drivers/net/ixgb/ixgb_main.c

   1 /*******************************************************************************
   2
   3   Intel PRO/10GbE Linux driver
   4   Copyright(c) 1999 - 2008 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include "ixgb.h"
  30
  31 char ixgb_driver_name[] = "ixgb";
  32 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
  33
  34 #define DRIVERNAPI "-NAPI"
  35 #define DRV_VERSION "1.0.135-k2" DRIVERNAPI
  36 const char ixgb_driver_version[] = DRV_VERSION;
  37 static const char ixgb_copyright[] = "Copyright (c) 1999-2008 Intel Corporation.";
  38
  39 #define IXGB_CB_LENGTH 256
  40 static unsigned int copybreak __read_mostly = IXGB_CB_LENGTH;
  41 module_param(copybreak, uint, 0644);
  42 MODULE_PARM_DESC(copybreak,
  43         "Maximum size of packet that is copied to a new buffer on receive");
  44
  45 /* ixgb_pci_tbl - PCI Device ID Table
  46  *
  47  * Wildcard entries (PCI_ANY_ID) should come last
  48  * Last entry must be all 0s
  49  *
  50  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  51  *   Class, Class Mask, private data (not used) }
  52  */
  53 static DEFINE_PCI_DEVICE_TABLE(ixgb_pci_tbl) = {
  54         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX,
  55          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  56         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_CX4,
  57          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  58         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_SR,
  59          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  60         {INTEL_VENDOR_ID, IXGB_DEVICE_ID_82597EX_LR,
  61          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  62
  63         /* required last entry */
  64         {0,}
  65 };
  66
  67 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
  68
  69 /* Local Function Prototypes */
  70 static int ixgb_init_module(void);
  71 static void ixgb_exit_module(void);
  72 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
  73 static void __devexit ixgb_remove(struct pci_dev *pdev);
  74 static int ixgb_sw_init(struct ixgb_adapter *adapter);
  75 static int ixgb_open(struct net_device *netdev);
  76 static int ixgb_close(struct net_device *netdev);
  77 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
  78 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
  79 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
  80 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
  81 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
  82 static void ixgb_set_multi(struct net_device *netdev);
  83 static void ixgb_watchdog(unsigned long data);
  84 static netdev_tx_t ixgb_xmit_frame(struct sk_buff *skb,
  85                                    struct net_device *netdev);
  86 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
  87 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
  88 static int ixgb_set_mac(struct net_device *netdev, void *p);
  89 static irqreturn_t ixgb_intr(int irq, void *data);
  90 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
  91
  92 static int ixgb_clean(struct napi_struct *, int);
  93 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
  94 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
  95
  96 static void ixgb_tx_timeout(struct net_device *dev);
  97 static void ixgb_tx_timeout_task(struct work_struct *work);
  98
  99 static void ixgb_vlan_rx_register(struct net_device *netdev,
 100                                   struct vlan_group *grp);
 101 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
 102 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
 103 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
 104
 105 #ifdef CONFIG_NET_POLL_CONTROLLER
 106 /* for netdump / net console */
 107 static void ixgb_netpoll(struct net_device *dev);
 108 #endif
 109
 110 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
 111                              enum pci_channel_state state);
 112 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
 113 static void ixgb_io_resume (struct pci_dev *pdev);
 114
 115 static struct pci_error_handlers ixgb_err_handler = {
 116         .error_detected = ixgb_io_error_detected,
 117         .slot_reset = ixgb_io_slot_reset,
 118         .resume = ixgb_io_resume,
 119 };
 120
 121 static struct pci_driver ixgb_driver = {
 122         .name     = ixgb_driver_name,
 123         .id_table = ixgb_pci_tbl,
 124         .probe    = ixgb_probe,
 125         .remove   = __devexit_p(ixgb_remove),
 126         .err_handler = &ixgb_err_handler
 127 };
 128
 129 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 130 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
 131 MODULE_LICENSE("GPL");
 132 MODULE_VERSION(DRV_VERSION);
 133
 134 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
 135 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
 136 module_param(debug, int, 0);
 137 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 138
 139 /**
 140  * ixgb_init_module - Driver Registration Routine
 141  *
 142  * ixgb_init_module is the first routine called when the driver is
 143  * loaded. All it does is register with the PCI subsystem.
 144  **/
 145
 146 static int __init
 147 ixgb_init_module(void)
 148 {
 149         printk(KERN_INFO "%s - version %s\n",
 150                ixgb_driver_string, ixgb_driver_version);
 151
 152         printk(KERN_INFO "%s\n", ixgb_copyright);
 153
 154         return pci_register_driver(&ixgb_driver);
 155 }
 156
 157 module_init(ixgb_init_module);
 158
 159 /**
 160  * ixgb_exit_module - Driver Exit Cleanup Routine
 161  *
 162  * ixgb_exit_module is called just before the driver is removed
 163  * from memory.
