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

   1 /* Intel(R) Gigabit Ethernet Linux driver
   2  * Copyright(c) 2007-2014 Intel Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, see <http://www.gnu.org/licenses/>.
  15  *
  16  * The full GNU General Public License is included in this distribution in
  17  * the file called "COPYING".
  18  *
  19  * Contact Information:
  20  * e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  21  * Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  22  */
  23
  24 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  25
  26 #include <linux/module.h>
  27 #include <linux/types.h>
  28 #include <linux/init.h>
  29 #include <linux/bitops.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/netdevice.h>
  33 #include <linux/ipv6.h>
  34 #include <linux/slab.h>
  35 #include <net/checksum.h>
  36 #include <net/ip6_checksum.h>
  37 #include <linux/net_tstamp.h>
  38 #include <linux/mii.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/if.h>
  41 #include <linux/if_vlan.h>
  42 #include <linux/pci.h>
  43 #include <linux/pci-aspm.h>
  44 #include <linux/delay.h>
  45 #include <linux/interrupt.h>
  46 #include <linux/ip.h>
  47 #include <linux/tcp.h>
  48 #include <linux/sctp.h>
  49 #include <linux/if_ether.h>
  50 #include <linux/aer.h>
  51 #include <linux/prefetch.h>
  52 #include <linux/pm_runtime.h>
  53 #include <linux/etherdevice.h>
  54 #ifdef CONFIG_IGB_DCA
  55 #include <linux/dca.h>
  56 #endif
  57 #include <linux/i2c.h>
  58 #include "igb.h"
  59
  60 #define MAJ 5
  61 #define MIN 4
  62 #define BUILD 0
  63 #define DRV_VERSION __stringify(MAJ) "." __stringify(MIN) "." \
  64 __stringify(BUILD) "-k"
  65 char igb_driver_name[] = "igb";
  66 char igb_driver_version[] = DRV_VERSION;
  67 static const char igb_driver_string[] =
  68                                 "Intel(R) Gigabit Ethernet Network Driver";
  69 static const char igb_copyright[] =
  70                                 "Copyright (c) 2007-2014 Intel Corporation.";
  71
  72 static const struct e1000_info *igb_info_tbl[] = {
  73         [board_82575] = &e1000_82575_info,
  74 };
  75
  76 static const struct pci_device_id igb_pci_tbl[] = {
  77         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
  78         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
  79         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
  80         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
  81         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
  82         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
  83         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
  84         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
  85         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
  86         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
  87         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
  88         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
  89         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
  90         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
  91         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
  92         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
  93         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
  94         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
  95         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
  96         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
  97         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
  98         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
  99         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
 100         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
 101         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
 102         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
 103         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
 104         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
 105         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
 106         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
 107         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
 108         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
 109         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
 110         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
 111         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
 112         /* required last entry */
 113         {0, }
 114 };
 115
 116 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 117
 118 static int igb_setup_all_tx_resources(struct igb_adapter *);
 119 static int igb_setup_all_rx_resources(struct igb_adapter *);
 120 static void igb_free_all_tx_resources(struct igb_adapter *);
 121 static void igb_free_all_rx_resources(struct igb_adapter *);
 122 static void igb_setup_mrqc(struct igb_adapter *);
 123 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 124 static void igb_remove(struct pci_dev *pdev);
 125 static int igb_sw_init(struct igb_adapter *);
 126 int igb_open(struct net_device *);
 127 int igb_close(struct net_device *);
 128 static void igb_configure(struct igb_adapter *);
 129 static void igb_configure_tx(struct igb_adapter *);
 130 static void igb_configure_rx(struct igb_adapter *);
 131 static void igb_clean_all_tx_rings(struct igb_adapter *);
 132 static void igb_clean_all_rx_rings(struct igb_adapter *);
 133 static void igb_clean_tx_ring(struct igb_ring *);
 134 static void igb_clean_rx_ring(struct igb_ring *);
 135 static void igb_set_rx_mode(struct net_device *);
 136 static void igb_update_phy_info(unsigned long);
 137 static void igb_watchdog(unsigned long);
 138 static void igb_watchdog_task(struct work_struct *);
 139 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 140 static void igb_get_stats64(struct net_device *dev,
 141                             struct rtnl_link_stats64 *stats);
 142 static int igb_change_mtu(struct net_device *, int);
 143 static int igb_set_mac(struct net_device *, void *);
 144 static void igb_set_uta(struct igb_adapter *adapter, bool set);
 145 static irqreturn_t igb_intr(int irq, void *);
 146 static irqreturn_t igb_intr_msi(int irq, void *);
 147 static irqreturn_t igb_msix_other(int irq, void *);
 148 static irqreturn_t igb_msix_ring(int irq, void *);
 149 #ifdef CONFIG_IGB_DCA
 150 static void igb_update_dca(struct igb_q_vector *);
 151 static void igb_setup_dca(struct igb_adapter *);
 152 #endif /* CONFIG_IGB_DCA */
 153 static int igb_poll(struct napi_struct *, int);
 154 static bool igb_clean_tx_irq(struct igb_q_vector *, int);
 155 static int igb_clean_rx_irq(struct igb_q_vector *, int);
 156 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 157 static void igb_tx_timeout(struct net_device *);
 158 static void igb_reset_task(struct work_struct *);
 159 static void igb_vlan_mode(struct net_device *netdev,
 160                           netdev_features_t features);
 161 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
 162 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
 163 static void igb_restore_vlan(struct igb_adapter *);
 164 static void igb_rar_set_index(struct igb_adapter *, u32);
 165 static void igb_ping_all_vfs(struct igb_adapter *);
 166 static void igb_msg_task(struct igb_adapter *);
 167 static void igb_vmm_control(struct igb_adapter *);
 168 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 169 static void igb_flush_mac_table(struct igb_adapter *);
 170 static int igb_available_rars(struct igb_adapter *, u8);
 171 static void igb_set_default_mac_filter(struct igb_adapter *);
 172 static int igb_uc_sync(struct net_device *, const unsigned char *);
 173 static int igb_uc_unsync(struct net_device *, const unsigned char *);
 174 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 175 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 176 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 177                                int vf, u16 vlan, u8 qos, __be16 vlan_proto);
 178 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
 179 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 180                                    bool setting);
 181 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 182                                  struct ifla_vf_info *ivi);
 183 static void igb_check_vf_rate_limit(struct igb_adapter *);
 184 static void igb_nfc_filter_exit(struct igb_adapter *adapter);
 185 static void igb_nfc_filter_restore(struct igb_adapter *adapter);
 186
 187 #ifdef CONFIG_PCI_IOV
 188 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 189 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
 190 static int igb_disable_sriov(struct pci_dev *dev);
 191 static int igb_pci_disable_sriov(struct pci_dev *dev);
 192 #endif
 193
 194 static int igb_suspend(struct device *);
 195 static int igb_resume(struct device *);
 196 static int igb_runtime_suspend(struct device *dev);
 197 static int igb_runtime_resume(struct device *dev);
 198 static int igb_runtime_idle(struct device *dev);
 199 static const struct dev_pm_ops igb_pm_ops = {
 200         SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
 201         SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
 202                         igb_runtime_idle)
 203 };
 204 static void igb_shutdown(struct pci_dev *);
 205 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 206 #ifdef CONFIG_IGB_DCA
 207 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 208 static struct notifier_block dca_notifier = {
 209         .notifier_call  = igb_notify_dca,
 210         .next           = NULL,
 211         .priority       = 0
 212 };
 213 #endif
 214 #ifdef CONFIG_NET_POLL_CONTROLLER
 215 /* for netdump / net console */
 216 static void igb_netpoll(struct net_device *);
 217 #endif
 218 #ifdef CONFIG_PCI_IOV
 219 static unsigned int max_vfs;
 220 module_param(max_vfs, uint, 0);
 221 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
 222 #endif /* CONFIG_PCI_IOV */
 223
 224 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 225                      pci_channel_state_t);
 226 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 227 static void igb_io_resume(struct pci_dev *);
 228
 229 static 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
 235 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 236
 237 static 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
 250 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 251 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 252 MODULE_LICENSE("GPL");
 253 MODULE_VERSION(DRV_VERSION);
 254
 255 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 256 static int debug = -1;
 257 module_param(debug, int, 0);
 258 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 259
 260 struct igb_reg_info {
 261         u32 ofs;
 262         char *name;
 263 };
 264
 265 static 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 */
 302 static 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 */
 368 static 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\n");
 388                 pr_info("%-15s %016lX %016lX\n", netdev->name,
 389                         netdev->state, dev_trans_start(netdev));
 390         }
 391
 392         /* Print Registers */
 393         dev_info(&adapter->pdev->dev, "Register Dump\n");
 394         pr_info(" Register Name   Value\n");
 395         for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
 396              reginfo->name; reginfo++) {
 397                 igb_regdump(hw, reginfo);
 398         }
 399
 400         /* Print TX Ring Summary */
 401         if (!netdev || !netif_running(netdev))
 402                 goto exit;
 403
 404         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 405         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 406         for (n = 0; n < adapter->num_tx_queues; n++) {
 407                 struct igb_tx_buffer *buffer_info;
 408                 tx_ring = adapter->tx_ring[n];
 409                 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 410                 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
 411                         n, tx_ring->next_to_use, tx_ring->next_to_clean,
 412                         (u64)dma_unmap_addr(buffer_info, dma),
 413                         dma_unmap_len(buffer_info, len),
 414                         buffer_info->next_to_watch,
 415                         (u64)buffer_info->time_stamp);
 416         }
 417
 418         /* Print TX Rings */
 419         if (!netif_msg_tx_done(adapter))
 420                 goto rx_ring_summary;
 421
 422         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 423
 424         /* Transmit Descriptor Formats
 425          *
 426          * Advanced Transmit Descriptor
 427          *   +--------------------------------------------------------------+
 428          * 0 |         Buffer Address [63:0]                                |
 429          *   +--------------------------------------------------------------+
 430          * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
 431          *   +--------------------------------------------------------------+
 432          *   63      46 45    40 39 38 36 35 32 31   24             15       0
 433          */
 434
 435         for (n = 0; n < adapter->num_tx_queues; n++) {
 436                 tx_ring = adapter->tx_ring[n];
 437                 pr_info("------------------------------------\n");
 438                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 439                 pr_info("------------------------------------\n");
 440                 pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] [bi->dma       ] leng  ntw timestamp        bi->skb\n");
 441
 442                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 443                         const char *next_desc;
 444                         struct igb_tx_buffer *buffer_info;
 445                         tx_desc = IGB_TX_DESC(tx_ring, i);
 446                         buffer_info = &tx_ring->tx_buffer_info[i];
 447                         u0 = (struct my_u0 *)tx_desc;
 448                         if (i == tx_ring->next_to_use &&
 449                             i == tx_ring->next_to_clean)
 450                                 next_desc = " NTC/U";
 451                         else if (i == tx_ring->next_to_use)
 452                                 next_desc = " NTU";
 453                         else if (i == tx_ring->next_to_clean)
 454                                 next_desc = " NTC";
 455                         else
 456                                 next_desc = "";
 457
 458                         pr_info("T [0x%03X]    %016llX %016llX %016llX %04X  %p %016llX %p%s\n",
 459                                 i, le64_to_cpu(u0->a),
 460                                 le64_to_cpu(u0->b),
 461                                 (u64)dma_unmap_addr(buffer_info, dma),
 462                                 dma_unmap_len(buffer_info, len),
 463                                 buffer_info->next_to_watch,
 464                                 (u64)buffer_info->time_stamp,
 465                                 buffer_info->skb, next_desc);
 466
 467                         if (netif_msg_pktdata(adapter) && buffer_info->skb)
 468                                 print_hex_dump(KERN_INFO, "",
 469                                         DUMP_PREFIX_ADDRESS,
 470                                         16, 1, buffer_info->skb->data,
 471                                         dma_unmap_len(buffer_info, len),
 472                                         true);
 473                 }
 474         }
 475
 476         /* Print RX Rings Summary */
 477 rx_ring_summary:
 478         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 479         pr_info("Queue [NTU] [NTC]\n");
 480         for (n = 0; n < adapter->num_rx_queues; n++) {
 481                 rx_ring = adapter->rx_ring[n];
 482                 pr_info(" %5d %5X %5X\n",
 483                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 484         }
 485
 486         /* Print RX Rings */
 487         if (!netif_msg_rx_status(adapter))
 488                 goto exit;
 489
 490         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 491
 492         /* Advanced Receive Descriptor (Read) Format
 493          *    63                                           1        0
 494          *    +-----------------------------------------------------+
 495          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 496          *    +----------------------------------------------+------+
 497          *  8 |       Header Buffer Address [63:1]           |  DD  |
 498          *    +-----------------------------------------------------+
 499          *
 500          *
 501          * Advanced Receive Descriptor (Write-Back) Format
 502          *
 503          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 504          *   +------------------------------------------------------+
 505          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 506          *   | Checksum   Ident  |   |           |    | Type | Type |
 507          *   +------------------------------------------------------+
 508          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 509          *   +------------------------------------------------------+
 510          *   63       48 47    32 31            20 19               0
 511          */
 512
 513         for (n = 0; n < adapter->num_rx_queues; n++) {
 514                 rx_ring = adapter->rx_ring[n];
 515                 pr_info("------------------------------------\n");
 516                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 517                 pr_info("------------------------------------\n");
 518                 pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] [bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
 519                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
 520
 521                 for (i = 0; i < rx_ring->count; i++) {
 522                         const char *next_desc;
 523                         struct igb_rx_buffer *buffer_info;
 524                         buffer_info = &rx_ring->rx_buffer_info[i];
 525                         rx_desc = IGB_RX_DESC(rx_ring, i);
 526                         u0 = (struct my_u0 *)rx_desc;
 527                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 528
 529                         if (i == rx_ring->next_to_use)
 530                                 next_desc = " NTU";
 531                         else if (i == rx_ring->next_to_clean)
 532                                 next_desc = " NTC";
 533                         else
 534                                 next_desc = "";
 535
 536                         if (staterr & E1000_RXD_STAT_DD) {
 537                                 /* Descriptor Done */
 538                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
 539                                         "RWB", i,
 540                                         le64_to_cpu(u0->a),
 541                                         le64_to_cpu(u0->b),
 542                                         next_desc);
 543                         } else {
 544                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
 545                                         "R  ", i,
 546                                         le64_to_cpu(u0->a),
 547                                         le64_to_cpu(u0->b),
 548                                         (u64)buffer_info->dma,
 549                                         next_desc);
 550
 551                                 if (netif_msg_pktdata(adapter) &&
 552                                     buffer_info->dma && buffer_info->page) {
 553                                         print_hex_dump(KERN_INFO, "",
 554                                           DUMP_PREFIX_ADDRESS,
 555                                           16, 1,
 556                                           page_address(buffer_info->page) +
 557                                                       buffer_info->page_offset,
 558                                           igb_rx_bufsz(rx_ring), true);
 559                                 }
 560                         }
 561                 }
 562         }
 563
 564 exit:
 565         return;
 566 }
 567
 568 /**
 569  *  igb_get_i2c_data - Reads the I2C SDA data bit
 570  *  @hw: pointer to hardware structure
 571  *  @i2cctl: Current value of I2CCTL register
 572  *
 573  *  Returns the I2C data bit value
 574  **/
 575 static int igb_get_i2c_data(void *data)
 576 {
 577         struct igb_adapter *adapter = (struct igb_adapter *)data;
 578         struct e1000_hw *hw = &adapter->hw;
 579         s32 i2cctl = rd32(E1000_I2CPARAMS);
 580
 581         return !!(i2cctl & E1000_I2C_DATA_IN);
 582 }
 583
 584 /**
 585  *  igb_set_i2c_data - Sets the I2C data bit
 586  *  @data: pointer to hardware structure
 587  *  @state: I2C data value (0 or 1) to set
 588  *
 589  *  Sets the I2C data bit
 590  **/
 591 static void igb_set_i2c_data(void *data, int state)
 592 {
 593         struct igb_adapter *adapter = (struct igb_adapter *)data;
 594         struct e1000_hw *hw = &adapter->hw;
 595         s32 i2cctl = rd32(E1000_I2CPARAMS);
 596
 597         if (state)
 598                 i2cctl |= E1000_I2C_DATA_OUT;
 599         else
 600                 i2cctl &= ~E1000_I2C_DATA_OUT;
 601
 602         i2cctl &= ~E1000_I2C_DATA_OE_N;
 603         i2cctl |= E1000_I2C_CLK_OE_N;
 604         wr32(E1000_I2CPARAMS, i2cctl);
 605         wrfl();
 606
 607 }
 608
 609 /**
 610  *  igb_set_i2c_clk - Sets the I2C SCL clock
 611  *  @data: pointer to hardware structure
 612  *  @state: state to set clock
 613  *
 614  *  Sets the I2C clock line to state
 615  **/
 616 static void igb_set_i2c_clk(void *data, int state)
 617 {
 618         struct igb_adapter *adapter = (struct igb_adapter *)data;
 619         struct e1000_hw *hw = &adapter->hw;
 620         s32 i2cctl = rd32(E1000_I2CPARAMS);
 621
 622         if (state) {
 623                 i2cctl |= E1000_I2C_CLK_OUT;
 624                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 625         } else {
 626                 i2cctl &= ~E1000_I2C_CLK_OUT;
 627                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 628         }
 629         wr32(E1000_I2CPARAMS, i2cctl);
 630         wrfl();
 631 }
 632
 633 /**
 634  *  igb_get_i2c_clk - Gets the I2C SCL clock state
 635  *  @data: pointer to hardware structure
 636  *
 637  *  Gets the I2C clock state
 638  **/
 639 static int igb_get_i2c_clk(void *data)
 640 {
 641         struct igb_adapter *adapter = (struct igb_adapter *)data;
 642         struct e1000_hw *hw = &adapter->hw;
 643         s32 i2cctl = rd32(E1000_I2CPARAMS);
 644
 645         return !!(i2cctl & E1000_I2C_CLK_IN);
 646 }
 647
 648 static const struct i2c_algo_bit_data igb_i2c_algo = {
 649         .setsda         = igb_set_i2c_data,
 650         .setscl         = igb_set_i2c_clk,
 651         .getsda         = igb_get_i2c_data,
 652         .getscl         = igb_get_i2c_clk,
 653         .udelay         = 5,
 654         .timeout        = 20,
 655 };
 656
 657 /**
 658  *  igb_get_hw_dev - return device
 659  *  @hw: pointer to hardware structure
 660  *
 661  *  used by hardware layer to print debugging information
 662  **/
 663 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
 664 {
 665         struct igb_adapter *adapter = hw->back;
 666         return adapter->netdev;
 667 }
 668
 669 /**
 670  *  igb_init_module - Driver Registration Routine
 671  *
 672  *  igb_init_module is the first routine called when the driver is
 673  *  loaded. All it does is register with the PCI subsystem.
 674  **/
 675 static int __init igb_init_module(void)
 676 {
 677         int ret;
 678
 679         pr_info("%s - version %s\n",
 680                igb_driver_string, igb_driver_version);
 681         pr_info("%s\n", igb_copyright);
 682
 683 #ifdef CONFIG_IGB_DCA
 684         dca_register_notify(&dca_notifier);
 685 #endif
 686         ret = pci_register_driver(&igb_driver);
 687         return ret;
 688 }
 689
 690 module_init(igb_init_module);
 691
 692 /**
 693  *  igb_exit_module - Driver Exit Cleanup Routine
 694  *
 695  *  igb_exit_module is called just before the driver is removed
 696  *  from memory.
 697  **/
 698 static void __exit igb_exit_module(void)
 699 {
 700 #ifdef CONFIG_IGB_DCA
 701         dca_unregister_notify(&dca_notifier);
 702 #endif
 703         pci_unregister_driver(&igb_driver);
 704 }
 705
 706 module_exit(igb_exit_module);
 707
 708 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 709 /**
 710  *  igb_cache_ring_register - Descriptor ring to register mapping
 711  *  @adapter: board private structure to initialize
 712  *
 713  *  Once we know the feature-set enabled for the device, we'll cache
 714  *  the register offset the descriptor ring is assigned to.
 715  **/
 716 static void igb_cache_ring_register(struct igb_adapter *adapter)
 717 {
 718         int i = 0, j = 0;
 719         u32 rbase_offset = adapter->vfs_allocated_count;
 720
 721         switch (adapter->hw.mac.type) {
 722         case e1000_82576:
 723                 /* The queues are allocated for virtualization such that VF 0
 724                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 725                  * In order to avoid collision we start at the first free queue
 726                  * and continue consuming queues in the same sequence
 727                  */
 728                 if (adapter->vfs_allocated_count) {
 729                         for (; i < adapter->rss_queues; i++)
 730                                 adapter->rx_ring[i]->reg_idx = rbase_offset +
 731                                                                Q_IDX_82576(i);
 732                 }
 733                 /* Fall through */
 734         case e1000_82575:
 735         case e1000_82580:
 736         case e1000_i350:
 737         case e1000_i354:
 738         case e1000_i210:
 739         case e1000_i211:
 740                 /* Fall through */
 741         default:
 742                 for (; i < adapter->num_rx_queues; i++)
 743                         adapter->rx_ring[i]->reg_idx = rbase_offset + i;
 744                 for (; j < adapter->num_tx_queues; j++)
 745                         adapter->tx_ring[j]->reg_idx = rbase_offset + j;
 746                 break;
 747         }
 748 }
 749
 750 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
 751 {
 752         struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
 753         u8 __iomem *hw_addr = ACCESS_ONCE(hw->hw_addr);
 754         u32 value = 0;
 755
 756         if (E1000_REMOVED(hw_addr))
 757                 return ~value;
 758
 759         value = readl(&hw_addr[reg]);
 760
 761         /* reads should not return all F's */
 762         if (!(~value) && (!reg || !(~readl(hw_addr)))) {
 763                 struct net_device *netdev = igb->netdev;
 764                 hw->hw_addr = NULL;
 765                 netif_device_detach(netdev);
 766                 netdev_err(netdev, "PCIe link lost, device now detached\n");
 767         }
 768
 769         return value;
 770 }
 771
 772 /**
 773  *  igb_write_ivar - configure ivar for given MSI-X vector
 774  *  @hw: pointer to the HW structure
 775  *  @msix_vector: vector number we are allocating to a given ring
 776  *  @index: row index of IVAR register to write within IVAR table
 777  *  @offset: column offset of in IVAR, should be multiple of 8
 778  *
 779  *  This function is intended to handle the writing of the IVAR register
 780  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 781  *  each containing an cause allocation for an Rx and Tx ring, and a
 782  *  variable number of rows depending on the number of queues supported.
 783  **/
 784 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 785                            int index, int offset)
 786 {
 787         u32 ivar = array_rd32(E1000_IVAR0, index);
 788
 789         /* clear any bits that are currently set */
 790         ivar &= ~((u32)0xFF << offset);
 791
 792         /* write vector and valid bit */
 793         ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 794
 795         array_wr32(E1000_IVAR0, index, ivar);
 796 }
 797
 798 #define IGB_N0_QUEUE -1
 799 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 800 {
 801         struct igb_adapter *adapter = q_vector->adapter;
 802         struct e1000_hw *hw = &adapter->hw;
 803         int rx_queue = IGB_N0_QUEUE;
 804         int tx_queue = IGB_N0_QUEUE;
 805         u32 msixbm = 0;
 806
 807         if (q_vector->rx.ring)
 808                 rx_queue = q_vector->rx.ring->reg_idx;
 809         if (q_vector->tx.ring)
 810                 tx_queue = q_vector->tx.ring->reg_idx;
 811
 812         switch (hw->mac.type) {
 813         case e1000_82575:
 814                 /* The 82575 assigns vectors using a bitmask, which matches the
 815                  * bitmask for the EICR/EIMS/EIMC registers.  To assign one
 816                  * or more queues to a vector, we write the appropriate bits
 817                  * into the MSIXBM register for that vector.
 818                  */
 819                 if (rx_queue > IGB_N0_QUEUE)
 820                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 821                 if (tx_queue > IGB_N0_QUEUE)
 822                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 823                 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
 824                         msixbm |= E1000_EIMS_OTHER;
 825                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 826                 q_vector->eims_value = msixbm;
 827                 break;
 828         case e1000_82576:
 829                 /* 82576 uses a table that essentially consists of 2 columns
 830                  * with 8 rows.  The ordering is column-major so we use the
 831                  * lower 3 bits as the row index, and the 4th bit as the
 832                  * column offset.
 833                  */
 834                 if (rx_queue > IGB_N0_QUEUE)
 835                         igb_write_ivar(hw, msix_vector,
 836                                        rx_queue & 0x7,
 837                                        (rx_queue & 0x8) << 1);
 838                 if (tx_queue > IGB_N0_QUEUE)
 839                         igb_write_ivar(hw, msix_vector,
 840                                        tx_queue & 0x7,
 841                                        ((tx_queue & 0x8) << 1) + 8);
 842                 q_vector->eims_value = BIT(msix_vector);
 843                 break;
 844         case e1000_82580:
 845         case e1000_i350:
 846         case e1000_i354:
 847         case e1000_i210:
 848         case e1000_i211:
 849                 /* On 82580 and newer adapters the scheme is similar to 82576
 850                  * however instead of ordering column-major we have things
 851                  * ordered row-major.  So we traverse the table by using
 852                  * bit 0 as the column offset, and the remaining bits as the
 853                  * row index.
 854                  */
 855                 if (rx_queue > IGB_N0_QUEUE)
 856                         igb_write_ivar(hw, msix_vector,
 857                                        rx_queue >> 1,
 858                                        (rx_queue & 0x1) << 4);
 859                 if (tx_queue > IGB_N0_QUEUE)
 860                         igb_write_ivar(hw, msix_vector,
 861                                        tx_queue >> 1,
 862                                        ((tx_queue & 0x1) << 4) + 8);
 863                 q_vector->eims_value = BIT(msix_vector);
 864                 break;
 865         default:
 866                 BUG();
 867                 break;
 868         }
 869
 870         /* add q_vector eims value to global eims_enable_mask */
 871         adapter->eims_enable_mask |= q_vector->eims_value;
 872
 873         /* configure q_vector to set itr on first interrupt */
 874         q_vector->set_itr = 1;
 875 }
 876
 877 /**
 878  *  igb_configure_msix - Configure MSI-X hardware
 879  *  @adapter: board private structure to initialize
 880  *
 881  *  igb_configure_msix sets up the hardware to properly
 882  *  generate MSI-X interrupts.
 883  **/
 884 static void igb_configure_msix(struct igb_adapter *adapter)
 885 {
 886         u32 tmp;
 887         int i, vector = 0;
 888         struct e1000_hw *hw = &adapter->hw;
 889
 890         adapter->eims_enable_mask = 0;
 891
 892         /* set vector for other causes, i.e. link changes */
 893         switch (hw->mac.type) {
 894         case e1000_82575:
 895                 tmp = rd32(E1000_CTRL_EXT);
 896                 /* enable MSI-X PBA support*/
 897                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 898
 899                 /* Auto-Mask interrupts upon ICR read. */
 900                 tmp |= E1000_CTRL_EXT_EIAME;
 901                 tmp |= E1000_CTRL_EXT_IRCA;
 902
 903                 wr32(E1000_CTRL_EXT, tmp);
 904
 905                 /* enable msix_other interrupt */
 906                 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
 907                 adapter->eims_other = E1000_EIMS_OTHER;
 908
 909                 break;
 910
 911         case e1000_82576:
 912         case e1000_82580:
 913         case e1000_i350:
 914         case e1000_i354:
 915         case e1000_i210:
 916         case e1000_i211:
 917                 /* Turn on MSI-X capability first, or our settings
 918                  * won't stick.  And it will take days to debug.
 919                  */
 920                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 921                      E1000_GPIE_PBA | E1000_GPIE_EIAME |
 922                      E1000_GPIE_NSICR);
 923
 924                 /* enable msix_other interrupt */
 925                 adapter->eims_other = BIT(vector);
 926                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 927
 928                 wr32(E1000_IVAR_MISC, tmp);
 929                 break;
 930         default:
 931                 /* do nothing, since nothing else supports MSI-X */
 932                 break;
 933         } /* switch (hw->mac.type) */
 934
 935         adapter->eims_enable_mask |= adapter->eims_other;
 936
 937         for (i = 0; i < adapter->num_q_vectors; i++)
 938                 igb_assign_vector(adapter->q_vector[i], vector++);
 939
 940         wrfl();
 941 }
 942
 943 /**
 944  *  igb_request_msix - Initialize MSI-X interrupts
 945  *  @adapter: board private structure to initialize
 946  *
 947  *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
 948  *  kernel.
 949  **/
 950 static int igb_request_msix(struct igb_adapter *adapter)
 951 {
 952         struct net_device *netdev = adapter->netdev;
 953         int i, err = 0, vector = 0, free_vector = 0;
 954
 955         err = request_irq(adapter->msix_entries[vector].vector,
 956                           igb_msix_other, 0, netdev->name, adapter);
 957         if (err)
 958                 goto err_out;
 959
 960         for (i = 0; i < adapter->num_q_vectors; i++) {
 961                 struct igb_q_vector *q_vector = adapter->q_vector[i];
 962
 963                 vector++;
 964
 965                 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
 966
 967                 if (q_vector->rx.ring && q_vector->tx.ring)
 968                         sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 969                                 q_vector->rx.ring->queue_index);
 970                 else if (q_vector->tx.ring)
 971                         sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 972                                 q_vector->tx.ring->queue_index);
 973                 else if (q_vector->rx.ring)
 974                         sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 975                                 q_vector->rx.ring->queue_index);
 976                 else
 977                         sprintf(q_vector->name, "%s-unused", netdev->name);
 978
 979                 err = request_irq(adapter->msix_entries[vector].vector,
 980                                   igb_msix_ring, 0, q_vector->name,
 981                                   q_vector);
 982                 if (err)
 983                         goto err_free;
 984         }
 985
 986         igb_configure_msix(adapter);
 987         return 0;
 988
 989 err_free:
 990         /* free already assigned IRQs */
 991         free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 992
 993         vector--;
 994         for (i = 0; i < vector; i++) {
 995                 free_irq(adapter->msix_entries[free_vector++].vector,
 996                          adapter->q_vector[i]);
 997         }
 998 err_out:
 999         return err;
1000 }
1001
1002 /**
1003  *  igb_free_q_vector - Free memory allocated for specific interrupt vector
1004  *  @adapter: board private structure to initialize
1005  *  @v_idx: Index of vector to be freed
1006  *
1007  *  This function frees the memory allocated to the q_vector.
1008  **/
1009 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
1010 {
1011         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1012
1013         adapter->q_vector[v_idx] = NULL;
1014
1015         /* igb_get_stats64() might access the rings on this vector,
1016          * we must wait a grace period before freeing it.
1017          */
1018         if (q_vector)
1019                 kfree_rcu(q_vector, rcu);
1020 }
1021
1022 /**
1023  *  igb_reset_q_vector - Reset config for interrupt vector
1024  *  @adapter: board private structure to initialize
1025  *  @v_idx: Index of vector to be reset
1026  *
1027  *  If NAPI is enabled it will delete any references to the
1028  *  NAPI struct. This is preparation for igb_free_q_vector.
1029  **/
1030 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1031 {
1032         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1033
1034         /* Coming from igb_set_interrupt_capability, the vectors are not yet
1035          * allocated. So, q_vector is NULL so we should stop here.
1036          */
1037         if (!q_vector)
1038                 return;
1039
1040         if (q_vector->tx.ring)
1041                 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1042
1043         if (q_vector->rx.ring)
1044                 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1045
1046         netif_napi_del(&q_vector->napi);
1047
1048 }
1049
1050 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1051 {
1052         int v_idx = adapter->num_q_vectors;
1053
1054         if (adapter->flags & IGB_FLAG_HAS_MSIX)
1055                 pci_disable_msix(adapter->pdev);
1056         else if (adapter->flags & IGB_FLAG_HAS_MSI)
1057                 pci_disable_msi(adapter->pdev);
1058
1059         while (v_idx--)
1060                 igb_reset_q_vector(adapter, v_idx);
1061 }
1062
1063 /**
1064  *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1065  *  @adapter: board private structure to initialize
1066  *
1067  *  This function frees the memory allocated to the q_vectors.  In addition if
1068  *  NAPI is enabled it will delete any references to the NAPI struct prior
1069  *  to freeing the q_vector.
1070  **/
1071 static void igb_free_q_vectors(struct igb_adapter *adapter)
1072 {
1073         int v_idx = adapter->num_q_vectors;
1074
1075         adapter->num_tx_queues = 0;
1076         adapter->num_rx_queues = 0;
1077         adapter->num_q_vectors = 0;
1078
1079         while (v_idx--) {
1080                 igb_reset_q_vector(adapter, v_idx);
1081                 igb_free_q_vector(adapter, v_idx);
1082         }
1083 }
1084
1085 /**
1086  *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1087  *  @adapter: board private structure to initialize
1088  *
1089  *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1090  *  MSI-X interrupts allocated.
1091  */
1092 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1093 {
1094         igb_free_q_vectors(adapter);
1095         igb_reset_interrupt_capability(adapter);
1096 }
1097
1098 /**
1099  *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1100  *  @adapter: board private structure to initialize
1101  *  @msix: boolean value of MSIX capability
1102  *
1103  *  Attempt to configure interrupts using the best available
1104  *  capabilities of the hardware and kernel.
