return 0;
3876 }
3877
3878 /**
3879  *  igb_open - Called when a network interface is made active
3880  *  @netdev: network interface device structure
3881  *
3882  *  Returns 0 on success, negative value on failure
3883  *
3884  *  The open entry point is called when a network interface is made
3885  *  active by the system (IFF_UP).  At this point all resources needed
3886  *  for transmit and receive operations are allocated, the interrupt
3887  *  handler is registered with the OS, the watchdog timer is started,
3888  *  and the stack is notified that the interface is ready.
3889  **/
3890 static int __igb_open(struct net_device *netdev, bool resuming)
3891 {
3892         struct igb_adapter *adapter = netdev_priv(netdev);
3893         struct e1000_hw *hw = &adapter->hw;
3894         struct pci_dev *pdev = adapter->pdev;
3895         int err;
3896         int i;
3897
3898         /* disallow open during test */
3899         if (test_bit(__IGB_TESTING, &adapter->state)) {
3900                 WARN_ON(resuming);
3901                 return -EBUSY;
3902         }
3903
3904         if (!resuming)
3905                 pm_runtime_get_sync(&pdev->dev);
3906
3907         netif_carrier_off(netdev);
3908
3909         /* allocate transmit descriptors */
3910         err = igb_setup_all_tx_resources(adapter);
3911         if (err)
3912                 goto err_setup_tx;
3913
3914         /* allocate receive descriptors */
3915         err = igb_setup_all_rx_resources(adapter);
3916         if (err)
3917                 goto err_setup_rx;
3918
3919         igb_power_up_link(adapter);
3920
3921         /* before we allocate an interrupt, we must be ready to handle it.
3922          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3923          * as soon as we call pci_request_irq, so we have to setup our
3924          * clean_rx handler before we do so.
3925          */
3926         igb_configure(adapter);
3927
3928         err = igb_request_irq(adapter);
3929         if (err)
3930                 goto err_req_irq;
3931
3932         /* Notify the stack of the actual queue counts. */
3933         err = netif_set_real_num_tx_queues(adapter->netdev,
3934                                            adapter->num_tx_queues);
3935         if (err)
3936                 goto err_set_queues;
3937
3938         err = netif_set_real_num_rx_queues(adapter->netdev,
3939                                            adapter->num_rx_queues);
3940         if (err)
3941                 goto err_set_queues;
3942
3943         /* From here on the code is the same as igb_up() */
3944         clear_bit(__IGB_DOWN, &adapter->state);
3945
3946         for (i = 0; i < adapter->num_q_vectors; i++)
3947                 napi_enable(&(adapter->q_vector[i]->napi));
3948
3949         /* Clear any pending interrupts. */
3950         rd32(E1000_TSICR);
3951         rd32(E1000_ICR);
3952
3953         igb_irq_enable(adapter);
3954
3955         /* notify VFs that reset has been completed */
3956         if (adapter->vfs_allocated_count) {
3957                 u32 reg_data = rd32(E1000_CTRL_EXT);
3958
3959                 reg_data |= E1000_CTRL_EXT_PFRSTD;
3960                 wr32(E1000_CTRL_EXT, reg_data);
3961         }
3962
3963         netif_tx_start_all_queues(netdev);
3964
3965         if (!resuming)
3966                 pm_runtime_put(&pdev->dev);
3967
3968         /* start the watchdog. */
3969         hw->mac.get_link_status = 1;
3970         schedule_work(&adapter->watchdog_task);
3971
3972         return 0;
3973
3974 err_set_queues:
3975         igb_free_irq(adapter);
3976 err_req_irq:
3977         igb_release_hw_control(adapter);
3978         igb_power_down_link(adapter);
3979         igb_free_all_rx_resources(adapter);
3980 err_setup_rx:
3981         igb_free_all_tx_resources(adapter);
3982 err_setup_tx:
3983         igb_reset(adapter);
3984         if (!resuming)
3985                 pm_runtime_put(&pdev->dev);
3986
3987         return err;
3988 }
3989
3990 int igb_open(struct net_device *netdev)
3991 {
3992         return __igb_open(netdev, false);
3993 }
3994
3995 /**
3996  *  igb_close - Disables a network interface
3997  *  @netdev: network interface device structure
3998  *
3999  *  Returns 0, this is not allowed to fail
4000  *
4001  *  The close entry point is called when an interface is de-activated
4002  *  by the OS.  The hardware is still under the driver's control, but
4003  *  needs to be disabled.  A global MAC reset is issued to stop the
4004  *  hardware, and all transmit and receive resources are freed.
4005  **/
4006 static int __igb_close(struct net_device *netdev, bool suspending)
4007 {
4008         struct igb_adapter *adapter = netdev_priv(netdev);
4009         struct pci_dev *pdev = adapter->pdev;
4010
4011         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4012
4013         if (!suspending)
4014                 pm_runtime_get_sync(&pdev->dev);
4015
4016         igb_down(adapter);
4017         igb_free_irq(adapter);
4018
4019         igb_free_all_tx_resources(adapter);
4020         igb_free_all_rx_resources(adapter);
4021
4022         if (!suspending)
4023                 pm_runtime_put_sync(&pdev->dev);
4024         return 0;
4025 }
4026
4027 int igb_close(struct net_device *netdev)
4028 {
4029         if (netif_device_present(netdev) || netdev->dismantle)
4030                 return __igb_close(netdev, false);
4031         return 0;
4032 }
4033
4034 /**
4035  *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
4036  *  @tx_ring: tx descriptor ring (for a specific queue) to setup
4037  *
4038  *  Return 0 on success, negative on failure
4039  **/
4040 int igb_setup_tx_resources(struct igb_ring *tx_ring)
4041 {
4042         struct device *dev = tx_ring->dev;
4043         int size;
4044
4045         size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4046
4047         tx_ring->tx_buffer_info = vmalloc(size);
4048         if (!tx_ring->tx_buffer_info)
4049                 goto err;
4050
4051         /* round up to nearest 4K */
4052         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4053         tx_ring->size = ALIGN(tx_ring->size, 4096);
4054
4055         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4056                                            &tx_ring->dma, GFP_KERNEL);
4057         if (!tx_ring->desc)
4058                 goto err;
4059
4060         tx_ring->next_to_use = 0;
4061         tx_ring->next_to_clean = 0;
4062
4063         return 0;
4064
4065 err:
4066         vfree(tx_ring->tx_buffer_info);
4067         tx_ring->tx_buffer_info = NULL;
4068         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4069         return -ENOMEM;
4070 }
4071
4072 /**
4073  *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
4074  *                               (Descriptors) for all queues
4075  *  @adapter: board private structure
4076  *
4077  *  Return 0 on success, negative on failure
4078  **/
4079 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4080 {
4081         struct pci_dev *pdev = adapter->pdev;
4082         int i, err = 0;
4083
4084         for (i = 0; i < adapter->num_tx_queues; i++) {
4085                 err = igb_setup_tx_resources(adapter->tx_ring[i]);
4086                 if (err) {
4087                         dev_err(&pdev->dev,
4088                                 "Allocation for Tx Queue %u failed\n", i);
4089                         for (i--; i >= 0; i--)
4090                                 igb_free_tx_resources(adapter->tx_ring[i]);
4091                         break;
4092                 }
4093         }
4094
4095         return err;
4096 }
4097
4098 /**
4099  *  igb_setup_tctl - configure the transmit control registers
4100  *  @adapter: Board private structure
4101  **/
4102 void igb_setup_tctl(struct igb_adapter *adapter)
4103 {
4104         struct e1000_hw *hw = &adapter->hw;
4105         u32 tctl;
4106
4107         /* disable queue 0 which is enabled by default on 82575 and 82576 */
4108         wr32(E1000_TXDCTL(0), 0);
4109
4110         /* Program the Transmit Control Register */
4111         tctl = rd32(E1000_TCTL);
4112         tctl &= ~E1000_TCTL_CT;
4113         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4114                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4115
4116         igb_config_collision_dist(hw);
4117
4118         /* Enable transmits */
4119         tctl |= E1000_TCTL_EN;
4120
4121         wr32(E1000_TCTL, tctl);
4122 }
4123
4124 /**
4125  *  igb_configure_tx_ring - Configure transmit ring after Reset
4126  *  @adapter: board private structure
4127  *  @ring: tx ring to configure
4128  *
4129  *  Configure a transmit ring after a reset.
4130  **/
4131 void igb_configure_tx_ring(struct igb_adapter *adapter,
4132                            struct igb_ring *ring)
4133 {
4134         struct e1000_hw *hw = &adapter->hw;
4135         u32 txdctl = 0;
4136         u64 tdba = ring->dma;
4137         int reg_idx = ring->reg_idx;
4138
4139         wr32(E1000_TDLEN(reg_idx),
4140              ring->count * sizeof(union e1000_adv_tx_desc));
4141         wr32(E1000_TDBAL(reg_idx),
4142              tdba & 0x00000000ffffffffULL);
4143         wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4144
4145         ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4146         wr32(E1000_TDH(reg_idx), 0);
4147         writel(0, ring->tail);
4148
4149         txdctl |= IGB_TX_PTHRESH;
4150         txdctl |= IGB_TX_HTHRESH << 8;
4151         txdctl |= IGB_TX_WTHRESH << 16;
4152
4153         /* reinitialize tx_buffer_info */
4154         memset(ring->tx_buffer_info, 0,
4155                sizeof(struct igb_tx_buffer) * ring->count);
4156
4157         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4158         wr32(E1000_TXDCTL(reg_idx), txdctl);
4159 }
4160
4161 /**
4162  *  igb_configure_tx - Configure transmit Unit after Reset
4163  *  @adapter: board private structure
4164  *
4165  *  Configure the Tx unit of the MAC after a reset.
4166  **/
4167 static void igb_configure_tx(struct igb_adapter *adapter)
4168 {
4169         struct e1000_hw *hw = &adapter->hw;
4170         int i;
4171
4172         /* disable the queues */
4173         for (i = 0; i < adapter->num_tx_queues; i++)
4174                 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4175
4176         wrfl();
4177         usleep_range(10000, 20000);
4178
4179         for (i = 0; i < adapter->num_tx_queues; i++)
4180                 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4181 }
4182
4183 /**
4184  *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
4185  *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
4186  *
4187  *  Returns 0 on success, negative on failure
4188  **/
4189 int igb_setup_rx_resources(struct igb_ring *rx_ring)
4190 {
4191         struct device *dev = rx_ring->dev;
4192         int size;
4193
4194         size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4195
4196         rx_ring->rx_buffer_info = vmalloc(size);
4197         if (!rx_ring->rx_buffer_info)
4198                 goto err;
4199
4200         /* Round up to nearest 4K */
4201         rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4202         rx_ring->size = ALIGN(rx_ring->size, 4096);
4203
4204         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4205                                            &rx_ring->dma, GFP_KERNEL);
4206         if (!rx_ring->desc)
4207                 goto err;
4208
4209         rx_ring->next_to_alloc = 0;
4210         rx_ring->next_to_clean = 0;
4211         rx_ring->next_to_use = 0;
4212
4213         return 0;
4214
4215 err:
4216         vfree(rx_ring->rx_buffer_info);
4217         rx_ring->rx_buffer_info = NULL;
4218         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4219         return -ENOMEM;
4220 }
4221
4222 /**
4223  *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
4224  *                               (Descriptors) for all queues
4225  *  @adapter: board private structure
4226  *
4227  *  Return 0 on success, negative on failure
4228  **/
4229 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4230 {
4231         struct pci_dev *pdev = adapter->pdev;
4232         int i, err = 0;
4233
4234         for (i = 0; i < adapter->num_rx_queues; i++) {
4235                 err = igb_setup_rx_resources(adapter->rx_ring[i]);
4236                 if (err) {
4237                         dev_err(&pdev->dev,
4238                                 "Allocation for Rx Queue %u failed\n", i);
4239                         for (i--; i >= 0; i--)
4240                                 igb_free_rx_resources(adapter->rx_ring[i]);
4241                         break;
4242                 }
4243         }
4244
4245         return err;
4246 }
4247
4248 /**
4249  *  igb_setup_mrqc - configure the multiple receive queue control registers
4250  *  @adapter: Board private structure
4251  **/
4252 static void igb_setup_mrqc(struct igb_adapter *adapter)
4253 {
4254         struct e1000_hw *hw = &adapter->hw;
4255         u32 mrqc, rxcsum;
4256         u32 j, num_rx_queues;
4257         u32 rss_key[10];
4258
4259         netdev_rss_key_fill(rss_key, sizeof(rss_key));
4260         for (j = 0; j < 10; j++)
4261                 wr32(E1000_RSSRK(j), rss_key[j]);
4262
4263         num_rx_queues = adapter->rss_queues;
4264
4265         switch (hw->mac.type) {
4266         case e1000_82576:
4267                 /* 82576 supports 2 RSS queues for SR-IOV */
4268                 if (adapter->vfs_allocated_count)
4269                         num_rx_queues = 2;
4270                 break;
4271         default:
4272                 break;
4273         }
4274
4275         if (adapter->rss_indir_tbl_init != num_rx_queues) {
4276                 for (j = 0; j < IGB_RETA_SIZE; j++)
4277                         adapter->rss_indir_tbl[j] =
4278                         (j * num_rx_queues) / IGB_RETA_SIZE;
4279                 adapter->rss_indir_tbl_init = num_rx_queues;
4280         }
4281         igb_write_rss_indir_tbl(adapter);
4282
4283         /* Disable raw packet checksumming so that RSS hash is placed in
4284          * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
4285          * offloads as they are enabled by default
4286          */
4287         rxcsum = rd32(E1000_RXCSUM);
4288         rxcsum |= E1000_RXCSUM_PCSD;
4289
4290         if (adapter->hw.mac.type >= e1000_82576)
4291                 /* Enable Receive Checksum Offload for SCTP */
4292                 rxcsum |= E1000_RXCSUM_CRCOFL;
4293
4294         /* Don't need to set TUOFL or IPOFL, they default to 1 */
4295         wr32(E1000_RXCSUM, rxcsum);
4296
4297         /* Generate RSS hash based on packet types, TCP/UDP
4298          * port numbers and/or IPv4/v6 src and dst addresses
4299          */
4300         mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4301                E1000_MRQC_RSS_FIELD_IPV4_TCP |
4302                E1000_MRQC_RSS_FIELD_IPV6 |
4303                E1000_MRQC_RSS_FIELD_IPV6_TCP |
4304                E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4305
4306         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4307                 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4308         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4309                 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4310
4311         /* If VMDq is enabled then we set the appropriate mode for that, else
4312          * we default to RSS so that an RSS hash is calculated per packet even
4313          * if we are only using one queue
4314          */
4315         if (adapter->vfs_allocated_count) {
4316                 if (hw->mac.type > e1000_82575) {
4317                         /* Set the default pool for the PF's first queue */
4318                         u32 vtctl = rd32(E1000_VT_CTL);
4319
4320                         vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4321                                    E1000_VT_CTL_DISABLE_DEF_POOL);
4322                         vtctl |= adapter->vfs_allocated_count <<
4323                                 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4324                         wr32(E1000_VT_CTL, vtctl);
4325                 }
4326                 if (adapter->rss_queues > 1)
4327                         mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4328                 else
4329                         mrqc |= E1000_MRQC_ENABLE_VMDQ;
4330         } else {
4331                 if (hw->mac.type != e1000_i211)
4332                         mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4333         }
4334         igb_vmm_control(adapter);
4335
4336         wr32(E1000_MRQC, mrqc);
4337 }
4338
4339 /**
4340  *  igb_setup_rctl - configure the receive control registers
4341  *  @adapter: Board private structure
4342  **/
4343 void igb_setup_rctl(struct igb_adapter *adapter)
4344 {
4345         struct e1000_hw *hw = &adapter->hw;
4346         u32 rctl;
4347
4348         rctl = rd32(E1000_RCTL);
4349
4350         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4351         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4352
4353         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4354                 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4355
4356         /* enable stripping of CRC. It's unlikely this will break BMC
4357          * redirection as it did with e1000. Newer features require
4358          * that the HW strips the CRC.
4359          */
4360         rctl |= E1000_RCTL_SECRC;
4361
4362         /* disable store bad packets and clear size bits. */
4363         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4364
4365         /* enable LPE to allow for reception of jumbo frames */
4366         rctl |= E1000_RCTL_LPE;
4367
4368         /* disable queue 0 to prevent tail write w/o re-config */
4369         wr32(E1000_RXDCTL(0), 0);
4370
4371         /* Attention!!!  For SR-IOV PF driver operations you must enable
4372          * queue drop for all VF and PF queues to prevent head of line blocking
4373          * if an un-trusted VF does not provide descriptors to hardware.
4374          */
4375         if (adapter->vfs_allocated_count) {
4376                 /* set all queue drop enable bits */
4377                 wr32(E1000_QDE, ALL_QUEUES);
4378         }
4379
4380         /* This is useful for sniffing bad packets. */
4381         if (adapter->netdev->features & NETIF_F_RXALL) {
4382                 /* UPE and MPE will be handled by normal PROMISC logic
4383                  * in e1000e_set_rx_mode
4384                  */
4385                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4386                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
4387                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4388
4389                 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4390                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4391                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4392                  * and that breaks VLANs.
4393                  */
4394         }
4395
4396         wr32(E1000_RCTL, rctl);
4397 }
4398
4399 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4400                                    int vfn)
4401 {
4402         struct e1000_hw *hw = &adapter->hw;
4403         u32 vmolr;
4404
4405         if (size > MAX_JUMBO_FRAME_SIZE)
4406                 size = MAX_JUMBO_FRAME_SIZE;
4407
4408         vmolr = rd32(E1000_VMOLR(vfn));
4409         vmolr &= ~E1000_VMOLR_RLPML_MASK;
4410         vmolr |= size | E1000_VMOLR_LPE;
4411         wr32(E1000_VMOLR(vfn), vmolr);
4412
4413         return 0;
4414 }
4415
4416 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4417                                          int vfn, bool enable)
4418 {
4419         struct e1000_hw *hw = &adapter->hw;
4420         u32 val, reg;
4421
4422         if (hw->mac.type < e1000_82576)
4423                 return;
4424
4425         if (hw->mac.type == e1000_i350)
4426                 reg = E1000_DVMOLR(vfn);
4427         else
4428                 reg = E1000_VMOLR(vfn);
4429
4430         val = rd32(reg);
4431         if (enable)
4432                 val |= E1000_VMOLR_STRVLAN;
4433         else
4434                 val &= ~(E1000_VMOLR_STRVLAN);
4435         wr32(reg, val);
4436 }
4437
4438 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4439                                  int vfn, bool aupe)
4440 {
4441         struct e1000_hw *hw = &adapter->hw;
4442         u32 vmolr;
4443
4444         /* This register exists only on 82576 and newer so if we are older then
4445          * we should exit and do nothing
4446          */
4447         if (hw->mac.type < e1000_82576)
4448                 return;
4449
4450         vmolr = rd32(E1000_VMOLR(vfn));
4451         if (aupe)
4452                 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4453         else
4454                 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4455
4456         /* clear all bits that might not be set */
4457         vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4458
4459         if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4460                 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4461         /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4462          * multicast packets
4463          */
4464         if (vfn <= adapter->vfs_allocated_count)
4465                 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4466
4467         wr32(E1000_VMOLR(vfn), vmolr);
4468 }
4469
4470 /**
4471  *  igb_setup_srrctl - configure the split and replication receive control
4472  *                     registers
4473  *  @adapter: Board private structure
4474  *  @ring: receive ring to be configured
4475  **/
4476 void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
4477 {
4478         struct e1000_hw *hw = &adapter->hw;
4479         int reg_idx = ring->reg_idx;
4480         u32 srrctl = 0;
4481
4482         srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4483         if (ring_uses_large_buffer(ring))
4484                 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4485         else
4486                 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4487         srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4488         if (hw->mac.type >= e1000_82580)
4489                 srrctl |= E1000_SRRCTL_TIMESTAMP;
4490         /* Only set Drop Enable if VFs allocated, or we are supporting multiple
4491          * queues and rx flow control is disabled
4492          */
4493         if (adapter->vfs_allocated_count ||
4494             (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
4495              adapter->num_rx_queues > 1))
4496                 srrctl |= E1000_SRRCTL_DROP_EN;
4497
4498         wr32(E1000_SRRCTL(reg_idx), srrctl);
4499 }
4500
4501 /**
4502  *  igb_configure_rx_ring - Configure a receive ring after Reset
4503  *  @adapter: board private structure
4504  *  @ring: receive ring to be configured
4505  *
4506  *  Configure the Rx unit of the MAC after a reset.
4507  **/
4508 void igb_configure_rx_ring(struct igb_adapter *adapter,
4509                            struct igb_ring *ring)
4510 {
4511         struct e1000_hw *hw = &adapter->hw;
4512         union e1000_adv_rx_desc *rx_desc;
4513         u64 rdba = ring->dma;
4514         int reg_idx = ring->reg_idx;
4515         u32 rxdctl = 0;
4516
4517         /* disable the queue */
4518         wr32(E1000_RXDCTL(reg_idx), 0);
4519
4520         /* Set DMA base address registers */
4521         wr32(E1000_RDBAL(reg_idx),
4522              rdba & 0x00000000ffffffffULL);
4523         wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4524         wr32(E1000_RDLEN(reg_idx),
4525              ring->count * sizeof(union e1000_adv_rx_desc));
4526
4527         /* initialize head and tail */
4528         ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4529         wr32(E1000_RDH(reg_idx), 0);
4530         writel(0, ring->tail);
4531
4532         /* set descriptor configuration */
4533         igb_setup_srrctl(adapter, ring);
4534
4535         /* set filtering for VMDQ pools */
4536         igb_set_vmolr(adapter, reg_idx & 0x7, true);
4537
4538         rxdctl |= IGB_RX_PTHRESH;
4539         rxdctl |= IGB_RX_HTHRESH << 8;
4540         rxdctl |= IGB_RX_WTHRESH << 16;
4541
4542         /* initialize rx_buffer_info */
4543         memset(ring->rx_buffer_info, 0,
4544                sizeof(struct igb_rx_buffer) * ring->count);
4545
4546         /* initialize Rx descriptor 0 */
4547         rx_desc = IGB_RX_DESC(ring, 0);
4548         rx_desc->wb.upper.length = 0;
4549
4550         /* enable receive descriptor fetching */
4551         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4552         wr32(E1000_RXDCTL(reg_idx), rxdctl);
4553 }
4554
4555 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4556                                   struct igb_ring *rx_ring)
4557 {
4558         /* set build_skb and buffer size flags */
4559         clear_ring_build_skb_enabled(rx_ring);
4560         clear_ring_uses_large_buffer(rx_ring);
4561
4562         if (adapter->flags & IGB_FLAG_RX_LEGACY)
4563                 return;
4564
4565         set_ring_build_skb_enabled(rx_ring);
4566
4567 #if (PAGE_SIZE < 8192)
4568         if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4569                 return;
4570
4571         set_ring_uses_large_buffer(rx_ring);
4572 #endif
4573 }
4574
4575 /**
4576  *  igb_configure_rx - Configure receive Unit after Reset
4577  *  @adapter: board private structure
4578  *
4579  *  Configure the Rx unit of the MAC after a reset.
