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

   1 // SPDX-License-Identifier: GPL-2.0
   2 /* Copyright(c) 2007 - 2018 Intel Corporation. */
   3
   4 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
   5
   6 #include <linux/module.h>
   7 #include <linux/types.h>
   8 #include <linux/init.h>
   9 #include <linux/bitops.h>
  10 #include <linux/vmalloc.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/netdevice.h>
  13 #include <linux/ipv6.h>
  14 #include <linux/slab.h>
  15 #include <net/checksum.h>
  16 #include <net/ip6_checksum.h>
  17 #include <net/pkt_sched.h>
  18 #include <net/pkt_cls.h>
  19 #include <linux/net_tstamp.h>
  20 #include <linux/mii.h>
  21 #include <linux/ethtool.h>
  22 #include <linux/if.h>
  23 #include <linux/if_vlan.h>
  24 #include <linux/pci.h>
  25 #include <linux/delay.h>
  26 #include <linux/interrupt.h>
  27 #include <linux/ip.h>
  28 #include <linux/tcp.h>
  29 #include <linux/sctp.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/aer.h>
  32 #include <linux/prefetch.h>
  33 #include <linux/bpf.h>
  34 #include <linux/bpf_trace.h>
  35 #include <linux/pm_runtime.h>
  36 #include <linux/etherdevice.h>
  37 #ifdef CONFIG_IGB_DCA
  38 #include <linux/dca.h>
  39 #endif
  40 #include <linux/i2c.h>
  41 #include "igb.h"
  42
  43 enum queue_mode {
  44         QUEUE_MODE_STRICT_PRIORITY,
  45         QUEUE_MODE_STREAM_RESERVATION,
  46 };
  47
  48 enum tx_queue_prio {
  49         TX_QUEUE_PRIO_HIGH,
  50         TX_QUEUE_PRIO_LOW,
  51 };
  52
  53 char igb_driver_name[] = "igb";
  54 static const char igb_driver_string[] =
  55                                 "Intel(R) Gigabit Ethernet Network Driver";
  56 static const char igb_copyright[] =
  57                                 "Copyright (c) 2007-2014 Intel Corporation.";
  58
  59 static const struct e1000_info *igb_info_tbl[] = {
  60         [board_82575] = &e1000_82575_info,
  61 };
  62
  63 static const struct pci_device_id igb_pci_tbl[] = {
  64         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_1GBPS) },
  65         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_SGMII) },
  66         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I354_BACKPLANE_2_5GBPS) },
  67         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I211_COPPER), board_82575 },
  68         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER), board_82575 },
  69         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_FIBER), board_82575 },
  70         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES), board_82575 },
  71         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SGMII), board_82575 },
  72         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_COPPER_FLASHLESS), board_82575 },
  73         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I210_SERDES_FLASHLESS), board_82575 },
  74         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_COPPER), board_82575 },
  75         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_FIBER), board_82575 },
  76         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SERDES), board_82575 },
  77         { PCI_VDEVICE(INTEL, E1000_DEV_ID_I350_SGMII), board_82575 },
  78         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER), board_82575 },
  79         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_FIBER), board_82575 },
  80         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_QUAD_FIBER), board_82575 },
  81         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SERDES), board_82575 },
  82         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_SGMII), board_82575 },
  83         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82580_COPPER_DUAL), board_82575 },
  84         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SGMII), board_82575 },
  85         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SERDES), board_82575 },
  86         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_BACKPLANE), board_82575 },
  87         { PCI_VDEVICE(INTEL, E1000_DEV_ID_DH89XXCC_SFP), board_82575 },
  88         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
  89         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
  90         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS_SERDES), board_82575 },
  91         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
  92         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
  93         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES_QUAD), board_82575 },
  94         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER_ET2), board_82575 },
  95         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
  96         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
  97         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
  98         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
  99         /* required last entry */
 100         {0, }
 101 };
 102
 103 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
 104
 105 static int igb_setup_all_tx_resources(struct igb_adapter *);
 106 static int igb_setup_all_rx_resources(struct igb_adapter *);
 107 static void igb_free_all_tx_resources(struct igb_adapter *);
 108 static void igb_free_all_rx_resources(struct igb_adapter *);
 109 static void igb_setup_mrqc(struct igb_adapter *);
 110 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
 111 static void igb_remove(struct pci_dev *pdev);
 112 static int igb_sw_init(struct igb_adapter *);
 113 int igb_open(struct net_device *);
 114 int igb_close(struct net_device *);
 115 static void igb_configure(struct igb_adapter *);
 116 static void igb_configure_tx(struct igb_adapter *);
 117 static void igb_configure_rx(struct igb_adapter *);
 118 static void igb_clean_all_tx_rings(struct igb_adapter *);
 119 static void igb_clean_all_rx_rings(struct igb_adapter *);
 120 static void igb_clean_tx_ring(struct igb_ring *);
 121 static void igb_clean_rx_ring(struct igb_ring *);
 122 static void igb_set_rx_mode(struct net_device *);
 123 static void igb_update_phy_info(struct timer_list *);
 124 static void igb_watchdog(struct timer_list *);
 125 static void igb_watchdog_task(struct work_struct *);
 126 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb, struct net_device *);
 127 static void igb_get_stats64(struct net_device *dev,
 128                             struct rtnl_link_stats64 *stats);
 129 static int igb_change_mtu(struct net_device *, int);
 130 static int igb_set_mac(struct net_device *, void *);
 131 static void igb_set_uta(struct igb_adapter *adapter, bool set);
 132 static irqreturn_t igb_intr(int irq, void *);
 133 static irqreturn_t igb_intr_msi(int irq, void *);
 134 static irqreturn_t igb_msix_other(int irq, void *);
 135 static irqreturn_t igb_msix_ring(int irq, void *);
 136 #ifdef CONFIG_IGB_DCA
 137 static void igb_update_dca(struct igb_q_vector *);
 138 static void igb_setup_dca(struct igb_adapter *);
 139 #endif /* CONFIG_IGB_DCA */
 140 static int igb_poll(struct napi_struct *, int);
 141 static bool igb_clean_tx_irq(struct igb_q_vector *, int);
 142 static int igb_clean_rx_irq(struct igb_q_vector *, int);
 143 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 144 static void igb_tx_timeout(struct net_device *, unsigned int txqueue);
 145 static void igb_reset_task(struct work_struct *);
 146 static void igb_vlan_mode(struct net_device *netdev,
 147                           netdev_features_t features);
 148 static int igb_vlan_rx_add_vid(struct net_device *, __be16, u16);
 149 static int igb_vlan_rx_kill_vid(struct net_device *, __be16, u16);
 150 static void igb_restore_vlan(struct igb_adapter *);
 151 static void igb_rar_set_index(struct igb_adapter *, u32);
 152 static void igb_ping_all_vfs(struct igb_adapter *);
 153 static void igb_msg_task(struct igb_adapter *);
 154 static void igb_vmm_control(struct igb_adapter *);
 155 static int igb_set_vf_mac(struct igb_adapter *, int, unsigned char *);
 156 static void igb_flush_mac_table(struct igb_adapter *);
 157 static int igb_available_rars(struct igb_adapter *, u8);
 158 static void igb_set_default_mac_filter(struct igb_adapter *);
 159 static int igb_uc_sync(struct net_device *, const unsigned char *);
 160 static int igb_uc_unsync(struct net_device *, const unsigned char *);
 161 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 162 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac);
 163 static int igb_ndo_set_vf_vlan(struct net_device *netdev,
 164                                int vf, u16 vlan, u8 qos, __be16 vlan_proto);
 165 static int igb_ndo_set_vf_bw(struct net_device *, int, int, int);
 166 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
 167                                    bool setting);
 168 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf,
 169                                 bool setting);
 170 static int igb_ndo_get_vf_config(struct net_device *netdev, int vf,
 171                                  struct ifla_vf_info *ivi);
 172 static void igb_check_vf_rate_limit(struct igb_adapter *);
 173 static void igb_nfc_filter_exit(struct igb_adapter *adapter);
 174 static void igb_nfc_filter_restore(struct igb_adapter *adapter);
 175
 176 #ifdef CONFIG_PCI_IOV
 177 static int igb_vf_configure(struct igb_adapter *adapter, int vf);
 178 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs);
 179 static int igb_disable_sriov(struct pci_dev *dev);
 180 static int igb_pci_disable_sriov(struct pci_dev *dev);
 181 #endif
 182
 183 static int igb_suspend(struct device *);
 184 static int igb_resume(struct device *);
 185 static int igb_runtime_suspend(struct device *dev);
 186 static int igb_runtime_resume(struct device *dev);
 187 static int igb_runtime_idle(struct device *dev);
 188 static const struct dev_pm_ops igb_pm_ops = {
 189         SET_SYSTEM_SLEEP_PM_OPS(igb_suspend, igb_resume)
 190         SET_RUNTIME_PM_OPS(igb_runtime_suspend, igb_runtime_resume,
 191                         igb_runtime_idle)
 192 };
 193 static void igb_shutdown(struct pci_dev *);
 194 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs);
 195 #ifdef CONFIG_IGB_DCA
 196 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 197 static struct notifier_block dca_notifier = {
 198         .notifier_call  = igb_notify_dca,
 199         .next           = NULL,
 200         .priority       = 0
 201 };
 202 #endif
 203 #ifdef CONFIG_PCI_IOV
 204 static unsigned int max_vfs;
 205 module_param(max_vfs, uint, 0);
 206 MODULE_PARM_DESC(max_vfs, "Maximum number of virtual functions to allocate per physical function");
 207 #endif /* CONFIG_PCI_IOV */
 208
 209 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 210                      pci_channel_state_t);
 211 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 212 static void igb_io_resume(struct pci_dev *);
 213
 214 static const struct pci_error_handlers igb_err_handler = {
 215         .error_detected = igb_io_error_detected,
 216         .slot_reset = igb_io_slot_reset,
 217         .resume = igb_io_resume,
 218 };
 219
 220 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba);
 221
 222 static struct pci_driver igb_driver = {
 223         .name     = igb_driver_name,
 224         .id_table = igb_pci_tbl,
 225         .probe    = igb_probe,
 226         .remove   = igb_remove,
 227 #ifdef CONFIG_PM
 228         .driver.pm = &igb_pm_ops,
 229 #endif
 230         .shutdown = igb_shutdown,
 231         .sriov_configure = igb_pci_sriov_configure,
 232         .err_handler = &igb_err_handler
 233 };
 234
 235 MODULE_AUTHOR("Intel Corporation, <[email protected]>");
 236 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 237 MODULE_LICENSE("GPL v2");
 238
 239 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 240 static int debug = -1;
 241 module_param(debug, int, 0);
 242 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 243
 244 struct igb_reg_info {
 245         u32 ofs;
 246         char *name;
 247 };
 248
 249 static const struct igb_reg_info igb_reg_info_tbl[] = {
 250
 251         /* General Registers */
 252         {E1000_CTRL, "CTRL"},
 253         {E1000_STATUS, "STATUS"},
 254         {E1000_CTRL_EXT, "CTRL_EXT"},
 255
 256         /* Interrupt Registers */
 257         {E1000_ICR, "ICR"},
 258
 259         /* RX Registers */
 260         {E1000_RCTL, "RCTL"},
 261         {E1000_RDLEN(0), "RDLEN"},
 262         {E1000_RDH(0), "RDH"},
 263         {E1000_RDT(0), "RDT"},
 264         {E1000_RXDCTL(0), "RXDCTL"},
 265         {E1000_RDBAL(0), "RDBAL"},
 266         {E1000_RDBAH(0), "RDBAH"},
 267
 268         /* TX Registers */
 269         {E1000_TCTL, "TCTL"},
 270         {E1000_TDBAL(0), "TDBAL"},
 271         {E1000_TDBAH(0), "TDBAH"},
 272         {E1000_TDLEN(0), "TDLEN"},
 273         {E1000_TDH(0), "TDH"},
 274         {E1000_TDT(0), "TDT"},
 275         {E1000_TXDCTL(0), "TXDCTL"},
 276         {E1000_TDFH, "TDFH"},
 277         {E1000_TDFT, "TDFT"},
 278         {E1000_TDFHS, "TDFHS"},
 279         {E1000_TDFPC, "TDFPC"},
 280
 281         /* List Terminator */
 282         {}
 283 };
 284
 285 /* igb_regdump - register printout routine */
 286 static void igb_regdump(struct e1000_hw *hw, struct igb_reg_info *reginfo)
 287 {
 288         int n = 0;
 289         char rname[16];
 290         u32 regs[8];
 291
 292         switch (reginfo->ofs) {
 293         case E1000_RDLEN(0):
 294                 for (n = 0; n < 4; n++)
 295                         regs[n] = rd32(E1000_RDLEN(n));
 296                 break;
 297         case E1000_RDH(0):
 298                 for (n = 0; n < 4; n++)
 299                         regs[n] = rd32(E1000_RDH(n));
 300                 break;
 301         case E1000_RDT(0):
 302                 for (n = 0; n < 4; n++)
 303                         regs[n] = rd32(E1000_RDT(n));
 304                 break;
 305         case E1000_RXDCTL(0):
 306                 for (n = 0; n < 4; n++)
 307                         regs[n] = rd32(E1000_RXDCTL(n));
 308                 break;
 309         case E1000_RDBAL(0):
 310                 for (n = 0; n < 4; n++)
 311                         regs[n] = rd32(E1000_RDBAL(n));
 312                 break;
 313         case E1000_RDBAH(0):
 314                 for (n = 0; n < 4; n++)
 315                         regs[n] = rd32(E1000_RDBAH(n));
 316                 break;
 317         case E1000_TDBAL(0):
 318                 for (n = 0; n < 4; n++)
 319                         regs[n] = rd32(E1000_TDBAL(n));
 320                 break;
 321         case E1000_TDBAH(0):
 322                 for (n = 0; n < 4; n++)
 323                         regs[n] = rd32(E1000_TDBAH(n));
 324                 break;
 325         case E1000_TDLEN(0):
 326                 for (n = 0; n < 4; n++)
 327                         regs[n] = rd32(E1000_TDLEN(n));
 328                 break;
 329         case E1000_TDH(0):
 330                 for (n = 0; n < 4; n++)
 331                         regs[n] = rd32(E1000_TDH(n));
 332                 break;
 333         case E1000_TDT(0):
 334                 for (n = 0; n < 4; n++)
 335                         regs[n] = rd32(E1000_TDT(n));
 336                 break;
 337         case E1000_TXDCTL(0):
 338                 for (n = 0; n < 4; n++)
 339                         regs[n] = rd32(E1000_TXDCTL(n));
 340                 break;
 341         default:
 342                 pr_info("%-15s %08x\n", reginfo->name, rd32(reginfo->ofs));
 343                 return;
 344         }
 345
 346         snprintf(rname, 16, "%s%s", reginfo->name, "[0-3]");
 347         pr_info("%-15s %08x %08x %08x %08x\n", rname, regs[0], regs[1],
 348                 regs[2], regs[3]);
 349 }
 350
 351 /* igb_dump - Print registers, Tx-rings and Rx-rings */
 352 static void igb_dump(struct igb_adapter *adapter)
 353 {
 354         struct net_device *netdev = adapter->netdev;
 355         struct e1000_hw *hw = &adapter->hw;
 356         struct igb_reg_info *reginfo;
 357         struct igb_ring *tx_ring;
 358         union e1000_adv_tx_desc *tx_desc;
 359         struct my_u0 { __le64 a; __le64 b; } *u0;
 360         struct igb_ring *rx_ring;
 361         union e1000_adv_rx_desc *rx_desc;
 362         u32 staterr;
 363         u16 i, n;
 364
 365         if (!netif_msg_hw(adapter))
 366                 return;
 367
 368         /* Print netdevice Info */
 369         if (netdev) {
 370                 dev_info(&adapter->pdev->dev, "Net device Info\n");
 371                 pr_info("Device Name     state            trans_start\n");
 372                 pr_info("%-15s %016lX %016lX\n", netdev->name,
 373                         netdev->state, dev_trans_start(netdev));
 374         }
 375
 376         /* Print Registers */
 377         dev_info(&adapter->pdev->dev, "Register Dump\n");
 378         pr_info(" Register Name   Value\n");
 379         for (reginfo = (struct igb_reg_info *)igb_reg_info_tbl;
 380              reginfo->name; reginfo++) {
 381                 igb_regdump(hw, reginfo);
 382         }
 383
 384         /* Print TX Ring Summary */
 385         if (!netdev || !netif_running(netdev))
 386                 goto exit;
 387
 388         dev_info(&adapter->pdev->dev, "TX Rings Summary\n");
 389         pr_info("Queue [NTU] [NTC] [bi(ntc)->dma  ] leng ntw timestamp\n");
 390         for (n = 0; n < adapter->num_tx_queues; n++) {
 391                 struct igb_tx_buffer *buffer_info;
 392                 tx_ring = adapter->tx_ring[n];
 393                 buffer_info = &tx_ring->tx_buffer_info[tx_ring->next_to_clean];
 394                 pr_info(" %5d %5X %5X %016llX %04X %p %016llX\n",
 395                         n, tx_ring->next_to_use, tx_ring->next_to_clean,
 396                         (u64)dma_unmap_addr(buffer_info, dma),
 397                         dma_unmap_len(buffer_info, len),
 398                         buffer_info->next_to_watch,
 399                         (u64)buffer_info->time_stamp);
 400         }
 401
 402         /* Print TX Rings */
 403         if (!netif_msg_tx_done(adapter))
 404                 goto rx_ring_summary;
 405
 406         dev_info(&adapter->pdev->dev, "TX Rings Dump\n");
 407
 408         /* Transmit Descriptor Formats
 409          *
 410          * Advanced Transmit Descriptor
 411          *   +--------------------------------------------------------------+
 412          * 0 |         Buffer Address [63:0]                                |
 413          *   +--------------------------------------------------------------+
 414          * 8 | PAYLEN  | PORTS  |CC|IDX | STA | DCMD  |DTYP|MAC|RSV| DTALEN |
 415          *   +--------------------------------------------------------------+
 416          *   63      46 45    40 39 38 36 35 32 31   24             15       0
 417          */
 418
 419         for (n = 0; n < adapter->num_tx_queues; n++) {
 420                 tx_ring = adapter->tx_ring[n];
 421                 pr_info("------------------------------------\n");
 422                 pr_info("TX QUEUE INDEX = %d\n", tx_ring->queue_index);
 423                 pr_info("------------------------------------\n");
 424                 pr_info("T [desc]     [address 63:0  ] [PlPOCIStDDM Ln] [bi->dma       ] leng  ntw timestamp        bi->skb\n");
 425
 426                 for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
 427                         const char *next_desc;
 428                         struct igb_tx_buffer *buffer_info;
 429                         tx_desc = IGB_TX_DESC(tx_ring, i);
 430                         buffer_info = &tx_ring->tx_buffer_info[i];
 431                         u0 = (struct my_u0 *)tx_desc;
 432                         if (i == tx_ring->next_to_use &&
 433                             i == tx_ring->next_to_clean)
 434                                 next_desc = " NTC/U";
 435                         else if (i == tx_ring->next_to_use)
 436                                 next_desc = " NTU";
 437                         else if (i == tx_ring->next_to_clean)
 438                                 next_desc = " NTC";
 439                         else
 440                                 next_desc = "";
 441
 442                         pr_info("T [0x%03X]    %016llX %016llX %016llX %04X  %p %016llX %p%s\n",
 443                                 i, le64_to_cpu(u0->a),
 444                                 le64_to_cpu(u0->b),
 445                                 (u64)dma_unmap_addr(buffer_info, dma),
 446                                 dma_unmap_len(buffer_info, len),
 447                                 buffer_info->next_to_watch,
 448                                 (u64)buffer_info->time_stamp,
 449                                 buffer_info->skb, next_desc);
 450
 451                         if (netif_msg_pktdata(adapter) && buffer_info->skb)
 452                                 print_hex_dump(KERN_INFO, "",
 453                                         DUMP_PREFIX_ADDRESS,
 454                                         16, 1, buffer_info->skb->data,
 455                                         dma_unmap_len(buffer_info, len),
 456                                         true);
 457                 }
 458         }
 459
 460         /* Print RX Rings Summary */
 461 rx_ring_summary:
 462         dev_info(&adapter->pdev->dev, "RX Rings Summary\n");
 463         pr_info("Queue [NTU] [NTC]\n");
 464         for (n = 0; n < adapter->num_rx_queues; n++) {
 465                 rx_ring = adapter->rx_ring[n];
 466                 pr_info(" %5d %5X %5X\n",
 467                         n, rx_ring->next_to_use, rx_ring->next_to_clean);
 468         }
 469
 470         /* Print RX Rings */
 471         if (!netif_msg_rx_status(adapter))
 472                 goto exit;
 473
 474         dev_info(&adapter->pdev->dev, "RX Rings Dump\n");
 475
 476         /* Advanced Receive Descriptor (Read) Format
 477          *    63                                           1        0
 478          *    +-----------------------------------------------------+
 479          *  0 |       Packet Buffer Address [63:1]           |A0/NSE|
 480          *    +----------------------------------------------+------+
 481          *  8 |       Header Buffer Address [63:1]           |  DD  |
 482          *    +-----------------------------------------------------+
 483          *
 484          *
 485          * Advanced Receive Descriptor (Write-Back) Format
 486          *
 487          *   63       48 47    32 31  30      21 20 17 16   4 3     0
 488          *   +------------------------------------------------------+
 489          * 0 | Packet     IP     |SPH| HDR_LEN   | RSV|Packet|  RSS |
 490          *   | Checksum   Ident  |   |           |    | Type | Type |
 491          *   +------------------------------------------------------+
 492          * 8 | VLAN Tag | Length | Extended Error | Extended Status |
 493          *   +------------------------------------------------------+
 494          *   63       48 47    32 31            20 19               0
 495          */
 496
 497         for (n = 0; n < adapter->num_rx_queues; n++) {
 498                 rx_ring = adapter->rx_ring[n];
 499                 pr_info("------------------------------------\n");
 500                 pr_info("RX QUEUE INDEX = %d\n", rx_ring->queue_index);
 501                 pr_info("------------------------------------\n");
 502                 pr_info("R  [desc]      [ PktBuf     A0] [  HeadBuf   DD] [bi->dma       ] [bi->skb] <-- Adv Rx Read format\n");
 503                 pr_info("RWB[desc]      [PcsmIpSHl PtRs] [vl er S cks ln] ---------------- [bi->skb] <-- Adv Rx Write-Back format\n");
 504
 505                 for (i = 0; i < rx_ring->count; i++) {
 506                         const char *next_desc;
 507                         struct igb_rx_buffer *buffer_info;
 508                         buffer_info = &rx_ring->rx_buffer_info[i];
 509                         rx_desc = IGB_RX_DESC(rx_ring, i);
 510                         u0 = (struct my_u0 *)rx_desc;
 511                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 512
 513                         if (i == rx_ring->next_to_use)
 514                                 next_desc = " NTU";
 515                         else if (i == rx_ring->next_to_clean)
 516                                 next_desc = " NTC";
 517                         else
 518                                 next_desc = "";
 519
 520                         if (staterr & E1000_RXD_STAT_DD) {
 521                                 /* Descriptor Done */
 522                                 pr_info("%s[0x%03X]     %016llX %016llX ---------------- %s\n",
 523                                         "RWB", i,
 524                                         le64_to_cpu(u0->a),
 525                                         le64_to_cpu(u0->b),
 526                                         next_desc);
 527                         } else {
 528                                 pr_info("%s[0x%03X]     %016llX %016llX %016llX %s\n",
 529                                         "R  ", i,
 530                                         le64_to_cpu(u0->a),
 531                                         le64_to_cpu(u0->b),
 532                                         (u64)buffer_info->dma,
 533                                         next_desc);
 534
 535                                 if (netif_msg_pktdata(adapter) &&
 536                                     buffer_info->dma && buffer_info->page) {
 537                                         print_hex_dump(KERN_INFO, "",
 538                                           DUMP_PREFIX_ADDRESS,
 539                                           16, 1,
 540                                           page_address(buffer_info->page) +
 541                                                       buffer_info->page_offset,
 542                                           igb_rx_bufsz(rx_ring), true);
 543                                 }
 544                         }
 545                 }
 546         }
 547
 548 exit:
 549         return;
 550 }
 551
 552 /**
 553  *  igb_get_i2c_data - Reads the I2C SDA data bit
 554  *  @data: opaque pointer to adapter struct
 555  *
 556  *  Returns the I2C data bit value
 557  **/
 558 static int igb_get_i2c_data(void *data)
 559 {
 560         struct igb_adapter *adapter = (struct igb_adapter *)data;
 561         struct e1000_hw *hw = &adapter->hw;
 562         s32 i2cctl = rd32(E1000_I2CPARAMS);
 563
 564         return !!(i2cctl & E1000_I2C_DATA_IN);
 565 }
 566
 567 /**
 568  *  igb_set_i2c_data - Sets the I2C data bit
 569  *  @data: pointer to hardware structure
 570  *  @state: I2C data value (0 or 1) to set
 571  *
 572  *  Sets the I2C data bit
 573  **/
 574 static void igb_set_i2c_data(void *data, int state)
 575 {
 576         struct igb_adapter *adapter = (struct igb_adapter *)data;
 577         struct e1000_hw *hw = &adapter->hw;
 578         s32 i2cctl = rd32(E1000_I2CPARAMS);
 579
 580         if (state)
 581                 i2cctl |= E1000_I2C_DATA_OUT;
 582         else
 583                 i2cctl &= ~E1000_I2C_DATA_OUT;
 584
 585         i2cctl &= ~E1000_I2C_DATA_OE_N;
 586         i2cctl |= E1000_I2C_CLK_OE_N;
 587         wr32(E1000_I2CPARAMS, i2cctl);
 588         wrfl();
 589
 590 }
 591
 592 /**
 593  *  igb_set_i2c_clk - Sets the I2C SCL clock
 594  *  @data: pointer to hardware structure
 595  *  @state: state to set clock
 596  *
 597  *  Sets the I2C clock line to state
 598  **/
 599 static void igb_set_i2c_clk(void *data, int state)
 600 {
 601         struct igb_adapter *adapter = (struct igb_adapter *)data;
 602         struct e1000_hw *hw = &adapter->hw;
 603         s32 i2cctl = rd32(E1000_I2CPARAMS);
 604
 605         if (state) {
 606                 i2cctl |= E1000_I2C_CLK_OUT;
 607                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 608         } else {
 609                 i2cctl &= ~E1000_I2C_CLK_OUT;
 610                 i2cctl &= ~E1000_I2C_CLK_OE_N;
 611         }
 612         wr32(E1000_I2CPARAMS, i2cctl);
 613         wrfl();
 614 }
 615
 616 /**
 617  *  igb_get_i2c_clk - Gets the I2C SCL clock state
 618  *  @data: pointer to hardware structure
 619  *
 620  *  Gets the I2C clock state
 621  **/
 622 static int igb_get_i2c_clk(void *data)
 623 {
 624         struct igb_adapter *adapter = (struct igb_adapter *)data;
 625         struct e1000_hw *hw = &adapter->hw;
 626         s32 i2cctl = rd32(E1000_I2CPARAMS);
 627
 628         return !!(i2cctl & E1000_I2C_CLK_IN);
 629 }
 630
 631 static const struct i2c_algo_bit_data igb_i2c_algo = {
 632         .setsda         = igb_set_i2c_data,
 633         .setscl         = igb_set_i2c_clk,
 634         .getsda         = igb_get_i2c_data,
 635         .getscl         = igb_get_i2c_clk,
 636         .udelay         = 5,
 637         .timeout        = 20,
 638 };
 639
 640 /**
 641  *  igb_get_hw_dev - return device
 642  *  @hw: pointer to hardware structure
 643  *
 644  *  used by hardware layer to print debugging information
 645  **/
 646 struct net_device *igb_get_hw_dev(struct e1000_hw *hw)
 647 {
 648         struct igb_adapter *adapter = hw->back;
 649         return adapter->netdev;
 650 }
 651
 652 /**
 653  *  igb_init_module - Driver Registration Routine
 654  *
 655  *  igb_init_module is the first routine called when the driver is
 656  *  loaded. All it does is register with the PCI subsystem.
 657  **/
 658 static int __init igb_init_module(void)
 659 {
 660         int ret;
 661
 662         pr_info("%s\n", igb_driver_string);
 663         pr_info("%s\n", igb_copyright);
 664
 665 #ifdef CONFIG_IGB_DCA
 666         dca_register_notify(&dca_notifier);
 667 #endif
 668         ret = pci_register_driver(&igb_driver);
 669         return ret;
 670 }
 671
 672 module_init(igb_init_module);
 673
 674 /**
 675  *  igb_exit_module - Driver Exit Cleanup Routine
 676  *
 677  *  igb_exit_module is called just before the driver is removed
 678  *  from memory.
 679  **/
 680 static void __exit igb_exit_module(void)
 681 {
 682 #ifdef CONFIG_IGB_DCA
 683         dca_unregister_notify(&dca_notifier);
 684 #endif
 685         pci_unregister_driver(&igb_driver);
 686 }
 687
 688 module_exit(igb_exit_module);
 689
 690 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 691 /**
 692  *  igb_cache_ring_register - Descriptor ring to register mapping
 693  *  @adapter: board private structure to initialize
 694  *
 695  *  Once we know the feature-set enabled for the device, we'll cache
 696  *  the register offset the descriptor ring is assigned to.
 697  **/
 698 static void igb_cache_ring_register(struct igb_adapter *adapter)
 699 {
 700         int i = 0, j = 0;
 701         u32 rbase_offset = adapter->vfs_allocated_count;
 702
 703         switch (adapter->hw.mac.type) {
 704         case e1000_82576:
 705                 /* The queues are allocated for virtualization such that VF 0
 706                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 707                  * In order to avoid collision we start at the first free queue
 708                  * and continue consuming queues in the same sequence
 709                  */
 710                 if (adapter->vfs_allocated_count) {
 711                         for (; i < adapter->rss_queues; i++)
 712                                 adapter->rx_ring[i]->reg_idx = rbase_offset +
 713                                                                Q_IDX_82576(i);
 714                 }
 715                 fallthrough;
 716         case e1000_82575:
 717         case e1000_82580:
 718         case e1000_i350:
 719         case e1000_i354:
 720         case e1000_i210:
 721         case e1000_i211:
 722         default:
 723                 for (; i < adapter->num_rx_queues; i++)
 724                         adapter->rx_ring[i]->reg_idx = rbase_offset + i;
 725                 for (; j < adapter->num_tx_queues; j++)
 726                         adapter->tx_ring[j]->reg_idx = rbase_offset + j;
 727                 break;
 728         }
 729 }
 730
 731 u32 igb_rd32(struct e1000_hw *hw, u32 reg)
 732 {
 733         struct igb_adapter *igb = container_of(hw, struct igb_adapter, hw);
 734         u8 __iomem *hw_addr = READ_ONCE(hw->hw_addr);
 735         u32 value = 0;
 736
 737         if (E1000_REMOVED(hw_addr))
 738                 return ~value;
 739
 740         value = readl(&hw_addr[reg]);
 741
 742         /* reads should not return all F's */
 743         if (!(~value) && (!reg || !(~readl(hw_addr)))) {
 744                 struct net_device *netdev = igb->netdev;
 745                 hw->hw_addr = NULL;
 746                 netdev_err(netdev, "PCIe link lost\n");
 747                 WARN(pci_device_is_present(igb->pdev),
 748                      "igb: Failed to read reg 0x%x!\n", reg);
 749         }
 750
 751         return value;
 752 }
 753
 754 /**
 755  *  igb_write_ivar - configure ivar for given MSI-X vector
 756  *  @hw: pointer to the HW structure
 757  *  @msix_vector: vector number we are allocating to a given ring
 758  *  @index: row index of IVAR register to write within IVAR table
 759  *  @offset: column offset of in IVAR, should be multiple of 8
 760  *
 761  *  This function is intended to handle the writing of the IVAR register
 762  *  for adapters 82576 and newer.  The IVAR table consists of 2 columns,
 763  *  each containing an cause allocation for an Rx and Tx ring, and a
 764  *  variable number of rows depending on the number of queues supported.
 765  **/
 766 static void igb_write_ivar(struct e1000_hw *hw, int msix_vector,
 767                            int index, int offset)
 768 {
 769         u32 ivar = array_rd32(E1000_IVAR0, index);
 770
 771         /* clear any bits that are currently set */
 772         ivar &= ~((u32)0xFF << offset);
 773
 774         /* write vector and valid bit */
 775         ivar |= (msix_vector | E1000_IVAR_VALID) << offset;
 776
 777         array_wr32(E1000_IVAR0, index, ivar);
 778 }
 779
 780 #define IGB_N0_QUEUE -1
 781 static void igb_assign_vector(struct igb_q_vector *q_vector, int msix_vector)
 782 {
 783         struct igb_adapter *adapter = q_vector->adapter;
 784         struct e1000_hw *hw = &adapter->hw;
 785         int rx_queue = IGB_N0_QUEUE;
 786         int tx_queue = IGB_N0_QUEUE;
 787         u32 msixbm = 0;
 788
 789         if (q_vector->rx.ring)
 790                 rx_queue = q_vector->rx.ring->reg_idx;
 791         if (q_vector->tx.ring)
 792                 tx_queue = q_vector->tx.ring->reg_idx;
 793
 794         switch (hw->mac.type) {
 795         case e1000_82575:
 796                 /* The 82575 assigns vectors using a bitmask, which matches the
 797                  * bitmask for the EICR/EIMS/EIMC registers.  To assign one
 798                  * or more queues to a vector, we write the appropriate bits
 799                  * into the MSIXBM register for that vector.
 800                  */
 801                 if (rx_queue > IGB_N0_QUEUE)
 802                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 803                 if (tx_queue > IGB_N0_QUEUE)
 804                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 805                 if (!(adapter->flags & IGB_FLAG_HAS_MSIX) && msix_vector == 0)
 806                         msixbm |= E1000_EIMS_OTHER;
 807                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 808                 q_vector->eims_value = msixbm;
 809                 break;
 810         case e1000_82576:
 811                 /* 82576 uses a table that essentially consists of 2 columns
 812                  * with 8 rows.  The ordering is column-major so we use the
 813                  * lower 3 bits as the row index, and the 4th bit as the
 814                  * column offset.
 815                  */
 816                 if (rx_queue > IGB_N0_QUEUE)
 817                         igb_write_ivar(hw, msix_vector,
 818                                        rx_queue & 0x7,
 819                                        (rx_queue & 0x8) << 1);
 820                 if (tx_queue > IGB_N0_QUEUE)
 821                         igb_write_ivar(hw, msix_vector,
 822                                        tx_queue & 0x7,
 823                                        ((tx_queue & 0x8) << 1) + 8);
 824                 q_vector->eims_value = BIT(msix_vector);
 825                 break;
 826         case e1000_82580:
 827         case e1000_i350:
 828         case e1000_i354:
 829         case e1000_i210:
 830         case e1000_i211:
 831                 /* On 82580 and newer adapters the scheme is similar to 82576
 832                  * however instead of ordering column-major we have things
 833                  * ordered row-major.  So we traverse the table by using
 834                  * bit 0 as the column offset, and the remaining bits as the
 835                  * row index.
 836                  */
 837                 if (rx_queue > IGB_N0_QUEUE)
 838                         igb_write_ivar(hw, msix_vector,
 839                                        rx_queue >> 1,
 840                                        (rx_queue & 0x1) << 4);
 841                 if (tx_queue > IGB_N0_QUEUE)
 842                         igb_write_ivar(hw, msix_vector,
 843                                        tx_queue >> 1,
 844                                        ((tx_queue & 0x1) << 4) + 8);
 845                 q_vector->eims_value = BIT(msix_vector);
 846                 break;
 847         default:
 848                 BUG();
 849                 break;
 850         }
 851
 852         /* add q_vector eims value to global eims_enable_mask */
 853         adapter->eims_enable_mask |= q_vector->eims_value;
 854
 855         /* configure q_vector to set itr on first interrupt */
 856         q_vector->set_itr = 1;
 857 }
 858
 859 /**
 860  *  igb_configure_msix - Configure MSI-X hardware
 861  *  @adapter: board private structure to initialize
 862  *
 863  *  igb_configure_msix sets up the hardware to properly
 864  *  generate MSI-X interrupts.
 865  **/
 866 static void igb_configure_msix(struct igb_adapter *adapter)
 867 {
 868         u32 tmp;
 869         int i, vector = 0;
 870         struct e1000_hw *hw = &adapter->hw;
 871
 872         adapter->eims_enable_mask = 0;
 873
 874         /* set vector for other causes, i.e. link changes */
 875         switch (hw->mac.type) {
 876         case e1000_82575:
 877                 tmp = rd32(E1000_CTRL_EXT);
 878                 /* enable MSI-X PBA support*/
 879                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 880
 881                 /* Auto-Mask interrupts upon ICR read. */
 882                 tmp |= E1000_CTRL_EXT_EIAME;
 883                 tmp |= E1000_CTRL_EXT_IRCA;
 884
 885                 wr32(E1000_CTRL_EXT, tmp);
 886
 887                 /* enable msix_other interrupt */
 888                 array_wr32(E1000_MSIXBM(0), vector++, E1000_EIMS_OTHER);
 889                 adapter->eims_other = E1000_EIMS_OTHER;
 890
 891                 break;
 892
 893         case e1000_82576:
 894         case e1000_82580:
 895         case e1000_i350:
 896         case e1000_i354:
 897         case e1000_i210:
 898         case e1000_i211:
 899                 /* Turn on MSI-X capability first, or our settings
 900                  * won't stick.  And it will take days to debug.
 901                  */
 902                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 903                      E1000_GPIE_PBA | E1000_GPIE_EIAME |
 904                      E1000_GPIE_NSICR);
 905
 906                 /* enable msix_other interrupt */
 907                 adapter->eims_other = BIT(vector);
 908                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 909
 910                 wr32(E1000_IVAR_MISC, tmp);
 911                 break;
 912         default:
 913                 /* do nothing, since nothing else supports MSI-X */
 914                 break;
 915         } /* switch (hw->mac.type) */
 916
 917         adapter->eims_enable_mask |= adapter->eims_other;
 918
 919         for (i = 0; i < adapter->num_q_vectors; i++)
 920                 igb_assign_vector(adapter->q_vector[i], vector++);
 921
 922         wrfl();
 923 }
 924
 925 /**
 926  *  igb_request_msix - Initialize MSI-X interrupts
 927  *  @adapter: board private structure to initialize
 928  *
 929  *  igb_request_msix allocates MSI-X vectors and requests interrupts from the
 930  *  kernel.
 931  **/
 932 static int igb_request_msix(struct igb_adapter *adapter)
 933 {
 934         struct net_device *netdev = adapter->netdev;
 935         int i, err = 0, vector = 0, free_vector = 0;
 936
 937         err = request_irq(adapter->msix_entries[vector].vector,
 938                           igb_msix_other, 0, netdev->name, adapter);
 939         if (err)
 940                 goto err_out;
 941
 942         for (i = 0; i < adapter->num_q_vectors; i++) {
 943                 struct igb_q_vector *q_vector = adapter->q_vector[i];
 944
 945                 vector++;
 946
 947                 q_vector->itr_register = adapter->io_addr + E1000_EITR(vector);
 948
 949                 if (q_vector->rx.ring && q_vector->tx.ring)
 950                         sprintf(q_vector->name, "%s-TxRx-%u", netdev->name,
 951                                 q_vector->rx.ring->queue_index);
 952                 else if (q_vector->tx.ring)
 953                         sprintf(q_vector->name, "%s-tx-%u", netdev->name,
 954                                 q_vector->tx.ring->queue_index);
 955                 else if (q_vector->rx.ring)
 956                         sprintf(q_vector->name, "%s-rx-%u", netdev->name,
 957                                 q_vector->rx.ring->queue_index);
 958                 else
 959                         sprintf(q_vector->name, "%s-unused", netdev->name);
 960
 961                 err = request_irq(adapter->msix_entries[vector].vector,
 962                                   igb_msix_ring, 0, q_vector->name,
 963                                   q_vector);
 964                 if (err)
 965                         goto err_free;
 966         }
 967
 968         igb_configure_msix(adapter);
 969         return 0;
 970
 971 err_free:
 972         /* free already assigned IRQs */
 973         free_irq(adapter->msix_entries[free_vector++].vector, adapter);
 974
 975         vector--;
 976         for (i = 0; i < vector; i++) {
 977                 free_irq(adapter->msix_entries[free_vector++].vector,
 978                          adapter->q_vector[i]);
 979         }
 980 err_out:
 981         return err;
 982 }
 983
 984 /**
 985  *  igb_free_q_vector - Free memory allocated for specific interrupt vector
 986  *  @adapter: board private structure to initialize
 987  *  @v_idx: Index of vector to be freed
 988  *
 989  *  This function frees the memory allocated to the q_vector.
 990  **/
 991 static void igb_free_q_vector(struct igb_adapter *adapter, int v_idx)
 992 {
 993         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
 994
 995         adapter->q_vector[v_idx] = NULL;
 996
 997         /* igb_get_stats64() might access the rings on this vector,
 998          * we must wait a grace period before freeing it.
 999          */
1000         if (q_vector)
1001                 kfree_rcu(q_vector, rcu);
1002 }
1003
1004 /**
1005  *  igb_reset_q_vector - Reset config for interrupt vector
1006  *  @adapter: board private structure to initialize
1007  *  @v_idx: Index of vector to be reset
1008  *
1009  *  If NAPI is enabled it will delete any references to the
1010  *  NAPI struct. This is preparation for igb_free_q_vector.
1011  **/
1012 static void igb_reset_q_vector(struct igb_adapter *adapter, int v_idx)
1013 {
1014         struct igb_q_vector *q_vector = adapter->q_vector[v_idx];
1015
1016         /* Coming from igb_set_interrupt_capability, the vectors are not yet
1017          * allocated. So, q_vector is NULL so we should stop here.
1018          */
1019         if (!q_vector)
1020                 return;
1021
1022         if (q_vector->tx.ring)
1023                 adapter->tx_ring[q_vector->tx.ring->queue_index] = NULL;
1024
1025         if (q_vector->rx.ring)
1026                 adapter->rx_ring[q_vector->rx.ring->queue_index] = NULL;
1027
1028         netif_napi_del(&q_vector->napi);
1029
1030 }
1031
1032 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
1033 {
1034         int v_idx = adapter->num_q_vectors;
1035
1036         if (adapter->flags & IGB_FLAG_HAS_MSIX)
1037                 pci_disable_msix(adapter->pdev);
1038         else if (adapter->flags & IGB_FLAG_HAS_MSI)
1039                 pci_disable_msi(adapter->pdev);
1040
1041         while (v_idx--)
1042                 igb_reset_q_vector(adapter, v_idx);
1043 }
1044
1045 /**
1046  *  igb_free_q_vectors - Free memory allocated for interrupt vectors
1047  *  @adapter: board private structure to initialize
1048  *
1049  *  This function frees the memory allocated to the q_vectors.  In addition if
1050  *  NAPI is enabled it will delete any references to the NAPI struct prior
1051  *  to freeing the q_vector.
1052  **/
1053 static void igb_free_q_vectors(struct igb_adapter *adapter)
1054 {
1055         int v_idx = adapter->num_q_vectors;
1056
1057         adapter->num_tx_queues = 0;
1058         adapter->num_rx_queues = 0;
1059         adapter->num_q_vectors = 0;
1060
1061         while (v_idx--) {
1062                 igb_reset_q_vector(adapter, v_idx);
1063                 igb_free_q_vector(adapter, v_idx);
1064         }
1065 }
1066
1067 /**
1068  *  igb_clear_interrupt_scheme - reset the device to a state of no interrupts
1069  *  @adapter: board private structure to initialize
1070  *
1071  *  This function resets the device so that it has 0 Rx queues, Tx queues, and
1072  *  MSI-X interrupts allocated.
1073  */
1074 static void igb_clear_interrupt_scheme(struct igb_adapter *adapter)
1075 {
1076         igb_free_q_vectors(adapter);
1077         igb_reset_interrupt_capability(adapter);
1078 }
1079
1080 /**
1081  *  igb_set_interrupt_capability - set MSI or MSI-X if supported
1082  *  @adapter: board private structure to initialize
1083  *  @msix: boolean value of MSIX capability
1084  *
1085  *  Attempt to configure interrupts using the best available
1086  *  capabilities of the hardware and kernel.
1087  **/
1088 static void igb_set_interrupt_capability(struct igb_adapter *adapter, bool msix)
1089 {
1090         int err;
1091         int numvecs, i;
1092
1093         if (!msix)
1094                 goto msi_only;
1095         adapter->flags |= IGB_FLAG_HAS_MSIX;
1096
1097         /* Number of supported queues. */
1098         adapter->num_rx_queues = adapter->rss_queues;
1099         if (adapter->vfs_allocated_count)
1100                 adapter->num_tx_queues = 1;
1101         else
1102                 adapter->num_tx_queues = adapter->rss_queues;
1103
1104         /* start with one vector for every Rx queue */
1105         numvecs = adapter->num_rx_queues;
1106
1107         /* if Tx handler is separate add 1 for every Tx queue */
1108         if (!(adapter->flags & IGB_FLAG_QUEUE_PAIRS))
1109                 numvecs += adapter->num_tx_queues;
1110
1111         /* store the number of vectors reserved for queues */
1112         adapter->num_q_vectors = numvecs;
1113
1114         /* add 1 vector for link status interrupts */
1115         numvecs++;
1116         for (i = 0; i < numvecs; i++)
1117                 adapter->msix_entries[i].entry = i;
1118
1119         err = pci_enable_msix_range(adapter->pdev,
1120                                     adapter->msix_entries,
1121                                     numvecs,
1122                                     numvecs);
1123         if (err > 0)
1124                 return;
1125
1126         igb_reset_interrupt_capability(adapter);
1127
1128         /* If we can't do MSI-X, try MSI */
1129 msi_only:
1130         adapter->flags &= ~IGB_FLAG_HAS_MSIX;
1131 #ifdef CONFIG_PCI_IOV
1132         /* disable SR-IOV for non MSI-X configurations */
1133         if (adapter->vf_data) {
1134                 struct e1000_hw *hw = &adapter->hw;
1135                 /* disable iov and allow time for transactions to clear */
1136                 pci_disable_sriov(adapter->pdev);
1137                 msleep(500);
1138
1139                 kfree(adapter->vf_mac_list);
1140                 adapter->vf_mac_list = NULL;
1141                 kfree(adapter->vf_data);
1142                 adapter->vf_data = NULL;
1143                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1144                 wrfl();
1145                 msleep(100);
1146                 dev_info(&adapter->pdev->dev, "IOV Disabled\n");
1147         }
1148 #endif
1149         adapter->vfs_allocated_count = 0;
1150         adapter->rss_queues = 1;
1151         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
1152         adapter->num_rx_queues = 1;
1153         adapter->num_tx_queues = 1;
1154         adapter->num_q_vectors = 1;
1155         if (!pci_enable_msi(adapter->pdev))
1156                 adapter->flags |= IGB_FLAG_HAS_MSI;
1157 }
1158
1159 static void igb_add_ring(struct igb_ring *ring,
1160                          struct igb_ring_container *head)
1161 {
1162         head->ring = ring;
1163         head->count++;
1164 }
1165
1166 /**
1167  *  igb_alloc_q_vector - Allocate memory for a single interrupt vector
1168  *  @adapter: board private structure to initialize
1169  *  @v_count: q_vectors allocated on adapter, used for ring interleaving
1170  *  @v_idx: index of vector in adapter struct
1171  *  @txr_count: total number of Tx rings to allocate
1172  *  @txr_idx: index of first Tx ring to allocate
1173  *  @rxr_count: total number of Rx rings to allocate
1174  *  @rxr_idx: index of first Rx ring to allocate
1175  *
1176  *  We allocate one q_vector.  If allocation fails we return -ENOMEM.