1105  **/
1106 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1107 {
1108         int err;
1109         int numvecs, i;
1110
1111         if (!msix)
1112                 goto msi_only;
1113         adapter->flags |= IGB_FLAG_HAS_MSIX;
1114
1115         /* Number of supported queues. */
1116         adapter->num_rx_queues = adapter->rss_queues;
1117         if (adapter->vfs_allocated_count)
1118                 adapter->num_tx_queues = 1;
1119         else
1120                 adapter->num_tx_queues = adapter->rss_queues;
1121
1122         /* start with one vector for every Rx queue */
1123         numvecs = adapter->num_rx_queues;
1124
1125         /* if Tx handler is separate add 1 for every Tx queue */
1126         if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1127                 numvecs += adapter->num_tx_queues;
1128
1129         /* store the number of vectors reserved for queues */
1130         adapter->num_q_vectors = numvecs;
1131
1132         /* add 1 vector for link status interrupts */
1133         numvecs++;
1134         for (i = 0; i < numvecs; i++)
1135                 adapter->msix_entries[i].entry = i;
1136
1137         err = pci_enable_msix_range(adapter->pdev,
1138                                     adapter->msix_entries,
1139                                     numvecs,
1140                                     numvecs);
1141         if (err > 0)
1142                 return;
1143
1144         igb_reset_interrupt_capability(adapter);
1145
1146         /* If we can't do MSI-X, try MSI */
1147 msi_only:
1148         adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1149 #ifdef CONFIG_PCI_IOV
1150         /* disable SR-IOV for non MSI-X configurations */
1151         if (adapter->vf_data) {
1152                 struct e1000_hw *hw = &adapter->hw;
1153                 /* disable iov and allow time for transactions to clear */
1154                 pci_disable_sriov(adapter->pdev);
1155                 msleep(500);
1156
1157                 kfree(adapter->vf_mac_list);
1158                 adapter->vf_mac_list = NULL;
1159                 kfree(adapter->vf_data);
1160                 adapter->vf_data = NULL;
1161                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1162                 wrfl();
1163                 msleep(100);
1164                 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1165         }
1166 #endif
1167         adapter->vfs_allocated_count = 0;
1168         adapter->rss_queues = 1;
1169         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1170         adapter->num_rx_queues = 1;
1171         adapter->num_tx_queues = 1;
1172         adapter->num_q_vectors = 1;
1173         if (!pci_enable_msi(adapter->pdev))
1174                 adapter->flags |= IGB_FLAG_HAS_MSI;
1175 }
1176
1177 static void igb_add_ring(struct igb_ring *ring,
1178                          struct igb_ring_container *head)
1179 {
1180         head->ring = ring;
1181         head->count++;
1182 }
1183
1184 /**
1185  *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1186  *  @adapter: board private structure to initialize
1187  *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1188  *  @v_idx: index of vector in adapter struct
1189  *  @txr_count: total number of Tx rings to allocate
1190  *  @txr_idx: index of first Tx ring to allocate
1191  *  @rxr_count: total number of Rx rings to allocate
1192  *  @rxr_idx: index of first Rx ring to allocate
1193  *
1194  *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1195  **/
1196 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1197                               int v_count, int v_idx,
1198                               int txr_count, int txr_idx,
1199                               int rxr_count, int rxr_idx)
1200 {
1201         struct igb_q_vector *q_vector;
1202         struct igb_ring *ring;
1203         int ring_count, size;
1204
1205         /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1206         if (txr_count > 1 || rxr_count > 1)
1207                 return -ENOMEM;
1208
1209         ring_count = txr_count + rxr_count;
1210         size = sizeof(struct igb_q_vector) +
1211                (sizeof(struct igb_ring) * ring_count);
1212
1213         /* allocate q_vector and rings */
1214         q_vector = adapter->q_vector[v_idx];
1215         if (!q_vector) {
1216                 q_vector = kzalloc(size, GFP_KERNEL);
1217         } else if (size > ksize(q_vector)) {
1218                 kfree_rcu(q_vector, rcu);
1219                 q_vector = kzalloc(size, GFP_KERNEL);
1220         } else {
1221                 memset(q_vector, 0, size);
1222         }
1223         if (!q_vector)
1224                 return -ENOMEM;
1225
1226         /* initialize NAPI */
1227         netif_napi_add(adapter->netdev, &q_vector->napi,
1228                        igb_poll, 64);
1229
1230         /* tie q_vector and adapter together */
1231         adapter->q_vector[v_idx] = q_vector;
1232         q_vector->adapter = adapter;
1233
1234         /* initialize work limits */
1235         q_vector->tx.work_limit = adapter->tx_work_limit;
1236
1237         /* initialize ITR configuration */
1238         q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1239         q_vector->itr_val = IGB_START_ITR;
1240
1241         /* initialize pointer to rings */
1242         ring = q_vector->ring;
1243
1244         /* intialize ITR */
1245         if (rxr_count) {
1246                 /* rx or rx/tx vector */
1247                 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1248                         q_vector->itr_val = adapter->rx_itr_setting;
1249         } else {
1250                 /* tx only vector */
1251                 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1252                         q_vector->itr_val = adapter->tx_itr_setting;
1253         }
1254
1255         if (txr_count) {
1256                 /* assign generic ring traits */
1257                 ring->dev = &adapter->pdev->dev;
1258                 ring->netdev = adapter->netdev;
1259
1260                 /* configure backlink on ring */
1261                 ring->q_vector = q_vector;
1262
1263                 /* update q_vector Tx values */
1264                 igb_add_ring(ring, &q_vector->tx);
1265
1266                 /* For 82575, context index must be unique per ring. */
1267                 if (adapter->hw.mac.type == e1000_82575)
1268                         set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1269
1270                 /* apply Tx specific ring traits */
1271                 ring->count = adapter->tx_ring_count;
1272                 ring->queue_index = txr_idx;
1273
1274                 u64_stats_init(&ring->tx_syncp);
1275                 u64_stats_init(&ring->tx_syncp2);
1276
1277                 /* assign ring to adapter */
1278                 adapter->tx_ring[txr_idx] = ring;
1279
1280                 /* push pointer to next ring */
1281                 ring++;
1282         }
1283
1284         if (rxr_count) {
1285                 /* assign generic ring traits */
1286                 ring->dev = &adapter->pdev->dev;
1287                 ring->netdev = adapter->netdev;
1288
1289                 /* configure backlink on ring */
1290                 ring->q_vector = q_vector;
1291
1292                 /* update q_vector Rx values */
1293                 igb_add_ring(ring, &q_vector->rx);
1294
1295                 /* set flag indicating ring supports SCTP checksum offload */
1296                 if (adapter->hw.mac.type >= e1000_82576)
1297                         set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1298
1299                 /* On i350, i354, i210, and i211, loopback VLAN packets
1300                  * have the tag byte-swapped.
1301                  */
1302                 if (adapter->hw.mac.type >= e1000_i350)
1303                         set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1304
1305                 /* apply Rx specific ring traits */
1306                 ring->count = adapter->rx_ring_count;
1307                 ring->queue_index = rxr_idx;
1308
1309                 u64_stats_init(&ring->rx_syncp);
1310
1311                 /* assign ring to adapter */
1312                 adapter->rx_ring[rxr_idx] = ring;
1313         }
1314
1315         return 0;
1316 }
1317
1318
1319 /**
1320  *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1321  *  @adapter: board private structure to initialize
1322  *
1323  *  We allocate one q_vector per queue interrupt.  If allocation fails we
1324  *  return -ENOMEM.
1325  **/
1326 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1327 {
1328         int q_vectors = adapter->num_q_vectors;
1329         int rxr_remaining = adapter->num_rx_queues;
1330         int txr_remaining = adapter->num_tx_queues;
1331         int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1332         int err;
1333
1334         if (q_vectors >= (rxr_remaining + txr_remaining)) {
1335                 for (; rxr_remaining; v_idx++) {
1336                         err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1337                                                  0, 0, 1, rxr_idx);
1338
1339                         if (err)
1340                                 goto err_out;
1341
1342                         /* update counts and index */
1343                         rxr_remaining--;
1344                         rxr_idx++;
1345                 }
1346         }
1347
1348         for (; v_idx < q_vectors; v_idx++) {
1349                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1350                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1351
1352                 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1353                                          tqpv, txr_idx, rqpv, rxr_idx);
1354
1355                 if (err)
1356                         goto err_out;
1357
1358                 /* update counts and index */
1359                 rxr_remaining -= rqpv;
1360                 txr_remaining -= tqpv;
1361                 rxr_idx++;
1362                 txr_idx++;
1363         }
1364
1365         return 0;
1366
1367 err_out:
1368         adapter->num_tx_queues = 0;
1369         adapter->num_rx_queues = 0;
1370         adapter->num_q_vectors = 0;
1371
1372         while (v_idx--)
1373                 igb_free_q_vector(adapter, v_idx);
1374
1375         return -ENOMEM;
1376 }
1377
1378 /**
1379  *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1380  *  @adapter: board private structure to initialize
1381  *  @msix: boolean value of MSIX capability
1382  *
1383  *  This function initializes the interrupts and allocates all of the queues.
1384  **/
1385 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1386 {
1387         struct pci_dev *pdev = adapter->pdev;
1388         int err;
1389
1390         igb_set_interrupt_capability(adapter, msix);
1391
1392         err = igb_alloc_q_vectors(adapter);
1393         if (err) {
1394                 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1395                 goto err_alloc_q_vectors;
1396         }
1397
1398         igb_cache_ring_register(adapter);
1399
1400         return 0;
1401
1402 err_alloc_q_vectors:
1403         igb_reset_interrupt_capability(adapter);
1404         return err;
1405 }
1406
1407 /**
1408  *  igb_request_irq - initialize interrupts
1409  *  @adapter: board private structure to initialize
1410  *
1411  *  Attempts to configure interrupts using the best available
1412  *  capabilities of the hardware and kernel.
1413  **/
1414 static int igb_request_irq(struct igb_adapter *adapter)
1415 {
1416         struct net_device *netdev = adapter->netdev;
1417         struct pci_dev *pdev = adapter->pdev;
1418         int err = 0;
1419
1420         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1421                 err = igb_request_msix(adapter);
1422                 if (!err)
1423                         goto request_done;
1424                 /* fall back to MSI */
1425                 igb_free_all_tx_resources(adapter);
1426                 igb_free_all_rx_resources(adapter);
1427
1428                 igb_clear_interrupt_scheme(adapter);
1429                 err = igb_init_interrupt_scheme(adapter, false);
1430                 if (err)
1431                         goto request_done;
1432
1433                 igb_setup_all_tx_resources(adapter);
1434                 igb_setup_all_rx_resources(adapter);
1435                 igb_configure(adapter);
1436         }
1437
1438         igb_assign_vector(adapter->q_vector[0], 0);
1439
1440         if (adapter->flags & IGB_FLAG_HAS_MSI) {
1441                 err = request_irq(pdev->irq, igb_intr_msi, 0,
1442                                   netdev->name, adapter);
1443                 if (!err)
1444                         goto request_done;
1445
1446                 /* fall back to legacy interrupts */
1447                 igb_reset_interrupt_capability(adapter);
1448                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1449         }
1450
1451         err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1452                           netdev->name, adapter);
1453
1454         if (err)
1455                 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1456                         err);
1457
1458 request_done:
1459         return err;
1460 }
1461
1462 static void igb_free_irq(struct igb_adapter *adapter)
1463 {
1464         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1465                 int vector = 0, i;
1466
1467                 free_irq(adapter->msix_entries[vector++].vector, adapter);
1468
1469                 for (i = 0; i < adapter->num_q_vectors; i++)
1470                         free_irq(adapter->msix_entries[vector++].vector,
1471                                  adapter->q_vector[i]);
1472         } else {
1473                 free_irq(adapter->pdev->irq, adapter);
1474         }
1475 }
1476
1477 /**
1478  *  igb_irq_disable - Mask off interrupt generation on the NIC
1479  *  @adapter: board private structure
1480  **/
1481 static void igb_irq_disable(struct igb_adapter *adapter)
1482 {
1483         struct e1000_hw *hw = &adapter->hw;
1484
1485         /* we need to be careful when disabling interrupts.  The VFs are also
1486          * mapped into these registers and so clearing the bits can cause
1487          * issues on the VF drivers so we only need to clear what we set
1488          */
1489         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1490                 u32 regval = rd32(E1000_EIAM);
1491
1492                 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1493                 wr32(E1000_EIMC, adapter->eims_enable_mask);
1494                 regval = rd32(E1000_EIAC);
1495                 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1496         }
1497
1498         wr32(E1000_IAM, 0);
1499         wr32(E1000_IMC, ~0);
1500         wrfl();
1501         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1502                 int i;
1503
1504                 for (i = 0; i < adapter->num_q_vectors; i++)
1505                         synchronize_irq(adapter->msix_entries[i].vector);
1506         } else {
1507                 synchronize_irq(adapter->pdev->irq);
1508         }
1509 }
1510
1511 /**
1512  *  igb_irq_enable - Enable default interrupt generation settings
1513  *  @adapter: board private structure
1514  **/
1515 static void igb_irq_enable(struct igb_adapter *adapter)
1516 {
1517         struct e1000_hw *hw = &adapter->hw;
1518
1519         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1520                 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1521                 u32 regval = rd32(E1000_EIAC);
1522
1523                 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1524                 regval = rd32(E1000_EIAM);
1525                 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1526                 wr32(E1000_EIMS, adapter->eims_enable_mask);
1527                 if (adapter->vfs_allocated_count) {
1528                         wr32(E1000_MBVFIMR, 0xFF);
1529                         ims |= E1000_IMS_VMMB;
1530                 }
1531                 wr32(E1000_IMS, ims);
1532         } else {
1533                 wr32(E1000_IMS, IMS_ENABLE_MASK |
1534                                 E1000_IMS_DRSTA);
1535                 wr32(E1000_IAM, IMS_ENABLE_MASK |
1536                                 E1000_IMS_DRSTA);
1537         }
1538 }
1539
1540 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1541 {
1542         struct e1000_hw *hw = &adapter->hw;
1543         u16 pf_id = adapter->vfs_allocated_count;
1544         u16 vid = adapter->hw.mng_cookie.vlan_id;
1545         u16 old_vid = adapter->mng_vlan_id;
1546
1547         if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1548                 /* add VID to filter table */
1549                 igb_vfta_set(hw, vid, pf_id, true, true);
1550                 adapter->mng_vlan_id = vid;
1551         } else {
1552                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1553         }
1554
1555         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1556             (vid != old_vid) &&
1557             !test_bit(old_vid, adapter->active_vlans)) {
1558                 /* remove VID from filter table */
1559                 igb_vfta_set(hw, vid, pf_id, false, true);
1560         }
1561 }
1562
1563 /**
1564  *  igb_release_hw_control - release control of the h/w to f/w
1565  *  @adapter: address of board private structure
1566  *
1567  *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1568  *  For ASF and Pass Through versions of f/w this means that the
1569  *  driver is no longer loaded.
1570  **/
1571 static void igb_release_hw_control(struct igb_adapter *adapter)
1572 {
1573         struct e1000_hw *hw = &adapter->hw;
1574         u32 ctrl_ext;
1575
1576         /* Let firmware take over control of h/w */
1577         ctrl_ext = rd32(E1000_CTRL_EXT);
1578         wr32(E1000_CTRL_EXT,
1579                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1580 }
1581
1582 /**
1583  *  igb_get_hw_control - get control of the h/w from f/w
1584  *  @adapter: address of board private structure
1585  *
1586  *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1587  *  For ASF and Pass Through versions of f/w this means that
1588  *  the driver is loaded.
1589  **/
1590 static void igb_get_hw_control(struct igb_adapter *adapter)
1591 {
1592         struct e1000_hw *hw = &adapter->hw;
1593         u32 ctrl_ext;
1594
1595         /* Let firmware know the driver has taken over */
1596         ctrl_ext = rd32(E1000_CTRL_EXT);
1597         wr32(E1000_CTRL_EXT,
1598                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1599 }
1600
1601 /**
1602  *  igb_configure - configure the hardware for RX and TX
1603  *  @adapter: private board structure
1604  **/
1605 static void igb_configure(struct igb_adapter *adapter)
1606 {
1607         struct net_device *netdev = adapter->netdev;
1608         int i;
1609
1610         igb_get_hw_control(adapter);
1611         igb_set_rx_mode(netdev);
1612
1613         igb_restore_vlan(adapter);
1614
1615         igb_setup_tctl(adapter);
1616         igb_setup_mrqc(adapter);
1617         igb_setup_rctl(adapter);
1618
1619         igb_nfc_filter_restore(adapter);
1620         igb_configure_tx(adapter);
1621         igb_configure_rx(adapter);
1622
1623         igb_rx_fifo_flush_82575(&adapter->hw);
1624
1625         /* call igb_desc_unused which always leaves
1626          * at least 1 descriptor unused to make sure
1627          * next_to_use != next_to_clean
1628          */
1629         for (i = 0; i < adapter->num_rx_queues; i++) {
1630                 struct igb_ring *ring = adapter->rx_ring[i];
1631                 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
1632         }
1633 }
1634
1635 /**
1636  *  igb_power_up_link - Power up the phy/serdes link
1637  *  @adapter: address of board private structure
1638  **/
1639 void igb_power_up_link(struct igb_adapter *adapter)
1640 {
1641         igb_reset_phy(&adapter->hw);
1642
1643         if (adapter->hw.phy.media_type == e1000_media_type_copper)
1644                 igb_power_up_phy_copper(&adapter->hw);
1645         else
1646                 igb_power_up_serdes_link_82575(&adapter->hw);
1647
1648         igb_setup_link(&adapter->hw);
1649 }
1650
1651 /**
1652  *  igb_power_down_link - Power down the phy/serdes link
1653  *  @adapter: address of board private structure
1654  */
1655 static void igb_power_down_link(struct igb_adapter *adapter)
1656 {
1657         if (adapter->hw.phy.media_type == e1000_media_type_copper)
1658                 igb_power_down_phy_copper_82575(&adapter->hw);
1659         else
1660                 igb_shutdown_serdes_link_82575(&adapter->hw);
1661 }
1662
1663 /**
1664  * Detect and switch function for Media Auto Sense
1665  * @adapter: address of the board private structure
1666  **/
1667 static void igb_check_swap_media(struct igb_adapter *adapter)
1668 {
1669         struct e1000_hw *hw = &adapter->hw;
1670         u32 ctrl_ext, connsw;
1671         bool swap_now = false;
1672
1673         ctrl_ext = rd32(E1000_CTRL_EXT);
1674         connsw = rd32(E1000_CONNSW);
1675
1676         /* need to live swap if current media is copper and we have fiber/serdes
1677          * to go to.
1678          */
1679
1680         if ((hw->phy.media_type == e1000_media_type_copper) &&
1681             (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
1682                 swap_now = true;
1683         } else if (!(connsw & E1000_CONNSW_SERDESD)) {
1684                 /* copper signal takes time to appear */
1685                 if (adapter->copper_tries < 4) {
1686                         adapter->copper_tries++;
1687                         connsw |= E1000_CONNSW_AUTOSENSE_CONF;
1688                         wr32(E1000_CONNSW, connsw);
1689                         return;
1690                 } else {
1691                         adapter->copper_tries = 0;
1692                         if ((connsw & E1000_CONNSW_PHYSD) &&
1693                             (!(connsw & E1000_CONNSW_PHY_PDN))) {
1694                                 swap_now = true;
1695                                 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
1696                                 wr32(E1000_CONNSW, connsw);
1697                         }
1698                 }
1699         }
1700
1701         if (!swap_now)
1702                 return;
1703
1704         switch (hw->phy.media_type) {
1705         case e1000_media_type_copper:
1706                 netdev_info(adapter->netdev,
1707                         "MAS: changing media to fiber/serdes\n");
1708                 ctrl_ext |=
1709                         E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1710                 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1711                 adapter->copper_tries = 0;
1712                 break;
1713         case e1000_media_type_internal_serdes:
1714         case e1000_media_type_fiber:
1715                 netdev_info(adapter->netdev,
1716                         "MAS: changing media to copper\n");
1717                 ctrl_ext &=
1718                         ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
1719                 adapter->flags |= IGB_FLAG_MEDIA_RESET;
1720                 break;
1721         default:
1722                 /* shouldn't get here during regular operation */
1723                 netdev_err(adapter->netdev,
1724                         "AMS: Invalid media type found, returning\n");
1725                 break;
1726         }
1727         wr32(E1000_CTRL_EXT, ctrl_ext);
1728 }
1729
1730 /**
1731  *  igb_up - Open the interface and prepare it to handle traffic
1732  *  @adapter: board private structure
1733  **/
1734 int igb_up(struct igb_adapter *adapter)
1735 {
1736         struct e1000_hw *hw = &adapter->hw;
1737         int i;
1738
1739         /* hardware has been reset, we need to reload some things */
1740         igb_configure(adapter);
1741
1742         clear_bit(__IGB_DOWN, &adapter->state);
1743
1744         for (i = 0; i < adapter->num_q_vectors; i++)
1745                 napi_enable(&(adapter->q_vector[i]->napi));
1746
1747         if (adapter->flags & IGB_FLAG_HAS_MSIX)
1748                 igb_configure_msix(adapter);
1749         else
1750                 igb_assign_vector(adapter->q_vector[0], 0);
1751
1752         /* Clear any pending interrupts. */
1753         rd32(E1000_ICR);
1754         igb_irq_enable(adapter);
1755
1756         /* notify VFs that reset has been completed */
1757         if (adapter->vfs_allocated_count) {
1758                 u32 reg_data = rd32(E1000_CTRL_EXT);
1759
1760                 reg_data |= E1000_CTRL_EXT_PFRSTD;
1761                 wr32(E1000_CTRL_EXT, reg_data);
1762         }
1763
1764         netif_tx_start_all_queues(adapter->netdev);
1765
1766         /* start the watchdog. */
1767         hw->mac.get_link_status = 1;
1768         schedule_work(&adapter->watchdog_task);
1769
1770         if ((adapter->flags & IGB_FLAG_EEE) &&
1771             (!hw->dev_spec._82575.eee_disable))
1772                 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
1773
1774         return 0;
1775 }
1776
1777 void igb_down(struct igb_adapter *adapter)
1778 {
1779         struct net_device *netdev = adapter->netdev;
1780         struct e1000_hw *hw = &adapter->hw;
1781         u32 tctl, rctl;
1782         int i;
1783
1784         /* signal that we're down so the interrupt handler does not
1785          * reschedule our watchdog timer
1786          */
1787         set_bit(__IGB_DOWN, &adapter->state);
1788
1789         /* disable receives in the hardware */
1790         rctl = rd32(E1000_RCTL);
1791         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1792         /* flush and sleep below */
1793
1794         igb_nfc_filter_exit(adapter);
1795
1796         netif_carrier_off(netdev);
1797         netif_tx_stop_all_queues(netdev);
1798
1799         /* disable transmits in the hardware */
1800         tctl = rd32(E1000_TCTL);
1801         tctl &= ~E1000_TCTL_EN;
1802         wr32(E1000_TCTL, tctl);
1803         /* flush both disables and wait for them to finish */
1804         wrfl();
1805         usleep_range(10000, 11000);
1806
1807         igb_irq_disable(adapter);
1808
1809         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
1810
1811         for (i = 0; i < adapter->num_q_vectors; i++) {
1812                 if (adapter->q_vector[i]) {
1813                         napi_synchronize(&adapter->q_vector[i]->napi);
1814                         napi_disable(&adapter->q_vector[i]->napi);
1815                 }
1816         }
1817
1818         del_timer_sync(&adapter->watchdog_timer);
1819         del_timer_sync(&adapter->phy_info_timer);
1820
1821         /* record the stats before reset*/
1822         spin_lock(&adapter->stats64_lock);
1823         igb_update_stats(adapter);
1824         spin_unlock(&adapter->stats64_lock);
1825
1826         adapter->link_speed = 0;
1827         adapter->link_duplex = 0;
1828
1829         if (!pci_channel_offline(adapter->pdev))
1830                 igb_reset(adapter);
1831
1832         /* clear VLAN promisc flag so VFTA will be updated if necessary */
1833         adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
1834
1835         igb_clean_all_tx_rings(adapter);
1836         igb_clean_all_rx_rings(adapter);
1837 #ifdef CONFIG_IGB_DCA
1838
1839         /* since we reset the hardware DCA settings were cleared */
1840         igb_setup_dca(adapter);
1841 #endif
1842 }
1843
1844 void igb_reinit_locked(struct igb_adapter *adapter)
1845 {
1846         WARN_ON(in_interrupt());
1847         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
1848                 usleep_range(1000, 2000);
1849         igb_down(adapter);
1850         igb_up(adapter);
1851         clear_bit(__IGB_RESETTING, &adapter->state);
1852 }
1853
1854 /** igb_enable_mas - Media Autosense re-enable after swap
1855  *
1856  * @adapter: adapter struct
1857  **/
1858 static void igb_enable_mas(struct igb_adapter *adapter)
1859 {
1860         struct e1000_hw *hw = &adapter->hw;
1861         u32 connsw = rd32(E1000_CONNSW);
1862
1863         /* configure for SerDes media detect */
1864         if ((hw->phy.media_type == e1000_media_type_copper) &&
1865             (!(connsw & E1000_CONNSW_SERDESD))) {
1866                 connsw |= E1000_CONNSW_ENRGSRC;
1867                 connsw |= E1000_CONNSW_AUTOSENSE_EN;
1868                 wr32(E1000_CONNSW, connsw);
1869                 wrfl();
1870         }
1871 }
1872
1873 void igb_reset(struct igb_adapter *adapter)
1874 {
1875         struct pci_dev *pdev = adapter->pdev;
1876         struct e1000_hw *hw = &adapter->hw;
1877         struct e1000_mac_info *mac = &hw->mac;
1878         struct e1000_fc_info *fc = &hw->fc;
1879         u32 pba, hwm;
1880
1881         /* Repartition Pba for greater than 9k mtu
1882          * To take effect CTRL.RST is required.
1883          */
1884         switch (mac->type) {
1885         case e1000_i350:
1886         case e1000_i354:
1887         case e1000_82580:
1888                 pba = rd32(E1000_RXPBS);
1889                 pba = igb_rxpbs_adjust_82580(pba);
1890                 break;
1891         case e1000_82576:
1892                 pba = rd32(E1000_RXPBS);
1893                 pba &= E1000_RXPBS_SIZE_MASK_82576;
1894                 break;
1895         case e1000_82575:
1896         case e1000_i210:
1897         case e1000_i211:
1898         default:
1899                 pba = E1000_PBA_34K;
1900                 break;
1901         }
1902
1903         if (mac->type == e1000_82575) {
1904                 u32 min_rx_space, min_tx_space, needed_tx_space;
1905
1906                 /* write Rx PBA so that hardware can report correct Tx PBA */
1907                 wr32(E1000_PBA, pba);
1908
1909                 /* To maintain wire speed transmits, the Tx FIFO should be
1910                  * large enough to accommodate two full transmit packets,
1911                  * rounded up to the next 1KB and expressed in KB.  Likewise,
1912                  * the Rx FIFO should be large enough to accommodate at least
1913                  * one full receive packet and is similarly rounded up and
1914                  * expressed in KB.
1915                  */
1916                 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
1917
1918                 /* The Tx FIFO also stores 16 bytes of information about the Tx
1919                  * but don't include Ethernet FCS because hardware appends it.
1920                  * We only need to round down to the nearest 512 byte block
1921                  * count since the value we care about is 2 frames, not 1.
1922                  */
1923                 min_tx_space = adapter->max_frame_size;
1924                 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
1925                 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
1926
1927                 /* upper 16 bits has Tx packet buffer allocation size in KB */
1928                 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
1929
1930                 /* If current Tx allocation is less than the min Tx FIFO size,
1931                  * and the min Tx FIFO size is less than the current Rx FIFO
1932                  * allocation, take space away from current Rx allocation.
1933                  */
1934                 if (needed_tx_space < pba) {
1935                         pba -= needed_tx_space;
1936
1937                         /* if short on Rx space, Rx wins and must trump Tx
1938                          * adjustment
1939                          */
1940                         if (pba < min_rx_space)
1941                                 pba = min_rx_space;
1942                 }
1943
1944                 /* adjust PBA for jumbo frames */
1945                 wr32(E1000_PBA, pba);
1946         }
1947
1948         /* flow control settings
1949          * The high water mark must be low enough to fit one full frame
1950          * after transmitting the pause frame.  As such we must have enough
1951          * space to allow for us to complete our current transmit and then
1952          * receive the frame that is in progress from the link partner.
1953          * Set it to:
1954          * - the full Rx FIFO size minus one full Tx plus one full Rx frame
1955          */
1956         hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
1957
1958         fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
1959         fc->low_water = fc->high_water - 16;
1960         fc->pause_time = 0xFFFF;
1961         fc->send_xon = 1;
1962         fc->current_mode = fc->requested_mode;
1963
1964         /* disable receive for all VFs and wait one second */
1965         if (adapter->vfs_allocated_count) {
1966                 int i;
1967
1968                 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1969                         adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
1970
1971                 /* ping all the active vfs to let them know we are going down */
1972                 igb_ping_all_vfs(adapter);
1973
1974                 /* disable transmits and receives */
1975                 wr32(E1000_VFRE, 0);
1976                 wr32(E1000_VFTE, 0);
1977         }
1978
1979         /* Allow time for pending master requests to run */
1980         hw->mac.ops.reset_hw(hw);
1981         wr32(E1000_WUC, 0);
1982
1983         if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
1984                 /* need to resetup here after media swap */
1985                 adapter->ei.get_invariants(hw);
1986                 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
1987         }
1988         if ((mac->type == e1000_82575) &&
1989             (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
1990                 igb_enable_mas(adapter);
1991         }
1992         if (hw->mac.ops.init_hw(hw))
1993                 dev_err(&pdev->dev, "Hardware Error\n");
1994
1995         /* RAR registers were cleared during init_hw, clear mac table */
1996         igb_flush_mac_table(adapter);
1997         __dev_uc_unsync(adapter->netdev, NULL);
1998
1999         /* Recover default RAR entry */
2000         igb_set_default_mac_filter(adapter);
2001
2002         /* Flow control settings reset on hardware reset, so guarantee flow
2003          * control is off when forcing speed.
2004          */
2005         if (!hw->mac.autoneg)
2006                 igb_force_mac_fc(hw);
2007
2008         igb_init_dmac(adapter, pba);
2009 #ifdef CONFIG_IGB_HWMON
2010         /* Re-initialize the thermal sensor on i350 devices. */
2011         if (!test_bit(__IGB_DOWN, &adapter->state)) {
2012                 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2013                         /* If present, re-initialize the external thermal sensor
2014                          * interface.
2015                          */
2016                         if (adapter->ets)
2017                                 mac->ops.init_thermal_sensor_thresh(hw);
2018                 }
2019         }
2020 #endif
2021         /* Re-establish EEE setting */
2022         if (hw->phy.media_type == e1000_media_type_copper) {
2023                 switch (mac->type) {
2024                 case e1000_i350:
2025                 case e1000_i210:
2026                 case e1000_i211:
2027                         igb_set_eee_i350(hw, true, true);
2028                         break;
2029                 case e1000_i354:
2030                         igb_set_eee_i354(hw, true, true);
2031                         break;
2032                 default:
2033                         break;
2034                 }
2035         }
2036         if (!netif_running(adapter->netdev))
2037                 igb_power_down_link(adapter);
2038
2039         igb_update_mng_vlan(adapter);
2040
2041         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2042         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2043
2044         /* Re-enable PTP, where applicable. */
2045         if (adapter->ptp_flags & IGB_PTP_ENABLED)
2046                 igb_ptp_reset(adapter);
2047
2048         igb_get_phy_info(hw);
2049 }
2050
2051 static netdev_features_t igb_fix_features(struct net_device *netdev,
2052         netdev_features_t features)
2053 {
2054         /* Since there is no support for separate Rx/Tx vlan accel
2055          * enable/disable make sure Tx flag is always in same state as Rx.
2056          */
2057         if (features & NETIF_F_HW_VLAN_CTAG_RX)
2058                 features |= NETIF_F_HW_VLAN_CTAG_TX;
2059         else
2060                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2061
2062         return features;
2063 }
2064
2065 static int igb_set_features(struct net_device *netdev,
2066         netdev_features_t features)
2067 {
2068         netdev_features_t changed = netdev->features ^ features;
2069         struct igb_adapter *adapter = netdev_priv(netdev);
2070
2071         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2072                 igb_vlan_mode(netdev, features);
2073
2074         if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2075                 return 0;
2076
2077         if (!(features & NETIF_F_NTUPLE)) {
2078                 struct hlist_node *node2;
2079                 struct igb_nfc_filter *rule;
2080
2081                 spin_lock(&adapter->nfc_lock);
2082                 hlist_for_each_entry_safe(rule, node2,
2083                                           &adapter->nfc_filter_list, nfc_node) {
2084                         igb_erase_filter(adapter, rule);
2085                         hlist_del(&rule->nfc_node);
2086                         kfree(rule);
2087                 }
2088                 spin_unlock(&adapter->nfc_lock);
2089                 adapter->nfc_filter_count = 0;
2090         }
2091
2092         netdev->features = features;
2093
2094         if (netif_running(netdev))
2095                 igb_reinit_locked(adapter);
2096         else
2097                 igb_reset(adapter);
2098
2099         return 0;
2100 }
2101
2102 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2103                            struct net_device *dev,
2104                            const unsigned char *addr, u16 vid,
2105                            u16 flags)
2106 {
2107         /* guarantee we can provide a unique filter for the unicast address */
2108         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2109                 struct igb_adapter *adapter = netdev_priv(dev);
2110                 int vfn = adapter->vfs_allocated_count;
2111
2112                 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2113                         return -ENOMEM;
2114         }
2115
2116         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2117 }
2118
2119 #define IGB_MAX_MAC_HDR_LEN     127
2120 #define IGB_MAX_NETWORK_HDR_LEN 511
2121
2122 static netdev_features_t
2123 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2124                    netdev_features_t features)
2125 {
2126         unsigned int network_hdr_len, mac_hdr_len;
2127
2128         /* Make certain the headers can be described by a context descriptor */
2129         mac_hdr_len = skb_network_header(skb) - skb->data;
2130         if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2131                 return features & ~(NETIF_F_HW_CSUM |
2132                                     NETIF_F_SCTP_CRC |
2133                                     NETIF_F_HW_VLAN_CTAG_TX |
2134                                     NETIF_F_TSO |
2135                                     NETIF_F_TSO6);
2136
2137         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2138         if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
2139                 return features & ~(NETIF_F_HW_CSUM |
2140                                     NETIF_F_SCTP_CRC |
2141                                     NETIF_F_TSO |
2142                                     NETIF_F_TSO6);
2143
2144         /* We can only support IPV4 TSO in tunnels if we can mangle the
2145          * inner IP ID field, so strip TSO if MANGLEID is not supported.
2146          */
2147         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2148                 features &= ~NETIF_F_TSO;
2149
2150         return features;
2151 }
2152
2153 static const struct net_device_ops igb_netdev_ops = {
2154         .ndo_open               = igb_open,
2155         .ndo_stop               = igb_close,
2156         .ndo_start_xmit         = igb_xmit_frame,
2157         .ndo_get_stats64        = igb_get_stats64,
2158         .ndo_set_rx_mode        = igb_set_rx_mode,
2159         .ndo_set_mac_address    = igb_set_mac,
2160         .ndo_change_mtu         = igb_change_mtu,
2161         .ndo_do_ioctl           = igb_ioctl,
2162         .ndo_tx_timeout         = igb_tx_timeout,
2163         .ndo_validate_addr      = eth_validate_addr,
2164         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
2165         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
2166         .ndo_set_vf_mac         = igb_ndo_set_vf_mac,
2167         .ndo_set_vf_vlan        = igb_ndo_set_vf_vlan,
2168         .ndo_set_vf_rate        = igb_ndo_set_vf_bw,
2169         .ndo_set_vf_spoofchk    = igb_ndo_set_vf_spoofchk,
2170         .ndo_get_vf_config      = igb_ndo_get_vf_config,
2171 #ifdef CONFIG_NET_POLL_CONTROLLER
2172         .ndo_poll_controller    = igb_netpoll,
2173 #endif
2174         .ndo_fix_features       = igb_fix_features,
2175         .ndo_set_features       = igb_set_features,
2176         .ndo_fdb_add            = igb_ndo_fdb_add,
2177         .ndo_features_check     = igb_features_check,
2178 };
2179
2180 /**
2181  * igb_set_fw_version - Configure version string for ethtool
2182  * @adapter: adapter struct
2183  **/
2184 void igb_set_fw_version(struct igb_adapter *adapter)
2185 {
2186         struct e1000_hw *hw = &adapter->hw;
2187         struct e1000_fw_version fw;
2188
2189         igb_get_fw_version(hw, &fw);
2190
2191         switch (hw->mac.type) {
2192         case e1000_i210:
2193         case e1000_i211:
2194                 if (!(igb_get_flash_presence_i210(hw))) {
2195                         snprintf(adapter->fw_version,
2196                                  sizeof(adapter->fw_version),
2197                                  "%2d.%2d-%d",
2198                                  fw.invm_major, fw.invm_minor,
2199                                  fw.invm_img_type);
2200                         break;
2201                 }
2202                 /* fall through */
2203         default:
2204                 /* if option is rom valid, display its version too */
2205                 if (fw.or_valid) {
2206                         snprintf(adapter->fw_version,
2207                                  sizeof(adapter->fw_version),
2208                                  "%d.%d, 0x%08x, %d.%d.%d",
2209                                  fw.eep_major, fw.eep_minor, fw.etrack_id,
2210                                  fw.or_major, fw.or_build, fw.or_patch);
2211                 /* no option rom */
2212                 } else if (fw.etrack_id != 0X0000) {
2213                         snprintf(adapter->fw_version,
2214                             sizeof(adapter->fw_version),
2215                             "%d.%d, 0x%08x",
2216                             fw.eep_major, fw.eep_minor, fw.etrack_id);
2217                 } else {
2218                 snprintf(adapter->fw_version,
2219                     sizeof(adapter->fw_version),
2220                     "%d.%d.%d",
2221                     fw.eep_major, fw.eep_minor, fw.eep_build);
2222                 }
2223                 break;
2224         }
2225 }
2226
2227 /**
2228  * igb_init_mas - init Media Autosense feature if enabled in the NVM
2229  *
2230  * @adapter: adapter struct
2231  **/
2232 static void igb_init_mas(struct igb_adapter *adapter)
2233 {
2234         struct e1000_hw *hw = &adapter->hw;
2235         u16 eeprom_data;
2236
2237         hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
2238         switch (hw->bus.func) {
2239         case E1000_FUNC_0:
2240                 if (eeprom_data & IGB_MAS_ENABLE_0) {
2241                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2242                         netdev_info(adapter->netdev,
2243                                 "MAS: Enabling Media Autosense for port %d\n",
2244                                 hw->bus.func);
2245                 }
2246                 break;
2247         case E1000_FUNC_1:
2248                 if (eeprom_data & IGB_MAS_ENABLE_1) {
2249                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2250                         netdev_info(adapter->netdev,
2251                                 "MAS: Enabling Media Autosense for port %d\n",
2252                                 hw->bus.func);
2253                 }
2254                 break;
2255         case E1000_FUNC_2:
2256                 if (eeprom_data & IGB_MAS_ENABLE_2) {
2257                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2258                         netdev_info(adapter->netdev,
2259                                 "MAS: Enabling Media Autosense for port %d\n",
2260                                 hw->bus.func);
2261                 }
2262                 break;
2263         case E1000_FUNC_3:
2264                 if (eeprom_data & IGB_MAS_ENABLE_3) {
2265                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
2266                         netdev_info(adapter->netdev,
2267                                 "MAS: Enabling Media Autosense for port %d\n",
2268                                 hw->bus.func);
2269                 }
2270                 break;
2271         default:
2272                 /* Shouldn't get here */
2273                 netdev_err(adapter->netdev,
2274                         "MAS: Invalid port configuration, returning\n");
2275                 break;
2276         }
2277 }
2278
2279 /**
2280  *  igb_init_i2c - Init I2C interface
2281  *  @adapter: pointer to adapter structure
2282  **/
2283 static s32 igb_init_i2c(struct igb_adapter *adapter)
2284 {
2285         s32 status = 0;
2286
2287         /* I2C interface supported on i350 devices */
2288         if (adapter->hw.mac.type != e1000_i350)
2289                 return 0;
2290
2291         /* Initialize the i2c bus which is controlled by the registers.
