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