1177  **/
1178 static int igb_alloc_q_vector(struct igb_adapter *adapter,
1179                               int v_count, int v_idx,
1180                               int txr_count, int txr_idx,
1181                               int rxr_count, int rxr_idx)
1182 {
1183         struct igb_q_vector *q_vector;
1184         struct igb_ring *ring;
1185         int ring_count;
1186         size_t size;
1187
1188         /* igb only supports 1 Tx and/or 1 Rx queue per vector */
1189         if (txr_count > 1 || rxr_count > 1)
1190                 return -ENOMEM;
1191
1192         ring_count = txr_count + rxr_count;
1193         size = struct_size(q_vector, ring, ring_count);
1194
1195         /* allocate q_vector and rings */
1196         q_vector = adapter->q_vector[v_idx];
1197         if (!q_vector) {
1198                 q_vector = kzalloc(size, GFP_KERNEL);
1199         } else if (size > ksize(q_vector)) {
1200                 kfree_rcu(q_vector, rcu);
1201                 q_vector = kzalloc(size, GFP_KERNEL);
1202         } else {
1203                 memset(q_vector, 0, size);
1204         }
1205         if (!q_vector)
1206                 return -ENOMEM;
1207
1208         /* initialize NAPI */
1209         netif_napi_add(adapter->netdev, &q_vector->napi,
1210                        igb_poll, 64);
1211
1212         /* tie q_vector and adapter together */
1213         adapter->q_vector[v_idx] = q_vector;
1214         q_vector->adapter = adapter;
1215
1216         /* initialize work limits */
1217         q_vector->tx.work_limit = adapter->tx_work_limit;
1218
1219         /* initialize ITR configuration */
1220         q_vector->itr_register = adapter->io_addr + E1000_EITR(0);
1221         q_vector->itr_val = IGB_START_ITR;
1222
1223         /* initialize pointer to rings */
1224         ring = q_vector->ring;
1225
1226         /* intialize ITR */
1227         if (rxr_count) {
1228                 /* rx or rx/tx vector */
1229                 if (!adapter->rx_itr_setting || adapter->rx_itr_setting > 3)
1230                         q_vector->itr_val = adapter->rx_itr_setting;
1231         } else {
1232                 /* tx only vector */
1233                 if (!adapter->tx_itr_setting || adapter->tx_itr_setting > 3)
1234                         q_vector->itr_val = adapter->tx_itr_setting;
1235         }
1236
1237         if (txr_count) {
1238                 /* assign generic ring traits */
1239                 ring->dev = &adapter->pdev->dev;
1240                 ring->netdev = adapter->netdev;
1241
1242                 /* configure backlink on ring */
1243                 ring->q_vector = q_vector;
1244
1245                 /* update q_vector Tx values */
1246                 igb_add_ring(ring, &q_vector->tx);
1247
1248                 /* For 82575, context index must be unique per ring. */
1249                 if (adapter->hw.mac.type == e1000_82575)
1250                         set_bit(IGB_RING_FLAG_TX_CTX_IDX, &ring->flags);
1251
1252                 /* apply Tx specific ring traits */
1253                 ring->count = adapter->tx_ring_count;
1254                 ring->queue_index = txr_idx;
1255
1256                 ring->cbs_enable = false;
1257                 ring->idleslope = 0;
1258                 ring->sendslope = 0;
1259                 ring->hicredit = 0;
1260                 ring->locredit = 0;
1261
1262                 u64_stats_init(&ring->tx_syncp);
1263                 u64_stats_init(&ring->tx_syncp2);
1264
1265                 /* assign ring to adapter */
1266                 adapter->tx_ring[txr_idx] = ring;
1267
1268                 /* push pointer to next ring */
1269                 ring++;
1270         }
1271
1272         if (rxr_count) {
1273                 /* assign generic ring traits */
1274                 ring->dev = &adapter->pdev->dev;
1275                 ring->netdev = adapter->netdev;
1276
1277                 /* configure backlink on ring */
1278                 ring->q_vector = q_vector;
1279
1280                 /* update q_vector Rx values */
1281                 igb_add_ring(ring, &q_vector->rx);
1282
1283                 /* set flag indicating ring supports SCTP checksum offload */
1284                 if (adapter->hw.mac.type >= e1000_82576)
1285                         set_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags);
1286
1287                 /* On i350, i354, i210, and i211, loopback VLAN packets
1288                  * have the tag byte-swapped.
1289                  */
1290                 if (adapter->hw.mac.type >= e1000_i350)
1291                         set_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &ring->flags);
1292
1293                 /* apply Rx specific ring traits */
1294                 ring->count = adapter->rx_ring_count;
1295                 ring->queue_index = rxr_idx;
1296
1297                 u64_stats_init(&ring->rx_syncp);
1298
1299                 /* assign ring to adapter */
1300                 adapter->rx_ring[rxr_idx] = ring;
1301         }
1302
1303         return 0;
1304 }
1305
1306
1307 /**
1308  *  igb_alloc_q_vectors - Allocate memory for interrupt vectors
1309  *  @adapter: board private structure to initialize
1310  *
1311  *  We allocate one q_vector per queue interrupt.  If allocation fails we
1312  *  return -ENOMEM.
1313  **/
1314 static int igb_alloc_q_vectors(struct igb_adapter *adapter)
1315 {
1316         int q_vectors = adapter->num_q_vectors;
1317         int rxr_remaining = adapter->num_rx_queues;
1318         int txr_remaining = adapter->num_tx_queues;
1319         int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1320         int err;
1321
1322         if (q_vectors >= (rxr_remaining + txr_remaining)) {
1323                 for (; rxr_remaining; v_idx++) {
1324                         err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1325                                                  0, 0, 1, rxr_idx);
1326
1327                         if (err)
1328                                 goto err_out;
1329
1330                         /* update counts and index */
1331                         rxr_remaining--;
1332                         rxr_idx++;
1333                 }
1334         }
1335
1336         for (; v_idx < q_vectors; v_idx++) {
1337                 int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1338                 int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1339
1340                 err = igb_alloc_q_vector(adapter, q_vectors, v_idx,
1341                                          tqpv, txr_idx, rqpv, rxr_idx);
1342
1343                 if (err)
1344                         goto err_out;
1345
1346                 /* update counts and index */
1347                 rxr_remaining -= rqpv;
1348                 txr_remaining -= tqpv;
1349                 rxr_idx++;
1350                 txr_idx++;
1351         }
1352
1353         return 0;
1354
1355 err_out:
1356         adapter->num_tx_queues = 0;
1357         adapter->num_rx_queues = 0;
1358         adapter->num_q_vectors = 0;
1359
1360         while (v_idx--)
1361                 igb_free_q_vector(adapter, v_idx);
1362
1363         return -ENOMEM;
1364 }
1365
1366 /**
1367  *  igb_init_interrupt_scheme - initialize interrupts, allocate queues/vectors
1368  *  @adapter: board private structure to initialize
1369  *  @msix: boolean value of MSIX capability
1370  *
1371  *  This function initializes the interrupts and allocates all of the queues.
1372  **/
1373 static int igb_init_interrupt_scheme(struct igb_adapter *adapter, bool msix)
1374 {
1375         struct pci_dev *pdev = adapter->pdev;
1376         int err;
1377
1378         igb_set_interrupt_capability(adapter, msix);
1379
1380         err = igb_alloc_q_vectors(adapter);
1381         if (err) {
1382                 dev_err(&pdev->dev, "Unable to allocate memory for vectors\n");
1383                 goto err_alloc_q_vectors;
1384         }
1385
1386         igb_cache_ring_register(adapter);
1387
1388         return 0;
1389
1390 err_alloc_q_vectors:
1391         igb_reset_interrupt_capability(adapter);
1392         return err;
1393 }
1394
1395 /**
1396  *  igb_request_irq - initialize interrupts
1397  *  @adapter: board private structure to initialize
1398  *
1399  *  Attempts to configure interrupts using the best available
1400  *  capabilities of the hardware and kernel.
1401  **/
1402 static int igb_request_irq(struct igb_adapter *adapter)
1403 {
1404         struct net_device *netdev = adapter->netdev;
1405         struct pci_dev *pdev = adapter->pdev;
1406         int err = 0;
1407
1408         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1409                 err = igb_request_msix(adapter);
1410                 if (!err)
1411                         goto request_done;
1412                 /* fall back to MSI */
1413                 igb_free_all_tx_resources(adapter);
1414                 igb_free_all_rx_resources(adapter);
1415
1416                 igb_clear_interrupt_scheme(adapter);
1417                 err = igb_init_interrupt_scheme(adapter, false);
1418                 if (err)
1419                         goto request_done;
1420
1421                 igb_setup_all_tx_resources(adapter);
1422                 igb_setup_all_rx_resources(adapter);
1423                 igb_configure(adapter);
1424         }
1425
1426         igb_assign_vector(adapter->q_vector[0], 0);
1427
1428         if (adapter->flags & IGB_FLAG_HAS_MSI) {
1429                 err = request_irq(pdev->irq, igb_intr_msi, 0,
1430                                   netdev->name, adapter);
1431                 if (!err)
1432                         goto request_done;
1433
1434                 /* fall back to legacy interrupts */
1435                 igb_reset_interrupt_capability(adapter);
1436                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
1437         }
1438
1439         err = request_irq(pdev->irq, igb_intr, IRQF_SHARED,
1440                           netdev->name, adapter);
1441
1442         if (err)
1443                 dev_err(&pdev->dev, "Error %d getting interrupt\n",
1444                         err);
1445
1446 request_done:
1447         return err;
1448 }
1449
1450 static void igb_free_irq(struct igb_adapter *adapter)
1451 {
1452         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1453                 int vector = 0, i;
1454
1455                 free_irq(adapter->msix_entries[vector++].vector, adapter);
1456
1457                 for (i = 0; i < adapter->num_q_vectors; i++)
1458                         free_irq(adapter->msix_entries[vector++].vector,
1459                                  adapter->q_vector[i]);
1460         } else {
1461                 free_irq(adapter->pdev->irq, adapter);
1462         }
1463 }
1464
1465 /**
1466  *  igb_irq_disable - Mask off interrupt generation on the NIC
1467  *  @adapter: board private structure
1468  **/
1469 static void igb_irq_disable(struct igb_adapter *adapter)
1470 {
1471         struct e1000_hw *hw = &adapter->hw;
1472
1473         /* we need to be careful when disabling interrupts.  The VFs are also
1474          * mapped into these registers and so clearing the bits can cause
1475          * issues on the VF drivers so we only need to clear what we set
1476          */
1477         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1478                 u32 regval = rd32(E1000_EIAM);
1479
1480                 wr32(E1000_EIAM, regval & ~adapter->eims_enable_mask);
1481                 wr32(E1000_EIMC, adapter->eims_enable_mask);
1482                 regval = rd32(E1000_EIAC);
1483                 wr32(E1000_EIAC, regval & ~adapter->eims_enable_mask);
1484         }
1485
1486         wr32(E1000_IAM, 0);
1487         wr32(E1000_IMC, ~0);
1488         wrfl();
1489         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1490                 int i;
1491
1492                 for (i = 0; i < adapter->num_q_vectors; i++)
1493                         synchronize_irq(adapter->msix_entries[i].vector);
1494         } else {
1495                 synchronize_irq(adapter->pdev->irq);
1496         }
1497 }
1498
1499 /**
1500  *  igb_irq_enable - Enable default interrupt generation settings
1501  *  @adapter: board private structure
1502  **/
1503 static void igb_irq_enable(struct igb_adapter *adapter)
1504 {
1505         struct e1000_hw *hw = &adapter->hw;
1506
1507         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
1508                 u32 ims = E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_DRSTA;
1509                 u32 regval = rd32(E1000_EIAC);
1510
1511                 wr32(E1000_EIAC, regval | adapter->eims_enable_mask);
1512                 regval = rd32(E1000_EIAM);
1513                 wr32(E1000_EIAM, regval | adapter->eims_enable_mask);
1514                 wr32(E1000_EIMS, adapter->eims_enable_mask);
1515                 if (adapter->vfs_allocated_count) {
1516                         wr32(E1000_MBVFIMR, 0xFF);
1517                         ims |= E1000_IMS_VMMB;
1518                 }
1519                 wr32(E1000_IMS, ims);
1520         } else {
1521                 wr32(E1000_IMS, IMS_ENABLE_MASK |
1522                                 E1000_IMS_DRSTA);
1523                 wr32(E1000_IAM, IMS_ENABLE_MASK |
1524                                 E1000_IMS_DRSTA);
1525         }
1526 }
1527
1528 static void igb_update_mng_vlan(struct igb_adapter *adapter)
1529 {
1530         struct e1000_hw *hw = &adapter->hw;
1531         u16 pf_id = adapter->vfs_allocated_count;
1532         u16 vid = adapter->hw.mng_cookie.vlan_id;
1533         u16 old_vid = adapter->mng_vlan_id;
1534
1535         if (hw->mng_cookie.status & E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
1536                 /* add VID to filter table */
1537                 igb_vfta_set(hw, vid, pf_id, true, true);
1538                 adapter->mng_vlan_id = vid;
1539         } else {
1540                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1541         }
1542
1543         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
1544             (vid != old_vid) &&
1545             !test_bit(old_vid, adapter->active_vlans)) {
1546                 /* remove VID from filter table */
1547                 igb_vfta_set(hw, vid, pf_id, false, true);
1548         }
1549 }
1550
1551 /**
1552  *  igb_release_hw_control - release control of the h/w to f/w
1553  *  @adapter: address of board private structure
1554  *
1555  *  igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
1556  *  For ASF and Pass Through versions of f/w this means that the
1557  *  driver is no longer loaded.
1558  **/
1559 static void igb_release_hw_control(struct igb_adapter *adapter)
1560 {
1561         struct e1000_hw *hw = &adapter->hw;
1562         u32 ctrl_ext;
1563
1564         /* Let firmware take over control of h/w */
1565         ctrl_ext = rd32(E1000_CTRL_EXT);
1566         wr32(E1000_CTRL_EXT,
1567                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
1568 }
1569
1570 /**
1571  *  igb_get_hw_control - get control of the h/w from f/w
1572  *  @adapter: address of board private structure
1573  *
1574  *  igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
1575  *  For ASF and Pass Through versions of f/w this means that
1576  *  the driver is loaded.
1577  **/
1578 static void igb_get_hw_control(struct igb_adapter *adapter)
1579 {
1580         struct e1000_hw *hw = &adapter->hw;
1581         u32 ctrl_ext;
1582
1583         /* Let firmware know the driver has taken over */
1584         ctrl_ext = rd32(E1000_CTRL_EXT);
1585         wr32(E1000_CTRL_EXT,
1586                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
1587 }
1588
1589 static void enable_fqtss(struct igb_adapter *adapter, bool enable)
1590 {
1591         struct net_device *netdev = adapter->netdev;
1592         struct e1000_hw *hw = &adapter->hw;
1593
1594         WARN_ON(hw->mac.type != e1000_i210);
1595
1596         if (enable)
1597                 adapter->flags |= IGB_FLAG_FQTSS;
1598         else
1599                 adapter->flags &= ~IGB_FLAG_FQTSS;
1600
1601         if (netif_running(netdev))
1602                 schedule_work(&adapter->reset_task);
1603 }
1604
1605 static bool is_fqtss_enabled(struct igb_adapter *adapter)
1606 {
1607         return (adapter->flags & IGB_FLAG_FQTSS) ? true : false;
1608 }
1609
1610 static void set_tx_desc_fetch_prio(struct e1000_hw *hw, int queue,
1611                                    enum tx_queue_prio prio)
1612 {
1613         u32 val;
1614
1615         WARN_ON(hw->mac.type != e1000_i210);
1616         WARN_ON(queue < 0 || queue > 4);
1617
1618         val = rd32(E1000_I210_TXDCTL(queue));
1619
1620         if (prio == TX_QUEUE_PRIO_HIGH)
1621                 val |= E1000_TXDCTL_PRIORITY;
1622         else
1623                 val &= ~E1000_TXDCTL_PRIORITY;
1624
1625         wr32(E1000_I210_TXDCTL(queue), val);
1626 }
1627
1628 static void set_queue_mode(struct e1000_hw *hw, int queue, enum queue_mode mode)
1629 {
1630         u32 val;
1631
1632         WARN_ON(hw->mac.type != e1000_i210);
1633         WARN_ON(queue < 0 || queue > 1);
1634
1635         val = rd32(E1000_I210_TQAVCC(queue));
1636
1637         if (mode == QUEUE_MODE_STREAM_RESERVATION)
1638                 val |= E1000_TQAVCC_QUEUEMODE;
1639         else
1640                 val &= ~E1000_TQAVCC_QUEUEMODE;
1641
1642         wr32(E1000_I210_TQAVCC(queue), val);
1643 }
1644
1645 static bool is_any_cbs_enabled(struct igb_adapter *adapter)
1646 {
1647         int i;
1648
1649         for (i = 0; i < adapter->num_tx_queues; i++) {
1650                 if (adapter->tx_ring[i]->cbs_enable)
1651                         return true;
1652         }
1653
1654         return false;
1655 }
1656
1657 static bool is_any_txtime_enabled(struct igb_adapter *adapter)
1658 {
1659         int i;
1660
1661         for (i = 0; i < adapter->num_tx_queues; i++) {
1662                 if (adapter->tx_ring[i]->launchtime_enable)
1663                         return true;
1664         }
1665
1666         return false;
1667 }
1668
1669 /**
1670  *  igb_config_tx_modes - Configure "Qav Tx mode" features on igb
1671  *  @adapter: pointer to adapter struct
1672  *  @queue: queue number
1673  *
1674  *  Configure CBS and Launchtime for a given hardware queue.
1675  *  Parameters are retrieved from the correct Tx ring, so
1676  *  igb_save_cbs_params() and igb_save_txtime_params() should be used
1677  *  for setting those correctly prior to this function being called.
1678  **/
1679 static void igb_config_tx_modes(struct igb_adapter *adapter, int queue)
1680 {
1681         struct igb_ring *ring = adapter->tx_ring[queue];
1682         struct net_device *netdev = adapter->netdev;
1683         struct e1000_hw *hw = &adapter->hw;
1684         u32 tqavcc, tqavctrl;
1685         u16 value;
1686
1687         WARN_ON(hw->mac.type != e1000_i210);
1688         WARN_ON(queue < 0 || queue > 1);
1689
1690         /* If any of the Qav features is enabled, configure queues as SR and
1691          * with HIGH PRIO. If none is, then configure them with LOW PRIO and
1692          * as SP.
1693          */
1694         if (ring->cbs_enable || ring->launchtime_enable) {
1695                 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_HIGH);
1696                 set_queue_mode(hw, queue, QUEUE_MODE_STREAM_RESERVATION);
1697         } else {
1698                 set_tx_desc_fetch_prio(hw, queue, TX_QUEUE_PRIO_LOW);
1699                 set_queue_mode(hw, queue, QUEUE_MODE_STRICT_PRIORITY);
1700         }
1701
1702         /* If CBS is enabled, set DataTranARB and config its parameters. */
1703         if (ring->cbs_enable || queue == 0) {
1704                 /* i210 does not allow the queue 0 to be in the Strict
1705                  * Priority mode while the Qav mode is enabled, so,
1706                  * instead of disabling strict priority mode, we give
1707                  * queue 0 the maximum of credits possible.
1708                  *
1709                  * See section 8.12.19 of the i210 datasheet, "Note:
1710                  * Queue0 QueueMode must be set to 1b when
1711                  * TransmitMode is set to Qav."
1712                  */
1713                 if (queue == 0 && !ring->cbs_enable) {
1714                         /* max "linkspeed" idleslope in kbps */
1715                         ring->idleslope = 1000000;
1716                         ring->hicredit = ETH_FRAME_LEN;
1717                 }
1718
1719                 /* Always set data transfer arbitration to credit-based
1720                  * shaper algorithm on TQAVCTRL if CBS is enabled for any of
1721                  * the queues.
1722                  */
1723                 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1724                 tqavctrl |= E1000_TQAVCTRL_DATATRANARB;
1725                 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1726
1727                 /* According to i210 datasheet section 7.2.7.7, we should set
1728                  * the 'idleSlope' field from TQAVCC register following the
1729                  * equation:
1730                  *
1731                  * For 100 Mbps link speed:
1732                  *
1733                  *     value = BW * 0x7735 * 0.2                          (E1)
1734                  *
1735                  * For 1000Mbps link speed:
1736                  *
1737                  *     value = BW * 0x7735 * 2                            (E2)
1738                  *
1739                  * E1 and E2 can be merged into one equation as shown below.
1740                  * Note that 'link-speed' is in Mbps.
1741                  *
1742                  *     value = BW * 0x7735 * 2 * link-speed
1743                  *                           --------------               (E3)
1744                  *                                1000
1745                  *
1746                  * 'BW' is the percentage bandwidth out of full link speed
1747                  * which can be found with the following equation. Note that
1748                  * idleSlope here is the parameter from this function which
1749                  * is in kbps.
1750                  *
1751                  *     BW =     idleSlope
1752                  *          -----------------                             (E4)
1753                  *          link-speed * 1000
1754                  *
1755                  * That said, we can come up with a generic equation to
1756                  * calculate the value we should set it TQAVCC register by
1757                  * replacing 'BW' in E3 by E4. The resulting equation is:
1758                  *
1759                  * value =     idleSlope     * 0x7735 * 2 * link-speed
1760                  *         -----------------            --------------    (E5)
1761                  *         link-speed * 1000                 1000
1762                  *
1763                  * 'link-speed' is present in both sides of the fraction so
1764                  * it is canceled out. The final equation is the following:
1765                  *
1766                  *     value = idleSlope * 61034
1767                  *             -----------------                          (E6)
1768                  *                  1000000
1769                  *
1770                  * NOTE: For i210, given the above, we can see that idleslope
1771                  *       is represented in 16.38431 kbps units by the value at
1772                  *       the TQAVCC register (1Gbps / 61034), which reduces
1773                  *       the granularity for idleslope increments.
1774                  *       For instance, if you want to configure a 2576kbps
1775                  *       idleslope, the value to be written on the register
1776                  *       would have to be 157.23. If rounded down, you end
1777                  *       up with less bandwidth available than originally
1778                  *       required (~2572 kbps). If rounded up, you end up
1779                  *       with a higher bandwidth (~2589 kbps). Below the
1780                  *       approach we take is to always round up the
1781                  *       calculated value, so the resulting bandwidth might
1782                  *       be slightly higher for some configurations.
1783                  */
1784                 value = DIV_ROUND_UP_ULL(ring->idleslope * 61034ULL, 1000000);
1785
1786                 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1787                 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1788                 tqavcc |= value;
1789                 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1790
1791                 wr32(E1000_I210_TQAVHC(queue),
1792                      0x80000000 + ring->hicredit * 0x7735);
1793         } else {
1794
1795                 /* Set idleSlope to zero. */
1796                 tqavcc = rd32(E1000_I210_TQAVCC(queue));
1797                 tqavcc &= ~E1000_TQAVCC_IDLESLOPE_MASK;
1798                 wr32(E1000_I210_TQAVCC(queue), tqavcc);
1799
1800                 /* Set hiCredit to zero. */
1801                 wr32(E1000_I210_TQAVHC(queue), 0);
1802
1803                 /* If CBS is not enabled for any queues anymore, then return to
1804                  * the default state of Data Transmission Arbitration on
1805                  * TQAVCTRL.
1806                  */
1807                 if (!is_any_cbs_enabled(adapter)) {
1808                         tqavctrl = rd32(E1000_I210_TQAVCTRL);
1809                         tqavctrl &= ~E1000_TQAVCTRL_DATATRANARB;
1810                         wr32(E1000_I210_TQAVCTRL, tqavctrl);
1811                 }
1812         }
1813
1814         /* If LaunchTime is enabled, set DataTranTIM. */
1815         if (ring->launchtime_enable) {
1816                 /* Always set DataTranTIM on TQAVCTRL if LaunchTime is enabled
1817                  * for any of the SR queues, and configure fetchtime delta.
1818                  * XXX NOTE:
1819                  *     - LaunchTime will be enabled for all SR queues.
1820                  *     - A fixed offset can be added relative to the launch
1821                  *       time of all packets if configured at reg LAUNCH_OS0.
1822                  *       We are keeping it as 0 for now (default value).
1823                  */
1824                 tqavctrl = rd32(E1000_I210_TQAVCTRL);
1825                 tqavctrl |= E1000_TQAVCTRL_DATATRANTIM |
1826                        E1000_TQAVCTRL_FETCHTIME_DELTA;
1827                 wr32(E1000_I210_TQAVCTRL, tqavctrl);
1828         } else {
1829                 /* If Launchtime is not enabled for any SR queues anymore,
1830                  * then clear DataTranTIM on TQAVCTRL and clear fetchtime delta,
1831                  * effectively disabling Launchtime.
1832                  */
1833                 if (!is_any_txtime_enabled(adapter)) {
1834                         tqavctrl = rd32(E1000_I210_TQAVCTRL);
1835                         tqavctrl &= ~E1000_TQAVCTRL_DATATRANTIM;
1836                         tqavctrl &= ~E1000_TQAVCTRL_FETCHTIME_DELTA;
1837                         wr32(E1000_I210_TQAVCTRL, tqavctrl);
1838                 }
1839         }
1840
1841         /* XXX: In i210 controller the sendSlope and loCredit parameters from
1842          * CBS are not configurable by software so we don't do any 'controller
1843          * configuration' in respect to these parameters.
1844          */
1845
1846         netdev_dbg(netdev, "Qav Tx mode: cbs %s, launchtime %s, queue %d idleslope %d sendslope %d hiCredit %d locredit %d\n",
1847                    ring->cbs_enable ? "enabled" : "disabled",
1848                    ring->launchtime_enable ? "enabled" : "disabled",
1849                    queue,
1850                    ring->idleslope, ring->sendslope,
1851                    ring->hicredit, ring->locredit);
1852 }
1853
1854 static int igb_save_txtime_params(struct igb_adapter *adapter, int queue,
1855                                   bool enable)
1856 {
1857         struct igb_ring *ring;
1858
1859         if (queue < 0 || queue > adapter->num_tx_queues)
1860                 return -EINVAL;
1861
1862         ring = adapter->tx_ring[queue];
1863         ring->launchtime_enable = enable;
1864
1865         return 0;
1866 }
1867
1868 static int igb_save_cbs_params(struct igb_adapter *adapter, int queue,
1869                                bool enable, int idleslope, int sendslope,
1870                                int hicredit, int locredit)
1871 {
1872         struct igb_ring *ring;
1873
1874         if (queue < 0 || queue > adapter->num_tx_queues)
1875                 return -EINVAL;
1876
1877         ring = adapter->tx_ring[queue];
1878
1879         ring->cbs_enable = enable;
1880         ring->idleslope = idleslope;
1881         ring->sendslope = sendslope;
1882         ring->hicredit = hicredit;
1883         ring->locredit = locredit;
1884
1885         return 0;
1886 }
1887
1888 /**
1889  *  igb_setup_tx_mode - Switch to/from Qav Tx mode when applicable
1890  *  @adapter: pointer to adapter struct
1891  *
1892  *  Configure TQAVCTRL register switching the controller's Tx mode
1893  *  if FQTSS mode is enabled or disabled. Additionally, will issue
1894  *  a call to igb_config_tx_modes() per queue so any previously saved
1895  *  Tx parameters are applied.
1896  **/
1897 static void igb_setup_tx_mode(struct igb_adapter *adapter)
1898 {
1899         struct net_device *netdev = adapter->netdev;
1900         struct e1000_hw *hw = &adapter->hw;
1901         u32 val;
1902
1903         /* Only i210 controller supports changing the transmission mode. */
1904         if (hw->mac.type != e1000_i210)
1905                 return;
1906
1907         if (is_fqtss_enabled(adapter)) {
1908                 int i, max_queue;
1909
1910                 /* Configure TQAVCTRL register: set transmit mode to 'Qav',
1911                  * set data fetch arbitration to 'round robin', set SP_WAIT_SR
1912                  * so SP queues wait for SR ones.
1913                  */
1914                 val = rd32(E1000_I210_TQAVCTRL);
1915                 val |= E1000_TQAVCTRL_XMIT_MODE | E1000_TQAVCTRL_SP_WAIT_SR;
1916                 val &= ~E1000_TQAVCTRL_DATAFETCHARB;
1917                 wr32(E1000_I210_TQAVCTRL, val);
1918
1919                 /* Configure Tx and Rx packet buffers sizes as described in
1920                  * i210 datasheet section 7.2.7.7.
1921                  */
1922                 val = rd32(E1000_TXPBS);
1923                 val &= ~I210_TXPBSIZE_MASK;
1924                 val |= I210_TXPBSIZE_PB0_6KB | I210_TXPBSIZE_PB1_6KB |
1925                         I210_TXPBSIZE_PB2_6KB | I210_TXPBSIZE_PB3_6KB;
1926                 wr32(E1000_TXPBS, val);
1927
1928                 val = rd32(E1000_RXPBS);
1929                 val &= ~I210_RXPBSIZE_MASK;
1930                 val |= I210_RXPBSIZE_PB_30KB;
1931                 wr32(E1000_RXPBS, val);
1932
1933                 /* Section 8.12.9 states that MAX_TPKT_SIZE from DTXMXPKTSZ
1934                  * register should not exceed the buffer size programmed in
1935                  * TXPBS. The smallest buffer size programmed in TXPBS is 4kB
1936                  * so according to the datasheet we should set MAX_TPKT_SIZE to
1937                  * 4kB / 64.
1938                  *
1939                  * However, when we do so, no frame from queue 2 and 3 are
1940                  * transmitted.  It seems the MAX_TPKT_SIZE should not be great
1941                  * or _equal_ to the buffer size programmed in TXPBS. For this
1942                  * reason, we set set MAX_ TPKT_SIZE to (4kB - 1) / 64.
1943                  */
1944                 val = (4096 - 1) / 64;
1945                 wr32(E1000_I210_DTXMXPKTSZ, val);
1946
1947                 /* Since FQTSS mode is enabled, apply any CBS configuration
1948                  * previously set. If no previous CBS configuration has been
1949                  * done, then the initial configuration is applied, which means
1950                  * CBS is disabled.
1951                  */
1952                 max_queue = (adapter->num_tx_queues < I210_SR_QUEUES_NUM) ?
1953                             adapter->num_tx_queues : I210_SR_QUEUES_NUM;
1954
1955                 for (i = 0; i < max_queue; i++) {
1956                         igb_config_tx_modes(adapter, i);
1957                 }
1958         } else {
1959                 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
1960                 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
1961                 wr32(E1000_I210_DTXMXPKTSZ, I210_DTXMXPKTSZ_DEFAULT);
1962
1963                 val = rd32(E1000_I210_TQAVCTRL);
1964                 /* According to Section 8.12.21, the other flags we've set when
1965                  * enabling FQTSS are not relevant when disabling FQTSS so we
1966                  * don't set they here.
1967                  */
1968                 val &= ~E1000_TQAVCTRL_XMIT_MODE;
1969                 wr32(E1000_I210_TQAVCTRL, val);
1970         }
1971
1972         netdev_dbg(netdev, "FQTSS %s\n", (is_fqtss_enabled(adapter)) ?
1973                    "enabled" : "disabled");
1974 }
1975
1976 /**
1977  *  igb_configure - configure the hardware for RX and TX
1978  *  @adapter: private board structure
1979  **/
1980 static void igb_configure(struct igb_adapter *adapter)
1981 {
1982         struct net_device *netdev = adapter->netdev;
1983         int i;
1984
1985         igb_get_hw_control(adapter);
1986         igb_set_rx_mode(netdev);
1987         igb_setup_tx_mode(adapter);
1988
1989         igb_restore_vlan(adapter);
1990
1991         igb_setup_tctl(adapter);
1992         igb_setup_mrqc(adapter);
1993         igb_setup_rctl(adapter);
1994
1995         igb_nfc_filter_restore(adapter);
1996         igb_configure_tx(adapter);
1997         igb_configure_rx(adapter);
1998
1999         igb_rx_fifo_flush_82575(&adapter->hw);
2000
2001         /* call igb_desc_unused which always leaves
2002          * at least 1 descriptor unused to make sure
2003          * next_to_use != next_to_clean
2004          */
2005         for (i = 0; i < adapter->num_rx_queues; i++) {
2006                 struct igb_ring *ring = adapter->rx_ring[i];
2007                 igb_alloc_rx_buffers(ring, igb_desc_unused(ring));
2008         }
2009 }
2010
2011 /**
2012  *  igb_power_up_link - Power up the phy/serdes link
2013  *  @adapter: address of board private structure
2014  **/
2015 void igb_power_up_link(struct igb_adapter *adapter)
2016 {
2017         igb_reset_phy(&adapter->hw);
2018
2019         if (adapter->hw.phy.media_type == e1000_media_type_copper)
2020                 igb_power_up_phy_copper(&adapter->hw);
2021         else
2022                 igb_power_up_serdes_link_82575(&adapter->hw);
2023
2024         igb_setup_link(&adapter->hw);
2025 }
2026
2027 /**
2028  *  igb_power_down_link - Power down the phy/serdes link
2029  *  @adapter: address of board private structure
2030  */
2031 static void igb_power_down_link(struct igb_adapter *adapter)
2032 {
2033         if (adapter->hw.phy.media_type == e1000_media_type_copper)
2034                 igb_power_down_phy_copper_82575(&adapter->hw);
2035         else
2036                 igb_shutdown_serdes_link_82575(&adapter->hw);
2037 }
2038
2039 /**
2040  * igb_check_swap_media -  Detect and switch function for Media Auto Sense
2041  * @adapter: address of the board private structure
2042  **/
2043 static void igb_check_swap_media(struct igb_adapter *adapter)
2044 {
2045         struct e1000_hw *hw = &adapter->hw;
2046         u32 ctrl_ext, connsw;
2047         bool swap_now = false;
2048
2049         ctrl_ext = rd32(E1000_CTRL_EXT);
2050         connsw = rd32(E1000_CONNSW);
2051
2052         /* need to live swap if current media is copper and we have fiber/serdes
2053          * to go to.
2054          */
2055
2056         if ((hw->phy.media_type == e1000_media_type_copper) &&
2057             (!(connsw & E1000_CONNSW_AUTOSENSE_EN))) {
2058                 swap_now = true;
2059         } else if ((hw->phy.media_type != e1000_media_type_copper) &&
2060                    !(connsw & E1000_CONNSW_SERDESD)) {
2061                 /* copper signal takes time to appear */
2062                 if (adapter->copper_tries < 4) {
2063                         adapter->copper_tries++;
2064                         connsw |= E1000_CONNSW_AUTOSENSE_CONF;
2065                         wr32(E1000_CONNSW, connsw);
2066                         return;
2067                 } else {
2068                         adapter->copper_tries = 0;
2069                         if ((connsw & E1000_CONNSW_PHYSD) &&
2070                             (!(connsw & E1000_CONNSW_PHY_PDN))) {
2071                                 swap_now = true;
2072                                 connsw &= ~E1000_CONNSW_AUTOSENSE_CONF;
2073                                 wr32(E1000_CONNSW, connsw);
2074                         }
2075                 }
2076         }
2077
2078         if (!swap_now)
2079                 return;
2080
2081         switch (hw->phy.media_type) {
2082         case e1000_media_type_copper:
2083                 netdev_info(adapter->netdev,
2084                         "MAS: changing media to fiber/serdes\n");
2085                 ctrl_ext |=
2086                         E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2087                 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2088                 adapter->copper_tries = 0;
2089                 break;
2090         case e1000_media_type_internal_serdes:
2091         case e1000_media_type_fiber:
2092                 netdev_info(adapter->netdev,
2093                         "MAS: changing media to copper\n");
2094                 ctrl_ext &=
2095                         ~E1000_CTRL_EXT_LINK_MODE_PCIE_SERDES;
2096                 adapter->flags |= IGB_FLAG_MEDIA_RESET;
2097                 break;
2098         default:
2099                 /* shouldn't get here during regular operation */
2100                 netdev_err(adapter->netdev,
2101                         "AMS: Invalid media type found, returning\n");
2102                 break;
2103         }
2104         wr32(E1000_CTRL_EXT, ctrl_ext);
2105 }
2106
2107 /**
2108  *  igb_up - Open the interface and prepare it to handle traffic
2109  *  @adapter: board private structure
2110  **/
2111 int igb_up(struct igb_adapter *adapter)
2112 {
2113         struct e1000_hw *hw = &adapter->hw;
2114         int i;
2115
2116         /* hardware has been reset, we need to reload some things */
2117         igb_configure(adapter);
2118
2119         clear_bit(__IGB_DOWN, &adapter->state);
2120
2121         for (i = 0; i < adapter->num_q_vectors; i++)
2122                 napi_enable(&(adapter->q_vector[i]->napi));
2123
2124         if (adapter->flags & IGB_FLAG_HAS_MSIX)
2125                 igb_configure_msix(adapter);
2126         else
2127                 igb_assign_vector(adapter->q_vector[0], 0);
2128
2129         /* Clear any pending interrupts. */
2130         rd32(E1000_TSICR);
2131         rd32(E1000_ICR);
2132         igb_irq_enable(adapter);
2133
2134         /* notify VFs that reset has been completed */
2135         if (adapter->vfs_allocated_count) {
2136                 u32 reg_data = rd32(E1000_CTRL_EXT);
2137
2138                 reg_data |= E1000_CTRL_EXT_PFRSTD;
2139                 wr32(E1000_CTRL_EXT, reg_data);
2140         }
2141
2142         netif_tx_start_all_queues(adapter->netdev);
2143
2144         /* start the watchdog. */
2145         hw->mac.get_link_status = 1;
2146         schedule_work(&adapter->watchdog_task);
2147
2148         if ((adapter->flags & IGB_FLAG_EEE) &&
2149             (!hw->dev_spec._82575.eee_disable))
2150                 adapter->eee_advert = MDIO_EEE_100TX | MDIO_EEE_1000T;
2151
2152         return 0;
2153 }
2154
2155 void igb_down(struct igb_adapter *adapter)
2156 {
2157         struct net_device *netdev = adapter->netdev;
2158         struct e1000_hw *hw = &adapter->hw;
2159         u32 tctl, rctl;
2160         int i;
2161
2162         /* signal that we're down so the interrupt handler does not
2163          * reschedule our watchdog timer
2164          */
2165         set_bit(__IGB_DOWN, &adapter->state);
2166
2167         /* disable receives in the hardware */
2168         rctl = rd32(E1000_RCTL);
2169         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2170         /* flush and sleep below */
2171
2172         igb_nfc_filter_exit(adapter);
2173
2174         netif_carrier_off(netdev);
2175         netif_tx_stop_all_queues(netdev);
2176
2177         /* disable transmits in the hardware */
2178         tctl = rd32(E1000_TCTL);
2179         tctl &= ~E1000_TCTL_EN;
2180         wr32(E1000_TCTL, tctl);
2181         /* flush both disables and wait for them to finish */
2182         wrfl();
2183         usleep_range(10000, 11000);
2184
2185         igb_irq_disable(adapter);
2186
2187         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
2188
2189         for (i = 0; i < adapter->num_q_vectors; i++) {
2190                 if (adapter->q_vector[i]) {
2191                         napi_synchronize(&adapter->q_vector[i]->napi);
2192                         napi_disable(&adapter->q_vector[i]->napi);
2193                 }
2194         }
2195
2196         del_timer_sync(&adapter->watchdog_timer);
2197         del_timer_sync(&adapter->phy_info_timer);
2198
2199         /* record the stats before reset*/
2200         spin_lock(&adapter->stats64_lock);
2201         igb_update_stats(adapter);
2202         spin_unlock(&adapter->stats64_lock);
2203
2204         adapter->link_speed = 0;
2205         adapter->link_duplex = 0;
2206
2207         if (!pci_channel_offline(adapter->pdev))
2208                 igb_reset(adapter);
2209
2210         /* clear VLAN promisc flag so VFTA will be updated if necessary */
2211         adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
2212
2213         igb_clean_all_tx_rings(adapter);
2214         igb_clean_all_rx_rings(adapter);
2215 #ifdef CONFIG_IGB_DCA
2216
2217         /* since we reset the hardware DCA settings were cleared */
2218         igb_setup_dca(adapter);
2219 #endif
2220 }
2221
2222 void igb_reinit_locked(struct igb_adapter *adapter)
2223 {
2224         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2225                 usleep_range(1000, 2000);
2226         igb_down(adapter);
2227         igb_up(adapter);
2228         clear_bit(__IGB_RESETTING, &adapter->state);
2229 }
2230
2231 /** igb_enable_mas - Media Autosense re-enable after swap
2232  *
2233  * @adapter: adapter struct
2234  **/
2235 static void igb_enable_mas(struct igb_adapter *adapter)
2236 {
2237         struct e1000_hw *hw = &adapter->hw;
2238         u32 connsw = rd32(E1000_CONNSW);
2239
2240         /* configure for SerDes media detect */
2241         if ((hw->phy.media_type == e1000_media_type_copper) &&
2242             (!(connsw & E1000_CONNSW_SERDESD))) {
2243                 connsw |= E1000_CONNSW_ENRGSRC;
2244                 connsw |= E1000_CONNSW_AUTOSENSE_EN;
2245                 wr32(E1000_CONNSW, connsw);
2246                 wrfl();
2247         }
2248 }
2249
2250 void igb_reset(struct igb_adapter *adapter)
2251 {
2252         struct pci_dev *pdev = adapter->pdev;
2253         struct e1000_hw *hw = &adapter->hw;
2254         struct e1000_mac_info *mac = &hw->mac;
2255         struct e1000_fc_info *fc = &hw->fc;
2256         u32 pba, hwm;
2257
2258         /* Repartition Pba for greater than 9k mtu
2259          * To take effect CTRL.RST is required.
2260          */
2261         switch (mac->type) {
2262         case e1000_i350:
2263         case e1000_i354:
2264         case e1000_82580:
2265                 pba = rd32(E1000_RXPBS);
2266                 pba = igb_rxpbs_adjust_82580(pba);
2267                 break;
2268         case e1000_82576:
2269                 pba = rd32(E1000_RXPBS);
2270                 pba &= E1000_RXPBS_SIZE_MASK_82576;
2271                 break;
2272         case e1000_82575:
2273         case e1000_i210:
2274         case e1000_i211:
2275         default:
2276                 pba = E1000_PBA_34K;
2277                 break;
2278         }
2279
2280         if (mac->type == e1000_82575) {
2281                 u32 min_rx_space, min_tx_space, needed_tx_space;
2282
2283                 /* write Rx PBA so that hardware can report correct Tx PBA */
2284                 wr32(E1000_PBA, pba);
2285
2286                 /* To maintain wire speed transmits, the Tx FIFO should be
2287                  * large enough to accommodate two full transmit packets,
2288                  * rounded up to the next 1KB and expressed in KB.  Likewise,
2289                  * the Rx FIFO should be large enough to accommodate at least
2290                  * one full receive packet and is similarly rounded up and
2291                  * expressed in KB.
2292                  */
2293                 min_rx_space = DIV_ROUND_UP(MAX_JUMBO_FRAME_SIZE, 1024);
2294
2295                 /* The Tx FIFO also stores 16 bytes of information about the Tx
2296                  * but don't include Ethernet FCS because hardware appends it.
2297                  * We only need to round down to the nearest 512 byte block
2298                  * count since the value we care about is 2 frames, not 1.
2299                  */
2300                 min_tx_space = adapter->max_frame_size;
2301                 min_tx_space += sizeof(union e1000_adv_tx_desc) - ETH_FCS_LEN;
2302                 min_tx_space = DIV_ROUND_UP(min_tx_space, 512);
2303
2304                 /* upper 16 bits has Tx packet buffer allocation size in KB */
2305                 needed_tx_space = min_tx_space - (rd32(E1000_PBA) >> 16);
2306
2307                 /* If current Tx allocation is less than the min Tx FIFO size,
2308                  * and the min Tx FIFO size is less than the current Rx FIFO
2309                  * allocation, take space away from current Rx allocation.
2310                  */
2311                 if (needed_tx_space < pba) {
2312                         pba -= needed_tx_space;
2313
2314                         /* if short on Rx space, Rx wins and must trump Tx
2315                          * adjustment
2316                          */
2317                         if (pba < min_rx_space)
2318                                 pba = min_rx_space;
2319                 }
2320
2321                 /* adjust PBA for jumbo frames */
2322                 wr32(E1000_PBA, pba);
2323         }
2324
2325         /* flow control settings
2326          * The high water mark must be low enough to fit one full frame
2327          * after transmitting the pause frame.  As such we must have enough
2328          * space to allow for us to complete our current transmit and then
2329          * receive the frame that is in progress from the link partner.
2330          * Set it to:
2331          * - the full Rx FIFO size minus one full Tx plus one full Rx frame
2332          */
2333         hwm = (pba << 10) - (adapter->max_frame_size + MAX_JUMBO_FRAME_SIZE);
2334
2335         fc->high_water = hwm & 0xFFFFFFF0;      /* 16-byte granularity */
2336         fc->low_water = fc->high_water - 16;
2337         fc->pause_time = 0xFFFF;
2338         fc->send_xon = 1;
2339         fc->current_mode = fc->requested_mode;
2340
2341         /* disable receive for all VFs and wait one second */
2342         if (adapter->vfs_allocated_count) {
2343                 int i;
2344
2345                 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
2346                         adapter->vf_data[i].flags &= IGB_VF_FLAG_PF_SET_MAC;
2347
2348                 /* ping all the active vfs to let them know we are going down */
2349                 igb_ping_all_vfs(adapter);
2350
2351                 /* disable transmits and receives */
2352                 wr32(E1000_VFRE, 0);
2353                 wr32(E1000_VFTE, 0);
2354         }
2355
2356         /* Allow time for pending master requests to run */
2357         hw->mac.ops.reset_hw(hw);
2358         wr32(E1000_WUC, 0);
2359
2360         if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
2361                 /* need to resetup here after media swap */
2362                 adapter->ei.get_invariants(hw);
2363                 adapter->flags &= ~IGB_FLAG_MEDIA_RESET;
2364         }
2365         if ((mac->type == e1000_82575 || mac->type == e1000_i350) &&
2366             (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
2367                 igb_enable_mas(adapter);
2368         }
2369         if (hw->mac.ops.init_hw(hw))
2370                 dev_err(&pdev->dev, "Hardware Error\n");
2371
2372         /* RAR registers were cleared during init_hw, clear mac table */
2373         igb_flush_mac_table(adapter);
2374         __dev_uc_unsync(adapter->netdev, NULL);
2375
2376         /* Recover default RAR entry */
2377         igb_set_default_mac_filter(adapter);
2378
2379         /* Flow control settings reset on hardware reset, so guarantee flow
2380          * control is off when forcing speed.
2381          */
2382         if (!hw->mac.autoneg)
2383                 igb_force_mac_fc(hw);
2384
2385         igb_init_dmac(adapter, pba);
2386 #ifdef CONFIG_IGB_HWMON
2387         /* Re-initialize the thermal sensor on i350 devices. */
2388         if (!test_bit(__IGB_DOWN, &adapter->state)) {
2389                 if (mac->type == e1000_i350 && hw->bus.func == 0) {
2390                         /* If present, re-initialize the external thermal sensor
2391                          * interface.
2392                          */
2393                         if (adapter->ets)
2394                                 mac->ops.init_thermal_sensor_thresh(hw);
2395                 }
2396         }
2397 #endif
2398         /* Re-establish EEE setting */
2399         if (hw->phy.media_type == e1000_media_type_copper) {
2400                 switch (mac->type) {
2401                 case e1000_i350:
2402                 case e1000_i210:
2403                 case e1000_i211:
2404                         igb_set_eee_i350(hw, true, true);
2405                         break;
2406                 case e1000_i354:
2407                         igb_set_eee_i354(hw, true, true);
2408                         break;
2409                 default:
2410                         break;
2411                 }
2412         }
2413         if (!netif_running(adapter->netdev))
2414                 igb_power_down_link(adapter);
2415
2416         igb_update_mng_vlan(adapter);
2417
2418         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
2419         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
2420
2421         /* Re-enable PTP, where applicable. */
2422         if (adapter->ptp_flags & IGB_PTP_ENABLED)
2423                 igb_ptp_reset(adapter);
2424
2425         igb_get_phy_info(hw);
2426 }
2427
2428 static netdev_features_t igb_fix_features(struct net_device *netdev,
2429         netdev_features_t features)
2430 {
2431         /* Since there is no support for separate Rx/Tx vlan accel
2432          * enable/disable make sure Tx flag is always in same state as Rx.