4580  **/
4581 static void igb_configure_rx(struct igb_adapter *adapter)
4582 {
4583         int i;
4584
4585         /* set the correct pool for the PF default MAC address in entry 0 */
4586         igb_set_default_mac_filter(adapter);
4587
4588         /* Setup the HW Rx Head and Tail Descriptor Pointers and
4589          * the Base and Length of the Rx Descriptor Ring
4590          */
4591         for (i = 0; i < adapter->num_rx_queues; i++) {
4592                 struct igb_ring *rx_ring = adapter->rx_ring[i];
4593
4594                 igb_set_rx_buffer_len(adapter, rx_ring);
4595                 igb_configure_rx_ring(adapter, rx_ring);
4596         }
4597 }
4598
4599 /**
4600  *  igb_free_tx_resources - Free Tx Resources per Queue
4601  *  @tx_ring: Tx descriptor ring for a specific queue
4602  *
4603  *  Free all transmit software resources
4604  **/
4605 void igb_free_tx_resources(struct igb_ring *tx_ring)
4606 {
4607         igb_clean_tx_ring(tx_ring);
4608
4609         vfree(tx_ring->tx_buffer_info);
4610         tx_ring->tx_buffer_info = NULL;
4611
4612         /* if not set, then don't free */
4613         if (!tx_ring->desc)
4614                 return;
4615
4616         dma_free_coherent(tx_ring->dev, tx_ring->size,
4617                           tx_ring->desc, tx_ring->dma);
4618
4619         tx_ring->desc = NULL;
4620 }
4621
4622 /**
4623  *  igb_free_all_tx_resources - Free Tx Resources for All Queues
4624  *  @adapter: board private structure
4625  *
4626  *  Free all transmit software resources
4627  **/
4628 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4629 {
4630         int i;
4631
4632         for (i = 0; i < adapter->num_tx_queues; i++)
4633                 if (adapter->tx_ring[i])
4634                         igb_free_tx_resources(adapter->tx_ring[i]);
4635 }
4636
4637 /**
4638  *  igb_clean_tx_ring - Free Tx Buffers
4639  *  @tx_ring: ring to be cleaned
4640  **/
4641 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4642 {
4643         u16 i = tx_ring->next_to_clean;
4644         struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4645
4646         while (i != tx_ring->next_to_use) {
4647                 union e1000_adv_tx_desc *eop_desc, *tx_desc;
4648
4649                 /* Free all the Tx ring sk_buffs */
4650                 dev_kfree_skb_any(tx_buffer->skb);
4651
4652                 /* unmap skb header data */
4653                 dma_unmap_single(tx_ring->dev,
4654                                  dma_unmap_addr(tx_buffer, dma),
4655                                  dma_unmap_len(tx_buffer, len),
4656                                  DMA_TO_DEVICE);
4657
4658                 /* check for eop_desc to determine the end of the packet */
4659                 eop_desc = tx_buffer->next_to_watch;
4660                 tx_desc = IGB_TX_DESC(tx_ring, i);
4661
4662                 /* unmap remaining buffers */
4663                 while (tx_desc != eop_desc) {
4664                         tx_buffer++;
4665                         tx_desc++;
4666                         i++;
4667                         if (unlikely(i == tx_ring->count)) {
4668                                 i = 0;
4669                                 tx_buffer = tx_ring->tx_buffer_info;
4670                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
4671                         }
4672
4673                         /* unmap any remaining paged data */
4674                         if (dma_unmap_len(tx_buffer, len))
4675                                 dma_unmap_page(tx_ring->dev,
4676                                                dma_unmap_addr(tx_buffer, dma),
4677                                                dma_unmap_len(tx_buffer, len),
4678                                                DMA_TO_DEVICE);
4679                 }
4680
4681                 /* move us one more past the eop_desc for start of next pkt */
4682                 tx_buffer++;
4683                 i++;
4684                 if (unlikely(i == tx_ring->count)) {
4685                         i = 0;
4686                         tx_buffer = tx_ring->tx_buffer_info;
4687                 }
4688         }
4689
4690         /* reset BQL for queue */
4691         netdev_tx_reset_queue(txring_txq(tx_ring));
4692
4693         /* reset next_to_use and next_to_clean */
4694         tx_ring->next_to_use = 0;
4695         tx_ring->next_to_clean = 0;
4696 }
4697
4698 /**
4699  *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
4700  *  @adapter: board private structure
4701  **/
4702 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4703 {
4704         int i;
4705
4706         for (i = 0; i < adapter->num_tx_queues; i++)
4707                 if (adapter->tx_ring[i])
4708                         igb_clean_tx_ring(adapter->tx_ring[i]);
4709 }
4710
4711 /**
4712  *  igb_free_rx_resources - Free Rx Resources
4713  *  @rx_ring: ring to clean the resources from
4714  *
4715  *  Free all receive software resources
4716  **/
4717 void igb_free_rx_resources(struct igb_ring *rx_ring)
4718 {
4719         igb_clean_rx_ring(rx_ring);
4720
4721         vfree(rx_ring->rx_buffer_info);
4722         rx_ring->rx_buffer_info = NULL;
4723
4724         /* if not set, then don't free */
4725         if (!rx_ring->desc)
4726                 return;
4727
4728         dma_free_coherent(rx_ring->dev, rx_ring->size,
4729                           rx_ring->desc, rx_ring->dma);
4730
4731         rx_ring->desc = NULL;
4732 }
4733
4734 /**
4735  *  igb_free_all_rx_resources - Free Rx Resources for All Queues
4736  *  @adapter: board private structure
4737  *
4738  *  Free all receive software resources
4739  **/
4740 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4741 {
4742         int i;
4743
4744         for (i = 0; i < adapter->num_rx_queues; i++)
4745                 if (adapter->rx_ring[i])
4746                         igb_free_rx_resources(adapter->rx_ring[i]);
4747 }
4748
4749 /**
4750  *  igb_clean_rx_ring - Free Rx Buffers per Queue
4751  *  @rx_ring: ring to free buffers from
4752  **/
4753 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4754 {
4755         u16 i = rx_ring->next_to_clean;
4756
4757         dev_kfree_skb(rx_ring->skb);
4758         rx_ring->skb = NULL;
4759
4760         /* Free all the Rx ring sk_buffs */
4761         while (i != rx_ring->next_to_alloc) {
4762                 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4763
4764                 /* Invalidate cache lines that may have been written to by
4765                  * device so that we avoid corrupting memory.
4766                  */
4767                 dma_sync_single_range_for_cpu(rx_ring->dev,
4768                                               buffer_info->dma,
4769                                               buffer_info->page_offset,
4770                                               igb_rx_bufsz(rx_ring),
4771                                               DMA_FROM_DEVICE);
4772
4773                 /* free resources associated with mapping */
4774                 dma_unmap_page_attrs(rx_ring->dev,
4775                                      buffer_info->dma,
4776                                      igb_rx_pg_size(rx_ring),
4777                                      DMA_FROM_DEVICE,
4778                                      IGB_RX_DMA_ATTR);
4779                 __page_frag_cache_drain(buffer_info->page,
4780                                         buffer_info->pagecnt_bias);
4781
4782                 i++;
4783                 if (i == rx_ring->count)
4784                         i = 0;
4785         }
4786
4787         rx_ring->next_to_alloc = 0;
4788         rx_ring->next_to_clean = 0;
4789         rx_ring->next_to_use = 0;
4790 }
4791
4792 /**
4793  *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
4794  *  @adapter: board private structure
4795  **/
4796 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4797 {
4798         int i;
4799
4800         for (i = 0; i < adapter->num_rx_queues; i++)
4801                 if (adapter->rx_ring[i])
4802                         igb_clean_rx_ring(adapter->rx_ring[i]);
4803 }
4804
4805 /**
4806  *  igb_set_mac - Change the Ethernet Address of the NIC
4807  *  @netdev: network interface device structure
4808  *  @p: pointer to an address structure
4809  *
4810  *  Returns 0 on success, negative on failure
4811  **/
4812 static int igb_set_mac(struct net_device *netdev, void *p)
4813 {
4814         struct igb_adapter *adapter = netdev_priv(netdev);
4815         struct e1000_hw *hw = &adapter->hw;
4816         struct sockaddr *addr = p;
4817
4818         if (!is_valid_ether_addr(addr->sa_data))
4819                 return -EADDRNOTAVAIL;
4820
4821         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4822         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4823
4824         /* set the correct pool for the new PF MAC address in entry 0 */
4825         igb_set_default_mac_filter(adapter);
4826
4827         return 0;
4828 }
4829
4830 /**
4831  *  igb_write_mc_addr_list - write multicast addresses to MTA
4832  *  @netdev: network interface device structure
4833  *
4834  *  Writes multicast address list to the MTA hash table.
4835  *  Returns: -ENOMEM on failure
4836  *           0 on no addresses written
4837  *           X on writing X addresses to MTA
4838  **/
4839 static int igb_write_mc_addr_list(struct net_device *netdev)
4840 {
4841         struct igb_adapter *adapter = netdev_priv(netdev);
4842         struct e1000_hw *hw = &adapter->hw;
4843         struct netdev_hw_addr *ha;
4844         u8  *mta_list;
4845         int i;
4846
4847         if (netdev_mc_empty(netdev)) {
4848                 /* nothing to program, so clear mc list */
4849                 igb_update_mc_addr_list(hw, NULL, 0);
4850                 igb_restore_vf_multicasts(adapter);
4851                 return 0;
4852         }
4853
4854         mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
4855         if (!mta_list)
4856                 return -ENOMEM;
4857
4858         /* The shared function expects a packed array of only addresses. */
4859         i = 0;
4860         netdev_for_each_mc_addr(ha, netdev)
4861                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
4862
4863         igb_update_mc_addr_list(hw, mta_list, i);
4864         kfree(mta_list);
4865
4866         return netdev_mc_count(netdev);
4867 }
4868
4869 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
4870 {
4871         struct e1000_hw *hw = &adapter->hw;
4872         u32 i, pf_id;
4873
4874         switch (hw->mac.type) {
4875         case e1000_i210:
4876         case e1000_i211:
4877         case e1000_i350:
4878                 /* VLAN filtering needed for VLAN prio filter */
4879                 if (adapter->netdev->features & NETIF_F_NTUPLE)
4880                         break;
4881                 /* fall through */
4882         case e1000_82576:
4883         case e1000_82580:
4884         case e1000_i354:
4885                 /* VLAN filtering needed for pool filtering */
4886                 if (adapter->vfs_allocated_count)
4887                         break;
4888                 /* fall through */
4889         default:
4890                 return 1;
4891         }
4892
4893         /* We are already in VLAN promisc, nothing to do */
4894         if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
4895                 return 0;
4896
4897         if (!adapter->vfs_allocated_count)
4898                 goto set_vfta;
4899
4900         /* Add PF to all active pools */
4901         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4902
4903         for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4904                 u32 vlvf = rd32(E1000_VLVF(i));
4905
4906                 vlvf |= BIT(pf_id);
4907                 wr32(E1000_VLVF(i), vlvf);
4908         }
4909
4910 set_vfta:
4911         /* Set all bits in the VLAN filter table array */
4912         for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
4913                 hw->mac.ops.write_vfta(hw, i, ~0U);
4914
4915         /* Set flag so we don't redo unnecessary work */
4916         adapter->flags |= IGB_FLAG_VLAN_PROMISC;
4917
4918         return 0;
4919 }
4920
4921 #define VFTA_BLOCK_SIZE 8
4922 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
4923 {
4924         struct e1000_hw *hw = &adapter->hw;
4925         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4926         u32 vid_start = vfta_offset * 32;
4927         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4928         u32 i, vid, word, bits, pf_id;
4929
4930         /* guarantee that we don't scrub out management VLAN */
4931         vid = adapter->mng_vlan_id;
4932         if (vid >= vid_start && vid < vid_end)
4933                 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4934
4935         if (!adapter->vfs_allocated_count)
4936                 goto set_vfta;
4937
4938         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4939
4940         for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4941                 u32 vlvf = rd32(E1000_VLVF(i));
4942
4943                 /* pull VLAN ID from VLVF */
4944                 vid = vlvf & VLAN_VID_MASK;
4945
4946                 /* only concern ourselves with a certain range */
4947                 if (vid < vid_start || vid >= vid_end)
4948                         continue;
4949
4950                 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
4951                         /* record VLAN ID in VFTA */
4952                         vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4953
4954                         /* if PF is part of this then continue */
4955                         if (test_bit(vid, adapter->active_vlans))
4956                                 continue;
4957                 }
4958
4959                 /* remove PF from the pool */
4960                 bits = ~BIT(pf_id);
4961                 bits &= rd32(E1000_VLVF(i));
4962                 wr32(E1000_VLVF(i), bits);
4963         }
4964
4965 set_vfta:
4966         /* extract values from active_vlans and write back to VFTA */
4967         for (i = VFTA_BLOCK_SIZE; i--;) {
4968                 vid = (vfta_offset + i) * 32;
4969                 word = vid / BITS_PER_LONG;
4970                 bits = vid % BITS_PER_LONG;
4971
4972                 vfta[i] |= adapter->active_vlans[word] >> bits;
4973
4974                 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
4975         }
4976 }
4977
4978 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
4979 {
4980         u32 i;
4981
4982         /* We are not in VLAN promisc, nothing to do */
4983         if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
4984                 return;
4985
4986         /* Set flag so we don't redo unnecessary work */
4987         adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
4988
4989         for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
4990                 igb_scrub_vfta(adapter, i);
4991 }
4992
4993 /**
4994  *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
4995  *  @netdev: network interface device structure
4996  *
4997  *  The set_rx_mode entry point is called whenever the unicast or multicast
4998  *  address lists or the network interface flags are updated.  This routine is
4999  *  responsible for configuring the hardware for proper unicast, multicast,
5000  *  promiscuous mode, and all-multi behavior.
5001  **/
5002 static void igb_set_rx_mode(struct net_device *netdev)
5003 {
5004         struct igb_adapter *adapter = netdev_priv(netdev);
5005         struct e1000_hw *hw = &adapter->hw;
5006         unsigned int vfn = adapter->vfs_allocated_count;
5007         u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5008         int count;
5009
5010         /* Check for Promiscuous and All Multicast modes */
5011         if (netdev->flags & IFF_PROMISC) {
5012                 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5013                 vmolr |= E1000_VMOLR_MPME;
5014
5015                 /* enable use of UTA filter to force packets to default pool */
5016                 if (hw->mac.type == e1000_82576)
5017                         vmolr |= E1000_VMOLR_ROPE;
5018         } else {
5019                 if (netdev->flags & IFF_ALLMULTI) {
5020                         rctl |= E1000_RCTL_MPE;
5021                         vmolr |= E1000_VMOLR_MPME;
5022                 } else {
5023                         /* Write addresses to the MTA, if the attempt fails
5024                          * then we should just turn on promiscuous mode so
5025                          * that we can at least receive multicast traffic
5026                          */
5027                         count = igb_write_mc_addr_list(netdev);
5028                         if (count < 0) {
5029                                 rctl |= E1000_RCTL_MPE;
5030                                 vmolr |= E1000_VMOLR_MPME;
5031                         } else if (count) {
5032                                 vmolr |= E1000_VMOLR_ROMPE;
5033                         }
5034                 }
5035         }
5036
5037         /* Write addresses to available RAR registers, if there is not
5038          * sufficient space to store all the addresses then enable
5039          * unicast promiscuous mode
5040          */
5041         if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5042                 rctl |= E1000_RCTL_UPE;
5043                 vmolr |= E1000_VMOLR_ROPE;
5044         }
5045
5046         /* enable VLAN filtering by default */
5047         rctl |= E1000_RCTL_VFE;
5048
5049         /* disable VLAN filtering for modes that require it */
5050         if ((netdev->flags & IFF_PROMISC) ||
5051             (netdev->features & NETIF_F_RXALL)) {
5052                 /* if we fail to set all rules then just clear VFE */
5053                 if (igb_vlan_promisc_enable(adapter))
5054                         rctl &= ~E1000_RCTL_VFE;
5055         } else {
5056                 igb_vlan_promisc_disable(adapter);
5057         }
5058
5059         /* update state of unicast, multicast, and VLAN filtering modes */
5060         rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5061                                      E1000_RCTL_VFE);
5062         wr32(E1000_RCTL, rctl);
5063
5064 #if (PAGE_SIZE < 8192)
5065         if (!adapter->vfs_allocated_count) {
5066                 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5067                         rlpml = IGB_MAX_FRAME_BUILD_SKB;
5068         }
5069 #endif
5070         wr32(E1000_RLPML, rlpml);
5071
5072         /* In order to support SR-IOV and eventually VMDq it is necessary to set
5073          * the VMOLR to enable the appropriate modes.  Without this workaround
5074          * we will have issues with VLAN tag stripping not being done for frames
5075          * that are only arriving because we are the default pool
5076          */
5077         if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5078                 return;
5079
5080         /* set UTA to appropriate mode */
5081         igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5082
5083         vmolr |= rd32(E1000_VMOLR(vfn)) &
5084                  ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5085
5086         /* enable Rx jumbo frames, restrict as needed to support build_skb */
5087         vmolr &= ~E1000_VMOLR_RLPML_MASK;
5088 #if (PAGE_SIZE < 8192)
5089         if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5090                 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5091         else
5092 #endif
5093                 vmolr |= MAX_JUMBO_FRAME_SIZE;
5094         vmolr |= E1000_VMOLR_LPE;
5095
5096         wr32(E1000_VMOLR(vfn), vmolr);
5097
5098         igb_restore_vf_multicasts(adapter);
5099 }
5100
5101 static void igb_check_wvbr(struct igb_adapter *adapter)
5102 {
5103         struct e1000_hw *hw = &adapter->hw;
5104         u32 wvbr = 0;
5105
5106         switch (hw->mac.type) {
5107         case e1000_82576:
5108         case e1000_i350:
5109                 wvbr = rd32(E1000_WVBR);
5110                 if (!wvbr)
5111                         return;
5112                 break;
5113         default:
5114                 break;
5115         }
5116
5117         adapter->wvbr |= wvbr;
5118 }
5119
5120 #define IGB_STAGGERED_QUEUE_OFFSET 8
5121
5122 static void igb_spoof_check(struct igb_adapter *adapter)
5123 {
5124         int j;
5125
5126         if (!adapter->wvbr)
5127                 return;
5128
5129         for (j = 0; j < adapter->vfs_allocated_count; j++) {
5130                 if (adapter->wvbr & BIT(j) ||
5131                     adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5132                         dev_warn(&adapter->pdev->dev,
5133                                 "Spoof event(s) detected on VF %d\n", j);
5134                         adapter->wvbr &=
5135                                 ~(BIT(j) |
5136                                   BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5137                 }
5138         }
5139 }
5140
5141 /* Need to wait a few seconds after link up to get diagnostic information from
5142  * the phy
5143  */
5144 static void igb_update_phy_info(struct timer_list *t)
5145 {
5146         struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5147         igb_get_phy_info(&adapter->hw);
5148 }
5149
5150 /**
5151  *  igb_has_link - check shared code for link and determine up/down
5152  *  @adapter: pointer to driver private info
5153  **/
5154 bool igb_has_link(struct igb_adapter *adapter)
5155 {
5156         struct e1000_hw *hw = &adapter->hw;
5157         bool link_active = false;
5158
5159         /* get_link_status is set on LSC (link status) interrupt or
5160          * rx sequence error interrupt.  get_link_status will stay
5161          * false until the e1000_check_for_link establishes link
5162          * for copper adapters ONLY
5163          */
5164         switch (hw->phy.media_type) {
5165         case e1000_media_type_copper:
5166                 if (!hw->mac.get_link_status)
5167                         return true;
5168                 /* fall through */
5169         case e1000_media_type_internal_serdes:
5170                 hw->mac.ops.check_for_link(hw);
5171                 link_active = !hw->mac.get_link_status;
5172                 break;
5173         default:
5174         case e1000_media_type_unknown:
5175                 break;
5176         }
5177
5178         if (((hw->mac.type == e1000_i210) ||
5179              (hw->mac.type == e1000_i211)) &&
5180              (hw->phy.id == I210_I_PHY_ID)) {
5181                 if (!netif_carrier_ok(adapter->netdev)) {
5182                         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5183                 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5184                         adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5185                         adapter->link_check_timeout = jiffies;
5186                 }
5187         }
5188
5189         return link_active;
5190 }
5191
5192 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5193 {
5194         bool ret = false;
5195         u32 ctrl_ext, thstat;
5196
5197         /* check for thermal sensor event on i350 copper only */
5198         if (hw->mac.type == e1000_i350) {
5199                 thstat = rd32(E1000_THSTAT);
5200                 ctrl_ext = rd32(E1000_CTRL_EXT);
5201
5202                 if ((hw->phy.media_type == e1000_media_type_copper) &&
5203                     !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5204                         ret = !!(thstat & event);
5205         }
5206
5207         return ret;
5208 }
5209
5210 /**
5211  *  igb_check_lvmmc - check for malformed packets received
5212  *  and indicated in LVMMC register
5213  *  @adapter: pointer to adapter
5214  **/
5215 static void igb_check_lvmmc(struct igb_adapter *adapter)
5216 {
5217         struct e1000_hw *hw = &adapter->hw;
5218         u32 lvmmc;
5219
5220         lvmmc = rd32(E1000_LVMMC);
5221         if (lvmmc) {
5222                 if (unlikely(net_ratelimit())) {
5223                         netdev_warn(adapter->netdev,
5224                                     "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5225                                     lvmmc);
5226                 }
5227         }
5228 }
5229
5230 /**
5231  *  igb_watchdog - Timer Call-back
5232  *  @data: pointer to adapter cast into an unsigned long
5233  **/
5234 static void igb_watchdog(struct timer_list *t)
5235 {
5236         struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5237         /* Do the rest outside of interrupt context */
5238         schedule_work(&adapter->watchdog_task);
5239 }
5240
5241 static void igb_watchdog_task(struct work_struct *work)
5242 {
5243         struct igb_adapter *adapter = container_of(work,
5244                                                    struct igb_adapter,
5245                                                    watchdog_task);
5246         struct e1000_hw *hw = &adapter->hw;
5247         struct e1000_phy_info *phy = &hw->phy;
5248         struct net_device *netdev = adapter->netdev;
5249         u32 link;
5250         int i;
5251         u32 connsw;
5252         u16 phy_data, retry_count = 20;
5253
5254         link = igb_has_link(adapter);
5255
5256         if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5257                 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5258                         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5259                 else
5260                         link = false;
5261         }
5262
5263         /* Force link down if we have fiber to swap to */
5264         if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5265                 if (hw->phy.media_type == e1000_media_type_copper) {
5266                         connsw = rd32(E1000_CONNSW);
5267                         if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5268                                 link = 0;
5269                 }
5270         }
5271         if (link) {
5272                 /* Perform a reset if the media type changed. */
5273                 if (hw->dev_spec._82575.media_changed) {
5274                         hw->dev_spec._82575.media_changed = false;
5275                         adapter->flags |= IGB_FLAG_MEDIA_RESET;
5276                         igb_reset(adapter);
5277                 }
5278                 /* Cancel scheduled suspend requests. */
5279                 pm_runtime_resume(netdev->dev.parent);
5280
5281                 if (!netif_carrier_ok(netdev)) {
5282                         u32 ctrl;
5283
5284                         hw->mac.ops.get_speed_and_duplex(hw,
5285                                                          &adapter->link_speed,
5286                                                          &adapter->link_duplex);
5287
5288                         ctrl = rd32(E1000_CTRL);
5289                         /* Links status message must follow this format */
5290                         netdev_info(netdev,
5291                                "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5292                                netdev->name,
5293                                adapter->link_speed,
5294                                adapter->link_duplex == FULL_DUPLEX ?
