linux/drivers/net/ethernet/intel/igb/igb_main.c
<<
>>
Prefs
   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 4
  43#define BUILD 0
  44#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
  45__stringify(BUILD) "-k"
  46
  47enum queue_mode {
  48        QUEUE_MODE_STRICT_PRIORITY,
  49        QUEUE_MODE_STREAM_RESERVATION,
  50};
  51
  52enum tx_queue_prio {
  53        TX_QUEUE_PRIO_HIGH,
  54        TX_QUEUE_PRIO_LOW,
  55};
  56
  57char igb_driver_name[] = "igb";
  58char igb_driver_version[] = DRV_VERSION;
  59static const char igb_driver_string[] =
  60                                "Intel(R) Gigabit Ethernet Network Driver";
  61static const char igb_copyright[] =
  62                                "Copyright (c) 2007-2014 Intel Corporation.";
  63
  64static const struct e1000_info *igb_info_tbl[] = {
  65        [board_82575] = &e1000_82575_info,
  66};
  67
  68static 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
 108MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 109
 110static int igb_setup_all_tx_resources(struct igb_adapter *);
 111static int igb_setup_all_rx_resources(struct igb_adapter *);
 112static void igb_free_all_tx_resources(struct igb_adapter *);
 113static void igb_free_all_rx_resources(struct igb_adapter *);
 114static void igb_setup_mrqc(struct igb_adapter *);
 115static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 116static void igb_remove(struct pci_dev *pdev);
 117static int igb_sw_init(struct igb_adapter *);
 118int igb_open(struct net_device *);
 119int igb_close(struct net_device *);
 120static void igb_configure(struct igb_adapter *);
 121static void igb_configure_tx(struct igb_adapter *);
 122static void igb_configure_rx(struct igb_adapter *);
 123static void igb_clean_all_tx_rings(struct igb_adapter *);
 124static void igb_clean_all_rx_rings(struct igb_adapter *);
 125static void igb_clean_tx_ring(struct igb_ring *);
 126static void igb_clean_rx_ring(struct igb_ring *);
 127static void igb_set_rx_mode(struct net_device *);
 128static void igb_update_phy_info(struct timer_list *);
 129static void igb_watchdog(struct timer_list *);
 130static void igb_watchdog_task(struct work_struct *);
 131static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 132static void igb_get_stats64(struct net_device *dev,
 133                            struct rtnl_link_stats64 *stats);
 134static int igb_change_mtu(struct net_device *, int);
 135static int igb_set_mac(struct net_device *, void *);
 136static void igb_set_uta(struct igb_adapter *adapter, bool set);
 137static irqreturn_t igb_intr(int irq, void *);
 138static irqreturn_t igb_intr_msi(int irq, void *);
 139static irqreturn_t igb_msix_other(int irq, void *);
 140static irqreturn_t igb_msix_ring(int irq, void *);
 141#ifdef CONFIG_IGB_DCA
 142static void igb_update_dca(struct igb_q_vector *);
 143static void igb_setup_dca(struct igb_adapter *);
 144#endif /* CONFIG_IGB_DCA */
 145static int igb_poll(struct napi_struct *, int);
 146static bool igb_clean_tx_irq(struct igb_q_vector *, int);
 147static int igb_clean_rx_irq(struct igb_q_vector *, int);
 148static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 149static void igb_tx_timeout(struct net_device *);
 150static void igb_reset_task(struct work_struct *);
 151static void igb_vlan_mode(struct net_device *netdev,
 152                          netdev_features_t features);
 153static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
 154static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
 155static void igb_restore_vlan(struct igb_adapter *);
 156static void igb_rar_set_index(struct igb_adapter *, u32);
 157static void igb_ping_all_vfs(struct igb_adapter *);
 158static void igb_msg_task(struct igb_adapter *);
 159static void igb_vmm_control(struct igb_adapter *);
 160static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 161static void igb_flush_mac_table(struct igb_adapter *);
 162static int igb_available_rars(struct igb_adapter *, u8);
 163static void igb_set_default_mac_filter(struct igb_adapter *);
 164static int igb_uc_sync(struct net_device *, const unsigned char *);
 165static int igb_uc_unsync(struct net_device *, const unsigned char *);
 166static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 167static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 168static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 169                               int vf, u16 vlan, u8 qos, __be16 vlan_proto);
 170static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
 171static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 172                                   bool setting);
 173static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
 174                                bool setting);
 175static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 176                                 struct ifla_vf_info *ivi);
 177static void igb_check_vf_rate_limit(struct igb_adapter *);
 178static void igb_nfc_filter_exit(struct igb_adapter *adapter);
 179static void igb_nfc_filter_restore(struct igb_adapter *adapter);
 180
 181#ifdef CONFIG_PCI_IOV
 182static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 183static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
 184static int igb_disable_sriov(struct pci_dev *dev);
 185static int igb_pci_disable_sriov(struct pci_dev *dev);
 186#endif
 187
 188static int igb_suspend(struct device *);
 189static int igb_resume(struct device *);
 190static int igb_runtime_suspend(struct device *dev);
 191static int igb_runtime_resume(struct device *dev);
 192static int igb_runtime_idle(struct device *dev);
 193static 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};
 198static void igb_shutdown(struct pci_dev *);
 199static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 200#ifdef CONFIG_IGB_DCA
 201static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 202static 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
 209static unsigned int max_vfs;
 210module_param(max_vfs, uint, 0);
 211MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
 212#endif /* CONFIG_PCI_IOV */
 213
 214static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 215                     pci_channel_state_t);
 216static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 217static void igb_io_resume(struct pci_dev *);
 218
 219static 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
 225static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 226
 227static 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
 240MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 241MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 242MODULE_LICENSE("GPL v2");
 243MODULE_VERSION(DRV_VERSION);
 244
 245#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 246static int debug = -1;
 247module_param(debug, int, 0);
 248MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 249
 250struct igb_reg_info {
 251        u32 ofs;
 252        char *name;
 253};
 254
 255static 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 */
 292static 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 */
 358static 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 */
 467rx_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
 554exit:
 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 **/
 565static 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 **/
 581static 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 **/
 606static 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 **/
 629static 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
 638static 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 **/
 653struct 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 **/
 665static 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
 680module_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 **/
 688static 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
 696module_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 **/
 706static 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
 740u32 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        }
 757
 758        return value;
 759}
 760
 761/**
 762 *  igb_write_ivar - configure ivar for given MSI-X vector
 763 *  @hw: pointer to the HW structure
 764 *  @msix_vector: vector number we are allocating to a given ring
 765 *  @index: row index of IVAR register to write within IVAR table
 766 *  @offset: column offset of in IVAR, should be multiple of 8
 767 *
 768 *  This function is intended to handle the writing of the IVAR register
 769 *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 770 *  each containing an cause allocation for an Rx and Tx ring, and a
 771 *  variable number of rows depending on the number of queues supported.
 772 **/
 773static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 774                           int index, int offset)
 775{
 776        u32 ivar = array_rd32(E1000_IVAR0, index);
 777
 778        /* clear any bits that are currently set */
 779        ivar &= ~((u32)0xFF << offset);
 780
 781        /* write vector and valid bit */
 782        ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 783
 784        array_wr32(E1000_IVAR0, index, ivar);
 785}
 786
 787#define IGB_N0_QUEUE -1
 788static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 789{
 790        struct igb_adapter *adapter = q_vector->adapter;
 791        struct e1000_hw *hw = &adapter->hw;
 792        int rx_queue = IGB_N0_QUEUE;
 793        int tx_queue = IGB_N0_QUEUE;
 794        u32 msixbm = 0;
 795
 796        if (q_vector->rx.ring)
 797                rx_queue = q_vector->rx.ring->reg_idx;
 798        if (q_vector->tx.ring)
 799                tx_queue = q_vector->tx.ring->reg_idx;
 800
 801        switch (hw->mac.type) {
 802        case e1000_82575:
 803                /* The 82575 assigns vectors using a bitmask, which matches the
 804                 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
 805                 * or more queues to a vector, we write the appropriate bits
 806                 * into the MSIXBM register for that vector.
 807                 */
 808                if (rx_queue > IGB_N0_QUEUE)
 809                        msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 810                if (tx_queue > IGB_N0_QUEUE)
 811                        msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 812                if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
 813                        msixbm |= E1000_EIMS_OTHER;
 814                array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 815                q_vector->eims_value = msixbm;
 816                break;
 817        case e1000_82576:
 818                /* 82576 uses a table that essentially consists of 2 columns
 819                 * with 8 rows.  The ordering is column-major so we use the
 820                 * lower 3 bits as the row index, and the 4th bit as the
 821                 * column offset.
 822                 */
 823                if (rx_queue > IGB_N0_QUEUE)
 824                        igb_write_ivar(hw, msix_vector,
 825                                       rx_queue & 0x7,
 826                                       (rx_queue & 0x8) << 1);
 827                if (tx_queue > IGB_N0_QUEUE)
 828                        igb_write_ivar(hw, msix_vector,
 829                                       tx_queue & 0x7,
 830                                       ((tx_queue & 0x8) << 1) + 8);
 831                q_vector->eims_value = BIT(msix_vector);
 832                break;
 833        case e1000_82580:
 834        case e1000_i350:
 835        case e1000_i354:
 836        case e1000_i210:
 837        case e1000_i211:
 838                /* On 82580 and newer adapters the scheme is similar to 82576
 839                 * however instead of ordering column-major we have things
 840                 * ordered row-major.  So we traverse the table by using
 841                 * bit 0 as the column offset, and the remaining bits as the
 842                 * row index.
 843                 */
 844                if (rx_queue > IGB_N0_QUEUE)
 845                        igb_write_ivar(hw, msix_vector,
 846                                       rx_queue >> 1,
 847                                       (rx_queue & 0x1) << 4);
 848                if (tx_queue > IGB_N0_QUEUE)
 849                        igb_write_ivar(hw, msix_vector,
 850                                       tx_queue >> 1,
 851                                       ((tx_queue & 0x1) << 4) + 8);
 852                q_vector->eims_value = BIT(msix_vector);
 853                break;
 854        default:
 855                BUG();
 856                break;
 857        }
 858
 859        /* add q_vector eims value to global eims_enable_mask */
 860        adapter->eims_enable_mask |= q_vector->eims_value;
 861
 862        /* configure q_vector to set itr on first interrupt */
 863        q_vector->set_itr = 1;
 864}
 865
 866/**
 867 *  igb_configure_msix - Configure MSI-X hardware
 868 *  @adapter: board private structure to initialize
 869 *
 870 *  igb_configure_msix sets up the hardware to properly
 871 *  generate MSI-X interrupts.
 872 **/
 873static void igb_configure_msix(struct igb_adapter *adapter)
 874{
 875        u32 tmp;
 876        int i, vector = 0;
 877        struct e1000_hw *hw = &adapter->hw;
 878
 879        adapter->eims_enable_mask = 0;
 880
 881        /* set vector for other causes, i.e. link changes */
 882        switch (hw->mac.type) {
 883        case e1000_82575:
 884                tmp = rd32(E1000_CTRL_EXT);
 885                /* enable MSI-X PBA support*/
 886                tmp |= E1000_CTRL_EXT_PBA_CLR;
 887
 888                /* Auto-Mask interrupts upon ICR read. */
 889                tmp |= E1000_CTRL_EXT_EIAME;
 890                tmp |= E1000_CTRL_EXT_IRCA;
 891
 892                wr32(E1000_CTRL_EXT, tmp);
 893
 894                /* enable msix_other interrupt */
 895                array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
 896                adapter->eims_other = E1000_EIMS_OTHER;
 897
 898                break;
 899
 900        case e1000_82576:
 901        case e1000_82580:
 902        case e1000_i350:
 903        case e1000_i354:
 904        case e1000_i210:
 905        case e1000_i211:
 906                /* Turn on MSI-X capability first, or our settings
 907                 * won't stick.  And it will take days to debug.
 908                 */
 909                wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 910                     E1000_GPIE_PBA | E1000_GPIE_EIAME |
 911                     E1000_GPIE_NSICR);
 912
 913                /* enable msix_other interrupt */
 914                adapter->eims_other = BIT(vector);
 915                tmp = (vector++ | E1000_IVAR_VALID) << 8;
 916
 917                wr32(E1000_IVAR_MISC, tmp);
 918                break;
 919        default:
 920                /* do nothing, since nothing else supports MSI-X */
 921                break;
 922        } /* switch (hw->mac.type) */
 923
 924        adapter->eims_enable_mask |= adapter->eims_other;
 925
 926        for (i = 0; i < adapter->num_q_vectors; i++)
 927                igb_assign_vector(adapter->q_vector[i], vector++);
 928
 929        wrfl();
 930}
 931
 932/**
 933 *  igb_request_msix - Initialize MSI-X interrupts
 934 *  @adapter: board private structure to initialize
 935 *
 936 *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
 937 *  kernel.
 938 **/
 939static int igb_request_msix(struct igb_adapter *adapter)
 940{
 941        struct net_device *netdev = adapter->netdev;
 942        int i, err = 0, vector = 0, free_vector = 0;
 943
 944        err = request_irq(adapter->msix_entries[vector].vector,
 945                          igb_msix_other, 0, netdev->name, adapter);
 946        if (err)
 947                goto err_out;
 948
 949        for (i = 0; i < adapter->num_q_vectors; i++) {
 950                struct igb_q_vector *q_vector = adapter->q_vector[i];
 951
 952                vector++;
 953
 954                q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
 955
 956                if (q_vector->rx.ring && q_vector->tx.ring)
 957                        sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 958                                q_vector->rx.ring->queue_index);
 959                else if (q_vector->tx.ring)
 960                        sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 961                                q_vector->tx.ring->queue_index);
 962                else if (q_vector->rx.ring)
 963                        sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 964                                q_vector->rx.ring->queue_index);
 965                else
 966                        sprintf(q_vector->name, "%s-unused", netdev->name);
 967
 968                err = request_irq(adapter->msix_entries[vector].vector,
 969                                  igb_msix_ring, 0, q_vector->name,
 970                                  q_vector);
 971                if (err)
 972                        goto err_free;
 973        }
 974
 975        igb_configure_msix(adapter);
 976        return 0;
 977
 978err_free:
 979        /* free already assigned IRQs */
 980        free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 981
 982        vector--;
 983        for (i = 0; i < vector; i++) {
 984                free_irq(adapter->msix_entries[free_vector++].vector,
 985                         adapter->q_vector[i]);
 986        }
 987err_out:
 988        return err;
 989}
 990
 991/**
 992 *  igb_free_q_vector - Free memory allocated for specific interrupt vector
 993 *  @adapter: board private structure to initialize
 994 *  @v_idx: Index of vector to be freed
 995 *
 996 *  This function frees the memory allocated to the q_vector.
 997 **/
 998static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
 999{
1000        struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1001
1002        adapter->q_vector[v_idx] = NULL;
1003
1004        /* igb_get_stats64() might access the rings on this vector,
1005         * we must wait a grace period before freeing it.
1006         */
1007        if (q_vector)
1008                kfree_rcu(q_vector, rcu);
1009}
1010
1011/**
1012 *  igb_reset_q_vector - Reset config for interrupt vector
1013 *  @adapter: board private structure to initialize
1014 *  @v_idx: Index of vector to be reset
1015 *
1016 *  If NAPI is enabled it will delete any references to the
1017 *  NAPI struct. This is preparation for igb_free_q_vector.
1018 **/
1019static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1020{
1021        struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1022
1023        /* Coming from igb_set_interrupt_capability, the vectors are not yet
1024         * allocated. So, q_vector is NULL so we should stop here.
1025         */
1026        if (!q_vector)
1027                return;
1028
1029        if (q_vector->tx.ring)
1030                adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1031
1032        if (q_vector->rx.ring)
1033                adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1034
1035        netif_napi_del(&q_vector->napi);
1036
1037}
1038
1039static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1040{
1041        int v_idx = adapter->num_q_vectors;
1042
1043        if (adapter->flags & IGB_FLAG_HAS_MSIX)
1044                pci_disable_msix(adapter->pdev);
1045        else if (adapter->flags & IGB_FLAG_HAS_MSI)
1046                pci_disable_msi(adapter->pdev);
1047
1048        while (v_idx--)
1049                igb_reset_q_vector(adapter, v_idx);
1050}
1051
1052/**
1053 *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1054 *  @adapter: board private structure to initialize
1055 *
1056 *  This function frees the memory allocated to the q_vectors.  In addition if
1057 *  NAPI is enabled it will delete any references to the NAPI struct prior
1058 *  to freeing the q_vector.
1059 **/
1060static void igb_free_q_vectors(struct igb_adapter *adapter)
1061{
1062        int v_idx = adapter->num_q_vectors;
1063
1064        adapter->num_tx_queues = 0;
1065        adapter->num_rx_queues = 0;
1066        adapter->num_q_vectors = 0;
1067
1068        while (v_idx--) {
1069                igb_reset_q_vector(adapter, v_idx);
1070                igb_free_q_vector(adapter, v_idx);
1071        }
1072}
1073
1074/**
1075 *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1076 *  @adapter: board private structure to initialize
1077 *
1078 *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1079 *  MSI-X interrupts allocated.
1080 */
1081static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1082{
1083        igb_free_q_vectors(adapter);
1084        igb_reset_interrupt_capability(adapter);
1085}
1086
1087/**
1088 *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1089 *  @adapter: board private structure to initialize
1090 *  @msix: boolean value of MSIX capability
1091 *
1092 *  Attempt to configure interrupts using the best available
1093 *  capabilities of the hardware and kernel.
1094 **/
1095static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1096{
1097        int err;
1098        int numvecs, i;
1099
1100        if (!msix)
1101                goto msi_only;
1102        adapter->flags |= IGB_FLAG_HAS_MSIX;
1103
1104        /* Number of supported queues. */
1105        adapter->num_rx_queues = adapter->rss_queues;
1106        if (adapter->vfs_allocated_count)
1107                adapter->num_tx_queues = 1;
1108        else
1109                adapter->num_tx_queues = adapter->rss_queues;
1110
1111        /* start with one vector for every Rx queue */
1112        numvecs = adapter->num_rx_queues;
1113
1114        /* if Tx handler is separate add 1 for every Tx queue */
1115        if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1116                numvecs += adapter->num_tx_queues;
1117
1118        /* store the number of vectors reserved for queues */
1119        adapter->num_q_vectors = numvecs;
1120
1121        /* add 1 vector for link status interrupts */
1122        numvecs++;
1123        for (i = 0; i < numvecs; i++)
1124                adapter->msix_entries[i].entry = i;
1125
1126        err = pci_enable_msix_range(adapter->pdev,
1127                                    adapter->msix_entries,
1128                                    numvecs,
1129                                    numvecs);
1130        if (err > 0)
1131                return;
1132
1133        igb_reset_interrupt_capability(adapter);
1134
1135        /* If we can't do MSI-X, try MSI */
1136msi_only:
1137        adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1138#ifdef CONFIG_PCI_IOV
1139        /* disable SR-IOV for non MSI-X configurations */
1140        if (adapter->vf_data) {
1141                struct e1000_hw *hw = &adapter->hw;
1142                /* disable iov and allow time for transactions to clear */
1143                pci_disable_sriov(adapter->pdev);
1144                msleep(500);
1145
1146                kfree(adapter->vf_mac_list);
1147                adapter->vf_mac_list = NULL;
1148                kfree(adapter->vf_data);
1149                adapter->vf_data = NULL;
1150                wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1151                wrfl();
1152                msleep(100);
1153                dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1154        }
1155#endif
1156        adapter->vfs_allocated_count = 0;
1157        adapter->rss_queues = 1;
1158        adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1159        adapter->num_rx_queues = 1;
1160        adapter->num_tx_queues = 1;
1161        adapter->num_q_vectors = 1;
1162        if (!pci_enable_msi(adapter->pdev))
1163                adapter->flags |= IGB_FLAG_HAS_MSI;
1164}
1165
1166static void igb_add_ring(struct igb_ring *ring,
1167                         struct igb_ring_container *head)
1168{
1169        head->ring = ring;
1170        head->count++;
1171}
1172
1173/**
1174 *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1175 *  @adapter: board private structure to initialize
1176 *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1177 *  @v_idx: index of vector in adapter struct
1178 *  @txr_count: total number of Tx rings to allocate
1179 *  @txr_idx: index of first Tx ring to allocate
1180 *  @rxr_count: total number of Rx rings to allocate
1181 *  @rxr_idx: index of first Rx ring to allocate
1182 *
1183 *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1184 **/
1185static int igb_alloc_q_vector(struct igb_adapter *adapter,
1186                              int v_count, int v_idx,
1187                              int txr_count, int txr_idx,
1188                              int rxr_count, int rxr_idx)
1189{
1190        struct igb_q_vector *q_vector;
1191        struct igb_ring *ring;
1192        int ring_count, size;
1193
1194        /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1195        if (txr_count > 1 || rxr_count > 1)
1196                return -ENOMEM;
1197
1198        ring_count = txr_count + rxr_count;
1199        size = sizeof(struct igb_q_vector) +
1200               (sizeof(struct igb_ring) * ring_count);
1201
1202        /* allocate q_vector and rings */
1203        q_vector = adapter->q_vector[v_idx];
1204        if (!q_vector) {
1205                q_vector = kzalloc(size, GFP_KERNEL);
1206        } else if (size > ksize(q_vector)) {
1207                kfree_rcu(q_vector, rcu);
1208                q_vector = kzalloc(size, GFP_KERNEL);
1209        } else {
1210                memset(q_vector, 0, size);
1211        }
1212        if (!q_vector)
1213                return -ENOMEM;
1214
1215        /* initialize NAPI */
1216        netif_napi_add(adapter->netdev, &q_vector->napi,
1217                       igb_poll, 64);
1218
1219        /* tie q_vector and adapter together */
1220        adapter->q_vector[v_idx] = q_vector;
1221        q_vector->adapter = adapter;
1222
1223        /* initialize work limits */
1224        q_vector->tx.work_limit = adapter->tx_work_limit;
1225
1226        /* initialize ITR configuration */
1227        q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1228        q_vector->itr_val = IGB_START_ITR;
1229
1230        /* initialize pointer to rings */
1231        ring = q_vector->ring;
1232
1233        /* intialize ITR */
1234        if (rxr_count) {
1235                /* rx or rx/tx vector */
1236                if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1237                        q_vector->itr_val = adapter->rx_itr_setting;
1238        } else {
1239                /* tx only vector */
1240                if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1241                        q_vector->itr_val = adapter->tx_itr_setting;
1242        }
1243
1244        if (txr_count) {
1245                /* assign generic ring traits */
1246                ring->dev = &adapter->pdev->dev;
1247                ring->netdev = adapter->netdev;
1248
1249                /* configure backlink on ring */
1250                ring->q_vector = q_vector;
1251
1252                /* update q_vector Tx values */
1253                igb_add_ring(ring, &q_vector->tx);
1254
1255                /* For 82575, context index must be unique per ring. */
1256                if (adapter->hw.mac.type == e1000_82575)
1257                        set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1258
1259                /* apply Tx specific ring traits */
1260                ring->count = adapter->tx_ring_count;
1261                ring->queue_index = txr_idx;
1262
1263                ring->cbs_enable = false;
1264                ring->idleslope = 0;
1265                ring->sendslope = 0;
1266                ring->hicredit = 0;
1267                ring->locredit = 0;
1268
1269                u64_stats_init(&ring->tx_syncp);
1270                u64_stats_init(&ring->tx_syncp2);
1271
1272                /* assign ring to adapter */
1273                adapter->tx_ring[txr_idx] = ring;
1274
1275                /* push pointer to next ring */
1276                ring++;
1277        }
1278
1279        if (rxr_count) {
1280                /* assign generic ring traits */
1281                ring->dev = &adapter->pdev->dev;
1282                ring->netdev = adapter->netdev;
1283
1284                /* configure backlink on ring */
1285                ring->q_vector = q_vector;
1286
1287                /* update q_vector Rx values */
1288                igb_add_ring(ring, &q_vector->rx);
1289
1290                /* set flag indicating ring supports SCTP checksum offload */
1291                if (adapter->hw.mac.type >= e1000_82576)
1292                        set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1293
1294                /* On i350, i354, i210, and i211, loopback VLAN packets
1295                 * have the tag byte-swapped.
1296                 */
1297                if (adapter->hw.mac.type >= e1000_i350)
1298                        set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1299
1300                /* apply Rx specific ring traits */
1301                ring->count = adapter->rx_ring_count;
1302                ring->queue_index = rxr_idx;
1303
1304                u64_stats_init(&ring->rx_syncp);
1305
1306                /* assign ring to adapter */
1307                adapter->rx_ring[rxr_idx] = ring;
1308        }
1309
1310        return 0;
1311}
1312
1313
1314/**
1315 *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1316 *  @adapter: board private structure to initialize
1317 *
1318 *  We allocate one q_vector per queue interrupt.  If allocation fails we
1319 *  return -ENOMEM.
1320 **/
1321static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1322{
1323        int q_vectors = adapter->num_q_vectors;
1324        int rxr_remaining = adapter->num_rx_queues;
1325        int txr_remaining = adapter->num_tx_queues;
1326        int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1327        int err;
1328
1329        if (q_vectors >= (rxr_remaining + txr_remaining)) {
1330                for (; rxr_remaining; v_idx++) {
1331                        err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1332                                                 0, 0, 1, rxr_idx);
1333
1334                        if (err)
1335                                goto err_out;
1336
1337                        /* update counts and index */
1338                        rxr_remaining--;
1339                        rxr_idx++;
1340                }
1341        }
1342
1343        for (; v_idx < q_vectors; v_idx++) {
1344                int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1345                int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1346
1347                err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1348                                         tqpv, txr_idx, rqpv, rxr_idx);
1349
1350                if (err)
1351                        goto err_out;
1352
1353                /* update counts and index */
1354                rxr_remaining -= rqpv;
1355                txr_remaining -= tqpv;
1356                rxr_idx++;
1357                txr_idx++;
1358        }
1359
1360        return 0;
1361
1362err_out:
1363        adapter->num_tx_queues = 0;
1364        adapter->num_rx_queues = 0;
1365        adapter->num_q_vectors = 0;
1366
1367        while (v_idx--)
1368                igb_free_q_vector(adapter, v_idx);
1369
1370        return -ENOMEM;
1371}
1372
1373/**
1374 *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1375 *  @adapter: board private structure to initialize
1376 *  @msix: boolean value of MSIX capability
1377 *
1378 *  This function initializes the interrupts and allocates all of the queues.
1379 **/
1380static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1381{
1382        struct pci_dev *pdev = adapter->pdev;
1383        int err;
1384
1385        igb_set_interrupt_capability(adapter, msix);
1386
1387        err = igb_alloc_q_vectors(adapter);
1388        if (err) {
1389                dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1390                goto err_alloc_q_vectors;
1391        }
1392
1393        igb_cache_ring_register(adapter);
1394
1395        return 0;
1396
1397err_alloc_q_vectors:
1398        igb_reset_interrupt_capability(adapter);
1399        return err;
1400}
1401
1402/**
1403 *  igb_request_irq - initialize interrupts
1404 *  @adapter: board private structure to initialize
1405 *
1406 *  Attempts to configure interrupts using the best available
1407 *  capabilities of the hardware and kernel.
1408 **/
1409static int igb_request_irq(struct igb_adapter *adapter)
1410{
1411        struct net_device *netdev = adapter->netdev;
1412        struct pci_dev *pdev = adapter->pdev;
1413        int err = 0;
1414
1415        if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1416                err = igb_request_msix(adapter);
1417                if (!err)
1418                        goto request_done;
1419                /* fall back to MSI */
1420                igb_free_all_tx_resources(adapter);
1421                igb_free_all_rx_resources(adapter);
1422
1423                igb_clear_interrupt_scheme(adapter);
1424                err = igb_init_interrupt_scheme(adapter, false);
1425                if (err)
1426                        goto request_done;
1427
1428                igb_setup_all_tx_resources(adapter);
1429                igb_setup_all_rx_resources(adapter);
1430                igb_configure(adapter);
1431        }
1432
1433        igb_assign_vector(adapter->q_vector[0], 0);
1434
1435        if (adapter->flags & IGB_FLAG_HAS_MSI) {
1436                err = request_irq(pdev->irq, igb_intr_msi, 0,
1437                                  netdev->name, adapter);
1438                if (!err)
1439                        goto request_done;
1440
1441                /* fall back to legacy interrupts */
1442                igb_reset_interrupt_capability(adapter);
1443                adapter->flags &= ~IGB_FLAG_HAS_MSI;
1444        }
1445
1446        err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1447                          netdev->name, adapter);
1448
1449        if (err)
1450                dev_err(&pdev->dev, "Error %d getting interrupt\n",
1451                        err);
1452
1453request_done:
1454        return err;
1455}
1456
1457static void igb_free_irq(struct igb_adapter *adapter)
1458{
1459        if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1460                int vector = 0, i;
1461
1462                free_irq(adapter->msix_entries[vector++].vector, adapter);
1463
1464                for (i = 0; i < adapter->num_q_vectors; i++)
1465                        free_irq(adapter->msix_entries[vector++].vector,
1466                                 adapter->q_vector[i]);
1467        } else {
1468                free_irq(adapter->pdev->irq, adapter);
1469        }
1470}
1471
1472/**
1473 *  igb_irq_disable - Mask off interrupt generation on the NIC
1474 *  @adapter: board private structure
1475 **/
1476static void igb_irq_disable(struct igb_adapter *adapter)
1477{
1478        struct e1000_hw *hw = &adapter->hw;
1479
1480        /* we need to be careful when disabling interrupts.  The VFs are also
1481         * mapped into these registers and so clearing the bits can cause
1482         * issues on the VF drivers so we only need to clear what we set
1483         */
1484        if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1485                u32 regval = rd32(E1000_EIAM);
1486
1487                wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1488                wr32(E1000_EIMC, adapter->eims_enable_mask);
1489                regval = rd32(E1000_EIAC);
1490                wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1491        }
1492
1493        wr32(E1000_IAM, 0);
1494        wr32(E1000_IMC, ~0);
1495        wrfl();
1496        if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1497                int i;
1498
1499                for (i = 0; i < adapter->num_q_vectors; i++)
1500                        synchronize_irq(adapter->msix_entries[i].vector);
1501        } else {
1502                synchronize_irq(adapter->pdev->irq);
1503        }
1504}
1505
1506/**
1507 *  igb_irq_enable - Enable default interrupt generation settings
1508 *  @adapter: board private structure
1509 **/
1510static void igb_irq_enable(struct igb_adapter *adapter)
1511{
1512        struct e1000_hw *hw = &adapter->hw;
1513
1514        if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1515                u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1516                u32 regval = rd32(E1000_EIAC);
1517
1518                wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1519                regval = rd32(E1000_EIAM);
1520                wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1521                wr32(E1000_EIMS, adapter->eims_enable_mask);
1522                if (adapter->vfs_allocated_count) {
1523                        wr32(E1000_MBVFIMR, 0xFF);
1524                        ims |= E1000_IMS_VMMB;
1525                }
1526                wr32(E1000_IMS, ims);
1527        } else {
1528                wr32(E1000_IMS, IMS_ENABLE_MASK |
1529                                E1000_IMS_DRSTA);
1530                wr32(E1000_IAM, IMS_ENABLE_MASK |
1531                                E1000_IMS_DRSTA);
1532        }
1533}
1534
1535static void igb_update_mng_vlan(struct igb_adapter *adapter)
1536{
1537        struct e1000_hw *hw = &adapter->hw;
1538        u16 pf_id = adapter->vfs_allocated_count;
1539        u16 vid = adapter->hw.mng_cookie.vlan_id;
1540        u16 old_vid = adapter->mng_vlan_id;
1541
1542        if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1543                /* add VID to filter table */
1544                igb_vfta_set(hw, vid, pf_id, true, true);
1545                adapter->mng_vlan_id = vid;
1546        } else {
1547                adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1548        }
1549
1550        if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1551            (vid != old_vid) &&
1552            !test_bit(old_vid, adapter->active_vlans)) {
1553                /* remove VID from filter table */
1554                igb_vfta_set(hw, vid, pf_id, false, true);
1555        }
1556}
1557
1558/**
1559 *  igb_release_hw_control - release control of the h/w to f/w
1560 *  @adapter: address of board private structure
1561 *
1562 *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1563 *  For ASF and Pass Through versions of f/w this means that the
1564 *  driver is no longer loaded.
1565 **/
1566static void igb_release_hw_control(struct igb_adapter *adapter)
1567{
1568        struct e1000_hw *hw = &adapter->hw;
1569        u32 ctrl_ext;
1570
1571        /* Let firmware take over control of h/w */
1572        ctrl_ext = rd32(E1000_CTRL_EXT);
1573        wr32(E1000_CTRL_EXT,
1574                        ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1575}
1576
1577/**
1578 *  igb_get_hw_control - get control of the h/w from f/w
1579 *  @adapter: address of board private structure
1580 *
1581 *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1582 *  For ASF and Pass Through versions of f/w this means that
1583 *  the driver is loaded.
1584 **/
1585static void igb_get_hw_control(struct igb_adapter *adapter)
1586{
1587        struct e1000_hw *hw = &adapter->hw;
1588        u32 ctrl_ext;
1589
1590        /* Let firmware know the driver has taken over */
1591        ctrl_ext = rd32(E1000_CTRL_EXT);
1592        wr32(E1000_CTRL_EXT,
1593                        ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1594}
1595
1596static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1597{
1598        struct net_device *netdev = adapter->netdev;
1599        struct e1000_hw *hw = &adapter->hw;
1600
1601        WARN_ON(hw->mac.type != e1000_i210);
1602
1603        if (enable)
1604                adapter->flags |= IGB_FLAG_FQTSS;
1605        else
1606                adapter->flags &= ~IGB_FLAG_FQTSS;
1607
1608        if (netif_running(netdev))
1609                schedule_work(&adapter->reset_task);
1610}
1611
1612static bool is_fqtss_enabled(struct igb_adapter *adapter)
1613{
1614        return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1615}
1616
1617static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1618                                   enum tx_queue_prio prio)
1619{
1620        u32 val;
1621
1622        WARN_ON(hw->mac.type != e1000_i210);
1623        WARN_ON(queue < 0 || queue > 4);
1624
1625        val = rd32(E1000_I210_TXDCTL(queue));
1626
1627        if (prio == TX_QUEUE_PRIO_HIGH)
1628                val |= E1000_TXDCTL_PRIORITY;
1629        else
1630                val &= ~E1000_TXDCTL_PRIORITY;
1631
1632        wr32(E1000_I210_TXDCTL(queue), val);
1633}
1634
1635static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1636{
1637        u32 val;
1638
1639        WARN_ON(hw->mac.type != e1000_i210);
1640        WARN_ON(queue < 0 || queue > 1);
1641
1642        val = rd32(E1000_I210_TQAVCC(queue));
1643
1644        if (mode == QUEUE_MODE_STREAM_RESERVATION)
1645                val |= E1000_TQAVCC_QUEUEMODE;
1646        else
1647                val &= ~E1000_TQAVCC_QUEUEMODE;
1648
1649        wr32(E1000_I210_TQAVCC(queue), val);
1650}
1651
1652static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1653{
1654        int i;
1655
1656        for (i = 0; i < adapter->num_tx_queues; i++) {
1657                if (adapter->tx_ring[i]->cbs_enable)
1658                        return true;
1659        }
1660
1661        return false;
1662}
1663
1664static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1665{
1666        int i;
1667
1668        for (i = 0; i < adapter->num_tx_queues; i++) {
1669                if (adapter->tx_ring[i]->launchtime_enable)
1670                        return true;
1671        }
1672
1673        return false;
1674}
1675
1676/**
1677 *  igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1678 *  @adapter: pointer to adapter struct
1679 *  @queue: queue number
1680 *
1681 *  Configure CBS and Launchtime for a given hardware queue.
1682 *  Parameters are retrieved from the correct Tx ring, so
1683 *  igb_save_cbs_params() and igb_save_txtime_params() should be used
1684 *  for setting those correctly prior to this function being called.
1685 **/
1686static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1687{
1688        struct igb_ring *ring = adapter->tx_ring[queue];
1689        struct net_device *netdev = adapter->netdev;
1690        struct e1000_hw *hw = &adapter->hw;
1691        u32 tqavcc, tqavctrl;
1692        u16 value;
1693
1694        WARN_ON(hw->mac.type != e1000_i210);
1695        WARN_ON(queue < 0 || queue > 1);
1696
1697        /* If any of the Qav features is enabled, configure queues as SR and
1698         * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1699         * as SP.
1700         */
1701        if (ring->cbs_enable || ring->launchtime_enable) {
1702                set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1703                set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1704        } else {
1705                set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1706                set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1707        }
1708
1709        /* If CBS is enabled, set DataTranARB and config its parameters. */
1710        if (ring->cbs_enable || queue == 0) {
1711                /* i210 does not allow the queue 0 to be in the Strict
1712                 * Priority mode while the Qav mode is enabled, so,
1713                 * instead of disabling strict priority mode, we give
1714                 * queue 0 the maximum of credits possible.
1715                 *
1716                 * See section 8.12.19 of the i210 datasheet, "Note:
1717                 * Queue0 QueueMode must be set to 1b when
1718                 * TransmitMode is set to Qav."