2433          */
2434         if (features & NETIF_F_HW_VLAN_CTAG_RX)
2435                 features |= NETIF_F_HW_VLAN_CTAG_TX;
2436         else
2437                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
2438
2439         return features;
2440 }
2441
2442 static int igb_set_features(struct net_device *netdev,
2443         netdev_features_t features)
2444 {
2445         netdev_features_t changed = netdev->features ^ features;
2446         struct igb_adapter *adapter = netdev_priv(netdev);
2447
2448         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
2449                 igb_vlan_mode(netdev, features);
2450
2451         if (!(changed & (NETIF_F_RXALL | NETIF_F_NTUPLE)))
2452                 return 0;
2453
2454         if (!(features & NETIF_F_NTUPLE)) {
2455                 struct hlist_node *node2;
2456                 struct igb_nfc_filter *rule;
2457
2458                 spin_lock(&adapter->nfc_lock);
2459                 hlist_for_each_entry_safe(rule, node2,
2460                                           &adapter->nfc_filter_list, nfc_node) {
2461                         igb_erase_filter(adapter, rule);
2462                         hlist_del(&rule->nfc_node);
2463                         kfree(rule);
2464                 }
2465                 spin_unlock(&adapter->nfc_lock);
2466                 adapter->nfc_filter_count = 0;
2467         }
2468
2469         netdev->features = features;
2470
2471         if (netif_running(netdev))
2472                 igb_reinit_locked(adapter);
2473         else
2474                 igb_reset(adapter);
2475
2476         return 1;
2477 }
2478
2479 static int igb_ndo_fdb_add(struct ndmsg *ndm, struct nlattr *tb[],
2480                            struct net_device *dev,
2481                            const unsigned char *addr, u16 vid,
2482                            u16 flags,
2483                            struct netlink_ext_ack *extack)
2484 {
2485         /* guarantee we can provide a unique filter for the unicast address */
2486         if (is_unicast_ether_addr(addr) || is_link_local_ether_addr(addr)) {
2487                 struct igb_adapter *adapter = netdev_priv(dev);
2488                 int vfn = adapter->vfs_allocated_count;
2489
2490                 if (netdev_uc_count(dev) >= igb_available_rars(adapter, vfn))
2491                         return -ENOMEM;
2492         }
2493
2494         return ndo_dflt_fdb_add(ndm, tb, dev, addr, vid, flags);
2495 }
2496
2497 #define IGB_MAX_MAC_HDR_LEN     127
2498 #define IGB_MAX_NETWORK_HDR_LEN 511
2499
2500 static netdev_features_t
2501 igb_features_check(struct sk_buff *skb, struct net_device *dev,
2502                    netdev_features_t features)
2503 {
2504         unsigned int network_hdr_len, mac_hdr_len;
2505
2506         /* Make certain the headers can be described by a context descriptor */
2507         mac_hdr_len = skb_network_header(skb) - skb->data;
2508         if (unlikely(mac_hdr_len > IGB_MAX_MAC_HDR_LEN))
2509                 return features & ~(NETIF_F_HW_CSUM |
2510                                     NETIF_F_SCTP_CRC |
2511                                     NETIF_F_GSO_UDP_L4 |
2512                                     NETIF_F_HW_VLAN_CTAG_TX |
2513                                     NETIF_F_TSO |
2514                                     NETIF_F_TSO6);
2515
2516         network_hdr_len = skb_checksum_start(skb) - skb_network_header(skb);
2517         if (unlikely(network_hdr_len >  IGB_MAX_NETWORK_HDR_LEN))
2518                 return features & ~(NETIF_F_HW_CSUM |
2519                                     NETIF_F_SCTP_CRC |
2520                                     NETIF_F_GSO_UDP_L4 |
2521                                     NETIF_F_TSO |
2522                                     NETIF_F_TSO6);
2523
2524         /* We can only support IPV4 TSO in tunnels if we can mangle the
2525          * inner IP ID field, so strip TSO if MANGLEID is not supported.
2526          */
2527         if (skb->encapsulation && !(features & NETIF_F_TSO_MANGLEID))
2528                 features &= ~NETIF_F_TSO;
2529
2530         return features;
2531 }
2532
2533 static void igb_offload_apply(struct igb_adapter *adapter, s32 queue)
2534 {
2535         if (!is_fqtss_enabled(adapter)) {
2536                 enable_fqtss(adapter, true);
2537                 return;
2538         }
2539
2540         igb_config_tx_modes(adapter, queue);
2541
2542         if (!is_any_cbs_enabled(adapter) && !is_any_txtime_enabled(adapter))
2543                 enable_fqtss(adapter, false);
2544 }
2545
2546 static int igb_offload_cbs(struct igb_adapter *adapter,
2547                            struct tc_cbs_qopt_offload *qopt)
2548 {
2549         struct e1000_hw *hw = &adapter->hw;
2550         int err;
2551
2552         /* CBS offloading is only supported by i210 controller. */
2553         if (hw->mac.type != e1000_i210)
2554                 return -EOPNOTSUPP;
2555
2556         /* CBS offloading is only supported by queue 0 and queue 1. */
2557         if (qopt->queue < 0 || qopt->queue > 1)
2558                 return -EINVAL;
2559
2560         err = igb_save_cbs_params(adapter, qopt->queue, qopt->enable,
2561                                   qopt->idleslope, qopt->sendslope,
2562                                   qopt->hicredit, qopt->locredit);
2563         if (err)
2564                 return err;
2565
2566         igb_offload_apply(adapter, qopt->queue);
2567
2568         return 0;
2569 }
2570
2571 #define ETHER_TYPE_FULL_MASK ((__force __be16)~0)
2572 #define VLAN_PRIO_FULL_MASK (0x07)
2573
2574 static int igb_parse_cls_flower(struct igb_adapter *adapter,
2575                                 struct flow_cls_offload *f,
2576                                 int traffic_class,
2577                                 struct igb_nfc_filter *input)
2578 {
2579         struct flow_rule *rule = flow_cls_offload_flow_rule(f);
2580         struct flow_dissector *dissector = rule->match.dissector;
2581         struct netlink_ext_ack *extack = f->common.extack;
2582
2583         if (dissector->used_keys &
2584             ~(BIT(FLOW_DISSECTOR_KEY_BASIC) |
2585               BIT(FLOW_DISSECTOR_KEY_CONTROL) |
2586               BIT(FLOW_DISSECTOR_KEY_ETH_ADDRS) |
2587               BIT(FLOW_DISSECTOR_KEY_VLAN))) {
2588                 NL_SET_ERR_MSG_MOD(extack,
2589                                    "Unsupported key used, only BASIC, CONTROL, ETH_ADDRS and VLAN are supported");
2590                 return -EOPNOTSUPP;
2591         }
2592
2593         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_ETH_ADDRS)) {
2594                 struct flow_match_eth_addrs match;
2595
2596                 flow_rule_match_eth_addrs(rule, &match);
2597                 if (!is_zero_ether_addr(match.mask->dst)) {
2598                         if (!is_broadcast_ether_addr(match.mask->dst)) {
2599                                 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for destination MAC address");
2600                                 return -EINVAL;
2601                         }
2602
2603                         input->filter.match_flags |=
2604                                 IGB_FILTER_FLAG_DST_MAC_ADDR;
2605                         ether_addr_copy(input->filter.dst_addr, match.key->dst);
2606                 }
2607
2608                 if (!is_zero_ether_addr(match.mask->src)) {
2609                         if (!is_broadcast_ether_addr(match.mask->src)) {
2610                                 NL_SET_ERR_MSG_MOD(extack, "Only full masks are supported for source MAC address");
2611                                 return -EINVAL;
2612                         }
2613
2614                         input->filter.match_flags |=
2615                                 IGB_FILTER_FLAG_SRC_MAC_ADDR;
2616                         ether_addr_copy(input->filter.src_addr, match.key->src);
2617                 }
2618         }
2619
2620         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_BASIC)) {
2621                 struct flow_match_basic match;
2622
2623                 flow_rule_match_basic(rule, &match);
2624                 if (match.mask->n_proto) {
2625                         if (match.mask->n_proto != ETHER_TYPE_FULL_MASK) {
2626                                 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for EtherType filter");
2627                                 return -EINVAL;
2628                         }
2629
2630                         input->filter.match_flags |= IGB_FILTER_FLAG_ETHER_TYPE;
2631                         input->filter.etype = match.key->n_proto;
2632                 }
2633         }
2634
2635         if (flow_rule_match_key(rule, FLOW_DISSECTOR_KEY_VLAN)) {
2636                 struct flow_match_vlan match;
2637
2638                 flow_rule_match_vlan(rule, &match);
2639                 if (match.mask->vlan_priority) {
2640                         if (match.mask->vlan_priority != VLAN_PRIO_FULL_MASK) {
2641                                 NL_SET_ERR_MSG_MOD(extack, "Only full mask is supported for VLAN priority");
2642                                 return -EINVAL;
2643                         }
2644
2645                         input->filter.match_flags |= IGB_FILTER_FLAG_VLAN_TCI;
2646                         input->filter.vlan_tci =
2647                                 (__force __be16)match.key->vlan_priority;
2648                 }
2649         }
2650
2651         input->action = traffic_class;
2652         input->cookie = f->cookie;
2653
2654         return 0;
2655 }
2656
2657 static int igb_configure_clsflower(struct igb_adapter *adapter,
2658                                    struct flow_cls_offload *cls_flower)
2659 {
2660         struct netlink_ext_ack *extack = cls_flower->common.extack;
2661         struct igb_nfc_filter *filter, *f;
2662         int err, tc;
2663
2664         tc = tc_classid_to_hwtc(adapter->netdev, cls_flower->classid);
2665         if (tc < 0) {
2666                 NL_SET_ERR_MSG_MOD(extack, "Invalid traffic class");
2667                 return -EINVAL;
2668         }
2669
2670         filter = kzalloc(sizeof(*filter), GFP_KERNEL);
2671         if (!filter)
2672                 return -ENOMEM;
2673
2674         err = igb_parse_cls_flower(adapter, cls_flower, tc, filter);
2675         if (err < 0)
2676                 goto err_parse;
2677
2678         spin_lock(&adapter->nfc_lock);
2679
2680         hlist_for_each_entry(f, &adapter->nfc_filter_list, nfc_node) {
2681                 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2682                         err = -EEXIST;
2683                         NL_SET_ERR_MSG_MOD(extack,
2684                                            "This filter is already set in ethtool");
2685                         goto err_locked;
2686                 }
2687         }
2688
2689         hlist_for_each_entry(f, &adapter->cls_flower_list, nfc_node) {
2690                 if (!memcmp(&f->filter, &filter->filter, sizeof(f->filter))) {
2691                         err = -EEXIST;
2692                         NL_SET_ERR_MSG_MOD(extack,
2693                                            "This filter is already set in cls_flower");
2694                         goto err_locked;
2695                 }
2696         }
2697
2698         err = igb_add_filter(adapter, filter);
2699         if (err < 0) {
2700                 NL_SET_ERR_MSG_MOD(extack, "Could not add filter to the adapter");
2701                 goto err_locked;
2702         }
2703
2704         hlist_add_head(&filter->nfc_node, &adapter->cls_flower_list);
2705
2706         spin_unlock(&adapter->nfc_lock);
2707
2708         return 0;
2709
2710 err_locked:
2711         spin_unlock(&adapter->nfc_lock);
2712
2713 err_parse:
2714         kfree(filter);
2715
2716         return err;
2717 }
2718
2719 static int igb_delete_clsflower(struct igb_adapter *adapter,
2720                                 struct flow_cls_offload *cls_flower)
2721 {
2722         struct igb_nfc_filter *filter;
2723         int err;
2724
2725         spin_lock(&adapter->nfc_lock);
2726
2727         hlist_for_each_entry(filter, &adapter->cls_flower_list, nfc_node)
2728                 if (filter->cookie == cls_flower->cookie)
2729                         break;
2730
2731         if (!filter) {
2732                 err = -ENOENT;
2733                 goto out;
2734         }
2735
2736         err = igb_erase_filter(adapter, filter);
2737         if (err < 0)
2738                 goto out;
2739
2740         hlist_del(&filter->nfc_node);
2741         kfree(filter);
2742
2743 out:
2744         spin_unlock(&adapter->nfc_lock);
2745
2746         return err;
2747 }
2748
2749 static int igb_setup_tc_cls_flower(struct igb_adapter *adapter,
2750                                    struct flow_cls_offload *cls_flower)
2751 {
2752         switch (cls_flower->command) {
2753         case FLOW_CLS_REPLACE:
2754                 return igb_configure_clsflower(adapter, cls_flower);
2755         case FLOW_CLS_DESTROY:
2756                 return igb_delete_clsflower(adapter, cls_flower);
2757         case FLOW_CLS_STATS:
2758                 return -EOPNOTSUPP;
2759         default:
2760                 return -EOPNOTSUPP;
2761         }
2762 }
2763
2764 static int igb_setup_tc_block_cb(enum tc_setup_type type, void *type_data,
2765                                  void *cb_priv)
2766 {
2767         struct igb_adapter *adapter = cb_priv;
2768
2769         if (!tc_cls_can_offload_and_chain0(adapter->netdev, type_data))
2770                 return -EOPNOTSUPP;
2771
2772         switch (type) {
2773         case TC_SETUP_CLSFLOWER:
2774                 return igb_setup_tc_cls_flower(adapter, type_data);
2775
2776         default:
2777                 return -EOPNOTSUPP;
2778         }
2779 }
2780
2781 static int igb_offload_txtime(struct igb_adapter *adapter,
2782                               struct tc_etf_qopt_offload *qopt)
2783 {
2784         struct e1000_hw *hw = &adapter->hw;
2785         int err;
2786
2787         /* Launchtime offloading is only supported by i210 controller. */
2788         if (hw->mac.type != e1000_i210)
2789                 return -EOPNOTSUPP;
2790
2791         /* Launchtime offloading is only supported by queues 0 and 1. */
2792         if (qopt->queue < 0 || qopt->queue > 1)
2793                 return -EINVAL;
2794
2795         err = igb_save_txtime_params(adapter, qopt->queue, qopt->enable);
2796         if (err)
2797                 return err;
2798
2799         igb_offload_apply(adapter, qopt->queue);
2800
2801         return 0;
2802 }
2803
2804 static LIST_HEAD(igb_block_cb_list);
2805
2806 static int igb_setup_tc(struct net_device *dev, enum tc_setup_type type,
2807                         void *type_data)
2808 {
2809         struct igb_adapter *adapter = netdev_priv(dev);
2810
2811         switch (type) {
2812         case TC_SETUP_QDISC_CBS:
2813                 return igb_offload_cbs(adapter, type_data);
2814         case TC_SETUP_BLOCK:
2815                 return flow_block_cb_setup_simple(type_data,
2816                                                   &igb_block_cb_list,
2817                                                   igb_setup_tc_block_cb,
2818                                                   adapter, adapter, true);
2819
2820         case TC_SETUP_QDISC_ETF:
2821                 return igb_offload_txtime(adapter, type_data);
2822
2823         default:
2824                 return -EOPNOTSUPP;
2825         }
2826 }
2827
2828 static int igb_xdp_setup(struct net_device *dev, struct netdev_bpf *bpf)
2829 {
2830         int i, frame_size = dev->mtu + IGB_ETH_PKT_HDR_PAD;
2831         struct igb_adapter *adapter = netdev_priv(dev);
2832         struct bpf_prog *prog = bpf->prog, *old_prog;
2833         bool running = netif_running(dev);
2834         bool need_reset;
2835
2836         /* verify igb ring attributes are sufficient for XDP */
2837         for (i = 0; i < adapter->num_rx_queues; i++) {
2838                 struct igb_ring *ring = adapter->rx_ring[i];
2839
2840                 if (frame_size > igb_rx_bufsz(ring)) {
2841                         NL_SET_ERR_MSG_MOD(bpf->extack,
2842                                            "The RX buffer size is too small for the frame size");
2843                         netdev_warn(dev, "XDP RX buffer size %d is too small for the frame size %d\n",
2844                                     igb_rx_bufsz(ring), frame_size);
2845                         return -EINVAL;
2846                 }
2847         }
2848
2849         old_prog = xchg(&adapter->xdp_prog, prog);
2850         need_reset = (!!prog != !!old_prog);
2851
2852         /* device is up and bpf is added/removed, must setup the RX queues */
2853         if (need_reset && running) {
2854                 igb_close(dev);
2855         } else {
2856                 for (i = 0; i < adapter->num_rx_queues; i++)
2857                         (void)xchg(&adapter->rx_ring[i]->xdp_prog,
2858                             adapter->xdp_prog);
2859         }
2860
2861         if (old_prog)
2862                 bpf_prog_put(old_prog);
2863
2864         /* bpf is just replaced, RXQ and MTU are already setup */
2865         if (!need_reset)
2866                 return 0;
2867
2868         if (running)
2869                 igb_open(dev);
2870
2871         return 0;
2872 }
2873
2874 static int igb_xdp(struct net_device *dev, struct netdev_bpf *xdp)
2875 {
2876         switch (xdp->command) {
2877         case XDP_SETUP_PROG:
2878                 return igb_xdp_setup(dev, xdp);
2879         default:
2880                 return -EINVAL;
2881         }
2882 }
2883
2884 static void igb_xdp_ring_update_tail(struct igb_ring *ring)
2885 {
2886         /* Force memory writes to complete before letting h/w know there
2887          * are new descriptors to fetch.
2888          */
2889         wmb();
2890         writel(ring->next_to_use, ring->tail);
2891 }
2892
2893 static struct igb_ring *igb_xdp_tx_queue_mapping(struct igb_adapter *adapter)
2894 {
2895         unsigned int r_idx = smp_processor_id();
2896
2897         if (r_idx >= adapter->num_tx_queues)
2898                 r_idx = r_idx % adapter->num_tx_queues;
2899
2900         return adapter->tx_ring[r_idx];
2901 }
2902
2903 static int igb_xdp_xmit_back(struct igb_adapter *adapter, struct xdp_buff *xdp)
2904 {
2905         struct xdp_frame *xdpf = xdp_convert_buff_to_frame(xdp);
2906         int cpu = smp_processor_id();
2907         struct igb_ring *tx_ring;
2908         struct netdev_queue *nq;
2909         u32 ret;
2910
2911         if (unlikely(!xdpf))
2912                 return IGB_XDP_CONSUMED;
2913
2914         /* During program transitions its possible adapter->xdp_prog is assigned
2915          * but ring has not been configured yet. In this case simply abort xmit.
2916          */
2917         tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2918         if (unlikely(!tx_ring))
2919                 return IGB_XDP_CONSUMED;
2920
2921         nq = txring_txq(tx_ring);
2922         __netif_tx_lock(nq, cpu);
2923         /* Avoid transmit queue timeout since we share it with the slow path */
2924         nq->trans_start = jiffies;
2925         ret = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2926         __netif_tx_unlock(nq);
2927
2928         return ret;
2929 }
2930
2931 static int igb_xdp_xmit(struct net_device *dev, int n,
2932                         struct xdp_frame **frames, u32 flags)
2933 {
2934         struct igb_adapter *adapter = netdev_priv(dev);
2935         int cpu = smp_processor_id();
2936         struct igb_ring *tx_ring;
2937         struct netdev_queue *nq;
2938         int nxmit = 0;
2939         int i;
2940
2941         if (unlikely(test_bit(__IGB_DOWN, &adapter->state)))
2942                 return -ENETDOWN;
2943
2944         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
2945                 return -EINVAL;
2946
2947         /* During program transitions its possible adapter->xdp_prog is assigned
2948          * but ring has not been configured yet. In this case simply abort xmit.
2949          */
2950         tx_ring = adapter->xdp_prog ? igb_xdp_tx_queue_mapping(adapter) : NULL;
2951         if (unlikely(!tx_ring))
2952                 return -ENXIO;
2953
2954         nq = txring_txq(tx_ring);
2955         __netif_tx_lock(nq, cpu);
2956
2957         /* Avoid transmit queue timeout since we share it with the slow path */
2958         nq->trans_start = jiffies;
2959
2960         for (i = 0; i < n; i++) {
2961                 struct xdp_frame *xdpf = frames[i];
2962                 int err;
2963
2964                 err = igb_xmit_xdp_ring(adapter, tx_ring, xdpf);
2965                 if (err != IGB_XDP_TX)
2966                         break;
2967                 nxmit++;
2968         }
2969
2970         __netif_tx_unlock(nq);
2971
2972         if (unlikely(flags & XDP_XMIT_FLUSH))
2973                 igb_xdp_ring_update_tail(tx_ring);
2974
2975         return nxmit;
2976 }
2977
2978 static const struct net_device_ops igb_netdev_ops = {
2979         .ndo_open               = igb_open,
2980         .ndo_stop               = igb_close,
2981         .ndo_start_xmit         = igb_xmit_frame,
2982         .ndo_get_stats64        = igb_get_stats64,
2983         .ndo_set_rx_mode        = igb_set_rx_mode,
2984         .ndo_set_mac_address    = igb_set_mac,
2985         .ndo_change_mtu         = igb_change_mtu,
2986         .ndo_do_ioctl           = igb_ioctl,
2987         .ndo_tx_timeout         = igb_tx_timeout,
2988         .ndo_validate_addr      = eth_validate_addr,
2989         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
2990         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
2991         .ndo_set_vf_mac         = igb_ndo_set_vf_mac,
2992         .ndo_set_vf_vlan        = igb_ndo_set_vf_vlan,
2993         .ndo_set_vf_rate        = igb_ndo_set_vf_bw,
2994         .ndo_set_vf_spoofchk    = igb_ndo_set_vf_spoofchk,
2995         .ndo_set_vf_trust       = igb_ndo_set_vf_trust,
2996         .ndo_get_vf_config      = igb_ndo_get_vf_config,
2997         .ndo_fix_features       = igb_fix_features,
2998         .ndo_set_features       = igb_set_features,
2999         .ndo_fdb_add            = igb_ndo_fdb_add,
3000         .ndo_features_check     = igb_features_check,
3001         .ndo_setup_tc           = igb_setup_tc,
3002         .ndo_bpf                = igb_xdp,
3003         .ndo_xdp_xmit           = igb_xdp_xmit,
3004 };
3005
3006 /**
3007  * igb_set_fw_version - Configure version string for ethtool
3008  * @adapter: adapter struct
3009  **/
3010 void igb_set_fw_version(struct igb_adapter *adapter)
3011 {
3012         struct e1000_hw *hw = &adapter->hw;
3013         struct e1000_fw_version fw;
3014
3015         igb_get_fw_version(hw, &fw);
3016
3017         switch (hw->mac.type) {
3018         case e1000_i210:
3019         case e1000_i211:
3020                 if (!(igb_get_flash_presence_i210(hw))) {
3021                         snprintf(adapter->fw_version,
3022                                  sizeof(adapter->fw_version),
3023                                  "%2d.%2d-%d",
3024                                  fw.invm_major, fw.invm_minor,
3025                                  fw.invm_img_type);
3026                         break;
3027                 }
3028                 fallthrough;
3029         default:
3030                 /* if option is rom valid, display its version too */
3031                 if (fw.or_valid) {
3032                         snprintf(adapter->fw_version,
3033                                  sizeof(adapter->fw_version),
3034                                  "%d.%d, 0x%08x, %d.%d.%d",
3035                                  fw.eep_major, fw.eep_minor, fw.etrack_id,
3036                                  fw.or_major, fw.or_build, fw.or_patch);
3037                 /* no option rom */
3038                 } else if (fw.etrack_id != 0X0000) {
3039                         snprintf(adapter->fw_version,
3040                             sizeof(adapter->fw_version),
3041                             "%d.%d, 0x%08x",
3042                             fw.eep_major, fw.eep_minor, fw.etrack_id);
3043                 } else {
3044                 snprintf(adapter->fw_version,
3045                     sizeof(adapter->fw_version),
3046                     "%d.%d.%d",
3047                     fw.eep_major, fw.eep_minor, fw.eep_build);
3048                 }
3049                 break;
3050         }
3051 }
3052
3053 /**
3054  * igb_init_mas - init Media Autosense feature if enabled in the NVM
3055  *
3056  * @adapter: adapter struct
3057  **/
3058 static void igb_init_mas(struct igb_adapter *adapter)
3059 {
3060         struct e1000_hw *hw = &adapter->hw;
3061         u16 eeprom_data;
3062
3063         hw->nvm.ops.read(hw, NVM_COMPAT, 1, &eeprom_data);
3064         switch (hw->bus.func) {
3065         case E1000_FUNC_0:
3066                 if (eeprom_data & IGB_MAS_ENABLE_0) {
3067                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
3068                         netdev_info(adapter->netdev,
3069                                 "MAS: Enabling Media Autosense for port %d\n",
3070                                 hw->bus.func);
3071                 }
3072                 break;
3073         case E1000_FUNC_1:
3074                 if (eeprom_data & IGB_MAS_ENABLE_1) {
3075                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
3076                         netdev_info(adapter->netdev,
3077                                 "MAS: Enabling Media Autosense for port %d\n",
3078                                 hw->bus.func);
3079                 }
3080                 break;
3081         case E1000_FUNC_2:
3082                 if (eeprom_data & IGB_MAS_ENABLE_2) {
3083                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
3084                         netdev_info(adapter->netdev,
3085                                 "MAS: Enabling Media Autosense for port %d\n",
3086                                 hw->bus.func);
3087                 }
3088                 break;
3089         case E1000_FUNC_3:
3090                 if (eeprom_data & IGB_MAS_ENABLE_3) {
3091                         adapter->flags |= IGB_FLAG_MAS_ENABLE;
3092                         netdev_info(adapter->netdev,
3093                                 "MAS: Enabling Media Autosense for port %d\n",
3094                                 hw->bus.func);
3095                 }
3096                 break;
3097         default:
3098                 /* Shouldn't get here */
3099                 netdev_err(adapter->netdev,
3100                         "MAS: Invalid port configuration, returning\n");
3101                 break;
3102         }
3103 }
3104
3105 /**
3106  *  igb_init_i2c - Init I2C interface
3107  *  @adapter: pointer to adapter structure
3108  **/
3109 static s32 igb_init_i2c(struct igb_adapter *adapter)
3110 {
3111         s32 status = 0;
3112
3113         /* I2C interface supported on i350 devices */
3114         if (adapter->hw.mac.type != e1000_i350)
3115                 return 0;
3116
3117         /* Initialize the i2c bus which is controlled by the registers.
3118          * This bus will use the i2c_algo_bit structure that implements
3119          * the protocol through toggling of the 4 bits in the register.
3120          */
3121         adapter->i2c_adap.owner = THIS_MODULE;
3122         adapter->i2c_algo = igb_i2c_algo;
3123         adapter->i2c_algo.data = adapter;
3124         adapter->i2c_adap.algo_data = &adapter->i2c_algo;
3125         adapter->i2c_adap.dev.parent = &adapter->pdev->dev;
3126         strlcpy(adapter->i2c_adap.name, "igb BB",
3127                 sizeof(adapter->i2c_adap.name));
3128         status = i2c_bit_add_bus(&adapter->i2c_adap);
3129         return status;
3130 }
3131
3132 /**
3133  *  igb_probe - Device Initialization Routine
3134  *  @pdev: PCI device information struct
3135  *  @ent: entry in igb_pci_tbl
3136  *
3137  *  Returns 0 on success, negative on failure
3138  *
3139  *  igb_probe initializes an adapter identified by a pci_dev structure.
3140  *  The OS initialization, configuring of the adapter private structure,
3141  *  and a hardware reset occur.
3142  **/
3143 static int igb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3144 {
3145         struct net_device *netdev;
3146         struct igb_adapter *adapter;
3147         struct e1000_hw *hw;
3148         u16 eeprom_data = 0;
3149         s32 ret_val;
3150         static int global_quad_port_a; /* global quad port a indication */
3151         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
3152         int err, pci_using_dac;
3153         u8 part_str[E1000_PBANUM_LENGTH];
3154
3155         /* Catch broken hardware that put the wrong VF device ID in
3156          * the PCIe SR-IOV capability.
3157          */
3158         if (pdev->is_virtfn) {
3159                 WARN(1, KERN_ERR "%s (%x:%x) should not be a VF!\n",
3160                         pci_name(pdev), pdev->vendor, pdev->device);
3161                 return -EINVAL;
3162         }
3163
3164         err = pci_enable_device_mem(pdev);
3165         if (err)
3166                 return err;
3167
3168         pci_using_dac = 0;
3169         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
3170         if (!err) {
3171                 pci_using_dac = 1;
3172         } else {
3173                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3174                 if (err) {
3175                         dev_err(&pdev->dev,
3176                                 "No usable DMA configuration, aborting\n");
3177                         goto err_dma;
3178                 }
3179         }
3180
3181         err = pci_request_mem_regions(pdev, igb_driver_name);
3182         if (err)
3183                 goto err_pci_reg;
3184
3185         pci_enable_pcie_error_reporting(pdev);
3186
3187         pci_set_master(pdev);
3188         pci_save_state(pdev);
3189
3190         err = -ENOMEM;
3191         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
3192                                    IGB_MAX_TX_QUEUES);
3193         if (!netdev)
3194                 goto err_alloc_etherdev;
3195
3196         SET_NETDEV_DEV(netdev, &pdev->dev);
3197
3198         pci_set_drvdata(pdev, netdev);
3199         adapter = netdev_priv(netdev);
3200         adapter->netdev = netdev;
3201         adapter->pdev = pdev;
3202         hw = &adapter->hw;
3203         hw->back = adapter;
3204         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3205
3206         err = -EIO;
3207         adapter->io_addr = pci_iomap(pdev, 0, 0);
3208         if (!adapter->io_addr)
3209                 goto err_ioremap;
3210         /* hw->hw_addr can be altered, we'll use adapter->io_addr for unmap */
3211         hw->hw_addr = adapter->io_addr;
3212
3213         netdev->netdev_ops = &igb_netdev_ops;
3214         igb_set_ethtool_ops(netdev);
3215         netdev->watchdog_timeo = 5 * HZ;
3216
3217         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
3218
3219         netdev->mem_start = pci_resource_start(pdev, 0);
3220         netdev->mem_end = pci_resource_end(pdev, 0);
3221
3222         /* PCI config space info */
3223         hw->vendor_id = pdev->vendor;
3224         hw->device_id = pdev->device;
3225         hw->revision_id = pdev->revision;
3226         hw->subsystem_vendor_id = pdev->subsystem_vendor;
3227         hw->subsystem_device_id = pdev->subsystem_device;
3228
3229         /* Copy the default MAC, PHY and NVM function pointers */
3230         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
3231         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
3232         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
3233         /* Initialize skew-specific constants */
3234         err = ei->get_invariants(hw);
3235         if (err)
3236                 goto err_sw_init;
3237
3238         /* setup the private structure */
3239         err = igb_sw_init(adapter);
3240         if (err)
3241                 goto err_sw_init;
3242
3243         igb_get_bus_info_pcie(hw);
3244
3245         hw->phy.autoneg_wait_to_complete = false;
3246
3247         /* Copper options */
3248         if (hw->phy.media_type == e1000_media_type_copper) {
3249                 hw->phy.mdix = AUTO_ALL_MODES;
3250                 hw->phy.disable_polarity_correction = false;
3251                 hw->phy.ms_type = e1000_ms_hw_default;
3252         }
3253
3254         if (igb_check_reset_block(hw))
3255                 dev_info(&pdev->dev,
3256                         "PHY reset is blocked due to SOL/IDER session.\n");
3257
3258         /* features is initialized to 0 in allocation, it might have bits
3259          * set by igb_sw_init so we should use an or instead of an
3260          * assignment.
3261          */
3262         netdev->features |= NETIF_F_SG |
3263                             NETIF_F_TSO |
3264                             NETIF_F_TSO6 |
3265                             NETIF_F_RXHASH |
3266                             NETIF_F_RXCSUM |
3267                             NETIF_F_HW_CSUM;
3268
3269         if (hw->mac.type >= e1000_82576)
3270                 netdev->features |= NETIF_F_SCTP_CRC | NETIF_F_GSO_UDP_L4;
3271
3272         if (hw->mac.type >= e1000_i350)
3273                 netdev->features |= NETIF_F_HW_TC;
3274
3275 #define IGB_GSO_PARTIAL_FEATURES (NETIF_F_GSO_GRE | \
3276                                   NETIF_F_GSO_GRE_CSUM | \
3277                                   NETIF_F_GSO_IPXIP4 | \
3278                                   NETIF_F_GSO_IPXIP6 | \
3279                                   NETIF_F_GSO_UDP_TUNNEL | \
3280                                   NETIF_F_GSO_UDP_TUNNEL_CSUM)
3281
3282         netdev->gso_partial_features = IGB_GSO_PARTIAL_FEATURES;
3283         netdev->features |= NETIF_F_GSO_PARTIAL | IGB_GSO_PARTIAL_FEATURES;
3284
3285         /* copy netdev features into list of user selectable features */
3286         netdev->hw_features |= netdev->features |
3287                                NETIF_F_HW_VLAN_CTAG_RX |
3288                                NETIF_F_HW_VLAN_CTAG_TX |
3289                                NETIF_F_RXALL;
3290
3291         if (hw->mac.type >= e1000_i350)
3292                 netdev->hw_features |= NETIF_F_NTUPLE;
3293
3294         if (pci_using_dac)
3295                 netdev->features |= NETIF_F_HIGHDMA;
3296
3297         netdev->vlan_features |= netdev->features | NETIF_F_TSO_MANGLEID;
3298         netdev->mpls_features |= NETIF_F_HW_CSUM;
3299         netdev->hw_enc_features |= netdev->vlan_features;
3300
3301         /* set this bit last since it cannot be part of vlan_features */
3302         netdev->features |= NETIF_F_HW_VLAN_CTAG_FILTER |
3303                             NETIF_F_HW_VLAN_CTAG_RX |
3304                             NETIF_F_HW_VLAN_CTAG_TX;
3305
3306         netdev->priv_flags |= IFF_SUPP_NOFCS;
3307
3308         netdev->priv_flags |= IFF_UNICAST_FLT;
3309
3310         /* MTU range: 68 - 9216 */
3311         netdev->min_mtu = ETH_MIN_MTU;
3312         netdev->max_mtu = MAX_STD_JUMBO_FRAME_SIZE;
3313
3314         adapter->en_mng_pt = igb_enable_mng_pass_thru(hw);
3315
3316         /* before reading the NVM, reset the controller to put the device in a
3317          * known good starting state
3318          */
3319         hw->mac.ops.reset_hw(hw);
3320
3321         /* make sure the NVM is good , i211/i210 parts can have special NVM
3322          * that doesn't contain a checksum
3323          */
3324         switch (hw->mac.type) {
3325         case e1000_i210:
3326         case e1000_i211:
3327                 if (igb_get_flash_presence_i210(hw)) {
3328                         if (hw->nvm.ops.validate(hw) < 0) {
3329                                 dev_err(&pdev->dev,
3330                                         "The NVM Checksum Is Not Valid\n");
3331                                 err = -EIO;
3332                                 goto err_eeprom;
3333                         }
3334                 }
3335                 break;
3336         default:
3337                 if (hw->nvm.ops.validate(hw) < 0) {
3338                         dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
3339                         err = -EIO;
3340                         goto err_eeprom;
3341                 }
3342                 break;
3343         }
3344
3345         if (eth_platform_get_mac_address(&pdev->dev, hw->mac.addr)) {
3346                 /* copy the MAC address out of the NVM */
3347                 if (hw->mac.ops.read_mac_addr(hw))
3348                         dev_err(&pdev->dev, "NVM Read Error\n");
3349         }
3350
3351         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
3352
3353         if (!is_valid_ether_addr(netdev->dev_addr)) {
3354                 dev_err(&pdev->dev, "Invalid MAC Address\n");
3355                 err = -EIO;
3356                 goto err_eeprom;
3357         }
3358
3359         igb_set_default_mac_filter(adapter);
3360
3361         /* get firmware version for ethtool -i */
3362         igb_set_fw_version(adapter);
3363
3364         /* configure RXPBSIZE and TXPBSIZE */
3365         if (hw->mac.type == e1000_i210) {
3366                 wr32(E1000_RXPBS, I210_RXPBSIZE_DEFAULT);
3367                 wr32(E1000_TXPBS, I210_TXPBSIZE_DEFAULT);
3368         }
3369
3370         timer_setup(&adapter->watchdog_timer, igb_watchdog, 0);
3371         timer_setup(&adapter->phy_info_timer, igb_update_phy_info, 0);
3372
3373         INIT_WORK(&adapter->reset_task, igb_reset_task);
3374         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
3375
3376         /* Initialize link properties that are user-changeable */
3377         adapter->fc_autoneg = true;
3378         hw->mac.autoneg = true;
3379         hw->phy.autoneg_advertised = 0x2f;
3380
3381         hw->fc.requested_mode = e1000_fc_default;
3382         hw->fc.current_mode = e1000_fc_default;
3383
3384         igb_validate_mdi_setting(hw);
3385
3386         /* By default, support wake on port A */
3387         if (hw->bus.func == 0)
3388                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3389
3390         /* Check the NVM for wake support on non-port A ports */
3391         if (hw->mac.type >= e1000_82580)
3392                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A +
3393                                  NVM_82580_LAN_FUNC_OFFSET(hw->bus.func), 1,
3394                                  &eeprom_data);
3395         else if (hw->bus.func == 1)
3396                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
3397
3398         if (eeprom_data & IGB_EEPROM_APME)
3399                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3400
3401         /* now that we have the eeprom settings, apply the special cases where
3402          * the eeprom may be wrong or the board simply won't support wake on
3403          * lan on a particular port
3404          */
3405         switch (pdev->device) {
3406         case E1000_DEV_ID_82575GB_QUAD_COPPER:
3407                 adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3408                 break;
3409         case E1000_DEV_ID_82575EB_FIBER_SERDES:
3410         case E1000_DEV_ID_82576_FIBER:
3411         case E1000_DEV_ID_82576_SERDES:
3412                 /* Wake events only supported on port A for dual fiber
3413                  * regardless of eeprom setting
3414                  */
3415                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
3416                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3417                 break;
3418         case E1000_DEV_ID_82576_QUAD_COPPER:
3419         case E1000_DEV_ID_82576_QUAD_COPPER_ET2:
3420                 /* if quad port adapter, disable WoL on all but port A */
3421                 if (global_quad_port_a != 0)
3422                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3423                 else
3424                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
3425                 /* Reset for multiple quad port adapters */
3426                 if (++global_quad_port_a == 4)
3427                         global_quad_port_a = 0;
3428                 break;
3429         default:
3430                 /* If the device can't wake, don't set software support */
3431                 if (!device_can_wakeup(&adapter->pdev->dev))
3432                         adapter->flags &= ~IGB_FLAG_WOL_SUPPORTED;
3433         }
3434
3435         /* initialize the wol settings based on the eeprom settings */
3436         if (adapter->flags & IGB_FLAG_WOL_SUPPORTED)
3437                 adapter->wol |= E1000_WUFC_MAG;
3438
3439         /* Some vendors want WoL disabled by default, but still supported */
3440         if ((hw->mac.type == e1000_i350) &&
3441             (pdev->subsystem_vendor == PCI_VENDOR_ID_HP)) {
3442                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3443                 adapter->wol = 0;
3444         }
3445
3446         /* Some vendors want the ability to Use the EEPROM setting as
3447          * enable/disable only, and not for capability
3448          */
3449         if (((hw->mac.type == e1000_i350) ||
3450              (hw->mac.type == e1000_i354)) &&
3451             (pdev->subsystem_vendor == PCI_VENDOR_ID_DELL)) {
3452                 adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3453                 adapter->wol = 0;
3454         }
3455         if (hw->mac.type == e1000_i350) {
3456                 if (((pdev->subsystem_device == 0x5001) ||
3457                      (pdev->subsystem_device == 0x5002)) &&
3458                                 (hw->bus.func == 0)) {
3459                         adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3460                         adapter->wol = 0;
3461                 }
3462                 if (pdev->subsystem_device == 0x1F52)
3463                         adapter->flags |= IGB_FLAG_WOL_SUPPORTED;
3464         }
3465
3466         device_set_wakeup_enable(&adapter->pdev->dev,
3467                                  adapter->flags & IGB_FLAG_WOL_SUPPORTED);
3468
3469         /* reset the hardware with the new settings */
3470         igb_reset(adapter);
3471
3472         /* Init the I2C interface */
3473         err = igb_init_i2c(adapter);
3474         if (err) {
3475                 dev_err(&pdev->dev, "failed to init i2c interface\n");
3476                 goto err_eeprom;
3477         }
3478
3479         /* let the f/w know that the h/w is now under the control of the
3480          * driver.
3481          */
3482         igb_get_hw_control(adapter);
3483
3484         strcpy(netdev->name, "eth%d");
3485         err = register_netdev(netdev);
3486         if (err)
3487                 goto err_register;
3488
3489         /* carrier off reporting is important to ethtool even BEFORE open */
3490         netif_carrier_off(netdev);
3491
3492 #ifdef CONFIG_IGB_DCA
3493         if (dca_add_requester(&pdev->dev) == 0) {
3494                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3495                 dev_info(&pdev->dev, "DCA enabled\n");
3496                 igb_setup_dca(adapter);
3497         }
3498
3499 #endif
3500 #ifdef CONFIG_IGB_HWMON
3501         /* Initialize the thermal sensor on i350 devices. */
3502         if (hw->mac.type == e1000_i350 && hw->bus.func == 0) {
3503                 u16 ets_word;
3504
3505                 /* Read the NVM to determine if this i350 device supports an
3506                  * external thermal sensor.
3507                  */
3508                 hw->nvm.ops.read(hw, NVM_ETS_CFG, 1, &ets_word);
3509                 if (ets_word != 0x0000 && ets_word != 0xFFFF)
3510                         adapter->ets = true;
3511                 else
3512                         adapter->ets = false;
3513                 if (igb_sysfs_init(adapter))
3514                         dev_err(&pdev->dev,
3515                                 "failed to allocate sysfs resources\n");
3516         } else {
3517                 adapter->ets = false;
3518         }
3519 #endif
3520         /* Check if Media Autosense is enabled */
3521         adapter->ei = *ei;
3522         if (hw->dev_spec._82575.mas_capable)
3523                 igb_init_mas(adapter);
3524
3525         /* do hw tstamp init after resetting */
3526         igb_ptp_init(adapter);
3527
3528         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
3529         /* print bus type/speed/width info, not applicable to i354 */
3530         if (hw->mac.type != e1000_i354) {
3531                 dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
3532                          netdev->name,
3533                          ((hw->bus.speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
3534                           (hw->bus.speed == e1000_bus_speed_5000) ? "5.0Gb/s" :
3535                            "unknown"),
3536                          ((hw->bus.width == e1000_bus_width_pcie_x4) ?
3537                           "Width x4" :
3538                           (hw->bus.width == e1000_bus_width_pcie_x2) ?
3539                           "Width x2" :
3540                           (hw->bus.width == e1000_bus_width_pcie_x1) ?
3541                           "Width x1" : "unknown"), netdev->dev_addr);
3542         }
3543
3544         if ((hw->mac.type == e1000_82576 &&
3545              rd32(E1000_EECD) & E1000_EECD_PRES) ||
3546             (hw->mac.type >= e1000_i210 ||
3547              igb_get_flash_presence_i210(hw))) {
3548                 ret_val = igb_read_part_string(hw, part_str,
3549                                                E1000_PBANUM_LENGTH);
3550         } else {
3551                 ret_val = -E1000_ERR_INVM_VALUE_NOT_FOUND;
3552         }
3553
3554         if (ret_val)
3555                 strcpy(part_str, "Unknown");
3556         dev_info(&pdev->dev, "%s: PBA No: %s\n", netdev->name, part_str);
3557         dev_info(&pdev->dev,
3558                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
3559                 (adapter->flags & IGB_FLAG_HAS_MSIX) ? "MSI-X" :
3560                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
3561                 adapter->num_rx_queues, adapter->num_tx_queues);
3562         if (hw->phy.media_type == e1000_media_type_copper) {
3563                 switch (hw->mac.type) {
3564                 case e1000_i350:
3565                 case e1000_i210:
3566                 case e1000_i211:
3567                         /* Enable EEE for internal copper PHY devices */
3568                         err = igb_set_eee_i350(hw, true, true);
3569                         if ((!err) &&
3570                             (!hw->dev_spec._82575.eee_disable)) {
3571                                 adapter->eee_advert =
3572                                         MDIO_EEE_100TX | MDIO_EEE_1000T;
3573                                 adapter->flags |= IGB_FLAG_EEE;
3574                         }
3575                         break;
3576                 case e1000_i354:
3577                         if ((rd32(E1000_CTRL_EXT) &
3578                             E1000_CTRL_EXT_LINK_MODE_SGMII)) {
3579                                 err = igb_set_eee_i354(hw, true, true);
3580                                 if ((!err) &&
3581                                         (!hw->dev_spec._82575.eee_disable)) {
3582                                         adapter->eee_advert =
3583                                            MDIO_EEE_100TX | MDIO_EEE_1000T;
3584                                         adapter->flags |= IGB_FLAG_EEE;
3585                                 }
3586                         }
3587                         break;
3588                 default:
3589                         break;
3590                 }
3591         }
3592
3593         dev_pm_set_driver_flags(&pdev->dev, DPM_FLAG_NO_DIRECT_COMPLETE);
3594
3595         pm_runtime_put_noidle(&pdev->dev);
3596         return 0;
3597
3598 err_register:
3599         igb_release_hw_control(adapter);
3600         memset(&adapter->i2c_adap, 0, sizeof(adapter->i2c_adap));
3601 err_eeprom:
3602         if (!igb_check_reset_block(hw))
3603                 igb_reset_phy(hw);
3604
3605         if (hw->flash_address)
3606                 iounmap(hw->flash_address);
3607 err_sw_init:
3608         kfree(adapter->mac_table);
3609         kfree(adapter->shadow_vfta);
3610         igb_clear_interrupt_scheme(adapter);
3611 #ifdef CONFIG_PCI_IOV
3612         igb_disable_sriov(pdev);
3613 #endif
3614         pci_iounmap(pdev, adapter->io_addr);
3615 err_ioremap:
3616         free_netdev(netdev);
3617 err_alloc_etherdev:
3618         pci_release_mem_regions(pdev);
3619 err_pci_reg:
3620 err_dma:
3621         pci_disable_device(pdev);
3622         return err;
3623 }
3624
3625 #ifdef CONFIG_PCI_IOV
3626 static int igb_disable_sriov(struct pci_dev *pdev)
3627 {
3628         struct net_device *netdev = pci_get_drvdata(pdev);
3629         struct igb_adapter *adapter = netdev_priv(netdev);
3630         struct e1000_hw *hw = &adapter->hw;
3631
3632         /* reclaim resources allocated to VFs */
3633         if (adapter->vf_data) {
3634                 /* disable iov and allow time for transactions to clear */
3635                 if (pci_vfs_assigned(pdev)) {
3636                         dev_warn(&pdev->dev,
3637                                  "Cannot deallocate SR-IOV virtual functions while they are assigned - VFs will not be deallocated\n");
3638                         return -EPERM;
3639                 } else {
3640                         pci_disable_sriov(pdev);
3641                         msleep(500);
3642                 }
3643
3644                 kfree(adapter->vf_mac_list);
3645                 adapter->vf_mac_list = NULL;
3646                 kfree(adapter->vf_data);
3647                 adapter->vf_data = NULL;
3648                 adapter->vfs_allocated_count = 0;
3649                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
3650                 wrfl();
3651                 msleep(100);
3652                 dev_info(&pdev->dev, "IOV Disabled\n");
3653
3654                 /* Re-enable DMA Coalescing flag since IOV is turned off */
3655                 adapter->flags |= IGB_FLAG_DMAC;
3656         }
3657
3658         return 0;
3659 }
3660
3661 static int igb_enable_sriov(struct pci_dev *pdev, int num_vfs)
3662 {
3663         struct net_device *netdev = pci_get_drvdata(pdev);
3664         struct igb_adapter *adapter = netdev_priv(netdev);
3665         int old_vfs = pci_num_vf(pdev);
3666         struct vf_mac_filter *mac_list;
3667         int err = 0;
3668         int num_vf_mac_filters, i;
3669
3670         if (!(adapter->flags & IGB_FLAG_HAS_MSIX) || num_vfs > 7) {
3671                 err = -EPERM;
3672                 goto out;
3673         }
3674         if (!num_vfs)
3675                 goto out;
3676
3677         if (old_vfs) {
3678                 dev_info(&pdev->dev, "%d pre-allocated VFs found - override max_vfs setting of %d\n",
3679                          old_vfs, max_vfs);
3680                 adapter->vfs_allocated_count = old_vfs;
3681         } else
3682                 adapter->vfs_allocated_count = num_vfs;
3683
3684         adapter->vf_data = kcalloc(adapter->vfs_allocated_count,
3685                                 sizeof(struct vf_data_storage), GFP_KERNEL);
3686
3687         /* if allocation failed then we do not support SR-IOV */
3688         if (!adapter->vf_data) {
3689                 adapter->vfs_allocated_count = 0;
3690                 err = -ENOMEM;
3691                 goto out;
3692         }
3693
3694         /* Due to the limited number of RAR entries calculate potential
3695          * number of MAC filters available for the VFs. Reserve entries
3696          * for PF default MAC, PF MAC filters and at least one RAR entry
3697          * for each VF for VF MAC.