 164  **/
 165
 166 static void __exit
 167 ixgb_exit_module(void)
 168 {
 169         pci_unregister_driver(&ixgb_driver);
 170 }
 171
 172 module_exit(ixgb_exit_module);
 173
 174 /**
 175  * ixgb_irq_disable - Mask off interrupt generation on the NIC
 176  * @adapter: board private structure
 177  **/
 178
 179 static void
 180 ixgb_irq_disable(struct ixgb_adapter *adapter)
 181 {
 182         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
 183         IXGB_WRITE_FLUSH(&adapter->hw);
 184         synchronize_irq(adapter->pdev->irq);
 185 }
 186
 187 /**
 188  * ixgb_irq_enable - Enable default interrupt generation settings
 189  * @adapter: board private structure
 190  **/
 191
 192 static void
 193 ixgb_irq_enable(struct ixgb_adapter *adapter)
 194 {
 195         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
 196                   IXGB_INT_TXDW | IXGB_INT_LSC;
 197         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
 198                 val |= IXGB_INT_GPI0;
 199         IXGB_WRITE_REG(&adapter->hw, IMS, val);
 200         IXGB_WRITE_FLUSH(&adapter->hw);
 201 }
 202
 203 int
 204 ixgb_up(struct ixgb_adapter *adapter)
 205 {
 206         struct net_device *netdev = adapter->netdev;
 207         int err, irq_flags = IRQF_SHARED;
 208         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 209         struct ixgb_hw *hw = &adapter->hw;
 210
 211         /* hardware has been reset, we need to reload some things */
 212
 213         ixgb_rar_set(hw, netdev->dev_addr, 0);
 214         ixgb_set_multi(netdev);
 215
 216         ixgb_restore_vlan(adapter);
 217
 218         ixgb_configure_tx(adapter);
 219         ixgb_setup_rctl(adapter);
 220         ixgb_configure_rx(adapter);
 221         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
 222
 223         /* disable interrupts and get the hardware into a known state */
 224         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
 225
 226         /* only enable MSI if bus is in PCI-X mode */
 227         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
 228                 err = pci_enable_msi(adapter->pdev);
 229                 if (!err) {
 230                         adapter->have_msi = 1;
 231                         irq_flags = 0;
 232                 }
 233                 /* proceed to try to request regular interrupt */
 234         }
 235
 236         err = request_irq(adapter->pdev->irq, ixgb_intr, irq_flags,
 237                           netdev->name, netdev);
 238         if (err) {
 239                 if (adapter->have_msi)
 240                         pci_disable_msi(adapter->pdev);
 241                 netif_err(adapter, probe, adapter->netdev,
 242                           "Unable to allocate interrupt Error: %d\n", err);
 243                 return err;
 244         }
 245
 246         if ((hw->max_frame_size != max_frame) ||
 247                 (hw->max_frame_size !=
 248                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
 249
 250                 hw->max_frame_size = max_frame;
 251
 252                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
 253
 254                 if (hw->max_frame_size >
 255                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
 256                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
 257
 258                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
 259                                 ctrl0 |= IXGB_CTRL0_JFE;
 260                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
 261                         }
 262                 }
 263         }
 264
 265         clear_bit(__IXGB_DOWN, &adapter->flags);
 266
 267         napi_enable(&adapter->napi);
 268         ixgb_irq_enable(adapter);
 269
 270         netif_wake_queue(netdev);
 271
 272         mod_timer(&adapter->watchdog_timer, jiffies);
 273
 274         return 0;
 275 }
 276
 277 void
 278 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
 279 {
 280         struct net_device *netdev = adapter->netdev;
 281
 282         /* prevent the interrupt handler from restarting watchdog */
 283         set_bit(__IXGB_DOWN, &adapter->flags);
 284
 285         napi_disable(&adapter->napi);
 286         /* waiting for NAPI to complete can re-enable interrupts */
 287         ixgb_irq_disable(adapter);
 288         free_irq(adapter->pdev->irq, netdev);
 289
 290         if (adapter->have_msi)
 291                 pci_disable_msi(adapter->pdev);
 292
 293         if (kill_watchdog)
 294                 del_timer_sync(&adapter->watchdog_timer);
 295
 296         adapter->link_speed = 0;
 297         adapter->link_duplex = 0;
 298         netif_carrier_off(netdev);
 299         netif_stop_queue(netdev);
 300
 301         ixgb_reset(adapter);
 302         ixgb_clean_tx_ring(adapter);
 303         ixgb_clean_rx_ring(adapter);
 304 }
 305
 306 void
 307 ixgb_reset(struct ixgb_adapter *adapter)
 308 {
 309         struct ixgb_hw *hw = &adapter->hw;
 310
 311         ixgb_adapter_stop(hw);
 312         if (!ixgb_init_hw(hw))
 313                 netif_err(adapter, probe, adapter->netdev, "ixgb_init_hw failed\n");
 314
 315         /* restore frame size information */
 316         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
 317         if (hw->max_frame_size >
 318             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
 319                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
 320                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
 321                         ctrl0 |= IXGB_CTRL0_JFE;
 322                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
 323                 }
 324         }
 325 }
 326
 327 static const struct net_device_ops ixgb_netdev_ops = {
 328         .ndo_open               = ixgb_open,
 329         .ndo_stop               = ixgb_close,
 330         .ndo_start_xmit         = ixgb_xmit_frame,
 331         .ndo_get_stats          = ixgb_get_stats,
 332         .ndo_set_multicast_list = ixgb_set_multi,
 333         .ndo_validate_addr      = eth_validate_addr,
 334         .ndo_set_mac_address    = ixgb_set_mac,
 335         .ndo_change_mtu         = ixgb_change_mtu,
 336         .ndo_tx_timeout         = ixgb_tx_timeout,
 337         .ndo_vlan_rx_register   = ixgb_vlan_rx_register,
 338         .ndo_vlan_rx_add_vid    = ixgb_vlan_rx_add_vid,
 339         .ndo_vlan_rx_kill_vid   = ixgb_vlan_rx_kill_vid,
 340 #ifdef CONFIG_NET_POLL_CONTROLLER
 341         .ndo_poll_controller    = ixgb_netpoll,
 342 #endif
 343 };
 344
 345 /**
 346  * ixgb_probe - Device Initialization Routine
 347  * @pdev: PCI device information struct
 348  * @ent: entry in ixgb_pci_tbl
 349  *
 350  * Returns 0 on success, negative on failure
 351  *
 352  * ixgb_probe initializes an adapter identified by a pci_dev structure.
 353  * The OS initialization, configuring of the adapter private structure,
 354  * and a hardware reset occur.
