linux/drivers/net/ethernet/intel/igb/igb_main.c
<<
>>
Prefs
   1/*******************************************************************************
   2
   3  Intel(R) Gigabit Ethernet Linux driver
   4  Copyright(c) 2007-2013 Intel Corporation.
   5
   6  This program is free software; you can redistribute it and/or modify it
   7  under the terms and conditions of the GNU General Public License,
   8  version 2, as published by the Free Software Foundation.
   9
  10  This program is distributed in the hope it will be useful, but WITHOUT
  11  ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12  FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13  more details.
  14
  15  You should have received a copy of the GNU General Public License along with
  16  this program; if not, write to the Free Software Foundation, Inc.,
  17  51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19  The full GNU General Public License is included in this distribution in
  20  the file called "COPYING".
  21
  22  Contact Information:
  23  e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24  Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26*******************************************************************************/
  27
  28#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  29
  30#include <linux/module.h>
  31#include <linux/types.h>
  32#include <linux/init.h>
  33#include <linux/bitops.h>
  34#include <linux/vmalloc.h>
  35#include <linux/pagemap.h>
  36#include <linux/netdevice.h>
  37#include <linux/ipv6.h>
  38#include <linux/slab.h>
  39#include <net/checksum.h>
  40#include <net/ip6_checksum.h>
  41#include <linux/net_tstamp.h>
  42#include <linux/mii.h>
  43#include <linux/ethtool.h>
  44#include <linux/if.h>
  45#include <linux/if_vlan.h>
  46#include <linux/pci.h>
  47#include <linux/pci-aspm.h>
  48#include <linux/delay.h>
  49#include <linux/interrupt.h>
  50#include <linux/ip.h>
  51#include <linux/tcp.h>
  52#include <linux/sctp.h>
  53#include <linux/if_ether.h>
  54#include <linux/aer.h>
  55#include <linux/prefetch.h>
  56#include <linux/pm_runtime.h>
  57#ifdef CONFIG_IGB_DCA
  58#include <linux/dca.h>
  59#endif
  60#include <linux/i2c.h>
  61#include "igb.h"
  62
  63#define MAJ 5
  64#define MIN 0
  65#define BUILD 5
  66#define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
  67__stringify(BUILD) "-k"
  68char igb_driver_name[] = "igb";
  69char igb_driver_version[] = DRV_VERSION;
  70static const char igb_driver_string[] =
  71                                "Intel(R) Gigabit Ethernet Network Driver";
  72static const char igb_copyright[] =
  73                                "Copyright (c) 2007-2013 Intel Corporation.";
  74
  75static const struct e1000_info *igb_info_tbl[] = {
  76        [board_82575] = &e1000_82575_info,
  77};
  78
  79static DEFINE_PCI_DEVICE_TABLE(igb_pci_tbl) = {
  80        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
  81        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
  82        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
  83        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
  84        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
  85        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
  86        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
  87        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
  88        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
  89        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
  90        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
  91        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
  92        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
  93        { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
  94        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
  95        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
  96        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
  97        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
  98        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
  99        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
 100        { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
 101        { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
 102        { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
 103        { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
 104        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
 105        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
 106        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
 107        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
 108        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
 109        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
 110        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
 111        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
 112        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
 113        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
 114        { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
 115        /* required last entry */
 116        {0, }
 117};
 118
 119MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 120
 121void igb_reset(struct igb_adapter *);
 122static int igb_setup_all_tx_resources(struct igb_adapter *);
 123static int igb_setup_all_rx_resources(struct igb_adapter *);
 124static void igb_free_all_tx_resources(struct igb_adapter *);
 125static void igb_free_all_rx_resources(struct igb_adapter *);
 126static void igb_setup_mrqc(struct igb_adapter *);
 127static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 128static void igb_remove(struct pci_dev *pdev);
 129static int igb_sw_init(struct igb_adapter *);
 130static int igb_open(struct net_device *);
 131static int igb_close(struct net_device *);
 132static void igb_configure(struct igb_adapter *);
 133static void igb_configure_tx(struct igb_adapter *);
 134static void igb_configure_rx(struct igb_adapter *);
 135static void igb_clean_all_tx_rings(struct igb_adapter *);
 136static void igb_clean_all_rx_rings(struct igb_adapter *);
 137static void igb_clean_tx_ring(struct igb_ring *);
 138static void igb_clean_rx_ring(struct igb_ring *);
 139static void igb_set_rx_mode(struct net_device *);
 140static void igb_update_phy_info(unsigned long);
 141static void igb_watchdog(unsigned long);
 142static void igb_watchdog_task(struct work_struct *);
 143static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 144static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *dev,
 145                                                 struct rtnl_link_stats64 *stats);
 146static int igb_change_mtu(struct net_device *, int);
 147static int igb_set_mac(struct net_device *, void *);
 148static void igb_set_uta(struct igb_adapter *adapter);
 149static irqreturn_t igb_intr(int irq, void *);
 150static irqreturn_t igb_intr_msi(int irq, void *);
 151static irqreturn_t igb_msix_other(int irq, void *);
 152static irqreturn_t igb_msix_ring(int irq, void *);
 153#ifdef CONFIG_IGB_DCA
 154static void igb_update_dca(struct igb_q_vector *);
 155static void igb_setup_dca(struct igb_adapter *);
 156#endif /* CONFIG_IGB_DCA */
 157static int igb_poll(struct napi_struct *, int);
 158static bool igb_clean_tx_irq(struct igb_q_vector *);
 159static bool igb_clean_rx_irq(struct igb_q_vector *, int);
 160static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 161static void igb_tx_timeout(struct net_device *);
 162static void igb_reset_task(struct work_struct *);
 163static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features);
 164static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
 165static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
 166static void igb_restore_vlan(struct igb_adapter *);
 167static void igb_rar_set_qsel(struct igb_adapter *, u8 *, u32 , u8);
 168static void igb_ping_all_vfs(struct igb_adapter *);
 169static void igb_msg_task(struct igb_adapter *);
 170static void igb_vmm_control(struct igb_adapter *);
 171static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 172static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 173static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 174static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 175                               int vf, u16 vlan, u8 qos);
 176static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate);
 177static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 178                                   bool setting);
 179static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 180                                 struct ifla_vf_info *ivi);
 181static void igb_check_vf_rate_limit(struct igb_adapter *);
 182
 183#ifdef CONFIG_PCI_IOV
 184static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 185#endif
 186
 187#ifdef CONFIG_PM
 188#ifdef CONFIG_PM_SLEEP
 189static int igb_suspend(struct device *);
 190#endif
 191static int igb_resume(struct device *);
 192#ifdef CONFIG_PM_RUNTIME
 193static int igb_runtime_suspend(struct device *dev);
 194static int igb_runtime_resume(struct device *dev);
 195static int igb_runtime_idle(struct device *dev);
 196#endif
 197static const struct dev_pm_ops igb_pm_ops = {
 198        SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
 199        SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
 200                        igb_runtime_idle)
 201};
 202#endif
 203static void igb_shutdown(struct pci_dev *);
 204static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 205#ifdef CONFIG_IGB_DCA
 206static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 207static struct notifier_block dca_notifier = {
 208        .notifier_call  = igb_notify_dca,
 209        .next           = NULL,
 210        .priority       = 0
 211};
 212#endif
 213#ifdef CONFIG_NET_POLL_CONTROLLER
 214/* for netdump / net console */
 215static void igb_netpoll(struct net_device *);
 216#endif
 217#ifdef CONFIG_PCI_IOV
 218static unsigned int max_vfs = 0;
 219module_param(max_vfs, uint, 0);
 220MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate "
 221                 "per physical function");
 222#endif /* CONFIG_PCI_IOV */
 223
 224static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 225                     pci_channel_state_t);
 226static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 227static void igb_io_resume(struct pci_dev *);
 228
 229static const struct pci_error_handlers igb_err_handler = {
 230        .error_detected = igb_io_error_detected,
 231        .slot_reset = igb_io_slot_reset,
 232        .resume = igb_io_resume,
 233};
 234
 235static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 236
 237static struct pci_driver igb_driver = {
 238        .name     = igb_driver_name,
 239        .id_table = igb_pci_tbl,
 240        .probe    = igb_probe,
 241        .remove   = igb_remove,
 242#ifdef CONFIG_PM
 243        .driver.pm = &igb_pm_ops,
 244#endif
 245        .shutdown = igb_shutdown,
 246        .sriov_configure = igb_pci_sriov_configure,
 247        .err_handler = &igb_err_handler
 248};
 249
 250MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 251MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 252MODULE_LICENSE("GPL");
 253MODULE_VERSION(DRV_VERSION);
 254
 255#define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 256static int debug = -1;
 257module_param(debug, int, 0);
 258MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 259
 260struct igb_reg_info {
 261        u32 ofs;
 262        char *name;
 263};
 264
 265static const struct igb_reg_info igb_reg_info_tbl[] = {
 266
 267        /* General Registers */
 268        {E1000_CTRL, "CTRL"},
 269        {E1000_STATUS, "STATUS"},
 270        {E1000_CTRL_EXT, "CTRL_EXT"},
 271
 272        /* Interrupt Registers */
 273        {E1000_ICR, "ICR"},
 274
 275        /* RX Registers */
 276        {E1000_RCTL, "RCTL"},
 277        {E1000_RDLEN(0), "RDLEN"},
 278        {E1000_RDH(0), "RDH"},
 279        {E1000_RDT(0), "RDT"},
 280        {E1000_RXDCTL(0), "RXDCTL"},
 281        {E1000_RDBAL(0), "RDBAL"},
 282        {E1000_RDBAH(0), "RDBAH"},
 283
 284        /* TX Registers */
 285        {E1000_TCTL, "TCTL"},
 286        {E1000_TDBAL(0), "TDBAL"},
 287        {E1000_TDBAH(0), "TDBAH"},
 288        {E1000_TDLEN(0), "TDLEN"},
 289        {E1000_TDH(0), "TDH"},
 290        {E1000_TDT(0), "TDT"},
 291        {E1000_TXDCTL(0), "TXDCTL"},
 292        {E1000_TDFH, "TDFH"},
 293        {E1000_TDFT, "TDFT"},
 294        {E1000_TDFHS, "TDFHS"},
 295        {E1000_TDFPC, "TDFPC"},
 296
 297        /* List Terminator */
 298        {}
 299};
 300
 301/* igb_regdump - register printout routine */
 302static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
 303{
 304        int n = 0;
 305        char rname[16];
 306        u32 regs[8];
 307
 308        switch (reginfo->ofs) {
 309        case E1000_RDLEN(0):
 310                for (n = 0; n < 4; n++)
 311                        regs[n] = rd32(E1000_RDLEN(n));
 312                break;
 313        case E1000_RDH(0):
 314                for (n = 0; n < 4; n++)
 315                        regs[n] = rd32(E1000_RDH(n));
 316                break;
 317        case E1000_RDT(0):
 318                for (n = 0; n < 4; n++)
 319                        regs[n] = rd32(E1000_RDT(n));
 320                break;
 321        case E1000_RXDCTL(0):
 322                for (n = 0; n < 4; n++)
 323                        regs[n] = rd32(E1000_RXDCTL(n));
 324                break;
 325        case E1000_RDBAL(0):
 326                for (n = 0; n < 4; n++)
 327                        regs[n] = rd32(E1000_RDBAL(n));
 328                break;
 329        case E1000_RDBAH(0):
 330                for (n = 0; n < 4; n++)
 331                        regs[n] = rd32(E1000_RDBAH(n));
 332                break;
 333        case E1000_TDBAL(0):
 334                for (n = 0; n < 4; n++)
 335                        regs[n] = rd32(E1000_RDBAL(n));
 336                break;
 337        case E1000_TDBAH(0):
 338                for (n = 0; n < 4; n++)
 339                        regs[n] = rd32(E1000_TDBAH(n));
 340                break;
 341        case E1000_TDLEN(0):
 342                for (n = 0; n < 4; n++)
 343                        regs[n] = rd32(E1000_TDLEN(n));
 344                break;
 345        case E1000_TDH(0):
 346                for (n = 0; n < 4; n++)
 347                        regs[n] = rd32(E1000_TDH(n));
 348                break;
 349        case E1000_TDT(0):
 350                for (n = 0; n < 4; n++)
 351                        regs[n] = rd32(E1000_TDT(n));
 352                break;
 353        case E1000_TXDCTL(0):
 354                for (n = 0; n < 4; n++)
 355                        regs[n] = rd32(E1000_TXDCTL(n));
 356                break;
 357        default:
 358                pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
 359                return;
 360        }
 361
 362        snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
 363        pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
 364                regs[2], regs[3]);
 365}
 366
 367/* igb_dump - Print registers, Tx-rings and Rx-rings */
 368static void igb_dump(struct igb_adapter *adapter)
 369{
 370        struct net_device *netdev = adapter->netdev;
 371        struct e1000_hw *hw = &adapter->hw;
 372        struct igb_reg_info *reginfo;
 373        struct igb_ring *tx_ring;
 374        union e1000_adv_tx_desc *tx_desc;
 375        struct my_u0 { u64 a; u64 b; } *u0;
 376        struct igb_ring *rx_ring;
 377        union e1000_adv_rx_desc *rx_desc;
 378        u32 staterr;
 379        u16 i, n;
 380
 381        if (!netif_msg_hw(adapter))
 382                return;
 383
 384        /* Print netdevice Info */
 385        if (netdev) {
 386                dev_info(&adapter->pdev->dev, "Net device Info\n");
 387                pr_info("Device Name     state            trans_start      "
 388                        "last_rx\n");
 389                pr_info("%-15s %016lX %016lX %016lX\n", netdev->name,
 390                        netdev->state, netdev->trans_start, netdev->last_rx);
 391        }
 392
 393        /* Print Registers */
 394        dev_info(&adapter->pdev->dev, "Register Dump\n");
 395        pr_info(" Register Name   Value\n");
 396        for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
 397             reginfo->name; reginfo++) {
 398                igb_regdump(hw, reginfo);
 399        }
 400
 401        /* Print TX Ring Summary */
 402        if (!netdev || !netif_running(netdev))
 403                goto exit;
 404
 405        dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 406        pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 407        for (n = 0; n < adapter->num_tx_queues; n++) {
 408                struct igb_tx_buffer *buffer_info;
 409                tx_ring = adapter->tx_ring[n];
 410                buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 411                pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
 412                        n, tx_ring->next_to_use, tx_ring->next_to_clean,
 413                        (u64)dma_unmap_addr(buffer_info, dma),
 414                        dma_unmap_len(buffer_info, len),
 415                        buffer_info->next_to_watch,
 416                        (u64)buffer_info->time_stamp);
 417        }
 418
 419        /* Print TX Rings */
 420        if (!netif_msg_tx_done(adapter))
 421                goto rx_ring_summary;
 422
 423        dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 424
 425        /* Transmit Descriptor Formats
 426         *
 427         * Advanced Transmit Descriptor
 428         *   +--------------------------------------------------------------+
 429         * 0 |         Buffer Address [63:0]                                |
 430         *   +--------------------------------------------------------------+
 431         * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
 432         *   +--------------------------------------------------------------+
 433         *   63      46 45    40 39 38 36 35 32 31   24             15       0
 434         */
 435
 436        for (n = 0; n < adapter->num_tx_queues; n++) {
 437                tx_ring = adapter->tx_ring[n];
 438                pr_info("------------------------------------\n");
 439                pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 440                pr_info("------------------------------------\n");
 441                pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] "
 442                        "[bi->dma       ] leng  ntw timestamp        "
 443                        "bi->skb\n");
 444
 445                for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 446                        const char *next_desc;
 447                        struct igb_tx_buffer *buffer_info;
 448                        tx_desc = IGB_TX_DESC(tx_ring, i);
 449                        buffer_info = &tx_ring->tx_buffer_info[i];
 450                        u0 = (struct my_u0 *)tx_desc;
 451                        if (i == tx_ring->next_to_use &&
 452                            i == tx_ring->next_to_clean)
 453                                next_desc = " NTC/U";
 454                        else if (i == tx_ring->next_to_use)
 455                                next_desc = " NTU";
 456                        else if (i == tx_ring->next_to_clean)
 457                                next_desc = " NTC";
 458                        else
 459                                next_desc = "";
 460
 461                        pr_info("T [0x%03X]    %016llX %016llX %016llX"
 462                                " %04X  %p %016llX %p%s\n", i,
 463                                le64_to_cpu(u0->a),
 464                                le64_to_cpu(u0->b),
 465                                (u64)dma_unmap_addr(buffer_info, dma),
 466                                dma_unmap_len(buffer_info, len),
 467                                buffer_info->next_to_watch,
 468                                (u64)buffer_info->time_stamp,
 469                                buffer_info->skb, next_desc);
 470
 471                        if (netif_msg_pktdata(adapter) && buffer_info->skb)
 472                                print_hex_dump(KERN_INFO, "",
 473                                        DUMP_PREFIX_ADDRESS,
 474                                        16, 1, buffer_info->skb->data,
 475                                        dma_unmap_len(buffer_info, len),
 476                                        true);
 477                }
 478        }
 479
 480        /* Print RX Rings Summary */
 481rx_ring_summary:
 482        dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 483        pr_info("Queue [NTU] [NTC]\n");
 484        for (n = 0; n < adapter->num_rx_queues; n++) {
 485                rx_ring = adapter->rx_ring[n];
 486                pr_info(" %5d %5X %5X\n",
 487                        n, rx_ring->next_to_use, rx_ring->next_to_clean);
 488        }
 489
 490        /* Print RX Rings */
 491        if (!netif_msg_rx_status(adapter))
 492                goto exit;
 493
 494        dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 495
 496        /* Advanced Receive Descriptor (Read) Format
 497         *    63                                           1        0
 498         *    +-----------------------------------------------------+
 499         *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 500         *    +----------------------------------------------+------+
 501         *  8 |       Header Buffer Address [63:1]           |  DD  |
 502         *    +-----------------------------------------------------+
 503         *
 504         *
 505         * Advanced Receive Descriptor (Write-Back) Format
 506         *
 507         *   63       48 47    32 31  30      21 20 17 16   4 3     0
 508         *   +------------------------------------------------------+
 509         * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 510         *   | Checksum   Ident  |   |           |    | Type | Type |
 511         *   +------------------------------------------------------+
 512         * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 513         *   +------------------------------------------------------+
 514         *   63       48 47    32 31            20 19               0
 515         */
 516
 517        for (n = 0; n < adapter->num_rx_queues; n++) {
 518                rx_ring = adapter->rx_ring[n];
 519                pr_info("------------------------------------\n");
 520                pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 521                pr_info("------------------------------------\n");
 522                pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] "
 523                        "[bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
 524                pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] -----"
 525                        "----------- [bi->skb] <-- Adv Rx Write-Back format\n");
 526
 527                for (i = 0; i < rx_ring->count; i++) {
 528                        const char *next_desc;
 529                        struct igb_rx_buffer *buffer_info;
 530                        buffer_info = &rx_ring->rx_buffer_info[i];
 531                        rx_desc = IGB_RX_DESC(rx_ring, i);
 532                        u0 = (struct my_u0 *)rx_desc;
 533                        staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 534
 535                        if (i == rx_ring->next_to_use)
 536                                next_desc = " NTU";
 537                        else if (i == rx_ring->next_to_clean)
 538                                next_desc = " NTC";
 539                        else
 540                                next_desc = "";
 541
 542                        if (staterr & E1000_RXD_STAT_DD) {
 543                                /* Descriptor Done */
 544                                pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
 545                                        "RWB", i,
 546                                        le64_to_cpu(u0->a),
 547                                        le64_to_cpu(u0->b),
 548                                        next_desc);
 549                        } else {
 550                                pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
 551                                        "R  ", i,
 552                                        le64_to_cpu(u0->a),
 553                                        le64_to_cpu(u0->b),
 554                                        (u64)buffer_info->dma,
 555                                        next_desc);
 556
 557                                if (netif_msg_pktdata(adapter) &&
 558                                    buffer_info->dma && buffer_info->page) {
 559                                        print_hex_dump(KERN_INFO, "",
 560                                          DUMP_PREFIX_ADDRESS,
 561                                          16, 1,
 562                                          page_address(buffer_info->page) +
 563                                                      buffer_info->page_offset,
 564                                          IGB_RX_BUFSZ, true);
 565                                }
 566                        }
 567                }
 568        }
 569
 570exit:
 571        return;
 572}
 573
 574/**
 575 *  igb_get_i2c_data - Reads the I2C SDA data bit
 576 *  @hw: pointer to hardware structure
 577 *  @i2cctl: Current value of I2CCTL register
 578 *
 579 *  Returns the I2C data bit value
 580 **/
 581static int igb_get_i2c_data(void *data)
 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        return ((i2cctl & E1000_I2C_DATA_IN) != 0);
 588}
 589
 590/**
 591 *  igb_set_i2c_data - Sets the I2C data bit
 592 *  @data: pointer to hardware structure
 593 *  @state: I2C data value (0 or 1) to set
 594 *
 595 *  Sets the I2C data bit
 596 **/
 597static void igb_set_i2c_data(void *data, int state)
 598{
 599        struct igb_adapter *adapter = (struct igb_adapter *)data;
 600        struct e1000_hw *hw = &adapter->hw;
 601        s32 i2cctl = rd32(E1000_I2CPARAMS);
 602
 603        if (state)
 604                i2cctl |= E1000_I2C_DATA_OUT;
 605        else
 606                i2cctl &= ~E1000_I2C_DATA_OUT;
 607
 608        i2cctl &= ~E1000_I2C_DATA_OE_N;
 609        i2cctl |= E1000_I2C_CLK_OE_N;
 610        wr32(E1000_I2CPARAMS, i2cctl);
 611        wrfl();
 612
 613}
 614
 615/**
 616 *  igb_set_i2c_clk - Sets the I2C SCL clock
 617 *  @data: pointer to hardware structure
 618 *  @state: state to set clock
 619 *
 620 *  Sets the I2C clock line to state
 621 **/
 622static void igb_set_i2c_clk(void *data, int state)
 623{
 624        struct igb_adapter *adapter = (struct igb_adapter *)data;
 625        struct e1000_hw *hw = &adapter->hw;
 626        s32 i2cctl = rd32(E1000_I2CPARAMS);
 627
 628        if (state) {
 629                i2cctl |= E1000_I2C_CLK_OUT;
 630                i2cctl &= ~E1000_I2C_CLK_OE_N;
 631        } else {
 632                i2cctl &= ~E1000_I2C_CLK_OUT;
 633                i2cctl &= ~E1000_I2C_CLK_OE_N;
 634        }
 635        wr32(E1000_I2CPARAMS, i2cctl);
 636        wrfl();
 637}
 638
 639/**
 640 *  igb_get_i2c_clk - Gets the I2C SCL clock state
 641 *  @data: pointer to hardware structure
 642 *
 643 *  Gets the I2C clock state
 644 **/
 645static int igb_get_i2c_clk(void *data)
 646{
 647        struct igb_adapter *adapter = (struct igb_adapter *)data;
 648        struct e1000_hw *hw = &adapter->hw;
 649        s32 i2cctl = rd32(E1000_I2CPARAMS);
 650
 651        return ((i2cctl & E1000_I2C_CLK_IN) != 0);
 652}
 653
 654static const struct i2c_algo_bit_data igb_i2c_algo = {
 655        .setsda         = igb_set_i2c_data,
 656        .setscl         = igb_set_i2c_clk,
 657        .getsda         = igb_get_i2c_data,
 658        .getscl         = igb_get_i2c_clk,
 659        .udelay         = 5,
 660        .timeout        = 20,
 661};
 662
 663/**
 664 *  igb_get_hw_dev - return device
 665 *  @hw: pointer to hardware structure
 666 *
 667 *  used by hardware layer to print debugging information
 668 **/
 669struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
 670{
 671        struct igb_adapter *adapter = hw->back;
 672        return adapter->netdev;
 673}
 674
 675/**
 676 *  igb_init_module - Driver Registration Routine
 677 *
 678 *  igb_init_module is the first routine called when the driver is
 679 *  loaded. All it does is register with the PCI subsystem.
 680 **/
 681static int __init igb_init_module(void)
 682{
 683        int ret;
 684        pr_info("%s - version %s\n",
 685               igb_driver_string, igb_driver_version);
 686
 687        pr_info("%s\n", igb_copyright);
 688
 689#ifdef CONFIG_IGB_DCA
 690        dca_register_notify(&dca_notifier);
 691#endif
 692        ret = pci_register_driver(&igb_driver);
 693        return ret;
 694}
 695
 696module_init(igb_init_module);
 697
 698/**
 699 *  igb_exit_module - Driver Exit Cleanup Routine
 700 *
 701 *  igb_exit_module is called just before the driver is removed
 702 *  from memory.
 703 **/
 704static void __exit igb_exit_module(void)
 705{
 706#ifdef CONFIG_IGB_DCA
 707        dca_unregister_notify(&dca_notifier);
 708#endif
 709        pci_unregister_driver(&igb_driver);
 710}
 711
 712module_exit(igb_exit_module);
 713
 714#define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 715/**
 716 *  igb_cache_ring_register - Descriptor ring to register mapping
 717 *  @adapter: board private structure to initialize
 718 *
 719 *  Once we know the feature-set enabled for the device, we'll cache
 720 *  the register offset the descriptor ring is assigned to.
 721 **/
 722static void igb_cache_ring_register(struct igb_adapter *adapter)
 723{
 724        int i = 0, j = 0;
 725        u32 rbase_offset = adapter->vfs_allocated_count;
 726
 727        switch (adapter->hw.mac.type) {
 728        case e1000_82576:
 729                /* The queues are allocated for virtualization such that VF 0
 730                 * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 731                 * In order to avoid collision we start at the first free queue
 732                 * and continue consuming queues in the same sequence
 733                 */
 734                if (adapter->vfs_allocated_count) {
 735                        for (; i < adapter->rss_queues; i++)
 736                                adapter->rx_ring[i]->reg_idx = rbase_offset +
 737                                                               Q_IDX_82576(i);
 738                }
 739        case e1000_82575:
 740        case e1000_82580:
 741        case e1000_i350:
 742        case e1000_i354:
 743        case e1000_i210:
 744        case e1000_i211:
 745        default:
 746                for (; i < adapter->num_rx_queues; i++)
 747                        adapter->rx_ring[i]->reg_idx = rbase_offset + i;
 748                for (; j < adapter->num_tx_queues; j++)
 749                        adapter->tx_ring[j]->reg_idx = rbase_offset + j;
 750                break;
 751        }
 752}
 753
 754/**
 755 *  igb_write_ivar - configure ivar for given MSI-X vector
 756 *  @hw: pointer to the HW structure
 757 *  @msix_vector: vector number we are allocating to a given ring
 758 *  @index: row index of IVAR register to write within IVAR table
 759 *  @offset: column offset of in IVAR, should be multiple of 8
 760 *
 761 *  This function is intended to handle the writing of the IVAR register
 762 *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 763 *  each containing an cause allocation for an Rx and Tx ring, and a
 764 *  variable number of rows depending on the number of queues supported.
 765 **/
 766static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 767                           int index, int offset)
 768{
 769        u32 ivar = array_rd32(E1000_IVAR0, index);
 770
 771        /* clear any bits that are currently set */
 772        ivar &= ~((u32)0xFF << offset);
 773
 774        /* write vector and valid bit */
 775        ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 776
 777        array_wr32(E1000_IVAR0, index, ivar);
 778}
 779
 780#define IGB_N0_QUEUE -1
 781static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 782{
 783        struct igb_adapter *adapter = q_vector->adapter;
 784        struct e1000_hw *hw = &adapter->hw;
 785        int rx_queue = IGB_N0_QUEUE;
 786        int tx_queue = IGB_N0_QUEUE;
 787        u32 msixbm = 0;
 788
 789        if (q_vector->rx.ring)
 790                rx_queue = q_vector->rx.ring->reg_idx;
 791        if (q_vector->tx.ring)
 792                tx_queue = q_vector->tx.ring->reg_idx;
 793
 794        switch (hw->mac.type) {
 795        case e1000_82575:
 796                /* The 82575 assigns vectors using a bitmask, which matches the
 797                 * bitmask for the EICR/EIMS/EIMC registers.  To assign one
 798                 * or more queues to a vector, we write the appropriate bits
 799                 * into the MSIXBM register for that vector.
 800                 */
 801                if (rx_queue > IGB_N0_QUEUE)
 802                        msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 803                if (tx_queue > IGB_N0_QUEUE)
 804                        msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 805                if (!adapter->msix_entries && msix_vector == 0)
 806                        msixbm |= E1000_EIMS_OTHER;
 807                array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 808                q_vector->eims_value = msixbm;
 809                break;
 810        case e1000_82576:
 811                /* 82576 uses a table that essentially consists of 2 columns
 812                 * with 8 rows.  The ordering is column-major so we use the
 813                 * lower 3 bits as the row index, and the 4th bit as the
 814                 * column offset.
 815                 */
 816                if (rx_queue > IGB_N0_QUEUE)
 817                        igb_write_ivar(hw, msix_vector,
 818                                       rx_queue & 0x7,
 819                                       (rx_queue & 0x8) << 1);
 820                if (tx_queue > IGB_N0_QUEUE)
 821                        igb_write_ivar(hw, msix_vector,
 822                                       tx_queue & 0x7,
 823                                       ((tx_queue & 0x8) << 1) + 8);
 824                q_vector->eims_value = 1 << msix_vector;
 825                break;
 826        case e1000_82580:
 827        case e1000_i350:
 828        case e1000_i354:
 829        case e1000_i210:
 830        case e1000_i211:
 831                /* On 82580 and newer adapters the scheme is similar to 82576
 832                 * however instead of ordering column-major we have things
 833                 * ordered row-major.  So we traverse the table by using
 834                 * bit 0 as the column offset, and the remaining bits as the
 835                 * row index.
 836                 */
 837                if (rx_queue > IGB_N0_QUEUE)
 838                        igb_write_ivar(hw, msix_vector,
 839                                       rx_queue >> 1,
 840                                       (rx_queue & 0x1) << 4);
 841                if (tx_queue > IGB_N0_QUEUE)
 842                        igb_write_ivar(hw, msix_vector,
 843                                       tx_queue >> 1,
 844                                       ((tx_queue & 0x1) << 4) + 8);
 845                q_vector->eims_value = 1 << msix_vector;
 846                break;
 847        default:
 848                BUG();
 849                break;
 850        }
 851
 852        /* add q_vector eims value to global eims_enable_mask */
 853        adapter->eims_enable_mask |= q_vector->eims_value;
 854
 855        /* configure q_vector to set itr on first interrupt */
 856        q_vector->set_itr = 1;
 857}
 858
 859/**
 860 *  igb_configure_msix - Configure MSI-X hardware
 861 *  @adapter: board private structure to initialize
 862 *
 863 *  igb_configure_msix sets up the hardware to properly
 864 *  generate MSI-X interrupts.
 865 **/
 866static void igb_configure_msix(struct igb_adapter *adapter)
 867{
 868        u32 tmp;
 869        int i, vector = 0;
 870        struct e1000_hw *hw = &adapter->hw;
 871
 872        adapter->eims_enable_mask = 0;
 873
 874        /* set vector for other causes, i.e. link changes */
 875        switch (hw->mac.type) {
 876        case e1000_82575:
 877                tmp = rd32(E1000_CTRL_EXT);
 878                /* enable MSI-X PBA support*/
 879                tmp |= E1000_CTRL_EXT_PBA_CLR;
 880
 881                /* Auto-Mask interrupts upon ICR read. */
 882                tmp |= E1000_CTRL_EXT_EIAME;
 883                tmp |= E1000_CTRL_EXT_IRCA;
 884
 885                wr32(E1000_CTRL_EXT, tmp);
 886
 887                /* enable msix_other interrupt */
 888                array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
 889                adapter->eims_other = E1000_EIMS_OTHER;
 890
 891                break;
 892
 893        case e1000_82576:
 894        case e1000_82580:
 895        case e1000_i350:
 896        case e1000_i354:
 897        case e1000_i210:
 898        case e1000_i211:
 899                /* Turn on MSI-X capability first, or our settings
 900                 * won't stick.  And it will take days to debug.
 901                 */
 902                wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 903                     E1000_GPIE_PBA | E1000_GPIE_EIAME |
 904                     E1000_GPIE_NSICR);
 905
 906                /* enable msix_other interrupt */
 907                adapter->eims_other = 1 << vector;
 908                tmp = (vector++ | E1000_IVAR_VALID) << 8;
 909
 910                wr32(E1000_IVAR_MISC, tmp);
 911                break;
 912        default:
 913                /* do nothing, since nothing else supports MSI-X */
 914                break;
 915        } /* switch (hw->mac.type) */
 916
 917        adapter->eims_enable_mask |= adapter->eims_other;
 918
 919        for (i = 0; i < adapter->num_q_vectors; i++)
 920                igb_assign_vector(adapter->q_vector[i], vector++);
 921
 922        wrfl();
 923}
 924
 925/**
 926 *  igb_request_msix - Initialize MSI-X interrupts
 927 *  @adapter: board private structure to initialize
 928 *
 929 *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
 930 *  kernel.