3698          */
3699         num_vf_mac_filters = adapter->hw.mac.rar_entry_count -
3700                              (1 + IGB_PF_MAC_FILTERS_RESERVED +
3701                               adapter->vfs_allocated_count);
3702
3703         adapter->vf_mac_list = kcalloc(num_vf_mac_filters,
3704                                        sizeof(struct vf_mac_filter),
3705                                        GFP_KERNEL);
3706
3707         mac_list = adapter->vf_mac_list;
3708         INIT_LIST_HEAD(&adapter->vf_macs.l);
3709
3710         if (adapter->vf_mac_list) {
3711                 /* Initialize list of VF MAC filters */
3712                 for (i = 0; i < num_vf_mac_filters; i++) {
3713                         mac_list->vf = -1;
3714                         mac_list->free = true;
3715                         list_add(&mac_list->l, &adapter->vf_macs.l);
3716                         mac_list++;
3717                 }
3718         } else {
3719                 /* If we could not allocate memory for the VF MAC filters
3720                  * we can continue without this feature but warn user.
3721                  */
3722                 dev_err(&pdev->dev,
3723                         "Unable to allocate memory for VF MAC filter list\n");
3724         }
3725
3726         /* only call pci_enable_sriov() if no VFs are allocated already */
3727         if (!old_vfs) {
3728                 err = pci_enable_sriov(pdev, adapter->vfs_allocated_count);
3729                 if (err)
3730                         goto err_out;
3731         }
3732         dev_info(&pdev->dev, "%d VFs allocated\n",
3733                  adapter->vfs_allocated_count);
3734         for (i = 0; i < adapter->vfs_allocated_count; i++)
3735                 igb_vf_configure(adapter, i);
3736
3737         /* DMA Coalescing is not supported in IOV mode. */
3738         adapter->flags &= ~IGB_FLAG_DMAC;
3739         goto out;
3740
3741 err_out:
3742         kfree(adapter->vf_mac_list);
3743         adapter->vf_mac_list = NULL;
3744         kfree(adapter->vf_data);
3745         adapter->vf_data = NULL;
3746         adapter->vfs_allocated_count = 0;
3747 out:
3748         return err;
3749 }
3750
3751 #endif
3752 /**
3753  *  igb_remove_i2c - Cleanup  I2C interface
3754  *  @adapter: pointer to adapter structure
3755  **/
3756 static void igb_remove_i2c(struct igb_adapter *adapter)
3757 {
3758         /* free the adapter bus structure */
3759         i2c_del_adapter(&adapter->i2c_adap);
3760 }
3761
3762 /**
3763  *  igb_remove - Device Removal Routine
3764  *  @pdev: PCI device information struct
3765  *
3766  *  igb_remove is called by the PCI subsystem to alert the driver
3767  *  that it should release a PCI device.  The could be caused by a
3768  *  Hot-Plug event, or because the driver is going to be removed from
3769  *  memory.
3770  **/
3771 static void igb_remove(struct pci_dev *pdev)
3772 {
3773         struct net_device *netdev = pci_get_drvdata(pdev);
3774         struct igb_adapter *adapter = netdev_priv(netdev);
3775         struct e1000_hw *hw = &adapter->hw;
3776
3777         pm_runtime_get_noresume(&pdev->dev);
3778 #ifdef CONFIG_IGB_HWMON
3779         igb_sysfs_exit(adapter);
3780 #endif
3781         igb_remove_i2c(adapter);
3782         igb_ptp_stop(adapter);
3783         /* The watchdog timer may be rescheduled, so explicitly
3784          * disable watchdog from being rescheduled.
3785          */
3786         set_bit(__IGB_DOWN, &adapter->state);
3787         del_timer_sync(&adapter->watchdog_timer);
3788         del_timer_sync(&adapter->phy_info_timer);
3789
3790         cancel_work_sync(&adapter->reset_task);
3791         cancel_work_sync(&adapter->watchdog_task);
3792
3793 #ifdef CONFIG_IGB_DCA
3794         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3795                 dev_info(&pdev->dev, "DCA disabled\n");
3796                 dca_remove_requester(&pdev->dev);
3797                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3798                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3799         }
3800 #endif
3801
3802         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3803          * would have already happened in close and is redundant.
3804          */
3805         igb_release_hw_control(adapter);
3806
3807 #ifdef CONFIG_PCI_IOV
3808         igb_disable_sriov(pdev);
3809 #endif
3810
3811         unregister_netdev(netdev);
3812
3813         igb_clear_interrupt_scheme(adapter);
3814
3815         pci_iounmap(pdev, adapter->io_addr);
3816         if (hw->flash_address)
3817                 iounmap(hw->flash_address);
3818         pci_release_mem_regions(pdev);
3819
3820         kfree(adapter->mac_table);
3821         kfree(adapter->shadow_vfta);
3822         free_netdev(netdev);
3823
3824         pci_disable_pcie_error_reporting(pdev);
3825
3826         pci_disable_device(pdev);
3827 }
3828
3829 /**
3830  *  igb_probe_vfs - Initialize vf data storage and add VFs to pci config space
3831  *  @adapter: board private structure to initialize
3832  *
3833  *  This function initializes the vf specific data storage and then attempts to
3834  *  allocate the VFs.  The reason for ordering it this way is because it is much
3835  *  mor expensive time wise to disable SR-IOV than it is to allocate and free
3836  *  the memory for the VFs.
3837  **/
3838 static void igb_probe_vfs(struct igb_adapter *adapter)
3839 {
3840 #ifdef CONFIG_PCI_IOV
3841         struct pci_dev *pdev = adapter->pdev;
3842         struct e1000_hw *hw = &adapter->hw;
3843
3844         /* Virtualization features not supported on i210 family. */
3845         if ((hw->mac.type == e1000_i210) || (hw->mac.type == e1000_i211))
3846                 return;
3847
3848         /* Of the below we really only want the effect of getting
3849          * IGB_FLAG_HAS_MSIX set (if available), without which
3850          * igb_enable_sriov() has no effect.
3851          */
3852         igb_set_interrupt_capability(adapter, true);
3853         igb_reset_interrupt_capability(adapter);
3854
3855         pci_sriov_set_totalvfs(pdev, 7);
3856         igb_enable_sriov(pdev, max_vfs);
3857
3858 #endif /* CONFIG_PCI_IOV */
3859 }
3860
3861 unsigned int igb_get_max_rss_queues(struct igb_adapter *adapter)
3862 {
3863         struct e1000_hw *hw = &adapter->hw;
3864         unsigned int max_rss_queues;
3865
3866         /* Determine the maximum number of RSS queues supported. */
3867         switch (hw->mac.type) {
3868         case e1000_i211:
3869                 max_rss_queues = IGB_MAX_RX_QUEUES_I211;
3870                 break;
3871         case e1000_82575:
3872         case e1000_i210:
3873                 max_rss_queues = IGB_MAX_RX_QUEUES_82575;
3874                 break;
3875         case e1000_i350:
3876                 /* I350 cannot do RSS and SR-IOV at the same time */
3877                 if (!!adapter->vfs_allocated_count) {
3878                         max_rss_queues = 1;
3879                         break;
3880                 }
3881                 fallthrough;
3882         case e1000_82576:
3883                 if (!!adapter->vfs_allocated_count) {
3884                         max_rss_queues = 2;
3885                         break;
3886                 }
3887                 fallthrough;
3888         case e1000_82580:
3889         case e1000_i354:
3890         default:
3891                 max_rss_queues = IGB_MAX_RX_QUEUES;
3892                 break;
3893         }
3894
3895         return max_rss_queues;
3896 }
3897
3898 static void igb_init_queue_configuration(struct igb_adapter *adapter)
3899 {
3900         u32 max_rss_queues;
3901
3902         max_rss_queues = igb_get_max_rss_queues(adapter);
3903         adapter->rss_queues = min_t(u32, max_rss_queues, num_online_cpus());
3904
3905         igb_set_flag_queue_pairs(adapter, max_rss_queues);
3906 }
3907
3908 void igb_set_flag_queue_pairs(struct igb_adapter *adapter,
3909                               const u32 max_rss_queues)
3910 {
3911         struct e1000_hw *hw = &adapter->hw;
3912
3913         /* Determine if we need to pair queues. */
3914         switch (hw->mac.type) {
3915         case e1000_82575:
3916         case e1000_i211:
3917                 /* Device supports enough interrupts without queue pairing. */
3918                 break;
3919         case e1000_82576:
3920         case e1000_82580:
3921         case e1000_i350:
3922         case e1000_i354:
3923         case e1000_i210:
3924         default:
3925                 /* If rss_queues > half of max_rss_queues, pair the queues in
3926                  * order to conserve interrupts due to limited supply.
3927                  */
3928                 if (adapter->rss_queues > (max_rss_queues / 2))
3929                         adapter->flags |= IGB_FLAG_QUEUE_PAIRS;
3930                 else
3931                         adapter->flags &= ~IGB_FLAG_QUEUE_PAIRS;
3932                 break;
3933         }
3934 }
3935
3936 /**
3937  *  igb_sw_init - Initialize general software structures (struct igb_adapter)
3938  *  @adapter: board private structure to initialize
3939  *
3940  *  igb_sw_init initializes the Adapter private data structure.
3941  *  Fields are initialized based on PCI device information and
3942  *  OS network device settings (MTU size).
3943  **/
3944 static int igb_sw_init(struct igb_adapter *adapter)
3945 {
3946         struct e1000_hw *hw = &adapter->hw;
3947         struct net_device *netdev = adapter->netdev;
3948         struct pci_dev *pdev = adapter->pdev;
3949
3950         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
3951
3952         /* set default ring sizes */
3953         adapter->tx_ring_count = IGB_DEFAULT_TXD;
3954         adapter->rx_ring_count = IGB_DEFAULT_RXD;
3955
3956         /* set default ITR values */
3957         adapter->rx_itr_setting = IGB_DEFAULT_ITR;
3958         adapter->tx_itr_setting = IGB_DEFAULT_ITR;
3959
3960         /* set default work limits */
3961         adapter->tx_work_limit = IGB_DEFAULT_TX_WORK;
3962
3963         adapter->max_frame_size = netdev->mtu + IGB_ETH_PKT_HDR_PAD;
3964         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
3965
3966         spin_lock_init(&adapter->nfc_lock);
3967         spin_lock_init(&adapter->stats64_lock);
3968 #ifdef CONFIG_PCI_IOV
3969         switch (hw->mac.type) {
3970         case e1000_82576:
3971         case e1000_i350:
3972                 if (max_vfs > 7) {
3973                         dev_warn(&pdev->dev,
3974                                  "Maximum of 7 VFs per PF, using max\n");
3975                         max_vfs = adapter->vfs_allocated_count = 7;
3976                 } else
3977                         adapter->vfs_allocated_count = max_vfs;
3978                 if (adapter->vfs_allocated_count)
3979                         dev_warn(&pdev->dev,
3980                                  "Enabling SR-IOV VFs using the module parameter is deprecated - please use the pci sysfs interface.\n");
3981                 break;
3982         default:
3983                 break;
3984         }
3985 #endif /* CONFIG_PCI_IOV */
3986
3987         /* Assume MSI-X interrupts, will be checked during IRQ allocation */
3988         adapter->flags |= IGB_FLAG_HAS_MSIX;
3989
3990         adapter->mac_table = kcalloc(hw->mac.rar_entry_count,
3991                                      sizeof(struct igb_mac_addr),
3992                                      GFP_KERNEL);
3993         if (!adapter->mac_table)
3994                 return -ENOMEM;
3995
3996         igb_probe_vfs(adapter);
3997
3998         igb_init_queue_configuration(adapter);
3999
4000         /* Setup and initialize a copy of the hw vlan table array */
4001         adapter->shadow_vfta = kcalloc(E1000_VLAN_FILTER_TBL_SIZE, sizeof(u32),
4002                                        GFP_KERNEL);
4003         if (!adapter->shadow_vfta)
4004                 return -ENOMEM;
4005
4006         /* This call may decrease the number of queues */
4007         if (igb_init_interrupt_scheme(adapter, true)) {
4008                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4009                 return -ENOMEM;
4010         }
4011
4012         /* Explicitly disable IRQ since the NIC can be in any state. */
4013         igb_irq_disable(adapter);
4014
4015         if (hw->mac.type >= e1000_i350)
4016                 adapter->flags &= ~IGB_FLAG_DMAC;
4017
4018         set_bit(__IGB_DOWN, &adapter->state);
4019         return 0;
4020 }
4021
4022 /**
4023  *  __igb_open - Called when a network interface is made active
4024  *  @netdev: network interface device structure
4025  *  @resuming: indicates whether we are in a resume call
4026  *
4027  *  Returns 0 on success, negative value on failure
4028  *
4029  *  The open entry point is called when a network interface is made
4030  *  active by the system (IFF_UP).  At this point all resources needed
4031  *  for transmit and receive operations are allocated, the interrupt
4032  *  handler is registered with the OS, the watchdog timer is started,
4033  *  and the stack is notified that the interface is ready.
4034  **/
4035 static int __igb_open(struct net_device *netdev, bool resuming)
4036 {
4037         struct igb_adapter *adapter = netdev_priv(netdev);
4038         struct e1000_hw *hw = &adapter->hw;
4039         struct pci_dev *pdev = adapter->pdev;
4040         int err;
4041         int i;
4042
4043         /* disallow open during test */
4044         if (test_bit(__IGB_TESTING, &adapter->state)) {
4045                 WARN_ON(resuming);
4046                 return -EBUSY;
4047         }
4048
4049         if (!resuming)
4050                 pm_runtime_get_sync(&pdev->dev);
4051
4052         netif_carrier_off(netdev);
4053
4054         /* allocate transmit descriptors */
4055         err = igb_setup_all_tx_resources(adapter);
4056         if (err)
4057                 goto err_setup_tx;
4058
4059         /* allocate receive descriptors */
4060         err = igb_setup_all_rx_resources(adapter);
4061         if (err)
4062                 goto err_setup_rx;
4063
4064         igb_power_up_link(adapter);
4065
4066         /* before we allocate an interrupt, we must be ready to handle it.
4067          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
4068          * as soon as we call pci_request_irq, so we have to setup our
4069          * clean_rx handler before we do so.
4070          */
4071         igb_configure(adapter);
4072
4073         err = igb_request_irq(adapter);
4074         if (err)
4075                 goto err_req_irq;
4076
4077         /* Notify the stack of the actual queue counts. */
4078         err = netif_set_real_num_tx_queues(adapter->netdev,
4079                                            adapter->num_tx_queues);
4080         if (err)
4081                 goto err_set_queues;
4082
4083         err = netif_set_real_num_rx_queues(adapter->netdev,
4084                                            adapter->num_rx_queues);
4085         if (err)
4086                 goto err_set_queues;
4087
4088         /* From here on the code is the same as igb_up() */
4089         clear_bit(__IGB_DOWN, &adapter->state);
4090
4091         for (i = 0; i < adapter->num_q_vectors; i++)
4092                 napi_enable(&(adapter->q_vector[i]->napi));
4093
4094         /* Clear any pending interrupts. */
4095         rd32(E1000_TSICR);
4096         rd32(E1000_ICR);
4097
4098         igb_irq_enable(adapter);
4099
4100         /* notify VFs that reset has been completed */
4101         if (adapter->vfs_allocated_count) {
4102                 u32 reg_data = rd32(E1000_CTRL_EXT);
4103
4104                 reg_data |= E1000_CTRL_EXT_PFRSTD;
4105                 wr32(E1000_CTRL_EXT, reg_data);
4106         }
4107
4108         netif_tx_start_all_queues(netdev);
4109
4110         if (!resuming)
4111                 pm_runtime_put(&pdev->dev);
4112
4113         /* start the watchdog. */
4114         hw->mac.get_link_status = 1;
4115         schedule_work(&adapter->watchdog_task);
4116
4117         return 0;
4118
4119 err_set_queues:
4120         igb_free_irq(adapter);
4121 err_req_irq:
4122         igb_release_hw_control(adapter);
4123         igb_power_down_link(adapter);
4124         igb_free_all_rx_resources(adapter);
4125 err_setup_rx:
4126         igb_free_all_tx_resources(adapter);
4127 err_setup_tx:
4128         igb_reset(adapter);
4129         if (!resuming)
4130                 pm_runtime_put(&pdev->dev);
4131
4132         return err;
4133 }
4134
4135 int igb_open(struct net_device *netdev)
4136 {
4137         return __igb_open(netdev, false);
4138 }
4139
4140 /**
4141  *  __igb_close - Disables a network interface
4142  *  @netdev: network interface device structure
4143  *  @suspending: indicates we are in a suspend call
4144  *
4145  *  Returns 0, this is not allowed to fail
4146  *
4147  *  The close entry point is called when an interface is de-activated
4148  *  by the OS.  The hardware is still under the driver's control, but
4149  *  needs to be disabled.  A global MAC reset is issued to stop the
4150  *  hardware, and all transmit and receive resources are freed.
4151  **/
4152 static int __igb_close(struct net_device *netdev, bool suspending)
4153 {
4154         struct igb_adapter *adapter = netdev_priv(netdev);
4155         struct pci_dev *pdev = adapter->pdev;
4156
4157         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
4158
4159         if (!suspending)
4160                 pm_runtime_get_sync(&pdev->dev);
4161
4162         igb_down(adapter);
4163         igb_free_irq(adapter);
4164
4165         igb_free_all_tx_resources(adapter);
4166         igb_free_all_rx_resources(adapter);
4167
4168         if (!suspending)
4169                 pm_runtime_put_sync(&pdev->dev);
4170         return 0;
4171 }
4172
4173 int igb_close(struct net_device *netdev)
4174 {
4175         if (netif_device_present(netdev) || netdev->dismantle)
4176                 return __igb_close(netdev, false);
4177         return 0;
4178 }
4179
4180 /**
4181  *  igb_setup_tx_resources - allocate Tx resources (Descriptors)
4182  *  @tx_ring: tx descriptor ring (for a specific queue) to setup
4183  *
4184  *  Return 0 on success, negative on failure
4185  **/
4186 int igb_setup_tx_resources(struct igb_ring *tx_ring)
4187 {
4188         struct device *dev = tx_ring->dev;
4189         int size;
4190
4191         size = sizeof(struct igb_tx_buffer) * tx_ring->count;
4192
4193         tx_ring->tx_buffer_info = vmalloc(size);
4194         if (!tx_ring->tx_buffer_info)
4195                 goto err;
4196
4197         /* round up to nearest 4K */
4198         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
4199         tx_ring->size = ALIGN(tx_ring->size, 4096);
4200
4201         tx_ring->desc = dma_alloc_coherent(dev, tx_ring->size,
4202                                            &tx_ring->dma, GFP_KERNEL);
4203         if (!tx_ring->desc)
4204                 goto err;
4205
4206         tx_ring->next_to_use = 0;
4207         tx_ring->next_to_clean = 0;
4208
4209         return 0;
4210
4211 err:
4212         vfree(tx_ring->tx_buffer_info);
4213         tx_ring->tx_buffer_info = NULL;
4214         dev_err(dev, "Unable to allocate memory for the Tx descriptor ring\n");
4215         return -ENOMEM;
4216 }
4217
4218 /**
4219  *  igb_setup_all_tx_resources - wrapper to allocate Tx resources
4220  *                               (Descriptors) for all queues
4221  *  @adapter: board private structure
4222  *
4223  *  Return 0 on success, negative on failure
4224  **/
4225 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
4226 {
4227         struct pci_dev *pdev = adapter->pdev;
4228         int i, err = 0;
4229
4230         for (i = 0; i < adapter->num_tx_queues; i++) {
4231                 err = igb_setup_tx_resources(adapter->tx_ring[i]);
4232                 if (err) {
4233                         dev_err(&pdev->dev,
4234                                 "Allocation for Tx Queue %u failed\n", i);
4235                         for (i--; i >= 0; i--)
4236                                 igb_free_tx_resources(adapter->tx_ring[i]);
4237                         break;
4238                 }
4239         }
4240
4241         return err;
4242 }
4243
4244 /**
4245  *  igb_setup_tctl - configure the transmit control registers
4246  *  @adapter: Board private structure
4247  **/
4248 void igb_setup_tctl(struct igb_adapter *adapter)
4249 {
4250         struct e1000_hw *hw = &adapter->hw;
4251         u32 tctl;
4252
4253         /* disable queue 0 which is enabled by default on 82575 and 82576 */
4254         wr32(E1000_TXDCTL(0), 0);
4255
4256         /* Program the Transmit Control Register */
4257         tctl = rd32(E1000_TCTL);
4258         tctl &= ~E1000_TCTL_CT;
4259         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
4260                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
4261
4262         igb_config_collision_dist(hw);
4263
4264         /* Enable transmits */
4265         tctl |= E1000_TCTL_EN;
4266
4267         wr32(E1000_TCTL, tctl);
4268 }
4269
4270 /**
4271  *  igb_configure_tx_ring - Configure transmit ring after Reset
4272  *  @adapter: board private structure
4273  *  @ring: tx ring to configure
4274  *
4275  *  Configure a transmit ring after a reset.
4276  **/
4277 void igb_configure_tx_ring(struct igb_adapter *adapter,
4278                            struct igb_ring *ring)
4279 {
4280         struct e1000_hw *hw = &adapter->hw;
4281         u32 txdctl = 0;
4282         u64 tdba = ring->dma;
4283         int reg_idx = ring->reg_idx;
4284
4285         wr32(E1000_TDLEN(reg_idx),
4286              ring->count * sizeof(union e1000_adv_tx_desc));
4287         wr32(E1000_TDBAL(reg_idx),
4288              tdba & 0x00000000ffffffffULL);
4289         wr32(E1000_TDBAH(reg_idx), tdba >> 32);
4290
4291         ring->tail = adapter->io_addr + E1000_TDT(reg_idx);
4292         wr32(E1000_TDH(reg_idx), 0);
4293         writel(0, ring->tail);
4294
4295         txdctl |= IGB_TX_PTHRESH;
4296         txdctl |= IGB_TX_HTHRESH << 8;
4297         txdctl |= IGB_TX_WTHRESH << 16;
4298
4299         /* reinitialize tx_buffer_info */
4300         memset(ring->tx_buffer_info, 0,
4301                sizeof(struct igb_tx_buffer) * ring->count);
4302
4303         txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
4304         wr32(E1000_TXDCTL(reg_idx), txdctl);
4305 }
4306
4307 /**
4308  *  igb_configure_tx - Configure transmit Unit after Reset
4309  *  @adapter: board private structure
4310  *
4311  *  Configure the Tx unit of the MAC after a reset.
4312  **/
4313 static void igb_configure_tx(struct igb_adapter *adapter)
4314 {
4315         struct e1000_hw *hw = &adapter->hw;
4316         int i;
4317
4318         /* disable the queues */
4319         for (i = 0; i < adapter->num_tx_queues; i++)
4320                 wr32(E1000_TXDCTL(adapter->tx_ring[i]->reg_idx), 0);
4321
4322         wrfl();
4323         usleep_range(10000, 20000);
4324
4325         for (i = 0; i < adapter->num_tx_queues; i++)
4326                 igb_configure_tx_ring(adapter, adapter->tx_ring[i]);
4327 }
4328
4329 /**
4330  *  igb_setup_rx_resources - allocate Rx resources (Descriptors)
4331  *  @rx_ring: Rx descriptor ring (for a specific queue) to setup
4332  *
4333  *  Returns 0 on success, negative on failure
4334  **/
4335 int igb_setup_rx_resources(struct igb_ring *rx_ring)
4336 {
4337         struct igb_adapter *adapter = netdev_priv(rx_ring->netdev);
4338         struct device *dev = rx_ring->dev;
4339         int size;
4340
4341         size = sizeof(struct igb_rx_buffer) * rx_ring->count;
4342
4343         rx_ring->rx_buffer_info = vmalloc(size);
4344         if (!rx_ring->rx_buffer_info)
4345                 goto err;
4346
4347         /* Round up to nearest 4K */
4348         rx_ring->size = rx_ring->count * sizeof(union e1000_adv_rx_desc);
4349         rx_ring->size = ALIGN(rx_ring->size, 4096);
4350
4351         rx_ring->desc = dma_alloc_coherent(dev, rx_ring->size,
4352                                            &rx_ring->dma, GFP_KERNEL);
4353         if (!rx_ring->desc)
4354                 goto err;
4355
4356         rx_ring->next_to_alloc = 0;
4357         rx_ring->next_to_clean = 0;
4358         rx_ring->next_to_use = 0;
4359
4360         rx_ring->xdp_prog = adapter->xdp_prog;
4361
4362         /* XDP RX-queue info */
4363         if (xdp_rxq_info_reg(&rx_ring->xdp_rxq, rx_ring->netdev,
4364                              rx_ring->queue_index, 0) < 0)
4365                 goto err;
4366
4367         return 0;
4368
4369 err:
4370         vfree(rx_ring->rx_buffer_info);
4371         rx_ring->rx_buffer_info = NULL;
4372         dev_err(dev, "Unable to allocate memory for the Rx descriptor ring\n");
4373         return -ENOMEM;
4374 }
4375
4376 /**
4377  *  igb_setup_all_rx_resources - wrapper to allocate Rx resources
4378  *                               (Descriptors) for all queues
4379  *  @adapter: board private structure
4380  *
4381  *  Return 0 on success, negative on failure
4382  **/
4383 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
4384 {
4385         struct pci_dev *pdev = adapter->pdev;
4386         int i, err = 0;
4387
4388         for (i = 0; i < adapter->num_rx_queues; i++) {
4389                 err = igb_setup_rx_resources(adapter->rx_ring[i]);
4390                 if (err) {
4391                         dev_err(&pdev->dev,
4392                                 "Allocation for Rx Queue %u failed\n", i);
4393                         for (i--; i >= 0; i--)
4394                                 igb_free_rx_resources(adapter->rx_ring[i]);
4395                         break;
4396                 }
4397         }
4398
4399         return err;
4400 }
4401
4402 /**
4403  *  igb_setup_mrqc - configure the multiple receive queue control registers
4404  *  @adapter: Board private structure
4405  **/
4406 static void igb_setup_mrqc(struct igb_adapter *adapter)
4407 {
4408         struct e1000_hw *hw = &adapter->hw;
4409         u32 mrqc, rxcsum;
4410         u32 j, num_rx_queues;
4411         u32 rss_key[10];
4412
4413         netdev_rss_key_fill(rss_key, sizeof(rss_key));
4414         for (j = 0; j < 10; j++)
4415                 wr32(E1000_RSSRK(j), rss_key[j]);
4416
4417         num_rx_queues = adapter->rss_queues;
4418
4419         switch (hw->mac.type) {
4420         case e1000_82576:
4421                 /* 82576 supports 2 RSS queues for SR-IOV */
4422                 if (adapter->vfs_allocated_count)
4423                         num_rx_queues = 2;
4424                 break;
4425         default:
4426                 break;
4427         }
4428
4429         if (adapter->rss_indir_tbl_init != num_rx_queues) {
4430                 for (j = 0; j < IGB_RETA_SIZE; j++)
4431                         adapter->rss_indir_tbl[j] =
4432                         (j * num_rx_queues) / IGB_RETA_SIZE;
4433                 adapter->rss_indir_tbl_init = num_rx_queues;
4434         }
4435         igb_write_rss_indir_tbl(adapter);
4436
4437         /* Disable raw packet checksumming so that RSS hash is placed in
4438          * descriptor on writeback.  No need to enable TCP/UDP/IP checksum
4439          * offloads as they are enabled by default
4440          */
4441         rxcsum = rd32(E1000_RXCSUM);
4442         rxcsum |= E1000_RXCSUM_PCSD;
4443
4444         if (adapter->hw.mac.type >= e1000_82576)
4445                 /* Enable Receive Checksum Offload for SCTP */
4446                 rxcsum |= E1000_RXCSUM_CRCOFL;
4447
4448         /* Don't need to set TUOFL or IPOFL, they default to 1 */
4449         wr32(E1000_RXCSUM, rxcsum);
4450
4451         /* Generate RSS hash based on packet types, TCP/UDP
4452          * port numbers and/or IPv4/v6 src and dst addresses
4453          */
4454         mrqc = E1000_MRQC_RSS_FIELD_IPV4 |
4455                E1000_MRQC_RSS_FIELD_IPV4_TCP |
4456                E1000_MRQC_RSS_FIELD_IPV6 |
4457                E1000_MRQC_RSS_FIELD_IPV6_TCP |
4458                E1000_MRQC_RSS_FIELD_IPV6_TCP_EX;
4459
4460         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV4_UDP)
4461                 mrqc |= E1000_MRQC_RSS_FIELD_IPV4_UDP;
4462         if (adapter->flags & IGB_FLAG_RSS_FIELD_IPV6_UDP)
4463                 mrqc |= E1000_MRQC_RSS_FIELD_IPV6_UDP;
4464
4465         /* If VMDq is enabled then we set the appropriate mode for that, else
4466          * we default to RSS so that an RSS hash is calculated per packet even
4467          * if we are only using one queue
4468          */
4469         if (adapter->vfs_allocated_count) {
4470                 if (hw->mac.type > e1000_82575) {
4471                         /* Set the default pool for the PF's first queue */
4472                         u32 vtctl = rd32(E1000_VT_CTL);
4473
4474                         vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
4475                                    E1000_VT_CTL_DISABLE_DEF_POOL);
4476                         vtctl |= adapter->vfs_allocated_count <<
4477                                 E1000_VT_CTL_DEFAULT_POOL_SHIFT;
4478                         wr32(E1000_VT_CTL, vtctl);
4479                 }
4480                 if (adapter->rss_queues > 1)
4481                         mrqc |= E1000_MRQC_ENABLE_VMDQ_RSS_MQ;
4482                 else
4483                         mrqc |= E1000_MRQC_ENABLE_VMDQ;
4484         } else {
4485                 mrqc |= E1000_MRQC_ENABLE_RSS_MQ;
4486         }
4487         igb_vmm_control(adapter);
4488
4489         wr32(E1000_MRQC, mrqc);
4490 }
4491
4492 /**
4493  *  igb_setup_rctl - configure the receive control registers
4494  *  @adapter: Board private structure
4495  **/
4496 void igb_setup_rctl(struct igb_adapter *adapter)
4497 {
4498         struct e1000_hw *hw = &adapter->hw;
4499         u32 rctl;
4500
4501         rctl = rd32(E1000_RCTL);
4502
4503         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
4504         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
4505
4506         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
4507                 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
4508
4509         /* enable stripping of CRC. It's unlikely this will break BMC
4510          * redirection as it did with e1000. Newer features require
4511          * that the HW strips the CRC.
4512          */
4513         rctl |= E1000_RCTL_SECRC;
4514
4515         /* disable store bad packets and clear size bits. */
4516         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
4517
4518         /* enable LPE to allow for reception of jumbo frames */
4519         rctl |= E1000_RCTL_LPE;
4520
4521         /* disable queue 0 to prevent tail write w/o re-config */
4522         wr32(E1000_RXDCTL(0), 0);
4523
4524         /* Attention!!!  For SR-IOV PF driver operations you must enable
4525          * queue drop for all VF and PF queues to prevent head of line blocking
4526          * if an un-trusted VF does not provide descriptors to hardware.
4527          */
4528         if (adapter->vfs_allocated_count) {
4529                 /* set all queue drop enable bits */
4530                 wr32(E1000_QDE, ALL_QUEUES);
4531         }
4532
4533         /* This is useful for sniffing bad packets. */
4534         if (adapter->netdev->features & NETIF_F_RXALL) {
4535                 /* UPE and MPE will be handled by normal PROMISC logic
4536                  * in e1000e_set_rx_mode
4537                  */
4538                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
4539                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
4540                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
4541
4542                 rctl &= ~(E1000_RCTL_DPF | /* Allow filtered pause */
4543                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
4544                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
4545                  * and that breaks VLANs.
4546                  */
4547         }
4548
4549         wr32(E1000_RCTL, rctl);
4550 }
4551
4552 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
4553                                    int vfn)
4554 {
4555         struct e1000_hw *hw = &adapter->hw;
4556         u32 vmolr;
4557
4558         if (size > MAX_JUMBO_FRAME_SIZE)
4559                 size = MAX_JUMBO_FRAME_SIZE;
4560
4561         vmolr = rd32(E1000_VMOLR(vfn));
4562         vmolr &= ~E1000_VMOLR_RLPML_MASK;
4563         vmolr |= size | E1000_VMOLR_LPE;
4564         wr32(E1000_VMOLR(vfn), vmolr);
4565
4566         return 0;
4567 }
4568
4569 static inline void igb_set_vf_vlan_strip(struct igb_adapter *adapter,
4570                                          int vfn, bool enable)
4571 {
4572         struct e1000_hw *hw = &adapter->hw;
4573         u32 val, reg;
4574
4575         if (hw->mac.type < e1000_82576)
4576                 return;
4577
4578         if (hw->mac.type == e1000_i350)
4579                 reg = E1000_DVMOLR(vfn);
4580         else
4581                 reg = E1000_VMOLR(vfn);
4582
4583         val = rd32(reg);
4584         if (enable)
4585                 val |= E1000_VMOLR_STRVLAN;
4586         else
4587                 val &= ~(E1000_VMOLR_STRVLAN);
4588         wr32(reg, val);
4589 }
4590
4591 static inline void igb_set_vmolr(struct igb_adapter *adapter,
4592                                  int vfn, bool aupe)
4593 {
4594         struct e1000_hw *hw = &adapter->hw;
4595         u32 vmolr;
4596
4597         /* This register exists only on 82576 and newer so if we are older then
4598          * we should exit and do nothing
4599          */
4600         if (hw->mac.type < e1000_82576)
4601                 return;
4602
4603         vmolr = rd32(E1000_VMOLR(vfn));
4604         if (aupe)
4605                 vmolr |= E1000_VMOLR_AUPE; /* Accept untagged packets */
4606         else
4607                 vmolr &= ~(E1000_VMOLR_AUPE); /* Tagged packets ONLY */
4608
4609         /* clear all bits that might not be set */
4610         vmolr &= ~(E1000_VMOLR_BAM | E1000_VMOLR_RSSE);
4611
4612         if (adapter->rss_queues > 1 && vfn == adapter->vfs_allocated_count)
4613                 vmolr |= E1000_VMOLR_RSSE; /* enable RSS */
4614         /* for VMDq only allow the VFs and pool 0 to accept broadcast and
4615          * multicast packets
4616          */
4617         if (vfn <= adapter->vfs_allocated_count)
4618                 vmolr |= E1000_VMOLR_BAM; /* Accept broadcast */
4619
4620         wr32(E1000_VMOLR(vfn), vmolr);
4621 }
4622
4623 /**
4624  *  igb_setup_srrctl - configure the split and replication receive control
4625  *                     registers
4626  *  @adapter: Board private structure
4627  *  @ring: receive ring to be configured
4628  **/
4629 void igb_setup_srrctl(struct igb_adapter *adapter, struct igb_ring *ring)
4630 {
4631         struct e1000_hw *hw = &adapter->hw;
4632         int reg_idx = ring->reg_idx;
4633         u32 srrctl = 0;
4634
4635         srrctl = IGB_RX_HDR_LEN << E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
4636         if (ring_uses_large_buffer(ring))
4637                 srrctl |= IGB_RXBUFFER_3072 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4638         else
4639                 srrctl |= IGB_RXBUFFER_2048 >> E1000_SRRCTL_BSIZEPKT_SHIFT;
4640         srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
4641         if (hw->mac.type >= e1000_82580)
4642                 srrctl |= E1000_SRRCTL_TIMESTAMP;
4643         /* Only set Drop Enable if VFs allocated, or we are supporting multiple
4644          * queues and rx flow control is disabled
4645          */
4646         if (adapter->vfs_allocated_count ||
4647             (!(hw->fc.current_mode & e1000_fc_rx_pause) &&
4648              adapter->num_rx_queues > 1))
4649                 srrctl |= E1000_SRRCTL_DROP_EN;
4650
4651         wr32(E1000_SRRCTL(reg_idx), srrctl);
4652 }
4653
4654 /**
4655  *  igb_configure_rx_ring - Configure a receive ring after Reset
4656  *  @adapter: board private structure
4657  *  @ring: receive ring to be configured
4658  *
4659  *  Configure the Rx unit of the MAC after a reset.
4660  **/
4661 void igb_configure_rx_ring(struct igb_adapter *adapter,
4662                            struct igb_ring *ring)
4663 {
4664         struct e1000_hw *hw = &adapter->hw;
4665         union e1000_adv_rx_desc *rx_desc;
4666         u64 rdba = ring->dma;
4667         int reg_idx = ring->reg_idx;
4668         u32 rxdctl = 0;
4669
4670         xdp_rxq_info_unreg_mem_model(&ring->xdp_rxq);
4671         WARN_ON(xdp_rxq_info_reg_mem_model(&ring->xdp_rxq,
4672                                            MEM_TYPE_PAGE_SHARED, NULL));
4673
4674         /* disable the queue */
4675         wr32(E1000_RXDCTL(reg_idx), 0);
4676
4677         /* Set DMA base address registers */
4678         wr32(E1000_RDBAL(reg_idx),
4679              rdba & 0x00000000ffffffffULL);
4680         wr32(E1000_RDBAH(reg_idx), rdba >> 32);
4681         wr32(E1000_RDLEN(reg_idx),
4682              ring->count * sizeof(union e1000_adv_rx_desc));
4683
4684         /* initialize head and tail */
4685         ring->tail = adapter->io_addr + E1000_RDT(reg_idx);
4686         wr32(E1000_RDH(reg_idx), 0);
4687         writel(0, ring->tail);
4688
4689         /* set descriptor configuration */
4690         igb_setup_srrctl(adapter, ring);
4691
4692         /* set filtering for VMDQ pools */
4693         igb_set_vmolr(adapter, reg_idx & 0x7, true);
4694
4695         rxdctl |= IGB_RX_PTHRESH;
4696         rxdctl |= IGB_RX_HTHRESH << 8;
4697         rxdctl |= IGB_RX_WTHRESH << 16;
4698
4699         /* initialize rx_buffer_info */
4700         memset(ring->rx_buffer_info, 0,
4701                sizeof(struct igb_rx_buffer) * ring->count);
4702
4703         /* initialize Rx descriptor 0 */
4704         rx_desc = IGB_RX_DESC(ring, 0);
4705         rx_desc->wb.upper.length = 0;
4706
4707         /* enable receive descriptor fetching */
4708         rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
4709         wr32(E1000_RXDCTL(reg_idx), rxdctl);
4710 }
4711
4712 static void igb_set_rx_buffer_len(struct igb_adapter *adapter,
4713                                   struct igb_ring *rx_ring)
4714 {
4715         /* set build_skb and buffer size flags */
4716         clear_ring_build_skb_enabled(rx_ring);
4717         clear_ring_uses_large_buffer(rx_ring);
4718
4719         if (adapter->flags & IGB_FLAG_RX_LEGACY)
4720                 return;
4721
4722         set_ring_build_skb_enabled(rx_ring);
4723
4724 #if (PAGE_SIZE < 8192)
4725         if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
4726                 return;
4727
4728         set_ring_uses_large_buffer(rx_ring);
4729 #endif
4730 }
4731
4732 /**
4733  *  igb_configure_rx - Configure receive Unit after Reset
4734  *  @adapter: board private structure
4735  *
4736  *  Configure the Rx unit of the MAC after a reset.
4737  **/
4738 static void igb_configure_rx(struct igb_adapter *adapter)
4739 {
4740         int i;
4741
4742         /* set the correct pool for the PF default MAC address in entry 0 */
4743         igb_set_default_mac_filter(adapter);
4744
4745         /* Setup the HW Rx Head and Tail Descriptor Pointers and
4746          * the Base and Length of the Rx Descriptor Ring
4747          */
4748         for (i = 0; i < adapter->num_rx_queues; i++) {
4749                 struct igb_ring *rx_ring = adapter->rx_ring[i];
4750
4751                 igb_set_rx_buffer_len(adapter, rx_ring);
4752                 igb_configure_rx_ring(adapter, rx_ring);
4753         }
4754 }
4755
4756 /**
4757  *  igb_free_tx_resources - Free Tx Resources per Queue
4758  *  @tx_ring: Tx descriptor ring for a specific queue
4759  *
4760  *  Free all transmit software resources
4761  **/
4762 void igb_free_tx_resources(struct igb_ring *tx_ring)
4763 {
4764         igb_clean_tx_ring(tx_ring);
4765
4766         vfree(tx_ring->tx_buffer_info);
4767         tx_ring->tx_buffer_info = NULL;
4768
4769         /* if not set, then don't free */
4770         if (!tx_ring->desc)
4771                 return;
4772
4773         dma_free_coherent(tx_ring->dev, tx_ring->size,
4774                           tx_ring->desc, tx_ring->dma);
4775
4776         tx_ring->desc = NULL;
4777 }
4778
4779 /**
4780  *  igb_free_all_tx_resources - Free Tx Resources for All Queues
4781  *  @adapter: board private structure
4782  *
4783  *  Free all transmit software resources
4784  **/
4785 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
4786 {
4787         int i;
4788
4789         for (i = 0; i < adapter->num_tx_queues; i++)
4790                 if (adapter->tx_ring[i])
4791                         igb_free_tx_resources(adapter->tx_ring[i]);
4792 }
4793
4794 /**
4795  *  igb_clean_tx_ring - Free Tx Buffers
4796  *  @tx_ring: ring to be cleaned
4797  **/
4798 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
4799 {
4800         u16 i = tx_ring->next_to_clean;
4801         struct igb_tx_buffer *tx_buffer = &tx_ring->tx_buffer_info[i];
4802
4803         while (i != tx_ring->next_to_use) {
4804                 union e1000_adv_tx_desc *eop_desc, *tx_desc;
4805
4806                 /* Free all the Tx ring sk_buffs */
4807                 dev_kfree_skb_any(tx_buffer->skb);
4808
4809                 /* unmap skb header data */
4810                 dma_unmap_single(tx_ring->dev,
4811                                  dma_unmap_addr(tx_buffer, dma),
4812                                  dma_unmap_len(tx_buffer, len),
4813                                  DMA_TO_DEVICE);
4814
4815                 /* check for eop_desc to determine the end of the packet */
4816                 eop_desc = tx_buffer->next_to_watch;
4817                 tx_desc = IGB_TX_DESC(tx_ring, i);
4818
4819                 /* unmap remaining buffers */
4820                 while (tx_desc != eop_desc) {
4821                         tx_buffer++;
4822                         tx_desc++;
4823                         i++;
4824                         if (unlikely(i == tx_ring->count)) {
4825                                 i = 0;
4826                                 tx_buffer = tx_ring->tx_buffer_info;
4827                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
4828                         }
4829
4830                         /* unmap any remaining paged data */
4831                         if (dma_unmap_len(tx_buffer, len))
4832                                 dma_unmap_page(tx_ring->dev,
4833                                                dma_unmap_addr(tx_buffer, dma),
4834                                                dma_unmap_len(tx_buffer, len),
4835                                                DMA_TO_DEVICE);
4836                 }
4837
4838                 /* move us one more past the eop_desc for start of next pkt */
4839                 tx_buffer++;
4840                 i++;
4841                 if (unlikely(i == tx_ring->count)) {
4842                         i = 0;
4843                         tx_buffer = tx_ring->tx_buffer_info;
4844                 }
4845         }
4846
4847         /* reset BQL for queue */
4848         netdev_tx_reset_queue(txring_txq(tx_ring));
4849
4850         /* reset next_to_use and next_to_clean */
4851         tx_ring->next_to_use = 0;
4852         tx_ring->next_to_clean = 0;
4853 }
4854
4855 /**
4856  *  igb_clean_all_tx_rings - Free Tx Buffers for all queues
4857  *  @adapter: board private structure
4858  **/
4859 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
4860 {
4861         int i;
4862
4863         for (i = 0; i < adapter->num_tx_queues; i++)
4864                 if (adapter->tx_ring[i])
4865                         igb_clean_tx_ring(adapter->tx_ring[i]);
4866 }
4867
4868 /**
4869  *  igb_free_rx_resources - Free Rx Resources
4870  *  @rx_ring: ring to clean the resources from
4871  *
4872  *  Free all receive software resources
4873  **/
4874 void igb_free_rx_resources(struct igb_ring *rx_ring)
4875 {
4876         igb_clean_rx_ring(rx_ring);
4877
4878         rx_ring->xdp_prog = NULL;
4879         xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
4880         vfree(rx_ring->rx_buffer_info);
4881         rx_ring->rx_buffer_info = NULL;
4882
4883         /* if not set, then don't free */
4884         if (!rx_ring->desc)
4885                 return;
4886
4887         dma_free_coherent(rx_ring->dev, rx_ring->size,
4888                           rx_ring->desc, rx_ring->dma);
4889
4890         rx_ring->desc = NULL;
4891 }
4892
4893 /**
4894  *  igb_free_all_rx_resources - Free Rx Resources for All Queues
4895  *  @adapter: board private structure
4896  *
4897  *  Free all receive software resources
4898  **/
4899 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
4900 {
4901         int i;
4902
4903         for (i = 0; i < adapter->num_rx_queues; i++)
4904                 if (adapter->rx_ring[i])
4905                         igb_free_rx_resources(adapter->rx_ring[i]);
4906 }
4907
4908 /**
4909  *  igb_clean_rx_ring - Free Rx Buffers per Queue
4910  *  @rx_ring: ring to free buffers from
4911  **/
4912 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
4913 {
4914         u16 i = rx_ring->next_to_clean;
4915
4916         dev_kfree_skb(rx_ring->skb);
4917         rx_ring->skb = NULL;
4918
4919         /* Free all the Rx ring sk_buffs */
4920         while (i != rx_ring->next_to_alloc) {
4921                 struct igb_rx_buffer *buffer_info = &rx_ring->rx_buffer_info[i];
4922
4923                 /* Invalidate cache lines that may have been written to by
4924                  * device so that we avoid corrupting memory.
4925                  */
4926                 dma_sync_single_range_for_cpu(rx_ring->dev,
4927                                               buffer_info->dma,
4928                                               buffer_info->page_offset,
4929                                               igb_rx_bufsz(rx_ring),
4930                                               DMA_FROM_DEVICE);
4931
4932                 /* free resources associated with mapping */
4933                 dma_unmap_page_attrs(rx_ring->dev,
4934                                      buffer_info->dma,
4935                                      igb_rx_pg_size(rx_ring),
4936                                      DMA_FROM_DEVICE,
4937                                      IGB_RX_DMA_ATTR);
4938                 __page_frag_cache_drain(buffer_info->page,
4939                                         buffer_info->pagecnt_bias);
4940
4941                 i++;
4942                 if (i == rx_ring->count)
4943                         i = 0;
4944         }
4945
4946         rx_ring->next_to_alloc = 0;
4947         rx_ring->next_to_clean = 0;
4948         rx_ring->next_to_use = 0;
4949 }
4950
4951 /**
4952  *  igb_clean_all_rx_rings - Free Rx Buffers for all queues
4953  *  @adapter: board private structure
4954  **/
4955 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
4956 {
4957         int i;
4958
4959         for (i = 0; i < adapter->num_rx_queues; i++)
4960                 if (adapter->rx_ring[i])
4961                         igb_clean_rx_ring(adapter->rx_ring[i]);
4962 }
4963
4964 /**
4965  *  igb_set_mac - Change the Ethernet Address of the NIC
4966  *  @netdev: network interface device structure
4967  *  @p: pointer to an address structure
4968  *
4969  *  Returns 0 on success, negative on failure
4970  **/
4971 static int igb_set_mac(struct net_device *netdev, void *p)
4972 {
4973         struct igb_adapter *adapter = netdev_priv(netdev);
4974         struct e1000_hw *hw = &adapter->hw;
4975         struct sockaddr *addr = p;
4976
4977         if (!is_valid_ether_addr(addr->sa_data))
4978                 return -EADDRNOTAVAIL;
4979
4980         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
4981         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
4982
4983         /* set the correct pool for the new PF MAC address in entry 0 */
4984         igb_set_default_mac_filter(adapter);
4985
4986         return 0;
4987 }
4988
4989 /**
4990  *  igb_write_mc_addr_list - write multicast addresses to MTA
4991  *  @netdev: network interface device structure
4992  *
4993  *  Writes multicast address list to the MTA hash table.