5295                                "Full" : "Half",
5296                                (ctrl & E1000_CTRL_TFCE) &&
5297                                (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5298                                (ctrl & E1000_CTRL_RFCE) ?  "RX" :
5299                                (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
5300
5301                         /* disable EEE if enabled */
5302                         if ((adapter->flags & IGB_FLAG_EEE) &&
5303                                 (adapter->link_duplex == HALF_DUPLEX)) {
5304                                 dev_info(&adapter->pdev->dev,
5305                                 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5306                                 adapter->hw.dev_spec._82575.eee_disable = true;
5307                                 adapter->flags &= ~IGB_FLAG_EEE;
5308                         }
5309
5310                         /* check if SmartSpeed worked */
5311                         igb_check_downshift(hw);
5312                         if (phy->speed_downgraded)
5313                                 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5314
5315                         /* check for thermal sensor event */
5316                         if (igb_thermal_sensor_event(hw,
5317                             E1000_THSTAT_LINK_THROTTLE))
5318                                 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5319
5320                         /* adjust timeout factor according to speed/duplex */
5321                         adapter->tx_timeout_factor = 1;
5322                         switch (adapter->link_speed) {
5323                         case SPEED_10:
5324                                 adapter->tx_timeout_factor = 14;
5325                                 break;
5326                         case SPEED_100:
5327                                 /* maybe add some timeout factor ? */
5328                                 break;
5329                         }
5330
5331                         if (adapter->link_speed != SPEED_1000)
5332                                 goto no_wait;
5333
5334                         /* wait for Remote receiver status OK */
5335 retry_read_status:
5336                         if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5337                                               &phy_data)) {
5338                                 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5339                                     retry_count) {
5340                                         msleep(100);
5341                                         retry_count--;
5342                                         goto retry_read_status;
5343                                 } else if (!retry_count) {
5344                                         dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5345                                 }
5346                         } else {
5347                                 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5348                         }
5349 no_wait:
5350                         netif_carrier_on(netdev);
5351
5352                         igb_ping_all_vfs(adapter);
5353                         igb_check_vf_rate_limit(adapter);
5354
5355                         /* link state has changed, schedule phy info update */
5356                         if (!test_bit(__IGB_DOWN, &adapter->state))
5357                                 mod_timer(&adapter->phy_info_timer,
5358                                           round_jiffies(jiffies + 2 * HZ));
5359                 }
5360         } else {
5361                 if (netif_carrier_ok(netdev)) {
5362                         adapter->link_speed = 0;
5363                         adapter->link_duplex = 0;
5364
5365                         /* check for thermal sensor event */
5366                         if (igb_thermal_sensor_event(hw,
5367                             E1000_THSTAT_PWR_DOWN)) {
5368                                 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5369                         }
5370
5371                         /* Links status message must follow this format */
5372                         netdev_info(netdev, "igb: %s NIC Link is Down\n",
5373                                netdev->name);
5374                         netif_carrier_off(netdev);
5375
5376                         igb_ping_all_vfs(adapter);
5377
5378                         /* link state has changed, schedule phy info update */
5379                         if (!test_bit(__IGB_DOWN, &adapter->state))
5380                                 mod_timer(&adapter->phy_info_timer,
5381                                           round_jiffies(jiffies + 2 * HZ));
5382
5383                         /* link is down, time to check for alternate media */
5384                         if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5385                                 igb_check_swap_media(adapter);
5386                                 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5387                                         schedule_work(&adapter->reset_task);
5388                                         /* return immediately */
5389                                         return;
5390                                 }
5391                         }
5392                         pm_schedule_suspend(netdev->dev.parent,
5393                                             MSEC_PER_SEC * 5);
5394
5395                 /* also check for alternate media here */
5396                 } else if (!netif_carrier_ok(netdev) &&
5397                            (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5398                         igb_check_swap_media(adapter);
5399                         if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5400                                 schedule_work(&adapter->reset_task);
5401                                 /* return immediately */
5402                                 return;
5403                         }
5404                 }
5405         }
5406
5407         spin_lock(&adapter->stats64_lock);
5408         igb_update_stats(adapter);
5409         spin_unlock(&adapter->stats64_lock);
5410
5411         for (i = 0; i < adapter->num_tx_queues; i++) {
5412                 struct igb_ring *tx_ring = adapter->tx_ring[i];
5413                 if (!netif_carrier_ok(netdev)) {
5414                         /* We've lost link, so the controller stops DMA,
5415                          * but we've got queued Tx work that's never going
5416                          * to get done, so reset controller to flush Tx.
5417                          * (Do the reset outside of interrupt context).
5418                          */
5419                         if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5420                                 adapter->tx_timeout_count++;
5421                                 schedule_work(&adapter->reset_task);
5422                                 /* return immediately since reset is imminent */
5423                                 return;
5424                         }
5425                 }
5426
5427                 /* Force detection of hung controller every watchdog period */
5428                 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5429         }
5430
5431         /* Cause software interrupt to ensure Rx ring is cleaned */
5432         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5433                 u32 eics = 0;
5434
5435                 for (i = 0; i < adapter->num_q_vectors; i++)
5436                         eics |= adapter->q_vector[i]->eims_value;
5437                 wr32(E1000_EICS, eics);
5438         } else {
5439                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
5440         }
5441
5442         igb_spoof_check(adapter);
5443         igb_ptp_rx_hang(adapter);
5444         igb_ptp_tx_hang(adapter);
5445
5446         /* Check LVMMC register on i350/i354 only */
5447         if ((adapter->hw.mac.type == e1000_i350) ||
5448             (adapter->hw.mac.type == e1000_i354))
5449                 igb_check_lvmmc(adapter);
5450
5451         /* Reset the timer */
5452         if (!test_bit(__IGB_DOWN, &adapter->state)) {
5453                 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5454                         mod_timer(&adapter->watchdog_timer,
5455                                   round_jiffies(jiffies +  HZ));
5456                 else
5457                         mod_timer(&adapter->watchdog_timer,
5458                                   round_jiffies(jiffies + 2 * HZ));
5459         }
5460 }
5461
5462 enum latency_range {
5463         lowest_latency = 0,
5464         low_latency = 1,
5465         bulk_latency = 2,
5466         latency_invalid = 255
5467 };
5468
5469 /**
5470  *  igb_update_ring_itr - update the dynamic ITR value based on packet size
5471  *  @q_vector: pointer to q_vector
5472  *
5473  *  Stores a new ITR value based on strictly on packet size.  This
5474  *  algorithm is less sophisticated than that used in igb_update_itr,
5475  *  due to the difficulty of synchronizing statistics across multiple
5476  *  receive rings.  The divisors and thresholds used by this function
5477  *  were determined based on theoretical maximum wire speed and testing
5478  *  data, in order to minimize response time while increasing bulk
5479  *  throughput.
5480  *  This functionality is controlled by ethtool's coalescing settings.
5481  *  NOTE:  This function is called only when operating in a multiqueue
5482  *         receive environment.
5483  **/
5484 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5485 {
5486         int new_val = q_vector->itr_val;
5487         int avg_wire_size = 0;
5488         struct igb_adapter *adapter = q_vector->adapter;
5489         unsigned int packets;
5490
5491         /* For non-gigabit speeds, just fix the interrupt rate at 4000
5492          * ints/sec - ITR timer value of 120 ticks.
5493          */
5494         if (adapter->link_speed != SPEED_1000) {
5495                 new_val = IGB_4K_ITR;
5496                 goto set_itr_val;
5497         }
5498
5499         packets = q_vector->rx.total_packets;
5500         if (packets)
5501                 avg_wire_size = q_vector->rx.total_bytes / packets;
5502
5503         packets = q_vector->tx.total_packets;
5504         if (packets)
5505                 avg_wire_size = max_t(u32, avg_wire_size,
5506                                       q_vector->tx.total_bytes / packets);
5507
5508         /* if avg_wire_size isn't set no work was done */
5509         if (!avg_wire_size)
5510                 goto clear_counts;
5511
5512         /* Add 24 bytes to size to account for CRC, preamble, and gap */
5513         avg_wire_size += 24;
5514
5515         /* Don't starve jumbo frames */
5516         avg_wire_size = min(avg_wire_size, 3000);
5517
5518         /* Give a little boost to mid-size frames */
5519         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5520                 new_val = avg_wire_size / 3;
5521         else
5522                 new_val = avg_wire_size / 2;
5523
5524         /* conservative mode (itr 3) eliminates the lowest_latency setting */
5525         if (new_val < IGB_20K_ITR &&
5526             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5527              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5528                 new_val = IGB_20K_ITR;
5529
5530 set_itr_val:
5531         if (new_val != q_vector->itr_val) {
5532                 q_vector->itr_val = new_val;
5533                 q_vector->set_itr = 1;
5534         }
5535 clear_counts:
5536         q_vector->rx.total_bytes = 0;
5537         q_vector->rx.total_packets = 0;
5538         q_vector->tx.total_bytes = 0;
5539         q_vector->tx.total_packets = 0;
5540 }
5541
5542 /**
5543  *  igb_update_itr - update the dynamic ITR value based on statistics
5544  *  @q_vector: pointer to q_vector
5545  *  @ring_container: ring info to update the itr for
5546  *
5547  *  Stores a new ITR value based on packets and byte
5548  *  counts during the last interrupt.  The advantage of per interrupt
5549  *  computation is faster updates and more accurate ITR for the current
5550  *  traffic pattern.  Constants in this function were computed
5551  *  based on theoretical maximum wire speed and thresholds were set based
5552  *  on testing data as well as attempting to minimize response time
5553  *  while increasing bulk throughput.
5554  *  This functionality is controlled by ethtool's coalescing settings.
5555  *  NOTE:  These calculations are only valid when operating in a single-
5556  *         queue environment.
5557  **/
5558 static void igb_update_itr(struct igb_q_vector *q_vector,
5559                            struct igb_ring_container *ring_container)
5560 {
5561         unsigned int packets = ring_container->total_packets;
5562         unsigned int bytes = ring_container->total_bytes;
5563         u8 itrval = ring_container->itr;
5564
5565         /* no packets, exit with status unchanged */
5566         if (packets == 0)
5567                 return;
5568
5569         switch (itrval) {
5570         case lowest_latency:
5571                 /* handle TSO and jumbo frames */
5572                 if (bytes/packets > 8000)
5573                         itrval = bulk_latency;
5574                 else if ((packets < 5) && (bytes > 512))
5575                         itrval = low_latency;
5576                 break;
5577         case low_latency:  /* 50 usec aka 20000 ints/s */
5578                 if (bytes > 10000) {
5579                         /* this if handles the TSO accounting */
5580                         if (bytes/packets > 8000)
5581                                 itrval = bulk_latency;
5582                         else if ((packets < 10) || ((bytes/packets) > 1200))
5583                                 itrval = bulk_latency;
5584                         else if ((packets > 35))
5585                                 itrval = lowest_latency;
5586                 } else if (bytes/packets > 2000) {
5587                         itrval = bulk_latency;
5588                 } else if (packets <= 2 && bytes < 512) {
5589                         itrval = lowest_latency;
5590                 }
5591                 break;
5592         case bulk_latency: /* 250 usec aka 4000 ints/s */
5593                 if (bytes > 25000) {
5594                         if (packets > 35)
5595                                 itrval = low_latency;
5596                 } else if (bytes < 1500) {
5597                         itrval = low_latency;
5598                 }
5599                 break;
5600         }
5601
5602         /* clear work counters since we have the values we need */
5603         ring_container->total_bytes = 0;
5604         ring_container->total_packets = 0;
5605
5606         /* write updated itr to ring container */
5607         ring_container->itr = itrval;
5608 }
5609
5610 static void igb_set_itr(struct igb_q_vector *q_vector)
5611 {
5612         struct igb_adapter *adapter = q_vector->adapter;
5613         u32 new_itr = q_vector->itr_val;
5614         u8 current_itr = 0;
5615
5616         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5617         if (adapter->link_speed != SPEED_1000) {
5618                 current_itr = 0;
5619                 new_itr = IGB_4K_ITR;
5620                 goto set_itr_now;
5621         }
5622
5623         igb_update_itr(q_vector, &q_vector->tx);
5624         igb_update_itr(q_vector, &q_vector->rx);
5625
5626         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5627
5628         /* conservative mode (itr 3) eliminates the lowest_latency setting */
5629         if (current_itr == lowest_latency &&
5630             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5631              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5632                 current_itr = low_latency;
5633
5634         switch (current_itr) {
5635         /* counts and packets in update_itr are dependent on these numbers */
5636         case lowest_latency:
5637                 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5638                 break;
5639         case low_latency:
5640                 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5641                 break;
5642         case bulk_latency:
5643                 new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
5644                 break;
5645         default:
5646                 break;
5647         }
5648
5649 set_itr_now:
5650         if (new_itr != q_vector->itr_val) {
5651                 /* this attempts to bias the interrupt rate towards Bulk
5652                  * by adding intermediate steps when interrupt rate is
5653                  * increasing
5654                  */
5655                 new_itr = new_itr > q_vector->itr_val ?
5656                           max((new_itr * q_vector->itr_val) /
5657                           (new_itr + (q_vector->itr_val >> 2)),
5658                           new_itr) : new_itr;
5659                 /* Don't write the value here; it resets the adapter's
5660                  * internal timer, and causes us to delay far longer than
5661                  * we should between interrupts.  Instead, we write the ITR
5662                  * value at the beginning of the next interrupt so the timing
5663                  * ends up being correct.
5664                  */
5665                 q_vector->itr_val = new_itr;
5666                 q_vector->set_itr = 1;
5667         }
5668 }
5669
5670 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5671                             struct igb_tx_buffer *first,
5672                             u32 vlan_macip_lens, u32 type_tucmd,
5673                             u32 mss_l4len_idx)
5674 {
5675         struct e1000_adv_tx_context_desc *context_desc;
5676         u16 i = tx_ring->next_to_use;
5677         struct timespec64 ts;
5678
5679         context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5680
5681         i++;
5682         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5683
5684         /* set bits to identify this as an advanced context descriptor */
5685         type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5686
5687         /* For 82575, context index must be unique per ring. */
5688         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5689                 mss_l4len_idx |= tx_ring->reg_idx << 4;
5690
5691         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
5692         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
5693         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
5694
5695         /* We assume there is always a valid tx time available. Invalid times
5696          * should have been handled by the upper layers.
5697          */
5698         if (tx_ring->launchtime_enable) {
5699                 ts = ktime_to_timespec64(first->skb->tstamp);
5700                 first->skb->tstamp = ktime_set(0, 0);
5701                 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5702         } else {
5703                 context_desc->seqnum_seed = 0;
5704         }
5705 }
5706
5707 static int igb_tso(struct igb_ring *tx_ring,
5708                    struct igb_tx_buffer *first,
5709                    u8 *hdr_len)
5710 {
5711         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5712         struct sk_buff *skb = first->skb;
5713         union {
5714                 struct iphdr *v4;
5715                 struct ipv6hdr *v6;
5716                 unsigned char *hdr;
5717         } ip;
5718         union {
5719                 struct tcphdr *tcp;
5720                 struct udphdr *udp;
5721                 unsigned char *hdr;
5722         } l4;
5723         u32 paylen, l4_offset;
5724         int err;
5725
5726         if (skb->ip_summed != CHECKSUM_PARTIAL)
5727                 return 0;
5728
5729         if (!skb_is_gso(skb))
5730                 return 0;
5731
5732         err = skb_cow_head(skb, 0);
5733         if (err < 0)
5734                 return err;
5735
5736         ip.hdr = skb_network_header(skb);
5737         l4.hdr = skb_checksum_start(skb);
5738
5739         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5740         type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
5741                       E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
5742
5743         /* initialize outer IP header fields */
5744         if (ip.v4->version == 4) {
5745                 unsigned char *csum_start = skb_checksum_start(skb);
5746                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5747
5748                 /* IP header will have to cancel out any data that
5749                  * is not a part of the outer IP header
5750                  */
5751                 ip.v4->check = csum_fold(csum_partial(trans_start,
5752                                                       csum_start - trans_start,
5753                                                       0));
5754                 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5755
5756                 ip.v4->tot_len = 0;
5757                 first->tx_flags |= IGB_TX_FLAGS_TSO |
5758                                    IGB_TX_FLAGS_CSUM |
5759                                    IGB_TX_FLAGS_IPV4;
5760         } else {
5761                 ip.v6->payload_len = 0;
5762                 first->tx_flags |= IGB_TX_FLAGS_TSO |
5763                                    IGB_TX_FLAGS_CSUM;
5764         }
5765
5766         /* determine offset of inner transport header */
5767         l4_offset = l4.hdr - skb->data;
5768
5769         /* remove payload length from inner checksum */
5770         paylen = skb->len - l4_offset;
5771         if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
5772                 /* compute length of segmentation header */
5773                 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
5774                 csum_replace_by_diff(&l4.tcp->check,
5775                         (__force __wsum)htonl(paylen));
5776         } else {
5777                 /* compute length of segmentation header */
5778                 *hdr_len = sizeof(*l4.udp) + l4_offset;
5779                 csum_replace_by_diff(&l4.udp->check,
5780                                      (__force __wsum)htonl(paylen));
5781         }
5782
5783         /* update gso size and bytecount with header size */
5784         first->gso_segs = skb_shinfo(skb)->gso_segs;
5785         first->bytecount += (first->gso_segs - 1) * *hdr_len;
5786
5787         /* MSS L4LEN IDX */
5788         mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5789         mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5790
5791         /* VLAN MACLEN IPLEN */
5792         vlan_macip_lens = l4.hdr - ip.hdr;
5793         vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5794         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5795
5796         igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
5797                         type_tucmd, mss_l4len_idx);
5798
5799         return 1;
5800 }
5801
5802 static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb)
5803 {
5804         unsigned int offset = 0;
5805
5806         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
5807
5808         return offset == skb_checksum_start_offset(skb);
5809 }
5810
5811 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5812 {
5813         struct sk_buff *skb = first->skb;
5814         u32 vlan_macip_lens = 0;
5815         u32 type_tucmd = 0;
5816
5817         if (skb->ip_summed != CHECKSUM_PARTIAL) {
5818 csum_failed:
5819                 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
5820                     !tx_ring->launchtime_enable)
5821                         return;
5822                 goto no_csum;
5823         }
5824
5825         switch (skb->csum_offset) {
5826         case offsetof(struct tcphdr, check):
5827                 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5828                 /* fall through */
5829         case offsetof(struct udphdr, check):
5830                 break;
5831         case offsetof(struct sctphdr, checksum):
5832                 /* validate that this is actually an SCTP request */
5833                 if (((first->protocol == htons(ETH_P_IP)) &&
5834                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
5835                     ((first->protocol == htons(ETH_P_IPV6)) &&
5836                      igb_ipv6_csum_is_sctp(skb))) {
5837                         type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
5838                         break;
5839                 }
5840                 /* fall through */
5841         default:
5842                 skb_checksum_help(skb);
5843                 goto csum_failed;
5844         }
5845
5846         /* update TX checksum flag */
5847         first->tx_flags |= IGB_TX_FLAGS_CSUM;
5848         vlan_macip_lens = skb_checksum_start_offset(skb) -
5849                           skb_network_offset(skb);
5850 no_csum:
5851         vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
5852         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5853
5854         igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
5855 }
5856
5857 #define IGB_SET_FLAG(_input, _flag, _result) \
5858         ((_flag <= _result) ? \
5859          ((u32)(_input & _flag) * (_result / _flag)) : \
5860          ((u32)(_input & _flag) / (_flag / _result)))
5861
5862 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
5863 {
5864         /* set type for advanced descriptor with frame checksum insertion */
5865         u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
5866                        E1000_ADVTXD_DCMD_DEXT |
5867                        E1000_ADVTXD_DCMD_IFCS;
5868
5869         /* set HW vlan bit if vlan is present */
5870         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
5871                                  (E1000_ADVTXD_DCMD_VLE));
5872
5873         /* set segmentation bits for TSO */
5874         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
5875                                  (E1000_ADVTXD_DCMD_TSE));
5876
5877         /* set timestamp bit if present */
5878         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
5879                                  (E1000_ADVTXD_MAC_TSTAMP));
5880
5881         /* insert frame checksum */
5882         cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
5883
5884         return cmd_type;
5885 }
5886
5887 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
5888                                  union e1000_adv_tx_desc *tx_desc,
5889                                  u32 tx_flags, unsigned int paylen)
5890 {
5891         u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
5892
5893         /* 82575 requires a unique index per ring */
5894         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5895                 olinfo_status |= tx_ring->reg_idx << 4;
5896
5897         /* insert L4 checksum */
5898         olinfo_status |= IGB_SET_FLAG(tx_flags,
5899                                       IGB_TX_FLAGS_CSUM,
5900                                       (E1000_TXD_POPTS_TXSM << 8));
5901
5902         /* insert IPv4 checksum */
5903         olinfo_status |= IGB_SET_FLAG(tx_flags,
5904                                       IGB_TX_FLAGS_IPV4,
5905                                       (E1000_TXD_POPTS_IXSM << 8));
5906
5907         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5908 }
5909
5910 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5911 {
5912         struct net_device *netdev = tx_ring->netdev;
5913
5914         netif_stop_subqueue(netdev, tx_ring->queue_index);
5915
5916         /* Herbert's original patch had:
5917          *  smp_mb__after_netif_stop_queue();
5918          * but since that doesn't exist yet, just open code it.
5919          */
5920         smp_mb();
5921
5922         /* We need to check again in a case another CPU has just
5923          * made room available.
5924          */
5925         if (igb_desc_unused(tx_ring) < size)
5926                 return -EBUSY;
5927
5928         /* A reprieve! */
5929         netif_wake_subqueue(netdev, tx_ring->queue_index);
5930
5931         u64_stats_update_begin(&tx_ring->tx_syncp2);
5932         tx_ring->tx_stats.restart_queue2++;
5933         u64_stats_update_end(&tx_ring->tx_syncp2);
5934
5935         return 0;
5936 }
5937
5938 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5939 {
5940         if (igb_desc_unused(tx_ring) >= size)
5941                 return 0;
5942         return __igb_maybe_stop_tx(tx_ring, size);
5943 }
5944
5945 static int igb_tx_map(struct igb_ring *tx_ring,
5946                       struct igb_tx_buffer *first,
5947                       const u8 hdr_len)
5948 {
5949         struct sk_buff *skb = first->skb;
5950         struct igb_tx_buffer *tx_buffer;
5951         union e1000_adv_tx_desc *tx_desc;
5952         skb_frag_t *frag;
5953         dma_addr_t dma;
5954         unsigned int data_len, size;
5955         u32 tx_flags = first->tx_flags;
5956         u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
5957         u16 i = tx_ring->next_to_use;
5958
5959         tx_desc = IGB_TX_DESC(tx_ring, i);
5960
5961         igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
5962
5963         size = skb_headlen(skb);
5964         data_len = skb->data_len;
5965
5966         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5967
5968         tx_buffer = first;
5969
5970         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
5971                 if (dma_mapping_error(tx_ring->dev, dma))
5972                         goto dma_error;
5973
5974                 /* record length, and DMA address */
5975                 dma_unmap_len_set(tx_buffer, len, size);
5976                 dma_unmap_addr_set(tx_buffer, dma, dma);
5977
5978                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5979
5980                 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
5981                         tx_desc->read.cmd_type_len =
5982                                 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
5983
5984                         i++;
5985                         tx_desc++;
5986                         if (i == tx_ring->count) {
5987                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
5988                                 i = 0;
5989                         }
5990                         tx_desc->read.olinfo_status = 0;
5991
5992                         dma += IGB_MAX_DATA_PER_TXD;
5993                         size -= IGB_MAX_DATA_PER_TXD;
5994
5995                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
5996                 }
5997
5998                 if (likely(!data_len))
5999                         break;
6000
6001                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6002
6003                 i++;
6004                 tx_desc++;
6005                 if (i == tx_ring->count) {
6006                         tx_desc = IGB_TX_DESC(tx_ring, 0);
6007                         i = 0;
6008                 }
6009                 tx_desc->read.olinfo_status = 0;
6010
6011                 size = skb_frag_size(frag);
6012                 data_len -= size;
6013
6014                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6015                                        size, DMA_TO_DEVICE);
6016
6017                 tx_buffer = &tx_ring->tx_buffer_info[i];
6018         }
6019
6020         /* write last descriptor with RS and EOP bits */
6021         cmd_type |= size | IGB_TXD_DCMD;
6022         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6023
6024         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6025
6026         /* set the timestamp */
6027         first->time_stamp = jiffies;
6028
6029         skb_tx_timestamp(skb);
6030
6031         /* Force memory writes to complete before letting h/w know there
6032          * are new descriptors to fetch.  (Only applicable for weak-ordered
6033          * memory model archs, such as IA-64).