 931 **/
 932static int igb_request_msix(struct igb_adapter *adapter)
 933{
 934        struct net_device *netdev = adapter->netdev;
 935        struct e1000_hw *hw = &adapter->hw;
 936        int i, err = 0, vector = 0, free_vector = 0;
 937
 938        err = request_irq(adapter->msix_entries[vector].vector,
 939                          igb_msix_other, 0, netdev->name, adapter);
 940        if (err)
 941                goto err_out;
 942
 943        for (i = 0; i < adapter->num_q_vectors; i++) {
 944                struct igb_q_vector *q_vector = adapter->q_vector[i];
 945
 946                vector++;
 947
 948                q_vector->itr_register = hw->hw_addr + E1000_EITR(vector);
 949
 950                if (q_vector->rx.ring && q_vector->tx.ring)
 951                        sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 952                                q_vector->rx.ring->queue_index);
 953                else if (q_vector->tx.ring)
 954                        sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 955                                q_vector->tx.ring->queue_index);
 956                else if (q_vector->rx.ring)
 957                        sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 958                                q_vector->rx.ring->queue_index);
 959                else
 960                        sprintf(q_vector->name, "%s-unused", netdev->name);
 961
 962                err = request_irq(adapter->msix_entries[vector].vector,
 963                                  igb_msix_ring, 0, q_vector->name,
 964                                  q_vector);
 965                if (err)
 966                        goto err_free;
 967        }
 968
 969        igb_configure_msix(adapter);
 970        return 0;
 971
 972err_free:
 973        /* free already assigned IRQs */
 974        free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 975
 976        vector--;
 977        for (i = 0; i < vector; i++) {
 978                free_irq(adapter->msix_entries[free_vector++].vector,
 979                         adapter->q_vector[i]);
 980        }
 981err_out:
 982        return err;
 983}
 984
 985static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 986{
 987        if (adapter->msix_entries) {
 988                pci_disable_msix(adapter->pdev);
 989                kfree(adapter->msix_entries);
 990                adapter->msix_entries = NULL;
 991        } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
 992                pci_disable_msi(adapter->pdev);
 993        }
 994}
 995
 996/**
 997 *  igb_free_q_vector - Free memory allocated for specific interrupt vector
 998 *  @adapter: board private structure to initialize
 999 *  @v_idx: Index of vector to be freed
1000 *
1001 *  This function frees the memory allocated to the q_vector.  In addition if
1002 *  NAPI is enabled it will delete any references to the NAPI struct prior
1003 *  to freeing the q_vector.
1004 **/
1005static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1006{
1007        struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1008
1009        if (q_vector->tx.ring)
1010                adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1011
1012        if (q_vector->rx.ring)
1013                adapter->tx_ring[q_vector->rx.ring->queue_index] = NULL;
1014
1015        adapter->q_vector[v_idx] = NULL;
1016        netif_napi_del(&q_vector->napi);
1017
1018        /* igb_get_stats64() might access the rings on this vector,
1019         * we must wait a grace period before freeing it.
1020         */
1021        kfree_rcu(q_vector, rcu);
1022}
1023
1024/**
1025 *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1026 *  @adapter: board private structure to initialize
1027 *
1028 *  This function frees the memory allocated to the q_vectors.  In addition if
1029 *  NAPI is enabled it will delete any references to the NAPI struct prior
1030 *  to freeing the q_vector.
1031 **/
1032static void igb_free_q_vectors(struct igb_adapter *adapter)
1033{
1034        int v_idx = adapter->num_q_vectors;
1035
1036        adapter->num_tx_queues = 0;
1037        adapter->num_rx_queues = 0;
1038        adapter->num_q_vectors = 0;
1039
1040        while (v_idx--)
1041                igb_free_q_vector(adapter, v_idx);
1042}
1043
1044/**
1045 *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1046 *  @adapter: board private structure to initialize
1047 *
1048 *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1049 *  MSI-X interrupts allocated.
1050 */
1051static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1052{
1053        igb_free_q_vectors(adapter);
1054        igb_reset_interrupt_capability(adapter);
1055}
1056
1057/**
1058 *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1059 *  @adapter: board private structure to initialize
1060 *  @msix: boolean value of MSIX capability
1061 *
1062 *  Attempt to configure interrupts using the best available
1063 *  capabilities of the hardware and kernel.
1064 **/
1065static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1066{
1067        int err;
1068        int numvecs, i;
1069
1070        if (!msix)
1071                goto msi_only;
1072
1073        /* Number of supported queues. */
1074        adapter->num_rx_queues = adapter->rss_queues;
1075        if (adapter->vfs_allocated_count)
1076                adapter->num_tx_queues = 1;
1077        else
1078                adapter->num_tx_queues = adapter->rss_queues;
1079
1080        /* start with one vector for every Rx queue */
1081        numvecs = adapter->num_rx_queues;
1082
1083        /* if Tx handler is separate add 1 for every Tx queue */
1084        if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1085                numvecs += adapter->num_tx_queues;
1086
1087        /* store the number of vectors reserved for queues */
1088        adapter->num_q_vectors = numvecs;
1089
1090        /* add 1 vector for link status interrupts */
1091        numvecs++;
1092        adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
1093                                        GFP_KERNEL);
1094
1095        if (!adapter->msix_entries)
1096                goto msi_only;
1097
1098        for (i = 0; i < numvecs; i++)
1099                adapter->msix_entries[i].entry = i;
1100
1101        err = pci_enable_msix(adapter->pdev,
1102                              adapter->msix_entries,
1103                              numvecs);
1104        if (err == 0)
1105                return;
1106
1107        igb_reset_interrupt_capability(adapter);
1108
1109        /* If we can't do MSI-X, try MSI */
1110msi_only:
1111#ifdef CONFIG_PCI_IOV
1112        /* disable SR-IOV for non MSI-X configurations */
1113        if (adapter->vf_data) {
1114                struct e1000_hw *hw = &adapter->hw;
1115                /* disable iov and allow time for transactions to clear */
1116                pci_disable_sriov(adapter->pdev);
1117                msleep(500);
1118
1119                kfree(adapter->vf_data);
1120                adapter->vf_data = NULL;
1121                wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1122                wrfl();
1123                msleep(100);
1124                dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1125        }
1126#endif
1127        adapter->vfs_allocated_count = 0;
1128        adapter->rss_queues = 1;
1129        adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1130        adapter->num_rx_queues = 1;
1131        adapter->num_tx_queues = 1;
1132        adapter->num_q_vectors = 1;
1133        if (!pci_enable_msi(adapter->pdev))
1134                adapter->flags |= IGB_FLAG_HAS_MSI;
1135}
1136
1137static void igb_add_ring(struct igb_ring *ring,
1138                         struct igb_ring_container *head)
1139{
1140        head->ring = ring;
1141        head->count++;
1142}
1143
1144/**
1145 *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1146 *  @adapter: board private structure to initialize
1147 *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1148 *  @v_idx: index of vector in adapter struct
1149 *  @txr_count: total number of Tx rings to allocate
1150 *  @txr_idx: index of first Tx ring to allocate
1151 *  @rxr_count: total number of Rx rings to allocate
1152 *  @rxr_idx: index of first Rx ring to allocate
1153 *
1154 *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1155 **/
1156static int igb_alloc_q_vector(struct igb_adapter *adapter,
1157                              int v_count, int v_idx,
1158                              int txr_count, int txr_idx,
1159                              int rxr_count, int rxr_idx)
1160{
1161        struct igb_q_vector *q_vector;
1162        struct igb_ring *ring;
1163        int ring_count, size;
1164
1165        /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1166        if (txr_count > 1 || rxr_count > 1)
1167                return -ENOMEM;
1168
1169        ring_count = txr_count + rxr_count;
1170        size = sizeof(struct igb_q_vector) +
1171               (sizeof(struct igb_ring) * ring_count);
1172
1173        /* allocate q_vector and rings */
1174        q_vector = kzalloc(size, GFP_KERNEL);
1175        if (!q_vector)
1176                return -ENOMEM;
1177
1178        /* initialize NAPI */
1179        netif_napi_add(adapter->netdev, &q_vector->napi,
1180                       igb_poll, 64);
1181
1182        /* tie q_vector and adapter together */
1183        adapter->q_vector[v_idx] = q_vector;
1184        q_vector->adapter = adapter;
1185
1186        /* initialize work limits */
1187        q_vector->tx.work_limit = adapter->tx_work_limit;
1188
1189        /* initialize ITR configuration */
1190        q_vector->itr_register = adapter->hw.hw_addr + E1000_EITR(0);
1191        q_vector->itr_val = IGB_START_ITR;
1192
1193        /* initialize pointer to rings */
1194        ring = q_vector->ring;
1195
1196        /* intialize ITR */
1197        if (rxr_count) {
1198                /* rx or rx/tx vector */
1199                if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1200                        q_vector->itr_val = adapter->rx_itr_setting;
1201        } else {
1202                /* tx only vector */
1203                if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1204                        q_vector->itr_val = adapter->tx_itr_setting;
1205        }
1206
1207        if (txr_count) {
1208                /* assign generic ring traits */
1209                ring->dev = &adapter->pdev->dev;
1210                ring->netdev = adapter->netdev;
1211
1212                /* configure backlink on ring */
1213                ring->q_vector = q_vector;
1214
1215                /* update q_vector Tx values */
1216                igb_add_ring(ring, &q_vector->tx);
1217
1218                /* For 82575, context index must be unique per ring. */
1219                if (adapter->hw.mac.type == e1000_82575)
1220                        set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1221
1222                /* apply Tx specific ring traits */
1223                ring->count = adapter->tx_ring_count;
1224                ring->queue_index = txr_idx;
1225
1226                /* assign ring to adapter */
1227                adapter->tx_ring[txr_idx] = ring;
1228
1229                /* push pointer to next ring */
1230                ring++;
1231        }
1232
1233        if (rxr_count) {
1234                /* assign generic ring traits */
1235                ring->dev = &adapter->pdev->dev;
1236                ring->netdev = adapter->netdev;
1237
1238                /* configure backlink on ring */
1239                ring->q_vector = q_vector;
1240
1241                /* update q_vector Rx values */
1242                igb_add_ring(ring, &q_vector->rx);
1243
1244                /* set flag indicating ring supports SCTP checksum offload */
1245                if (adapter->hw.mac.type >= e1000_82576)
1246                        set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1247
1248                /*
1249                 * On i350, i354, i210, and i211, loopback VLAN packets
1250                 * have the tag byte-swapped.
1251                 */
1252                if (adapter->hw.mac.type >= e1000_i350)
1253                        set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1254
1255                /* apply Rx specific ring traits */
1256                ring->count = adapter->rx_ring_count;
1257                ring->queue_index = rxr_idx;
1258
1259                /* assign ring to adapter */
1260                adapter->rx_ring[rxr_idx] = ring;
1261        }
1262
1263        return 0;
1264}
1265
1266
1267/**
1268 *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1269 *  @adapter: board private structure to initialize
1270 *
1271 *  We allocate one q_vector per queue interrupt.  If allocation fails we
1272 *  return -ENOMEM.
1273 **/
1274static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1275{
1276        int q_vectors = adapter->num_q_vectors;
1277        int rxr_remaining = adapter->num_rx_queues;
1278        int txr_remaining = adapter->num_tx_queues;
1279        int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1280        int err;
1281
1282        if (q_vectors >= (rxr_remaining + txr_remaining)) {
1283                for (; rxr_remaining; v_idx++) {
1284                        err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1285                                                 0, 0, 1, rxr_idx);
1286
1287                        if (err)
1288                                goto err_out;
1289
1290                        /* update counts and index */
1291                        rxr_remaining--;
1292                        rxr_idx++;
1293                }
1294        }
1295
1296        for (; v_idx < q_vectors; v_idx++) {
1297                int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1298                int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1299                err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1300                                         tqpv, txr_idx, rqpv, rxr_idx);
1301
1302                if (err)
1303                        goto err_out;
1304
1305                /* update counts and index */
1306                rxr_remaining -= rqpv;
1307                txr_remaining -= tqpv;
1308                rxr_idx++;
1309                txr_idx++;
1310        }
1311
1312        return 0;
1313
1314err_out:
1315        adapter->num_tx_queues = 0;
1316        adapter->num_rx_queues = 0;
1317        adapter->num_q_vectors = 0;
1318
1319        while (v_idx--)
1320                igb_free_q_vector(adapter, v_idx);
1321
1322        return -ENOMEM;
1323}
1324
1325/**
1326 *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1327 *  @adapter: board private structure to initialize
1328 *  @msix: boolean value of MSIX capability
1329 *
1330 *  This function initializes the interrupts and allocates all of the queues.
1331 **/
1332static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1333{
1334        struct pci_dev *pdev = adapter->pdev;
1335        int err;
1336
1337        igb_set_interrupt_capability(adapter, msix);
1338
1339        err = igb_alloc_q_vectors(adapter);
1340        if (err) {
1341                dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1342                goto err_alloc_q_vectors;
1343        }
1344
1345        igb_cache_ring_register(adapter);
1346
1347        return 0;
1348
1349err_alloc_q_vectors:
1350        igb_reset_interrupt_capability(adapter);
1351        return err;
1352}
1353
1354/**
1355 *  igb_request_irq - initialize interrupts
1356 *  @adapter: board private structure to initialize
1357 *
1358 *  Attempts to configure interrupts using the best available
1359 *  capabilities of the hardware and kernel.
1360 **/
1361static int igb_request_irq(struct igb_adapter *adapter)
1362{
1363        struct net_device *netdev = adapter->netdev;
1364        struct pci_dev *pdev = adapter->pdev;
1365        int err = 0;
1366
1367        if (adapter->msix_entries) {
1368                err = igb_request_msix(adapter);
1369                if (!err)
1370                        goto request_done;
1371                /* fall back to MSI */
1372                igb_free_all_tx_resources(adapter);
1373                igb_free_all_rx_resources(adapter);
1374
1375                igb_clear_interrupt_scheme(adapter);
1376                err = igb_init_interrupt_scheme(adapter, false);
1377                if (err)
1378                        goto request_done;
1379
1380                igb_setup_all_tx_resources(adapter);
1381                igb_setup_all_rx_resources(adapter);
1382                igb_configure(adapter);
1383        }
1384
1385        igb_assign_vector(adapter->q_vector[0], 0);
1386
1387        if (adapter->flags & IGB_FLAG_HAS_MSI) {
1388                err = request_irq(pdev->irq, igb_intr_msi, 0,
1389                                  netdev->name, adapter);
1390                if (!err)
1391                        goto request_done;
1392
1393                /* fall back to legacy interrupts */
1394                igb_reset_interrupt_capability(adapter);
1395                adapter->flags &= ~IGB_FLAG_HAS_MSI;
1396        }
1397
1398        err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1399                          netdev->name, adapter);
1400
1401        if (err)
1402                dev_err(&pdev->dev, "Error %d getting interrupt\n",
1403                        err);
1404
1405request_done:
1406        return err;
1407}
1408
1409static void igb_free_irq(struct igb_adapter *adapter)
1410{
1411        if (adapter->msix_entries) {
1412                int vector = 0, i;
1413
1414                free_irq(adapter->msix_entries[vector++].vector, adapter);
1415
1416                for (i = 0; i < adapter->num_q_vectors; i++)
1417                        free_irq(adapter->msix_entries[vector++].vector,
1418                                 adapter->q_vector[i]);
1419        } else {
1420                free_irq(adapter->pdev->irq, adapter);
1421        }
1422}
1423
1424/**
1425 *  igb_irq_disable - Mask off interrupt generation on the NIC
1426 *  @adapter: board private structure
1427 **/
1428static void igb_irq_disable(struct igb_adapter *adapter)
1429{
1430        struct e1000_hw *hw = &adapter->hw;
1431
1432        /* we need to be careful when disabling interrupts.  The VFs are also
1433         * mapped into these registers and so clearing the bits can cause
1434         * issues on the VF drivers so we only need to clear what we set
1435         */
1436        if (adapter->msix_entries) {
1437                u32 regval = rd32(E1000_EIAM);
1438                wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1439                wr32(E1000_EIMC, adapter->eims_enable_mask);
1440                regval = rd32(E1000_EIAC);
1441                wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1442        }
1443
1444        wr32(E1000_IAM, 0);
1445        wr32(E1000_IMC, ~0);
1446        wrfl();
1447        if (adapter->msix_entries) {
1448                int i;
1449                for (i = 0; i < adapter->num_q_vectors; i++)
1450                        synchronize_irq(adapter->msix_entries[i].vector);
1451        } else {
1452                synchronize_irq(adapter->pdev->irq);
1453        }
1454}
1455
1456/**
1457 *  igb_irq_enable - Enable default interrupt generation settings
1458 *  @adapter: board private structure
1459 **/
1460static void igb_irq_enable(struct igb_adapter *adapter)
1461{
1462        struct e1000_hw *hw = &adapter->hw;
1463
1464        if (adapter->msix_entries) {
1465                u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1466                u32 regval = rd32(E1000_EIAC);
1467                wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1468                regval = rd32(E1000_EIAM);
1469                wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1470                wr32(E1000_EIMS, adapter->eims_enable_mask);
1471                if (adapter->vfs_allocated_count) {
1472                        wr32(E1000_MBVFIMR, 0xFF);
1473                        ims |= E1000_IMS_VMMB;
1474                }
1475                wr32(E1000_IMS, ims);
1476        } else {
1477                wr32(E1000_IMS, IMS_ENABLE_MASK |
1478                                E1000_IMS_DRSTA);
1479                wr32(E1000_IAM, IMS_ENABLE_MASK |
1480                                E1000_IMS_DRSTA);
1481        }
1482}
1483
1484static void igb_update_mng_vlan(struct igb_adapter *adapter)
1485{
1486        struct e1000_hw *hw = &adapter->hw;
1487        u16 vid = adapter->hw.mng_cookie.vlan_id;
1488        u16 old_vid = adapter->mng_vlan_id;
1489
1490        if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1491                /* add VID to filter table */
1492                igb_vfta_set(hw, vid, true);
1493                adapter->mng_vlan_id = vid;
1494        } else {
1495                adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1496        }
1497
1498        if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1499            (vid != old_vid) &&
1500            !test_bit(old_vid, adapter->active_vlans)) {
1501                /* remove VID from filter table */
1502                igb_vfta_set(hw, old_vid, false);
1503        }
1504}
1505
1506/**
1507 *  igb_release_hw_control - release control of the h/w to f/w
1508 *  @adapter: address of board private structure
1509 *
1510 *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1511 *  For ASF and Pass Through versions of f/w this means that the
1512 *  driver is no longer loaded.
1513 **/
1514static void igb_release_hw_control(struct igb_adapter *adapter)
1515{
1516        struct e1000_hw *hw = &adapter->hw;
1517        u32 ctrl_ext;
1518
1519        /* Let firmware take over control of h/w */
1520        ctrl_ext = rd32(E1000_CTRL_EXT);
1521        wr32(E1000_CTRL_EXT,
1522                        ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1523}
1524
1525/**
1526 *  igb_get_hw_control - get control of the h/w from f/w
1527 *  @adapter: address of board private structure
1528 *
1529 *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1530 *  For ASF and Pass Through versions of f/w this means that
1531 *  the driver is loaded.
1532 **/
1533static void igb_get_hw_control(struct igb_adapter *adapter)
1534{
1535        struct e1000_hw *hw = &adapter->hw;
1536        u32 ctrl_ext;
1537
1538        /* Let firmware know the driver has taken over */
1539        ctrl_ext = rd32(E1000_CTRL_EXT);
1540        wr32(E1000_CTRL_EXT,
1541                        ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1542}
1543
1544/**
1545 *  igb_configure - configure the hardware for RX and TX
1546 *  @adapter: private board structure
1547 **/
1548static void igb_configure(struct igb_adapter *adapter)
1549{
1550        struct net_device *netdev = adapter->netdev;
1551        int i;
1552
1553        igb_get_hw_control(adapter);
1554        igb_set_rx_mode(netdev);
1555
1556        igb_restore_vlan(adapter);
1557
1558        igb_setup_tctl(adapter);
1559        igb_setup_mrqc(adapter);
1560        igb_setup_rctl(adapter);
1561
1562        igb_configure_tx(adapter);
1563        igb_configure_rx(adapter);
1564
1565        igb_rx_fifo_flush_82575(&adapter->hw);
1566
1567        /* call igb_desc_unused which always leaves
1568         * at least 1 descriptor unused to make sure
1569         * next_to_use != next_to_clean
1570         */
1571        for (i = 0; i < adapter->num_rx_queues; i++) {
1572                struct igb_ring *ring = adapter->rx_ring[i];
1573                igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1574        }
1575}
1576
1577/**
1578 *  igb_power_up_link - Power up the phy/serdes link
1579 *  @adapter: address of board private structure
1580 **/
1581void igb_power_up_link(struct igb_adapter *adapter)
1582{
1583        igb_reset_phy(&adapter->hw);
1584
1585        if (adapter->hw.phy.media_type == e1000_media_type_copper)
1586                igb_power_up_phy_copper(&adapter->hw);
1587        else
1588                igb_power_up_serdes_link_82575(&adapter->hw);
1589}
1590
1591/**
1592 *  igb_power_down_link - Power down the phy/serdes link
1593 *  @adapter: address of board private structure
1594 */
1595static void igb_power_down_link(struct igb_adapter *adapter)
1596{
1597        if (adapter->hw.phy.media_type == e1000_media_type_copper)
1598                igb_power_down_phy_copper_82575(&adapter->hw);
1599        else
1600                igb_shutdown_serdes_link_82575(&adapter->hw);
1601}
1602
1603/**
1604 *  igb_up - Open the interface and prepare it to handle traffic
1605 *  @adapter: board private structure
1606 **/
1607int igb_up(struct igb_adapter *adapter)
1608{
1609        struct e1000_hw *hw = &adapter->hw;
1610        int i;
1611
1612        /* hardware has been reset, we need to reload some things */
1613        igb_configure(adapter);
1614
1615        clear_bit(__IGB_DOWN, &adapter->state);
1616
1617        for (i = 0; i < adapter->num_q_vectors; i++)
1618                napi_enable(&(adapter->q_vector[i]->napi));
1619
1620        if (adapter->msix_entries)
1621                igb_configure_msix(adapter);
1622        else
1623                igb_assign_vector(adapter->q_vector[0], 0);
1624
1625        /* Clear any pending interrupts. */
1626        rd32(E1000_ICR);
1627        igb_irq_enable(adapter);
1628
1629        /* notify VFs that reset has been completed */
1630        if (adapter->vfs_allocated_count) {
1631                u32 reg_data = rd32(E1000_CTRL_EXT);
1632                reg_data |= E1000_CTRL_EXT_PFRSTD;
1633                wr32(E1000_CTRL_EXT, reg_data);
1634        }
1635
1636        netif_tx_start_all_queues(adapter->netdev);
1637
1638        /* start the watchdog. */
1639        hw->mac.get_link_status = 1;
1640        schedule_work(&adapter->watchdog_task);
1641
1642        return 0;
1643}
1644
1645void igb_down(struct igb_adapter *adapter)
1646{
1647        struct net_device *netdev = adapter->netdev;
1648        struct e1000_hw *hw = &adapter->hw;
1649        u32 tctl, rctl;
1650        int i;
1651
1652        /* signal that we're down so the interrupt handler does not
1653         * reschedule our watchdog timer
1654         */
1655        set_bit(__IGB_DOWN, &adapter->state);
1656
1657        /* disable receives in the hardware */
1658        rctl = rd32(E1000_RCTL);
1659        wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1660        /* flush and sleep below */
1661
1662        netif_tx_stop_all_queues(netdev);
1663
1664        /* disable transmits in the hardware */
1665        tctl = rd32(E1000_TCTL);
1666        tctl &= ~E1000_TCTL_EN;
1667        wr32(E1000_TCTL, tctl);
1668        /* flush both disables and wait for them to finish */
1669        wrfl();
1670        msleep(10);
1671
1672        igb_irq_disable(adapter);
1673
1674        adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1675
1676        for (i = 0; i < adapter->num_q_vectors; i++) {
1677                napi_synchronize(&(adapter->q_vector[i]->napi));
1678                napi_disable(&(adapter->q_vector[i]->napi));
1679        }
1680
1681
1682        del_timer_sync(&adapter->watchdog_timer);
1683        del_timer_sync(&adapter->phy_info_timer);
1684
1685        netif_carrier_off(netdev);
1686
1687        /* record the stats before reset*/
1688        spin_lock(&adapter->stats64_lock);
1689        igb_update_stats(adapter, &adapter->stats64);
1690        spin_unlock(&adapter->stats64_lock);
1691
1692        adapter->link_speed = 0;
1693        adapter->link_duplex = 0;
1694
1695        if (!pci_channel_offline(adapter->pdev))
1696                igb_reset(adapter);
1697        igb_clean_all_tx_rings(adapter);
1698        igb_clean_all_rx_rings(adapter);
1699#ifdef CONFIG_IGB_DCA
1700
1701        /* since we reset the hardware DCA settings were cleared */
1702        igb_setup_dca(adapter);
1703#endif
1704}
1705
1706void igb_reinit_locked(struct igb_adapter *adapter)
1707{
1708        WARN_ON(in_interrupt());
1709        while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1710                msleep(1);
1711        igb_down(adapter);
1712        igb_up(adapter);
1713        clear_bit(__IGB_RESETTING, &adapter->state);
1714}
1715
1716void igb_reset(struct igb_adapter *adapter)
1717{
1718        struct pci_dev *pdev = adapter->pdev;
1719        struct e1000_hw *hw = &adapter->hw;
1720        struct e1000_mac_info *mac = &hw->mac;
1721        struct e1000_fc_info *fc = &hw->fc;
1722        u32 pba = 0, tx_space, min_tx_space, min_rx_space, hwm;
1723
1724        /* Repartition Pba for greater than 9k mtu
1725         * To take effect CTRL.RST is required.
1726         */
1727        switch (mac->type) {
1728        case e1000_i350:
1729        case e1000_i354:
1730        case e1000_82580:
1731                pba = rd32(E1000_RXPBS);
1732                pba = igb_rxpbs_adjust_82580(pba);
1733                break;
1734        case e1000_82576:
1735                pba = rd32(E1000_RXPBS);
1736                pba &= E1000_RXPBS_SIZE_MASK_82576;
1737                break;
1738        case e1000_82575:
1739        case e1000_i210:
1740        case e1000_i211:
1741        default:
1742                pba = E1000_PBA_34K;
1743                break;
1744        }
1745
1746        if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1747            (mac->type < e1000_82576)) {
1748                /* adjust PBA for jumbo frames */
1749                wr32(E1000_PBA, pba);
1750
1751                /* To maintain wire speed transmits, the Tx FIFO should be
1752                 * large enough to accommodate two full transmit packets,
1753                 * rounded up to the next 1KB and expressed in KB.  Likewise,
1754                 * the Rx FIFO should be large enough to accommodate at least
1755                 * one full receive packet and is similarly rounded up and
1756                 * expressed in KB.
1757                 */
1758                pba = rd32(E1000_PBA);
1759                /* upper 16 bits has Tx packet buffer allocation size in KB */
1760                tx_space = pba >> 16;
1761                /* lower 16 bits has Rx packet buffer allocation size in KB */
1762                pba &= 0xffff;
1763                /* the Tx fifo also stores 16 bytes of information about the Tx
1764                 * but don't include ethernet FCS because hardware appends it
1765                 */
1766                min_tx_space = (adapter->max_frame_size +
1767                                sizeof(union e1000_adv_tx_desc) -
1768                                ETH_FCS_LEN) * 2;
1769                min_tx_space = ALIGN(min_tx_space, 1024);
1770                min_tx_space >>= 10;
1771                /* software strips receive CRC, so leave room for it */
1772                min_rx_space = adapter->max_frame_size;
1773                min_rx_space = ALIGN(min_rx_space, 1024);
1774                min_rx_space >>= 10;
1775
1776                /* If current Tx allocation is less than the min Tx FIFO size,
1777                 * and the min Tx FIFO size is less than the current Rx FIFO
1778                 * allocation, take space away from current Rx allocation
1779                 */
1780                if (tx_space < min_tx_space &&
1781                    ((min_tx_space - tx_space) < pba)) {
1782                        pba = pba - (min_tx_space - tx_space);
1783
1784                        /* if short on Rx space, Rx wins and must trump Tx
1785                         * adjustment
1786                         */
1787                        if (pba < min_rx_space)
1788                                pba = min_rx_space;
1789                }
1790                wr32(E1000_PBA, pba);
1791        }
1792
1793        /* flow control settings */
1794        /* The high water mark must be low enough to fit one full frame
1795         * (or the size used for early receive) above it in the Rx FIFO.
1796         * Set it to the lower of:
1797         * - 90% of the Rx FIFO size, or
1798         * - the full Rx FIFO size minus one full frame
1799         */
1800        hwm = min(((pba << 10) * 9 / 10),
1801                        ((pba << 10) - 2 * adapter->max_frame_size));
1802
1803        fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
1804        fc->low_water = fc->high_water - 16;
1805        fc->pause_time = 0xFFFF;
1806        fc->send_xon = 1;
1807        fc->current_mode = fc->requested_mode;
1808
1809        /* disable receive for all VFs and wait one second */
1810        if (adapter->vfs_allocated_count) {
1811                int i;
1812                for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1813                        adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1814
1815                /* ping all the active vfs to let them know we are going down */
1816                igb_ping_all_vfs(adapter);
1817
1818                /* disable transmits and receives */
1819                wr32(E1000_VFRE, 0);
1820                wr32(E1000_VFTE, 0);
1821        }
1822
1823        /* Allow time for pending master requests to run */
1824        hw->mac.ops.reset_hw(hw);
1825        wr32(E1000_WUC, 0);
1826
1827        if (hw->mac.ops.init_hw(hw))
1828                dev_err(&pdev->dev, "Hardware Error\n");
1829
1830        /* Flow control settings reset on hardware reset, so guarantee flow
1831         * control is off when forcing speed.
1832         */
1833        if (!hw->mac.autoneg)
1834                igb_force_mac_fc(hw);
1835
1836        igb_init_dmac(adapter, pba);
1837#ifdef CONFIG_IGB_HWMON
1838        /* Re-initialize the thermal sensor on i350 devices. */
1839        if (!test_bit(__IGB_DOWN, &adapter->state)) {
1840                if (mac->type == e1000_i350 && hw->bus.func == 0) {
1841                        /* If present, re-initialize the external thermal sensor
1842                         * interface.
1843                         */
1844                        if (adapter->ets)
1845                                mac->ops.init_thermal_sensor_thresh(hw);
1846                }
1847        }
1848#endif
1849        if (!netif_running(adapter->netdev))
1850                igb_power_down_link(adapter);
1851
1852        igb_update_mng_vlan(adapter);
1853
1854        /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1855        wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1856
1857        /* Re-enable PTP, where applicable. */
1858        igb_ptp_reset(adapter);
1859
1860        igb_get_phy_info(hw);
1861}
1862
1863static netdev_features_t igb_fix_features(struct net_device *netdev,
1864        netdev_features_t features)
1865{
1866        /* Since there is no support for separate Rx/Tx vlan accel
1867         * enable/disable make sure Tx flag is always in same state as Rx.
1868         */
1869        if (features & NETIF_F_HW_VLAN_CTAG_RX)
1870                features |= NETIF_F_HW_VLAN_CTAG_TX;
1871        else
1872                features &= ~NETIF_F_HW_VLAN_CTAG_TX;
1873
1874        return features;
1875}
1876
1877static int igb_set_features(struct net_device *netdev,
1878        netdev_features_t features)
1879{
1880        netdev_features_t changed = netdev->features ^ features;
1881        struct igb_adapter *adapter = netdev_priv(netdev);
1882
1883        if (changed & NETIF_F_HW_VLAN_CTAG_RX)
1884                igb_vlan_mode(netdev, features);
1885
1886        if (!(changed & NETIF_F_RXALL))
1887                return 0;
1888
1889        netdev->features = features;
1890
1891        if (netif_running(netdev))
1892                igb_reinit_locked(adapter);
1893        else
1894                igb_reset(adapter);
1895
1896        return 0;
1897}
1898
1899static const struct net_device_ops igb_netdev_ops = {
1900        .ndo_open               = igb_open,
1901        .ndo_stop               = igb_close,
1902        .ndo_start_xmit         = igb_xmit_frame,
1903        .ndo_get_stats64        = igb_get_stats64,
1904        .ndo_set_rx_mode        = igb_set_rx_mode,
1905        .ndo_set_mac_address    = igb_set_mac,
1906        .ndo_change_mtu         = igb_change_mtu,
1907        .ndo_do_ioctl           = igb_ioctl,
1908        .ndo_tx_timeout         = igb_tx_timeout,
1909        .ndo_validate_addr      = eth_validate_addr,
1910        .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
1911        .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
1912        .ndo_set_vf_mac         = igb_ndo_set_vf_mac,
1913        .ndo_set_vf_vlan        = igb_ndo_set_vf_vlan,
1914        .ndo_set_vf_tx_rate     = igb_ndo_set_vf_bw,
1915        .ndo_set_vf_spoofchk    = igb_ndo_set_vf_spoofchk,
1916        .ndo_get_vf_config      = igb_ndo_get_vf_config,
1917#ifdef CONFIG_NET_POLL_CONTROLLER
1918        .ndo_poll_controller    = igb_netpoll,
1919#endif
1920        .ndo_fix_features       = igb_fix_features,
1921        .ndo_set_features       = igb_set_features,
1922};
1923
1924/**
1925 * igb_set_fw_version - Configure version string for ethtool
1926 * @adapter: adapter struct
1927 **/
1928void igb_set_fw_version(struct igb_adapter *adapter)
1929{
1930        struct e1000_hw *hw = &adapter->hw;
1931        struct e1000_fw_version fw;
1932
1933        igb_get_fw_version(hw, &fw);
1934
1935        switch (hw->mac.type) {
1936        case e1000_i210:
1937        case e1000_i211:
1938                if (!(igb_get_flash_presence_i210(hw))) {
1939                        snprintf(adapter->fw_version,
1940                                 sizeof(adapter->fw_version),
1941                                 "%2d.%2d-%d",
1942                                 fw.invm_major, fw.invm_minor,
1943                                 fw.invm_img_type);
1944                        break;
1945                }
1946                /* fall through */
1947        default:
1948                /* if option is rom valid, display its version too */
1949                if (fw.or_valid) {
1950                        snprintf(adapter->fw_version,
1951                                 sizeof(adapter->fw_version),
1952                                 "%d.%d, 0x%08x, %d.%d.%d",
1953                                 fw.eep_major, fw.eep_minor, fw.etrack_id,
1954                                 fw.or_major, fw.or_build, fw.or_patch);
1955                /* no option rom */
1956                } else if (fw.etrack_id != 0X0000) {
1957                        snprintf(adapter->fw_version,
1958                            sizeof(adapter->fw_version),
1959                            "%d.%d, 0x%08x",
1960                            fw.eep_major, fw.eep_minor, fw.etrack_id);
1961                } else {
1962                snprintf(adapter->fw_version,
1963                    sizeof(adapter->fw_version),
1964                    "%d.%d.%d",
1965                    fw.eep_major, fw.eep_minor, fw.eep_build);
1966                }
1967                break;
1968        }
1969        return;
1970}
1971
1972/**
1973 *  igb_init_i2c - Init I2C interface
1974 *  @adapter: pointer to adapter structure
1975 **/
1976static s32 igb_init_i2c(struct igb_adapter *adapter)
1977{
1978        s32 status = E1000_SUCCESS;
1979
1980        /* I2C interface supported on i350 devices */
1981        if (adapter->hw.mac.type != e1000_i350)
1982                return E1000_SUCCESS;
1983
1984        /* Initialize the i2c bus which is controlled by the registers.