4994  *  Returns: -ENOMEM on failure
4995  *           0 on no addresses written
4996  *           X on writing X addresses to MTA
4997  **/
4998 static int igb_write_mc_addr_list(struct net_device *netdev)
4999 {
5000         struct igb_adapter *adapter = netdev_priv(netdev);
5001         struct e1000_hw *hw = &adapter->hw;
5002         struct netdev_hw_addr *ha;
5003         u8  *mta_list;
5004         int i;
5005
5006         if (netdev_mc_empty(netdev)) {
5007                 /* nothing to program, so clear mc list */
5008                 igb_update_mc_addr_list(hw, NULL, 0);
5009                 igb_restore_vf_multicasts(adapter);
5010                 return 0;
5011         }
5012
5013         mta_list = kcalloc(netdev_mc_count(netdev), 6, GFP_ATOMIC);
5014         if (!mta_list)
5015                 return -ENOMEM;
5016
5017         /* The shared function expects a packed array of only addresses. */
5018         i = 0;
5019         netdev_for_each_mc_addr(ha, netdev)
5020                 memcpy(mta_list + (i++ * ETH_ALEN), ha->addr, ETH_ALEN);
5021
5022         igb_update_mc_addr_list(hw, mta_list, i);
5023         kfree(mta_list);
5024
5025         return netdev_mc_count(netdev);
5026 }
5027
5028 static int igb_vlan_promisc_enable(struct igb_adapter *adapter)
5029 {
5030         struct e1000_hw *hw = &adapter->hw;
5031         u32 i, pf_id;
5032
5033         switch (hw->mac.type) {
5034         case e1000_i210:
5035         case e1000_i211:
5036         case e1000_i350:
5037                 /* VLAN filtering needed for VLAN prio filter */
5038                 if (adapter->netdev->features & NETIF_F_NTUPLE)
5039                         break;
5040                 fallthrough;
5041         case e1000_82576:
5042         case e1000_82580:
5043         case e1000_i354:
5044                 /* VLAN filtering needed for pool filtering */
5045                 if (adapter->vfs_allocated_count)
5046                         break;
5047                 fallthrough;
5048         default:
5049                 return 1;
5050         }
5051
5052         /* We are already in VLAN promisc, nothing to do */
5053         if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
5054                 return 0;
5055
5056         if (!adapter->vfs_allocated_count)
5057                 goto set_vfta;
5058
5059         /* Add PF to all active pools */
5060         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5061
5062         for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5063                 u32 vlvf = rd32(E1000_VLVF(i));
5064
5065                 vlvf |= BIT(pf_id);
5066                 wr32(E1000_VLVF(i), vlvf);
5067         }
5068
5069 set_vfta:
5070         /* Set all bits in the VLAN filter table array */
5071         for (i = E1000_VLAN_FILTER_TBL_SIZE; i--;)
5072                 hw->mac.ops.write_vfta(hw, i, ~0U);
5073
5074         /* Set flag so we don't redo unnecessary work */
5075         adapter->flags |= IGB_FLAG_VLAN_PROMISC;
5076
5077         return 0;
5078 }
5079
5080 #define VFTA_BLOCK_SIZE 8
5081 static void igb_scrub_vfta(struct igb_adapter *adapter, u32 vfta_offset)
5082 {
5083         struct e1000_hw *hw = &adapter->hw;
5084         u32 vfta[VFTA_BLOCK_SIZE] = { 0 };
5085         u32 vid_start = vfta_offset * 32;
5086         u32 vid_end = vid_start + (VFTA_BLOCK_SIZE * 32);
5087         u32 i, vid, word, bits, pf_id;
5088
5089         /* guarantee that we don't scrub out management VLAN */
5090         vid = adapter->mng_vlan_id;
5091         if (vid >= vid_start && vid < vid_end)
5092                 vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5093
5094         if (!adapter->vfs_allocated_count)
5095                 goto set_vfta;
5096
5097         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
5098
5099         for (i = E1000_VLVF_ARRAY_SIZE; --i;) {
5100                 u32 vlvf = rd32(E1000_VLVF(i));
5101
5102                 /* pull VLAN ID from VLVF */
5103                 vid = vlvf & VLAN_VID_MASK;
5104
5105                 /* only concern ourselves with a certain range */
5106                 if (vid < vid_start || vid >= vid_end)
5107                         continue;
5108
5109                 if (vlvf & E1000_VLVF_VLANID_ENABLE) {
5110                         /* record VLAN ID in VFTA */
5111                         vfta[(vid - vid_start) / 32] |= BIT(vid % 32);
5112
5113                         /* if PF is part of this then continue */
5114                         if (test_bit(vid, adapter->active_vlans))
5115                                 continue;
5116                 }
5117
5118                 /* remove PF from the pool */
5119                 bits = ~BIT(pf_id);
5120                 bits &= rd32(E1000_VLVF(i));
5121                 wr32(E1000_VLVF(i), bits);
5122         }
5123
5124 set_vfta:
5125         /* extract values from active_vlans and write back to VFTA */
5126         for (i = VFTA_BLOCK_SIZE; i--;) {
5127                 vid = (vfta_offset + i) * 32;
5128                 word = vid / BITS_PER_LONG;
5129                 bits = vid % BITS_PER_LONG;
5130
5131                 vfta[i] |= adapter->active_vlans[word] >> bits;
5132
5133                 hw->mac.ops.write_vfta(hw, vfta_offset + i, vfta[i]);
5134         }
5135 }
5136
5137 static void igb_vlan_promisc_disable(struct igb_adapter *adapter)
5138 {
5139         u32 i;
5140
5141         /* We are not in VLAN promisc, nothing to do */
5142         if (!(adapter->flags & IGB_FLAG_VLAN_PROMISC))
5143                 return;
5144
5145         /* Set flag so we don't redo unnecessary work */
5146         adapter->flags &= ~IGB_FLAG_VLAN_PROMISC;
5147
5148         for (i = 0; i < E1000_VLAN_FILTER_TBL_SIZE; i += VFTA_BLOCK_SIZE)
5149                 igb_scrub_vfta(adapter, i);
5150 }
5151
5152 /**
5153  *  igb_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
5154  *  @netdev: network interface device structure
5155  *
5156  *  The set_rx_mode entry point is called whenever the unicast or multicast
5157  *  address lists or the network interface flags are updated.  This routine is
5158  *  responsible for configuring the hardware for proper unicast, multicast,
5159  *  promiscuous mode, and all-multi behavior.
5160  **/
5161 static void igb_set_rx_mode(struct net_device *netdev)
5162 {
5163         struct igb_adapter *adapter = netdev_priv(netdev);
5164         struct e1000_hw *hw = &adapter->hw;
5165         unsigned int vfn = adapter->vfs_allocated_count;
5166         u32 rctl = 0, vmolr = 0, rlpml = MAX_JUMBO_FRAME_SIZE;
5167         int count;
5168
5169         /* Check for Promiscuous and All Multicast modes */
5170         if (netdev->flags & IFF_PROMISC) {
5171                 rctl |= E1000_RCTL_UPE | E1000_RCTL_MPE;
5172                 vmolr |= E1000_VMOLR_MPME;
5173
5174                 /* enable use of UTA filter to force packets to default pool */
5175                 if (hw->mac.type == e1000_82576)
5176                         vmolr |= E1000_VMOLR_ROPE;
5177         } else {
5178                 if (netdev->flags & IFF_ALLMULTI) {
5179                         rctl |= E1000_RCTL_MPE;
5180                         vmolr |= E1000_VMOLR_MPME;
5181                 } else {
5182                         /* Write addresses to the MTA, if the attempt fails
5183                          * then we should just turn on promiscuous mode so
5184                          * that we can at least receive multicast traffic
5185                          */
5186                         count = igb_write_mc_addr_list(netdev);
5187                         if (count < 0) {
5188                                 rctl |= E1000_RCTL_MPE;
5189                                 vmolr |= E1000_VMOLR_MPME;
5190                         } else if (count) {
5191                                 vmolr |= E1000_VMOLR_ROMPE;
5192                         }
5193                 }
5194         }
5195
5196         /* Write addresses to available RAR registers, if there is not
5197          * sufficient space to store all the addresses then enable
5198          * unicast promiscuous mode
5199          */
5200         if (__dev_uc_sync(netdev, igb_uc_sync, igb_uc_unsync)) {
5201                 rctl |= E1000_RCTL_UPE;
5202                 vmolr |= E1000_VMOLR_ROPE;
5203         }
5204
5205         /* enable VLAN filtering by default */
5206         rctl |= E1000_RCTL_VFE;
5207
5208         /* disable VLAN filtering for modes that require it */
5209         if ((netdev->flags & IFF_PROMISC) ||
5210             (netdev->features & NETIF_F_RXALL)) {
5211                 /* if we fail to set all rules then just clear VFE */
5212                 if (igb_vlan_promisc_enable(adapter))
5213                         rctl &= ~E1000_RCTL_VFE;
5214         } else {
5215                 igb_vlan_promisc_disable(adapter);
5216         }
5217
5218         /* update state of unicast, multicast, and VLAN filtering modes */
5219         rctl |= rd32(E1000_RCTL) & ~(E1000_RCTL_UPE | E1000_RCTL_MPE |
5220                                      E1000_RCTL_VFE);
5221         wr32(E1000_RCTL, rctl);
5222
5223 #if (PAGE_SIZE < 8192)
5224         if (!adapter->vfs_allocated_count) {
5225                 if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5226                         rlpml = IGB_MAX_FRAME_BUILD_SKB;
5227         }
5228 #endif
5229         wr32(E1000_RLPML, rlpml);
5230
5231         /* In order to support SR-IOV and eventually VMDq it is necessary to set
5232          * the VMOLR to enable the appropriate modes.  Without this workaround
5233          * we will have issues with VLAN tag stripping not being done for frames
5234          * that are only arriving because we are the default pool
5235          */
5236         if ((hw->mac.type < e1000_82576) || (hw->mac.type > e1000_i350))
5237                 return;
5238
5239         /* set UTA to appropriate mode */
5240         igb_set_uta(adapter, !!(vmolr & E1000_VMOLR_ROPE));
5241
5242         vmolr |= rd32(E1000_VMOLR(vfn)) &
5243                  ~(E1000_VMOLR_ROPE | E1000_VMOLR_MPME | E1000_VMOLR_ROMPE);
5244
5245         /* enable Rx jumbo frames, restrict as needed to support build_skb */
5246         vmolr &= ~E1000_VMOLR_RLPML_MASK;
5247 #if (PAGE_SIZE < 8192)
5248         if (adapter->max_frame_size <= IGB_MAX_FRAME_BUILD_SKB)
5249                 vmolr |= IGB_MAX_FRAME_BUILD_SKB;
5250         else
5251 #endif
5252                 vmolr |= MAX_JUMBO_FRAME_SIZE;
5253         vmolr |= E1000_VMOLR_LPE;
5254
5255         wr32(E1000_VMOLR(vfn), vmolr);
5256
5257         igb_restore_vf_multicasts(adapter);
5258 }
5259
5260 static void igb_check_wvbr(struct igb_adapter *adapter)
5261 {
5262         struct e1000_hw *hw = &adapter->hw;
5263         u32 wvbr = 0;
5264
5265         switch (hw->mac.type) {
5266         case e1000_82576:
5267         case e1000_i350:
5268                 wvbr = rd32(E1000_WVBR);
5269                 if (!wvbr)
5270                         return;
5271                 break;
5272         default:
5273                 break;
5274         }
5275
5276         adapter->wvbr |= wvbr;
5277 }
5278
5279 #define IGB_STAGGERED_QUEUE_OFFSET 8
5280
5281 static void igb_spoof_check(struct igb_adapter *adapter)
5282 {
5283         int j;
5284
5285         if (!adapter->wvbr)
5286                 return;
5287
5288         for (j = 0; j < adapter->vfs_allocated_count; j++) {
5289                 if (adapter->wvbr & BIT(j) ||
5290                     adapter->wvbr & BIT(j + IGB_STAGGERED_QUEUE_OFFSET)) {
5291                         dev_warn(&adapter->pdev->dev,
5292                                 "Spoof event(s) detected on VF %d\n", j);
5293                         adapter->wvbr &=
5294                                 ~(BIT(j) |
5295                                   BIT(j + IGB_STAGGERED_QUEUE_OFFSET));
5296                 }
5297         }
5298 }
5299
5300 /* Need to wait a few seconds after link up to get diagnostic information from
5301  * the phy
5302  */
5303 static void igb_update_phy_info(struct timer_list *t)
5304 {
5305         struct igb_adapter *adapter = from_timer(adapter, t, phy_info_timer);
5306         igb_get_phy_info(&adapter->hw);
5307 }
5308
5309 /**
5310  *  igb_has_link - check shared code for link and determine up/down
5311  *  @adapter: pointer to driver private info
5312  **/
5313 bool igb_has_link(struct igb_adapter *adapter)
5314 {
5315         struct e1000_hw *hw = &adapter->hw;
5316         bool link_active = false;
5317
5318         /* get_link_status is set on LSC (link status) interrupt or
5319          * rx sequence error interrupt.  get_link_status will stay
5320          * false until the e1000_check_for_link establishes link
5321          * for copper adapters ONLY
5322          */
5323         switch (hw->phy.media_type) {
5324         case e1000_media_type_copper:
5325                 if (!hw->mac.get_link_status)
5326                         return true;
5327                 fallthrough;
5328         case e1000_media_type_internal_serdes:
5329                 hw->mac.ops.check_for_link(hw);
5330                 link_active = !hw->mac.get_link_status;
5331                 break;
5332         default:
5333         case e1000_media_type_unknown:
5334                 break;
5335         }
5336
5337         if (((hw->mac.type == e1000_i210) ||
5338              (hw->mac.type == e1000_i211)) &&
5339              (hw->phy.id == I210_I_PHY_ID)) {
5340                 if (!netif_carrier_ok(adapter->netdev)) {
5341                         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5342                 } else if (!(adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)) {
5343                         adapter->flags |= IGB_FLAG_NEED_LINK_UPDATE;
5344                         adapter->link_check_timeout = jiffies;
5345                 }
5346         }
5347
5348         return link_active;
5349 }
5350
5351 static bool igb_thermal_sensor_event(struct e1000_hw *hw, u32 event)
5352 {
5353         bool ret = false;
5354         u32 ctrl_ext, thstat;
5355
5356         /* check for thermal sensor event on i350 copper only */
5357         if (hw->mac.type == e1000_i350) {
5358                 thstat = rd32(E1000_THSTAT);
5359                 ctrl_ext = rd32(E1000_CTRL_EXT);
5360
5361                 if ((hw->phy.media_type == e1000_media_type_copper) &&
5362                     !(ctrl_ext & E1000_CTRL_EXT_LINK_MODE_SGMII))
5363                         ret = !!(thstat & event);
5364         }
5365
5366         return ret;
5367 }
5368
5369 /**
5370  *  igb_check_lvmmc - check for malformed packets received
5371  *  and indicated in LVMMC register
5372  *  @adapter: pointer to adapter
5373  **/
5374 static void igb_check_lvmmc(struct igb_adapter *adapter)
5375 {
5376         struct e1000_hw *hw = &adapter->hw;
5377         u32 lvmmc;
5378
5379         lvmmc = rd32(E1000_LVMMC);
5380         if (lvmmc) {
5381                 if (unlikely(net_ratelimit())) {
5382                         netdev_warn(adapter->netdev,
5383                                     "malformed Tx packet detected and dropped, LVMMC:0x%08x\n",
5384                                     lvmmc);
5385                 }
5386         }
5387 }
5388
5389 /**
5390  *  igb_watchdog - Timer Call-back
5391  *  @t: pointer to timer_list containing our private info pointer
5392  **/
5393 static void igb_watchdog(struct timer_list *t)
5394 {
5395         struct igb_adapter *adapter = from_timer(adapter, t, watchdog_timer);
5396         /* Do the rest outside of interrupt context */
5397         schedule_work(&adapter->watchdog_task);
5398 }
5399
5400 static void igb_watchdog_task(struct work_struct *work)
5401 {
5402         struct igb_adapter *adapter = container_of(work,
5403                                                    struct igb_adapter,
5404                                                    watchdog_task);
5405         struct e1000_hw *hw = &adapter->hw;
5406         struct e1000_phy_info *phy = &hw->phy;
5407         struct net_device *netdev = adapter->netdev;
5408         u32 link;
5409         int i;
5410         u32 connsw;
5411         u16 phy_data, retry_count = 20;
5412
5413         link = igb_has_link(adapter);
5414
5415         if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE) {
5416                 if (time_after(jiffies, (adapter->link_check_timeout + HZ)))
5417                         adapter->flags &= ~IGB_FLAG_NEED_LINK_UPDATE;
5418                 else
5419                         link = false;
5420         }
5421
5422         /* Force link down if we have fiber to swap to */
5423         if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5424                 if (hw->phy.media_type == e1000_media_type_copper) {
5425                         connsw = rd32(E1000_CONNSW);
5426                         if (!(connsw & E1000_CONNSW_AUTOSENSE_EN))
5427                                 link = 0;
5428                 }
5429         }
5430         if (link) {
5431                 /* Perform a reset if the media type changed. */
5432                 if (hw->dev_spec._82575.media_changed) {
5433                         hw->dev_spec._82575.media_changed = false;
5434                         adapter->flags |= IGB_FLAG_MEDIA_RESET;
5435                         igb_reset(adapter);
5436                 }
5437                 /* Cancel scheduled suspend requests. */
5438                 pm_runtime_resume(netdev->dev.parent);
5439
5440                 if (!netif_carrier_ok(netdev)) {
5441                         u32 ctrl;
5442
5443                         hw->mac.ops.get_speed_and_duplex(hw,
5444                                                          &adapter->link_speed,
5445                                                          &adapter->link_duplex);
5446
5447                         ctrl = rd32(E1000_CTRL);
5448                         /* Links status message must follow this format */
5449                         netdev_info(netdev,
5450                                "igb: %s NIC Link is Up %d Mbps %s Duplex, Flow Control: %s\n",
5451                                netdev->name,
5452                                adapter->link_speed,
5453                                adapter->link_duplex == FULL_DUPLEX ?
5454                                "Full" : "Half",
5455                                (ctrl & E1000_CTRL_TFCE) &&
5456                                (ctrl & E1000_CTRL_RFCE) ? "RX/TX" :
5457                                (ctrl & E1000_CTRL_RFCE) ?  "RX" :
5458                                (ctrl & E1000_CTRL_TFCE) ?  "TX" : "None");
5459
5460                         /* disable EEE if enabled */
5461                         if ((adapter->flags & IGB_FLAG_EEE) &&
5462                                 (adapter->link_duplex == HALF_DUPLEX)) {
5463                                 dev_info(&adapter->pdev->dev,
5464                                 "EEE Disabled: unsupported at half duplex. Re-enable using ethtool when at full duplex.\n");
5465                                 adapter->hw.dev_spec._82575.eee_disable = true;
5466                                 adapter->flags &= ~IGB_FLAG_EEE;
5467                         }
5468
5469                         /* check if SmartSpeed worked */
5470                         igb_check_downshift(hw);
5471                         if (phy->speed_downgraded)
5472                                 netdev_warn(netdev, "Link Speed was downgraded by SmartSpeed\n");
5473
5474                         /* check for thermal sensor event */
5475                         if (igb_thermal_sensor_event(hw,
5476                             E1000_THSTAT_LINK_THROTTLE))
5477                                 netdev_info(netdev, "The network adapter link speed was downshifted because it overheated\n");
5478
5479                         /* adjust timeout factor according to speed/duplex */
5480                         adapter->tx_timeout_factor = 1;
5481                         switch (adapter->link_speed) {
5482                         case SPEED_10:
5483                                 adapter->tx_timeout_factor = 14;
5484                                 break;
5485                         case SPEED_100:
5486                                 /* maybe add some timeout factor ? */
5487                                 break;
5488                         }
5489
5490                         if (adapter->link_speed != SPEED_1000)
5491                                 goto no_wait;
5492
5493                         /* wait for Remote receiver status OK */
5494 retry_read_status:
5495                         if (!igb_read_phy_reg(hw, PHY_1000T_STATUS,
5496                                               &phy_data)) {
5497                                 if (!(phy_data & SR_1000T_REMOTE_RX_STATUS) &&
5498                                     retry_count) {
5499                                         msleep(100);
5500                                         retry_count--;
5501                                         goto retry_read_status;
5502                                 } else if (!retry_count) {
5503                                         dev_err(&adapter->pdev->dev, "exceed max 2 second\n");
5504                                 }
5505                         } else {
5506                                 dev_err(&adapter->pdev->dev, "read 1000Base-T Status Reg\n");
5507                         }
5508 no_wait:
5509                         netif_carrier_on(netdev);
5510
5511                         igb_ping_all_vfs(adapter);
5512                         igb_check_vf_rate_limit(adapter);
5513
5514                         /* link state has changed, schedule phy info update */
5515                         if (!test_bit(__IGB_DOWN, &adapter->state))
5516                                 mod_timer(&adapter->phy_info_timer,
5517                                           round_jiffies(jiffies + 2 * HZ));
5518                 }
5519         } else {
5520                 if (netif_carrier_ok(netdev)) {
5521                         adapter->link_speed = 0;
5522                         adapter->link_duplex = 0;
5523
5524                         /* check for thermal sensor event */
5525                         if (igb_thermal_sensor_event(hw,
5526                             E1000_THSTAT_PWR_DOWN)) {
5527                                 netdev_err(netdev, "The network adapter was stopped because it overheated\n");
5528                         }
5529
5530                         /* Links status message must follow this format */
5531                         netdev_info(netdev, "igb: %s NIC Link is Down\n",
5532                                netdev->name);
5533                         netif_carrier_off(netdev);
5534
5535                         igb_ping_all_vfs(adapter);
5536
5537                         /* link state has changed, schedule phy info update */
5538                         if (!test_bit(__IGB_DOWN, &adapter->state))
5539                                 mod_timer(&adapter->phy_info_timer,
5540                                           round_jiffies(jiffies + 2 * HZ));
5541
5542                         /* link is down, time to check for alternate media */
5543                         if (adapter->flags & IGB_FLAG_MAS_ENABLE) {
5544                                 igb_check_swap_media(adapter);
5545                                 if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5546                                         schedule_work(&adapter->reset_task);
5547                                         /* return immediately */
5548                                         return;
5549                                 }
5550                         }
5551                         pm_schedule_suspend(netdev->dev.parent,
5552                                             MSEC_PER_SEC * 5);
5553
5554                 /* also check for alternate media here */
5555                 } else if (!netif_carrier_ok(netdev) &&
5556                            (adapter->flags & IGB_FLAG_MAS_ENABLE)) {
5557                         igb_check_swap_media(adapter);
5558                         if (adapter->flags & IGB_FLAG_MEDIA_RESET) {
5559                                 schedule_work(&adapter->reset_task);
5560                                 /* return immediately */
5561                                 return;
5562                         }
5563                 }
5564         }
5565
5566         spin_lock(&adapter->stats64_lock);
5567         igb_update_stats(adapter);
5568         spin_unlock(&adapter->stats64_lock);
5569
5570         for (i = 0; i < adapter->num_tx_queues; i++) {
5571                 struct igb_ring *tx_ring = adapter->tx_ring[i];
5572                 if (!netif_carrier_ok(netdev)) {
5573                         /* We've lost link, so the controller stops DMA,
5574                          * but we've got queued Tx work that's never going
5575                          * to get done, so reset controller to flush Tx.
5576                          * (Do the reset outside of interrupt context).
5577                          */
5578                         if (igb_desc_unused(tx_ring) + 1 < tx_ring->count) {
5579                                 adapter->tx_timeout_count++;
5580                                 schedule_work(&adapter->reset_task);
5581                                 /* return immediately since reset is imminent */
5582                                 return;
5583                         }
5584                 }
5585
5586                 /* Force detection of hung controller every watchdog period */
5587                 set_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
5588         }
5589
5590         /* Cause software interrupt to ensure Rx ring is cleaned */
5591         if (adapter->flags & IGB_FLAG_HAS_MSIX) {
5592                 u32 eics = 0;
5593
5594                 for (i = 0; i < adapter->num_q_vectors; i++)
5595                         eics |= adapter->q_vector[i]->eims_value;
5596                 wr32(E1000_EICS, eics);
5597         } else {
5598                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
5599         }
5600
5601         igb_spoof_check(adapter);
5602         igb_ptp_rx_hang(adapter);
5603         igb_ptp_tx_hang(adapter);
5604
5605         /* Check LVMMC register on i350/i354 only */
5606         if ((adapter->hw.mac.type == e1000_i350) ||
5607             (adapter->hw.mac.type == e1000_i354))
5608                 igb_check_lvmmc(adapter);
5609
5610         /* Reset the timer */
5611         if (!test_bit(__IGB_DOWN, &adapter->state)) {
5612                 if (adapter->flags & IGB_FLAG_NEED_LINK_UPDATE)
5613                         mod_timer(&adapter->watchdog_timer,
5614                                   round_jiffies(jiffies +  HZ));
5615                 else
5616                         mod_timer(&adapter->watchdog_timer,
5617                                   round_jiffies(jiffies + 2 * HZ));
5618         }
5619 }
5620
5621 enum latency_range {
5622         lowest_latency = 0,
5623         low_latency = 1,
5624         bulk_latency = 2,
5625         latency_invalid = 255
5626 };
5627
5628 /**
5629  *  igb_update_ring_itr - update the dynamic ITR value based on packet size
5630  *  @q_vector: pointer to q_vector
5631  *
5632  *  Stores a new ITR value based on strictly on packet size.  This
5633  *  algorithm is less sophisticated than that used in igb_update_itr,
5634  *  due to the difficulty of synchronizing statistics across multiple
5635  *  receive rings.  The divisors and thresholds used by this function
5636  *  were determined based on theoretical maximum wire speed and testing
5637  *  data, in order to minimize response time while increasing bulk
5638  *  throughput.
5639  *  This functionality is controlled by ethtool's coalescing settings.
5640  *  NOTE:  This function is called only when operating in a multiqueue
5641  *         receive environment.
5642  **/
5643 static void igb_update_ring_itr(struct igb_q_vector *q_vector)
5644 {
5645         int new_val = q_vector->itr_val;
5646         int avg_wire_size = 0;
5647         struct igb_adapter *adapter = q_vector->adapter;
5648         unsigned int packets;
5649
5650         /* For non-gigabit speeds, just fix the interrupt rate at 4000
5651          * ints/sec - ITR timer value of 120 ticks.
5652          */
5653         if (adapter->link_speed != SPEED_1000) {
5654                 new_val = IGB_4K_ITR;
5655                 goto set_itr_val;
5656         }
5657
5658         packets = q_vector->rx.total_packets;
5659         if (packets)
5660                 avg_wire_size = q_vector->rx.total_bytes / packets;
5661
5662         packets = q_vector->tx.total_packets;
5663         if (packets)
5664                 avg_wire_size = max_t(u32, avg_wire_size,
5665                                       q_vector->tx.total_bytes / packets);
5666
5667         /* if avg_wire_size isn't set no work was done */
5668         if (!avg_wire_size)
5669                 goto clear_counts;
5670
5671         /* Add 24 bytes to size to account for CRC, preamble, and gap */
5672         avg_wire_size += 24;
5673
5674         /* Don't starve jumbo frames */
5675         avg_wire_size = min(avg_wire_size, 3000);
5676
5677         /* Give a little boost to mid-size frames */
5678         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
5679                 new_val = avg_wire_size / 3;
5680         else
5681                 new_val = avg_wire_size / 2;
5682
5683         /* conservative mode (itr 3) eliminates the lowest_latency setting */
5684         if (new_val < IGB_20K_ITR &&
5685             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5686              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5687                 new_val = IGB_20K_ITR;
5688
5689 set_itr_val:
5690         if (new_val != q_vector->itr_val) {
5691                 q_vector->itr_val = new_val;
5692                 q_vector->set_itr = 1;
5693         }
5694 clear_counts:
5695         q_vector->rx.total_bytes = 0;
5696         q_vector->rx.total_packets = 0;
5697         q_vector->tx.total_bytes = 0;
5698         q_vector->tx.total_packets = 0;
5699 }
5700
5701 /**
5702  *  igb_update_itr - update the dynamic ITR value based on statistics
5703  *  @q_vector: pointer to q_vector
5704  *  @ring_container: ring info to update the itr for
5705  *
5706  *  Stores a new ITR value based on packets and byte
5707  *  counts during the last interrupt.  The advantage of per interrupt
5708  *  computation is faster updates and more accurate ITR for the current
5709  *  traffic pattern.  Constants in this function were computed
5710  *  based on theoretical maximum wire speed and thresholds were set based
5711  *  on testing data as well as attempting to minimize response time
5712  *  while increasing bulk throughput.
5713  *  This functionality is controlled by ethtool's coalescing settings.
5714  *  NOTE:  These calculations are only valid when operating in a single-
5715  *         queue environment.
5716  **/
5717 static void igb_update_itr(struct igb_q_vector *q_vector,
5718                            struct igb_ring_container *ring_container)
5719 {
5720         unsigned int packets = ring_container->total_packets;
5721         unsigned int bytes = ring_container->total_bytes;
5722         u8 itrval = ring_container->itr;
5723
5724         /* no packets, exit with status unchanged */
5725         if (packets == 0)
5726                 return;
5727
5728         switch (itrval) {
5729         case lowest_latency:
5730                 /* handle TSO and jumbo frames */
5731                 if (bytes/packets > 8000)
5732                         itrval = bulk_latency;
5733                 else if ((packets < 5) && (bytes > 512))
5734                         itrval = low_latency;
5735                 break;
5736         case low_latency:  /* 50 usec aka 20000 ints/s */
5737                 if (bytes > 10000) {
5738                         /* this if handles the TSO accounting */
5739                         if (bytes/packets > 8000)
5740                                 itrval = bulk_latency;
5741                         else if ((packets < 10) || ((bytes/packets) > 1200))
5742                                 itrval = bulk_latency;
5743                         else if ((packets > 35))
5744                                 itrval = lowest_latency;
5745                 } else if (bytes/packets > 2000) {
5746                         itrval = bulk_latency;
5747                 } else if (packets <= 2 && bytes < 512) {
5748                         itrval = lowest_latency;
5749                 }
5750                 break;
5751         case bulk_latency: /* 250 usec aka 4000 ints/s */
5752                 if (bytes > 25000) {
5753                         if (packets > 35)
5754                                 itrval = low_latency;
5755                 } else if (bytes < 1500) {
5756                         itrval = low_latency;
5757                 }
5758                 break;
5759         }
5760
5761         /* clear work counters since we have the values we need */
5762         ring_container->total_bytes = 0;
5763         ring_container->total_packets = 0;
5764
5765         /* write updated itr to ring container */
5766         ring_container->itr = itrval;
5767 }
5768
5769 static void igb_set_itr(struct igb_q_vector *q_vector)
5770 {
5771         struct igb_adapter *adapter = q_vector->adapter;
5772         u32 new_itr = q_vector->itr_val;
5773         u8 current_itr = 0;
5774
5775         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
5776         if (adapter->link_speed != SPEED_1000) {
5777                 current_itr = 0;
5778                 new_itr = IGB_4K_ITR;
5779                 goto set_itr_now;
5780         }
5781
5782         igb_update_itr(q_vector, &q_vector->tx);
5783         igb_update_itr(q_vector, &q_vector->rx);
5784
5785         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
5786
5787         /* conservative mode (itr 3) eliminates the lowest_latency setting */
5788         if (current_itr == lowest_latency &&
5789             ((q_vector->rx.ring && adapter->rx_itr_setting == 3) ||
5790              (!q_vector->rx.ring && adapter->tx_itr_setting == 3)))
5791                 current_itr = low_latency;
5792
5793         switch (current_itr) {
5794         /* counts and packets in update_itr are dependent on these numbers */
5795         case lowest_latency:
5796                 new_itr = IGB_70K_ITR; /* 70,000 ints/sec */
5797                 break;
5798         case low_latency:
5799                 new_itr = IGB_20K_ITR; /* 20,000 ints/sec */
5800                 break;
5801         case bulk_latency:
5802                 new_itr = IGB_4K_ITR;  /* 4,000 ints/sec */
5803                 break;
5804         default:
5805                 break;
5806         }
5807
5808 set_itr_now:
5809         if (new_itr != q_vector->itr_val) {
5810                 /* this attempts to bias the interrupt rate towards Bulk
5811                  * by adding intermediate steps when interrupt rate is
5812                  * increasing
5813                  */
5814                 new_itr = new_itr > q_vector->itr_val ?
5815                           max((new_itr * q_vector->itr_val) /
5816                           (new_itr + (q_vector->itr_val >> 2)),
5817                           new_itr) : new_itr;
5818                 /* Don't write the value here; it resets the adapter's
5819                  * internal timer, and causes us to delay far longer than
5820                  * we should between interrupts.  Instead, we write the ITR
5821                  * value at the beginning of the next interrupt so the timing
5822                  * ends up being correct.
5823                  */
5824                 q_vector->itr_val = new_itr;
5825                 q_vector->set_itr = 1;
5826         }
5827 }
5828
5829 static void igb_tx_ctxtdesc(struct igb_ring *tx_ring,
5830                             struct igb_tx_buffer *first,
5831                             u32 vlan_macip_lens, u32 type_tucmd,
5832                             u32 mss_l4len_idx)
5833 {
5834         struct e1000_adv_tx_context_desc *context_desc;
5835         u16 i = tx_ring->next_to_use;
5836         struct timespec64 ts;
5837
5838         context_desc = IGB_TX_CTXTDESC(tx_ring, i);
5839
5840         i++;
5841         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
5842
5843         /* set bits to identify this as an advanced context descriptor */
5844         type_tucmd |= E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT;
5845
5846         /* For 82575, context index must be unique per ring. */
5847         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
5848                 mss_l4len_idx |= tx_ring->reg_idx << 4;
5849
5850         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
5851         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
5852         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
5853
5854         /* We assume there is always a valid tx time available. Invalid times
5855          * should have been handled by the upper layers.
5856          */
5857         if (tx_ring->launchtime_enable) {
5858                 ts = ktime_to_timespec64(first->skb->tstamp);
5859                 skb_txtime_consumed(first->skb);
5860                 context_desc->seqnum_seed = cpu_to_le32(ts.tv_nsec / 32);
5861         } else {
5862                 context_desc->seqnum_seed = 0;
5863         }
5864 }
5865
5866 static int igb_tso(struct igb_ring *tx_ring,
5867                    struct igb_tx_buffer *first,
5868                    u8 *hdr_len)
5869 {
5870         u32 vlan_macip_lens, type_tucmd, mss_l4len_idx;
5871         struct sk_buff *skb = first->skb;
5872         union {
5873                 struct iphdr *v4;
5874                 struct ipv6hdr *v6;
5875                 unsigned char *hdr;
5876         } ip;
5877         union {
5878                 struct tcphdr *tcp;
5879                 struct udphdr *udp;
5880                 unsigned char *hdr;
5881         } l4;
5882         u32 paylen, l4_offset;
5883         int err;
5884
5885         if (skb->ip_summed != CHECKSUM_PARTIAL)
5886                 return 0;
5887
5888         if (!skb_is_gso(skb))
5889                 return 0;
5890
5891         err = skb_cow_head(skb, 0);
5892         if (err < 0)
5893                 return err;
5894
5895         ip.hdr = skb_network_header(skb);
5896         l4.hdr = skb_checksum_start(skb);
5897
5898         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
5899         type_tucmd = (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) ?
5900                       E1000_ADVTXD_TUCMD_L4T_UDP : E1000_ADVTXD_TUCMD_L4T_TCP;
5901
5902         /* initialize outer IP header fields */
5903         if (ip.v4->version == 4) {
5904                 unsigned char *csum_start = skb_checksum_start(skb);
5905                 unsigned char *trans_start = ip.hdr + (ip.v4->ihl * 4);
5906
5907                 /* IP header will have to cancel out any data that
5908                  * is not a part of the outer IP header
5909                  */
5910                 ip.v4->check = csum_fold(csum_partial(trans_start,
5911                                                       csum_start - trans_start,
5912                                                       0));
5913                 type_tucmd |= E1000_ADVTXD_TUCMD_IPV4;
5914
5915                 ip.v4->tot_len = 0;
5916                 first->tx_flags |= IGB_TX_FLAGS_TSO |
5917                                    IGB_TX_FLAGS_CSUM |
5918                                    IGB_TX_FLAGS_IPV4;
5919         } else {
5920                 ip.v6->payload_len = 0;
5921                 first->tx_flags |= IGB_TX_FLAGS_TSO |
5922                                    IGB_TX_FLAGS_CSUM;
5923         }
5924
5925         /* determine offset of inner transport header */
5926         l4_offset = l4.hdr - skb->data;
5927
5928         /* remove payload length from inner checksum */
5929         paylen = skb->len - l4_offset;
5930         if (type_tucmd & E1000_ADVTXD_TUCMD_L4T_TCP) {
5931                 /* compute length of segmentation header */
5932                 *hdr_len = (l4.tcp->doff * 4) + l4_offset;
5933                 csum_replace_by_diff(&l4.tcp->check,
5934                         (__force __wsum)htonl(paylen));
5935         } else {
5936                 /* compute length of segmentation header */
5937                 *hdr_len = sizeof(*l4.udp) + l4_offset;
5938                 csum_replace_by_diff(&l4.udp->check,
5939                                      (__force __wsum)htonl(paylen));
5940         }
5941
5942         /* update gso size and bytecount with header size */
5943         first->gso_segs = skb_shinfo(skb)->gso_segs;
5944         first->bytecount += (first->gso_segs - 1) * *hdr_len;
5945
5946         /* MSS L4LEN IDX */
5947         mss_l4len_idx = (*hdr_len - l4_offset) << E1000_ADVTXD_L4LEN_SHIFT;
5948         mss_l4len_idx |= skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT;
5949
5950         /* VLAN MACLEN IPLEN */
5951         vlan_macip_lens = l4.hdr - ip.hdr;
5952         vlan_macip_lens |= (ip.hdr - skb->data) << E1000_ADVTXD_MACLEN_SHIFT;
5953         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
5954
5955         igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens,
5956                         type_tucmd, mss_l4len_idx);
5957
5958         return 1;
5959 }
5960
5961 static void igb_tx_csum(struct igb_ring *tx_ring, struct igb_tx_buffer *first)
5962 {
5963         struct sk_buff *skb = first->skb;
5964         u32 vlan_macip_lens = 0;
5965         u32 type_tucmd = 0;
5966
5967         if (skb->ip_summed != CHECKSUM_PARTIAL) {
5968 csum_failed:
5969                 if (!(first->tx_flags & IGB_TX_FLAGS_VLAN) &&
5970                     !tx_ring->launchtime_enable)
5971                         return;
5972                 goto no_csum;
5973         }
5974
5975         switch (skb->csum_offset) {
5976         case offsetof(struct tcphdr, check):
5977                 type_tucmd = E1000_ADVTXD_TUCMD_L4T_TCP;
5978                 fallthrough;
5979         case offsetof(struct udphdr, check):
5980                 break;
5981         case offsetof(struct sctphdr, checksum):
5982                 /* validate that this is actually an SCTP request */
5983                 if (skb_csum_is_sctp(skb)) {
5984                         type_tucmd = E1000_ADVTXD_TUCMD_L4T_SCTP;
5985                         break;
5986                 }
5987                 fallthrough;
5988         default:
5989                 skb_checksum_help(skb);
5990                 goto csum_failed;
5991         }
5992
5993         /* update TX checksum flag */
5994         first->tx_flags |= IGB_TX_FLAGS_CSUM;
5995         vlan_macip_lens = skb_checksum_start_offset(skb) -
5996                           skb_network_offset(skb);
5997 no_csum:
5998         vlan_macip_lens |= skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT;
5999         vlan_macip_lens |= first->tx_flags & IGB_TX_FLAGS_VLAN_MASK;
6000
6001         igb_tx_ctxtdesc(tx_ring, first, vlan_macip_lens, type_tucmd, 0);
6002 }
6003
6004 #define IGB_SET_FLAG(_input, _flag, _result) \
6005         ((_flag <= _result) ? \
6006          ((u32)(_input & _flag) * (_result / _flag)) : \
6007          ((u32)(_input & _flag) / (_flag / _result)))
6008
6009 static u32 igb_tx_cmd_type(struct sk_buff *skb, u32 tx_flags)
6010 {
6011         /* set type for advanced descriptor with frame checksum insertion */
6012         u32 cmd_type = E1000_ADVTXD_DTYP_DATA |
6013                        E1000_ADVTXD_DCMD_DEXT |
6014                        E1000_ADVTXD_DCMD_IFCS;
6015
6016         /* set HW vlan bit if vlan is present */
6017         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_VLAN,
6018                                  (E1000_ADVTXD_DCMD_VLE));
6019
6020         /* set segmentation bits for TSO */
6021         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSO,
6022                                  (E1000_ADVTXD_DCMD_TSE));
6023
6024         /* set timestamp bit if present */
6025         cmd_type |= IGB_SET_FLAG(tx_flags, IGB_TX_FLAGS_TSTAMP,
6026                                  (E1000_ADVTXD_MAC_TSTAMP));
6027
6028         /* insert frame checksum */
6029         cmd_type ^= IGB_SET_FLAG(skb->no_fcs, 1, E1000_ADVTXD_DCMD_IFCS);
6030
6031         return cmd_type;
6032 }
6033
6034 static void igb_tx_olinfo_status(struct igb_ring *tx_ring,
6035                                  union e1000_adv_tx_desc *tx_desc,
6036                                  u32 tx_flags, unsigned int paylen)
6037 {
6038         u32 olinfo_status = paylen << E1000_ADVTXD_PAYLEN_SHIFT;
6039
6040         /* 82575 requires a unique index per ring */
6041         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6042                 olinfo_status |= tx_ring->reg_idx << 4;
6043
6044         /* insert L4 checksum */
6045         olinfo_status |= IGB_SET_FLAG(tx_flags,
6046                                       IGB_TX_FLAGS_CSUM,
6047                                       (E1000_TXD_POPTS_TXSM << 8));
6048
6049         /* insert IPv4 checksum */
6050         olinfo_status |= IGB_SET_FLAG(tx_flags,
6051                                       IGB_TX_FLAGS_IPV4,
6052                                       (E1000_TXD_POPTS_IXSM << 8));
6053
6054         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6055 }
6056
6057 static int __igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6058 {
6059         struct net_device *netdev = tx_ring->netdev;
6060
6061         netif_stop_subqueue(netdev, tx_ring->queue_index);
6062
6063         /* Herbert's original patch had:
6064          *  smp_mb__after_netif_stop_queue();
6065          * but since that doesn't exist yet, just open code it.
6066          */
6067         smp_mb();
6068
6069         /* We need to check again in a case another CPU has just
6070          * made room available.
6071          */
6072         if (igb_desc_unused(tx_ring) < size)
6073                 return -EBUSY;
6074
6075         /* A reprieve! */
6076         netif_wake_subqueue(netdev, tx_ring->queue_index);
6077
6078         u64_stats_update_begin(&tx_ring->tx_syncp2);
6079         tx_ring->tx_stats.restart_queue2++;
6080         u64_stats_update_end(&tx_ring->tx_syncp2);
6081
6082         return 0;
6083 }
6084
6085 static inline int igb_maybe_stop_tx(struct igb_ring *tx_ring, const u16 size)
6086 {
6087         if (igb_desc_unused(tx_ring) >= size)
6088                 return 0;
6089         return __igb_maybe_stop_tx(tx_ring, size);
6090 }
6091
6092 static int igb_tx_map(struct igb_ring *tx_ring,
6093                       struct igb_tx_buffer *first,
6094                       const u8 hdr_len)
6095 {
6096         struct sk_buff *skb = first->skb;
6097         struct igb_tx_buffer *tx_buffer;
6098         union e1000_adv_tx_desc *tx_desc;
6099         skb_frag_t *frag;
6100         dma_addr_t dma;
6101         unsigned int data_len, size;
6102         u32 tx_flags = first->tx_flags;
6103         u32 cmd_type = igb_tx_cmd_type(skb, tx_flags);
6104         u16 i = tx_ring->next_to_use;
6105
6106         tx_desc = IGB_TX_DESC(tx_ring, i);
6107
6108         igb_tx_olinfo_status(tx_ring, tx_desc, tx_flags, skb->len - hdr_len);
6109
6110         size = skb_headlen(skb);
6111         data_len = skb->data_len;
6112
6113         dma = dma_map_single(tx_ring->dev, skb->data, size, DMA_TO_DEVICE);
6114
6115         tx_buffer = first;
6116
6117         for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
6118                 if (dma_mapping_error(tx_ring->dev, dma))
6119                         goto dma_error;
6120
6121                 /* record length, and DMA address */
6122                 dma_unmap_len_set(tx_buffer, len, size);
6123                 dma_unmap_addr_set(tx_buffer, dma, dma);
6124
6125                 tx_desc->read.buffer_addr = cpu_to_le64(dma);
6126
6127                 while (unlikely(size > IGB_MAX_DATA_PER_TXD)) {
6128                         tx_desc->read.cmd_type_len =
6129                                 cpu_to_le32(cmd_type ^ IGB_MAX_DATA_PER_TXD);
6130
6131                         i++;
6132                         tx_desc++;
6133                         if (i == tx_ring->count) {
6134                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
6135                                 i = 0;
6136                         }
6137                         tx_desc->read.olinfo_status = 0;
6138
6139                         dma += IGB_MAX_DATA_PER_TXD;
6140                         size -= IGB_MAX_DATA_PER_TXD;
6141
6142                         tx_desc->read.buffer_addr = cpu_to_le64(dma);
6143                 }
6144
6145                 if (likely(!data_len))
6146                         break;
6147
6148                 tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type ^ size);
6149
6150                 i++;
6151                 tx_desc++;
6152                 if (i == tx_ring->count) {
6153                         tx_desc = IGB_TX_DESC(tx_ring, 0);
6154                         i = 0;
6155                 }
6156                 tx_desc->read.olinfo_status = 0;
6157
6158                 size = skb_frag_size(frag);
6159                 data_len -= size;
6160
6161                 dma = skb_frag_dma_map(tx_ring->dev, frag, 0,
6162                                        size, DMA_TO_DEVICE);
6163
6164                 tx_buffer = &tx_ring->tx_buffer_info[i];
6165         }
6166
6167         /* write last descriptor with RS and EOP bits */
6168         cmd_type |= size | IGB_TXD_DCMD;
6169         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6170
6171         netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
6172
6173         /* set the timestamp */
6174         first->time_stamp = jiffies;
6175
6176         skb_tx_timestamp(skb);
6177
6178         /* Force memory writes to complete before letting h/w know there
6179          * are new descriptors to fetch.  (Only applicable for weak-ordered
6180          * memory model archs, such as IA-64).
6181          *
6182          * We also need this memory barrier to make certain all of the
6183          * status bits have been updated before next_to_watch is written.