2292          * This bus will use the i2c_algo_bit structue that implements
2293          * the protocol through toggling of the 4 bits in the register.
2294          */
2295         adapter->i2c_adap.owner = THIS_MODULE;
2296         adapter->i2c_algo = igb_i2c_algo;
2297         adapter->i2c_algo.data = adapter;
2298         adapter->i2c_adap.algo_data = &adapter->i2c_algo;
2299         adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
2300         strlcpy(adapter->i2c_adap.name, "igb BB",
2301                 sizeof(adapter->i2c_adap.name));
2302         status = i2c_bit_add_bus(&adapter->i2c_adap);
2303         return status;
2304 }
2305
2306 /**
2307  *  igb_probe - Device Initialization Routine
2308  *  @pdev: PCI device information struct
2309  *  @ent: entry in igb_pci_tbl
2310  *
2311  *  Returns 0 on success, negative on failure
2312  *
2313  *  igb_probe initializes an adapter identified by a pci_dev structure.
2314  *  The OS initialization, configuring of the adapter private structure,
2315  *  and a hardware reset occur.
2316  **/
2317 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
2318 {
2319         struct net_device *netdev;
2320         struct igb_adapter *adapter;
2321         struct e1000_hw *hw;
2322         u16 eeprom_data = 0;
2323         s32 ret_val;
2324         static int global_quad_port_a; /* global quad port a indication */
2325         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
2326         int err, pci_using_dac;
2327         u8 part_str[E1000_PBANUM_LENGTH];
2328
2329         /* Catch broken hardware that put the wrong VF device ID in
2330          * the PCIe SR-IOV capability.
2331          */
2332         if (pdev->is_virtfn) {
2333                 WARN(1, KERN_ERR "%s (%hx:%hx) should not be a VF!\n",
2334                         pci_name(pdev), pdev->vendor, pdev->device);
2335                 return -EINVAL;
2336         }
2337
2338         err = pci_enable_device_mem(pdev);
2339         if (err)
2340                 return err;
2341
2342         pci_using_dac = 0;
2343         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2344         if (!err) {
2345                 pci_using_dac = 1;
2346         } else {
2347                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
2348                 if (err) {
2349                         dev_err(&pdev->dev,
2350                                 "No usable DMA configuration, aborting\n");
2351                         goto err_dma;
2352                 }
2353         }
2354
2355         err = pci_request_mem_regions(pdev, igb_driver_name);
2356         if (err)
2357                 goto err_pci_reg;
2358
2359         pci_enable_pcie_error_reporting(pdev);
2360
2361         pci_set_master(pdev);
2362         pci_save_state(pdev);
2363
2364         err = -ENOMEM;
2365         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
2366                                    IGB_MAX_TX_QUEUES);
2367         if (!netdev)
2368                 goto err_alloc_etherdev;
2369
2370         SET_NETDEV_DEV(netdev, &pdev->dev);
2371
2372         pci_set_drvdata(pdev, netdev);
2373         adapter = netdev_priv(netdev);
2374         adapter->netdev = netdev;
2375         adapter->pdev = pdev;
2376         hw = &adapter->hw;
2377         hw->back = adapter;
2378         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
2379
2380         err = -EIO;
2381         adapter->io_addr = pci_iomap(pdev, 0, 0);
2382         if (!adapter->io_addr)
2383                 goto err_ioremap;
2384         /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
2385         hw->hw_addr = adapter->io_addr;
2386
2387         netdev->netdev_ops = &igb_netdev_ops;
2388         igb_set_ethtool_ops(netdev);
2389         netdev->watchdog_timeo = 5 * HZ;
2390
2391         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
2392
2393         netdev->mem_start = pci_resource_start(pdev, 0);
2394         netdev->mem_end = pci_resource_end(pdev, 0);
2395
2396         /* PCI config space info */
2397         hw->vendor_id = pdev->vendor;
2398         hw->device_id = pdev->device;
2399         hw->revision_id = pdev->revision;
2400         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2401         hw->subsystem_device_id = pdev->subsystem_device;
2402
2403         /* Copy the default MAC, PHY and NVM function pointers */
2404         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
2405         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
2406         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
2407         /* Initialize skew-specific constants */
2408         err = ei->get_invariants(hw);
2409         if (err)
2410                 goto err_sw_init;
2411
2412         /* setup the private structure */
2413         err = igb_sw_init(adapter);
2414         if (err)
2415                 goto err_sw_init;
2416
2417         igb_get_bus_info_pcie(hw);
2418
2419         hw->phy.autoneg_wait_to_complete = false;
2420
2421         /* Copper options */
2422         if (hw->phy.media_type == e1000_media_type_copper) {
2423                 hw->phy.mdix = AUTO_ALL_MODES;
2424                 hw->phy.disable_polarity_correction = false;
2425                 hw->phy.ms_type = e1000_ms_hw_default;
2426         }
2427
2428         if (igb_check_reset_block(hw))
2429                 dev_info(&pdev->dev,
2430                         "PHY reset is blocked due to SOL/IDER session.\n");
2431
2432         /* features is initialized to 0 in allocation, it might have bits
2433          * set by igb_sw_init so we should use an or instead of an
2434          * assignment.
2435          */
2436         netdev->features |= NETIF_F_SG |
2437                             NETIF_F_TSO |
2438                             NETIF_F_TSO6 |
2439                             NETIF_F_RXHASH |
2440                             NETIF_F_RXCSUM |
2441                             NETIF_F_HW_CSUM;
2442
2443         if (hw->mac.type >= e1000_82576)
2444                 netdev->features |= NETIF_F_SCTP_CRC;
2445
2446 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
2447                                   NETIF_F_GSO_GRE_CSUM | \
2448                                   NETIF_F_GSO_IPXIP4 | \
2449                                   NETIF_F_GSO_IPXIP6 | \
2450                                   NETIF_F_GSO_UDP_TUNNEL | \
2451                                   NETIF_F_GSO_UDP_TUNNEL_CSUM)
2452
2453         netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
2454         netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
2455
2456         /* copy netdev features into list of user selectable features */
2457         netdev->hw_features |= netdev->features |
2458                                NETIF_F_HW_VLAN_CTAG_RX |
2459                                NETIF_F_HW_VLAN_CTAG_TX |
2460                                NETIF_F_RXALL;
2461
2462         if (hw->mac.type >= e1000_i350)
2463                 netdev->hw_features |= NETIF_F_NTUPLE;
2464
2465         if (pci_using_dac)
2466                 netdev->features |= NETIF_F_HIGHDMA;
2467
2468         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
2469         netdev->mpls_features |= NETIF_F_HW_CSUM;
2470         netdev->hw_enc_features |= netdev->vlan_features;
2471
2472         /* set this bit last since it cannot be part of vlan_features */
2473         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
2474                             NETIF_F_HW_VLAN_CTAG_RX |
2475                             NETIF_F_HW_VLAN_CTAG_TX;
2476
2477         netdev->priv_flags |= IFF_SUPP_NOFCS;
2478
2479         netdev->priv_flags |= IFF_UNICAST_FLT;
2480
2481         /* MTU range: 68 - 9216 */
2482         netdev->min_mtu = ETH_MIN_MTU;
2483         netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
2484
2485         adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
2486
2487         /* before reading the NVM, reset the controller to put the device in a
2488          * known good starting state
2489          */
2490         hw->mac.ops.reset_hw(hw);
2491
2492         /* make sure the NVM is good , i211/i210 parts can have special NVM
2493          * that doesn't contain a checksum
2494          */
2495         switch (hw->mac.type) {
2496         case e1000_i210:
2497         case e1000_i211:
2498                 if (igb_get_flash_presence_i210(hw)) {
2499                         if (hw->nvm.ops.validate(hw) < 0) {
2500                                 dev_err(&pdev->dev,
2501                                         "The NVM Checksum Is Not Valid\n");
2502                                 err = -EIO;
2503                                 goto err_eeprom;
2504                         }
2505                 }
2506                 break;
2507         default:
2508                 if (hw->nvm.ops.validate(hw) < 0) {
2509                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
2510                         err = -EIO;
2511                         goto err_eeprom;
2512                 }
2513                 break;
2514         }
2515
2516         if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
2517                 /* copy the MAC address out of the NVM */
2518                 if (hw->mac.ops.read_mac_addr(hw))
2519                         dev_err(&pdev->dev, "NVM Read Error\n");
2520         }
2521
2522         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2523
2524         if (!is_valid_ether_addr(netdev->dev_addr)) {
2525                 dev_err(&pdev->dev, "Invalid MAC Address\n");
2526                 err = -EIO;
2527                 goto err_eeprom;
2528         }
2529
2530         igb_set_default_mac_filter(adapter);
2531
2532         /* get firmware version for ethtool -i */
2533         igb_set_fw_version(adapter);
2534
2535         /* configure RXPBSIZE and TXPBSIZE */
2536         if (hw->mac.type == e1000_i210) {
2537                 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
2538                 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
2539         }
2540
2541         setup_timer(&adapter->watchdog_timer, igb_watchdog,
2542                     (unsigned long) adapter);
2543         setup_timer(&adapter->phy_info_timer, igb_update_phy_info,
2544                     (unsigned long) adapter);
2545
2546         INIT_WORK(&adapter->reset_task, igb_reset_task);
2547         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
2548
2549         /* Initialize link properties that are user-changeable */
2550         adapter->fc_autoneg = true;
2551         hw->mac.autoneg = true;
2552         hw->phy.autoneg_advertised = 0x2f;
2553
2554         hw->fc.requested_mode = e1000_fc_default;
2555         hw->fc.current_mode = e1000_fc_default;
2556
2557         igb_validate_mdi_setting(hw);
2558
2559         /* By default, support wake on port A */
2560         if (hw->bus.func == 0)
2561                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2562
2563         /* Check the NVM for wake support on non-port A ports */
2564         if (hw->mac.type >= e1000_82580)
2565                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
2566                                  NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
2567                                  &eeprom_data);
2568         else if (hw->bus.func == 1)
2569                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
2570
2571         if (eeprom_data & IGB_EEPROM_APME)
2572                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2573
2574         /* now that we have the eeprom settings, apply the special cases where
2575          * the eeprom may be wrong or the board simply won't support wake on
2576          * lan on a particular port
2577          */
2578         switch (pdev->device) {
2579         case E1000_DEV_ID_82575GB_QUAD_COPPER:
2580                 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2581                 break;
2582         case E1000_DEV_ID_82575EB_FIBER_SERDES:
2583         case E1000_DEV_ID_82576_FIBER:
2584         case E1000_DEV_ID_82576_SERDES:
2585                 /* Wake events only supported on port A for dual fiber
2586                  * regardless of eeprom setting
2587                  */
2588                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
2589                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2590                 break;
2591         case E1000_DEV_ID_82576_QUAD_COPPER:
2592         case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
2593                 /* if quad port adapter, disable WoL on all but port A */
2594                 if (global_quad_port_a != 0)
2595                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2596                 else
2597                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
2598                 /* Reset for multiple quad port adapters */
2599                 if (++global_quad_port_a == 4)
2600                         global_quad_port_a = 0;
2601                 break;
2602         default:
2603                 /* If the device can't wake, don't set software support */
2604                 if (!device_can_wakeup(&adapter->pdev->dev))
2605                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
2606         }
2607
2608         /* initialize the wol settings based on the eeprom settings */
2609         if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
2610                 adapter->wol |= E1000_WUFC_MAG;
2611
2612         /* Some vendors want WoL disabled by default, but still supported */
2613         if ((hw->mac.type == e1000_i350) &&
2614             (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
2615                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2616                 adapter->wol = 0;
2617         }
2618
2619         /* Some vendors want the ability to Use the EEPROM setting as
2620          * enable/disable only, and not for capability
2621          */
2622         if (((hw->mac.type == e1000_i350) ||
2623              (hw->mac.type == e1000_i354)) &&
2624             (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
2625                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2626                 adapter->wol = 0;
2627         }
2628         if (hw->mac.type == e1000_i350) {
2629                 if (((pdev->subsystem_device == 0x5001) ||
2630                      (pdev->subsystem_device == 0x5002)) &&
2631                                 (hw->bus.func == 0)) {
2632                         adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2633                         adapter->wol = 0;
2634                 }
2635                 if (pdev->subsystem_device == 0x1F52)
2636                         adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
2637         }
2638
2639         device_set_wakeup_enable(&adapter->pdev->dev,
2640                                  adapter->flags & IGB_FLAG_WOL_SUPPORTED);
2641
2642         /* reset the hardware with the new settings */
2643         igb_reset(adapter);
2644
2645         /* Init the I2C interface */
2646         err = igb_init_i2c(adapter);
2647         if (err) {
2648                 dev_err(&pdev->dev, "failed to init i2c interface\n");
2649                 goto err_eeprom;
2650         }
2651
2652         /* let the f/w know that the h/w is now under the control of the
2653          * driver.
2654          */
2655         igb_get_hw_control(adapter);
2656
2657         strcpy(netdev->name, "eth%d");
2658         err = register_netdev(netdev);
2659         if (err)
2660                 goto err_register;
2661
2662         /* carrier off reporting is important to ethtool even BEFORE open */
2663         netif_carrier_off(netdev);
2664
2665 #ifdef CONFIG_IGB_DCA
2666         if (dca_add_requester(&pdev->dev) == 0) {
2667                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
2668                 dev_info(&pdev->dev, "DCA enabled\n");
2669                 igb_setup_dca(adapter);
2670         }
2671
2672 #endif
2673 #ifdef CONFIG_IGB_HWMON
2674         /* Initialize the thermal sensor on i350 devices. */
2675         if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
2676                 u16 ets_word;
2677
2678                 /* Read the NVM to determine if this i350 device supports an
2679                  * external thermal sensor.
2680                  */
2681                 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
2682                 if (ets_word != 0x0000 && ets_word != 0xFFFF)
2683                         adapter->ets = true;
2684                 else
2685                         adapter->ets = false;
2686                 if (igb_sysfs_init(adapter))
2687                         dev_err(&pdev->dev,
2688                                 "failed to allocate sysfs resources\n");
2689         } else {
2690                 adapter->ets = false;
2691         }
2692 #endif
2693         /* Check if Media Autosense is enabled */
2694         adapter->ei = *ei;
2695         if (hw->dev_spec._82575.mas_capable)
2696                 igb_init_mas(adapter);
2697
2698         /* do hw tstamp init after resetting */
2699         igb_ptp_init(adapter);
2700
2701         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
2702         /* print bus type/speed/width info, not applicable to i354 */
2703         if (hw->mac.type != e1000_i354) {
2704                 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
2705                          netdev->name,
2706                          ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
2707                           (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
2708                            "unknown"),
2709                          ((hw->bus.width == e1000_bus_width_pcie_x4) ?
2710                           "Width x4" :
2711                           (hw->bus.width == e1000_bus_width_pcie_x2) ?
2712                           "Width x2" :
2713                           (hw->bus.width == e1000_bus_width_pcie_x1) ?
2714                           "Width x1" : "unknown"), netdev->dev_addr);
2715         }
2716
2717         if ((hw->mac.type >= e1000_i210 ||
2718              igb_get_flash_presence_i210(hw))) {
2719                 ret_val = igb_read_part_string(hw, part_str,
2720                                                E1000_PBANUM_LENGTH);
2721         } else {
2722                 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
2723         }
2724
2725         if (ret_val)
2726                 strcpy(part_str, "Unknown");
2727         dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
2728         dev_info(&pdev->dev,
2729                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
2730                 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
2731                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
2732                 adapter->num_rx_queues, adapter->num_tx_queues);
2733         if (hw->phy.media_type == e1000_media_type_copper) {
2734                 switch (hw->mac.type) {
2735                 case e1000_i350:
2736                 case e1000_i210:
2737                 case e1000_i211:
2738                         /* Enable EEE for internal copper PHY devices */
2739                         err = igb_set_eee_i350(hw, true, true);
2740                         if ((!err) &&
2741                             (!hw->dev_spec._82575.eee_disable)) {
2742                                 adapter->eee_advert =
2743                                         MDIO_EEE_100TX | MDIO_EEE_1000T;
2744                                 adapter->flags |= IGB_FLAG_EEE;
2745                         }
2746                         break;
2747                 case e1000_i354:
2748                         if ((rd32(E1000_CTRL_EXT) &
2749                             E1000_CTRL_EXT_LINK_MODE_SGMII)) {
2750                                 err = igb_set_eee_i354(hw, true, true);
2751                                 if ((!err) &&
2752                                         (!hw->dev_spec._82575.eee_disable)) {
2753                                         adapter->eee_advert =
2754                                            MDIO_EEE_100TX | MDIO_EEE_1000T;
2755                                         adapter->flags |= IGB_FLAG_EEE;
2756                                 }
2757                         }
2758                         break;
2759                 default:
2760                         break;
2761                 }
2762         }
2763         pm_runtime_put_noidle(&pdev->dev);
2764         return 0;
2765
2766 err_register:
2767         igb_release_hw_control(adapter);
2768         memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
2769 err_eeprom:
2770         if (!igb_check_reset_block(hw))
2771                 igb_reset_phy(hw);
2772
2773         if (hw->flash_address)
2774                 iounmap(hw->flash_address);
2775 err_sw_init:
2776         kfree(adapter->mac_table);
2777         kfree(adapter->shadow_vfta);
2778         igb_clear_interrupt_scheme(adapter);
2779 #ifdef CONFIG_PCI_IOV
2780         igb_disable_sriov(pdev);
2781 #endif
2782         pci_iounmap(pdev, adapter->io_addr);
2783 err_ioremap:
2784         free_netdev(netdev);
2785 err_alloc_etherdev:
2786         pci_release_mem_regions(pdev);
2787 err_pci_reg:
2788 err_dma:
2789         pci_disable_device(pdev);
2790         return err;
2791 }
2792
2793 #ifdef CONFIG_PCI_IOV
2794 static int igb_disable_sriov(struct pci_dev *pdev)
2795 {
2796         struct net_device *netdev = pci_get_drvdata(pdev);
2797         struct igb_adapter *adapter = netdev_priv(netdev);
2798         struct e1000_hw *hw = &adapter->hw;
2799
2800         /* reclaim resources allocated to VFs */
2801         if (adapter->vf_data) {
2802                 /* disable iov and allow time for transactions to clear */
2803                 if (pci_vfs_assigned(pdev)) {
2804                         dev_warn(&pdev->dev,
2805                                  "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
2806                         return -EPERM;
2807                 } else {
2808                         pci_disable_sriov(pdev);
2809                         msleep(500);
2810                 }
2811
2812                 kfree(adapter->vf_mac_list);
2813                 adapter->vf_mac_list = NULL;
2814                 kfree(adapter->vf_data);
2815                 adapter->vf_data = NULL;
2816                 adapter->vfs_allocated_count = 0;
2817                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
2818                 wrfl();
2819                 msleep(100);
2820                 dev_info(&pdev->dev, "IOV Disabled\n");
2821
2822                 /* Re-enable DMA Coalescing flag since IOV is turned off */
2823                 adapter->flags |= IGB_FLAG_DMAC;
2824         }
2825
2826         return 0;
2827 }
2828
2829 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
2830 {
2831         struct net_device *netdev = pci_get_drvdata(pdev);
2832         struct igb_adapter *adapter = netdev_priv(netdev);
2833         int old_vfs = pci_num_vf(pdev);
2834         struct vf_mac_filter *mac_list;
2835         int err = 0;
2836         int num_vf_mac_filters, i;
2837
2838         if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
2839                 err = -EPERM;
2840                 goto out;
2841         }
2842         if (!num_vfs)
2843                 goto out;
2844
2845         if (old_vfs) {
2846                 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
2847                          old_vfs, max_vfs);
2848                 adapter->vfs_allocated_count = old_vfs;
2849         } else
2850                 adapter->vfs_allocated_count = num_vfs;
2851
2852         adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
2853                                 sizeof(struct vf_data_storage), GFP_KERNEL);
2854
2855         /* if allocation failed then we do not support SR-IOV */
2856         if (!adapter->vf_data) {
2857                 adapter->vfs_allocated_count = 0;
2858                 dev_err(&pdev->dev,
2859                         "Unable to allocate memory for VF Data Storage\n");
2860                 err = -ENOMEM;
2861                 goto out;
2862         }
2863
2864         /* Due to the limited number of RAR entries calculate potential
2865          * number of MAC filters available for the VFs. Reserve entries
2866          * for PF default MAC, PF MAC filters and at least one RAR entry
2867          * for each VF for VF MAC.
2868          */
2869         num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
2870                              (1 + IGB_PF_MAC_FILTERS_RESERVED +
2871                               adapter->vfs_allocated_count);
2872
2873         adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
2874                                        sizeof(struct vf_mac_filter),
2875                                        GFP_KERNEL);
2876
2877         mac_list = adapter->vf_mac_list;
2878         INIT_LIST_HEAD(&adapter->vf_macs.l);
2879
2880         if (adapter->vf_mac_list) {
2881                 /* Initialize list of VF MAC filters */
2882                 for (i = 0; i < num_vf_mac_filters; i++) {
2883                         mac_list->vf = -1;
2884                         mac_list->free = true;
2885                         list_add(&mac_list->l, &adapter->vf_macs.l);
2886                         mac_list++;
2887                 }
2888         } else {
2889                 /* If we could not allocate memory for the VF MAC filters
2890                  * we can continue without this feature but warn user.
2891                  */
2892                 dev_err(&pdev->dev,
2893                         "Unable to allocate memory for VF MAC filter list\n");
2894         }
2895
2896         /* only call pci_enable_sriov() if no VFs are allocated already */
2897         if (!old_vfs) {
2898                 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
2899                 if (err)
2900                         goto err_out;
2901         }
2902         dev_info(&pdev->dev, "%d VFs allocated\n",
2903                  adapter->vfs_allocated_count);
2904         for (i = 0; i < adapter->vfs_allocated_count; i++)
2905                 igb_vf_configure(adapter, i);
2906
2907         /* DMA Coalescing is not supported in IOV mode. */
2908         adapter->flags &= ~IGB_FLAG_DMAC;
2909         goto out;
2910
2911 err_out:
2912         kfree(adapter->vf_mac_list);
2913         adapter->vf_mac_list = NULL;
2914         kfree(adapter->vf_data);
2915         adapter->vf_data = NULL;
2916         adapter->vfs_allocated_count = 0;
2917 out:
2918         return err;
2919 }
2920
2921 #endif
2922 /**
2923  *  igb_remove_i2c - Cleanup  I2C interface
2924  *  @adapter: pointer to adapter structure
2925  **/
2926 static void igb_remove_i2c(struct igb_adapter *adapter)
2927 {
2928         /* free the adapter bus structure */
2929         i2c_del_adapter(&adapter->i2c_adap);
2930 }
2931
2932 /**
2933  *  igb_remove - Device Removal Routine
2934  *  @pdev: PCI device information struct
2935  *
2936  *  igb_remove is called by the PCI subsystem to alert the driver
2937  *  that it should release a PCI device.  The could be caused by a
2938  *  Hot-Plug event, or because the driver is going to be removed from
2939  *  memory.
2940  **/
2941 static void igb_remove(struct pci_dev *pdev)
2942 {
2943         struct net_device *netdev = pci_get_drvdata(pdev);
2944         struct igb_adapter *adapter = netdev_priv(netdev);
2945         struct e1000_hw *hw = &adapter->hw;
2946
2947         pm_runtime_get_noresume(&pdev->dev);
2948 #ifdef CONFIG_IGB_HWMON
2949         igb_sysfs_exit(adapter);
2950 #endif
2951         igb_remove_i2c(adapter);
2952         igb_ptp_stop(adapter);
2953         /* The watchdog timer may be rescheduled, so explicitly
2954          * disable watchdog from being rescheduled.
2955          */
2956         set_bit(__IGB_DOWN, &adapter->state);
2957         del_timer_sync(&adapter->watchdog_timer);
2958         del_timer_sync(&adapter->phy_info_timer);
2959
2960         cancel_work_sync(&adapter->reset_task);
2961         cancel_work_sync(&adapter->watchdog_task);
2962
2963 #ifdef CONFIG_IGB_DCA
2964         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
2965                 dev_info(&pdev->dev, "DCA disabled\n");
2966                 dca_remove_requester(&pdev->dev);
2967                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
2968                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
2969         }
2970 #endif
2971
2972         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
2973          * would have already happened in close and is redundant.
2974          */
2975         igb_release_hw_control(adapter);
2976
2977 #ifdef CONFIG_PCI_IOV
2978         igb_disable_sriov(pdev);
2979 #endif
2980
2981         unregister_netdev(netdev);
2982
2983         igb_clear_interrupt_scheme(adapter);
2984
2985         pci_iounmap(pdev, adapter->io_addr);
2986         if (hw->flash_address)
2987                 iounmap(hw->flash_address);
2988         pci_release_mem_regions(pdev);
2989
2990         kfree(adapter->mac_table);
2991         kfree(adapter->shadow_vfta);
2992         free_netdev(netdev);
2993
2994         pci_disable_pcie_error_reporting(pdev);
2995
2996         pci_disable_device(pdev);
2997 }
2998
2999 /**
3000  *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3001  *  @adapter: board private structure to initialize
3002  *
3003  *  This function initializes the vf specific data storage and then attempts to
3004  *  allocate the VFs.  The reason for ordering it this way is because it is much
3005  *  mor expensive time wise to disable SR-IOV than it is to allocate and free
3006  *  the memory for the VFs.
3007  **/
3008 static void igb_probe_vfs(struct igb_adapter *adapter)
3009 {
3010 #ifdef CONFIG_PCI_IOV
3011         struct pci_dev *pdev = adapter->pdev;
3012         struct e1000_hw *hw = &adapter->hw;
3013
3014         /* Virtualization features not supported on i210 family. */
3015         if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3016                 return;
3017
3018         /* Of the below we really only want the effect of getting
3019          * IGB_FLAG_HAS_MSIX set (if available), without which
3020          * igb_enable_sriov() has no effect.
3021          */
3022         igb_set_interrupt_capability(adapter, true);
3023         igb_reset_interrupt_capability(adapter);
3024
3025         pci_sriov_set_totalvfs(pdev, 7);
3026         igb_enable_sriov(pdev, max_vfs);
3027
3028 #endif /* CONFIG_PCI_IOV */
3029 }
3030
3031 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3032 {
3033         struct e1000_hw *hw = &adapter->hw;
3034         u32 max_rss_queues;
3035
3036         /* Determine the maximum number of RSS queues supported. */
3037         switch (hw->mac.type) {
3038         case e1000_i211:
3039                 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3040                 break;
3041         case e1000_82575:
3042         case e1000_i210:
3043                 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3044                 break;
3045         case e1000_i350:
3046                 /* I350 cannot do RSS and SR-IOV at the same time */
3047                 if (!!adapter->vfs_allocated_count) {
3048                         max_rss_queues = 1;
3049                         break;
3050                 }
3051                 /* fall through */
3052         case e1000_82576:
3053                 if (!!adapter->vfs_allocated_count) {
3054                         max_rss_queues = 2;
3055                         break;
3056                 }
3057                 /* fall through */
3058         case e1000_82580:
3059         case e1000_i354:
3060         default:
3061                 max_rss_queues = IGB_MAX_RX_QUEUES;
3062                 break;
3063         }
3064
3065         adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3066
3067         igb_set_flag_queue_pairs(adapter, max_rss_queues);
3068 }
3069
3070 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3071                               const u32 max_rss_queues)
3072 {
3073         struct e1000_hw *hw = &adapter->hw;
3074
3075         /* Determine if we need to pair queues. */
3076         switch (hw->mac.type) {
3077         case e1000_82575:
3078         case e1000_i211:
3079                 /* Device supports enough interrupts without queue pairing. */
3080                 break;
3081         case e1000_82576:
3082         case e1000_82580:
3083         case e1000_i350:
3084         case e1000_i354:
3085         case e1000_i210:
3086         default:
3087                 /* If rss_queues > half of max_rss_queues, pair the queues in
3088                  * order to conserve interrupts due to limited supply.
3089                  */
3090                 if (adapter->rss_queues > (max_rss_queues / 2))
3091                         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3092                 else
3093                         adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3094                 break;
3095         }
3096 }
3097
3098 /**
3099  *  igb_sw_init - Initialize general software structures (struct igb_adapter)
3100  *  @adapter: board private structure to initialize
3101  *
3102  *  igb_sw_init initializes the Adapter private data structure.
3103  *  Fields are initialized based on PCI device information and
3104  *  OS network device settings (MTU size).
3105  **/
3106 static int igb_sw_init(struct igb_adapter *adapter)
3107 {
3108         struct e1000_hw *hw = &adapter->hw;
3109         struct net_device *netdev = adapter->netdev;
3110         struct pci_dev *pdev = adapter->pdev;
3111
3112         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3113
3114         /* set default ring sizes */
3115         adapter->tx_ring_count = IGB_DEFAULT_TXD;
3116         adapter->rx_ring_count = IGB_DEFAULT_RXD;
3117
3118         /* set default ITR values */
3119         adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3120         adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3121
3122         /* set default work limits */
3123         adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3124
3125         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN +
3126                                   VLAN_HLEN;
3127         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3128
3129         spin_lock_init(&adapter->nfc_lock);
3130         spin_lock_init(&adapter->stats64_lock);
3131 #ifdef CONFIG_PCI_IOV
3132         switch (hw->mac.type) {
3133         case e1000_82576:
3134         case e1000_i350:
3135                 if (max_vfs > 7) {
3136                         dev_warn(&pdev->dev,
3137                                  "Maximum of 7 VFs per PF, using max\n");
3138                         max_vfs = adapter->vfs_allocated_count = 7;
3139                 } else
3140                         adapter->vfs_allocated_count = max_vfs;
3141                 if (adapter->vfs_allocated_count)
3142                         dev_warn(&pdev->dev,
3143                                  "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3144                 break;
3145         default:
3146                 break;
3147         }
3148 #endif /* CONFIG_PCI_IOV */
3149
3150         /* Assume MSI-X interrupts, will be checked during IRQ allocation */
3151         adapter->flags |= IGB_FLAG_HAS_MSIX;
3152
3153         adapter->mac_table = kzalloc(sizeof(struct igb_mac_addr) *
3154                                      hw->mac.rar_entry_count, GFP_ATOMIC);
3155         if (!adapter->mac_table)
3156                 return -ENOMEM;
3157
3158         igb_probe_vfs(adapter);
3159
3160         igb_init_queue_configuration(adapter);
3161
3162         /* Setup and initialize a copy of the hw vlan table array */
3163         adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
3164                                        GFP_ATOMIC);
3165
3166         /* This call may decrease the number of queues */
3167         if (igb_init_interrupt_scheme(adapter, true)) {
3168                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
3169                 return -ENOMEM;
3170         }
3171
3172         /* Explicitly disable IRQ since the NIC can be in any state. */
3173         igb_irq_disable(adapter);
3174
3175         if (hw->mac.type >= e1000_i350)
3176                 adapter->flags &= ~IGB_FLAG_DMAC;
3177
3178         set_bit(__IGB_DOWN, &adapter->state);
3179         return 0;
3180 }
3181
3182 /**
3183  *  igb_open - Called when a network interface is made active
3184  *  @netdev: network interface device structure
3185  *
3186  *  Returns 0 on success, negative value on failure
3187  *
3188  *  The open entry point is called when a network interface is made
3189  *  active by the system (IFF_UP).  At this point all resources needed
3190  *  for transmit and receive operations are allocated, the interrupt
3191  *  handler is registered with the OS, the watchdog timer is started,
3192  *  and the stack is notified that the interface is ready.
3193  **/
3194 static int __igb_open(struct net_device *netdev, bool resuming)
3195 {
3196         struct igb_adapter *adapter = netdev_priv(netdev);
3197         struct e1000_hw *hw = &adapter->hw;
3198         struct pci_dev *pdev = adapter->pdev;
3199         int err;
3200         int i;
3201
3202         /* disallow open during test */
3203         if (test_bit(__IGB_TESTING, &adapter->state)) {
3204                 WARN_ON(resuming);
3205                 return -EBUSY;
3206         }
3207
3208         if (!resuming)
3209                 pm_runtime_get_sync(&pdev->dev);
3210
3211         netif_carrier_off(netdev);
3212
3213         /* allocate transmit descriptors */
3214         err = igb_setup_all_tx_resources(adapter);
3215         if (err)
3216                 goto err_setup_tx;
3217
3218         /* allocate receive descriptors */
3219         err = igb_setup_all_rx_resources(adapter);
3220         if (err)
3221                 goto err_setup_rx;
3222
3223         igb_power_up_link(adapter);
3224
3225         /* before we allocate an interrupt, we must be ready to handle it.
3226          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
3227          * as soon as we call pci_request_irq, so we have to setup our
3228          * clean_rx handler before we do so.
3229          */
3230         igb_configure(adapter);
3231
3232         err = igb_request_irq(adapter);
3233         if (err)
3234                 goto err_req_irq;
3235
3236         /* Notify the stack of the actual queue counts. */
3237         err = netif_set_real_num_tx_queues(adapter->netdev,
3238                                            adapter->num_tx_queues);
3239         if (err)
3240                 goto err_set_queues;
3241
3242         err = netif_set_real_num_rx_queues(adapter->netdev,
3243                                            adapter->num_rx_queues);
3244         if (err)
3245                 goto err_set_queues;
3246
3247         /* From here on the code is the same as igb_up() */
3248         clear_bit(__IGB_DOWN, &adapter->state);
3249
3250         for (i = 0; i < adapter->num_q_vectors; i++)
3251                 napi_enable(&(adapter->q_vector[i]->napi));
3252
3253         /* Clear any pending interrupts. */
3254         rd32(E1000_ICR);
3255
3256         igb_irq_enable(adapter);
3257
3258         /* notify VFs that reset has been completed */
3259         if (adapter->vfs_allocated_count) {
3260                 u32 reg_data = rd32(E1000_CTRL_EXT);
3261
3262                 reg_data |= E1000_CTRL_EXT_PFRSTD;
3263                 wr32(E1000_CTRL_EXT, reg_data);
3264         }
3265
3266         netif_tx_start_all_queues(netdev);
3267
3268         if (!resuming)
3269                 pm_runtime_put(&pdev->dev);
3270
3271         /* start the watchdog. */
3272         hw->mac.get_link_status = 1;
3273         schedule_work(&adapter->watchdog_task);
3274
3275         return 0;
3276
3277 err_set_queues:
3278         igb_free_irq(adapter);
3279 err_req_irq:
3280         igb_release_hw_control(adapter);
3281         igb_power_down_link(adapter);
3282         igb_free_all_rx_resources(adapter);
3283 err_setup_rx:
3284         igb_free_all_tx_resources(adapter);
3285 err_setup_tx:
3286         igb_reset(adapter);
3287         if (!resuming)
3288                 pm_runtime_put(&pdev->dev);
3289
3290         return err;
3291 }
3292
3293 int igb_open(struct net_device *netdev)
3294 {
3295         return __igb_open(netdev, false);
3296 }
3297
3298 /**
3299  *  igb_close - Disables a network interface
3300  *  @netdev: network interface device structure
3301  *
3302  *  Returns 0, this is not allowed to fail
3303  *
3304  *  The close entry point is called when an interface is de-activated
3305  *  by the OS.  The hardware is still under the driver's control, but
3306  *  needs to be disabled.  A global MAC reset is issued to stop the
3307  *  hardware, and all transmit and receive resources are freed.