1719                 */
1720                if (queue == 0 && !ring->cbs_enable) {
1721                        /* max "linkspeed" idleslope in kbps */
1722                        ring->idleslope = 1000000;
1723                        ring->hicredit = ETH_FRAME_LEN;
1724                }
1725
1726                /* Always set data transfer arbitration to credit-based
1727                 * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1728                 * the queues.
1729                 */
1730                tqavctrl = rd32(E1000_I210_TQAVCTRL);
1731                tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1732                wr32(E1000_I210_TQAVCTRL, tqavctrl);
1733
1734                /* According to i210 datasheet section 7.2.7.7, we should set
1735                 * the 'idleSlope' field from TQAVCC register following the
1736                 * equation:
1737                 *
1738                 * For 100 Mbps link speed:
1739                 *
1740                 *     value = BW * 0x7735 * 0.2                          (E1)
1741                 *
1742                 * For 1000Mbps link speed:
1743                 *
1744                 *     value = BW * 0x7735 * 2                            (E2)
1745                 *
1746                 * E1 and E2 can be merged into one equation as shown below.
1747                 * Note that 'link-speed' is in Mbps.
1748                 *
1749                 *     value = BW * 0x7735 * 2 * link-speed
1750                 *                           --------------               (E3)
1751                 *                                1000
1752                 *
1753                 * 'BW' is the percentage bandwidth out of full link speed
1754                 * which can be found with the following equation. Note that
1755                 * idleSlope here is the parameter from this function which
1756                 * is in kbps.
1757                 *
1758                 *     BW =     idleSlope
1759                 *          -----------------                             (E4)
1760                 *          link-speed * 1000
1761                 *
1762                 * That said, we can come up with a generic equation to
1763                 * calculate the value we should set it TQAVCC register by
1764                 * replacing 'BW' in E3 by E4. The resulting equation is:
1765                 *
1766                 * value =     idleSlope     * 0x7735 * 2 * link-speed
1767                 *         -----------------            --------------    (E5)
1768                 *         link-speed * 1000                 1000
1769                 *
1770                 * 'link-speed' is present in both sides of the fraction so
1771                 * it is canceled out. The final equation is the following:
1772                 *
1773                 *     value = idleSlope * 61034
1774                 *             -----------------                          (E6)
1775                 *                  1000000
1776                 *
1777                 * NOTE: For i210, given the above, we can see that idleslope
1778                 *       is represented in 16.38431 kbps units by the value at
1779                 *       the TQAVCC register (1Gbps / 61034), which reduces
1780                 *       the granularity for idleslope increments.
1781                 *       For instance, if you want to configure a 2576kbps
1782                 *       idleslope, the value to be written on the register
1783                 *       would have to be 157.23. If rounded down, you end
1784                 *       up with less bandwidth available than originally
1785                 *       required (~2572 kbps). If rounded up, you end up
1786                 *       with a higher bandwidth (~2589 kbps). Below the
1787                 *       approach we take is to always round up the
1788                 *       calculated value, so the resulting bandwidth might
1789                 *       be slightly higher for some configurations.
1790                 */
1791                value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1792
1793                tqavcc = rd32(E1000_I210_TQAVCC(queue));
1794                tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1795                tqavcc |= value;
1796                wr32(E1000_I210_TQAVCC(queue), tqavcc);
1797
1798                wr32(E1000_I210_TQAVHC(queue),
1799                     0x80000000 + ring->hicredit * 0x7735);
1800        } else {
1801
1802                /* Set idleSlope to zero. */
1803                tqavcc = rd32(E1000_I210_TQAVCC(queue));
1804                tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1805                wr32(E1000_I210_TQAVCC(queue), tqavcc);
1806
1807                /* Set hiCredit to zero. */
1808                wr32(E1000_I210_TQAVHC(queue), 0);
1809
1810                /* If CBS is not enabled for any queues anymore, then return to
1811                 * the default state of Data Transmission Arbitration on
1812                 * TQAVCTRL.
1813                 */
1814                if (!is_any_cbs_enabled(adapter)) {
1815                        tqavctrl = rd32(E1000_I210_TQAVCTRL);
1816                        tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1817                        wr32(E1000_I210_TQAVCTRL, tqavctrl);
1818                }
1819        }
1820
1821        /* If LaunchTime is enabled, set DataTranTIM. */
1822        if (ring->launchtime_enable) {
1823                /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1824                 * for any of the SR queues, and configure fetchtime delta.
1825                 * XXX NOTE:
1826                 *     - LaunchTime will be enabled for all SR queues.
1827                 *     - A fixed offset can be added relative to the launch
1828                 *       time of all packets if configured at reg LAUNCH_OS0.
1829                 *       We are keeping it as 0 for now (default value).
1830                 */
1831                tqavctrl = rd32(E1000_I210_TQAVCTRL);
1832                tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1833                       E1000_TQAVCTRL_FETCHTIME_DELTA;
1834                wr32(E1000_I210_TQAVCTRL, tqavctrl);
1835        } else {
1836                /* If Launchtime is not enabled for any SR queues anymore,
1837                 * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1838                 * effectively disabling Launchtime.
1839                 */
1840                if (!is_any_txtime_enabled(adapter)) {
1841                        tqavctrl = rd32(E1000_I210_TQAVCTRL);
1842                        tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1843                        tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1844                        wr32(E1000_I210_TQAVCTRL, tqavctrl);
1845                }
1846        }
1847
1848        /* XXX: In i210 controller the sendSlope and loCredit parameters from
1849         * CBS are not configurable by software so we don't do any 'controller
1850         * configuration' in respect to these parameters.
1851         */
1852
1853        netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1854                   ring->cbs_enable ? "enabled" : "disabled",
1855                   ring->launchtime_enable ? "enabled" : "disabled",
1856                   queue,
1857                   ring->idleslope, ring->sendslope,
1858                   ring->hicredit, ring->locredit);
1859}
1860
1861static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1862                                  bool enable)
1863{
1864        struct igb_ring *ring;
1865
1866        if (queue < 0 || queue > adapter->num_tx_queues)
1867                return -EINVAL;
1868
1869        ring = adapter->tx_ring[queue];
1870        ring->launchtime_enable = enable;
1871
1872        return 0;
1873}
1874
1875static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1876                               bool enable, int idleslope, int sendslope,
1877                               int hicredit, int locredit)
1878{
1879        struct igb_ring *ring;
1880
1881        if (queue < 0 || queue > adapter->num_tx_queues)
1882                return -EINVAL;
1883
1884        ring = adapter->tx_ring[queue];
1885
1886        ring->cbs_enable = enable;
1887        ring->idleslope = idleslope;
1888        ring->sendslope = sendslope;
1889        ring->hicredit = hicredit;
1890        ring->locredit = locredit;
1891
1892        return 0;
1893}
1894
1895/**
1896 *  igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1897 *  @adapter: pointer to adapter struct
1898 *
1899 *  Configure TQAVCTRL register switching the controller's Tx mode
1900 *  if FQTSS mode is enabled or disabled. Additionally, will issue
1901 *  a call to igb_config_tx_modes() per queue so any previously saved
1902 *  Tx parameters are applied.
1903 **/
1904static void igb_setup_tx_mode(struct igb_adapter *adapter)
1905{
1906        struct net_device *netdev = adapter->netdev;
1907        struct e1000_hw *hw = &adapter->hw;
1908        u32 val;
1909
1910        /* Only i210 controller supports changing the transmission mode. */
1911        if (hw->mac.type != e1000_i210)
1912                return;
1913
1914        if (is_fqtss_enabled(adapter)) {
1915                int i, max_queue;
1916
1917                /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1918                 * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1919                 * so SP queues wait for SR ones.
1920                 */
1921                val = rd32(E1000_I210_TQAVCTRL);
1922                val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1923                val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1924                wr32(E1000_I210_TQAVCTRL, val);
1925
1926                /* Configure Tx and Rx packet buffers sizes as described in
1927                 * i210 datasheet section 7.2.7.7.
1928                 */
1929                val = rd32(E1000_TXPBS);
1930                val &= ~I210_TXPBSIZE_MASK;
1931                val |= I210_TXPBSIZE_PB0_8KB | I210_TXPBSIZE_PB1_8KB |
1932                        I210_TXPBSIZE_PB2_4KB | I210_TXPBSIZE_PB3_4KB;
1933                wr32(E1000_TXPBS, val);
1934
1935                val = rd32(E1000_RXPBS);
1936                val &= ~I210_RXPBSIZE_MASK;
1937                val |= I210_RXPBSIZE_PB_30KB;
1938                wr32(E1000_RXPBS, val);
1939
1940                /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1941                 * register should not exceed the buffer size programmed in
1942                 * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1943                 * so according to the datasheet we should set MAX_TPKT_SIZE to
1944                 * 4kB / 64.
1945                 *
1946                 * However, when we do so, no frame from queue 2 and 3 are
1947                 * transmitted.  It seems the MAX_TPKT_SIZE should not be great
1948                 * or _equal_ to the buffer size programmed in TXPBS. For this
1949                 * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1950                 */
1951                val = (4096 - 1) / 64;
1952                wr32(E1000_I210_DTXMXPKTSZ, val);
1953
1954                /* Since FQTSS mode is enabled, apply any CBS configuration
1955                 * previously set. If no previous CBS configuration has been
1956                 * done, then the initial configuration is applied, which means
1957                 * CBS is disabled.
1958                 */
1959                max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1960                            adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1961
1962                for (i = 0; i < max_queue; i++) {
1963                        igb_config_tx_modes(adapter, i);
1964                }
1965        } else {
1966                wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1967                wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1968                wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1969
1970                val = rd32(E1000_I210_TQAVCTRL);
1971                /* According to Section 8.12.21, the other flags we've set when
1972                 * enabling FQTSS are not relevant when disabling FQTSS so we
1973                 * don't set they here.
1974                 */
1975                val &= ~E1000_TQAVCTRL_XMIT_MODE;
1976                wr32(E1000_I210_TQAVCTRL, val);
1977        }
1978
1979        netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1980                   "enabled" : "disabled");
1981}
1982
1983/**
1984 *  igb_configure - configure the hardware for RX and TX
1985 *  @adapter: private board structure
1986 **/
1987static void igb_configure(struct igb_adapter *adapter)
1988{
1989        struct net_device *netdev = adapter->netdev;
1990        int i;
1991
1992        igb_get_hw_control(adapter);
1993        igb_set_rx_mode(netdev);
1994        igb_setup_tx_mode(adapter);
1995
1996        igb_restore_vlan(adapter);
1997
1998        igb_setup_tctl(adapter);
1999        igb_setup_mrqc(adapter);
2000        igb_setup_rctl(adapter);
2001
2002        igb_nfc_filter_restore(adapter);
2003        igb_configure_tx(adapter);
2004        igb_configure_rx(adapter);
2005
2006        igb_rx_fifo_flush_82575(&adapter->hw);
2007
2008        /* call igb_desc_unused which always leaves
2009         * at least 1 descriptor unused to make sure
2010         * next_to_use != next_to_clean
2011         */
2012        for (i = 0; i < adapter->num_rx_queues; i++) {
2013                struct igb_ring *ring = adapter->rx_ring[i];
2014                igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
2015        }
2016}
2017
2018/**
2019 *  igb_power_up_link - Power up the phy/serdes link
2020 *  @adapter: address of board private structure
2021 **/
2022void igb_power_up_link(struct igb_adapter *adapter)
2023{
2024        igb_reset_phy(&adapter->hw);
2025
2026        if (adapter->hw.phy.media_type == e1000_media_type_copper)
2027                igb_power_up_phy_copper(&adapter->hw);
2028        else
2029                igb_power_up_serdes_link_82575(&adapter->hw);
2030
2031        igb_setup_link(&adapter->hw);
2032}
2033
2034/**
2035 *  igb_power_down_link - Power down the phy/serdes link
2036 *  @adapter: address of board private structure
2037 */
2038static void igb_power_down_link(struct igb_adapter *adapter)
2039{
2040        if (adapter->hw.phy.media_type == e1000_media_type_copper)
2041                igb_power_down_phy_copper_82575(&adapter->hw);
2042        else
2043                igb_shutdown_serdes_link_82575(&adapter->hw);
2044}
2045
2046/**
2047 * Detect and switch function for Media Auto Sense
2048 * @adapter: address of the board private structure
2049 **/
2050static void igb_check_swap_media(struct igb_adapter *adapter)
2051{
2052        struct e1000_hw *hw = &adapter->hw;
2053        u32 ctrl_ext, connsw;
2054        bool swap_now = false;
2055
2056        ctrl_ext = rd32(E1000_CTRL_EXT);
2057        connsw = rd32(E1000_CONNSW);
2058
2059        /* need to live swap if current media is copper and we have fiber/serdes
2060         * to go to.
2061         */
2062
2063        if ((hw->phy.media_type == e1000_media_type_copper) &&
2064            (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2065                swap_now = true;
2066        } else if (!(connsw & E1000_CONNSW_SERDESD)) {
2067                /* copper signal takes time to appear */
2068                if (adapter->copper_tries < 4) {
2069                        adapter->copper_tries++;
2070                        connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2071                        wr32(E1000_CONNSW, connsw);
2072                        return;
2073                } else {
2074                        adapter->copper_tries = 0;
2075                        if ((connsw & E1000_CONNSW_PHYSD) &&
2076                            (!(connsw & E1000_CONNSW_PHY_PDN))) {
2077                                swap_now = true;
2078                                connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2079                                wr32(E1000_CONNSW, connsw);
2080                        }
2081                }
2082        }
2083
2084        if (!swap_now)
2085                return;
2086
2087        switch (hw->phy.media_type) {
2088        case e1000_media_type_copper:
2089                netdev_info(adapter->netdev,
2090                        "MAS: changing media to fiber/serdes\n");
2091                ctrl_ext |=
2092                        E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2093                adapter->flags |= IGB_FLAG_MEDIA_RESET;
2094                adapter->copper_tries = 0;
2095                break;
2096        case e1000_media_type_internal_serdes:
2097        case e1000_media_type_fiber:
2098                netdev_info(adapter->netdev,
2099                        "MAS: changing media to copper\n");
2100                ctrl_ext &=
2101                        ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2102                adapter->flags |= IGB_FLAG_MEDIA_RESET;
2103                break;
2104        default:
2105                /* shouldn't get here during regular operation */
2106                netdev_err(adapter->netdev,
2107                        "AMS: Invalid media type found, returning\n");
2108                break;
2109        }
2110        wr32(E1000_CTRL_EXT, ctrl_ext);
2111}
2112
2113/**
2114 *  igb_up - Open the interface and prepare it to handle traffic
2115 *  @adapter: board private structure
2116 **/
2117int igb_up(struct igb_adapter *adapter)
2118{
2119        struct e1000_hw *hw = &adapter->hw;
2120        int i;
2121
2122        /* hardware has been reset, we need to reload some things */
2123        igb_configure(adapter);
2124
2125        clear_bit(__IGB_DOWN, &adapter->state);
2126
2127        for (i = 0; i < adapter->num_q_vectors; i++)
2128                napi_enable(&(adapter->q_vector[i]->napi));
2129
2130        if (adapter->flags & IGB_FLAG_HAS_MSIX)
2131                igb_configure_msix(adapter);
2132        else
2133                igb_assign_vector(adapter->q_vector[0], 0);
2134
2135        /* Clear any pending interrupts. */
2136        rd32(E1000_TSICR);
2137        rd32(E1000_ICR);
2138        igb_irq_enable(adapter);
2139
2140        /* notify VFs that reset has been completed */
2141        if (adapter->vfs_allocated_count) {
2142                u32 reg_data = rd32(E1000_CTRL_EXT);
2143
2144                reg_data |= E1000_CTRL_EXT_PFRSTD;
2145                wr32(E1000_CTRL_EXT, reg_data);
2146        }
2147
2148        netif_tx_start_all_queues(adapter->netdev);
2149
2150        /* start the watchdog. */
2151        hw->mac.get_link_status = 1;
2152        schedule_work(&adapter->watchdog_task);
2153
2154        if ((adapter->flags & IGB_FLAG_EEE) &&
2155            (!hw->dev_spec._82575.eee_disable))
2156                adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2157
2158        return 0;
2159}
2160
2161void igb_down(struct igb_adapter *adapter)
2162{
2163        struct net_device *netdev = adapter->netdev;
2164        struct e1000_hw *hw = &adapter->hw;
2165        u32 tctl, rctl;
2166        int i;
2167
2168        /* signal that we're down so the interrupt handler does not
2169         * reschedule our watchdog timer
2170         */
2171        set_bit(__IGB_DOWN, &adapter->state);
2172
2173        /* disable receives in the hardware */
2174        rctl = rd32(E1000_RCTL);
2175        wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2176        /* flush and sleep below */
2177
2178        igb_nfc_filter_exit(adapter);
2179
2180        netif_carrier_off(netdev);
2181        netif_tx_stop_all_queues(netdev);
2182
2183        /* disable transmits in the hardware */
2184        tctl = rd32(E1000_TCTL);
2185        tctl &= ~E1000_TCTL_EN;
2186        wr32(E1000_TCTL, tctl);
2187        /* flush both disables and wait for them to finish */
2188        wrfl();
2189        usleep_range(10000, 11000);
2190
2191        igb_irq_disable(adapter);
2192
2193        adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2194
2195        for (i = 0; i < adapter->num_q_vectors; i++) {
2196                if (adapter->q_vector[i]) {
2197                        napi_synchronize(&adapter->q_vector[i]->napi);
2198                        napi_disable(&adapter->q_vector[i]->napi);
2199                }
2200        }
2201
2202        del_timer_sync(&adapter->watchdog_timer);
2203        del_timer_sync(&adapter->phy_info_timer);
2204
2205        /* record the stats before reset*/
2206        spin_lock(&adapter->stats64_lock);
2207        igb_update_stats(adapter);
2208        spin_unlock(&adapter->stats64_lock);
2209
2210        adapter->link_speed = 0;
2211        adapter->link_duplex = 0;
2212
2213        if (!pci_channel_offline(adapter->pdev))
2214                igb_reset(adapter);
2215
2216        /* clear VLAN promisc flag so VFTA will be updated if necessary */
2217        adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2218
2219        igb_clean_all_tx_rings(adapter);
2220        igb_clean_all_rx_rings(adapter);
2221#ifdef CONFIG_IGB_DCA
2222
2223        /* since we reset the hardware DCA settings were cleared */
2224        igb_setup_dca(adapter);
2225#endif
2226}
2227
2228void igb_reinit_locked(struct igb_adapter *adapter)
2229{
2230        WARN_ON(in_interrupt());
2231        while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2232                usleep_range(1000, 2000);
2233        igb_down(adapter);
2234        igb_up(adapter);
2235        clear_bit(__IGB_RESETTING, &adapter->state);
2236}
2237
2238/** igb_enable_mas - Media Autosense re-enable after swap
2239 *
2240 * @adapter: adapter struct
2241 **/
2242static void igb_enable_mas(struct igb_adapter *adapter)
2243{
2244        struct e1000_hw *hw = &adapter->hw;
2245        u32 connsw = rd32(E1000_CONNSW);
2246
2247        /* configure for SerDes media detect */
2248        if ((hw->phy.media_type == e1000_media_type_copper) &&
2249            (!(connsw & E1000_CONNSW_SERDESD))) {
2250                connsw |= E1000_CONNSW_ENRGSRC;
2251                connsw |= E1000_CONNSW_AUTOSENSE_EN;
2252                wr32(E1000_CONNSW, connsw);
2253                wrfl();
2254        }
2255}
2256
2257void igb_reset(struct igb_adapter *adapter)
2258{
2259        struct pci_dev *pdev = adapter->pdev;
2260        struct e1000_hw *hw = &adapter->hw;
2261        struct e1000_mac_info *mac = &hw->mac;
2262        struct e1000_fc_info *fc = &hw->fc;
2263        u32 pba, hwm;
2264
2265        /* Repartition Pba for greater than 9k mtu
2266         * To take effect CTRL.RST is required.
2267         */
2268        switch (mac->type) {
2269        case e1000_i350:
2270        case e1000_i354:
2271        case e1000_82580:
2272                pba = rd32(E1000_RXPBS);
2273                pba = igb_rxpbs_adjust_82580(pba);
2274                break;
2275        case e1000_82576:
2276                pba = rd32(E1000_RXPBS);
2277                pba &= E1000_RXPBS_SIZE_MASK_82576;
2278                break;
2279        case e1000_82575:
2280        case e1000_i210:
2281        case e1000_i211:
2282        default:
2283                pba = E1000_PBA_34K;
2284                break;
2285        }
2286
2287        if (mac->type == e1000_82575) {
2288                u32 min_rx_space, min_tx_space, needed_tx_space;
2289
2290                /* write Rx PBA so that hardware can report correct Tx PBA */
2291                wr32(E1000_PBA, pba);
2292
2293                /* To maintain wire speed transmits, the Tx FIFO should be
2294                 * large enough to accommodate two full transmit packets,
2295                 * rounded up to the next 1KB and expressed in KB.  Likewise,
2296                 * the Rx FIFO should be large enough to accommodate at least
2297                 * one full receive packet and is similarly rounded up and
2298                 * expressed in KB.
2299                 */
2300                min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2301
2302                /* The Tx FIFO also stores 16 bytes of information about the Tx
2303                 * but don't include Ethernet FCS because hardware appends it.
2304                 * We only need to round down to the nearest 512 byte block
2305                 * count since the value we care about is 2 frames, not 1.
2306                 */
2307                min_tx_space = adapter->max_frame_size;
2308                min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2309                min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2310
2311                /* upper 16 bits has Tx packet buffer allocation size in KB */
2312                needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2313
2314                /* If current Tx allocation is less than the min Tx FIFO size,
2315                 * and the min Tx FIFO size is less than the current Rx FIFO
2316                 * allocation, take space away from current Rx allocation.
2317                 */
2318                if (needed_tx_space < pba) {
2319                        pba -= needed_tx_space;
2320
2321                        /* if short on Rx space, Rx wins and must trump Tx
2322                         * adjustment
2323                         */
2324                        if (pba < min_rx_space)
2325                                pba = min_rx_space;
2326                }
2327
2328                /* adjust PBA for jumbo frames */
2329                wr32(E1000_PBA, pba);
2330        }
2331
2332        /* flow control settings
2333         * The high water mark must be low enough to fit one full frame
2334         * after transmitting the pause frame.  As such we must have enough
2335         * space to allow for us to complete our current transmit and then
2336         * receive the frame that is in progress from the link partner.
2337         * Set it to:
2338         * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2339         */
2340        hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2341
2342        fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
2343        fc->low_water = fc->high_water - 16;
2344        fc->pause_time = 0xFFFF;
2345        fc->send_xon = 1;
2346        fc->current_mode = fc->requested_mode;
2347
2348        /* disable receive for all VFs and wait one second */
2349        if (adapter->vfs_allocated_count) {
2350                int i;
2351
2352                for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2353                        adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2354
2355                /* ping all the active vfs to let them know we are going down */
2356                igb_ping_all_vfs(adapter);
2357
2358                /* disable transmits and receives */
2359                wr32(E1000_VFRE, 0);
2360                wr32(E1000_VFTE, 0);
2361        }
2362
2363        /* Allow time for pending master requests to run */
2364        hw->mac.ops.reset_hw(hw);
2365        wr32(E1000_WUC, 0);
2366
2367        if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2368                /* need to resetup here after media swap */
2369                adapter->ei.get_invariants(hw);
2370                adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2371        }
2372        if ((mac->type == e1000_82575) &&
2373            (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2374                igb_enable_mas(adapter);
2375        }
2376        if (hw->mac.ops.init_hw(hw))
2377                dev_err(&pdev->dev, "Hardware Error\n");
2378
2379        /* RAR registers were cleared during init_hw, clear mac table */
2380        igb_flush_mac_table(adapter);
2381        __dev_uc_unsync(adapter->netdev, NULL);
2382
2383        /* Recover default RAR entry */
2384        igb_set_default_mac_filter(adapter);
2385
2386        /* Flow control settings reset on hardware reset, so guarantee flow
2387         * control is off when forcing speed.
2388         */
2389        if (!hw->mac.autoneg)
2390                igb_force_mac_fc(hw);
2391
2392        igb_init_dmac(adapter, pba);
2393#ifdef CONFIG_IGB_HWMON
2394        /* Re-initialize the thermal sensor on i350 devices. */
2395        if (!test_bit(__IGB_DOWN, &adapter->state)) {
2396                if (mac->type == e1000_i350 && hw->bus.func == 0) {
2397                        /* If present, re-initialize the external thermal sensor
2398                         * interface.
2399                         */
2400                        if (adapter->ets)
2401                                mac->ops.init_thermal_sensor_thresh(hw);
2402                }
2403        }
2404#endif
2405        /* Re-establish EEE setting */
2406        if (hw->phy.media_type == e1000_media_type_copper) {
2407                switch (mac->type) {
2408                case e1000_i350:
2409                case e1000_i210:
2410                case e1000_i211:
2411                        igb_set_eee_i350(hw, true, true);
2412                        break;
2413                case e1000_i354:
2414                        igb_set_eee_i354(hw, true, true);
2415                        break;
2416                default:
2417                        break;
2418                }
2419        }
2420        if (!netif_running(adapter->netdev))
2421                igb_power_down_link(adapter);
2422
2423        igb_update_mng_vlan(adapter);
2424
2425        /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2426        wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2427
2428        /* Re-enable PTP, where applicable. */
2429        if (adapter->ptp_flags & IGB_PTP_ENABLED)
2430                igb_ptp_reset(adapter);
2431
2432        igb_get_phy_info(hw);
2433}
2434
2435static netdev_features_t igb_fix_features(struct net_device *netdev,
2436        netdev_features_t features)
2437{
2438        /* Since there is no support for separate Rx/Tx vlan accel
2439         * enable/disable make sure Tx flag is always in same state as Rx.
2440         */
2441        if (features & NETIF_F_HW_VLAN_CTAG_RX)
2442                features |= NETIF_F_HW_VLAN_CTAG_TX;
2443        else
2444                features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2445
2446        return features;
2447}
2448
2449static int igb_set_features(struct net_device *netdev,
2450        netdev_features_t features)
2451{
2452        netdev_features_t changed = netdev->features ^ features;
2453        struct igb_adapter *adapter = netdev_priv(netdev);
2454
2455        if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2456                igb_vlan_mode(netdev, features);
2457
2458        if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2459                return 0;
2460
2461        if (!(features & NETIF_F_NTUPLE)) {
2462                struct hlist_node *node2;
2463                struct igb_nfc_filter *rule;
2464
2465                spin_lock(&adapter->nfc_lock);
2466                hlist_for_each_entry_safe(rule, node2,
2467                                          &adapter->nfc_filter_list, nfc_node) {
2468                        igb_erase_filter(adapter, rule);
2469                        hlist_del(&rule->nfc_node);
2470                        kfree(rule);
2471                }
2472                spin_unlock(&adapter->nfc_lock);
2473                adapter->nfc_filter_count = 0;
2474        }
2475
2476        netdev->features = features;
2477
2478        if (netif_running(netdev))
2479                igb_reinit_locked(adapter);
2480        else
2481                igb_reset(adapter);
2482
2483        return 0;
2484}
2485
2486static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2487                           struct net_device *dev,
2488                           const unsigned char *addr, u16 vid,
2489                           u16 flags)
2490{
2491        /* guarantee we can provide a unique filter for the unicast address */
2492        if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2493                struct igb_adapter *adapter = netdev_priv(dev);
2494                int vfn = adapter->vfs_allocated_count;
2495
2496                if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2497                        return -ENOMEM;
2498        }
2499
2500        return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2501}
2502
2503#define IGB_MAX_MAC_HDR_LEN     127
2504#define IGB_MAX_NETWORK_HDR_LEN 511
2505
2506static netdev_features_t
2507igb_features_check(struct sk_buff *skb, struct net_device *dev,
2508                   netdev_features_t features)
2509{
2510        unsigned int network_hdr_len, mac_hdr_len;
2511
2512        /* Make certain the headers can be described by a context descriptor */
2513        mac_hdr_len = skb_network_header(skb) - skb->data;
2514        if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2515                return features & ~(NETIF_F_HW_CSUM |
2516                                    NETIF_F_SCTP_CRC |
2517                                    NETIF_F_HW_VLAN_CTAG_TX |
2518                                    NETIF_F_TSO |
2519                                    NETIF_F_TSO6);
2520
2521        network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2522        if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
2523                return features & ~(NETIF_F_HW_CSUM |
2524                                    NETIF_F_SCTP_CRC |
2525                                    NETIF_F_TSO |
2526                                    NETIF_F_TSO6);
2527
2528        /* We can only support IPV4 TSO in tunnels if we can mangle the
2529         * inner IP ID field, so strip TSO if MANGLEID is not supported.
2530         */
2531        if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2532                features &= ~NETIF_F_TSO;
2533
2534        return features;
2535}
2536
2537static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2538{
2539        if (!is_fqtss_enabled(adapter)) {
2540                enable_fqtss(adapter, true);
2541                return;
2542        }
2543
2544        igb_config_tx_modes(adapter, queue);
2545
2546        if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2547                enable_fqtss(adapter, false);
2548}
2549
2550static int igb_offload_cbs(struct igb_adapter *adapter,
2551                           struct tc_cbs_qopt_offload *qopt)
2552{
2553        struct e1000_hw *hw = &adapter->hw;
2554        int err;
2555
2556        /* CBS offloading is only supported by i210 controller. */
2557        if (hw->mac.type != e1000_i210)
2558                return -EOPNOTSUPP;
2559
2560        /* CBS offloading is only supported by queue 0 and queue 1. */
2561        if (qopt->queue < 0 || qopt->queue > 1)
2562                return -EINVAL;
2563
2564        err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2565                                  qopt->idleslope, qopt->sendslope,
2566                                  qopt->hicredit, qopt->locredit);
2567        if (err)
2568                return err;
2569
2570        igb_offload_apply(adapter, qopt->queue);
2571
2572        return 0;
2573}
2574
2575#define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2576#define VLAN_PRIO_FULL_MASK (0x07)
2577
2578static int igb_parse_cls_flower(struct igb_adapter *adapter,
2579                                struct tc_cls_flower_offload *f,
2580                                int traffic_class,
2581                                struct igb_nfc_filter *input)
2582{
2583        struct netlink_ext_ack *extack = f->common.extack;
2584
2585        if (f->dissector->used_keys &
2586            ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
2587              BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2588              BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2589              BIT(FLOW_DISSECTOR_KEY_VLAN))) {
2590                NL_SET_ERR_MSG_MOD(extack,
2591                                   "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2592                return -EOPNOTSUPP;
2593        }
2594
2595        if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2596                struct flow_dissector_key_eth_addrs *key, *mask;
2597
2598                key = skb_flow_dissector_target(f->dissector,
2599                                                FLOW_DISSECTOR_KEY_ETH_ADDRS,
2600                                                f->key);
2601                mask = skb_flow_dissector_target(f->dissector,
2602                                                 FLOW_DISSECTOR_KEY_ETH_ADDRS,
2603                                                 f->mask);
2604
2605                if (!is_zero_ether_addr(mask->dst)) {
2606                        if (!is_broadcast_ether_addr(mask->dst)) {
2607                                NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2608                                return -EINVAL;
2609                        }
2610
2611                        input->filter.match_flags |=
2612                                IGB_FILTER_FLAG_DST_MAC_ADDR;
2613                        ether_addr_copy(input->filter.dst_addr, key->dst);
2614                }
2615
2616                if (!is_zero_ether_addr(mask->src)) {
2617                        if (!is_broadcast_ether_addr(mask->src)) {
2618                                NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2619                                return -EINVAL;
2620                        }
2621
2622                        input->filter.match_flags |=
2623                                IGB_FILTER_FLAG_SRC_MAC_ADDR;
2624                        ether_addr_copy(input->filter.src_addr, key->src);
2625                }
2626        }
2627
2628        if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_BASIC)) {
2629                struct flow_dissector_key_basic *key, *mask;
2630
2631                key = skb_flow_dissector_target(f->dissector,
2632                                                FLOW_DISSECTOR_KEY_BASIC,
2633                                                f->key);
2634                mask = skb_flow_dissector_target(f->dissector,
2635                                                 FLOW_DISSECTOR_KEY_BASIC,
2636                                                 f->mask);
2637
2638                if (mask->n_proto) {
2639                        if (mask->n_proto != ETHER_TYPE_FULL_MASK) {
2640                                NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2641                                return -EINVAL;
2642                        }
2643
2644                        input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2645                        input->filter.etype = key->n_proto;
2646                }
2647        }
2648
2649        if (dissector_uses_key(f->dissector, FLOW_DISSECTOR_KEY_VLAN)) {
2650                struct flow_dissector_key_vlan *key, *mask;
2651
2652                key = skb_flow_dissector_target(f->dissector,
2653                                                FLOW_DISSECTOR_KEY_VLAN,
2654                                                f->key);
2655                mask = skb_flow_dissector_target(f->dissector,
2656                                                 FLOW_DISSECTOR_KEY_VLAN,
2657                                                 f->mask);
2658
2659                if (mask->vlan_priority) {
2660                        if (mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2661                                NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2662                                return -EINVAL;
2663                        }
2664
2665                        input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2666                        input->filter.vlan_tci = key->vlan_priority;
2667                }
2668        }
2669
2670        input->action = traffic_class;
2671        input->cookie = f->cookie;
2672
2673        return 0;
2674}
2675
2676static int igb_configure_clsflower(struct igb_adapter *adapter,
2677                                   struct tc_cls_flower_offload *cls_flower)
2678{
2679        struct netlink_ext_ack *extack = cls_flower->common.extack;
2680        struct igb_nfc_filter *filter, *f;
2681        int err, tc;
2682
2683        tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2684        if (tc < 0) {
2685                NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2686                return -EINVAL;
2687        }
2688
2689        filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2690        if (!filter)
2691                return -ENOMEM;
2692
2693        err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2694        if (err < 0)
2695                goto err_parse;
2696
2697        spin_lock(&adapter->nfc_lock);
2698
2699        hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2700                if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2701                        err = -EEXIST;
2702                        NL_SET_ERR_MSG_MOD(extack,
2703                                           "This filter is already set in ethtool");
2704                        goto err_locked;
2705                }
2706        }
2707
2708        hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2709                if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2710                        err = -EEXIST;
2711                        NL_SET_ERR_MSG_MOD(extack,
2712                                           "This filter is already set in cls_flower");
2713                        goto err_locked;
2714                }
2715        }
2716
2717        err = igb_add_filter(adapter, filter);
2718        if (err < 0) {
2719                NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2720                goto err_locked;
2721        }
2722
2723        hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2724
2725        spin_unlock(&adapter->nfc_lock);
2726
2727        return 0;
2728
2729err_locked:
2730        spin_unlock(&adapter->nfc_lock);
2731
2732err_parse:
2733        kfree(filter);
2734
2735        return err;
2736}
2737
2738static int igb_delete_clsflower(struct igb_adapter *adapter,
2739                                struct tc_cls_flower_offload *cls_flower)
2740{
2741        struct igb_nfc_filter *filter;
2742        int err;
2743
2744        spin_lock(&adapter->nfc_lock);
2745
2746        hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2747                if (filter->cookie == cls_flower->cookie)
2748                        break;
2749
2750        if (!filter) {
2751                err = -ENOENT;
2752                goto out;
2753        }
2754
2755        err = igb_erase_filter(adapter, filter);
2756        if (err < 0)
2757                goto out;
2758
2759        hlist_del(&filter->nfc_node);
2760        kfree(filter);
2761
2762out:
2763        spin_unlock(&adapter->nfc_lock);
2764
2765        return err;
2766}
2767
2768static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2769                                   struct tc_cls_flower_offload *cls_flower)
2770{
2771        switch (cls_flower->command) {
2772        case TC_CLSFLOWER_REPLACE:
2773                return igb_configure_clsflower(adapter, cls_flower);
2774        case TC_CLSFLOWER_DESTROY:
2775                return igb_delete_clsflower(adapter, cls_flower);
2776        case TC_CLSFLOWER_STATS:
2777                return -EOPNOTSUPP;
2778        default:
2779                return -EOPNOTSUPP;
2780        }
2781}
2782
2783static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2784                                 void *cb_priv)
2785{
2786        struct igb_adapter *adapter = cb_priv;
2787
2788        if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2789                return -EOPNOTSUPP;
2790
2791        switch (type) {
2792        case TC_SETUP_CLSFLOWER:
2793                return igb_setup_tc_cls_flower(adapter, type_data);
2794
2795        default:
2796                return -EOPNOTSUPP;
2797        }
2798}
2799
2800static int igb_setup_tc_block(struct igb_adapter *adapter,
2801                              struct tc_block_offload *f)
2802{
2803        if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
2804                return -EOPNOTSUPP;
2805
2806        switch (f->command) {
2807        case TC_BLOCK_BIND:
2808                return tcf_block_cb_register(f->block, igb_setup_tc_block_cb,
2809                                             adapter, adapter, f->extack);
2810        case TC_BLOCK_UNBIND:
2811                tcf_block_cb_unregister(f->block, igb_setup_tc_block_cb,
2812                                        adapter);
2813                return 0;
2814        default:
2815                return -EOPNOTSUPP;
2816        }
2817}
2818
2819static int igb_offload_txtime(struct igb_adapter *adapter,
2820                              struct tc_etf_qopt_offload *qopt)
2821{
2822        struct e1000_hw *hw = &adapter->hw;
2823        int err;
2824
2825        /* Launchtime offloading is only supported by i210 controller. */
2826        if (hw->mac.type != e1000_i210)
2827                return -EOPNOTSUPP;
2828
2829        /* Launchtime offloading is only supported by queues 0 and 1. */
2830        if (qopt->queue < 0 || qopt->queue > 1)
2831                return -EINVAL;
2832
2833        err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2834        if (err)
2835                return err;
2836
2837        igb_offload_apply(adapter, qopt->queue);
2838
2839        return 0;
2840}
2841
2842static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2843                        void *type_data)
2844{
2845        struct igb_adapter *adapter = netdev_priv(dev);
2846
2847        switch (type) {
2848        case TC_SETUP_QDISC_CBS:
2849                return igb_offload_cbs(adapter, type_data);
2850        case TC_SETUP_BLOCK:
2851                return igb_setup_tc_block(adapter, type_data);
2852        case TC_SETUP_QDISC_ETF:
2853                return igb_offload_txtime(adapter, type_data);
2854
2855        default:
2856                return -EOPNOTSUPP;
2857        }
2858}
2859
2860static const struct net_device_ops igb_netdev_ops = {
2861        .ndo_open               = igb_open,
2862        .ndo_stop               = igb_close,
2863        .ndo_start_xmit         = igb_xmit_frame,
2864        .ndo_get_stats64        = igb_get_stats64,
2865        .ndo_set_rx_mode        = igb_set_rx_mode,
2866        .ndo_set_mac_address    = igb_set_mac,
2867        .ndo_change_mtu         = igb_change_mtu,
2868        .ndo_do_ioctl           = igb_ioctl,
2869        .ndo_tx_timeout         = igb_tx_timeout,
2870        .ndo_validate_addr      = eth_validate_addr,
2871        .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
2872        .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
2873        .ndo_set_vf_mac         = igb_ndo_set_vf_mac,
2874        .ndo_set_vf_vlan        = igb_ndo_set_vf_vlan,
2875        .ndo_set_vf_rate        = igb_ndo_set_vf_bw,
2876        .ndo_set_vf_spoofchk    = igb_ndo_set_vf_spoofchk,
2877        .ndo_set_vf_trust       = igb_ndo_set_vf_trust,
2878        .ndo_get_vf_config      = igb_ndo_get_vf_config,
2879        .ndo_fix_features       = igb_fix_features,
2880        .ndo_set_features       = igb_set_features,
2881        .ndo_fdb_add            = igb_ndo_fdb_add,
2882        .ndo_features_check     = igb_features_check,
2883        .ndo_setup_tc           = igb_setup_tc,
2884};
2885
2886/**
2887 * igb_set_fw_version - Configure version string for ethtool
2888 * @adapter: adapter struct
2889 **/
2890void igb_set_fw_version(struct igb_adapter *adapter)
2891{
2892        struct e1000_hw *hw = &adapter->hw;
2893        struct e1000_fw_version fw;
2894
2895        igb_get_fw_version(hw, &fw);
2896
2897        switch (hw->mac.type) {
2898        case e1000_i210:
2899        case e1000_i211:
2900                if (!(igb_get_flash_presence_i210(hw))) {
2901                        snprintf(adapter->fw_version,
2902                                 sizeof(adapter->fw_version),
2903                                 "%2d.%2d-%d",
2904                                 fw.invm_major, fw.invm_minor,
2905                                 fw.invm_img_type);
2906                        break;
2907                }
2908                /* fall through */
2909        default:
2910                /* if option is rom valid, display its version too */
2911                if (fw.or_valid) {
2912                        snprintf(adapter->fw_version,
2913                                 sizeof(adapter->fw_version),
2914                                 "%d.%d, 0x%08x, %d.%d.%d",
2915                                 fw.eep_major, fw.eep_minor, fw.etrack_id,
2916                                 fw.or_major, fw.or_build, fw.or_patch);
2917                /* no option rom */
2918                } else if (fw.etrack_id != 0X0000) {
2919                        snprintf(adapter->fw_version,
2920                            sizeof(adapter->fw_version),
2921                            "%d.%d, 0x%08x",
2922                            fw.eep_major, fw.eep_minor, fw.etrack_id);
2923                } else {
2924                snprintf(adapter->fw_version,
2925                    sizeof(adapter->fw_version),
2926                    "%d.%d.%d",
2927                    fw.eep_major, fw.eep_minor, fw.eep_build);
2928                }
2929                break;
2930        }
2931}
2932
2933/**
2934 * igb_init_mas - init Media Autosense feature if enabled in the NVM
2935 *
2936 * @adapter: adapter struct
2937 **/
2938static void igb_init_mas(struct igb_adapter *adapter)
2939{
2940        struct e1000_hw *hw = &adapter->hw;
2941        u16 eeprom_data;
2942
2943        hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2944        switch (hw->bus.func) {
2945        case E1000_FUNC_0:
2946                if (eeprom_data & IGB_MAS_ENABLE_0) {
2947                        adapter->flags |= IGB_FLAG_MAS_ENABLE;
2948                        netdev_info(adapter->netdev,
2949                                "MAS: Enabling Media Autosense for port %d\n",
2950                                hw->bus.func);
2951                }
2952                break;
2953        case E1000_FUNC_1:
2954                if (eeprom_data & IGB_MAS_ENABLE_1) {
2955                        adapter->flags |= IGB_FLAG_MAS_ENABLE;
2956                        netdev_info(adapter->netdev,
2957                                "MAS: Enabling Media Autosense for port %d\n",
2958                                hw->bus.func);
2959                }
2960                break;
2961        case E1000_FUNC_2:
2962                if (eeprom_data & IGB_MAS_ENABLE_2) {
2963                        adapter->flags |= IGB_FLAG_MAS_ENABLE;
2964                        netdev_info(adapter->netdev,
2965                                "MAS: Enabling Media Autosense for port %d\n",
2966                                hw->bus.func);
2967                }
2968                break;
2969        case E1000_FUNC_3:
2970                if (eeprom_data & IGB_MAS_ENABLE_3) {
2971                        adapter->flags |= IGB_FLAG_MAS_ENABLE;
2972                        netdev_info(adapter->netdev,
2973                                "MAS: Enabling Media Autosense for port %d\n",
2974                                hw->bus.func);
2975                }
2976                break;
2977        default:
2978                /* Shouldn't get here */
2979                netdev_err(adapter->netdev,
2980                        "MAS: Invalid port configuration, returning\n");
2981                break;
2982        }
2983}
2984
2985/**
2986 *  igb_init_i2c - Init I2C interface
2987 *  @adapter: pointer to adapter structure
2988 **/
2989static s32 igb_init_i2c(struct igb_adapter *adapter)
2990{
2991        s32 status = 0;
2992
2993        /* I2C interface supported on i350 devices */
2994        if (adapter->hw.mac.type != e1000_i350)
2995                return 0;
2996
2997        /* Initialize the i2c bus which is controlled by the registers.