 355  **/
 356
 357 static int __devinit
 358 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 359 {
 360         struct net_device *netdev = NULL;
 361         struct ixgb_adapter *adapter;
 362         static int cards_found = 0;
 363         int pci_using_dac;
 364         int i;
 365         int err;
 366
 367         err = pci_enable_device(pdev);
 368         if (err)
 369                 return err;
 370
 371         if (!(err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) &&
 372             !(err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)))) {
 373                 pci_using_dac = 1;
 374         } else {
 375                 if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) ||
 376                     (err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)))) {
 377                         printk(KERN_ERR
 378                          "ixgb: No usable DMA configuration, aborting\n");
 379                         goto err_dma_mask;
 380                 }
 381                 pci_using_dac = 0;
 382         }
 383
 384         err = pci_request_regions(pdev, ixgb_driver_name);
 385         if (err)
 386                 goto err_request_regions;
 387
 388         pci_set_master(pdev);
 389
 390         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
 391         if (!netdev) {
 392                 err = -ENOMEM;
 393                 goto err_alloc_etherdev;
 394         }
 395
 396         SET_NETDEV_DEV(netdev, &pdev->dev);
 397
 398         pci_set_drvdata(pdev, netdev);
 399         adapter = netdev_priv(netdev);
 400         adapter->netdev = netdev;
 401         adapter->pdev = pdev;
 402         adapter->hw.back = adapter;
 403         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
 404
 405         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
 406         if (!adapter->hw.hw_addr) {
 407                 err = -EIO;
 408                 goto err_ioremap;
 409         }
 410
 411         for (i = BAR_1; i <= BAR_5; i++) {
 412                 if (pci_resource_len(pdev, i) == 0)
 413                         continue;
 414                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
 415                         adapter->hw.io_base = pci_resource_start(pdev, i);
 416                         break;
 417                 }
 418         }
 419
 420         netdev->netdev_ops = &ixgb_netdev_ops;
 421         ixgb_set_ethtool_ops(netdev);
 422         netdev->watchdog_timeo = 5 * HZ;
 423         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
 424
 425         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 426
 427         adapter->bd_number = cards_found;
 428         adapter->link_speed = 0;
 429         adapter->link_duplex = 0;
 430
 431         /* setup the private structure */
 432
 433         err = ixgb_sw_init(adapter);
 434         if (err)
 435                 goto err_sw_init;
 436
 437         netdev->features = NETIF_F_SG |
 438                            NETIF_F_HW_CSUM |
 439                            NETIF_F_HW_VLAN_TX |
 440                            NETIF_F_HW_VLAN_RX |
 441                            NETIF_F_HW_VLAN_FILTER;
 442         netdev->features |= NETIF_F_TSO;
 443
 444         if (pci_using_dac)
 445                 netdev->features |= NETIF_F_HIGHDMA;
 446
 447         /* make sure the EEPROM is good */
 448
 449         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
 450                 netif_err(adapter, probe, adapter->netdev,
 451                           "The EEPROM Checksum Is Not Valid\n");
 452                 err = -EIO;
 453                 goto err_eeprom;
 454         }
 455
 456         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
 457         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
 458
 459         if (!is_valid_ether_addr(netdev->perm_addr)) {
 460                 netif_err(adapter, probe, adapter->netdev, "Invalid MAC Address\n");
 461                 err = -EIO;
 462                 goto err_eeprom;
 463         }
 464
 465         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
 466
 467         init_timer(&adapter->watchdog_timer);
 468         adapter->watchdog_timer.function = &ixgb_watchdog;
 469         adapter->watchdog_timer.data = (unsigned long)adapter;
 470
 471         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
 472
 473         strcpy(netdev->name, "eth%d");
 474         err = register_netdev(netdev);
 475         if (err)
 476                 goto err_register;
 477
 478         /* carrier off reporting is important to ethtool even BEFORE open */
 479         netif_carrier_off(netdev);
 480
 481         netif_info(adapter, probe, adapter->netdev,
 482                    "Intel(R) PRO/10GbE Network Connection\n");
 483         ixgb_check_options(adapter);
 484         /* reset the hardware with the new settings */
 485
 486         ixgb_reset(adapter);
 487
 488         cards_found++;
 489         return 0;
 490
 491 err_register:
 492 err_sw_init:
 493 err_eeprom:
 494         iounmap(adapter->hw.hw_addr);
 495 err_ioremap:
 496         free_netdev(netdev);
 497 err_alloc_etherdev:
 498         pci_release_regions(pdev);
 499 err_request_regions:
 500 err_dma_mask:
 501         pci_disable_device(pdev);
 502         return err;
 503 }
 504
 505 /**
 506  * ixgb_remove - Device Removal Routine
 507  * @pdev: PCI device information struct
 508  *
 509  * ixgb_remove is called by the PCI subsystem to alert the driver
 510  * that it should release a PCI device.  The could be caused by a
 511  * Hot-Plug event, or because the driver is going to be removed from
 512  * memory.
 513  **/
 514
 515 static void __devexit
 516 ixgb_remove(struct pci_dev *pdev)
 517 {
 518         struct net_device *netdev = pci_get_drvdata(pdev);
 519         struct ixgb_adapter *adapter = netdev_priv(netdev);
 520
 521         flush_scheduled_work();
 522
 523         unregister_netdev(netdev);
 524
 525         iounmap(adapter->hw.hw_addr);
 526         pci_release_regions(pdev);
 527
 528         free_netdev(netdev);
 529 }
 530
 531 /**
 532  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
 533  * @adapter: board private structure to initialize
 534  *
 535  * ixgb_sw_init initializes the Adapter private data structure.
 536  * Fields are initialized based on PCI device information and
 537  * OS network device settings (MTU size).
 538  **/
 539
 540 static int __devinit
 541 ixgb_sw_init(struct ixgb_adapter *adapter)
 542 {
 543         struct ixgb_hw *hw = &adapter->hw;
 544         struct net_device *netdev = adapter->netdev;
 545         struct pci_dev *pdev = adapter->pdev;
 546
 547         /* PCI config space info */
 548
 549         hw->vendor_id = pdev->vendor;
 550         hw->device_id = pdev->device;
 551         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 552         hw->subsystem_id = pdev->subsystem_device;
 553
 554         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 555         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
 556
 557         if ((hw->device_id == IXGB_DEVICE_ID_82597EX) ||
 558             (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4) ||
 559             (hw->device_id == IXGB_DEVICE_ID_82597EX_LR) ||
 560             (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
 561                 hw->mac_type = ixgb_82597;
 562         else {
 563                 /* should never have loaded on this device */
 564                 netif_err(adapter, probe, adapter->netdev, "unsupported device id\n");
 565         }
 566
 567         /* enable flow control to be programmed */
 568         hw->fc.send_xon = 1;
 569
 570         set_bit(__IXGB_DOWN, &adapter->flags);
 571         return 0;
 572 }
 573
 574 /**
 575  * ixgb_open - Called when a network interface is made active
 576  * @netdev: network interface device structure
 577  *
 578  * Returns 0 on success, negative value on failure
 579  *
 580  * The open entry point is called when a network interface is made
 581  * active by the system (IFF_UP).  At this point all resources needed
 582  * for transmit and receive operations are allocated, the interrupt
 583  * handler is registered with the OS, the watchdog timer is started,
 584  * and the stack is notified that the interface is ready.
 585  **/
 586
 587 static int
 588 ixgb_open(struct net_device *netdev)
 589 {
 590         struct ixgb_adapter *adapter = netdev_priv(netdev);
 591         int err;
 592
 593         /* allocate transmit descriptors */
 594         err = ixgb_setup_tx_resources(adapter);
 595         if (err)
 596                 goto err_setup_tx;
 597
 598         netif_carrier_off(netdev);
 599
 600         /* allocate receive descriptors */
 601
 602         err = ixgb_setup_rx_resources(adapter);
 603         if (err)
 604                 goto err_setup_rx;
 605
 606         err = ixgb_up(adapter);
 607         if (err)
 608                 goto err_up;
 609
 610         netif_start_queue(netdev);
 611
 612         return 0;
 613
 614 err_up:
 615         ixgb_free_rx_resources(adapter);
 616 err_setup_rx:
 617         ixgb_free_tx_resources(adapter);
 618 err_setup_tx:
 619         ixgb_reset(adapter);
 620
 621         return err;
 622 }
 623
 624 /**
 625  * ixgb_close - Disables a network interface
 626  * @netdev: network interface device structure
 627  *
 628  * Returns 0, this is not allowed to fail
 629  *
 630  * The close entry point is called when an interface is de-activated
 631  * by the OS.  The hardware is still under the drivers control, but
 632  * needs to be disabled.  A global MAC reset is issued to stop the
 633  * hardware, and all transmit and receive resources are freed.