3308  **/
3309 static int __igb_close(struct net_device *netdev, bool suspending)
3310 {
3311         struct igb_adapter *adapter = netdev_priv(netdev);
3312         struct pci_dev *pdev = adapter->pdev;
3313
3314         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3315
3316         if (!suspending)
3317                 pm_runtime_get_sync(&pdev->dev);
3318
3319         igb_down(adapter);
3320         igb_free_irq(adapter);
3321
3322         igb_free_all_tx_resources(adapter);
3323         igb_free_all_rx_resources(adapter);
3324
3325         if (!suspending)
3326                 pm_runtime_put_sync(&pdev->dev);
3327         return 0;
3328 }
3329
3330 int igb_close(struct net_device *netdev)
3331 {
3332         if (netif_device_present(netdev))
3333                 return __igb_close(netdev, false);
3334         return 0;
3335 }
3336
3337 /**
3338  *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
3339  *  @tx_ring: tx descriptor ring (for a specific queue) to setup
3340  *
3341  *  Return 0 on success, negative on failure
3342  **/
3343 int igb_setup_tx_resources(struct igb_ring *tx_ring)
3344 {
3345         struct device *dev = tx_ring->dev;
3346         int size;
3347
3348         size = sizeof(struct igb_tx_buffer) * tx_ring->count;
3349
3350         tx_ring->tx_buffer_info = vmalloc(size);
3351         if (!tx_ring->tx_buffer_info)
3352                 goto err;
3353
3354         /* round up to nearest 4K */
3355         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
3356         tx_ring->size = ALIGN(tx_ring->size, 4096);
3357
3358         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
3359                                            &tx_ring->dma, GFP_KERNEL);
3360         if (!tx_ring->desc)
3361                 goto err;
3362
3363         tx_ring->next_to_use = 0;
3364         tx_ring->next_to_clean = 0;
3365
3366         return 0;
3367
3368 err:
3369         vfree(tx_ring->tx_buffer_info);
3370         tx_ring->tx_buffer_info = NULL;
3371         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
3372         return -ENOMEM;
3373 }
3374
3375 /**
3376  *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
3377  *                               (Descriptors) for all queues
3378  *  @adapter: board private structure
3379  *
3380  *  Return 0 on success, negative on failure
3381  **/
3382 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
3383 {
3384         struct pci_dev *pdev = adapter->pdev;
3385         int i, err = 0;
3386
3387         for (i = 0; i < adapter->num_tx_queues; i++) {
3388                 err = igb_setup_tx_resources(adapter->tx_ring[i]);
3389                 if (err) {
3390                         dev_err(&pdev->dev,
3391                                 "Allocation for Tx Queue %u failed\n", i);
3392                         for (i--; i >= 0; i--)
3393                                 igb_free_tx_resources(adapter->tx_ring[i]);
3394                         break;
3395                 }
3396         }
3397
3398         return err;
3399 }
3400
3401 /**
3402  *  igb_setup_tctl - configure the transmit control registers
3403  *  @adapter: Board private structure
3404  **/
3405 void igb_setup_tctl(struct igb_adapter *adapter)
3406 {
3407         struct e1000_hw *hw = &adapter->hw;
3408         u32 tctl;
3409
3410         /* disable queue 0 which is enabled by default on 82575 and 82576 */
3411         wr32(E1000_TXDCTL(0), 0);
3412
3413         /* Program the Transmit Control Register */
3414         tctl = rd32(E1000_TCTL);
3415         tctl &= ~E1000_TCTL_CT;
3416         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
3417                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
3418
3419         igb_config_collision_dist(hw);
3420
3421         /* Enable transmits */
3422         tctl |= E1000_TCTL_EN;
3423
3424         wr32(E1000_TCTL, tctl);
3425 }
3426
3427 /**
3428  *  igb_configure_tx_ring - Configure transmit ring after Reset
3429  *  @adapter: board private structure
3430  *  @ring: tx ring to configure
3431  *
3432  *  Configure a transmit ring after a reset.
3433  **/
3434 void igb_configure_tx_ring(struct igb_adapter *adapter,
3435                            struct igb_ring *ring)
3436 {
3437         struct e1000_hw *hw = &adapter->hw;
3438         u32 txdctl = 0;
3439         u64 tdba = ring->dma;
3440         int reg_idx = ring->reg_idx;
3441
3442         /* disable the queue */
3443         wr32(E1000_TXDCTL(reg_idx), 0);
3444         wrfl();
3445         mdelay(10);
3446
3447         wr32(E1000_TDLEN(reg_idx),
3448              ring->count * sizeof(union e1000_adv_tx_desc));
3449         wr32(E1000_TDBAL(reg_idx),
3450              tdba & 0x00000000ffffffffULL);
3451         wr32(E1000_TDBAH(reg_idx), tdba >> 32);
3452
3453         ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
3454         wr32(E1000_TDH(reg_idx), 0);
3455         writel(0, ring->tail);
3456
3457         txdctl |= IGB_TX_PTHRESH;
3458         txdctl |= IGB_TX_HTHRESH << 8;
3459         txdctl |= IGB_TX_WTHRESH << 16;
3460
3461         /* reinitialize tx_buffer_info */
3462         memset(ring->tx_buffer_info, 0,
3463                sizeof(struct igb_tx_buffer) * ring->count);
3464
3465         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
3466         wr32(E1000_TXDCTL(reg_idx), txdctl);
3467 }
3468
3469 /**
3470  *  igb_configure_tx - Configure transmit Unit after Reset
3471  *  @adapter: board private structure
3472  *
3473  *  Configure the Tx unit of the MAC after a reset.
3474  **/
3475 static void igb_configure_tx(struct igb_adapter *adapter)
3476 {
3477         int i;
3478
3479         for (i = 0; i < adapter->num_tx_queues; i++)
3480                 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
3481 }
3482
3483 /**
3484  *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
3485  *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
3486  *
3487  *  Returns 0 on success, negative on failure
3488  **/
3489 int igb_setup_rx_resources(struct igb_ring *rx_ring)
3490 {
3491         struct device *dev = rx_ring->dev;
3492         int size;
3493
3494         size = sizeof(struct igb_rx_buffer) * rx_ring->count;
3495
3496         rx_ring->rx_buffer_info = vmalloc(size);
3497         if (!rx_ring->rx_buffer_info)
3498                 goto err;
3499
3500         /* Round up to nearest 4K */
3501         rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
3502         rx_ring->size = ALIGN(rx_ring->size, 4096);
3503
3504         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
3505                                            &rx_ring->dma, GFP_KERNEL);
3506         if (!rx_ring->desc)
3507                 goto err;
3508
3509         rx_ring->next_to_alloc = 0;
3510         rx_ring->next_to_clean = 0;
3511         rx_ring->next_to_use = 0;
3512
3513         return 0;
3514
3515 err:
3516         vfree(rx_ring->rx_buffer_info);
3517         rx_ring->rx_buffer_info = NULL;
3518         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
3519         return -ENOMEM;
3520 }
3521
3522 /**
3523  *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
3524  *                               (Descriptors) for all queues
3525  *  @adapter: board private structure
3526  *
3527  *  Return 0 on success, negative on failure
3528  **/
3529 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
3530 {
3531         struct pci_dev *pdev = adapter->pdev;
3532         int i, err = 0;
3533
3534         for (i = 0; i < adapter->num_rx_queues; i++) {
3535                 err = igb_setup_rx_resources(adapter->rx_ring[i]);
3536                 if (err) {
3537                         dev_err(&pdev->dev,
3538                                 "Allocation for Rx Queue %u failed\n", i);
3539                         for (i--; i >= 0; i--)
3540                                 igb_free_rx_resources(adapter->rx_ring[i]);
3541                         break;
3542                 }
3543         }
3544
3545         return err;
3546 }
3547
3548 /**
3549  *  igb_setup_mrqc - configure the multiple receive queue control registers
3550  *  @adapter: Board private structure
3551  **/
3552 static void igb_setup_mrqc(struct igb_adapter *adapter)
3553 {
3554         struct e1000_hw *hw = &adapter->hw;
3555         u32 mrqc, rxcsum;
3556         u32 j, num_rx_queues;
3557         u32 rss_key[10];
3558
3559         netdev_rss_key_fill(rss_key, sizeof(rss_key));
3560         for (j = 0; j < 10; j++)
3561                 wr32(E1000_RSSRK(j), rss_key[j]);
3562
3563         num_rx_queues = adapter->rss_queues;
3564
3565         switch (hw->mac.type) {
3566         case e1000_82576:
3567                 /* 82576 supports 2 RSS queues for SR-IOV */
3568                 if (adapter->vfs_allocated_count)
3569                         num_rx_queues = 2;
3570                 break;
3571         default:
3572                 break;
3573         }
3574
3575         if (adapter->rss_indir_tbl_init != num_rx_queues) {
3576                 for (j = 0; j < IGB_RETA_SIZE; j++)
3577                         adapter->rss_indir_tbl[j] =
3578                         (j * num_rx_queues) / IGB_RETA_SIZE;
3579                 adapter->rss_indir_tbl_init = num_rx_queues;
3580         }
3581         igb_write_rss_indir_tbl(adapter);
3582
3583         /* Disable raw packet checksumming so that RSS hash is placed in
3584          * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
3585          * offloads as they are enabled by default
3586          */
3587         rxcsum = rd32(E1000_RXCSUM);
3588         rxcsum |= E1000_RXCSUM_PCSD;
3589
3590         if (adapter->hw.mac.type >= e1000_82576)
3591                 /* Enable Receive Checksum Offload for SCTP */
3592                 rxcsum |= E1000_RXCSUM_CRCOFL;
3593
3594         /* Don't need to set TUOFL or IPOFL, they default to 1 */
3595         wr32(E1000_RXCSUM, rxcsum);
3596
3597         /* Generate RSS hash based on packet types, TCP/UDP
3598          * port numbers and/or IPv4/v6 src and dst addresses
3599          */
3600         mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
3601                E1000_MRQC_RSS_FIELD_IPV4_TCP |
3602                E1000_MRQC_RSS_FIELD_IPV6 |
3603                E1000_MRQC_RSS_FIELD_IPV6_TCP |
3604                E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
3605
3606         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
3607                 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
3608         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
3609                 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
3610
3611         /* If VMDq is enabled then we set the appropriate mode for that, else
3612          * we default to RSS so that an RSS hash is calculated per packet even
3613          * if we are only using one queue
3614          */
3615         if (adapter->vfs_allocated_count) {
3616                 if (hw->mac.type > e1000_82575) {
3617                         /* Set the default pool for the PF's first queue */
3618                         u32 vtctl = rd32(E1000_VT_CTL);
3619
3620                         vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
3621                                    E1000_VT_CTL_DISABLE_DEF_POOL);
3622                         vtctl |= adapter->vfs_allocated_count <<
3623                                 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
3624                         wr32(E1000_VT_CTL, vtctl);
3625                 }
3626                 if (adapter->rss_queues > 1)
3627                         mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
3628                 else
3629                         mrqc |= E1000_MRQC_ENABLE_VMDQ;
3630         } else {
3631                 if (hw->mac.type != e1000_i211)
3632                         mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
3633         }
3634         igb_vmm_control(adapter);
3635
3636         wr32(E1000_MRQC, mrqc);
3637 }
3638
3639 /**
3640  *  igb_setup_rctl - configure the receive control registers
3641  *  @adapter: Board private structure
3642  **/
3643 void igb_setup_rctl(struct igb_adapter *adapter)
3644 {
3645         struct e1000_hw *hw = &adapter->hw;
3646         u32 rctl;
3647
3648         rctl = rd32(E1000_RCTL);
3649
3650         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
3651         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
3652
3653         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
3654                 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
3655
3656         /* enable stripping of CRC. It's unlikely this will break BMC
3657          * redirection as it did with e1000. Newer features require
3658          * that the HW strips the CRC.
3659          */
3660         rctl |= E1000_RCTL_SECRC;
3661
3662         /* disable store bad packets and clear size bits. */
3663         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
3664
3665         /* enable LPE to allow for reception of jumbo frames */
3666         rctl |= E1000_RCTL_LPE;
3667
3668         /* disable queue 0 to prevent tail write w/o re-config */
3669         wr32(E1000_RXDCTL(0), 0);
3670
3671         /* Attention!!!  For SR-IOV PF driver operations you must enable
3672          * queue drop for all VF and PF queues to prevent head of line blocking
3673          * if an un-trusted VF does not provide descriptors to hardware.
3674          */
3675         if (adapter->vfs_allocated_count) {
3676                 /* set all queue drop enable bits */
3677                 wr32(E1000_QDE, ALL_QUEUES);
3678         }
3679
3680         /* This is useful for sniffing bad packets. */
3681         if (adapter->netdev->features & NETIF_F_RXALL) {
3682                 /* UPE and MPE will be handled by normal PROMISC logic
3683                  * in e1000e_set_rx_mode
3684                  */
3685                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
3686                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
3687                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
3688
3689                 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
3690                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
3691                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
3692                  * and that breaks VLANs.
3693                  */
3694         }
3695
3696         wr32(E1000_RCTL, rctl);
3697 }
3698
3699 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
3700                                    int vfn)
3701 {
3702         struct e1000_hw *hw = &adapter->hw;
3703         u32 vmolr;
3704
3705         if (size > MAX_JUMBO_FRAME_SIZE)
3706                 size = MAX_JUMBO_FRAME_SIZE;
3707
3708         vmolr = rd32(E1000_VMOLR(vfn));
3709         vmolr &= ~E1000_VMOLR_RLPML_MASK;
3710         vmolr |= size | E1000_VMOLR_LPE;
3711         wr32(E1000_VMOLR(vfn), vmolr);
3712
3713         return 0;
3714 }
3715
3716 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
3717                                          int vfn, bool enable)
3718 {
3719         struct e1000_hw *hw = &adapter->hw;
3720         u32 val, reg;
3721
3722         if (hw->mac.type < e1000_82576)
3723                 return;
3724
3725         if (hw->mac.type == e1000_i350)
3726                 reg = E1000_DVMOLR(vfn);
3727         else
3728                 reg = E1000_VMOLR(vfn);
3729
3730         val = rd32(reg);
3731         if (enable)
3732                 val |= E1000_VMOLR_STRVLAN;
3733         else
3734                 val &= ~(E1000_VMOLR_STRVLAN);
3735         wr32(reg, val);
3736 }
3737
3738 static inline void igb_set_vmolr(struct igb_adapter *adapter,
3739                                  int vfn, bool aupe)
3740 {
3741         struct e1000_hw *hw = &adapter->hw;
3742         u32 vmolr;
3743
3744         /* This register exists only on 82576 and newer so if we are older then
3745          * we should exit and do nothing
3746          */
3747         if (hw->mac.type < e1000_82576)
3748                 return;
3749
3750         vmolr = rd32(E1000_VMOLR(vfn));
3751         if (aupe)
3752                 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
3753         else
3754                 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
3755
3756         /* clear all bits that might not be set */
3757         vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
3758
3759         if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
3760                 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
3761         /* for VMDq only allow the VFs and pool 0 to accept broadcast and
3762          * multicast packets
3763          */
3764         if (vfn <= adapter->vfs_allocated_count)
3765                 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
3766
3767         wr32(E1000_VMOLR(vfn), vmolr);
3768 }
3769
3770 /**
3771  *  igb_configure_rx_ring - Configure a receive ring after Reset
3772  *  @adapter: board private structure
3773  *  @ring: receive ring to be configured
3774  *
3775  *  Configure the Rx unit of the MAC after a reset.
3776  **/
3777 void igb_configure_rx_ring(struct igb_adapter *adapter,
3778                            struct igb_ring *ring)
3779 {
3780         struct e1000_hw *hw = &adapter->hw;
3781         union e1000_adv_rx_desc *rx_desc;
3782         u64 rdba = ring->dma;
3783         int reg_idx = ring->reg_idx;
3784         u32 srrctl = 0, rxdctl = 0;
3785
3786         /* disable the queue */
3787         wr32(E1000_RXDCTL(reg_idx), 0);
3788
3789         /* Set DMA base address registers */
3790         wr32(E1000_RDBAL(reg_idx),
3791              rdba & 0x00000000ffffffffULL);
3792         wr32(E1000_RDBAH(reg_idx), rdba >> 32);
3793         wr32(E1000_RDLEN(reg_idx),
3794              ring->count * sizeof(union e1000_adv_rx_desc));
3795
3796         /* initialize head and tail */
3797         ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
3798         wr32(E1000_RDH(reg_idx), 0);
3799         writel(0, ring->tail);
3800
3801         /* set descriptor configuration */
3802         srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
3803         if (ring_uses_large_buffer(ring))
3804                 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3805         else
3806                 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
3807         srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
3808         if (hw->mac.type >= e1000_82580)
3809                 srrctl |= E1000_SRRCTL_TIMESTAMP;
3810         /* Only set Drop Enable if we are supporting multiple queues */
3811         if (adapter->vfs_allocated_count || adapter->num_rx_queues > 1)
3812                 srrctl |= E1000_SRRCTL_DROP_EN;
3813
3814         wr32(E1000_SRRCTL(reg_idx), srrctl);
3815
3816         /* set filtering for VMDQ pools */
3817         igb_set_vmolr(adapter, reg_idx & 0x7, true);
3818
3819         rxdctl |= IGB_RX_PTHRESH;
3820         rxdctl |= IGB_RX_HTHRESH << 8;
3821         rxdctl |= IGB_RX_WTHRESH << 16;
3822
3823         /* initialize rx_buffer_info */
3824         memset(ring->rx_buffer_info, 0,
3825                sizeof(struct igb_rx_buffer) * ring->count);
3826
3827         /* initialize Rx descriptor 0 */
3828         rx_desc = IGB_RX_DESC(ring, 0);
3829         rx_desc->wb.upper.length = 0;
3830
3831         /* enable receive descriptor fetching */
3832         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
3833         wr32(E1000_RXDCTL(reg_idx), rxdctl);
3834 }
3835
3836 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
3837                                   struct igb_ring *rx_ring)
3838 {
3839         /* set build_skb and buffer size flags */
3840         clear_ring_build_skb_enabled(rx_ring);
3841         clear_ring_uses_large_buffer(rx_ring);
3842
3843         if (adapter->flags & IGB_FLAG_RX_LEGACY)
3844                 return;
3845
3846         set_ring_build_skb_enabled(rx_ring);
3847
3848 #if (PAGE_SIZE < 8192)
3849         if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
3850                 return;
3851
3852         set_ring_uses_large_buffer(rx_ring);
3853 #endif
3854 }
3855
3856 /**
3857  *  igb_configure_rx - Configure receive Unit after Reset
3858  *  @adapter: board private structure
3859  *
3860  *  Configure the Rx unit of the MAC after a reset.
3861  **/
3862 static void igb_configure_rx(struct igb_adapter *adapter)
3863 {
3864         int i;
3865
3866         /* set the correct pool for the PF default MAC address in entry 0 */
3867         igb_set_default_mac_filter(adapter);
3868
3869         /* Setup the HW Rx Head and Tail Descriptor Pointers and
3870          * the Base and Length of the Rx Descriptor Ring
3871          */
3872         for (i = 0; i < adapter->num_rx_queues; i++) {
3873                 struct igb_ring *rx_ring = adapter->rx_ring[i];
3874
3875                 igb_set_rx_buffer_len(adapter, rx_ring);
3876                 igb_configure_rx_ring(adapter, rx_ring);
3877         }
3878 }
3879
3880 /**
3881  *  igb_free_tx_resources - Free Tx Resources per Queue
3882  *  @tx_ring: Tx descriptor ring for a specific queue
3883  *
3884  *  Free all transmit software resources
3885  **/
3886 void igb_free_tx_resources(struct igb_ring *tx_ring)
3887 {
3888         igb_clean_tx_ring(tx_ring);
3889
3890         vfree(tx_ring->tx_buffer_info);
3891         tx_ring->tx_buffer_info = NULL;
3892
3893         /* if not set, then don't free */
3894         if (!tx_ring->desc)
3895                 return;
3896
3897         dma_free_coherent(tx_ring->dev, tx_ring->size,
3898                           tx_ring->desc, tx_ring->dma);
3899
3900         tx_ring->desc = NULL;
3901 }
3902
3903 /**
3904  *  igb_free_all_tx_resources - Free Tx Resources for All Queues
3905  *  @adapter: board private structure
3906  *
3907  *  Free all transmit software resources
3908  **/
3909 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
3910 {
3911         int i;
3912
3913         for (i = 0; i < adapter->num_tx_queues; i++)
3914                 if (adapter->tx_ring[i])
3915                         igb_free_tx_resources(adapter->tx_ring[i]);
3916 }
3917
3918 /**
3919  *  igb_clean_tx_ring - Free Tx Buffers
3920  *  @tx_ring: ring to be cleaned
3921  **/
3922 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
3923 {
3924         u16 i = tx_ring->next_to_clean;
3925         struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
3926
3927         while (i != tx_ring->next_to_use) {
3928                 union e1000_adv_tx_desc *eop_desc, *tx_desc;
3929
3930                 /* Free all the Tx ring sk_buffs */
3931                 dev_kfree_skb_any(tx_buffer->skb);
3932
3933                 /* unmap skb header data */
3934                 dma_unmap_single(tx_ring->dev,
3935                                  dma_unmap_addr(tx_buffer, dma),
3936                                  dma_unmap_len(tx_buffer, len),
3937                                  DMA_TO_DEVICE);
3938
3939                 /* check for eop_desc to determine the end of the packet */
3940                 eop_desc = tx_buffer->next_to_watch;
3941                 tx_desc = IGB_TX_DESC(tx_ring, i);
3942
3943                 /* unmap remaining buffers */
3944                 while (tx_desc != eop_desc) {
3945                         tx_buffer++;
3946                         tx_desc++;
3947                         i++;
3948                         if (unlikely(i == tx_ring->count)) {
3949                                 i = 0;
3950                                 tx_buffer = tx_ring->tx_buffer_info;
3951                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
3952                         }
3953
3954                         /* unmap any remaining paged data */
3955                         if (dma_unmap_len(tx_buffer, len))
3956                                 dma_unmap_page(tx_ring->dev,
3957                                                dma_unmap_addr(tx_buffer, dma),
3958                                                dma_unmap_len(tx_buffer, len),
3959                                                DMA_TO_DEVICE);
3960                 }
3961
3962                 /* move us one more past the eop_desc for start of next pkt */
3963                 tx_buffer++;
3964                 i++;
3965                 if (unlikely(i == tx_ring->count)) {
3966                         i = 0;
3967                         tx_buffer = tx_ring->tx_buffer_info;
3968                 }
3969         }
3970
3971         /* reset BQL for queue */
3972         netdev_tx_reset_queue(txring_txq(tx_ring));
3973
3974         /* reset next_to_use and next_to_clean */
3975         tx_ring->next_to_use = 0;
3976         tx_ring->next_to_clean = 0;
3977 }
3978
3979 /**
3980  *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
3981  *  @adapter: board private structure
3982  **/
3983 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
3984 {
3985         int i;
3986
3987         for (i = 0; i < adapter->num_tx_queues; i++)
3988                 if (adapter->tx_ring[i])
3989                         igb_clean_tx_ring(adapter->tx_ring[i]);
3990 }
3991
3992 /**
3993  *  igb_free_rx_resources - Free Rx Resources
3994  *  @rx_ring: ring to clean the resources from
3995  *
3996  *  Free all receive software resources
3997  **/
3998 void igb_free_rx_resources(struct igb_ring *rx_ring)
3999 {
4000         igb_clean_rx_ring(rx_ring);
4001
4002         vfree(rx_ring->rx_buffer_info);
4003         rx_ring->rx_buffer_info = NULL;
4004
4005         /* if not set, then don't free */
4006         if (!rx_ring->desc)
4007                 return;
4008
4009         dma_free_coherent(rx_ring->dev, rx_ring->size,
4010                           rx_ring->desc, rx_ring->dma);
4011
4012         rx_ring->desc = NULL;
4013 }
4014
4015 /**
4016  *  igb_free_all_rx_resources - Free Rx Resources for All Queues
4017  *  @adapter: board private structure
4018  *
4019  *  Free all receive software resources
4020  **/
4021 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4022 {
4023         int i;
4024
4025         for (i = 0; i < adapter->num_rx_queues; i++)
4026                 if (adapter->rx_ring[i])
4027                         igb_free_rx_resources(adapter->rx_ring[i]);
4028 }
4029
4030 /**
4031  *  igb_clean_rx_ring - Free Rx Buffers per Queue
4032  *  @rx_ring: ring to free buffers from
4033  **/
4034 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4035 {
4036         u16 i = rx_ring->next_to_clean;
4037
4038         if (rx_ring->skb)
4039                 dev_kfree_skb(rx_ring->skb);
4040         rx_ring->skb = NULL;
4041
4042         /* Free all the Rx ring sk_buffs */
4043         while (i != rx_ring->next_to_alloc) {
4044                 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4045
4046                 /* Invalidate cache lines that may have been written to by
4047                  * device so that we avoid corrupting memory.
4048                  */
4049                 dma_sync_single_range_for_cpu(rx_ring->dev,
4050                                               buffer_info->dma,
4051                                               buffer_info->page_offset,
4052                                               igb_rx_bufsz(rx_ring),
4053                                               DMA_FROM_DEVICE);
4054
4055                 /* free resources associated with mapping */
4056                 dma_unmap_page_attrs(rx_ring->dev,
4057                                      buffer_info->dma,
4058                                      igb_rx_pg_size(rx_ring),
4059                                      DMA_FROM_DEVICE,
4060                                      IGB_RX_DMA_ATTR);
4061                 __page_frag_cache_drain(buffer_info->page,
4062                                         buffer_info->pagecnt_bias);
4063
4064                 i++;
4065                 if (i == rx_ring->count)
4066                         i = 0;
4067         }
4068
4069         rx_ring->next_to_alloc = 0;
4070         rx_ring->next_to_clean = 0;
4071         rx_ring->next_to_use = 0;
4072 }
4073
4074 /**
4075  *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
4076  *  @adapter: board private structure
4077  **/
4078 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4079 {
4080         int i;
4081
4082         for (i = 0; i < adapter->num_rx_queues; i++)
4083                 if (adapter->rx_ring[i])
4084                         igb_clean_rx_ring(adapter->rx_ring[i]);
4085 }
4086
4087 /**
4088  *  igb_set_mac - Change the Ethernet Address of the NIC
4089  *  @netdev: network interface device structure
4090  *  @p: pointer to an address structure
4091  *
4092  *  Returns 0 on success, negative on failure
4093  **/
4094 static int igb_set_mac(struct net_device *netdev, void *p)
4095 {
4096         struct igb_adapter *adapter = netdev_priv(netdev);
4097         struct e1000_hw *hw = &adapter->hw;
4098         struct sockaddr *addr = p;
4099
4100         if (!is_valid_ether_addr(addr->sa_data))
4101                 return -EADDRNOTAVAIL;
4102
4103         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4104         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4105
4106         /* set the correct pool for the new PF MAC address in entry 0 */
4107         igb_set_default_mac_filter(adapter);
4108
4109         return 0;
4110 }
4111
4112 /**
4113  *  igb_write_mc_addr_list - write multicast addresses to MTA
4114  *  @netdev: network interface device structure
4115  *
4116  *  Writes multicast address list to the MTA hash table.
4117  *  Returns: -ENOMEM on failure
4118  *           0 on no addresses written
4119  *           X on writing X addresses to MTA
4120  **/
4121 static int igb_write_mc_addr_list(struct net_device *netdev)
4122 {
4123         struct igb_adapter *adapter = netdev_priv(netdev);
4124         struct e1000_hw *hw = &adapter->hw;
4125         struct netdev_hw_addr *ha;
4126         u8  *mta_list;
4127         int i;
4128
4129         if (netdev_mc_empty(netdev)) {
4130                 /* nothing to program, so clear mc list */
4131                 igb_update_mc_addr_list(hw, NULL, 0);
4132                 igb_restore_vf_multicasts(adapter);
4133                 return 0;
4134         }
4135
4136         mta_list = kzalloc(netdev_mc_count(netdev) * 6, GFP_ATOMIC);
4137         if (!mta_list)
4138                 return -ENOMEM;
4139
4140         /* The shared function expects a packed array of only addresses. */
4141         i = 0;
4142         netdev_for_each_mc_addr(ha, netdev)
4143                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
4144
4145         igb_update_mc_addr_list(hw, mta_list, i);
4146         kfree(mta_list);
4147
4148         return netdev_mc_count(netdev);
4149 }
4150
4151 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
4152 {
4153         struct e1000_hw *hw = &adapter->hw;
4154         u32 i, pf_id;
4155
4156         switch (hw->mac.type) {
4157         case e1000_i210:
4158         case e1000_i211:
4159         case e1000_i350:
4160                 /* VLAN filtering needed for VLAN prio filter */
4161                 if (adapter->netdev->features & NETIF_F_NTUPLE)
4162                         break;
4163                 /* fall through */
4164         case e1000_82576:
4165         case e1000_82580:
4166         case e1000_i354:
4167                 /* VLAN filtering needed for pool filtering */
4168                 if (adapter->vfs_allocated_count)
4169                         break;
4170                 /* fall through */
4171         default:
4172                 return 1;
4173         }
4174
4175         /* We are already in VLAN promisc, nothing to do */
4176         if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
4177                 return 0;
4178
4179         if (!adapter->vfs_allocated_count)
4180                 goto set_vfta;
4181
4182         /* Add PF to all active pools */
4183         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4184
4185         for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4186                 u32 vlvf = rd32(E1000_VLVF(i));
4187
4188                 vlvf |= BIT(pf_id);
4189                 wr32(E1000_VLVF(i), vlvf);
4190         }
4191
4192 set_vfta:
4193         /* Set all bits in the VLAN filter table array */
4194         for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
4195                 hw->mac.ops.write_vfta(hw, i, ~0U);
4196
4197         /* Set flag so we don't redo unnecessary work */
4198         adapter->flags |= IGB_FLAG_VLAN_PROMISC;
4199
4200         return 0;
4201 }
4202
4203 #define VFTA_BLOCK_SIZE 8
4204 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
4205 {
4206         struct e1000_hw *hw = &adapter->hw;
4207         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
4208         u32 vid_start = vfta_offset * 32;
4209         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
4210         u32 i, vid, word, bits, pf_id;
4211
4212         /* guarantee that we don't scrub out management VLAN */
4213         vid = adapter->mng_vlan_id;
4214         if (vid >= vid_start && vid < vid_end)
4215                 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4216
4217         if (!adapter->vfs_allocated_count)
4218                 goto set_vfta;
4219
4220         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
4221
4222         for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
4223                 u32 vlvf = rd32(E1000_VLVF(i));
4224
4225                 /* pull VLAN ID from VLVF */
4226                 vid = vlvf & VLAN_VID_MASK;
4227
4228                 /* only concern ourselves with a certain range */
4229                 if (vid < vid_start || vid >= vid_end)
4230                         continue;
4231
4232                 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
4233                         /* record VLAN ID in VFTA */
4234                         vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
4235
4236                         /* if PF is part of this then continue */
4237                         if (test_bit(vid, adapter->active_vlans))
4238                                 continue;
4239                 }
4240
4241                 /* remove PF from the pool */
4242                 bits = ~BIT(pf_id);
4243                 bits &= rd32(E1000_VLVF(i));
4244                 wr32(E1000_VLVF(i), bits);
4245         }
4246
4247 set_vfta:
4248         /* extract values from active_vlans and write back to VFTA */
4249         for (i = VFTA_BLOCK_SIZE; i--;) {
4250                 vid = (vfta_offset + i) * 32;
4251                 word = vid / BITS_PER_LONG;
4252                 bits = vid % BITS_PER_LONG;
4253
4254                 vfta[i] |= adapter->active_vlans[word] >> bits;
4255
4256                 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
4257         }
4258 }
4259
4260 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
4261 {
4262         u32 i;
4263
4264         /* We are not in VLAN promisc, nothing to do */
4265         if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
4266                 return;
4267
4268         /* Set flag so we don't redo unnecessary work */
4269         adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
4270
4271         for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
4272                 igb_scrub_vfta(adapter, i);
4273 }
4274
4275 /**
4276  *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
4277  *  @netdev: network interface device structure
4278  *
4279  *  The set_rx_mode entry point is called whenever the unicast or multicast
4280  *  address lists or the network interface flags are updated.  This routine is
4281  *  responsible for configuring the hardware for proper unicast, multicast,
4282  *  promiscuous mode, and all-multi behavior.