2998         * This bus will use the i2c_algo_bit structue that implements
2999         * the protocol through toggling of the 4 bits in the register.
3000         */
3001        adapter->i2c_adap.owner = THIS_MODULE;
3002        adapter->i2c_algo = igb_i2c_algo;
3003        adapter->i2c_algo.data = adapter;
3004        adapter->i2c_adap.algo_data = &adapter->i2c_algo;
3005        adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
3006        strlcpy(adapter->i2c_adap.name, "igb BB",
3007                sizeof(adapter->i2c_adap.name));
3008        status = i2c_bit_add_bus(&adapter->i2c_adap);
3009        return status;
3010}
3011
3012/**
3013 *  igb_probe - Device Initialization Routine
3014 *  @pdev: PCI device information struct
3015 *  @ent: entry in igb_pci_tbl
3016 *
3017 *  Returns 0 on success, negative on failure
3018 *
3019 *  igb_probe initializes an adapter identified by a pci_dev structure.
3020 *  The OS initialization, configuring of the adapter private structure,
3021 *  and a hardware reset occur.
3022 **/
3023static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3024{
3025        struct net_device *netdev;
3026        struct igb_adapter *adapter;
3027        struct e1000_hw *hw;
3028        u16 eeprom_data = 0;
3029        s32 ret_val;
3030        static int global_quad_port_a; /* global quad port a indication */
3031        const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3032        int err, pci_using_dac;
3033        u8 part_str[E1000_PBANUM_LENGTH];
3034
3035        /* Catch broken hardware that put the wrong VF device ID in
3036         * the PCIe SR-IOV capability.
3037         */
3038        if (pdev->is_virtfn) {
3039                WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
3040                        pci_name(pdev), pdev->vendor, pdev->device);
3041                return -EINVAL;
3042        }
3043
3044        err = pci_enable_device_mem(pdev);
3045        if (err)
3046                return err;
3047
3048        pci_using_dac = 0;
3049        err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3050        if (!err) {
3051                pci_using_dac = 1;
3052        } else {
3053                err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3054                if (err) {
3055                        dev_err(&pdev->dev,
3056                                "No usable DMA configuration, aborting\n");
3057                        goto err_dma;
3058                }
3059        }
3060
3061        err = pci_request_mem_regions(pdev, igb_driver_name);
3062        if (err)
3063                goto err_pci_reg;
3064
3065        pci_enable_pcie_error_reporting(pdev);
3066
3067        pci_set_master(pdev);
3068        pci_save_state(pdev);
3069
3070        err = -ENOMEM;
3071        netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3072                                   IGB_MAX_TX_QUEUES);
3073        if (!netdev)
3074                goto err_alloc_etherdev;
3075
3076        SET_NETDEV_DEV(netdev, &pdev->dev);
3077
3078        pci_set_drvdata(pdev, netdev);
3079        adapter = netdev_priv(netdev);
3080        adapter->netdev = netdev;
3081        adapter->pdev = pdev;
3082        hw = &adapter->hw;
3083        hw->back = adapter;
3084        adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3085
3086        err = -EIO;
3087        adapter->io_addr = pci_iomap(pdev, 0, 0);
3088        if (!adapter->io_addr)
3089                goto err_ioremap;
3090        /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3091        hw->hw_addr = adapter->io_addr;
3092
3093        netdev->netdev_ops = &igb_netdev_ops;
3094        igb_set_ethtool_ops(netdev);
3095        netdev->watchdog_timeo = 5 * HZ;
3096
3097        strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3098
3099        netdev->mem_start = pci_resource_start(pdev, 0);
3100        netdev->mem_end = pci_resource_end(pdev, 0);
3101
3102        /* PCI config space info */
3103        hw->vendor_id = pdev->vendor;
3104        hw->device_id = pdev->device;
3105        hw->revision_id = pdev->revision;
3106        hw->subsystem_vendor_id = pdev->subsystem_vendor;
3107        hw->subsystem_device_id = pdev->subsystem_device;
3108
3109        /* Copy the default MAC, PHY and NVM function pointers */
3110        memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3111        memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3112        memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3113        /* Initialize skew-specific constants */
3114        err = ei->get_invariants(hw);
3115        if (err)
3116                goto err_sw_init;
3117
3118        /* setup the private structure */
3119        err = igb_sw_init(adapter);
3120        if (err)
3121                goto err_sw_init;
3122
3123        igb_get_bus_info_pcie(hw);
3124
3125        hw->phy.autoneg_wait_to_complete = false;
3126
3127        /* Copper options */
3128        if (hw->phy.media_type == e1000_media_type_copper) {
3129                hw->phy.mdix = AUTO_ALL_MODES;
3130                hw->phy.disable_polarity_correction = false;
3131                hw->phy.ms_type = e1000_ms_hw_default;
3132        }
3133
3134        if (igb_check_reset_block(hw))
3135                dev_info(&pdev->dev,
3136                        "PHY reset is blocked due to SOL/IDER session.\n");
3137
3138        /* features is initialized to 0 in allocation, it might have bits
3139         * set by igb_sw_init so we should use an or instead of an
3140         * assignment.
3141         */
3142        netdev->features |= NETIF_F_SG |
3143                            NETIF_F_TSO |
3144                            NETIF_F_TSO6 |
3145                            NETIF_F_RXHASH |
3146                            NETIF_F_RXCSUM |
3147                            NETIF_F_HW_CSUM;
3148
3149        if (hw->mac.type >= e1000_82576)
3150                netdev->features |= NETIF_F_SCTP_CRC;
3151
3152        if (hw->mac.type >= e1000_i350)
3153                netdev->features |= NETIF_F_HW_TC;
3154
3155#define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3156                                  NETIF_F_GSO_GRE_CSUM | \
3157                                  NETIF_F_GSO_IPXIP4 | \
3158                                  NETIF_F_GSO_IPXIP6 | \
3159                                  NETIF_F_GSO_UDP_TUNNEL | \
3160                                  NETIF_F_GSO_UDP_TUNNEL_CSUM)
3161
3162        netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3163        netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3164
3165        /* copy netdev features into list of user selectable features */
3166        netdev->hw_features |= netdev->features |
3167                               NETIF_F_HW_VLAN_CTAG_RX |
3168                               NETIF_F_HW_VLAN_CTAG_TX |
3169                               NETIF_F_RXALL;
3170
3171        if (hw->mac.type >= e1000_i350)
3172                netdev->hw_features |= NETIF_F_NTUPLE;
3173
3174        if (pci_using_dac)
3175                netdev->features |= NETIF_F_HIGHDMA;
3176
3177        netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3178        netdev->mpls_features |= NETIF_F_HW_CSUM;
3179        netdev->hw_enc_features |= netdev->vlan_features;
3180
3181        /* set this bit last since it cannot be part of vlan_features */
3182        netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3183                            NETIF_F_HW_VLAN_CTAG_RX |
3184                            NETIF_F_HW_VLAN_CTAG_TX;
3185
3186        netdev->priv_flags |= IFF_SUPP_NOFCS;
3187
3188        netdev->priv_flags |= IFF_UNICAST_FLT;
3189
3190        /* MTU range: 68 - 9216 */
3191        netdev->min_mtu = ETH_MIN_MTU;
3192        netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3193
3194        adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3195
3196        /* before reading the NVM, reset the controller to put the device in a
3197         * known good starting state
3198         */
3199        hw->mac.ops.reset_hw(hw);
3200
3201        /* make sure the NVM is good , i211/i210 parts can have special NVM
3202         * that doesn't contain a checksum
3203         */
3204        switch (hw->mac.type) {
3205        case e1000_i210:
3206        case e1000_i211:
3207                if (igb_get_flash_presence_i210(hw)) {
3208                        if (hw->nvm.ops.validate(hw) < 0) {
3209                                dev_err(&pdev->dev,
3210                                        "The NVM Checksum Is Not Valid\n");
3211                                err = -EIO;
3212                                goto err_eeprom;
3213                        }
3214                }
3215                break;
3216        default:
3217                if (hw->nvm.ops.validate(hw) < 0) {
3218                        dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3219                        err = -EIO;
3220                        goto err_eeprom;
3221                }
3222                break;
3223        }
3224
3225        if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3226                /* copy the MAC address out of the NVM */
3227                if (hw->mac.ops.read_mac_addr(hw))
3228                        dev_err(&pdev->dev, "NVM Read Error\n");
3229        }
3230
3231        memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
3232
3233        if (!is_valid_ether_addr(netdev->dev_addr)) {
3234                dev_err(&pdev->dev, "Invalid MAC Address\n");
3235                err = -EIO;
3236                goto err_eeprom;
3237        }
3238
3239        igb_set_default_mac_filter(adapter);
3240
3241        /* get firmware version for ethtool -i */
3242        igb_set_fw_version(adapter);
3243
3244        /* configure RXPBSIZE and TXPBSIZE */
3245        if (hw->mac.type == e1000_i210) {
3246                wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3247                wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3248        }
3249
3250        timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3251        timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3252
3253        INIT_WORK(&adapter->reset_task, igb_reset_task);
3254        INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3255
3256        /* Initialize link properties that are user-changeable */
3257        adapter->fc_autoneg = true;
3258        hw->mac.autoneg = true;
3259        hw->phy.autoneg_advertised = 0x2f;
3260
3261        hw->fc.requested_mode = e1000_fc_default;
3262        hw->fc.current_mode = e1000_fc_default;
3263
3264        igb_validate_mdi_setting(hw);
3265
3266        /* By default, support wake on port A */
3267        if (hw->bus.func == 0)
3268                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3269
3270        /* Check the NVM for wake support on non-port A ports */
3271        if (hw->mac.type >= e1000_82580)
3272                hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3273                                 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3274                                 &eeprom_data);
3275        else if (hw->bus.func == 1)
3276                hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3277
3278        if (eeprom_data & IGB_EEPROM_APME)
3279                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3280
3281        /* now that we have the eeprom settings, apply the special cases where
3282         * the eeprom may be wrong or the board simply won't support wake on
3283         * lan on a particular port
3284         */
3285        switch (pdev->device) {
3286        case E1000_DEV_ID_82575GB_QUAD_COPPER:
3287                adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3288                break;
3289        case E1000_DEV_ID_82575EB_FIBER_SERDES:
3290        case E1000_DEV_ID_82576_FIBER:
3291        case E1000_DEV_ID_82576_SERDES:
3292                /* Wake events only supported on port A for dual fiber
3293                 * regardless of eeprom setting
3294                 */
3295                if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3296                        adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3297                break;
3298        case E1000_DEV_ID_82576_QUAD_COPPER:
3299        case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3300                /* if quad port adapter, disable WoL on all but port A */
3301                if (global_quad_port_a != 0)
3302                        adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3303                else
3304                        adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3305                /* Reset for multiple quad port adapters */
3306                if (++global_quad_port_a == 4)
3307                        global_quad_port_a = 0;
3308                break;
3309        default:
3310                /* If the device can't wake, don't set software support */
3311                if (!device_can_wakeup(&adapter->pdev->dev))
3312                        adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3313        }
3314
3315        /* initialize the wol settings based on the eeprom settings */
3316        if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3317                adapter->wol |= E1000_WUFC_MAG;
3318
3319        /* Some vendors want WoL disabled by default, but still supported */
3320        if ((hw->mac.type == e1000_i350) &&
3321            (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3322                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3323                adapter->wol = 0;
3324        }
3325
3326        /* Some vendors want the ability to Use the EEPROM setting as
3327         * enable/disable only, and not for capability
3328         */
3329        if (((hw->mac.type == e1000_i350) ||
3330             (hw->mac.type == e1000_i354)) &&
3331            (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3332                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3333                adapter->wol = 0;
3334        }
3335        if (hw->mac.type == e1000_i350) {
3336                if (((pdev->subsystem_device == 0x5001) ||
3337                     (pdev->subsystem_device == 0x5002)) &&
3338                                (hw->bus.func == 0)) {
3339                        adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3340                        adapter->wol = 0;
3341                }
3342                if (pdev->subsystem_device == 0x1F52)
3343                        adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3344        }
3345
3346        device_set_wakeup_enable(&adapter->pdev->dev,
3347                                 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3348
3349        /* reset the hardware with the new settings */
3350        igb_reset(adapter);
3351
3352        /* Init the I2C interface */
3353        err = igb_init_i2c(adapter);
3354        if (err) {
3355                dev_err(&pdev->dev, "failed to init i2c interface\n");
3356                goto err_eeprom;
3357        }
3358
3359        /* let the f/w know that the h/w is now under the control of the
3360         * driver.
3361         */
3362        igb_get_hw_control(adapter);
3363
3364        strcpy(netdev->name, "eth%d");
3365        err = register_netdev(netdev);
3366        if (err)
3367                goto err_register;
3368
3369        /* carrier off reporting is important to ethtool even BEFORE open */
3370        netif_carrier_off(netdev);
3371
3372#ifdef CONFIG_IGB_DCA
3373        if (dca_add_requester(&pdev->dev) == 0) {
3374                adapter->flags |= IGB_FLAG_DCA_ENABLED;
3375                dev_info(&pdev->dev, "DCA enabled\n");
3376                igb_setup_dca(adapter);
3377        }
3378
3379#endif
3380#ifdef CONFIG_IGB_HWMON
3381        /* Initialize the thermal sensor on i350 devices. */
3382        if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3383                u16 ets_word;
3384
3385                /* Read the NVM to determine if this i350 device supports an
3386                 * external thermal sensor.
3387                 */
3388                hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3389                if (ets_word != 0x0000 && ets_word != 0xFFFF)
3390                        adapter->ets = true;
3391                else
3392                        adapter->ets = false;
3393                if (igb_sysfs_init(adapter))
3394                        dev_err(&pdev->dev,
3395                                "failed to allocate sysfs resources\n");
3396        } else {
3397                adapter->ets = false;
3398        }
3399#endif
3400        /* Check if Media Autosense is enabled */
3401        adapter->ei = *ei;
3402        if (hw->dev_spec._82575.mas_capable)
3403                igb_init_mas(adapter);
3404
3405        /* do hw tstamp init after resetting */
3406        igb_ptp_init(adapter);
3407
3408        dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3409        /* print bus type/speed/width info, not applicable to i354 */
3410        if (hw->mac.type != e1000_i354) {
3411                dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3412                         netdev->name,
3413                         ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3414                          (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3415                           "unknown"),
3416                         ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3417                          "Width x4" :
3418                          (hw->bus.width == e1000_bus_width_pcie_x2) ?
3419                          "Width x2" :
3420                          (hw->bus.width == e1000_bus_width_pcie_x1) ?
3421                          "Width x1" : "unknown"), netdev->dev_addr);
3422        }
3423
3424        if ((hw->mac.type >= e1000_i210 ||
3425             igb_get_flash_presence_i210(hw))) {
3426                ret_val = igb_read_part_string(hw, part_str,
3427                                               E1000_PBANUM_LENGTH);
3428        } else {
3429                ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3430        }
3431
3432        if (ret_val)
3433                strcpy(part_str, "Unknown");
3434        dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3435        dev_info(&pdev->dev,
3436                "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3437                (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3438                (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3439                adapter->num_rx_queues, adapter->num_tx_queues);
3440        if (hw->phy.media_type == e1000_media_type_copper) {
3441                switch (hw->mac.type) {
3442                case e1000_i350:
3443                case e1000_i210:
3444                case e1000_i211:
3445                        /* Enable EEE for internal copper PHY devices */
3446                        err = igb_set_eee_i350(hw, true, true);
3447                        if ((!err) &&
3448                            (!hw->dev_spec._82575.eee_disable)) {
3449                                adapter->eee_advert =
3450                                        MDIO_EEE_100TX | MDIO_EEE_1000T;
3451                                adapter->flags |= IGB_FLAG_EEE;
3452                        }
3453                        break;
3454                case e1000_i354:
3455                        if ((rd32(E1000_CTRL_EXT) &
3456                            E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3457                                err = igb_set_eee_i354(hw, true, true);
3458                                if ((!err) &&
3459                                        (!hw->dev_spec._82575.eee_disable)) {
3460                                        adapter->eee_advert =
3461                                           MDIO_EEE_100TX | MDIO_EEE_1000T;
3462                                        adapter->flags |= IGB_FLAG_EEE;
3463                                }
3464                        }
3465                        break;
3466                default:
3467                        break;
3468                }
3469        }
3470        pm_runtime_put_noidle(&pdev->dev);
3471        return 0;
3472
3473err_register:
3474        igb_release_hw_control(adapter);
3475        memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3476err_eeprom:
3477        if (!igb_check_reset_block(hw))
3478                igb_reset_phy(hw);
3479
3480        if (hw->flash_address)
3481                iounmap(hw->flash_address);
3482err_sw_init:
3483        kfree(adapter->mac_table);
3484        kfree(adapter->shadow_vfta);
3485        igb_clear_interrupt_scheme(adapter);
3486#ifdef CONFIG_PCI_IOV
3487        igb_disable_sriov(pdev);
3488#endif
3489        pci_iounmap(pdev, adapter->io_addr);
3490err_ioremap:
3491        free_netdev(netdev);
3492err_alloc_etherdev:
3493        pci_release_mem_regions(pdev);
3494err_pci_reg:
3495err_dma:
3496        pci_disable_device(pdev);
3497        return err;
3498}
3499
3500#ifdef CONFIG_PCI_IOV
3501static int igb_disable_sriov(struct pci_dev *pdev)
3502{
3503        struct net_device *netdev = pci_get_drvdata(pdev);
3504        struct igb_adapter *adapter = netdev_priv(netdev);
3505        struct e1000_hw *hw = &adapter->hw;
3506
3507        /* reclaim resources allocated to VFs */
3508        if (adapter->vf_data) {
3509                /* disable iov and allow time for transactions to clear */
3510                if (pci_vfs_assigned(pdev)) {
3511                        dev_warn(&pdev->dev,
3512                                 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3513                        return -EPERM;
3514                } else {
3515                        pci_disable_sriov(pdev);
3516                        msleep(500);
3517                }
3518
3519                kfree(adapter->vf_mac_list);
3520                adapter->vf_mac_list = NULL;
3521                kfree(adapter->vf_data);
3522                adapter->vf_data = NULL;
3523                adapter->vfs_allocated_count = 0;
3524                wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3525                wrfl();
3526                msleep(100);
3527                dev_info(&pdev->dev, "IOV Disabled\n");
3528
3529                /* Re-enable DMA Coalescing flag since IOV is turned off */
3530                adapter->flags |= IGB_FLAG_DMAC;
3531        }
3532
3533        return 0;
3534}
3535
3536static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
3537{
3538        struct net_device *netdev = pci_get_drvdata(pdev);
3539        struct igb_adapter *adapter = netdev_priv(netdev);
3540        int old_vfs = pci_num_vf(pdev);
3541        struct vf_mac_filter *mac_list;
3542        int err = 0;
3543        int num_vf_mac_filters, i;
3544
3545        if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3546                err = -EPERM;
3547                goto out;
3548        }
3549        if (!num_vfs)
3550                goto out;
3551
3552        if (old_vfs) {
3553                dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3554                         old_vfs, max_vfs);
3555                adapter->vfs_allocated_count = old_vfs;
3556        } else
3557                adapter->vfs_allocated_count = num_vfs;
3558
3559        adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3560                                sizeof(struct vf_data_storage), GFP_KERNEL);
3561
3562        /* if allocation failed then we do not support SR-IOV */
3563        if (!adapter->vf_data) {
3564                adapter->vfs_allocated_count = 0;
3565                err = -ENOMEM;
3566                goto out;
3567        }
3568
3569        /* Due to the limited number of RAR entries calculate potential
3570         * number of MAC filters available for the VFs. Reserve entries
3571         * for PF default MAC, PF MAC filters and at least one RAR entry
3572         * for each VF for VF MAC.
3573         */
3574        num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3575                             (1 + IGB_PF_MAC_FILTERS_RESERVED +
3576                              adapter->vfs_allocated_count);
3577
3578        adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3579                                       sizeof(struct vf_mac_filter),
3580                                       GFP_KERNEL);
3581
3582        mac_list = adapter->vf_mac_list;
3583        INIT_LIST_HEAD(&adapter->vf_macs.l);
3584
3585        if (adapter->vf_mac_list) {
3586                /* Initialize list of VF MAC filters */
3587                for (i = 0; i < num_vf_mac_filters; i++) {
3588                        mac_list->vf = -1;
3589                        mac_list->free = true;
3590                        list_add(&mac_list->l, &adapter->vf_macs.l);
3591                        mac_list++;
3592                }
3593        } else {
3594                /* If we could not allocate memory for the VF MAC filters
3595                 * we can continue without this feature but warn user.
3596                 */
3597                dev_err(&pdev->dev,
3598                        "Unable to allocate memory for VF MAC filter list\n");
3599        }
3600
3601        /* only call pci_enable_sriov() if no VFs are allocated already */
3602        if (!old_vfs) {
3603                err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3604                if (err)
3605                        goto err_out;
3606        }
3607        dev_info(&pdev->dev, "%d VFs allocated\n",
3608                 adapter->vfs_allocated_count);
3609        for (i = 0; i < adapter->vfs_allocated_count; i++)
3610                igb_vf_configure(adapter, i);
3611
3612        /* DMA Coalescing is not supported in IOV mode. */
3613        adapter->flags &= ~IGB_FLAG_DMAC;
3614        goto out;
3615
3616err_out:
3617        kfree(adapter->vf_mac_list);
3618        adapter->vf_mac_list = NULL;
3619        kfree(adapter->vf_data);
3620        adapter->vf_data = NULL;
3621        adapter->vfs_allocated_count = 0;
3622out:
3623        return err;
3624}
3625
3626#endif
3627/**
3628 *  igb_remove_i2c - Cleanup  I2C interface
3629 *  @adapter: pointer to adapter structure
3630 **/
3631static void igb_remove_i2c(struct igb_adapter *adapter)
3632{
3633        /* free the adapter bus structure */
3634        i2c_del_adapter(&adapter->i2c_adap);
3635}
3636
3637/**
3638 *  igb_remove - Device Removal Routine
3639 *  @pdev: PCI device information struct
3640 *
3641 *  igb_remove is called by the PCI subsystem to alert the driver
3642 *  that it should release a PCI device.  The could be caused by a
3643 *  Hot-Plug event, or because the driver is going to be removed from
3644 *  memory.
3645 **/
3646static void igb_remove(struct pci_dev *pdev)
3647{
3648        struct net_device *netdev = pci_get_drvdata(pdev);
3649        struct igb_adapter *adapter = netdev_priv(netdev);
3650        struct e1000_hw *hw = &adapter->hw;
3651
3652        pm_runtime_get_noresume(&pdev->dev);
3653#ifdef CONFIG_IGB_HWMON
3654        igb_sysfs_exit(adapter);
3655#endif
3656        igb_remove_i2c(adapter);
3657        igb_ptp_stop(adapter);
3658        /* The watchdog timer may be rescheduled, so explicitly
3659         * disable watchdog from being rescheduled.
3660         */
3661        set_bit(__IGB_DOWN, &adapter->state);
3662        del_timer_sync(&adapter->watchdog_timer);
3663        del_timer_sync(&adapter->phy_info_timer);
3664
3665        cancel_work_sync(&adapter->reset_task);
3666        cancel_work_sync(&adapter->watchdog_task);
3667
3668#ifdef CONFIG_IGB_DCA
3669        if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3670                dev_info(&pdev->dev, "DCA disabled\n");
3671                dca_remove_requester(&pdev->dev);
3672                adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3673                wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3674        }
3675#endif
3676
3677        /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3678         * would have already happened in close and is redundant.
3679         */
3680        igb_release_hw_control(adapter);
3681
3682#ifdef CONFIG_PCI_IOV
3683        igb_disable_sriov(pdev);
3684#endif
3685
3686        unregister_netdev(netdev);
3687
3688        igb_clear_interrupt_scheme(adapter);
3689
3690        pci_iounmap(pdev, adapter->io_addr);
3691        if (hw->flash_address)
3692                iounmap(hw->flash_address);
3693        pci_release_mem_regions(pdev);
3694
3695        kfree(adapter->mac_table);
3696        kfree(adapter->shadow_vfta);
3697        free_netdev(netdev);
3698
3699        pci_disable_pcie_error_reporting(pdev);
3700
3701        pci_disable_device(pdev);
3702}
3703
3704/**
3705 *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3706 *  @adapter: board private structure to initialize
3707 *
3708 *  This function initializes the vf specific data storage and then attempts to
3709 *  allocate the VFs.  The reason for ordering it this way is because it is much
3710 *  mor expensive time wise to disable SR-IOV than it is to allocate and free
3711 *  the memory for the VFs.
3712 **/
3713static void igb_probe_vfs(struct igb_adapter *adapter)
3714{
3715#ifdef CONFIG_PCI_IOV
3716        struct pci_dev *pdev = adapter->pdev;
3717        struct e1000_hw *hw = &adapter->hw;
3718
3719        /* Virtualization features not supported on i210 family. */
3720        if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3721                return;
3722
3723        /* Of the below we really only want the effect of getting
3724         * IGB_FLAG_HAS_MSIX set (if available), without which
3725         * igb_enable_sriov() has no effect.
3726         */
3727        igb_set_interrupt_capability(adapter, true);
3728        igb_reset_interrupt_capability(adapter);
3729
3730        pci_sriov_set_totalvfs(pdev, 7);
3731        igb_enable_sriov(pdev, max_vfs);
3732
3733#endif /* CONFIG_PCI_IOV */
3734}
3735
3736unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3737{
3738        struct e1000_hw *hw = &adapter->hw;
3739        unsigned int max_rss_queues;
3740
3741        /* Determine the maximum number of RSS queues supported. */
3742        switch (hw->mac.type) {
3743        case e1000_i211:
3744                max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3745                break;
3746        case e1000_82575:
3747        case e1000_i210:
3748                max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3749                break;
3750        case e1000_i350:
3751                /* I350 cannot do RSS and SR-IOV at the same time */
3752                if (!!adapter->vfs_allocated_count) {
3753                        max_rss_queues = 1;
3754                        break;
3755                }
3756                /* fall through */
3757        case e1000_82576:
3758                if (!!adapter->vfs_allocated_count) {
3759                        max_rss_queues = 2;
3760                        break;
3761                }
3762                /* fall through */
3763        case e1000_82580:
3764        case e1000_i354:
3765        default:
3766                max_rss_queues = IGB_MAX_RX_QUEUES;
3767                break;
3768        }
3769
3770        return max_rss_queues;
3771}
3772
3773static void igb_init_queue_configuration(struct igb_adapter *adapter)
3774{
3775        u32 max_rss_queues;
3776
3777        max_rss_queues = igb_get_max_rss_queues(adapter);
3778        adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3779
3780        igb_set_flag_queue_pairs(adapter, max_rss_queues);
3781}
3782
3783void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3784                              const u32 max_rss_queues)
3785{
3786        struct e1000_hw *hw = &adapter->hw;
3787
3788        /* Determine if we need to pair queues. */
3789        switch (hw->mac.type) {
3790        case e1000_82575:
3791        case e1000_i211:
3792                /* Device supports enough interrupts without queue pairing. */
3793                break;
3794        case e1000_82576:
3795        case e1000_82580:
3796        case e1000_i350:
3797        case e1000_i354:
3798        case e1000_i210:
3799        default:
3800                /* If rss_queues > half of max_rss_queues, pair the queues in
3801                 * order to conserve interrupts due to limited supply.
3802                 */
3803                if (adapter->rss_queues > (max_rss_queues / 2))
3804                        adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3805                else
3806                        adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3807                break;
3808        }
3809}
3810
3811/**
3812 *  igb_sw_init - Initialize general software structures (struct igb_adapter)
3813 *  @adapter: board private structure to initialize
3814 *
3815 *  igb_sw_init initializes the Adapter private data structure.
3816 *  Fields are initialized based on PCI device information and
3817 *  OS network device settings (MTU size).
3818 **/
3819static int igb_sw_init(struct igb_adapter *adapter)
3820{
3821        struct e1000_hw *hw = &adapter->hw;
3822        struct net_device *netdev = adapter->netdev;
3823        struct pci_dev *pdev = adapter->pdev;
3824
3825        pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3826
3827        /* set default ring sizes */
3828        adapter->tx_ring_count = IGB_DEFAULT_TXD;
3829        adapter->rx_ring_count = IGB_DEFAULT_RXD;
3830
3831        /* set default ITR values */
3832        adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3833        adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3834
3835        /* set default work limits */
3836        adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3837
3838        adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3839                                  VLAN_HLEN;
3840        adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3841
3842        spin_lock_init(&adapter->nfc_lock);
3843        spin_lock_init(&adapter->stats64_lock);
3844#ifdef CONFIG_PCI_IOV
3845        switch (hw->mac.type) {
3846        case e1000_82576:
3847        case e1000_i350:
3848                if (max_vfs > 7) {
3849                        dev_warn(&pdev->dev,
3850                                 "Maximum of 7 VFs per PF, using max\n");
3851                        max_vfs = adapter->vfs_allocated_count = 7;
3852                } else
3853                        adapter->vfs_allocated_count = max_vfs;
3854                if (adapter->vfs_allocated_count)
3855                        dev_warn(&pdev->dev,
3856                                 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3857                break;
3858        default:
3859                break;
3860        }
3861#endif /* CONFIG_PCI_IOV */
3862
3863        /* Assume MSI-X interrupts, will be checked during IRQ allocation */
3864        adapter->flags |= IGB_FLAG_HAS_MSIX;
3865
3866        adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
3867                                     sizeof(struct igb_mac_addr),
3868                                     GFP_KERNEL);
3869        if (!adapter->mac_table)
3870                return -ENOMEM;
3871
3872        igb_probe_vfs(adapter);
3873
3874        igb_init_queue_configuration(adapter);
3875
3876        /* Setup and initialize a copy of the hw vlan table array */
3877        adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
3878                                       GFP_KERNEL);
3879        if (!adapter->shadow_vfta)
3880                return -ENOMEM;
3881
3882        /* This call may decrease the number of queues */
3883        if (igb_init_interrupt_scheme(adapter, true)) {
3884                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3885                return -ENOMEM;
3886        }
3887
3888        /* Explicitly disable IRQ since the NIC can be in any state. */
3889        igb_irq_disable(adapter);
3890
3891        if (hw->mac.type >= e1000_i350)
3892                adapter->flags &= ~IGB_FLAG_DMAC;
3893
3894        set_bit(__IGB_DOWN, &adapter->state);
3895        return 0;
3896}
3897
3898/**
3899 *  igb_open - Called when a network interface is made active
3900 *  @netdev: network interface device structure
3901 *
3902 *  Returns 0 on success, negative value on failure
3903 *
3904 *  The open entry point is called when a network interface is made
3905 *  active by the system (IFF_UP).  At this point all resources needed
3906 *  for transmit and receive operations are allocated, the interrupt
3907 *  handler is registered with the OS, the watchdog timer is started,
3908 *  and the stack is notified that the interface is ready.