1985         * This bus will use the i2c_algo_bit structue that implements
1986         * the protocol through toggling of the 4 bits in the register.
1987         */
1988        adapter->i2c_adap.owner = THIS_MODULE;
1989        adapter->i2c_algo = igb_i2c_algo;
1990        adapter->i2c_algo.data = adapter;
1991        adapter->i2c_adap.algo_data = &adapter->i2c_algo;
1992        adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
1993        strlcpy(adapter->i2c_adap.name, "igb BB",
1994                sizeof(adapter->i2c_adap.name));
1995        status = i2c_bit_add_bus(&adapter->i2c_adap);
1996        return status;
1997}
1998
1999/**
2000 *  igb_probe - Device Initialization Routine
2001 *  @pdev: PCI device information struct
2002 *  @ent: entry in igb_pci_tbl
2003 *
2004 *  Returns 0 on success, negative on failure
2005 *
2006 *  igb_probe initializes an adapter identified by a pci_dev structure.
2007 *  The OS initialization, configuring of the adapter private structure,
2008 *  and a hardware reset occur.
2009 **/
2010static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2011{
2012        struct net_device *netdev;
2013        struct igb_adapter *adapter;
2014        struct e1000_hw *hw;
2015        u16 eeprom_data = 0;
2016        s32 ret_val;
2017        static int global_quad_port_a; /* global quad port a indication */
2018        const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2019        unsigned long mmio_start, mmio_len;
2020        int err, pci_using_dac;
2021        u8 part_str[E1000_PBANUM_LENGTH];
2022
2023        /* Catch broken hardware that put the wrong VF device ID in
2024         * the PCIe SR-IOV capability.
2025         */
2026        if (pdev->is_virtfn) {
2027                WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2028                        pci_name(pdev), pdev->vendor, pdev->device);
2029                return -EINVAL;
2030        }
2031
2032        err = pci_enable_device_mem(pdev);
2033        if (err)
2034                return err;
2035
2036        pci_using_dac = 0;
2037        err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(64));
2038        if (!err) {
2039                err = dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64));
2040                if (!err)
2041                        pci_using_dac = 1;
2042        } else {
2043                err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2044                if (err) {
2045                        err = dma_set_coherent_mask(&pdev->dev,
2046                                                    DMA_BIT_MASK(32));
2047                        if (err) {
2048                                dev_err(&pdev->dev,
2049                                        "No usable DMA configuration, aborting\n");
2050                                goto err_dma;
2051                        }
2052                }
2053        }
2054
2055        err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
2056                                           IORESOURCE_MEM),
2057                                           igb_driver_name);
2058        if (err)
2059                goto err_pci_reg;
2060
2061        pci_enable_pcie_error_reporting(pdev);
2062
2063        pci_set_master(pdev);
2064        pci_save_state(pdev);
2065
2066        err = -ENOMEM;
2067        netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2068                                   IGB_MAX_TX_QUEUES);
2069        if (!netdev)
2070                goto err_alloc_etherdev;
2071
2072        SET_NETDEV_DEV(netdev, &pdev->dev);
2073
2074        pci_set_drvdata(pdev, netdev);
2075        adapter = netdev_priv(netdev);
2076        adapter->netdev = netdev;
2077        adapter->pdev = pdev;
2078        hw = &adapter->hw;
2079        hw->back = adapter;
2080        adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2081
2082        mmio_start = pci_resource_start(pdev, 0);
2083        mmio_len = pci_resource_len(pdev, 0);
2084
2085        err = -EIO;
2086        hw->hw_addr = ioremap(mmio_start, mmio_len);
2087        if (!hw->hw_addr)
2088                goto err_ioremap;
2089
2090        netdev->netdev_ops = &igb_netdev_ops;
2091        igb_set_ethtool_ops(netdev);
2092        netdev->watchdog_timeo = 5 * HZ;
2093
2094        strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2095
2096        netdev->mem_start = mmio_start;
2097        netdev->mem_end = mmio_start + mmio_len;
2098
2099        /* PCI config space info */
2100        hw->vendor_id = pdev->vendor;
2101        hw->device_id = pdev->device;
2102        hw->revision_id = pdev->revision;
2103        hw->subsystem_vendor_id = pdev->subsystem_vendor;
2104        hw->subsystem_device_id = pdev->subsystem_device;
2105
2106        /* Copy the default MAC, PHY and NVM function pointers */
2107        memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2108        memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2109        memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2110        /* Initialize skew-specific constants */
2111        err = ei->get_invariants(hw);
2112        if (err)
2113                goto err_sw_init;
2114
2115        /* setup the private structure */
2116        err = igb_sw_init(adapter);
2117        if (err)
2118                goto err_sw_init;
2119
2120        igb_get_bus_info_pcie(hw);
2121
2122        hw->phy.autoneg_wait_to_complete = false;
2123
2124        /* Copper options */
2125        if (hw->phy.media_type == e1000_media_type_copper) {
2126                hw->phy.mdix = AUTO_ALL_MODES;
2127                hw->phy.disable_polarity_correction = false;
2128                hw->phy.ms_type = e1000_ms_hw_default;
2129        }
2130
2131        if (igb_check_reset_block(hw))
2132                dev_info(&pdev->dev,
2133                        "PHY reset is blocked due to SOL/IDER session.\n");
2134
2135        /* features is initialized to 0 in allocation, it might have bits
2136         * set by igb_sw_init so we should use an or instead of an
2137         * assignment.
2138         */
2139        netdev->features |= NETIF_F_SG |
2140                            NETIF_F_IP_CSUM |
2141                            NETIF_F_IPV6_CSUM |
2142                            NETIF_F_TSO |
2143                            NETIF_F_TSO6 |
2144                            NETIF_F_RXHASH |
2145                            NETIF_F_RXCSUM |
2146                            NETIF_F_HW_VLAN_CTAG_RX |
2147                            NETIF_F_HW_VLAN_CTAG_TX;
2148
2149        /* copy netdev features into list of user selectable features */
2150        netdev->hw_features |= netdev->features;
2151        netdev->hw_features |= NETIF_F_RXALL;
2152
2153        /* set this bit last since it cannot be part of hw_features */
2154        netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2155
2156        netdev->vlan_features |= NETIF_F_TSO |
2157                                 NETIF_F_TSO6 |
2158                                 NETIF_F_IP_CSUM |
2159                                 NETIF_F_IPV6_CSUM |
2160                                 NETIF_F_SG;
2161
2162        netdev->priv_flags |= IFF_SUPP_NOFCS;
2163
2164        if (pci_using_dac) {
2165                netdev->features |= NETIF_F_HIGHDMA;
2166                netdev->vlan_features |= NETIF_F_HIGHDMA;
2167        }
2168
2169        if (hw->mac.type >= e1000_82576) {
2170                netdev->hw_features |= NETIF_F_SCTP_CSUM;
2171                netdev->features |= NETIF_F_SCTP_CSUM;
2172        }
2173
2174        netdev->priv_flags |= IFF_UNICAST_FLT;
2175
2176        adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2177
2178        /* before reading the NVM, reset the controller to put the device in a
2179         * known good starting state
2180         */
2181        hw->mac.ops.reset_hw(hw);
2182
2183        /* make sure the NVM is good , i211/i210 parts can have special NVM
2184         * that doesn't contain a checksum
2185         */
2186        switch (hw->mac.type) {
2187        case e1000_i210:
2188        case e1000_i211:
2189                if (igb_get_flash_presence_i210(hw)) {
2190                        if (hw->nvm.ops.validate(hw) < 0) {
2191                                dev_err(&pdev->dev,
2192                                        "The NVM Checksum Is Not Valid\n");
2193                                err = -EIO;
2194                                goto err_eeprom;
2195                        }
2196                }
2197                break;
2198        default:
2199                if (hw->nvm.ops.validate(hw) < 0) {
2200                        dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2201                        err = -EIO;
2202                        goto err_eeprom;
2203                }
2204                break;
2205        }
2206
2207        /* copy the MAC address out of the NVM */
2208        if (hw->mac.ops.read_mac_addr(hw))
2209                dev_err(&pdev->dev, "NVM Read Error\n");
2210
2211        memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2212
2213        if (!is_valid_ether_addr(netdev->dev_addr)) {
2214                dev_err(&pdev->dev, "Invalid MAC Address\n");
2215                err = -EIO;
2216                goto err_eeprom;
2217        }
2218
2219        /* get firmware version for ethtool -i */
2220        igb_set_fw_version(adapter);
2221
2222        setup_timer(&adapter->watchdog_timer, igb_watchdog,
2223                    (unsigned long) adapter);
2224        setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2225                    (unsigned long) adapter);
2226
2227        INIT_WORK(&adapter->reset_task, igb_reset_task);
2228        INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2229
2230        /* Initialize link properties that are user-changeable */
2231        adapter->fc_autoneg = true;
2232        hw->mac.autoneg = true;
2233        hw->phy.autoneg_advertised = 0x2f;
2234
2235        hw->fc.requested_mode = e1000_fc_default;
2236        hw->fc.current_mode = e1000_fc_default;
2237
2238        igb_validate_mdi_setting(hw);
2239
2240        /* By default, support wake on port A */
2241        if (hw->bus.func == 0)
2242                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2243
2244        /* Check the NVM for wake support on non-port A ports */
2245        if (hw->mac.type >= e1000_82580)
2246                hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2247                                 NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2248                                 &eeprom_data);
2249        else if (hw->bus.func == 1)
2250                hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2251
2252        if (eeprom_data & IGB_EEPROM_APME)
2253                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2254
2255        /* now that we have the eeprom settings, apply the special cases where
2256         * the eeprom may be wrong or the board simply won't support wake on
2257         * lan on a particular port
2258         */
2259        switch (pdev->device) {
2260        case E1000_DEV_ID_82575GB_QUAD_COPPER:
2261                adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2262                break;
2263        case E1000_DEV_ID_82575EB_FIBER_SERDES:
2264        case E1000_DEV_ID_82576_FIBER:
2265        case E1000_DEV_ID_82576_SERDES:
2266                /* Wake events only supported on port A for dual fiber
2267                 * regardless of eeprom setting
2268                 */
2269                if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2270                        adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2271                break;
2272        case E1000_DEV_ID_82576_QUAD_COPPER:
2273        case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2274                /* if quad port adapter, disable WoL on all but port A */
2275                if (global_quad_port_a != 0)
2276                        adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2277                else
2278                        adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2279                /* Reset for multiple quad port adapters */
2280                if (++global_quad_port_a == 4)
2281                        global_quad_port_a = 0;
2282                break;
2283        default:
2284                /* If the device can't wake, don't set software support */
2285                if (!device_can_wakeup(&adapter->pdev->dev))
2286                        adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2287        }
2288
2289        /* initialize the wol settings based on the eeprom settings */
2290        if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2291                adapter->wol |= E1000_WUFC_MAG;
2292
2293        /* Some vendors want WoL disabled by default, but still supported */
2294        if ((hw->mac.type == e1000_i350) &&
2295            (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2296                adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2297                adapter->wol = 0;
2298        }
2299
2300        device_set_wakeup_enable(&adapter->pdev->dev,
2301                                 adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2302
2303        /* reset the hardware with the new settings */
2304        igb_reset(adapter);
2305
2306        /* Init the I2C interface */
2307        err = igb_init_i2c(adapter);
2308        if (err) {
2309                dev_err(&pdev->dev, "failed to init i2c interface\n");
2310                goto err_eeprom;
2311        }
2312
2313        /* let the f/w know that the h/w is now under the control of the
2314         * driver. */
2315        igb_get_hw_control(adapter);
2316
2317        strcpy(netdev->name, "eth%d");
2318        err = register_netdev(netdev);
2319        if (err)
2320                goto err_register;
2321
2322        /* carrier off reporting is important to ethtool even BEFORE open */
2323        netif_carrier_off(netdev);
2324
2325#ifdef CONFIG_IGB_DCA
2326        if (dca_add_requester(&pdev->dev) == 0) {
2327                adapter->flags |= IGB_FLAG_DCA_ENABLED;
2328                dev_info(&pdev->dev, "DCA enabled\n");
2329                igb_setup_dca(adapter);
2330        }
2331
2332#endif
2333#ifdef CONFIG_IGB_HWMON
2334        /* Initialize the thermal sensor on i350 devices. */
2335        if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2336                u16 ets_word;
2337
2338                /* Read the NVM to determine if this i350 device supports an
2339                 * external thermal sensor.
2340                 */
2341                hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2342                if (ets_word != 0x0000 && ets_word != 0xFFFF)
2343                        adapter->ets = true;
2344                else
2345                        adapter->ets = false;
2346                if (igb_sysfs_init(adapter))
2347                        dev_err(&pdev->dev,
2348                                "failed to allocate sysfs resources\n");
2349        } else {
2350                adapter->ets = false;
2351        }
2352#endif
2353        /* do hw tstamp init after resetting */
2354        igb_ptp_init(adapter);
2355
2356        dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2357        /* print bus type/speed/width info, not applicable to i354 */
2358        if (hw->mac.type != e1000_i354) {
2359                dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2360                         netdev->name,
2361                         ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2362                          (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2363                           "unknown"),
2364                         ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2365                          "Width x4" :
2366                          (hw->bus.width == e1000_bus_width_pcie_x2) ?
2367                          "Width x2" :
2368                          (hw->bus.width == e1000_bus_width_pcie_x1) ?
2369                          "Width x1" : "unknown"), netdev->dev_addr);
2370        }
2371
2372        if ((hw->mac.type >= e1000_i210 ||
2373             igb_get_flash_presence_i210(hw))) {
2374                ret_val = igb_read_part_string(hw, part_str,
2375                                               E1000_PBANUM_LENGTH);
2376        } else {
2377                ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2378        }
2379
2380        if (ret_val)
2381                strcpy(part_str, "Unknown");
2382        dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2383        dev_info(&pdev->dev,
2384                "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2385                adapter->msix_entries ? "MSI-X" :
2386                (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2387                adapter->num_rx_queues, adapter->num_tx_queues);
2388        switch (hw->mac.type) {
2389        case e1000_i350:
2390        case e1000_i210:
2391        case e1000_i211:
2392                igb_set_eee_i350(hw);
2393                break;
2394        case e1000_i354:
2395                if (hw->phy.media_type == e1000_media_type_copper) {
2396                        if ((rd32(E1000_CTRL_EXT) &
2397                            E1000_CTRL_EXT_LINK_MODE_SGMII))
2398                                igb_set_eee_i354(hw);
2399                }
2400                break;
2401        default:
2402                break;
2403        }
2404
2405        pm_runtime_put_noidle(&pdev->dev);
2406        return 0;
2407
2408err_register:
2409        igb_release_hw_control(adapter);
2410        memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2411err_eeprom:
2412        if (!igb_check_reset_block(hw))
2413                igb_reset_phy(hw);
2414
2415        if (hw->flash_address)
2416                iounmap(hw->flash_address);
2417err_sw_init:
2418        igb_clear_interrupt_scheme(adapter);
2419        iounmap(hw->hw_addr);
2420err_ioremap:
2421        free_netdev(netdev);
2422err_alloc_etherdev:
2423        pci_release_selected_regions(pdev,
2424                                     pci_select_bars(pdev, IORESOURCE_MEM));
2425err_pci_reg:
2426err_dma:
2427        pci_disable_device(pdev);
2428        return err;
2429}
2430
2431#ifdef CONFIG_PCI_IOV
2432static int  igb_disable_sriov(struct pci_dev *pdev)
2433{
2434        struct net_device *netdev = pci_get_drvdata(pdev);
2435        struct igb_adapter *adapter = netdev_priv(netdev);
2436        struct e1000_hw *hw = &adapter->hw;
2437
2438        /* reclaim resources allocated to VFs */
2439        if (adapter->vf_data) {
2440                /* disable iov and allow time for transactions to clear */
2441                if (pci_vfs_assigned(pdev)) {
2442                        dev_warn(&pdev->dev,
2443                                 "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2444                        return -EPERM;
2445                } else {
2446                        pci_disable_sriov(pdev);
2447                        msleep(500);
2448                }
2449
2450                kfree(adapter->vf_data);
2451                adapter->vf_data = NULL;
2452                adapter->vfs_allocated_count = 0;
2453                wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2454                wrfl();
2455                msleep(100);
2456                dev_info(&pdev->dev, "IOV Disabled\n");
2457
2458                /* Re-enable DMA Coalescing flag since IOV is turned off */
2459                adapter->flags |= IGB_FLAG_DMAC;
2460        }
2461
2462        return 0;
2463}
2464
2465static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2466{
2467        struct net_device *netdev = pci_get_drvdata(pdev);
2468        struct igb_adapter *adapter = netdev_priv(netdev);
2469        int old_vfs = pci_num_vf(pdev);
2470        int err = 0;
2471        int i;
2472
2473        if (!adapter->msix_entries) {
2474                err = -EPERM;
2475                goto out;
2476        }
2477
2478        if (!num_vfs)
2479                goto out;
2480        else if (old_vfs && old_vfs == num_vfs)
2481                goto out;
2482        else if (old_vfs && old_vfs != num_vfs)
2483                err = igb_disable_sriov(pdev);
2484
2485        if (err)
2486                goto out;
2487
2488        if (num_vfs > 7) {
2489                err = -EPERM;
2490                goto out;
2491        }
2492
2493        adapter->vfs_allocated_count = num_vfs;
2494
2495        adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2496                                sizeof(struct vf_data_storage), GFP_KERNEL);
2497
2498        /* if allocation failed then we do not support SR-IOV */
2499        if (!adapter->vf_data) {
2500                adapter->vfs_allocated_count = 0;
2501                dev_err(&pdev->dev,
2502                        "Unable to allocate memory for VF Data Storage\n");
2503                err = -ENOMEM;
2504                goto out;
2505        }
2506
2507        err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2508        if (err)
2509                goto err_out;
2510
2511        dev_info(&pdev->dev, "%d VFs allocated\n",
2512                 adapter->vfs_allocated_count);
2513        for (i = 0; i < adapter->vfs_allocated_count; i++)
2514                igb_vf_configure(adapter, i);
2515
2516        /* DMA Coalescing is not supported in IOV mode. */
2517        adapter->flags &= ~IGB_FLAG_DMAC;
2518        goto out;
2519
2520err_out:
2521        kfree(adapter->vf_data);
2522        adapter->vf_data = NULL;
2523        adapter->vfs_allocated_count = 0;
2524out:
2525        return err;
2526}
2527
2528#endif
2529/**
2530 *  igb_remove_i2c - Cleanup  I2C interface
2531 *  @adapter: pointer to adapter structure
2532 **/
2533static void igb_remove_i2c(struct igb_adapter *adapter)
2534{
2535        /* free the adapter bus structure */
2536        i2c_del_adapter(&adapter->i2c_adap);
2537}
2538
2539/**
2540 *  igb_remove - Device Removal Routine
2541 *  @pdev: PCI device information struct
2542 *
2543 *  igb_remove is called by the PCI subsystem to alert the driver
2544 *  that it should release a PCI device.  The could be caused by a
2545 *  Hot-Plug event, or because the driver is going to be removed from
2546 *  memory.
2547 **/
2548static void igb_remove(struct pci_dev *pdev)
2549{
2550        struct net_device *netdev = pci_get_drvdata(pdev);
2551        struct igb_adapter *adapter = netdev_priv(netdev);
2552        struct e1000_hw *hw = &adapter->hw;
2553
2554        pm_runtime_get_noresume(&pdev->dev);
2555#ifdef CONFIG_IGB_HWMON
2556        igb_sysfs_exit(adapter);
2557#endif
2558        igb_remove_i2c(adapter);
2559        igb_ptp_stop(adapter);
2560        /* The watchdog timer may be rescheduled, so explicitly
2561         * disable watchdog from being rescheduled.
2562         */
2563        set_bit(__IGB_DOWN, &adapter->state);
2564        del_timer_sync(&adapter->watchdog_timer);
2565        del_timer_sync(&adapter->phy_info_timer);
2566
2567        cancel_work_sync(&adapter->reset_task);
2568        cancel_work_sync(&adapter->watchdog_task);
2569
2570#ifdef CONFIG_IGB_DCA
2571        if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2572                dev_info(&pdev->dev, "DCA disabled\n");
2573                dca_remove_requester(&pdev->dev);
2574                adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2575                wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2576        }
2577#endif
2578
2579        /* Release control of h/w to f/w.  If f/w is AMT enabled, this
2580         * would have already happened in close and is redundant.
2581         */
2582        igb_release_hw_control(adapter);
2583
2584        unregister_netdev(netdev);
2585
2586        igb_clear_interrupt_scheme(adapter);
2587
2588#ifdef CONFIG_PCI_IOV
2589        igb_disable_sriov(pdev);
2590#endif
2591
2592        iounmap(hw->hw_addr);
2593        if (hw->flash_address)
2594                iounmap(hw->flash_address);
2595        pci_release_selected_regions(pdev,
2596                                     pci_select_bars(pdev, IORESOURCE_MEM));
2597
2598        kfree(adapter->shadow_vfta);
2599        free_netdev(netdev);
2600
2601        pci_disable_pcie_error_reporting(pdev);
2602
2603        pci_disable_device(pdev);
2604}
2605
2606/**
2607 *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
2608 *  @adapter: board private structure to initialize
2609 *
2610 *  This function initializes the vf specific data storage and then attempts to
2611 *  allocate the VFs.  The reason for ordering it this way is because it is much
2612 *  mor expensive time wise to disable SR-IOV than it is to allocate and free
2613 *  the memory for the VFs.
2614 **/
2615static void igb_probe_vfs(struct igb_adapter *adapter)
2616{
2617#ifdef CONFIG_PCI_IOV
2618        struct pci_dev *pdev = adapter->pdev;
2619        struct e1000_hw *hw = &adapter->hw;
2620
2621        /* Virtualization features not supported on i210 family. */
2622        if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
2623                return;
2624
2625        pci_sriov_set_totalvfs(pdev, 7);
2626        igb_enable_sriov(pdev, max_vfs);
2627
2628#endif /* CONFIG_PCI_IOV */
2629}
2630
2631static void igb_init_queue_configuration(struct igb_adapter *adapter)
2632{
2633        struct e1000_hw *hw = &adapter->hw;
2634        u32 max_rss_queues;
2635
2636        /* Determine the maximum number of RSS queues supported. */
2637        switch (hw->mac.type) {
2638        case e1000_i211:
2639                max_rss_queues = IGB_MAX_RX_QUEUES_I211;
2640                break;
2641        case e1000_82575:
2642        case e1000_i210:
2643                max_rss_queues = IGB_MAX_RX_QUEUES_82575;
2644                break;
2645        case e1000_i350:
2646                /* I350 cannot do RSS and SR-IOV at the same time */
2647                if (!!adapter->vfs_allocated_count) {
2648                        max_rss_queues = 1;
2649                        break;
2650                }
2651                /* fall through */
2652        case e1000_82576:
2653                if (!!adapter->vfs_allocated_count) {
2654                        max_rss_queues = 2;
2655                        break;
2656                }
2657                /* fall through */
2658        case e1000_82580:
2659        case e1000_i354:
2660        default:
2661                max_rss_queues = IGB_MAX_RX_QUEUES;
2662                break;
2663        }
2664
2665        adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
2666
2667        /* Determine if we need to pair queues. */
2668        switch (hw->mac.type) {
2669        case e1000_82575:
2670        case e1000_i211:
2671                /* Device supports enough interrupts without queue pairing. */
2672                break;
2673        case e1000_82576:
2674                /* If VFs are going to be allocated with RSS queues then we
2675                 * should pair the queues in order to conserve interrupts due
2676                 * to limited supply.
2677                 */
2678                if ((adapter->rss_queues > 1) &&
2679                    (adapter->vfs_allocated_count > 6))
2680                        adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2681                /* fall through */
2682        case e1000_82580:
2683        case e1000_i350:
2684        case e1000_i354:
2685        case e1000_i210:
2686        default:
2687                /* If rss_queues > half of max_rss_queues, pair the queues in
2688                 * order to conserve interrupts due to limited supply.
2689                 */
2690                if (adapter->rss_queues > (max_rss_queues / 2))
2691                        adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
2692                break;
2693        }
2694}
2695
2696/**
2697 *  igb_sw_init - Initialize general software structures (struct igb_adapter)
2698 *  @adapter: board private structure to initialize
2699 *
2700 *  igb_sw_init initializes the Adapter private data structure.
2701 *  Fields are initialized based on PCI device information and
2702 *  OS network device settings (MTU size).
2703 **/
2704static int igb_sw_init(struct igb_adapter *adapter)
2705{
2706        struct e1000_hw *hw = &adapter->hw;
2707        struct net_device *netdev = adapter->netdev;
2708        struct pci_dev *pdev = adapter->pdev;
2709
2710        pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
2711
2712        /* set default ring sizes */
2713        adapter->tx_ring_count = IGB_DEFAULT_TXD;
2714        adapter->rx_ring_count = IGB_DEFAULT_RXD;
2715
2716        /* set default ITR values */
2717        adapter->rx_itr_setting = IGB_DEFAULT_ITR;
2718        adapter->tx_itr_setting = IGB_DEFAULT_ITR;
2719
2720        /* set default work limits */
2721        adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
2722
2723        adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
2724                                  VLAN_HLEN;
2725        adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
2726
2727        spin_lock_init(&adapter->stats64_lock);
2728#ifdef CONFIG_PCI_IOV
2729        switch (hw->mac.type) {
2730        case e1000_82576:
2731        case e1000_i350:
2732                if (max_vfs > 7) {
2733                        dev_warn(&pdev->dev,
2734                                 "Maximum of 7 VFs per PF, using max\n");
2735                        max_vfs = adapter->vfs_allocated_count = 7;
2736                } else
2737                        adapter->vfs_allocated_count = max_vfs;
2738                if (adapter->vfs_allocated_count)
2739                        dev_warn(&pdev->dev,
2740                                 "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
2741                break;
2742        default:
2743                break;
2744        }
2745#endif /* CONFIG_PCI_IOV */
2746
2747        igb_init_queue_configuration(adapter);
2748
2749        /* Setup and initialize a copy of the hw vlan table array */
2750        adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
2751                                       GFP_ATOMIC);
2752
2753        /* This call may decrease the number of queues */
2754        if (igb_init_interrupt_scheme(adapter, true)) {
2755                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
2756                return -ENOMEM;
2757        }
2758
2759        igb_probe_vfs(adapter);
2760
2761        /* Explicitly disable IRQ since the NIC can be in any state. */
2762        igb_irq_disable(adapter);
2763
2764        if (hw->mac.type >= e1000_i350)
2765                adapter->flags &= ~IGB_FLAG_DMAC;
2766
2767        set_bit(__IGB_DOWN, &adapter->state);
2768        return 0;
2769}
2770
2771/**
2772 *  igb_open - Called when a network interface is made active
2773 *  @netdev: network interface device structure
2774 *
2775 *  Returns 0 on success, negative value on failure
2776 *
2777 *  The open entry point is called when a network interface is made
2778 *  active by the system (IFF_UP).  At this point all resources needed
2779 *  for transmit and receive operations are allocated, the interrupt
2780 *  handler is registered with the OS, the watchdog timer is started,
2781 *  and the stack is notified that the interface is ready.
2782 **/
2783static int __igb_open(struct net_device *netdev, bool resuming)
2784{
2785        struct igb_adapter *adapter = netdev_priv(netdev);
2786        struct e1000_hw *hw = &adapter->hw;
2787        struct pci_dev *pdev = adapter->pdev;
2788        int err;
2789        int i;
2790
2791        /* disallow open during test */
2792        if (test_bit(__IGB_TESTING, &adapter->state)) {
2793                WARN_ON(resuming);
2794                return -EBUSY;
2795        }
2796
2797        if (!resuming)
2798                pm_runtime_get_sync(&pdev->dev);
2799
2800        netif_carrier_off(netdev);
2801
2802        /* allocate transmit descriptors */
2803        err = igb_setup_all_tx_resources(adapter);
2804        if (err)
2805                goto err_setup_tx;
2806
2807        /* allocate receive descriptors */
2808        err = igb_setup_all_rx_resources(adapter);
2809        if (err)
2810                goto err_setup_rx;
2811
2812        igb_power_up_link(adapter);
2813
2814        /* before we allocate an interrupt, we must be ready to handle it.
2815         * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
2816         * as soon as we call pci_request_irq, so we have to setup our
2817         * clean_rx handler before we do so.
2818         */
2819        igb_configure(adapter);
2820
2821        err = igb_request_irq(adapter);
2822        if (err)
2823                goto err_req_irq;
2824
2825        /* Notify the stack of the actual queue counts. */
2826        err = netif_set_real_num_tx_queues(adapter->netdev,
2827                                           adapter->num_tx_queues);
2828        if (err)
2829                goto err_set_queues;
2830
2831        err = netif_set_real_num_rx_queues(adapter->netdev,
2832                                           adapter->num_rx_queues);
2833        if (err)
2834                goto err_set_queues;
2835
2836        /* From here on the code is the same as igb_up() */
2837        clear_bit(__IGB_DOWN, &adapter->state);
2838
2839        for (i = 0; i < adapter->num_q_vectors; i++)
2840                napi_enable(&(adapter->q_vector[i]->napi));
2841
2842        /* Clear any pending interrupts. */
2843        rd32(E1000_ICR);
2844
2845        igb_irq_enable(adapter);
2846
2847        /* notify VFs that reset has been completed */
2848        if (adapter->vfs_allocated_count) {
2849                u32 reg_data = rd32(E1000_CTRL_EXT);
2850                reg_data |= E1000_CTRL_EXT_PFRSTD;
2851                wr32(E1000_CTRL_EXT, reg_data);
2852        }
2853
2854        netif_tx_start_all_queues(netdev);
2855
2856        if (!resuming)
2857                pm_runtime_put(&pdev->dev);
2858
2859        /* start the watchdog. */
2860        hw->mac.get_link_status = 1;
2861        schedule_work(&adapter->watchdog_task);
2862
2863        return 0;
2864
2865err_set_queues:
2866        igb_free_irq(adapter);
2867err_req_irq:
2868        igb_release_hw_control(adapter);
2869        igb_power_down_link(adapter);
2870        igb_free_all_rx_resources(adapter);
2871err_setup_rx:
2872        igb_free_all_tx_resources(adapter);
2873err_setup_tx:
2874        igb_reset(adapter);
2875        if (!resuming)
2876                pm_runtime_put(&pdev->dev);
2877
2878        return err;
2879}
2880
2881static int igb_open(struct net_device *netdev)
2882{
2883        return __igb_open(netdev, false);
2884}
2885
2886/**
2887 *  igb_close - Disables a network interface
2888 *  @netdev: network interface device structure
2889 *
2890 *  Returns 0, this is not allowed to fail
2891 *
2892 *  The close entry point is called when an interface is de-activated
2893 *  by the OS.  The hardware is still under the driver's control, but
2894 *  needs to be disabled.  A global MAC reset is issued to stop the
2895 *  hardware, and all transmit and receive resources are freed.