 634  **/
 635
 636 static int
 637 ixgb_close(struct net_device *netdev)
 638 {
 639         struct ixgb_adapter *adapter = netdev_priv(netdev);
 640
 641         ixgb_down(adapter, true);
 642
 643         ixgb_free_tx_resources(adapter);
 644         ixgb_free_rx_resources(adapter);
 645
 646         return 0;
 647 }
 648
 649 /**
 650  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
 651  * @adapter: board private structure
 652  *
 653  * Return 0 on success, negative on failure
 654  **/
 655
 656 int
 657 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
 658 {
 659         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
 660         struct pci_dev *pdev = adapter->pdev;
 661         int size;
 662
 663         size = sizeof(struct ixgb_buffer) * txdr->count;
 664         txdr->buffer_info = vmalloc(size);
 665         if (!txdr->buffer_info) {
 666                 netif_err(adapter, probe, adapter->netdev,
 667                           "Unable to allocate transmit descriptor ring memory\n");
 668                 return -ENOMEM;
 669         }
 670         memset(txdr->buffer_info, 0, size);
 671
 672         /* round up to nearest 4K */
 673
 674         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
 675         txdr->size = ALIGN(txdr->size, 4096);
 676
 677         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
 678         if (!txdr->desc) {
 679                 vfree(txdr->buffer_info);
 680                 netif_err(adapter, probe, adapter->netdev,
 681                           "Unable to allocate transmit descriptor memory\n");
 682                 return -ENOMEM;
 683         }
 684         memset(txdr->desc, 0, txdr->size);
 685
 686         txdr->next_to_use = 0;
 687         txdr->next_to_clean = 0;
 688
 689         return 0;
 690 }
 691
 692 /**
 693  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
 694  * @adapter: board private structure
 695  *
 696  * Configure the Tx unit of the MAC after a reset.
 697  **/
 698
 699 static void
 700 ixgb_configure_tx(struct ixgb_adapter *adapter)
 701 {
 702         u64 tdba = adapter->tx_ring.dma;
 703         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
 704         u32 tctl;
 705         struct ixgb_hw *hw = &adapter->hw;
 706
 707         /* Setup the Base and Length of the Tx Descriptor Ring
 708          * tx_ring.dma can be either a 32 or 64 bit value
 709          */
 710
 711         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
 712         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
 713
 714         IXGB_WRITE_REG(hw, TDLEN, tdlen);
 715
 716         /* Setup the HW Tx Head and Tail descriptor pointers */
 717
 718         IXGB_WRITE_REG(hw, TDH, 0);
 719         IXGB_WRITE_REG(hw, TDT, 0);
 720
 721         /* don't set up txdctl, it induces performance problems if configured
 722          * incorrectly */
 723         /* Set the Tx Interrupt Delay register */
 724
 725         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
 726
 727         /* Program the Transmit Control Register */
 728
 729         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
 730         IXGB_WRITE_REG(hw, TCTL, tctl);
 731
 732         /* Setup Transmit Descriptor Settings for this adapter */
 733         adapter->tx_cmd_type =
 734                 IXGB_TX_DESC_TYPE |
 735                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
 736 }
 737
 738 /**
 739  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
 740  * @adapter: board private structure
 741  *
 742  * Returns 0 on success, negative on failure
 743  **/
 744
 745 int
 746 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
 747 {
 748         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
 749         struct pci_dev *pdev = adapter->pdev;
 750         int size;
 751
 752         size = sizeof(struct ixgb_buffer) * rxdr->count;
 753         rxdr->buffer_info = vmalloc(size);
 754         if (!rxdr->buffer_info) {
 755                 netif_err(adapter, probe, adapter->netdev,
 756                           "Unable to allocate receive descriptor ring\n");
 757                 return -ENOMEM;
 758         }
 759         memset(rxdr->buffer_info, 0, size);
 760
 761         /* Round up to nearest 4K */
 762
 763         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
 764         rxdr->size = ALIGN(rxdr->size, 4096);
 765
 766         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
 767
 768         if (!rxdr->desc) {
 769                 vfree(rxdr->buffer_info);
 770                 netif_err(adapter, probe, adapter->netdev,
 771                           "Unable to allocate receive descriptors\n");
 772                 return -ENOMEM;
 773         }
 774         memset(rxdr->desc, 0, rxdr->size);
 775
 776         rxdr->next_to_clean = 0;
 777         rxdr->next_to_use = 0;
 778
 779         return 0;
 780 }
 781
 782 /**
 783  * ixgb_setup_rctl - configure the receive control register
 784  * @adapter: Board private structure
 785  **/
 786
 787 static void
 788 ixgb_setup_rctl(struct ixgb_adapter *adapter)
 789 {
 790         u32 rctl;
 791
 792         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
 793
 794         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
 795
 796         rctl |=
 797                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
 798                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
 799                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
 800
 801         rctl |= IXGB_RCTL_SECRC;
 802
 803         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
 804                 rctl |= IXGB_RCTL_BSIZE_2048;
 805         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
 806                 rctl |= IXGB_RCTL_BSIZE_4096;
 807         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
 808                 rctl |= IXGB_RCTL_BSIZE_8192;
 809         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
 810                 rctl |= IXGB_RCTL_BSIZE_16384;
 811
 812         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
 813 }
 814
 815 /**
 816  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
 817  * @adapter: board private structure
 818  *
 819  * Configure the Rx unit of the MAC after a reset.