3909 **/
3910static int __igb_open(struct net_device *netdev, bool resuming)
3911{
3912        struct igb_adapter *adapter = netdev_priv(netdev);
3913        struct e1000_hw *hw = &adapter->hw;
3914        struct pci_dev *pdev = adapter->pdev;
3915        int err;
3916        int i;
3917
3918        /* disallow open during test */
3919        if (test_bit(__IGB_TESTING, &adapter->state)) {
3920                WARN_ON(resuming);
3921                return -EBUSY;
3922        }
3923
3924        if (!resuming)
3925                pm_runtime_get_sync(&pdev->dev);
3926
3927        netif_carrier_off(netdev);
3928
3929        /* allocate transmit descriptors */
3930        err = igb_setup_all_tx_resources(adapter);
3931        if (err)
3932                goto err_setup_tx;
3933
3934        /* allocate receive descriptors */
3935        err = igb_setup_all_rx_resources(adapter);
3936        if (err)
3937                goto err_setup_rx;
3938
3939        igb_power_up_link(adapter);
3940
3941        /* before we allocate an interrupt, we must be ready to handle it.
3942         * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3943         * as soon as we call pci_request_irq, so we have to setup our
3944         * clean_rx handler before we do so.
3945         */
3946        igb_configure(adapter);
3947
3948        err = igb_request_irq(adapter);
3949        if (err)
3950                goto err_req_irq;
3951
3952        /* Notify the stack of the actual queue counts. */
3953        err = netif_set_real_num_tx_queues(adapter->netdev,
3954                                           adapter->num_tx_queues);
3955        if (err)
3956                goto err_set_queues;
3957
3958        err = netif_set_real_num_rx_queues(adapter->netdev,
3959                                           adapter->num_rx_queues);
3960        if (err)
3961                goto err_set_queues;
3962
3963        /* From here on the code is the same as igb_up() */
3964        clear_bit(__IGB_DOWN, &adapter->state);
3965
3966        for (i = 0; i < adapter->num_q_vectors; i++)
3967                napi_enable(&(adapter->q_vector[i]->napi));
3968
3969        /* Clear any pending interrupts. */
3970        rd32(E1000_TSICR);
3971        rd32(E1000_ICR);
3972
3973        igb_irq_enable(adapter);
3974
3975        /* notify VFs that reset has been completed */
3976        if (adapter->vfs_allocated_count) {
3977                u32 reg_data = rd32(E1000_CTRL_EXT);
3978
3979                reg_data |= E1000_CTRL_EXT_PFRSTD;
3980                wr32(E1000_CTRL_EXT, reg_data);
3981        }
3982
3983        netif_tx_start_all_queues(netdev);
3984
3985        if (!resuming)
3986                pm_runtime_put(&pdev->dev);
3987
3988        /* start the watchdog. */
3989        hw->mac.get_link_status = 1;
3990        schedule_work(&adapter->watchdog_task);
3991
3992        return 0;
3993
3994err_set_queues:
3995        igb_free_irq(adapter);
3996err_req_irq:
3997        igb_release_hw_control(adapter);
3998        igb_power_down_link(adapter);
3999        igb_free_all_rx_resources(adapter);
4000err_setup_rx:
4001        igb_free_all_tx_resources(adapter);
4002err_setup_tx:
4003        igb_reset(adapter);
4004        if (!resuming)
4005                pm_runtime_put(&pdev->dev);
4006
4007        return err;
4008}
4009
4010int igb_open(struct net_device *netdev)
4011{
4012        return __igb_open(netdev, false);
4013}
4014
4015/**
4016 *  igb_close - Disables a network interface
4017 *  @netdev: network interface device structure
4018 *
4019 *  Returns 0, this is not allowed to fail
4020 *
4021 *  The close entry point is called when an interface is de-activated
4022 *  by the OS.  The hardware is still under the driver's control, but
4023 *  needs to be disabled.  A global MAC reset is issued to stop the
4024 *  hardware, and all transmit and receive resources are freed.
4025 **/
4026static int __igb_close(struct net_device *netdev, bool suspending)
4027{
4028        struct igb_adapter *adapter = netdev_priv(netdev);
4029        struct pci_dev *pdev = adapter->pdev;
4030
4031        WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4032
4033        if (!suspending)
4034                pm_runtime_get_sync(&pdev->dev);
4035
4036        igb_down(adapter);
4037        igb_free_irq(adapter);
4038
4039        igb_free_all_tx_resources(adapter);
4040        igb_free_all_rx_resources(adapter);
4041
4042        if (!suspending)
4043                pm_runtime_put_sync(&pdev->dev);
4044        return 0;
4045}
4046
4047int igb_close(struct net_device *netdev)
4048{
4049        if (netif_device_present(netdev) || netdev->dismantle)
4050                return __igb_close(netdev, false);
4051        return 0;
4052}
4053
4054/**
4055 *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
4056 *  @tx_ring: tx descriptor ring (for a specific queue) to setup
4057 *
4058 *  Return 0 on success, negative on failure
4059 **/
4060int igb_setup_tx_resources(struct igb_ring *tx_ring)
4061{
4062        struct device *dev = tx_ring->dev;
4063        int size;
4064
4065        size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4066
4067        tx_ring->tx_buffer_info = vmalloc(size);
4068        if (!tx_ring->tx_buffer_info)
4069                goto err;
4070
4071        /* round up to nearest 4K */
4072        tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4073        tx_ring->size = ALIGN(tx_ring->size, 4096);
4074
4075        tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4076                                           &tx_ring->dma, GFP_KERNEL);
4077        if (!tx_ring->desc)
4078                goto err;
4079
4080        tx_ring->next_to_use = 0;
4081        tx_ring->next_to_clean = 0;
4082
4083        return 0;
4084
4085err:
4086        vfree(tx_ring->tx_buffer_info);
4087        tx_ring->tx_buffer_info = NULL;
4088        dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4089        return -ENOMEM;
4090}
4091
4092/**
4093 *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
4094 *                               (Descriptors) for all queues
4095 *  @adapter: board private structure
4096 *
4097 *  Return 0 on success, negative on failure
4098 **/
4099static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4100{
4101        struct pci_dev *pdev = adapter->pdev;
4102        int i, err = 0;
4103
4104        for (i = 0; i < adapter->num_tx_queues; i++) {
4105                err = igb_setup_tx_resources(adapter->tx_ring[i]);
4106                if (err) {
4107                        dev_err(&pdev->dev,
4108                                "Allocation for Tx Queue %u failed\n", i);
4109                        for (i--; i >= 0; i--)
4110                                igb_free_tx_resources(adapter->tx_ring[i]);
4111                        break;
4112                }
4113        }
4114
4115        return err;
4116}
4117
4118/**
4119 *  igb_setup_tctl - configure the transmit control registers
4120 *  @adapter: Board private structure
4121 **/
4122void igb_setup_tctl(struct igb_adapter *adapter)
4123{
4124        struct e1000_hw *hw = &adapter->hw;
4125        u32 tctl;
4126
4127        /* disable queue 0 which is enabled by default on 82575 and 82576 */
4128        wr32(E1000_TXDCTL(0), 0);
4129
4130        /* Program the Transmit Control Register */
4131        tctl = rd32(E1000_TCTL);
4132        tctl &= ~E1000_TCTL_CT;
4133        tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4134                (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4135
4136        igb_config_collision_dist(hw);
4137
4138        /* Enable transmits */
4139        tctl |= E1000_TCTL_EN;
4140
4141        wr32(E1000_TCTL, tctl);
4142}
4143
4144/**
4145 *  igb_configure_tx_ring - Configure transmit ring after Reset
4146 *  @adapter: board private structure
4147 *  @ring: tx ring to configure
4148 *
4149 *  Configure a transmit ring after a reset.
4150 **/
4151void igb_configure_tx_ring(struct igb_adapter *adapter,
4152                           struct igb_ring *ring)
4153{
4154        struct e1000_hw *hw = &adapter->hw;
4155        u32 txdctl = 0;
4156        u64 tdba = ring->dma;
4157        int reg_idx = ring->reg_idx;
4158
4159        wr32(E1000_TDLEN(reg_idx),
4160             ring->count * sizeof(union e1000_adv_tx_desc));
4161        wr32(E1000_TDBAL(reg_idx),
4162             tdba & 0x00000000ffffffffULL);
4163        wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4164
4165        ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4166        wr32(E1000_TDH(reg_idx), 0);
4167        writel(0, ring->tail);
4168
4169        txdctl |= IGB_TX_PTHRESH;
4170        txdctl |= IGB_TX_HTHRESH << 8;
4171        txdctl |= IGB_TX_WTHRESH << 16;
4172
4173        /* reinitialize tx_buffer_info */
4174        memset(ring->tx_buffer_info, 0,
4175               sizeof(struct igb_tx_buffer) * ring->count);
4176
4177        txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4178        wr32(E1000_TXDCTL(reg_idx), txdctl);
4179}
4180
4181/**
4182 *  igb_configure_tx - Configure transmit Unit after Reset
4183 *  @adapter: board private structure
4184 *
4185 *  Configure the Tx unit of the MAC after a reset.
4186 **/
4187static void igb_configure_tx(struct igb_adapter *adapter)
4188{
4189        struct e1000_hw *hw = &adapter->hw;
4190        int i;
4191
4192        /* disable the queues */
4193        for (i = 0; i < adapter->num_tx_queues; i++)
4194                wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4195
4196        wrfl();
4197        usleep_range(10000, 20000);
4198
4199        for (i = 0; i < adapter->num_tx_queues; i++)
4200                igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4201}
4202
4203/**
4204 *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
4205 *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
4206 *
4207 *  Returns 0 on success, negative on failure
4208 **/
4209int igb_setup_rx_resources(struct igb_ring *rx_ring)
4210{
4211        struct device *dev = rx_ring->dev;
4212        int size;
4213
4214        size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4215
4216        rx_ring->rx_buffer_info = vmalloc(size);
4217        if (!rx_ring->rx_buffer_info)
4218                goto err;
4219
4220        /* Round up to nearest 4K */
4221        rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4222        rx_ring->size = ALIGN(rx_ring->size, 4096);
4223
4224        rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4225                                           &rx_ring->dma, GFP_KERNEL);
4226        if (!rx_ring->desc)
4227                goto err;
4228
4229        rx_ring->next_to_alloc = 0;
4230        rx_ring->next_to_clean = 0;
4231        rx_ring->next_to_use = 0;
4232
4233        return 0;
4234
4235err:
4236        vfree(rx_ring->rx_buffer_info);
4237        rx_ring->rx_buffer_info = NULL;
4238        dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4239        return -ENOMEM;
4240}
4241
4242/**
4243 *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
4244 *                               (Descriptors) for all queues
4245 *  @adapter: board private structure
4246 *
4247 *  Return 0 on success, negative on failure
4248 **/
4249static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4250{
4251        struct pci_dev *pdev = adapter->pdev;
4252        int i, err = 0;
4253
4254        for (i = 0; i < adapter->num_rx_queues; i++) {
4255                err = igb_setup_rx_resources(adapter->rx_ring[i]);
4256                if (err) {
4257                        dev_err(&pdev->dev,
4258                                "Allocation for Rx Queue %u failed\n", i);
4259                        for (i--; i >= 0; i--)
4260                                igb_free_rx_resources(adapter->rx_ring[i]);
4261                        break;
4262                }
4263        }
4264
4265        return err;
4266}
4267
4268/**
4269 *  igb_setup_mrqc - configure the multiple receive queue control registers
4270 *  @adapter: Board private structure
4271 **/
4272static void igb_setup_mrqc(struct igb_adapter *adapter)
4273{
4274        struct e1000_hw *hw = &adapter->hw;
4275        u32 mrqc, rxcsum;
4276        u32 j, num_rx_queues;
4277        u32 rss_key[10];
4278
4279        netdev_rss_key_fill(rss_key, sizeof(rss_key));
4280        for (j = 0; j < 10; j++)
4281                wr32(E1000_RSSRK(j), rss_key[j]);
4282
4283        num_rx_queues = adapter->rss_queues;
4284
4285        switch (hw->mac.type) {
4286        case e1000_82576:
4287                /* 82576 supports 2 RSS queues for SR-IOV */
4288                if (adapter->vfs_allocated_count)
4289                        num_rx_queues = 2;
4290                break;
4291        default:
4292                break;
4293        }
4294
4295        if (adapter->rss_indir_tbl_init != num_rx_queues) {
4296                for (j = 0; j < IGB_RETA_SIZE; j++)
4297                        adapter->rss_indir_tbl[j] =
4298                        (j * num_rx_queues) / IGB_RETA_SIZE;
4299                adapter->rss_indir_tbl_init = num_rx_queues;
4300        }
4301        igb_write_rss_indir_tbl(adapter);
4302
4303        /* Disable raw packet checksumming so that RSS hash is placed in
4304         * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
4305         * offloads as they are enabled by default
4306         */
4307        rxcsum = rd32(E1000_RXCSUM);
4308        rxcsum |= E1000_RXCSUM_PCSD;
4309
4310        if (adapter->hw.mac.type >= e1000_82576)
4311                /* Enable Receive Checksum Offload for SCTP */
4312                rxcsum |= E1000_RXCSUM_CRCOFL;
4313
4314        /* Don't need to set TUOFL or IPOFL, they default to 1 */
4315        wr32(E1000_RXCSUM, rxcsum);
4316
4317        /* Generate RSS hash based on packet types, TCP/UDP
4318         * port numbers and/or IPv4/v6 src and dst addresses
4319         */
4320        mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4321               E1000_MRQC_RSS_FIELD_IPV4_TCP |
4322               E1000_MRQC_RSS_FIELD_IPV6 |
4323               E1000_MRQC_RSS_FIELD_IPV6_TCP |
4324               E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4325
4326        if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4327                mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4328        if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4329                mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4330
4331        /* If VMDq is enabled then we set the appropriate mode for that, else
4332         * we default to RSS so that an RSS hash is calculated per packet even
4333         * if we are only using one queue
4334         */
4335        if (adapter->vfs_allocated_count) {
4336                if (hw->mac.type > e1000_82575) {
4337                        /* Set the default pool for the PF's first queue */
4338                        u32 vtctl = rd32(E1000_VT_CTL);
4339
4340                        vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4341                                   E1000_VT_CTL_DISABLE_DEF_POOL);
4342                        vtctl |= adapter->vfs_allocated_count <<
4343                                E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4344                        wr32(E1000_VT_CTL, vtctl);
4345                }
4346                if (adapter->rss_queues > 1)
4347                        mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4348                else
4349                        mrqc |= E1000_MRQC_ENABLE_VMDQ;
4350        } else {
4351                if (hw->mac.type != e1000_i211)
4352                        mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4353        }
4354        igb_vmm_control(adapter);
4355
4356        wr32(E1000_MRQC, mrqc);
4357}
4358
4359/**
4360 *  igb_setup_rctl - configure the receive control registers
4361 *  @adapter: Board private structure
4362 **/
4363void igb_setup_rctl(struct igb_adapter *adapter)
4364{
4365        struct e1000_hw *hw = &adapter->hw;
4366        u32 rctl;
4367
4368        rctl = rd32(E1000_RCTL);
4369
4370        rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4371        rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4372
4373        rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4374                (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4375
4376        /* enable stripping of CRC. It's unlikely this will break BMC
4377         * redirection as it did with e1000. Newer features require
4378         * that the HW strips the CRC.
4379         */
4380        rctl |= E1000_RCTL_SECRC;
4381
4382        /* disable store bad packets and clear size bits. */
4383        rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4384
4385        /* enable LPE to allow for reception of jumbo frames */
4386        rctl |= E1000_RCTL_LPE;
4387
4388        /* disable queue 0 to prevent tail write w/o re-config */
4389        wr32(E1000_RXDCTL(0), 0);
4390
4391        /* Attention!!!  For SR-IOV PF driver operations you must enable
4392         * queue drop for all VF and PF queues to prevent head of line blocking
4393         * if an un-trusted VF does not provide descriptors to hardware.
4394         */
4395        if (adapter->vfs_allocated_count) {
4396                /* set all queue drop enable bits */
4397                wr32(E1000_QDE, ALL_QUEUES);
4398        }
4399
4400        /* This is useful for sniffing bad packets. */
4401        if (adapter->netdev->features & NETIF_F_RXALL) {
4402                /* UPE and MPE will be handled by normal PROMISC logic
4403                 * in e1000e_set_rx_mode
4404                 */
4405                rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4406                         E1000_RCTL_BAM | /* RX All Bcast Pkts */
4407                         E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4408
4409                rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4410                          E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4411                /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4412                 * and that breaks VLANs.
4413                 */
4414        }
4415
4416        wr32(E1000_RCTL, rctl);
4417}
4418
4419static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4420                                   int vfn)
4421{
4422        struct e1000_hw *hw = &adapter->hw;
4423        u32 vmolr;
4424
4425        if (size > MAX_JUMBO_FRAME_SIZE)
4426                size = MAX_JUMBO_FRAME_SIZE;
4427
4428        vmolr = rd32(E1000_VMOLR(vfn));
4429        vmolr &= ~E1000_VMOLR_RLPML_MASK;
4430        vmolr |= size | E1000_VMOLR_LPE;
4431        wr32(E1000_VMOLR(vfn), vmolr);
4432
4433        return 0;
4434}
4435
4436static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4437                                         int vfn, bool enable)
4438{
4439        struct e1000_hw *hw = &adapter->hw;
4440        u32 val, reg;
4441
4442        if (hw->mac.type < e1000_82576)
4443                return;
4444
4445        if (hw->mac.type == e1000_i350)
4446                reg = E1000_DVMOLR(vfn);
4447        else
4448                reg = E1000_VMOLR(vfn);
4449
4450        val = rd32(reg);
4451        if (enable)
4452                val |= E1000_VMOLR_STRVLAN;
4453        else
4454                val &= ~(E1000_VMOLR_STRVLAN);
4455        wr32(reg, val);
4456}
4457
4458static inline void igb_set_vmolr(struct igb_adapter *adapter,
4459                                 int vfn, bool aupe)
4460{
4461        struct e1000_hw *hw = &adapter->hw;
4462        u32 vmolr;
4463
4464        /* This register exists only on 82576 and newer so if we are older then
4465         * we should exit and do nothing
4466         */
4467        if (hw->mac.type < e1000_82576)
4468                return;
4469
4470        vmolr = rd32(E1000_VMOLR(vfn));
4471        if (aupe)
4472                vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4473        else
4474                vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4475
4476        /* clear all bits that might not be set */
4477        vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4478
4479        if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4480                vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4481        /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4482         * multicast packets
4483         */
4484        if (vfn <= adapter->vfs_allocated_count)
4485                vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4486
4487        wr32(E1000_VMOLR(vfn), vmolr);
4488}
4489
4490/**
4491 *  igb_configure_rx_ring - Configure a receive ring after Reset
4492 *  @adapter: board private structure
4493 *  @ring: receive ring to be configured
4494 *
4495 *  Configure the Rx unit of the MAC after a reset.
4496 **/
4497void igb_configure_rx_ring(struct igb_adapter *adapter,
4498                           struct igb_ring *ring)
4499{
4500        struct e1000_hw *hw = &adapter->hw;
4501        union e1000_adv_rx_desc *rx_desc;
4502        u64 rdba = ring->dma;
4503        int reg_idx = ring->reg_idx;
4504        u32 srrctl = 0, rxdctl = 0;
4505
4506        /* disable the queue */
4507        wr32(E1000_RXDCTL(reg_idx), 0);
4508
4509        /* Set DMA base address registers */
4510        wr32(E1000_RDBAL(reg_idx),
4511             rdba & 0x00000000ffffffffULL);
4512        wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4513        wr32(E1000_RDLEN(reg_idx),
4514             ring->count * sizeof(union e1000_adv_rx_desc));
4515
4516        /* initialize head and tail */
4517        ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4518        wr32(E1000_RDH(reg_idx), 0);
4519        writel(0, ring->tail);
4520
4521        /* set descriptor configuration */
4522        srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4523        if (ring_uses_large_buffer(ring))
4524                srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4525        else
4526                srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4527        srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4528        if (hw->mac.type >= e1000_82580)
4529                srrctl |= E1000_SRRCTL_TIMESTAMP;
4530        /* Only set Drop Enable if we are supporting multiple queues */
4531        if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
4532                srrctl |= E1000_SRRCTL_DROP_EN;
4533
4534        wr32(E1000_SRRCTL(reg_idx), srrctl);
4535
4536        /* set filtering for VMDQ pools */
4537        igb_set_vmolr(adapter, reg_idx & 0x7, true);
4538
4539        rxdctl |= IGB_RX_PTHRESH;
4540        rxdctl |= IGB_RX_HTHRESH << 8;
4541        rxdctl |= IGB_RX_WTHRESH << 16;
4542
4543        /* initialize rx_buffer_info */
4544        memset(ring->rx_buffer_info, 0,
4545               sizeof(struct igb_rx_buffer) * ring->count);
4546
4547        /* initialize Rx descriptor 0 */
4548        rx_desc = IGB_RX_DESC(ring, 0);
4549        rx_desc->wb.upper.length = 0;
4550
4551        /* enable receive descriptor fetching */
4552        rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4553        wr32(E1000_RXDCTL(reg_idx), rxdctl);
4554}
4555
4556static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4557                                  struct igb_ring *rx_ring)
4558{
4559        /* set build_skb and buffer size flags */
4560        clear_ring_build_skb_enabled(rx_ring);
4561        clear_ring_uses_large_buffer(rx_ring);
4562
4563        if (adapter->flags & IGB_FLAG_RX_LEGACY)
4564                return;
4565
4566        set_ring_build_skb_enabled(rx_ring);
4567
4568#if (PAGE_SIZE < 8192)
4569        if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4570                return;
4571
4572        set_ring_uses_large_buffer(rx_ring);
4573#endif
4574}
4575
4576/**
4577 *  igb_configure_rx - Configure receive Unit after Reset
4578 *  @adapter: board private structure
4579 *
4580 *  Configure the Rx unit of the MAC after a reset.
4581 **/
4582static void igb_configure_rx(struct igb_adapter *adapter)
4583{
4584        int i;
4585
4586        /* set the correct pool for the PF default MAC address in entry 0 */
4587        igb_set_default_mac_filter(adapter);
4588
4589        /* Setup the HW Rx Head and Tail Descriptor Pointers and
4590         * the Base and Length of the Rx Descriptor Ring
4591         */
4592        for (i = 0; i < adapter->num_rx_queues; i++) {
4593                struct igb_ring *rx_ring = adapter->rx_ring[i];
4594
4595                igb_set_rx_buffer_len(adapter, rx_ring);
4596                igb_configure_rx_ring(adapter, rx_ring);
4597        }
4598}
4599
4600/**
4601 *  igb_free_tx_resources - Free Tx Resources per Queue
4602 *  @tx_ring: Tx descriptor ring for a specific queue
4603 *
4604 *  Free all transmit software resources
4605 **/
4606void igb_free_tx_resources(struct igb_ring *tx_ring)
4607{
4608        igb_clean_tx_ring(tx_ring);
4609
4610        vfree(tx_ring->tx_buffer_info);
4611        tx_ring->tx_buffer_info = NULL;
4612
4613        /* if not set, then don't free */
4614        if (!tx_ring->desc)
4615                return;
4616
4617        dma_free_coherent(tx_ring->dev, tx_ring->size,
4618                          tx_ring->desc, tx_ring->dma);
4619
4620        tx_ring->desc = NULL;
4621}
4622
4623/**
4624 *  igb_free_all_tx_resources - Free Tx Resources for All Queues
4625 *  @adapter: board private structure
4626 *
4627 *  Free all transmit software resources
4628 **/
4629static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4630{
4631        int i;
4632
4633        for (i = 0; i < adapter->num_tx_queues; i++)
4634                if (adapter->tx_ring[i])
4635                        igb_free_tx_resources(adapter->tx_ring[i]);
4636}
4637
4638/**
4639 *  igb_clean_tx_ring - Free Tx Buffers
4640 *  @tx_ring: ring to be cleaned
4641 **/
4642static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4643{
4644        u16 i = tx_ring->next_to_clean;
4645        struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4646
4647        while (i != tx_ring->next_to_use) {
4648                union e1000_adv_tx_desc *eop_desc, *tx_desc;
4649
4650                /* Free all the Tx ring sk_buffs */
4651                dev_kfree_skb_any(tx_buffer->skb);
4652
4653                /* unmap skb header data */
4654                dma_unmap_single(tx_ring->dev,
4655                                 dma_unmap_addr(tx_buffer, dma),
4656                                 dma_unmap_len(tx_buffer, len),
4657                                 DMA_TO_DEVICE);
4658
4659                /* check for eop_desc to determine the end of the packet */
4660                eop_desc = tx_buffer->next_to_watch;
4661                tx_desc = IGB_TX_DESC(tx_ring, i);
4662
4663                /* unmap remaining buffers */
4664                while (tx_desc != eop_desc) {
4665                        tx_buffer++;
4666                        tx_desc++;
4667                        i++;
4668                        if (unlikely(i == tx_ring->count)) {
4669                                i = 0;
4670                                tx_buffer = tx_ring->tx_buffer_info;
4671                                tx_desc = IGB_TX_DESC(tx_ring, 0);
4672                        }
4673
4674                        /* unmap any remaining paged data */
4675                        if (dma_unmap_len(tx_buffer, len))
4676                                dma_unmap_page(tx_ring->dev,
4677                                               dma_unmap_addr(tx_buffer, dma),
4678                                               dma_unmap_len(tx_buffer, len),
4679                                               DMA_TO_DEVICE);
4680                }
4681
4682                /* move us one more past the eop_desc for start of next pkt */
4683                tx_buffer++;
4684                i++;
4685                if (unlikely(i == tx_ring->count)) {
4686                        i = 0;
4687                        tx_buffer = tx_ring->tx_buffer_info;
4688                }
4689        }
4690
4691        /* reset BQL for queue */
4692        netdev_tx_reset_queue(txring_txq(tx_ring));
4693
4694        /* reset next_to_use and next_to_clean */
4695        tx_ring->next_to_use = 0;
4696        tx_ring->next_to_clean = 0;
4697}
4698
4699/**
4700 *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
4701 *  @adapter: board private structure
4702 **/
4703static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4704{
4705        int i;
4706
4707        for (i = 0; i < adapter->num_tx_queues; i++)
4708                if (adapter->tx_ring[i])
4709                        igb_clean_tx_ring(adapter->tx_ring[i]);
4710}
4711
4712/**
4713 *  igb_free_rx_resources - Free Rx Resources
4714 *  @rx_ring: ring to clean the resources from
4715 *
4716 *  Free all receive software resources
4717 **/
4718void igb_free_rx_resources(struct igb_ring *rx_ring)
4719{
4720        igb_clean_rx_ring(rx_ring);
4721
4722        vfree(rx_ring->rx_buffer_info);
4723        rx_ring->rx_buffer_info = NULL;
4724
4725        /* if not set, then don't free */
4726        if (!rx_ring->desc)
4727                return;
4728
4729        dma_free_coherent(rx_ring->dev, rx_ring->size,
4730                          rx_ring->desc, rx_ring->dma);
4731
4732        rx_ring->desc = NULL;
4733}
4734
4735/**
4736 *  igb_free_all_rx_resources - Free Rx Resources for All Queues
4737 *  @adapter: board private structure
4738 *
4739 *  Free all receive software resources
4740 **/
4741static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4742{
4743        int i;
4744
4745        for (i = 0; i < adapter->num_rx_queues; i++)
4746                if (adapter->rx_ring[i])
4747                        igb_free_rx_resources(adapter->rx_ring[i]);
4748}
4749
4750/**
4751 *  igb_clean_rx_ring - Free Rx Buffers per Queue
4752 *  @rx_ring: ring to free buffers from
4753 **/
4754static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4755{
4756        u16 i = rx_ring->next_to_clean;
4757
4758        if (rx_ring->skb)
4759                dev_kfree_skb(rx_ring->skb);
4760        rx_ring->skb = NULL;
4761
4762        /* Free all the Rx ring sk_buffs */
4763        while (i != rx_ring->next_to_alloc) {
4764                struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4765
4766                /* Invalidate cache lines that may have been written to by
4767                 * device so that we avoid corrupting memory.
4768                 */
4769                dma_sync_single_range_for_cpu(rx_ring->dev,
4770                                              buffer_info->dma,
4771                                              buffer_info->page_offset,
4772                                              igb_rx_bufsz(rx_ring),
4773                                              DMA_FROM_DEVICE);
4774
4775                /* free resources associated with mapping */
4776                dma_unmap_page_attrs(rx_ring->dev,
4777                                     buffer_info->dma,
4778                                     igb_rx_pg_size(rx_ring),
4779                                     DMA_FROM_DEVICE,
4780                                     IGB_RX_DMA_ATTR);
4781                __page_frag_cache_drain(buffer_info->page,
4782                                        buffer_info->pagecnt_bias);
4783
4784                i++;
4785                if (i == rx_ring->count)
4786                        i = 0;
4787        }
4788
4789        rx_ring->next_to_alloc = 0;
4790        rx_ring->next_to_clean = 0;
4791        rx_ring->next_to_use = 0;
4792}
4793
4794/**
4795 *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
4796 *  @adapter: board private structure
4797 **/
4798static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4799{
4800        int i;
4801
4802        for (i = 0; i < adapter->num_rx_queues; i++)
4803                if (adapter->rx_ring[i])
4804                        igb_clean_rx_ring(adapter->rx_ring[i]);
4805}
4806
4807/**
4808 *  igb_set_mac - Change the Ethernet Address of the NIC
4809 *  @netdev: network interface device structure
4810 *  @p: pointer to an address structure
4811 *
4812 *  Returns 0 on success, negative on failure
4813 **/
4814static int igb_set_mac(struct net_device *netdev, void *p)
4815{
4816        struct igb_adapter *adapter = netdev_priv(netdev);
4817        struct e1000_hw *hw = &adapter->hw;
4818        struct sockaddr *addr = p;
4819
4820        if (!is_valid_ether_addr(addr->sa_data))
4821                return -EADDRNOTAVAIL;
4822
4823        memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4824        memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4825
4826        /* set the correct pool for the new PF MAC address in entry 0 */
4827        igb_set_default_mac_filter(adapter);
4828
4829        return 0;
4830}
4831
4832/**
4833 *  igb_write_mc_addr_list - write multicast addresses to MTA
4834 *  @netdev: network interface device structure
4835 *
4836 *  Writes multicast address list to the MTA hash table.
4837 *  Returns: -ENOMEM on failure
4838 *           0 on no addresses written
4839 *           X on writing X addresses to MTA
4840 **/
4841static int igb_write_mc_addr_list(struct net_device *netdev)
4842{
4843        struct igb_adapter *adapter = netdev_priv(netdev);
4844        struct e1000_hw *hw = &adapter->hw;
4845        struct netdev_hw_addr *ha;
4846        u8  *mta_list;
4847        int i;
4848
4849        if (netdev_mc_empty(netdev)) {
4850                /* nothing to program, so clear mc list */
4851                igb_update_mc_addr_list(hw, NULL, 0);
4852                igb_restore_vf_multicasts(adapter);
4853                return 0;
4854        }
4855
4856        mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
4857        if (!mta_list)
4858                return -ENOMEM;
4859
4860        /* The shared function expects a packed array of only addresses. */
4861        i = 0;
4862        netdev_for_each_mc_addr(ha, netdev)
4863                memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
4864
4865        igb_update_mc_addr_list(hw, mta_list, i);
4866        kfree(mta_list);
4867
4868        return netdev_mc_count(netdev);
4869}
4870
4871static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
4872{
4873        struct e1000_hw *hw = &adapter->hw;
4874        u32 i, pf_id;
4875
4876        switch (hw->mac.type) {
4877        case e1000_i210:
4878        case e1000_i211:
4879        case e1000_i350:
4880                /* VLAN filtering needed for VLAN prio filter */
4881                if (adapter->netdev->features & NETIF_F_NTUPLE)
4882                        break;
4883                /* fall through */
4884        case e1000_82576:
4885        case e1000_82580:
4886        case e1000_i354:
4887                /* VLAN filtering needed for pool filtering */
4888                if (adapter->vfs_allocated_count)
4889                        break;
4890                /* fall through */
4891        default:
4892                return 1;
4893        }
4894
4895        /* We are already in VLAN promisc, nothing to do */
4896        if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
4897                return 0;
4898
4899        if (!adapter->vfs_allocated_count)
4900                goto set_vfta;
4901
4902        /* Add PF to all active pools */
4903        pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4904
4905        for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4906                u32 vlvf = rd32(E1000_VLVF(i));
4907
4908                vlvf |= BIT(pf_id);
4909                wr32(E1000_VLVF(i), vlvf);
4910        }
4911
4912set_vfta:
4913        /* Set all bits in the VLAN filter table array */
4914        for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
4915                hw->mac.ops.write_vfta(hw, i, ~0U);
4916
4917        /* Set flag so we don't redo unnecessary work */
4918        adapter->flags |= IGB_FLAG_VLAN_PROMISC;
4919
4920        return 0;
4921}
4922
4923#define VFTA_BLOCK_SIZE 8
4924static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
4925{
4926        struct e1000_hw *hw = &adapter->hw;
4927        u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4928        u32 vid_start = vfta_offset * 32;
4929        u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4930        u32 i, vid, word, bits, pf_id;
4931
4932        /* guarantee that we don't scrub out management VLAN */
4933        vid = adapter->mng_vlan_id;
4934        if (vid >= vid_start && vid < vid_end)
4935                vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4936
4937        if (!adapter->vfs_allocated_count)
4938                goto set_vfta;
4939
4940        pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4941
4942        for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4943                u32 vlvf = rd32(E1000_VLVF(i));
4944
4945                /* pull VLAN ID from VLVF */
4946                vid = vlvf & VLAN_VID_MASK;
4947
4948                /* only concern ourselves with a certain range */
4949                if (vid < vid_start || vid >= vid_end)
4950                        continue;
4951
4952                if (vlvf & E1000_VLVF_VLANID_ENABLE) {
4953                        /* record VLAN ID in VFTA */
4954                        vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4955
4956                        /* if PF is part of this then continue */
4957                        if (test_bit(vid, adapter->active_vlans))
4958                                continue;
4959                }
4960
4961                /* remove PF from the pool */
4962                bits = ~BIT(pf_id);
4963                bits &= rd32(E1000_VLVF(i));
4964                wr32(E1000_VLVF(i), bits);
4965        }
4966
4967set_vfta:
4968        /* extract values from active_vlans and write back to VFTA */
4969        for (i = VFTA_BLOCK_SIZE; i--;) {
4970                vid = (vfta_offset + i) * 32;
4971                word = vid / BITS_PER_LONG;
4972                bits = vid % BITS_PER_LONG;
4973
4974                vfta[i] |= adapter->active_vlans[word] >> bits;
4975
4976                hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
4977        }
4978}
4979
4980static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
4981{
4982        u32 i;
4983
4984        /* We are not in VLAN promisc, nothing to do */
4985        if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
4986                return;
4987
4988        /* Set flag so we don't redo unnecessary work */
4989        adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
4990
4991        for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
4992                igb_scrub_vfta(adapter, i);
4993}
4994
4995/**
4996 *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
4997 *  @netdev: network interface device structure
4998 *
4999 *  The set_rx_mode entry point is called whenever the unicast or multicast
5000 *  address lists or the network interface flags are updated.  This routine is
5001 *  responsible for configuring the hardware for proper unicast, multicast,
5002 *  promiscuous mode, and all-multi behavior.