6034          *
6035          * We also need this memory barrier to make certain all of the
6036          * status bits have been updated before next_to_watch is written.
6037          */
6038         dma_wmb();
6039
6040         /* set next_to_watch value indicating a packet is present */
6041         first->next_to_watch = tx_desc;
6042
6043         i++;
6044         if (i == tx_ring->count)
6045                 i = 0;
6046
6047         tx_ring->next_to_use = i;
6048
6049         /* Make sure there is space in the ring for the next send. */
6050         igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6051
6052         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
6053                 writel(i, tx_ring->tail);
6054         }
6055         return 0;
6056
6057 dma_error:
6058         dev_err(tx_ring->dev, "TX DMA map failed\n");
6059         tx_buffer = &tx_ring->tx_buffer_info[i];
6060
6061         /* clear dma mappings for failed tx_buffer_info map */
6062         while (tx_buffer != first) {
6063                 if (dma_unmap_len(tx_buffer, len))
6064                         dma_unmap_page(tx_ring->dev,
6065                                        dma_unmap_addr(tx_buffer, dma),
6066                                        dma_unmap_len(tx_buffer, len),
6067                                        DMA_TO_DEVICE);
6068                 dma_unmap_len_set(tx_buffer, len, 0);
6069
6070                 if (i-- == 0)
6071                         i += tx_ring->count;
6072                 tx_buffer = &tx_ring->tx_buffer_info[i];
6073         }
6074
6075         if (dma_unmap_len(tx_buffer, len))
6076                 dma_unmap_single(tx_ring->dev,
6077                                  dma_unmap_addr(tx_buffer, dma),
6078                                  dma_unmap_len(tx_buffer, len),
6079                                  DMA_TO_DEVICE);
6080         dma_unmap_len_set(tx_buffer, len, 0);
6081
6082         dev_kfree_skb_any(tx_buffer->skb);
6083         tx_buffer->skb = NULL;
6084
6085         tx_ring->next_to_use = i;
6086
6087         return -1;
6088 }
6089
6090 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6091                                 struct igb_ring *tx_ring)
6092 {
6093         struct igb_tx_buffer *first;
6094         int tso;
6095         u32 tx_flags = 0;
6096         unsigned short f;
6097         u16 count = TXD_USE_COUNT(skb_headlen(skb));
6098         __be16 protocol = vlan_get_protocol(skb);
6099         u8 hdr_len = 0;
6100
6101         /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6102          *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6103          *       + 2 desc gap to keep tail from touching head,
6104          *       + 1 desc for context descriptor,
6105          * otherwise try next time
6106          */
6107         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6108                 count += TXD_USE_COUNT(skb_frag_size(
6109                                                 &skb_shinfo(skb)->frags[f]));
6110
6111         if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6112                 /* this is a hard error */
6113                 return NETDEV_TX_BUSY;
6114         }
6115
6116         /* record the location of the first descriptor for this packet */
6117         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6118         first->skb = skb;
6119         first->bytecount = skb->len;
6120         first->gso_segs = 1;
6121
6122         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6123                 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6124
6125                 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6126                     !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6127                                            &adapter->state)) {
6128                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6129                         tx_flags |= IGB_TX_FLAGS_TSTAMP;
6130
6131                         adapter->ptp_tx_skb = skb_get(skb);
6132                         adapter->ptp_tx_start = jiffies;
6133                         if (adapter->hw.mac.type == e1000_82576)
6134                                 schedule_work(&adapter->ptp_tx_work);
6135                 } else {
6136                         adapter->tx_hwtstamp_skipped++;
6137                 }
6138         }
6139
6140         if (skb_vlan_tag_present(skb)) {
6141                 tx_flags |= IGB_TX_FLAGS_VLAN;
6142                 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6143         }
6144
6145         /* record initial flags and protocol */
6146         first->tx_flags = tx_flags;
6147         first->protocol = protocol;
6148
6149         tso = igb_tso(tx_ring, first, &hdr_len);
6150         if (tso < 0)
6151                 goto out_drop;
6152         else if (!tso)
6153                 igb_tx_csum(tx_ring, first);
6154
6155         if (igb_tx_map(tx_ring, first, hdr_len))
6156                 goto cleanup_tx_tstamp;
6157
6158         return NETDEV_TX_OK;
6159
6160 out_drop:
6161         dev_kfree_skb_any(first->skb);
6162         first->skb = NULL;
6163 cleanup_tx_tstamp:
6164         if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6165                 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6166
6167                 dev_kfree_skb_any(adapter->ptp_tx_skb);
6168                 adapter->ptp_tx_skb = NULL;
6169                 if (adapter->hw.mac.type == e1000_82576)
6170                         cancel_work_sync(&adapter->ptp_tx_work);
6171                 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6172         }
6173
6174         return NETDEV_TX_OK;
6175 }
6176
6177 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6178                                                     struct sk_buff *skb)
6179 {
6180         unsigned int r_idx = skb->queue_mapping;
6181
6182         if (r_idx >= adapter->num_tx_queues)
6183                 r_idx = r_idx % adapter->num_tx_queues;
6184
6185         return adapter->tx_ring[r_idx];
6186 }
6187
6188 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6189                                   struct net_device *netdev)
6190 {
6191         struct igb_adapter *adapter = netdev_priv(netdev);
6192
6193         /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6194          * in order to meet this minimum size requirement.
6195          */
6196         if (skb_put_padto(skb, 17))
6197                 return NETDEV_TX_OK;
6198
6199         return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6200 }
6201
6202 /**
6203  *  igb_tx_timeout - Respond to a Tx Hang
6204  *  @netdev: network interface device structure
6205  **/
6206 static void igb_tx_timeout(struct net_device *netdev, unsigned int txqueue)
6207 {
6208         struct igb_adapter *adapter = netdev_priv(netdev);
6209         struct e1000_hw *hw = &adapter->hw;
6210
6211         /* Do the reset outside of interrupt context */
6212         adapter->tx_timeout_count++;
6213
6214         if (hw->mac.type >= e1000_82580)
6215                 hw->dev_spec._82575.global_device_reset = true;
6216
6217         schedule_work(&adapter->reset_task);
6218         wr32(E1000_EICS,
6219              (adapter->eims_enable_mask & ~adapter->eims_other));
6220 }
6221
6222 static void igb_reset_task(struct work_struct *work)
6223 {
6224         struct igb_adapter *adapter;
6225         adapter = container_of(work, struct igb_adapter, reset_task);
6226
6227         igb_dump(adapter);
6228         netdev_err(adapter->netdev, "Reset adapter\n");
6229         igb_reinit_locked(adapter);
6230 }
6231
6232 /**
6233  *  igb_get_stats64 - Get System Network Statistics
6234  *  @netdev: network interface device structure
6235  *  @stats: rtnl_link_stats64 pointer
6236  **/
6237 static void igb_get_stats64(struct net_device *netdev,
6238                             struct rtnl_link_stats64 *stats)
6239 {
6240         struct igb_adapter *adapter = netdev_priv(netdev);
6241
6242         spin_lock(&adapter->stats64_lock);
6243         igb_update_stats(adapter);
6244         memcpy(stats, &adapter->stats64, sizeof(*stats));
6245         spin_unlock(&adapter->stats64_lock);
6246 }
6247
6248 /**
6249  *  igb_change_mtu - Change the Maximum Transfer Unit
6250  *  @netdev: network interface device structure
6251  *  @new_mtu: new value for maximum frame size
6252  *
6253  *  Returns 0 on success, negative on failure
6254  **/
6255 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6256 {
6257         struct igb_adapter *adapter = netdev_priv(netdev);
6258         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
6259
6260         /* adjust max frame to be at least the size of a standard frame */
6261         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6262                 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6263
6264         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6265                 usleep_range(1000, 2000);
6266
6267         /* igb_down has a dependency on max_frame_size */
6268         adapter->max_frame_size = max_frame;
6269
6270         if (netif_running(netdev))
6271                 igb_down(adapter);
6272
6273         netdev_dbg(netdev, "changing MTU from %d to %d\n",
6274                    netdev->mtu, new_mtu);
6275         netdev->mtu = new_mtu;
6276
6277         if (netif_running(netdev))
6278                 igb_up(adapter);
6279         else
6280                 igb_reset(adapter);
6281
6282         clear_bit(__IGB_RESETTING, &adapter->state);
6283
6284         return 0;
6285 }
6286
6287 /**
6288  *  igb_update_stats - Update the board statistics counters
6289  *  @adapter: board private structure
6290  **/
6291 void igb_update_stats(struct igb_adapter *adapter)
6292 {
6293         struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6294         struct e1000_hw *hw = &adapter->hw;
6295         struct pci_dev *pdev = adapter->pdev;
6296         u32 reg, mpc;
6297         int i;
6298         u64 bytes, packets;
6299         unsigned int start;
6300         u64 _bytes, _packets;
6301
6302         /* Prevent stats update while adapter is being reset, or if the pci
6303          * connection is down.
6304          */
6305         if (adapter->link_speed == 0)
6306                 return;
6307         if (pci_channel_offline(pdev))
6308                 return;
6309
6310         bytes = 0;
6311         packets = 0;
6312
6313         rcu_read_lock();
6314         for (i = 0; i < adapter->num_rx_queues; i++) {
6315                 struct igb_ring *ring = adapter->rx_ring[i];
6316                 u32 rqdpc = rd32(E1000_RQDPC(i));
6317                 if (hw->mac.type >= e1000_i210)
6318                         wr32(E1000_RQDPC(i), 0);
6319
6320                 if (rqdpc) {
6321                         ring->rx_stats.drops += rqdpc;
6322                         net_stats->rx_fifo_errors += rqdpc;
6323                 }
6324
6325                 do {
6326                         start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
6327                         _bytes = ring->rx_stats.bytes;
6328                         _packets = ring->rx_stats.packets;
6329                 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
6330                 bytes += _bytes;
6331                 packets += _packets;
6332         }
6333
6334         net_stats->rx_bytes = bytes;
6335         net_stats->rx_packets = packets;
6336
6337         bytes = 0;
6338         packets = 0;
6339         for (i = 0; i < adapter->num_tx_queues; i++) {
6340                 struct igb_ring *ring = adapter->tx_ring[i];
6341                 do {
6342                         start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
6343                         _bytes = ring->tx_stats.bytes;
6344                         _packets = ring->tx_stats.packets;
6345                 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
6346                 bytes += _bytes;
6347                 packets += _packets;
6348         }
6349         net_stats->tx_bytes = bytes;
6350         net_stats->tx_packets = packets;
6351         rcu_read_unlock();
6352
6353         /* read stats registers */
6354         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6355         adapter->stats.gprc += rd32(E1000_GPRC);
6356         adapter->stats.gorc += rd32(E1000_GORCL);
6357         rd32(E1000_GORCH); /* clear GORCL */
6358         adapter->stats.bprc += rd32(E1000_BPRC);
6359         adapter->stats.mprc += rd32(E1000_MPRC);
6360         adapter->stats.roc += rd32(E1000_ROC);
6361
6362         adapter->stats.prc64 += rd32(E1000_PRC64);
6363         adapter->stats.prc127 += rd32(E1000_PRC127);
6364         adapter->stats.prc255 += rd32(E1000_PRC255);
6365         adapter->stats.prc511 += rd32(E1000_PRC511);
6366         adapter->stats.prc1023 += rd32(E1000_PRC1023);
6367         adapter->stats.prc1522 += rd32(E1000_PRC1522);
6368         adapter->stats.symerrs += rd32(E1000_SYMERRS);
6369         adapter->stats.sec += rd32(E1000_SEC);
6370
6371         mpc = rd32(E1000_MPC);
6372         adapter->stats.mpc += mpc;
6373         net_stats->rx_fifo_errors += mpc;
6374         adapter->stats.scc += rd32(E1000_SCC);
6375         adapter->stats.ecol += rd32(E1000_ECOL);
6376         adapter->stats.mcc += rd32(E1000_MCC);
6377         adapter->stats.latecol += rd32(E1000_LATECOL);
6378         adapter->stats.dc += rd32(E1000_DC);
6379         adapter->stats.rlec += rd32(E1000_RLEC);
6380         adapter->stats.xonrxc += rd32(E1000_XONRXC);
6381         adapter->stats.xontxc += rd32(E1000_XONTXC);
6382         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6383         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6384         adapter->stats.fcruc += rd32(E1000_FCRUC);
6385         adapter->stats.gptc += rd32(E1000_GPTC);
6386         adapter->stats.gotc += rd32(E1000_GOTCL);
6387         rd32(E1000_GOTCH); /* clear GOTCL */
6388         adapter->stats.rnbc += rd32(E1000_RNBC);
6389         adapter->stats.ruc += rd32(E1000_RUC);
6390         adapter->stats.rfc += rd32(E1000_RFC);
6391         adapter->stats.rjc += rd32(E1000_RJC);
6392         adapter->stats.tor += rd32(E1000_TORH);
6393         adapter->stats.tot += rd32(E1000_TOTH);
6394         adapter->stats.tpr += rd32(E1000_TPR);
6395
6396         adapter->stats.ptc64 += rd32(E1000_PTC64);
6397         adapter->stats.ptc127 += rd32(E1000_PTC127);
6398         adapter->stats.ptc255 += rd32(E1000_PTC255);
6399         adapter->stats.ptc511 += rd32(E1000_PTC511);
6400         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6401         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6402
6403         adapter->stats.mptc += rd32(E1000_MPTC);
6404         adapter->stats.bptc += rd32(E1000_BPTC);
6405
6406         adapter->stats.tpt += rd32(E1000_TPT);
6407         adapter->stats.colc += rd32(E1000_COLC);
6408
6409         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6410         /* read internal phy specific stats */
6411         reg = rd32(E1000_CTRL_EXT);
6412         if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6413                 adapter->stats.rxerrc += rd32(E1000_RXERRC);
6414
6415                 /* this stat has invalid values on i210/i211 */
6416                 if ((hw->mac.type != e1000_i210) &&
6417                     (hw->mac.type != e1000_i211))
6418                         adapter->stats.tncrs += rd32(E1000_TNCRS);
6419         }
6420
6421         adapter->stats.tsctc += rd32(E1000_TSCTC);
6422         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6423
6424         adapter->stats.iac += rd32(E1000_IAC);
6425         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6426         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6427         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6428         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6429         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6430         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6431         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6432         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6433
6434         /* Fill out the OS statistics structure */
6435         net_stats->multicast = adapter->stats.mprc;
6436         net_stats->collisions = adapter->stats.colc;
6437
6438         /* Rx Errors */
6439
6440         /* RLEC on some newer hardware can be incorrect so build
6441          * our own version based on RUC and ROC
6442          */
6443         net_stats->rx_errors = adapter->stats.rxerrc +
6444                 adapter->stats.crcerrs + adapter->stats.algnerrc +
6445                 adapter->stats.ruc + adapter->stats.roc +
6446                 adapter->stats.cexterr;
6447         net_stats->rx_length_errors = adapter->stats.ruc +
6448                                       adapter->stats.roc;
6449         net_stats->rx_crc_errors = adapter->stats.crcerrs;
6450         net_stats->rx_frame_errors = adapter->stats.algnerrc;
6451         net_stats->rx_missed_errors = adapter->stats.mpc;
6452
6453         /* Tx Errors */
6454         net_stats->tx_errors = adapter->stats.ecol +
6455                                adapter->stats.latecol;
6456         net_stats->tx_aborted_errors = adapter->stats.ecol;
6457         net_stats->tx_window_errors = adapter->stats.latecol;
6458         net_stats->tx_carrier_errors = adapter->stats.tncrs;
6459
6460         /* Tx Dropped needs to be maintained elsewhere */
6461
6462         /* Management Stats */
6463         adapter->stats.mgptc += rd32(E1000_MGTPTC);
6464         adapter->stats.mgprc += rd32(E1000_MGTPRC);
6465         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6466
6467         /* OS2BMC Stats */
6468         reg = rd32(E1000_MANC);
6469         if (reg & E1000_MANC_EN_BMC2OS) {
6470                 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6471                 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6472                 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6473                 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6474         }
6475 }
6476
6477 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6478 {
6479         struct e1000_hw *hw = &adapter->hw;
6480         struct ptp_clock_event event;
6481         struct timespec64 ts;
6482         u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
6483
6484         if (tsicr & TSINTR_SYS_WRAP) {
6485                 event.type = PTP_CLOCK_PPS;
6486                 if (adapter->ptp_caps.pps)
6487                         ptp_clock_event(adapter->ptp_clock, &event);
6488                 ack |= TSINTR_SYS_WRAP;
6489         }
6490
6491         if (tsicr & E1000_TSICR_TXTS) {
6492                 /* retrieve hardware timestamp */
6493                 schedule_work(&adapter->ptp_tx_work);
6494                 ack |= E1000_TSICR_TXTS;
6495         }
6496
6497         if (tsicr & TSINTR_TT0) {
6498                 spin_lock(&adapter->tmreg_lock);
6499                 ts = timespec64_add(adapter->perout[0].start,
6500                                     adapter->perout[0].period);
6501                 /* u32 conversion of tv_sec is safe until y2106 */
6502                 wr32(E1000_TRGTTIML0, ts.tv_nsec);
6503                 wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
6504                 tsauxc = rd32(E1000_TSAUXC);
6505                 tsauxc |= TSAUXC_EN_TT0;
6506                 wr32(E1000_TSAUXC, tsauxc);
6507                 adapter->perout[0].start = ts;
6508                 spin_unlock(&adapter->tmreg_lock);
6509                 ack |= TSINTR_TT0;
6510         }
6511
6512         if (tsicr & TSINTR_TT1) {
6513                 spin_lock(&adapter->tmreg_lock);
6514                 ts = timespec64_add(adapter->perout[1].start,
6515                                     adapter->perout[1].period);
6516                 wr32(E1000_TRGTTIML1, ts.tv_nsec);
6517                 wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
6518                 tsauxc = rd32(E1000_TSAUXC);
6519                 tsauxc |= TSAUXC_EN_TT1;
6520                 wr32(E1000_TSAUXC, tsauxc);
6521                 adapter->perout[1].start = ts;
6522                 spin_unlock(&adapter->tmreg_lock);
6523                 ack |= TSINTR_TT1;
6524         }
6525
6526         if (tsicr & TSINTR_AUTT0) {
6527                 nsec = rd32(E1000_AUXSTMPL0);
6528                 sec  = rd32(E1000_AUXSTMPH0);
6529                 event.type = PTP_CLOCK_EXTTS;
6530                 event.index = 0;
6531                 event.timestamp = sec * 1000000000ULL + nsec;
6532                 ptp_clock_event(adapter->ptp_clock, &event);
6533                 ack |= TSINTR_AUTT0;
6534         }
6535
6536         if (tsicr & TSINTR_AUTT1) {
6537                 nsec = rd32(E1000_AUXSTMPL1);
6538                 sec  = rd32(E1000_AUXSTMPH1);
6539                 event.type = PTP_CLOCK_EXTTS;
6540                 event.index = 1;
6541                 event.timestamp = sec * 1000000000ULL + nsec;
6542                 ptp_clock_event(adapter->ptp_clock, &event);
6543                 ack |= TSINTR_AUTT1;
6544         }
6545
6546         /* acknowledge the interrupts */
6547         wr32(E1000_TSICR, ack);
6548 }
6549
6550 static irqreturn_t igb_msix_other(int irq, void *data)
6551 {
6552         struct igb_adapter *adapter = data;
6553         struct e1000_hw *hw = &adapter->hw;
6554         u32 icr = rd32(E1000_ICR);
6555         /* reading ICR causes bit 31 of EICR to be cleared */
6556
6557         if (icr & E1000_ICR_DRSTA)
6558                 schedule_work(&adapter->reset_task);
6559
6560         if (icr & E1000_ICR_DOUTSYNC) {
6561                 /* HW is reporting DMA is out of sync */
6562                 adapter->stats.doosync++;
6563                 /* The DMA Out of Sync is also indication of a spoof event
6564                  * in IOV mode. Check the Wrong VM Behavior register to
6565                  * see if it is really a spoof event.
6566                  */
6567                 igb_check_wvbr(adapter);
6568         }
6569
6570         /* Check for a mailbox event */
6571         if (icr & E1000_ICR_VMMB)
6572                 igb_msg_task(adapter);
6573
6574         if (icr & E1000_ICR_LSC) {
6575                 hw->mac.get_link_status = 1;
6576                 /* guard against interrupt when we're going down */
6577                 if (!test_bit(__IGB_DOWN, &adapter->state))
6578                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
6579         }
6580
6581         if (icr & E1000_ICR_TS)
6582                 igb_tsync_interrupt(adapter);
6583
6584         wr32(E1000_EIMS, adapter->eims_other);
6585
6586         return IRQ_HANDLED;
6587 }
6588
6589 static void igb_write_itr(struct igb_q_vector *q_vector)
6590 {
6591         struct igb_adapter *adapter = q_vector->adapter;
6592         u32 itr_val = q_vector->itr_val & 0x7FFC;
6593
6594         if (!q_vector->set_itr)
6595                 return;
6596
6597         if (!itr_val)
6598                 itr_val = 0x4;
6599
6600         if (adapter->hw.mac.type == e1000_82575)
6601                 itr_val |= itr_val << 16;
6602         else
6603                 itr_val |= E1000_EITR_CNT_IGNR;
6604
6605         writel(itr_val, q_vector->itr_register);
6606         q_vector->set_itr = 0;
6607 }
6608
6609 static irqreturn_t igb_msix_ring(int irq, void *data)
6610 {
6611         struct igb_q_vector *q_vector = data;
6612
6613         /* Write the ITR value calculated from the previous interrupt. */
6614         igb_write_itr(q_vector);
6615
6616         napi_schedule(&q_vector->napi);
6617
6618         return IRQ_HANDLED;
6619 }
6620
6621 #ifdef CONFIG_IGB_DCA
6622 static void igb_update_tx_dca(struct igb_adapter *adapter,
6623                               struct igb_ring *tx_ring,
6624                               int cpu)
6625 {
6626         struct e1000_hw *hw = &adapter->hw;
6627         u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
6628
6629         if (hw->mac.type != e1000_82575)
6630                 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
6631
6632         /* We can enable relaxed ordering for reads, but not writes when
6633          * DCA is enabled.  This is due to a known issue in some chipsets
6634          * which will cause the DCA tag to be cleared.