 820  **/
 821
 822 static void
 823 ixgb_configure_rx(struct ixgb_adapter *adapter)
 824 {
 825         u64 rdba = adapter->rx_ring.dma;
 826         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
 827         struct ixgb_hw *hw = &adapter->hw;
 828         u32 rctl;
 829         u32 rxcsum;
 830
 831         /* make sure receives are disabled while setting up the descriptors */
 832
 833         rctl = IXGB_READ_REG(hw, RCTL);
 834         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
 835
 836         /* set the Receive Delay Timer Register */
 837
 838         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
 839
 840         /* Setup the Base and Length of the Rx Descriptor Ring */
 841
 842         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
 843         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
 844
 845         IXGB_WRITE_REG(hw, RDLEN, rdlen);
 846
 847         /* Setup the HW Rx Head and Tail Descriptor Pointers */
 848         IXGB_WRITE_REG(hw, RDH, 0);
 849         IXGB_WRITE_REG(hw, RDT, 0);
 850
 851         /* due to the hardware errata with RXDCTL, we are unable to use any of
 852          * the performance enhancing features of it without causing other
 853          * subtle bugs, some of the bugs could include receive length
 854          * corruption at high data rates (WTHRESH > 0) and/or receive
 855          * descriptor ring irregularites (particularly in hardware cache) */
 856         IXGB_WRITE_REG(hw, RXDCTL, 0);
 857
 858         /* Enable Receive Checksum Offload for TCP and UDP */
 859         if (adapter->rx_csum) {
 860                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
 861                 rxcsum |= IXGB_RXCSUM_TUOFL;
 862                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
 863         }
 864
 865         /* Enable Receives */
 866
 867         IXGB_WRITE_REG(hw, RCTL, rctl);
 868 }
 869
 870 /**
 871  * ixgb_free_tx_resources - Free Tx Resources
 872  * @adapter: board private structure
 873  *
 874  * Free all transmit software resources
 875  **/
 876
 877 void
 878 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
 879 {
 880         struct pci_dev *pdev = adapter->pdev;
 881
 882         ixgb_clean_tx_ring(adapter);
 883
 884         vfree(adapter->tx_ring.buffer_info);
 885         adapter->tx_ring.buffer_info = NULL;
 886
 887         pci_free_consistent(pdev, adapter->tx_ring.size,
 888                             adapter->tx_ring.desc, adapter->tx_ring.dma);
 889
 890         adapter->tx_ring.desc = NULL;
 891 }
 892
 893 static void
 894 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
 895                                 struct ixgb_buffer *buffer_info)
 896 {
 897         if (buffer_info->dma) {
 898                 if (buffer_info->mapped_as_page)
 899                         pci_unmap_page(adapter->pdev, buffer_info->dma,
 900                                        buffer_info->length, PCI_DMA_TODEVICE);
 901                 else
 902                         pci_unmap_single(adapter->pdev, buffer_info->dma,
 903                                          buffer_info->length,
 904                                          PCI_DMA_TODEVICE);
 905                 buffer_info->dma = 0;
 906         }
 907
 908         if (buffer_info->skb) {
 909                 dev_kfree_skb_any(buffer_info->skb);
 910                 buffer_info->skb = NULL;
 911         }
 912         buffer_info->time_stamp = 0;
 913         /* these fields must always be initialized in tx
 914          * buffer_info->length = 0;
 915          * buffer_info->next_to_watch = 0; */
 916 }
 917
 918 /**
 919  * ixgb_clean_tx_ring - Free Tx Buffers
 920  * @adapter: board private structure
 921  **/
 922
 923 static void
 924 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
 925 {
 926         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
 927         struct ixgb_buffer *buffer_info;
 928         unsigned long size;
 929         unsigned int i;
 930
 931         /* Free all the Tx ring sk_buffs */
 932
 933         for (i = 0; i < tx_ring->count; i++) {
 934                 buffer_info = &tx_ring->buffer_info[i];
 935                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
 936         }
 937
 938         size = sizeof(struct ixgb_buffer) * tx_ring->count;
 939         memset(tx_ring->buffer_info, 0, size);
 940
 941         /* Zero out the descriptor ring */
 942
 943         memset(tx_ring->desc, 0, tx_ring->size);
 944
 945         tx_ring->next_to_use = 0;
 946         tx_ring->next_to_clean = 0;
 947
 948         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
 949         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
 950 }
 951
 952 /**
 953  * ixgb_free_rx_resources - Free Rx Resources
 954  * @adapter: board private structure
 955  *
 956  * Free all receive software resources
 957  **/
 958
 959 void
 960 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
 961 {
 962         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 963         struct pci_dev *pdev = adapter->pdev;
 964
 965         ixgb_clean_rx_ring(adapter);
 966
 967         vfree(rx_ring->buffer_info);
 968         rx_ring->buffer_info = NULL;
 969
 970         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
 971
 972         rx_ring->desc = NULL;
 973 }
 974
 975 /**
 976  * ixgb_clean_rx_ring - Free Rx Buffers
 977  * @adapter: board private structure
 978  **/
 979
 980 static void
 981 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
 982 {
 983         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 984         struct ixgb_buffer *buffer_info;
 985         struct pci_dev *pdev = adapter->pdev;
 986         unsigned long size;
 987         unsigned int i;
 988
 989         /* Free all the Rx ring sk_buffs */
 990
 991         for (i = 0; i < rx_ring->count; i++) {
 992                 buffer_info = &rx_ring->buffer_info[i];
 993                 if (buffer_info->dma) {
 994                         pci_unmap_single(pdev,
 995                                          buffer_info->dma,
 996                                          buffer_info->length,
 997                                          PCI_DMA_FROMDEVICE);
 998                         buffer_info->dma = 0;
 999                         buffer_info->length = 0;
1000                 }
1001
1002                 if (buffer_info->skb) {
1003                         dev_kfree_skb(buffer_info->skb);
1004                         buffer_info->skb = NULL;
1005                 }
1006         }
1007
1008         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1009         memset(rx_ring->buffer_info, 0, size);
1010
1011         /* Zero out the descriptor ring */
1012
1013         memset(rx_ring->desc, 0, rx_ring->size);
1014
1015         rx_ring->next_to_clean = 0;
1016         rx_ring->next_to_use = 0;
1017
1018         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1019         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1020 }
1021
1022 /**
1023  * ixgb_set_mac - Change the Ethernet Address of the NIC
1024  * @netdev: network interface device structure
1025  * @p: pointer to an address structure
1026  *
1027  * Returns 0 on success, negative on failure
1028  **/
1029
1030 static int
1031 ixgb_set_mac(struct net_device *netdev, void *p)
1032 {
1033         struct ixgb_adapter *adapter = netdev_priv(netdev);
1034         struct sockaddr *addr = p;
1035
1036         if (!is_valid_ether_addr(addr->sa_data))
1037                 return -EADDRNOTAVAIL;
1038
1039         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1040
1041         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1042
1043         return 0;
1044 }
1045
1046 /**
1047  * ixgb_set_multi - Multicast and Promiscuous mode set
1048  * @netdev: network interface device structure
1049  *
1050  * The set_multi entry point is called whenever the multicast address
1051  * list or the network interface flags are updated.  This routine is
1052  * responsible for configuring the hardware for proper multicast,
1053  * promiscuous mode, and all-multi behavior.