6184          */
6185         dma_wmb();
6186
6187         /* set next_to_watch value indicating a packet is present */
6188         first->next_to_watch = tx_desc;
6189
6190         i++;
6191         if (i == tx_ring->count)
6192                 i = 0;
6193
6194         tx_ring->next_to_use = i;
6195
6196         /* Make sure there is space in the ring for the next send. */
6197         igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6198
6199         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
6200                 writel(i, tx_ring->tail);
6201         }
6202         return 0;
6203
6204 dma_error:
6205         dev_err(tx_ring->dev, "TX DMA map failed\n");
6206         tx_buffer = &tx_ring->tx_buffer_info[i];
6207
6208         /* clear dma mappings for failed tx_buffer_info map */
6209         while (tx_buffer != first) {
6210                 if (dma_unmap_len(tx_buffer, len))
6211                         dma_unmap_page(tx_ring->dev,
6212                                        dma_unmap_addr(tx_buffer, dma),
6213                                        dma_unmap_len(tx_buffer, len),
6214                                        DMA_TO_DEVICE);
6215                 dma_unmap_len_set(tx_buffer, len, 0);
6216
6217                 if (i-- == 0)
6218                         i += tx_ring->count;
6219                 tx_buffer = &tx_ring->tx_buffer_info[i];
6220         }
6221
6222         if (dma_unmap_len(tx_buffer, len))
6223                 dma_unmap_single(tx_ring->dev,
6224                                  dma_unmap_addr(tx_buffer, dma),
6225                                  dma_unmap_len(tx_buffer, len),
6226                                  DMA_TO_DEVICE);
6227         dma_unmap_len_set(tx_buffer, len, 0);
6228
6229         dev_kfree_skb_any(tx_buffer->skb);
6230         tx_buffer->skb = NULL;
6231
6232         tx_ring->next_to_use = i;
6233
6234         return -1;
6235 }
6236
6237 int igb_xmit_xdp_ring(struct igb_adapter *adapter,
6238                       struct igb_ring *tx_ring,
6239                       struct xdp_frame *xdpf)
6240 {
6241         union e1000_adv_tx_desc *tx_desc;
6242         u32 len, cmd_type, olinfo_status;
6243         struct igb_tx_buffer *tx_buffer;
6244         dma_addr_t dma;
6245         u16 i;
6246
6247         len = xdpf->len;
6248
6249         if (unlikely(!igb_desc_unused(tx_ring)))
6250                 return IGB_XDP_CONSUMED;
6251
6252         dma = dma_map_single(tx_ring->dev, xdpf->data, len, DMA_TO_DEVICE);
6253         if (dma_mapping_error(tx_ring->dev, dma))
6254                 return IGB_XDP_CONSUMED;
6255
6256         /* record the location of the first descriptor for this packet */
6257         tx_buffer = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6258         tx_buffer->bytecount = len;
6259         tx_buffer->gso_segs = 1;
6260         tx_buffer->protocol = 0;
6261
6262         i = tx_ring->next_to_use;
6263         tx_desc = IGB_TX_DESC(tx_ring, i);
6264
6265         dma_unmap_len_set(tx_buffer, len, len);
6266         dma_unmap_addr_set(tx_buffer, dma, dma);
6267         tx_buffer->type = IGB_TYPE_XDP;
6268         tx_buffer->xdpf = xdpf;
6269
6270         tx_desc->read.buffer_addr = cpu_to_le64(dma);
6271
6272         /* put descriptor type bits */
6273         cmd_type = E1000_ADVTXD_DTYP_DATA |
6274                    E1000_ADVTXD_DCMD_DEXT |
6275                    E1000_ADVTXD_DCMD_IFCS;
6276         cmd_type |= len | IGB_TXD_DCMD;
6277         tx_desc->read.cmd_type_len = cpu_to_le32(cmd_type);
6278
6279         olinfo_status = len << E1000_ADVTXD_PAYLEN_SHIFT;
6280         /* 82575 requires a unique index per ring */
6281         if (test_bit(IGB_RING_FLAG_TX_CTX_IDX, &tx_ring->flags))
6282                 olinfo_status |= tx_ring->reg_idx << 4;
6283
6284         tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
6285
6286         netdev_tx_sent_queue(txring_txq(tx_ring), tx_buffer->bytecount);
6287
6288         /* set the timestamp */
6289         tx_buffer->time_stamp = jiffies;
6290
6291         /* Avoid any potential race with xdp_xmit and cleanup */
6292         smp_wmb();
6293
6294         /* set next_to_watch value indicating a packet is present */
6295         i++;
6296         if (i == tx_ring->count)
6297                 i = 0;
6298
6299         tx_buffer->next_to_watch = tx_desc;
6300         tx_ring->next_to_use = i;
6301
6302         /* Make sure there is space in the ring for the next send. */
6303         igb_maybe_stop_tx(tx_ring, DESC_NEEDED);
6304
6305         if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more())
6306                 writel(i, tx_ring->tail);
6307
6308         return IGB_XDP_TX;
6309 }
6310
6311 netdev_tx_t igb_xmit_frame_ring(struct sk_buff *skb,
6312                                 struct igb_ring *tx_ring)
6313 {
6314         struct igb_tx_buffer *first;
6315         int tso;
6316         u32 tx_flags = 0;
6317         unsigned short f;
6318         u16 count = TXD_USE_COUNT(skb_headlen(skb));
6319         __be16 protocol = vlan_get_protocol(skb);
6320         u8 hdr_len = 0;
6321
6322         /* need: 1 descriptor per page * PAGE_SIZE/IGB_MAX_DATA_PER_TXD,
6323          *       + 1 desc for skb_headlen/IGB_MAX_DATA_PER_TXD,
6324          *       + 2 desc gap to keep tail from touching head,
6325          *       + 1 desc for context descriptor,
6326          * otherwise try next time
6327          */
6328         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
6329                 count += TXD_USE_COUNT(skb_frag_size(
6330                                                 &skb_shinfo(skb)->frags[f]));
6331
6332         if (igb_maybe_stop_tx(tx_ring, count + 3)) {
6333                 /* this is a hard error */
6334                 return NETDEV_TX_BUSY;
6335         }
6336
6337         /* record the location of the first descriptor for this packet */
6338         first = &tx_ring->tx_buffer_info[tx_ring->next_to_use];
6339         first->type = IGB_TYPE_SKB;
6340         first->skb = skb;
6341         first->bytecount = skb->len;
6342         first->gso_segs = 1;
6343
6344         if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
6345                 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6346
6347                 if (adapter->tstamp_config.tx_type == HWTSTAMP_TX_ON &&
6348                     !test_and_set_bit_lock(__IGB_PTP_TX_IN_PROGRESS,
6349                                            &adapter->state)) {
6350                         skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
6351                         tx_flags |= IGB_TX_FLAGS_TSTAMP;
6352
6353                         adapter->ptp_tx_skb = skb_get(skb);
6354                         adapter->ptp_tx_start = jiffies;
6355                         if (adapter->hw.mac.type == e1000_82576)
6356                                 schedule_work(&adapter->ptp_tx_work);
6357                 } else {
6358                         adapter->tx_hwtstamp_skipped++;
6359                 }
6360         }
6361
6362         if (skb_vlan_tag_present(skb)) {
6363                 tx_flags |= IGB_TX_FLAGS_VLAN;
6364                 tx_flags |= (skb_vlan_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
6365         }
6366
6367         /* record initial flags and protocol */
6368         first->tx_flags = tx_flags;
6369         first->protocol = protocol;
6370
6371         tso = igb_tso(tx_ring, first, &hdr_len);
6372         if (tso < 0)
6373                 goto out_drop;
6374         else if (!tso)
6375                 igb_tx_csum(tx_ring, first);
6376
6377         if (igb_tx_map(tx_ring, first, hdr_len))
6378                 goto cleanup_tx_tstamp;
6379
6380         return NETDEV_TX_OK;
6381
6382 out_drop:
6383         dev_kfree_skb_any(first->skb);
6384         first->skb = NULL;
6385 cleanup_tx_tstamp:
6386         if (unlikely(tx_flags & IGB_TX_FLAGS_TSTAMP)) {
6387                 struct igb_adapter *adapter = netdev_priv(tx_ring->netdev);
6388
6389                 dev_kfree_skb_any(adapter->ptp_tx_skb);
6390                 adapter->ptp_tx_skb = NULL;
6391                 if (adapter->hw.mac.type == e1000_82576)
6392                         cancel_work_sync(&adapter->ptp_tx_work);
6393                 clear_bit_unlock(__IGB_PTP_TX_IN_PROGRESS, &adapter->state);
6394         }
6395
6396         return NETDEV_TX_OK;
6397 }
6398
6399 static inline struct igb_ring *igb_tx_queue_mapping(struct igb_adapter *adapter,
6400                                                     struct sk_buff *skb)
6401 {
6402         unsigned int r_idx = skb->queue_mapping;
6403
6404         if (r_idx >= adapter->num_tx_queues)
6405                 r_idx = r_idx % adapter->num_tx_queues;
6406
6407         return adapter->tx_ring[r_idx];
6408 }
6409
6410 static netdev_tx_t igb_xmit_frame(struct sk_buff *skb,
6411                                   struct net_device *netdev)
6412 {
6413         struct igb_adapter *adapter = netdev_priv(netdev);
6414
6415         /* The minimum packet size with TCTL.PSP set is 17 so pad the skb
6416          * in order to meet this minimum size requirement.
6417          */
6418         if (skb_put_padto(skb, 17))
6419                 return NETDEV_TX_OK;
6420
6421         return igb_xmit_frame_ring(skb, igb_tx_queue_mapping(adapter, skb));
6422 }
6423
6424 /**
6425  *  igb_tx_timeout - Respond to a Tx Hang
6426  *  @netdev: network interface device structure
6427  *  @txqueue: number of the Tx queue that hung (unused)
6428  **/
6429 static void igb_tx_timeout(struct net_device *netdev, unsigned int __always_unused txqueue)
6430 {
6431         struct igb_adapter *adapter = netdev_priv(netdev);
6432         struct e1000_hw *hw = &adapter->hw;
6433
6434         /* Do the reset outside of interrupt context */
6435         adapter->tx_timeout_count++;
6436
6437         if (hw->mac.type >= e1000_82580)
6438                 hw->dev_spec._82575.global_device_reset = true;
6439
6440         schedule_work(&adapter->reset_task);
6441         wr32(E1000_EICS,
6442              (adapter->eims_enable_mask & ~adapter->eims_other));
6443 }
6444
6445 static void igb_reset_task(struct work_struct *work)
6446 {
6447         struct igb_adapter *adapter;
6448         adapter = container_of(work, struct igb_adapter, reset_task);
6449
6450         rtnl_lock();
6451         /* If we're already down or resetting, just bail */
6452         if (test_bit(__IGB_DOWN, &adapter->state) ||
6453             test_bit(__IGB_RESETTING, &adapter->state)) {
6454                 rtnl_unlock();
6455                 return;
6456         }
6457
6458         igb_dump(adapter);
6459         netdev_err(adapter->netdev, "Reset adapter\n");
6460         igb_reinit_locked(adapter);
6461         rtnl_unlock();
6462 }
6463
6464 /**
6465  *  igb_get_stats64 - Get System Network Statistics
6466  *  @netdev: network interface device structure
6467  *  @stats: rtnl_link_stats64 pointer
6468  **/
6469 static void igb_get_stats64(struct net_device *netdev,
6470                             struct rtnl_link_stats64 *stats)
6471 {
6472         struct igb_adapter *adapter = netdev_priv(netdev);
6473
6474         spin_lock(&adapter->stats64_lock);
6475         igb_update_stats(adapter);
6476         memcpy(stats, &adapter->stats64, sizeof(*stats));
6477         spin_unlock(&adapter->stats64_lock);
6478 }
6479
6480 /**
6481  *  igb_change_mtu - Change the Maximum Transfer Unit
6482  *  @netdev: network interface device structure
6483  *  @new_mtu: new value for maximum frame size
6484  *
6485  *  Returns 0 on success, negative on failure
6486  **/
6487 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
6488 {
6489         struct igb_adapter *adapter = netdev_priv(netdev);
6490         int max_frame = new_mtu + IGB_ETH_PKT_HDR_PAD;
6491
6492         if (adapter->xdp_prog) {
6493                 int i;
6494
6495                 for (i = 0; i < adapter->num_rx_queues; i++) {
6496                         struct igb_ring *ring = adapter->rx_ring[i];
6497
6498                         if (max_frame > igb_rx_bufsz(ring)) {
6499                                 netdev_warn(adapter->netdev,
6500                                             "Requested MTU size is not supported with XDP. Max frame size is %d\n",
6501                                             max_frame);
6502                                 return -EINVAL;
6503                         }
6504                 }
6505         }
6506
6507         /* adjust max frame to be at least the size of a standard frame */
6508         if (max_frame < (ETH_FRAME_LEN + ETH_FCS_LEN))
6509                 max_frame = ETH_FRAME_LEN + ETH_FCS_LEN;
6510
6511         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
6512                 usleep_range(1000, 2000);
6513
6514         /* igb_down has a dependency on max_frame_size */
6515         adapter->max_frame_size = max_frame;
6516
6517         if (netif_running(netdev))
6518                 igb_down(adapter);
6519
6520         netdev_dbg(netdev, "changing MTU from %d to %d\n",
6521                    netdev->mtu, new_mtu);
6522         netdev->mtu = new_mtu;
6523
6524         if (netif_running(netdev))
6525                 igb_up(adapter);
6526         else
6527                 igb_reset(adapter);
6528
6529         clear_bit(__IGB_RESETTING, &adapter->state);
6530
6531         return 0;
6532 }
6533
6534 /**
6535  *  igb_update_stats - Update the board statistics counters
6536  *  @adapter: board private structure
6537  **/
6538 void igb_update_stats(struct igb_adapter *adapter)
6539 {
6540         struct rtnl_link_stats64 *net_stats = &adapter->stats64;
6541         struct e1000_hw *hw = &adapter->hw;
6542         struct pci_dev *pdev = adapter->pdev;
6543         u32 reg, mpc;
6544         int i;
6545         u64 bytes, packets;
6546         unsigned int start;
6547         u64 _bytes, _packets;
6548
6549         /* Prevent stats update while adapter is being reset, or if the pci
6550          * connection is down.
6551          */
6552         if (adapter->link_speed == 0)
6553                 return;
6554         if (pci_channel_offline(pdev))
6555                 return;
6556
6557         bytes = 0;
6558         packets = 0;
6559
6560         rcu_read_lock();
6561         for (i = 0; i < adapter->num_rx_queues; i++) {
6562                 struct igb_ring *ring = adapter->rx_ring[i];
6563                 u32 rqdpc = rd32(E1000_RQDPC(i));
6564                 if (hw->mac.type >= e1000_i210)
6565                         wr32(E1000_RQDPC(i), 0);
6566
6567                 if (rqdpc) {
6568                         ring->rx_stats.drops += rqdpc;
6569                         net_stats->rx_fifo_errors += rqdpc;
6570                 }
6571
6572                 do {
6573                         start = u64_stats_fetch_begin_irq(&ring->rx_syncp);
6574                         _bytes = ring->rx_stats.bytes;
6575                         _packets = ring->rx_stats.packets;
6576                 } while (u64_stats_fetch_retry_irq(&ring->rx_syncp, start));
6577                 bytes += _bytes;
6578                 packets += _packets;
6579         }
6580
6581         net_stats->rx_bytes = bytes;
6582         net_stats->rx_packets = packets;
6583
6584         bytes = 0;
6585         packets = 0;
6586         for (i = 0; i < adapter->num_tx_queues; i++) {
6587                 struct igb_ring *ring = adapter->tx_ring[i];
6588                 do {
6589                         start = u64_stats_fetch_begin_irq(&ring->tx_syncp);
6590                         _bytes = ring->tx_stats.bytes;
6591                         _packets = ring->tx_stats.packets;
6592                 } while (u64_stats_fetch_retry_irq(&ring->tx_syncp, start));
6593                 bytes += _bytes;
6594                 packets += _packets;
6595         }
6596         net_stats->tx_bytes = bytes;
6597         net_stats->tx_packets = packets;
6598         rcu_read_unlock();
6599
6600         /* read stats registers */
6601         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
6602         adapter->stats.gprc += rd32(E1000_GPRC);
6603         adapter->stats.gorc += rd32(E1000_GORCL);
6604         rd32(E1000_GORCH); /* clear GORCL */
6605         adapter->stats.bprc += rd32(E1000_BPRC);
6606         adapter->stats.mprc += rd32(E1000_MPRC);
6607         adapter->stats.roc += rd32(E1000_ROC);
6608
6609         adapter->stats.prc64 += rd32(E1000_PRC64);
6610         adapter->stats.prc127 += rd32(E1000_PRC127);
6611         adapter->stats.prc255 += rd32(E1000_PRC255);
6612         adapter->stats.prc511 += rd32(E1000_PRC511);
6613         adapter->stats.prc1023 += rd32(E1000_PRC1023);
6614         adapter->stats.prc1522 += rd32(E1000_PRC1522);
6615         adapter->stats.symerrs += rd32(E1000_SYMERRS);
6616         adapter->stats.sec += rd32(E1000_SEC);
6617
6618         mpc = rd32(E1000_MPC);
6619         adapter->stats.mpc += mpc;
6620         net_stats->rx_fifo_errors += mpc;
6621         adapter->stats.scc += rd32(E1000_SCC);
6622         adapter->stats.ecol += rd32(E1000_ECOL);
6623         adapter->stats.mcc += rd32(E1000_MCC);
6624         adapter->stats.latecol += rd32(E1000_LATECOL);
6625         adapter->stats.dc += rd32(E1000_DC);
6626         adapter->stats.rlec += rd32(E1000_RLEC);
6627         adapter->stats.xonrxc += rd32(E1000_XONRXC);
6628         adapter->stats.xontxc += rd32(E1000_XONTXC);
6629         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
6630         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
6631         adapter->stats.fcruc += rd32(E1000_FCRUC);
6632         adapter->stats.gptc += rd32(E1000_GPTC);
6633         adapter->stats.gotc += rd32(E1000_GOTCL);
6634         rd32(E1000_GOTCH); /* clear GOTCL */
6635         adapter->stats.rnbc += rd32(E1000_RNBC);
6636         adapter->stats.ruc += rd32(E1000_RUC);
6637         adapter->stats.rfc += rd32(E1000_RFC);
6638         adapter->stats.rjc += rd32(E1000_RJC);
6639         adapter->stats.tor += rd32(E1000_TORH);
6640         adapter->stats.tot += rd32(E1000_TOTH);
6641         adapter->stats.tpr += rd32(E1000_TPR);
6642
6643         adapter->stats.ptc64 += rd32(E1000_PTC64);
6644         adapter->stats.ptc127 += rd32(E1000_PTC127);
6645         adapter->stats.ptc255 += rd32(E1000_PTC255);
6646         adapter->stats.ptc511 += rd32(E1000_PTC511);
6647         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
6648         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
6649
6650         adapter->stats.mptc += rd32(E1000_MPTC);
6651         adapter->stats.bptc += rd32(E1000_BPTC);
6652
6653         adapter->stats.tpt += rd32(E1000_TPT);
6654         adapter->stats.colc += rd32(E1000_COLC);
6655
6656         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
6657         /* read internal phy specific stats */
6658         reg = rd32(E1000_CTRL_EXT);
6659         if (!(reg & E1000_CTRL_EXT_LINK_MODE_MASK)) {
6660                 adapter->stats.rxerrc += rd32(E1000_RXERRC);
6661
6662                 /* this stat has invalid values on i210/i211 */
6663                 if ((hw->mac.type != e1000_i210) &&
6664                     (hw->mac.type != e1000_i211))
6665                         adapter->stats.tncrs += rd32(E1000_TNCRS);
6666         }
6667
6668         adapter->stats.tsctc += rd32(E1000_TSCTC);
6669         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
6670
6671         adapter->stats.iac += rd32(E1000_IAC);
6672         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
6673         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
6674         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
6675         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
6676         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
6677         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
6678         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
6679         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
6680
6681         /* Fill out the OS statistics structure */
6682         net_stats->multicast = adapter->stats.mprc;
6683         net_stats->collisions = adapter->stats.colc;
6684
6685         /* Rx Errors */
6686
6687         /* RLEC on some newer hardware can be incorrect so build
6688          * our own version based on RUC and ROC
6689          */
6690         net_stats->rx_errors = adapter->stats.rxerrc +
6691                 adapter->stats.crcerrs + adapter->stats.algnerrc +
6692                 adapter->stats.ruc + adapter->stats.roc +
6693                 adapter->stats.cexterr;
6694         net_stats->rx_length_errors = adapter->stats.ruc +
6695                                       adapter->stats.roc;
6696         net_stats->rx_crc_errors = adapter->stats.crcerrs;
6697         net_stats->rx_frame_errors = adapter->stats.algnerrc;
6698         net_stats->rx_missed_errors = adapter->stats.mpc;
6699
6700         /* Tx Errors */
6701         net_stats->tx_errors = adapter->stats.ecol +
6702                                adapter->stats.latecol;
6703         net_stats->tx_aborted_errors = adapter->stats.ecol;
6704         net_stats->tx_window_errors = adapter->stats.latecol;
6705         net_stats->tx_carrier_errors = adapter->stats.tncrs;
6706
6707         /* Tx Dropped needs to be maintained elsewhere */
6708
6709         /* Management Stats */
6710         adapter->stats.mgptc += rd32(E1000_MGTPTC);
6711         adapter->stats.mgprc += rd32(E1000_MGTPRC);
6712         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
6713
6714         /* OS2BMC Stats */
6715         reg = rd32(E1000_MANC);
6716         if (reg & E1000_MANC_EN_BMC2OS) {
6717                 adapter->stats.o2bgptc += rd32(E1000_O2BGPTC);
6718                 adapter->stats.o2bspc += rd32(E1000_O2BSPC);
6719                 adapter->stats.b2ospc += rd32(E1000_B2OSPC);
6720                 adapter->stats.b2ogprc += rd32(E1000_B2OGPRC);
6721         }
6722 }
6723
6724 static void igb_tsync_interrupt(struct igb_adapter *adapter)
6725 {
6726         struct e1000_hw *hw = &adapter->hw;
6727         struct ptp_clock_event event;
6728         struct timespec64 ts;
6729         u32 ack = 0, tsauxc, sec, nsec, tsicr = rd32(E1000_TSICR);
6730
6731         if (tsicr & TSINTR_SYS_WRAP) {
6732                 event.type = PTP_CLOCK_PPS;
6733                 if (adapter->ptp_caps.pps)
6734                         ptp_clock_event(adapter->ptp_clock, &event);
6735                 ack |= TSINTR_SYS_WRAP;
6736         }
6737
6738         if (tsicr & E1000_TSICR_TXTS) {
6739                 /* retrieve hardware timestamp */
6740                 schedule_work(&adapter->ptp_tx_work);
6741                 ack |= E1000_TSICR_TXTS;
6742         }
6743
6744         if (tsicr & TSINTR_TT0) {
6745                 spin_lock(&adapter->tmreg_lock);
6746                 ts = timespec64_add(adapter->perout[0].start,
6747                                     adapter->perout[0].period);
6748                 /* u32 conversion of tv_sec is safe until y2106 */
6749                 wr32(E1000_TRGTTIML0, ts.tv_nsec);
6750                 wr32(E1000_TRGTTIMH0, (u32)ts.tv_sec);
6751                 tsauxc = rd32(E1000_TSAUXC);
6752                 tsauxc |= TSAUXC_EN_TT0;
6753                 wr32(E1000_TSAUXC, tsauxc);
6754                 adapter->perout[0].start = ts;
6755                 spin_unlock(&adapter->tmreg_lock);
6756                 ack |= TSINTR_TT0;
6757         }
6758
6759         if (tsicr & TSINTR_TT1) {
6760                 spin_lock(&adapter->tmreg_lock);
6761                 ts = timespec64_add(adapter->perout[1].start,
6762                                     adapter->perout[1].period);
6763                 wr32(E1000_TRGTTIML1, ts.tv_nsec);
6764                 wr32(E1000_TRGTTIMH1, (u32)ts.tv_sec);
6765                 tsauxc = rd32(E1000_TSAUXC);
6766                 tsauxc |= TSAUXC_EN_TT1;
6767                 wr32(E1000_TSAUXC, tsauxc);
6768                 adapter->perout[1].start = ts;
6769                 spin_unlock(&adapter->tmreg_lock);
6770                 ack |= TSINTR_TT1;
6771         }
6772
6773         if (tsicr & TSINTR_AUTT0) {
6774                 nsec = rd32(E1000_AUXSTMPL0);
6775                 sec  = rd32(E1000_AUXSTMPH0);
6776                 event.type = PTP_CLOCK_EXTTS;
6777                 event.index = 0;
6778                 event.timestamp = sec * 1000000000ULL + nsec;
6779                 ptp_clock_event(adapter->ptp_clock, &event);
6780                 ack |= TSINTR_AUTT0;
6781         }
6782
6783         if (tsicr & TSINTR_AUTT1) {
6784                 nsec = rd32(E1000_AUXSTMPL1);
6785                 sec  = rd32(E1000_AUXSTMPH1);
6786                 event.type = PTP_CLOCK_EXTTS;
6787                 event.index = 1;
6788                 event.timestamp = sec * 1000000000ULL + nsec;
6789                 ptp_clock_event(adapter->ptp_clock, &event);
6790                 ack |= TSINTR_AUTT1;
6791         }
6792
6793         /* acknowledge the interrupts */
6794         wr32(E1000_TSICR, ack);
6795 }
6796
6797 static irqreturn_t igb_msix_other(int irq, void *data)
6798 {
6799         struct igb_adapter *adapter = data;
6800         struct e1000_hw *hw = &adapter->hw;
6801         u32 icr = rd32(E1000_ICR);
6802         /* reading ICR causes bit 31 of EICR to be cleared */
6803
6804         if (icr & E1000_ICR_DRSTA)
6805                 schedule_work(&adapter->reset_task);
6806
6807         if (icr & E1000_ICR_DOUTSYNC) {
6808                 /* HW is reporting DMA is out of sync */
6809                 adapter->stats.doosync++;
6810                 /* The DMA Out of Sync is also indication of a spoof event
6811                  * in IOV mode. Check the Wrong VM Behavior register to
6812                  * see if it is really a spoof event.
6813                  */
6814                 igb_check_wvbr(adapter);
6815         }
6816
6817         /* Check for a mailbox event */
6818         if (icr & E1000_ICR_VMMB)
6819                 igb_msg_task(adapter);
6820
6821         if (icr & E1000_ICR_LSC) {
6822                 hw->mac.get_link_status = 1;
6823                 /* guard against interrupt when we're going down */
6824                 if (!test_bit(__IGB_DOWN, &adapter->state))
6825                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
6826         }
6827
6828         if (icr & E1000_ICR_TS)
6829                 igb_tsync_interrupt(adapter);
6830
6831         wr32(E1000_EIMS, adapter->eims_other);
6832
6833         return IRQ_HANDLED;
6834 }
6835
6836 static void igb_write_itr(struct igb_q_vector *q_vector)
6837 {
6838         struct igb_adapter *adapter = q_vector->adapter;
6839         u32 itr_val = q_vector->itr_val & 0x7FFC;
6840
6841         if (!q_vector->set_itr)
6842                 return;
6843
6844         if (!itr_val)
6845                 itr_val = 0x4;
6846
6847         if (adapter->hw.mac.type == e1000_82575)
6848                 itr_val |= itr_val << 16;
6849         else
6850                 itr_val |= E1000_EITR_CNT_IGNR;
6851
6852         writel(itr_val, q_vector->itr_register);
6853         q_vector->set_itr = 0;
6854 }
6855
6856 static irqreturn_t igb_msix_ring(int irq, void *data)
6857 {
6858         struct igb_q_vector *q_vector = data;
6859
6860         /* Write the ITR value calculated from the previous interrupt. */
6861         igb_write_itr(q_vector);
6862
6863         napi_schedule(&q_vector->napi);
6864
6865         return IRQ_HANDLED;
6866 }
6867
6868 #ifdef CONFIG_IGB_DCA
6869 static void igb_update_tx_dca(struct igb_adapter *adapter,
6870                               struct igb_ring *tx_ring,
6871                               int cpu)
6872 {
6873         struct e1000_hw *hw = &adapter->hw;
6874         u32 txctrl = dca3_get_tag(tx_ring->dev, cpu);
6875
6876         if (hw->mac.type != e1000_82575)
6877                 txctrl <<= E1000_DCA_TXCTRL_CPUID_SHIFT;
6878
6879         /* We can enable relaxed ordering for reads, but not writes when
6880          * DCA is enabled.  This is due to a known issue in some chipsets
6881          * which will cause the DCA tag to be cleared.
6882          */
6883         txctrl |= E1000_DCA_TXCTRL_DESC_RRO_EN |
6884                   E1000_DCA_TXCTRL_DATA_RRO_EN |
6885                   E1000_DCA_TXCTRL_DESC_DCA_EN;
6886
6887         wr32(E1000_DCA_TXCTRL(tx_ring->reg_idx), txctrl);
6888 }
6889
6890 static void igb_update_rx_dca(struct igb_adapter *adapter,
6891                               struct igb_ring *rx_ring,
6892                               int cpu)
6893 {
6894         struct e1000_hw *hw = &adapter->hw;
6895         u32 rxctrl = dca3_get_tag(&adapter->pdev->dev, cpu);
6896
6897         if (hw->mac.type != e1000_82575)
6898                 rxctrl <<= E1000_DCA_RXCTRL_CPUID_SHIFT;
6899
6900         /* We can enable relaxed ordering for reads, but not writes when
6901          * DCA is enabled.  This is due to a known issue in some chipsets
6902          * which will cause the DCA tag to be cleared.
6903          */
6904         rxctrl |= E1000_DCA_RXCTRL_DESC_RRO_EN |
6905                   E1000_DCA_RXCTRL_DESC_DCA_EN;
6906
6907         wr32(E1000_DCA_RXCTRL(rx_ring->reg_idx), rxctrl);
6908 }
6909
6910 static void igb_update_dca(struct igb_q_vector *q_vector)
6911 {
6912         struct igb_adapter *adapter = q_vector->adapter;
6913         int cpu = get_cpu();
6914
6915         if (q_vector->cpu == cpu)
6916                 goto out_no_update;
6917
6918         if (q_vector->tx.ring)
6919                 igb_update_tx_dca(adapter, q_vector->tx.ring, cpu);
6920
6921         if (q_vector->rx.ring)
6922                 igb_update_rx_dca(adapter, q_vector->rx.ring, cpu);
6923
6924         q_vector->cpu = cpu;
6925 out_no_update:
6926         put_cpu();
6927 }
6928
6929 static void igb_setup_dca(struct igb_adapter *adapter)
6930 {
6931         struct e1000_hw *hw = &adapter->hw;
6932         int i;
6933
6934         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
6935                 return;
6936
6937         /* Always use CB2 mode, difference is masked in the CB driver. */
6938         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
6939
6940         for (i = 0; i < adapter->num_q_vectors; i++) {
6941                 adapter->q_vector[i]->cpu = -1;
6942                 igb_update_dca(adapter->q_vector[i]);
6943         }
6944 }
6945
6946 static int __igb_notify_dca(struct device *dev, void *data)
6947 {
6948         struct net_device *netdev = dev_get_drvdata(dev);
6949         struct igb_adapter *adapter = netdev_priv(netdev);
6950         struct pci_dev *pdev = adapter->pdev;
6951         struct e1000_hw *hw = &adapter->hw;
6952         unsigned long event = *(unsigned long *)data;
6953
6954         switch (event) {
6955         case DCA_PROVIDER_ADD:
6956                 /* if already enabled, don't do it again */
6957                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
6958                         break;
6959                 if (dca_add_requester(dev) == 0) {
6960                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
6961                         dev_info(&pdev->dev, "DCA enabled\n");
6962                         igb_setup_dca(adapter);
6963                         break;
6964                 }
6965                 fallthrough; /* since DCA is disabled. */
6966         case DCA_PROVIDER_REMOVE:
6967                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
6968                         /* without this a class_device is left
6969                          * hanging around in the sysfs model
6970                          */
6971                         dca_remove_requester(dev);
6972                         dev_info(&pdev->dev, "DCA disabled\n");
6973                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
6974                         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
6975                 }
6976                 break;
6977         }
6978
6979         return 0;
6980 }
6981
6982 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
6983                           void *p)
6984 {
6985         int ret_val;
6986
6987         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
6988                                          __igb_notify_dca);
6989
6990         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
6991 }
6992 #endif /* CONFIG_IGB_DCA */
6993
6994 #ifdef CONFIG_PCI_IOV
6995 static int igb_vf_configure(struct igb_adapter *adapter, int vf)
6996 {
6997         unsigned char mac_addr[ETH_ALEN];
6998
6999         eth_zero_addr(mac_addr);
7000         igb_set_vf_mac(adapter, vf, mac_addr);
7001
7002         /* By default spoof check is enabled for all VFs */
7003         adapter->vf_data[vf].spoofchk_enabled = true;
7004
7005         /* By default VFs are not trusted */
7006         adapter->vf_data[vf].trusted = false;
7007
7008         return 0;
7009 }
7010
7011 #endif
7012 static void igb_ping_all_vfs(struct igb_adapter *adapter)
7013 {
7014         struct e1000_hw *hw = &adapter->hw;
7015         u32 ping;
7016         int i;
7017
7018         for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
7019                 ping = E1000_PF_CONTROL_MSG;
7020                 if (adapter->vf_data[i].flags & IGB_VF_FLAG_CTS)
7021                         ping |= E1000_VT_MSGTYPE_CTS;
7022                 igb_write_mbx(hw, &ping, 1, i);
7023         }
7024 }
7025
7026 static int igb_set_vf_promisc(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7027 {
7028         struct e1000_hw *hw = &adapter->hw;
7029         u32 vmolr = rd32(E1000_VMOLR(vf));
7030         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7031
7032         vf_data->flags &= ~(IGB_VF_FLAG_UNI_PROMISC |
7033                             IGB_VF_FLAG_MULTI_PROMISC);
7034         vmolr &= ~(E1000_VMOLR_ROPE | E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7035
7036         if (*msgbuf & E1000_VF_SET_PROMISC_MULTICAST) {
7037                 vmolr |= E1000_VMOLR_MPME;
7038                 vf_data->flags |= IGB_VF_FLAG_MULTI_PROMISC;
7039                 *msgbuf &= ~E1000_VF_SET_PROMISC_MULTICAST;
7040         } else {
7041                 /* if we have hashes and we are clearing a multicast promisc
7042                  * flag we need to write the hashes to the MTA as this step
7043                  * was previously skipped
7044                  */
7045                 if (vf_data->num_vf_mc_hashes > 30) {
7046                         vmolr |= E1000_VMOLR_MPME;
7047                 } else if (vf_data->num_vf_mc_hashes) {
7048                         int j;
7049
7050                         vmolr |= E1000_VMOLR_ROMPE;
7051                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7052                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7053                 }
7054         }
7055
7056         wr32(E1000_VMOLR(vf), vmolr);
7057
7058         /* there are flags left unprocessed, likely not supported */
7059         if (*msgbuf & E1000_VT_MSGINFO_MASK)
7060                 return -EINVAL;
7061
7062         return 0;
7063 }
7064
7065 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
7066                                   u32 *msgbuf, u32 vf)
7067 {
7068         int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7069         u16 *hash_list = (u16 *)&msgbuf[1];
7070         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7071         int i;
7072
7073         /* salt away the number of multicast addresses assigned
7074          * to this VF for later use to restore when the PF multi cast
7075          * list changes
7076          */
7077         vf_data->num_vf_mc_hashes = n;
7078
7079         /* only up to 30 hash values supported */
7080         if (n > 30)
7081                 n = 30;
7082
7083         /* store the hashes for later use */
7084         for (i = 0; i < n; i++)
7085                 vf_data->vf_mc_hashes[i] = hash_list[i];
7086
7087         /* Flush and reset the mta with the new values */
7088         igb_set_rx_mode(adapter->netdev);
7089
7090         return 0;
7091 }
7092
7093 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
7094 {
7095         struct e1000_hw *hw = &adapter->hw;
7096         struct vf_data_storage *vf_data;
7097         int i, j;
7098
7099         for (i = 0; i < adapter->vfs_allocated_count; i++) {
7100                 u32 vmolr = rd32(E1000_VMOLR(i));
7101
7102                 vmolr &= ~(E1000_VMOLR_ROMPE | E1000_VMOLR_MPME);
7103
7104                 vf_data = &adapter->vf_data[i];
7105
7106                 if ((vf_data->num_vf_mc_hashes > 30) ||
7107                     (vf_data->flags & IGB_VF_FLAG_MULTI_PROMISC)) {
7108                         vmolr |= E1000_VMOLR_MPME;
7109                 } else if (vf_data->num_vf_mc_hashes) {
7110                         vmolr |= E1000_VMOLR_ROMPE;
7111                         for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
7112                                 igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
7113                 }
7114                 wr32(E1000_VMOLR(i), vmolr);
7115         }
7116 }
7117
7118 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
7119 {
7120         struct e1000_hw *hw = &adapter->hw;
7121         u32 pool_mask, vlvf_mask, i;
7122
7123         /* create mask for VF and other pools */
7124         pool_mask = E1000_VLVF_POOLSEL_MASK;
7125         vlvf_mask = BIT(E1000_VLVF_POOLSEL_SHIFT + vf);
7126
7127         /* drop PF from pool bits */
7128         pool_mask &= ~BIT(E1000_VLVF_POOLSEL_SHIFT +
7129                              adapter->vfs_allocated_count);
7130
7131         /* Find the vlan filter for this id */
7132         for (i = E1000_VLVF_ARRAY_SIZE; i--;) {
7133                 u32 vlvf = rd32(E1000_VLVF(i));
7134                 u32 vfta_mask, vid, vfta;
7135
7136                 /* remove the vf from the pool */
7137                 if (!(vlvf & vlvf_mask))
7138                         continue;
7139
7140                 /* clear out bit from VLVF */
7141                 vlvf ^= vlvf_mask;
7142
7143                 /* if other pools are present, just remove ourselves */
7144                 if (vlvf & pool_mask)
7145                         goto update_vlvfb;
7146
7147                 /* if PF is present, leave VFTA */
7148                 if (vlvf & E1000_VLVF_POOLSEL_MASK)
7149                         goto update_vlvf;
7150
7151                 vid = vlvf & E1000_VLVF_VLANID_MASK;
7152                 vfta_mask = BIT(vid % 32);
7153
7154                 /* clear bit from VFTA */
7155                 vfta = adapter->shadow_vfta[vid / 32];
7156                 if (vfta & vfta_mask)
7157                         hw->mac.ops.write_vfta(hw, vid / 32, vfta ^ vfta_mask);
7158 update_vlvf:
7159                 /* clear pool selection enable */
7160                 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7161                         vlvf &= E1000_VLVF_POOLSEL_MASK;
7162                 else
7163                         vlvf = 0;
7164 update_vlvfb:
7165                 /* clear pool bits */
7166                 wr32(E1000_VLVF(i), vlvf);
7167         }
7168 }
7169
7170 static int igb_find_vlvf_entry(struct e1000_hw *hw, u32 vlan)
7171 {
7172         u32 vlvf;
7173         int idx;
7174
7175         /* short cut the special case */
7176         if (vlan == 0)
7177                 return 0;
7178
7179         /* Search for the VLAN id in the VLVF entries */
7180         for (idx = E1000_VLVF_ARRAY_SIZE; --idx;) {
7181                 vlvf = rd32(E1000_VLVF(idx));
7182                 if ((vlvf & VLAN_VID_MASK) == vlan)
7183                         break;
7184         }
7185
7186         return idx;
7187 }
7188
7189 static void igb_update_pf_vlvf(struct igb_adapter *adapter, u32 vid)
7190 {
7191         struct e1000_hw *hw = &adapter->hw;
7192         u32 bits, pf_id;
7193         int idx;
7194
7195         idx = igb_find_vlvf_entry(hw, vid);
7196         if (!idx)
7197                 return;
7198
7199         /* See if any other pools are set for this VLAN filter
7200          * entry other than the PF.
7201          */
7202         pf_id = adapter->vfs_allocated_count + E1000_VLVF_POOLSEL_SHIFT;
7203         bits = ~BIT(pf_id) & E1000_VLVF_POOLSEL_MASK;
7204         bits &= rd32(E1000_VLVF(idx));
7205
7206         /* Disable the filter so this falls into the default pool. */
7207         if (!bits) {
7208                 if (adapter->flags & IGB_FLAG_VLAN_PROMISC)
7209                         wr32(E1000_VLVF(idx), BIT(pf_id));
7210                 else
7211                         wr32(E1000_VLVF(idx), 0);
7212         }
7213 }
7214
7215 static s32 igb_set_vf_vlan(struct igb_adapter *adapter, u32 vid,
7216                            bool add, u32 vf)
7217 {
7218         int pf_id = adapter->vfs_allocated_count;
7219         struct e1000_hw *hw = &adapter->hw;
7220         int err;
7221
7222         /* If VLAN overlaps with one the PF is currently monitoring make
7223          * sure that we are able to allocate a VLVF entry.  This may be
7224          * redundant but it guarantees PF will maintain visibility to
7225          * the VLAN.
7226          */
7227         if (add && test_bit(vid, adapter->active_vlans)) {
7228                 err = igb_vfta_set(hw, vid, pf_id, true, false);
7229                 if (err)
7230                         return err;
7231         }
7232
7233         err = igb_vfta_set(hw, vid, vf, add, false);
7234
7235         if (add && !err)
7236                 return err;
7237
7238         /* If we failed to add the VF VLAN or we are removing the VF VLAN
7239          * we may need to drop the PF pool bit in order to allow us to free
7240          * up the VLVF resources.
7241          */
7242         if (test_bit(vid, adapter->active_vlans) ||
7243             (adapter->flags & IGB_FLAG_VLAN_PROMISC))
7244                 igb_update_pf_vlvf(adapter, vid);
7245
7246         return err;
7247 }
7248
7249 static void igb_set_vmvir(struct igb_adapter *adapter, u32 vid, u32 vf)
7250 {
7251         struct e1000_hw *hw = &adapter->hw;
7252
7253         if (vid)
7254                 wr32(E1000_VMVIR(vf), (vid | E1000_VMVIR_VLANA_DEFAULT));
7255         else
7256                 wr32(E1000_VMVIR(vf), 0);
7257 }
7258
7259 static int igb_enable_port_vlan(struct igb_adapter *adapter, int vf,
7260                                 u16 vlan, u8 qos)
7261 {
7262         int err;
7263
7264         err = igb_set_vf_vlan(adapter, vlan, true, vf);
7265         if (err)
7266                 return err;
7267
7268         igb_set_vmvir(adapter, vlan | (qos << VLAN_PRIO_SHIFT), vf);
7269         igb_set_vmolr(adapter, vf, !vlan);
7270
7271         /* revoke access to previous VLAN */
7272         if (vlan != adapter->vf_data[vf].pf_vlan)
7273                 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7274                                 false, vf);
7275
7276         adapter->vf_data[vf].pf_vlan = vlan;
7277         adapter->vf_data[vf].pf_qos = qos;
7278         igb_set_vf_vlan_strip(adapter, vf, true);
7279         dev_info(&adapter->pdev->dev,
7280                  "Setting VLAN %d, QOS 0x%x on VF %d\n", vlan, qos, vf);
7281         if (test_bit(__IGB_DOWN, &adapter->state)) {
7282                 dev_warn(&adapter->pdev->dev,
7283                          "The VF VLAN has been set, but the PF device is not up.\n");
7284                 dev_warn(&adapter->pdev->dev,
7285                          "Bring the PF device up before attempting to use the VF device.\n");
7286         }
7287
7288         return err;
7289 }
7290
7291 static int igb_disable_port_vlan(struct igb_adapter *adapter, int vf)
7292 {
7293         /* Restore tagless access via VLAN 0 */
7294         igb_set_vf_vlan(adapter, 0, true, vf);
7295
7296         igb_set_vmvir(adapter, 0, vf);
7297         igb_set_vmolr(adapter, vf, true);
7298
7299         /* Remove any PF assigned VLAN */
7300         if (adapter->vf_data[vf].pf_vlan)
7301                 igb_set_vf_vlan(adapter, adapter->vf_data[vf].pf_vlan,
7302                                 false, vf);
7303
7304         adapter->vf_data[vf].pf_vlan = 0;
7305         adapter->vf_data[vf].pf_qos = 0;
7306         igb_set_vf_vlan_strip(adapter, vf, false);
7307
7308         return 0;
7309 }
7310
7311 static int igb_ndo_set_vf_vlan(struct net_device *netdev, int vf,
7312                                u16 vlan, u8 qos, __be16 vlan_proto)
7313 {
7314         struct igb_adapter *adapter = netdev_priv(netdev);
7315
7316         if ((vf >= adapter->vfs_allocated_count) || (vlan > 4095) || (qos > 7))
7317                 return -EINVAL;
7318
7319         if (vlan_proto != htons(ETH_P_8021Q))
7320                 return -EPROTONOSUPPORT;
7321
7322         return (vlan || qos) ? igb_enable_port_vlan(adapter, vf, vlan, qos) :
7323                                igb_disable_port_vlan(adapter, vf);
7324 }
7325
7326 static int igb_set_vf_vlan_msg(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
7327 {
7328         int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
7329         int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
7330         int ret;
7331
7332         if (adapter->vf_data[vf].pf_vlan)
7333                 return -1;
7334
7335         /* VLAN 0 is a special case, don't allow it to be removed */
7336         if (!vid && !add)
7337                 return 0;
7338
7339         ret = igb_set_vf_vlan(adapter, vid, !!add, vf);
7340         if (!ret)
7341                 igb_set_vf_vlan_strip(adapter, vf, !!vid);
7342         return ret;
7343 }
7344
7345 static inline void igb_vf_reset(struct igb_adapter *adapter, u32 vf)
7346 {
7347         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7348
7349         /* clear flags - except flag that indicates PF has set the MAC */
7350         vf_data->flags &= IGB_VF_FLAG_PF_SET_MAC;
7351         vf_data->last_nack = jiffies;
7352
7353         /* reset vlans for device */
7354         igb_clear_vf_vfta(adapter, vf);
7355         igb_set_vf_vlan(adapter, vf_data->pf_vlan, true, vf);
7356         igb_set_vmvir(adapter, vf_data->pf_vlan |
7357                                (vf_data->pf_qos << VLAN_PRIO_SHIFT), vf);
7358         igb_set_vmolr(adapter, vf, !vf_data->pf_vlan);
7359         igb_set_vf_vlan_strip(adapter, vf, !!(vf_data->pf_vlan));
7360
7361         /* reset multicast table array for vf */
7362         adapter->vf_data[vf].num_vf_mc_hashes = 0;
7363
7364         /* Flush and reset the mta with the new values */
7365         igb_set_rx_mode(adapter->netdev);
7366 }
7367
7368 static void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
7369 {
7370         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7371
7372         /* clear mac address as we were hotplug removed/added */
7373         if (!(adapter->vf_data[vf].flags & IGB_VF_FLAG_PF_SET_MAC))
7374                 eth_zero_addr(vf_mac);
7375
7376         /* process remaining reset events */
7377         igb_vf_reset(adapter, vf);
7378 }
7379
7380 static void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
7381 {
7382         struct e1000_hw *hw = &adapter->hw;
7383         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
7384         u32 reg, msgbuf[3];
7385         u8 *addr = (u8 *)(&msgbuf[1]);
7386
7387         /* process all the same items cleared in a function level reset */
7388         igb_vf_reset(adapter, vf);
7389
7390         /* set vf mac address */
7391         igb_set_vf_mac(adapter, vf, vf_mac);
7392
7393         /* enable transmit and receive for vf */
7394         reg = rd32(E1000_VFTE);
7395         wr32(E1000_VFTE, reg | BIT(vf));
7396         reg = rd32(E1000_VFRE);
7397         wr32(E1000_VFRE, reg | BIT(vf));
7398
7399         adapter->vf_data[vf].flags |= IGB_VF_FLAG_CTS;
7400
7401         /* reply to reset with ack and vf mac address */
7402         if (!is_zero_ether_addr(vf_mac)) {
7403                 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
7404                 memcpy(addr, vf_mac, ETH_ALEN);
7405         } else {
7406                 msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_NACK;
7407         }
7408         igb_write_mbx(hw, msgbuf, 3, vf);
7409 }
7410
7411 static void igb_flush_mac_table(struct igb_adapter *adapter)
7412 {
7413         struct e1000_hw *hw = &adapter->hw;
7414         int i;
7415
7416         for (i = 0; i < hw->mac.rar_entry_count; i++) {
7417                 adapter->mac_table[i].state &= ~IGB_MAC_STATE_IN_USE;
7418                 eth_zero_addr(adapter->mac_table[i].addr);
7419                 adapter->mac_table[i].queue = 0;
7420                 igb_rar_set_index(adapter, i);
7421         }
7422 }
7423
7424 static int igb_available_rars(struct igb_adapter *adapter, u8 queue)
7425 {
7426         struct e1000_hw *hw = &adapter->hw;
7427         /* do not count rar entries reserved for VFs MAC addresses */
7428         int rar_entries = hw->mac.rar_entry_count -
7429                           adapter->vfs_allocated_count;
7430         int i, count = 0;
7431
7432         for (i = 0; i < rar_entries; i++) {
7433                 /* do not count default entries */
7434                 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT)
7435                         continue;
7436
7437                 /* do not count "in use" entries for different queues */
7438                 if ((adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE) &&
7439                     (adapter->mac_table[i].queue != queue))
7440                         continue;
7441
7442                 count++;
7443         }
7444
7445         return count;
7446 }
7447
7448 /* Set default MAC address for the PF in the first RAR entry */
7449 static void igb_set_default_mac_filter(struct igb_adapter *adapter)
7450 {
7451         struct igb_mac_addr *mac_table = &adapter->mac_table[0];
7452
7453         ether_addr_copy(mac_table->addr, adapter->hw.mac.addr);
7454         mac_table->queue = adapter->vfs_allocated_count;
7455         mac_table->state = IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7456
7457         igb_rar_set_index(adapter, 0);
7458 }
7459
7460 /* If the filter to be added and an already existing filter express
7461  * the same address and address type, it should be possible to only
7462  * override the other configurations, for example the queue to steer
7463  * traffic.