6635          */
6636         txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
6637                   E1000_DCA_TXCTRL_DATA_RRO_EN |
6638                   E1000_DCA_TXCTRL_DESC_DCA_EN;
6639
6640         wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
6641 }
6642
6643 static void igb_update_rx_dca(struct igb_adapter *adapter,
6644                               struct igb_ring *rx_ring,
6645                               int cpu)
6646 {
6647         struct e1000_hw *hw = &adapter->hw;
6648         u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
6649
6650         if (hw->mac.type != e1000_82575)
6651                 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
6652
6653         /* We can enable relaxed ordering for reads, but not writes when
6654          * DCA is enabled.  This is due to a known issue in some chipsets
6655          * which will cause the DCA tag to be cleared.
6656          */
6657         rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
6658                   E1000_DCA_RXCTRL_DESC_DCA_EN;
6659
6660         wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
6661 }
6662
6663 static void igb_update_dca(struct igb_q_vector *q_vector)
6664 {
6665         struct igb_adapter *adapter = q_vector->adapter;
6666         int cpu = get_cpu();
6667
6668         if (q_vector->cpu == cpu)
6669                 goto out_no_update;
6670
6671         if (q_vector->tx.ring)
6672                 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
6673
6674         if (q_vector->rx.ring)
6675                 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
6676
6677         q_vector->cpu = cpu;
6678 out_no_update:
6679         put_cpu();
6680 }
6681
6682 static void igb_setup_dca(struct igb_adapter *adapter)
6683 {
6684         struct e1000_hw *hw = &adapter->hw;
6685         int i;
6686
6687         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
6688                 return;
6689
6690         /* Always use CB2 mode, difference is masked in the CB driver. */
6691         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
6692
6693         for (i = 0; i < adapter->num_q_vectors; i++) {
6694                 adapter->q_vector[i]->cpu = -1;
6695                 igb_update_dca(adapter->q_vector[i]);
6696         }
6697 }
6698
6699 static int __igb_notify_dca(struct device *dev, void *data)
6700 {
6701         struct net_device *netdev = dev_get_drvdata(dev);
6702         struct igb_adapter *adapter = netdev_priv(netdev);
6703         struct pci_dev *pdev = adapter->pdev;
6704         struct e1000_hw *hw = &adapter->hw;
6705         unsigned long event = *(unsigned long *)data;
6706
6707         switch (event) {
6708         case DCA_PROVIDER_ADD:
6709                 /* if already enabled, don't do it again */
6710                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
6711                         break;
6712                 if (dca_add_requester(dev) == 0) {
6713                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
6714                         dev_info(&pdev->dev, "DCA enabled\n");
6715                         igb_setup_dca(adapter);
6716                         break;
6717                 }
6718                 /* Fall Through - since DCA is disabled. */
6719         case DCA_PROVIDER_REMOVE:
6720                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
6721                         /* without this a class_device is left
6722                          * hanging around in the sysfs model
6723                          */
6724                         dca_remove_requester(dev);
6725                         dev_info(&pdev->dev, "DCA disabled\n");
6726                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
6727                         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
6728                 }
6729                 break;
6730         }
6731
6732         return 0;
6733 }
6734
6735 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
6736                           void *p)
6737 {
6738         int ret_val;
6739
6740         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
6741                                          __igb_notify_dca);
6742
6743         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6744 }
6745 #endif /* CONFIG_IGB_DCA */
6746
6747 #ifdef CONFIG_PCI_IOV
6748 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
6749 {
6750         unsigned char mac_addr[ETH_ALEN];
6751
6752         eth_zero_addr(mac_addr);
6753         igb_set_vf_mac(adapter, vf, mac_addr);
6754
6755         /* By default spoof check is enabled for all VFs */
6756         adapter->vf_data[vf].spoofchk_enabled = true;
6757
6758         /* By default VFs are not trusted */
6759         adapter->vf_data[vf].trusted = false;
6760
6761         return 0;
6762 }
6763
6764 #endif
6765 static void igb_ping_all_vfs(struct igb_adapter *adapter)
6766 {
6767         struct e1000_hw *hw = &adapter->hw;
6768         u32 ping;
6769         int i;
6770
6771         for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
6772                 ping = E1000_PF_CONTROL_MSG;
6773                 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
6774                         ping |= E1000_VT_MSGTYPE_CTS;
6775                 igb_write_mbx(hw, &ping, 1, i);
6776         }
6777 }
6778
6779 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6780 {
6781         struct e1000_hw *hw = &adapter->hw;
6782         u32 vmolr = rd32(E1000_VMOLR(vf));
6783         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6784
6785         vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
6786                             IGB_VF_FLAG_MULTI_PROMISC);
6787         vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6788
6789         if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
6790                 vmolr |= E1000_VMOLR_MPME;
6791                 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
6792                 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
6793         } else {
6794                 /* if we have hashes and we are clearing a multicast promisc
6795                  * flag we need to write the hashes to the MTA as this step
6796                  * was previously skipped
6797                  */
6798                 if (vf_data->num_vf_mc_hashes > 30) {
6799                         vmolr |= E1000_VMOLR_MPME;
6800                 } else if (vf_data->num_vf_mc_hashes) {
6801                         int j;
6802
6803                         vmolr |= E1000_VMOLR_ROMPE;
6804                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6805                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6806                 }
6807         }
6808
6809         wr32(E1000_VMOLR(vf), vmolr);
6810
6811         /* there are flags left unprocessed, likely not supported */
6812         if (*msgbuf & E1000_VT_MSGINFO_MASK)
6813                 return -EINVAL;
6814
6815         return 0;
6816 }
6817
6818 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
6819                                   u32 *msgbuf, u32 vf)
6820 {
6821         int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6822         u16 *hash_list = (u16 *)&msgbuf[1];
6823         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6824         int i;
6825
6826         /* salt away the number of multicast addresses assigned
6827          * to this VF for later use to restore when the PF multi cast
6828          * list changes
6829          */
6830         vf_data->num_vf_mc_hashes = n;
6831
6832         /* only up to 30 hash values supported */
6833         if (n > 30)
6834                 n = 30;
6835
6836         /* store the hashes for later use */
6837         for (i = 0; i < n; i++)
6838                 vf_data->vf_mc_hashes[i] = hash_list[i];
6839
6840         /* Flush and reset the mta with the new values */
6841         igb_set_rx_mode(adapter->netdev);
6842
6843         return 0;
6844 }
6845
6846 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
6847 {
6848         struct e1000_hw *hw = &adapter->hw;
6849         struct vf_data_storage *vf_data;
6850         int i, j;
6851
6852         for (i = 0; i < adapter->vfs_allocated_count; i++) {
6853                 u32 vmolr = rd32(E1000_VMOLR(i));
6854
6855                 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6856
6857                 vf_data = &adapter->vf_data[i];
6858
6859                 if ((vf_data->num_vf_mc_hashes > 30) ||
6860                     (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
6861                         vmolr |= E1000_VMOLR_MPME;
6862                 } else if (vf_data->num_vf_mc_hashes) {
6863                         vmolr |= E1000_VMOLR_ROMPE;
6864                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6865                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6866                 }
6867                 wr32(E1000_VMOLR(i), vmolr);
6868         }
6869 }
6870
6871 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
6872 {
6873         struct e1000_hw *hw = &adapter->hw;
6874         u32 pool_mask, vlvf_mask, i;
6875
6876         /* create mask for VF and other pools */
6877         pool_mask = E1000_VLVF_POOLSEL_MASK;
6878         vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
6879
6880         /* drop PF from pool bits */
6881         pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
6882                              adapter->vfs_allocated_count);
6883
6884         /* Find the vlan filter for this id */
6885         for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
6886                 u32 vlvf = rd32(E1000_VLVF(i));
6887                 u32 vfta_mask, vid, vfta;
6888
6889                 /* remove the vf from the pool */
6890                 if (!(vlvf & vlvf_mask))
6891                         continue;
6892
6893                 /* clear out bit from VLVF */
6894                 vlvf ^= vlvf_mask;
6895
6896                 /* if other pools are present, just remove ourselves */
6897                 if (vlvf & pool_mask)
6898                         goto update_vlvfb;
6899
6900                 /* if PF is present, leave VFTA */
6901                 if (vlvf & E1000_VLVF_POOLSEL_MASK)
6902                         goto update_vlvf;
6903
6904                 vid = vlvf & E1000_VLVF_VLANID_MASK;
6905                 vfta_mask = BIT(vid % 32);
6906
6907                 /* clear bit from VFTA */
6908                 vfta = adapter->shadow_vfta[vid / 32];
6909                 if (vfta & vfta_mask)
6910                         hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
6911 update_vlvf:
6912                 /* clear pool selection enable */
6913                 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6914                         vlvf &= E1000_VLVF_POOLSEL_MASK;
6915                 else
6916                         vlvf = 0;
6917 update_vlvfb:
6918                 /* clear pool bits */
6919                 wr32(E1000_VLVF(i), vlvf);
6920         }
6921 }
6922
6923 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
6924 {
6925         u32 vlvf;
6926         int idx;
6927
6928         /* short cut the special case */
6929         if (vlan == 0)
6930                 return 0;
6931
6932         /* Search for the VLAN id in the VLVF entries */
6933         for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
6934                 vlvf = rd32(E1000_VLVF(idx));
6935                 if ((vlvf & VLAN_VID_MASK) == vlan)
6936                         break;
6937         }
6938
6939         return idx;
6940 }
6941
6942 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
6943 {
6944         struct e1000_hw *hw = &adapter->hw;
6945         u32 bits, pf_id;
6946         int idx;
6947
6948         idx = igb_find_vlvf_entry(hw, vid);
6949         if (!idx)
6950                 return;
6951
6952         /* See if any other pools are set for this VLAN filter
6953          * entry other than the PF.
6954          */
6955         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
6956         bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
6957         bits &= rd32(E1000_VLVF(idx));
6958
6959         /* Disable the filter so this falls into the default pool. */
6960         if (!bits) {
6961                 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6962                         wr32(E1000_VLVF(idx), BIT(pf_id));
6963                 else
6964                         wr32(E1000_VLVF(idx), 0);
6965         }
6966 }
6967
6968 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
6969                            bool add, u32 vf)
6970 {
6971         int pf_id = adapter->vfs_allocated_count;
6972         struct e1000_hw *hw = &adapter->hw;
6973         int err;
6974
6975         /* If VLAN overlaps with one the PF is currently monitoring make
6976          * sure that we are able to allocate a VLVF entry.  This may be
6977          * redundant but it guarantees PF will maintain visibility to
6978          * the VLAN.
6979          */
6980         if (add && test_bit(vid, adapter->active_vlans)) {
6981                 err = igb_vfta_set(hw, vid, pf_id, true, false);
6982                 if (err)
6983                         return err;
6984         }
6985
6986         err = igb_vfta_set(hw, vid, vf, add, false);
6987
6988         if (add && !err)
6989                 return err;
6990
6991         /* If we failed to add the VF VLAN or we are removing the VF VLAN
6992          * we may need to drop the PF pool bit in order to allow us to free
6993          * up the VLVF resources.
6994          */
6995         if (test_bit(vid, adapter->active_vlans) ||
6996             (adapter->flags & IGB_FLAG_VLAN_PROMISC))
6997                 igb_update_pf_vlvf(adapter, vid);
6998
6999         return err;
7000 }
7001
7002 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7003 {
7004         struct e1000_hw *hw = &adapter->hw;
7005
7006         if (vid)
7007                 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7008         else
7009                 wr32(E1000_VMVIR(vf), 0);
7010 }
7011
7012 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7013                                 u16 vlan, u8 qos)
7014 {
7015         int err;
7016
7017         err = igb_set_vf_vlan(adapter, vlan, true, vf);
7018         if (err)
7019                 return err;
7020
7021         igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7022         igb_set_vmolr(adapter, vf, !vlan);
7023
7024         /* revoke access to previous VLAN */
7025         if (vlan != adapter->vf_data[vf].pf_vlan)
7026                 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7027                                 false, vf);
7028
7029         adapter->vf_data[vf].pf_vlan = vlan;
7030         adapter->vf_data[vf].pf_qos = qos;
7031         igb_set_vf_vlan_strip(adapter, vf, true);
7032         dev_info(&adapter->pdev->dev,
7033                  "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7034         if (test_bit(__IGB_DOWN, &adapter->state)) {
7035                 dev_warn(&adapter->pdev->dev,
7036                          "The VF VLAN has been set, but the PF device is not up.\n");
7037                 dev_warn(&adapter->pdev->dev,
7038                          "Bring the PF device up before attempting to use the VF device.\n");
7039         }
7040
7041         return err;
7042 }
7043
7044 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7045 {
7046         /* Restore tagless access via VLAN 0 */
7047         igb_set_vf_vlan(adapter, 0, true, vf);
7048
7049         igb_set_vmvir(adapter, 0, vf);
7050         igb_set_vmolr(adapter, vf, true);
7051
7052         /* Remove any PF assigned VLAN */
7053         if (adapter->vf_data[vf].pf_vlan)
7054                 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7055                                 false, vf);
7056
7057         adapter->vf_data[vf].pf_vlan = 0;
7058         adapter->vf_data[vf].pf_qos = 0;
7059         igb_set_vf_vlan_strip(adapter, vf, false);
7060
7061         return 0;
7062 }
7063
7064 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7065                                u16 vlan, u8 qos, __be16 vlan_proto)
7066 {
7067         struct igb_adapter *adapter = netdev_priv(netdev);
7068
7069         if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7070                 return -EINVAL;
7071
7072         if (vlan_proto != htons(ETH_P_8021Q))
7073                 return -EPROTONOSUPPORT;
7074
7075         return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7076                                igb_disable_port_vlan(adapter, vf);
7077 }
7078
7079 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7080 {
7081         int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7082         int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7083         int ret;
7084
7085         if (adapter->vf_data[vf].pf_vlan)
7086                 return -1;
7087
7088         /* VLAN 0 is a special case, don't allow it to be removed */
7089         if (!vid && !add)
7090                 return 0;
7091
7092         ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7093         if (!ret)
7094                 igb_set_vf_vlan_strip(adapter, vf, !!vid);
7095         return ret;
7096 }
7097
7098 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7099 {
7100         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7101
7102         /* clear flags - except flag that indicates PF has set the MAC */
7103         vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7104         vf_data->last_nack = jiffies;
7105
7106         /* reset vlans for device */
7107         igb_clear_vf_vfta(adapter, vf);
7108         igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7109         igb_set_vmvir(adapter, vf_data->pf_vlan |
7110                                (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7111         igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7112         igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7113
7114         /* reset multicast table array for vf */
7115         adapter->vf_data[vf].num_vf_mc_hashes = 0;
7116
7117         /* Flush and reset the mta with the new values */
7118         igb_set_rx_mode(adapter->netdev);
7119 }
7120
7121 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7122 {
7123         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7124
7125         /* clear mac address as we were hotplug removed/added */
7126         if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7127                 eth_zero_addr(vf_mac);
7128
7129         /* process remaining reset events */
7130         igb_vf_reset(adapter, vf);
7131 }
7132
7133 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7134 {
7135         struct e1000_hw *hw = &adapter->hw;
7136         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7137         u32 reg, msgbuf[3];
7138         u8 *addr = (u8 *)(&msgbuf[1]);
7139
7140         /* process all the same items cleared in a function level reset */
7141         igb_vf_reset(adapter, vf);
7142
7143         /* set vf mac address */
7144         igb_set_vf_mac(adapter, vf, vf_mac);
7145
7146         /* enable transmit and receive for vf */
7147         reg = rd32(E1000_VFTE);
7148         wr32(E1000_VFTE, reg | BIT(vf));
7149         reg = rd32(E1000_VFRE);
7150         wr32(E1000_VFRE, reg | BIT(vf));
7151
7152         adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7153
7154         /* reply to reset with ack and vf mac address */
7155         if (!is_zero_ether_addr(vf_mac)) {
7156                 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7157                 memcpy(addr, vf_mac, ETH_ALEN);
7158         } else {
7159                 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7160         }
7161         igb_write_mbx(hw, msgbuf, 3, vf);
7162 }
7163
7164 static void igb_flush_mac_table(struct igb_adapter *adapter)
7165 {
7166         struct e1000_hw *hw = &adapter->hw;
7167         int i;
7168
7169         for (i = 0; i < hw->mac.rar_entry_count; i++) {
7170                 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7171                 memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
7172                 adapter->mac_table[i].queue = 0;
7173                 igb_rar_set_index(adapter, i);
7174         }
7175 }
7176
7177 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7178 {
7179         struct e1000_hw *hw = &adapter->hw;
7180         /* do not count rar entries reserved for VFs MAC addresses */
7181         int rar_entries = hw->mac.rar_entry_count -
7182                           adapter->vfs_allocated_count;
7183         int i, count = 0;
7184
7185         for (i = 0; i < rar_entries; i++) {
7186                 /* do not count default entries */
7187                 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7188                         continue;
7189
7190                 /* do not count "in use" entries for different queues */
7191                 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7192                     (adapter->mac_table[i].queue != queue))
7193                         continue;
7194
7195                 count++;
7196         }
7197
7198         return count;
7199 }
7200
7201 /* Set default MAC address for the PF in the first RAR entry */
7202 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7203 {
7204         struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7205
7206         ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7207         mac_table->queue = adapter->vfs_allocated_count;
7208         mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7209
7210         igb_rar_set_index(adapter, 0);
7211 }
7212
7213 /* If the filter to be added and an already existing filter express
7214  * the same address and address type, it should be possible to only
7215  * override the other configurations, for example the queue to steer
7216  * traffic.
7217  */
7218 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7219                                       const u8 *addr, const u8 flags)
7220 {
7221         if (!(entry->state & IGB_MAC_STATE_IN_USE))
7222                 return true;
7223
7224         if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7225             (flags & IGB_MAC_STATE_SRC_ADDR))
7226                 return false;
7227
7228         if (!ether_addr_equal(addr, entry->addr))
7229                 return false;
7230
7231         return true;
7232 }
7233
7234 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7235  * 'flags' is used to indicate what kind of match is made, match is by
7236  * default for the destination address, if matching by source address
7237  * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7238  */
7239 static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7240                                     const u8 *addr, const u8 queue,
7241                                     const u8 flags)
7242 {
7243         struct e1000_hw *hw = &adapter->hw;
7244         int rar_entries = hw->mac.rar_entry_count -
7245                           adapter->vfs_allocated_count;
7246         int i;
7247
7248         if (is_zero_ether_addr(addr))
7249                 return -EINVAL;
7250
7251         /* Search for the first empty entry in the MAC table.
7252          * Do not touch entries at the end of the table reserved for the VF MAC
7253          * addresses.
7254          */
7255         for (i = 0; i < rar_entries; i++) {
7256                 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7257                                                addr, flags))
7258                         continue;
7259
7260                 ether_addr_copy(adapter->mac_table[i].addr, addr);
7261                 adapter->mac_table[i].queue = queue;
7262                 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7263
7264                 igb_rar_set_index(adapter, i);
7265                 return i;
7266         }
7267
7268         return -ENOSPC;
7269 }
7270
7271 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7272                               const u8 queue)
7273 {
7274         return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7275 }
7276
7277 /* Remove a MAC filter for 'addr' directing matching traffic to
7278  * 'queue', 'flags' is used to indicate what kind of match need to be
7279  * removed, match is by default for the destination address, if
7280  * matching by source address is to be removed the flag
7281  * IGB_MAC_STATE_SRC_ADDR can be used.
7282  */
7283 static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7284                                     const u8 *addr, const u8 queue,
7285                                     const u8 flags)
7286 {
7287         struct e1000_hw *hw = &adapter->hw;
7288         int rar_entries = hw->mac.rar_entry_count -
7289                           adapter->vfs_allocated_count;
7290         int i;
7291
7292         if (is_zero_ether_addr(addr))
7293                 return -EINVAL;
7294
7295         /* Search for matching entry in the MAC table based on given address
7296          * and queue. Do not touch entries at the end of the table reserved
7297          * for the VF MAC addresses.
7298          */
7299         for (i = 0; i < rar_entries; i++) {
7300                 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7301                         continue;
7302                 if ((adapter->mac_table[i].state & flags) != flags)
7303                         continue;
7304                 if (adapter->mac_table[i].queue != queue)
7305                         continue;
7306                 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7307                         continue;
7308
7309                 /* When a filter for the default address is "deleted",
7310                  * we return it to its initial configuration
7311                  */
7312                 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7313                         adapter->mac_table[i].state =
7314                                 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7315                         adapter->mac_table[i].queue =
7316                                 adapter->vfs_allocated_count;
7317                 } else {
7318                         adapter->mac_table[i].state = 0;
7319                         adapter->mac_table[i].queue = 0;
7320                         memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
7321                 }
7322
7323                 igb_rar_set_index(adapter, i);
7324                 return 0;
7325         }
7326
7327         return -ENOENT;
7328 }
7329
7330 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7331                               const u8 queue)
7332 {
7333         return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7334 }
7335
7336 int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7337                                 const u8 *addr, u8 queue, u8 flags)
7338 {
7339         struct e1000_hw *hw = &adapter->hw;
7340
7341         /* In theory, this should be supported on 82575 as well, but
7342          * that part wasn't easily accessible during development.