4283  **/
4284 static void igb_set_rx_mode(struct net_device *netdev)
4285 {
4286         struct igb_adapter *adapter = netdev_priv(netdev);
4287         struct e1000_hw *hw = &adapter->hw;
4288         unsigned int vfn = adapter->vfs_allocated_count;
4289         u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
4290         int count;
4291
4292         /* Check for Promiscuous and All Multicast modes */
4293         if (netdev->flags & IFF_PROMISC) {
4294                 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
4295                 vmolr |= E1000_VMOLR_MPME;
4296
4297                 /* enable use of UTA filter to force packets to default pool */
4298                 if (hw->mac.type == e1000_82576)
4299                         vmolr |= E1000_VMOLR_ROPE;
4300         } else {
4301                 if (netdev->flags & IFF_ALLMULTI) {
4302                         rctl |= E1000_RCTL_MPE;
4303                         vmolr |= E1000_VMOLR_MPME;
4304                 } else {
4305                         /* Write addresses to the MTA, if the attempt fails
4306                          * then we should just turn on promiscuous mode so
4307                          * that we can at least receive multicast traffic
4308                          */
4309                         count = igb_write_mc_addr_list(netdev);
4310                         if (count < 0) {
4311                                 rctl |= E1000_RCTL_MPE;
4312                                 vmolr |= E1000_VMOLR_MPME;
4313                         } else if (count) {
4314                                 vmolr |= E1000_VMOLR_ROMPE;
4315                         }
4316                 }
4317         }
4318
4319         /* Write addresses to available RAR registers, if there is not
4320          * sufficient space to store all the addresses then enable
4321          * unicast promiscuous mode
4322          */
4323         if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
4324                 rctl |= E1000_RCTL_UPE;
4325                 vmolr |= E1000_VMOLR_ROPE;
4326         }
4327
4328         /* enable VLAN filtering by default */
4329         rctl |= E1000_RCTL_VFE;
4330
4331         /* disable VLAN filtering for modes that require it */
4332         if ((netdev->flags & IFF_PROMISC) ||
4333             (netdev->features & NETIF_F_RXALL)) {
4334                 /* if we fail to set all rules then just clear VFE */
4335                 if (igb_vlan_promisc_enable(adapter))
4336                         rctl &= ~E1000_RCTL_VFE;
4337         } else {
4338                 igb_vlan_promisc_disable(adapter);
4339         }
4340
4341         /* update state of unicast, multicast, and VLAN filtering modes */
4342         rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
4343                                      E1000_RCTL_VFE);
4344         wr32(E1000_RCTL, rctl);
4345
4346 #if (PAGE_SIZE < 8192)
4347         if (!adapter->vfs_allocated_count) {
4348                 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4349                         rlpml = IGB_MAX_FRAME_BUILD_SKB;
4350         }
4351 #endif
4352         wr32(E1000_RLPML, rlpml);
4353
4354         /* In order to support SR-IOV and eventually VMDq it is necessary to set
4355          * the VMOLR to enable the appropriate modes.  Without this workaround
4356          * we will have issues with VLAN tag stripping not being done for frames
4357          * that are only arriving because we are the default pool
4358          */
4359         if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
4360                 return;
4361
4362         /* set UTA to appropriate mode */
4363         igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
4364
4365         vmolr |= rd32(E1000_VMOLR(vfn)) &
4366                  ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
4367
4368         /* enable Rx jumbo frames, restrict as needed to support build_skb */
4369         vmolr &= ~E1000_VMOLR_RLPML_MASK;
4370 #if (PAGE_SIZE < 8192)
4371         if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4372                 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
4373         else
4374 #endif
4375                 vmolr |= MAX_JUMBO_FRAME_SIZE;
4376         vmolr |= E1000_VMOLR_LPE;
4377
4378         wr32(E1000_VMOLR(vfn), vmolr);
4379
4380         igb_restore_vf_multicasts(adapter);
4381 }
4382
4383 static void igb_check_wvbr(struct igb_adapter *adapter)
4384 {
4385         struct e1000_hw *hw = &adapter->hw;
4386         u32 wvbr = 0;
4387
4388         switch (hw->mac.type) {
4389         case e1000_82576:
4390         case e1000_i350:
4391                 wvbr = rd32(E1000_WVBR);
4392                 if (!wvbr)
4393                         return;
4394                 break;
4395         default:
4396                 break;
4397         }
4398
4399         adapter->wvbr |= wvbr;
4400 }
4401
4402 #define IGB_STAGGERED_QUEUE_OFFSET 8
4403
4404 static void igb_spoof_check(struct igb_adapter *adapter)
4405 {
4406         int j;
4407
4408         if (!adapter->wvbr)
4409                 return;
4410
4411         for (j = 0; j < adapter->vfs_allocated_count; j++) {
4412                 if (adapter->wvbr & BIT(j) ||
4413                     adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
4414                         dev_warn(&adapter->pdev->dev,
4415                                 "Spoof event(s) detected on VF %d\n", j);
4416                         adapter->wvbr &=
4417                                 ~(BIT(j) |
4418                                   BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
4419                 }
4420         }
4421 }
4422
4423 /* Need to wait a few seconds after link up to get diagnostic information from
4424  * the phy
4425  */
4426 static void igb_update_phy_info(unsigned long data)
4427 {
4428         struct igb_adapter *adapter = (struct igb_adapter *) data;
4429         igb_get_phy_info(&adapter->hw);
4430 }
4431
4432 /**
4433  *  igb_has_link - check shared code for link and determine up/down
4434  *  @adapter: pointer to driver private info
4435  **/
4436 bool igb_has_link(struct igb_adapter *adapter)
4437 {
4438         struct e1000_hw *hw = &adapter->hw;
4439         bool link_active = false;
4440
4441         /* get_link_status is set on LSC (link status) interrupt or
4442          * rx sequence error interrupt.  get_link_status will stay
4443          * false until the e1000_check_for_link establishes link
4444          * for copper adapters ONLY
4445          */
4446         switch (hw->phy.media_type) {
4447         case e1000_media_type_copper:
4448                 if (!hw->mac.get_link_status)
4449                         return true;
4450         case e1000_media_type_internal_serdes:
4451                 hw->mac.ops.check_for_link(hw);
4452                 link_active = !hw->mac.get_link_status;
4453                 break;
4454         default:
4455         case e1000_media_type_unknown:
4456                 break;
4457         }
4458
4459         if (((hw->mac.type == e1000_i210) ||
4460              (hw->mac.type == e1000_i211)) &&
4461              (hw->phy.id == I210_I_PHY_ID)) {
4462                 if (!netif_carrier_ok(adapter->netdev)) {
4463                         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4464                 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
4465                         adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
4466                         adapter->link_check_timeout = jiffies;
4467                 }
4468         }
4469
4470         return link_active;
4471 }
4472
4473 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
4474 {
4475         bool ret = false;
4476         u32 ctrl_ext, thstat;
4477
4478         /* check for thermal sensor event on i350 copper only */
4479         if (hw->mac.type == e1000_i350) {
4480                 thstat = rd32(E1000_THSTAT);
4481                 ctrl_ext = rd32(E1000_CTRL_EXT);
4482
4483                 if ((hw->phy.media_type == e1000_media_type_copper) &&
4484                     !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
4485                         ret = !!(thstat & event);
4486         }
4487
4488         return ret;
4489 }
4490
4491 /**
4492  *  igb_check_lvmmc - check for malformed packets received
4493  *  and indicated in LVMMC register
4494  *  @adapter: pointer to adapter
4495  **/
4496 static void igb_check_lvmmc(struct igb_adapter *adapter)
4497 {
4498         struct e1000_hw *hw = &adapter->hw;
4499         u32 lvmmc;
4500
4501         lvmmc = rd32(E1000_LVMMC);
4502         if (lvmmc) {
4503                 if (unlikely(net_ratelimit())) {
4504                         netdev_warn(adapter->netdev,
4505                                     "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
4506                                     lvmmc);
4507                 }
4508         }
4509 }
4510
4511 /**
4512  *  igb_watchdog - Timer Call-back
4513  *  @data: pointer to adapter cast into an unsigned long
4514  **/
4515 static void igb_watchdog(unsigned long data)
4516 {
4517         struct igb_adapter *adapter = (struct igb_adapter *)data;
4518         /* Do the rest outside of interrupt context */
4519         schedule_work(&adapter->watchdog_task);
4520 }
4521
4522 static void igb_watchdog_task(struct work_struct *work)
4523 {
4524         struct igb_adapter *adapter = container_of(work,
4525                                                    struct igb_adapter,
4526                                                    watchdog_task);
4527         struct e1000_hw *hw = &adapter->hw;
4528         struct e1000_phy_info *phy = &hw->phy;
4529         struct net_device *netdev = adapter->netdev;
4530         u32 link;
4531         int i;
4532         u32 connsw;
4533         u16 phy_data, retry_count = 20;
4534
4535         link = igb_has_link(adapter);
4536
4537         if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
4538                 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
4539                         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
4540                 else
4541                         link = false;
4542         }
4543
4544         /* Force link down if we have fiber to swap to */
4545         if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4546                 if (hw->phy.media_type == e1000_media_type_copper) {
4547                         connsw = rd32(E1000_CONNSW);
4548                         if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
4549                                 link = 0;
4550                 }
4551         }
4552         if (link) {
4553                 /* Perform a reset if the media type changed. */
4554                 if (hw->dev_spec._82575.media_changed) {
4555                         hw->dev_spec._82575.media_changed = false;
4556                         adapter->flags |= IGB_FLAG_MEDIA_RESET;
4557                         igb_reset(adapter);
4558                 }
4559                 /* Cancel scheduled suspend requests. */
4560                 pm_runtime_resume(netdev->dev.parent);
4561
4562                 if (!netif_carrier_ok(netdev)) {
4563                         u32 ctrl;
4564
4565                         hw->mac.ops.get_speed_and_duplex(hw,
4566                                                          &adapter->link_speed,
4567                                                          &adapter->link_duplex);
4568
4569                         ctrl = rd32(E1000_CTRL);
4570                         /* Links status message must follow this format */
4571                         netdev_info(netdev,
4572                                "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
4573                                netdev->name,
4574                                adapter->link_speed,
4575                                adapter->link_duplex == FULL_DUPLEX ?
4576                                "Full" : "Half",
4577                                (ctrl & E1000_CTRL_TFCE) &&
4578                                (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
4579                                (ctrl & E1000_CTRL_RFCE) ?  "RX" :
4580                                (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
4581
4582                         /* disable EEE if enabled */
4583                         if ((adapter->flags & IGB_FLAG_EEE) &&
4584                                 (adapter->link_duplex == HALF_DUPLEX)) {
4585                                 dev_info(&adapter->pdev->dev,
4586                                 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
4587                                 adapter->hw.dev_spec._82575.eee_disable = true;
4588                                 adapter->flags &= ~IGB_FLAG_EEE;
4589                         }
4590
4591                         /* check if SmartSpeed worked */
4592                         igb_check_downshift(hw);
4593                         if (phy->speed_downgraded)
4594                                 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
4595
4596                         /* check for thermal sensor event */
4597                         if (igb_thermal_sensor_event(hw,
4598                             E1000_THSTAT_LINK_THROTTLE))
4599                                 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
4600
4601                         /* adjust timeout factor according to speed/duplex */
4602                         adapter->tx_timeout_factor = 1;
4603                         switch (adapter->link_speed) {
4604                         case SPEED_10:
4605                                 adapter->tx_timeout_factor = 14;
4606                                 break;
4607                         case SPEED_100:
4608                                 /* maybe add some timeout factor ? */
4609                                 break;
4610                         }
4611
4612                         if (adapter->link_speed != SPEED_1000)
4613                                 goto no_wait;
4614
4615                         /* wait for Remote receiver status OK */
4616 retry_read_status:
4617                         if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
4618                                               &phy_data)) {
4619                                 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
4620                                     retry_count) {
4621                                         msleep(100);
4622                                         retry_count--;
4623                                         goto retry_read_status;
4624                                 } else if (!retry_count) {
4625                                         dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
4626                                 }
4627                         } else {
4628                                 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
4629                         }
4630 no_wait:
4631                         netif_carrier_on(netdev);
4632
4633                         igb_ping_all_vfs(adapter);
4634                         igb_check_vf_rate_limit(adapter);
4635
4636                         /* link state has changed, schedule phy info update */
4637                         if (!test_bit(__IGB_DOWN, &adapter->state))
4638                                 mod_timer(&adapter->phy_info_timer,
4639                                           round_jiffies(jiffies + 2 * HZ));
4640                 }
4641         } else {
4642                 if (netif_carrier_ok(netdev)) {
4643                         adapter->link_speed = 0;
4644                         adapter->link_duplex = 0;
4645
4646                         /* check for thermal sensor event */
4647                         if (igb_thermal_sensor_event(hw,
4648                             E1000_THSTAT_PWR_DOWN)) {
4649                                 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
4650                         }
4651
4652                         /* Links status message must follow this format */
4653                         netdev_info(netdev, "igb: %s NIC Link is Down\n",
4654                                netdev->name);
4655                         netif_carrier_off(netdev);
4656
4657                         igb_ping_all_vfs(adapter);
4658
4659                         /* link state has changed, schedule phy info update */
4660                         if (!test_bit(__IGB_DOWN, &adapter->state))
4661                                 mod_timer(&adapter->phy_info_timer,
4662                                           round_jiffies(jiffies + 2 * HZ));
4663
4664                         /* link is down, time to check for alternate media */
4665                         if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
4666                                 igb_check_swap_media(adapter);
4667                                 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4668                                         schedule_work(&adapter->reset_task);
4669                                         /* return immediately */
4670                                         return;
4671                                 }
4672                         }
4673                         pm_schedule_suspend(netdev->dev.parent,
4674                                             MSEC_PER_SEC * 5);
4675
4676                 /* also check for alternate media here */
4677                 } else if (!netif_carrier_ok(netdev) &&
4678                            (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
4679                         igb_check_swap_media(adapter);
4680                         if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
4681                                 schedule_work(&adapter->reset_task);
4682                                 /* return immediately */
4683                                 return;
4684                         }
4685                 }
4686         }
4687
4688         spin_lock(&adapter->stats64_lock);
4689         igb_update_stats(adapter);
4690         spin_unlock(&adapter->stats64_lock);
4691
4692         for (i = 0; i < adapter->num_tx_queues; i++) {
4693                 struct igb_ring *tx_ring = adapter->tx_ring[i];
4694                 if (!netif_carrier_ok(netdev)) {
4695                         /* We've lost link, so the controller stops DMA,
4696                          * but we've got queued Tx work that's never going
4697                          * to get done, so reset controller to flush Tx.
4698                          * (Do the reset outside of interrupt context).
4699                          */
4700                         if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
4701                                 adapter->tx_timeout_count++;
4702                                 schedule_work(&adapter->reset_task);
4703                                 /* return immediately since reset is imminent */
4704                                 return;
4705                         }
4706                 }
4707
4708                 /* Force detection of hung controller every watchdog period */
4709                 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
4710         }
4711
4712         /* Cause software interrupt to ensure Rx ring is cleaned */
4713         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
4714                 u32 eics = 0;
4715
4716                 for (i = 0; i < adapter->num_q_vectors; i++)
4717                         eics |= adapter->q_vector[i]->eims_value;
4718                 wr32(E1000_EICS, eics);
4719         } else {
4720                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
4721         }
4722
4723         igb_spoof_check(adapter);
4724         igb_ptp_rx_hang(adapter);
4725         igb_ptp_tx_hang(adapter);
4726
4727         /* Check LVMMC register on i350/i354 only */
4728         if ((adapter->hw.mac.type == e1000_i350) ||
4729             (adapter->hw.mac.type == e1000_i354))
4730                 igb_check_lvmmc(adapter);
4731
4732         /* Reset the timer */
4733         if (!test_bit(__IGB_DOWN, &adapter->state)) {
4734                 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
4735                         mod_timer(&adapter->watchdog_timer,
4736                                   round_jiffies(jiffies +  HZ));
4737                 else
4738                         mod_timer(&adapter->watchdog_timer,
4739                                   round_jiffies(jiffies + 2 * HZ));
4740         }
4741 }
4742
4743 enum latency_range {
4744         lowest_latency = 0,
4745         low_latency = 1,
4746         bulk_latency = 2,
4747         latency_invalid = 255
4748 };
4749
4750 /**
4751  *  igb_update_ring_itr - update the dynamic ITR value based on packet size
4752  *  @q_vector: pointer to q_vector
4753  *
4754  *  Stores a new ITR value based on strictly on packet size.  This
4755  *  algorithm is less sophisticated than that used in igb_update_itr,
4756  *  due to the difficulty of synchronizing statistics across multiple
4757  *  receive rings.  The divisors and thresholds used by this function
4758  *  were determined based on theoretical maximum wire speed and testing
4759  *  data, in order to minimize response time while increasing bulk
4760  *  throughput.
4761  *  This functionality is controlled by ethtool's coalescing settings.
4762  *  NOTE:  This function is called only when operating in a multiqueue
4763  *         receive environment.
4764  **/
4765 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
4766 {
4767         int new_val = q_vector->itr_val;
4768         int avg_wire_size = 0;
4769         struct igb_adapter *adapter = q_vector->adapter;
4770         unsigned int packets;
4771
4772         /* For non-gigabit speeds, just fix the interrupt rate at 4000
4773          * ints/sec - ITR timer value of 120 ticks.
4774          */
4775         if (adapter->link_speed != SPEED_1000) {
4776                 new_val = IGB_4K_ITR;
4777                 goto set_itr_val;
4778         }
4779
4780         packets = q_vector->rx.total_packets;
4781         if (packets)
4782                 avg_wire_size = q_vector->rx.total_bytes / packets;
4783
4784         packets = q_vector->tx.total_packets;
4785         if (packets)
4786                 avg_wire_size = max_t(u32, avg_wire_size,
4787                                       q_vector->tx.total_bytes / packets);
4788
4789         /* if avg_wire_size isn't set no work was done */
4790         if (!avg_wire_size)
4791                 goto clear_counts;
4792
4793         /* Add 24 bytes to size to account for CRC, preamble, and gap */
4794         avg_wire_size += 24;
4795
4796         /* Don't starve jumbo frames */
4797         avg_wire_size = min(avg_wire_size, 3000);
4798
4799         /* Give a little boost to mid-size frames */
4800         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
4801                 new_val = avg_wire_size / 3;
4802         else
4803                 new_val = avg_wire_size / 2;
4804
4805         /* conservative mode (itr 3) eliminates the lowest_latency setting */
4806         if (new_val < IGB_20K_ITR &&
4807             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4808              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4809                 new_val = IGB_20K_ITR;
4810
4811 set_itr_val:
4812         if (new_val != q_vector->itr_val) {
4813                 q_vector->itr_val = new_val;
4814                 q_vector->set_itr = 1;
4815         }
4816 clear_counts:
4817         q_vector->rx.total_bytes = 0;
4818         q_vector->rx.total_packets = 0;
4819         q_vector->tx.total_bytes = 0;
4820         q_vector->tx.total_packets = 0;
4821 }
4822
4823 /**
4824  *  igb_update_itr - update the dynamic ITR value based on statistics
4825  *  @q_vector: pointer to q_vector
4826  *  @ring_container: ring info to update the itr for
4827  *
4828  *  Stores a new ITR value based on packets and byte
4829  *  counts during the last interrupt.  The advantage of per interrupt
4830  *  computation is faster updates and more accurate ITR for the current
4831  *  traffic pattern.  Constants in this function were computed
4832  *  based on theoretical maximum wire speed and thresholds were set based
4833  *  on testing data as well as attempting to minimize response time
4834  *  while increasing bulk throughput.
4835  *  This functionality is controlled by ethtool's coalescing settings.
4836  *  NOTE:  These calculations are only valid when operating in a single-
4837  *         queue environment.
4838  **/
4839 static void igb_update_itr(struct igb_q_vector *q_vector,
4840                            struct igb_ring_container *ring_container)
4841 {
4842         unsigned int packets = ring_container->total_packets;
4843         unsigned int bytes = ring_container->total_bytes;
4844         u8 itrval = ring_container->itr;
4845
4846         /* no packets, exit with status unchanged */
4847         if (packets == 0)
4848                 return;
4849
4850         switch (itrval) {
4851         case lowest_latency:
4852                 /* handle TSO and jumbo frames */
4853                 if (bytes/packets > 8000)
4854                         itrval = bulk_latency;
4855                 else if ((packets < 5) && (bytes > 512))
4856                         itrval = low_latency;
4857                 break;
4858         case low_latency:  /* 50 usec aka 20000 ints/s */
4859                 if (bytes > 10000) {
4860                         /* this if handles the TSO accounting */
4861                         if (bytes/packets > 8000)
4862                                 itrval = bulk_latency;
4863                         else if ((packets < 10) || ((bytes/packets) > 1200))
4864                                 itrval = bulk_latency;
4865                         else if ((packets > 35))
4866                                 itrval = lowest_latency;
4867                 } else if (bytes/packets > 2000) {
4868                         itrval = bulk_latency;
4869                 } else if (packets <= 2 && bytes < 512) {
4870                         itrval = lowest_latency;
4871                 }
4872                 break;
4873         case bulk_latency: /* 250 usec aka 4000 ints/s */
4874                 if (bytes > 25000) {
4875                         if (packets > 35)
4876                                 itrval = low_latency;
4877                 } else if (bytes < 1500) {
4878                         itrval = low_latency;
4879                 }
4880                 break;
4881         }
4882
4883         /* clear work counters since we have the values we need */
4884         ring_container->total_bytes = 0;
4885         ring_container->total_packets = 0;
4886
4887         /* write updated itr to ring container */
4888         ring_container->itr = itrval;
4889 }
4890
4891 static void igb_set_itr(struct igb_q_vector *q_vector)
4892 {
4893         struct igb_adapter *adapter = q_vector->adapter;
4894         u32 new_itr = q_vector->itr_val;
4895         u8 current_itr = 0;
4896
4897         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
4898         if (adapter->link_speed != SPEED_1000) {
4899                 current_itr = 0;
4900                 new_itr = IGB_4K_ITR;
4901                 goto set_itr_now;
4902         }
4903
4904         igb_update_itr(q_vector, &q_vector->tx);
4905         igb_update_itr(q_vector, &q_vector->rx);
4906
4907         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
4908
4909         /* conservative mode (itr 3) eliminates the lowest_latency setting */
4910         if (current_itr == lowest_latency &&
4911             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
4912              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
4913                 current_itr = low_latency;
4914
4915         switch (current_itr) {
4916         /* counts and packets in update_itr are dependent on these numbers */
4917         case lowest_latency:
4918                 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
4919                 break;
4920         case low_latency:
4921                 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
4922                 break;
4923         case bulk_latency:
4924                 new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
4925                 break;
4926         default:
4927                 break;
4928         }
4929
4930 set_itr_now:
4931         if (new_itr != q_vector->itr_val) {
4932                 /* this attempts to bias the interrupt rate towards Bulk
4933                  * by adding intermediate steps when interrupt rate is
4934                  * increasing
4935                  */
4936                 new_itr = new_itr > q_vector->itr_val ?
4937                           max((new_itr * q_vector->itr_val) /
4938                           (new_itr + (q_vector->itr_val >> 2)),
4939                           new_itr) : new_itr;
4940                 /* Don't write the value here; it resets the adapter's
4941                  * internal timer, and causes us to delay far longer than
4942                  * we should between interrupts.  Instead, we write the ITR
4943                  * value at the beginning of the next interrupt so the timing
4944                  * ends up being correct.
4945                  */
4946                 q_vector->itr_val = new_itr;
4947                 q_vector->set_itr = 1;
4948         }
4949 }
4950
4951 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring, u32 vlan_macip_lens,
4952                             u32 type_tucmd, u32 mss_l4len_idx)
4953 {
4954         struct e1000_adv_tx_context_desc *context_desc;
4955         u16 i = tx_ring->next_to_use;
4956
4957         context_desc = IGB_TX_CTXTDESC(tx_ring, i);
4958
4959         i++;
4960         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
4961
4962         /* set bits to identify this as an advanced context descriptor */
4963         type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
4964
4965         /* For 82575, context index must be unique per ring. */
4966         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
4967                 mss_l4len_idx |= tx_ring->reg_idx << 4;
4968
4969         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
4970         context_desc->seqnum_seed       = 0;
4971         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
4972         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
4973 }
4974
4975 static int igb_tso(struct igb_ring *tx_ring,
4976                    struct igb_tx_buffer *first,
4977                    u8 *hdr_len)
4978 {
4979         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
4980         struct sk_buff *skb = first->skb;
4981         union {
4982                 struct iphdr *v4;
4983                 struct ipv6hdr *v6;
4984                 unsigned char *hdr;
4985         } ip;
4986         union {
4987                 struct tcphdr *tcp;
4988                 unsigned char *hdr;
4989         } l4;
4990         u32 paylen, l4_offset;
4991         int err;
4992
4993         if (skb->ip_summed != CHECKSUM_PARTIAL)
4994                 return 0;
4995
4996         if (!skb_is_gso(skb))
4997                 return 0;
4998
4999         err = skb_cow_head(skb, 0);
5000         if (err < 0)
5001                 return err;
5002
5003         ip.hdr = skb_network_header(skb);
5004         l4.hdr = skb_checksum_start(skb);
5005
5006         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5007         type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5008
5009         /* initialize outer IP header fields */
5010         if (ip.v4->version == 4) {
5011                 unsigned char *csum_start = skb_checksum_start(skb);
5012                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5013
5014                 /* IP header will have to cancel out any data that
5015                  * is not a part of the outer IP header
5016                  */
5017                 ip.v4->check = csum_fold(csum_partial(trans_start,
5018                                                       csum_start - trans_start,
5019                                                       0));
5020                 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5021
5022                 ip.v4->tot_len = 0;
5023                 first->tx_flags |= IGB_TX_FLAGS_TSO |
5024                                    IGB_TX_FLAGS_CSUM |
5025                                    IGB_TX_FLAGS_IPV4;
5026         } else {
5027                 ip.v6->payload_len = 0;
5028                 first->tx_flags |= IGB_TX_FLAGS_TSO |
5029                                    IGB_TX_FLAGS_CSUM;
5030         }
5031
5032         /* determine offset of inner transport header */
5033         l4_offset = l4.hdr - skb->data;
5034
5035         /* compute length of segmentation header */
5036         *hdr_len = (l4.tcp->doff * 4) + l4_offset;
5037
5038         /* remove payload length from inner checksum */
5039         paylen = skb->len - l4_offset;
5040         csum_replace_by_diff(&l4.tcp->check, htonl(paylen));
5041
5042         /* update gso size and bytecount with header size */
5043         first->gso_segs = skb_shinfo(skb)->gso_segs;
5044         first->bytecount += (first->gso_segs - 1) * *hdr_len;
5045
5046         /* MSS L4LEN IDX */
5047         mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5048         mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5049
5050         /* VLAN MACLEN IPLEN */
5051         vlan_macip_lens = l4.hdr - ip.hdr;
5052         vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5053         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5054
5055         igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, mss_l4len_idx);
5056
5057         return 1;
5058 }
5059
5060 static inline bool igb_ipv6_csum_is_sctp(struct sk_buff *skb)
5061 {
5062         unsigned int offset = 0;
5063
5064         ipv6_find_hdr(skb, &offset, IPPROTO_SCTP, NULL, NULL);
5065
5066         return offset == skb_checksum_start_offset(skb);
5067 }
5068
5069 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5070 {
5071         struct sk_buff *skb = first->skb;
5072         u32 vlan_macip_lens = 0;
5073         u32 type_tucmd = 0;
5074
5075         if (skb->ip_summed != CHECKSUM_PARTIAL) {
5076 csum_failed:
5077                 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN))
5078                         return;
5079                 goto no_csum;
5080         }
5081
5082         switch (skb->csum_offset) {
5083         case offsetof(struct tcphdr, check):
5084                 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5085                 /* fall through */
5086         case offsetof(struct udphdr, check):
5087                 break;
5088         case offsetof(struct sctphdr, checksum):
5089                 /* validate that this is actually an SCTP request */
5090                 if (((first->protocol == htons(ETH_P_IP)) &&
5091                      (ip_hdr(skb)->protocol == IPPROTO_SCTP)) ||
5092                     ((first->protocol == htons(ETH_P_IPV6)) &&
5093                      igb_ipv6_csum_is_sctp(skb))) {
5094                         type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
5095                         break;
5096                 }
5097         default:
5098                 skb_checksum_help(skb);
5099                 goto csum_failed;
5100         }
5101
5102         /* update TX checksum flag */
5103         first->tx_flags |= IGB_TX_FLAGS_CSUM;
5104         vlan_macip_lens = skb_checksum_start_offset(skb) -
5105                           skb_network_offset(skb);
5106 no_csum:
5107         vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
5108         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5109
5110         igb_tx_ctxtdesc(tx_ring, vlan_macip_lens, type_tucmd, 0);
5111 }
5112
5113 #define IGB_SET_FLAG(_input, _flag, _result) \
5114         ((_flag <= _result) ? \
5115          ((u32)(_input & _flag) * (_result / _flag)) : \
5116          ((u32)(_input & _flag) / (_flag / _result)))
5117
5118 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
5119 {
5120         /* set type for advanced descriptor with frame checksum insertion */
5121         u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
5122                        E1000_ADVTXD_DCMD_DEXT |
5123                        E1000_ADVTXD_DCMD_IFCS;
5124
5125         /* set HW vlan bit if vlan is present */
5126         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
5127                                  (E1000_ADVTXD_DCMD_VLE));
5128
5129         /* set segmentation bits for TSO */
5130         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
5131                                  (E1000_ADVTXD_DCMD_TSE));
5132
5133         /* set timestamp bit if present */
5134         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
5135                                  (E1000_ADVTXD_MAC_TSTAMP));
5136
5137         /* insert frame checksum */
5138         cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
5139
5140         return cmd_type;
5141 }
5142
5143 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
5144                                  union e1000_adv_tx_desc *tx_desc,
5145                                  u32 tx_flags, unsigned int paylen)
5146 {
5147         u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
5148
5149         /* 82575 requires a unique index per ring */
5150         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5151                 olinfo_status |= tx_ring->reg_idx << 4;
5152
5153         /* insert L4 checksum */
5154         olinfo_status |= IGB_SET_FLAG(tx_flags,
5155                                       IGB_TX_FLAGS_CSUM,
5156                                       (E1000_TXD_POPTS_TXSM << 8));
5157
5158         /* insert IPv4 checksum */
5159         olinfo_status |= IGB_SET_FLAG(tx_flags,
5160                                       IGB_TX_FLAGS_IPV4,
5161                                       (E1000_TXD_POPTS_IXSM << 8));
5162
5163         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
5164 }
5165
5166 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5167 {
5168         struct net_device *netdev = tx_ring->netdev;
5169
5170         netif_stop_subqueue(netdev, tx_ring->queue_index);
5171
5172         /* Herbert's original patch had:
5173          *  smp_mb__after_netif_stop_queue();
5174          * but since that doesn't exist yet, just open code it.
5175          */
5176         smp_mb();
5177
5178         /* We need to check again in a case another CPU has just
5179          * made room available.
5180          */
5181         if (igb_desc_unused(tx_ring) < size)
5182                 return -EBUSY;
5183
5184         /* A reprieve! */
5185         netif_wake_subqueue(netdev, tx_ring->queue_index);
5186
5187         u64_stats_update_begin(&tx_ring->tx_syncp2);
5188         tx_ring->tx_stats.restart_queue2++;
5189         u64_stats_update_end(&tx_ring->tx_syncp2);
5190
5191         return 0;
5192 }
5193
5194 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
5195 {
5196         if (igb_desc_unused(tx_ring) >= size)
5197                 return 0;
5198         return __igb_maybe_stop_tx(tx_ring, size);
5199 }
5200
5201 static int igb_tx_map(struct igb_ring *tx_ring,
5202                       struct igb_tx_buffer *first,
5203                       const u8 hdr_len)
5204 {
5205         struct sk_buff *skb = first->skb;
5206         struct igb_tx_buffer *tx_buffer;
5207         union e1000_adv_tx_desc *tx_desc;
5208         struct skb_frag_struct *frag;
5209         dma_addr_t dma;
5210         unsigned int data_len, size;
5211         u32 tx_flags = first->tx_flags;
5212         u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
5213         u16 i = tx_ring->next_to_use;
5214
5215         tx_desc = IGB_TX_DESC(tx_ring, i);
5216
5217         igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
5218
5219         size = skb_headlen(skb);
5220         data_len = skb->data_len;
5221
5222         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
5223
5224         tx_buffer = first;
5225
5226         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
5227                 if (dma_mapping_error(tx_ring->dev, dma))
5228                         goto dma_error;
5229
5230                 /* record length, and DMA address */
5231                 dma_unmap_len_set(tx_buffer, len, size);
5232                 dma_unmap_addr_set(tx_buffer, dma, dma);
5233
5234                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
5235
5236                 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
5237                         tx_desc->read.cmd_type_len =
5238                                 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
5239
5240                         i++;
5241                         tx_desc++;
5242                         if (i == tx_ring->count) {
5243                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
5244                                 i = 0;
5245                         }
5246                         tx_desc->read.olinfo_status = 0;
5247
5248                         dma += IGB_MAX_DATA_PER_TXD;
5249                         size -= IGB_MAX_DATA_PER_TXD;
5250
5251                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
5252                 }
5253
5254                 if (likely(!data_len))
5255                         break;
5256
5257                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
5258
5259                 i++;
5260                 tx_desc++;
5261                 if (i == tx_ring->count) {
5262                         tx_desc = IGB_TX_DESC(tx_ring, 0);
5263                         i = 0;
5264                 }
5265                 tx_desc->read.olinfo_status = 0;
5266
5267                 size = skb_frag_size(frag);
5268                 data_len -= size;
5269
5270                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
5271                                        size, DMA_TO_DEVICE);
5272
5273                 tx_buffer = &tx_ring->tx_buffer_info[i];
5274         }
5275
5276         /* write last descriptor with RS and EOP bits */
5277         cmd_type |= size | IGB_TXD_DCMD;
5278         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
5279
5280         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
5281
5282         /* set the timestamp */
5283         first->time_stamp = jiffies;
5284
5285         /* Force memory writes to complete before letting h/w know there
5286          * are new descriptors to fetch.  (Only applicable for weak-ordered
5287          * memory model archs, such as IA-64).
5288          *
5289          * We also need this memory barrier to make certain all of the
5290          * status bits have been updated before next_to_watch is written.
5291          */
5292         wmb();
5293
5294         /* set next_to_watch value indicating a packet is present */
5295         first->next_to_watch = tx_desc;
5296
5297         i++;
5298         if (i == tx_ring->count)
5299                 i = 0;
5300
5301         tx_ring->next_to_use = i;
5302
5303         /* Make sure there is space in the ring for the next send. */
5304         igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
5305
5306         if (netif_xmit_stopped(txring_txq(tx_ring)) || !skb->xmit_more) {
5307                 writel(i, tx_ring->tail);
5308
5309                 /* we need this if more than one processor can write to our tail
5310                  * at a time, it synchronizes IO on IA64/Altix systems
5311                  */
5312                 mmiowb();
5313         }
5314         return 0;
5315
5316 dma_error:
5317         dev_err(tx_ring->dev, "TX DMA map failed\n");
5318         tx_buffer = &tx_ring->tx_buffer_info[i];
5319
5320         /* clear dma mappings for failed tx_buffer_info map */
5321         while (tx_buffer != first) {
5322                 if (dma_unmap_len(tx_buffer, len))
5323                         dma_unmap_page(tx_ring->dev,
5324                                        dma_unmap_addr(tx_buffer, dma),
5325                                        dma_unmap_len(tx_buffer, len),
5326                                        DMA_TO_DEVICE);
5327                 dma_unmap_len_set(tx_buffer, len, 0);
5328
5329                 if (i--)
5330                         i += tx_ring->count;
5331                 tx_buffer = &tx_ring->tx_buffer_info[i];
5332         }
5333
5334         if (dma_unmap_len(tx_buffer, len))
5335                 dma_unmap_single(tx_ring->dev,
5336                                  dma_unmap_addr(tx_buffer, dma),
5337                                  dma_unmap_len(tx_buffer, len),
5338                                  DMA_TO_DEVICE);
5339         dma_unmap_len_set(tx_buffer, len, 0);
5340
5341         dev_kfree_skb_any(tx_buffer->skb);
5342         tx_buffer->skb = NULL;
5343
5344         tx_ring->next_to_use = i;
5345
5346         return -1;
5347 }
5348
5349 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
5350                                 struct igb_ring *tx_ring)
5351 {
5352         struct igb_tx_buffer *first;
5353         int tso;
5354         u32 tx_flags = 0;
5355         unsigned short f;
5356         u16 count = TXD_USE_COUNT(skb_headlen(skb));
5357         __be16 protocol = vlan_get_protocol(skb);
5358         u8 hdr_len = 0;
5359
5360         /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
5361          *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
5362          *       + 2 desc gap to keep tail from touching head,
5363          *       + 1 desc for context descriptor,
5364          * otherwise try next time
5365          */
5366         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
5367                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
5368
5369         if (igb_maybe_stop_tx(tx_ring, count + 3)) {
5370                 /* this is a hard error */
5371                 return NETDEV_TX_BUSY;
5372         }
5373
5374         /* record the location of the first descriptor for this packet */
5375         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
5376         first->skb = skb;
5377         first->bytecount = skb->len;
5378         first->gso_segs = 1;
5379
5380         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
5381                 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5382
5383                 if (adapter->tstamp_config.tx_type & HWTSTAMP_TX_ON &&
5384                     !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
5385                                            &adapter->state)) {
5386                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
5387                         tx_flags |= IGB_TX_FLAGS_TSTAMP;
5388
5389                         adapter->ptp_tx_skb = skb_get(skb);
5390                         adapter->ptp_tx_start = jiffies;
5391                         if (adapter->hw.mac.type == e1000_82576)
5392                                 schedule_work(&adapter->ptp_tx_work);
5393                 } else {
5394                         adapter->tx_hwtstamp_skipped++;
5395                 }
5396         }
5397
5398         skb_tx_timestamp(skb);
5399
5400         if (skb_vlan_tag_present(skb)) {
5401                 tx_flags |= IGB_TX_FLAGS_VLAN;
5402                 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
5403         }
5404
5405         /* record initial flags and protocol */
5406         first->tx_flags = tx_flags;
5407         first->protocol = protocol;
5408
5409         tso = igb_tso(tx_ring, first, &hdr_len);
5410         if (tso < 0)
5411                 goto out_drop;
5412         else if (!tso)
5413                 igb_tx_csum(tx_ring, first);
5414
5415         if (igb_tx_map(tx_ring, first, hdr_len))
5416                 goto cleanup_tx_tstamp;
5417
5418         return NETDEV_TX_OK;
5419
5420 out_drop:
5421         dev_kfree_skb_any(first->skb);
5422         first->skb = NULL;
5423 cleanup_tx_tstamp:
5424         if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
5425                 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
5426
5427                 dev_kfree_skb_any(adapter->ptp_tx_skb);
5428                 adapter->ptp_tx_skb = NULL;
5429                 if (adapter->hw.mac.type == e1000_82576)
5430                         cancel_work_sync(&adapter->ptp_tx_work);
5431                 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
5432         }
5433
5434         return NETDEV_TX_OK;
5435 }
5436
5437 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
5438                                                     struct sk_buff *skb)
5439 {
5440         unsigned int r_idx = skb->queue_mapping;
5441
5442         if (r_idx >= adapter->num_tx_queues)
5443                 r_idx = r_idx % adapter->num_tx_queues;
5444
5445         return adapter->tx_ring[r_idx];
5446 }
5447
5448 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
5449                                   struct net_device *netdev)
5450 {
5451         struct igb_adapter *adapter = netdev_priv(netdev);
5452
5453         /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
5454          * in order to meet this minimum size requirement.