1054  **/
1055
1056 static void
1057 ixgb_set_multi(struct net_device *netdev)
1058 {
1059         struct ixgb_adapter *adapter = netdev_priv(netdev);
1060         struct ixgb_hw *hw = &adapter->hw;
1061         struct dev_mc_list *mc_ptr;
1062         u32 rctl;
1063         int i;
1064
1065         /* Check for Promiscuous and All Multicast modes */
1066
1067         rctl = IXGB_READ_REG(hw, RCTL);
1068
1069         if (netdev->flags & IFF_PROMISC) {
1070                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1071                 rctl &= ~IXGB_RCTL_VFE;
1072         } else {
1073                 if (netdev->flags & IFF_ALLMULTI) {
1074                         rctl |= IXGB_RCTL_MPE;
1075                         rctl &= ~IXGB_RCTL_UPE;
1076                 } else {
1077                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1078                 }
1079                 rctl |= IXGB_RCTL_VFE;
1080         }
1081
1082         if (netdev_mc_count(netdev) > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1083                 rctl |= IXGB_RCTL_MPE;
1084                 IXGB_WRITE_REG(hw, RCTL, rctl);
1085         } else {
1086                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1087                             IXGB_ETH_LENGTH_OF_ADDRESS];
1088
1089                 IXGB_WRITE_REG(hw, RCTL, rctl);
1090
1091                 for (i = 0, mc_ptr = netdev->mc_list;
1092                      mc_ptr;
1093                      i++, mc_ptr = mc_ptr->next)
1094                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1095                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1096
1097                 ixgb_mc_addr_list_update(hw, mta, netdev_mc_count(netdev), 0);
1098         }
1099 }
1100
1101 /**
1102  * ixgb_watchdog - Timer Call-back
1103  * @data: pointer to netdev cast into an unsigned long
1104  **/
1105
1106 static void
1107 ixgb_watchdog(unsigned long data)
1108 {
1109         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1110         struct net_device *netdev = adapter->netdev;
1111         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1112
1113         ixgb_check_for_link(&adapter->hw);
1114
1115         if (ixgb_check_for_bad_link(&adapter->hw)) {
1116                 /* force the reset path */
1117                 netif_stop_queue(netdev);
1118         }
1119
1120         if (adapter->hw.link_up) {
1121                 if (!netif_carrier_ok(netdev)) {
1122                         printk(KERN_INFO "ixgb: %s NIC Link is Up 10 Gbps "
1123                                "Full Duplex, Flow Control: %s\n",
1124                                netdev->name,
1125                                (adapter->hw.fc.type == ixgb_fc_full) ?
1126                                 "RX/TX" :
1127                                 ((adapter->hw.fc.type == ixgb_fc_rx_pause) ?
1128                                  "RX" :
1129                                  ((adapter->hw.fc.type == ixgb_fc_tx_pause) ?
1130                                   "TX" : "None")));
1131                         adapter->link_speed = 10000;
1132                         adapter->link_duplex = FULL_DUPLEX;
1133                         netif_carrier_on(netdev);
1134                 }
1135         } else {
1136                 if (netif_carrier_ok(netdev)) {
1137                         adapter->link_speed = 0;
1138                         adapter->link_duplex = 0;
1139                         printk(KERN_INFO "ixgb: %s NIC Link is Down\n",
1140                                netdev->name);
1141                         netif_carrier_off(netdev);
1142                 }
1143         }
1144
1145         ixgb_update_stats(adapter);
1146
1147         if (!netif_carrier_ok(netdev)) {
1148                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1149                         /* We've lost link, so the controller stops DMA,
1150                          * but we've got queued Tx work that's never going
1151                          * to get done, so reset controller to flush Tx.
1152                          * (Do the reset outside of interrupt context). */
1153                         schedule_work(&adapter->tx_timeout_task);
1154                         /* return immediately since reset is imminent */
1155                         return;
1156                 }
1157         }
1158
1159         /* Force detection of hung controller every watchdog period */
1160         adapter->detect_tx_hung = true;
1161
1162         /* generate an interrupt to force clean up of any stragglers */
1163         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1164
1165         /* Reset the timer */
1166         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1167 }
1168
1169 #define IXGB_TX_FLAGS_CSUM              0x00000001
1170 #define IXGB_TX_FLAGS_VLAN              0x00000002
1171 #define IXGB_TX_FLAGS_TSO               0x00000004
1172
1173 static int
1174 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1175 {
1176         struct ixgb_context_desc *context_desc;
1177         unsigned int i;
1178         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1179         u16 ipcse, tucse, mss;
1180         int err;
1181
1182         if (likely(skb_is_gso(skb))) {
1183                 struct ixgb_buffer *buffer_info;
1184                 struct iphdr *iph;
1185
1186                 if (skb_header_cloned(skb)) {
1187                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1188                         if (err)
1189                                 return err;
1190                 }
1191
1192                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1193                 mss = skb_shinfo(skb)->gso_size;
1194                 iph = ip_hdr(skb);
1195                 iph->tot_len = 0;
1196                 iph->check = 0;
1197                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1198                                                          iph->daddr, 0,
1199                                                          IPPROTO_TCP, 0);
1200                 ipcss = skb_network_offset(skb);
1201                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1202                 ipcse = skb_transport_offset(skb) - 1;
1203                 tucss = skb_transport_offset(skb);
1204                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1205                 tucse = 0;
1206
1207                 i = adapter->tx_ring.next_to_use;
1208                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1209                 buffer_info = &adapter->tx_ring.buffer_info[i];
1210                 WARN_ON(buffer_info->dma != 0);
1211
1212                 context_desc->ipcss = ipcss;
1213                 context_desc->ipcso = ipcso;
1214                 context_desc->ipcse = cpu_to_le16(ipcse);
1215                 context_desc->tucss = tucss;
1216                 context_desc->tucso = tucso;
1217                 context_desc->tucse = cpu_to_le16(tucse);
1218                 context_desc->mss = cpu_to_le16(mss);
1219                 context_desc->hdr_len = hdr_len;
1220                 context_desc->status = 0;
1221                 context_desc->cmd_type_len = cpu_to_le32(
1222                                                   IXGB_CONTEXT_DESC_TYPE
1223                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1224                                                 | IXGB_CONTEXT_DESC_CMD_IP
1225                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1226                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1227                                                 | (skb->len - (hdr_len)));
1228
1229
1230                 if (++i == adapter->tx_ring.count) i = 0;
1231                 adapter->tx_ring.next_to_use = i;
1232
1233                 return 1;
1234         }
1235
1236         return 0;
1237 }
1238
1239 static bool
1240 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1241 {
1242         struct ixgb_context_desc *context_desc;
1243         unsigned int i;
1244         u8 css, cso;
1245
1246         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1247                 struct ixgb_buffer *buffer_info;
1248                 css = skb_transport_offset(skb);
1249                 cso = css + skb->csum_offset;
1250
1251                 i = adapter->tx_ring.next_to_use;
1252                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1253                 buffer_info = &adapter->tx_ring.buffer_info[i];
1254                 WARN_ON(buffer_info->dma != 0);
1255
1256                 context_desc->tucss = css;
1257                 context_desc->tucso = cso;
1258                 context_desc->tucse = 0;
1259                 /* zero out any previously existing data in one instruction */
1260                 *(u32 *)&(context_desc->ipcss) = 0;
1261                 context_desc->status = 0;
1262                 context_desc->hdr_len = 0;
1263                 context_desc->mss = 0;
1264                 context_desc->cmd_type_len =
1265                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1266                                     | IXGB_TX_DESC_CMD_IDE);
1267
1268                 if (++i == adapter->tx_ring.count) i = 0;
1269                 adapter->tx_ring.next_to_use = i;
1270
1271                 return true;
1272         }
1273
1274         return false;
1275 }
1276
1277 #define IXGB_MAX_TXD_PWR        14
1278 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1279