7343          */
7344         if (hw->mac.type != e1000_i210)
7345                 return -EOPNOTSUPP;
7346
7347         return igb_add_mac_filter_flags(adapter, addr, queue,
7348                                         IGB_MAC_STATE_QUEUE_STEERING | flags);
7349 }
7350
7351 int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7352                                 const u8 *addr, u8 queue, u8 flags)
7353 {
7354         return igb_del_mac_filter_flags(adapter, addr, queue,
7355                                         IGB_MAC_STATE_QUEUE_STEERING | flags);
7356 }
7357
7358 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7359 {
7360         struct igb_adapter *adapter = netdev_priv(netdev);
7361         int ret;
7362
7363         ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7364
7365         return min_t(int, ret, 0);
7366 }
7367
7368 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7369 {
7370         struct igb_adapter *adapter = netdev_priv(netdev);
7371
7372         igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7373
7374         return 0;
7375 }
7376
7377 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7378                                  const u32 info, const u8 *addr)
7379 {
7380         struct pci_dev *pdev = adapter->pdev;
7381         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7382         struct list_head *pos;
7383         struct vf_mac_filter *entry = NULL;
7384         int ret = 0;
7385
7386         switch (info) {
7387         case E1000_VF_MAC_FILTER_CLR:
7388                 /* remove all unicast MAC filters related to the current VF */
7389                 list_for_each(pos, &adapter->vf_macs.l) {
7390                         entry = list_entry(pos, struct vf_mac_filter, l);
7391                         if (entry->vf == vf) {
7392                                 entry->vf = -1;
7393                                 entry->free = true;
7394                                 igb_del_mac_filter(adapter, entry->vf_mac, vf);
7395                         }
7396                 }
7397                 break;
7398         case E1000_VF_MAC_FILTER_ADD:
7399                 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7400                     !vf_data->trusted) {
7401                         dev_warn(&pdev->dev,
7402                                  "VF %d requested MAC filter but is administratively denied\n",
7403                                  vf);
7404                         return -EINVAL;
7405                 }
7406                 if (!is_valid_ether_addr(addr)) {
7407                         dev_warn(&pdev->dev,
7408                                  "VF %d attempted to set invalid MAC filter\n",
7409                                  vf);
7410                         return -EINVAL;
7411                 }
7412
7413                 /* try to find empty slot in the list */
7414                 list_for_each(pos, &adapter->vf_macs.l) {
7415                         entry = list_entry(pos, struct vf_mac_filter, l);
7416                         if (entry->free)
7417                                 break;
7418                 }
7419
7420                 if (entry && entry->free) {
7421                         entry->free = false;
7422                         entry->vf = vf;
7423                         ether_addr_copy(entry->vf_mac, addr);
7424
7425                         ret = igb_add_mac_filter(adapter, addr, vf);
7426                         ret = min_t(int, ret, 0);
7427                 } else {
7428                         ret = -ENOSPC;
7429                 }
7430
7431                 if (ret == -ENOSPC)
7432                         dev_warn(&pdev->dev,
7433                                  "VF %d has requested MAC filter but there is no space for it\n",
7434                                  vf);
7435                 break;
7436         default:
7437                 ret = -EINVAL;
7438                 break;
7439         }
7440
7441         return ret;
7442 }
7443
7444 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7445 {
7446         struct pci_dev *pdev = adapter->pdev;
7447         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7448         u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7449
7450         /* The VF MAC Address is stored in a packed array of bytes
7451          * starting at the second 32 bit word of the msg array
7452          */
7453         unsigned char *addr = (unsigned char *)&msg[1];
7454         int ret = 0;
7455
7456         if (!info) {
7457                 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7458                     !vf_data->trusted) {
7459                         dev_warn(&pdev->dev,
7460                                  "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7461                                  vf);
7462                         return -EINVAL;
7463                 }
7464
7465                 if (!is_valid_ether_addr(addr)) {
7466                         dev_warn(&pdev->dev,
7467                                  "VF %d attempted to set invalid MAC\n",
7468                                  vf);
7469                         return -EINVAL;
7470                 }
7471
7472                 ret = igb_set_vf_mac(adapter, vf, addr);
7473         } else {
7474                 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7475         }
7476
7477         return ret;
7478 }
7479
7480 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7481 {
7482         struct e1000_hw *hw = &adapter->hw;
7483         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7484         u32 msg = E1000_VT_MSGTYPE_NACK;
7485
7486         /* if device isn't clear to send it shouldn't be reading either */
7487         if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7488             time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7489                 igb_write_mbx(hw, &msg, 1, vf);
7490                 vf_data->last_nack = jiffies;
7491         }
7492 }
7493
7494 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7495 {
7496         struct pci_dev *pdev = adapter->pdev;
7497         u32 msgbuf[E1000_VFMAILBOX_SIZE];
7498         struct e1000_hw *hw = &adapter->hw;
7499         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7500         s32 retval;
7501
7502         retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7503
7504         if (retval) {
7505                 /* if receive failed revoke VF CTS stats and restart init */
7506                 dev_err(&pdev->dev, "Error receiving message from VF\n");
7507                 vf_data->flags &= ~IGB_VF_FLAG_CTS;
7508                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7509                         goto unlock;
7510                 goto out;
7511         }
7512
7513         /* this is a message we already processed, do nothing */
7514         if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7515                 goto unlock;
7516
7517         /* until the vf completes a reset it should not be
7518          * allowed to start any configuration.
7519          */
7520         if (msgbuf[0] == E1000_VF_RESET) {
7521                 /* unlocks mailbox */
7522                 igb_vf_reset_msg(adapter, vf);
7523                 return;
7524         }
7525
7526         if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7527                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7528                         goto unlock;
7529                 retval = -1;
7530                 goto out;
7531         }
7532
7533         switch ((msgbuf[0] & 0xFFFF)) {
7534         case E1000_VF_SET_MAC_ADDR:
7535                 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7536                 break;
7537         case E1000_VF_SET_PROMISC:
7538                 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7539                 break;
7540         case E1000_VF_SET_MULTICAST:
7541                 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
7542                 break;
7543         case E1000_VF_SET_LPE:
7544                 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
7545                 break;
7546         case E1000_VF_SET_VLAN:
7547                 retval = -1;
7548                 if (vf_data->pf_vlan)
7549                         dev_warn(&pdev->dev,
7550                                  "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
7551                                  vf);
7552                 else
7553                         retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
7554                 break;
7555         default:
7556                 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
7557                 retval = -1;
7558                 break;
7559         }
7560
7561         msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
7562 out:
7563         /* notify the VF of the results of what it sent us */
7564         if (retval)
7565                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
7566         else
7567                 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
7568
7569         /* unlocks mailbox */
7570         igb_write_mbx(hw, msgbuf, 1, vf);
7571         return;
7572
7573 unlock:
7574         igb_unlock_mbx(hw, vf);
7575 }
7576
7577 static void igb_msg_task(struct igb_adapter *adapter)
7578 {
7579         struct e1000_hw *hw = &adapter->hw;
7580         u32 vf;
7581
7582         for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
7583                 /* process any reset requests */
7584                 if (!igb_check_for_rst(hw, vf))
7585                         igb_vf_reset_event(adapter, vf);
7586
7587                 /* process any messages pending */
7588                 if (!igb_check_for_msg(hw, vf))
7589                         igb_rcv_msg_from_vf(adapter, vf);
7590
7591                 /* process any acks */
7592                 if (!igb_check_for_ack(hw, vf))
7593                         igb_rcv_ack_from_vf(adapter, vf);
7594         }
7595 }
7596
7597 /**
7598  *  igb_set_uta - Set unicast filter table address
7599  *  @adapter: board private structure
7600  *  @set: boolean indicating if we are setting or clearing bits
7601  *
7602  *  The unicast table address is a register array of 32-bit registers.
7603  *  The table is meant to be used in a way similar to how the MTA is used
7604  *  however due to certain limitations in the hardware it is necessary to
7605  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
7606  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
7607  **/
7608 static void igb_set_uta(struct igb_adapter *adapter, bool set)
7609 {
7610         struct e1000_hw *hw = &adapter->hw;
7611         u32 uta = set ? ~0 : 0;
7612         int i;
7613
7614         /* we only need to do this if VMDq is enabled */
7615         if (!adapter->vfs_allocated_count)
7616                 return;
7617
7618         for (i = hw->mac.uta_reg_count; i--;)
7619                 array_wr32(E1000_UTA, i, uta);
7620 }
7621
7622 /**
7623  *  igb_intr_msi - Interrupt Handler
7624  *  @irq: interrupt number
7625  *  @data: pointer to a network interface device structure
7626  **/
7627 static irqreturn_t igb_intr_msi(int irq, void *data)
7628 {
7629         struct igb_adapter *adapter = data;
7630         struct igb_q_vector *q_vector = adapter->q_vector[0];
7631         struct e1000_hw *hw = &adapter->hw;
7632         /* read ICR disables interrupts using IAM */
7633         u32 icr = rd32(E1000_ICR);
7634
7635         igb_write_itr(q_vector);
7636
7637         if (icr & E1000_ICR_DRSTA)
7638                 schedule_work(&adapter->reset_task);
7639
7640         if (icr & E1000_ICR_DOUTSYNC) {
7641                 /* HW is reporting DMA is out of sync */
7642                 adapter->stats.doosync++;
7643         }
7644
7645         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7646                 hw->mac.get_link_status = 1;
7647                 if (!test_bit(__IGB_DOWN, &adapter->state))
7648                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
7649         }
7650
7651         if (icr & E1000_ICR_TS)
7652                 igb_tsync_interrupt(adapter);
7653
7654         napi_schedule(&q_vector->napi);
7655
7656         return IRQ_HANDLED;
7657 }
7658
7659 /**
7660  *  igb_intr - Legacy Interrupt Handler
7661  *  @irq: interrupt number
7662  *  @data: pointer to a network interface device structure
7663  **/
7664 static irqreturn_t igb_intr(int irq, void *data)
7665 {
7666         struct igb_adapter *adapter = data;
7667         struct igb_q_vector *q_vector = adapter->q_vector[0];
7668         struct e1000_hw *hw = &adapter->hw;
7669         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
7670          * need for the IMC write
7671          */
7672         u32 icr = rd32(E1000_ICR);
7673
7674         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
7675          * not set, then the adapter didn't send an interrupt
7676          */
7677         if (!(icr & E1000_ICR_INT_ASSERTED))
7678                 return IRQ_NONE;
7679
7680         igb_write_itr(q_vector);
7681
7682         if (icr & E1000_ICR_DRSTA)
7683                 schedule_work(&adapter->reset_task);
7684
7685         if (icr & E1000_ICR_DOUTSYNC) {
7686                 /* HW is reporting DMA is out of sync */
7687                 adapter->stats.doosync++;
7688         }
7689
7690         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7691                 hw->mac.get_link_status = 1;
7692                 /* guard against interrupt when we're going down */
7693                 if (!test_bit(__IGB_DOWN, &adapter->state))
7694                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
7695         }
7696
7697         if (icr & E1000_ICR_TS)
7698                 igb_tsync_interrupt(adapter);
7699
7700         napi_schedule(&q_vector->napi);
7701
7702         return IRQ_HANDLED;
7703 }
7704
7705 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
7706 {
7707         struct igb_adapter *adapter = q_vector->adapter;
7708         struct e1000_hw *hw = &adapter->hw;
7709
7710         if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
7711             (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
7712                 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
7713                         igb_set_itr(q_vector);
7714                 else
7715                         igb_update_ring_itr(q_vector);
7716         }
7717
7718         if (!test_bit(__IGB_DOWN, &adapter->state)) {
7719                 if (adapter->flags & IGB_FLAG_HAS_MSIX)
7720                         wr32(E1000_EIMS, q_vector->eims_value);
7721                 else
7722                         igb_irq_enable(adapter);
7723         }
7724 }
7725
7726 /**
7727  *  igb_poll - NAPI Rx polling callback
7728  *  @napi: napi polling structure
7729  *  @budget: count of how many packets we should handle
7730  **/
7731 static int igb_poll(struct napi_struct *napi, int budget)
7732 {
7733         struct igb_q_vector *q_vector = container_of(napi,
7734                                                      struct igb_q_vector,
7735                                                      napi);
7736         bool clean_complete = true;
7737         int work_done = 0;
7738
7739 #ifdef CONFIG_IGB_DCA
7740         if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
7741                 igb_update_dca(q_vector);
7742 #endif
7743         if (q_vector->tx.ring)
7744                 clean_complete = igb_clean_tx_irq(q_vector, budget);
7745
7746         if (q_vector->rx.ring) {
7747                 int cleaned = igb_clean_rx_irq(q_vector, budget);
7748
7749                 work_done += cleaned;
7750                 if (cleaned >= budget)
7751                         clean_complete = false;
7752         }
7753
7754         /* If all work not completed, return budget and keep polling */
7755         if (!clean_complete)
7756                 return budget;
7757
7758         /* Exit the polling mode, but don't re-enable interrupts if stack might
7759          * poll us due to busy-polling
7760          */
7761         if (likely(napi_complete_done(napi, work_done)))
7762                 igb_ring_irq_enable(q_vector);
7763
7764         return min(work_done, budget - 1);
7765 }
7766
7767 /**
7768  *  igb_clean_tx_irq - Reclaim resources after transmit completes
7769  *  @q_vector: pointer to q_vector containing needed info
7770  *  @napi_budget: Used to determine if we are in netpoll
7771  *
7772  *  returns true if ring is completely cleaned
7773  **/
7774 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
7775 {
7776         struct igb_adapter *adapter = q_vector->adapter;
7777         struct igb_ring *tx_ring = q_vector->tx.ring;
7778         struct igb_tx_buffer *tx_buffer;
7779         union e1000_adv_tx_desc *tx_desc;
7780         unsigned int total_bytes = 0, total_packets = 0;
7781         unsigned int budget = q_vector->tx.work_limit;
7782         unsigned int i = tx_ring->next_to_clean;
7783
7784         if (test_bit(__IGB_DOWN, &adapter->state))
7785                 return true;
7786
7787         tx_buffer = &tx_ring->tx_buffer_info[i];
7788         tx_desc = IGB_TX_DESC(tx_ring, i);
7789         i -= tx_ring->count;
7790
7791         do {
7792                 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
7793
7794                 /* if next_to_watch is not set then there is no work pending */
7795                 if (!eop_desc)
7796                         break;
7797
7798                 /* prevent any other reads prior to eop_desc */
7799                 smp_rmb();
7800
7801                 /* if DD is not set pending work has not been completed */
7802                 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
7803                         break;
7804
7805                 /* clear next_to_watch to prevent false hangs */
7806                 tx_buffer->next_to_watch = NULL;
7807
7808                 /* update the statistics for this packet */
7809                 total_bytes += tx_buffer->bytecount;
7810                 total_packets += tx_buffer->gso_segs;
7811
7812                 /* free the skb */
7813                 napi_consume_skb(tx_buffer->skb, napi_budget);
7814
7815                 /* unmap skb header data */
7816                 dma_unmap_single(tx_ring->dev,
7817                                  dma_unmap_addr(tx_buffer, dma),
7818                                  dma_unmap_len(tx_buffer, len),
7819                                  DMA_TO_DEVICE);
7820
7821                 /* clear tx_buffer data */
7822                 dma_unmap_len_set(tx_buffer, len, 0);
7823
7824                 /* clear last DMA location and unmap remaining buffers */
7825                 while (tx_desc != eop_desc) {
7826                         tx_buffer++;
7827                         tx_desc++;
7828                         i++;
7829                         if (unlikely(!i)) {
7830                                 i -= tx_ring->count;
7831                                 tx_buffer = tx_ring->tx_buffer_info;
7832                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
7833                         }
7834
7835                         /* unmap any remaining paged data */
7836                         if (dma_unmap_len(tx_buffer, len)) {
7837                                 dma_unmap_page(tx_ring->dev,
7838                                                dma_unmap_addr(tx_buffer, dma),
7839                                                dma_unmap_len(tx_buffer, len),
7840                                                DMA_TO_DEVICE);
7841                                 dma_unmap_len_set(tx_buffer, len, 0);
7842                         }
7843                 }
7844
7845                 /* move us one more past the eop_desc for start of next pkt */
7846                 tx_buffer++;
7847                 tx_desc++;
7848                 i++;
7849                 if (unlikely(!i)) {
7850                         i -= tx_ring->count;
7851                         tx_buffer = tx_ring->tx_buffer_info;
7852                         tx_desc = IGB_TX_DESC(tx_ring, 0);
7853                 }
7854
7855                 /* issue prefetch for next Tx descriptor */
7856                 prefetch(tx_desc);
7857
7858                 /* update budget accounting */
7859                 budget--;
7860         } while (likely(budget));
7861
7862         netdev_tx_completed_queue(txring_txq(tx_ring),
7863                                   total_packets, total_bytes);
7864         i += tx_ring->count;
7865         tx_ring->next_to_clean = i;
7866         u64_stats_update_begin(&tx_ring->tx_syncp);
7867         tx_ring->tx_stats.bytes += total_bytes;
7868         tx_ring->tx_stats.packets += total_packets;
7869         u64_stats_update_end(&tx_ring->tx_syncp);
7870         q_vector->tx.total_bytes += total_bytes;
7871         q_vector->tx.total_packets += total_packets;
7872
7873         if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
7874                 struct e1000_hw *hw = &adapter->hw;
7875
7876                 /* Detect a transmit hang in hardware, this serializes the
7877                  * check with the clearing of time_stamp and movement of i
7878                  */
7879                 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
7880                 if (tx_buffer->next_to_watch &&
7881                     time_after(jiffies, tx_buffer->time_stamp +
7882                                (adapter->tx_timeout_factor * HZ)) &&
7883                     !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
7884
7885                         /* detected Tx unit hang */
7886                         dev_err(tx_ring->dev,
7887                                 "Detected Tx Unit Hang\n"
7888                                 "  Tx Queue             <%d>\n"
7889                                 "  TDH                  <%x>\n"
7890                                 "  TDT                  <%x>\n"
7891                                 "  next_to_use          <%x>\n"
7892                                 "  next_to_clean        <%x>\n"
7893                                 "buffer_info[next_to_clean]\n"
7894                                 "  time_stamp           <%lx>\n"
7895                                 "  next_to_watch        <%p>\n"
7896                                 "  jiffies              <%lx>\n"
7897                                 "  desc.status          <%x>\n",
7898                                 tx_ring->queue_index,
7899                                 rd32(E1000_TDH(tx_ring->reg_idx)),
7900                                 readl(tx_ring->tail),
7901                                 tx_ring->next_to_use,
7902                                 tx_ring->next_to_clean,
7903                                 tx_buffer->time_stamp,
7904                                 tx_buffer->next_to_watch,
7905                                 jiffies,
7906                                 tx_buffer->next_to_watch->wb.status);
7907                         netif_stop_subqueue(tx_ring->netdev,
7908                                             tx_ring->queue_index);
7909
7910                         /* we are about to reset, no point in enabling stuff */
7911                         return true;
7912                 }
7913         }
7914
7915 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
7916         if (unlikely(total_packets &&
7917             netif_carrier_ok(tx_ring->netdev) &&
7918             igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
7919                 /* Make sure that anybody stopping the queue after this
7920                  * sees the new next_to_clean.
7921                  */
7922                 smp_mb();
7923                 if (__netif_subqueue_stopped(tx_ring->netdev,
7924                                              tx_ring->queue_index) &&
7925                     !(test_bit(__IGB_DOWN, &adapter->state))) {
7926                         netif_wake_subqueue(tx_ring->netdev,
7927                                             tx_ring->queue_index);
7928
7929                         u64_stats_update_begin(&tx_ring->tx_syncp);
7930                         tx_ring->tx_stats.restart_queue++;
7931                         u64_stats_update_end(&tx_ring->tx_syncp);
7932                 }
7933         }
7934
7935         return !!budget;
7936 }
7937
7938 /**
7939  *  igb_reuse_rx_page - page flip buffer and store it back on the ring
7940  *  @rx_ring: rx descriptor ring to store buffers on
7941  *  @old_buff: donor buffer to have page reused
7942  *
7943  *  Synchronizes page for reuse by the adapter
7944  **/
7945 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
7946                               struct igb_rx_buffer *old_buff)
7947 {
7948         struct igb_rx_buffer *new_buff;
7949         u16 nta = rx_ring->next_to_alloc;
7950
7951         new_buff = &rx_ring->rx_buffer_info[nta];
7952
7953         /* update, and store next to alloc */
7954         nta++;
7955         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
7956
7957         /* Transfer page from old buffer to new buffer.
7958          * Move each member individually to avoid possible store
7959          * forwarding stalls.
7960          */
7961         new_buff->dma           = old_buff->dma;
7962         new_buff->page          = old_buff->page;
7963         new_buff->page_offset   = old_buff->page_offset;
7964         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
7965 }
7966
7967 static inline bool igb_page_is_reserved(struct page *page)
7968 {
7969         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
7970 }
7971
7972 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer)
7973 {
7974         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
7975         struct page *page = rx_buffer->page;
7976
7977         /* avoid re-using remote pages */
7978         if (unlikely(igb_page_is_reserved(page)))
7979                 return false;
7980
7981 #if (PAGE_SIZE < 8192)
7982         /* if we are only owner of page we can reuse it */
7983         if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
7984                 return false;
7985 #else
7986 #define IGB_LAST_OFFSET \
7987         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
7988
7989         if (rx_buffer->page_offset > IGB_LAST_OFFSET)
7990                 return false;
7991 #endif
7992
7993         /* If we have drained the page fragment pool we need to update
7994          * the pagecnt_bias and page count so that we fully restock the
7995          * number of references the driver holds.
7996          */
7997         if (unlikely(!pagecnt_bias)) {
7998                 page_ref_add(page, USHRT_MAX);
7999                 rx_buffer->pagecnt_bias = USHRT_MAX;
8000         }
8001
8002         return true;
8003 }
8004
8005 /**
8006  *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8007  *  @rx_ring: rx descriptor ring to transact packets on
8008  *  @rx_buffer: buffer containing page to add
8009  *  @skb: sk_buff to place the data into
8010  *  @size: size of buffer to be added
8011  *
8012  *  This function will add the data contained in rx_buffer->page to the skb.
8013  **/
8014 static void igb_add_rx_frag(struct igb_ring *rx_ring,
8015                             struct igb_rx_buffer *rx_buffer,
8016                             struct sk_buff *skb,
8017                             unsigned int size)
8018 {
8019 #if (PAGE_SIZE < 8192)
8020         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8021 #else
8022         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8023                                 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8024                                 SKB_DATA_ALIGN(size);
8025 #endif
8026         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8027                         rx_buffer->page_offset, size, truesize);
8028 #if (PAGE_SIZE < 8192)
8029         rx_buffer->page_offset ^= truesize;
8030 #else
8031         rx_buffer->page_offset += truesize;
8032 #endif
8033 }
8034
8035 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8036                                          struct igb_rx_buffer *rx_buffer,
8037                                          union e1000_adv_rx_desc *rx_desc,
8038                                          unsigned int size)
8039 {
8040         void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
8041 #if (PAGE_SIZE < 8192)
8042         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8043 #else
8044         unsigned int truesize = SKB_DATA_ALIGN(size);
8045 #endif
8046         unsigned int headlen;
8047         struct sk_buff *skb;
8048
8049         /* prefetch first cache line of first page */
8050         prefetch(va);
8051 #if L1_CACHE_BYTES < 128
8052         prefetch(va + L1_CACHE_BYTES);
8053 #endif
8054
8055         /* allocate a skb to store the frags */
8056         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8057         if (unlikely(!skb))
8058                 return NULL;
8059
8060         if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) {
8061                 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
8062                 va += IGB_TS_HDR_LEN;
8063                 size -= IGB_TS_HDR_LEN;
8064         }
8065
8066         /* Determine available headroom for copy */
8067         headlen = size;
8068         if (headlen > IGB_RX_HDR_LEN)
8069                 headlen = eth_get_headlen(skb->dev, va, IGB_RX_HDR_LEN);
8070
8071         /* align pull length to size of long to optimize memcpy performance */
8072         memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
8073
8074         /* update all of the pointers */
8075         size -= headlen;
8076         if (size) {
8077                 skb_add_rx_frag(skb, 0, rx_buffer->page,
8078                                 (va + headlen) - page_address(rx_buffer->page),
8079                                 size, truesize);
8080 #if (PAGE_SIZE < 8192)
8081                 rx_buffer->page_offset ^= truesize;
8082 #else
8083                 rx_buffer->page_offset += truesize;
8084 #endif
8085         } else {
8086                 rx_buffer->pagecnt_bias++;
8087         }
8088
8089         return skb;
8090 }
8091
8092 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8093                                      struct igb_rx_buffer *rx_buffer,
8094                                      union e1000_adv_rx_desc *rx_desc,
8095                                      unsigned int size)
8096 {
8097         void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
8098 #if (PAGE_SIZE < 8192)
8099         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8100 #else
8101         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8102                                 SKB_DATA_ALIGN(IGB_SKB_PAD + size);
8103 #endif
8104         struct sk_buff *skb;
8105
8106         /* prefetch first cache line of first page */
8107         prefetch(va);
8108 #if L1_CACHE_BYTES < 128
8109         prefetch(va + L1_CACHE_BYTES);
8110 #endif
8111
8112         /* build an skb around the page buffer */
8113         skb = build_skb(va - IGB_SKB_PAD, truesize);
8114         if (unlikely(!skb))
8115                 return NULL;
8116
8117         /* update pointers within the skb to store the data */
8118         skb_reserve(skb, IGB_SKB_PAD);
8119         __skb_put(skb, size);
8120
8121         /* pull timestamp out of packet data */
8122         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8123                 igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
8124                 __skb_pull(skb, IGB_TS_HDR_LEN);
8125         }
8126
8127         /* update buffer offset */
8128 #if (PAGE_SIZE < 8192)
8129         rx_buffer->page_offset ^= truesize;
8130 #else
8131         rx_buffer->page_offset += truesize;
8132 #endif
8133
8134         return skb;
8135 }
8136
8137 static inline void igb_rx_checksum(struct igb_ring *ring,
8138                                    union e1000_adv_rx_desc *rx_desc,
8139                                    struct sk_buff *skb)
8140 {
8141         skb_checksum_none_assert(skb);
8142
8143         /* Ignore Checksum bit is set */
8144         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8145                 return;
8146
8147         /* Rx checksum disabled via ethtool */
8148         if (!(ring->netdev->features & NETIF_F_RXCSUM))
8149                 return;
8150
8151         /* TCP/UDP checksum error bit is set */
8152         if (igb_test_staterr(rx_desc,
8153                              E1000_RXDEXT_STATERR_TCPE |
8154                              E1000_RXDEXT_STATERR_IPE)) {
8155                 /* work around errata with sctp packets where the TCPE aka
8156                  * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8157                  * packets, (aka let the stack check the crc32c)
8158                  */
8159                 if (!((skb->len == 60) &&
8160                       test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8161                         u64_stats_update_begin(&ring->rx_syncp);
8162                         ring->rx_stats.csum_err++;
8163                         u64_stats_update_end(&ring->rx_syncp);
8164                 }
8165                 /* let the stack verify checksum errors */
8166                 return;
8167         }
8168         /* It must be a TCP or UDP packet with a valid checksum */
8169         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8170                                       E1000_RXD_STAT_UDPCS))
8171                 skb->ip_summed = CHECKSUM_UNNECESSARY;
8172
8173         dev_dbg(ring->dev, "cksum success: bits %08X\n",
8174                 le32_to_cpu(rx_desc->wb.upper.status_error));
8175 }
8176
8177 static inline void igb_rx_hash(struct igb_ring *ring,
8178                                union e1000_adv_rx_desc *rx_desc,
8179                                struct sk_buff *skb)
8180 {
8181         if (ring->netdev->features & NETIF_F_RXHASH)
8182                 skb_set_hash(skb,
8183                              le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8184                              PKT_HASH_TYPE_L3);
8185 }
8186
8187 /**
8188  *  igb_is_non_eop - process handling of non-EOP buffers
8189  *  @rx_ring: Rx ring being processed
8190  *  @rx_desc: Rx descriptor for current buffer
8191  *  @skb: current socket buffer containing buffer in progress
8192  *
8193  *  This function updates next to clean.  If the buffer is an EOP buffer
8194  *  this function exits returning false, otherwise it will place the
8195  *  sk_buff in the next buffer to be chained and return true indicating
8196  *  that this is in fact a non-EOP buffer.