5455          */
5456         if (skb_put_padto(skb, 17))
5457                 return NETDEV_TX_OK;
5458
5459         return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
5460 }
5461
5462 /**
5463  *  igb_tx_timeout - Respond to a Tx Hang
5464  *  @netdev: network interface device structure
5465  **/
5466 static void igb_tx_timeout(struct net_device *netdev)
5467 {
5468         struct igb_adapter *adapter = netdev_priv(netdev);
5469         struct e1000_hw *hw = &adapter->hw;
5470
5471         /* Do the reset outside of interrupt context */
5472         adapter->tx_timeout_count++;
5473
5474         if (hw->mac.type >= e1000_82580)
5475                 hw->dev_spec._82575.global_device_reset = true;
5476
5477         schedule_work(&adapter->reset_task);
5478         wr32(E1000_EICS,
5479              (adapter->eims_enable_mask & ~adapter->eims_other));
5480 }
5481
5482 static void igb_reset_task(struct work_struct *work)
5483 {
5484         struct igb_adapter *adapter;
5485         adapter = container_of(work, struct igb_adapter, reset_task);
5486
5487         igb_dump(adapter);
5488         netdev_err(adapter->netdev, "Reset adapter\n");
5489         igb_reinit_locked(adapter);
5490 }
5491
5492 /**
5493  *  igb_get_stats64 - Get System Network Statistics
5494  *  @netdev: network interface device structure
5495  *  @stats: rtnl_link_stats64 pointer
5496  **/
5497 static void igb_get_stats64(struct net_device *netdev,
5498                             struct rtnl_link_stats64 *stats)
5499 {
5500         struct igb_adapter *adapter = netdev_priv(netdev);
5501
5502         spin_lock(&adapter->stats64_lock);
5503         igb_update_stats(adapter);
5504         memcpy(stats, &adapter->stats64, sizeof(*stats));
5505         spin_unlock(&adapter->stats64_lock);
5506 }
5507
5508 /**
5509  *  igb_change_mtu - Change the Maximum Transfer Unit
5510  *  @netdev: network interface device structure
5511  *  @new_mtu: new value for maximum frame size
5512  *
5513  *  Returns 0 on success, negative on failure
5514  **/
5515 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
5516 {
5517         struct igb_adapter *adapter = netdev_priv(netdev);
5518         struct pci_dev *pdev = adapter->pdev;
5519         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
5520
5521         /* adjust max frame to be at least the size of a standard frame */
5522         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
5523                 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
5524
5525         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
5526                 usleep_range(1000, 2000);
5527
5528         /* igb_down has a dependency on max_frame_size */
5529         adapter->max_frame_size = max_frame;
5530
5531         if (netif_running(netdev))
5532                 igb_down(adapter);
5533
5534         dev_info(&pdev->dev, "changing MTU from %d to %d\n",
5535                  netdev->mtu, new_mtu);
5536         netdev->mtu = new_mtu;
5537
5538         if (netif_running(netdev))
5539                 igb_up(adapter);
5540         else
5541                 igb_reset(adapter);
5542
5543         clear_bit(__IGB_RESETTING, &adapter->state);
5544
5545         return 0;
5546 }
5547
5548 /**
5549  *  igb_update_stats - Update the board statistics counters
5550  *  @adapter: board private structure
5551  **/
5552 void igb_update_stats(struct igb_adapter *adapter)
5553 {
5554         struct rtnl_link_stats64 *net_stats = &adapter->stats64;
5555         struct e1000_hw *hw = &adapter->hw;
5556         struct pci_dev *pdev = adapter->pdev;
5557         u32 reg, mpc;
5558         int i;
5559         u64 bytes, packets;
5560         unsigned int start;
5561         u64 _bytes, _packets;
5562
5563         /* Prevent stats update while adapter is being reset, or if the pci
5564          * connection is down.
5565          */
5566         if (adapter->link_speed == 0)
5567                 return;
5568         if (pci_channel_offline(pdev))
5569                 return;
5570
5571         bytes = 0;
5572         packets = 0;
5573
5574         rcu_read_lock();
5575         for (i = 0; i < adapter->num_rx_queues; i++) {
5576                 struct igb_ring *ring = adapter->rx_ring[i];
5577                 u32 rqdpc = rd32(E1000_RQDPC(i));
5578                 if (hw->mac.type >= e1000_i210)
5579                         wr32(E1000_RQDPC(i), 0);
5580
5581                 if (rqdpc) {
5582                         ring->rx_stats.drops += rqdpc;
5583                         net_stats->rx_fifo_errors += rqdpc;
5584                 }
5585
5586                 do {
5587                         start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
5588                         _bytes = ring->rx_stats.bytes;
5589                         _packets = ring->rx_stats.packets;
5590                 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
5591                 bytes += _bytes;
5592                 packets += _packets;
5593         }
5594
5595         net_stats->rx_bytes = bytes;
5596         net_stats->rx_packets = packets;
5597
5598         bytes = 0;
5599         packets = 0;
5600         for (i = 0; i < adapter->num_tx_queues; i++) {
5601                 struct igb_ring *ring = adapter->tx_ring[i];
5602                 do {
5603                         start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
5604                         _bytes = ring->tx_stats.bytes;
5605                         _packets = ring->tx_stats.packets;
5606                 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
5607                 bytes += _bytes;
5608                 packets += _packets;
5609         }
5610         net_stats->tx_bytes = bytes;
5611         net_stats->tx_packets = packets;
5612         rcu_read_unlock();
5613
5614         /* read stats registers */
5615         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
5616         adapter->stats.gprc += rd32(E1000_GPRC);
5617         adapter->stats.gorc += rd32(E1000_GORCL);
5618         rd32(E1000_GORCH); /* clear GORCL */
5619         adapter->stats.bprc += rd32(E1000_BPRC);
5620         adapter->stats.mprc += rd32(E1000_MPRC);
5621         adapter->stats.roc += rd32(E1000_ROC);
5622
5623         adapter->stats.prc64 += rd32(E1000_PRC64);
5624         adapter->stats.prc127 += rd32(E1000_PRC127);
5625         adapter->stats.prc255 += rd32(E1000_PRC255);
5626         adapter->stats.prc511 += rd32(E1000_PRC511);
5627         adapter->stats.prc1023 += rd32(E1000_PRC1023);
5628         adapter->stats.prc1522 += rd32(E1000_PRC1522);
5629         adapter->stats.symerrs += rd32(E1000_SYMERRS);
5630         adapter->stats.sec += rd32(E1000_SEC);
5631
5632         mpc = rd32(E1000_MPC);
5633         adapter->stats.mpc += mpc;
5634         net_stats->rx_fifo_errors += mpc;
5635         adapter->stats.scc += rd32(E1000_SCC);
5636         adapter->stats.ecol += rd32(E1000_ECOL);
5637         adapter->stats.mcc += rd32(E1000_MCC);
5638         adapter->stats.latecol += rd32(E1000_LATECOL);
5639         adapter->stats.dc += rd32(E1000_DC);
5640         adapter->stats.rlec += rd32(E1000_RLEC);
5641         adapter->stats.xonrxc += rd32(E1000_XONRXC);
5642         adapter->stats.xontxc += rd32(E1000_XONTXC);
5643         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
5644         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
5645         adapter->stats.fcruc += rd32(E1000_FCRUC);
5646         adapter->stats.gptc += rd32(E1000_GPTC);
5647         adapter->stats.gotc += rd32(E1000_GOTCL);
5648         rd32(E1000_GOTCH); /* clear GOTCL */
5649         adapter->stats.rnbc += rd32(E1000_RNBC);
5650         adapter->stats.ruc += rd32(E1000_RUC);
5651         adapter->stats.rfc += rd32(E1000_RFC);
5652         adapter->stats.rjc += rd32(E1000_RJC);
5653         adapter->stats.tor += rd32(E1000_TORH);
5654         adapter->stats.tot += rd32(E1000_TOTH);
5655         adapter->stats.tpr += rd32(E1000_TPR);
5656
5657         adapter->stats.ptc64 += rd32(E1000_PTC64);
5658         adapter->stats.ptc127 += rd32(E1000_PTC127);
5659         adapter->stats.ptc255 += rd32(E1000_PTC255);
5660         adapter->stats.ptc511 += rd32(E1000_PTC511);
5661         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
5662         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
5663
5664         adapter->stats.mptc += rd32(E1000_MPTC);
5665         adapter->stats.bptc += rd32(E1000_BPTC);
5666
5667         adapter->stats.tpt += rd32(E1000_TPT);
5668         adapter->stats.colc += rd32(E1000_COLC);
5669
5670         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
5671         /* read internal phy specific stats */
5672         reg = rd32(E1000_CTRL_EXT);
5673         if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
5674                 adapter->stats.rxerrc += rd32(E1000_RXERRC);
5675
5676                 /* this stat has invalid values on i210/i211 */
5677                 if ((hw->mac.type != e1000_i210) &&
5678                     (hw->mac.type != e1000_i211))
5679                         adapter->stats.tncrs += rd32(E1000_TNCRS);
5680         }
5681
5682         adapter->stats.tsctc += rd32(E1000_TSCTC);
5683         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
5684
5685         adapter->stats.iac += rd32(E1000_IAC);
5686         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
5687         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
5688         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
5689         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
5690         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
5691         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
5692         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
5693         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
5694
5695         /* Fill out the OS statistics structure */
5696         net_stats->multicast = adapter->stats.mprc;
5697         net_stats->collisions = adapter->stats.colc;
5698
5699         /* Rx Errors */
5700
5701         /* RLEC on some newer hardware can be incorrect so build
5702          * our own version based on RUC and ROC
5703          */
5704         net_stats->rx_errors = adapter->stats.rxerrc +
5705                 adapter->stats.crcerrs + adapter->stats.algnerrc +
5706                 adapter->stats.ruc + adapter->stats.roc +
5707                 adapter->stats.cexterr;
5708         net_stats->rx_length_errors = adapter->stats.ruc +
5709                                       adapter->stats.roc;
5710         net_stats->rx_crc_errors = adapter->stats.crcerrs;
5711         net_stats->rx_frame_errors = adapter->stats.algnerrc;
5712         net_stats->rx_missed_errors = adapter->stats.mpc;
5713
5714         /* Tx Errors */
5715         net_stats->tx_errors = adapter->stats.ecol +
5716                                adapter->stats.latecol;
5717         net_stats->tx_aborted_errors = adapter->stats.ecol;
5718         net_stats->tx_window_errors = adapter->stats.latecol;
5719         net_stats->tx_carrier_errors = adapter->stats.tncrs;
5720
5721         /* Tx Dropped needs to be maintained elsewhere */
5722
5723         /* Management Stats */
5724         adapter->stats.mgptc += rd32(E1000_MGTPTC);
5725         adapter->stats.mgprc += rd32(E1000_MGTPRC);
5726         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
5727
5728         /* OS2BMC Stats */
5729         reg = rd32(E1000_MANC);
5730         if (reg & E1000_MANC_EN_BMC2OS) {
5731                 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
5732                 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
5733                 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
5734                 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
5735         }
5736 }
5737
5738 static void igb_tsync_interrupt(struct igb_adapter *adapter)
5739 {
5740         struct e1000_hw *hw = &adapter->hw;
5741         struct ptp_clock_event event;
5742         struct timespec64 ts;
5743         u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
5744
5745         if (tsicr & TSINTR_SYS_WRAP) {
5746                 event.type = PTP_CLOCK_PPS;
5747                 if (adapter->ptp_caps.pps)
5748                         ptp_clock_event(adapter->ptp_clock, &event);
5749                 ack |= TSINTR_SYS_WRAP;
5750         }
5751
5752         if (tsicr & E1000_TSICR_TXTS) {
5753                 /* retrieve hardware timestamp */
5754                 schedule_work(&adapter->ptp_tx_work);
5755                 ack |= E1000_TSICR_TXTS;
5756         }
5757
5758         if (tsicr & TSINTR_TT0) {
5759                 spin_lock(&adapter->tmreg_lock);
5760                 ts = timespec64_add(adapter->perout[0].start,
5761                                     adapter->perout[0].period);
5762                 /* u32 conversion of tv_sec is safe until y2106 */
5763                 wr32(E1000_TRGTTIML0, ts.tv_nsec);
5764                 wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
5765                 tsauxc = rd32(E1000_TSAUXC);
5766                 tsauxc |= TSAUXC_EN_TT0;
5767                 wr32(E1000_TSAUXC, tsauxc);
5768                 adapter->perout[0].start = ts;
5769                 spin_unlock(&adapter->tmreg_lock);
5770                 ack |= TSINTR_TT0;
5771         }
5772
5773         if (tsicr & TSINTR_TT1) {
5774                 spin_lock(&adapter->tmreg_lock);
5775                 ts = timespec64_add(adapter->perout[1].start,
5776                                     adapter->perout[1].period);
5777                 wr32(E1000_TRGTTIML1, ts.tv_nsec);
5778                 wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
5779                 tsauxc = rd32(E1000_TSAUXC);
5780                 tsauxc |= TSAUXC_EN_TT1;
5781                 wr32(E1000_TSAUXC, tsauxc);
5782                 adapter->perout[1].start = ts;
5783                 spin_unlock(&adapter->tmreg_lock);
5784                 ack |= TSINTR_TT1;
5785         }
5786
5787         if (tsicr & TSINTR_AUTT0) {
5788                 nsec = rd32(E1000_AUXSTMPL0);
5789                 sec  = rd32(E1000_AUXSTMPH0);
5790                 event.type = PTP_CLOCK_EXTTS;
5791                 event.index = 0;
5792                 event.timestamp = sec * 1000000000ULL + nsec;
5793                 ptp_clock_event(adapter->ptp_clock, &event);
5794                 ack |= TSINTR_AUTT0;
5795         }
5796
5797         if (tsicr & TSINTR_AUTT1) {
5798                 nsec = rd32(E1000_AUXSTMPL1);
5799                 sec  = rd32(E1000_AUXSTMPH1);
5800                 event.type = PTP_CLOCK_EXTTS;
5801                 event.index = 1;
5802                 event.timestamp = sec * 1000000000ULL + nsec;
5803                 ptp_clock_event(adapter->ptp_clock, &event);
5804                 ack |= TSINTR_AUTT1;
5805         }
5806
5807         /* acknowledge the interrupts */
5808         wr32(E1000_TSICR, ack);
5809 }
5810
5811 static irqreturn_t igb_msix_other(int irq, void *data)
5812 {
5813         struct igb_adapter *adapter = data;
5814         struct e1000_hw *hw = &adapter->hw;
5815         u32 icr = rd32(E1000_ICR);
5816         /* reading ICR causes bit 31 of EICR to be cleared */
5817
5818         if (icr & E1000_ICR_DRSTA)
5819                 schedule_work(&adapter->reset_task);
5820
5821         if (icr & E1000_ICR_DOUTSYNC) {
5822                 /* HW is reporting DMA is out of sync */
5823                 adapter->stats.doosync++;
5824                 /* The DMA Out of Sync is also indication of a spoof event
5825                  * in IOV mode. Check the Wrong VM Behavior register to
5826                  * see if it is really a spoof event.
5827                  */
5828                 igb_check_wvbr(adapter);
5829         }
5830
5831         /* Check for a mailbox event */
5832         if (icr & E1000_ICR_VMMB)
5833                 igb_msg_task(adapter);
5834
5835         if (icr & E1000_ICR_LSC) {
5836                 hw->mac.get_link_status = 1;
5837                 /* guard against interrupt when we're going down */
5838                 if (!test_bit(__IGB_DOWN, &adapter->state))
5839                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
5840         }
5841
5842         if (icr & E1000_ICR_TS)
5843                 igb_tsync_interrupt(adapter);
5844
5845         wr32(E1000_EIMS, adapter->eims_other);
5846
5847         return IRQ_HANDLED;
5848 }
5849
5850 static void igb_write_itr(struct igb_q_vector *q_vector)
5851 {
5852         struct igb_adapter *adapter = q_vector->adapter;
5853         u32 itr_val = q_vector->itr_val & 0x7FFC;
5854
5855         if (!q_vector->set_itr)
5856                 return;
5857
5858         if (!itr_val)
5859                 itr_val = 0x4;
5860
5861         if (adapter->hw.mac.type == e1000_82575)
5862                 itr_val |= itr_val << 16;
5863         else
5864                 itr_val |= E1000_EITR_CNT_IGNR;
5865
5866         writel(itr_val, q_vector->itr_register);
5867         q_vector->set_itr = 0;
5868 }
5869
5870 static irqreturn_t igb_msix_ring(int irq, void *data)
5871 {
5872         struct igb_q_vector *q_vector = data;
5873
5874         /* Write the ITR value calculated from the previous interrupt. */
5875         igb_write_itr(q_vector);
5876
5877         napi_schedule(&q_vector->napi);
5878
5879         return IRQ_HANDLED;
5880 }
5881
5882 #ifdef CONFIG_IGB_DCA
5883 static void igb_update_tx_dca(struct igb_adapter *adapter,
5884                               struct igb_ring *tx_ring,
5885                               int cpu)
5886 {
5887         struct e1000_hw *hw = &adapter->hw;
5888         u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
5889
5890         if (hw->mac.type != e1000_82575)
5891                 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
5892
5893         /* We can enable relaxed ordering for reads, but not writes when
5894          * DCA is enabled.  This is due to a known issue in some chipsets
5895          * which will cause the DCA tag to be cleared.
5896          */
5897         txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
5898                   E1000_DCA_TXCTRL_DATA_RRO_EN |
5899                   E1000_DCA_TXCTRL_DESC_DCA_EN;
5900
5901         wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
5902 }
5903
5904 static void igb_update_rx_dca(struct igb_adapter *adapter,
5905                               struct igb_ring *rx_ring,
5906                               int cpu)
5907 {
5908         struct e1000_hw *hw = &adapter->hw;
5909         u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
5910
5911         if (hw->mac.type != e1000_82575)
5912                 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
5913
5914         /* We can enable relaxed ordering for reads, but not writes when
5915          * DCA is enabled.  This is due to a known issue in some chipsets
5916          * which will cause the DCA tag to be cleared.
5917          */
5918         rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
5919                   E1000_DCA_RXCTRL_DESC_DCA_EN;
5920
5921         wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
5922 }
5923
5924 static void igb_update_dca(struct igb_q_vector *q_vector)
5925 {
5926         struct igb_adapter *adapter = q_vector->adapter;
5927         int cpu = get_cpu();
5928
5929         if (q_vector->cpu == cpu)
5930                 goto out_no_update;
5931
5932         if (q_vector->tx.ring)
5933                 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
5934
5935         if (q_vector->rx.ring)
5936                 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
5937
5938         q_vector->cpu = cpu;
5939 out_no_update:
5940         put_cpu();
5941 }
5942
5943 static void igb_setup_dca(struct igb_adapter *adapter)
5944 {
5945         struct e1000_hw *hw = &adapter->hw;
5946         int i;
5947
5948         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
5949                 return;
5950
5951         /* Always use CB2 mode, difference is masked in the CB driver. */
5952         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
5953
5954         for (i = 0; i < adapter->num_q_vectors; i++) {
5955                 adapter->q_vector[i]->cpu = -1;
5956                 igb_update_dca(adapter->q_vector[i]);
5957         }
5958 }
5959
5960 static int __igb_notify_dca(struct device *dev, void *data)
5961 {
5962         struct net_device *netdev = dev_get_drvdata(dev);
5963         struct igb_adapter *adapter = netdev_priv(netdev);
5964         struct pci_dev *pdev = adapter->pdev;
5965         struct e1000_hw *hw = &adapter->hw;
5966         unsigned long event = *(unsigned long *)data;
5967
5968         switch (event) {
5969         case DCA_PROVIDER_ADD:
5970                 /* if already enabled, don't do it again */
5971                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
5972                         break;
5973                 if (dca_add_requester(dev) == 0) {
5974                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
5975                         dev_info(&pdev->dev, "DCA enabled\n");
5976                         igb_setup_dca(adapter);
5977                         break;
5978                 }
5979                 /* Fall Through since DCA is disabled. */
5980         case DCA_PROVIDER_REMOVE:
5981                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
5982                         /* without this a class_device is left
5983                          * hanging around in the sysfs model
5984                          */
5985                         dca_remove_requester(dev);
5986                         dev_info(&pdev->dev, "DCA disabled\n");
5987                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
5988                         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
5989                 }
5990                 break;
5991         }
5992
5993         return 0;
5994 }
5995
5996 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
5997                           void *p)
5998 {
5999         int ret_val;
6000
6001         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
6002                                          __igb_notify_dca);
6003
6004         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6005 }
6006 #endif /* CONFIG_IGB_DCA */
6007
6008 #ifdef CONFIG_PCI_IOV
6009 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
6010 {
6011         unsigned char mac_addr[ETH_ALEN];
6012
6013         eth_zero_addr(mac_addr);
6014         igb_set_vf_mac(adapter, vf, mac_addr);
6015
6016         /* By default spoof check is enabled for all VFs */
6017         adapter->vf_data[vf].spoofchk_enabled = true;
6018
6019         return 0;
6020 }
6021
6022 #endif
6023 static void igb_ping_all_vfs(struct igb_adapter *adapter)
6024 {
6025         struct e1000_hw *hw = &adapter->hw;
6026         u32 ping;
6027         int i;
6028
6029         for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
6030                 ping = E1000_PF_CONTROL_MSG;
6031                 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
6032                         ping |= E1000_VT_MSGTYPE_CTS;
6033                 igb_write_mbx(hw, &ping, 1, i);
6034         }
6035 }
6036
6037 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6038 {
6039         struct e1000_hw *hw = &adapter->hw;
6040         u32 vmolr = rd32(E1000_VMOLR(vf));
6041         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6042
6043         vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
6044                             IGB_VF_FLAG_MULTI_PROMISC);
6045         vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6046
6047         if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
6048                 vmolr |= E1000_VMOLR_MPME;
6049                 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
6050                 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
6051         } else {
6052                 /* if we have hashes and we are clearing a multicast promisc
6053                  * flag we need to write the hashes to the MTA as this step
6054                  * was previously skipped
6055                  */
6056                 if (vf_data->num_vf_mc_hashes > 30) {
6057                         vmolr |= E1000_VMOLR_MPME;
6058                 } else if (vf_data->num_vf_mc_hashes) {
6059                         int j;
6060
6061                         vmolr |= E1000_VMOLR_ROMPE;
6062                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6063                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6064                 }
6065         }
6066
6067         wr32(E1000_VMOLR(vf), vmolr);
6068
6069         /* there are flags left unprocessed, likely not supported */
6070         if (*msgbuf & E1000_VT_MSGINFO_MASK)
6071                 return -EINVAL;
6072
6073         return 0;
6074 }
6075
6076 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
6077                                   u32 *msgbuf, u32 vf)
6078 {
6079         int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6080         u16 *hash_list = (u16 *)&msgbuf[1];
6081         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6082         int i;
6083
6084         /* salt away the number of multicast addresses assigned
6085          * to this VF for later use to restore when the PF multi cast
6086          * list changes
6087          */
6088         vf_data->num_vf_mc_hashes = n;
6089
6090         /* only up to 30 hash values supported */
6091         if (n > 30)
6092                 n = 30;
6093
6094         /* store the hashes for later use */
6095         for (i = 0; i < n; i++)
6096                 vf_data->vf_mc_hashes[i] = hash_list[i];
6097
6098         /* Flush and reset the mta with the new values */
6099         igb_set_rx_mode(adapter->netdev);
6100
6101         return 0;
6102 }
6103
6104 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
6105 {
6106         struct e1000_hw *hw = &adapter->hw;
6107         struct vf_data_storage *vf_data;
6108         int i, j;
6109
6110         for (i = 0; i < adapter->vfs_allocated_count; i++) {
6111                 u32 vmolr = rd32(E1000_VMOLR(i));
6112
6113                 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
6114
6115                 vf_data = &adapter->vf_data[i];
6116
6117                 if ((vf_data->num_vf_mc_hashes > 30) ||
6118                     (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
6119                         vmolr |= E1000_VMOLR_MPME;
6120                 } else if (vf_data->num_vf_mc_hashes) {
6121                         vmolr |= E1000_VMOLR_ROMPE;
6122                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
6123                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
6124                 }
6125                 wr32(E1000_VMOLR(i), vmolr);
6126         }
6127 }
6128
6129 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
6130 {
6131         struct e1000_hw *hw = &adapter->hw;
6132         u32 pool_mask, vlvf_mask, i;
6133
6134         /* create mask for VF and other pools */
6135         pool_mask = E1000_VLVF_POOLSEL_MASK;
6136         vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
6137
6138         /* drop PF from pool bits */
6139         pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
6140                              adapter->vfs_allocated_count);
6141
6142         /* Find the vlan filter for this id */
6143         for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
6144                 u32 vlvf = rd32(E1000_VLVF(i));
6145                 u32 vfta_mask, vid, vfta;
6146
6147                 /* remove the vf from the pool */
6148                 if (!(vlvf & vlvf_mask))
6149                         continue;
6150
6151                 /* clear out bit from VLVF */
6152                 vlvf ^= vlvf_mask;
6153
6154                 /* if other pools are present, just remove ourselves */
6155                 if (vlvf & pool_mask)
6156                         goto update_vlvfb;
6157
6158                 /* if PF is present, leave VFTA */
6159                 if (vlvf & E1000_VLVF_POOLSEL_MASK)
6160                         goto update_vlvf;
6161
6162                 vid = vlvf & E1000_VLVF_VLANID_MASK;
6163                 vfta_mask = BIT(vid % 32);
6164
6165                 /* clear bit from VFTA */
6166                 vfta = adapter->shadow_vfta[vid / 32];
6167                 if (vfta & vfta_mask)
6168                         hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
6169 update_vlvf:
6170                 /* clear pool selection enable */
6171                 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6172                         vlvf &= E1000_VLVF_POOLSEL_MASK;
6173                 else
6174                         vlvf = 0;
6175 update_vlvfb:
6176                 /* clear pool bits */
6177                 wr32(E1000_VLVF(i), vlvf);
6178         }
6179 }
6180
6181 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
6182 {
6183         u32 vlvf;
6184         int idx;
6185
6186         /* short cut the special case */
6187         if (vlan == 0)
6188                 return 0;
6189
6190         /* Search for the VLAN id in the VLVF entries */
6191         for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
6192                 vlvf = rd32(E1000_VLVF(idx));
6193                 if ((vlvf & VLAN_VID_MASK) == vlan)
6194                         break;
6195         }
6196
6197         return idx;
6198 }
6199
6200 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
6201 {
6202         struct e1000_hw *hw = &adapter->hw;
6203         u32 bits, pf_id;
6204         int idx;
6205
6206         idx = igb_find_vlvf_entry(hw, vid);
6207         if (!idx)
6208                 return;
6209
6210         /* See if any other pools are set for this VLAN filter
6211          * entry other than the PF.
6212          */
6213         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
6214         bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
6215         bits &= rd32(E1000_VLVF(idx));
6216
6217         /* Disable the filter so this falls into the default pool. */
6218         if (!bits) {
6219                 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
6220                         wr32(E1000_VLVF(idx), BIT(pf_id));
6221                 else
6222                         wr32(E1000_VLVF(idx), 0);
6223         }
6224 }
6225
6226 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
6227                            bool add, u32 vf)
6228 {
6229         int pf_id = adapter->vfs_allocated_count;
6230         struct e1000_hw *hw = &adapter->hw;
6231         int err;
6232
6233         /* If VLAN overlaps with one the PF is currently monitoring make
6234          * sure that we are able to allocate a VLVF entry.  This may be
6235          * redundant but it guarantees PF will maintain visibility to
6236          * the VLAN.
6237          */
6238         if (add && test_bit(vid, adapter->active_vlans)) {
6239                 err = igb_vfta_set(hw, vid, pf_id, true, false);
6240                 if (err)
6241                         return err;
6242         }
6243
6244         err = igb_vfta_set(hw, vid, vf, add, false);
6245
6246         if (add && !err)
6247                 return err;
6248
6249         /* If we failed to add the VF VLAN or we are removing the VF VLAN
6250          * we may need to drop the PF pool bit in order to allow us to free
6251          * up the VLVF resources.
6252          */
6253         if (test_bit(vid, adapter->active_vlans) ||
6254             (adapter->flags & IGB_FLAG_VLAN_PROMISC))
6255                 igb_update_pf_vlvf(adapter, vid);
6256
6257         return err;
6258 }
6259
6260 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
6261 {
6262         struct e1000_hw *hw = &adapter->hw;
6263
6264         if (vid)
6265                 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
6266         else
6267                 wr32(E1000_VMVIR(vf), 0);
6268 }
6269
6270 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
6271                                 u16 vlan, u8 qos)
6272 {
6273         int err;
6274
6275         err = igb_set_vf_vlan(adapter, vlan, true, vf);
6276         if (err)
6277                 return err;
6278
6279         igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
6280         igb_set_vmolr(adapter, vf, !vlan);
6281
6282         /* revoke access to previous VLAN */
6283         if (vlan != adapter->vf_data[vf].pf_vlan)
6284                 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6285                                 false, vf);
6286
6287         adapter->vf_data[vf].pf_vlan = vlan;
6288         adapter->vf_data[vf].pf_qos = qos;
6289         igb_set_vf_vlan_strip(adapter, vf, true);
6290         dev_info(&adapter->pdev->dev,
6291                  "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
6292         if (test_bit(__IGB_DOWN, &adapter->state)) {
6293                 dev_warn(&adapter->pdev->dev,
6294                          "The VF VLAN has been set, but the PF device is not up.\n");
6295                 dev_warn(&adapter->pdev->dev,
6296                          "Bring the PF device up before attempting to use the VF device.\n");
6297         }
6298
6299         return err;
6300 }
6301
6302 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
6303 {
6304         /* Restore tagless access via VLAN 0 */
6305         igb_set_vf_vlan(adapter, 0, true, vf);
6306
6307         igb_set_vmvir(adapter, 0, vf);
6308         igb_set_vmolr(adapter, vf, true);
6309
6310         /* Remove any PF assigned VLAN */
6311         if (adapter->vf_data[vf].pf_vlan)
6312                 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
6313                                 false, vf);
6314
6315         adapter->vf_data[vf].pf_vlan = 0;
6316         adapter->vf_data[vf].pf_qos = 0;
6317         igb_set_vf_vlan_strip(adapter, vf, false);
6318
6319         return 0;
6320 }
6321
6322 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
6323                                u16 vlan, u8 qos, __be16 vlan_proto)
6324 {
6325         struct igb_adapter *adapter = netdev_priv(netdev);
6326
6327         if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
6328                 return -EINVAL;
6329
6330         if (vlan_proto != htons(ETH_P_8021Q))
6331                 return -EPROTONOSUPPORT;
6332
6333         return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
6334                                igb_disable_port_vlan(adapter, vf);
6335 }
6336
6337 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
6338 {
6339         int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
6340         int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
6341         int ret;
6342
6343         if (adapter->vf_data[vf].pf_vlan)
6344                 return -1;
6345
6346         /* VLAN 0 is a special case, don't allow it to be removed */
6347         if (!vid && !add)
6348                 return 0;
6349
6350         ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
6351         if (!ret)
6352                 igb_set_vf_vlan_strip(adapter, vf, !!vid);
6353         return ret;
6354 }
6355
6356 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
6357 {
6358         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6359
6360         /* clear flags - except flag that indicates PF has set the MAC */
6361         vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
6362         vf_data->last_nack = jiffies;
6363
6364         /* reset vlans for device */
6365         igb_clear_vf_vfta(adapter, vf);
6366         igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
6367         igb_set_vmvir(adapter, vf_data->pf_vlan |
6368                                (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
6369         igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
6370         igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
6371
6372         /* reset multicast table array for vf */
6373         adapter->vf_data[vf].num_vf_mc_hashes = 0;
6374
6375         /* Flush and reset the mta with the new values */
6376         igb_set_rx_mode(adapter->netdev);
6377 }
6378
6379 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
6380 {
6381         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6382
6383         /* clear mac address as we were hotplug removed/added */
6384         if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
6385                 eth_zero_addr(vf_mac);
6386
6387         /* process remaining reset events */
6388         igb_vf_reset(adapter, vf);
6389 }
6390
6391 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
6392 {
6393         struct e1000_hw *hw = &adapter->hw;
6394         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
6395         u32 reg, msgbuf[3];
6396         u8 *addr = (u8 *)(&msgbuf[1]);
6397
6398         /* process all the same items cleared in a function level reset */
6399         igb_vf_reset(adapter, vf);
6400
6401         /* set vf mac address */
6402         igb_set_vf_mac(adapter, vf, vf_mac);
6403
6404         /* enable transmit and receive for vf */
6405         reg = rd32(E1000_VFTE);
6406         wr32(E1000_VFTE, reg | BIT(vf));
6407         reg = rd32(E1000_VFRE);
6408         wr32(E1000_VFRE, reg | BIT(vf));
6409
6410         adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
6411
6412         /* reply to reset with ack and vf mac address */
6413         if (!is_zero_ether_addr(vf_mac)) {
6414                 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
6415                 memcpy(addr, vf_mac, ETH_ALEN);
6416         } else {
6417                 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
6418         }
6419         igb_write_mbx(hw, msgbuf, 3, vf);
6420 }
6421
6422 static void igb_flush_mac_table(struct igb_adapter *adapter)
6423 {
6424         struct e1000_hw *hw = &adapter->hw;
6425         int i;
6426
6427         for (i = 0; i < hw->mac.rar_entry_count; i++) {
6428                 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
6429                 memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
6430                 adapter->mac_table[i].queue = 0;
6431                 igb_rar_set_index(adapter, i);
6432         }
6433 }
6434
6435 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
6436 {
6437         struct e1000_hw *hw = &adapter->hw;
6438         /* do not count rar entries reserved for VFs MAC addresses */
6439         int rar_entries = hw->mac.rar_entry_count -
6440                           adapter->vfs_allocated_count;
6441         int i, count = 0;
6442
6443         for (i = 0; i < rar_entries; i++) {
6444                 /* do not count default entries */
6445                 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
6446                         continue;
6447
6448                 /* do not count "in use" entries for different queues */
6449                 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
6450                     (adapter->mac_table[i].queue != queue))
6451                         continue;
6452
6453                 count++;
6454         }
6455
6456         return count;
6457 }
6458
6459 /* Set default MAC address for the PF in the first RAR entry */
6460 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
6461 {
6462         struct igb_mac_addr *mac_table = &adapter->mac_table[0];
6463
6464         ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
6465         mac_table->queue = adapter->vfs_allocated_count;
6466         mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
6467
6468         igb_rar_set_index(adapter, 0);
6469 }
6470
6471 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
6472                               const u8 queue)
6473 {
6474         struct e1000_hw *hw = &adapter->hw;
6475         int rar_entries = hw->mac.rar_entry_count -
6476                           adapter->vfs_allocated_count;
6477         int i;
6478
6479         if (is_zero_ether_addr(addr))
6480                 return -EINVAL;
6481
6482         /* Search for the first empty entry in the MAC table.
6483          * Do not touch entries at the end of the table reserved for the VF MAC
6484          * addresses.
6485          */
6486         for (i = 0; i < rar_entries; i++) {
6487                 if (adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE)
6488                         continue;
6489
6490                 ether_addr_copy(adapter->mac_table[i].addr, addr);
6491                 adapter->mac_table[i].queue = queue;
6492                 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE;
6493
6494                 igb_rar_set_index(adapter, i);
6495                 return i;
6496         }
6497
6498         return -ENOSPC;
6499 }
6500
6501 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
6502                               const u8 queue)
6503 {
6504         struct e1000_hw *hw = &adapter->hw;
6505         int rar_entries = hw->mac.rar_entry_count -
6506                           adapter->vfs_allocated_count;
6507         int i;
6508
6509         if (is_zero_ether_addr(addr))
6510                 return -EINVAL;
6511
6512         /* Search for matching entry in the MAC table based on given address
6513          * and queue. Do not touch entries at the end of the table reserved
6514          * for the VF MAC addresses.