1280 static int
1281 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1282             unsigned int first)
1283 {
1284         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1285         struct pci_dev *pdev = adapter->pdev;
1286         struct ixgb_buffer *buffer_info;
1287         int len = skb_headlen(skb);
1288         unsigned int offset = 0, size, count = 0, i;
1289         unsigned int mss = skb_shinfo(skb)->gso_size;
1290         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1291         unsigned int f;
1292
1293         i = tx_ring->next_to_use;
1294
1295         while (len) {
1296                 buffer_info = &tx_ring->buffer_info[i];
1297                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1298                 /* Workaround for premature desc write-backs
1299                  * in TSO mode.  Append 4-byte sentinel desc */
1300                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1301                         size -= 4;
1302
1303                 buffer_info->length = size;
1304                 WARN_ON(buffer_info->dma != 0);
1305                 buffer_info->time_stamp = jiffies;
1306                 buffer_info->mapped_as_page = false;
1307                 buffer_info->dma = pci_map_single(pdev, skb->data + offset,
1308                                                   size, PCI_DMA_TODEVICE);
1309                 if (pci_dma_mapping_error(pdev, buffer_info->dma))
1310                         goto dma_error;
1311                 buffer_info->next_to_watch = 0;
1312
1313                 len -= size;
1314                 offset += size;
1315                 count++;
1316                 if (len) {
1317                         i++;
1318                         if (i == tx_ring->count)
1319                                 i = 0;
1320                 }
1321         }
1322
1323         for (f = 0; f < nr_frags; f++) {
1324                 struct skb_frag_struct *frag;
1325
1326                 frag = &skb_shinfo(skb)->frags[f];
1327                 len = frag->size;
1328                 offset = frag->page_offset;
1329
1330                 while (len) {
1331                         i++;
1332                         if (i == tx_ring->count)
1333                                 i = 0;
1334
1335                         buffer_info = &tx_ring->buffer_info[i];
1336                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1337
1338                         /* Workaround for premature desc write-backs
1339                          * in TSO mode.  Append 4-byte sentinel desc */
1340                         if (unlikely(mss && (f == (nr_frags - 1))
1341                                      && size == len && size > 8))
1342                                 size -= 4;
1343
1344                         buffer_info->length = size;
1345                         buffer_info->time_stamp = jiffies;
1346                         buffer_info->mapped_as_page = true;
1347                         buffer_info->dma =
1348                                 pci_map_page(pdev, frag->page,
1349                                              offset, size,
1350                                              PCI_DMA_TODEVICE);
1351                         if (pci_dma_mapping_error(pdev, buffer_info->dma))
1352                                 goto dma_error;
1353                         buffer_info->next_to_watch = 0;
1354
1355                         len -= size;
1356                         offset += size;
1357                         count++;
1358                 }
1359         }
1360         tx_ring->buffer_info[i].skb = skb;
1361         tx_ring->buffer_info[first].next_to_watch = i;
1362
1363         return count;
1364
1365 dma_error:
1366         dev_err(&pdev->dev, "TX DMA map failed\n");
1367         buffer_info->dma = 0;
1368         if (count)
1369                 count--;
1370
1371         while (count--) {
1372                 if (i==0)
1373                         i += tx_ring->count;
1374                 i--;
1375                 buffer_info = &tx_ring->buffer_info[i];
1376                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1377         }
1378
1379         return 0;
1380 }
1381
1382 static void
1383 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1384 {
1385         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1386         struct ixgb_tx_desc *tx_desc = NULL;
1387         struct ixgb_buffer *buffer_info;
1388         u32 cmd_type_len = adapter->tx_cmd_type;
1389         u8 status = 0;
1390         u8 popts = 0;
1391         unsigned int i;
1392
1393         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1394                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1395                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1396         }
1397
1398         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1399                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1400
1401         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1402                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1403
1404         i = tx_ring->next_to_use;
1405
1406         while (count--) {
1407                 buffer_info = &tx_ring->buffer_info[i];
1408                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1409                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1410                 tx_desc->cmd_type_len =
1411                         cpu_to_le32(cmd_type_len | buffer_info->length);
1412                 tx_desc->status = status;
1413                 tx_desc->popts = popts;
1414                 tx_desc->vlan = cpu_to_le16(vlan_id);
1415
1416                 if (++i == tx_ring->count) i = 0;
1417         }
1418
1419         tx_desc->cmd_type_len |=
1420                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1421
1422         /* Force memory writes to complete before letting h/w
1423          * know there are new descriptors to fetch.  (Only
1424          * applicable for weak-ordered memory model archs,
1425          * such as IA-64). */
1426         wmb();
1427
1428         tx_ring->next_to_use = i;
1429         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1430 }
1431
1432 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1433 {
1434         struct ixgb_adapter *adapter = netdev_priv(netdev);
1435         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1436
1437         netif_stop_queue(netdev);
1438         /* Herbert's original patch had:
1439          *  smp_mb__after_netif_stop_queue();
1440          * but since that doesn't exist yet, just open code it. */
1441         smp_mb();
1442
1443         /* We need to check again in a case another CPU has just
1444          * made room available. */
1445         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1446                 return -EBUSY;
1447
1448         /* A reprieve! */
1449         netif_start_queue(netdev);
1450         ++adapter->restart_queue;
1451         return 0;
1452 }
1453
1454 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1455                               struct ixgb_desc_ring *tx_ring, int size)
1456 {
1457         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1458                 return 0;
1459         return __ixgb_maybe_stop_tx(netdev, size);
1460 }
1461
1462
1463 /* Tx Descriptors needed, worst case */
1464 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1465                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1466 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1467         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1468         + 1 /* one more needed for sentinel TSO workaround */
1469
1470 static netdev_tx_t
1471 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1472 {
1473         struct ixgb_adapter *adapter = netdev_priv(netdev);
1474         unsigned int first;
1475         unsigned int tx_flags = 0;
1476         int vlan_id = 0;
1477         int count = 0;
1478         int tso;
1479
1480         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1481                 dev_kfree_skb(skb);
1482                 return NETDEV_TX_OK;
1483         }
1484
1485         if (skb->len <= 0) {
1486                 dev_kfree_skb(skb);
1487                 return NETDEV_TX_OK;
1488         }
1489
1490         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1491                      DESC_NEEDED)))
1492                 return NETDEV_TX_BUSY;
1493
1494         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1495                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1496                 vlan_id = vlan_tx_tag_get(skb);
1497         }
1498
1499         first = adapter->tx_ring.next_to_use;
1500
1501         tso = ixgb_tso(adapter, skb);
1502         if (tso < 0) {
1503                 dev_kfree_skb(skb);
1504                 return NETDEV_TX_OK;
1505         }
1506
1507         if (likely(tso))
1508                 tx_flags |= IXGB_TX_FLAGS_TSO;
1509         else if (ixgb_tx_csum(adapter, skb))
1510                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1511
1512         count = ixgb_tx_map(adapter, skb, first);
1513
1514         if (count) {
1515                 ixgb_tx_queue(adapter, count, vlan_id, tx_flags);
1516                 /* Make sure there is space in the ring for the next send. */
1517                 ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1518
1519         } else {
1520                 dev_kfree_skb_any(skb);