2896 **/
2897static int __igb_close(struct net_device *netdev, bool suspending)
2898{
2899        struct igb_adapter *adapter = netdev_priv(netdev);
2900        struct pci_dev *pdev = adapter->pdev;
2901
2902        WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
2903
2904        if (!suspending)
2905                pm_runtime_get_sync(&pdev->dev);
2906
2907        igb_down(adapter);
2908        igb_free_irq(adapter);
2909
2910        igb_free_all_tx_resources(adapter);
2911        igb_free_all_rx_resources(adapter);
2912
2913        if (!suspending)
2914                pm_runtime_put_sync(&pdev->dev);
2915        return 0;
2916}
2917
2918static int igb_close(struct net_device *netdev)
2919{
2920        return __igb_close(netdev, false);
2921}
2922
2923/**
2924 *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
2925 *  @tx_ring: tx descriptor ring (for a specific queue) to setup
2926 *
2927 *  Return 0 on success, negative on failure
2928 **/
2929int igb_setup_tx_resources(struct igb_ring *tx_ring)
2930{
2931        struct device *dev = tx_ring->dev;
2932        int size;
2933
2934        size = sizeof(struct igb_tx_buffer) * tx_ring->count;
2935
2936        tx_ring->tx_buffer_info = vzalloc(size);
2937        if (!tx_ring->tx_buffer_info)
2938                goto err;
2939
2940        /* round up to nearest 4K */
2941        tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
2942        tx_ring->size = ALIGN(tx_ring->size, 4096);
2943
2944        tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
2945                                           &tx_ring->dma, GFP_KERNEL);
2946        if (!tx_ring->desc)
2947                goto err;
2948
2949        tx_ring->next_to_use = 0;
2950        tx_ring->next_to_clean = 0;
2951
2952        return 0;
2953
2954err:
2955        vfree(tx_ring->tx_buffer_info);
2956        tx_ring->tx_buffer_info = NULL;
2957        dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
2958        return -ENOMEM;
2959}
2960
2961/**
2962 *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
2963 *                               (Descriptors) for all queues
2964 *  @adapter: board private structure
2965 *
2966 *  Return 0 on success, negative on failure
2967 **/
2968static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
2969{
2970        struct pci_dev *pdev = adapter->pdev;
2971        int i, err = 0;
2972
2973        for (i = 0; i < adapter->num_tx_queues; i++) {
2974                err = igb_setup_tx_resources(adapter->tx_ring[i]);
2975                if (err) {
2976                        dev_err(&pdev->dev,
2977                                "Allocation for Tx Queue %u failed\n", i);
2978                        for (i--; i >= 0; i--)
2979                                igb_free_tx_resources(adapter->tx_ring[i]);
2980                        break;
2981                }
2982        }
2983
2984        return err;
2985}
2986
2987/**
2988 *  igb_setup_tctl - configure the transmit control registers
2989 *  @adapter: Board private structure
2990 **/
2991void igb_setup_tctl(struct igb_adapter *adapter)
2992{
2993        struct e1000_hw *hw = &adapter->hw;
2994        u32 tctl;
2995
2996        /* disable queue 0 which is enabled by default on 82575 and 82576 */
2997        wr32(E1000_TXDCTL(0), 0);
2998
2999        /* Program the Transmit Control Register */
3000        tctl = rd32(E1000_TCTL);
3001        tctl &= ~E1000_TCTL_CT;
3002        tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3003                (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3004
3005        igb_config_collision_dist(hw);
3006
3007        /* Enable transmits */
3008        tctl |= E1000_TCTL_EN;
3009
3010        wr32(E1000_TCTL, tctl);
3011}
3012
3013/**
3014 *  igb_configure_tx_ring - Configure transmit ring after Reset
3015 *  @adapter: board private structure
3016 *  @ring: tx ring to configure
3017 *
3018 *  Configure a transmit ring after a reset.
3019 **/
3020void igb_configure_tx_ring(struct igb_adapter *adapter,
3021                           struct igb_ring *ring)
3022{
3023        struct e1000_hw *hw = &adapter->hw;
3024        u32 txdctl = 0;
3025        u64 tdba = ring->dma;
3026        int reg_idx = ring->reg_idx;
3027
3028        /* disable the queue */
3029        wr32(E1000_TXDCTL(reg_idx), 0);
3030        wrfl();
3031        mdelay(10);
3032
3033        wr32(E1000_TDLEN(reg_idx),
3034             ring->count * sizeof(union e1000_adv_tx_desc));
3035        wr32(E1000_TDBAL(reg_idx),
3036             tdba & 0x00000000ffffffffULL);
3037        wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3038
3039        ring->tail = hw->hw_addr + E1000_TDT(reg_idx);
3040        wr32(E1000_TDH(reg_idx), 0);
3041        writel(0, ring->tail);
3042
3043        txdctl |= IGB_TX_PTHRESH;
3044        txdctl |= IGB_TX_HTHRESH << 8;
3045        txdctl |= IGB_TX_WTHRESH << 16;
3046
3047        txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3048        wr32(E1000_TXDCTL(reg_idx), txdctl);
3049}
3050
3051/**
3052 *  igb_configure_tx - Configure transmit Unit after Reset
3053 *  @adapter: board private structure
3054 *
3055 *  Configure the Tx unit of the MAC after a reset.
3056 **/
3057static void igb_configure_tx(struct igb_adapter *adapter)
3058{
3059        int i;
3060
3061        for (i = 0; i < adapter->num_tx_queues; i++)
3062                igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3063}
3064
3065/**
3066 *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
3067 *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
3068 *
3069 *  Returns 0 on success, negative on failure
3070 **/
3071int igb_setup_rx_resources(struct igb_ring *rx_ring)
3072{
3073        struct device *dev = rx_ring->dev;
3074        int size;
3075
3076        size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3077
3078        rx_ring->rx_buffer_info = vzalloc(size);
3079        if (!rx_ring->rx_buffer_info)
3080                goto err;
3081
3082        /* Round up to nearest 4K */
3083        rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3084        rx_ring->size = ALIGN(rx_ring->size, 4096);
3085
3086        rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3087                                           &rx_ring->dma, GFP_KERNEL);
3088        if (!rx_ring->desc)
3089                goto err;
3090
3091        rx_ring->next_to_alloc = 0;
3092        rx_ring->next_to_clean = 0;
3093        rx_ring->next_to_use = 0;
3094
3095        return 0;
3096
3097err:
3098        vfree(rx_ring->rx_buffer_info);
3099        rx_ring->rx_buffer_info = NULL;
3100        dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3101        return -ENOMEM;
3102}
3103
3104/**
3105 *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
3106 *                               (Descriptors) for all queues
3107 *  @adapter: board private structure
3108 *
3109 *  Return 0 on success, negative on failure
3110 **/
3111static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3112{
3113        struct pci_dev *pdev = adapter->pdev;
3114        int i, err = 0;
3115
3116        for (i = 0; i < adapter->num_rx_queues; i++) {
3117                err = igb_setup_rx_resources(adapter->rx_ring[i]);
3118                if (err) {
3119                        dev_err(&pdev->dev,
3120                                "Allocation for Rx Queue %u failed\n", i);
3121                        for (i--; i >= 0; i--)
3122                                igb_free_rx_resources(adapter->rx_ring[i]);
3123                        break;
3124                }
3125        }
3126
3127        return err;
3128}
3129
3130/**
3131 *  igb_setup_mrqc - configure the multiple receive queue control registers
3132 *  @adapter: Board private structure
3133 **/
3134static void igb_setup_mrqc(struct igb_adapter *adapter)
3135{
3136        struct e1000_hw *hw = &adapter->hw;
3137        u32 mrqc, rxcsum;
3138        u32 j, num_rx_queues;
3139        static const u32 rsskey[10] = { 0xDA565A6D, 0xC20E5B25, 0x3D256741,
3140                                        0xB08FA343, 0xCB2BCAD0, 0xB4307BAE,
3141                                        0xA32DCB77, 0x0CF23080, 0x3BB7426A,
3142                                        0xFA01ACBE };
3143
3144        /* Fill out hash function seeds */
3145        for (j = 0; j < 10; j++)
3146                wr32(E1000_RSSRK(j), rsskey[j]);
3147
3148        num_rx_queues = adapter->rss_queues;
3149
3150        switch (hw->mac.type) {
3151        case e1000_82576:
3152                /* 82576 supports 2 RSS queues for SR-IOV */
3153                if (adapter->vfs_allocated_count)
3154                        num_rx_queues = 2;
3155                break;
3156        default:
3157                break;
3158        }
3159
3160        if (adapter->rss_indir_tbl_init != num_rx_queues) {
3161                for (j = 0; j < IGB_RETA_SIZE; j++)
3162                        adapter->rss_indir_tbl[j] = (j * num_rx_queues) / IGB_RETA_SIZE;
3163                adapter->rss_indir_tbl_init = num_rx_queues;
3164        }
3165        igb_write_rss_indir_tbl(adapter);
3166
3167        /* Disable raw packet checksumming so that RSS hash is placed in
3168         * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3169         * offloads as they are enabled by default
3170         */
3171        rxcsum = rd32(E1000_RXCSUM);
3172        rxcsum |= E1000_RXCSUM_PCSD;
3173
3174        if (adapter->hw.mac.type >= e1000_82576)
3175                /* Enable Receive Checksum Offload for SCTP */
3176                rxcsum |= E1000_RXCSUM_CRCOFL;
3177
3178        /* Don't need to set TUOFL or IPOFL, they default to 1 */
3179        wr32(E1000_RXCSUM, rxcsum);
3180
3181        /* Generate RSS hash based on packet types, TCP/UDP
3182         * port numbers and/or IPv4/v6 src and dst addresses
3183         */
3184        mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3185               E1000_MRQC_RSS_FIELD_IPV4_TCP |
3186               E1000_MRQC_RSS_FIELD_IPV6 |
3187               E1000_MRQC_RSS_FIELD_IPV6_TCP |
3188               E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3189
3190        if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3191                mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3192        if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3193                mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3194
3195        /* If VMDq is enabled then we set the appropriate mode for that, else
3196         * we default to RSS so that an RSS hash is calculated per packet even
3197         * if we are only using one queue
3198         */
3199        if (adapter->vfs_allocated_count) {
3200                if (hw->mac.type > e1000_82575) {
3201                        /* Set the default pool for the PF's first queue */
3202                        u32 vtctl = rd32(E1000_VT_CTL);
3203                        vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3204                                   E1000_VT_CTL_DISABLE_DEF_POOL);
3205                        vtctl |= adapter->vfs_allocated_count <<
3206                                E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3207                        wr32(E1000_VT_CTL, vtctl);
3208                }
3209                if (adapter->rss_queues > 1)
3210                        mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
3211                else
3212                        mrqc |= E1000_MRQC_ENABLE_VMDQ;
3213        } else {
3214                if (hw->mac.type != e1000_i211)
3215                        mrqc |= E1000_MRQC_ENABLE_RSS_4Q;
3216        }
3217        igb_vmm_control(adapter);
3218
3219        wr32(E1000_MRQC, mrqc);
3220}
3221
3222/**
3223 *  igb_setup_rctl - configure the receive control registers
3224 *  @adapter: Board private structure
3225 **/
3226void igb_setup_rctl(struct igb_adapter *adapter)
3227{
3228        struct e1000_hw *hw = &adapter->hw;
3229        u32 rctl;
3230
3231        rctl = rd32(E1000_RCTL);
3232
3233        rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3234        rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3235
3236        rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3237                (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3238
3239        /* enable stripping of CRC. It's unlikely this will break BMC
3240         * redirection as it did with e1000. Newer features require
3241         * that the HW strips the CRC.
3242         */
3243        rctl |= E1000_RCTL_SECRC;
3244
3245        /* disable store bad packets and clear size bits. */
3246        rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3247
3248        /* enable LPE to prevent packets larger than max_frame_size */
3249        rctl |= E1000_RCTL_LPE;
3250
3251        /* disable queue 0 to prevent tail write w/o re-config */
3252        wr32(E1000_RXDCTL(0), 0);
3253
3254        /* Attention!!!  For SR-IOV PF driver operations you must enable
3255         * queue drop for all VF and PF queues to prevent head of line blocking
3256         * if an un-trusted VF does not provide descriptors to hardware.
3257         */
3258        if (adapter->vfs_allocated_count) {
3259                /* set all queue drop enable bits */
3260                wr32(E1000_QDE, ALL_QUEUES);
3261        }
3262
3263        /* This is useful for sniffing bad packets. */
3264        if (adapter->netdev->features & NETIF_F_RXALL) {
3265                /* UPE and MPE will be handled by normal PROMISC logic
3266                 * in e1000e_set_rx_mode
3267                 */
3268                rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3269                         E1000_RCTL_BAM | /* RX All Bcast Pkts */
3270                         E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3271
3272                rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
3273                          E1000_RCTL_DPF | /* Allow filtered pause */
3274                          E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3275                /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3276                 * and that breaks VLANs.
3277                 */
3278        }
3279
3280        wr32(E1000_RCTL, rctl);
3281}
3282
3283static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3284                                   int vfn)
3285{
3286        struct e1000_hw *hw = &adapter->hw;
3287        u32 vmolr;
3288
3289        /* if it isn't the PF check to see if VFs are enabled and
3290         * increase the size to support vlan tags
3291         */
3292        if (vfn < adapter->vfs_allocated_count &&
3293            adapter->vf_data[vfn].vlans_enabled)
3294                size += VLAN_TAG_SIZE;
3295
3296        vmolr = rd32(E1000_VMOLR(vfn));
3297        vmolr &= ~E1000_VMOLR_RLPML_MASK;
3298        vmolr |= size | E1000_VMOLR_LPE;
3299        wr32(E1000_VMOLR(vfn), vmolr);
3300
3301        return 0;
3302}
3303
3304/**
3305 *  igb_rlpml_set - set maximum receive packet size
3306 *  @adapter: board private structure
3307 *
3308 *  Configure maximum receivable packet size.
3309 **/
3310static void igb_rlpml_set(struct igb_adapter *adapter)
3311{
3312        u32 max_frame_size = adapter->max_frame_size;
3313        struct e1000_hw *hw = &adapter->hw;
3314        u16 pf_id = adapter->vfs_allocated_count;
3315
3316        if (pf_id) {
3317                igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
3318                /* If we're in VMDQ or SR-IOV mode, then set global RLPML
3319                 * to our max jumbo frame size, in case we need to enable
3320                 * jumbo frames on one of the rings later.
3321                 * This will not pass over-length frames into the default
3322                 * queue because it's gated by the VMOLR.RLPML.
3323                 */
3324                max_frame_size = MAX_JUMBO_FRAME_SIZE;
3325        }
3326
3327        wr32(E1000_RLPML, max_frame_size);
3328}
3329
3330static inline void igb_set_vmolr(struct igb_adapter *adapter,
3331                                 int vfn, bool aupe)
3332{
3333        struct e1000_hw *hw = &adapter->hw;
3334        u32 vmolr;
3335
3336        /* This register exists only on 82576 and newer so if we are older then
3337         * we should exit and do nothing
3338         */
3339        if (hw->mac.type < e1000_82576)
3340                return;
3341
3342        vmolr = rd32(E1000_VMOLR(vfn));
3343        vmolr |= E1000_VMOLR_STRVLAN; /* Strip vlan tags */
3344        if (aupe)
3345                vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3346        else
3347                vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3348
3349        /* clear all bits that might not be set */
3350        vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3351
3352        if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3353                vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3354        /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3355         * multicast packets
3356         */
3357        if (vfn <= adapter->vfs_allocated_count)
3358                vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3359
3360        wr32(E1000_VMOLR(vfn), vmolr);
3361}
3362
3363/**
3364 *  igb_configure_rx_ring - Configure a receive ring after Reset
3365 *  @adapter: board private structure
3366 *  @ring: receive ring to be configured
3367 *
3368 *  Configure the Rx unit of the MAC after a reset.
3369 **/
3370void igb_configure_rx_ring(struct igb_adapter *adapter,
3371                           struct igb_ring *ring)
3372{
3373        struct e1000_hw *hw = &adapter->hw;
3374        u64 rdba = ring->dma;
3375        int reg_idx = ring->reg_idx;
3376        u32 srrctl = 0, rxdctl = 0;
3377
3378        /* disable the queue */
3379        wr32(E1000_RXDCTL(reg_idx), 0);
3380
3381        /* Set DMA base address registers */
3382        wr32(E1000_RDBAL(reg_idx),
3383             rdba & 0x00000000ffffffffULL);
3384        wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3385        wr32(E1000_RDLEN(reg_idx),
3386             ring->count * sizeof(union e1000_adv_rx_desc));
3387
3388        /* initialize head and tail */
3389        ring->tail = hw->hw_addr + E1000_RDT(reg_idx);
3390        wr32(E1000_RDH(reg_idx), 0);
3391        writel(0, ring->tail);
3392
3393        /* set descriptor configuration */
3394        srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3395        srrctl |= IGB_RX_BUFSZ >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3396        srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3397        if (hw->mac.type >= e1000_82580)
3398                srrctl |= E1000_SRRCTL_TIMESTAMP;
3399        /* Only set Drop Enable if we are supporting multiple queues */
3400        if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3401                srrctl |= E1000_SRRCTL_DROP_EN;
3402
3403        wr32(E1000_SRRCTL(reg_idx), srrctl);
3404
3405        /* set filtering for VMDQ pools */
3406        igb_set_vmolr(adapter, reg_idx & 0x7, true);
3407
3408        rxdctl |= IGB_RX_PTHRESH;
3409        rxdctl |= IGB_RX_HTHRESH << 8;
3410        rxdctl |= IGB_RX_WTHRESH << 16;
3411
3412        /* enable receive descriptor fetching */
3413        rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3414        wr32(E1000_RXDCTL(reg_idx), rxdctl);
3415}
3416
3417/**
3418 *  igb_configure_rx - Configure receive Unit after Reset
3419 *  @adapter: board private structure
3420 *
3421 *  Configure the Rx unit of the MAC after a reset.
3422 **/
3423static void igb_configure_rx(struct igb_adapter *adapter)
3424{
3425        int i;
3426
3427        /* set UTA to appropriate mode */
3428        igb_set_uta(adapter);
3429
3430        /* set the correct pool for the PF default MAC address in entry 0 */
3431        igb_rar_set_qsel(adapter, adapter->hw.mac.addr, 0,
3432                         adapter->vfs_allocated_count);
3433
3434        /* Setup the HW Rx Head and Tail Descriptor Pointers and
3435         * the Base and Length of the Rx Descriptor Ring
3436         */
3437        for (i = 0; i < adapter->num_rx_queues; i++)
3438                igb_configure_rx_ring(adapter, adapter->rx_ring[i]);
3439}
3440
3441/**
3442 *  igb_free_tx_resources - Free Tx Resources per Queue
3443 *  @tx_ring: Tx descriptor ring for a specific queue
3444 *
3445 *  Free all transmit software resources
3446 **/
3447void igb_free_tx_resources(struct igb_ring *tx_ring)
3448{
3449        igb_clean_tx_ring(tx_ring);
3450
3451        vfree(tx_ring->tx_buffer_info);
3452        tx_ring->tx_buffer_info = NULL;
3453
3454        /* if not set, then don't free */
3455        if (!tx_ring->desc)
3456                return;
3457
3458        dma_free_coherent(tx_ring->dev, tx_ring->size,
3459                          tx_ring->desc, tx_ring->dma);
3460
3461        tx_ring->desc = NULL;
3462}
3463
3464/**
3465 *  igb_free_all_tx_resources - Free Tx Resources for All Queues
3466 *  @adapter: board private structure
3467 *
3468 *  Free all transmit software resources
3469 **/
3470static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3471{
3472        int i;
3473
3474        for (i = 0; i < adapter->num_tx_queues; i++)
3475                igb_free_tx_resources(adapter->tx_ring[i]);
3476}
3477
3478void igb_unmap_and_free_tx_resource(struct igb_ring *ring,
3479                                    struct igb_tx_buffer *tx_buffer)
3480{
3481        if (tx_buffer->skb) {
3482                dev_kfree_skb_any(tx_buffer->skb);
3483                if (dma_unmap_len(tx_buffer, len))
3484                        dma_unmap_single(ring->dev,
3485                                         dma_unmap_addr(tx_buffer, dma),
3486                                         dma_unmap_len(tx_buffer, len),
3487                                         DMA_TO_DEVICE);
3488        } else if (dma_unmap_len(tx_buffer, len)) {
3489                dma_unmap_page(ring->dev,
3490                               dma_unmap_addr(tx_buffer, dma),
3491                               dma_unmap_len(tx_buffer, len),
3492                               DMA_TO_DEVICE);
3493        }
3494        tx_buffer->next_to_watch = NULL;
3495        tx_buffer->skb = NULL;
3496        dma_unmap_len_set(tx_buffer, len, 0);
3497        /* buffer_info must be completely set up in the transmit path */
3498}
3499
3500/**
3501 *  igb_clean_tx_ring - Free Tx Buffers
3502 *  @tx_ring: ring to be cleaned
3503 **/
3504static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3505{
3506        struct igb_tx_buffer *buffer_info;
3507        unsigned long size;
3508        u16 i;
3509
3510        if (!tx_ring->tx_buffer_info)
3511                return;
3512        /* Free all the Tx ring sk_buffs */
3513
3514        for (i = 0; i < tx_ring->count; i++) {
3515                buffer_info = &tx_ring->tx_buffer_info[i];
3516                igb_unmap_and_free_tx_resource(tx_ring, buffer_info);
3517        }
3518
3519        netdev_tx_reset_queue(txring_txq(tx_ring));
3520
3521        size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3522        memset(tx_ring->tx_buffer_info, 0, size);
3523
3524        /* Zero out the descriptor ring */
3525        memset(tx_ring->desc, 0, tx_ring->size);
3526
3527        tx_ring->next_to_use = 0;
3528        tx_ring->next_to_clean = 0;
3529}
3530
3531/**
3532 *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
3533 *  @adapter: board private structure
3534 **/
3535static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3536{
3537        int i;
3538
3539        for (i = 0; i < adapter->num_tx_queues; i++)
3540                igb_clean_tx_ring(adapter->tx_ring[i]);
3541}
3542
3543/**
3544 *  igb_free_rx_resources - Free Rx Resources
3545 *  @rx_ring: ring to clean the resources from
3546 *
3547 *  Free all receive software resources
3548 **/
3549void igb_free_rx_resources(struct igb_ring *rx_ring)
3550{
3551        igb_clean_rx_ring(rx_ring);
3552
3553        vfree(rx_ring->rx_buffer_info);
3554        rx_ring->rx_buffer_info = NULL;
3555
3556        /* if not set, then don't free */
3557        if (!rx_ring->desc)
3558                return;
3559
3560        dma_free_coherent(rx_ring->dev, rx_ring->size,
3561                          rx_ring->desc, rx_ring->dma);
3562
3563        rx_ring->desc = NULL;
3564}
3565
3566/**
3567 *  igb_free_all_rx_resources - Free Rx Resources for All Queues
3568 *  @adapter: board private structure
3569 *
3570 *  Free all receive software resources
3571 **/
3572static void igb_free_all_rx_resources(struct igb_adapter *adapter)
3573{
3574        int i;
3575
3576        for (i = 0; i < adapter->num_rx_queues; i++)
3577                igb_free_rx_resources(adapter->rx_ring[i]);
3578}
3579
3580/**
3581 *  igb_clean_rx_ring - Free Rx Buffers per Queue
3582 *  @rx_ring: ring to free buffers from
3583 **/
3584static void igb_clean_rx_ring(struct igb_ring *rx_ring)
3585{
3586        unsigned long size;
3587        u16 i;
3588
3589        if (rx_ring->skb)
3590                dev_kfree_skb(rx_ring->skb);
3591        rx_ring->skb = NULL;
3592
3593        if (!rx_ring->rx_buffer_info)
3594                return;
3595
3596        /* Free all the Rx ring sk_buffs */
3597        for (i = 0; i < rx_ring->count; i++) {
3598                struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
3599
3600                if (!buffer_info->page)
3601                        continue;
3602
3603                dma_unmap_page(rx_ring->dev,
3604                               buffer_info->dma,
3605                               PAGE_SIZE,
3606                               DMA_FROM_DEVICE);
3607                __free_page(buffer_info->page);
3608
3609                buffer_info->page = NULL;
3610        }
3611
3612        size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3613        memset(rx_ring->rx_buffer_info, 0, size);
3614
3615        /* Zero out the descriptor ring */
3616        memset(rx_ring->desc, 0, rx_ring->size);
3617
3618        rx_ring->next_to_alloc = 0;
3619        rx_ring->next_to_clean = 0;
3620        rx_ring->next_to_use = 0;
3621}
3622
3623/**
3624 *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
3625 *  @adapter: board private structure
3626 **/
3627static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
3628{
3629        int i;
3630
3631        for (i = 0; i < adapter->num_rx_queues; i++)
3632                igb_clean_rx_ring(adapter->rx_ring[i]);
3633}
3634
3635/**
3636 *  igb_set_mac - Change the Ethernet Address of the NIC
3637 *  @netdev: network interface device structure
3638 *  @p: pointer to an address structure
3639 *
3640 *  Returns 0 on success, negative on failure
3641 **/
3642static int igb_set_mac(struct net_device *netdev, void *p)
3643{
3644        struct igb_adapter *adapter = netdev_priv(netdev);
3645        struct e1000_hw *hw = &adapter->hw;
3646        struct sockaddr *addr = p;
3647
3648        if (!is_valid_ether_addr(addr->sa_data))
3649                return -EADDRNOTAVAIL;
3650
3651        memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3652        memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3653
3654        /* set the correct pool for the new PF MAC address in entry 0 */
3655        igb_rar_set_qsel(adapter, hw->mac.addr, 0,
3656                         adapter->vfs_allocated_count);
3657
3658        return 0;
3659}
3660
3661/**
3662 *  igb_write_mc_addr_list - write multicast addresses to MTA
3663 *  @netdev: network interface device structure
3664 *
3665 *  Writes multicast address list to the MTA hash table.
3666 *  Returns: -ENOMEM on failure
3667 *           0 on no addresses written
3668 *           X on writing X addresses to MTA
3669 **/
3670static int igb_write_mc_addr_list(struct net_device *netdev)
3671{
3672        struct igb_adapter *adapter = netdev_priv(netdev);
3673        struct e1000_hw *hw = &adapter->hw;
3674        struct netdev_hw_addr *ha;
3675        u8  *mta_list;
3676        int i;
3677
3678        if (netdev_mc_empty(netdev)) {
3679                /* nothing to program, so clear mc list */
3680                igb_update_mc_addr_list(hw, NULL, 0);
3681                igb_restore_vf_multicasts(adapter);
3682                return 0;
3683        }
3684
3685        mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
3686        if (!mta_list)
3687                return -ENOMEM;
3688
3689        /* The shared function expects a packed array of only addresses. */
3690        i = 0;
3691        netdev_for_each_mc_addr(ha, netdev)
3692                memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
3693
3694        igb_update_mc_addr_list(hw, mta_list, i);
3695        kfree(mta_list);
3696
3697        return netdev_mc_count(netdev);
3698}
3699
3700/**
3701 *  igb_write_uc_addr_list - write unicast addresses to RAR table
3702 *  @netdev: network interface device structure
3703 *
3704 *  Writes unicast address list to the RAR table.
3705 *  Returns: -ENOMEM on failure/insufficient address space
3706 *           0 on no addresses written
3707 *           X on writing X addresses to the RAR table
3708 **/
3709static int igb_write_uc_addr_list(struct net_device *netdev)
3710{
3711        struct igb_adapter *adapter = netdev_priv(netdev);
3712        struct e1000_hw *hw = &adapter->hw;
3713        unsigned int vfn = adapter->vfs_allocated_count;
3714        unsigned int rar_entries = hw->mac.rar_entry_count - (vfn + 1);
3715        int count = 0;
3716
3717        /* return ENOMEM indicating insufficient memory for addresses */
3718        if (netdev_uc_count(netdev) > rar_entries)
3719                return -ENOMEM;
3720
3721        if (!netdev_uc_empty(netdev) && rar_entries) {
3722                struct netdev_hw_addr *ha;
3723
3724                netdev_for_each_uc_addr(ha, netdev) {
3725                        if (!rar_entries)
3726                                break;
3727                        igb_rar_set_qsel(adapter, ha->addr,
3728                                         rar_entries--,
3729                                         vfn);
3730                        count++;
3731                }
3732        }
3733        /* write the addresses in reverse order to avoid write combining */
3734        for (; rar_entries > 0 ; rar_entries--) {
3735                wr32(E1000_RAH(rar_entries), 0);
3736                wr32(E1000_RAL(rar_entries), 0);
3737        }
3738        wrfl();
3739
3740        return count;
3741}
3742
3743/**
3744 *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
3745 *  @netdev: network interface device structure
3746 *
3747 *  The set_rx_mode entry point is called whenever the unicast or multicast
3748 *  address lists or the network interface flags are updated.  This routine is
3749 *  responsible for configuring the hardware for proper unicast, multicast,
3750 *  promiscuous mode, and all-multi behavior.
3751 **/
3752static void igb_set_rx_mode(struct net_device *netdev)
3753{
3754        struct igb_adapter *adapter = netdev_priv(netdev);
3755        struct e1000_hw *hw = &adapter->hw;
3756        unsigned int vfn = adapter->vfs_allocated_count;
3757        u32 rctl, vmolr = 0;
3758        int count;
3759
3760        /* Check for Promiscuous and All Multicast modes */
3761        rctl = rd32(E1000_RCTL);
3762
3763        /* clear the effected bits */
3764        rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE | E1000_RCTL_VFE);
3765
3766        if (netdev->flags & IFF_PROMISC) {
3767                /* retain VLAN HW filtering if in VT mode */
3768                if (adapter->vfs_allocated_count)
3769                        rctl |= E1000_RCTL_VFE;
3770                rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
3771                vmolr |= (E1000_VMOLR_ROPE | E1000_VMOLR_MPME);
3772        } else {
3773                if (netdev->flags & IFF_ALLMULTI) {
3774                        rctl |= E1000_RCTL_MPE;
3775                        vmolr |= E1000_VMOLR_MPME;
3776                } else {
3777                        /* Write addresses to the MTA, if the attempt fails
3778                         * then we should just turn on promiscuous mode so
3779                         * that we can at least receive multicast traffic
3780                         */
3781                        count = igb_write_mc_addr_list(netdev);
3782                        if (count < 0) {
3783                                rctl |= E1000_RCTL_MPE;
3784                                vmolr |= E1000_VMOLR_MPME;
3785                        } else if (count) {
3786                                vmolr |= E1000_VMOLR_ROMPE;
3787                        }
3788                }
3789                /* Write addresses to available RAR registers, if there is not
3790                 * sufficient space to store all the addresses then enable
3791                 * unicast promiscuous mode
3792                 */
3793                count = igb_write_uc_addr_list(netdev);
3794                if (count < 0) {
3795                        rctl |= E1000_RCTL_UPE;
3796                        vmolr |= E1000_VMOLR_ROPE;
3797                }
3798                rctl |= E1000_RCTL_VFE;
3799        }
3800        wr32(E1000_RCTL, rctl);
3801
3802        /* In order to support SR-IOV and eventually VMDq it is necessary to set
3803         * the VMOLR to enable the appropriate modes.  Without this workaround
3804         * we will have issues with VLAN tag stripping not being done for frames
3805         * that are only arriving because we are the default pool
3806         */
3807        if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
3808                return;
3809
3810        vmolr |= rd32(E1000_VMOLR(vfn)) &
3811                 ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
3812        wr32(E1000_VMOLR(vfn), vmolr);
3813        igb_restore_vf_multicasts(adapter);
3814}
3815
3816static void igb_check_wvbr(struct igb_adapter *adapter)
3817{
3818        struct e1000_hw *hw = &adapter->hw;
3819        u32 wvbr = 0;
3820
3821        switch (hw->mac.type) {
3822        case e1000_82576:
3823        case e1000_i350:
3824                if (!(wvbr = rd32(E1000_WVBR)))
3825                        return;
3826                break;
3827        default:
3828                break;
3829        }
3830
3831        adapter->wvbr |= wvbr;
3832}
3833
3834#define IGB_STAGGERED_QUEUE_OFFSET 8
3835
3836static void igb_spoof_check(struct igb_adapter *adapter)
3837{
3838        int j;
3839
3840        if (!adapter->wvbr)
3841                return;
3842
3843        for(j = 0; j < adapter->vfs_allocated_count; j++) {
3844                if (adapter->wvbr & (1 << j) ||
3845                    adapter->wvbr & (1 << (j + IGB_STAGGERED_QUEUE_OFFSET))) {
3846                        dev_warn(&adapter->pdev->dev,
3847                                "Spoof event(s) detected on VF %d\n", j);
3848                        adapter->wvbr &=
3849                                ~((1 << j) |
3850                                  (1 << (j + IGB_STAGGERED_QUEUE_OFFSET)));
3851                }
3852        }
3853}
3854
3855/* Need to wait a few seconds after link up to get diagnostic information from
3856 * the phy
3857 */
3858static void igb_update_phy_info(unsigned long data)
3859{
3860        struct igb_adapter *adapter = (struct igb_adapter *) data;
3861        igb_get_phy_info(&adapter->hw);
3862}
3863
3864/**
3865 *  igb_has_link - check shared code for link and determine up/down
3866 *  @adapter: pointer to driver private info
3867 **/
3868bool igb_has_link(struct igb_adapter *adapter)
3869{
3870        struct e1000_hw *hw = &adapter->hw;
3871        bool link_active = false;
3872
3873        /* get_link_status is set on LSC (link status) interrupt or
3874         * rx sequence error interrupt.  get_link_status will stay
3875         * false until the e1000_check_for_link establishes link
3876         * for copper adapters ONLY
3877         */
3878        switch (hw->phy.media_type) {
3879        case e1000_media_type_copper:
3880                if (!hw->mac.get_link_status)
3881                        return true;
3882        case e1000_media_type_internal_serdes:
3883                hw->mac.ops.check_for_link(hw);
3884                link_active = !hw->mac.get_link_status;
3885                break;
3886        default:
3887        case e1000_media_type_unknown:
3888                break;
3889        }
3890
3891        if (((hw->mac.type == e1000_i210) ||
3892             (hw->mac.type == e1000_i211)) &&
3893             (hw->phy.id == I210_I_PHY_ID)) {
3894                if (!netif_carrier_ok(adapter->netdev)) {
3895                        adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
3896                } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
3897                        adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
3898                        adapter->link_check_timeout = jiffies;
3899                }
3900        }
3901
3902        return link_active;
3903}
3904
3905static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
3906{
3907        bool ret = false;
3908        u32 ctrl_ext, thstat;
3909
3910        /* check for thermal sensor event on i350 copper only */
3911        if (hw->mac.type == e1000_i350) {
3912                thstat = rd32(E1000_THSTAT);
3913                ctrl_ext = rd32(E1000_CTRL_EXT);
3914
3915                if ((hw->phy.media_type == e1000_media_type_copper) &&
3916                    !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
3917                        ret = !!(thstat & event);
3918        }
3919
3920        return ret;
3921}
3922
3923/**
3924 *  igb_watchdog - Timer Call-back
3925 *  @data: pointer to adapter cast into an unsigned long
3926 **/
3927static void igb_watchdog(unsigned long data)
3928{
3929        struct igb_adapter *adapter = (struct igb_adapter *)data;
3930        /* Do the rest outside of interrupt context */
3931        schedule_work(&adapter->watchdog_task);
3932}
3933
3934static void igb_watchdog_task(struct work_struct *work)
3935{
3936        struct igb_adapter *adapter = container_of(work,
3937                                                   struct igb_adapter,
3938                                                   watchdog_task);
3939        struct e1000_hw *hw = &adapter->hw;
3940        struct e1000_phy_info *phy = &hw->phy;
3941        struct net_device *netdev = adapter->netdev;
3942        u32 link;
3943        int i;
3944
3945        link = igb_has_link(adapter);
3946
3947        if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
3948                if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
3949                        adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
3950                else
3951                        link = false;
3952        }
3953
3954        if (link) {
3955                /* Cancel scheduled suspend requests. */
3956                pm_runtime_resume(netdev->dev.parent);
3957
3958                if (!netif_carrier_ok(netdev)) {
3959                        u32 ctrl;
3960                        hw->mac.ops.get_speed_and_duplex(hw,
3961                                                         &adapter->link_speed,
3962                                                         &adapter->link_duplex);
3963
3964                        ctrl = rd32(E1000_CTRL);
3965                        /* Links status message must follow this format */
3966                        printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s "
3967                               "Duplex, Flow Control: %s\n",
3968                               netdev->name,
3969                               adapter->link_speed,
3970                               adapter->link_duplex == FULL_DUPLEX ?