7464  */
7465 static bool igb_mac_entry_can_be_used(const struct igb_mac_addr *entry,
7466                                       const u8 *addr, const u8 flags)
7467 {
7468         if (!(entry->state & IGB_MAC_STATE_IN_USE))
7469                 return true;
7470
7471         if ((entry->state & IGB_MAC_STATE_SRC_ADDR) !=
7472             (flags & IGB_MAC_STATE_SRC_ADDR))
7473                 return false;
7474
7475         if (!ether_addr_equal(addr, entry->addr))
7476                 return false;
7477
7478         return true;
7479 }
7480
7481 /* Add a MAC filter for 'addr' directing matching traffic to 'queue',
7482  * 'flags' is used to indicate what kind of match is made, match is by
7483  * default for the destination address, if matching by source address
7484  * is desired the flag IGB_MAC_STATE_SRC_ADDR can be used.
7485  */
7486 static int igb_add_mac_filter_flags(struct igb_adapter *adapter,
7487                                     const u8 *addr, const u8 queue,
7488                                     const u8 flags)
7489 {
7490         struct e1000_hw *hw = &adapter->hw;
7491         int rar_entries = hw->mac.rar_entry_count -
7492                           adapter->vfs_allocated_count;
7493         int i;
7494
7495         if (is_zero_ether_addr(addr))
7496                 return -EINVAL;
7497
7498         /* Search for the first empty entry in the MAC table.
7499          * Do not touch entries at the end of the table reserved for the VF MAC
7500          * addresses.
7501          */
7502         for (i = 0; i < rar_entries; i++) {
7503                 if (!igb_mac_entry_can_be_used(&adapter->mac_table[i],
7504                                                addr, flags))
7505                         continue;
7506
7507                 ether_addr_copy(adapter->mac_table[i].addr, addr);
7508                 adapter->mac_table[i].queue = queue;
7509                 adapter->mac_table[i].state |= IGB_MAC_STATE_IN_USE | flags;
7510
7511                 igb_rar_set_index(adapter, i);
7512                 return i;
7513         }
7514
7515         return -ENOSPC;
7516 }
7517
7518 static int igb_add_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7519                               const u8 queue)
7520 {
7521         return igb_add_mac_filter_flags(adapter, addr, queue, 0);
7522 }
7523
7524 /* Remove a MAC filter for 'addr' directing matching traffic to
7525  * 'queue', 'flags' is used to indicate what kind of match need to be
7526  * removed, match is by default for the destination address, if
7527  * matching by source address is to be removed the flag
7528  * IGB_MAC_STATE_SRC_ADDR can be used.
7529  */
7530 static int igb_del_mac_filter_flags(struct igb_adapter *adapter,
7531                                     const u8 *addr, const u8 queue,
7532                                     const u8 flags)
7533 {
7534         struct e1000_hw *hw = &adapter->hw;
7535         int rar_entries = hw->mac.rar_entry_count -
7536                           adapter->vfs_allocated_count;
7537         int i;
7538
7539         if (is_zero_ether_addr(addr))
7540                 return -EINVAL;
7541
7542         /* Search for matching entry in the MAC table based on given address
7543          * and queue. Do not touch entries at the end of the table reserved
7544          * for the VF MAC addresses.
7545          */
7546         for (i = 0; i < rar_entries; i++) {
7547                 if (!(adapter->mac_table[i].state & IGB_MAC_STATE_IN_USE))
7548                         continue;
7549                 if ((adapter->mac_table[i].state & flags) != flags)
7550                         continue;
7551                 if (adapter->mac_table[i].queue != queue)
7552                         continue;
7553                 if (!ether_addr_equal(adapter->mac_table[i].addr, addr))
7554                         continue;
7555
7556                 /* When a filter for the default address is "deleted",
7557                  * we return it to its initial configuration
7558                  */
7559                 if (adapter->mac_table[i].state & IGB_MAC_STATE_DEFAULT) {
7560                         adapter->mac_table[i].state =
7561                                 IGB_MAC_STATE_DEFAULT | IGB_MAC_STATE_IN_USE;
7562                         adapter->mac_table[i].queue =
7563                                 adapter->vfs_allocated_count;
7564                 } else {
7565                         adapter->mac_table[i].state = 0;
7566                         adapter->mac_table[i].queue = 0;
7567                         eth_zero_addr(adapter->mac_table[i].addr);
7568                 }
7569
7570                 igb_rar_set_index(adapter, i);
7571                 return 0;
7572         }
7573
7574         return -ENOENT;
7575 }
7576
7577 static int igb_del_mac_filter(struct igb_adapter *adapter, const u8 *addr,
7578                               const u8 queue)
7579 {
7580         return igb_del_mac_filter_flags(adapter, addr, queue, 0);
7581 }
7582
7583 int igb_add_mac_steering_filter(struct igb_adapter *adapter,
7584                                 const u8 *addr, u8 queue, u8 flags)
7585 {
7586         struct e1000_hw *hw = &adapter->hw;
7587
7588         /* In theory, this should be supported on 82575 as well, but
7589          * that part wasn't easily accessible during development.
7590          */
7591         if (hw->mac.type != e1000_i210)
7592                 return -EOPNOTSUPP;
7593
7594         return igb_add_mac_filter_flags(adapter, addr, queue,
7595                                         IGB_MAC_STATE_QUEUE_STEERING | flags);
7596 }
7597
7598 int igb_del_mac_steering_filter(struct igb_adapter *adapter,
7599                                 const u8 *addr, u8 queue, u8 flags)
7600 {
7601         return igb_del_mac_filter_flags(adapter, addr, queue,
7602                                         IGB_MAC_STATE_QUEUE_STEERING | flags);
7603 }
7604
7605 static int igb_uc_sync(struct net_device *netdev, const unsigned char *addr)
7606 {
7607         struct igb_adapter *adapter = netdev_priv(netdev);
7608         int ret;
7609
7610         ret = igb_add_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7611
7612         return min_t(int, ret, 0);
7613 }
7614
7615 static int igb_uc_unsync(struct net_device *netdev, const unsigned char *addr)
7616 {
7617         struct igb_adapter *adapter = netdev_priv(netdev);
7618
7619         igb_del_mac_filter(adapter, addr, adapter->vfs_allocated_count);
7620
7621         return 0;
7622 }
7623
7624 static int igb_set_vf_mac_filter(struct igb_adapter *adapter, const int vf,
7625                                  const u32 info, const u8 *addr)
7626 {
7627         struct pci_dev *pdev = adapter->pdev;
7628         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7629         struct list_head *pos;
7630         struct vf_mac_filter *entry = NULL;
7631         int ret = 0;
7632
7633         switch (info) {
7634         case E1000_VF_MAC_FILTER_CLR:
7635                 /* remove all unicast MAC filters related to the current VF */
7636                 list_for_each(pos, &adapter->vf_macs.l) {
7637                         entry = list_entry(pos, struct vf_mac_filter, l);
7638                         if (entry->vf == vf) {
7639                                 entry->vf = -1;
7640                                 entry->free = true;
7641                                 igb_del_mac_filter(adapter, entry->vf_mac, vf);
7642                         }
7643                 }
7644                 break;
7645         case E1000_VF_MAC_FILTER_ADD:
7646                 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7647                     !vf_data->trusted) {
7648                         dev_warn(&pdev->dev,
7649                                  "VF %d requested MAC filter but is administratively denied\n",
7650                                  vf);
7651                         return -EINVAL;
7652                 }
7653                 if (!is_valid_ether_addr(addr)) {
7654                         dev_warn(&pdev->dev,
7655                                  "VF %d attempted to set invalid MAC filter\n",
7656                                  vf);
7657                         return -EINVAL;
7658                 }
7659
7660                 /* try to find empty slot in the list */
7661                 list_for_each(pos, &adapter->vf_macs.l) {
7662                         entry = list_entry(pos, struct vf_mac_filter, l);
7663                         if (entry->free)
7664                                 break;
7665                 }
7666
7667                 if (entry && entry->free) {
7668                         entry->free = false;
7669                         entry->vf = vf;
7670                         ether_addr_copy(entry->vf_mac, addr);
7671
7672                         ret = igb_add_mac_filter(adapter, addr, vf);
7673                         ret = min_t(int, ret, 0);
7674                 } else {
7675                         ret = -ENOSPC;
7676                 }
7677
7678                 if (ret == -ENOSPC)
7679                         dev_warn(&pdev->dev,
7680                                  "VF %d has requested MAC filter but there is no space for it\n",
7681                                  vf);
7682                 break;
7683         default:
7684                 ret = -EINVAL;
7685                 break;
7686         }
7687
7688         return ret;
7689 }
7690
7691 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
7692 {
7693         struct pci_dev *pdev = adapter->pdev;
7694         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7695         u32 info = msg[0] & E1000_VT_MSGINFO_MASK;
7696
7697         /* The VF MAC Address is stored in a packed array of bytes
7698          * starting at the second 32 bit word of the msg array
7699          */
7700         unsigned char *addr = (unsigned char *)&msg[1];
7701         int ret = 0;
7702
7703         if (!info) {
7704                 if ((vf_data->flags & IGB_VF_FLAG_PF_SET_MAC) &&
7705                     !vf_data->trusted) {
7706                         dev_warn(&pdev->dev,
7707                                  "VF %d attempted to override administratively set MAC address\nReload the VF driver to resume operations\n",
7708                                  vf);
7709                         return -EINVAL;
7710                 }
7711
7712                 if (!is_valid_ether_addr(addr)) {
7713                         dev_warn(&pdev->dev,
7714                                  "VF %d attempted to set invalid MAC\n",
7715                                  vf);
7716                         return -EINVAL;
7717                 }
7718
7719                 ret = igb_set_vf_mac(adapter, vf, addr);
7720         } else {
7721                 ret = igb_set_vf_mac_filter(adapter, vf, info, addr);
7722         }
7723
7724         return ret;
7725 }
7726
7727 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
7728 {
7729         struct e1000_hw *hw = &adapter->hw;
7730         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7731         u32 msg = E1000_VT_MSGTYPE_NACK;
7732
7733         /* if device isn't clear to send it shouldn't be reading either */
7734         if (!(vf_data->flags & IGB_VF_FLAG_CTS) &&
7735             time_after(jiffies, vf_data->last_nack + (2 * HZ))) {
7736                 igb_write_mbx(hw, &msg, 1, vf);
7737                 vf_data->last_nack = jiffies;
7738         }
7739 }
7740
7741 static void igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
7742 {
7743         struct pci_dev *pdev = adapter->pdev;
7744         u32 msgbuf[E1000_VFMAILBOX_SIZE];
7745         struct e1000_hw *hw = &adapter->hw;
7746         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
7747         s32 retval;
7748
7749         retval = igb_read_mbx(hw, msgbuf, E1000_VFMAILBOX_SIZE, vf, false);
7750
7751         if (retval) {
7752                 /* if receive failed revoke VF CTS stats and restart init */
7753                 dev_err(&pdev->dev, "Error receiving message from VF\n");
7754                 vf_data->flags &= ~IGB_VF_FLAG_CTS;
7755                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7756                         goto unlock;
7757                 goto out;
7758         }
7759
7760         /* this is a message we already processed, do nothing */
7761         if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
7762                 goto unlock;
7763
7764         /* until the vf completes a reset it should not be
7765          * allowed to start any configuration.
7766          */
7767         if (msgbuf[0] == E1000_VF_RESET) {
7768                 /* unlocks mailbox */
7769                 igb_vf_reset_msg(adapter, vf);
7770                 return;
7771         }
7772
7773         if (!(vf_data->flags & IGB_VF_FLAG_CTS)) {
7774                 if (!time_after(jiffies, vf_data->last_nack + (2 * HZ)))
7775                         goto unlock;
7776                 retval = -1;
7777                 goto out;
7778         }
7779
7780         switch ((msgbuf[0] & 0xFFFF)) {
7781         case E1000_VF_SET_MAC_ADDR:
7782                 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
7783                 break;
7784         case E1000_VF_SET_PROMISC:
7785                 retval = igb_set_vf_promisc(adapter, msgbuf, vf);
7786                 break;
7787         case E1000_VF_SET_MULTICAST:
7788                 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
7789                 break;
7790         case E1000_VF_SET_LPE:
7791                 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
7792                 break;
7793         case E1000_VF_SET_VLAN:
7794                 retval = -1;
7795                 if (vf_data->pf_vlan)
7796                         dev_warn(&pdev->dev,
7797                                  "VF %d attempted to override administratively set VLAN tag\nReload the VF driver to resume operations\n",
7798                                  vf);
7799                 else
7800                         retval = igb_set_vf_vlan_msg(adapter, msgbuf, vf);
7801                 break;
7802         default:
7803                 dev_err(&pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
7804                 retval = -1;
7805                 break;
7806         }
7807
7808         msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
7809 out:
7810         /* notify the VF of the results of what it sent us */
7811         if (retval)
7812                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
7813         else
7814                 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
7815
7816         /* unlocks mailbox */
7817         igb_write_mbx(hw, msgbuf, 1, vf);
7818         return;
7819
7820 unlock:
7821         igb_unlock_mbx(hw, vf);
7822 }
7823
7824 static void igb_msg_task(struct igb_adapter *adapter)
7825 {
7826         struct e1000_hw *hw = &adapter->hw;
7827         u32 vf;
7828
7829         for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
7830                 /* process any reset requests */
7831                 if (!igb_check_for_rst(hw, vf))
7832                         igb_vf_reset_event(adapter, vf);
7833
7834                 /* process any messages pending */
7835                 if (!igb_check_for_msg(hw, vf))
7836                         igb_rcv_msg_from_vf(adapter, vf);
7837
7838                 /* process any acks */
7839                 if (!igb_check_for_ack(hw, vf))
7840                         igb_rcv_ack_from_vf(adapter, vf);
7841         }
7842 }
7843
7844 /**
7845  *  igb_set_uta - Set unicast filter table address
7846  *  @adapter: board private structure
7847  *  @set: boolean indicating if we are setting or clearing bits
7848  *
7849  *  The unicast table address is a register array of 32-bit registers.
7850  *  The table is meant to be used in a way similar to how the MTA is used
7851  *  however due to certain limitations in the hardware it is necessary to
7852  *  set all the hash bits to 1 and use the VMOLR ROPE bit as a promiscuous
7853  *  enable bit to allow vlan tag stripping when promiscuous mode is enabled
7854  **/
7855 static void igb_set_uta(struct igb_adapter *adapter, bool set)
7856 {
7857         struct e1000_hw *hw = &adapter->hw;
7858         u32 uta = set ? ~0 : 0;
7859         int i;
7860
7861         /* we only need to do this if VMDq is enabled */
7862         if (!adapter->vfs_allocated_count)
7863                 return;
7864
7865         for (i = hw->mac.uta_reg_count; i--;)
7866                 array_wr32(E1000_UTA, i, uta);
7867 }
7868
7869 /**
7870  *  igb_intr_msi - Interrupt Handler
7871  *  @irq: interrupt number
7872  *  @data: pointer to a network interface device structure
7873  **/
7874 static irqreturn_t igb_intr_msi(int irq, void *data)
7875 {
7876         struct igb_adapter *adapter = data;
7877         struct igb_q_vector *q_vector = adapter->q_vector[0];
7878         struct e1000_hw *hw = &adapter->hw;
7879         /* read ICR disables interrupts using IAM */
7880         u32 icr = rd32(E1000_ICR);
7881
7882         igb_write_itr(q_vector);
7883
7884         if (icr & E1000_ICR_DRSTA)
7885                 schedule_work(&adapter->reset_task);
7886
7887         if (icr & E1000_ICR_DOUTSYNC) {
7888                 /* HW is reporting DMA is out of sync */
7889                 adapter->stats.doosync++;
7890         }
7891
7892         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7893                 hw->mac.get_link_status = 1;
7894                 if (!test_bit(__IGB_DOWN, &adapter->state))
7895                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
7896         }
7897
7898         if (icr & E1000_ICR_TS)
7899                 igb_tsync_interrupt(adapter);
7900
7901         napi_schedule(&q_vector->napi);
7902
7903         return IRQ_HANDLED;
7904 }
7905
7906 /**
7907  *  igb_intr - Legacy Interrupt Handler
7908  *  @irq: interrupt number
7909  *  @data: pointer to a network interface device structure
7910  **/
7911 static irqreturn_t igb_intr(int irq, void *data)
7912 {
7913         struct igb_adapter *adapter = data;
7914         struct igb_q_vector *q_vector = adapter->q_vector[0];
7915         struct e1000_hw *hw = &adapter->hw;
7916         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
7917          * need for the IMC write
7918          */
7919         u32 icr = rd32(E1000_ICR);
7920
7921         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
7922          * not set, then the adapter didn't send an interrupt
7923          */
7924         if (!(icr & E1000_ICR_INT_ASSERTED))
7925                 return IRQ_NONE;
7926
7927         igb_write_itr(q_vector);
7928
7929         if (icr & E1000_ICR_DRSTA)
7930                 schedule_work(&adapter->reset_task);
7931
7932         if (icr & E1000_ICR_DOUTSYNC) {
7933                 /* HW is reporting DMA is out of sync */
7934                 adapter->stats.doosync++;
7935         }
7936
7937         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
7938                 hw->mac.get_link_status = 1;
7939                 /* guard against interrupt when we're going down */
7940                 if (!test_bit(__IGB_DOWN, &adapter->state))
7941                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
7942         }
7943
7944         if (icr & E1000_ICR_TS)
7945                 igb_tsync_interrupt(adapter);
7946
7947         napi_schedule(&q_vector->napi);
7948
7949         return IRQ_HANDLED;
7950 }
7951
7952 static void igb_ring_irq_enable(struct igb_q_vector *q_vector)
7953 {
7954         struct igb_adapter *adapter = q_vector->adapter;
7955         struct e1000_hw *hw = &adapter->hw;
7956
7957         if ((q_vector->rx.ring && (adapter->rx_itr_setting & 3)) ||
7958             (!q_vector->rx.ring && (adapter->tx_itr_setting & 3))) {
7959                 if ((adapter->num_q_vectors == 1) && !adapter->vf_data)
7960                         igb_set_itr(q_vector);
7961                 else
7962                         igb_update_ring_itr(q_vector);
7963         }
7964
7965         if (!test_bit(__IGB_DOWN, &adapter->state)) {
7966                 if (adapter->flags & IGB_FLAG_HAS_MSIX)
7967                         wr32(E1000_EIMS, q_vector->eims_value);
7968                 else
7969                         igb_irq_enable(adapter);
7970         }
7971 }
7972
7973 /**
7974  *  igb_poll - NAPI Rx polling callback
7975  *  @napi: napi polling structure
7976  *  @budget: count of how many packets we should handle
7977  **/
7978 static int igb_poll(struct napi_struct *napi, int budget)
7979 {
7980         struct igb_q_vector *q_vector = container_of(napi,
7981                                                      struct igb_q_vector,
7982                                                      napi);
7983         bool clean_complete = true;
7984         int work_done = 0;
7985
7986 #ifdef CONFIG_IGB_DCA
7987         if (q_vector->adapter->flags & IGB_FLAG_DCA_ENABLED)
7988                 igb_update_dca(q_vector);
7989 #endif
7990         if (q_vector->tx.ring)
7991                 clean_complete = igb_clean_tx_irq(q_vector, budget);
7992
7993         if (q_vector->rx.ring) {
7994                 int cleaned = igb_clean_rx_irq(q_vector, budget);
7995
7996                 work_done += cleaned;
7997                 if (cleaned >= budget)
7998                         clean_complete = false;
7999         }
8000
8001         /* If all work not completed, return budget and keep polling */
8002         if (!clean_complete)
8003                 return budget;
8004
8005         /* Exit the polling mode, but don't re-enable interrupts if stack might
8006          * poll us due to busy-polling
8007          */
8008         if (likely(napi_complete_done(napi, work_done)))
8009                 igb_ring_irq_enable(q_vector);
8010
8011         return min(work_done, budget - 1);
8012 }
8013
8014 /**
8015  *  igb_clean_tx_irq - Reclaim resources after transmit completes
8016  *  @q_vector: pointer to q_vector containing needed info
8017  *  @napi_budget: Used to determine if we are in netpoll
8018  *
8019  *  returns true if ring is completely cleaned
8020  **/
8021 static bool igb_clean_tx_irq(struct igb_q_vector *q_vector, int napi_budget)
8022 {
8023         struct igb_adapter *adapter = q_vector->adapter;
8024         struct igb_ring *tx_ring = q_vector->tx.ring;
8025         struct igb_tx_buffer *tx_buffer;
8026         union e1000_adv_tx_desc *tx_desc;
8027         unsigned int total_bytes = 0, total_packets = 0;
8028         unsigned int budget = q_vector->tx.work_limit;
8029         unsigned int i = tx_ring->next_to_clean;
8030
8031         if (test_bit(__IGB_DOWN, &adapter->state))
8032                 return true;
8033
8034         tx_buffer = &tx_ring->tx_buffer_info[i];
8035         tx_desc = IGB_TX_DESC(tx_ring, i);
8036         i -= tx_ring->count;
8037
8038         do {
8039                 union e1000_adv_tx_desc *eop_desc = tx_buffer->next_to_watch;
8040
8041                 /* if next_to_watch is not set then there is no work pending */
8042                 if (!eop_desc)
8043                         break;
8044
8045                 /* prevent any other reads prior to eop_desc */
8046                 smp_rmb();
8047
8048                 /* if DD is not set pending work has not been completed */
8049                 if (!(eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)))
8050                         break;
8051
8052                 /* clear next_to_watch to prevent false hangs */
8053                 tx_buffer->next_to_watch = NULL;
8054
8055                 /* update the statistics for this packet */
8056                 total_bytes += tx_buffer->bytecount;
8057                 total_packets += tx_buffer->gso_segs;
8058
8059                 /* free the skb */
8060                 if (tx_buffer->type == IGB_TYPE_SKB)
8061                         napi_consume_skb(tx_buffer->skb, napi_budget);
8062                 else
8063                         xdp_return_frame(tx_buffer->xdpf);
8064
8065                 /* unmap skb header data */
8066                 dma_unmap_single(tx_ring->dev,
8067                                  dma_unmap_addr(tx_buffer, dma),
8068                                  dma_unmap_len(tx_buffer, len),
8069                                  DMA_TO_DEVICE);
8070
8071                 /* clear tx_buffer data */
8072                 dma_unmap_len_set(tx_buffer, len, 0);
8073
8074                 /* clear last DMA location and unmap remaining buffers */
8075                 while (tx_desc != eop_desc) {
8076                         tx_buffer++;
8077                         tx_desc++;
8078                         i++;
8079                         if (unlikely(!i)) {
8080                                 i -= tx_ring->count;
8081                                 tx_buffer = tx_ring->tx_buffer_info;
8082                                 tx_desc = IGB_TX_DESC(tx_ring, 0);
8083                         }
8084
8085                         /* unmap any remaining paged data */
8086                         if (dma_unmap_len(tx_buffer, len)) {
8087                                 dma_unmap_page(tx_ring->dev,
8088                                                dma_unmap_addr(tx_buffer, dma),
8089                                                dma_unmap_len(tx_buffer, len),
8090                                                DMA_TO_DEVICE);
8091                                 dma_unmap_len_set(tx_buffer, len, 0);
8092                         }
8093                 }
8094
8095                 /* move us one more past the eop_desc for start of next pkt */
8096                 tx_buffer++;
8097                 tx_desc++;
8098                 i++;
8099                 if (unlikely(!i)) {
8100                         i -= tx_ring->count;
8101                         tx_buffer = tx_ring->tx_buffer_info;
8102                         tx_desc = IGB_TX_DESC(tx_ring, 0);
8103                 }
8104
8105                 /* issue prefetch for next Tx descriptor */
8106                 prefetch(tx_desc);
8107
8108                 /* update budget accounting */
8109                 budget--;
8110         } while (likely(budget));
8111
8112         netdev_tx_completed_queue(txring_txq(tx_ring),
8113                                   total_packets, total_bytes);
8114         i += tx_ring->count;
8115         tx_ring->next_to_clean = i;
8116         u64_stats_update_begin(&tx_ring->tx_syncp);
8117         tx_ring->tx_stats.bytes += total_bytes;
8118         tx_ring->tx_stats.packets += total_packets;
8119         u64_stats_update_end(&tx_ring->tx_syncp);
8120         q_vector->tx.total_bytes += total_bytes;
8121         q_vector->tx.total_packets += total_packets;
8122
8123         if (test_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags)) {
8124                 struct e1000_hw *hw = &adapter->hw;
8125
8126                 /* Detect a transmit hang in hardware, this serializes the
8127                  * check with the clearing of time_stamp and movement of i
8128                  */
8129                 clear_bit(IGB_RING_FLAG_TX_DETECT_HANG, &tx_ring->flags);
8130                 if (tx_buffer->next_to_watch &&
8131                     time_after(jiffies, tx_buffer->time_stamp +
8132                                (adapter->tx_timeout_factor * HZ)) &&
8133                     !(rd32(E1000_STATUS) & E1000_STATUS_TXOFF)) {
8134
8135                         /* detected Tx unit hang */
8136                         dev_err(tx_ring->dev,
8137                                 "Detected Tx Unit Hang\n"
8138                                 "  Tx Queue             <%d>\n"
8139                                 "  TDH                  <%x>\n"
8140                                 "  TDT                  <%x>\n"
8141                                 "  next_to_use          <%x>\n"
8142                                 "  next_to_clean        <%x>\n"
8143                                 "buffer_info[next_to_clean]\n"
8144                                 "  time_stamp           <%lx>\n"
8145                                 "  next_to_watch        <%p>\n"
8146                                 "  jiffies              <%lx>\n"
8147                                 "  desc.status          <%x>\n",
8148                                 tx_ring->queue_index,
8149                                 rd32(E1000_TDH(tx_ring->reg_idx)),
8150                                 readl(tx_ring->tail),
8151                                 tx_ring->next_to_use,
8152                                 tx_ring->next_to_clean,
8153                                 tx_buffer->time_stamp,
8154                                 tx_buffer->next_to_watch,
8155                                 jiffies,
8156                                 tx_buffer->next_to_watch->wb.status);
8157                         netif_stop_subqueue(tx_ring->netdev,
8158                                             tx_ring->queue_index);
8159
8160                         /* we are about to reset, no point in enabling stuff */
8161                         return true;
8162                 }
8163         }
8164
8165 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
8166         if (unlikely(total_packets &&
8167             netif_carrier_ok(tx_ring->netdev) &&
8168             igb_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD)) {
8169                 /* Make sure that anybody stopping the queue after this
8170                  * sees the new next_to_clean.
8171                  */
8172                 smp_mb();
8173                 if (__netif_subqueue_stopped(tx_ring->netdev,
8174                                              tx_ring->queue_index) &&
8175                     !(test_bit(__IGB_DOWN, &adapter->state))) {
8176                         netif_wake_subqueue(tx_ring->netdev,
8177                                             tx_ring->queue_index);
8178
8179                         u64_stats_update_begin(&tx_ring->tx_syncp);
8180                         tx_ring->tx_stats.restart_queue++;
8181                         u64_stats_update_end(&tx_ring->tx_syncp);
8182                 }
8183         }
8184
8185         return !!budget;
8186 }
8187
8188 /**
8189  *  igb_reuse_rx_page - page flip buffer and store it back on the ring
8190  *  @rx_ring: rx descriptor ring to store buffers on
8191  *  @old_buff: donor buffer to have page reused
8192  *
8193  *  Synchronizes page for reuse by the adapter
8194  **/
8195 static void igb_reuse_rx_page(struct igb_ring *rx_ring,
8196                               struct igb_rx_buffer *old_buff)
8197 {
8198         struct igb_rx_buffer *new_buff;
8199         u16 nta = rx_ring->next_to_alloc;
8200
8201         new_buff = &rx_ring->rx_buffer_info[nta];
8202
8203         /* update, and store next to alloc */
8204         nta++;
8205         rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
8206
8207         /* Transfer page from old buffer to new buffer.
8208          * Move each member individually to avoid possible store
8209          * forwarding stalls.
8210          */
8211         new_buff->dma           = old_buff->dma;
8212         new_buff->page          = old_buff->page;
8213         new_buff->page_offset   = old_buff->page_offset;
8214         new_buff->pagecnt_bias  = old_buff->pagecnt_bias;
8215 }
8216
8217 static bool igb_can_reuse_rx_page(struct igb_rx_buffer *rx_buffer,
8218                                   int rx_buf_pgcnt)
8219 {
8220         unsigned int pagecnt_bias = rx_buffer->pagecnt_bias;
8221         struct page *page = rx_buffer->page;
8222
8223         /* avoid re-using remote and pfmemalloc pages */
8224         if (!dev_page_is_reusable(page))
8225                 return false;
8226
8227 #if (PAGE_SIZE < 8192)
8228         /* if we are only owner of page we can reuse it */
8229         if (unlikely((rx_buf_pgcnt - pagecnt_bias) > 1))
8230                 return false;
8231 #else
8232 #define IGB_LAST_OFFSET \
8233         (SKB_WITH_OVERHEAD(PAGE_SIZE) - IGB_RXBUFFER_2048)
8234
8235         if (rx_buffer->page_offset > IGB_LAST_OFFSET)
8236                 return false;
8237 #endif
8238
8239         /* If we have drained the page fragment pool we need to update
8240          * the pagecnt_bias and page count so that we fully restock the
8241          * number of references the driver holds.
8242          */
8243         if (unlikely(pagecnt_bias == 1)) {
8244                 page_ref_add(page, USHRT_MAX - 1);
8245                 rx_buffer->pagecnt_bias = USHRT_MAX;
8246         }
8247
8248         return true;
8249 }
8250
8251 /**
8252  *  igb_add_rx_frag - Add contents of Rx buffer to sk_buff
8253  *  @rx_ring: rx descriptor ring to transact packets on
8254  *  @rx_buffer: buffer containing page to add
8255  *  @skb: sk_buff to place the data into
8256  *  @size: size of buffer to be added
8257  *
8258  *  This function will add the data contained in rx_buffer->page to the skb.
8259  **/
8260 static void igb_add_rx_frag(struct igb_ring *rx_ring,
8261                             struct igb_rx_buffer *rx_buffer,
8262                             struct sk_buff *skb,
8263                             unsigned int size)
8264 {
8265 #if (PAGE_SIZE < 8192)
8266         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8267 #else
8268         unsigned int truesize = ring_uses_build_skb(rx_ring) ?
8269                                 SKB_DATA_ALIGN(IGB_SKB_PAD + size) :
8270                                 SKB_DATA_ALIGN(size);
8271 #endif
8272         skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, rx_buffer->page,
8273                         rx_buffer->page_offset, size, truesize);
8274 #if (PAGE_SIZE < 8192)
8275         rx_buffer->page_offset ^= truesize;
8276 #else
8277         rx_buffer->page_offset += truesize;
8278 #endif
8279 }
8280
8281 static struct sk_buff *igb_construct_skb(struct igb_ring *rx_ring,
8282                                          struct igb_rx_buffer *rx_buffer,
8283                                          struct xdp_buff *xdp,
8284                                          ktime_t timestamp)
8285 {
8286 #if (PAGE_SIZE < 8192)
8287         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8288 #else
8289         unsigned int truesize = SKB_DATA_ALIGN(xdp->data_end -
8290                                                xdp->data_hard_start);
8291 #endif
8292         unsigned int size = xdp->data_end - xdp->data;
8293         unsigned int headlen;
8294         struct sk_buff *skb;
8295
8296         /* prefetch first cache line of first page */
8297         net_prefetch(xdp->data);
8298
8299         /* allocate a skb to store the frags */
8300         skb = napi_alloc_skb(&rx_ring->q_vector->napi, IGB_RX_HDR_LEN);
8301         if (unlikely(!skb))
8302                 return NULL;
8303
8304         if (timestamp)
8305                 skb_hwtstamps(skb)->hwtstamp = timestamp;
8306
8307         /* Determine available headroom for copy */
8308         headlen = size;
8309         if (headlen > IGB_RX_HDR_LEN)
8310                 headlen = eth_get_headlen(skb->dev, xdp->data, IGB_RX_HDR_LEN);
8311
8312         /* align pull length to size of long to optimize memcpy performance */
8313         memcpy(__skb_put(skb, headlen), xdp->data, ALIGN(headlen, sizeof(long)));
8314
8315         /* update all of the pointers */
8316         size -= headlen;
8317         if (size) {
8318                 skb_add_rx_frag(skb, 0, rx_buffer->page,
8319                                 (xdp->data + headlen) - page_address(rx_buffer->page),
8320                                 size, truesize);
8321 #if (PAGE_SIZE < 8192)
8322                 rx_buffer->page_offset ^= truesize;
8323 #else
8324                 rx_buffer->page_offset += truesize;
8325 #endif
8326         } else {
8327                 rx_buffer->pagecnt_bias++;
8328         }
8329
8330         return skb;
8331 }
8332
8333 static struct sk_buff *igb_build_skb(struct igb_ring *rx_ring,
8334                                      struct igb_rx_buffer *rx_buffer,
8335                                      struct xdp_buff *xdp,
8336                                      ktime_t timestamp)
8337 {
8338 #if (PAGE_SIZE < 8192)
8339         unsigned int truesize = igb_rx_pg_size(rx_ring) / 2;
8340 #else
8341         unsigned int truesize = SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) +
8342                                 SKB_DATA_ALIGN(xdp->data_end -
8343                                                xdp->data_hard_start);
8344 #endif
8345         unsigned int metasize = xdp->data - xdp->data_meta;
8346         struct sk_buff *skb;
8347
8348         /* prefetch first cache line of first page */
8349         net_prefetch(xdp->data_meta);
8350
8351         /* build an skb around the page buffer */
8352         skb = build_skb(xdp->data_hard_start, truesize);
8353         if (unlikely(!skb))
8354                 return NULL;
8355
8356         /* update pointers within the skb to store the data */
8357         skb_reserve(skb, xdp->data - xdp->data_hard_start);
8358         __skb_put(skb, xdp->data_end - xdp->data);
8359
8360         if (metasize)
8361                 skb_metadata_set(skb, metasize);
8362
8363         if (timestamp)
8364                 skb_hwtstamps(skb)->hwtstamp = timestamp;
8365
8366         /* update buffer offset */
8367 #if (PAGE_SIZE < 8192)
8368         rx_buffer->page_offset ^= truesize;
8369 #else
8370         rx_buffer->page_offset += truesize;
8371 #endif
8372
8373         return skb;
8374 }
8375
8376 static struct sk_buff *igb_run_xdp(struct igb_adapter *adapter,
8377                                    struct igb_ring *rx_ring,
8378                                    struct xdp_buff *xdp)
8379 {
8380         int err, result = IGB_XDP_PASS;
8381         struct bpf_prog *xdp_prog;
8382         u32 act;
8383
8384         xdp_prog = READ_ONCE(rx_ring->xdp_prog);
8385
8386         if (!xdp_prog)
8387                 goto xdp_out;
8388
8389         prefetchw(xdp->data_hard_start); /* xdp_frame write */
8390
8391         act = bpf_prog_run_xdp(xdp_prog, xdp);
8392         switch (act) {
8393         case XDP_PASS:
8394                 break;
8395         case XDP_TX:
8396                 result = igb_xdp_xmit_back(adapter, xdp);
8397                 if (result == IGB_XDP_CONSUMED)
8398                         goto out_failure;
8399                 break;
8400         case XDP_REDIRECT:
8401                 err = xdp_do_redirect(adapter->netdev, xdp, xdp_prog);
8402                 if (err)
8403                         goto out_failure;
8404                 result = IGB_XDP_REDIR;
8405                 break;
8406         default:
8407                 bpf_warn_invalid_xdp_action(act);
8408                 fallthrough;
8409         case XDP_ABORTED:
8410 out_failure:
8411                 trace_xdp_exception(rx_ring->netdev, xdp_prog, act);
8412                 fallthrough;
8413         case XDP_DROP:
8414                 result = IGB_XDP_CONSUMED;
8415                 break;
8416         }
8417 xdp_out:
8418         return ERR_PTR(-result);
8419 }
8420
8421 static unsigned int igb_rx_frame_truesize(struct igb_ring *rx_ring,
8422                                           unsigned int size)
8423 {
8424         unsigned int truesize;
8425
8426 #if (PAGE_SIZE < 8192)
8427         truesize = igb_rx_pg_size(rx_ring) / 2; /* Must be power-of-2 */
8428 #else
8429         truesize = ring_uses_build_skb(rx_ring) ?
8430                 SKB_DATA_ALIGN(IGB_SKB_PAD + size) +
8431                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)) :
8432                 SKB_DATA_ALIGN(size);
8433 #endif
8434         return truesize;
8435 }
8436
8437 static void igb_rx_buffer_flip(struct igb_ring *rx_ring,
8438                                struct igb_rx_buffer *rx_buffer,
8439                                unsigned int size)
8440 {
8441         unsigned int truesize = igb_rx_frame_truesize(rx_ring, size);
8442 #if (PAGE_SIZE < 8192)
8443         rx_buffer->page_offset ^= truesize;
8444 #else
8445         rx_buffer->page_offset += truesize;
8446 #endif
8447 }
8448
8449 static inline void igb_rx_checksum(struct igb_ring *ring,
8450                                    union e1000_adv_rx_desc *rx_desc,
8451                                    struct sk_buff *skb)
8452 {
8453         skb_checksum_none_assert(skb);
8454
8455         /* Ignore Checksum bit is set */
8456         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_IXSM))
8457                 return;
8458
8459         /* Rx checksum disabled via ethtool */
8460         if (!(ring->netdev->features & NETIF_F_RXCSUM))
8461                 return;
8462
8463         /* TCP/UDP checksum error bit is set */
8464         if (igb_test_staterr(rx_desc,
8465                              E1000_RXDEXT_STATERR_TCPE |
8466                              E1000_RXDEXT_STATERR_IPE)) {
8467                 /* work around errata with sctp packets where the TCPE aka
8468                  * L4E bit is set incorrectly on 64 byte (60 byte w/o crc)
8469                  * packets, (aka let the stack check the crc32c)
8470                  */
8471                 if (!((skb->len == 60) &&
8472                       test_bit(IGB_RING_FLAG_RX_SCTP_CSUM, &ring->flags))) {
8473                         u64_stats_update_begin(&ring->rx_syncp);
8474                         ring->rx_stats.csum_err++;
8475                         u64_stats_update_end(&ring->rx_syncp);
8476                 }
8477                 /* let the stack verify checksum errors */
8478                 return;
8479         }
8480         /* It must be a TCP or UDP packet with a valid checksum */
8481         if (igb_test_staterr(rx_desc, E1000_RXD_STAT_TCPCS |
8482                                       E1000_RXD_STAT_UDPCS))
8483                 skb->ip_summed = CHECKSUM_UNNECESSARY;
8484
8485         dev_dbg(ring->dev, "cksum success: bits %08X\n",
8486                 le32_to_cpu(rx_desc->wb.upper.status_error));
8487 }
8488
8489 static inline void igb_rx_hash(struct igb_ring *ring,
8490                                union e1000_adv_rx_desc *rx_desc,
8491                                struct sk_buff *skb)
8492 {
8493         if (ring->netdev->features & NETIF_F_RXHASH)
8494                 skb_set_hash(skb,
8495                              le32_to_cpu(rx_desc->wb.lower.hi_dword.rss),
8496                              PKT_HASH_TYPE_L3);
8497 }
8498
8499 /**
8500  *  igb_is_non_eop - process handling of non-EOP buffers
8501  *  @rx_ring: Rx ring being processed
8502  *  @rx_desc: Rx descriptor for current buffer
8503  *
8504  *  This function updates next to clean.  If the buffer is an EOP buffer
8505  *  this function exits returning false, otherwise it will place the
8506  *  sk_buff in the next buffer to be chained and return true indicating
8507  *  that this is in fact a non-EOP buffer.
8508  **/
8509 static bool igb_is_non_eop(struct igb_ring *rx_ring,
8510                            union e1000_adv_rx_desc *rx_desc)
8511 {
8512         u32 ntc = rx_ring->next_to_clean + 1;
8513
8514         /* fetch, update, and store next to clean */
8515         ntc = (ntc < rx_ring->count) ? ntc : 0;
8516         rx_ring->next_to_clean = ntc;
8517
8518         prefetch(IGB_RX_DESC(rx_ring, ntc));
8519
8520         if (likely(igb_test_staterr(rx_desc, E1000_RXD_STAT_EOP)))
8521                 return false;
8522
8523         return true;
8524 }
8525
8526 /**
8527  *  igb_cleanup_headers - Correct corrupted or empty headers
8528  *  @rx_ring: rx descriptor ring packet is being transacted on
8529  *  @rx_desc: pointer to the EOP Rx descriptor
8530  *  @skb: pointer to current skb being fixed
8531  *
8532  *  Address the case where we are pulling data in on pages only
8533  *  and as such no data is present in the skb header.
8534  *
8535  *  In addition if skb is not at least 60 bytes we need to pad it so that
8536  *  it is large enough to qualify as a valid Ethernet frame.
8537  *
8538  *  Returns true if an error was encountered and skb was freed.
8539  **/
8540 static bool igb_cleanup_headers(struct igb_ring *rx_ring,
8541                                 union e1000_adv_rx_desc *rx_desc,
8542                                 struct sk_buff *skb)
8543 {
8544         /* XDP packets use error pointer so abort at this point */
8545         if (IS_ERR(skb))
8546                 return true;
8547
8548         if (unlikely((igb_test_staterr(rx_desc,
8549                                        E1000_RXDEXT_ERR_FRAME_ERR_MASK)))) {
8550                 struct net_device *netdev = rx_ring->netdev;
8551                 if (!(netdev->features & NETIF_F_RXALL)) {
8552                         dev_kfree_skb_any(skb);
8553                         return true;
8554                 }
8555         }
8556
8557         /* if eth_skb_pad returns an error the skb was freed */
8558         if (eth_skb_pad(skb))
8559                 return true;
8560
8561         return false;
8562 }
8563
8564 /**
8565  *  igb_process_skb_fields - Populate skb header fields from Rx descriptor
8566  *  @rx_ring: rx descriptor ring packet is being transacted on
8567  *  @rx_desc: pointer to the EOP Rx descriptor
8568  *  @skb: pointer to current skb being populated
8569  *
8570  *  This function checks the ring, descriptor, and packet information in
8571  *  order to populate the hash, checksum, VLAN, timestamp, protocol, and
8572  *  other fields within the skb.