6515          */
6516         for (i = 0; i < rar_entries; i++) {
6517                 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
6518                         continue;
6519                 if (adapter->mac_table[i].queue != queue)
6520                         continue;
6521                 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
6522                         continue;
6523
6524                 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
6525                 memset(adapter->mac_table[i].addr, 0, ETH_ALEN);
6526                 adapter->mac_table[i].queue = 0;
6527
6528                 igb_rar_set_index(adapter, i);
6529                 return 0;
6530         }
6531
6532         return -ENOENT;
6533 }
6534
6535 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
6536 {
6537         struct igb_adapter *adapter = netdev_priv(netdev);
6538         int ret;
6539
6540         ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
6541
6542         return min_t(int, ret, 0);
6543 }
6544
6545 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
6546 {
6547         struct igb_adapter *adapter = netdev_priv(netdev);
6548
6549         igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
6550
6551         return 0;
6552 }
6553
6554 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
6555                                  const u32 info, const u8 *addr)
6556 {
6557         struct pci_dev *pdev = adapter->pdev;
6558         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6559         struct list_head *pos;
6560         struct vf_mac_filter *entry = NULL;
6561         int ret = 0;
6562
6563         switch (info) {
6564         case E1000_VF_MAC_FILTER_CLR:
6565                 /* remove all unicast MAC filters related to the current VF */
6566                 list_for_each(pos, &adapter->vf_macs.l) {
6567                         entry = list_entry(pos, struct vf_mac_filter, l);
6568                         if (entry->vf == vf) {
6569                                 entry->vf = -1;
6570                                 entry->free = true;
6571                                 igb_del_mac_filter(adapter, entry->vf_mac, vf);
6572                         }
6573                 }
6574                 break;
6575         case E1000_VF_MAC_FILTER_ADD:
6576                 if (vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) {
6577                         dev_warn(&pdev->dev,
6578                                  "VF %d requested MAC filter but is administratively denied\n",
6579                                  vf);
6580                         return -EINVAL;
6581                 }
6582
6583                 if (!is_valid_ether_addr(addr)) {
6584                         dev_warn(&pdev->dev,
6585                                  "VF %d attempted to set invalid MAC filter\n",
6586                                  vf);
6587                         return -EINVAL;
6588                 }
6589
6590                 /* try to find empty slot in the list */
6591                 list_for_each(pos, &adapter->vf_macs.l) {
6592                         entry = list_entry(pos, struct vf_mac_filter, l);
6593                         if (entry->free)
6594                                 break;
6595                 }
6596
6597                 if (entry && entry->free) {
6598                         entry->free = false;
6599                         entry->vf = vf;
6600                         ether_addr_copy(entry->vf_mac, addr);
6601
6602                         ret = igb_add_mac_filter(adapter, addr, vf);
6603                         ret = min_t(int, ret, 0);
6604                 } else {
6605                         ret = -ENOSPC;
6606                 }
6607
6608                 if (ret == -ENOSPC)
6609                         dev_warn(&pdev->dev,
6610                                  "VF %d has requested MAC filter but there is no space for it\n",
6611                                  vf);
6612                 break;
6613         default:
6614                 ret = -EINVAL;
6615                 break;
6616         }
6617
6618         return ret;
6619 }
6620
6621 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
6622 {
6623         struct pci_dev *pdev = adapter->pdev;
6624         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6625         u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
6626
6627         /* The VF MAC Address is stored in a packed array of bytes
6628          * starting at the second 32 bit word of the msg array
6629          */
6630         unsigned char *addr = (unsigned char *)&msg[1];
6631         int ret = 0;
6632
6633         if (!info) {
6634                 if (vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) {
6635                         dev_warn(&pdev->dev,
6636                                  "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
6637                                  vf);
6638                         return -EINVAL;
6639                 }
6640
6641                 if (!is_valid_ether_addr(addr)) {
6642                         dev_warn(&pdev->dev,
6643                                  "VF %d attempted to set invalid MAC\n",
6644                                  vf);
6645                         return -EINVAL;
6646                 }
6647
6648                 ret = igb_set_vf_mac(adapter, vf, addr);
6649         } else {
6650                 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
6651         }
6652
6653         return ret;
6654 }
6655
6656 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
6657 {
6658         struct e1000_hw *hw = &adapter->hw;
6659         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6660         u32 msg = E1000_VT_MSGTYPE_NACK;
6661
6662         /* if device isn't clear to send it shouldn't be reading either */
6663         if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
6664             time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
6665                 igb_write_mbx(hw, &msg, 1, vf);
6666                 vf_data->last_nack = jiffies;
6667         }
6668 }
6669
6670 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
6671 {
6672         struct pci_dev *pdev = adapter->pdev;
6673         u32 msgbuf[E1000_VFMAILBOX_SIZE];
6674         struct e1000_hw *hw = &adapter->hw;
6675         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
6676         s32 retval;
6677
6678         retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
6679
6680         if (retval) {
6681                 /* if receive failed revoke VF CTS stats and restart init */
6682                 dev_err(&pdev->dev, "Error receiving message from VF\n");
6683                 vf_data->flags &= ~IGB_VF_FLAG_CTS;
6684                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6685                         goto unlock;
6686                 goto out;
6687         }
6688
6689         /* this is a message we already processed, do nothing */
6690         if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
6691                 goto unlock;
6692
6693         /* until the vf completes a reset it should not be
6694          * allowed to start any configuration.
6695          */
6696         if (msgbuf[0] == E1000_VF_RESET) {
6697                 /* unlocks mailbox */
6698                 igb_vf_reset_msg(adapter, vf);
6699                 return;
6700         }
6701
6702         if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
6703                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
6704                         goto unlock;
6705                 retval = -1;
6706                 goto out;
6707         }
6708
6709         switch ((msgbuf[0] & 0xFFFF)) {
6710         case E1000_VF_SET_MAC_ADDR:
6711                 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
6712                 break;
6713         case E1000_VF_SET_PROMISC:
6714                 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
6715                 break;
6716         case E1000_VF_SET_MULTICAST:
6717                 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
6718                 break;
6719         case E1000_VF_SET_LPE:
6720                 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
6721                 break;
6722         case E1000_VF_SET_VLAN:
6723                 retval = -1;
6724                 if (vf_data->pf_vlan)
6725                         dev_warn(&pdev->dev,
6726                                  "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
6727                                  vf);
6728                 else
6729                         retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
6730                 break;
6731         default:
6732                 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
6733                 retval = -1;
6734                 break;
6735         }
6736
6737         msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
6738 out:
6739         /* notify the VF of the results of what it sent us */
6740         if (retval)
6741                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
6742         else
6743                 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
6744
6745         /* unlocks mailbox */
6746         igb_write_mbx(hw, msgbuf, 1, vf);
6747         return;
6748
6749 unlock:
6750         igb_unlock_mbx(hw, vf);
6751 }
6752
6753 static void igb_msg_task(struct igb_adapter *adapter)
6754 {
6755         struct e1000_hw *hw = &adapter->hw;
6756         u32 vf;
6757
6758         for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
6759                 /* process any reset requests */
6760                 if (!igb_check_for_rst(hw, vf))
6761                         igb_vf_reset_event(adapter, vf);
6762
6763                 /* process any messages pending */
6764                 if (!igb_check_for_msg(hw, vf))
6765                         igb_rcv_msg_from_vf(adapter, vf);
6766
6767                 /* process any acks */
6768                 if (!igb_check_for_ack(hw, vf))
6769                         igb_rcv_ack_from_vf(adapter, vf);
6770         }
6771 }
6772
6773 /**
6774  *  igb_set_uta - Set unicast filter table address
6775  *  @adapter: board private structure
6776  *  @set: boolean indicating if we are setting or clearing bits
6777  *
6778  *  The unicast table address is a register array of 32-bit registers.
6779  *  The table is meant to be used in a way similar to how the MTA is used
6780  *  however due to certain limitations in the hardware it is necessary to
6781  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
6782  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
6783  **/
6784 static void igb_set_uta(struct igb_adapter *adapter, bool set)
6785 {
6786         struct e1000_hw *hw = &adapter->hw;
6787         u32 uta = set ? ~0 : 0;
6788         int i;
6789
6790         /* we only need to do this if VMDq is enabled */
6791         if (!adapter->vfs_allocated_count)
6792                 return;
6793
6794         for (i = hw->mac.uta_reg_count; i--;)
6795                 array_wr32(E1000_UTA, i, uta);
6796 }
6797
6798 /**
6799  *  igb_intr_msi - Interrupt Handler
6800  *  @irq: interrupt number
6801  *  @data: pointer to a network interface device structure
6802  **/
6803 static irqreturn_t igb_intr_msi(int irq, void *data)
6804 {
6805         struct igb_adapter *adapter = data;
6806         struct igb_q_vector *q_vector = adapter->q_vector[0];
6807         struct e1000_hw *hw = &adapter->hw;
6808         /* read ICR disables interrupts using IAM */
6809         u32 icr = rd32(E1000_ICR);
6810
6811         igb_write_itr(q_vector);
6812
6813         if (icr & E1000_ICR_DRSTA)
6814                 schedule_work(&adapter->reset_task);
6815
6816         if (icr & E1000_ICR_DOUTSYNC) {
6817                 /* HW is reporting DMA is out of sync */
6818                 adapter->stats.doosync++;
6819         }
6820
6821         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6822                 hw->mac.get_link_status = 1;
6823                 if (!test_bit(__IGB_DOWN, &adapter->state))
6824                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
6825         }
6826
6827         if (icr & E1000_ICR_TS)
6828                 igb_tsync_interrupt(adapter);
6829
6830         napi_schedule(&q_vector->napi);
6831
6832         return IRQ_HANDLED;
6833 }
6834
6835 /**
6836  *  igb_intr - Legacy Interrupt Handler
6837  *  @irq: interrupt number
6838  *  @data: pointer to a network interface device structure
6839  **/
6840 static irqreturn_t igb_intr(int irq, void *data)
6841 {
6842         struct igb_adapter *adapter = data;
6843         struct igb_q_vector *q_vector = adapter->q_vector[0];
6844         struct e1000_hw *hw = &adapter->hw;
6845         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
6846          * need for the IMC write
6847          */
6848         u32 icr = rd32(E1000_ICR);
6849
6850         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
6851          * not set, then the adapter didn't send an interrupt
6852          */
6853         if (!(icr & E1000_ICR_INT_ASSERTED))
6854                 return IRQ_NONE;
6855
6856         igb_write_itr(q_vector);
6857
6858         if (icr & E1000_ICR_DRSTA)
6859                 schedule_work(&adapter->reset_task);
6860
6861         if (icr & E1000_ICR_DOUTSYNC) {
6862                 /* HW is reporting DMA is out of sync */
6863                 adapter->stats.doosync++;
6864         }
6865
6866         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
6867                 hw->mac.get_link_status = 1;
6868                 /* guard against interrupt when we're going down */
6869                 if (!test_bit(__IGB_DOWN, &adapter->state))
6870                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
6871         }
6872
6873         if (icr & E1000_ICR_TS)
6874                 igb_tsync_interrupt(adapter);
6875
6876         napi_schedule(&q_vector->napi);
6877
6878         return IRQ_HANDLED;
6879 }
6880
6881 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
6882 {
6883         struct igb_adapter *adapter = q_vector->adapter;
6884         struct e1000_hw *hw = &adapter->hw;
6885
6886         if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
6887             (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
6888                 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
6889                         igb_set_itr(q_vector);
6890                 else
6891                         igb_update_ring_itr(q_vector);
6892         }
6893
6894         if (!test_bit(__IGB_DOWN, &adapter->state)) {
6895                 if (adapter->flags & IGB_FLAG_HAS_MSIX)
6896                         wr32(E1000_EIMS, q_vector->eims_value);
6897                 else
6898                         igb_irq_enable(adapter);
6899         }
6900 }
6901
6902 /**
6903  *  igb_poll - NAPI Rx polling callback
6904  *  @napi: napi polling structure
6905  *  @budget: count of how many packets we should handle
6906  **/
6907 static int igb_poll(struct napi_struct *napi, int budget)
6908 {
6909         struct igb_q_vector *q_vector = container_of(napi,
6910                                                      struct igb_q_vector,
6911                                                      napi);
6912         bool clean_complete = true;
6913         int work_done = 0;
6914
6915 #ifdef CONFIG_IGB_DCA
6916         if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
6917                 igb_update_dca(q_vector);
6918 #endif
6919         if (q_vector->tx.ring)
6920                 clean_complete = igb_clean_tx_irq(q_vector, budget);
6921
6922         if (q_vector->rx.ring) {
6923                 int cleaned = igb_clean_rx_irq(q_vector, budget);
6924
6925                 work_done += cleaned;
6926                 if (cleaned >= budget)
6927                         clean_complete = false;
6928         }
6929
6930         /* If all work not completed, return budget and keep polling */
6931         if (!clean_complete)
6932                 return budget;
6933
6934         /* If not enough Rx work done, exit the polling mode */
6935         napi_complete_done(napi, work_done);
6936         igb_ring_irq_enable(q_vector);
6937
6938         return 0;
6939 }
6940
6941 /**
6942  *  igb_clean_tx_irq - Reclaim resources after transmit completes
6943  *  @q_vector: pointer to q_vector containing needed info
6944  *  @napi_budget: Used to determine if we are in netpoll
6945  *
6946  *  returns true if ring is completely cleaned
6947  **/
6948 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
6949 {
6950         struct igb_adapter *adapter = q_vector->adapter;
6951         struct igb_ring *tx_ring = q_vector->tx.ring;
6952         struct igb_tx_buffer *tx_buffer;
6953         union e1000_adv_tx_desc *tx_desc;
6954         unsigned int total_bytes = 0, total_packets = 0;
6955         unsigned int budget = q_vector->tx.work_limit;
6956         unsigned int i = tx_ring->next_to_clean;
6957
6958         if (test_bit(__IGB_DOWN, &adapter->state))
6959                 return true;
6960
6961         tx_buffer = &tx_ring->tx_buffer_info[i];
6962         tx_desc = IGB_TX_DESC(tx_ring, i);
6963         i -= tx_ring->count;
6964
6965         do {
6966                 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
6967
6968                 /* if next_to_watch is not set then there is no work pending */
6969                 if (!eop_desc)
6970                         break;
6971
6972                 /* prevent any other reads prior to eop_desc */
6973                 read_barrier_depends();
6974
6975                 /* if DD is not set pending work has not been completed */
6976                 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
6977                         break;
6978
6979                 /* clear next_to_watch to prevent false hangs */
6980                 tx_buffer->next_to_watch = NULL;
6981
6982                 /* update the statistics for this packet */
6983                 total_bytes += tx_buffer->bytecount;
6984                 total_packets += tx_buffer->gso_segs;
6985
6986                 /* free the skb */
6987                 napi_consume_skb(tx_buffer->skb, napi_budget);
6988
6989                 /* unmap skb header data */
6990                 dma_unmap_single(tx_ring->dev,
6991                                  dma_unmap_addr(tx_buffer, dma),
6992                                  dma_unmap_len(tx_buffer, len),
6993                                  DMA_TO_DEVICE);
6994
6995                 /* clear tx_buffer data */
6996                 dma_unmap_len_set(tx_buffer, len, 0);
6997
6998                 /* clear last DMA location and unmap remaining buffers */
6999                 while (tx_desc != eop_desc) {
7000                         tx_buffer++;
7001                         tx_desc++;
7002                         i++;
7003                         if (unlikely(!i)) {
7004                                 i -= tx_ring->count;
7005                                 tx_buffer = tx_ring->tx_buffer_info;
7006                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
7007                         }
7008
7009                         /* unmap any remaining paged data */
7010                         if (dma_unmap_len(tx_buffer, len)) {
7011                                 dma_unmap_page(tx_ring->dev,
7012                                                dma_unmap_addr(tx_buffer, dma),
7013                                                dma_unmap_len(tx_buffer, len),
7014                                                DMA_TO_DEVICE);
7015                                 dma_unmap_len_set(tx_buffer, len, 0);
7016                         }
7017                 }
7018
7019                 /* move us one more past the eop_desc for start of next pkt */
7020                 tx_buffer++;
7021                 tx_desc++;
7022                 i++;
7023                 if (unlikely(!i)) {
7024                         i -= tx_ring->count;
7025                         tx_buffer = tx_ring->tx_buffer_info;
7026                         tx_desc = IGB_TX_DESC(tx_ring, 0);
7027                 }
7028
7029                 /* issue prefetch for next Tx descriptor */
7030                 prefetch(tx_desc);
7031
7032                 /* update budget accounting */
7033                 budget--;
7034         } while (likely(budget));
7035
7036         netdev_tx_completed_queue(txring_txq(tx_ring),
7037                                   total_packets, total_bytes);
7038         i += tx_ring->count;
7039         tx_ring->next_to_clean = i;
7040         u64_stats_update_begin(&tx_ring->tx_syncp);
7041         tx_ring->tx_stats.bytes += total_bytes;
7042         tx_ring->tx_stats.packets += total_packets;
7043         u64_stats_update_end(&tx_ring->tx_syncp);
7044         q_vector->tx.total_bytes += total_bytes;
7045         q_vector->tx.total_packets += total_packets;
7046
7047         if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
7048                 struct e1000_hw *hw = &adapter->hw;
7049
7050                 /* Detect a transmit hang in hardware, this serializes the
7051                  * check with the clearing of time_stamp and movement of i
7052                  */
7053                 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
7054                 if (tx_buffer->next_to_watch &&
7055                     time_after(jiffies, tx_buffer->time_stamp +
7056                                (adapter->tx_timeout_factor * HZ)) &&
7057                     !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
7058
7059                         /* detected Tx unit hang */
7060                         dev_err(tx_ring->dev,
7061                                 "Detected Tx Unit Hang\n"
7062                                 "  Tx Queue             <%d>\n"
7063                                 "  TDH                  <%x>\n"
7064                                 "  TDT                  <%x>\n"
7065                                 "  next_to_use          <%x>\n"
7066                                 "  next_to_clean        <%x>\n"
7067                                 "buffer_info[next_to_clean]\n"
7068                                 "  time_stamp           <%lx>\n"
7069                                 "  next_to_watch        <%p>\n"
7070                                 "  jiffies              <%lx>\n"
7071                                 "  desc.status          <%x>\n",
7072                                 tx_ring->queue_index,
7073                                 rd32(E1000_TDH(tx_ring->reg_idx)),
7074                                 readl(tx_ring->tail),
7075                                 tx_ring->next_to_use,
7076                                 tx_ring->next_to_clean,
7077                                 tx_buffer->time_stamp,
7078                                 tx_buffer->next_to_watch,
7079                                 jiffies,
7080                                 tx_buffer->next_to_watch->wb.status);
7081                         netif_stop_subqueue(tx_ring->netdev,
7082                                             tx_ring->queue_index);
7083
7084                         /* we are about to reset, no point in enabling stuff */
7085                         return true;
7086                 }
7087         }
7088
7089 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
7090         if (unlikely(total_packets &&
7091             netif_carrier_ok(tx_ring->netdev) &&
7092             igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
7093                 /* Make sure that anybody stopping the queue after this
7094                  * sees the new next_to_clean.
7095                  */
7096                 smp_mb();
7097                 if (__netif_subqueue_stopped(tx_ring->netdev,
7098                                              tx_ring->queue_index) &&
7099                     !(test_bit(__IGB_DOWN, &adapter->state))) {
7100                         netif_wake_subqueue(tx_ring->netdev,
7101                                             tx_ring->queue_index);
7102
7103                         u64_stats_update_begin(&tx_ring->tx_syncp);
7104                         tx_ring->tx_stats.restart_queue++;
7105                         u64_stats_update_end(&tx_ring->tx_syncp);
7106                 }
7107         }
7108
7109         return !!budget;
7110 }
7111
7112 /**
7113  *  igb_reuse_rx_page - page flip buffer and store it back on the ring
7114  *  @rx_ring: rx descriptor ring to store buffers on
7115  *  @old_buff: donor buffer to have page reused
7116  *
7117  *  Synchronizes page for reuse by the adapter
7118  **/
7119 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
7120                               struct igb_rx_buffer *old_buff)
7121 {
7122         struct igb_rx_buffer *new_buff;
7123         u16 nta = rx_ring->next_to_alloc;
7124
7125         new_buff = &rx_ring->rx_buffer_info[nta];
7126
7127         /* update, and store next to alloc */
7128         nta++;
7129         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
7130
7131         /* Transfer page from old buffer to new buffer.
7132          * Move each member individually to avoid possible store
7133          * forwarding stalls.
7134          */
7135         new_buff->dma           = old_buff->dma;
7136         new_buff->page          = old_buff->page;
7137         new_buff->page_offset   = old_buff->page_offset;
7138         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
7139 }
7140
7141 static inline bool igb_page_is_reserved(struct page *page)
7142 {
7143         return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
7144 }
7145
7146 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer)
7147 {
7148         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
7149         struct page *page = rx_buffer->page;
7150
7151         /* avoid re-using remote pages */
7152         if (unlikely(igb_page_is_reserved(page)))
7153                 return false;
7154
7155 #if (PAGE_SIZE < 8192)
7156         /* if we are only owner of page we can reuse it */
7157         if (unlikely((page_ref_count(page) - pagecnt_bias) > 1))
7158                 return false;
7159 #else
7160 #define IGB_LAST_OFFSET \
7161         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
7162
7163         if (rx_buffer->page_offset > IGB_LAST_OFFSET)
7164                 return false;
7165 #endif
7166
7167         /* If we have drained the page fragment pool we need to update
7168          * the pagecnt_bias and page count so that we fully restock the
7169          * number of references the driver holds.
7170          */
7171         if (unlikely(!pagecnt_bias)) {
7172                 page_ref_add(page, USHRT_MAX);
7173                 rx_buffer->pagecnt_bias = USHRT_MAX;
7174         }
7175
7176         return true;
7177 }
7178
7179 /**
7180  *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
7181  *  @rx_ring: rx descriptor ring to transact packets on
7182  *  @rx_buffer: buffer containing page to add
7183  *  @skb: sk_buff to place the data into
7184  *  @size: size of buffer to be added
7185  *
7186  *  This function will add the data contained in rx_buffer->page to the skb.
7187  **/
7188 static void igb_add_rx_frag(struct igb_ring *rx_ring,
7189                             struct igb_rx_buffer *rx_buffer,
7190                             struct sk_buff *skb,
7191                             unsigned int size)
7192 {
7193 #if (PAGE_SIZE < 8192)
7194         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7195 #else
7196         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
7197                                 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
7198                                 SKB_DATA_ALIGN(size);
7199 #endif
7200         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
7201                         rx_buffer->page_offset, size, truesize);
7202 #if (PAGE_SIZE < 8192)
7203         rx_buffer->page_offset ^= truesize;
7204 #else
7205         rx_buffer->page_offset += truesize;
7206 #endif
7207 }
7208
7209 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
7210                                          struct igb_rx_buffer *rx_buffer,
7211                                          union e1000_adv_rx_desc *rx_desc,
7212                                          unsigned int size)
7213 {
7214         void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
7215 #if (PAGE_SIZE < 8192)
7216         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7217 #else
7218         unsigned int truesize = SKB_DATA_ALIGN(size);
7219 #endif
7220         unsigned int headlen;
7221         struct sk_buff *skb;
7222
7223         /* prefetch first cache line of first page */
7224         prefetch(va);
7225 #if L1_CACHE_BYTES < 128
7226         prefetch(va + L1_CACHE_BYTES);
7227 #endif
7228
7229         /* allocate a skb to store the frags */
7230         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
7231         if (unlikely(!skb))
7232                 return NULL;
7233
7234         if (unlikely(igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))) {
7235                 igb_ptp_rx_pktstamp(rx_ring->q_vector, va, skb);
7236                 va += IGB_TS_HDR_LEN;
7237                 size -= IGB_TS_HDR_LEN;
7238         }
7239
7240         /* Determine available headroom for copy */
7241         headlen = size;
7242         if (headlen > IGB_RX_HDR_LEN)
7243                 headlen = eth_get_headlen(va, IGB_RX_HDR_LEN);
7244
7245         /* align pull length to size of long to optimize memcpy performance */
7246         memcpy(__skb_put(skb, headlen), va, ALIGN(headlen, sizeof(long)));
7247
7248         /* update all of the pointers */
7249         size -= headlen;
7250         if (size) {
7251                 skb_add_rx_frag(skb, 0, rx_buffer->page,
7252                                 (va + headlen) - page_address(rx_buffer->page),
7253                                 size, truesize);
7254 #if (PAGE_SIZE < 8192)
7255                 rx_buffer->page_offset ^= truesize;
7256 #else
7257                 rx_buffer->page_offset += truesize;
7258 #endif
7259         } else {
7260                 rx_buffer->pagecnt_bias++;
7261         }
7262
7263         return skb;
7264 }
7265
7266 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
7267                                      struct igb_rx_buffer *rx_buffer,
7268                                      union e1000_adv_rx_desc *rx_desc,
7269                                      unsigned int size)
7270 {
7271         void *va = page_address(rx_buffer->page) + rx_buffer->page_offset;
7272 #if (PAGE_SIZE < 8192)
7273         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
7274 #else
7275         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
7276                                 SKB_DATA_ALIGN(IGB_SKB_PAD + size);
7277 #endif
7278         struct sk_buff *skb;
7279
7280         /* prefetch first cache line of first page */
7281         prefetch(va);
7282 #if L1_CACHE_BYTES < 128
7283         prefetch(va + L1_CACHE_BYTES);
7284 #endif
7285
7286         /* build an skb around the page buffer */
7287         skb = build_skb(va - IGB_SKB_PAD, truesize);
7288         if (unlikely(!skb))
7289                 return NULL;
7290
7291         /* update pointers within the skb to store the data */
7292         skb_reserve(skb, IGB_SKB_PAD);
7293         __skb_put(skb, size);
7294
7295         /* pull timestamp out of packet data */
7296         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
7297                 igb_ptp_rx_pktstamp(rx_ring->q_vector, skb->data, skb);
7298                 __skb_pull(skb, IGB_TS_HDR_LEN);
7299         }
7300
7301         /* update buffer offset */
7302 #if (PAGE_SIZE < 8192)
7303         rx_buffer->page_offset ^= truesize;
7304 #else
7305         rx_buffer->page_offset += truesize;
7306 #endif
7307
7308         return skb;
7309 }
7310
7311 static inline void igb_rx_checksum(struct igb_ring *ring,
7312                                    union e1000_adv_rx_desc *rx_desc,
7313                                    struct sk_buff *skb)
7314 {
7315         skb_checksum_none_assert(skb);
7316
7317         /* Ignore Checksum bit is set */
7318         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
7319                 return;
7320
7321         /* Rx checksum disabled via ethtool */
7322         if (!(ring->netdev->features & NETIF_F_RXCSUM))
7323                 return;
7324
7325         /* TCP/UDP checksum error bit is set */
7326         if (igb_test_staterr(rx_desc,
7327                              E1000_RXDEXT_STATERR_TCPE |
7328                              E1000_RXDEXT_STATERR_IPE)) {
7329                 /* work around errata with sctp packets where the TCPE aka
7330                  * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
7331                  * packets, (aka let the stack check the crc32c)
7332                  */
7333                 if (!((skb->len == 60) &&
7334                       test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
7335                         u64_stats_update_begin(&ring->rx_syncp);
7336                         ring->rx_stats.csum_err++;
7337                         u64_stats_update_end(&ring->rx_syncp);
7338                 }
7339                 /* let the stack verify checksum errors */
7340                 return;
7341         }
7342         /* It must be a TCP or UDP packet with a valid checksum */
7343         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
7344                                       E1000_RXD_STAT_UDPCS))
7345                 skb->ip_summed = CHECKSUM_UNNECESSARY;
7346
7347         dev_dbg(ring->dev, "cksum success: bits %08X\n",
7348                 le32_to_cpu(rx_desc->wb.upper.status_error));
7349 }
7350
7351 static inline void igb_rx_hash(struct igb_ring *ring,
7352                                union e1000_adv_rx_desc *rx_desc,
7353                                struct sk_buff *skb)
7354 {
7355         if (ring->netdev->features & NETIF_F_RXHASH)
7356                 skb_set_hash(skb,
7357                              le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
7358                              PKT_HASH_TYPE_L3);
7359 }
7360
7361 /**
7362  *  igb_is_non_eop - process handling of non-EOP buffers
7363  *  @rx_ring: Rx ring being processed
7364  *  @rx_desc: Rx descriptor for current buffer
7365  *  @skb: current socket buffer containing buffer in progress
7366  *
7367  *  This function updates next to clean.  If the buffer is an EOP buffer
7368  *  this function exits returning false, otherwise it will place the
7369  *  sk_buff in the next buffer to be chained and return true indicating
7370  *  that this is in fact a non-EOP buffer.
7371  **/
7372 static bool igb_is_non_eop(struct igb_ring *rx_ring,
7373                            union e1000_adv_rx_desc *rx_desc)
7374 {
7375         u32 ntc = rx_ring->next_to_clean + 1;
7376
7377         /* fetch, update, and store next to clean */
7378         ntc = (ntc < rx_ring->count) ? ntc : 0;
7379         rx_ring->next_to_clean = ntc;
7380
7381         prefetch(IGB_RX_DESC(rx_ring, ntc));
7382
7383         if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
7384                 return false;
7385
7386         return true;
7387 }
7388
7389 /**
7390  *  igb_cleanup_headers - Correct corrupted or empty headers
7391  *  @rx_ring: rx descriptor ring packet is being transacted on
7392  *  @rx_desc: pointer to the EOP Rx descriptor
7393  *  @skb: pointer to current skb being fixed
7394  *
7395  *  Address the case where we are pulling data in on pages only
7396  *  and as such no data is present in the skb header.
7397  *
7398  *  In addition if skb is not at least 60 bytes we need to pad it so that
7399  *  it is large enough to qualify as a valid Ethernet frame.
7400  *
7401  *  Returns true if an error was encountered and skb was freed.
7402  **/
7403 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
7404                                 union e1000_adv_rx_desc *rx_desc,
7405                                 struct sk_buff *skb)
7406 {
7407         if (unlikely((igb_test_staterr(rx_desc,
7408                                        E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
7409                 struct net_device *netdev = rx_ring->netdev;
7410                 if (!(netdev->features & NETIF_F_RXALL)) {
7411                         dev_kfree_skb_any(skb);
7412                         return true;
7413                 }
7414         }
7415
7416         /* if eth_skb_pad returns an error the skb was freed */
7417         if (eth_skb_pad(skb))
7418                 return true;
7419
7420         return false;
7421 }
7422
7423 /**
7424  *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
7425  *  @rx_ring: rx descriptor ring packet is being transacted on
7426  *  @rx_desc: pointer to the EOP Rx descriptor
7427  *  @skb: pointer to current skb being populated
7428  *
7429  *  This function checks the ring, descriptor, and packet information in
7430  *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
7431  *  other fields within the skb.
7432  **/
7433 static void igb_process_skb_fields(struct igb_ring *rx_ring,
7434                                    union e1000_adv_rx_desc *rx_desc,
7435                                    struct sk_buff *skb)
7436 {
7437         struct net_device *dev = rx_ring->netdev;
7438
7439         igb_rx_hash(rx_ring, rx_desc, skb);
7440
7441         igb_rx_checksum(rx_ring, rx_desc, skb);
7442
7443         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
7444             !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
7445                 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
7446
7447         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
7448             igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
7449                 u16 vid;
7450
7451                 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
7452                     test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
7453                         vid = be16_to_cpu(rx_desc->wb.upper.vlan);
7454                 else
7455                         vid = le16_to_cpu(rx_desc->wb.upper.vlan);
7456
7457                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
7458         }
7459
7460         skb_record_rx_queue(skb, rx_ring->queue_index);
7461
7462         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
7463 }
7464
7465 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
7466                                                const unsigned int size)
7467 {
7468         struct igb_rx_buffer *rx_buffer;
7469
7470         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
7471         prefetchw(rx_buffer->page);
7472
7473         /* we are reusing so sync this buffer for CPU use */
7474         dma_sync_single_range_for_cpu(rx_ring->dev,
7475                                       rx_buffer->dma,
7476                                       rx_buffer->page_offset,
7477                                       size,
7478                                       DMA_FROM_DEVICE);
7479
7480         rx_buffer->pagecnt_bias--;
7481
7482         return rx_buffer;
7483 }
7484
7485 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
7486                               struct igb_rx_buffer *rx_buffer)
7487 {
7488         if (igb_can_reuse_rx_page(rx_buffer)) {
7489                 /* hand second half of page back to the ring */
7490                 igb_reuse_rx_page(rx_ring, rx_buffer);
7491         } else {
7492                 /* We are not reusing the buffer so unmap it and free
7493                  * any references we are holding to it
7494                  */
7495                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
7496                                      igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
7497                                      IGB_RX_DMA_ATTR);
7498                 __page_frag_cache_drain(rx_buffer->page,
7499                                         rx_buffer->pagecnt_bias);
7500         }
7501
7502         /* clear contents of rx_buffer */
7503         rx_buffer->page = NULL;
7504 }
7505
7506 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
7507 {
7508         struct igb_ring *rx_ring = q_vector->rx.ring;
7509         struct sk_buff *skb = rx_ring->skb;
7510         unsigned int total_bytes = 0, total_packets = 0;
7511         u16 cleaned_count = igb_desc_unused(rx_ring);
7512
7513         while (likely(total_packets < budget)) {
7514                 union e1000_adv_rx_desc *rx_desc;
7515                 struct igb_rx_buffer *rx_buffer;
7516                 unsigned int size;
7517
7518                 /* return some buffers to hardware, one at a time is too slow */
7519                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
7520                         igb_alloc_rx_buffers(rx_ring, cleaned_count);
7521                         cleaned_count = 0;
7522                 }
7523
7524                 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
7525                 size = le16_to_cpu(rx_desc->wb.upper.length);
7526                 if (!size)
7527                         break;
7528
7529                 /* This memory barrier is needed to keep us from reading
7530                  * any other fields out of the rx_desc until we know the
7531                  * descriptor has been written back
7532                  */
7533                 dma_rmb();
7534
7535                 rx_buffer = igb_get_rx_buffer(rx_ring, size);
7536
7537                 /* retrieve a buffer from the ring */
7538                 if (skb)
7539                         igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
7540                 else if (ring_uses_build_skb(rx_ring))
7541                         skb = igb_build_skb(rx_ring, rx_buffer, rx_desc, size);
7542                 else
7543                         skb = igb_construct_skb(rx_ring, rx_buffer,
7544                                                 rx_desc, size);
7545
7546                 /* exit if we failed to retrieve a buffer */
7547                 if (!skb) {
7548                         rx_ring->rx_stats.alloc_failed++;
7549                         rx_buffer->pagecnt_bias++;
7550                         break;
7551                 }
7552
7553                 igb_put_rx_buffer(rx_ring, rx_buffer);
7554                 cleaned_count++;
7555
7556                 /* fetch next buffer in frame if non-eop */
7557                 if (igb_is_non_eop(rx_ring, rx_desc))
7558                         continue;
7559
7560                 /* verify the packet layout is correct */
7561                 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
7562                         skb = NULL;
7563                         continue;
7564                 }
7565
7566                 /* probably a little skewed due to removing CRC */
7567                 total_bytes += skb->len;
7568
7569                 /* populate checksum, timestamp, VLAN, and protocol */
7570                 igb_process_skb_fields(rx_ring, rx_desc, skb);
7571
7572                 napi_gro_receive(&q_vector->napi, skb);
7573
7574                 /* reset skb pointer */
7575                 skb = NULL;
7576
7577                 /* update budget accounting */
7578                 total_packets++;
7579         }
7580
7581         /* place incomplete frames back on ring for completion */
7582         rx_ring->skb = skb;
7583
7584         u64_stats_update_begin(&rx_ring->rx_syncp);
7585         rx_ring->rx_stats.packets += total_packets;
7586         rx_ring->rx_stats.bytes += total_bytes;
7587         u64_stats_update_end(&rx_ring->rx_syncp);
7588         q_vector->rx.total_packets += total_packets;
7589         q_vector->rx.total_bytes += total_bytes;
7590
7591         if (cleaned_count)
7592                 igb_alloc_rx_buffers(rx_ring, cleaned_count);
7593
7594         return total_packets;
7595 }
7596
7597 static inline unsigned int igb_rx_offset(struct igb_ring *rx_ring)
7598 {
7599         return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
7600 }
7601
7602 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
7603                                   struct igb_rx_buffer *bi)
7604 {
7605         struct page *page = bi->page;
7606         dma_addr_t dma;
7607
7608         /* since we are recycling buffers we should seldom need to alloc */
7609         if (likely(page))
7610                 return true;
7611
7612         /* alloc new page for storage */
7613         page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
7614         if (unlikely(!page)) {
7615                 rx_ring->rx_stats.alloc_failed++;
7616                 return false;
7617         }
7618
7619         /* map page for use */
7620         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
7621                                  igb_rx_pg_size(rx_ring),
7622                                  DMA_FROM_DEVICE,
7623                                  IGB_RX_DMA_ATTR);
7624
7625         /* if mapping failed free memory back to system since
7626          * there isn't much point in holding memory we can't use
7627          */
7628         if (dma_mapping_error(rx_ring->dev, dma)) {
7629                 __free_pages(page, igb_rx_pg_order(rx_ring));
7630
7631                 rx_ring->rx_stats.alloc_failed++;
7632                 return false;
7633         }
7634
7635         bi->dma = dma;
7636         bi->page = page;
7637         bi->page_offset = igb_rx_offset(rx_ring);
7638         bi->pagecnt_bias = 1;
7639
7640         return true;
7641 }
7642
7643 /**
7644  *  igb_alloc_rx_buffers - Replace used receive buffers; packet split
7645  *  @adapter: address of board private structure
7646  **/
7647 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
7648 {
7649         union e1000_adv_rx_desc *rx_desc;
7650         struct igb_rx_buffer *bi;
7651         u16 i = rx_ring->next_to_use;
7652         u16 bufsz;
7653
7654         /* nothing to do */
7655         if (!cleaned_count)
7656                 return;
7657
7658         rx_desc = IGB_RX_DESC(rx_ring, i);
7659         bi = &rx_ring->rx_buffer_info[i];
7660         i -= rx_ring->count;
7661
7662         bufsz = igb_rx_bufsz(rx_ring);
7663
7664         do {
7665                 if (!igb_alloc_mapped_page(rx_ring, bi))
7666                         break;
7667
7668                 /* sync the buffer for use by the device */
7669                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
7670                                                  bi->page_offset, bufsz,
7671                                                  DMA_FROM_DEVICE);
7672
7673                 /* Refresh the desc even if buffer_addrs didn't change
7674                  * because each write-back erases this info.