3971                               "Full" : "Half",
3972                               (ctrl & E1000_CTRL_TFCE) &&
3973                               (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
3974                               (ctrl & E1000_CTRL_RFCE) ?  "RX" :
3975                               (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
3976
3977                        /* check if SmartSpeed worked */
3978                        igb_check_downshift(hw);
3979                        if (phy->speed_downgraded)
3980                                netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
3981
3982                        /* check for thermal sensor event */
3983                        if (igb_thermal_sensor_event(hw,
3984                            E1000_THSTAT_LINK_THROTTLE)) {
3985                                netdev_info(netdev, "The network adapter link "
3986                                            "speed was downshifted because it "
3987                                            "overheated\n");
3988                        }
3989
3990                        /* adjust timeout factor according to speed/duplex */
3991                        adapter->tx_timeout_factor = 1;
3992                        switch (adapter->link_speed) {
3993                        case SPEED_10:
3994                                adapter->tx_timeout_factor = 14;
3995                                break;
3996                        case SPEED_100:
3997                                /* maybe add some timeout factor ? */
3998                                break;
3999                        }
4000
4001                        netif_carrier_on(netdev);
4002
4003                        igb_ping_all_vfs(adapter);
4004                        igb_check_vf_rate_limit(adapter);
4005
4006                        /* link state has changed, schedule phy info update */
4007                        if (!test_bit(__IGB_DOWN, &adapter->state))
4008                                mod_timer(&adapter->phy_info_timer,
4009                                          round_jiffies(jiffies + 2 * HZ));
4010                }
4011        } else {
4012                if (netif_carrier_ok(netdev)) {
4013                        adapter->link_speed = 0;
4014                        adapter->link_duplex = 0;
4015
4016                        /* check for thermal sensor event */
4017                        if (igb_thermal_sensor_event(hw,
4018                            E1000_THSTAT_PWR_DOWN)) {
4019                                netdev_err(netdev, "The network adapter was "
4020                                           "stopped because it overheated\n");
4021                        }
4022
4023                        /* Links status message must follow this format */
4024                        printk(KERN_INFO "igb: %s NIC Link is Down\n",
4025                               netdev->name);
4026                        netif_carrier_off(netdev);
4027
4028                        igb_ping_all_vfs(adapter);
4029
4030                        /* link state has changed, schedule phy info update */
4031                        if (!test_bit(__IGB_DOWN, &adapter->state))
4032                                mod_timer(&adapter->phy_info_timer,
4033                                          round_jiffies(jiffies + 2 * HZ));
4034
4035                        pm_schedule_suspend(netdev->dev.parent,
4036                                            MSEC_PER_SEC * 5);
4037                }
4038        }
4039
4040        spin_lock(&adapter->stats64_lock);
4041        igb_update_stats(adapter, &adapter->stats64);
4042        spin_unlock(&adapter->stats64_lock);
4043
4044        for (i = 0; i < adapter->num_tx_queues; i++) {
4045                struct igb_ring *tx_ring = adapter->tx_ring[i];
4046                if (!netif_carrier_ok(netdev)) {
4047                        /* We've lost link, so the controller stops DMA,
4048                         * but we've got queued Tx work that's never going
4049                         * to get done, so reset controller to flush Tx.
4050                         * (Do the reset outside of interrupt context).
4051                         */
4052                        if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4053                                adapter->tx_timeout_count++;
4054                                schedule_work(&adapter->reset_task);
4055                                /* return immediately since reset is imminent */
4056                                return;
4057                        }
4058                }
4059
4060                /* Force detection of hung controller every watchdog period */
4061                set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4062        }
4063
4064        /* Cause software interrupt to ensure Rx ring is cleaned */
4065        if (adapter->msix_entries) {
4066                u32 eics = 0;
4067                for (i = 0; i < adapter->num_q_vectors; i++)
4068                        eics |= adapter->q_vector[i]->eims_value;
4069                wr32(E1000_EICS, eics);
4070        } else {
4071                wr32(E1000_ICS, E1000_ICS_RXDMT0);
4072        }
4073
4074        igb_spoof_check(adapter);
4075        igb_ptp_rx_hang(adapter);
4076
4077        /* Reset the timer */
4078        if (!test_bit(__IGB_DOWN, &adapter->state)) {
4079                if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4080                        mod_timer(&adapter->watchdog_timer,
4081                                  round_jiffies(jiffies +  HZ));
4082                else
4083                        mod_timer(&adapter->watchdog_timer,
4084                                  round_jiffies(jiffies + 2 * HZ));
4085        }
4086}
4087
4088enum latency_range {
4089        lowest_latency = 0,
4090        low_latency = 1,
4091        bulk_latency = 2,
4092        latency_invalid = 255
4093};
4094
4095/**
4096 *  igb_update_ring_itr - update the dynamic ITR value based on packet size
4097 *  @q_vector: pointer to q_vector
4098 *
4099 *  Stores a new ITR value based on strictly on packet size.  This
4100 *  algorithm is less sophisticated than that used in igb_update_itr,
4101 *  due to the difficulty of synchronizing statistics across multiple
4102 *  receive rings.  The divisors and thresholds used by this function
4103 *  were determined based on theoretical maximum wire speed and testing
4104 *  data, in order to minimize response time while increasing bulk
4105 *  throughput.
4106 *  This functionality is controlled by the InterruptThrottleRate module
4107 *  parameter (see igb_param.c)
4108 *  NOTE:  This function is called only when operating in a multiqueue
4109 *         receive environment.
4110 **/
4111static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4112{
4113        int new_val = q_vector->itr_val;
4114        int avg_wire_size = 0;
4115        struct igb_adapter *adapter = q_vector->adapter;
4116        unsigned int packets;
4117
4118        /* For non-gigabit speeds, just fix the interrupt rate at 4000
4119         * ints/sec - ITR timer value of 120 ticks.
4120         */
4121        if (adapter->link_speed != SPEED_1000) {
4122                new_val = IGB_4K_ITR;
4123                goto set_itr_val;
4124        }
4125
4126        packets = q_vector->rx.total_packets;
4127        if (packets)
4128                avg_wire_size = q_vector->rx.total_bytes / packets;
4129
4130        packets = q_vector->tx.total_packets;
4131        if (packets)
4132                avg_wire_size = max_t(u32, avg_wire_size,
4133                                      q_vector->tx.total_bytes / packets);
4134
4135        /* if avg_wire_size isn't set no work was done */
4136        if (!avg_wire_size)
4137                goto clear_counts;
4138
4139        /* Add 24 bytes to size to account for CRC, preamble, and gap */
4140        avg_wire_size += 24;
4141
4142        /* Don't starve jumbo frames */
4143        avg_wire_size = min(avg_wire_size, 3000);
4144
4145        /* Give a little boost to mid-size frames */
4146        if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4147                new_val = avg_wire_size / 3;
4148        else
4149                new_val = avg_wire_size / 2;
4150
4151        /* conservative mode (itr 3) eliminates the lowest_latency setting */
4152        if (new_val < IGB_20K_ITR &&
4153            ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4154             (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4155                new_val = IGB_20K_ITR;
4156
4157set_itr_val:
4158        if (new_val != q_vector->itr_val) {
4159                q_vector->itr_val = new_val;
4160                q_vector->set_itr = 1;
4161        }
4162clear_counts:
4163        q_vector->rx.total_bytes = 0;
4164        q_vector->rx.total_packets = 0;
4165        q_vector->tx.total_bytes = 0;
4166        q_vector->tx.total_packets = 0;
4167}
4168
4169/**
4170 *  igb_update_itr - update the dynamic ITR value based on statistics
4171 *  @q_vector: pointer to q_vector
4172 *  @ring_container: ring info to update the itr for
4173 *
4174 *  Stores a new ITR value based on packets and byte
4175 *  counts during the last interrupt.  The advantage of per interrupt
4176 *  computation is faster updates and more accurate ITR for the current
4177 *  traffic pattern.  Constants in this function were computed
4178 *  based on theoretical maximum wire speed and thresholds were set based
4179 *  on testing data as well as attempting to minimize response time
4180 *  while increasing bulk throughput.
4181 *  this functionality is controlled by the InterruptThrottleRate module
4182 *  parameter (see igb_param.c)
4183 *  NOTE:  These calculations are only valid when operating in a single-
4184 *         queue environment.
4185 **/
4186static void igb_update_itr(struct igb_q_vector *q_vector,
4187                           struct igb_ring_container *ring_container)
4188{
4189        unsigned int packets = ring_container->total_packets;
4190        unsigned int bytes = ring_container->total_bytes;
4191        u8 itrval = ring_container->itr;
4192
4193        /* no packets, exit with status unchanged */
4194        if (packets == 0)
4195                return;
4196
4197        switch (itrval) {
4198        case lowest_latency:
4199                /* handle TSO and jumbo frames */
4200                if (bytes/packets > 8000)
4201                        itrval = bulk_latency;
4202                else if ((packets < 5) && (bytes > 512))
4203                        itrval = low_latency;
4204                break;
4205        case low_latency:  /* 50 usec aka 20000 ints/s */
4206                if (bytes > 10000) {
4207                        /* this if handles the TSO accounting */
4208                        if (bytes/packets > 8000) {
4209                                itrval = bulk_latency;
4210                        } else if ((packets < 10) || ((bytes/packets) > 1200)) {
4211                                itrval = bulk_latency;
4212                        } else if ((packets > 35)) {
4213                                itrval = lowest_latency;
4214                        }
4215                } else if (bytes/packets > 2000) {
4216                        itrval = bulk_latency;
4217                } else if (packets <= 2 && bytes < 512) {
4218                        itrval = lowest_latency;
4219                }
4220                break;
4221        case bulk_latency: /* 250 usec aka 4000 ints/s */
4222                if (bytes > 25000) {
4223                        if (packets > 35)
4224                                itrval = low_latency;
4225                } else if (bytes < 1500) {
4226                        itrval = low_latency;
4227                }
4228                break;
4229        }
4230
4231        /* clear work counters since we have the values we need */
4232        ring_container->total_bytes = 0;
4233        ring_container->total_packets = 0;
4234
4235        /* write updated itr to ring container */
4236        ring_container->itr = itrval;
4237}
4238
4239static void igb_set_itr(struct igb_q_vector *q_vector)
4240{
4241        struct igb_adapter *adapter = q_vector->adapter;
4242        u32 new_itr = q_vector->itr_val;
4243        u8 current_itr = 0;
4244
4245        /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4246        if (adapter->link_speed != SPEED_1000) {
4247                current_itr = 0;
4248                new_itr = IGB_4K_ITR;
4249                goto set_itr_now;
4250        }
4251
4252        igb_update_itr(q_vector, &q_vector->tx);
4253        igb_update_itr(q_vector, &q_vector->rx);
4254
4255        current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4256
4257        /* conservative mode (itr 3) eliminates the lowest_latency setting */
4258        if (current_itr == lowest_latency &&
4259            ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4260             (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4261                current_itr = low_latency;
4262
4263        switch (current_itr) {
4264        /* counts and packets in update_itr are dependent on these numbers */
4265        case lowest_latency:
4266                new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4267                break;
4268        case low_latency:
4269                new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4270                break;
4271        case bulk_latency:
4272                new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
4273                break;
4274        default:
4275                break;
4276        }
4277
4278set_itr_now:
4279        if (new_itr != q_vector->itr_val) {
4280                /* this attempts to bias the interrupt rate towards Bulk
4281                 * by adding intermediate steps when interrupt rate is
4282                 * increasing
4283                 */
4284                new_itr = new_itr > q_vector->itr_val ?
4285                          max((new_itr * q_vector->itr_val) /
4286                          (new_itr + (q_vector->itr_val >> 2)),
4287                          new_itr) : new_itr;
4288                /* Don't write the value here; it resets the adapter's
4289                 * internal timer, and causes us to delay far longer than
4290                 * we should between interrupts.  Instead, we write the ITR
4291                 * value at the beginning of the next interrupt so the timing
4292                 * ends up being correct.
4293                 */
4294                q_vector->itr_val = new_itr;
4295                q_vector->set_itr = 1;
4296        }
4297}
4298
4299static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4300                            u32 type_tucmd, u32 mss_l4len_idx)
4301{
4302        struct e1000_adv_tx_context_desc *context_desc;
4303        u16 i = tx_ring->next_to_use;
4304
4305        context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4306
4307        i++;
4308        tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4309
4310        /* set bits to identify this as an advanced context descriptor */
4311        type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4312
4313        /* For 82575, context index must be unique per ring. */
4314        if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4315                mss_l4len_idx |= tx_ring->reg_idx << 4;
4316
4317        context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
4318        context_desc->seqnum_seed       = 0;
4319        context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
4320        context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
4321}
4322
4323static int igb_tso(struct igb_ring *tx_ring,
4324                   struct igb_tx_buffer *first,
4325                   u8 *hdr_len)
4326{
4327        struct sk_buff *skb = first->skb;
4328        u32 vlan_macip_lens, type_tucmd;
4329        u32 mss_l4len_idx, l4len;
4330
4331        if (skb->ip_summed != CHECKSUM_PARTIAL)
4332                return 0;
4333
4334        if (!skb_is_gso(skb))
4335                return 0;
4336
4337        if (skb_header_cloned(skb)) {
4338                int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4339                if (err)
4340                        return err;
4341        }
4342
4343        /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
4344        type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
4345
4346        if (first->protocol == __constant_htons(ETH_P_IP)) {
4347                struct iphdr *iph = ip_hdr(skb);
4348                iph->tot_len = 0;
4349                iph->check = 0;
4350                tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
4351                                                         iph->daddr, 0,
4352                                                         IPPROTO_TCP,
4353                                                         0);
4354                type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4355                first->tx_flags |= IGB_TX_FLAGS_TSO |
4356                                   IGB_TX_FLAGS_CSUM |
4357                                   IGB_TX_FLAGS_IPV4;
4358        } else if (skb_is_gso_v6(skb)) {
4359                ipv6_hdr(skb)->payload_len = 0;
4360                tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4361                                                       &ipv6_hdr(skb)->daddr,
4362                                                       0, IPPROTO_TCP, 0);
4363                first->tx_flags |= IGB_TX_FLAGS_TSO |
4364                                   IGB_TX_FLAGS_CSUM;
4365        }
4366
4367        /* compute header lengths */
4368        l4len = tcp_hdrlen(skb);
4369        *hdr_len = skb_transport_offset(skb) + l4len;
4370
4371        /* update gso size and bytecount with header size */
4372        first->gso_segs = skb_shinfo(skb)->gso_segs;
4373        first->bytecount += (first->gso_segs - 1) * *hdr_len;
4374
4375        /* MSS L4LEN IDX */
4376        mss_l4len_idx = l4len << E1000_ADVTXD_L4LEN_SHIFT;
4377        mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
4378
4379        /* VLAN MACLEN IPLEN */
4380        vlan_macip_lens = skb_network_header_len(skb);
4381        vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4382        vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4383
4384        igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4385
4386        return 1;
4387}
4388
4389static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
4390{
4391        struct sk_buff *skb = first->skb;
4392        u32 vlan_macip_lens = 0;
4393        u32 mss_l4len_idx = 0;
4394        u32 type_tucmd = 0;
4395
4396        if (skb->ip_summed != CHECKSUM_PARTIAL) {
4397                if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
4398                        return;
4399        } else {
4400                u8 l4_hdr = 0;
4401                switch (first->protocol) {
4402                case __constant_htons(ETH_P_IP):
4403                        vlan_macip_lens |= skb_network_header_len(skb);
4404                        type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
4405                        l4_hdr = ip_hdr(skb)->protocol;
4406                        break;
4407                case __constant_htons(ETH_P_IPV6):
4408                        vlan_macip_lens |= skb_network_header_len(skb);
4409                        l4_hdr = ipv6_hdr(skb)->nexthdr;
4410                        break;
4411                default:
4412                        if (unlikely(net_ratelimit())) {
4413                                dev_warn(tx_ring->dev,
4414                                         "partial checksum but proto=%x!\n",
4415                                         first->protocol);
4416                        }
4417                        break;
4418                }
4419
4420                switch (l4_hdr) {
4421                case IPPROTO_TCP:
4422                        type_tucmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
4423                        mss_l4len_idx = tcp_hdrlen(skb) <<
4424                                        E1000_ADVTXD_L4LEN_SHIFT;
4425                        break;
4426                case IPPROTO_SCTP:
4427                        type_tucmd |= E1000_ADVTXD_TUCMD_L4T_SCTP;
4428                        mss_l4len_idx = sizeof(struct sctphdr) <<
4429                                        E1000_ADVTXD_L4LEN_SHIFT;
4430                        break;
4431                case IPPROTO_UDP:
4432                        mss_l4len_idx = sizeof(struct udphdr) <<
4433                                        E1000_ADVTXD_L4LEN_SHIFT;
4434                        break;
4435                default:
4436                        if (unlikely(net_ratelimit())) {
4437                                dev_warn(tx_ring->dev,
4438                                         "partial checksum but l4 proto=%x!\n",
4439                                         l4_hdr);
4440                        }
4441                        break;
4442                }
4443
4444                /* update TX checksum flag */
4445                first->tx_flags |= IGB_TX_FLAGS_CSUM;
4446        }
4447
4448        vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
4449        vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
4450
4451        igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
4452}
4453
4454#define IGB_SET_FLAG(_input, _flag, _result) \
4455        ((_flag <= _result) ? \
4456         ((u32)(_input & _flag) * (_result / _flag)) : \
4457         ((u32)(_input & _flag) / (_flag / _result)))
4458
4459static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
4460{
4461        /* set type for advanced descriptor with frame checksum insertion */
4462        u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
4463                       E1000_ADVTXD_DCMD_DEXT |
4464                       E1000_ADVTXD_DCMD_IFCS;
4465
4466        /* set HW vlan bit if vlan is present */
4467        cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
4468                                 (E1000_ADVTXD_DCMD_VLE));
4469
4470        /* set segmentation bits for TSO */
4471        cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
4472                                 (E1000_ADVTXD_DCMD_TSE));
4473
4474        /* set timestamp bit if present */
4475        cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
4476                                 (E1000_ADVTXD_MAC_TSTAMP));
4477
4478        /* insert frame checksum */
4479        cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
4480
4481        return cmd_type;
4482}
4483
4484static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
4485                                 union e1000_adv_tx_desc *tx_desc,
4486                                 u32 tx_flags, unsigned int paylen)
4487{
4488        u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
4489
4490        /* 82575 requires a unique index per ring */
4491        if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4492                olinfo_status |= tx_ring->reg_idx << 4;
4493
4494        /* insert L4 checksum */
4495        olinfo_status |= IGB_SET_FLAG(tx_flags,
4496                                      IGB_TX_FLAGS_CSUM,
4497                                      (E1000_TXD_POPTS_TXSM << 8));
4498
4499        /* insert IPv4 checksum */
4500        olinfo_status |= IGB_SET_FLAG(tx_flags,
4501                                      IGB_TX_FLAGS_IPV4,
4502                                      (E1000_TXD_POPTS_IXSM << 8));
4503
4504        tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
4505}
4506
4507static void igb_tx_map(struct igb_ring *tx_ring,
4508                       struct igb_tx_buffer *first,
4509                       const u8 hdr_len)
4510{
4511        struct sk_buff *skb = first->skb;
4512        struct igb_tx_buffer *tx_buffer;
4513        union e1000_adv_tx_desc *tx_desc;
4514        struct skb_frag_struct *frag;
4515        dma_addr_t dma;
4516        unsigned int data_len, size;
4517        u32 tx_flags = first->tx_flags;
4518        u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
4519        u16 i = tx_ring->next_to_use;
4520
4521        tx_desc = IGB_TX_DESC(tx_ring, i);
4522
4523        igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
4524
4525        size = skb_headlen(skb);
4526        data_len = skb->data_len;
4527
4528        dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
4529
4530        tx_buffer = first;
4531
4532        for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
4533                if (dma_mapping_error(tx_ring->dev, dma))
4534                        goto dma_error;
4535
4536                /* record length, and DMA address */
4537                dma_unmap_len_set(tx_buffer, len, size);
4538                dma_unmap_addr_set(tx_buffer, dma, dma);
4539
4540                tx_desc->read.buffer_addr = cpu_to_le64(dma);
4541
4542                while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
4543                        tx_desc->read.cmd_type_len =
4544                                cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
4545
4546                        i++;
4547                        tx_desc++;
4548                        if (i == tx_ring->count) {
4549                                tx_desc = IGB_TX_DESC(tx_ring, 0);
4550                                i = 0;
4551                        }
4552                        tx_desc->read.olinfo_status = 0;
4553
4554                        dma += IGB_MAX_DATA_PER_TXD;
4555                        size -= IGB_MAX_DATA_PER_TXD;
4556
4557                        tx_desc->read.buffer_addr = cpu_to_le64(dma);
4558                }
4559
4560                if (likely(!data_len))
4561                        break;
4562
4563                tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
4564
4565                i++;
4566                tx_desc++;
4567                if (i == tx_ring->count) {
4568                        tx_desc = IGB_TX_DESC(tx_ring, 0);
4569                        i = 0;
4570                }
4571                tx_desc->read.olinfo_status = 0;
4572
4573                size = skb_frag_size(frag);
4574                data_len -= size;
4575
4576                dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
4577                                       size, DMA_TO_DEVICE);
4578
4579                tx_buffer = &tx_ring->tx_buffer_info[i];
4580        }
4581
4582        /* write last descriptor with RS and EOP bits */
4583        cmd_type |= size | IGB_TXD_DCMD;
4584        tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
4585
4586        netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
4587
4588        /* set the timestamp */
4589        first->time_stamp = jiffies;
4590
4591        /* Force memory writes to complete before letting h/w know there
4592         * are new descriptors to fetch.  (Only applicable for weak-ordered
4593         * memory model archs, such as IA-64).
4594         *
4595         * We also need this memory barrier to make certain all of the
4596         * status bits have been updated before next_to_watch is written.
4597         */
4598        wmb();
4599
4600        /* set next_to_watch value indicating a packet is present */
4601        first->next_to_watch = tx_desc;
4602
4603        i++;
4604        if (i == tx_ring->count)
4605                i = 0;
4606
4607        tx_ring->next_to_use = i;
4608
4609        writel(i, tx_ring->tail);
4610
4611        /* we need this if more than one processor can write to our tail
4612         * at a time, it synchronizes IO on IA64/Altix systems
4613         */
4614        mmiowb();
4615
4616        return;
4617
4618dma_error:
4619        dev_err(tx_ring->dev, "TX DMA map failed\n");
4620
4621        /* clear dma mappings for failed tx_buffer_info map */
4622        for (;;) {
4623                tx_buffer = &tx_ring->tx_buffer_info[i];
4624                igb_unmap_and_free_tx_resource(tx_ring, tx_buffer);
4625                if (tx_buffer == first)
4626                        break;
4627                if (i == 0)
4628                        i = tx_ring->count;
4629                i--;
4630        }
4631
4632        tx_ring->next_to_use = i;
4633}
4634
4635static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4636{
4637        struct net_device *netdev = tx_ring->netdev;
4638
4639        netif_stop_subqueue(netdev, tx_ring->queue_index);
4640
4641        /* Herbert's original patch had:
4642         *  smp_mb__after_netif_stop_queue();
4643         * but since that doesn't exist yet, just open code it.
4644         */
4645        smp_mb();
4646
4647        /* We need to check again in a case another CPU has just
4648         * made room available.
4649         */
4650        if (igb_desc_unused(tx_ring) < size)
4651                return -EBUSY;
4652
4653        /* A reprieve! */
4654        netif_wake_subqueue(netdev, tx_ring->queue_index);
4655
4656        u64_stats_update_begin(&tx_ring->tx_syncp2);
4657        tx_ring->tx_stats.restart_queue2++;
4658        u64_stats_update_end(&tx_ring->tx_syncp2);
4659
4660        return 0;
4661}
4662
4663static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
4664{
4665        if (igb_desc_unused(tx_ring) >= size)
4666                return 0;
4667        return __igb_maybe_stop_tx(tx_ring, size);
4668}
4669
4670netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
4671                                struct igb_ring *tx_ring)
4672{
4673        struct igb_tx_buffer *first;
4674        int tso;
4675        u32 tx_flags = 0;
4676        u16 count = TXD_USE_COUNT(skb_headlen(skb));
4677        __be16 protocol = vlan_get_protocol(skb);
4678        u8 hdr_len = 0;
4679
4680        /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
4681         *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
4682         *       + 2 desc gap to keep tail from touching head,
4683         *       + 1 desc for context descriptor,
4684         * otherwise try next time
4685         */
4686        if (NETDEV_FRAG_PAGE_MAX_SIZE > IGB_MAX_DATA_PER_TXD) {
4687                unsigned short f;
4688                for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
4689                        count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
4690        } else {
4691                count += skb_shinfo(skb)->nr_frags;
4692        }
4693
4694        if (igb_maybe_stop_tx(tx_ring, count + 3)) {
4695                /* this is a hard error */
4696                return NETDEV_TX_BUSY;
4697        }
4698
4699        /* record the location of the first descriptor for this packet */
4700        first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
4701        first->skb = skb;
4702        first->bytecount = skb->len;
4703        first->gso_segs = 1;
4704
4705        skb_tx_timestamp(skb);
4706
4707        if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
4708                struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
4709
4710                if (!(adapter->ptp_tx_skb)) {
4711                        skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
4712                        tx_flags |= IGB_TX_FLAGS_TSTAMP;
4713
4714                        adapter->ptp_tx_skb = skb_get(skb);
4715                        adapter->ptp_tx_start = jiffies;
4716                        if (adapter->hw.mac.type == e1000_82576)
4717                                schedule_work(&adapter->ptp_tx_work);
4718                }
4719        }
4720
4721        if (vlan_tx_tag_present(skb)) {
4722                tx_flags |= IGB_TX_FLAGS_VLAN;
4723                tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
4724        }
4725
4726        /* record initial flags and protocol */
4727        first->tx_flags = tx_flags;
4728        first->protocol = protocol;
4729
4730        tso = igb_tso(tx_ring, first, &hdr_len);
4731        if (tso < 0)
4732                goto out_drop;
4733        else if (!tso)
4734                igb_tx_csum(tx_ring, first);
4735
4736        igb_tx_map(tx_ring, first, hdr_len);
4737
4738        /* Make sure there is space in the ring for the next send. */
4739        igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
4740
4741        return NETDEV_TX_OK;
4742
4743out_drop:
4744        igb_unmap_and_free_tx_resource(tx_ring, first);
4745
4746        return NETDEV_TX_OK;
4747}
4748
4749static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
4750                                                    struct sk_buff *skb)
4751{
4752        unsigned int r_idx = skb->queue_mapping;
4753
4754        if (r_idx >= adapter->num_tx_queues)
4755                r_idx = r_idx % adapter->num_tx_queues;
4756
4757        return adapter->tx_ring[r_idx];
4758}
4759
4760static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
4761                                  struct net_device *netdev)
4762{
4763        struct igb_adapter *adapter = netdev_priv(netdev);
4764
4765        if (test_bit(__IGB_DOWN, &adapter->state)) {
4766                dev_kfree_skb_any(skb);
4767                return NETDEV_TX_OK;
4768        }
4769
4770        if (skb->len <= 0) {
4771                dev_kfree_skb_any(skb);
4772                return NETDEV_TX_OK;
4773        }
4774
4775        /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
4776         * in order to meet this minimum size requirement.
4777         */
4778        if (unlikely(skb->len < 17)) {
4779                if (skb_pad(skb, 17 - skb->len))
4780                        return NETDEV_TX_OK;
4781                skb->len = 17;
4782                skb_set_tail_pointer(skb, 17);
4783        }
4784
4785        return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
4786}
4787
4788/**
4789 *  igb_tx_timeout - Respond to a Tx Hang
4790 *  @netdev: network interface device structure
4791 **/
4792static void igb_tx_timeout(struct net_device *netdev)
4793{
4794        struct igb_adapter *adapter = netdev_priv(netdev);
4795        struct e1000_hw *hw = &adapter->hw;
4796
4797        /* Do the reset outside of interrupt context */
4798        adapter->tx_timeout_count++;
4799
4800        if (hw->mac.type >= e1000_82580)
4801                hw->dev_spec._82575.global_device_reset = true;
4802
4803        schedule_work(&adapter->reset_task);
4804        wr32(E1000_EICS,
4805             (adapter->eims_enable_mask & ~adapter->eims_other));
4806}
4807
4808static void igb_reset_task(struct work_struct *work)
4809{
4810        struct igb_adapter *adapter;
4811        adapter = container_of(work, struct igb_adapter, reset_task);
4812
4813        igb_dump(adapter);
4814        netdev_err(adapter->netdev, "Reset adapter\n");
4815        igb_reinit_locked(adapter);
4816}
4817
4818/**
4819 *  igb_get_stats64 - Get System Network Statistics
4820 *  @netdev: network interface device structure
4821 *  @stats: rtnl_link_stats64 pointer
4822 **/
4823static struct rtnl_link_stats64 *igb_get_stats64(struct net_device *netdev,
4824                                                struct rtnl_link_stats64 *stats)
4825{
4826        struct igb_adapter *adapter = netdev_priv(netdev);
4827
4828        spin_lock(&adapter->stats64_lock);
4829        igb_update_stats(adapter, &adapter->stats64);
4830        memcpy(stats, &adapter->stats64, sizeof(*stats));
4831        spin_unlock(&adapter->stats64_lock);
4832
4833        return stats;
4834}
4835
4836/**
4837 *  igb_change_mtu - Change the Maximum Transfer Unit
4838 *  @netdev: network interface device structure
4839 *  @new_mtu: new value for maximum frame size
4840 *
4841 *  Returns 0 on success, negative on failure
4842 **/
4843static int igb_change_mtu(struct net_device *netdev, int new_mtu)
4844{
4845        struct igb_adapter *adapter = netdev_priv(netdev);
4846        struct pci_dev *pdev = adapter->pdev;
4847        int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
4848
4849        if ((new_mtu < 68) || (max_frame > MAX_JUMBO_FRAME_SIZE)) {
4850                dev_err(&pdev->dev, "Invalid MTU setting\n");
4851                return -EINVAL;
4852        }
4853
4854#define MAX_STD_JUMBO_FRAME_SIZE 9238
4855        if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
4856                dev_err(&pdev->dev, "MTU > 9216 not supported.\n");
4857                return -EINVAL;
4858        }
4859
4860        /* adjust max frame to be at least the size of a standard frame */
4861        if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
4862                max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
4863
4864        while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
4865                msleep(1);
4866
4867        /* igb_down has a dependency on max_frame_size */
4868        adapter->max_frame_size = max_frame;
4869
4870        if (netif_running(netdev))
4871                igb_down(adapter);
4872
4873        dev_info(&pdev->dev, "changing MTU from %d to %d\n",
4874                 netdev->mtu, new_mtu);
4875        netdev->mtu = new_mtu;
4876
4877        if (netif_running(netdev))
4878                igb_up(adapter);
4879        else
4880                igb_reset(adapter);
4881
4882        clear_bit(__IGB_RESETTING, &adapter->state);
4883
4884        return 0;
4885}
4886
4887/**
4888 *  igb_update_stats - Update the board statistics counters
4889 *  @adapter: board private structure
4890 **/
4891void igb_update_stats(struct igb_adapter *adapter,
4892                      struct rtnl_link_stats64 *net_stats)
4893{
4894        struct e1000_hw *hw = &adapter->hw;
4895        struct pci_dev *pdev = adapter->pdev;
4896        u32 reg, mpc;
4897        u16 phy_tmp;
4898        int i;
4899        u64 bytes, packets;
4900        unsigned int start;
4901        u64 _bytes, _packets;
4902
4903#define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
4904
4905        /* Prevent stats update while adapter is being reset, or if the pci
4906         * connection is down.