5003 **/
5004static void igb_set_rx_mode(struct net_device *netdev)
5005{
5006        struct igb_adapter *adapter = netdev_priv(netdev);
5007        struct e1000_hw *hw = &adapter->hw;
5008        unsigned int vfn = adapter->vfs_allocated_count;
5009        u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5010        int count;
5011
5012        /* Check for Promiscuous and All Multicast modes */
5013        if (netdev->flags & IFF_PROMISC) {
5014                rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5015                vmolr |= E1000_VMOLR_MPME;
5016
5017                /* enable use of UTA filter to force packets to default pool */
5018                if (hw->mac.type == e1000_82576)
5019                        vmolr |= E1000_VMOLR_ROPE;
5020        } else {
5021                if (netdev->flags & IFF_ALLMULTI) {
5022                        rctl |= E1000_RCTL_MPE;
5023                        vmolr |= E1000_VMOLR_MPME;
5024                } else {
5025                        /* Write addresses to the MTA, if the attempt fails
5026                         * then we should just turn on promiscuous mode so
5027                         * that we can at least receive multicast traffic
5028                         */
5029                        count = igb_write_mc_addr_list(netdev);
5030                        if (count < 0) {
5031                                rctl |= E1000_RCTL_MPE;
5032                                vmolr |= E1000_VMOLR_MPME;
5033                        } else if (count) {
5034                                vmolr |= E1000_VMOLR_ROMPE;
5035                        }
5036                }
5037        }
5038
5039        /* Write addresses to available RAR registers, if there is not
5040         * sufficient space to store all the addresses then enable
5041         * unicast promiscuous mode
5042         */
5043        if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5044                rctl |= E1000_RCTL_UPE;
5045                vmolr |= E1000_VMOLR_ROPE;
5046        }
5047
5048        /* enable VLAN filtering by default */
5049        rctl |= E1000_RCTL_VFE;
5050
5051        /* disable VLAN filtering for modes that require it */
5052        if ((netdev->flags & IFF_PROMISC) ||
5053            (netdev->features & NETIF_F_RXALL)) {
5054                /* if we fail to set all rules then just clear VFE */
5055                if (igb_vlan_promisc_enable(adapter))
5056                        rctl &= ~E1000_RCTL_VFE;
5057        } else {
5058                igb_vlan_promisc_disable(adapter);
5059        }
5060
5061        /* update state of unicast, multicast, and VLAN filtering modes */
5062        rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5063                                     E1000_RCTL_VFE);
5064        wr32(E1000_RCTL, rctl);
5065
5066#if (PAGE_SIZE < 8192)
5067        if (!adapter->vfs_allocated_count) {
5068                if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5069                        rlpml = IGB_MAX_FRAME_BUILD_SKB;
5070        }
5071#endif
5072        wr32(E1000_RLPML, rlpml);
5073
5074        /* In order to support SR-IOV and eventually VMDq it is necessary to set
5075         * the VMOLR to enable the appropriate modes.  Without this workaround
5076         * we will have issues with VLAN tag stripping not being done for frames
5077         * that are only arriving because we are the default pool
5078         */
5079        if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5080                return;
5081
5082        /* set UTA to appropriate mode */
5083        igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5084
5085        vmolr |= rd32(E1000_VMOLR(vfn)) &
5086                 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5087
5088        /* enable Rx jumbo frames, restrict as needed to support build_skb */
5089        vmolr &= ~E1000_VMOLR_RLPML_MASK;
5090#if (PAGE_SIZE < 8192)
5091        if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5092                vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5093        else
5094#endif
5095                vmolr |= MAX_JUMBO_FRAME_SIZE;
5096        vmolr |= E1000_VMOLR_LPE;
5097
5098        wr32(E1000_VMOLR(vfn), vmolr);
5099
5100        igb_restore_vf_multicasts(adapter);
5101}
5102
5103static void igb_check_wvbr(struct igb_adapter *adapter)
5104{
5105        struct e1000_hw *hw = &adapter->hw;
5106        u32 wvbr = 0;
5107
5108        switch (hw->mac.type) {
5109        case e1000_82576:
5110        case e1000_i350:
5111                wvbr = rd32(E1000_WVBR);
5112                if (!wvbr)
5113                        return;
5114                break;
5115        default:
5116                break;
5117        }
5118
5119        adapter->wvbr |= wvbr;
5120}
5121
5122#define IGB_STAGGERED_QUEUE_OFFSET 8
5123
5124static void igb_spoof_check(struct igb_adapter *adapter)
5125{
5126        int j;
5127
5128        if (!adapter->wvbr)
5129                return;
5130
5131        for (j = 0; j < adapter->vfs_allocated_count; j++) {
5132                if (adapter->wvbr & BIT(j) ||
5133                    adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5134                        dev_warn(&adapter->pdev->dev,
5135                                "Spoof event(s) detected on VF %d\n", j);
5136                        adapter->wvbr &=
5137                                ~(BIT(j) |
5138                                  BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5139                }
5140        }
5141}
5142
5143/* Need to wait a few seconds after link up to get diagnostic information from
5144 * the phy
5145 */
5146static void igb_update_phy_info(struct timer_list *t)
5147{
5148        struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5149        igb_get_phy_info(&adapter->hw);
5150}
5151
5152/**
5153 *  igb_has_link - check shared code for link and determine up/down
5154 *  @adapter: pointer to driver private info
5155 **/
5156bool igb_has_link(struct igb_adapter *adapter)
5157{
5158        struct e1000_hw *hw = &adapter->hw;
5159        bool link_active = false;
5160
5161        /* get_link_status is set on LSC (link status) interrupt or
5162         * rx sequence error interrupt.  get_link_status will stay
5163         * false until the e1000_check_for_link establishes link
5164         * for copper adapters ONLY
5165         */
5166        switch (hw->phy.media_type) {
5167        case e1000_media_type_copper:
5168                if (!hw->mac.get_link_status)
5169                        return true;
5170                /* fall through */
5171        case e1000_media_type_internal_serdes:
5172                hw->mac.ops.check_for_link(hw);
5173                link_active = !hw->mac.get_link_status;
5174                break;
5175        default:
5176        case e1000_media_type_unknown:
5177                break;
5178        }
5179
5180        if (((hw->mac.type == e1000_i210) ||
5181             (hw->mac.type == e1000_i211)) &&
5182             (hw->phy.id == I210_I_PHY_ID)) {
5183                if (!netif_carrier_ok(adapter->netdev)) {
5184                        adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5185                } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5186                        adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5187                        adapter->link_check_timeout = jiffies;
5188                }
5189        }
5190
5191        return link_active;
5192}
5193
5194static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5195{
5196        bool ret = false;
5197        u32 ctrl_ext, thstat;
5198
5199        /* check for thermal sensor event on i350 copper only */
5200        if (hw->mac.type == e1000_i350) {
5201                thstat = rd32(E1000_THSTAT);
5202                ctrl_ext = rd32(E1000_CTRL_EXT);
5203
5204                if ((hw->phy.media_type == e1000_media_type_copper) &&
5205                    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5206                        ret = !!(thstat & event);
5207        }
5208
5209        return ret;
5210}
5211
5212/**
5213 *  igb_check_lvmmc - check for malformed packets received
5214 *  and indicated in LVMMC register
5215 *  @adapter: pointer to adapter
5216 **/
5217static void igb_check_lvmmc(struct igb_adapter *adapter)
5218{
5219        struct e1000_hw *hw = &adapter->hw;
5220        u32 lvmmc;
5221
5222        lvmmc = rd32(E1000_LVMMC);
5223        if (lvmmc) {
5224                if (unlikely(net_ratelimit())) {
5225                        netdev_warn(adapter->netdev,
5226                                    "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5227                                    lvmmc);
5228                }
5229        }
5230}
5231
5232/**
5233 *  igb_watchdog - Timer Call-back
5234 *  @data: pointer to adapter cast into an unsigned long
5235 **/
5236static void igb_watchdog(struct timer_list *t)
5237{
5238        struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5239        /* Do the rest outside of interrupt context */
5240        schedule_work(&adapter->watchdog_task);
5241}
5242
5243static void igb_watchdog_task(struct work_struct *work)
5244{
5245        struct igb_adapter *adapter = container_of(work,
5246                                                   struct igb_adapter,
5247                                                   watchdog_task);
5248        struct e1000_hw *hw = &adapter->hw;
5249        struct e1000_phy_info *phy = &hw->phy;
5250        struct net_device *netdev = adapter->netdev;
5251        u32 link;
5252        int i;
5253        u32 connsw;
5254        u16 phy_data, retry_count = 20;
5255
5256        link = igb_has_link(adapter);
5257
5258        if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5259                if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5260                        adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5261                else
5262                        link = false;
5263        }
5264
5265        /* Force link down if we have fiber to swap to */
5266        if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5267                if (hw->phy.media_type == e1000_media_type_copper) {
5268                        connsw = rd32(E1000_CONNSW);
5269                        if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5270                                link = 0;
5271                }
5272        }
5273        if (link) {
5274                /* Perform a reset if the media type changed. */
5275                if (hw->dev_spec._82575.media_changed) {
5276                        hw->dev_spec._82575.media_changed = false;
5277                        adapter->flags |= IGB_FLAG_MEDIA_RESET;
5278                        igb_reset(adapter);
5279                }
5280                /* Cancel scheduled suspend requests. */
5281                pm_runtime_resume(netdev->dev.parent);
5282
5283                if (!netif_carrier_ok(netdev)) {
5284                        u32 ctrl;
5285
5286                        hw->mac.ops.get_speed_and_duplex(hw,
5287                                                         &adapter->link_speed,
5288                                                         &adapter->link_duplex);
5289
5290                        ctrl = rd32(E1000_CTRL);
5291                        /* Links status message must follow this format */
5292                        netdev_info(netdev,
5293                               "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5294                               netdev->name,
5295                               adapter->link_speed,
5296                               adapter->link_duplex == FULL_DUPLEX ?
5297                               "Full" : "Half",
5298                               (ctrl & E1000_CTRL_TFCE) &&
5299                               (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5300                               (ctrl & E1000_CTRL_RFCE) ?  "RX" :
5301                               (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
5302
5303                        /* disable EEE if enabled */
5304                        if ((adapter->flags & IGB_FLAG_EEE) &&
5305                                (adapter->link_duplex == HALF_DUPLEX)) {
5306                                dev_info(&adapter->pdev->dev,
5307                                "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5308                                adapter->hw.dev_spec._82575.eee_disable = true;
5309                                adapter->flags &= ~IGB_FLAG_EEE;
5310                        }
5311
5312                        /* check if SmartSpeed worked */
5313                        igb_check_downshift(hw);
5314                        if (phy->speed_downgraded)
5315                                netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5316
5317                        /* check for thermal sensor event */
5318                        if (igb_thermal_sensor_event(hw,
5319                            E1000_THSTAT_LINK_THROTTLE))
5320                                netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5321
5322                        /* adjust timeout factor according to speed/duplex */
5323                        adapter->tx_timeout_factor = 1;
5324                        switch (adapter->link_speed) {
5325                        case SPEED_10:
5326                                adapter->tx_timeout_factor = 14;
5327                                break;
5328                        case SPEED_100:
5329                                /* maybe add some timeout factor ? */
5330                                break;
5331                        }
5332
5333                        if (adapter->link_speed != SPEED_1000)
5334                                goto no_wait;
5335
5336                        /* wait for Remote receiver status OK */
5337retry_read_status:
5338                        if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5339                                              &phy_data)) {
5340                                if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5341                                    retry_count) {
5342                                        msleep(100);
5343                                        retry_count--;
5344                                        goto retry_read_status;
5345                                } else if (!retry_count) {
5346                                        dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5347                                }
5348                        } else {
5349                                dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5350                        }
5351no_wait:
5352                        netif_carrier_on(netdev);
5353
5354                        igb_ping_all_vfs(adapter);
5355                        igb_check_vf_rate_limit(adapter);
5356
5357                        /* link state has changed, schedule phy info update */
5358                        if (!test_bit(__IGB_DOWN, &adapter->state))
5359                                mod_timer(&adapter->phy_info_timer,
5360                                          round_jiffies(jiffies + 2 * HZ));
5361                }
5362        } else {
5363                if (netif_carrier_ok(netdev)) {
5364                        adapter->link_speed = 0;
5365                        adapter->link_duplex = 0;
5366
5367                        /* check for thermal sensor event */
5368                        if (igb_thermal_sensor_event(hw,
5369                            E1000_THSTAT_PWR_DOWN)) {
5370                                netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5371                        }
5372
5373                        /* Links status message must follow this format */
5374                        netdev_info(netdev, "igb: %s NIC Link is Down\n",
5375                               netdev->name);
5376                        netif_carrier_off(netdev);
5377
5378                        igb_ping_all_vfs(adapter);
5379
5380                        /* link state has changed, schedule phy info update */
5381                        if (!test_bit(__IGB_DOWN, &adapter->state))
5382                                mod_timer(&adapter->phy_info_timer,
5383                                          round_jiffies(jiffies + 2 * HZ));
5384
5385                        /* link is down, time to check for alternate media */
5386                        if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5387                                igb_check_swap_media(adapter);
5388                                if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5389                                        schedule_work(&adapter->reset_task);
5390                                        /* return immediately */
5391                                        return;
5392                                }
5393                        }
5394                        pm_schedule_suspend(netdev->dev.parent,
5395                                            MSEC_PER_SEC * 5);
5396
5397                /* also check for alternate media here */
5398                } else if (!netif_carrier_ok(netdev) &&
5399                           (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5400                        igb_check_swap_media(adapter);
5401                        if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5402                                schedule_work(&adapter->reset_task);
5403                                /* return immediately */
5404                                return;
5405                        }
5406                }
5407        }
5408
5409        spin_lock(&adapter->stats64_lock);
5410        igb_update_stats(adapter);
5411        spin_unlock(&adapter->stats64_lock);
5412
5413        for (i = 0; i < adapter->num_tx_queues; i++) {
5414                struct igb_ring *tx_ring = adapter->tx_ring[i];
5415                if (!netif_carrier_ok(netdev)) {
5416                        /* We've lost link, so the controller stops DMA,
5417                         * but we've got queued Tx work that's never going
5418                         * to get done, so reset controller to flush Tx.
5419                         * (Do the reset outside of interrupt context).
5420                         */
5421                        if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5422                                adapter->tx_timeout_count++;
5423                                schedule_work(&adapter->reset_task);
5424                                /* return immediately since reset is imminent */
5425                                return;
5426                        }
5427                }
5428
5429                /* Force detection of hung controller every watchdog period */
5430                set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5431        }
5432
5433        /* Cause software interrupt to ensure Rx ring is cleaned */
5434        if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5435                u32 eics = 0;
5436
5437                for (i = 0; i < adapter->num_q_vectors; i++)
5438                        eics |= adapter->q_vector[i]->eims_value;
5439                wr32(E1000_EICS, eics);
5440        } else {
5441                wr32(E1000_ICS, E1000_ICS_RXDMT0);
5442        }
5443
5444        igb_spoof_check(adapter);
5445        igb_ptp_rx_hang(adapter);
5446        igb_ptp_tx_hang(adapter);
5447
5448        /* Check LVMMC register on i350/i354 only */
5449        if ((adapter->hw.mac.type == e1000_i350) ||
5450            (adapter->hw.mac.type == e1000_i354))
5451                igb_check_lvmmc(adapter);
5452
5453        /* Reset the timer */
5454        if (!test_bit(__IGB_DOWN, &adapter->state)) {
5455                if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5456                        mod_timer(&adapter->watchdog_timer,
5457                                  round_jiffies(jiffies +  HZ));
5458                else
5459                        mod_timer(&adapter->watchdog_timer,
5460                                  round_jiffies(jiffies + 2 * HZ));
5461        }
5462}
5463
5464enum latency_range {
5465        lowest_latency = 0,
5466        low_latency = 1,
5467        bulk_latency = 2,
5468        latency_invalid = 255
5469};
5470
5471/**
5472 *  igb_update_ring_itr - update the dynamic ITR value based on packet size
5473 *  @q_vector: pointer to q_vector
5474 *
5475 *  Stores a new ITR value based on strictly on packet size.  This
5476 *  algorithm is less sophisticated than that used in igb_update_itr,
5477 *  due to the difficulty of synchronizing statistics across multiple
5478 *  receive rings.  The divisors and thresholds used by this function
5479 *  were determined based on theoretical maximum wire speed and testing
5480 *  data, in order to minimize response time while increasing bulk
5481 *  throughput.
5482 *  This functionality is controlled by ethtool's coalescing settings.
5483 *  NOTE:  This function is called only when operating in a multiqueue
5484 *         receive environment.
5485 **/
5486static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5487{
5488        int new_val = q_vector->itr_val;
5489        int avg_wire_size = 0;
5490        struct igb_adapter *adapter = q_vector->adapter;
5491        unsigned int packets;
5492
5493        /* For non-gigabit speeds, just fix the interrupt rate at 4000
5494         * ints/sec - ITR timer value of 120 ticks.
5495         */
5496        if (adapter->link_speed != SPEED_1000) {
5497                new_val = IGB_4K_ITR;
5498                goto set_itr_val;
5499        }
5500
5501        packets = q_vector->rx.total_packets;
5502        if (packets)
5503                avg_wire_size = q_vector->rx.total_bytes / packets;
5504
5505        packets = q_vector->tx.total_packets;
5506        if (packets)
5507                avg_wire_size = max_t(u32, avg_wire_size,
5508                                      q_vector->tx.total_bytes / packets);
5509
5510        /* if avg_wire_size isn't set no work was done */
5511        if (!avg_wire_size)
5512                goto clear_counts;
5513
5514        /* Add 24 bytes to size to account for CRC, preamble, and gap */
5515        avg_wire_size += 24;
5516
5517        /* Don't starve jumbo frames */
5518        avg_wire_size = min(avg_wire_size, 3000);
5519
5520        /* Give a little boost to mid-size frames */
5521        if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5522                new_val = avg_wire_size / 3;
5523        else
5524                new_val = avg_wire_size / 2;
5525
5526        /* conservative mode (itr 3) eliminates the lowest_latency setting */
5527        if (new_val < IGB_20K_ITR &&
5528            ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5529             (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5530                new_val = IGB_20K_ITR;
5531
5532set_itr_val:
5533        if (new_val != q_vector->itr_val) {
5534                q_vector->itr_val = new_val;
5535                q_vector->set_itr = 1;
5536        }
5537clear_counts:
5538        q_vector->rx.total_bytes = 0;
5539        q_vector->rx.total_packets = 0;
5540        q_vector->tx.total_bytes = 0;
5541        q_vector->tx.total_packets = 0;
5542}
5543
5544/**
5545 *  igb_update_itr - update the dynamic ITR value based on statistics
5546 *  @q_vector: pointer to q_vector
5547 *  @ring_container: ring info to update the itr for
5548 *
5549 *  Stores a new ITR value based on packets and byte
5550 *  counts during the last interrupt.  The advantage of per interrupt
5551 *  computation is faster updates and more accurate ITR for the current
5552 *  traffic pattern.  Constants in this function were computed
5553 *  based on theoretical maximum wire speed and thresholds were set based
5554 *  on testing data as well as attempting to minimize response time
5555 *  while increasing bulk throughput.
5556 *  This functionality is controlled by ethtool's coalescing settings.
5557 *  NOTE:  These calculations are only valid when operating in a single-
5558 *         queue environment.
5559 **/
5560static void igb_update_itr(struct igb_q_vector *q_vector,
5561                           struct igb_ring_container *ring_container)
5562{
5563        unsigned int packets = ring_container->total_packets;
5564        unsigned int bytes = ring_container->total_bytes;
5565        u8 itrval = ring_container->itr;
5566
5567        /* no packets, exit with status unchanged */
5568        if (packets == 0)
5569                return;
5570
5571        switch (itrval) {
5572        case lowest_latency:
5573                /* handle TSO and jumbo frames */
5574                if (bytes/packets > 8000)
5575                        itrval = bulk_latency;
5576                else if ((packets < 5) && (bytes > 512))
5577                        itrval = low_latency;
5578                break;
5579        case low_latency:  /* 50 usec aka 20000 ints/s */
5580                if (bytes > 10000) {
5581                        /* this if handles the TSO accounting */
5582                        if (bytes/packets > 8000)
5583                                itrval = bulk_latency;
5584                        else if ((packets < 10) || ((bytes/packets) > 1200))
5585                                itrval = bulk_latency;
5586                        else if ((packets > 35))
5587                                itrval = lowest_latency;
5588                } else if (bytes/packets > 2000) {
5589                        itrval = bulk_latency;
5590                } else if (packets <= 2 && bytes < 512) {
5591                        itrval = lowest_latency;
5592                }
5593                break;
5594        case bulk_latency: /* 250 usec aka 4000 ints/s */
5595                if (bytes > 25000) {
5596                        if (packets > 35)
5597                                itrval = low_latency;
5598                } else if (bytes < 1500) {
5599                        itrval = low_latency;
5600                }
5601                break;
5602        }
5603
5604        /* clear work counters since we have the values we need */
5605        ring_container->total_bytes = 0;
5606        ring_container->total_packets = 0;
5607
5608        /* write updated itr to ring container */
5609        ring_container->itr = itrval;
5610}
5611
5612static void igb_set_itr(struct igb_q_vector *q_vector)
5613{
5614        struct igb_adapter *adapter = q_vector->adapter;
5615        u32 new_itr = q_vector->itr_val;
5616        u8 current_itr = 0;
5617
5618        /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5619        if (adapter->link_speed != SPEED_1000) {
5620                current_itr = 0;
5621                new_itr = IGB_4K_ITR;
5622                goto set_itr_now;
5623        }
5624
5625        igb_update_itr(q_vector, &q_vector->tx);
5626        igb_update_itr(q_vector, &q_vector->rx);
5627
5628        current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5629
5630        /* conservative mode (itr 3) eliminates the lowest_latency setting */
5631        if (current_itr == lowest_latency &&
5632            ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5633             (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5634                current_itr = low_latency;
5635
5636        switch (current_itr) {
5637        /* counts and packets in update_itr are dependent on these numbers */
5638        case lowest_latency:
5639                new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5640                break;
5641        case low_latency:
5642                new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5643                break;
5644        case bulk_latency:
5645                new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
5646                break;
5647        default:
5648                break;
5649        }
5650
5651set_itr_now:
5652        if (new_itr != q_vector->itr_val) {
5653                /* this attempts to bias the interrupt rate towards Bulk
5654                 * by adding intermediate steps when interrupt rate is
5655                 * increasing
5656                 */
5657                new_itr = new_itr > q_vector->itr_val ?
5658                          max((new_itr * q_vector->itr_val) /
5659                          (new_itr + (q_vector->itr_val >> 2)),
5660                          new_itr) : new_itr;
5661                /* Don't write the value here; it resets the adapter's
5662                 * internal timer, and causes us to delay far longer than
5663                 * we should between interrupts.  Instead, we write the ITR
5664                 * value at the beginning of the next interrupt so the timing
5665                 * ends up being correct.
5666                 */
5667                q_vector->itr_val = new_itr;
5668                q_vector->set_itr = 1;
5669        }
5670}
5671
5672static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5673                            struct igb_tx_buffer *first,
5674                            u32 vlan_macip_lens, u32 type_tucmd,
5675                            u32 mss_l4len_idx)
5676{
5677        struct e1000_adv_tx_context_desc *context_desc;
5678        u16 i = tx_ring->next_to_use;
5679        struct timespec64 ts;
5680
5681        context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5682
5683        i++;
5684        tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5685
5686        /* set bits to identify this as an advanced context descriptor */
5687        type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5688
5689        /* For 82575, context index must be unique per ring. */
5690        if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5691                mss_l4len_idx |= tx_ring->reg_idx << 4;
5692
5693        context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
5694        context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
5695        context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
5696
5697        /* We assume there is always a valid tx time available. Invalid times
5698         * should have been handled by the upper layers.
5699         */
5700        if (tx_ring->launchtime_enable) {
5701                ts = ns_to_timespec64(first->skb->tstamp);
5702                context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5703        } else {
5704                context_desc->seqnum_seed = 0;
5705        }
5706}
5707
5708static int igb_tso(struct igb_ring *tx_ring,
5709                   struct igb_tx_buffer *first,
5710                   u8 *hdr_len)
5711{
5712        u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5713        struct sk_buff *skb = first->skb;
5714        union {
5715                struct iphdr *v4;
5716                struct ipv6hdr *v6;
5717                unsigned char *hdr;
5718        } ip;
5719        union {
5720                struct tcphdr *tcp;
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 = E1000_ADVTXD_TUCMD_L4T_TCP;
5741
5742        /* initialize outer IP header fields */
5743        if (ip.v4->version == 4) {
5744                unsigned char *csum_start = skb_checksum_start(skb);
5745                unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5746
5747                /* IP header will have to cancel out any data that
5748                 * is not a part of the outer IP header
5749                 */
5750                ip.v4->check = csum_fold(csum_partial(trans_start,
5751                                                      csum_start - trans_start,
5752                                                      0));
5753                type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5754
5755                ip.v4->tot_len = 0;
5756                first->tx_flags |= IGB_TX_FLAGS_TSO |
5757                                   IGB_TX_FLAGS_CSUM |
5758                                   IGB_TX_FLAGS_IPV4;
5759        } else {
5760                ip.v6->payload_len = 0;
5761                first->tx_flags |= IGB_TX_FLAGS_TSO |
5762                                   IGB_TX_FLAGS_CSUM;
5763        }
5764
5765        /* determine offset of inner transport header */
5766        l4_offset = l4.hdr - skb->data;
5767
5768        /* compute length of segmentation header */
5769        *hdr_len = (l4.tcp->doff * 4) + l4_offset;
5770
5771        /* remove payload length from inner checksum */
5772        paylen = skb->len - l4_offset;
5773        csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
5774
5775        /* update gso size and bytecount with header size */
5776        first->gso_segs = skb_shinfo(skb)->gso_segs;
5777        first->bytecount += (first->gso_segs - 1) * *hdr_len;
5778
5779        /* MSS L4LEN IDX */
5780        mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5781        mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5782
5783        /* VLAN MACLEN IPLEN */
5784        vlan_macip_lens = l4.hdr - ip.hdr;
5785        vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5786        vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5787
5788        igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
5789                        type_tucmd, mss_l4len_idx);
5790
5791        return 1;
5792}
5793
5794static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb)
5795{
5796        unsigned int offset = 0;
5797
5798        ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
5799
5800        return offset == skb_checksum_start_offset(skb);
5801}
5802
5803static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5804{
5805        struct sk_buff *skb = first->skb;
5806        u32 vlan_macip_lens = 0;
5807        u32 type_tucmd = 0;
5808
5809        if (skb->ip_summed != CHECKSUM_PARTIAL) {
5810csum_failed:
5811                if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
5812                    !tx_ring->launchtime_enable)
5813                        return;
5814                goto no_csum;
5815        }
5816
5817        switch (skb->csum_offset) {
5818        case offsetof(struct tcphdr, check):
5819                type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5820                /* fall through */
5821        case offsetof(struct udphdr, check):
5822                break;
5823        case offsetof(struct sctphdr, checksum):
5824                /* validate that this is actually an SCTP request */
5825                if (((first->protocol == htons(ETH_P_IP)) &&
5826                     (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
5827                    ((first->protocol == htons(ETH_P_IPV6)) &&
5828                     igb_ipv6_csum_is_sctp(skb))) {
5829                        type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
5830                        break;
5831                }
5832                /* fall through */
5833        default:
5834                skb_checksum_help(skb);
5835                goto csum_failed;
5836        }
5837
5838        /* update TX checksum flag */
5839        first->tx_flags |= IGB_TX_FLAGS_CSUM;
5840        vlan_macip_lens = skb_checksum_start_offset(skb) -
5841                          skb_network_offset(skb);
5842no_csum:
5843        vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
5844        vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5845
5846        igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
5847}
5848
5849#define IGB_SET_FLAG(_input, _flag, _result) \
5850        ((_flag <= _result) ? \
5851         ((u32)(_input & _flag) * (_result / _flag)) : \
5852         ((u32)(_input & _flag) / (_flag / _result)))
5853
5854static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
5855{
5856        /* set type for advanced descriptor with frame checksum insertion */
5857        u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
5858                       E1000_ADVTXD_DCMD_DEXT |
5859                       E1000_ADVTXD_DCMD_IFCS;
5860
5861        /* set HW vlan bit if vlan is present */
5862        cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
5863                                 (E1000_ADVTXD_DCMD_VLE));
5864
5865        /* set segmentation bits for TSO */
5866        cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
5867                                 (E1000_ADVTXD_DCMD_TSE));
5868
5869        /* set timestamp bit if present */
5870        cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
5871                                 (E1000_ADVTXD_MAC_TSTAMP));
5872
5873        /* insert frame checksum */
5874        cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
5875
5876        return cmd_type;
5877}
5878
5879static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
5880                                 union e1000_adv_tx_desc *tx_desc,
5881                                 u32 tx_flags, unsigned int paylen)
5882{
5883        u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
5884
5885        /* 82575 requires a unique index per ring */
5886        if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5887                olinfo_status |= tx_ring->reg_idx << 4;
5888
5889        /* insert L4 checksum */
5890        olinfo_status |= IGB_SET_FLAG(tx_flags,
5891                                      IGB_TX_FLAGS_CSUM,
5892                                      (E1000_TXD_POPTS_TXSM << 8));
5893
5894        /* insert IPv4 checksum */
5895        olinfo_status |= IGB_SET_FLAG(tx_flags,
5896                                      IGB_TX_FLAGS_IPV4,
5897                                      (E1000_TXD_POPTS_IXSM << 8));
5898
5899        tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5900}
5901
5902static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5903{
5904        struct net_device *netdev = tx_ring->netdev;
5905
5906        netif_stop_subqueue(netdev, tx_ring->queue_index);
5907
5908        /* Herbert's original patch had:
5909         *  smp_mb__after_netif_stop_queue();
5910         * but since that doesn't exist yet, just open code it.
5911         */
5912        smp_mb();
5913
5914        /* We need to check again in a case another CPU has just
5915         * made room available.
5916         */
5917        if (igb_desc_unused(tx_ring) < size)
5918                return -EBUSY;
5919
5920        /* A reprieve! */
5921        netif_wake_subqueue(netdev, tx_ring->queue_index);
5922
5923        u64_stats_update_begin(&tx_ring->tx_syncp2);
5924        tx_ring->tx_stats.restart_queue2++;
5925        u64_stats_update_end(&tx_ring->tx_syncp2);
5926
5927        return 0;
5928}
5929
5930static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5931{
5932        if (igb_desc_unused(tx_ring) >= size)
5933                return 0;
5934        return __igb_maybe_stop_tx(tx_ring, size);
5935}
5936
5937static int igb_tx_map(struct igb_ring *tx_ring,
5938                      struct igb_tx_buffer *first,
5939                      const u8 hdr_len)
5940{
5941        struct sk_buff *skb = first->skb;
5942        struct igb_tx_buffer *tx_buffer;
5943        union e1000_adv_tx_desc *tx_desc;
5944        struct skb_frag_struct *frag;
5945        dma_addr_t dma;
5946        unsigned int data_len, size;
5947        u32 tx_flags = first->tx_flags;
5948        u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
5949        u16 i = tx_ring->next_to_use;
5950
5951        tx_desc = IGB_TX_DESC(tx_ring, i);
5952
5953        igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
5954
5955        size = skb_headlen(skb);
5956        data_len = skb->data_len;
5957
5958        dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5959
5960        tx_buffer = first;
5961
5962        for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
5963                if (dma_mapping_error(tx_ring->dev, dma))
5964                        goto dma_error;
5965
5966                /* record length, and DMA address */
5967                dma_unmap_len_set(tx_buffer, len, size);
5968                dma_unmap_addr_set(tx_buffer, dma, dma);
5969
5970                tx_desc->read.buffer_addr = cpu_to_le64(dma);
5971
5972                while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
5973                        tx_desc->read.cmd_type_len =
5974                                cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
5975
5976                        i++;
5977                        tx_desc++;
5978                        if (i == tx_ring->count) {
5979                                tx_desc = IGB_TX_DESC(tx_ring, 0);
5980                                i = 0;
5981                        }
5982                        tx_desc->read.olinfo_status = 0;
5983
5984                        dma += IGB_MAX_DATA_PER_TXD;
5985                        size -= IGB_MAX_DATA_PER_TXD;
5986
5987                        tx_desc->read.buffer_addr = cpu_to_le64(dma);
5988                }
5989
5990                if (likely(!data_len))
5991                        break;
5992
5993                tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
5994
5995                i++;
5996                tx_desc++;
5997                if (i == tx_ring->count) {
5998                        tx_desc = IGB_TX_DESC(tx_ring, 0);
5999                        i = 0;
6000                }
6001                tx_desc->read.olinfo_status = 0;
6002
6003                size = skb_frag_size(frag);
6004                data_len -= size;
6005
6006                dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6007                                       size, DMA_TO_DEVICE);
6008
6009                tx_buffer = &tx_ring->tx_buffer_info[i];
6010        }
6011
6012        /* write last descriptor with RS and EOP bits */
6013        cmd_type |= size | IGB_TXD_DCMD;
6014        tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6015
6016        netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6017
6018        /* set the timestamp */
6019        first->time_stamp = jiffies;
6020
6021        skb_tx_timestamp(skb);
6022
6023        /* Force memory writes to complete before letting h/w know there
6024         * are new descriptors to fetch.  (Only applicable for weak-ordered
6025         * memory model archs, such as IA-64).
6026         *
6027         * We also need this memory barrier to make certain all of the
6028         * status bits have been updated before next_to_watch is written.
6029         */
6030        dma_wmb();
6031
6032        /* set next_to_watch value indicating a packet is present */
6033        first->next_to_watch = tx_desc;
6034
6035        i++;
6036        if (i == tx_ring->count)
6037                i = 0;
6038
6039        tx_ring->next_to_use = i;
6040
6041        /* Make sure there is space in the ring for the next send. */
6042        igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6043
6044        if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
6045                writel(i, tx_ring->tail);
6046
6047                /* we need this if more than one processor can write to our tail
6048                 * at a time, it synchronizes IO on IA64/Altix systems
6049                 */
6050                mmiowb();
6051        }
6052        return 0;
6053
6054dma_error:
6055        dev_err(tx_ring->dev, "TX DMA map failed\n");
6056        tx_buffer = &tx_ring->tx_buffer_info[i];
6057
6058        /* clear dma mappings for failed tx_buffer_info map */
6059        while (tx_buffer != first) {
6060                if (dma_unmap_len(tx_buffer, len))
6061                        dma_unmap_page(tx_ring->dev,
6062                                       dma_unmap_addr(tx_buffer, dma),
6063                                       dma_unmap_len(tx_buffer, len),
6064                                       DMA_TO_DEVICE);
6065                dma_unmap_len_set(tx_buffer, len, 0);
6066
6067                if (i-- == 0)
6068                        i += tx_ring->count;
6069                tx_buffer = &tx_ring->tx_buffer_info[i];
6070        }
6071
6072        if (dma_unmap_len(tx_buffer, len))
6073                dma_unmap_single(tx_ring->dev,
6074                                 dma_unmap_addr(tx_buffer, dma),
6075                                 dma_unmap_len(tx_buffer, len),
6076                                 DMA_TO_DEVICE);
6077        dma_unmap_len_set(tx_buffer, len, 0);
6078
6079        dev_kfree_skb_any(tx_buffer->skb);
6080        tx_buffer->skb = NULL;
6081
6082        tx_ring->next_to_use = i;
6083
6084        return -1;
6085}
6086
6087netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6088                                struct igb_ring *tx_ring)
6089{
6090        struct igb_tx_buffer *first;
6091        int tso;
6092        u32 tx_flags = 0;
6093        unsigned short f;
6094        u16 count = TXD_USE_COUNT(skb_headlen(skb));
6095        __be16 protocol = vlan_get_protocol(skb);
6096        u8 hdr_len = 0;
6097
6098        /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6099         *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6100         *       + 2 desc gap to keep tail from touching head,
6101         *       + 1 desc for context descriptor,
6102         * otherwise try next time
6103         */
6104        for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6105                count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
6106
6107        if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6108                /* this is a hard error */
6109                return NETDEV_TX_BUSY;
6110        }
6111
6112        /* record the location of the first descriptor for this packet */
6113        first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6114        first->skb = skb;
6115        first->bytecount = skb->len;
6116        first->gso_segs = 1;
6117
6118        if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6119                struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6120
6121                if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6122                    !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6123                                           &adapter->state)) {
6124                        skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6125                        tx_flags |= IGB_TX_FLAGS_TSTAMP;
6126
6127                        adapter->ptp_tx_skb = skb_get(skb);
6128                        adapter->ptp_tx_start = jiffies;
6129                        if (adapter->hw.mac.type == e1000_82576)
6130                                schedule_work(&adapter->ptp_tx_work);
6131                } else {
6132                        adapter->tx_hwtstamp_skipped++;
6133                }
6134        }
6135
6136        if (skb_vlan_tag_present(skb)) {
6137                tx_flags |= IGB_TX_FLAGS_VLAN;
6138                tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6139        }
6140
6141        /* record initial flags and protocol */
6142        first->tx_flags = tx_flags;
6143        first->protocol = protocol;
6144
6145        tso = igb_tso(tx_ring, first, &hdr_len);
6146        if (tso < 0)
6147                goto out_drop;
6148        else if (!tso)
6149                igb_tx_csum(tx_ring, first);
6150
6151        if (igb_tx_map(tx_ring, first, hdr_len))
6152                goto cleanup_tx_tstamp;
6153
6154        return NETDEV_TX_OK;
6155
6156out_drop:
6157        dev_kfree_skb_any(first->skb);
6158        first->skb = NULL;
6159cleanup_tx_tstamp:
6160        if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6161                struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6162
6163                dev_kfree_skb_any(adapter->ptp_tx_skb);
6164                adapter->ptp_tx_skb = NULL;
6165                if (adapter->hw.mac.type == e1000_82576)
6166                        cancel_work_sync(&adapter->ptp_tx_work);
6167                clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6168        }
6169
6170        return NETDEV_TX_OK;
6171}
6172
6173static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6174                                                    struct sk_buff *skb)
6175{
6176        unsigned int r_idx = skb->queue_mapping;
6177
6178        if (r_idx >= adapter->num_tx_queues)
6179                r_idx = r_idx % adapter->num_tx_queues;
6180
6181        return adapter->tx_ring[r_idx];
6182}
6183
6184static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6185                                  struct net_device *netdev)
6186{
6187        struct igb_adapter *adapter = netdev_priv(netdev);
6188
6189        /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6190         * in order to meet this minimum size requirement.