8573  **/
8574 static void igb_process_skb_fields(struct igb_ring *rx_ring,
8575                                    union e1000_adv_rx_desc *rx_desc,
8576                                    struct sk_buff *skb)
8577 {
8578         struct net_device *dev = rx_ring->netdev;
8579
8580         igb_rx_hash(rx_ring, rx_desc, skb);
8581
8582         igb_rx_checksum(rx_ring, rx_desc, skb);
8583
8584         if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TS) &&
8585             !igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP))
8586                 igb_ptp_rx_rgtstamp(rx_ring->q_vector, skb);
8587
8588         if ((dev->features & NETIF_F_HW_VLAN_CTAG_RX) &&
8589             igb_test_staterr(rx_desc, E1000_RXD_STAT_VP)) {
8590                 u16 vid;
8591
8592                 if (igb_test_staterr(rx_desc, E1000_RXDEXT_STATERR_LB) &&
8593                     test_bit(IGB_RING_FLAG_RX_LB_VLAN_BSWAP, &rx_ring->flags))
8594                         vid = be16_to_cpu((__force __be16)rx_desc->wb.upper.vlan);
8595                 else
8596                         vid = le16_to_cpu(rx_desc->wb.upper.vlan);
8597
8598                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
8599         }
8600
8601         skb_record_rx_queue(skb, rx_ring->queue_index);
8602
8603         skb->protocol = eth_type_trans(skb, rx_ring->netdev);
8604 }
8605
8606 static unsigned int igb_rx_offset(struct igb_ring *rx_ring)
8607 {
8608         return ring_uses_build_skb(rx_ring) ? IGB_SKB_PAD : 0;
8609 }
8610
8611 static struct igb_rx_buffer *igb_get_rx_buffer(struct igb_ring *rx_ring,
8612                                                const unsigned int size, int *rx_buf_pgcnt)
8613 {
8614         struct igb_rx_buffer *rx_buffer;
8615
8616         rx_buffer = &rx_ring->rx_buffer_info[rx_ring->next_to_clean];
8617         *rx_buf_pgcnt =
8618 #if (PAGE_SIZE < 8192)
8619                 page_count(rx_buffer->page);
8620 #else
8621                 0;
8622 #endif
8623         prefetchw(rx_buffer->page);
8624
8625         /* we are reusing so sync this buffer for CPU use */
8626         dma_sync_single_range_for_cpu(rx_ring->dev,
8627                                       rx_buffer->dma,
8628                                       rx_buffer->page_offset,
8629                                       size,
8630                                       DMA_FROM_DEVICE);
8631
8632         rx_buffer->pagecnt_bias--;
8633
8634         return rx_buffer;
8635 }
8636
8637 static void igb_put_rx_buffer(struct igb_ring *rx_ring,
8638                               struct igb_rx_buffer *rx_buffer, int rx_buf_pgcnt)
8639 {
8640         if (igb_can_reuse_rx_page(rx_buffer, rx_buf_pgcnt)) {
8641                 /* hand second half of page back to the ring */
8642                 igb_reuse_rx_page(rx_ring, rx_buffer);
8643         } else {
8644                 /* We are not reusing the buffer so unmap it and free
8645                  * any references we are holding to it
8646                  */
8647                 dma_unmap_page_attrs(rx_ring->dev, rx_buffer->dma,
8648                                      igb_rx_pg_size(rx_ring), DMA_FROM_DEVICE,
8649                                      IGB_RX_DMA_ATTR);
8650                 __page_frag_cache_drain(rx_buffer->page,
8651                                         rx_buffer->pagecnt_bias);
8652         }
8653
8654         /* clear contents of rx_buffer */
8655         rx_buffer->page = NULL;
8656 }
8657
8658 static int igb_clean_rx_irq(struct igb_q_vector *q_vector, const int budget)
8659 {
8660         struct igb_adapter *adapter = q_vector->adapter;
8661         struct igb_ring *rx_ring = q_vector->rx.ring;
8662         struct sk_buff *skb = rx_ring->skb;
8663         unsigned int total_bytes = 0, total_packets = 0;
8664         u16 cleaned_count = igb_desc_unused(rx_ring);
8665         unsigned int xdp_xmit = 0;
8666         struct xdp_buff xdp;
8667         u32 frame_sz = 0;
8668         int rx_buf_pgcnt;
8669
8670         /* Frame size depend on rx_ring setup when PAGE_SIZE=4K */
8671 #if (PAGE_SIZE < 8192)
8672         frame_sz = igb_rx_frame_truesize(rx_ring, 0);
8673 #endif
8674         xdp_init_buff(&xdp, frame_sz, &rx_ring->xdp_rxq);
8675
8676         while (likely(total_packets < budget)) {
8677                 union e1000_adv_rx_desc *rx_desc;
8678                 struct igb_rx_buffer *rx_buffer;
8679                 ktime_t timestamp = 0;
8680                 int pkt_offset = 0;
8681                 unsigned int size;
8682                 void *pktbuf;
8683
8684                 /* return some buffers to hardware, one at a time is too slow */
8685                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
8686                         igb_alloc_rx_buffers(rx_ring, cleaned_count);
8687                         cleaned_count = 0;
8688                 }
8689
8690                 rx_desc = IGB_RX_DESC(rx_ring, rx_ring->next_to_clean);
8691                 size = le16_to_cpu(rx_desc->wb.upper.length);
8692                 if (!size)
8693                         break;
8694
8695                 /* This memory barrier is needed to keep us from reading
8696                  * any other fields out of the rx_desc until we know the
8697                  * descriptor has been written back
8698                  */
8699                 dma_rmb();
8700
8701                 rx_buffer = igb_get_rx_buffer(rx_ring, size, &rx_buf_pgcnt);
8702                 pktbuf = page_address(rx_buffer->page) + rx_buffer->page_offset;
8703
8704                 /* pull rx packet timestamp if available and valid */
8705                 if (igb_test_staterr(rx_desc, E1000_RXDADV_STAT_TSIP)) {
8706                         int ts_hdr_len;
8707
8708                         ts_hdr_len = igb_ptp_rx_pktstamp(rx_ring->q_vector,
8709                                                          pktbuf, &timestamp);
8710
8711                         pkt_offset += ts_hdr_len;
8712                         size -= ts_hdr_len;
8713                 }
8714
8715                 /* retrieve a buffer from the ring */
8716                 if (!skb) {
8717                         unsigned char *hard_start = pktbuf - igb_rx_offset(rx_ring);
8718                         unsigned int offset = pkt_offset + igb_rx_offset(rx_ring);
8719
8720                         xdp_prepare_buff(&xdp, hard_start, offset, size, true);
8721 #if (PAGE_SIZE > 4096)
8722                         /* At larger PAGE_SIZE, frame_sz depend on len size */
8723                         xdp.frame_sz = igb_rx_frame_truesize(rx_ring, size);
8724 #endif
8725                         skb = igb_run_xdp(adapter, rx_ring, &xdp);
8726                 }
8727
8728                 if (IS_ERR(skb)) {
8729                         unsigned int xdp_res = -PTR_ERR(skb);
8730
8731                         if (xdp_res & (IGB_XDP_TX | IGB_XDP_REDIR)) {
8732                                 xdp_xmit |= xdp_res;
8733                                 igb_rx_buffer_flip(rx_ring, rx_buffer, size);
8734                         } else {
8735                                 rx_buffer->pagecnt_bias++;
8736                         }
8737                         total_packets++;
8738                         total_bytes += size;
8739                 } else if (skb)
8740                         igb_add_rx_frag(rx_ring, rx_buffer, skb, size);
8741                 else if (ring_uses_build_skb(rx_ring))
8742                         skb = igb_build_skb(rx_ring, rx_buffer, &xdp,
8743                                             timestamp);
8744                 else
8745                         skb = igb_construct_skb(rx_ring, rx_buffer,
8746                                                 &xdp, timestamp);
8747
8748                 /* exit if we failed to retrieve a buffer */
8749                 if (!skb) {
8750                         rx_ring->rx_stats.alloc_failed++;
8751                         rx_buffer->pagecnt_bias++;
8752                         break;
8753                 }
8754
8755                 igb_put_rx_buffer(rx_ring, rx_buffer, rx_buf_pgcnt);
8756                 cleaned_count++;
8757
8758                 /* fetch next buffer in frame if non-eop */
8759                 if (igb_is_non_eop(rx_ring, rx_desc))
8760                         continue;
8761
8762                 /* verify the packet layout is correct */
8763                 if (igb_cleanup_headers(rx_ring, rx_desc, skb)) {
8764                         skb = NULL;
8765                         continue;
8766                 }
8767
8768                 /* probably a little skewed due to removing CRC */
8769                 total_bytes += skb->len;
8770
8771                 /* populate checksum, timestamp, VLAN, and protocol */
8772                 igb_process_skb_fields(rx_ring, rx_desc, skb);
8773
8774                 napi_gro_receive(&q_vector->napi, skb);
8775
8776                 /* reset skb pointer */
8777                 skb = NULL;
8778
8779                 /* update budget accounting */
8780                 total_packets++;
8781         }
8782
8783         /* place incomplete frames back on ring for completion */
8784         rx_ring->skb = skb;
8785
8786         if (xdp_xmit & IGB_XDP_REDIR)
8787                 xdp_do_flush();
8788
8789         if (xdp_xmit & IGB_XDP_TX) {
8790                 struct igb_ring *tx_ring = igb_xdp_tx_queue_mapping(adapter);
8791
8792                 igb_xdp_ring_update_tail(tx_ring);
8793         }
8794
8795         u64_stats_update_begin(&rx_ring->rx_syncp);
8796         rx_ring->rx_stats.packets += total_packets;
8797         rx_ring->rx_stats.bytes += total_bytes;
8798         u64_stats_update_end(&rx_ring->rx_syncp);
8799         q_vector->rx.total_packets += total_packets;
8800         q_vector->rx.total_bytes += total_bytes;
8801
8802         if (cleaned_count)
8803                 igb_alloc_rx_buffers(rx_ring, cleaned_count);
8804
8805         return total_packets;
8806 }
8807
8808 static bool igb_alloc_mapped_page(struct igb_ring *rx_ring,
8809                                   struct igb_rx_buffer *bi)
8810 {
8811         struct page *page = bi->page;
8812         dma_addr_t dma;
8813
8814         /* since we are recycling buffers we should seldom need to alloc */
8815         if (likely(page))
8816                 return true;
8817
8818         /* alloc new page for storage */
8819         page = dev_alloc_pages(igb_rx_pg_order(rx_ring));
8820         if (unlikely(!page)) {
8821                 rx_ring->rx_stats.alloc_failed++;
8822                 return false;
8823         }
8824
8825         /* map page for use */
8826         dma = dma_map_page_attrs(rx_ring->dev, page, 0,
8827                                  igb_rx_pg_size(rx_ring),
8828                                  DMA_FROM_DEVICE,
8829                                  IGB_RX_DMA_ATTR);
8830
8831         /* if mapping failed free memory back to system since
8832          * there isn't much point in holding memory we can't use
8833          */
8834         if (dma_mapping_error(rx_ring->dev, dma)) {
8835                 __free_pages(page, igb_rx_pg_order(rx_ring));
8836
8837                 rx_ring->rx_stats.alloc_failed++;
8838                 return false;
8839         }
8840
8841         bi->dma = dma;
8842         bi->page = page;
8843         bi->page_offset = igb_rx_offset(rx_ring);
8844         page_ref_add(page, USHRT_MAX - 1);
8845         bi->pagecnt_bias = USHRT_MAX;
8846
8847         return true;
8848 }
8849
8850 /**
8851  *  igb_alloc_rx_buffers - Replace used receive buffers
8852  *  @rx_ring: rx descriptor ring to allocate new receive buffers
8853  *  @cleaned_count: count of buffers to allocate
8854  **/
8855 void igb_alloc_rx_buffers(struct igb_ring *rx_ring, u16 cleaned_count)
8856 {
8857         union e1000_adv_rx_desc *rx_desc;
8858         struct igb_rx_buffer *bi;
8859         u16 i = rx_ring->next_to_use;
8860         u16 bufsz;
8861
8862         /* nothing to do */
8863         if (!cleaned_count)
8864                 return;
8865
8866         rx_desc = IGB_RX_DESC(rx_ring, i);
8867         bi = &rx_ring->rx_buffer_info[i];
8868         i -= rx_ring->count;
8869
8870         bufsz = igb_rx_bufsz(rx_ring);
8871
8872         do {
8873                 if (!igb_alloc_mapped_page(rx_ring, bi))
8874                         break;
8875
8876                 /* sync the buffer for use by the device */
8877                 dma_sync_single_range_for_device(rx_ring->dev, bi->dma,
8878                                                  bi->page_offset, bufsz,
8879                                                  DMA_FROM_DEVICE);
8880
8881                 /* Refresh the desc even if buffer_addrs didn't change
8882                  * because each write-back erases this info.
8883                  */
8884                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
8885
8886                 rx_desc++;
8887                 bi++;
8888                 i++;
8889                 if (unlikely(!i)) {
8890                         rx_desc = IGB_RX_DESC(rx_ring, 0);
8891                         bi = rx_ring->rx_buffer_info;
8892                         i -= rx_ring->count;
8893                 }
8894
8895                 /* clear the length for the next_to_use descriptor */
8896                 rx_desc->wb.upper.length = 0;
8897
8898                 cleaned_count--;
8899         } while (cleaned_count);
8900
8901         i += rx_ring->count;
8902
8903         if (rx_ring->next_to_use != i) {
8904                 /* record the next descriptor to use */
8905                 rx_ring->next_to_use = i;
8906
8907                 /* update next to alloc since we have filled the ring */
8908                 rx_ring->next_to_alloc = i;
8909
8910                 /* Force memory writes to complete before letting h/w
8911                  * know there are new descriptors to fetch.  (Only
8912                  * applicable for weak-ordered memory model archs,
8913                  * such as IA-64).
8914                  */
8915                 dma_wmb();
8916                 writel(i, rx_ring->tail);
8917         }
8918 }
8919
8920 /**
8921  * igb_mii_ioctl -
8922  * @netdev: pointer to netdev struct
8923  * @ifr: interface structure
8924  * @cmd: ioctl command to execute
8925  **/
8926 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8927 {
8928         struct igb_adapter *adapter = netdev_priv(netdev);
8929         struct mii_ioctl_data *data = if_mii(ifr);
8930
8931         if (adapter->hw.phy.media_type != e1000_media_type_copper)
8932                 return -EOPNOTSUPP;
8933
8934         switch (cmd) {
8935         case SIOCGMIIPHY:
8936                 data->phy_id = adapter->hw.phy.addr;
8937                 break;
8938         case SIOCGMIIREG:
8939                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
8940                                      &data->val_out))
8941                         return -EIO;
8942                 break;
8943         case SIOCSMIIREG:
8944         default:
8945                 return -EOPNOTSUPP;
8946         }
8947         return 0;
8948 }
8949
8950 /**
8951  * igb_ioctl -
8952  * @netdev: pointer to netdev struct
8953  * @ifr: interface structure
8954  * @cmd: ioctl command to execute
8955  **/
8956 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
8957 {
8958         switch (cmd) {
8959         case SIOCGMIIPHY:
8960         case SIOCGMIIREG:
8961         case SIOCSMIIREG:
8962                 return igb_mii_ioctl(netdev, ifr, cmd);
8963         case SIOCGHWTSTAMP:
8964                 return igb_ptp_get_ts_config(netdev, ifr);
8965         case SIOCSHWTSTAMP:
8966                 return igb_ptp_set_ts_config(netdev, ifr);
8967         default:
8968                 return -EOPNOTSUPP;
8969         }
8970 }
8971
8972 void igb_read_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8973 {
8974         struct igb_adapter *adapter = hw->back;
8975
8976         pci_read_config_word(adapter->pdev, reg, value);
8977 }
8978
8979 void igb_write_pci_cfg(struct e1000_hw *hw, u32 reg, u16 *value)
8980 {
8981         struct igb_adapter *adapter = hw->back;
8982
8983         pci_write_config_word(adapter->pdev, reg, *value);
8984 }
8985
8986 s32 igb_read_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8987 {
8988         struct igb_adapter *adapter = hw->back;
8989
8990         if (pcie_capability_read_word(adapter->pdev, reg, value))
8991                 return -E1000_ERR_CONFIG;
8992
8993         return 0;
8994 }
8995
8996 s32 igb_write_pcie_cap_reg(struct e1000_hw *hw, u32 reg, u16 *value)
8997 {
8998         struct igb_adapter *adapter = hw->back;
8999
9000         if (pcie_capability_write_word(adapter->pdev, reg, *value))
9001                 return -E1000_ERR_CONFIG;
9002
9003         return 0;
9004 }
9005
9006 static void igb_vlan_mode(struct net_device *netdev, netdev_features_t features)
9007 {
9008         struct igb_adapter *adapter = netdev_priv(netdev);
9009         struct e1000_hw *hw = &adapter->hw;
9010         u32 ctrl, rctl;
9011         bool enable = !!(features & NETIF_F_HW_VLAN_CTAG_RX);
9012
9013         if (enable) {
9014                 /* enable VLAN tag insert/strip */
9015                 ctrl = rd32(E1000_CTRL);
9016                 ctrl |= E1000_CTRL_VME;
9017                 wr32(E1000_CTRL, ctrl);
9018
9019                 /* Disable CFI check */
9020                 rctl = rd32(E1000_RCTL);
9021                 rctl &= ~E1000_RCTL_CFIEN;
9022                 wr32(E1000_RCTL, rctl);
9023         } else {
9024                 /* disable VLAN tag insert/strip */
9025                 ctrl = rd32(E1000_CTRL);
9026                 ctrl &= ~E1000_CTRL_VME;
9027                 wr32(E1000_CTRL, ctrl);
9028         }
9029
9030         igb_set_vf_vlan_strip(adapter, adapter->vfs_allocated_count, enable);
9031 }
9032
9033 static int igb_vlan_rx_add_vid(struct net_device *netdev,
9034                                __be16 proto, u16 vid)
9035 {
9036         struct igb_adapter *adapter = netdev_priv(netdev);
9037         struct e1000_hw *hw = &adapter->hw;
9038         int pf_id = adapter->vfs_allocated_count;
9039
9040         /* add the filter since PF can receive vlans w/o entry in vlvf */
9041         if (!vid || !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9042                 igb_vfta_set(hw, vid, pf_id, true, !!vid);
9043
9044         set_bit(vid, adapter->active_vlans);
9045
9046         return 0;
9047 }
9048
9049 static int igb_vlan_rx_kill_vid(struct net_device *netdev,
9050                                 __be16 proto, u16 vid)
9051 {
9052         struct igb_adapter *adapter = netdev_priv(netdev);
9053         int pf_id = adapter->vfs_allocated_count;
9054         struct e1000_hw *hw = &adapter->hw;
9055
9056         /* remove VID from filter table */
9057         if (vid && !(adapter->flags & IGB_FLAG_VLAN_PROMISC))
9058                 igb_vfta_set(hw, vid, pf_id, false, true);
9059
9060         clear_bit(vid, adapter->active_vlans);
9061
9062         return 0;
9063 }
9064
9065 static void igb_restore_vlan(struct igb_adapter *adapter)
9066 {
9067         u16 vid = 1;
9068
9069         igb_vlan_mode(adapter->netdev, adapter->netdev->features);
9070         igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), 0);
9071
9072         for_each_set_bit_from(vid, adapter->active_vlans, VLAN_N_VID)
9073                 igb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
9074 }
9075
9076 int igb_set_spd_dplx(struct igb_adapter *adapter, u32 spd, u8 dplx)
9077 {
9078         struct pci_dev *pdev = adapter->pdev;
9079         struct e1000_mac_info *mac = &adapter->hw.mac;
9080
9081         mac->autoneg = 0;
9082
9083         /* Make sure dplx is at most 1 bit and lsb of speed is not set
9084          * for the switch() below to work
9085          */
9086         if ((spd & 1) || (dplx & ~1))
9087                 goto err_inval;
9088
9089         /* Fiber NIC's only allow 1000 gbps Full duplex
9090          * and 100Mbps Full duplex for 100baseFx sfp
9091          */
9092         if (adapter->hw.phy.media_type == e1000_media_type_internal_serdes) {
9093                 switch (spd + dplx) {
9094                 case SPEED_10 + DUPLEX_HALF:
9095                 case SPEED_10 + DUPLEX_FULL:
9096                 case SPEED_100 + DUPLEX_HALF:
9097                         goto err_inval;
9098                 default:
9099                         break;
9100                 }
9101         }
9102
9103         switch (spd + dplx) {
9104         case SPEED_10 + DUPLEX_HALF:
9105                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
9106                 break;
9107         case SPEED_10 + DUPLEX_FULL:
9108                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
9109                 break;
9110         case SPEED_100 + DUPLEX_HALF:
9111                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
9112                 break;
9113         case SPEED_100 + DUPLEX_FULL:
9114                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
9115                 break;
9116         case SPEED_1000 + DUPLEX_FULL:
9117                 mac->autoneg = 1;
9118                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
9119                 break;
9120         case SPEED_1000 + DUPLEX_HALF: /* not supported */
9121         default:
9122                 goto err_inval;
9123         }
9124
9125         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
9126         adapter->hw.phy.mdix = AUTO_ALL_MODES;
9127
9128         return 0;
9129
9130 err_inval:
9131         dev_err(&pdev->dev, "Unsupported Speed/Duplex configuration\n");
9132         return -EINVAL;
9133 }
9134
9135 static int __igb_shutdown(struct pci_dev *pdev, bool *enable_wake,
9136                           bool runtime)
9137 {
9138         struct net_device *netdev = pci_get_drvdata(pdev);
9139         struct igb_adapter *adapter = netdev_priv(netdev);
9140         struct e1000_hw *hw = &adapter->hw;
9141         u32 ctrl, rctl, status;
9142         u32 wufc = runtime ? E1000_WUFC_LNKC : adapter->wol;
9143         bool wake;
9144
9145         rtnl_lock();
9146         netif_device_detach(netdev);
9147
9148         if (netif_running(netdev))
9149                 __igb_close(netdev, true);
9150
9151         igb_ptp_suspend(adapter);
9152
9153         igb_clear_interrupt_scheme(adapter);
9154         rtnl_unlock();
9155
9156         status = rd32(E1000_STATUS);
9157         if (status & E1000_STATUS_LU)
9158                 wufc &= ~E1000_WUFC_LNKC;
9159
9160         if (wufc) {
9161                 igb_setup_rctl(adapter);
9162                 igb_set_rx_mode(netdev);
9163
9164                 /* turn on all-multi mode if wake on multicast is enabled */
9165                 if (wufc & E1000_WUFC_MC) {
9166                         rctl = rd32(E1000_RCTL);
9167                         rctl |= E1000_RCTL_MPE;
9168                         wr32(E1000_RCTL, rctl);
9169                 }
9170
9171                 ctrl = rd32(E1000_CTRL);
9172                 ctrl |= E1000_CTRL_ADVD3WUC;
9173                 wr32(E1000_CTRL, ctrl);
9174
9175                 /* Allow time for pending master requests to run */
9176                 igb_disable_pcie_master(hw);
9177
9178                 wr32(E1000_WUC, E1000_WUC_PME_EN);
9179                 wr32(E1000_WUFC, wufc);
9180         } else {
9181                 wr32(E1000_WUC, 0);
9182                 wr32(E1000_WUFC, 0);
9183         }
9184
9185         wake = wufc || adapter->en_mng_pt;
9186         if (!wake)
9187                 igb_power_down_link(adapter);
9188         else
9189                 igb_power_up_link(adapter);
9190
9191         if (enable_wake)
9192                 *enable_wake = wake;
9193
9194         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
9195          * would have already happened in close and is redundant.
9196          */
9197         igb_release_hw_control(adapter);
9198
9199         pci_disable_device(pdev);
9200
9201         return 0;
9202 }
9203
9204 static void igb_deliver_wake_packet(struct net_device *netdev)
9205 {
9206         struct igb_adapter *adapter = netdev_priv(netdev);
9207         struct e1000_hw *hw = &adapter->hw;
9208         struct sk_buff *skb;
9209         u32 wupl;
9210
9211         wupl = rd32(E1000_WUPL) & E1000_WUPL_MASK;
9212
9213         /* WUPM stores only the first 128 bytes of the wake packet.
9214          * Read the packet only if we have the whole thing.
9215          */
9216         if ((wupl == 0) || (wupl > E1000_WUPM_BYTES))
9217                 return;
9218
9219         skb = netdev_alloc_skb_ip_align(netdev, E1000_WUPM_BYTES);
9220         if (!skb)
9221                 return;
9222
9223         skb_put(skb, wupl);
9224
9225         /* Ensure reads are 32-bit aligned */
9226         wupl = roundup(wupl, 4);
9227
9228         memcpy_fromio(skb->data, hw->hw_addr + E1000_WUPM_REG(0), wupl);
9229
9230         skb->protocol = eth_type_trans(skb, netdev);
9231         netif_rx(skb);
9232 }
9233
9234 static int __maybe_unused igb_suspend(struct device *dev)
9235 {
9236         return __igb_shutdown(to_pci_dev(dev), NULL, 0);
9237 }
9238
9239 static int __maybe_unused igb_resume(struct device *dev)
9240 {
9241         struct pci_dev *pdev = to_pci_dev(dev);
9242         struct net_device *netdev = pci_get_drvdata(pdev);
9243         struct igb_adapter *adapter = netdev_priv(netdev);
9244         struct e1000_hw *hw = &adapter->hw;
9245         u32 err, val;
9246
9247         pci_set_power_state(pdev, PCI_D0);
9248         pci_restore_state(pdev);
9249         pci_save_state(pdev);
9250
9251         if (!pci_device_is_present(pdev))
9252                 return -ENODEV;
9253         err = pci_enable_device_mem(pdev);
9254         if (err) {
9255                 dev_err(&pdev->dev,
9256                         "igb: Cannot enable PCI device from suspend\n");
9257                 return err;
9258         }
9259         pci_set_master(pdev);
9260
9261         pci_enable_wake(pdev, PCI_D3hot, 0);
9262         pci_enable_wake(pdev, PCI_D3cold, 0);
9263
9264         if (igb_init_interrupt_scheme(adapter, true)) {
9265                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9266                 return -ENOMEM;
9267         }
9268
9269         igb_reset(adapter);
9270
9271         /* let the f/w know that the h/w is now under the control of the
9272          * driver.
9273          */
9274         igb_get_hw_control(adapter);
9275
9276         val = rd32(E1000_WUS);
9277         if (val & WAKE_PKT_WUS)
9278                 igb_deliver_wake_packet(netdev);
9279
9280         wr32(E1000_WUS, ~0);
9281
9282         rtnl_lock();
9283         if (!err && netif_running(netdev))
9284                 err = __igb_open(netdev, true);
9285
9286         if (!err)
9287                 netif_device_attach(netdev);
9288         rtnl_unlock();
9289
9290         return err;
9291 }
9292
9293 static int __maybe_unused igb_runtime_idle(struct device *dev)
9294 {
9295         struct net_device *netdev = dev_get_drvdata(dev);
9296         struct igb_adapter *adapter = netdev_priv(netdev);
9297
9298         if (!igb_has_link(adapter))
9299                 pm_schedule_suspend(dev, MSEC_PER_SEC * 5);
9300
9301         return -EBUSY;
9302 }
9303
9304 static int __maybe_unused igb_runtime_suspend(struct device *dev)
9305 {
9306         return __igb_shutdown(to_pci_dev(dev), NULL, 1);
9307 }
9308
9309 static int __maybe_unused igb_runtime_resume(struct device *dev)
9310 {
9311         return igb_resume(dev);
9312 }
9313
9314 static void igb_shutdown(struct pci_dev *pdev)
9315 {
9316         bool wake;
9317
9318         __igb_shutdown(pdev, &wake, 0);
9319
9320         if (system_state == SYSTEM_POWER_OFF) {
9321                 pci_wake_from_d3(pdev, wake);
9322                 pci_set_power_state(pdev, PCI_D3hot);
9323         }
9324 }
9325
9326 #ifdef CONFIG_PCI_IOV
9327 static int igb_sriov_reinit(struct pci_dev *dev)
9328 {
9329         struct net_device *netdev = pci_get_drvdata(dev);
9330         struct igb_adapter *adapter = netdev_priv(netdev);
9331         struct pci_dev *pdev = adapter->pdev;
9332
9333         rtnl_lock();
9334
9335         if (netif_running(netdev))
9336                 igb_close(netdev);
9337         else
9338                 igb_reset(adapter);
9339
9340         igb_clear_interrupt_scheme(adapter);
9341
9342         igb_init_queue_configuration(adapter);
9343
9344         if (igb_init_interrupt_scheme(adapter, true)) {
9345                 rtnl_unlock();
9346                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9347                 return -ENOMEM;
9348         }
9349
9350         if (netif_running(netdev))
9351                 igb_open(netdev);
9352
9353         rtnl_unlock();
9354
9355         return 0;
9356 }
9357
9358 static int igb_pci_disable_sriov(struct pci_dev *dev)
9359 {
9360         int err = igb_disable_sriov(dev);
9361
9362         if (!err)
9363                 err = igb_sriov_reinit(dev);
9364
9365         return err;
9366 }
9367
9368 static int igb_pci_enable_sriov(struct pci_dev *dev, int num_vfs)
9369 {
9370         int err = igb_enable_sriov(dev, num_vfs);
9371
9372         if (err)
9373                 goto out;
9374
9375         err = igb_sriov_reinit(dev);
9376         if (!err)
9377                 return num_vfs;
9378
9379 out:
9380         return err;
9381 }
9382
9383 #endif
9384 static int igb_pci_sriov_configure(struct pci_dev *dev, int num_vfs)
9385 {
9386 #ifdef CONFIG_PCI_IOV
9387         if (num_vfs == 0)
9388                 return igb_pci_disable_sriov(dev);
9389         else
9390                 return igb_pci_enable_sriov(dev, num_vfs);
9391 #endif
9392         return 0;
9393 }
9394
9395 /**
9396  *  igb_io_error_detected - called when PCI error is detected
9397  *  @pdev: Pointer to PCI device
9398  *  @state: The current pci connection state
9399  *
9400  *  This function is called after a PCI bus error affecting
9401  *  this device has been detected.
9402  **/
9403 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
9404                                               pci_channel_state_t state)
9405 {
9406         struct net_device *netdev = pci_get_drvdata(pdev);
9407         struct igb_adapter *adapter = netdev_priv(netdev);
9408
9409         netif_device_detach(netdev);
9410
9411         if (state == pci_channel_io_perm_failure)
9412                 return PCI_ERS_RESULT_DISCONNECT;
9413
9414         if (netif_running(netdev))
9415                 igb_down(adapter);
9416         pci_disable_device(pdev);
9417
9418         /* Request a slot slot reset. */
9419         return PCI_ERS_RESULT_NEED_RESET;
9420 }
9421
9422 /**
9423  *  igb_io_slot_reset - called after the pci bus has been reset.
9424  *  @pdev: Pointer to PCI device
9425  *
9426  *  Restart the card from scratch, as if from a cold-boot. Implementation
9427  *  resembles the first-half of the igb_resume routine.
9428  **/
9429 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
9430 {
9431         struct net_device *netdev = pci_get_drvdata(pdev);
9432         struct igb_adapter *adapter = netdev_priv(netdev);
9433         struct e1000_hw *hw = &adapter->hw;
9434         pci_ers_result_t result;
9435
9436         if (pci_enable_device_mem(pdev)) {
9437                 dev_err(&pdev->dev,
9438                         "Cannot re-enable PCI device after reset.\n");
9439                 result = PCI_ERS_RESULT_DISCONNECT;
9440         } else {
9441                 pci_set_master(pdev);
9442                 pci_restore_state(pdev);
9443                 pci_save_state(pdev);
9444
9445                 pci_enable_wake(pdev, PCI_D3hot, 0);
9446                 pci_enable_wake(pdev, PCI_D3cold, 0);
9447
9448                 /* In case of PCI error, adapter lose its HW address
9449                  * so we should re-assign it here.
9450                  */
9451                 hw->hw_addr = adapter->io_addr;
9452
9453                 igb_reset(adapter);
9454                 wr32(E1000_WUS, ~0);
9455                 result = PCI_ERS_RESULT_RECOVERED;
9456         }
9457
9458         return result;
9459 }
9460
9461 /**
9462  *  igb_io_resume - called when traffic can start flowing again.
9463  *  @pdev: Pointer to PCI device
9464  *
9465  *  This callback is called when the error recovery driver tells us that
9466  *  its OK to resume normal operation. Implementation resembles the
9467  *  second-half of the igb_resume routine.
9468  */
9469 static void igb_io_resume(struct pci_dev *pdev)
9470 {
9471         struct net_device *netdev = pci_get_drvdata(pdev);
9472         struct igb_adapter *adapter = netdev_priv(netdev);
9473
9474         if (netif_running(netdev)) {
9475                 if (igb_up(adapter)) {
9476                         dev_err(&pdev->dev, "igb_up failed after reset\n");
9477                         return;
9478                 }
9479         }
9480
9481         netif_device_attach(netdev);
9482
9483         /* let the f/w know that the h/w is now under the control of the
9484          * driver.
9485          */
9486         igb_get_hw_control(adapter);
9487 }
9488
9489 /**
9490  *  igb_rar_set_index - Sync RAL[index] and RAH[index] registers with MAC table
9491  *  @adapter: Pointer to adapter structure
9492  *  @index: Index of the RAR entry which need to be synced with MAC table
9493  **/
9494 static void igb_rar_set_index(struct igb_adapter *adapter, u32 index)
9495 {
9496         struct e1000_hw *hw = &adapter->hw;
9497         u32 rar_low, rar_high;
9498         u8 *addr = adapter->mac_table[index].addr;
9499
9500         /* HW expects these to be in network order when they are plugged
9501          * into the registers which are little endian.  In order to guarantee
9502          * that ordering we need to do an leXX_to_cpup here in order to be
9503          * ready for the byteswap that occurs with writel
9504          */
9505         rar_low = le32_to_cpup((__le32 *)(addr));
9506         rar_high = le16_to_cpup((__le16 *)(addr + 4));
9507
9508         /* Indicate to hardware the Address is Valid. */
9509         if (adapter->mac_table[index].state & IGB_MAC_STATE_IN_USE) {
9510                 if (is_valid_ether_addr(addr))
9511                         rar_high |= E1000_RAH_AV;
9512
9513                 if (adapter->mac_table[index].state & IGB_MAC_STATE_SRC_ADDR)
9514                         rar_high |= E1000_RAH_ASEL_SRC_ADDR;
9515
9516                 switch (hw->mac.type) {
9517                 case e1000_82575:
9518                 case e1000_i210:
9519                         if (adapter->mac_table[index].state &
9520                             IGB_MAC_STATE_QUEUE_STEERING)
9521                                 rar_high |= E1000_RAH_QSEL_ENABLE;
9522
9523                         rar_high |= E1000_RAH_POOL_1 *
9524                                     adapter->mac_table[index].queue;
9525                         break;
9526                 default:
9527                         rar_high |= E1000_RAH_POOL_1 <<
9528                                     adapter->mac_table[index].queue;
9529                         break;
9530                 }
9531         }
9532
9533         wr32(E1000_RAL(index), rar_low);
9534         wrfl();
9535         wr32(E1000_RAH(index), rar_high);
9536         wrfl();
9537 }
9538
9539 static int igb_set_vf_mac(struct igb_adapter *adapter,
9540                           int vf, unsigned char *mac_addr)
9541 {
9542         struct e1000_hw *hw = &adapter->hw;
9543         /* VF MAC addresses start at end of receive addresses and moves
9544          * towards the first, as a result a collision should not be possible
9545          */
9546         int rar_entry = hw->mac.rar_entry_count - (vf + 1);
9547         unsigned char *vf_mac_addr = adapter->vf_data[vf].vf_mac_addresses;
9548
9549         ether_addr_copy(vf_mac_addr, mac_addr);
9550         ether_addr_copy(adapter->mac_table[rar_entry].addr, mac_addr);
9551         adapter->mac_table[rar_entry].queue = vf;
9552         adapter->mac_table[rar_entry].state |= IGB_MAC_STATE_IN_USE;
9553         igb_rar_set_index(adapter, rar_entry);
9554
9555         return 0;
9556 }
9557
9558 static int igb_ndo_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
9559 {
9560         struct igb_adapter *adapter = netdev_priv(netdev);
9561
9562         if (vf >= adapter->vfs_allocated_count)
9563                 return -EINVAL;
9564
9565         /* Setting the VF MAC to 0 reverts the IGB_VF_FLAG_PF_SET_MAC
9566          * flag and allows to overwrite the MAC via VF netdev.  This
9567          * is necessary to allow libvirt a way to restore the original
9568          * MAC after unbinding vfio-pci and reloading igbvf after shutting
9569          * down a VM.
9570          */
9571         if (is_zero_ether_addr(mac)) {
9572                 adapter->vf_data[vf].flags &= ~IGB_VF_FLAG_PF_SET_MAC;
9573                 dev_info(&adapter->pdev->dev,
9574                          "remove administratively set MAC on VF %d\n",
9575                          vf);
9576         } else if (is_valid_ether_addr(mac)) {
9577                 adapter->vf_data[vf].flags |= IGB_VF_FLAG_PF_SET_MAC;
9578                 dev_info(&adapter->pdev->dev, "setting MAC %pM on VF %d\n",
9579                          mac, vf);
9580                 dev_info(&adapter->pdev->dev,
9581                          "Reload the VF driver to make this change effective.");
9582                 /* Generate additional warning if PF is down */
9583                 if (test_bit(__IGB_DOWN, &adapter->state)) {
9584                         dev_warn(&adapter->pdev->dev,
9585                                  "The VF MAC address has been set, but the PF device is not up.\n");
9586                         dev_warn(&adapter->pdev->dev,
9587                                  "Bring the PF device up before attempting to use the VF device.\n");
9588                 }
9589         } else {
9590                 return -EINVAL;
9591         }
9592         return igb_set_vf_mac(adapter, vf, mac);
9593 }
9594
9595 static int igb_link_mbps(int internal_link_speed)
9596 {
9597         switch (internal_link_speed) {
9598         case SPEED_100:
9599                 return 100;
9600         case SPEED_1000:
9601                 return 1000;
9602         default:
9603                 return 0;
9604         }
9605 }
9606
9607 static void igb_set_vf_rate_limit(struct e1000_hw *hw, int vf, int tx_rate,
9608                                   int link_speed)
9609 {
9610         int rf_dec, rf_int;
9611         u32 bcnrc_val;
9612
9613         if (tx_rate != 0) {
9614                 /* Calculate the rate factor values to set */
9615                 rf_int = link_speed / tx_rate;
9616                 rf_dec = (link_speed - (rf_int * tx_rate));
9617                 rf_dec = (rf_dec * BIT(E1000_RTTBCNRC_RF_INT_SHIFT)) /
9618                          tx_rate;
9619
9620                 bcnrc_val = E1000_RTTBCNRC_RS_ENA;
9621                 bcnrc_val |= ((rf_int << E1000_RTTBCNRC_RF_INT_SHIFT) &
9622                               E1000_RTTBCNRC_RF_INT_MASK);
9623                 bcnrc_val |= (rf_dec & E1000_RTTBCNRC_RF_DEC_MASK);
9624         } else {
9625                 bcnrc_val = 0;
9626         }
9627
9628         wr32(E1000_RTTDQSEL, vf); /* vf X uses queue X */
9629         /* Set global transmit compensation time to the MMW_SIZE in RTTBCNRM
9630          * register. MMW_SIZE=0x014 if 9728-byte jumbo is supported.
9631          */
9632         wr32(E1000_RTTBCNRM, 0x14);
9633         wr32(E1000_RTTBCNRC, bcnrc_val);
9634 }
9635
9636 static void igb_check_vf_rate_limit(struct igb_adapter *adapter)
9637 {
9638         int actual_link_speed, i;
9639         bool reset_rate = false;
9640
9641         /* VF TX rate limit was not set or not supported */
9642         if ((adapter->vf_rate_link_speed == 0) ||
9643             (adapter->hw.mac.type != e1000_82576))
9644                 return;
9645
9646         actual_link_speed = igb_link_mbps(adapter->link_speed);
9647         if (actual_link_speed != adapter->vf_rate_link_speed) {
9648                 reset_rate = true;
9649                 adapter->vf_rate_link_speed = 0;
9650                 dev_info(&adapter->pdev->dev,
9651                          "Link speed has been changed. VF Transmit rate is disabled\n");
9652         }
9653
9654         for (i = 0; i < adapter->vfs_allocated_count; i++) {
9655                 if (reset_rate)
9656                         adapter->vf_data[i].tx_rate = 0;
9657
9658                 igb_set_vf_rate_limit(&adapter->hw, i,
9659                                       adapter->vf_data[i].tx_rate,
9660                                       actual_link_speed);
9661         }
9662 }
9663
9664 static int igb_ndo_set_vf_bw(struct net_device *netdev, int vf,
9665                              int min_tx_rate, int max_tx_rate)
9666 {
9667         struct igb_adapter *adapter = netdev_priv(netdev);
9668         struct e1000_hw *hw = &adapter->hw;
9669         int actual_link_speed;
9670
9671         if (hw->mac.type != e1000_82576)
9672                 return -EOPNOTSUPP;
9673
9674         if (min_tx_rate)
9675                 return -EINVAL;
9676
9677         actual_link_speed = igb_link_mbps(adapter->link_speed);
9678         if ((vf >= adapter->vfs_allocated_count) ||
9679             (!(rd32(E1000_STATUS) & E1000_STATUS_LU)) ||
9680             (max_tx_rate < 0) ||
9681             (max_tx_rate > actual_link_speed))
9682                 return -EINVAL;
9683
9684         adapter->vf_rate_link_speed = actual_link_speed;
9685         adapter->vf_data[vf].tx_rate = (u16)max_tx_rate;
9686         igb_set_vf_rate_limit(hw, vf, max_tx_rate, actual_link_speed);
9687
9688         return 0;
9689 }
9690
9691 static int igb_ndo_set_vf_spoofchk(struct net_device *netdev, int vf,
9692                                    bool setting)
9693 {
9694         struct igb_adapter *adapter = netdev_priv(netdev);
9695         struct e1000_hw *hw = &adapter->hw;
9696         u32 reg_val, reg_offset;
9697
9698         if (!adapter->vfs_allocated_count)
9699                 return -EOPNOTSUPP;
9700
9701         if (vf >= adapter->vfs_allocated_count)
9702                 return -EINVAL;
9703
9704         reg_offset = (hw->mac.type == e1000_82576) ? E1000_DTXSWC : E1000_TXSWC;
9705         reg_val = rd32(reg_offset);
9706         if (setting)
9707                 reg_val |= (BIT(vf) |
9708                             BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9709         else
9710                 reg_val &= ~(BIT(vf) |
9711                              BIT(vf + E1000_DTXSWC_VLAN_SPOOF_SHIFT));
9712         wr32(reg_offset, reg_val);
9713
9714         adapter->vf_data[vf].spoofchk_enabled = setting;
9715         return 0;
9716 }
9717
9718 static int igb_ndo_set_vf_trust(struct net_device *netdev, int vf, bool setting)
9719 {
9720         struct igb_adapter *adapter = netdev_priv(netdev);
9721
9722         if (vf >= adapter->vfs_allocated_count)
9723                 return -EINVAL;
9724         if (adapter->vf_data[vf].trusted == setting)
9725                 return 0;
9726
9727         adapter->vf_data[vf].trusted = setting;
9728
9729         dev_info(&adapter->pdev->dev, "VF %u is %strusted\n",
9730                  vf, setting ? "" : "not ");
9731         return 0;
9732 }
9733
9734 static int igb_ndo_get_vf_config(struct net_device *netdev,
9735                                  int vf, struct ifla_vf_info *ivi)
9736 {
9737         struct igb_adapter *adapter = netdev_priv(netdev);
9738         if (vf >= adapter->vfs_allocated_count)
9739                 return -EINVAL;
9740         ivi->vf = vf;
9741         memcpy(&ivi->mac, adapter->vf_data[vf].vf_mac_addresses, ETH_ALEN);
9742         ivi->max_tx_rate = adapter->vf_data[vf].tx_rate;
9743         ivi->min_tx_rate = 0;
9744         ivi->vlan = adapter->vf_data[vf].pf_vlan;
9745         ivi->qos = adapter->vf_data[vf].pf_qos;
9746         ivi->spoofchk = adapter->vf_data[vf].spoofchk_enabled;
9747         ivi->trusted = adapter->vf_data[vf].trusted;
9748         return 0;
9749 }
9750
9751 static void igb_vmm_control(struct igb_adapter *adapter)
9752 {
9753         struct e1000_hw *hw = &adapter->hw;
9754         u32 reg;
9755
9756         switch (hw->mac.type) {
9757         case e1000_82575:
9758         case e1000_i210:
9759         case e1000_i211:
9760         case e1000_i354:
9761         default:
9762                 /* replication is not supported for 82575 */
9763                 return;
9764         case e1000_82576:
9765                 /* notify HW that the MAC is adding vlan tags */
9766                 reg = rd32(E1000_DTXCTL);
9767                 reg |= E1000_DTXCTL_VLAN_ADDED;
9768                 wr32(E1000_DTXCTL, reg);
9769                 fallthrough;
9770         case e1000_82580:
9771                 /* enable replication vlan tag stripping */
9772                 reg = rd32(E1000_RPLOLR);
9773                 reg |= E1000_RPLOLR_STRVLAN;
9774                 wr32(E1000_RPLOLR, reg);
9775                 fallthrough;
9776         case e1000_i350:
9777                 /* none of the above registers are supported by i350 */
9778                 break;
9779         }
9780
9781         if (adapter->vfs_allocated_count) {
9782                 igb_vmdq_set_loopback_pf(hw, true);
9783                 igb_vmdq_set_replication_pf(hw, true);
9784                 igb_vmdq_set_anti_spoofing_pf(hw, true,
9785                                               adapter->vfs_allocated_count);
9786         } else {
9787                 igb_vmdq_set_loopback_pf(hw, false);
9788                 igb_vmdq_set_replication_pf(hw, false);
9789         }
9790 }
9791
9792 static void igb_init_dmac(struct igb_adapter *adapter, u32 pba)
9793 {
9794         struct e1000_hw *hw = &adapter->hw;
9795         u32 dmac_thr;
9796         u16 hwm;
9797
9798         if (hw->mac.type > e1000_82580) {
9799                 if (adapter->flags & IGB_FLAG_DMAC) {
9800                         u32 reg;
9801
9802                         /* force threshold to 0. */
9803                         wr32(E1000_DMCTXTH, 0);
9804
9805                         /* DMA Coalescing high water mark needs to be greater
9806                          * than the Rx threshold. Set hwm to PBA - max frame
9807                          * size in 16B units, capping it at PBA - 6KB.
9808                          */
9809                         hwm = 64 * (pba - 6);
9810                         reg = rd32(E1000_FCRTC);
9811                         reg &= ~E1000_FCRTC_RTH_COAL_MASK;
9812                         reg |= ((hwm << E1000_FCRTC_RTH_COAL_SHIFT)
9813                                 & E1000_FCRTC_RTH_COAL_MASK);
9814                         wr32(E1000_FCRTC, reg);
9815
9816                         /* Set the DMA Coalescing Rx threshold to PBA - 2 * max
9817                          * frame size, capping it at PBA - 10KB.
9818                          */
9819                         dmac_thr = pba - 10;
9820                         reg = rd32(E1000_DMACR);
9821                         reg &= ~E1000_DMACR_DMACTHR_MASK;
9822                         reg |= ((dmac_thr << E1000_DMACR_DMACTHR_SHIFT)
9823                                 & E1000_DMACR_DMACTHR_MASK);
9824
9825                         /* transition to L0x or L1 if available..*/
9826                         reg |= (E1000_DMACR_DMAC_EN | E1000_DMACR_DMAC_LX_MASK);
9827
9828                         /* watchdog timer= +-1000 usec in 32usec intervals */
9829                         reg |= (1000 >> 5);
9830
9831                         /* Disable BMC-to-OS Watchdog Enable */
9832                         if (hw->mac.type != e1000_i354)
9833                                 reg &= ~E1000_DMACR_DC_BMC2OSW_EN;
9834
9835                         wr32(E1000_DMACR, reg);
9836
9837                         /* no lower threshold to disable
9838                          * coalescing(smart fifb)-UTRESH=0
9839                          */
9840                         wr32(E1000_DMCRTRH, 0);
9841
9842                         reg = (IGB_DMCTLX_DCFLUSH_DIS | 0x4);
9843
9844                         wr32(E1000_DMCTLX, reg);
9845
9846                         /* free space in tx packet buffer to wake from
9847                          * DMA coal
9848                          */
9849                         wr32(E1000_DMCTXTH, (IGB_MIN_TXPBSIZE -
9850                              (IGB_TX_BUF_4096 + adapter->max_frame_size)) >> 6);
9851
9852                         /* make low power state decision controlled
9853                          * by DMA coal
9854                          */
9855                         reg = rd32(E1000_PCIEMISC);
9856                         reg &= ~E1000_PCIEMISC_LX_DECISION;
9857                         wr32(E1000_PCIEMISC, reg);
9858                 } /* endif adapter->dmac is not disabled */
9859         } else if (hw->mac.type == e1000_82580) {
9860                 u32 reg = rd32(E1000_PCIEMISC);
9861
9862                 wr32(E1000_PCIEMISC, reg & ~E1000_PCIEMISC_LX_DECISION);
9863                 wr32(E1000_DMACR, 0);
9864         }
9865 }
9866
9867 /**
9868  *  igb_read_i2c_byte - Reads 8 bit word over I2C
9869  *  @hw: pointer to hardware structure
9870  *  @byte_offset: byte offset to read
9871  *  @dev_addr: device address
9872  *  @data: value read
9873  *
9874  *  Performs byte read operation over I2C interface at
9875  *  a specified device address.
9876  **/
9877 s32 igb_read_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9878                       u8 dev_addr, u8 *data)
9879 {
9880         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9881         struct i2c_client *this_client = adapter->i2c_client;
9882         s32 status;
9883         u16 swfw_mask = 0;
9884
9885         if (!this_client)
9886                 return E1000_ERR_I2C;
9887
9888         swfw_mask = E1000_SWFW_PHY0_SM;
9889
9890         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9891                 return E1000_ERR_SWFW_SYNC;
9892
9893         status = i2c_smbus_read_byte_data(this_client, byte_offset);
9894         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9895
9896         if (status < 0)
9897                 return E1000_ERR_I2C;
9898         else {
9899                 *data = status;
9900                 return 0;
9901         }
9902 }
9903
9904 /**
9905  *  igb_write_i2c_byte - Writes 8 bit word over I2C
9906  *  @hw: pointer to hardware structure
9907  *  @byte_offset: byte offset to write
9908  *  @dev_addr: device address
9909  *  @data: value to write
9910  *
9911  *  Performs byte write operation over I2C interface at
9912  *  a specified device address.
9913  **/
9914 s32 igb_write_i2c_byte(struct e1000_hw *hw, u8 byte_offset,
9915                        u8 dev_addr, u8 data)
9916 {
9917         struct igb_adapter *adapter = container_of(hw, struct igb_adapter, hw);
9918         struct i2c_client *this_client = adapter->i2c_client;
9919         s32 status;
9920         u16 swfw_mask = E1000_SWFW_PHY0_SM;
9921
9922         if (!this_client)
9923                 return E1000_ERR_I2C;
9924
9925         if (hw->mac.ops.acquire_swfw_sync(hw, swfw_mask))
9926                 return E1000_ERR_SWFW_SYNC;
9927         status = i2c_smbus_write_byte_data(this_client, byte_offset, data);
9928         hw->mac.ops.release_swfw_sync(hw, swfw_mask);
9929
9930         if (status)
9931                 return E1000_ERR_I2C;
9932         else
9933                 return 0;
9934
9935 }
9936
9937 int igb_reinit_queues(struct igb_adapter *adapter)
9938 {
9939         struct net_device *netdev = adapter->netdev;
9940         struct pci_dev *pdev = adapter->pdev;
9941         int err = 0;
9942
9943         if (netif_running(netdev))
9944                 igb_close(netdev);
9945
9946         igb_reset_interrupt_capability(adapter);
9947
9948         if (igb_init_interrupt_scheme(adapter, true)) {
9949                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
9950                 return -ENOMEM;
9951         }
9952
9953         if (netif_running(netdev))
9954                 err = igb_open(netdev);
9955
9956         return err;
9957 }
9958
9959 static void igb_nfc_filter_exit(struct igb_adapter *adapter)
9960 {
9961         struct igb_nfc_filter *rule;
9962
9963         spin_lock(&adapter->nfc_lock);
9964
9965         hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9966                 igb_erase_filter(adapter, rule);
9967
9968         hlist_for_each_entry(rule, &adapter->cls_flower_list, nfc_node)
9969                 igb_erase_filter(adapter, rule);
9970
9971         spin_unlock(&adapter->nfc_lock);
9972 }
9973
9974 static void igb_nfc_filter_restore(struct igb_adapter *adapter)
9975 {
9976         struct igb_nfc_filter *rule;
9977
9978         spin_lock(&adapter->nfc_lock);
9979
9980         hlist_for_each_entry(rule, &adapter->nfc_filter_list, nfc_node)
9981                 igb_add_filter(adapter, rule);
9982
9983         spin_unlock(&adapter->nfc_lock);
9984 }
9985 /* igb_main.c */