4907         */
4908        if (adapter->link_speed == 0)
4909                return;
4910        if (pci_channel_offline(pdev))
4911                return;
4912
4913        bytes = 0;
4914        packets = 0;
4915
4916        rcu_read_lock();
4917        for (i = 0; i < adapter->num_rx_queues; i++) {
4918                u32 rqdpc = rd32(E1000_RQDPC(i));
4919                struct igb_ring *ring = adapter->rx_ring[i];
4920
4921                if (rqdpc) {
4922                        ring->rx_stats.drops += rqdpc;
4923                        net_stats->rx_fifo_errors += rqdpc;
4924                }
4925
4926                do {
4927                        start = u64_stats_fetch_begin_bh(&ring->rx_syncp);
4928                        _bytes = ring->rx_stats.bytes;
4929                        _packets = ring->rx_stats.packets;
4930                } while (u64_stats_fetch_retry_bh(&ring->rx_syncp, start));
4931                bytes += _bytes;
4932                packets += _packets;
4933        }
4934
4935        net_stats->rx_bytes = bytes;
4936        net_stats->rx_packets = packets;
4937
4938        bytes = 0;
4939        packets = 0;
4940        for (i = 0; i < adapter->num_tx_queues; i++) {
4941                struct igb_ring *ring = adapter->tx_ring[i];
4942                do {
4943                        start = u64_stats_fetch_begin_bh(&ring->tx_syncp);
4944                        _bytes = ring->tx_stats.bytes;
4945                        _packets = ring->tx_stats.packets;
4946                } while (u64_stats_fetch_retry_bh(&ring->tx_syncp, start));
4947                bytes += _bytes;
4948                packets += _packets;
4949        }
4950        net_stats->tx_bytes = bytes;
4951        net_stats->tx_packets = packets;
4952        rcu_read_unlock();
4953
4954        /* read stats registers */
4955        adapter->stats.crcerrs += rd32(E1000_CRCERRS);
4956        adapter->stats.gprc += rd32(E1000_GPRC);
4957        adapter->stats.gorc += rd32(E1000_GORCL);
4958        rd32(E1000_GORCH); /* clear GORCL */
4959        adapter->stats.bprc += rd32(E1000_BPRC);
4960        adapter->stats.mprc += rd32(E1000_MPRC);
4961        adapter->stats.roc += rd32(E1000_ROC);
4962
4963        adapter->stats.prc64 += rd32(E1000_PRC64);
4964        adapter->stats.prc127 += rd32(E1000_PRC127);
4965        adapter->stats.prc255 += rd32(E1000_PRC255);
4966        adapter->stats.prc511 += rd32(E1000_PRC511);
4967        adapter->stats.prc1023 += rd32(E1000_PRC1023);
4968        adapter->stats.prc1522 += rd32(E1000_PRC1522);
4969        adapter->stats.symerrs += rd32(E1000_SYMERRS);
4970        adapter->stats.sec += rd32(E1000_SEC);
4971
4972        mpc = rd32(E1000_MPC);
4973        adapter->stats.mpc += mpc;
4974        net_stats->rx_fifo_errors += mpc;
4975        adapter->stats.scc += rd32(E1000_SCC);
4976        adapter->stats.ecol += rd32(E1000_ECOL);
4977        adapter->stats.mcc += rd32(E1000_MCC);
4978        adapter->stats.latecol += rd32(E1000_LATECOL);
4979        adapter->stats.dc += rd32(E1000_DC);
4980        adapter->stats.rlec += rd32(E1000_RLEC);
4981        adapter->stats.xonrxc += rd32(E1000_XONRXC);
4982        adapter->stats.xontxc += rd32(E1000_XONTXC);
4983        adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
4984        adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
4985        adapter->stats.fcruc += rd32(E1000_FCRUC);
4986        adapter->stats.gptc += rd32(E1000_GPTC);
4987        adapter->stats.gotc += rd32(E1000_GOTCL);
4988        rd32(E1000_GOTCH); /* clear GOTCL */
4989        adapter->stats.rnbc += rd32(E1000_RNBC);
4990        adapter->stats.ruc += rd32(E1000_RUC);
4991        adapter->stats.rfc += rd32(E1000_RFC);
4992        adapter->stats.rjc += rd32(E1000_RJC);
4993        adapter->stats.tor += rd32(E1000_TORH);
4994        adapter->stats.tot += rd32(E1000_TOTH);
4995        adapter->stats.tpr += rd32(E1000_TPR);
4996
4997        adapter->stats.ptc64 += rd32(E1000_PTC64);
4998        adapter->stats.ptc127 += rd32(E1000_PTC127);
4999        adapter->stats.ptc255 += rd32(E1000_PTC255);
5000        adapter->stats.ptc511 += rd32(E1000_PTC511);
5001        adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5002        adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5003
5004        adapter->stats.mptc += rd32(E1000_MPTC);
5005        adapter->stats.bptc += rd32(E1000_BPTC);
5006
5007        adapter->stats.tpt += rd32(E1000_TPT);
5008        adapter->stats.colc += rd32(E1000_COLC);
5009
5010        adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5011        /* read internal phy specific stats */
5012        reg = rd32(E1000_CTRL_EXT);
5013        if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5014                adapter->stats.rxerrc += rd32(E1000_RXERRC);
5015
5016                /* this stat has invalid values on i210/i211 */
5017                if ((hw->mac.type != e1000_i210) &&
5018                    (hw->mac.type != e1000_i211))
5019                        adapter->stats.tncrs += rd32(E1000_TNCRS);
5020        }
5021
5022        adapter->stats.tsctc += rd32(E1000_TSCTC);
5023        adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5024
5025        adapter->stats.iac += rd32(E1000_IAC);
5026        adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5027        adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5028        adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5029        adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5030        adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5031        adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5032        adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5033        adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5034
5035        /* Fill out the OS statistics structure */
5036        net_stats->multicast = adapter->stats.mprc;
5037        net_stats->collisions = adapter->stats.colc;
5038
5039        /* Rx Errors */
5040
5041        /* RLEC on some newer hardware can be incorrect so build
5042         * our own version based on RUC and ROC
5043         */
5044        net_stats->rx_errors = adapter->stats.rxerrc +
5045                adapter->stats.crcerrs + adapter->stats.algnerrc +
5046                adapter->stats.ruc + adapter->stats.roc +
5047                adapter->stats.cexterr;
5048        net_stats->rx_length_errors = adapter->stats.ruc +
5049                                      adapter->stats.roc;
5050        net_stats->rx_crc_errors = adapter->stats.crcerrs;
5051        net_stats->rx_frame_errors = adapter->stats.algnerrc;
5052        net_stats->rx_missed_errors = adapter->stats.mpc;
5053
5054        /* Tx Errors */
5055        net_stats->tx_errors = adapter->stats.ecol +
5056                               adapter->stats.latecol;
5057        net_stats->tx_aborted_errors = adapter->stats.ecol;
5058        net_stats->tx_window_errors = adapter->stats.latecol;
5059        net_stats->tx_carrier_errors = adapter->stats.tncrs;
5060
5061        /* Tx Dropped needs to be maintained elsewhere */
5062
5063        /* Phy Stats */
5064        if (hw->phy.media_type == e1000_media_type_copper) {
5065                if ((adapter->link_speed == SPEED_1000) &&
5066                   (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
5067                        phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
5068                        adapter->phy_stats.idle_errors += phy_tmp;
5069                }
5070        }
5071
5072        /* Management Stats */
5073        adapter->stats.mgptc += rd32(E1000_MGTPTC);
5074        adapter->stats.mgprc += rd32(E1000_MGTPRC);
5075        adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5076
5077        /* OS2BMC Stats */
5078        reg = rd32(E1000_MANC);
5079        if (reg & E1000_MANC_EN_BMC2OS) {
5080                adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5081                adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5082                adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5083                adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5084        }
5085}
5086
5087static irqreturn_t igb_msix_other(int irq, void *data)
5088{
5089        struct igb_adapter *adapter = data;
5090        struct e1000_hw *hw = &adapter->hw;
5091        u32 icr = rd32(E1000_ICR);
5092        /* reading ICR causes bit 31 of EICR to be cleared */
5093
5094        if (icr & E1000_ICR_DRSTA)
5095                schedule_work(&adapter->reset_task);
5096
5097        if (icr & E1000_ICR_DOUTSYNC) {
5098                /* HW is reporting DMA is out of sync */
5099                adapter->stats.doosync++;
5100                /* The DMA Out of Sync is also indication of a spoof event
5101                 * in IOV mode. Check the Wrong VM Behavior register to
5102                 * see if it is really a spoof event.
5103                 */
5104                igb_check_wvbr(adapter);
5105        }
5106
5107        /* Check for a mailbox event */
5108        if (icr & E1000_ICR_VMMB)
5109                igb_msg_task(adapter);
5110
5111        if (icr & E1000_ICR_LSC) {
5112                hw->mac.get_link_status = 1;
5113                /* guard against interrupt when we're going down */
5114                if (!test_bit(__IGB_DOWN, &adapter->state))
5115                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
5116        }
5117
5118        if (icr & E1000_ICR_TS) {
5119                u32 tsicr = rd32(E1000_TSICR);
5120
5121                if (tsicr & E1000_TSICR_TXTS) {
5122                        /* acknowledge the interrupt */
5123                        wr32(E1000_TSICR, E1000_TSICR_TXTS);
5124                        /* retrieve hardware timestamp */
5125                        schedule_work(&adapter->ptp_tx_work);
5126                }
5127        }
5128
5129        wr32(E1000_EIMS, adapter->eims_other);
5130
5131        return IRQ_HANDLED;
5132}
5133
5134static void igb_write_itr(struct igb_q_vector *q_vector)
5135{
5136        struct igb_adapter *adapter = q_vector->adapter;
5137        u32 itr_val = q_vector->itr_val & 0x7FFC;
5138
5139        if (!q_vector->set_itr)
5140                return;
5141
5142        if (!itr_val)
5143                itr_val = 0x4;
5144
5145        if (adapter->hw.mac.type == e1000_82575)
5146                itr_val |= itr_val << 16;
5147        else
5148                itr_val |= E1000_EITR_CNT_IGNR;
5149
5150        writel(itr_val, q_vector->itr_register);
5151        q_vector->set_itr = 0;
5152}
5153
5154static irqreturn_t igb_msix_ring(int irq, void *data)
5155{
5156        struct igb_q_vector *q_vector = data;
5157
5158        /* Write the ITR value calculated from the previous interrupt. */
5159        igb_write_itr(q_vector);
5160
5161        napi_schedule(&q_vector->napi);
5162
5163        return IRQ_HANDLED;
5164}
5165
5166#ifdef CONFIG_IGB_DCA
5167static void igb_update_tx_dca(struct igb_adapter *adapter,
5168                              struct igb_ring *tx_ring,
5169                              int cpu)
5170{
5171        struct e1000_hw *hw = &adapter->hw;
5172        u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5173
5174        if (hw->mac.type != e1000_82575)
5175                txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5176
5177        /* We can enable relaxed ordering for reads, but not writes when
5178         * DCA is enabled.  This is due to a known issue in some chipsets
5179         * which will cause the DCA tag to be cleared.
5180         */
5181        txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5182                  E1000_DCA_TXCTRL_DATA_RRO_EN |
5183                  E1000_DCA_TXCTRL_DESC_DCA_EN;
5184
5185        wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5186}
5187
5188static void igb_update_rx_dca(struct igb_adapter *adapter,
5189                              struct igb_ring *rx_ring,
5190                              int cpu)
5191{
5192        struct e1000_hw *hw = &adapter->hw;
5193        u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5194
5195        if (hw->mac.type != e1000_82575)
5196                rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5197
5198        /* We can enable relaxed ordering for reads, but not writes when
5199         * DCA is enabled.  This is due to a known issue in some chipsets
5200         * which will cause the DCA tag to be cleared.
5201         */
5202        rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5203                  E1000_DCA_RXCTRL_DESC_DCA_EN;
5204
5205        wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5206}
5207
5208static void igb_update_dca(struct igb_q_vector *q_vector)
5209{
5210        struct igb_adapter *adapter = q_vector->adapter;
5211        int cpu = get_cpu();
5212
5213        if (q_vector->cpu == cpu)
5214                goto out_no_update;
5215
5216        if (q_vector->tx.ring)
5217                igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5218
5219        if (q_vector->rx.ring)
5220                igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5221
5222        q_vector->cpu = cpu;
5223out_no_update:
5224        put_cpu();
5225}
5226
5227static void igb_setup_dca(struct igb_adapter *adapter)
5228{
5229        struct e1000_hw *hw = &adapter->hw;
5230        int i;
5231
5232        if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5233                return;
5234
5235        /* Always use CB2 mode, difference is masked in the CB driver. */
5236        wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5237
5238        for (i = 0; i < adapter->num_q_vectors; i++) {
5239                adapter->q_vector[i]->cpu = -1;
5240                igb_update_dca(adapter->q_vector[i]);
5241        }
5242}
5243
5244static int __igb_notify_dca(struct device *dev, void *data)
5245{
5246        struct net_device *netdev = dev_get_drvdata(dev);
5247        struct igb_adapter *adapter = netdev_priv(netdev);
5248        struct pci_dev *pdev = adapter->pdev;
5249        struct e1000_hw *hw = &adapter->hw;
5250        unsigned long event = *(unsigned long *)data;
5251
5252        switch (event) {
5253        case DCA_PROVIDER_ADD:
5254                /* if already enabled, don't do it again */
5255                if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5256                        break;
5257                if (dca_add_requester(dev) == 0) {
5258                        adapter->flags |= IGB_FLAG_DCA_ENABLED;
5259                        dev_info(&pdev->dev, "DCA enabled\n");
5260                        igb_setup_dca(adapter);
5261                        break;
5262                }
5263                /* Fall Through since DCA is disabled. */
5264        case DCA_PROVIDER_REMOVE:
5265                if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5266                        /* without this a class_device is left
5267                         * hanging around in the sysfs model
5268                         */
5269                        dca_remove_requester(dev);
5270                        dev_info(&pdev->dev, "DCA disabled\n");
5271                        adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5272                        wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5273                }
5274                break;
5275        }
5276
5277        return 0;
5278}
5279
5280static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5281                          void *p)
5282{
5283        int ret_val;
5284
5285        ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
5286                                         __igb_notify_dca);
5287
5288        return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
5289}
5290#endif /* CONFIG_IGB_DCA */
5291
5292#ifdef CONFIG_PCI_IOV
5293static int igb_vf_configure(struct igb_adapter *adapter, int vf)
5294{
5295        unsigned char mac_addr[ETH_ALEN];
5296
5297        eth_zero_addr(mac_addr);
5298        igb_set_vf_mac(adapter, vf, mac_addr);
5299
5300        /* By default spoof check is enabled for all VFs */
5301        adapter->vf_data[vf].spoofchk_enabled = true;
5302
5303        return 0;
5304}
5305
5306#endif
5307static void igb_ping_all_vfs(struct igb_adapter *adapter)
5308{
5309        struct e1000_hw *hw = &adapter->hw;
5310        u32 ping;
5311        int i;
5312
5313        for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
5314                ping = E1000_PF_CONTROL_MSG;
5315                if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
5316                        ping |= E1000_VT_MSGTYPE_CTS;
5317                igb_write_mbx(hw, &ping, 1, i);
5318        }
5319}
5320
5321static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5322{
5323        struct e1000_hw *hw = &adapter->hw;
5324        u32 vmolr = rd32(E1000_VMOLR(vf));
5325        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5326
5327        vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
5328                            IGB_VF_FLAG_MULTI_PROMISC);
5329        vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5330
5331        if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
5332                vmolr |= E1000_VMOLR_MPME;
5333                vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
5334                *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
5335        } else {
5336                /* if we have hashes and we are clearing a multicast promisc
5337                 * flag we need to write the hashes to the MTA as this step
5338                 * was previously skipped
5339                 */
5340                if (vf_data->num_vf_mc_hashes > 30) {
5341                        vmolr |= E1000_VMOLR_MPME;
5342                } else if (vf_data->num_vf_mc_hashes) {
5343                        int j;
5344                        vmolr |= E1000_VMOLR_ROMPE;
5345                        for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5346                                igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5347                }
5348        }
5349
5350        wr32(E1000_VMOLR(vf), vmolr);
5351
5352        /* there are flags left unprocessed, likely not supported */
5353        if (*msgbuf & E1000_VT_MSGINFO_MASK)
5354                return -EINVAL;
5355
5356        return 0;
5357}
5358
5359static int igb_set_vf_multicasts(struct igb_adapter *adapter,
5360                                  u32 *msgbuf, u32 vf)
5361{
5362        int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5363        u16 *hash_list = (u16 *)&msgbuf[1];
5364        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5365        int i;
5366
5367        /* salt away the number of multicast addresses assigned
5368         * to this VF for later use to restore when the PF multi cast
5369         * list changes
5370         */
5371        vf_data->num_vf_mc_hashes = n;
5372
5373        /* only up to 30 hash values supported */
5374        if (n > 30)
5375                n = 30;
5376
5377        /* store the hashes for later use */
5378        for (i = 0; i < n; i++)
5379                vf_data->vf_mc_hashes[i] = hash_list[i];
5380
5381        /* Flush and reset the mta with the new values */
5382        igb_set_rx_mode(adapter->netdev);
5383
5384        return 0;
5385}
5386
5387static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
5388{
5389        struct e1000_hw *hw = &adapter->hw;
5390        struct vf_data_storage *vf_data;
5391        int i, j;
5392
5393        for (i = 0; i < adapter->vfs_allocated_count; i++) {
5394                u32 vmolr = rd32(E1000_VMOLR(i));
5395                vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
5396
5397                vf_data = &adapter->vf_data[i];
5398
5399                if ((vf_data->num_vf_mc_hashes > 30) ||
5400                    (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
5401                        vmolr |= E1000_VMOLR_MPME;
5402                } else if (vf_data->num_vf_mc_hashes) {
5403                        vmolr |= E1000_VMOLR_ROMPE;
5404                        for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
5405                                igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
5406                }
5407                wr32(E1000_VMOLR(i), vmolr);
5408        }
5409}
5410
5411static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
5412{
5413        struct e1000_hw *hw = &adapter->hw;
5414        u32 pool_mask, reg, vid;
5415        int i;
5416
5417        pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5418
5419        /* Find the vlan filter for this id */
5420        for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5421                reg = rd32(E1000_VLVF(i));
5422
5423                /* remove the vf from the pool */
5424                reg &= ~pool_mask;
5425
5426                /* if pool is empty then remove entry from vfta */
5427                if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
5428                    (reg & E1000_VLVF_VLANID_ENABLE)) {
5429                        reg = 0;
5430                        vid = reg & E1000_VLVF_VLANID_MASK;
5431                        igb_vfta_set(hw, vid, false);
5432                }
5433
5434                wr32(E1000_VLVF(i), reg);
5435        }
5436
5437        adapter->vf_data[vf].vlans_enabled = 0;
5438}
5439
5440static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
5441{
5442        struct e1000_hw *hw = &adapter->hw;
5443        u32 reg, i;
5444
5445        /* The vlvf table only exists on 82576 hardware and newer */
5446        if (hw->mac.type < e1000_82576)
5447                return -1;
5448
5449        /* we only need to do this if VMDq is enabled */
5450        if (!adapter->vfs_allocated_count)
5451                return -1;
5452
5453        /* Find the vlan filter for this id */
5454        for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5455                reg = rd32(E1000_VLVF(i));
5456                if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5457                    vid == (reg & E1000_VLVF_VLANID_MASK))
5458                        break;
5459        }
5460
5461        if (add) {
5462                if (i == E1000_VLVF_ARRAY_SIZE) {
5463                        /* Did not find a matching VLAN ID entry that was
5464                         * enabled.  Search for a free filter entry, i.e.
5465                         * one without the enable bit set
5466                         */
5467                        for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5468                                reg = rd32(E1000_VLVF(i));
5469                                if (!(reg & E1000_VLVF_VLANID_ENABLE))
5470                                        break;
5471                        }
5472                }
5473                if (i < E1000_VLVF_ARRAY_SIZE) {
5474                        /* Found an enabled/available entry */
5475                        reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
5476
5477                        /* if !enabled we need to set this up in vfta */
5478                        if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
5479                                /* add VID to filter table */
5480                                igb_vfta_set(hw, vid, true);
5481                                reg |= E1000_VLVF_VLANID_ENABLE;
5482                        }
5483                        reg &= ~E1000_VLVF_VLANID_MASK;
5484                        reg |= vid;
5485                        wr32(E1000_VLVF(i), reg);
5486
5487                        /* do not modify RLPML for PF devices */
5488                        if (vf >= adapter->vfs_allocated_count)
5489                                return 0;
5490
5491                        if (!adapter->vf_data[vf].vlans_enabled) {
5492                                u32 size;
5493                                reg = rd32(E1000_VMOLR(vf));
5494                                size = reg & E1000_VMOLR_RLPML_MASK;
5495                                size += 4;
5496                                reg &= ~E1000_VMOLR_RLPML_MASK;
5497                                reg |= size;
5498                                wr32(E1000_VMOLR(vf), reg);
5499                        }
5500
5501                        adapter->vf_data[vf].vlans_enabled++;
5502                }
5503        } else {
5504                if (i < E1000_VLVF_ARRAY_SIZE) {
5505                        /* remove vf from the pool */
5506                        reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
5507                        /* if pool is empty then remove entry from vfta */
5508                        if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
5509                                reg = 0;
5510                                igb_vfta_set(hw, vid, false);
5511                        }
5512                        wr32(E1000_VLVF(i), reg);
5513
5514                        /* do not modify RLPML for PF devices */
5515                        if (vf >= adapter->vfs_allocated_count)
5516                                return 0;
5517
5518                        adapter->vf_data[vf].vlans_enabled--;
5519                        if (!adapter->vf_data[vf].vlans_enabled) {
5520                                u32 size;
5521                                reg = rd32(E1000_VMOLR(vf));
5522                                size = reg & E1000_VMOLR_RLPML_MASK;
5523                                size -= 4;
5524                                reg &= ~E1000_VMOLR_RLPML_MASK;
5525                                reg |= size;
5526                                wr32(E1000_VMOLR(vf), reg);
5527                        }
5528                }
5529        }
5530        return 0;
5531}
5532
5533static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
5534{
5535        struct e1000_hw *hw = &adapter->hw;
5536
5537        if (vid)
5538                wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
5539        else
5540                wr32(E1000_VMVIR(vf), 0);
5541}
5542
5543static int igb_ndo_set_vf_vlan(struct net_device *netdev,
5544                               int vf, u16 vlan, u8 qos)
5545{
5546        int err = 0;
5547        struct igb_adapter *adapter = netdev_priv(netdev);
5548
5549        if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
5550                return -EINVAL;
5551        if (vlan || qos) {
5552                err = igb_vlvf_set(adapter, vlan, !!vlan, vf);
5553                if (err)
5554                        goto out;
5555                igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
5556                igb_set_vmolr(adapter, vf, !vlan);
5557                adapter->vf_data[vf].pf_vlan = vlan;
5558                adapter->vf_data[vf].pf_qos = qos;
5559                dev_info(&adapter->pdev->dev,
5560                         "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
5561                if (test_bit(__IGB_DOWN, &adapter->state)) {
5562                        dev_warn(&adapter->pdev->dev,
5563                                 "The VF VLAN has been set, but the PF device is not up.\n");
5564                        dev_warn(&adapter->pdev->dev,
5565                                 "Bring the PF device up before attempting to use the VF device.\n");
5566                }
5567        } else {
5568                igb_vlvf_set(adapter, adapter->vf_data[vf].pf_vlan,
5569                             false, vf);
5570                igb_set_vmvir(adapter, vlan, vf);
5571                igb_set_vmolr(adapter, vf, true);
5572                adapter->vf_data[vf].pf_vlan = 0;
5573                adapter->vf_data[vf].pf_qos = 0;
5574        }
5575out:
5576        return err;
5577}
5578
5579static int igb_find_vlvf_entry(struct igb_adapter *adapter, int vid)
5580{
5581        struct e1000_hw *hw = &adapter->hw;
5582        int i;
5583        u32 reg;
5584
5585        /* Find the vlan filter for this id */
5586        for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
5587                reg = rd32(E1000_VLVF(i));
5588                if ((reg & E1000_VLVF_VLANID_ENABLE) &&
5589                    vid == (reg & E1000_VLVF_VLANID_MASK))
5590                        break;
5591        }
5592
5593        if (i >= E1000_VLVF_ARRAY_SIZE)
5594                i = -1;
5595
5596        return i;
5597}
5598
5599static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
5600{
5601        struct e1000_hw *hw = &adapter->hw;
5602        int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
5603        int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
5604        int err = 0;
5605
5606        /* If in promiscuous mode we need to make sure the PF also has
5607         * the VLAN filter set.
5608         */
5609        if (add && (adapter->netdev->flags & IFF_PROMISC))
5610                err = igb_vlvf_set(adapter, vid, add,
5611                                   adapter->vfs_allocated_count);
5612        if (err)
5613                goto out;
5614
5615        err = igb_vlvf_set(adapter, vid, add, vf);
5616
5617        if (err)
5618                goto out;
5619
5620        /* Go through all the checks to see if the VLAN filter should
5621         * be wiped completely.
5622         */
5623        if (!add && (adapter->netdev->flags & IFF_PROMISC)) {
5624                u32 vlvf, bits;
5625
5626                int regndx = igb_find_vlvf_entry(adapter, vid);
5627                if (regndx < 0)
5628                        goto out;
5629                /* See if any other pools are set for this VLAN filter
5630                 * entry other than the PF.
5631                 */
5632                vlvf = bits = rd32(E1000_VLVF(regndx));
5633                bits &= 1 << (E1000_VLVF_POOLSEL_SHIFT +
5634                              adapter->vfs_allocated_count);
5635                /* If the filter was removed then ensure PF pool bit
5636                 * is cleared if the PF only added itself to the pool
5637                 * because the PF is in promiscuous mode.
5638                 */
5639                if ((vlvf & VLAN_VID_MASK) == vid &&
5640                    !test_bit(vid, adapter->active_vlans) &&
5641                    !bits)
5642                        igb_vlvf_set(adapter, vid, add,
5643                                     adapter->vfs_allocated_count);
5644        }
5645
5646out:
5647        return err;
5648}
5649
5650static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
5651{
5652        /* clear flags - except flag that indicates PF has set the MAC */
5653        adapter->vf_data[vf].flags &= IGB_VF_FLAG_PF_SET_MAC;
5654        adapter->vf_data[vf].last_nack = jiffies;
5655
5656        /* reset offloads to defaults */
5657        igb_set_vmolr(adapter, vf, true);
5658
5659        /* reset vlans for device */
5660        igb_clear_vf_vfta(adapter, vf);
5661        if (adapter->vf_data[vf].pf_vlan)
5662                igb_ndo_set_vf_vlan(adapter->netdev, vf,
5663                                    adapter->vf_data[vf].pf_vlan,
5664                                    adapter->vf_data[vf].pf_qos);
5665        else
5666                igb_clear_vf_vfta(adapter, vf);
5667
5668        /* reset multicast table array for vf */
5669        adapter->vf_data[vf].num_vf_mc_hashes = 0;
5670
5671        /* Flush and reset the mta with the new values */
5672        igb_set_rx_mode(adapter->netdev);
5673}
5674
5675static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
5676{
5677        unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5678
5679        /* clear mac address as we were hotplug removed/added */
5680        if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
5681                eth_zero_addr(vf_mac);
5682
5683        /* process remaining reset events */
5684        igb_vf_reset(adapter, vf);
5685}
5686
5687static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
5688{
5689        struct e1000_hw *hw = &adapter->hw;
5690        unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
5691        int rar_entry = hw->mac.rar_entry_count - (vf + 1);
5692        u32 reg, msgbuf[3];
5693        u8 *addr = (u8 *)(&msgbuf[1]);
5694
5695        /* process all the same items cleared in a function level reset */
5696        igb_vf_reset(adapter, vf);
5697
5698        /* set vf mac address */
5699        igb_rar_set_qsel(adapter, vf_mac, rar_entry, vf);
5700
5701        /* enable transmit and receive for vf */
5702        reg = rd32(E1000_VFTE);
5703        wr32(E1000_VFTE, reg | (1 << vf));
5704        reg = rd32(E1000_VFRE);
5705        wr32(E1000_VFRE, reg | (1 << vf));
5706
5707        adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
5708
5709        /* reply to reset with ack and vf mac address */
5710        msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
5711        memcpy(addr, vf_mac, 6);
5712        igb_write_mbx(hw, msgbuf, 3, vf);
5713}
5714
5715static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
5716{
5717        /* The VF MAC Address is stored in a packed array of bytes
5718         * starting at the second 32 bit word of the msg array
5719         */
5720        unsigned char *addr = (char *)&msg[1];
5721        int err = -1;
5722
5723        if (is_valid_ether_addr(addr))
5724                err = igb_set_vf_mac(adapter, vf, addr);
5725
5726        return err;
5727}
5728
5729static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
5730{
5731        struct e1000_hw *hw = &adapter->hw;
5732        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5733        u32 msg = E1000_VT_MSGTYPE_NACK;
5734
5735        /* if device isn't clear to send it shouldn't be reading either */
5736        if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
5737            time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
5738                igb_write_mbx(hw, &msg, 1, vf);
5739                vf_data->last_nack = jiffies;
5740        }
5741}
5742
5743static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
5744{
5745        struct pci_dev *pdev = adapter->pdev;
5746        u32 msgbuf[E1000_VFMAILBOX_SIZE];
5747        struct e1000_hw *hw = &adapter->hw;
5748        struct vf_data_storage *vf_data = &adapter->vf_data[vf];
5749        s32 retval;
5750
5751        retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf);
5752
5753        if (retval) {
5754                /* if receive failed revoke VF CTS stats and restart init */
5755                dev_err(&pdev->dev, "Error receiving message from VF\n");
5756                vf_data->flags &= ~IGB_VF_FLAG_CTS;
5757                if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5758                        return;
5759                goto out;
5760        }
5761
5762        /* this is a message we already processed, do nothing */
5763        if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
5764                return;
5765
5766        /* until the vf completes a reset it should not be
5767         * allowed to start any configuration.
5768         */
5769        if (msgbuf[0] == E1000_VF_RESET) {
5770                igb_vf_reset_msg(adapter, vf);
5771                return;
5772        }
5773
5774        if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
5775                if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
5776                        return;
5777                retval = -1;
5778                goto out;
5779        }
5780
5781        switch ((msgbuf[0] & 0xFFFF)) {
5782        case E1000_VF_SET_MAC_ADDR:
5783                retval = -EINVAL;
5784                if (!(vf_data->flags & IGB_VF_FLAG_PF_SET_MAC))
5785                        retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
5786                else
5787                        dev_warn(&pdev->dev,
5788                                 "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
5789                                 vf);
5790                break;
5791        case E1000_VF_SET_PROMISC:
5792                retval = igb_set_vf_promisc(adapter, msgbuf, vf);
5793                break;
5794        case E1000_VF_SET_MULTICAST:
5795                retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
5796                break;
5797        case E1000_VF_SET_LPE:
5798                retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
5799                break;
5800        case E1000_VF_SET_VLAN:
5801                retval = -1;
5802                if (vf_data->pf_vlan)
5803                        dev_warn(&pdev->dev,
5804                                 "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
5805                                 vf);
5806                else
5807                        retval = igb_set_vf_vlan(adapter, msgbuf, vf);
5808                break;
5809        default:
5810                dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
5811                retval = -1;
5812                break;
5813        }
5814
5815        msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
5816out:
5817        /* notify the VF of the results of what it sent us */
5818        if (retval)
5819                msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
5820        else
5821                msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
5822
5823        igb_write_mbx(hw, msgbuf, 1, vf);
5824}
5825
5826static void igb_msg_task(struct igb_adapter *adapter)
5827{
5828        struct e1000_hw *hw = &adapter->hw;
5829        u32 vf;
5830
5831        for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
5832                /* process any reset requests */
5833                if (!igb_check_for_rst(hw, vf))
5834                        igb_vf_reset_event(adapter, vf);
5835
5836                /* process any messages pending */
5837                if (!igb_check_for_msg(hw, vf))
5838                        igb_rcv_msg_from_vf(adapter, vf);
5839
5840                /* process any acks */
5841                if (!igb_check_for_ack(hw, vf))
5842                        igb_rcv_ack_from_vf(adapter, vf);
5843        }
5844}
5845
5846/**
5847 *  igb_set_uta - Set unicast filter table address
5848 *  @adapter: board private structure
5849 *
5850 *  The unicast table address is a register array of 32-bit registers.