6191         */
6192        if (skb_put_padto(skb, 17))
6193                return NETDEV_TX_OK;
6194
6195        return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6196}
6197
6198/**
6199 *  igb_tx_timeout - Respond to a Tx Hang
6200 *  @netdev: network interface device structure
6201 **/
6202static void igb_tx_timeout(struct net_device *netdev)
6203{
6204        struct igb_adapter *adapter = netdev_priv(netdev);
6205        struct e1000_hw *hw = &adapter->hw;
6206
6207        /* Do the reset outside of interrupt context */
6208        adapter->tx_timeout_count++;
6209
6210        if (hw->mac.type >= e1000_82580)
6211                hw->dev_spec._82575.global_device_reset = true;
6212
6213        schedule_work(&adapter->reset_task);
6214        wr32(E1000_EICS,
6215             (adapter->eims_enable_mask & ~adapter->eims_other));
6216}
6217
6218static void igb_reset_task(struct work_struct *work)
6219{
6220        struct igb_adapter *adapter;
6221        adapter = container_of(work, struct igb_adapter, reset_task);
6222
6223        igb_dump(adapter);
6224        netdev_err(adapter->netdev, "Reset adapter\n");
6225        igb_reinit_locked(adapter);
6226}
6227
6228/**
6229 *  igb_get_stats64 - Get System Network Statistics
6230 *  @netdev: network interface device structure
6231 *  @stats: rtnl_link_stats64 pointer
6232 **/
6233static void igb_get_stats64(struct net_device *netdev,
6234                            struct rtnl_link_stats64 *stats)
6235{
6236        struct igb_adapter *adapter = netdev_priv(netdev);
6237
6238        spin_lock(&adapter->stats64_lock);
6239        igb_update_stats(adapter);
6240        memcpy(stats, &adapter->stats64, sizeof(*stats));
6241        spin_unlock(&adapter->stats64_lock);
6242}
6243
6244/**
6245 *  igb_change_mtu - Change the Maximum Transfer Unit
6246 *  @netdev: network interface device structure
6247 *  @new_mtu: new value for maximum frame size
6248 *
6249 *  Returns 0 on success, negative on failure
6250 **/
6251static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6252{
6253        struct igb_adapter *adapter = netdev_priv(netdev);
6254        struct pci_dev *pdev = adapter->pdev;
6255        int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
6256
6257        /* adjust max frame to be at least the size of a standard frame */
6258        if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6259                max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6260
6261        while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6262                usleep_range(1000, 2000);
6263
6264        /* igb_down has a dependency on max_frame_size */
6265        adapter->max_frame_size = max_frame;
6266
6267        if (netif_running(netdev))
6268                igb_down(adapter);
6269
6270        dev_info(&pdev->dev, "changing MTU from %d to %d\n",
6271                 netdev->mtu, new_mtu);
6272        netdev->mtu = new_mtu;
6273
6274        if (netif_running(netdev))
6275                igb_up(adapter);
6276        else
6277                igb_reset(adapter);
6278
6279        clear_bit(__IGB_RESETTING, &adapter->state);
6280
6281        return 0;
6282}
6283
6284/**
6285 *  igb_update_stats - Update the board statistics counters
6286 *  @adapter: board private structure
6287 **/
6288void igb_update_stats(struct igb_adapter *adapter)
6289{
6290        struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6291        struct e1000_hw *hw = &adapter->hw;
6292        struct pci_dev *pdev = adapter->pdev;
6293        u32 reg, mpc;
6294        int i;
6295        u64 bytes, packets;
6296        unsigned int start;
6297        u64 _bytes, _packets;
6298
6299        /* Prevent stats update while adapter is being reset, or if the pci
6300         * connection is down.
6301         */
6302        if (adapter->link_speed == 0)
6303                return;
6304        if (pci_channel_offline(pdev))
6305                return;
6306
6307        bytes = 0;
6308        packets = 0;
6309
6310        rcu_read_lock();
6311        for (i = 0; i < adapter->num_rx_queues; i++) {
6312                struct igb_ring *ring = adapter->rx_ring[i];
6313                u32 rqdpc = rd32(E1000_RQDPC(i));
6314                if (hw->mac.type >= e1000_i210)
6315                        wr32(E1000_RQDPC(i), 0);
6316
6317                if (rqdpc) {
6318                        ring->rx_stats.drops += rqdpc;
6319                        net_stats->rx_fifo_errors += rqdpc;
6320                }
6321
6322                do {
6323                        start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
6324                        _bytes = ring->rx_stats.bytes;
6325                        _packets = ring->rx_stats.packets;
6326                } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
6327                bytes += _bytes;
6328                packets += _packets;
6329        }
6330
6331        net_stats->rx_bytes = bytes;
6332        net_stats->rx_packets = packets;
6333
6334        bytes = 0;
6335        packets = 0;
6336        for (i = 0; i < adapter->num_tx_queues; i++) {
6337                struct igb_ring *ring = adapter->tx_ring[i];
6338                do {
6339                        start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
6340                        _bytes = ring->tx_stats.bytes;
6341                        _packets = ring->tx_stats.packets;
6342                } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
6343                bytes += _bytes;
6344                packets += _packets;
6345        }
6346        net_stats->tx_bytes = bytes;
6347        net_stats->tx_packets = packets;
6348        rcu_read_unlock();
6349
6350        /* read stats registers */
6351        adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6352        adapter->stats.gprc += rd32(E1000_GPRC);
6353        adapter->stats.gorc += rd32(E1000_GORCL);
6354        rd32(E1000_GORCH); /* clear GORCL */
6355        adapter->stats.bprc += rd32(E1000_BPRC);
6356        adapter->stats.mprc += rd32(E1000_MPRC);
6357        adapter->stats.roc += rd32(E1000_ROC);
6358
6359        adapter->stats.prc64 += rd32(E1000_PRC64);
6360        adapter->stats.prc127 += rd32(E1000_PRC127);
6361        adapter->stats.prc255 += rd32(E1000_PRC255);
6362        adapter->stats.prc511 += rd32(E1000_PRC511);
6363        adapter->stats.prc1023 += rd32(E1000_PRC1023);
6364        adapter->stats.prc1522 += rd32(E1000_PRC1522);
6365        adapter->stats.symerrs += rd32(E1000_SYMERRS);
6366        adapter->stats.sec += rd32(E1000_SEC);
6367
6368        mpc = rd32(E1000_MPC);
6369        adapter->stats.mpc += mpc;
6370        net_stats->rx_fifo_errors += mpc;
6371        adapter->stats.scc += rd32(E1000_SCC);
6372        adapter->stats.ecol += rd32(E1000_ECOL);
6373        adapter->stats.mcc += rd32(E1000_MCC);
6374        adapter->stats.latecol += rd32(E1000_LATECOL);
6375        adapter->stats.dc += rd32(E1000_DC);
6376        adapter->stats.rlec += rd32(E1000_RLEC);
6377        adapter->stats.xonrxc += rd32(E1000_XONRXC);
6378        adapter->stats.xontxc += rd32(E1000_XONTXC);
6379        adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6380        adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6381        adapter->stats.fcruc += rd32(E1000_FCRUC);
6382        adapter->stats.gptc += rd32(E1000_GPTC);
6383        adapter->stats.gotc += rd32(E1000_GOTCL);
6384        rd32(E1000_GOTCH); /* clear GOTCL */
6385        adapter->stats.rnbc += rd32(E1000_RNBC);
6386        adapter->stats.ruc += rd32(E1000_RUC);
6387        adapter->stats.rfc += rd32(E1000_RFC);
6388        adapter->stats.rjc += rd32(E1000_RJC);
6389        adapter->stats.tor += rd32(E1000_TORH);
6390        adapter->stats.tot += rd32(E1000_TOTH);
6391        adapter->stats.tpr += rd32(E1000_TPR);
6392
6393        adapter->stats.ptc64 += rd32(E1000_PTC64);
6394        adapter->stats.ptc127 += rd32(E1000_PTC127);
6395        adapter->stats.ptc255 += rd32(E1000_PTC255);
6396        adapter->stats.ptc511 += rd32(E1000_PTC511);
6397        adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6398        adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6399
6400        adapter->stats.mptc += rd32(E1000_MPTC);
6401        adapter->stats.bptc += rd32(E1000_BPTC);
6402
6403        adapter->stats.tpt += rd32(E1000_TPT);
6404        adapter->stats.colc += rd32(E1000_COLC);
6405
6406        adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6407        /* read internal phy specific stats */
6408        reg = rd32(E1000_CTRL_EXT);
6409        if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6410                adapter->stats.rxerrc += rd32(E1000_RXERRC);
6411
6412                /* this stat has invalid values on i210/i211 */
6413                if ((hw->mac.type != e1000_i210) &&
6414                    (hw->mac.type != e1000_i211))
6415                        adapter->stats.tncrs += rd32(E1000_TNCRS);
6416        }
6417
6418        adapter->stats.tsctc += rd32(E1000_TSCTC);
6419        adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6420
6421        adapter->stats.iac += rd32(E1000_IAC);
6422        adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6423        adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6424        adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6425        adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6426        adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6427        adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6428        adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6429        adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6430
6431        /* Fill out the OS statistics structure */
6432        net_stats->multicast = adapter->stats.mprc;
6433        net_stats->collisions = adapter->stats.colc;
6434
6435        /* Rx Errors */
6436
6437        /* RLEC on some newer hardware can be incorrect so build
6438         * our own version based on RUC and ROC
6439         */
6440        net_stats->rx_errors = adapter->stats.rxerrc +
6441                adapter->stats.crcerrs + adapter->stats.algnerrc +
6442                adapter->stats.ruc + adapter->stats.roc +
6443                adapter->stats.cexterr;
6444        net_stats->rx_length_errors = adapter->stats.ruc +
6445                                      adapter->stats.roc;
6446        net_stats->rx_crc_errors = adapter->stats.crcerrs;
6447        net_stats->rx_frame_errors = adapter->stats.algnerrc;
6448        net_stats->rx_missed_errors = adapter->stats.mpc;
6449
6450        /* Tx Errors */
6451        net_stats->tx_errors = adapter->stats.ecol +
6452                               adapter->stats.latecol;
6453        net_stats->tx_aborted_errors = adapter->stats.ecol;
6454        net_stats->tx_window_errors = adapter->stats.latecol;
6455        net_stats->tx_carrier_errors = adapter->stats.tncrs;
6456
6457        /* Tx Dropped needs to be maintained elsewhere */
6458
6459        /* Management Stats */
6460        adapter->stats.mgptc += rd32(E1000_MGTPTC);
6461        adapter->stats.mgprc += rd32(E1000_MGTPRC);
6462        adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6463
6464        /* OS2BMC Stats */
6465        reg = rd32(E1000_MANC);
6466        if (reg & E1000_MANC_EN_BMC2OS) {
6467                adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6468                adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6469                adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6470                adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6471        }
6472}
6473
6474static void igb_tsync_interrupt(struct igb_adapter *adapter)
6475{
6476        struct e1000_hw *hw = &adapter->hw;
6477        struct ptp_clock_event event;
6478        struct timespec64 ts;
6479        u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
6480
6481        if (tsicr & TSINTR_SYS_WRAP) {
6482                event.type = PTP_CLOCK_PPS;
6483                if (adapter->ptp_caps.pps)
6484                        ptp_clock_event(adapter->ptp_clock, &event);
6485                ack |= TSINTR_SYS_WRAP;
6486        }
6487
6488        if (tsicr & E1000_TSICR_TXTS) {
6489                /* retrieve hardware timestamp */
6490                schedule_work(&adapter->ptp_tx_work);
6491                ack |= E1000_TSICR_TXTS;
6492        }
6493
6494        if (tsicr & TSINTR_TT0) {
6495                spin_lock(&adapter->tmreg_lock);
6496                ts = timespec64_add(adapter->perout[0].start,
6497                                    adapter->perout[0].period);
6498                /* u32 conversion of tv_sec is safe until y2106 */
6499                wr32(E1000_TRGTTIML0, ts.tv_nsec);
6500                wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
6501                tsauxc = rd32(E1000_TSAUXC);
6502                tsauxc |= TSAUXC_EN_TT0;
6503                wr32(E1000_TSAUXC, tsauxc);
6504                adapter->perout[0].start = ts;
6505                spin_unlock(&adapter->tmreg_lock);
6506                ack |= TSINTR_TT0;
6507        }
6508
6509        if (tsicr & TSINTR_TT1) {
6510                spin_lock(&adapter->tmreg_lock);
6511                ts = timespec64_add(adapter->perout[1].start,
6512                                    adapter->perout[1].period);
6513                wr32(E1000_TRGTTIML1, ts.tv_nsec);
6514                wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
6515                tsauxc = rd32(E1000_TSAUXC);
6516                tsauxc |= TSAUXC_EN_TT1;
6517                wr32(E1000_TSAUXC, tsauxc);
6518                adapter->perout[1].start = ts;
6519                spin_unlock(&adapter->tmreg_lock);
6520                ack |= TSINTR_TT1;
6521        }
6522
6523        if (tsicr & TSINTR_AUTT0) {
6524                nsec = rd32(E1000_AUXSTMPL0);
6525                sec  = rd32(E1000_AUXSTMPH0);
6526                event.type = PTP_CLOCK_EXTTS;
6527                event.index = 0;
6528                event.timestamp = sec * 1000000000ULL + nsec;
6529                ptp_clock_event(adapter->ptp_clock, &event);
6530                ack |= TSINTR_AUTT0;
6531        }
6532
6533        if (tsicr & TSINTR_AUTT1) {
6534                nsec = rd32(E1000_AUXSTMPL1);
6535                sec  = rd32(E1000_AUXSTMPH1);
6536                event.type = PTP_CLOCK_EXTTS;
6537                event.index = 1;
6538                event.timestamp = sec * 1000000000ULL + nsec;
6539                ptp_clock_event(adapter->ptp_clock, &event);
6540                ack |= TSINTR_AUTT1;
6541        }
6542
6543        /* acknowledge the interrupts */
6544        wr32(E1000_TSICR, ack);
6545}
6546
6547static irqreturn_t igb_msix_other(int irq, void *data)
6548{
6549        struct igb_adapter *adapter = data;
6550        struct e1000_hw *hw = &adapter->hw;
6551        u32 icr = rd32(E1000_ICR);
6552        /* reading ICR causes bit 31 of EICR to be cleared */
6553
6554        if (icr & E1000_ICR_DRSTA)
6555                schedule_work(&adapter->reset_task);
6556
6557        if (icr & E1000_ICR_DOUTSYNC) {
6558                /* HW is reporting DMA is out of sync */
6559                adapter->stats.doosync++;
6560                /* The DMA Out of Sync is also indication of a spoof event
6561                 * in IOV mode. Check the Wrong VM Behavior register to
6562                 * see if it is really a spoof event.
6563                 */
6564                igb_check_wvbr(adapter);
6565        }
6566
6567        /* Check for a mailbox event */
6568        if (icr & E1000_ICR_VMMB)
6569                igb_msg_task(adapter);
6570
6571        if (icr & E1000_ICR_LSC) {
6572                hw->mac.get_link_status = 1;
6573                /* guard against interrupt when we're going down */
6574                if (!test_bit(__IGB_DOWN, &adapter->state))
6575                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
6576        }
6577
6578        if (icr & E1000_ICR_TS)
6579                igb_tsync_interrupt(adapter);
6580
6581        wr32(E1000_EIMS, adapter->eims_other);
6582
6583        return IRQ_HANDLED;
6584}
6585
6586static void igb_write_itr(struct igb_q_vector *q_vector)
6587{
6588        struct igb_adapter *adapter = q_vector->adapter;
6589        u32 itr_val = q_vector->itr_val & 0x7FFC;
6590
6591        if (!q_vector->set_itr)
6592                return;
6593
6594        if (!itr_val)
6595                itr_val = 0x4;
6596
6597        if (adapter->hw.mac.type == e1000_82575)
6598                itr_val |= itr_val << 16;
6599        else
6600                itr_val |= E1000_EITR_CNT_IGNR;
6601
6602        writel(itr_val, q_vector->itr_register);
6603        q_vector->set_itr = 0;
6604}
6605
6606static irqreturn_t igb_msix_ring(int irq, void *data)
6607{
6608        struct igb_q_vector *q_vector = data;
6609
6610        /* Write the ITR value calculated from the previous interrupt. */
6611        igb_write_itr(q_vector);
6612
6613        napi_schedule(&q_vector->napi);
6614
6615        return IRQ_HANDLED;
6616}
6617
6618#ifdef CONFIG_IGB_DCA
6619static void igb_update_tx_dca(struct igb_adapter *adapter,
6620                              struct igb_ring *tx_ring,
6621                              int cpu)
6622{
6623        struct e1000_hw *hw = &adapter->hw;
6624        u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
6625
6626        if (hw->mac.type != e1000_82575)
6627                txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
6628
6629        /* We can enable relaxed ordering for reads, but not writes when
6630         * DCA is enabled.  This is due to a known issue in some chipsets
6631         * which will cause the DCA tag to be cleared.
6632         */
6633        txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
6634                  E1000_DCA_TXCTRL_DATA_RRO_EN |
6635                  E1000_DCA_TXCTRL_DESC_DCA_EN;
6636
6637        wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
6638}
6639
6640static void igb_update_rx_dca(struct igb_adapter *adapter,
6641                              struct igb_ring *rx_ring,
6642                              int cpu)
6643{
6644        struct e1000_hw *hw = &adapter->hw;
6645        u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
6646
6647        if (hw->mac.type != e1000_82575)
6648                rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
6649
6650        /* We can enable relaxed ordering for reads, but not writes when
6651         * DCA is enabled.  This is due to a known issue in some chipsets
6652         * which will cause the DCA tag to be cleared.
6653         */
6654        rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
6655                  E1000_DCA_RXCTRL_DESC_DCA_EN;
6656
6657        wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
6658}
6659
6660static void igb_update_dca(struct igb_q_vector *q_vector)
6661{
6662        struct igb_adapter *adapter = q_vector->adapter;
6663        int cpu = get_cpu();
6664
6665        if (q_vector->cpu == cpu)
6666                goto out_no_update;
6667
6668        if (q_vector->tx.ring)
6669                igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
6670
6671        if (q_vector->rx.ring)
6672                igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
6673
6674        q_vector->cpu = cpu;
6675out_no_update:
6676        put_cpu();
6677}
6678
6679static void igb_setup_dca(struct igb_adapter *adapter)
6680{
6681        struct e1000_hw *hw = &adapter->hw;
6682        int i;
6683
6684        if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
6685                return;
6686
6687        /* Always use CB2 mode, difference is masked in the CB driver. */
6688        wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
6689
6690        for (i = 0; i < adapter->num_q_vectors; i++) {
6691                adapter->q_vector[i]->cpu = -1;
6692                igb_update_dca(adapter->q_vector[i]);
6693        }
6694}
6695
6696static int __igb_notify_dca(struct device *dev, void *data)
6697{
6698        struct net_device *netdev = dev_get_drvdata(dev);
6699        struct igb_adapter *adapter = netdev_priv(netdev);
6700        struct pci_dev *pdev = adapter->pdev;
6701        struct e1000_hw *hw = &adapter->hw;
6702        unsigned long event = *(unsigned long *)data;
6703
6704        switch (event) {
6705        case DCA_PROVIDER_ADD:
6706                /* if already enabled, don't do it again */
6707                if (adapter->flags & IGB_FLAG_DCA_ENABLED)
6708                        break;
6709                if (dca_add_requester(dev) == 0) {
6710                        adapter->flags |= IGB_FLAG_DCA_ENABLED;
6711                        dev_info(&pdev->dev, "DCA enabled\n");
6712                        igb_setup_dca(adapter);
6713                        break;
6714                }
6715                /* Fall Through since DCA is disabled. */
6716        case DCA_PROVIDER_REMOVE:
6717                if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
6718                        /* without this a class_device is left
6719                         * hanging around in the sysfs model
6720                         */
6721                        dca_remove_requester(dev);
6722                        dev_info(&pdev->dev, "DCA disabled\n");
6723                        adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
6724                        wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
6725                }
6726                break;
6727        }
6728
6729        return 0;
6730}
6731
6732static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
6733                          void *p)
6734{
6735        int ret_val;
6736
6737        ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
6738                                         __igb_notify_dca);
6739
6740        return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6741}
6742#endif /* CONFIG_IGB_DCA */
6743
6744#ifdef CONFIG_PCI_IOV
6745static int igb_vf_configure(struct igb_adapter *adapter, int vf)
6746{
6747        unsigned char mac_addr[ETH_ALEN];
6748
6749        eth_zero_addr(mac_addr);
6750        igb_set_vf_mac(adapter, vf, mac_addr);
6751
6752        /* By default spoof check is enabled for all VFs */
6753        adapter->vf_data[vf].spoofchk_enabled = true;
6754
6755        /* By default VFs are not trusted */
6756        adapter->vf_data[vf].trusted = false;
6757
6758        return 0;
6759}
6760
6761#endif
6762static void igb_ping_all_vfs(struct igb_adapter *adapter)
6763{
6764        struct e1000_hw *hw = &adapter->hw;
6765        u32 ping;
6766        int i;
6767
6768        for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
6769                ping = E1000_PF_CONTROL_MSG;
6770                if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
6771                        ping |= E1000_VT_MSGTYPE_CTS;
6772                igb_write_mbx(hw, &ping, 1, i);
6773        }
6774}
6775
6776static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6777{
6778        struct e1000_hw *hw = &adapter->hw;
6779        u32 vmolr = rd32(E1000_VMOLR(vf));
6780        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6781
6782        vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
6783                            IGB_VF_FLAG_MULTI_PROMISC);
6784        vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6785
6786        if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
6787                vmolr |= E1000_VMOLR_MPME;
6788                vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
6789                *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
6790        } else {
6791                /* if we have hashes and we are clearing a multicast promisc
6792                 * flag we need to write the hashes to the MTA as this step
6793                 * was previously skipped
6794                 */
6795                if (vf_data->num_vf_mc_hashes > 30) {
6796                        vmolr |= E1000_VMOLR_MPME;
6797                } else if (vf_data->num_vf_mc_hashes) {
6798                        int j;
6799
6800                        vmolr |= E1000_VMOLR_ROMPE;
6801                        for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6802                                igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6803                }
6804        }
6805
6806        wr32(E1000_VMOLR(vf), vmolr);
6807
6808        /* there are flags left unprocessed, likely not supported */
6809        if (*msgbuf & E1000_VT_MSGINFO_MASK)
6810                return -EINVAL;
6811
6812        return 0;
6813}
6814
6815static int igb_set_vf_multicasts(struct igb_adapter *adapter,
6816                                  u32 *msgbuf, u32 vf)
6817{
6818        int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6819        u16 *hash_list = (u16 *)&msgbuf[1];
6820        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6821        int i;
6822
6823        /* salt away the number of multicast addresses assigned
6824         * to this VF for later use to restore when the PF multi cast
6825         * list changes
6826         */
6827        vf_data->num_vf_mc_hashes = n;
6828
6829        /* only up to 30 hash values supported */
6830        if (n > 30)
6831                n = 30;
6832
6833        /* store the hashes for later use */
6834        for (i = 0; i < n; i++)
6835                vf_data->vf_mc_hashes[i] = hash_list[i];
6836
6837        /* Flush and reset the mta with the new values */
6838        igb_set_rx_mode(adapter->netdev);
6839
6840        return 0;
6841}
6842
6843static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
6844{
6845        struct e1000_hw *hw = &adapter->hw;
6846        struct vf_data_storage *vf_data;
6847        int i, j;
6848
6849        for (i = 0; i < adapter->vfs_allocated_count; i++) {
6850                u32 vmolr = rd32(E1000_VMOLR(i));
6851
6852                vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6853
6854                vf_data = &adapter->vf_data[i];
6855
6856                if ((vf_data->num_vf_mc_hashes > 30) ||
6857                    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
6858                        vmolr |= E1000_VMOLR_MPME;
6859                } else if (vf_data->num_vf_mc_hashes) {
6860                        vmolr |= E1000_VMOLR_ROMPE;
6861                        for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6862                                igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6863                }
6864                wr32(E1000_VMOLR(i), vmolr);
6865        }
6866}
6867
6868static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
6869{
6870        struct e1000_hw *hw = &adapter->hw;
6871        u32 pool_mask, vlvf_mask, i;
6872
6873        /* create mask for VF and other pools */
6874        pool_mask = E1000_VLVF_POOLSEL_MASK;
6875        vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
6876
6877        /* drop PF from pool bits */
6878        pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
6879                             adapter->vfs_allocated_count);
6880
6881        /* Find the vlan filter for this id */
6882        for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
6883                u32 vlvf = rd32(E1000_VLVF(i));
6884                u32 vfta_mask, vid, vfta;
6885
6886                /* remove the vf from the pool */
6887                if (!(vlvf & vlvf_mask))
6888                        continue;
6889
6890                /* clear out bit from VLVF */
6891                vlvf ^= vlvf_mask;
6892
6893                /* if other pools are present, just remove ourselves */
6894                if (vlvf & pool_mask)
6895                        goto update_vlvfb;
6896
6897                /* if PF is present, leave VFTA */
6898                if (vlvf & E1000_VLVF_POOLSEL_MASK)
6899                        goto update_vlvf;
6900
6901                vid = vlvf & E1000_VLVF_VLANID_MASK;
6902                vfta_mask = BIT(vid % 32);
6903
6904                /* clear bit from VFTA */
6905                vfta = adapter->shadow_vfta[vid / 32];
6906                if (vfta & vfta_mask)
6907                        hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
6908update_vlvf:
6909                /* clear pool selection enable */
6910                if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6911                        vlvf &= E1000_VLVF_POOLSEL_MASK;
6912                else
6913                        vlvf = 0;
6914update_vlvfb:
6915                /* clear pool bits */
6916                wr32(E1000_VLVF(i), vlvf);
6917        }
6918}
6919
6920static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
6921{
6922        u32 vlvf;
6923        int idx;
6924
6925        /* short cut the special case */
6926        if (vlan == 0)
6927                return 0;
6928
6929        /* Search for the VLAN id in the VLVF entries */
6930        for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
6931                vlvf = rd32(E1000_VLVF(idx));
6932                if ((vlvf & VLAN_VID_MASK) == vlan)
6933                        break;
6934        }
6935
6936        return idx;
6937}
6938
6939static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
6940{
6941        struct e1000_hw *hw = &adapter->hw;
6942        u32 bits, pf_id;
6943        int idx;
6944
6945        idx = igb_find_vlvf_entry(hw, vid);
6946        if (!idx)
6947                return;
6948
6949        /* See if any other pools are set for this VLAN filter
6950         * entry other than the PF.
6951         */
6952        pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
6953        bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
6954        bits &= rd32(E1000_VLVF(idx));
6955
6956        /* Disable the filter so this falls into the default pool. */
6957        if (!bits) {
6958                if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6959                        wr32(E1000_VLVF(idx), BIT(pf_id));
6960                else
6961                        wr32(E1000_VLVF(idx), 0);
6962        }
6963}
6964
6965static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
6966                           bool add, u32 vf)
6967{
6968        int pf_id = adapter->vfs_allocated_count;
6969        struct e1000_hw *hw = &adapter->hw;
6970        int err;
6971
6972        /* If VLAN overlaps with one the PF is currently monitoring make
6973         * sure that we are able to allocate a VLVF entry.  This may be
6974         * redundant but it guarantees PF will maintain visibility to
6975         * the VLAN.
6976         */
6977        if (add && test_bit(vid, adapter->active_vlans)) {
6978                err = igb_vfta_set(hw, vid, pf_id, true, false);
6979                if (err)
6980                        return err;
6981        }
6982
6983        err = igb_vfta_set(hw, vid, vf, add, false);
6984
6985        if (add && !err)
6986                return err;
6987
6988        /* If we failed to add the VF VLAN or we are removing the VF VLAN
6989         * we may need to drop the PF pool bit in order to allow us to free
6990         * up the VLVF resources.
6991         */
6992        if (test_bit(vid, adapter->active_vlans) ||
6993            (adapter->flags & IGB_FLAG_VLAN_PROMISC))
6994                igb_update_pf_vlvf(adapter, vid);
6995
6996        return err;
6997}
6998
6999static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7000{
7001        struct e1000_hw *hw = &adapter->hw;
7002
7003        if (vid)
7004                wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7005        else
7006                wr32(E1000_VMVIR(vf), 0);
7007}
7008
7009static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7010                                u16 vlan, u8 qos)
7011{
7012        int err;
7013
7014        err = igb_set_vf_vlan(adapter, vlan, true, vf);
7015        if (err)
7016                return err;
7017
7018        igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7019        igb_set_vmolr(adapter, vf, !vlan);
7020
7021        /* revoke access to previous VLAN */
7022        if (vlan != adapter->vf_data[vf].pf_vlan)
7023                igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7024                                false, vf);
7025
7026        adapter->vf_data[vf].pf_vlan = vlan;
7027        adapter->vf_data[vf].pf_qos = qos;
7028        igb_set_vf_vlan_strip(adapter, vf, true);
7029        dev_info(&adapter->pdev->dev,
7030                 "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7031        if (test_bit(__IGB_DOWN, &adapter->state)) {
7032                dev_warn(&adapter->pdev->dev,
7033                         "The VF VLAN has been set, but the PF device is not up.\n");
7034                dev_warn(&adapter->pdev->dev,
7035                         "Bring the PF device up before attempting to use the VF device.\n");
7036        }
7037
7038        return err;
7039}
7040
7041static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7042{
7043        /* Restore tagless access via VLAN 0 */
7044        igb_set_vf_vlan(adapter, 0, true, vf);
7045
7046        igb_set_vmvir(adapter, 0, vf);
7047        igb_set_vmolr(adapter, vf, true);
7048
7049        /* Remove any PF assigned VLAN */
7050        if (adapter->vf_data[vf].pf_vlan)
7051                igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7052                                false, vf);
7053
7054        adapter->vf_data[vf].pf_vlan = 0;
7055        adapter->vf_data[vf].pf_qos = 0;
7056        igb_set_vf_vlan_strip(adapter, vf, false);
7057
7058        return 0;
7059}
7060
7061static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7062                               u16 vlan, u8 qos, __be16 vlan_proto)
7063{
7064        struct igb_adapter *adapter = netdev_priv(netdev);
7065
7066        if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7067                return -EINVAL;
7068
7069        if (vlan_proto != htons(ETH_P_8021Q))
7070                return -EPROTONOSUPPORT;
7071
7072        return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7073                               igb_disable_port_vlan(adapter, vf);
7074}
7075
7076static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7077{
7078        int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7079        int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7080        int ret;
7081
7082        if (adapter->vf_data[vf].pf_vlan)
7083                return -1;
7084
7085        /* VLAN 0 is a special case, don't allow it to be removed */
7086        if (!vid && !add)
7087                return 0;
7088
7089        ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7090        if (!ret)
7091                igb_set_vf_vlan_strip(adapter, vf, !!vid);
7092        return ret;
7093}
7094
7095static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7096{
7097        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7098
7099        /* clear flags - except flag that indicates PF has set the MAC */
7100        vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7101        vf_data->last_nack = jiffies;
7102
7103        /* reset vlans for device */
7104        igb_clear_vf_vfta(adapter, vf);
7105        igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7106        igb_set_vmvir(adapter, vf_data->pf_vlan |
7107                               (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7108        igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7109        igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7110
7111        /* reset multicast table array for vf */
7112        adapter->vf_data[vf].num_vf_mc_hashes = 0;
7113
7114        /* Flush and reset the mta with the new values */
7115        igb_set_rx_mode(adapter->netdev);
7116}
7117
7118static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7119{
7120        unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7121
7122        /* clear mac address as we were hotplug removed/added */
7123        if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7124                eth_zero_addr(vf_mac);
7125
7126        /* process remaining reset events */
7127        igb_vf_reset(adapter, vf);
7128}
7129
7130static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7131{
7132        struct e1000_hw *hw = &adapter->hw;
7133        unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7134        u32 reg, msgbuf[3];
7135        u8 *addr = (u8 *)(&msgbuf[1]);
7136
7137        /* process all the same items cleared in a function level reset */
7138        igb_vf_reset(adapter, vf);
7139
7140        /* set vf mac address */
7141        igb_set_vf_mac(adapter, vf, vf_mac);
7142
7143        /* enable transmit and receive for vf */
7144        reg = rd32(E1000_VFTE);
7145        wr32(E1000_VFTE, reg | BIT(vf));
7146        reg = rd32(E1000_VFRE);
7147        wr32(E1000_VFRE, reg | BIT(vf));
7148
7149        adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7150
7151        /* reply to reset with ack and vf mac address */
7152        if (!is_zero_ether_addr(vf_mac)) {
7153                msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7154                memcpy(addr, vf_mac, ETH_ALEN);
7155        } else {
7156                msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7157        }
7158        igb_write_mbx(hw, msgbuf, 3, vf);
7159}
7160
7161static void igb_flush_mac_table(struct igb_adapter *adapter)
7162{
7163        struct e1000_hw *hw = &adapter->hw;
7164        int i;
7165
7166        for (i = 0; i < hw->mac.rar_entry_count; i++) {
7167                adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7168                memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
7169                adapter->mac_table[i].queue = 0;
7170                igb_rar_set_index(adapter, i);
7171        }
7172}
7173
7174static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7175{
7176        struct e1000_hw *hw = &adapter->hw;
7177        /* do not count rar entries reserved for VFs MAC addresses */
7178        int rar_entries = hw->mac.rar_entry_count -
7179                          adapter->vfs_allocated_count;
7180        int i, count = 0;
7181
7182        for (i = 0; i < rar_entries; i++) {
7183                /* do not count default entries */
7184                if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7185                        continue;
7186
7187                /* do not count "in use" entries for different queues */
7188                if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7189                    (adapter->mac_table[i].queue != queue))
7190                        continue;
7191
7192                count++;
7193        }
7194
7195        return count;
7196}
7197
7198/* Set default MAC address for the PF in the first RAR entry */
7199static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7200{
7201        struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7202
7203        ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7204        mac_table->queue = adapter->vfs_allocated_count;
7205        mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7206
7207        igb_rar_set_index(adapter, 0);
7208}
7209
7210/* If the filter to be added and an already existing filter express
7211 * the same address and address type, it should be possible to only
7212 * override the other configurations, for example the queue to steer
7213 * traffic.
7214 */
7215static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7216                                      const u8 *addr, const u8 flags)
7217{
7218        if (!(entry->state & IGB_MAC_STATE_IN_USE))
7219                return true;
7220
7221        if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7222            (flags & IGB_MAC_STATE_SRC_ADDR))
7223                return false;
7224
7225        if (!ether_addr_equal(addr, entry->addr))
7226                return false;
7227
7228        return true;
7229}
7230
7231/* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7232 * 'flags' is used to indicate what kind of match is made, match is by
7233 * default for the destination address, if matching by source address
7234 * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7235 */
7236static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7237                                    const u8 *addr, const u8 queue,
7238                                    const u8 flags)
7239{
7240        struct e1000_hw *hw = &adapter->hw;
7241        int rar_entries = hw->mac.rar_entry_count -
7242                          adapter->vfs_allocated_count;
7243        int i;
7244
7245        if (is_zero_ether_addr(addr))
7246                return -EINVAL;
7247
7248        /* Search for the first empty entry in the MAC table.
7249         * Do not touch entries at the end of the table reserved for the VF MAC
7250         * addresses.
7251         */
7252        for (i = 0; i < rar_entries; i++) {
7253                if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7254                                               addr, flags))
7255                        continue;
7256
7257                ether_addr_copy(adapter->mac_table[i].addr, addr);
7258                adapter->mac_table[i].queue = queue;
7259                adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7260
7261                igb_rar_set_index(adapter, i);
7262                return i;
7263        }
7264
7265        return -ENOSPC;
7266}
7267
7268static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7269                              const u8 queue)
7270{
7271        return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7272}
7273
7274/* Remove a MAC filter for 'addr' directing matching traffic to
7275 * 'queue', 'flags' is used to indicate what kind of match need to be
7276 * removed, match is by default for the destination address, if
7277 * matching by source address is to be removed the flag
7278 * IGB_MAC_STATE_SRC_ADDR can be used.
7279 */
7280static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7281                                    const u8 *addr, const u8 queue,
7282                                    const u8 flags)
7283{
7284        struct e1000_hw *hw = &adapter->hw;
7285        int rar_entries = hw->mac.rar_entry_count -
7286                          adapter->vfs_allocated_count;
7287        int i;
7288
7289        if (is_zero_ether_addr(addr))
7290                return -EINVAL;
7291
7292        /* Search for matching entry in the MAC table based on given address
7293         * and queue. Do not touch entries at the end of the table reserved
7294         * for the VF MAC addresses.