8197  **/
8198 static bool igb_is_non_eop(struct igb_ring *rx_ring,
8199                            union e1000_adv_rx_desc *rx_desc)
8200 {
8201         u32 ntc = rx_ring->next_to_clean + 1;
8202
8203         /* fetch, update, and store next to clean */
8204         ntc = (ntc < rx_ring->count) ? ntc : 0;
8205         rx_ring->next_to_clean = ntc;
8206
8207         prefetch(IGB_RX_DESC(rx_ring, ntc));
8208
8209         if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8210                 return false;
8211
8212         return true;
8213 }
8214
8215 /**
8216  *  igb_cleanup_headers - Correct corrupted or empty headers
8217  *  @rx_ring: rx descriptor ring packet is being transacted on
8218  *  @rx_desc: pointer to the EOP Rx descriptor
8219  *  @skb: pointer to current skb being fixed
8220  *
8221  *  Address the case where we are pulling data in on pages only
8222  *  and as such no data is present in the skb header.
8223  *
8224  *  In addition if skb is not at least 60 bytes we need to pad it so that
8225  *  it is large enough to qualify as a valid Ethernet frame.
8226  *
8227  *  Returns true if an error was encountered and skb was freed.
8228  **/
8229 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8230                                 union e1000_adv_rx_desc *rx_desc,
8231                                 struct sk_buff *skb)
8232 {
8233         if (unlikely((igb_test_staterr(rx_desc,
8234                                        E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8235                 struct net_device *netdev = rx_ring->netdev;
8236                 if (!(netdev->features & NETIF_F_RXALL)) {
8237                         dev_kfree_skb_any(skb);
8238                         return true;
8239                 }
8240         }
8241
8242         /* if eth_skb_pad returns an error the skb was freed */
8243         if (eth_skb_pad(skb))
8244                 return true;
8245
8246         return false;
8247 }
8248
8249 /**
8250  *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
8251  *  @rx_ring: rx descriptor ring packet is being transacted on
8252  *  @rx_desc: pointer to the EOP Rx descriptor
8253  *  @skb: pointer to current skb being populated
8254  *
8255  *  This function checks the ring, descriptor, and packet information in
8256  *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
8257  *  other fields within the skb.
8258  **/
8259 static void igb_process_skb_fields(struct igb_ring *rx_ring,
8260                                    union e1000_adv_rx_desc *rx_desc,
8261                                    struct sk_buff *skb)
8262 {
8263         struct net_device *dev = rx_ring->netdev;
8264
8265         igb_rx_hash(rx_ring, rx_desc, skb);
8266
8267         igb_rx_checksum(rx_ring, rx_desc, skb);
8268
8269         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8270             !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8271                 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8272
8273         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8274             igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8275                 u16 vid;
8276
8277                 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8278                     test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8279                         vid = be16_to_cpu(rx_desc->wb.upper.vlan);
8280                 else
8281                         vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8282
8283                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8284         }
8285
8286         skb_record_rx_queue(skb, rx_ring->queue_index);
8287
8288         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8289 }
8290
8291 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8292                                                const unsigned int size)
8293 {
8294         struct igb_rx_buffer *rx_buffer;
8295
8296         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8297         prefetchw(rx_buffer->page);
8298
8299         /* we are reusing so sync this buffer for CPU use */
8300         dma_sync_single_range_for_cpu(rx_ring->dev,
8301                                       rx_buffer->dma,
8302                                       rx_buffer->page_offset,
8303                                       size,
8304                                       DMA_FROM_DEVICE);
8305
8306         rx_buffer->pagecnt_bias--;
8307
8308         return rx_buffer;
8309 }
8310
8311 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8312                               struct igb_rx_buffer *rx_buffer)
8313 {
8314         if (igb_can_reuse_rx_page(rx_buffer)) {
8315                 /* hand second half of page back to the ring */
8316                 igb_reuse_rx_page(rx_ring, rx_buffer);
8317         } else {
8318                 /* We are not reusing the buffer so unmap it and free
8319                  * any references we are holding to it
8320                  */
8321                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8322                                      igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8323                                      IGB_RX_DMA_ATTR);
8324                 __page_frag_cache_drain(rx_buffer->page,
8325                                         rx_buffer->pagecnt_bias);
8326         }
8327
8328         /* clear contents of rx_buffer */
8329         rx_buffer->page = NULL;
8330 }
8331
8332 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8333 {
8334         struct igb_ring *rx_ring = q_vector->rx.ring;
8335         struct sk_buff *skb = rx_ring->skb;
8336         unsigned int total_bytes = 0, total_packets = 0;
8337         u16 cleaned_count = igb_desc_unused(rx_ring);
8338
8339         while (likely(total_packets < budget)) {
8340                 union e1000_adv_rx_desc *rx_desc;
8341                 struct igb_rx_buffer *rx_buffer;
8342                 unsigned int size;
8343
8344                 /* return some buffers to hardware, one at a time is too slow */
8345                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8346                         igb_alloc_rx_buffers(rx_ring, cleaned_count);
8347                         cleaned_count = 0;
8348                 }
8349
8350                 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8351                 size = le16_to_cpu(rx_desc->wb.upper.length);
8352                 if (!size)
8353                         break;
8354
8355                 /* This memory barrier is needed to keep us from reading
8356                  * any other fields out of the rx_desc until we know the
8357                  * descriptor has been written back
8358                  */
8359                 dma_rmb();
8360
8361                 rx_buffer = igb_get_rx_buffer(rx_ring, size);
8362
8363                 /* retrieve a buffer from the ring */
8364                 if (skb)
8365                         igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8366                 else if (ring_uses_build_skb(rx_ring))
8367                         skb = igb_build_skb(rx_ring, rx_buffer, rx_desc, size);
8368                 else
8369                         skb = igb_construct_skb(rx_ring, rx_buffer,
8370                                                 rx_desc, size);
8371
8372                 /* exit if we failed to retrieve a buffer */
8373                 if (!skb) {
8374                         rx_ring->rx_stats.alloc_failed++;
8375                         rx_buffer->pagecnt_bias++;
8376                         break;
8377                 }
8378
8379                 igb_put_rx_buffer(rx_ring, rx_buffer);
8380                 cleaned_count++;
8381
8382                 /* fetch next buffer in frame if non-eop */
8383                 if (igb_is_non_eop(rx_ring, rx_desc))
8384                         continue;
8385
8386                 /* verify the packet layout is correct */
8387                 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8388                         skb = NULL;
8389                         continue;
8390                 }
8391
8392                 /* probably a little skewed due to removing CRC */
8393                 total_bytes += skb->len;
8394
8395                 /* populate checksum, timestamp, VLAN, and protocol */
8396                 igb_process_skb_fields(rx_ring, rx_desc, skb);
8397
8398                 napi_gro_receive(&q_vector->napi, skb);
8399
8400                 /* reset skb pointer */
8401                 skb = NULL;
8402
8403                 /* update budget accounting */
8404                 total_packets++;
8405         }
8406
8407         /* place incomplete frames back on ring for completion */
8408         rx_ring->skb = skb;
8409
8410         u64_stats_update_begin(&rx_ring->rx_syncp);
8411         rx_ring->rx_stats.packets += total_packets;
8412         rx_ring->rx_stats.bytes += total_bytes;
8413         u64_stats_update_end(&rx_ring->rx_syncp);
8414         q_vector->rx.total_packets += total_packets;
8415         q_vector->rx.total_bytes += total_bytes;
8416
8417         if (cleaned_count)
8418                 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8419
8420         return total_packets;
8421 }
8422
8423 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8424 {
8425         return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8426 }
8427
8428 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
8429                                   struct igb_rx_buffer *bi)
8430 {
8431         struct page *page = bi->page;
8432         dma_addr_t dma;
8433
8434         /* since we are recycling buffers we should seldom need to alloc */
8435         if (likely(page))
8436                 return true;
8437
8438         /* alloc new page for storage */
8439         page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
8440         if (unlikely(!page)) {
8441                 rx_ring->rx_stats.alloc_failed++;
8442                 return false;
8443         }
8444
8445         /* map page for use */
8446         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
8447                                  igb_rx_pg_size(rx_ring),
8448                                  DMA_FROM_DEVICE,
8449                                  IGB_RX_DMA_ATTR);
8450
8451         /* if mapping failed free memory back to system since
8452          * there isn't much point in holding memory we can't use
8453          */
8454         if (dma_mapping_error(rx_ring->dev, dma)) {
8455                 __free_pages(page, igb_rx_pg_order(rx_ring));
8456
8457                 rx_ring->rx_stats.alloc_failed++;
8458                 return false;
8459         }
8460
8461         bi->dma = dma;
8462         bi->page = page;
8463         bi->page_offset = igb_rx_offset(rx_ring);
8464         bi->pagecnt_bias = 1;
8465
8466         return true;
8467 }
8468
8469 /**
8470  *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
8471  *  @adapter: address of board private structure
8472  **/
8473 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
8474 {
8475         union e1000_adv_rx_desc *rx_desc;
8476         struct igb_rx_buffer *bi;
8477         u16 i = rx_ring->next_to_use;
8478         u16 bufsz;
8479
8480         /* nothing to do */
8481         if (!cleaned_count)
8482                 return;
8483
8484         rx_desc = IGB_RX_DESC(rx_ring, i);
8485         bi = &rx_ring->rx_buffer_info[i];
8486         i -= rx_ring->count;
8487
8488         bufsz = igb_rx_bufsz(rx_ring);
8489
8490         do {
8491                 if (!igb_alloc_mapped_page(rx_ring, bi))
8492                         break;
8493
8494                 /* sync the buffer for use by the device */
8495                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
8496                                                  bi->page_offset, bufsz,
8497                                                  DMA_FROM_DEVICE);
8498
8499                 /* Refresh the desc even if buffer_addrs didn't change
8500                  * because each write-back erases this info.
8501                  */
8502                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
8503
8504                 rx_desc++;
8505                 bi++;
8506                 i++;
8507                 if (unlikely(!i)) {
8508                         rx_desc = IGB_RX_DESC(rx_ring, 0);
8509                         bi = rx_ring->rx_buffer_info;
8510                         i -= rx_ring->count;
8511                 }
8512
8513                 /* clear the length for the next_to_use descriptor */
8514                 rx_desc->wb.upper.length = 0;
8515
8516                 cleaned_count--;
8517         } while (cleaned_count);
8518
8519         i += rx_ring->count;
8520
8521         if (rx_ring->next_to_use != i) {
8522                 /* record the next descriptor to use */
8523                 rx_ring->next_to_use = i;
8524
8525                 /* update next to alloc since we have filled the ring */
8526                 rx_ring->next_to_alloc = i;
8527
8528                 /* Force memory writes to complete before letting h/w
8529                  * know there are new descriptors to fetch.  (Only
8530                  * applicable for weak-ordered memory model archs,
8531                  * such as IA-64).
8532                  */
8533                 dma_wmb();
8534                 writel(i, rx_ring->tail);
8535         }
8536 }
8537
8538 /**
8539  * igb_mii_ioctl -
8540  * @netdev:
8541  * @ifreq:
8542  * @cmd:
8543  **/
8544 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8545 {
8546         struct igb_adapter *adapter = netdev_priv(netdev);
8547         struct mii_ioctl_data *data = if_mii(ifr);
8548
8549         if (adapter->hw.phy.media_type != e1000_media_type_copper)
8550                 return -EOPNOTSUPP;
8551
8552         switch (cmd) {
8553         case SIOCGMIIPHY:
8554                 data->phy_id = adapter->hw.phy.addr;
8555                 break;
8556         case SIOCGMIIREG:
8557                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
8558                                      &data->val_out))
8559                         return -EIO;
8560                 break;
8561         case SIOCSMIIREG:
8562         default:
8563                 return -EOPNOTSUPP;
8564         }
8565         return 0;
8566 }
8567
8568 /**
8569  * igb_ioctl -
8570  * @netdev:
8571  * @ifreq:
8572  * @cmd:
8573  **/
8574 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8575 {
8576         switch (cmd) {
8577         case SIOCGMIIPHY:
8578         case SIOCGMIIREG:
8579         case SIOCSMIIREG:
8580                 return igb_mii_ioctl(netdev, ifr, cmd);
8581         case SIOCGHWTSTAMP:
8582                 return igb_ptp_get_ts_config(netdev, ifr);
8583         case SIOCSHWTSTAMP:
8584                 return igb_ptp_set_ts_config(netdev, ifr);
8585         default:
8586                 return -EOPNOTSUPP;
8587         }
8588 }
8589
8590 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8591 {
8592         struct igb_adapter *adapter = hw->back;
8593
8594         pci_read_config_word(adapter->pdev, reg, value);
8595 }
8596
8597 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8598 {
8599         struct igb_adapter *adapter = hw->back;
8600
8601         pci_write_config_word(adapter->pdev, reg, *value);
8602 }
8603
8604 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8605 {
8606         struct igb_adapter *adapter = hw->back;
8607
8608         if (pcie_capability_read_word(adapter->pdev, reg, value))
8609                 return -E1000_ERR_CONFIG;
8610
8611         return 0;
8612 }
8613
8614 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8615 {
8616         struct igb_adapter *adapter = hw->back;
8617
8618         if (pcie_capability_write_word(adapter->pdev, reg, *value))
8619                 return -E1000_ERR_CONFIG;
8620
8621         return 0;
8622 }
8623
8624 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
8625 {
8626         struct igb_adapter *adapter = netdev_priv(netdev);
8627         struct e1000_hw *hw = &adapter->hw;
8628         u32 ctrl, rctl;
8629         bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
8630
8631         if (enable) {
8632                 /* enable VLAN tag insert/strip */
8633                 ctrl = rd32(E1000_CTRL);
8634                 ctrl |= E1000_CTRL_VME;
8635                 wr32(E1000_CTRL, ctrl);
8636
8637                 /* Disable CFI check */
8638                 rctl = rd32(E1000_RCTL);
8639                 rctl &= ~E1000_RCTL_CFIEN;
8640                 wr32(E1000_RCTL, rctl);
8641         } else {
8642                 /* disable VLAN tag insert/strip */
8643                 ctrl = rd32(E1000_CTRL);
8644                 ctrl &= ~E1000_CTRL_VME;
8645                 wr32(E1000_CTRL, ctrl);
8646         }
8647
8648         igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
8649 }
8650
8651 static int igb_vlan_rx_add_vid(struct net_device *netdev,
8652                                __be16 proto, u16 vid)
8653 {
8654         struct igb_adapter *adapter = netdev_priv(netdev);
8655         struct e1000_hw *hw = &adapter->hw;
8656         int pf_id = adapter->vfs_allocated_count;
8657
8658         /* add the filter since PF can receive vlans w/o entry in vlvf */
8659         if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8660                 igb_vfta_set(hw, vid, pf_id, true, !!vid);
8661
8662         set_bit(vid, adapter->active_vlans);
8663
8664         return 0;
8665 }
8666
8667 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
8668                                 __be16 proto, u16 vid)
8669 {
8670         struct igb_adapter *adapter = netdev_priv(netdev);
8671         int pf_id = adapter->vfs_allocated_count;
8672         struct e1000_hw *hw = &adapter->hw;
8673
8674         /* remove VID from filter table */
8675         if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8676                 igb_vfta_set(hw, vid, pf_id, false, true);
8677
8678         clear_bit(vid, adapter->active_vlans);
8679
8680         return 0;
8681 }
8682
8683 static void igb_restore_vlan(struct igb_adapter *adapter)
8684 {
8685         u16 vid = 1;
8686
8687         igb_vlan_mode(adapter->netdev, adapter->netdev->features);
8688         igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
8689
8690         for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
8691                 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
8692 }
8693
8694 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
8695 {
8696         struct pci_dev *pdev = adapter->pdev;
8697         struct e1000_mac_info *mac = &adapter->hw.mac;
8698
8699         mac->autoneg = 0;
8700
8701         /* Make sure dplx is at most 1 bit and lsb of speed is not set
8702          * for the switch() below to work
8703          */
8704         if ((spd & 1) || (dplx & ~1))
8705                 goto err_inval;
8706
8707         /* Fiber NIC's only allow 1000 gbps Full duplex
8708          * and 100Mbps Full duplex for 100baseFx sfp
8709          */
8710         if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
8711                 switch (spd + dplx) {
8712                 case SPEED_10 + DUPLEX_HALF:
8713                 case SPEED_10 + DUPLEX_FULL:
8714                 case SPEED_100 + DUPLEX_HALF:
8715                         goto err_inval;
8716                 default:
8717                         break;
8718                 }
8719         }
8720
8721         switch (spd + dplx) {
8722         case SPEED_10 + DUPLEX_HALF:
8723                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
8724                 break;
8725         case SPEED_10 + DUPLEX_FULL:
8726                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
8727                 break;
8728         case SPEED_100 + DUPLEX_HALF:
8729                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
8730                 break;
8731         case SPEED_100 + DUPLEX_FULL:
8732                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
8733                 break;
8734         case SPEED_1000 + DUPLEX_FULL:
8735                 mac->autoneg = 1;
8736                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
8737                 break;
8738         case SPEED_1000 + DUPLEX_HALF: /* not supported */
8739         default:
8740                 goto err_inval;
8741         }
8742
8743         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
8744         adapter->hw.phy.mdix = AUTO_ALL_MODES;
8745
8746         return 0;
8747
8748 err_inval:
8749         dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
8750         return -EINVAL;
8751 }
8752
8753 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
8754                           bool runtime)
8755 {
8756         struct net_device *netdev = pci_get_drvdata(pdev);
8757         struct igb_adapter *adapter = netdev_priv(netdev);
8758         struct e1000_hw *hw = &adapter->hw;
8759         u32 ctrl, rctl, status;
8760         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
8761         bool wake;
8762
8763         rtnl_lock();
8764         netif_device_detach(netdev);
8765
8766         if (netif_running(netdev))
8767                 __igb_close(netdev, true);
8768
8769         igb_ptp_suspend(adapter);
8770
8771         igb_clear_interrupt_scheme(adapter);
8772         rtnl_unlock();
8773
8774         status = rd32(E1000_STATUS);
8775         if (status & E1000_STATUS_LU)
8776                 wufc &= ~E1000_WUFC_LNKC;
8777
8778         if (wufc) {
8779                 igb_setup_rctl(adapter);
8780                 igb_set_rx_mode(netdev);
8781
8782                 /* turn on all-multi mode if wake on multicast is enabled */
8783                 if (wufc & E1000_WUFC_MC) {
8784                         rctl = rd32(E1000_RCTL);
8785                         rctl |= E1000_RCTL_MPE;
8786                         wr32(E1000_RCTL, rctl);
8787                 }
8788
8789                 ctrl = rd32(E1000_CTRL);
8790                 ctrl |= E1000_CTRL_ADVD3WUC;
8791                 wr32(E1000_CTRL, ctrl);
8792
8793                 /* Allow time for pending master requests to run */
8794                 igb_disable_pcie_master(hw);
8795
8796                 wr32(E1000_WUC, E1000_WUC_PME_EN);
8797                 wr32(E1000_WUFC, wufc);
8798         } else {
8799                 wr32(E1000_WUC, 0);
8800                 wr32(E1000_WUFC, 0);
8801         }
8802
8803         wake = wufc || adapter->en_mng_pt;
8804         if (!wake)
8805                 igb_power_down_link(adapter);
8806         else
8807                 igb_power_up_link(adapter);
8808
8809         if (enable_wake)
8810                 *enable_wake = wake;
8811
8812         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
8813          * would have already happened in close and is redundant.
8814          */
8815         igb_release_hw_control(adapter);
8816
8817         pci_disable_device(pdev);
8818
8819         return 0;
8820 }
8821
8822 static void igb_deliver_wake_packet(struct net_device *netdev)
8823 {
8824         struct igb_adapter *adapter = netdev_priv(netdev);
8825         struct e1000_hw *hw = &adapter->hw;
8826         struct sk_buff *skb;
8827         u32 wupl;
8828
8829         wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
8830
8831         /* WUPM stores only the first 128 bytes of the wake packet.
8832          * Read the packet only if we have the whole thing.
8833          */
8834         if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
8835                 return;
8836
8837         skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
8838         if (!skb)
8839                 return;
8840
8841         skb_put(skb, wupl);
8842
8843         /* Ensure reads are 32-bit aligned */
8844         wupl = roundup(wupl, 4);
8845
8846         memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
8847
8848         skb->protocol = eth_type_trans(skb, netdev);
8849         netif_rx(skb);
8850 }
8851
8852 static int __maybe_unused igb_suspend(struct device *dev)
8853 {
8854         return __igb_shutdown(to_pci_dev(dev), NULL, 0);
8855 }
8856
8857 static int __maybe_unused igb_resume(struct device *dev)
8858 {
8859         struct pci_dev *pdev = to_pci_dev(dev);
8860         struct net_device *netdev = pci_get_drvdata(pdev);
8861         struct igb_adapter *adapter = netdev_priv(netdev);
8862         struct e1000_hw *hw = &adapter->hw;
8863         u32 err, val;
8864
8865         pci_set_power_state(pdev, PCI_D0);
8866         pci_restore_state(pdev);
8867         pci_save_state(pdev);
8868
8869         if (!pci_device_is_present(pdev))
8870                 return -ENODEV;
8871         err = pci_enable_device_mem(pdev);
8872         if (err) {
8873                 dev_err(&pdev->dev,
8874                         "igb: Cannot enable PCI device from suspend\n");
8875                 return err;
8876         }
8877         pci_set_master(pdev);
8878
8879         pci_enable_wake(pdev, PCI_D3hot, 0);
8880         pci_enable_wake(pdev, PCI_D3cold, 0);
8881
8882         if (igb_init_interrupt_scheme(adapter, true)) {
8883                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8884                 return -ENOMEM;
8885         }
8886
8887         igb_reset(adapter);
8888
8889         /* let the f/w know that the h/w is now under the control of the
8890          * driver.