5851 *  The table is meant to be used in a way similar to how the MTA is used
5852 *  however due to certain limitations in the hardware it is necessary to
5853 *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
5854 *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
5855 **/
5856static void igb_set_uta(struct igb_adapter *adapter)
5857{
5858        struct e1000_hw *hw = &adapter->hw;
5859        int i;
5860
5861        /* The UTA table only exists on 82576 hardware and newer */
5862        if (hw->mac.type < e1000_82576)
5863                return;
5864
5865        /* we only need to do this if VMDq is enabled */
5866        if (!adapter->vfs_allocated_count)
5867                return;
5868
5869        for (i = 0; i < hw->mac.uta_reg_count; i++)
5870                array_wr32(E1000_UTA, i, ~0);
5871}
5872
5873/**
5874 *  igb_intr_msi - Interrupt Handler
5875 *  @irq: interrupt number
5876 *  @data: pointer to a network interface device structure
5877 **/
5878static irqreturn_t igb_intr_msi(int irq, void *data)
5879{
5880        struct igb_adapter *adapter = data;
5881        struct igb_q_vector *q_vector = adapter->q_vector[0];
5882        struct e1000_hw *hw = &adapter->hw;
5883        /* read ICR disables interrupts using IAM */
5884        u32 icr = rd32(E1000_ICR);
5885
5886        igb_write_itr(q_vector);
5887
5888        if (icr & E1000_ICR_DRSTA)
5889                schedule_work(&adapter->reset_task);
5890
5891        if (icr & E1000_ICR_DOUTSYNC) {
5892                /* HW is reporting DMA is out of sync */
5893                adapter->stats.doosync++;
5894        }
5895
5896        if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5897                hw->mac.get_link_status = 1;
5898                if (!test_bit(__IGB_DOWN, &adapter->state))
5899                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
5900        }
5901
5902        if (icr & E1000_ICR_TS) {
5903                u32 tsicr = rd32(E1000_TSICR);
5904
5905                if (tsicr & E1000_TSICR_TXTS) {
5906                        /* acknowledge the interrupt */
5907                        wr32(E1000_TSICR, E1000_TSICR_TXTS);
5908                        /* retrieve hardware timestamp */
5909                        schedule_work(&adapter->ptp_tx_work);
5910                }
5911        }
5912
5913        napi_schedule(&q_vector->napi);
5914
5915        return IRQ_HANDLED;
5916}
5917
5918/**
5919 *  igb_intr - Legacy Interrupt Handler
5920 *  @irq: interrupt number
5921 *  @data: pointer to a network interface device structure
5922 **/
5923static irqreturn_t igb_intr(int irq, void *data)
5924{
5925        struct igb_adapter *adapter = data;
5926        struct igb_q_vector *q_vector = adapter->q_vector[0];
5927        struct e1000_hw *hw = &adapter->hw;
5928        /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
5929         * need for the IMC write
5930         */
5931        u32 icr = rd32(E1000_ICR);
5932
5933        /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
5934         * not set, then the adapter didn't send an interrupt
5935         */
5936        if (!(icr & E1000_ICR_INT_ASSERTED))
5937                return IRQ_NONE;
5938
5939        igb_write_itr(q_vector);
5940
5941        if (icr & E1000_ICR_DRSTA)
5942                schedule_work(&adapter->reset_task);
5943
5944        if (icr & E1000_ICR_DOUTSYNC) {
5945                /* HW is reporting DMA is out of sync */
5946                adapter->stats.doosync++;
5947        }
5948
5949        if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
5950                hw->mac.get_link_status = 1;
5951                /* guard against interrupt when we're going down */
5952                if (!test_bit(__IGB_DOWN, &adapter->state))
5953                        mod_timer(&adapter->watchdog_timer, jiffies + 1);
5954        }
5955
5956        if (icr & E1000_ICR_TS) {
5957                u32 tsicr = rd32(E1000_TSICR);
5958
5959                if (tsicr & E1000_TSICR_TXTS) {
5960                        /* acknowledge the interrupt */
5961                        wr32(E1000_TSICR, E1000_TSICR_TXTS);
5962                        /* retrieve hardware timestamp */
5963                        schedule_work(&adapter->ptp_tx_work);
5964                }
5965        }
5966
5967        napi_schedule(&q_vector->napi);
5968
5969        return IRQ_HANDLED;
5970}
5971
5972static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
5973{
5974        struct igb_adapter *adapter = q_vector->adapter;
5975        struct e1000_hw *hw = &adapter->hw;
5976
5977        if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
5978            (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
5979                if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
5980                        igb_set_itr(q_vector);
5981                else
5982                        igb_update_ring_itr(q_vector);
5983        }
5984
5985        if (!test_bit(__IGB_DOWN, &adapter->state)) {
5986                if (adapter->msix_entries)
5987                        wr32(E1000_EIMS, q_vector->eims_value);
5988                else
5989                        igb_irq_enable(adapter);
5990        }
5991}
5992
5993/**
5994 *  igb_poll - NAPI Rx polling callback
5995 *  @napi: napi polling structure
5996 *  @budget: count of how many packets we should handle
5997 **/
5998static int igb_poll(struct napi_struct *napi, int budget)
5999{
6000        struct igb_q_vector *q_vector = container_of(napi,
6001                                                     struct igb_q_vector,
6002                                                     napi);
6003        bool clean_complete = true;
6004
6005#ifdef CONFIG_IGB_DCA
6006        if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6007                igb_update_dca(q_vector);
6008#endif
6009        if (q_vector->tx.ring)
6010                clean_complete = igb_clean_tx_irq(q_vector);
6011
6012        if (q_vector->rx.ring)
6013                clean_complete &= igb_clean_rx_irq(q_vector, budget);
6014
6015        /* If all work not completed, return budget and keep polling */
6016        if (!clean_complete)
6017                return budget;
6018
6019        /* If not enough Rx work done, exit the polling mode */
6020        napi_complete(napi);
6021        igb_ring_irq_enable(q_vector);
6022
6023        return 0;
6024}
6025
6026/**
6027 *  igb_clean_tx_irq - Reclaim resources after transmit completes
6028 *  @q_vector: pointer to q_vector containing needed info
6029 *
6030 *  returns true if ring is completely cleaned
6031 **/
6032static bool igb_clean_tx_irq(struct igb_q_vector *q_vector)
6033{
6034        struct igb_adapter *adapter = q_vector->adapter;
6035        struct igb_ring *tx_ring = q_vector->tx.ring;
6036        struct igb_tx_buffer *tx_buffer;
6037        union e1000_adv_tx_desc *tx_desc;
6038        unsigned int total_bytes = 0, total_packets = 0;
6039        unsigned int budget = q_vector->tx.work_limit;
6040        unsigned int i = tx_ring->next_to_clean;
6041
6042        if (test_bit(__IGB_DOWN, &adapter->state))
6043                return true;
6044
6045        tx_buffer = &tx_ring->tx_buffer_info[i];
6046        tx_desc = IGB_TX_DESC(tx_ring, i);
6047        i -= tx_ring->count;
6048
6049        do {
6050                union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6051
6052                /* if next_to_watch is not set then there is no work pending */
6053                if (!eop_desc)
6054                        break;
6055
6056                /* prevent any other reads prior to eop_desc */
6057                read_barrier_depends();
6058
6059                /* if DD is not set pending work has not been completed */
6060                if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6061                        break;
6062
6063                /* clear next_to_watch to prevent false hangs */
6064                tx_buffer->next_to_watch = NULL;
6065
6066                /* update the statistics for this packet */
6067                total_bytes += tx_buffer->bytecount;
6068                total_packets += tx_buffer->gso_segs;
6069
6070                /* free the skb */
6071                dev_kfree_skb_any(tx_buffer->skb);
6072
6073                /* unmap skb header data */
6074                dma_unmap_single(tx_ring->dev,
6075                                 dma_unmap_addr(tx_buffer, dma),
6076                                 dma_unmap_len(tx_buffer, len),
6077                                 DMA_TO_DEVICE);
6078
6079                /* clear tx_buffer data */
6080                tx_buffer->skb = NULL;
6081                dma_unmap_len_set(tx_buffer, len, 0);
6082
6083                /* clear last DMA location and unmap remaining buffers */
6084                while (tx_desc != eop_desc) {
6085                        tx_buffer++;
6086                        tx_desc++;
6087                        i++;
6088                        if (unlikely(!i)) {
6089                                i -= tx_ring->count;
6090                                tx_buffer = tx_ring->tx_buffer_info;
6091                                tx_desc = IGB_TX_DESC(tx_ring, 0);
6092                        }
6093
6094                        /* unmap any remaining paged data */
6095                        if (dma_unmap_len(tx_buffer, len)) {
6096                                dma_unmap_page(tx_ring->dev,
6097                                               dma_unmap_addr(tx_buffer, dma),
6098                                               dma_unmap_len(tx_buffer, len),
6099                                               DMA_TO_DEVICE);
6100                                dma_unmap_len_set(tx_buffer, len, 0);
6101                        }
6102                }
6103
6104                /* move us one more past the eop_desc for start of next pkt */
6105                tx_buffer++;
6106                tx_desc++;
6107                i++;
6108                if (unlikely(!i)) {
6109                        i -= tx_ring->count;
6110                        tx_buffer = tx_ring->tx_buffer_info;
6111                        tx_desc = IGB_TX_DESC(tx_ring, 0);
6112                }
6113
6114                /* issue prefetch for next Tx descriptor */
6115                prefetch(tx_desc);
6116
6117                /* update budget accounting */
6118                budget--;
6119        } while (likely(budget));
6120
6121        netdev_tx_completed_queue(txring_txq(tx_ring),
6122                                  total_packets, total_bytes);
6123        i += tx_ring->count;
6124        tx_ring->next_to_clean = i;
6125        u64_stats_update_begin(&tx_ring->tx_syncp);
6126        tx_ring->tx_stats.bytes += total_bytes;
6127        tx_ring->tx_stats.packets += total_packets;
6128        u64_stats_update_end(&tx_ring->tx_syncp);
6129        q_vector->tx.total_bytes += total_bytes;
6130        q_vector->tx.total_packets += total_packets;
6131
6132        if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
6133                struct e1000_hw *hw = &adapter->hw;
6134
6135                /* Detect a transmit hang in hardware, this serializes the
6136                 * check with the clearing of time_stamp and movement of i
6137                 */
6138                clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
6139                if (tx_buffer->next_to_watch &&
6140                    time_after(jiffies, tx_buffer->time_stamp +
6141                               (adapter->tx_timeout_factor * HZ)) &&
6142                    !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
6143
6144                        /* detected Tx unit hang */
6145                        dev_err(tx_ring->dev,
6146                                "Detected Tx Unit Hang\n"
6147                                "  Tx Queue             <%d>\n"
6148                                "  TDH                  <%x>\n"
6149                                "  TDT                  <%x>\n"
6150                                "  next_to_use          <%x>\n"
6151                                "  next_to_clean        <%x>\n"
6152                                "buffer_info[next_to_clean]\n"
6153                                "  time_stamp           <%lx>\n"
6154                                "  next_to_watch        <%p>\n"
6155                                "  jiffies              <%lx>\n"
6156                                "  desc.status          <%x>\n",
6157                                tx_ring->queue_index,
6158                                rd32(E1000_TDH(tx_ring->reg_idx)),
6159                                readl(tx_ring->tail),
6160                                tx_ring->next_to_use,
6161                                tx_ring->next_to_clean,
6162                                tx_buffer->time_stamp,
6163                                tx_buffer->next_to_watch,
6164                                jiffies,
6165                                tx_buffer->next_to_watch->wb.status);
6166                        netif_stop_subqueue(tx_ring->netdev,
6167                                            tx_ring->queue_index);
6168
6169                        /* we are about to reset, no point in enabling stuff */
6170                        return true;
6171                }
6172        }
6173
6174#define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
6175        if (unlikely(total_packets &&
6176            netif_carrier_ok(tx_ring->netdev) &&
6177            igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
6178                /* Make sure that anybody stopping the queue after this
6179                 * sees the new next_to_clean.
6180                 */
6181                smp_mb();
6182                if (__netif_subqueue_stopped(tx_ring->netdev,
6183                                             tx_ring->queue_index) &&
6184                    !(test_bit(__IGB_DOWN, &adapter->state))) {
6185                        netif_wake_subqueue(tx_ring->netdev,
6186                                            tx_ring->queue_index);
6187
6188                        u64_stats_update_begin(&tx_ring->tx_syncp);
6189                        tx_ring->tx_stats.restart_queue++;
6190                        u64_stats_update_end(&tx_ring->tx_syncp);
6191                }
6192        }
6193
6194        return !!budget;
6195}
6196
6197/**
6198 *  igb_reuse_rx_page - page flip buffer and store it back on the ring
6199 *  @rx_ring: rx descriptor ring to store buffers on
6200 *  @old_buff: donor buffer to have page reused
6201 *
6202 *  Synchronizes page for reuse by the adapter
6203 **/
6204static void igb_reuse_rx_page(struct igb_ring *rx_ring,
6205                              struct igb_rx_buffer *old_buff)
6206{
6207        struct igb_rx_buffer *new_buff;
6208        u16 nta = rx_ring->next_to_alloc;
6209
6210        new_buff = &rx_ring->rx_buffer_info[nta];
6211
6212        /* update, and store next to alloc */
6213        nta++;
6214        rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
6215
6216        /* transfer page from old buffer to new buffer */
6217        memcpy(new_buff, old_buff, sizeof(struct igb_rx_buffer));
6218
6219        /* sync the buffer for use by the device */
6220        dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
6221                                         old_buff->page_offset,
6222                                         IGB_RX_BUFSZ,
6223                                         DMA_FROM_DEVICE);
6224}
6225
6226static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
6227                                  struct page *page,
6228                                  unsigned int truesize)
6229{
6230        /* avoid re-using remote pages */
6231        if (unlikely(page_to_nid(page) != numa_node_id()))
6232                return false;
6233
6234#if (PAGE_SIZE < 8192)
6235        /* if we are only owner of page we can reuse it */
6236        if (unlikely(page_count(page) != 1))
6237                return false;
6238
6239        /* flip page offset to other buffer */
6240        rx_buffer->page_offset ^= IGB_RX_BUFSZ;
6241
6242        /* since we are the only owner of the page and we need to
6243         * increment it, just set the value to 2 in order to avoid
6244         * an unnecessary locked operation
6245         */
6246        atomic_set(&page->_count, 2);
6247#else
6248        /* move offset up to the next cache line */
6249        rx_buffer->page_offset += truesize;
6250
6251        if (rx_buffer->page_offset > (PAGE_SIZE - IGB_RX_BUFSZ))
6252                return false;
6253
6254        /* bump ref count on page before it is given to the stack */
6255        get_page(page);
6256#endif
6257
6258        return true;
6259}
6260
6261/**
6262 *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
6263 *  @rx_ring: rx descriptor ring to transact packets on
6264 *  @rx_buffer: buffer containing page to add
6265 *  @rx_desc: descriptor containing length of buffer written by hardware
6266 *  @skb: sk_buff to place the data into
6267 *
6268 *  This function will add the data contained in rx_buffer->page to the skb.
6269 *  This is done either through a direct copy if the data in the buffer is
6270 *  less than the skb header size, otherwise it will just attach the page as
6271 *  a frag to the skb.
6272 *
6273 *  The function will then update the page offset if necessary and return
6274 *  true if the buffer can be reused by the adapter.
6275 **/
6276static bool igb_add_rx_frag(struct igb_ring *rx_ring,
6277                            struct igb_rx_buffer *rx_buffer,
6278                            union e1000_adv_rx_desc *rx_desc,
6279                            struct sk_buff *skb)
6280{
6281        struct page *page = rx_buffer->page;
6282        unsigned int size = le16_to_cpu(rx_desc->wb.upper.length);
6283#if (PAGE_SIZE < 8192)
6284        unsigned int truesize = IGB_RX_BUFSZ;
6285#else
6286        unsigned int truesize = ALIGN(size, L1_CACHE_BYTES);
6287#endif
6288
6289        if ((size <= IGB_RX_HDR_LEN) && !skb_is_nonlinear(skb)) {
6290                unsigned char *va = page_address(page) + rx_buffer->page_offset;
6291
6292                if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6293                        igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6294                        va += IGB_TS_HDR_LEN;
6295                        size -= IGB_TS_HDR_LEN;
6296                }
6297
6298                memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
6299
6300                /* we can reuse buffer as-is, just make sure it is local */
6301                if (likely(page_to_nid(page) == numa_node_id()))
6302                        return true;
6303
6304                /* this page cannot be reused so discard it */
6305                put_page(page);
6306                return false;
6307        }
6308
6309        skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
6310                        rx_buffer->page_offset, size, truesize);
6311
6312        return igb_can_reuse_rx_page(rx_buffer, page, truesize);
6313}
6314
6315static struct sk_buff *igb_fetch_rx_buffer(struct igb_ring *rx_ring,
6316                                           union e1000_adv_rx_desc *rx_desc,
6317                                           struct sk_buff *skb)
6318{
6319        struct igb_rx_buffer *rx_buffer;
6320        struct page *page;
6321
6322        rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
6323
6324        page = rx_buffer->page;
6325        prefetchw(page);
6326
6327        if (likely(!skb)) {
6328                void *page_addr = page_address(page) +
6329                                  rx_buffer->page_offset;
6330
6331                /* prefetch first cache line of first page */
6332                prefetch(page_addr);
6333#if L1_CACHE_BYTES < 128
6334                prefetch(page_addr + L1_CACHE_BYTES);
6335#endif
6336
6337                /* allocate a skb to store the frags */
6338                skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
6339                                                IGB_RX_HDR_LEN);
6340                if (unlikely(!skb)) {
6341                        rx_ring->rx_stats.alloc_failed++;
6342                        return NULL;
6343                }
6344
6345                /* we will be copying header into skb->data in
6346                 * pskb_may_pull so it is in our interest to prefetch
6347                 * it now to avoid a possible cache miss
6348                 */
6349                prefetchw(skb->data);
6350        }
6351
6352        /* we are reusing so sync this buffer for CPU use */
6353        dma_sync_single_range_for_cpu(rx_ring->dev,
6354                                      rx_buffer->dma,
6355                                      rx_buffer->page_offset,
6356                                      IGB_RX_BUFSZ,
6357                                      DMA_FROM_DEVICE);
6358
6359        /* pull page into skb */
6360        if (igb_add_rx_frag(rx_ring, rx_buffer, rx_desc, skb)) {
6361                /* hand second half of page back to the ring */
6362                igb_reuse_rx_page(rx_ring, rx_buffer);
6363        } else {
6364                /* we are not reusing the buffer so unmap it */
6365                dma_unmap_page(rx_ring->dev, rx_buffer->dma,
6366                               PAGE_SIZE, DMA_FROM_DEVICE);
6367        }
6368
6369        /* clear contents of rx_buffer */
6370        rx_buffer->page = NULL;
6371
6372        return skb;
6373}
6374
6375static inline void igb_rx_checksum(struct igb_ring *ring,
6376                                   union e1000_adv_rx_desc *rx_desc,
6377                                   struct sk_buff *skb)
6378{
6379        skb_checksum_none_assert(skb);
6380
6381        /* Ignore Checksum bit is set */
6382        if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
6383                return;
6384
6385        /* Rx checksum disabled via ethtool */
6386        if (!(ring->netdev->features & NETIF_F_RXCSUM))
6387                return;
6388
6389        /* TCP/UDP checksum error bit is set */
6390        if (igb_test_staterr(rx_desc,
6391                             E1000_RXDEXT_STATERR_TCPE |
6392                             E1000_RXDEXT_STATERR_IPE)) {
6393                /* work around errata with sctp packets where the TCPE aka
6394                 * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
6395                 * packets, (aka let the stack check the crc32c)
6396                 */
6397                if (!((skb->len == 60) &&
6398                      test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
6399                        u64_stats_update_begin(&ring->rx_syncp);
6400                        ring->rx_stats.csum_err++;
6401                        u64_stats_update_end(&ring->rx_syncp);
6402                }
6403                /* let the stack verify checksum errors */
6404                return;
6405        }
6406        /* It must be a TCP or UDP packet with a valid checksum */
6407        if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
6408                                      E1000_RXD_STAT_UDPCS))
6409                skb->ip_summed = CHECKSUM_UNNECESSARY;
6410
6411        dev_dbg(ring->dev, "cksum success: bits %08X\n",
6412                le32_to_cpu(rx_desc->wb.upper.status_error));
6413}
6414
6415static inline void igb_rx_hash(struct igb_ring *ring,
6416                               union e1000_adv_rx_desc *rx_desc,
6417                               struct sk_buff *skb)
6418{
6419        if (ring->netdev->features & NETIF_F_RXHASH)
6420                skb->rxhash = le32_to_cpu(rx_desc->wb.lower.hi_dword.rss);
6421}
6422
6423/**
6424 *  igb_is_non_eop - process handling of non-EOP buffers
6425 *  @rx_ring: Rx ring being processed
6426 *  @rx_desc: Rx descriptor for current buffer
6427 *  @skb: current socket buffer containing buffer in progress
6428 *
6429 *  This function updates next to clean.  If the buffer is an EOP buffer
6430 *  this function exits returning false, otherwise it will place the
6431 *  sk_buff in the next buffer to be chained and return true indicating
6432 *  that this is in fact a non-EOP buffer.
6433 **/
6434static bool igb_is_non_eop(struct igb_ring *rx_ring,
6435                           union e1000_adv_rx_desc *rx_desc)
6436{
6437        u32 ntc = rx_ring->next_to_clean + 1;
6438
6439        /* fetch, update, and store next to clean */
6440        ntc = (ntc < rx_ring->count) ? ntc : 0;
6441        rx_ring->next_to_clean = ntc;
6442
6443        prefetch(IGB_RX_DESC(rx_ring, ntc));
6444
6445        if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
6446                return false;
6447
6448        return true;
6449}
6450
6451/**
6452 *  igb_get_headlen - determine size of header for LRO/GRO
6453 *  @data: pointer to the start of the headers
6454 *  @max_len: total length of section to find headers in
6455 *
6456 *  This function is meant to determine the length of headers that will
6457 *  be recognized by hardware for LRO, and GRO offloads.  The main
6458 *  motivation of doing this is to only perform one pull for IPv4 TCP
6459 *  packets so that we can do basic things like calculating the gso_size
6460 *  based on the average data per packet.
6461 **/
6462static unsigned int igb_get_headlen(unsigned char *data,
6463                                    unsigned int max_len)
6464{
6465        union {
6466                unsigned char *network;
6467                /* l2 headers */
6468                struct ethhdr *eth;
6469                struct vlan_hdr *vlan;
6470                /* l3 headers */
6471                struct iphdr *ipv4;
6472                struct ipv6hdr *ipv6;
6473        } hdr;
6474        __be16 protocol;
6475        u8 nexthdr = 0; /* default to not TCP */
6476        u8 hlen;
6477
6478        /* this should never happen, but better safe than sorry */
6479        if (max_len < ETH_HLEN)
6480                return max_len;
6481
6482        /* initialize network frame pointer */
6483        hdr.network = data;
6484
6485        /* set first protocol and move network header forward */
6486        protocol = hdr.eth->h_proto;
6487        hdr.network += ETH_HLEN;
6488
6489        /* handle any vlan tag if present */
6490        if (protocol == __constant_htons(ETH_P_8021Q)) {
6491                if ((hdr.network - data) > (max_len - VLAN_HLEN))
6492                        return max_len;
6493
6494                protocol = hdr.vlan->h_vlan_encapsulated_proto;
6495                hdr.network += VLAN_HLEN;
6496        }
6497
6498        /* handle L3 protocols */
6499        if (protocol == __constant_htons(ETH_P_IP)) {
6500                if ((hdr.network - data) > (max_len - sizeof(struct iphdr)))
6501                        return max_len;
6502
6503                /* access ihl as a u8 to avoid unaligned access on ia64 */
6504                hlen = (hdr.network[0] & 0x0F) << 2;
6505
6506                /* verify hlen meets minimum size requirements */
6507                if (hlen < sizeof(struct iphdr))
6508                        return hdr.network - data;
6509
6510                /* record next protocol if header is present */
6511                if (!(hdr.ipv4->frag_off & htons(IP_OFFSET)))
6512                        nexthdr = hdr.ipv4->protocol;
6513        } else if (protocol == __constant_htons(ETH_P_IPV6)) {
6514                if ((hdr.network - data) > (max_len - sizeof(struct ipv6hdr)))
6515                        return max_len;
6516
6517                /* record next protocol */
6518                nexthdr = hdr.ipv6->nexthdr;
6519                hlen = sizeof(struct ipv6hdr);
6520        } else {
6521                return hdr.network - data;
6522        }
6523
6524        /* relocate pointer to start of L4 header */
6525        hdr.network += hlen;
6526
6527        /* finally sort out TCP */
6528        if (nexthdr == IPPROTO_TCP) {
6529                if ((hdr.network - data) > (max_len - sizeof(struct tcphdr)))
6530                        return max_len;
6531
6532                /* access doff as a u8 to avoid unaligned access on ia64 */
6533                hlen = (hdr.network[12] & 0xF0) >> 2;
6534
6535                /* verify hlen meets minimum size requirements */
6536                if (hlen < sizeof(struct tcphdr))
6537                        return hdr.network - data;
6538
6539                hdr.network += hlen;
6540        } else if (nexthdr == IPPROTO_UDP) {
6541                if ((hdr.network - data) > (max_len - sizeof(struct udphdr)))
6542                        return max_len;
6543
6544                hdr.network += sizeof(struct udphdr);
6545        }
6546
6547        /* If everything has gone correctly hdr.network should be the
6548         * data section of the packet and will be the end of the header.
6549         * If not then it probably represents the end of the last recognized
6550         * header.
6551         */
6552        if ((hdr.network - data) < max_len)
6553                return hdr.network - data;
6554        else
6555                return max_len;
6556}
6557
6558/**
6559 *  igb_pull_tail - igb specific version of skb_pull_tail
6560 *  @rx_ring: rx descriptor ring packet is being transacted on
6561 *  @rx_desc: pointer to the EOP Rx descriptor
6562 *  @skb: pointer to current skb being adjusted
6563 *
6564 *  This function is an igb specific version of __pskb_pull_tail.  The
6565 *  main difference between this version and the original function is that
6566 *  this function can make several assumptions about the state of things
6567 *  that allow for significant optimizations versus the standard function.
6568 *  As a result we can do things like drop a frag and maintain an accurate
6569 *  truesize for the skb.
6570 */
6571static void igb_pull_tail(struct igb_ring *rx_ring,
6572                          union e1000_adv_rx_desc *rx_desc,
6573                          struct sk_buff *skb)
6574{
6575        struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[0];
6576        unsigned char *va;
6577        unsigned int pull_len;
6578
6579        /* it is valid to use page_address instead of kmap since we are
6580         * working with pages allocated out of the lomem pool per
6581         * alloc_page(GFP_ATOMIC)
6582         */
6583        va = skb_frag_address(frag);
6584
6585        if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
6586                /* retrieve timestamp from buffer */
6587                igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
6588
6589                /* update pointers to remove timestamp header */
6590                skb_frag_size_sub(frag, IGB_TS_HDR_LEN);
6591                frag->page_offset += IGB_TS_HDR_LEN;
6592                skb->data_len -= IGB_TS_HDR_LEN;
6593                skb->len -= IGB_TS_HDR_LEN;
6594
6595                /* move va to start of packet data */
6596                va += IGB_TS_HDR_LEN;
6597        }
6598
6599        /* we need the header to contain the greater of either ETH_HLEN or
6600         * 60 bytes if the skb->len is less than 60 for skb_pad.
6601         */
6602        pull_len = igb_get_headlen(va, IGB_RX_HDR_LEN);
6603
6604        /* align pull length to size of long to optimize memcpy performance */
6605        skb_copy_to_linear_data(skb, va, ALIGN(pull_len, sizeof(long)));
6606
6607        /* update all of the pointers */
6608        skb_frag_size_sub(frag, pull_len);
6609        frag->page_offset += pull_len;
6610        skb->data_len -= pull_len;
6611        skb->tail += pull_len;
6612}
6613
6614/**
6615 *  igb_cleanup_headers - Correct corrupted or empty headers
6616 *  @rx_ring: rx descriptor ring packet is being transacted on
6617 *  @rx_desc: pointer to the EOP Rx descriptor
6618 *  @skb: pointer to current skb being fixed
6619 *
6620 *  Address the case where we are pulling data in on pages only
6621 *  and as such no data is present in the skb header.
6622 *
6623 *  In addition if skb is not at least 60 bytes we need to pad it so that
6624 *  it is large enough to qualify as a valid Ethernet frame.
6625 *
6626 *  Returns true if an error was encountered and skb was freed.
6627 **/
6628static bool igb_cleanup_headers(struct igb_ring *rx_ring,
6629                                union e1000_adv_rx_desc *rx_desc,
6630                                struct sk_buff *skb)
6631{
6632        if (unlikely((igb_test_staterr(rx_desc,
6633                                       E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
6634                struct net_device *netdev = rx_ring->netdev;
6635                if (!(netdev->features & NETIF_F_RXALL)) {
6636                        dev_kfree_skb_any(skb);
6637                        return true;
6638                }
6639        }
6640
6641        /* place header in linear portion of buffer */
6642        if (skb_is_nonlinear(skb))
6643                igb_pull_tail(rx_ring, rx_desc, skb);
6644
6645        /* if skb_pad returns an error the skb was freed */
6646        if (unlikely(skb->len < 60)) {
6647                int pad_len = 60 - skb->len;
6648
6649                if (skb_pad(skb, pad_len))
6650                        return true;
6651                __skb_put(skb, pad_len);
6652        }
6653
6654        return false;
6655}
6656
6657/**
6658 *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
6659 *  @rx_ring: rx descriptor ring packet is being transacted on
6660 *  @rx_desc: pointer to the EOP Rx descriptor
6661 *  @skb: pointer to current skb being populated
6662 *
6663 *  This function checks the ring, descriptor, and packet information in
6664 *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
6665 *  other fields within the skb.
6666 **/
6667static void igb_process_skb_fields(struct igb_ring *rx_ring,
6668                                   union e1000_adv_rx_desc *rx_desc,
6669                                   struct sk_buff *skb)
6670{
6671        struct net_device *dev = rx_ring->netdev;
6672
6673        igb_rx_hash(rx_ring, rx_desc, skb);
6674
6675        igb_rx_checksum(rx_ring, rx_desc, skb);
6676
6677        igb_ptp_rx_hwtstamp(rx_ring, rx_desc, skb);
6678
6679        if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
6680            igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
6681                u16 vid;
6682                if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
6683                    test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
6684                        vid = be16_to_cpu(rx_desc->wb.upper.vlan);
6685                else
6686                        vid = le16_to_cpu(rx_desc->wb.upper.vlan);
6687
6688                __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
6689        }
6690
6691        skb_record_rx_queue(skb, rx_ring->queue_index);
6692
6693        skb->protocol = eth_type_trans(skb, rx_ring->netdev);
6694}
6695
6696static bool igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
6697{
6698        struct igb_ring *rx_ring = q_vector->rx.ring;
6699        struct sk_buff *skb = rx_ring->skb;
6700        unsigned int total_bytes = 0, total_packets = 0;
6701        u16 cleaned_count = igb_desc_unused(rx_ring);
6702
6703        do {
6704                union e1000_adv_rx_desc *rx_desc;
6705
6706                /* return some buffers to hardware, one at a time is too slow */
6707                if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
6708                        igb_alloc_rx_buffers(rx_ring, cleaned_count);
6709                        cleaned_count = 0;
6710                }
6711
6712                rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
6713
6714                if (!igb_test_staterr(rx_desc, E1000_RXD_STAT_DD))
6715                        break;
6716
6717                /* This memory barrier is needed to keep us from reading
6718                 * any other fields out of the rx_desc until we know the
6719                 * RXD_STAT_DD bit is set
6720                 */
6721                rmb();
6722
6723                /* retrieve a buffer from the ring */
6724                skb = igb_fetch_rx_buffer(rx_ring, rx_desc, skb);
6725
6726                /* exit if we failed to retrieve a buffer */
6727                if (!skb)
6728                        break;
6729
6730                cleaned_count++;
6731
6732                /* fetch next buffer in frame if non-eop */
6733                if (igb_is_non_eop(rx_ring, rx_desc))
6734                        continue;
6735
6736                /* verify the packet layout is correct */
6737                if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
6738                        skb = NULL;
6739                        continue;
6740                }
6741
6742                /* probably a little skewed due to removing CRC */
6743                total_bytes += skb->len;
6744
6745                /* populate checksum, timestamp, VLAN, and protocol */
6746                igb_process_skb_fields(rx_ring, rx_desc, skb);
6747
6748                napi_gro_receive(&q_vector->napi, skb);
6749
6750                /* reset skb pointer */
6751                skb = NULL;
6752
6753                /* update budget accounting */
6754                total_packets++;
6755        } while (likely(total_packets < budget));
6756
6757        /* place incomplete frames back on ring for completion */
6758        rx_ring->skb = skb;
6759
6760        u64_stats_update_begin(&rx_ring->rx_syncp);
6761        rx_ring->rx_stats.packets += total_packets;
6762        rx_ring->rx_stats.bytes += total_bytes;
6763        u64_stats_update_end(&rx_ring->rx_syncp);
6764        q_vector->rx.total_packets += total_packets;
6765        q_vector->rx.total_bytes += total_bytes;
6766
6767        if (cleaned_count)
6768                igb_alloc_rx_buffers(rx_ring, cleaned_count);
6769
6770        return (total_packets < budget);
6771}
6772
6773static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
6774                                  struct igb_rx_buffer *bi)
6775{
6776        struct page *page = bi->page;
6777        dma_addr_t dma;
6778
6779        /* since we are recycling buffers we should seldom need to alloc */
6780        if (likely(page))
6781                return true;
6782
6783        /* alloc new page for storage */
6784        page = __skb_alloc_page(GFP_ATOMIC | __GFP_COLD, NULL);
6785        if (unlikely(!page)) {
6786                rx_ring->rx_stats.alloc_failed++;
6787                return false;
6788        }
6789
6790        /* map page for use */
6791        dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
6792
6793        /* if mapping failed free memory back to system since
6794         * there isn't much point in holding memory we can't use
6795         */
6796        if (dma_mapping_error(rx_ring->dev, dma)) {
6797                __free_page(page);
6798
6799                rx_ring->rx_stats.alloc_failed++;
6800                return false;
6801        }
6802
6803        bi->dma = dma;
6804        bi->page = page;
6805        bi->page_offset = 0;
6806
6807        return true;
6808}
6809
6810/**
6811 *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
6812 *  @adapter: address of board private structure
6813 **/
6814void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
6815{
6816        union e1000_adv_rx_desc *rx_desc;
6817        struct igb_rx_buffer *bi;
6818        u16 i = rx_ring->next_to_use;
6819
6820        /* nothing to do */
6821        if (!cleaned_count)
6822                return;
6823
6824        rx_desc = IGB_RX_DESC(rx_ring, i);
6825        bi = &rx_ring->rx_buffer_info[i];
6826        i -= rx_ring->count;
6827
6828        do {
6829                if (!igb_alloc_mapped_page(rx_ring, bi))
6830                        break;
6831
6832                /* Refresh the desc even if buffer_addrs didn't change
6833                 * because each write-back erases this info.