7295         */
7296        for (i = 0; i < rar_entries; i++) {
7297                if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7298                        continue;
7299                if ((adapter->mac_table[i].state & flags) != flags)
7300                        continue;
7301                if (adapter->mac_table[i].queue != queue)
7302                        continue;
7303                if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7304                        continue;
7305
7306                /* When a filter for the default address is "deleted",
7307                 * we return it to its initial configuration
7308                 */
7309                if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7310                        adapter->mac_table[i].state =
7311                                IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7312                        adapter->mac_table[i].queue =
7313                                adapter->vfs_allocated_count;
7314                } else {
7315                        adapter->mac_table[i].state = 0;
7316                        adapter->mac_table[i].queue = 0;
7317                        memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
7318                }
7319
7320                igb_rar_set_index(adapter, i);
7321                return 0;
7322        }
7323
7324        return -ENOENT;
7325}
7326
7327static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7328                              const u8 queue)
7329{
7330        return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7331}
7332
7333int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7334                                const u8 *addr, u8 queue, u8 flags)
7335{
7336        struct e1000_hw *hw = &adapter->hw;
7337
7338        /* In theory, this should be supported on 82575 as well, but
7339         * that part wasn't easily accessible during development.
7340         */
7341        if (hw->mac.type != e1000_i210)
7342                return -EOPNOTSUPP;
7343
7344        return igb_add_mac_filter_flags(adapter, addr, queue,
7345                                        IGB_MAC_STATE_QUEUE_STEERING | flags);
7346}
7347
7348int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7349                                const u8 *addr, u8 queue, u8 flags)
7350{
7351        return igb_del_mac_filter_flags(adapter, addr, queue,
7352                                        IGB_MAC_STATE_QUEUE_STEERING | flags);
7353}
7354
7355static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7356{
7357        struct igb_adapter *adapter = netdev_priv(netdev);
7358        int ret;
7359
7360        ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7361
7362        return min_t(int, ret, 0);
7363}
7364
7365static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7366{
7367        struct igb_adapter *adapter = netdev_priv(netdev);
7368
7369        igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7370
7371        return 0;
7372}
7373
7374static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7375                                 const u32 info, const u8 *addr)
7376{
7377        struct pci_dev *pdev = adapter->pdev;
7378        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7379        struct list_head *pos;
7380        struct vf_mac_filter *entry = NULL;
7381        int ret = 0;
7382
7383        switch (info) {
7384        case E1000_VF_MAC_FILTER_CLR:
7385                /* remove all unicast MAC filters related to the current VF */
7386                list_for_each(pos, &adapter->vf_macs.l) {
7387                        entry = list_entry(pos, struct vf_mac_filter, l);
7388                        if (entry->vf == vf) {
7389                                entry->vf = -1;
7390                                entry->free = true;
7391                                igb_del_mac_filter(adapter, entry->vf_mac, vf);
7392                        }
7393                }
7394                break;
7395        case E1000_VF_MAC_FILTER_ADD:
7396                if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7397                    !vf_data->trusted) {
7398                        dev_warn(&pdev->dev,
7399                                 "VF %d requested MAC filter but is administratively denied\n",
7400                                 vf);
7401                        return -EINVAL;
7402                }
7403                if (!is_valid_ether_addr(addr)) {
7404                        dev_warn(&pdev->dev,
7405                                 "VF %d attempted to set invalid MAC filter\n",
7406                                 vf);
7407                        return -EINVAL;
7408                }
7409
7410                /* try to find empty slot in the list */
7411                list_for_each(pos, &adapter->vf_macs.l) {
7412                        entry = list_entry(pos, struct vf_mac_filter, l);
7413                        if (entry->free)
7414                                break;
7415                }
7416
7417                if (entry && entry->free) {
7418                        entry->free = false;
7419                        entry->vf = vf;
7420                        ether_addr_copy(entry->vf_mac, addr);
7421
7422                        ret = igb_add_mac_filter(adapter, addr, vf);
7423                        ret = min_t(int, ret, 0);
7424                } else {
7425                        ret = -ENOSPC;
7426                }
7427
7428                if (ret == -ENOSPC)
7429                        dev_warn(&pdev->dev,
7430                                 "VF %d has requested MAC filter but there is no space for it\n",
7431                                 vf);
7432                break;
7433        default:
7434                ret = -EINVAL;
7435                break;
7436        }
7437
7438        return ret;
7439}
7440
7441static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7442{
7443        struct pci_dev *pdev = adapter->pdev;
7444        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7445        u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7446
7447        /* The VF MAC Address is stored in a packed array of bytes
7448         * starting at the second 32 bit word of the msg array
7449         */
7450        unsigned char *addr = (unsigned char *)&msg[1];
7451        int ret = 0;
7452
7453        if (!info) {
7454                if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7455                    !vf_data->trusted) {
7456                        dev_warn(&pdev->dev,
7457                                 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7458                                 vf);
7459                        return -EINVAL;
7460                }
7461
7462                if (!is_valid_ether_addr(addr)) {
7463                        dev_warn(&pdev->dev,
7464                                 "VF %d attempted to set invalid MAC\n",
7465                                 vf);
7466                        return -EINVAL;
7467                }
7468
7469                ret = igb_set_vf_mac(adapter, vf, addr);
7470        } else {
7471                ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7472        }
7473
7474        return ret;
7475}
7476
7477static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7478{
7479        struct e1000_hw *hw = &adapter->hw;
7480        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7481        u32 msg = E1000_VT_MSGTYPE_NACK;
7482
7483        /* if device isn't clear to send it shouldn't be reading either */
7484        if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7485            time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7486                igb_write_mbx(hw, &msg, 1, vf);
7487                vf_data->last_nack = jiffies;
7488        }
7489}
7490
7491static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7492{
7493        struct pci_dev *pdev = adapter->pdev;
7494        u32 msgbuf[E1000_VFMAILBOX_SIZE];
7495        struct e1000_hw *hw = &adapter->hw;
7496        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7497        s32 retval;
7498
7499        retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7500
7501        if (retval) {
7502                /* if receive failed revoke VF CTS stats and restart init */
7503                dev_err(&pdev->dev, "Error receiving message from VF\n");
7504                vf_data->flags &= ~IGB_VF_FLAG_CTS;
7505                if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7506                        goto unlock;
7507                goto out;
7508        }
7509
7510        /* this is a message we already processed, do nothing */
7511        if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7512                goto unlock;
7513
7514        /* until the vf completes a reset it should not be
7515         * allowed to start any configuration.
7516         */
7517        if (msgbuf[0] == E1000_VF_RESET) {
7518                /* unlocks mailbox */
7519                igb_vf_reset_msg(adapter, vf);
7520                return;
7521        }
7522
7523        if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7524                if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7525                        goto unlock;
7526                retval = -1;
7527                goto out;
7528        }
7529
7530        switch ((msgbuf[0] & 0xFFFF)) {
7531        case E1000_VF_SET_MAC_ADDR:
7532                retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7533                break;
7534        case E1000_VF_SET_PROMISC:
7535                retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7536                break;
7537        case E1000_VF_SET_MULTICAST:
7538                retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
7539                break;
7540        case E1000_VF_SET_LPE:
7541                retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
7542                break;
7543        case E1000_VF_SET_VLAN:
7544                retval = -1;
7545                if (vf_data->pf_vlan)
7546                        dev_warn(&pdev->dev,
7547                                 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
7548                                 vf);
7549                else
7550                        retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
7551                break;
7552        default:
7553                dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
7554                retval = -1;
7555                break;
7556        }
7557
7558        msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
7559out:
7560        /* notify the VF of the results of what it sent us */
7561        if (retval)
7562                msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
7563        else
7564                msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
7565
7566        /* unlocks mailbox */
7567        igb_write_mbx(hw, msgbuf, 1, vf);
7568        return;
7569
7570unlock:
7571        igb_unlock_mbx(hw, vf);
7572}
7573
7574static void igb_msg_task(struct igb_adapter *adapter)
7575{
7576        struct e1000_hw *hw = &adapter->hw;
7577        u32 vf;
7578
7579        for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
7580                /* process any reset requests */
7581                if (!igb_check_for_rst(hw, vf))
7582                        igb_vf_reset_event(adapter, vf);
7583
7584                /* process any messages pending */
7585                if (!igb_check_for_msg(hw, vf))
7586                        igb_rcv_msg_from_vf(adapter, vf);
7587
7588                /* process any acks */
7589                if (!igb_check_for_ack(hw, vf))
7590                        igb_rcv_ack_from_vf(adapter, vf);
7591        }
7592}
7593
7594/**
7595 *  igb_set_uta - Set unicast filter table address
7596 *  @adapter: board private structure
7597 *  @set: boolean indicating if we are setting or clearing bits
7598 *
7599 *  The unicast table address is a register array of 32-bit registers.
7600 *  The table is meant to be used in a way similar to how the MTA is used
7601 *  however due to certain limitations in the hardware it is necessary to
7602 *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
7603 *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
7604 **/
7605static void igb_set_uta(struct igb_adapter *adapter, bool set)
7606{
7607        struct e1000_hw *hw = &adapter->hw;
7608        u32 uta = set ? ~0 : 0;
7609        int i;
7610
7611        /* we only need to do this if VMDq is enabled */
7612        if (!adapter->vfs_allocated_count)
7613                return;
7614
7615        for (i = hw->mac.uta_reg_count; i--;)
7616                array_wr32(E1000_UTA, i, uta);
7617}
7618
7619/**
7620 *  igb_intr_msi - Interrupt Handler
7621 *  @irq: interrupt number
7622 *  @data: pointer to a network interface device structure
7623 **/
7624static irqreturn_t igb_intr_msi(int irq, void *data)
7625{
7626        struct igb_adapter *adapter = data;
7627        struct igb_q_vector *q_vector = adapter->q_vector[0];
7628        struct e1000_hw *hw = &adapter->hw;
7629        /* read ICR disables interrupts using IAM */
7630        u32 icr = rd32(E1000_ICR);
7631
7632        igb_write_itr(q_vector);
7633
7634        if (icr & E1000_ICR_DRSTA)
7635                schedule_work(&adapter->reset_task);
7636
7637        if (icr & E1000_ICR_DOUTSYNC) {
7638                /* HW is reporting DMA is out of sync */
7639                adapter->stats.doosync++;
7640        }
7641
7642        if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7643                hw->mac.get_link_status = 1;
7644                if (!test_bit(__IGB_DOWN, &adapter->state))
7645                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
7646        }
7647
7648        if (icr & E1000_ICR_TS)
7649                igb_tsync_interrupt(adapter);
7650
7651        napi_schedule(&q_vector->napi);
7652
7653        return IRQ_HANDLED;
7654}
7655
7656/**
7657 *  igb_intr - Legacy Interrupt Handler
7658 *  @irq: interrupt number
7659 *  @data: pointer to a network interface device structure
7660 **/
7661static irqreturn_t igb_intr(int irq, void *data)
7662{
7663        struct igb_adapter *adapter = data;
7664        struct igb_q_vector *q_vector = adapter->q_vector[0];
7665        struct e1000_hw *hw = &adapter->hw;
7666        /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
7667         * need for the IMC write
7668         */
7669        u32 icr = rd32(E1000_ICR);
7670
7671        /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
7672         * not set, then the adapter didn't send an interrupt
7673         */
7674        if (!(icr & E1000_ICR_INT_ASSERTED))
7675                return IRQ_NONE;
7676
7677        igb_write_itr(q_vector);
7678
7679        if (icr & E1000_ICR_DRSTA)
7680                schedule_work(&adapter->reset_task);
7681
7682        if (icr & E1000_ICR_DOUTSYNC) {
7683                /* HW is reporting DMA is out of sync */
7684                adapter->stats.doosync++;
7685        }
7686
7687        if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7688                hw->mac.get_link_status = 1;
7689                /* guard against interrupt when we're going down */
7690                if (!test_bit(__IGB_DOWN, &adapter->state))
7691                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
7692        }
7693
7694        if (icr & E1000_ICR_TS)
7695                igb_tsync_interrupt(adapter);
7696
7697        napi_schedule(&q_vector->napi);
7698
7699        return IRQ_HANDLED;
7700}
7701
7702static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
7703{
7704        struct igb_adapter *adapter = q_vector->adapter;
7705        struct e1000_hw *hw = &adapter->hw;
7706
7707        if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
7708            (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
7709                if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
7710                        igb_set_itr(q_vector);
7711                else
7712                        igb_update_ring_itr(q_vector);
7713        }
7714
7715        if (!test_bit(__IGB_DOWN, &adapter->state)) {
7716                if (adapter->flags & IGB_FLAG_HAS_MSIX)
7717                        wr32(E1000_EIMS, q_vector->eims_value);
7718                else
7719                        igb_irq_enable(adapter);
7720        }
7721}
7722
7723/**
7724 *  igb_poll - NAPI Rx polling callback
7725 *  @napi: napi polling structure
7726 *  @budget: count of how many packets we should handle
7727 **/
7728static int igb_poll(struct napi_struct *napi, int budget)
7729{
7730        struct igb_q_vector *q_vector = container_of(napi,
7731                                                     struct igb_q_vector,
7732                                                     napi);
7733        bool clean_complete = true;
7734        int work_done = 0;
7735
7736#ifdef CONFIG_IGB_DCA
7737        if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
7738                igb_update_dca(q_vector);
7739#endif
7740        if (q_vector->tx.ring)
7741                clean_complete = igb_clean_tx_irq(q_vector, budget);
7742
7743        if (q_vector->rx.ring) {
7744                int cleaned = igb_clean_rx_irq(q_vector, budget);
7745
7746                work_done += cleaned;
7747                if (cleaned >= budget)
7748                        clean_complete = false;
7749        }
7750
7751        /* If all work not completed, return budget and keep polling */
7752        if (!clean_complete)
7753                return budget;
7754
7755        /* Exit the polling mode, but don't re-enable interrupts if stack might
7756         * poll us due to busy-polling
7757         */
7758        if (likely(napi_complete_done(napi, work_done)))
7759                igb_ring_irq_enable(q_vector);
7760
7761        return min(work_done, budget - 1);
7762}
7763
7764/**
7765 *  igb_clean_tx_irq - Reclaim resources after transmit completes
7766 *  @q_vector: pointer to q_vector containing needed info
7767 *  @napi_budget: Used to determine if we are in netpoll
7768 *
7769 *  returns true if ring is completely cleaned
7770 **/
7771static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
7772{
7773        struct igb_adapter *adapter = q_vector->adapter;
7774        struct igb_ring *tx_ring = q_vector->tx.ring;
7775        struct igb_tx_buffer *tx_buffer;
7776        union e1000_adv_tx_desc *tx_desc;
7777        unsigned int total_bytes = 0, total_packets = 0;
7778        unsigned int budget = q_vector->tx.work_limit;
7779        unsigned int i = tx_ring->next_to_clean;
7780
7781        if (test_bit(__IGB_DOWN, &adapter->state))
7782                return true;
7783
7784        tx_buffer = &tx_ring->tx_buffer_info[i];
7785        tx_desc = IGB_TX_DESC(tx_ring, i);
7786        i -= tx_ring->count;
7787
7788        do {
7789                union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
7790
7791                /* if next_to_watch is not set then there is no work pending */
7792                if (!eop_desc)
7793                        break;
7794
7795                /* prevent any other reads prior to eop_desc */
7796                smp_rmb();
7797
7798                /* if DD is not set pending work has not been completed */
7799                if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
7800                        break;
7801
7802                /* clear next_to_watch to prevent false hangs */
7803                tx_buffer->next_to_watch = NULL;
7804
7805                /* update the statistics for this packet */
7806                total_bytes += tx_buffer->bytecount;
7807                total_packets += tx_buffer->gso_segs;
7808
7809                /* free the skb */
7810                napi_consume_skb(tx_buffer->skb, napi_budget);
7811
7812                /* unmap skb header data */
7813                dma_unmap_single(tx_ring->dev,
7814                                 dma_unmap_addr(tx_buffer, dma),
7815                                 dma_unmap_len(tx_buffer, len),
7816                                 DMA_TO_DEVICE);
7817
7818                /* clear tx_buffer data */
7819                dma_unmap_len_set(tx_buffer, len, 0);
7820
7821                /* clear last DMA location and unmap remaining buffers */
7822                while (tx_desc != eop_desc) {
7823                        tx_buffer++;
7824                        tx_desc++;
7825                        i++;
7826                        if (unlikely(!i)) {
7827                                i -= tx_ring->count;
7828                                tx_buffer = tx_ring->tx_buffer_info;
7829                                tx_desc = IGB_TX_DESC(tx_ring, 0);
7830                        }
7831
7832                        /* unmap any remaining paged data */
7833                        if (dma_unmap_len(tx_buffer, len)) {
7834                                dma_unmap_page(tx_ring->dev,
7835                                               dma_unmap_addr(tx_buffer, dma),
7836                                               dma_unmap_len(tx_buffer, len),
7837                                               DMA_TO_DEVICE);
7838                                dma_unmap_len_set(tx_buffer, len, 0);
7839                        }
7840                }
7841
7842                /* move us one more past the eop_desc for start of next pkt */
7843                tx_buffer++;
7844                tx_desc++;
7845                i++;
7846                if (unlikely(!i)) {
7847                        i -= tx_ring->count;
7848                        tx_buffer = tx_ring->tx_buffer_info;
7849                        tx_desc = IGB_TX_DESC(tx_ring, 0);
7850                }
7851
7852                /* issue prefetch for next Tx descriptor */
7853                prefetch(tx_desc);
7854
7855                /* update budget accounting */
7856                budget--;
7857        } while (likely(budget));
7858
7859        netdev_tx_completed_queue(txring_txq(tx_ring),
7860                                  total_packets, total_bytes);
7861        i += tx_ring->count;
7862        tx_ring->next_to_clean = i;
7863        u64_stats_update_begin(&tx_ring->tx_syncp);
7864        tx_ring->tx_stats.bytes += total_bytes;
7865        tx_ring->tx_stats.packets += total_packets;
7866        u64_stats_update_end(&tx_ring->tx_syncp);
7867        q_vector->tx.total_bytes += total_bytes;
7868        q_vector->tx.total_packets += total_packets;
7869
7870        if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
7871                struct e1000_hw *hw = &adapter->hw;
7872
7873                /* Detect a transmit hang in hardware, this serializes the
7874                 * check with the clearing of time_stamp and movement of i
7875                 */
7876                clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
7877                if (tx_buffer->next_to_watch &&
7878                    time_after(jiffies, tx_buffer->time_stamp +
7879                               (adapter->tx_timeout_factor * HZ)) &&
7880                    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
7881
7882                        /* detected Tx unit hang */
7883                        dev_err(tx_ring->dev,
7884                                "Detected Tx Unit Hang\n"
7885                                "  Tx Queue             <%d>\n"
7886                                "  TDH                  <%x>\n"
7887                                "  TDT                  <%x>\n"
7888                                "  next_to_use          <%x>\n"
7889                                "  next_to_clean        <%x>\n"
7890                                "buffer_info[next_to_clean]\n"
7891                                "  time_stamp           <%lx>\n"
7892                                "  next_to_watch        <%p>\n"
7893                                "  jiffies              <%lx>\n"
7894                                "  desc.status          <%x>\n",
7895                                tx_ring->queue_index,
7896                                rd32(E1000_TDH(tx_ring->reg_idx)),
7897                                readl(tx_ring->tail),
7898                                tx_ring->next_to_use,
7899                                tx_ring->next_to_clean,
7900                                tx_buffer->time_stamp,
7901                                tx_buffer->next_to_watch,
7902                                jiffies,
7903                                tx_buffer->next_to_watch->wb.status);
7904                        netif_stop_subqueue(tx_ring->netdev,
7905                                            tx_ring->queue_index);
7906
7907                        /* we are about to reset, no point in enabling stuff */
7908                        return true;
7909                }
7910        }
7911
7912#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
7913        if (unlikely(total_packets &&
7914            netif_carrier_ok(tx_ring->netdev) &&
7915            igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
7916                /* Make sure that anybody stopping the queue after this
7917                 * sees the new next_to_clean.
7918                 */
7919                smp_mb();
7920                if (__netif_subqueue_stopped(tx_ring->netdev,
7921                                             tx_ring->queue_index) &&
7922                    !(test_bit(__IGB_DOWN, &adapter->state))) {
7923                        netif_wake_subqueue(tx_ring->netdev,
7924                                            tx_ring->queue_index);
7925
7926                        u64_stats_update_begin(&tx_ring->tx_syncp);
7927                        tx_ring->tx_stats.restart_queue++;
7928                        u64_stats_update_end(&tx_ring->tx_syncp);
7929                }
7930        }
7931
7932        return !!budget;
7933}
7934
7935/**
7936 *  igb_reuse_rx_page - page flip buffer and store it back on the ring
7937 *  @rx_ring: rx descriptor ring to store buffers on
7938 *  @old_buff: donor buffer to have page reused
7939 *
7940 *  Synchronizes page for reuse by the adapter
7941 **/
7942static void igb_reuse_rx_page(struct igb_ring *rx_ring,
7943                              struct igb_rx_buffer *old_buff)
7944{
7945        struct igb_rx_buffer *new_buff;
7946        u16 nta = rx_ring->next_to_alloc;
7947
7948        new_buff = &rx_ring->rx_buffer_info[nta];
7949
7950        /* update, and store next to alloc */
7951        nta++;
7952        rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
7953
7954        /* Transfer page from old buffer to new buffer.
7955         * Move each member individually to avoid possible store
7956         * forwarding stalls.
7957         */
7958        new_buff->dma           = old_buff->dma;
7959        new_buff->page          = old_buff->page;
7960        new_buff->page_offset   = old_buff->page_offset;
7961        new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
7962}
7963
7964static inline bool igb_page_is_reserved(struct page *page)
7965{
7966        return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
7967}
7968
7969static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer)
7970{
7971        unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
7972        struct page *page = rx_buffer->page;
7973
7974        /* avoid re-using remote pages */
7975        if (unlikely(igb_page_is_reserved(page)))
7976                return false;
7977
7978#if (PAGE_SIZE < 8192)
7979        /* if we are only owner of page we can reuse it */
7980        if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
7981                return false;
7982#else
7983#define IGB_LAST_OFFSET \
7984        (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
7985
7986        if (rx_buffer->page_offset > IGB_LAST_OFFSET)
7987                return false;
7988#endif
7989
7990        /* If we have drained the page fragment pool we need to update
7991         * the pagecnt_bias and page count so that we fully restock the
7992         * number of references the driver holds.
7993         */
7994        if (unlikely(!pagecnt_bias)) {
7995                page_ref_add(page, USHRT_MAX);
7996                rx_buffer->pagecnt_bias = USHRT_MAX;
7997        }
7998
7999        return true;
8000}
8001
8002/**
8003 *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8004 *  @rx_ring: rx descriptor ring to transact packets on
8005 *  @rx_buffer: buffer containing page to add
8006 *  @skb: sk_buff to place the data into
8007 *  @size: size of buffer to be added
8008 *
8009 *  This function will add the data contained in rx_buffer->page to the skb.
8010 **/
8011static void igb_add_rx_frag(struct igb_ring *rx_ring,
8012                            struct igb_rx_buffer *rx_buffer,
8013                            struct sk_buff *skb,
8014                            unsigned int size)
8015{
8016#if (PAGE_SIZE < 8192)
8017        unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8018#else
8019        unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8020                                SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8021                                SKB_DATA_ALIGN(size);
8022#endif
8023        skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8024                        rx_buffer->page_offset, size, truesize);
8025#if (PAGE_SIZE < 8192)
8026        rx_buffer->page_offset ^= truesize;
8027#else
8028        rx_buffer->page_offset += truesize;
8029#endif
8030}
8031
8032static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8033                                         struct igb_rx_buffer *rx_buffer,
8034                                         union e1000_adv_rx_desc *rx_desc,
8035                                         unsigned int size)
8036{
8037        void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
8038#if (PAGE_SIZE < 8192)
8039        unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8040#else
8041        unsigned int truesize = SKB_DATA_ALIGN(size);
8042#endif
8043        unsigned int headlen;
8044        struct sk_buff *skb;
8045
8046        /* prefetch first cache line of first page */
8047        prefetch(va);
8048#if L1_CACHE_BYTES < 128
8049        prefetch(va + L1_CACHE_BYTES);
8050#endif
8051
8052        /* allocate a skb to store the frags */
8053        skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8054        if (unlikely(!skb))
8055                return NULL;
8056
8057        if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) {
8058                igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
8059                va += IGB_TS_HDR_LEN;
8060                size -= IGB_TS_HDR_LEN;
8061        }
8062
8063        /* Determine available headroom for copy */
8064        headlen = size;
8065        if (headlen > IGB_RX_HDR_LEN)
8066                headlen = eth_get_headlen(va, IGB_RX_HDR_LEN);
8067
8068        /* align pull length to size of long to optimize memcpy performance */
8069        memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
8070
8071        /* update all of the pointers */
8072        size -= headlen;
8073        if (size) {
8074                skb_add_rx_frag(skb, 0, rx_buffer->page,
8075                                (va + headlen) - page_address(rx_buffer->page),
8076                                size, truesize);
8077#if (PAGE_SIZE < 8192)
8078                rx_buffer->page_offset ^= truesize;
8079#else
8080                rx_buffer->page_offset += truesize;
8081#endif
8082        } else {
8083                rx_buffer->pagecnt_bias++;
8084        }
8085
8086        return skb;
8087}
8088
8089static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8090                                     struct igb_rx_buffer *rx_buffer,
8091                                     union e1000_adv_rx_desc *rx_desc,
8092                                     unsigned int size)
8093{
8094        void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
8095#if (PAGE_SIZE < 8192)
8096        unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8097#else
8098        unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8099                                SKB_DATA_ALIGN(IGB_SKB_PAD + size);
8100#endif
8101        struct sk_buff *skb;
8102
8103        /* prefetch first cache line of first page */
8104        prefetch(va);
8105#if L1_CACHE_BYTES < 128
8106        prefetch(va + L1_CACHE_BYTES);
8107#endif
8108
8109        /* build an skb around the page buffer */
8110        skb = build_skb(va - IGB_SKB_PAD, truesize);
8111        if (unlikely(!skb))
8112                return NULL;
8113
8114        /* update pointers within the skb to store the data */
8115        skb_reserve(skb, IGB_SKB_PAD);
8116        __skb_put(skb, size);
8117
8118        /* pull timestamp out of packet data */
8119        if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8120                igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
8121                __skb_pull(skb, IGB_TS_HDR_LEN);
8122        }
8123
8124        /* update buffer offset */
8125#if (PAGE_SIZE < 8192)
8126        rx_buffer->page_offset ^= truesize;
8127#else
8128        rx_buffer->page_offset += truesize;
8129#endif
8130
8131        return skb;
8132}
8133
8134static inline void igb_rx_checksum(struct igb_ring *ring,
8135                                   union e1000_adv_rx_desc *rx_desc,
8136                                   struct sk_buff *skb)
8137{
8138        skb_checksum_none_assert(skb);
8139
8140        /* Ignore Checksum bit is set */
8141        if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8142                return;
8143
8144        /* Rx checksum disabled via ethtool */
8145        if (!(ring->netdev->features & NETIF_F_RXCSUM))
8146                return;
8147
8148        /* TCP/UDP checksum error bit is set */
8149        if (igb_test_staterr(rx_desc,
8150                             E1000_RXDEXT_STATERR_TCPE |
8151                             E1000_RXDEXT_STATERR_IPE)) {
8152                /* work around errata with sctp packets where the TCPE aka
8153                 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8154                 * packets, (aka let the stack check the crc32c)
8155                 */
8156                if (!((skb->len == 60) &&
8157                      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8158                        u64_stats_update_begin(&ring->rx_syncp);
8159                        ring->rx_stats.csum_err++;
8160                        u64_stats_update_end(&ring->rx_syncp);
8161                }
8162                /* let the stack verify checksum errors */
8163                return;
8164        }
8165        /* It must be a TCP or UDP packet with a valid checksum */
8166        if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8167                                      E1000_RXD_STAT_UDPCS))
8168                skb->ip_summed = CHECKSUM_UNNECESSARY;
8169
8170        dev_dbg(ring->dev, "cksum success: bits %08X\n",
8171                le32_to_cpu(rx_desc->wb.upper.status_error));
8172}
8173
8174static inline void igb_rx_hash(struct igb_ring *ring,
8175                               union e1000_adv_rx_desc *rx_desc,
8176                               struct sk_buff *skb)
8177{
8178        if (ring->netdev->features & NETIF_F_RXHASH)
8179                skb_set_hash(skb,
8180                             le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8181                             PKT_HASH_TYPE_L3);
8182}
8183
8184/**
8185 *  igb_is_non_eop - process handling of non-EOP buffers
8186 *  @rx_ring: Rx ring being processed
8187 *  @rx_desc: Rx descriptor for current buffer
8188 *  @skb: current socket buffer containing buffer in progress
8189 *
8190 *  This function updates next to clean.  If the buffer is an EOP buffer
8191 *  this function exits returning false, otherwise it will place the
8192 *  sk_buff in the next buffer to be chained and return true indicating
8193 *  that this is in fact a non-EOP buffer.
8194 **/
8195static bool igb_is_non_eop(struct igb_ring *rx_ring,
8196                           union e1000_adv_rx_desc *rx_desc)
8197{
8198        u32 ntc = rx_ring->next_to_clean + 1;
8199
8200        /* fetch, update, and store next to clean */
8201        ntc = (ntc < rx_ring->count) ? ntc : 0;
8202        rx_ring->next_to_clean = ntc;
8203
8204        prefetch(IGB_RX_DESC(rx_ring, ntc));
8205
8206        if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8207                return false;
8208
8209        return true;
8210}
8211
8212/**
8213 *  igb_cleanup_headers - Correct corrupted or empty headers
8214 *  @rx_ring: rx descriptor ring packet is being transacted on
8215 *  @rx_desc: pointer to the EOP Rx descriptor
8216 *  @skb: pointer to current skb being fixed
8217 *
8218 *  Address the case where we are pulling data in on pages only
8219 *  and as such no data is present in the skb header.
8220 *
8221 *  In addition if skb is not at least 60 bytes we need to pad it so that
8222 *  it is large enough to qualify as a valid Ethernet frame.
8223 *
8224 *  Returns true if an error was encountered and skb was freed.
8225 **/
8226static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8227                                union e1000_adv_rx_desc *rx_desc,
8228                                struct sk_buff *skb)
8229{
8230        if (unlikely((igb_test_staterr(rx_desc,
8231                                       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8232                struct net_device *netdev = rx_ring->netdev;
8233                if (!(netdev->features & NETIF_F_RXALL)) {
8234                        dev_kfree_skb_any(skb);
8235                        return true;
8236                }
8237        }
8238
8239        /* if eth_skb_pad returns an error the skb was freed */
8240        if (eth_skb_pad(skb))
8241                return true;
8242
8243        return false;
8244}
8245
8246/**
8247 *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
8248 *  @rx_ring: rx descriptor ring packet is being transacted on
8249 *  @rx_desc: pointer to the EOP Rx descriptor
8250 *  @skb: pointer to current skb being populated
8251 *
8252 *  This function checks the ring, descriptor, and packet information in
8253 *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
8254 *  other fields within the skb.
8255 **/
8256static void igb_process_skb_fields(struct igb_ring *rx_ring,
8257                                   union e1000_adv_rx_desc *rx_desc,
8258                                   struct sk_buff *skb)
8259{
8260        struct net_device *dev = rx_ring->netdev;
8261
8262        igb_rx_hash(rx_ring, rx_desc, skb);
8263
8264        igb_rx_checksum(rx_ring, rx_desc, skb);
8265
8266        if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8267            !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8268                igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8269
8270        if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8271            igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8272                u16 vid;
8273
8274                if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8275                    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8276                        vid = be16_to_cpu(rx_desc->wb.upper.vlan);
8277                else
8278                        vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8279
8280                __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8281        }
8282
8283        skb_record_rx_queue(skb, rx_ring->queue_index);
8284
8285        skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8286}
8287
8288static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8289                                               const unsigned int size)
8290{
8291        struct igb_rx_buffer *rx_buffer;
8292
8293        rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8294        prefetchw(rx_buffer->page);
8295
8296        /* we are reusing so sync this buffer for CPU use */
8297        dma_sync_single_range_for_cpu(rx_ring->dev,
8298                                      rx_buffer->dma,
8299                                      rx_buffer->page_offset,
8300                                      size,
8301                                      DMA_FROM_DEVICE);
8302
8303        rx_buffer->pagecnt_bias--;
8304
8305        return rx_buffer;
8306}
8307
8308static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8309                              struct igb_rx_buffer *rx_buffer)
8310{
8311        if (igb_can_reuse_rx_page(rx_buffer)) {
8312                /* hand second half of page back to the ring */
8313                igb_reuse_rx_page(rx_ring, rx_buffer);
8314        } else {
8315                /* We are not reusing the buffer so unmap it and free
8316                 * any references we are holding to it
8317                 */
8318                dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8319                                     igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8320                                     IGB_RX_DMA_ATTR);
8321                __page_frag_cache_drain(rx_buffer->page,
8322                                        rx_buffer->pagecnt_bias);
8323        }
8324
8325        /* clear contents of rx_buffer */
8326        rx_buffer->page = NULL;
8327}
8328
8329static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8330{
8331        struct igb_ring *rx_ring = q_vector->rx.ring;
8332        struct sk_buff *skb = rx_ring->skb;
8333        unsigned int total_bytes = 0, total_packets = 0;
8334        u16 cleaned_count = igb_desc_unused(rx_ring);
8335
8336        while (likely(total_packets < budget)) {
8337                union e1000_adv_rx_desc *rx_desc;
8338                struct igb_rx_buffer *rx_buffer;
8339                unsigned int size;
8340
8341                /* return some buffers to hardware, one at a time is too slow */
8342                if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8343                        igb_alloc_rx_buffers(rx_ring, cleaned_count);
8344                        cleaned_count = 0;
8345                }
8346
8347                rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8348                size = le16_to_cpu(rx_desc->wb.upper.length);
8349                if (!size)
8350                        break;
8351
8352                /* This memory barrier is needed to keep us from reading
8353                 * any other fields out of the rx_desc until we know the
8354                 * descriptor has been written back
8355                 */
8356                dma_rmb();
8357
8358                rx_buffer = igb_get_rx_buffer(rx_ring, size);
8359
8360                /* retrieve a buffer from the ring */
8361                if (skb)
8362                        igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8363                else if (ring_uses_build_skb(rx_ring))
8364                        skb = igb_build_skb(rx_ring, rx_buffer, rx_desc, size);
8365                else
8366                        skb = igb_construct_skb(rx_ring, rx_buffer,
8367                                                rx_desc, size);
8368
8369                /* exit if we failed to retrieve a buffer */
8370                if (!skb) {
8371                        rx_ring->rx_stats.alloc_failed++;
8372                        rx_buffer->pagecnt_bias++;
8373                        break;
8374                }
8375
8376                igb_put_rx_buffer(rx_ring, rx_buffer);
8377                cleaned_count++;
8378
8379                /* fetch next buffer in frame if non-eop */
8380                if (igb_is_non_eop(rx_ring, rx_desc))
8381                        continue;
8382
8383                /* verify the packet layout is correct */
8384                if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8385                        skb = NULL;
8386                        continue;
8387                }
8388
8389                /* probably a little skewed due to removing CRC */
8390                total_bytes += skb->len;
8391
8392                /* populate checksum, timestamp, VLAN, and protocol */
8393                igb_process_skb_fields(rx_ring, rx_desc, skb);
8394
8395                napi_gro_receive(&q_vector->napi, skb);
8396
8397                /* reset skb pointer */
8398                skb = NULL;
8399
8400                /* update budget accounting */
8401                total_packets++;
8402        }
8403
8404        /* place incomplete frames back on ring for completion */
8405        rx_ring->skb = skb;
8406
8407        u64_stats_update_begin(&rx_ring->rx_syncp);
8408        rx_ring->rx_stats.packets += total_packets;
8409        rx_ring->rx_stats.bytes += total_bytes;
8410        u64_stats_update_end(&rx_ring->rx_syncp);
8411        q_vector->rx.total_packets += total_packets;
8412        q_vector->rx.total_bytes += total_bytes;
8413
8414        if (cleaned_count)
8415                igb_alloc_rx_buffers(rx_ring, cleaned_count);
8416
8417        return total_packets;
8418}
8419
8420static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8421{
8422        return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8423}
8424
8425static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
8426                                  struct igb_rx_buffer *bi)
8427{
8428        struct page *page = bi->page;
8429        dma_addr_t dma;
8430
8431        /* since we are recycling buffers we should seldom need to alloc */
8432        if (likely(page))
8433                return true;
8434
8435        /* alloc new page for storage */
8436        page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
8437        if (unlikely(!page)) {
8438                rx_ring->rx_stats.alloc_failed++;
8439                return false;
8440        }
8441
8442        /* map page for use */
8443        dma = dma_map_page_attrs(rx_ring->dev, page, 0,
8444                                 igb_rx_pg_size(rx_ring),
8445                                 DMA_FROM_DEVICE,
8446                                 IGB_RX_DMA_ATTR);
8447
8448        /* if mapping failed free memory back to system since
8449         * there isn't much point in holding memory we can't use
8450         */
8451        if (dma_mapping_error(rx_ring->dev, dma)) {
8452                __free_pages(page, igb_rx_pg_order(rx_ring));
8453
8454                rx_ring->rx_stats.alloc_failed++;
8455                return false;
8456        }
8457
8458        bi->dma = dma;
8459        bi->page = page;
8460        bi->page_offset = igb_rx_offset(rx_ring);
8461        bi->pagecnt_bias = 1;
8462
8463        return true;
8464}
8465
8466/**
8467 *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
8468 *  @adapter: address of board private structure
8469 **/
8470void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
8471{
8472        union e1000_adv_rx_desc *rx_desc;
8473        struct igb_rx_buffer *bi;
8474        u16 i = rx_ring->next_to_use;
8475        u16 bufsz;
8476
8477        /* nothing to do */
8478        if (!cleaned_count)
8479                return;
8480
8481        rx_desc = IGB_RX_DESC(rx_ring, i);
8482        bi = &rx_ring->rx_buffer_info[i];
8483        i -= rx_ring->count;
8484
8485        bufsz = igb_rx_bufsz(rx_ring);
8486
8487        do {
8488                if (!igb_alloc_mapped_page(rx_ring, bi))
8489                        break;
8490
8491                /* sync the buffer for use by the device */
8492                dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
8493                                                 bi->page_offset, bufsz,
8494                                                 DMA_FROM_DEVICE);
8495
8496                /* Refresh the desc even if buffer_addrs didn't change
8497                 * because each write-back erases this info.