8891          */
8892         igb_get_hw_control(adapter);
8893
8894         val = rd32(E1000_WUS);
8895         if (val & WAKE_PKT_WUS)
8896                 igb_deliver_wake_packet(netdev);
8897
8898         wr32(E1000_WUS, ~0);
8899
8900         rtnl_lock();
8901         if (!err && netif_running(netdev))
8902                 err = __igb_open(netdev, true);
8903
8904         if (!err)
8905                 netif_device_attach(netdev);
8906         rtnl_unlock();
8907
8908         return err;
8909 }
8910
8911 static int __maybe_unused igb_runtime_idle(struct device *dev)
8912 {
8913         struct net_device *netdev = dev_get_drvdata(dev);
8914         struct igb_adapter *adapter = netdev_priv(netdev);
8915
8916         if (!igb_has_link(adapter))
8917                 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
8918
8919         return -EBUSY;
8920 }
8921
8922 static int __maybe_unused igb_runtime_suspend(struct device *dev)
8923 {
8924         return __igb_shutdown(to_pci_dev(dev), NULL, 1);
8925 }
8926
8927 static int __maybe_unused igb_runtime_resume(struct device *dev)
8928 {
8929         return igb_resume(dev);
8930 }
8931
8932 static void igb_shutdown(struct pci_dev *pdev)
8933 {
8934         bool wake;
8935
8936         __igb_shutdown(pdev, &wake, 0);
8937
8938         if (system_state == SYSTEM_POWER_OFF) {
8939                 pci_wake_from_d3(pdev, wake);
8940                 pci_set_power_state(pdev, PCI_D3hot);
8941         }
8942 }
8943
8944 #ifdef CONFIG_PCI_IOV
8945 static int igb_sriov_reinit(struct pci_dev *dev)
8946 {
8947         struct net_device *netdev = pci_get_drvdata(dev);
8948         struct igb_adapter *adapter = netdev_priv(netdev);
8949         struct pci_dev *pdev = adapter->pdev;
8950
8951         rtnl_lock();
8952
8953         if (netif_running(netdev))
8954                 igb_close(netdev);
8955         else
8956                 igb_reset(adapter);
8957
8958         igb_clear_interrupt_scheme(adapter);
8959
8960         igb_init_queue_configuration(adapter);
8961
8962         if (igb_init_interrupt_scheme(adapter, true)) {
8963                 rtnl_unlock();
8964                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8965                 return -ENOMEM;
8966         }
8967
8968         if (netif_running(netdev))
8969                 igb_open(netdev);
8970
8971         rtnl_unlock();
8972
8973         return 0;
8974 }
8975
8976 static int igb_pci_disable_sriov(struct pci_dev *dev)
8977 {
8978         int err = igb_disable_sriov(dev);
8979
8980         if (!err)
8981                 err = igb_sriov_reinit(dev);
8982
8983         return err;
8984 }
8985
8986 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
8987 {
8988         int err = igb_enable_sriov(dev, num_vfs);
8989
8990         if (err)
8991                 goto out;
8992
8993         err = igb_sriov_reinit(dev);
8994         if (!err)
8995                 return num_vfs;
8996
8997 out:
8998         return err;
8999 }
9000
9001 #endif
9002 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9003 {
9004 #ifdef CONFIG_PCI_IOV
9005         if (num_vfs == 0)
9006                 return igb_pci_disable_sriov(dev);
9007         else
9008                 return igb_pci_enable_sriov(dev, num_vfs);
9009 #endif
9010         return 0;
9011 }
9012
9013 /**
9014  *  igb_io_error_detected - called when PCI error is detected
9015  *  @pdev: Pointer to PCI device
9016  *  @state: The current pci connection state
9017  *
9018  *  This function is called after a PCI bus error affecting
9019  *  this device has been detected.
9020  **/
9021 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9022                                               pci_channel_state_t state)
9023 {
9024         struct net_device *netdev = pci_get_drvdata(pdev);
9025         struct igb_adapter *adapter = netdev_priv(netdev);
9026
9027         netif_device_detach(netdev);
9028
9029         if (state == pci_channel_io_perm_failure)
9030                 return PCI_ERS_RESULT_DISCONNECT;
9031
9032         if (netif_running(netdev))
9033                 igb_down(adapter);
9034         pci_disable_device(pdev);
9035
9036         /* Request a slot slot reset. */
9037         return PCI_ERS_RESULT_NEED_RESET;
9038 }
9039
9040 /**
9041  *  igb_io_slot_reset - called after the pci bus has been reset.
9042  *  @pdev: Pointer to PCI device
9043  *
9044  *  Restart the card from scratch, as if from a cold-boot. Implementation
9045  *  resembles the first-half of the igb_resume routine.
9046  **/
9047 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9048 {
9049         struct net_device *netdev = pci_get_drvdata(pdev);
9050         struct igb_adapter *adapter = netdev_priv(netdev);
9051         struct e1000_hw *hw = &adapter->hw;
9052         pci_ers_result_t result;
9053
9054         if (pci_enable_device_mem(pdev)) {
9055                 dev_err(&pdev->dev,
9056                         "Cannot re-enable PCI device after reset.\n");
9057                 result = PCI_ERS_RESULT_DISCONNECT;
9058         } else {
9059                 pci_set_master(pdev);
9060                 pci_restore_state(pdev);
9061                 pci_save_state(pdev);
9062
9063                 pci_enable_wake(pdev, PCI_D3hot, 0);
9064                 pci_enable_wake(pdev, PCI_D3cold, 0);
9065
9066                 /* In case of PCI error, adapter lose its HW address
9067                  * so we should re-assign it here.
9068                  */
9069                 hw->hw_addr = adapter->io_addr;
9070
9071                 igb_reset(adapter);
9072                 wr32(E1000_WUS, ~0);
9073                 result = PCI_ERS_RESULT_RECOVERED;
9074         }
9075
9076         return result;
9077 }
9078
9079 /**
9080  *  igb_io_resume - called when traffic can start flowing again.
9081  *  @pdev: Pointer to PCI device
9082  *
9083  *  This callback is called when the error recovery driver tells us that
9084  *  its OK to resume normal operation. Implementation resembles the
9085  *  second-half of the igb_resume routine.
9086  */
9087 static void igb_io_resume(struct pci_dev *pdev)
9088 {
9089         struct net_device *netdev = pci_get_drvdata(pdev);
9090         struct igb_adapter *adapter = netdev_priv(netdev);
9091
9092         if (netif_running(netdev)) {
9093                 if (igb_up(adapter)) {
9094                         dev_err(&pdev->dev, "igb_up failed after reset\n");
9095                         return;
9096                 }
9097         }
9098
9099         netif_device_attach(netdev);
9100
9101         /* let the f/w know that the h/w is now under the control of the
9102          * driver.
9103          */
9104         igb_get_hw_control(adapter);
9105 }
9106
9107 /**
9108  *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9109  *  @adapter: Pointer to adapter structure
9110  *  @index: Index of the RAR entry which need to be synced with MAC table
9111  **/
9112 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9113 {
9114         struct e1000_hw *hw = &adapter->hw;
9115         u32 rar_low, rar_high;
9116         u8 *addr = adapter->mac_table[index].addr;
9117
9118         /* HW expects these to be in network order when they are plugged
9119          * into the registers which are little endian.  In order to guarantee
9120          * that ordering we need to do an leXX_to_cpup here in order to be
9121          * ready for the byteswap that occurs with writel
9122          */
9123         rar_low = le32_to_cpup((__le32 *)(addr));
9124         rar_high = le16_to_cpup((__le16 *)(addr + 4));
9125
9126         /* Indicate to hardware the Address is Valid. */
9127         if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9128                 if (is_valid_ether_addr(addr))
9129                         rar_high |= E1000_RAH_AV;
9130
9131                 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9132                         rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9133
9134                 switch (hw->mac.type) {
9135                 case e1000_82575:
9136                 case e1000_i210:
9137                         if (adapter->mac_table[index].state &
9138                             IGB_MAC_STATE_QUEUE_STEERING)
9139                                 rar_high |= E1000_RAH_QSEL_ENABLE;
9140
9141                         rar_high |= E1000_RAH_POOL_1 *
9142                                     adapter->mac_table[index].queue;
9143                         break;
9144                 default:
9145                         rar_high |= E1000_RAH_POOL_1 <<
9146                                     adapter->mac_table[index].queue;
9147                         break;
9148                 }
9149         }
9150
9151         wr32(E1000_RAL(index), rar_low);
9152         wrfl();
9153         wr32(E1000_RAH(index), rar_high);
9154         wrfl();
9155 }
9156
9157 static int igb_set_vf_mac(struct igb_adapter *adapter,
9158                           int vf, unsigned char *mac_addr)
9159 {
9160         struct e1000_hw *hw = &adapter->hw;
9161         /* VF MAC addresses start at end of receive addresses and moves
9162          * towards the first, as a result a collision should not be possible
9163          */
9164         int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9165         unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9166
9167         ether_addr_copy(vf_mac_addr, mac_addr);
9168         ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9169         adapter->mac_table[rar_entry].queue = vf;
9170         adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9171         igb_rar_set_index(adapter, rar_entry);
9172
9173         return 0;
9174 }
9175
9176 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9177 {
9178         struct igb_adapter *adapter = netdev_priv(netdev);
9179
9180         if (vf >= adapter->vfs_allocated_count)
9181                 return -EINVAL;
9182
9183         /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9184          * flag and allows to overwrite the MAC via VF netdev.  This
9185          * is necessary to allow libvirt a way to restore the original
9186          * MAC after unbinding vfio-pci and reloading igbvf after shutting
9187          * down a VM.
9188          */
9189         if (is_zero_ether_addr(mac)) {
9190                 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9191                 dev_info(&adapter->pdev->dev,
9192                          "remove administratively set MAC on VF %d\n",
9193                          vf);
9194         } else if (is_valid_ether_addr(mac)) {
9195                 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9196                 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9197                          mac, vf);
9198                 dev_info(&adapter->pdev->dev,
9199                          "Reload the VF driver to make this change effective.");
9200                 /* Generate additional warning if PF is down */
9201                 if (test_bit(__IGB_DOWN, &adapter->state)) {
9202                         dev_warn(&adapter->pdev->dev,
9203                                  "The VF MAC address has been set, but the PF device is not up.\n");
9204                         dev_warn(&adapter->pdev->dev,
9205                                  "Bring the PF device up before attempting to use the VF device.\n");
9206                 }
9207         } else {
9208                 return -EINVAL;
9209         }
9210         return igb_set_vf_mac(adapter, vf, mac);
9211 }
9212
9213 static int igb_link_mbps(int internal_link_speed)
9214 {
9215         switch (internal_link_speed) {
9216         case SPEED_100:
9217                 return 100;
9218         case SPEED_1000:
9219                 return 1000;
9220         default:
9221                 return 0;
9222         }
9223 }
9224
9225 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9226                                   int link_speed)
9227 {
9228         int rf_dec, rf_int;
9229         u32 bcnrc_val;
9230
9231         if (tx_rate != 0) {
9232                 /* Calculate the rate factor values to set */
9233                 rf_int = link_speed / tx_rate;
9234                 rf_dec = (link_speed - (rf_int * tx_rate));
9235                 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9236                          tx_rate;
9237
9238                 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9239                 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
9240                               E1000_RTTBCNRC_RF_INT_MASK);
9241                 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9242         } else {
9243                 bcnrc_val = 0;
9244         }
9245
9246         wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9247         /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9248          * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9249          */
9250         wr32(E1000_RTTBCNRM, 0x14);
9251         wr32(E1000_RTTBCNRC, bcnrc_val);
9252 }
9253
9254 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9255 {
9256         int actual_link_speed, i;
9257         bool reset_rate = false;
9258
9259         /* VF TX rate limit was not set or not supported */
9260         if ((adapter->vf_rate_link_speed == 0) ||
9261             (adapter->hw.mac.type != e1000_82576))
9262                 return;
9263
9264         actual_link_speed = igb_link_mbps(adapter->link_speed);
9265         if (actual_link_speed != adapter->vf_rate_link_speed) {
9266                 reset_rate = true;
9267                 adapter->vf_rate_link_speed = 0;
9268                 dev_info(&adapter->pdev->dev,
9269                          "Link speed has been changed. VF Transmit rate is disabled\n");
9270         }
9271
9272         for (i = 0; i < adapter->vfs_allocated_count; i++) {
9273                 if (reset_rate)
9274                         adapter->vf_data[i].tx_rate = 0;
9275
9276                 igb_set_vf_rate_limit(&adapter->hw, i,
9277                                       adapter->vf_data[i].tx_rate,
9278                                       actual_link_speed);
9279         }
9280 }
9281
9282 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9283                              int min_tx_rate, int max_tx_rate)
9284 {
9285         struct igb_adapter *adapter = netdev_priv(netdev);
9286         struct e1000_hw *hw = &adapter->hw;
9287         int actual_link_speed;
9288
9289         if (hw->mac.type != e1000_82576)
9290                 return -EOPNOTSUPP;
9291
9292         if (min_tx_rate)
9293                 return -EINVAL;
9294
9295         actual_link_speed = igb_link_mbps(adapter->link_speed);
9296         if ((vf >= adapter->vfs_allocated_count) ||
9297             (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9298             (max_tx_rate < 0) ||
9299             (max_tx_rate > actual_link_speed))
9300                 return -EINVAL;
9301
9302         adapter->vf_rate_link_speed = actual_link_speed;
9303         adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9304         igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9305
9306         return 0;
9307 }
9308
9309 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9310                                    bool setting)
9311 {
9312         struct igb_adapter *adapter = netdev_priv(netdev);
9313         struct e1000_hw *hw = &adapter->hw;
9314         u32 reg_val, reg_offset;
9315
9316         if (!adapter->vfs_allocated_count)
9317                 return -EOPNOTSUPP;
9318
9319         if (vf >= adapter->vfs_allocated_count)
9320                 return -EINVAL;
9321
9322         reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9323         reg_val = rd32(reg_offset);
9324         if (setting)
9325                 reg_val |= (BIT(vf) |
9326                             BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9327         else
9328                 reg_val &= ~(BIT(vf) |
9329                              BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9330         wr32(reg_offset, reg_val);
9331
9332         adapter->vf_data[vf].spoofchk_enabled = setting;
9333         return 0;
9334 }
9335
9336 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9337 {
9338         struct igb_adapter *adapter = netdev_priv(netdev);
9339
9340         if (vf >= adapter->vfs_allocated_count)
9341                 return -EINVAL;
9342         if (adapter->vf_data[vf].trusted == setting)
9343                 return 0;
9344
9345         adapter->vf_data[vf].trusted = setting;
9346
9347         dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9348                  vf, setting ? "" : "not ");
9349         return 0;
9350 }
9351
9352 static int igb_ndo_get_vf_config(struct net_device *netdev,
9353                                  int vf, struct ifla_vf_info *ivi)
9354 {
9355         struct igb_adapter *adapter = netdev_priv(netdev);
9356         if (vf >= adapter->vfs_allocated_count)
9357                 return -EINVAL;
9358         ivi->vf = vf;
9359         memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9360         ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9361         ivi->min_tx_rate = 0;
9362         ivi->vlan = adapter->vf_data[vf].pf_vlan;
9363         ivi->qos = adapter->vf_data[vf].pf_qos;
9364         ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9365         ivi->trusted = adapter->vf_data[vf].trusted;
9366         return 0;
9367 }
9368
9369 static void igb_vmm_control(struct igb_adapter *adapter)
9370 {
9371         struct e1000_hw *hw = &adapter->hw;
9372         u32 reg;
9373
9374         switch (hw->mac.type) {
9375         case e1000_82575:
9376         case e1000_i210:
9377         case e1000_i211:
9378         case e1000_i354:
9379         default:
9380                 /* replication is not supported for 82575 */
9381                 return;
9382         case e1000_82576:
9383                 /* notify HW that the MAC is adding vlan tags */
9384                 reg = rd32(E1000_DTXCTL);
9385                 reg |= E1000_DTXCTL_VLAN_ADDED;
9386                 wr32(E1000_DTXCTL, reg);
9387                 /* Fall through */
9388         case e1000_82580:
9389                 /* enable replication vlan tag stripping */
9390                 reg = rd32(E1000_RPLOLR);
9391                 reg |= E1000_RPLOLR_STRVLAN;
9392                 wr32(E1000_RPLOLR, reg);
9393                 /* Fall through */
9394         case e1000_i350:
9395                 /* none of the above registers are supported by i350 */
9396                 break;
9397         }
9398
9399         if (adapter->vfs_allocated_count) {
9400                 igb_vmdq_set_loopback_pf(hw, true);
9401                 igb_vmdq_set_replication_pf(hw, true);
9402                 igb_vmdq_set_anti_spoofing_pf(hw, true,
9403                                               adapter->vfs_allocated_count);
9404         } else {
9405                 igb_vmdq_set_loopback_pf(hw, false);
9406                 igb_vmdq_set_replication_pf(hw, false);
9407         }
9408 }
9409
9410 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9411 {
9412         struct e1000_hw *hw = &adapter->hw;
9413         u32 dmac_thr;
9414         u16 hwm;
9415
9416         if (hw->mac.type > e1000_82580) {
9417                 if (adapter->flags & IGB_FLAG_DMAC) {
9418                         u32 reg;
9419
9420                         /* force threshold to 0. */
9421                         wr32(E1000_DMCTXTH, 0);
9422
9423                         /* DMA Coalescing high water mark needs to be greater
9424                          * than the Rx threshold. Set hwm to PBA - max frame
9425                          * size in 16B units, capping it at PBA - 6KB.
9426                          */
9427                         hwm = 64 * (pba - 6);
9428                         reg = rd32(E1000_FCRTC);
9429                         reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9430                         reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9431                                 & E1000_FCRTC_RTH_COAL_MASK);
9432                         wr32(E1000_FCRTC, reg);
9433
9434                         /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9435                          * frame size, capping it at PBA - 10KB.
9436                          */
9437                         dmac_thr = pba - 10;
9438                         reg = rd32(E1000_DMACR);
9439                         reg &= ~E1000_DMACR_DMACTHR_MASK;
9440                         reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9441                                 & E1000_DMACR_DMACTHR_MASK);
9442
9443                         /* transition to L0x or L1 if available..*/
9444                         reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
9445
9446                         /* watchdog timer= +-1000 usec in 32usec intervals */
9447                         reg |= (1000 >> 5);
9448
9449                         /* Disable BMC-to-OS Watchdog Enable */
9450                         if (hw->mac.type != e1000_i354)
9451                                 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
9452
9453                         wr32(E1000_DMACR, reg);
9454
9455                         /* no lower threshold to disable
9456                          * coalescing(smart fifb)-UTRESH=0
9457                          */
9458                         wr32(E1000_DMCRTRH, 0);
9459
9460                         reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
9461
9462                         wr32(E1000_DMCTLX, reg);
9463
9464                         /* free space in tx packet buffer to wake from
9465                          * DMA coal
9466                          */
9467                         wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
9468                              (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
9469
9470                         /* make low power state decision controlled
9471                          * by DMA coal
9472                          */
9473                         reg = rd32(E1000_PCIEMISC);
9474                         reg &= ~E1000_PCIEMISC_LX_DECISION;
9475                         wr32(E1000_PCIEMISC, reg);
9476                 } /* endif adapter->dmac is not disabled */
9477         } else if (hw->mac.type == e1000_82580) {
9478                 u32 reg = rd32(E1000_PCIEMISC);
9479
9480                 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
9481                 wr32(E1000_DMACR, 0);
9482         }
9483 }
9484
9485 /**
9486  *  igb_read_i2c_byte - Reads 8 bit word over I2C
9487  *  @hw: pointer to hardware structure
9488  *  @byte_offset: byte offset to read
9489  *  @dev_addr: device address
9490  *  @data: value read
9491  *
9492  *  Performs byte read operation over I2C interface at
9493  *  a specified device address.
9494  **/
9495 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9496                       u8 dev_addr, u8 *data)
9497 {
9498         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9499         struct i2c_client *this_client = adapter->i2c_client;
9500         s32 status;
9501         u16 swfw_mask = 0;
9502
9503         if (!this_client)
9504                 return E1000_ERR_I2C;
9505
9506         swfw_mask = E1000_SWFW_PHY0_SM;
9507
9508         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9509                 return E1000_ERR_SWFW_SYNC;
9510
9511         status = i2c_smbus_read_byte_data(this_client, byte_offset);
9512         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9513
9514         if (status < 0)
9515                 return E1000_ERR_I2C;
9516         else {
9517                 *data = status;
9518                 return 0;
9519         }
9520 }
9521
9522 /**
9523  *  igb_write_i2c_byte - Writes 8 bit word over I2C
9524  *  @hw: pointer to hardware structure
9525  *  @byte_offset: byte offset to write
9526  *  @dev_addr: device address
9527  *  @data: value to write
9528  *
9529  *  Performs byte write operation over I2C interface at
9530  *  a specified device address.
9531  **/
9532 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9533                        u8 dev_addr, u8 data)
9534 {
9535         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9536         struct i2c_client *this_client = adapter->i2c_client;
9537         s32 status;
9538         u16 swfw_mask = E1000_SWFW_PHY0_SM;
9539
9540         if (!this_client)
9541                 return E1000_ERR_I2C;
9542
9543         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9544                 return E1000_ERR_SWFW_SYNC;
9545         status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
9546         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9547
9548         if (status)
9549                 return E1000_ERR_I2C;
9550         else
9551                 return 0;
9552
9553 }
9554
9555 int igb_reinit_queues(struct igb_adapter *adapter)
9556 {
9557         struct net_device *netdev = adapter->netdev;
9558         struct pci_dev *pdev = adapter->pdev;
9559         int err = 0;
9560
9561         if (netif_running(netdev))
9562                 igb_close(netdev);
9563
9564         igb_reset_interrupt_capability(adapter);
9565
9566         if (igb_init_interrupt_scheme(adapter, true)) {
9567                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9568                 return -ENOMEM;
9569         }
9570
9571         if (netif_running(netdev))
9572                 err = igb_open(netdev);
9573
9574         return err;
9575 }
9576
9577 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
9578 {
9579         struct igb_nfc_filter *rule;
9580
9581         spin_lock(&adapter->nfc_lock);
9582
9583         hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9584                 igb_erase_filter(adapter, rule);
9585
9586         hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
9587                 igb_erase_filter(adapter, rule);
9588
9589         spin_unlock(&adapter->nfc_lock);
9590 }
9591
9592 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
9593 {
9594         struct igb_nfc_filter *rule;
9595
9596         spin_lock(&adapter->nfc_lock);
9597
9598         hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9599                 igb_add_filter(adapter, rule);
9600
9601         spin_unlock(&adapter->nfc_lock);
9602 }
9603 /* igb_main.c */