6834                 */
6835                rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
6836
6837                rx_desc++;
6838                bi++;
6839                i++;
6840                if (unlikely(!i)) {
6841                        rx_desc = IGB_RX_DESC(rx_ring, 0);
6842                        bi = rx_ring->rx_buffer_info;
6843                        i -= rx_ring->count;
6844                }
6845
6846                /* clear the hdr_addr for the next_to_use descriptor */
6847                rx_desc->read.hdr_addr = 0;
6848
6849                cleaned_count--;
6850        } while (cleaned_count);
6851
6852        i += rx_ring->count;
6853
6854        if (rx_ring->next_to_use != i) {
6855                /* record the next descriptor to use */
6856                rx_ring->next_to_use = i;
6857
6858                /* update next to alloc since we have filled the ring */
6859                rx_ring->next_to_alloc = i;
6860
6861                /* Force memory writes to complete before letting h/w
6862                 * know there are new descriptors to fetch.  (Only
6863                 * applicable for weak-ordered memory model archs,
6864                 * such as IA-64).
6865                 */
6866                wmb();
6867                writel(i, rx_ring->tail);
6868        }
6869}
6870
6871/**
6872 * igb_mii_ioctl -
6873 * @netdev:
6874 * @ifreq:
6875 * @cmd:
6876 **/
6877static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6878{
6879        struct igb_adapter *adapter = netdev_priv(netdev);
6880        struct mii_ioctl_data *data = if_mii(ifr);
6881
6882        if (adapter->hw.phy.media_type != e1000_media_type_copper)
6883                return -EOPNOTSUPP;
6884
6885        switch (cmd) {
6886        case SIOCGMIIPHY:
6887                data->phy_id = adapter->hw.phy.addr;
6888                break;
6889        case SIOCGMIIREG:
6890                if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
6891                                     &data->val_out))
6892                        return -EIO;
6893                break;
6894        case SIOCSMIIREG:
6895        default:
6896                return -EOPNOTSUPP;
6897        }
6898        return 0;
6899}
6900
6901/**
6902 * igb_ioctl -
6903 * @netdev:
6904 * @ifreq:
6905 * @cmd:
6906 **/
6907static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
6908{
6909        switch (cmd) {
6910        case SIOCGMIIPHY:
6911        case SIOCGMIIREG:
6912        case SIOCSMIIREG:
6913                return igb_mii_ioctl(netdev, ifr, cmd);
6914        case SIOCSHWTSTAMP:
6915                return igb_ptp_hwtstamp_ioctl(netdev, ifr, cmd);
6916        default:
6917                return -EOPNOTSUPP;
6918        }
6919}
6920
6921s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6922{
6923        struct igb_adapter *adapter = hw->back;
6924
6925        if (pcie_capability_read_word(adapter->pdev, reg, value))
6926                return -E1000_ERR_CONFIG;
6927
6928        return 0;
6929}
6930
6931s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
6932{
6933        struct igb_adapter *adapter = hw->back;
6934
6935        if (pcie_capability_write_word(adapter->pdev, reg, *value))
6936                return -E1000_ERR_CONFIG;
6937
6938        return 0;
6939}
6940
6941static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
6942{
6943        struct igb_adapter *adapter = netdev_priv(netdev);
6944        struct e1000_hw *hw = &adapter->hw;
6945        u32 ctrl, rctl;
6946        bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
6947
6948        if (enable) {
6949                /* enable VLAN tag insert/strip */
6950                ctrl = rd32(E1000_CTRL);
6951                ctrl |= E1000_CTRL_VME;
6952                wr32(E1000_CTRL, ctrl);
6953
6954                /* Disable CFI check */
6955                rctl = rd32(E1000_RCTL);
6956                rctl &= ~E1000_RCTL_CFIEN;
6957                wr32(E1000_RCTL, rctl);
6958        } else {
6959                /* disable VLAN tag insert/strip */
6960                ctrl = rd32(E1000_CTRL);
6961                ctrl &= ~E1000_CTRL_VME;
6962                wr32(E1000_CTRL, ctrl);
6963        }
6964
6965        igb_rlpml_set(adapter);
6966}
6967
6968static int igb_vlan_rx_add_vid(struct net_device *netdev,
6969                               __be16 proto, u16 vid)
6970{
6971        struct igb_adapter *adapter = netdev_priv(netdev);
6972        struct e1000_hw *hw = &adapter->hw;
6973        int pf_id = adapter->vfs_allocated_count;
6974
6975        /* attempt to add filter to vlvf array */
6976        igb_vlvf_set(adapter, vid, true, pf_id);
6977
6978        /* add the filter since PF can receive vlans w/o entry in vlvf */
6979        igb_vfta_set(hw, vid, true);
6980
6981        set_bit(vid, adapter->active_vlans);
6982
6983        return 0;
6984}
6985
6986static int igb_vlan_rx_kill_vid(struct net_device *netdev,
6987                                __be16 proto, u16 vid)
6988{
6989        struct igb_adapter *adapter = netdev_priv(netdev);
6990        struct e1000_hw *hw = &adapter->hw;
6991        int pf_id = adapter->vfs_allocated_count;
6992        s32 err;
6993
6994        /* remove vlan from VLVF table array */
6995        err = igb_vlvf_set(adapter, vid, false, pf_id);
6996
6997        /* if vid was not present in VLVF just remove it from table */
6998        if (err)
6999                igb_vfta_set(hw, vid, false);
7000
7001        clear_bit(vid, adapter->active_vlans);
7002
7003        return 0;
7004}
7005
7006static void igb_restore_vlan(struct igb_adapter *adapter)
7007{
7008        u16 vid;
7009
7010        igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7011
7012        for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
7013                igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7014}
7015
7016int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7017{
7018        struct pci_dev *pdev = adapter->pdev;
7019        struct e1000_mac_info *mac = &adapter->hw.mac;
7020
7021        mac->autoneg = 0;
7022
7023        /* Make sure dplx is at most 1 bit and lsb of speed is not set
7024         * for the switch() below to work
7025         */
7026        if ((spd & 1) || (dplx & ~1))
7027                goto err_inval;
7028
7029        /* Fiber NIC's only allow 1000 gbps Full duplex
7030         * and 100Mbps Full duplex for 100baseFx sfp
7031         */
7032        if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7033                switch (spd + dplx) {
7034                case SPEED_10 + DUPLEX_HALF:
7035                case SPEED_10 + DUPLEX_FULL:
7036                case SPEED_100 + DUPLEX_HALF:
7037                        goto err_inval;
7038                default:
7039                        break;
7040                }
7041        }
7042
7043        switch (spd + dplx) {
7044        case SPEED_10 + DUPLEX_HALF:
7045                mac->forced_speed_duplex = ADVERTISE_10_HALF;
7046                break;
7047        case SPEED_10 + DUPLEX_FULL:
7048                mac->forced_speed_duplex = ADVERTISE_10_FULL;
7049                break;
7050        case SPEED_100 + DUPLEX_HALF:
7051                mac->forced_speed_duplex = ADVERTISE_100_HALF;
7052                break;
7053        case SPEED_100 + DUPLEX_FULL:
7054                mac->forced_speed_duplex = ADVERTISE_100_FULL;
7055                break;
7056        case SPEED_1000 + DUPLEX_FULL:
7057                mac->autoneg = 1;
7058                adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7059                break;
7060        case SPEED_1000 + DUPLEX_HALF: /* not supported */
7061        default:
7062                goto err_inval;
7063        }
7064
7065        /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7066        adapter->hw.phy.mdix = AUTO_ALL_MODES;
7067
7068        return 0;
7069
7070err_inval:
7071        dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7072        return -EINVAL;
7073}
7074
7075static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7076                          bool runtime)
7077{
7078        struct net_device *netdev = pci_get_drvdata(pdev);
7079        struct igb_adapter *adapter = netdev_priv(netdev);
7080        struct e1000_hw *hw = &adapter->hw;
7081        u32 ctrl, rctl, status;
7082        u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7083#ifdef CONFIG_PM
7084        int retval = 0;
7085#endif
7086
7087        netif_device_detach(netdev);
7088
7089        if (netif_running(netdev))
7090                __igb_close(netdev, true);
7091
7092        igb_clear_interrupt_scheme(adapter);
7093
7094#ifdef CONFIG_PM
7095        retval = pci_save_state(pdev);
7096        if (retval)
7097                return retval;
7098#endif
7099
7100        status = rd32(E1000_STATUS);
7101        if (status & E1000_STATUS_LU)
7102                wufc &= ~E1000_WUFC_LNKC;
7103
7104        if (wufc) {
7105                igb_setup_rctl(adapter);
7106                igb_set_rx_mode(netdev);
7107
7108                /* turn on all-multi mode if wake on multicast is enabled */
7109                if (wufc & E1000_WUFC_MC) {
7110                        rctl = rd32(E1000_RCTL);
7111                        rctl |= E1000_RCTL_MPE;
7112                        wr32(E1000_RCTL, rctl);
7113                }
7114
7115                ctrl = rd32(E1000_CTRL);
7116                /* advertise wake from D3Cold */
7117                #define E1000_CTRL_ADVD3WUC 0x00100000
7118                /* phy power management enable */
7119                #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7120                ctrl |= E1000_CTRL_ADVD3WUC;
7121                wr32(E1000_CTRL, ctrl);
7122
7123                /* Allow time for pending master requests to run */
7124                igb_disable_pcie_master(hw);
7125
7126                wr32(E1000_WUC, E1000_WUC_PME_EN);
7127                wr32(E1000_WUFC, wufc);
7128        } else {
7129                wr32(E1000_WUC, 0);
7130                wr32(E1000_WUFC, 0);
7131        }
7132
7133        *enable_wake = wufc || adapter->en_mng_pt;
7134        if (!*enable_wake)
7135                igb_power_down_link(adapter);
7136        else
7137                igb_power_up_link(adapter);
7138
7139        /* Release control of h/w to f/w.  If f/w is AMT enabled, this
7140         * would have already happened in close and is redundant.
7141         */
7142        igb_release_hw_control(adapter);
7143
7144        pci_disable_device(pdev);
7145
7146        return 0;
7147}
7148
7149#ifdef CONFIG_PM
7150#ifdef CONFIG_PM_SLEEP
7151static int igb_suspend(struct device *dev)
7152{
7153        int retval;
7154        bool wake;
7155        struct pci_dev *pdev = to_pci_dev(dev);
7156
7157        retval = __igb_shutdown(pdev, &wake, 0);
7158        if (retval)
7159                return retval;
7160
7161        if (wake) {
7162                pci_prepare_to_sleep(pdev);
7163        } else {
7164                pci_wake_from_d3(pdev, false);
7165                pci_set_power_state(pdev, PCI_D3hot);
7166        }
7167
7168        return 0;
7169}
7170#endif /* CONFIG_PM_SLEEP */
7171
7172static int igb_resume(struct device *dev)
7173{
7174        struct pci_dev *pdev = to_pci_dev(dev);
7175        struct net_device *netdev = pci_get_drvdata(pdev);
7176        struct igb_adapter *adapter = netdev_priv(netdev);
7177        struct e1000_hw *hw = &adapter->hw;
7178        u32 err;
7179
7180        pci_set_power_state(pdev, PCI_D0);
7181        pci_restore_state(pdev);
7182        pci_save_state(pdev);
7183
7184        err = pci_enable_device_mem(pdev);
7185        if (err) {
7186                dev_err(&pdev->dev,
7187                        "igb: Cannot enable PCI device from suspend\n");
7188                return err;
7189        }
7190        pci_set_master(pdev);
7191
7192        pci_enable_wake(pdev, PCI_D3hot, 0);
7193        pci_enable_wake(pdev, PCI_D3cold, 0);
7194
7195        if (igb_init_interrupt_scheme(adapter, true)) {
7196                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7197                return -ENOMEM;
7198        }
7199
7200        igb_reset(adapter);
7201
7202        /* let the f/w know that the h/w is now under the control of the
7203         * driver.
7204         */
7205        igb_get_hw_control(adapter);
7206
7207        wr32(E1000_WUS, ~0);
7208
7209        if (netdev->flags & IFF_UP) {
7210                rtnl_lock();
7211                err = __igb_open(netdev, true);
7212                rtnl_unlock();
7213                if (err)
7214                        return err;
7215        }
7216
7217        netif_device_attach(netdev);
7218        return 0;
7219}
7220
7221#ifdef CONFIG_PM_RUNTIME
7222static int igb_runtime_idle(struct device *dev)
7223{
7224        struct pci_dev *pdev = to_pci_dev(dev);
7225        struct net_device *netdev = pci_get_drvdata(pdev);
7226        struct igb_adapter *adapter = netdev_priv(netdev);
7227
7228        if (!igb_has_link(adapter))
7229                pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
7230
7231        return -EBUSY;
7232}
7233
7234static int igb_runtime_suspend(struct device *dev)
7235{
7236        struct pci_dev *pdev = to_pci_dev(dev);
7237        int retval;
7238        bool wake;
7239
7240        retval = __igb_shutdown(pdev, &wake, 1);
7241        if (retval)
7242                return retval;
7243
7244        if (wake) {
7245                pci_prepare_to_sleep(pdev);
7246        } else {
7247                pci_wake_from_d3(pdev, false);
7248                pci_set_power_state(pdev, PCI_D3hot);
7249        }
7250
7251        return 0;
7252}
7253
7254static int igb_runtime_resume(struct device *dev)
7255{
7256        return igb_resume(dev);
7257}
7258#endif /* CONFIG_PM_RUNTIME */
7259#endif
7260
7261static void igb_shutdown(struct pci_dev *pdev)
7262{
7263        bool wake;
7264
7265        __igb_shutdown(pdev, &wake, 0);
7266
7267        if (system_state == SYSTEM_POWER_OFF) {
7268                pci_wake_from_d3(pdev, wake);
7269                pci_set_power_state(pdev, PCI_D3hot);
7270        }
7271}
7272
7273#ifdef CONFIG_PCI_IOV
7274static int igb_sriov_reinit(struct pci_dev *dev)
7275{
7276        struct net_device *netdev = pci_get_drvdata(dev);
7277        struct igb_adapter *adapter = netdev_priv(netdev);
7278        struct pci_dev *pdev = adapter->pdev;
7279
7280        rtnl_lock();
7281
7282        if (netif_running(netdev))
7283                igb_close(netdev);
7284
7285        igb_clear_interrupt_scheme(adapter);
7286
7287        igb_init_queue_configuration(adapter);
7288
7289        if (igb_init_interrupt_scheme(adapter, true)) {
7290                dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
7291                return -ENOMEM;
7292        }
7293
7294        if (netif_running(netdev))
7295                igb_open(netdev);
7296
7297        rtnl_unlock();
7298
7299        return 0;
7300}
7301
7302static int igb_pci_disable_sriov(struct pci_dev *dev)
7303{
7304        int err = igb_disable_sriov(dev);
7305
7306        if (!err)
7307                err = igb_sriov_reinit(dev);
7308
7309        return err;
7310}
7311
7312static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
7313{
7314        int err = igb_enable_sriov(dev, num_vfs);
7315
7316        if (err)
7317                goto out;
7318
7319        err = igb_sriov_reinit(dev);
7320        if (!err)
7321                return num_vfs;
7322
7323out:
7324        return err;
7325}
7326
7327#endif
7328static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
7329{
7330#ifdef CONFIG_PCI_IOV
7331        if (num_vfs == 0)
7332                return igb_pci_disable_sriov(dev);
7333        else
7334                return igb_pci_enable_sriov(dev, num_vfs);
7335#endif
7336        return 0;
7337}
7338
7339#ifdef CONFIG_NET_POLL_CONTROLLER
7340/* Polling 'interrupt' - used by things like netconsole to send skbs
7341 * without having to re-enable interrupts. It's not called while
7342 * the interrupt routine is executing.
7343 */
7344static void igb_netpoll(struct net_device *netdev)
7345{
7346        struct igb_adapter *adapter = netdev_priv(netdev);
7347        struct e1000_hw *hw = &adapter->hw;
7348        struct igb_q_vector *q_vector;
7349        int i;
7350
7351        for (i = 0; i < adapter->num_q_vectors; i++) {
7352                q_vector = adapter->q_vector[i];
7353                if (adapter->msix_entries)
7354                        wr32(E1000_EIMC, q_vector->eims_value);
7355                else
7356                        igb_irq_disable(adapter);
7357                napi_schedule(&q_vector->napi);
7358        }
7359}
7360#endif /* CONFIG_NET_POLL_CONTROLLER */
7361
7362/**
7363 *  igb_io_error_detected - called when PCI error is detected
7364 *  @pdev: Pointer to PCI device
7365 *  @state: The current pci connection state
7366 *
7367 *  This function is called after a PCI bus error affecting
7368 *  this device has been detected.
7369 **/
7370static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
7371                                              pci_channel_state_t state)
7372{
7373        struct net_device *netdev = pci_get_drvdata(pdev);
7374        struct igb_adapter *adapter = netdev_priv(netdev);
7375
7376        netif_device_detach(netdev);
7377
7378        if (state == pci_channel_io_perm_failure)
7379                return PCI_ERS_RESULT_DISCONNECT;
7380
7381        if (netif_running(netdev))
7382                igb_down(adapter);
7383        pci_disable_device(pdev);
7384
7385        /* Request a slot slot reset. */
7386        return PCI_ERS_RESULT_NEED_RESET;
7387}
7388
7389/**
7390 *  igb_io_slot_reset - called after the pci bus has been reset.
7391 *  @pdev: Pointer to PCI device
7392 *
7393 *  Restart the card from scratch, as if from a cold-boot. Implementation
7394 *  resembles the first-half of the igb_resume routine.
7395 **/
7396static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
7397{
7398        struct net_device *netdev = pci_get_drvdata(pdev);
7399        struct igb_adapter *adapter = netdev_priv(netdev);
7400        struct e1000_hw *hw = &adapter->hw;
7401        pci_ers_result_t result;
7402        int err;
7403
7404        if (pci_enable_device_mem(pdev)) {
7405                dev_err(&pdev->dev,
7406                        "Cannot re-enable PCI device after reset.\n");
7407                result = PCI_ERS_RESULT_DISCONNECT;
7408        } else {
7409                pci_set_master(pdev);
7410                pci_restore_state(pdev);
7411                pci_save_state(pdev);
7412
7413                pci_enable_wake(pdev, PCI_D3hot, 0);
7414                pci_enable_wake(pdev, PCI_D3cold, 0);
7415
7416                igb_reset(adapter);
7417                wr32(E1000_WUS, ~0);
7418                result = PCI_ERS_RESULT_RECOVERED;
7419        }
7420
7421        err = pci_cleanup_aer_uncorrect_error_status(pdev);
7422        if (err) {
7423                dev_err(&pdev->dev,
7424                        "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
7425                        err);
7426                /* non-fatal, continue */
7427        }
7428
7429        return result;
7430}
7431
7432/**
7433 *  igb_io_resume - called when traffic can start flowing again.
7434 *  @pdev: Pointer to PCI device
7435 *
7436 *  This callback is called when the error recovery driver tells us that
7437 *  its OK to resume normal operation. Implementation resembles the
7438 *  second-half of the igb_resume routine.
7439 */
7440static void igb_io_resume(struct pci_dev *pdev)
7441{
7442        struct net_device *netdev = pci_get_drvdata(pdev);
7443        struct igb_adapter *adapter = netdev_priv(netdev);
7444
7445        if (netif_running(netdev)) {
7446                if (igb_up(adapter)) {
7447                        dev_err(&pdev->dev, "igb_up failed after reset\n");
7448                        return;
7449                }
7450        }
7451
7452        netif_device_attach(netdev);
7453
7454        /* let the f/w know that the h/w is now under the control of the
7455         * driver.
7456         */
7457        igb_get_hw_control(adapter);
7458}
7459
7460static void igb_rar_set_qsel(struct igb_adapter *adapter, u8 *addr, u32 index,
7461                             u8 qsel)
7462{
7463        u32 rar_low, rar_high;
7464        struct e1000_hw *hw = &adapter->hw;
7465
7466        /* HW expects these in little endian so we reverse the byte order
7467         * from network order (big endian) to little endian
7468         */
7469        rar_low = ((u32) addr[0] | ((u32) addr[1] << 8) |
7470                   ((u32) addr[2] << 16) | ((u32) addr[3] << 24));
7471        rar_high = ((u32) addr[4] | ((u32) addr[5] << 8));
7472
7473        /* Indicate to hardware the Address is Valid. */
7474        rar_high |= E1000_RAH_AV;
7475
7476        if (hw->mac.type == e1000_82575)
7477                rar_high |= E1000_RAH_POOL_1 * qsel;
7478        else
7479                rar_high |= E1000_RAH_POOL_1 << qsel;
7480
7481        wr32(E1000_RAL(index), rar_low);
7482        wrfl();
7483        wr32(E1000_RAH(index), rar_high);
7484        wrfl();
7485}
7486
7487static int igb_set_vf_mac(struct igb_adapter *adapter,
7488                          int vf, unsigned char *mac_addr)
7489{
7490        struct e1000_hw *hw = &adapter->hw;
7491        /* VF MAC addresses start at end of receive addresses and moves
7492         * towards the first, as a result a collision should not be possible
7493         */
7494        int rar_entry = hw->mac.rar_entry_count - (vf + 1);
7495
7496        memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
7497
7498        igb_rar_set_qsel(adapter, mac_addr, rar_entry, vf);
7499
7500        return 0;
7501}
7502
7503static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
7504{
7505        struct igb_adapter *adapter = netdev_priv(netdev);
7506        if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
7507                return -EINVAL;
7508        adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
7509        dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
7510        dev_info(&adapter->pdev->dev,
7511                 "Reload the VF driver to make this change effective.");
7512        if (test_bit(__IGB_DOWN, &adapter->state)) {
7513                dev_warn(&adapter->pdev->dev,
7514                         "The VF MAC address has been set, but the PF device is not up.\n");
7515                dev_warn(&adapter->pdev->dev,
7516                         "Bring the PF device up before attempting to use the VF device.\n");
7517        }
7518        return igb_set_vf_mac(adapter, vf, mac);
7519}
7520
7521static int igb_link_mbps(int internal_link_speed)
7522{
7523        switch (internal_link_speed) {
7524        case SPEED_100:
7525                return 100;
7526        case SPEED_1000:
7527                return 1000;
7528        default:
7529                return 0;
7530        }
7531}
7532
7533static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
7534                                  int link_speed)
7535{
7536        int rf_dec, rf_int;
7537        u32 bcnrc_val;
7538
7539        if (tx_rate != 0) {
7540                /* Calculate the rate factor values to set */
7541                rf_int = link_speed / tx_rate;
7542                rf_dec = (link_speed - (rf_int * tx_rate));
7543                rf_dec = (rf_dec * (1 << E1000_RTTBCNRC_RF_INT_SHIFT)) /
7544                         tx_rate;
7545
7546                bcnrc_val = E1000_RTTBCNRC_RS_ENA;
7547                bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
7548                              E1000_RTTBCNRC_RF_INT_MASK);
7549                bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
7550        } else {
7551                bcnrc_val = 0;
7552        }
7553
7554        wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
7555        /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
7556         * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
7557         */
7558        wr32(E1000_RTTBCNRM, 0x14);
7559        wr32(E1000_RTTBCNRC, bcnrc_val);
7560}
7561
7562static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
7563{
7564        int actual_link_speed, i;
7565        bool reset_rate = false;
7566
7567        /* VF TX rate limit was not set or not supported */
7568        if ((adapter->vf_rate_link_speed == 0) ||
7569            (adapter->hw.mac.type != e1000_82576))
7570                return;
7571
7572        actual_link_speed = igb_link_mbps(adapter->link_speed);
7573        if (actual_link_speed != adapter->vf_rate_link_speed) {
7574                reset_rate = true;
7575                adapter->vf_rate_link_speed = 0;
7576                dev_info(&adapter->pdev->dev,
7577                         "Link speed has been changed. VF Transmit rate is disabled\n");
7578        }
7579
7580        for (i = 0; i < adapter->vfs_allocated_count; i++) {
7581                if (reset_rate)
7582                        adapter->vf_data[i].tx_rate = 0;
7583
7584                igb_set_vf_rate_limit(&adapter->hw, i,
7585                                      adapter->vf_data[i].tx_rate,
7586                                      actual_link_speed);
7587        }
7588}
7589
7590static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf, int tx_rate)
7591{
7592        struct igb_adapter *adapter = netdev_priv(netdev);
7593        struct e1000_hw *hw = &adapter->hw;
7594        int actual_link_speed;
7595
7596        if (hw->mac.type != e1000_82576)
7597                return -EOPNOTSUPP;
7598
7599        actual_link_speed = igb_link_mbps(adapter->link_speed);
7600        if ((vf >= adapter->vfs_allocated_count) ||
7601            (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
7602            (tx_rate < 0) || (tx_rate > actual_link_speed))
7603                return -EINVAL;
7604
7605        adapter->vf_rate_link_speed = actual_link_speed;
7606        adapter->vf_data[vf].tx_rate = (u16)tx_rate;
7607        igb_set_vf_rate_limit(hw, vf, tx_rate, actual_link_speed);
7608
7609        return 0;
7610}
7611
7612static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
7613                                   bool setting)
7614{
7615        struct igb_adapter *adapter = netdev_priv(netdev);
7616        struct e1000_hw *hw = &adapter->hw;
7617        u32 reg_val, reg_offset;
7618
7619        if (!adapter->vfs_allocated_count)
7620                return -EOPNOTSUPP;
7621
7622        if (vf >= adapter->vfs_allocated_count)
7623                return -EINVAL;
7624
7625        reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
7626        reg_val = rd32(reg_offset);
7627        if (setting)
7628                reg_val |= ((1 << vf) |
7629                            (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7630        else
7631                reg_val &= ~((1 << vf) |
7632                             (1 << (vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT)));
7633        wr32(reg_offset, reg_val);
7634
7635        adapter->vf_data[vf].spoofchk_enabled = setting;
7636        return E1000_SUCCESS;
7637}
7638
7639static int igb_ndo_get_vf_config(struct net_device *netdev,
7640                                 int vf, struct ifla_vf_info *ivi)
7641{
7642        struct igb_adapter *adapter = netdev_priv(netdev);
7643        if (vf >= adapter->vfs_allocated_count)
7644                return -EINVAL;
7645        ivi->vf = vf;
7646        memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
7647        ivi->tx_rate = adapter->vf_data[vf].tx_rate;
7648        ivi->vlan = adapter->vf_data[vf].pf_vlan;
7649        ivi->qos = adapter->vf_data[vf].pf_qos;
7650        ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
7651        return 0;
7652}
7653
7654static void igb_vmm_control(struct igb_adapter *adapter)
7655{
7656        struct e1000_hw *hw = &adapter->hw;
7657        u32 reg;
7658
7659        switch (hw->mac.type) {
7660        case e1000_82575:
7661        case e1000_i210:
7662        case e1000_i211:
7663        case e1000_i354:
7664        default:
7665                /* replication is not supported for 82575 */
7666                return;
7667        case e1000_82576:
7668                /* notify HW that the MAC is adding vlan tags */
7669                reg = rd32(E1000_DTXCTL);
7670                reg |= E1000_DTXCTL_VLAN_ADDED;
7671                wr32(E1000_DTXCTL, reg);
7672        case e1000_82580:
7673                /* enable replication vlan tag stripping */
7674                reg = rd32(E1000_RPLOLR);
7675                reg |= E1000_RPLOLR_STRVLAN;
7676                wr32(E1000_RPLOLR, reg);
7677        case e1000_i350:
7678                /* none of the above registers are supported by i350 */
7679                break;
7680        }
7681
7682        if (adapter->vfs_allocated_count) {
7683                igb_vmdq_set_loopback_pf(hw, true);
7684                igb_vmdq_set_replication_pf(hw, true);
7685                igb_vmdq_set_anti_spoofing_pf(hw, true,
7686                                              adapter->vfs_allocated_count);
7687        } else {
7688                igb_vmdq_set_loopback_pf(hw, false);
7689                igb_vmdq_set_replication_pf(hw, false);
7690        }
7691}
7692
7693static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
7694{
7695        struct e1000_hw *hw = &adapter->hw;
7696        u32 dmac_thr;
7697        u16 hwm;
7698
7699        if (hw->mac.type > e1000_82580) {
7700                if (adapter->flags & IGB_FLAG_DMAC) {
7701                        u32 reg;
7702
7703                        /* force threshold to 0. */
7704                        wr32(E1000_DMCTXTH, 0);
7705
7706                        /* DMA Coalescing high water mark needs to be greater
7707                         * than the Rx threshold. Set hwm to PBA - max frame
7708                         * size in 16B units, capping it at PBA - 6KB.
7709                         */
7710                        hwm = 64 * pba - adapter->max_frame_size / 16;
7711                        if (hwm < 64 * (pba - 6))
7712                                hwm = 64 * (pba - 6);
7713                        reg = rd32(E1000_FCRTC);
7714                        reg &= ~E1000_FCRTC_RTH_COAL_MASK;
7715                        reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
7716                                & E1000_FCRTC_RTH_COAL_MASK);
7717                        wr32(E1000_FCRTC, reg);
7718
7719                        /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
7720                         * frame size, capping it at PBA - 10KB.
7721                         */
7722                        dmac_thr = pba - adapter->max_frame_size / 512;
7723                        if (dmac_thr < pba - 10)
7724                                dmac_thr = pba - 10;
7725                        reg = rd32(E1000_DMACR);
7726                        reg &= ~E1000_DMACR_DMACTHR_MASK;
7727                        reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
7728                                & E1000_DMACR_DMACTHR_MASK);
7729
7730                        /* transition to L0x or L1 if available..*/
7731                        reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
7732
7733                        /* watchdog timer= +-1000 usec in 32usec intervals */
7734                        reg |= (1000 >> 5);
7735
7736                        /* Disable BMC-to-OS Watchdog Enable */
7737                        if (hw->mac.type != e1000_i354)
7738                                reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
7739
7740                        wr32(E1000_DMACR, reg);
7741
7742                        /* no lower threshold to disable
7743                         * coalescing(smart fifb)-UTRESH=0
7744                         */
7745                        wr32(E1000_DMCRTRH, 0);
7746
7747                        reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
7748
7749                        wr32(E1000_DMCTLX, reg);
7750
7751                        /* free space in tx packet buffer to wake from
7752                         * DMA coal
7753                         */
7754                        wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
7755                             (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
7756
7757                        /* make low power state decision controlled
7758                         * by DMA coal
7759                         */
7760                        reg = rd32(E1000_PCIEMISC);
7761                        reg &= ~E1000_PCIEMISC_LX_DECISION;
7762                        wr32(E1000_PCIEMISC, reg);
7763                } /* endif adapter->dmac is not disabled */
7764        } else if (hw->mac.type == e1000_82580) {
7765                u32 reg = rd32(E1000_PCIEMISC);
7766                wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
7767                wr32(E1000_DMACR, 0);
7768        }
7769}
7770
7771/**
7772 *  igb_read_i2c_byte - Reads 8 bit word over I2C
7773 *  @hw: pointer to hardware structure
7774 *  @byte_offset: byte offset to read
7775 *  @dev_addr: device address
7776 *  @data: value read
7777 *
7778 *  Performs byte read operation over I2C interface at
7779 *  a specified device address.
7780 **/
7781s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7782                      u8 dev_addr, u8 *data)
7783{
7784        struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7785        struct i2c_client *this_client = adapter->i2c_client;
7786        s32 status;
7787        u16 swfw_mask = 0;
7788
7789        if (!this_client)
7790                return E1000_ERR_I2C;
7791
7792        swfw_mask = E1000_SWFW_PHY0_SM;
7793
7794        if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask)
7795            != E1000_SUCCESS)
7796                return E1000_ERR_SWFW_SYNC;
7797
7798        status = i2c_smbus_read_byte_data(this_client, byte_offset);
7799        hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7800
7801        if (status < 0)
7802                return E1000_ERR_I2C;
7803        else {
7804                *data = status;
7805                return E1000_SUCCESS;
7806        }
7807}
7808
7809/**
7810 *  igb_write_i2c_byte - Writes 8 bit word over I2C
7811 *  @hw: pointer to hardware structure
7812 *  @byte_offset: byte offset to write
7813 *  @dev_addr: device address
7814 *  @data: value to write
7815 *
7816 *  Performs byte write operation over I2C interface at
7817 *  a specified device address.
7818 **/
7819s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
7820                       u8 dev_addr, u8 data)
7821{
7822        struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
7823        struct i2c_client *this_client = adapter->i2c_client;
7824        s32 status;
7825        u16 swfw_mask = E1000_SWFW_PHY0_SM;
7826
7827        if (!this_client)
7828                return E1000_ERR_I2C;
7829
7830        if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask) != E1000_SUCCESS)
7831                return E1000_ERR_SWFW_SYNC;
7832        status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
7833        hw->mac.ops.release_swfw_sync(hw, swfw_mask);
7834
7835        if (status)
7836                return E1000_ERR_I2C;
7837        else
7838                return E1000_SUCCESS;
7839
7840}
7841/* igb_main.c */
7842