8498                 */
8499                rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
8500
8501                rx_desc++;
8502                bi++;
8503                i++;
8504                if (unlikely(!i)) {
8505                        rx_desc = IGB_RX_DESC(rx_ring, 0);
8506                        bi = rx_ring->rx_buffer_info;
8507                        i -= rx_ring->count;
8508                }
8509
8510                /* clear the length for the next_to_use descriptor */
8511                rx_desc->wb.upper.length = 0;
8512
8513                cleaned_count--;
8514        } while (cleaned_count);
8515
8516        i += rx_ring->count;
8517
8518        if (rx_ring->next_to_use != i) {
8519                /* record the next descriptor to use */
8520                rx_ring->next_to_use = i;
8521
8522                /* update next to alloc since we have filled the ring */
8523                rx_ring->next_to_alloc = i;
8524
8525                /* Force memory writes to complete before letting h/w
8526                 * know there are new descriptors to fetch.  (Only
8527                 * applicable for weak-ordered memory model archs,
8528                 * such as IA-64).
8529                 */
8530                dma_wmb();
8531                writel(i, rx_ring->tail);
8532        }
8533}
8534
8535/**
8536 * igb_mii_ioctl -
8537 * @netdev:
8538 * @ifreq:
8539 * @cmd:
8540 **/
8541static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8542{
8543        struct igb_adapter *adapter = netdev_priv(netdev);
8544        struct mii_ioctl_data *data = if_mii(ifr);
8545
8546        if (adapter->hw.phy.media_type != e1000_media_type_copper)
8547                return -EOPNOTSUPP;
8548
8549        switch (cmd) {
8550        case SIOCGMIIPHY:
8551                data->phy_id = adapter->hw.phy.addr;
8552                break;
8553        case SIOCGMIIREG:
8554                if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
8555                                     &data->val_out))
8556                        return -EIO;
8557                break;
8558        case SIOCSMIIREG:
8559        default:
8560                return -EOPNOTSUPP;
8561        }
8562        return 0;
8563}
8564
8565/**
8566 * igb_ioctl -
8567 * @netdev:
8568 * @ifreq:
8569 * @cmd:
8570 **/
8571static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8572{
8573        switch (cmd) {
8574        case SIOCGMIIPHY:
8575        case SIOCGMIIREG:
8576        case SIOCSMIIREG:
8577                return igb_mii_ioctl(netdev, ifr, cmd);
8578        case SIOCGHWTSTAMP:
8579                return igb_ptp_get_ts_config(netdev, ifr);
8580        case SIOCSHWTSTAMP:
8581                return igb_ptp_set_ts_config(netdev, ifr);
8582        default:
8583                return -EOPNOTSUPP;
8584        }
8585}
8586
8587void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8588{
8589        struct igb_adapter *adapter = hw->back;
8590
8591        pci_read_config_word(adapter->pdev, reg, value);
8592}
8593
8594void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8595{
8596        struct igb_adapter *adapter = hw->back;
8597
8598        pci_write_config_word(adapter->pdev, reg, *value);
8599}
8600
8601s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8602{
8603        struct igb_adapter *adapter = hw->back;
8604
8605        if (pcie_capability_read_word(adapter->pdev, reg, value))
8606                return -E1000_ERR_CONFIG;
8607
8608        return 0;
8609}
8610
8611s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8612{
8613        struct igb_adapter *adapter = hw->back;
8614
8615        if (pcie_capability_write_word(adapter->pdev, reg, *value))
8616                return -E1000_ERR_CONFIG;
8617
8618        return 0;
8619}
8620
8621static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
8622{
8623        struct igb_adapter *adapter = netdev_priv(netdev);
8624        struct e1000_hw *hw = &adapter->hw;
8625        u32 ctrl, rctl;
8626        bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
8627
8628        if (enable) {
8629                /* enable VLAN tag insert/strip */
8630                ctrl = rd32(E1000_CTRL);
8631                ctrl |= E1000_CTRL_VME;
8632                wr32(E1000_CTRL, ctrl);
8633
8634                /* Disable CFI check */
8635                rctl = rd32(E1000_RCTL);
8636                rctl &= ~E1000_RCTL_CFIEN;
8637                wr32(E1000_RCTL, rctl);
8638        } else {
8639                /* disable VLAN tag insert/strip */
8640                ctrl = rd32(E1000_CTRL);
8641                ctrl &= ~E1000_CTRL_VME;
8642                wr32(E1000_CTRL, ctrl);
8643        }
8644
8645        igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
8646}
8647
8648static int igb_vlan_rx_add_vid(struct net_device *netdev,
8649                               __be16 proto, u16 vid)
8650{
8651        struct igb_adapter *adapter = netdev_priv(netdev);
8652        struct e1000_hw *hw = &adapter->hw;
8653        int pf_id = adapter->vfs_allocated_count;
8654
8655        /* add the filter since PF can receive vlans w/o entry in vlvf */
8656        if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8657                igb_vfta_set(hw, vid, pf_id, true, !!vid);
8658
8659        set_bit(vid, adapter->active_vlans);
8660
8661        return 0;
8662}
8663
8664static int igb_vlan_rx_kill_vid(struct net_device *netdev,
8665                                __be16 proto, u16 vid)
8666{
8667        struct igb_adapter *adapter = netdev_priv(netdev);
8668        int pf_id = adapter->vfs_allocated_count;
8669        struct e1000_hw *hw = &adapter->hw;
8670
8671        /* remove VID from filter table */
8672        if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
8673                igb_vfta_set(hw, vid, pf_id, false, true);
8674
8675        clear_bit(vid, adapter->active_vlans);
8676
8677        return 0;
8678}
8679
8680static void igb_restore_vlan(struct igb_adapter *adapter)
8681{
8682        u16 vid = 1;
8683
8684        igb_vlan_mode(adapter->netdev, adapter->netdev->features);
8685        igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
8686
8687        for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
8688                igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
8689}
8690
8691int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
8692{
8693        struct pci_dev *pdev = adapter->pdev;
8694        struct e1000_mac_info *mac = &adapter->hw.mac;
8695
8696        mac->autoneg = 0;
8697
8698        /* Make sure dplx is at most 1 bit and lsb of speed is not set
8699         * for the switch() below to work
8700         */
8701        if ((spd & 1) || (dplx & ~1))
8702                goto err_inval;
8703
8704        /* Fiber NIC's only allow 1000 gbps Full duplex
8705         * and 100Mbps Full duplex for 100baseFx sfp
8706         */
8707        if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
8708                switch (spd + dplx) {
8709                case SPEED_10 + DUPLEX_HALF:
8710                case SPEED_10 + DUPLEX_FULL:
8711                case SPEED_100 + DUPLEX_HALF:
8712                        goto err_inval;
8713                default:
8714                        break;
8715                }
8716        }
8717
8718        switch (spd + dplx) {
8719        case SPEED_10 + DUPLEX_HALF:
8720                mac->forced_speed_duplex = ADVERTISE_10_HALF;
8721                break;
8722        case SPEED_10 + DUPLEX_FULL:
8723                mac->forced_speed_duplex = ADVERTISE_10_FULL;
8724                break;
8725        case SPEED_100 + DUPLEX_HALF:
8726                mac->forced_speed_duplex = ADVERTISE_100_HALF;
8727                break;
8728        case SPEED_100 + DUPLEX_FULL:
8729                mac->forced_speed_duplex = ADVERTISE_100_FULL;
8730                break;
8731        case SPEED_1000 + DUPLEX_FULL:
8732                mac->autoneg = 1;
8733                adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
8734                break;
8735        case SPEED_1000 + DUPLEX_HALF: /* not supported */
8736        default:
8737                goto err_inval;
8738        }
8739
8740        /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
8741        adapter->hw.phy.mdix = AUTO_ALL_MODES;
8742
8743        return 0;
8744
8745err_inval:
8746        dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
8747        return -EINVAL;
8748}
8749
8750static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
8751                          bool runtime)
8752{
8753        struct net_device *netdev = pci_get_drvdata(pdev);
8754        struct igb_adapter *adapter = netdev_priv(netdev);
8755        struct e1000_hw *hw = &adapter->hw;
8756        u32 ctrl, rctl, status;
8757        u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
8758#ifdef CONFIG_PM
8759        int retval = 0;
8760#endif
8761
8762        rtnl_lock();
8763        netif_device_detach(netdev);
8764
8765        if (netif_running(netdev))
8766                __igb_close(netdev, true);
8767
8768        igb_ptp_suspend(adapter);
8769
8770        igb_clear_interrupt_scheme(adapter);
8771        rtnl_unlock();
8772
8773#ifdef CONFIG_PM
8774        if (!runtime) {
8775                retval = pci_save_state(pdev);
8776                if (retval)
8777                        return retval;
8778        }
8779#endif
8780
8781        status = rd32(E1000_STATUS);
8782        if (status & E1000_STATUS_LU)
8783                wufc &= ~E1000_WUFC_LNKC;
8784
8785        if (wufc) {
8786                igb_setup_rctl(adapter);
8787                igb_set_rx_mode(netdev);
8788
8789                /* turn on all-multi mode if wake on multicast is enabled */
8790                if (wufc & E1000_WUFC_MC) {
8791                        rctl = rd32(E1000_RCTL);
8792                        rctl |= E1000_RCTL_MPE;
8793                        wr32(E1000_RCTL, rctl);
8794                }
8795
8796                ctrl = rd32(E1000_CTRL);
8797                /* advertise wake from D3Cold */
8798                #define E1000_CTRL_ADVD3WUC 0x00100000
8799                /* phy power management enable */
8800                #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
8801                ctrl |= E1000_CTRL_ADVD3WUC;
8802                wr32(E1000_CTRL, ctrl);
8803
8804                /* Allow time for pending master requests to run */
8805                igb_disable_pcie_master(hw);
8806
8807                wr32(E1000_WUC, E1000_WUC_PME_EN);
8808                wr32(E1000_WUFC, wufc);
8809        } else {
8810                wr32(E1000_WUC, 0);
8811                wr32(E1000_WUFC, 0);
8812        }
8813
8814        *enable_wake = wufc || adapter->en_mng_pt;
8815        if (!*enable_wake)
8816                igb_power_down_link(adapter);
8817        else
8818                igb_power_up_link(adapter);
8819
8820        /* Release control of h/w to f/w.  If f/w is AMT enabled, this
8821         * would have already happened in close and is redundant.
8822         */
8823        igb_release_hw_control(adapter);
8824
8825        pci_disable_device(pdev);
8826
8827        return 0;
8828}
8829
8830static void igb_deliver_wake_packet(struct net_device *netdev)
8831{
8832        struct igb_adapter *adapter = netdev_priv(netdev);
8833        struct e1000_hw *hw = &adapter->hw;
8834        struct sk_buff *skb;
8835        u32 wupl;
8836
8837        wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
8838
8839        /* WUPM stores only the first 128 bytes of the wake packet.
8840         * Read the packet only if we have the whole thing.
8841         */
8842        if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
8843                return;
8844
8845        skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
8846        if (!skb)
8847                return;
8848
8849        skb_put(skb, wupl);
8850
8851        /* Ensure reads are 32-bit aligned */
8852        wupl = roundup(wupl, 4);
8853
8854        memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
8855
8856        skb->protocol = eth_type_trans(skb, netdev);
8857        netif_rx(skb);
8858}
8859
8860static int __maybe_unused igb_suspend(struct device *dev)
8861{
8862        int retval;
8863        bool wake;
8864        struct pci_dev *pdev = to_pci_dev(dev);
8865
8866        retval = __igb_shutdown(pdev, &wake, 0);
8867        if (retval)
8868                return retval;
8869
8870        if (wake) {
8871                pci_prepare_to_sleep(pdev);
8872        } else {
8873                pci_wake_from_d3(pdev, false);
8874                pci_set_power_state(pdev, PCI_D3hot);
8875        }
8876
8877        return 0;
8878}
8879
8880static int __maybe_unused igb_resume(struct device *dev)
8881{
8882        struct pci_dev *pdev = to_pci_dev(dev);
8883        struct net_device *netdev = pci_get_drvdata(pdev);
8884        struct igb_adapter *adapter = netdev_priv(netdev);
8885        struct e1000_hw *hw = &adapter->hw;
8886        u32 err, val;
8887
8888        pci_set_power_state(pdev, PCI_D0);
8889        pci_restore_state(pdev);
8890        pci_save_state(pdev);
8891
8892        if (!pci_device_is_present(pdev))
8893                return -ENODEV;
8894        err = pci_enable_device_mem(pdev);
8895        if (err) {
8896                dev_err(&pdev->dev,
8897                        "igb: Cannot enable PCI device from suspend\n");
8898                return err;
8899        }
8900        pci_set_master(pdev);
8901
8902        pci_enable_wake(pdev, PCI_D3hot, 0);
8903        pci_enable_wake(pdev, PCI_D3cold, 0);
8904
8905        if (igb_init_interrupt_scheme(adapter, true)) {
8906                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8907                return -ENOMEM;
8908        }
8909
8910        igb_reset(adapter);
8911
8912        /* let the f/w know that the h/w is now under the control of the
8913         * driver.
8914         */
8915        igb_get_hw_control(adapter);
8916
8917        val = rd32(E1000_WUS);
8918        if (val & WAKE_PKT_WUS)
8919                igb_deliver_wake_packet(netdev);
8920
8921        wr32(E1000_WUS, ~0);
8922
8923        rtnl_lock();
8924        if (!err && netif_running(netdev))
8925                err = __igb_open(netdev, true);
8926
8927        if (!err)
8928                netif_device_attach(netdev);
8929        rtnl_unlock();
8930
8931        return err;
8932}
8933
8934static int __maybe_unused igb_runtime_idle(struct device *dev)
8935{
8936        struct pci_dev *pdev = to_pci_dev(dev);
8937        struct net_device *netdev = pci_get_drvdata(pdev);
8938        struct igb_adapter *adapter = netdev_priv(netdev);
8939
8940        if (!igb_has_link(adapter))
8941                pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
8942
8943        return -EBUSY;
8944}
8945
8946static int __maybe_unused igb_runtime_suspend(struct device *dev)
8947{
8948        struct pci_dev *pdev = to_pci_dev(dev);
8949        int retval;
8950        bool wake;
8951
8952        retval = __igb_shutdown(pdev, &wake, 1);
8953        if (retval)
8954                return retval;
8955
8956        if (wake) {
8957                pci_prepare_to_sleep(pdev);
8958        } else {
8959                pci_wake_from_d3(pdev, false);
8960                pci_set_power_state(pdev, PCI_D3hot);
8961        }
8962
8963        return 0;
8964}
8965
8966static int __maybe_unused igb_runtime_resume(struct device *dev)
8967{
8968        return igb_resume(dev);
8969}
8970
8971static void igb_shutdown(struct pci_dev *pdev)
8972{
8973        bool wake;
8974
8975        __igb_shutdown(pdev, &wake, 0);
8976
8977        if (system_state == SYSTEM_POWER_OFF) {
8978                pci_wake_from_d3(pdev, wake);
8979                pci_set_power_state(pdev, PCI_D3hot);
8980        }
8981}
8982
8983#ifdef CONFIG_PCI_IOV
8984static int igb_sriov_reinit(struct pci_dev *dev)
8985{
8986        struct net_device *netdev = pci_get_drvdata(dev);
8987        struct igb_adapter *adapter = netdev_priv(netdev);
8988        struct pci_dev *pdev = adapter->pdev;
8989
8990        rtnl_lock();
8991
8992        if (netif_running(netdev))
8993                igb_close(netdev);
8994        else
8995                igb_reset(adapter);
8996
8997        igb_clear_interrupt_scheme(adapter);
8998
8999        igb_init_queue_configuration(adapter);
9000
9001        if (igb_init_interrupt_scheme(adapter, true)) {
9002                rtnl_unlock();
9003                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9004                return -ENOMEM;
9005        }
9006
9007        if (netif_running(netdev))
9008                igb_open(netdev);
9009
9010        rtnl_unlock();
9011
9012        return 0;
9013}
9014
9015static int igb_pci_disable_sriov(struct pci_dev *dev)
9016{
9017        int err = igb_disable_sriov(dev);
9018
9019        if (!err)
9020                err = igb_sriov_reinit(dev);
9021
9022        return err;
9023}
9024
9025static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
9026{
9027        int err = igb_enable_sriov(dev, num_vfs);
9028
9029        if (err)
9030                goto out;
9031
9032        err = igb_sriov_reinit(dev);
9033        if (!err)
9034                return num_vfs;
9035
9036out:
9037        return err;
9038}
9039
9040#endif
9041static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9042{
9043#ifdef CONFIG_PCI_IOV
9044        if (num_vfs == 0)
9045                return igb_pci_disable_sriov(dev);
9046        else
9047                return igb_pci_enable_sriov(dev, num_vfs);
9048#endif
9049        return 0;
9050}
9051
9052/**
9053 *  igb_io_error_detected - called when PCI error is detected
9054 *  @pdev: Pointer to PCI device
9055 *  @state: The current pci connection state
9056 *
9057 *  This function is called after a PCI bus error affecting
9058 *  this device has been detected.
9059 **/
9060static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9061                                              pci_channel_state_t state)
9062{
9063        struct net_device *netdev = pci_get_drvdata(pdev);
9064        struct igb_adapter *adapter = netdev_priv(netdev);
9065
9066        netif_device_detach(netdev);
9067
9068        if (state == pci_channel_io_perm_failure)
9069                return PCI_ERS_RESULT_DISCONNECT;
9070
9071        if (netif_running(netdev))
9072                igb_down(adapter);
9073        pci_disable_device(pdev);
9074
9075        /* Request a slot slot reset. */
9076        return PCI_ERS_RESULT_NEED_RESET;
9077}
9078
9079/**
9080 *  igb_io_slot_reset - called after the pci bus has been reset.
9081 *  @pdev: Pointer to PCI device
9082 *
9083 *  Restart the card from scratch, as if from a cold-boot. Implementation
9084 *  resembles the first-half of the igb_resume routine.
9085 **/
9086static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9087{
9088        struct net_device *netdev = pci_get_drvdata(pdev);
9089        struct igb_adapter *adapter = netdev_priv(netdev);
9090        struct e1000_hw *hw = &adapter->hw;
9091        pci_ers_result_t result;
9092
9093        if (pci_enable_device_mem(pdev)) {
9094                dev_err(&pdev->dev,
9095                        "Cannot re-enable PCI device after reset.\n");
9096                result = PCI_ERS_RESULT_DISCONNECT;
9097        } else {
9098                pci_set_master(pdev);
9099                pci_restore_state(pdev);
9100                pci_save_state(pdev);
9101
9102                pci_enable_wake(pdev, PCI_D3hot, 0);
9103                pci_enable_wake(pdev, PCI_D3cold, 0);
9104
9105                /* In case of PCI error, adapter lose its HW address
9106                 * so we should re-assign it here.
9107                 */
9108                hw->hw_addr = adapter->io_addr;
9109
9110                igb_reset(adapter);
9111                wr32(E1000_WUS, ~0);
9112                result = PCI_ERS_RESULT_RECOVERED;
9113        }
9114
9115        return result;
9116}
9117
9118/**
9119 *  igb_io_resume - called when traffic can start flowing again.
9120 *  @pdev: Pointer to PCI device
9121 *
9122 *  This callback is called when the error recovery driver tells us that
9123 *  its OK to resume normal operation. Implementation resembles the
9124 *  second-half of the igb_resume routine.
9125 */
9126static void igb_io_resume(struct pci_dev *pdev)
9127{
9128        struct net_device *netdev = pci_get_drvdata(pdev);
9129        struct igb_adapter *adapter = netdev_priv(netdev);
9130
9131        if (netif_running(netdev)) {
9132                if (igb_up(adapter)) {
9133                        dev_err(&pdev->dev, "igb_up failed after reset\n");
9134                        return;
9135                }
9136        }
9137
9138        netif_device_attach(netdev);
9139
9140        /* let the f/w know that the h/w is now under the control of the
9141         * driver.
9142         */
9143        igb_get_hw_control(adapter);
9144}
9145
9146/**
9147 *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9148 *  @adapter: Pointer to adapter structure
9149 *  @index: Index of the RAR entry which need to be synced with MAC table
9150 **/
9151static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9152{
9153        struct e1000_hw *hw = &adapter->hw;
9154        u32 rar_low, rar_high;
9155        u8 *addr = adapter->mac_table[index].addr;
9156
9157        /* HW expects these to be in network order when they are plugged
9158         * into the registers which are little endian.  In order to guarantee
9159         * that ordering we need to do an leXX_to_cpup here in order to be
9160         * ready for the byteswap that occurs with writel
9161         */
9162        rar_low = le32_to_cpup((__le32 *)(addr));
9163        rar_high = le16_to_cpup((__le16 *)(addr + 4));
9164
9165        /* Indicate to hardware the Address is Valid. */
9166        if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9167                if (is_valid_ether_addr(addr))
9168                        rar_high |= E1000_RAH_AV;
9169
9170                if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9171                        rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9172
9173                switch (hw->mac.type) {
9174                case e1000_82575:
9175                case e1000_i210:
9176                        if (adapter->mac_table[index].state &
9177                            IGB_MAC_STATE_QUEUE_STEERING)
9178                                rar_high |= E1000_RAH_QSEL_ENABLE;
9179
9180                        rar_high |= E1000_RAH_POOL_1 *
9181                                    adapter->mac_table[index].queue;
9182                        break;
9183                default:
9184                        rar_high |= E1000_RAH_POOL_1 <<
9185                                    adapter->mac_table[index].queue;
9186                        break;
9187                }
9188        }
9189
9190        wr32(E1000_RAL(index), rar_low);
9191        wrfl();
9192        wr32(E1000_RAH(index), rar_high);
9193        wrfl();
9194}
9195
9196static int igb_set_vf_mac(struct igb_adapter *adapter,
9197                          int vf, unsigned char *mac_addr)
9198{
9199        struct e1000_hw *hw = &adapter->hw;
9200        /* VF MAC addresses start at end of receive addresses and moves
9201         * towards the first, as a result a collision should not be possible
9202         */
9203        int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9204        unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9205
9206        ether_addr_copy(vf_mac_addr, mac_addr);
9207        ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9208        adapter->mac_table[rar_entry].queue = vf;
9209        adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9210        igb_rar_set_index(adapter, rar_entry);
9211
9212        return 0;
9213}
9214
9215static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9216{
9217        struct igb_adapter *adapter = netdev_priv(netdev);
9218
9219        if (vf >= adapter->vfs_allocated_count)
9220                return -EINVAL;
9221
9222        /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9223         * flag and allows to overwrite the MAC via VF netdev.  This
9224         * is necessary to allow libvirt a way to restore the original
9225         * MAC after unbinding vfio-pci and reloading igbvf after shutting
9226         * down a VM.
9227         */
9228        if (is_zero_ether_addr(mac)) {
9229                adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9230                dev_info(&adapter->pdev->dev,
9231                         "remove administratively set MAC on VF %d\n",
9232                         vf);
9233        } else if (is_valid_ether_addr(mac)) {
9234                adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9235                dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9236                         mac, vf);
9237                dev_info(&adapter->pdev->dev,
9238                         "Reload the VF driver to make this change effective.");
9239                /* Generate additional warning if PF is down */
9240                if (test_bit(__IGB_DOWN, &adapter->state)) {
9241                        dev_warn(&adapter->pdev->dev,
9242                                 "The VF MAC address has been set, but the PF device is not up.\n");
9243                        dev_warn(&adapter->pdev->dev,
9244                                 "Bring the PF device up before attempting to use the VF device.\n");
9245                }
9246        } else {
9247                return -EINVAL;
9248        }
9249        return igb_set_vf_mac(adapter, vf, mac);
9250}
9251
9252static int igb_link_mbps(int internal_link_speed)
9253{
9254        switch (internal_link_speed) {
9255        case SPEED_100:
9256                return 100;
9257        case SPEED_1000:
9258                return 1000;
9259        default:
9260                return 0;
9261        }
9262}
9263
9264static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9265                                  int link_speed)
9266{
9267        int rf_dec, rf_int;
9268        u32 bcnrc_val;
9269
9270        if (tx_rate != 0) {
9271                /* Calculate the rate factor values to set */
9272                rf_int = link_speed / tx_rate;
9273                rf_dec = (link_speed - (rf_int * tx_rate));
9274                rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9275                         tx_rate;
9276
9277                bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9278                bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
9279                              E1000_RTTBCNRC_RF_INT_MASK);
9280                bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9281        } else {
9282                bcnrc_val = 0;
9283        }
9284
9285        wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9286        /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9287         * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9288         */
9289        wr32(E1000_RTTBCNRM, 0x14);
9290        wr32(E1000_RTTBCNRC, bcnrc_val);
9291}
9292
9293static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9294{
9295        int actual_link_speed, i;
9296        bool reset_rate = false;
9297
9298        /* VF TX rate limit was not set or not supported */
9299        if ((adapter->vf_rate_link_speed == 0) ||
9300            (adapter->hw.mac.type != e1000_82576))
9301                return;
9302
9303        actual_link_speed = igb_link_mbps(adapter->link_speed);
9304        if (actual_link_speed != adapter->vf_rate_link_speed) {
9305                reset_rate = true;
9306                adapter->vf_rate_link_speed = 0;
9307                dev_info(&adapter->pdev->dev,
9308                         "Link speed has been changed. VF Transmit rate is disabled\n");
9309        }
9310
9311        for (i = 0; i < adapter->vfs_allocated_count; i++) {
9312                if (reset_rate)
9313                        adapter->vf_data[i].tx_rate = 0;
9314
9315                igb_set_vf_rate_limit(&adapter->hw, i,
9316                                      adapter->vf_data[i].tx_rate,
9317                                      actual_link_speed);
9318        }
9319}
9320
9321static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9322                             int min_tx_rate, int max_tx_rate)
9323{
9324        struct igb_adapter *adapter = netdev_priv(netdev);
9325        struct e1000_hw *hw = &adapter->hw;
9326        int actual_link_speed;
9327
9328        if (hw->mac.type != e1000_82576)
9329                return -EOPNOTSUPP;
9330
9331        if (min_tx_rate)
9332                return -EINVAL;
9333
9334        actual_link_speed = igb_link_mbps(adapter->link_speed);
9335        if ((vf >= adapter->vfs_allocated_count) ||
9336            (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9337            (max_tx_rate < 0) ||
9338            (max_tx_rate > actual_link_speed))
9339                return -EINVAL;
9340
9341        adapter->vf_rate_link_speed = actual_link_speed;
9342        adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9343        igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9344
9345        return 0;
9346}
9347
9348static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9349                                   bool setting)
9350{
9351        struct igb_adapter *adapter = netdev_priv(netdev);
9352        struct e1000_hw *hw = &adapter->hw;
9353        u32 reg_val, reg_offset;
9354
9355        if (!adapter->vfs_allocated_count)
9356                return -EOPNOTSUPP;
9357
9358        if (vf >= adapter->vfs_allocated_count)
9359                return -EINVAL;
9360
9361        reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9362        reg_val = rd32(reg_offset);
9363        if (setting)
9364                reg_val |= (BIT(vf) |
9365                            BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9366        else
9367                reg_val &= ~(BIT(vf) |
9368                             BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9369        wr32(reg_offset, reg_val);
9370
9371        adapter->vf_data[vf].spoofchk_enabled = setting;
9372        return 0;
9373}
9374
9375static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9376{
9377        struct igb_adapter *adapter = netdev_priv(netdev);
9378
9379        if (vf >= adapter->vfs_allocated_count)
9380                return -EINVAL;
9381        if (adapter->vf_data[vf].trusted == setting)
9382                return 0;
9383
9384        adapter->vf_data[vf].trusted = setting;
9385
9386        dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9387                 vf, setting ? "" : "not ");
9388        return 0;
9389}
9390
9391static int igb_ndo_get_vf_config(struct net_device *netdev,
9392                                 int vf, struct ifla_vf_info *ivi)
9393{
9394        struct igb_adapter *adapter = netdev_priv(netdev);
9395        if (vf >= adapter->vfs_allocated_count)
9396                return -EINVAL;
9397        ivi->vf = vf;
9398        memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9399        ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9400        ivi->min_tx_rate = 0;
9401        ivi->vlan = adapter->vf_data[vf].pf_vlan;
9402        ivi->qos = adapter->vf_data[vf].pf_qos;
9403        ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9404        ivi->trusted = adapter->vf_data[vf].trusted;
9405        return 0;
9406}
9407
9408static void igb_vmm_control(struct igb_adapter *adapter)
9409{
9410        struct e1000_hw *hw = &adapter->hw;
9411        u32 reg;
9412
9413        switch (hw->mac.type) {
9414        case e1000_82575:
9415        case e1000_i210:
9416        case e1000_i211:
9417        case e1000_i354:
9418        default:
9419                /* replication is not supported for 82575 */
9420                return;
9421        case e1000_82576:
9422                /* notify HW that the MAC is adding vlan tags */
9423                reg = rd32(E1000_DTXCTL);
9424                reg |= E1000_DTXCTL_VLAN_ADDED;
9425                wr32(E1000_DTXCTL, reg);
9426                /* Fall through */
9427        case e1000_82580:
9428                /* enable replication vlan tag stripping */
9429                reg = rd32(E1000_RPLOLR);
9430                reg |= E1000_RPLOLR_STRVLAN;
9431                wr32(E1000_RPLOLR, reg);
9432                /* Fall through */
9433        case e1000_i350:
9434                /* none of the above registers are supported by i350 */
9435                break;
9436        }
9437
9438        if (adapter->vfs_allocated_count) {
9439                igb_vmdq_set_loopback_pf(hw, true);
9440                igb_vmdq_set_replication_pf(hw, true);
9441                igb_vmdq_set_anti_spoofing_pf(hw, true,
9442                                              adapter->vfs_allocated_count);
9443        } else {
9444                igb_vmdq_set_loopback_pf(hw, false);
9445                igb_vmdq_set_replication_pf(hw, false);
9446        }
9447}
9448
9449static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9450{
9451        struct e1000_hw *hw = &adapter->hw;
9452        u32 dmac_thr;
9453        u16 hwm;
9454
9455        if (hw->mac.type > e1000_82580) {
9456                if (adapter->flags & IGB_FLAG_DMAC) {
9457                        u32 reg;
9458
9459                        /* force threshold to 0. */
9460                        wr32(E1000_DMCTXTH, 0);
9461
9462                        /* DMA Coalescing high water mark needs to be greater
9463                         * than the Rx threshold. Set hwm to PBA - max frame
9464                         * size in 16B units, capping it at PBA - 6KB.
9465                         */
9466                        hwm = 64 * (pba - 6);
9467                        reg = rd32(E1000_FCRTC);
9468                        reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9469                        reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9470                                & E1000_FCRTC_RTH_COAL_MASK);
9471                        wr32(E1000_FCRTC, reg);
9472
9473                        /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9474                         * frame size, capping it at PBA - 10KB.
9475                         */
9476                        dmac_thr = pba - 10;
9477                        reg = rd32(E1000_DMACR);
9478                        reg &= ~E1000_DMACR_DMACTHR_MASK;
9479                        reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9480                                & E1000_DMACR_DMACTHR_MASK);
9481
9482                        /* transition to L0x or L1 if available..*/
9483                        reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
9484
9485                        /* watchdog timer= +-1000 usec in 32usec intervals */
9486                        reg |= (1000 >> 5);
9487
9488                        /* Disable BMC-to-OS Watchdog Enable */
9489                        if (hw->mac.type != e1000_i354)
9490                                reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
9491
9492                        wr32(E1000_DMACR, reg);
9493
9494                        /* no lower threshold to disable
9495                         * coalescing(smart fifb)-UTRESH=0
9496                         */
9497                        wr32(E1000_DMCRTRH, 0);
9498
9499                        reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
9500
9501                        wr32(E1000_DMCTLX, reg);
9502
9503                        /* free space in tx packet buffer to wake from
9504                         * DMA coal
9505                         */
9506                        wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
9507                             (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
9508
9509                        /* make low power state decision controlled
9510                         * by DMA coal
9511                         */
9512                        reg = rd32(E1000_PCIEMISC);
9513                        reg &= ~E1000_PCIEMISC_LX_DECISION;
9514                        wr32(E1000_PCIEMISC, reg);
9515                } /* endif adapter->dmac is not disabled */
9516        } else if (hw->mac.type == e1000_82580) {
9517                u32 reg = rd32(E1000_PCIEMISC);
9518
9519                wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
9520                wr32(E1000_DMACR, 0);
9521        }
9522}
9523
9524/**
9525 *  igb_read_i2c_byte - Reads 8 bit word over I2C
9526 *  @hw: pointer to hardware structure
9527 *  @byte_offset: byte offset to read
9528 *  @dev_addr: device address
9529 *  @data: value read
9530 *
9531 *  Performs byte read operation over I2C interface at
9532 *  a specified device address.
9533 **/
9534s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9535                      u8 dev_addr, u8 *data)
9536{
9537        struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9538        struct i2c_client *this_client = adapter->i2c_client;
9539        s32 status;
9540        u16 swfw_mask = 0;
9541
9542        if (!this_client)
9543                return E1000_ERR_I2C;
9544
9545        swfw_mask = E1000_SWFW_PHY0_SM;
9546
9547        if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9548                return E1000_ERR_SWFW_SYNC;
9549
9550        status = i2c_smbus_read_byte_data(this_client, byte_offset);
9551        hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9552
9553        if (status < 0)
9554                return E1000_ERR_I2C;
9555        else {
9556                *data = status;
9557                return 0;
9558        }
9559}
9560
9561/**
9562 *  igb_write_i2c_byte - Writes 8 bit word over I2C
9563 *  @hw: pointer to hardware structure
9564 *  @byte_offset: byte offset to write
9565 *  @dev_addr: device address
9566 *  @data: value to write
9567 *
9568 *  Performs byte write operation over I2C interface at
9569 *  a specified device address.
9570 **/
9571s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9572                       u8 dev_addr, u8 data)
9573{
9574        struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9575        struct i2c_client *this_client = adapter->i2c_client;
9576        s32 status;
9577        u16 swfw_mask = E1000_SWFW_PHY0_SM;
9578
9579        if (!this_client)
9580                return E1000_ERR_I2C;
9581
9582        if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9583                return E1000_ERR_SWFW_SYNC;
9584        status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
9585        hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9586
9587        if (status)
9588                return E1000_ERR_I2C;
9589        else
9590                return 0;
9591
9592}
9593
9594int igb_reinit_queues(struct igb_adapter *adapter)
9595{
9596        struct net_device *netdev = adapter->netdev;
9597        struct pci_dev *pdev = adapter->pdev;
9598        int err = 0;
9599
9600        if (netif_running(netdev))
9601                igb_close(netdev);
9602
9603        igb_reset_interrupt_capability(adapter);
9604
9605        if (igb_init_interrupt_scheme(adapter, true)) {
9606                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9607                return -ENOMEM;
9608        }
9609
9610        if (netif_running(netdev))
9611                err = igb_open(netdev);
9612
9613        return err;
9614}
9615
9616static void igb_nfc_filter_exit(struct igb_adapter *adapter)
9617{
9618        struct igb_nfc_filter *rule;
9619
9620        spin_lock(&adapter->nfc_lock);
9621
9622        hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9623                igb_erase_filter(adapter, rule);
9624
9625        hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
9626                igb_erase_filter(adapter, rule);
9627
9628        spin_unlock(&adapter->nfc_lock);
9629}
9630
9631static void igb_nfc_filter_restore(struct igb_adapter *adapter)
9632{
9633        struct igb_nfc_filter *rule;
9634
9635        spin_lock(&adapter->nfc_lock);
9636
9637        hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9638                igb_add_filter(adapter, rule);
9639
9640        spin_unlock(&adapter->nfc_lock);
9641}
9642/* igb_main.c */
9643