7675                  */
7676                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
7677
7678                 rx_desc++;
7679                 bi++;
7680                 i++;
7681                 if (unlikely(!i)) {
7682                         rx_desc = IGB_RX_DESC(rx_ring, 0);
7683                         bi = rx_ring->rx_buffer_info;
7684                         i -= rx_ring->count;
7685                 }
7686
7687                 /* clear the length for the next_to_use descriptor */
7688                 rx_desc->wb.upper.length = 0;
7689
7690                 cleaned_count--;
7691         } while (cleaned_count);
7692
7693         i += rx_ring->count;
7694
7695         if (rx_ring->next_to_use != i) {
7696                 /* record the next descriptor to use */
7697                 rx_ring->next_to_use = i;
7698
7699                 /* update next to alloc since we have filled the ring */
7700                 rx_ring->next_to_alloc = i;
7701
7702                 /* Force memory writes to complete before letting h/w
7703                  * know there are new descriptors to fetch.  (Only
7704                  * applicable for weak-ordered memory model archs,
7705                  * such as IA-64).
7706                  */
7707                 wmb();
7708                 writel(i, rx_ring->tail);
7709         }
7710 }
7711
7712 /**
7713  * igb_mii_ioctl -
7714  * @netdev:
7715  * @ifreq:
7716  * @cmd:
7717  **/
7718 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7719 {
7720         struct igb_adapter *adapter = netdev_priv(netdev);
7721         struct mii_ioctl_data *data = if_mii(ifr);
7722
7723         if (adapter->hw.phy.media_type != e1000_media_type_copper)
7724                 return -EOPNOTSUPP;
7725
7726         switch (cmd) {
7727         case SIOCGMIIPHY:
7728                 data->phy_id = adapter->hw.phy.addr;
7729                 break;
7730         case SIOCGMIIREG:
7731                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
7732                                      &data->val_out))
7733                         return -EIO;
7734                 break;
7735         case SIOCSMIIREG:
7736         default:
7737                 return -EOPNOTSUPP;
7738         }
7739         return 0;
7740 }
7741
7742 /**
7743  * igb_ioctl -
7744  * @netdev:
7745  * @ifreq:
7746  * @cmd:
7747  **/
7748 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
7749 {
7750         switch (cmd) {
7751         case SIOCGMIIPHY:
7752         case SIOCGMIIREG:
7753         case SIOCSMIIREG:
7754                 return igb_mii_ioctl(netdev, ifr, cmd);
7755         case SIOCGHWTSTAMP:
7756                 return igb_ptp_get_ts_config(netdev, ifr);
7757         case SIOCSHWTSTAMP:
7758                 return igb_ptp_set_ts_config(netdev, ifr);
7759         default:
7760                 return -EOPNOTSUPP;
7761         }
7762 }
7763
7764 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
7765 {
7766         struct igb_adapter *adapter = hw->back;
7767
7768         pci_read_config_word(adapter->pdev, reg, value);
7769 }
7770
7771 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
7772 {
7773         struct igb_adapter *adapter = hw->back;
7774
7775         pci_write_config_word(adapter->pdev, reg, *value);
7776 }
7777
7778 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7779 {
7780         struct igb_adapter *adapter = hw->back;
7781
7782         if (pcie_capability_read_word(adapter->pdev, reg, value))
7783                 return -E1000_ERR_CONFIG;
7784
7785         return 0;
7786 }
7787
7788 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
7789 {
7790         struct igb_adapter *adapter = hw->back;
7791
7792         if (pcie_capability_write_word(adapter->pdev, reg, *value))
7793                 return -E1000_ERR_CONFIG;
7794
7795         return 0;
7796 }
7797
7798 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
7799 {
7800         struct igb_adapter *adapter = netdev_priv(netdev);
7801         struct e1000_hw *hw = &adapter->hw;
7802         u32 ctrl, rctl;
7803         bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
7804
7805         if (enable) {
7806                 /* enable VLAN tag insert/strip */
7807                 ctrl = rd32(E1000_CTRL);
7808                 ctrl |= E1000_CTRL_VME;
7809                 wr32(E1000_CTRL, ctrl);
7810
7811                 /* Disable CFI check */
7812                 rctl = rd32(E1000_RCTL);
7813                 rctl &= ~E1000_RCTL_CFIEN;
7814                 wr32(E1000_RCTL, rctl);
7815         } else {
7816                 /* disable VLAN tag insert/strip */
7817                 ctrl = rd32(E1000_CTRL);
7818                 ctrl &= ~E1000_CTRL_VME;
7819                 wr32(E1000_CTRL, ctrl);
7820         }
7821
7822         igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
7823 }
7824
7825 static int igb_vlan_rx_add_vid(struct net_device *netdev,
7826                                __be16 proto, u16 vid)
7827 {
7828         struct igb_adapter *adapter = netdev_priv(netdev);
7829         struct e1000_hw *hw = &adapter->hw;
7830         int pf_id = adapter->vfs_allocated_count;
7831
7832         /* add the filter since PF can receive vlans w/o entry in vlvf */
7833         if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
7834                 igb_vfta_set(hw, vid, pf_id, true, !!vid);
7835
7836         set_bit(vid, adapter->active_vlans);
7837
7838         return 0;
7839 }
7840
7841 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
7842                                 __be16 proto, u16 vid)
7843 {
7844         struct igb_adapter *adapter = netdev_priv(netdev);
7845         int pf_id = adapter->vfs_allocated_count;
7846         struct e1000_hw *hw = &adapter->hw;
7847
7848         /* remove VID from filter table */
7849         if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
7850                 igb_vfta_set(hw, vid, pf_id, false, true);
7851
7852         clear_bit(vid, adapter->active_vlans);
7853
7854         return 0;
7855 }
7856
7857 static void igb_restore_vlan(struct igb_adapter *adapter)
7858 {
7859         u16 vid = 1;
7860
7861         igb_vlan_mode(adapter->netdev, adapter->netdev->features);
7862         igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
7863
7864         for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
7865                 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
7866 }
7867
7868 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
7869 {
7870         struct pci_dev *pdev = adapter->pdev;
7871         struct e1000_mac_info *mac = &adapter->hw.mac;
7872
7873         mac->autoneg = 0;
7874
7875         /* Make sure dplx is at most 1 bit and lsb of speed is not set
7876          * for the switch() below to work
7877          */
7878         if ((spd & 1) || (dplx & ~1))
7879                 goto err_inval;
7880
7881         /* Fiber NIC's only allow 1000 gbps Full duplex
7882          * and 100Mbps Full duplex for 100baseFx sfp
7883          */
7884         if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
7885                 switch (spd + dplx) {
7886                 case SPEED_10 + DUPLEX_HALF:
7887                 case SPEED_10 + DUPLEX_FULL:
7888                 case SPEED_100 + DUPLEX_HALF:
7889                         goto err_inval;
7890                 default:
7891                         break;
7892                 }
7893         }
7894
7895         switch (spd + dplx) {
7896         case SPEED_10 + DUPLEX_HALF:
7897                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
7898                 break;
7899         case SPEED_10 + DUPLEX_FULL:
7900                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
7901                 break;
7902         case SPEED_100 + DUPLEX_HALF:
7903                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
7904                 break;
7905         case SPEED_100 + DUPLEX_FULL:
7906                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
7907                 break;
7908         case SPEED_1000 + DUPLEX_FULL:
7909                 mac->autoneg = 1;
7910                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
7911                 break;
7912         case SPEED_1000 + DUPLEX_HALF: /* not supported */
7913         default:
7914                 goto err_inval;
7915         }
7916
7917         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
7918         adapter->hw.phy.mdix = AUTO_ALL_MODES;
7919
7920         return 0;
7921
7922 err_inval:
7923         dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
7924         return -EINVAL;
7925 }
7926
7927 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
7928                           bool runtime)
7929 {
7930         struct net_device *netdev = pci_get_drvdata(pdev);
7931         struct igb_adapter *adapter = netdev_priv(netdev);
7932         struct e1000_hw *hw = &adapter->hw;
7933         u32 ctrl, rctl, status;
7934         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
7935 #ifdef CONFIG_PM
7936         int retval = 0;
7937 #endif
7938
7939         rtnl_lock();
7940         netif_device_detach(netdev);
7941
7942         if (netif_running(netdev))
7943                 __igb_close(netdev, true);
7944
7945         igb_ptp_suspend(adapter);
7946
7947         igb_clear_interrupt_scheme(adapter);
7948         rtnl_unlock();
7949
7950 #ifdef CONFIG_PM
7951         retval = pci_save_state(pdev);
7952         if (retval)
7953                 return retval;
7954 #endif
7955
7956         status = rd32(E1000_STATUS);
7957         if (status & E1000_STATUS_LU)
7958                 wufc &= ~E1000_WUFC_LNKC;
7959
7960         if (wufc) {
7961                 igb_setup_rctl(adapter);
7962                 igb_set_rx_mode(netdev);
7963
7964                 /* turn on all-multi mode if wake on multicast is enabled */
7965                 if (wufc & E1000_WUFC_MC) {
7966                         rctl = rd32(E1000_RCTL);
7967                         rctl |= E1000_RCTL_MPE;
7968                         wr32(E1000_RCTL, rctl);
7969                 }
7970
7971                 ctrl = rd32(E1000_CTRL);
7972                 /* advertise wake from D3Cold */
7973                 #define E1000_CTRL_ADVD3WUC 0x00100000
7974                 /* phy power management enable */
7975                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
7976                 ctrl |= E1000_CTRL_ADVD3WUC;
7977                 wr32(E1000_CTRL, ctrl);
7978
7979                 /* Allow time for pending master requests to run */
7980                 igb_disable_pcie_master(hw);
7981
7982                 wr32(E1000_WUC, E1000_WUC_PME_EN);
7983                 wr32(E1000_WUFC, wufc);
7984         } else {
7985                 wr32(E1000_WUC, 0);
7986                 wr32(E1000_WUFC, 0);
7987         }
7988
7989         *enable_wake = wufc || adapter->en_mng_pt;
7990         if (!*enable_wake)
7991                 igb_power_down_link(adapter);
7992         else
7993                 igb_power_up_link(adapter);
7994
7995         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
7996          * would have already happened in close and is redundant.
7997          */
7998         igb_release_hw_control(adapter);
7999
8000         pci_disable_device(pdev);
8001
8002         return 0;
8003 }
8004
8005 static void igb_deliver_wake_packet(struct net_device *netdev)
8006 {
8007         struct igb_adapter *adapter = netdev_priv(netdev);
8008         struct e1000_hw *hw = &adapter->hw;
8009         struct sk_buff *skb;
8010         u32 wupl;
8011
8012         wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
8013
8014         /* WUPM stores only the first 128 bytes of the wake packet.
8015          * Read the packet only if we have the whole thing.
8016          */
8017         if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
8018                 return;
8019
8020         skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
8021         if (!skb)
8022                 return;
8023
8024         skb_put(skb, wupl);
8025
8026         /* Ensure reads are 32-bit aligned */
8027         wupl = roundup(wupl, 4);
8028
8029         memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
8030
8031         skb->protocol = eth_type_trans(skb, netdev);
8032         netif_rx(skb);
8033 }
8034
8035 static int __maybe_unused igb_suspend(struct device *dev)
8036 {
8037         int retval;
8038         bool wake;
8039         struct pci_dev *pdev = to_pci_dev(dev);
8040
8041         retval = __igb_shutdown(pdev, &wake, 0);
8042         if (retval)
8043                 return retval;
8044
8045         if (wake) {
8046                 pci_prepare_to_sleep(pdev);
8047         } else {
8048                 pci_wake_from_d3(pdev, false);
8049                 pci_set_power_state(pdev, PCI_D3hot);
8050         }
8051
8052         return 0;
8053 }
8054
8055 static int __maybe_unused igb_resume(struct device *dev)
8056 {
8057         struct pci_dev *pdev = to_pci_dev(dev);
8058         struct net_device *netdev = pci_get_drvdata(pdev);
8059         struct igb_adapter *adapter = netdev_priv(netdev);
8060         struct e1000_hw *hw = &adapter->hw;
8061         u32 err, val;
8062
8063         pci_set_power_state(pdev, PCI_D0);
8064         pci_restore_state(pdev);
8065         pci_save_state(pdev);
8066
8067         if (!pci_device_is_present(pdev))
8068                 return -ENODEV;
8069         err = pci_enable_device_mem(pdev);
8070         if (err) {
8071                 dev_err(&pdev->dev,
8072                         "igb: Cannot enable PCI device from suspend\n");
8073                 return err;
8074         }
8075         pci_set_master(pdev);
8076
8077         pci_enable_wake(pdev, PCI_D3hot, 0);
8078         pci_enable_wake(pdev, PCI_D3cold, 0);
8079
8080         if (igb_init_interrupt_scheme(adapter, true)) {
8081                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8082                 return -ENOMEM;
8083         }
8084
8085         igb_reset(adapter);
8086
8087         /* let the f/w know that the h/w is now under the control of the
8088          * driver.
8089          */
8090         igb_get_hw_control(adapter);
8091
8092         val = rd32(E1000_WUS);
8093         if (val & WAKE_PKT_WUS)
8094                 igb_deliver_wake_packet(netdev);
8095
8096         wr32(E1000_WUS, ~0);
8097
8098         rtnl_lock();
8099         if (!err && netif_running(netdev))
8100                 err = __igb_open(netdev, true);
8101
8102         if (!err)
8103                 netif_device_attach(netdev);
8104         rtnl_unlock();
8105
8106         return err;
8107 }
8108
8109 static int __maybe_unused igb_runtime_idle(struct device *dev)
8110 {
8111         struct pci_dev *pdev = to_pci_dev(dev);
8112         struct net_device *netdev = pci_get_drvdata(pdev);
8113         struct igb_adapter *adapter = netdev_priv(netdev);
8114
8115         if (!igb_has_link(adapter))
8116                 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
8117
8118         return -EBUSY;
8119 }
8120
8121 static int __maybe_unused igb_runtime_suspend(struct device *dev)
8122 {
8123         struct pci_dev *pdev = to_pci_dev(dev);
8124         int retval;
8125         bool wake;
8126
8127         retval = __igb_shutdown(pdev, &wake, 1);
8128         if (retval)
8129                 return retval;
8130
8131         if (wake) {
8132                 pci_prepare_to_sleep(pdev);
8133         } else {
8134                 pci_wake_from_d3(pdev, false);
8135                 pci_set_power_state(pdev, PCI_D3hot);
8136         }
8137
8138         return 0;
8139 }
8140
8141 static int __maybe_unused igb_runtime_resume(struct device *dev)
8142 {
8143         return igb_resume(dev);
8144 }
8145
8146 static void igb_shutdown(struct pci_dev *pdev)
8147 {
8148         bool wake;
8149
8150         __igb_shutdown(pdev, &wake, 0);
8151
8152         if (system_state == SYSTEM_POWER_OFF) {
8153                 pci_wake_from_d3(pdev, wake);
8154                 pci_set_power_state(pdev, PCI_D3hot);
8155         }
8156 }
8157
8158 #ifdef CONFIG_PCI_IOV
8159 static int igb_sriov_reinit(struct pci_dev *dev)
8160 {
8161         struct net_device *netdev = pci_get_drvdata(dev);
8162         struct igb_adapter *adapter = netdev_priv(netdev);
8163         struct pci_dev *pdev = adapter->pdev;
8164
8165         rtnl_lock();
8166
8167         if (netif_running(netdev))
8168                 igb_close(netdev);
8169         else
8170                 igb_reset(adapter);
8171
8172         igb_clear_interrupt_scheme(adapter);
8173
8174         igb_init_queue_configuration(adapter);
8175
8176         if (igb_init_interrupt_scheme(adapter, true)) {
8177                 rtnl_unlock();
8178                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8179                 return -ENOMEM;
8180         }
8181
8182         if (netif_running(netdev))
8183                 igb_open(netdev);
8184
8185         rtnl_unlock();
8186
8187         return 0;
8188 }
8189
8190 static int igb_pci_disable_sriov(struct pci_dev *dev)
8191 {
8192         int err = igb_disable_sriov(dev);
8193
8194         if (!err)
8195                 err = igb_sriov_reinit(dev);
8196
8197         return err;
8198 }
8199
8200 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
8201 {
8202         int err = igb_enable_sriov(dev, num_vfs);
8203
8204         if (err)
8205                 goto out;
8206
8207         err = igb_sriov_reinit(dev);
8208         if (!err)
8209                 return num_vfs;
8210
8211 out:
8212         return err;
8213 }
8214
8215 #endif
8216 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
8217 {
8218 #ifdef CONFIG_PCI_IOV
8219         if (num_vfs == 0)
8220                 return igb_pci_disable_sriov(dev);
8221         else
8222                 return igb_pci_enable_sriov(dev, num_vfs);
8223 #endif
8224         return 0;
8225 }
8226
8227 #ifdef CONFIG_NET_POLL_CONTROLLER
8228 /* Polling 'interrupt' - used by things like netconsole to send skbs
8229  * without having to re-enable interrupts. It's not called while
8230  * the interrupt routine is executing.
8231  */
8232 static void igb_netpoll(struct net_device *netdev)
8233 {
8234         struct igb_adapter *adapter = netdev_priv(netdev);
8235         struct e1000_hw *hw = &adapter->hw;
8236         struct igb_q_vector *q_vector;
8237         int i;
8238
8239         for (i = 0; i < adapter->num_q_vectors; i++) {
8240                 q_vector = adapter->q_vector[i];
8241                 if (adapter->flags & IGB_FLAG_HAS_MSIX)
8242                         wr32(E1000_EIMC, q_vector->eims_value);
8243                 else
8244                         igb_irq_disable(adapter);
8245                 napi_schedule(&q_vector->napi);
8246         }
8247 }
8248 #endif /* CONFIG_NET_POLL_CONTROLLER */
8249
8250 /**
8251  *  igb_io_error_detected - called when PCI error is detected
8252  *  @pdev: Pointer to PCI device
8253  *  @state: The current pci connection state
8254  *
8255  *  This function is called after a PCI bus error affecting
8256  *  this device has been detected.
8257  **/
8258 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
8259                                               pci_channel_state_t state)
8260 {
8261         struct net_device *netdev = pci_get_drvdata(pdev);
8262         struct igb_adapter *adapter = netdev_priv(netdev);
8263
8264         netif_device_detach(netdev);
8265
8266         if (state == pci_channel_io_perm_failure)
8267                 return PCI_ERS_RESULT_DISCONNECT;
8268
8269         if (netif_running(netdev))
8270                 igb_down(adapter);
8271         pci_disable_device(pdev);
8272
8273         /* Request a slot slot reset. */
8274         return PCI_ERS_RESULT_NEED_RESET;
8275 }
8276
8277 /**
8278  *  igb_io_slot_reset - called after the pci bus has been reset.
8279  *  @pdev: Pointer to PCI device
8280  *
8281  *  Restart the card from scratch, as if from a cold-boot. Implementation
8282  *  resembles the first-half of the igb_resume routine.
8283  **/
8284 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
8285 {
8286         struct net_device *netdev = pci_get_drvdata(pdev);
8287         struct igb_adapter *adapter = netdev_priv(netdev);
8288         struct e1000_hw *hw = &adapter->hw;
8289         pci_ers_result_t result;
8290         int err;
8291
8292         if (pci_enable_device_mem(pdev)) {
8293                 dev_err(&pdev->dev,
8294                         "Cannot re-enable PCI device after reset.\n");
8295                 result = PCI_ERS_RESULT_DISCONNECT;
8296         } else {
8297                 pci_set_master(pdev);
8298                 pci_restore_state(pdev);
8299                 pci_save_state(pdev);
8300
8301                 pci_enable_wake(pdev, PCI_D3hot, 0);
8302                 pci_enable_wake(pdev, PCI_D3cold, 0);
8303
8304                 /* In case of PCI error, adapter lose its HW address
8305                  * so we should re-assign it here.
8306                  */
8307                 hw->hw_addr = adapter->io_addr;
8308
8309                 igb_reset(adapter);
8310                 wr32(E1000_WUS, ~0);
8311                 result = PCI_ERS_RESULT_RECOVERED;
8312         }
8313
8314         err = pci_cleanup_aer_uncorrect_error_status(pdev);
8315         if (err) {
8316                 dev_err(&pdev->dev,
8317                         "pci_cleanup_aer_uncorrect_error_status failed 0x%0x\n",
8318                         err);
8319                 /* non-fatal, continue */
8320         }
8321
8322         return result;
8323 }
8324
8325 /**
8326  *  igb_io_resume - called when traffic can start flowing again.
8327  *  @pdev: Pointer to PCI device
8328  *
8329  *  This callback is called when the error recovery driver tells us that
8330  *  its OK to resume normal operation. Implementation resembles the
8331  *  second-half of the igb_resume routine.
8332  */
8333 static void igb_io_resume(struct pci_dev *pdev)
8334 {
8335         struct net_device *netdev = pci_get_drvdata(pdev);
8336         struct igb_adapter *adapter = netdev_priv(netdev);
8337
8338         if (netif_running(netdev)) {
8339                 if (igb_up(adapter)) {
8340                         dev_err(&pdev->dev, "igb_up failed after reset\n");
8341                         return;
8342                 }
8343         }
8344
8345         netif_device_attach(netdev);
8346
8347         /* let the f/w know that the h/w is now under the control of the
8348          * driver.
8349          */
8350         igb_get_hw_control(adapter);
8351 }
8352
8353 /**
8354  *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
8355  *  @adapter: Pointer to adapter structure
8356  *  @index: Index of the RAR entry which need to be synced with MAC table
8357  **/
8358 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
8359 {
8360         struct e1000_hw *hw = &adapter->hw;
8361         u32 rar_low, rar_high;
8362         u8 *addr = adapter->mac_table[index].addr;
8363
8364         /* HW expects these to be in network order when they are plugged
8365          * into the registers which are little endian.  In order to guarantee
8366          * that ordering we need to do an leXX_to_cpup here in order to be
8367          * ready for the byteswap that occurs with writel
8368          */
8369         rar_low = le32_to_cpup((__le32 *)(addr));
8370         rar_high = le16_to_cpup((__le16 *)(addr + 4));
8371
8372         /* Indicate to hardware the Address is Valid. */
8373         if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
8374                 rar_high |= E1000_RAH_AV;
8375
8376                 if (hw->mac.type == e1000_82575)
8377                         rar_high |= E1000_RAH_POOL_1 *
8378                                     adapter->mac_table[index].queue;
8379                 else
8380                         rar_high |= E1000_RAH_POOL_1 <<
8381                                     adapter->mac_table[index].queue;
8382         }
8383
8384         wr32(E1000_RAL(index), rar_low);
8385         wrfl();
8386         wr32(E1000_RAH(index), rar_high);
8387         wrfl();
8388 }
8389
8390 static int igb_set_vf_mac(struct igb_adapter *adapter,
8391                           int vf, unsigned char *mac_addr)
8392 {
8393         struct e1000_hw *hw = &adapter->hw;
8394         /* VF MAC addresses start at end of receive addresses and moves
8395          * towards the first, as a result a collision should not be possible
8396          */
8397         int rar_entry = hw->mac.rar_entry_count - (vf + 1);
8398         unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
8399
8400         ether_addr_copy(vf_mac_addr, mac_addr);
8401         ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
8402         adapter->mac_table[rar_entry].queue = vf;
8403         adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
8404         igb_rar_set_index(adapter, rar_entry);
8405
8406         return 0;
8407 }
8408
8409 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
8410 {
8411         struct igb_adapter *adapter = netdev_priv(netdev);
8412         if (!is_valid_ether_addr(mac) || (vf >= adapter->vfs_allocated_count))
8413                 return -EINVAL;
8414         adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
8415         dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n", mac, vf);
8416         dev_info(&adapter->pdev->dev,
8417                  "Reload the VF driver to make this change effective.");
8418         if (test_bit(__IGB_DOWN, &adapter->state)) {
8419                 dev_warn(&adapter->pdev->dev,
8420                          "The VF MAC address has been set, but the PF device is not up.\n");
8421                 dev_warn(&adapter->pdev->dev,
8422                          "Bring the PF device up before attempting to use the VF device.\n");
8423         }
8424         return igb_set_vf_mac(adapter, vf, mac);
8425 }
8426
8427 static int igb_link_mbps(int internal_link_speed)
8428 {
8429         switch (internal_link_speed) {
8430         case SPEED_100:
8431                 return 100;
8432         case SPEED_1000:
8433                 return 1000;
8434         default:
8435                 return 0;
8436         }
8437 }
8438
8439 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
8440                                   int link_speed)
8441 {
8442         int rf_dec, rf_int;
8443         u32 bcnrc_val;
8444
8445         if (tx_rate != 0) {
8446                 /* Calculate the rate factor values to set */
8447                 rf_int = link_speed / tx_rate;
8448                 rf_dec = (link_speed - (rf_int * tx_rate));
8449                 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
8450                          tx_rate;
8451
8452                 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
8453                 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
8454                               E1000_RTTBCNRC_RF_INT_MASK);
8455                 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
8456         } else {
8457                 bcnrc_val = 0;
8458         }
8459
8460         wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
8461         /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
8462          * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
8463          */
8464         wr32(E1000_RTTBCNRM, 0x14);
8465         wr32(E1000_RTTBCNRC, bcnrc_val);
8466 }
8467
8468 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
8469 {
8470         int actual_link_speed, i;
8471         bool reset_rate = false;
8472
8473         /* VF TX rate limit was not set or not supported */
8474         if ((adapter->vf_rate_link_speed == 0) ||
8475             (adapter->hw.mac.type != e1000_82576))
8476                 return;
8477
8478         actual_link_speed = igb_link_mbps(adapter->link_speed);
8479         if (actual_link_speed != adapter->vf_rate_link_speed) {
8480                 reset_rate = true;
8481                 adapter->vf_rate_link_speed = 0;
8482                 dev_info(&adapter->pdev->dev,
8483                          "Link speed has been changed. VF Transmit rate is disabled\n");
8484         }
8485
8486         for (i = 0; i < adapter->vfs_allocated_count; i++) {
8487                 if (reset_rate)
8488                         adapter->vf_data[i].tx_rate = 0;
8489
8490                 igb_set_vf_rate_limit(&adapter->hw, i,
8491                                       adapter->vf_data[i].tx_rate,
8492                                       actual_link_speed);
8493         }
8494 }
8495
8496 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
8497                              int min_tx_rate, int max_tx_rate)
8498 {
8499         struct igb_adapter *adapter = netdev_priv(netdev);
8500         struct e1000_hw *hw = &adapter->hw;
8501         int actual_link_speed;
8502
8503         if (hw->mac.type != e1000_82576)
8504                 return -EOPNOTSUPP;
8505
8506         if (min_tx_rate)
8507                 return -EINVAL;
8508
8509         actual_link_speed = igb_link_mbps(adapter->link_speed);
8510         if ((vf >= adapter->vfs_allocated_count) ||
8511             (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
8512             (max_tx_rate < 0) ||
8513             (max_tx_rate > actual_link_speed))
8514                 return -EINVAL;
8515
8516         adapter->vf_rate_link_speed = actual_link_speed;
8517         adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
8518         igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
8519
8520         return 0;
8521 }
8522
8523 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
8524                                    bool setting)
8525 {
8526         struct igb_adapter *adapter = netdev_priv(netdev);
8527         struct e1000_hw *hw = &adapter->hw;
8528         u32 reg_val, reg_offset;
8529
8530         if (!adapter->vfs_allocated_count)
8531                 return -EOPNOTSUPP;
8532
8533         if (vf >= adapter->vfs_allocated_count)
8534                 return -EINVAL;
8535
8536         reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
8537         reg_val = rd32(reg_offset);
8538         if (setting)
8539                 reg_val |= (BIT(vf) |
8540                             BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
8541         else
8542                 reg_val &= ~(BIT(vf) |
8543                              BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
8544         wr32(reg_offset, reg_val);
8545
8546         adapter->vf_data[vf].spoofchk_enabled = setting;
8547         return 0;
8548 }
8549
8550 static int igb_ndo_get_vf_config(struct net_device *netdev,
8551                                  int vf, struct ifla_vf_info *ivi)
8552 {
8553         struct igb_adapter *adapter = netdev_priv(netdev);
8554         if (vf >= adapter->vfs_allocated_count)
8555                 return -EINVAL;
8556         ivi->vf = vf;
8557         memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
8558         ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
8559         ivi->min_tx_rate = 0;
8560         ivi->vlan = adapter->vf_data[vf].pf_vlan;
8561         ivi->qos = adapter->vf_data[vf].pf_qos;
8562         ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
8563         return 0;
8564 }
8565
8566 static void igb_vmm_control(struct igb_adapter *adapter)
8567 {
8568         struct e1000_hw *hw = &adapter->hw;
8569         u32 reg;
8570
8571         switch (hw->mac.type) {
8572         case e1000_82575:
8573         case e1000_i210:
8574         case e1000_i211:
8575         case e1000_i354:
8576         default:
8577                 /* replication is not supported for 82575 */
8578                 return;
8579         case e1000_82576:
8580                 /* notify HW that the MAC is adding vlan tags */
8581                 reg = rd32(E1000_DTXCTL);
8582                 reg |= E1000_DTXCTL_VLAN_ADDED;
8583                 wr32(E1000_DTXCTL, reg);
8584                 /* Fall through */
8585         case e1000_82580:
8586                 /* enable replication vlan tag stripping */
8587                 reg = rd32(E1000_RPLOLR);
8588                 reg |= E1000_RPLOLR_STRVLAN;
8589                 wr32(E1000_RPLOLR, reg);
8590                 /* Fall through */
8591         case e1000_i350:
8592                 /* none of the above registers are supported by i350 */
8593                 break;
8594         }
8595
8596         if (adapter->vfs_allocated_count) {
8597                 igb_vmdq_set_loopback_pf(hw, true);
8598                 igb_vmdq_set_replication_pf(hw, true);
8599                 igb_vmdq_set_anti_spoofing_pf(hw, true,
8600                                               adapter->vfs_allocated_count);
8601         } else {
8602                 igb_vmdq_set_loopback_pf(hw, false);
8603                 igb_vmdq_set_replication_pf(hw, false);
8604         }
8605 }
8606
8607 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
8608 {
8609         struct e1000_hw *hw = &adapter->hw;
8610         u32 dmac_thr;
8611         u16 hwm;
8612
8613         if (hw->mac.type > e1000_82580) {
8614                 if (adapter->flags & IGB_FLAG_DMAC) {
8615                         u32 reg;
8616
8617                         /* force threshold to 0. */
8618                         wr32(E1000_DMCTXTH, 0);
8619
8620                         /* DMA Coalescing high water mark needs to be greater
8621                          * than the Rx threshold. Set hwm to PBA - max frame
8622                          * size in 16B units, capping it at PBA - 6KB.
8623                          */
8624                         hwm = 64 * (pba - 6);
8625                         reg = rd32(E1000_FCRTC);
8626                         reg &= ~E1000_FCRTC_RTH_COAL_MASK;
8627                         reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
8628                                 & E1000_FCRTC_RTH_COAL_MASK);
8629                         wr32(E1000_FCRTC, reg);
8630
8631                         /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
8632                          * frame size, capping it at PBA - 10KB.
8633                          */
8634                         dmac_thr = pba - 10;
8635                         reg = rd32(E1000_DMACR);
8636                         reg &= ~E1000_DMACR_DMACTHR_MASK;
8637                         reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
8638                                 & E1000_DMACR_DMACTHR_MASK);
8639
8640                         /* transition to L0x or L1 if available..*/
8641                         reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
8642
8643                         /* watchdog timer= +-1000 usec in 32usec intervals */
8644                         reg |= (1000 >> 5);
8645
8646                         /* Disable BMC-to-OS Watchdog Enable */
8647                         if (hw->mac.type != e1000_i354)
8648                                 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
8649
8650                         wr32(E1000_DMACR, reg);
8651
8652                         /* no lower threshold to disable
8653                          * coalescing(smart fifb)-UTRESH=0
8654                          */
8655                         wr32(E1000_DMCRTRH, 0);
8656
8657                         reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
8658
8659                         wr32(E1000_DMCTLX, reg);
8660
8661                         /* free space in tx packet buffer to wake from
8662                          * DMA coal
8663                          */
8664                         wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
8665                              (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
8666
8667                         /* make low power state decision controlled
8668                          * by DMA coal
8669                          */
8670                         reg = rd32(E1000_PCIEMISC);
8671                         reg &= ~E1000_PCIEMISC_LX_DECISION;
8672                         wr32(E1000_PCIEMISC, reg);
8673                 } /* endif adapter->dmac is not disabled */
8674         } else if (hw->mac.type == e1000_82580) {
8675                 u32 reg = rd32(E1000_PCIEMISC);
8676
8677                 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
8678                 wr32(E1000_DMACR, 0);
8679         }
8680 }
8681
8682 /**
8683  *  igb_read_i2c_byte - Reads 8 bit word over I2C
8684  *  @hw: pointer to hardware structure
8685  *  @byte_offset: byte offset to read
8686  *  @dev_addr: device address
8687  *  @data: value read
8688  *
8689  *  Performs byte read operation over I2C interface at
8690  *  a specified device address.
8691  **/
8692 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8693                       u8 dev_addr, u8 *data)
8694 {
8695         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8696         struct i2c_client *this_client = adapter->i2c_client;
8697         s32 status;
8698         u16 swfw_mask = 0;
8699
8700         if (!this_client)
8701                 return E1000_ERR_I2C;
8702
8703         swfw_mask = E1000_SWFW_PHY0_SM;
8704
8705         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
8706                 return E1000_ERR_SWFW_SYNC;
8707
8708         status = i2c_smbus_read_byte_data(this_client, byte_offset);
8709         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8710
8711         if (status < 0)
8712                 return E1000_ERR_I2C;
8713         else {
8714                 *data = status;
8715                 return 0;
8716         }
8717 }
8718
8719 /**
8720  *  igb_write_i2c_byte - Writes 8 bit word over I2C
8721  *  @hw: pointer to hardware structure
8722  *  @byte_offset: byte offset to write
8723  *  @dev_addr: device address
8724  *  @data: value to write
8725  *
8726  *  Performs byte write operation over I2C interface at
8727  *  a specified device address.
8728  **/
8729 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
8730                        u8 dev_addr, u8 data)
8731 {
8732         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
8733         struct i2c_client *this_client = adapter->i2c_client;
8734         s32 status;
8735         u16 swfw_mask = E1000_SWFW_PHY0_SM;
8736
8737         if (!this_client)
8738                 return E1000_ERR_I2C;
8739
8740         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
8741                 return E1000_ERR_SWFW_SYNC;
8742         status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
8743         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
8744
8745         if (status)
8746                 return E1000_ERR_I2C;
8747         else
8748                 return 0;
8749
8750 }
8751
8752 int igb_reinit_queues(struct igb_adapter *adapter)
8753 {
8754         struct net_device *netdev = adapter->netdev;
8755         struct pci_dev *pdev = adapter->pdev;
8756         int err = 0;
8757
8758         if (netif_running(netdev))
8759                 igb_close(netdev);
8760
8761         igb_reset_interrupt_capability(adapter);
8762
8763         if (igb_init_interrupt_scheme(adapter, true)) {
8764                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
8765                 return -ENOMEM;
8766         }
8767
8768         if (netif_running(netdev))
8769                 err = igb_open(netdev);
8770
8771         return err;
8772 }
8773
8774 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
8775 {
8776         struct igb_nfc_filter *rule;
8777
8778         spin_lock(&adapter->nfc_lock);
8779
8780         hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
8781                 igb_erase_filter(adapter, rule);
8782
8783         spin_unlock(&adapter->nfc_lock);
8784 }
8785
8786 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
8787 {
8788         struct igb_nfc_filter *rule;
8789
8790         spin_lock(&adapter->nfc_lock);
8791
8792         hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
8793                 igb_add_filter(adapter, rule);
8794
8795         spin_unlock(&adapter->nfc_lock);
8796 }
8797 /* igb_main.c */