1521                 adapter->tx_ring.buffer_info[first].time_stamp = 0;
1522                 adapter->tx_ring.next_to_use = first;
1523         }
1524
1525         return NETDEV_TX_OK;
1526 }
1527
1528 /**
1529  * ixgb_tx_timeout - Respond to a Tx Hang
1530  * @netdev: network interface device structure
1531  **/
1532
1533 static void
1534 ixgb_tx_timeout(struct net_device *netdev)
1535 {
1536         struct ixgb_adapter *adapter = netdev_priv(netdev);
1537
1538         /* Do the reset outside of interrupt context */
1539         schedule_work(&adapter->tx_timeout_task);
1540 }
1541
1542 static void
1543 ixgb_tx_timeout_task(struct work_struct *work)
1544 {
1545         struct ixgb_adapter *adapter =
1546                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1547
1548         adapter->tx_timeout_count++;
1549         ixgb_down(adapter, true);
1550         ixgb_up(adapter);
1551 }
1552
1553 /**
1554  * ixgb_get_stats - Get System Network Statistics
1555  * @netdev: network interface device structure
1556  *
1557  * Returns the address of the device statistics structure.
1558  * The statistics are actually updated from the timer callback.
1559  **/
1560
1561 static struct net_device_stats *
1562 ixgb_get_stats(struct net_device *netdev)
1563 {
1564         return &netdev->stats;
1565 }
1566
1567 /**
1568  * ixgb_change_mtu - Change the Maximum Transfer Unit
1569  * @netdev: network interface device structure
1570  * @new_mtu: new value for maximum frame size
1571  *
1572  * Returns 0 on success, negative on failure
1573  **/
1574
1575 static int
1576 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1577 {
1578         struct ixgb_adapter *adapter = netdev_priv(netdev);
1579         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1580         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1581
1582         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1583         if ((new_mtu < 68) ||
1584             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1585                 netif_err(adapter, probe, adapter->netdev,
1586                           "Invalid MTU setting %d\n", new_mtu);
1587                 return -EINVAL;
1588         }
1589
1590         if (old_max_frame == max_frame)
1591                 return 0;
1592
1593         if (netif_running(netdev))
1594                 ixgb_down(adapter, true);
1595
1596         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1597
1598         netdev->mtu = new_mtu;
1599
1600         if (netif_running(netdev))
1601                 ixgb_up(adapter);
1602
1603         return 0;
1604 }
1605
1606 /**
1607  * ixgb_update_stats - Update the board statistics counters.
1608  * @adapter: board private structure
1609  **/
1610
1611 void
1612 ixgb_update_stats(struct ixgb_adapter *adapter)
1613 {
1614         struct net_device *netdev = adapter->netdev;
1615         struct pci_dev *pdev = adapter->pdev;
1616
1617         /* Prevent stats update while adapter is being reset */
1618         if (pci_channel_offline(pdev))
1619                 return;
1620
1621         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1622            (netdev_mc_count(netdev) > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1623                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1624                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1625                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1626                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1627
1628                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1629                 /* fix up multicast stats by removing broadcasts */
1630                 if (multi >= bcast)
1631                         multi -= bcast;
1632
1633                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1634                 adapter->stats.mprch += (multi >> 32);
1635                 adapter->stats.bprcl += bcast_l;
1636                 adapter->stats.bprch += bcast_h;
1637         } else {
1638                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1639                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1640                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1641                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1642         }
1643         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1644         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1645         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1646         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1647         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1648         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1649         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1650         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1651         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1652         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1653         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1654         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1655         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1656         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1657         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1658         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1659         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1660         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1661         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1662         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1663         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1664         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1665         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1666         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1667         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1668         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1669         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1670         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1671         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1672         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1673         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1674         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1675         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1676         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1677         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1678         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1679         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1680         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1681         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1682         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1683         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1684         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1685         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1686         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1687         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1688         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1689         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1690         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1691         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1692         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1693         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1694         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1695         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1696         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1697         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1698         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1699
1700         /* Fill out the OS statistics structure */
1701
1702         netdev->stats.rx_packets = adapter->stats.gprcl;
1703         netdev->stats.tx_packets = adapter->stats.gptcl;
1704         netdev->stats.rx_bytes = adapter->stats.gorcl;
1705         netdev->stats.tx_bytes = adapter->stats.gotcl;
1706         netdev->stats.multicast = adapter->stats.mprcl;
1707         netdev->stats.collisions = 0;
1708
1709         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1710          * with a length in the type/len field */
1711         netdev->stats.rx_errors =
1712             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1713             adapter->stats.ruc +
1714             adapter->stats.roc /*+ adapter->stats.rlec */  +
1715             adapter->stats.icbc +
1716             adapter->stats.ecbc + adapter->stats.mpc;
1717
1718         /* see above
1719          * netdev->stats.rx_length_errors = adapter->stats.rlec;
1720          */
1721
1722         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
1723         netdev->stats.rx_fifo_errors = adapter->stats.mpc;
1724         netdev->stats.rx_missed_errors = adapter->stats.mpc;
1725         netdev->stats.rx_over_errors = adapter->stats.mpc;
1726
1727         netdev->stats.tx_errors = 0;
1728         netdev->stats.rx_frame_errors = 0;
1729         netdev->stats.tx_aborted_errors = 0;
1730         netdev->stats.tx_carrier_errors = 0;
1731         netdev->stats.tx_fifo_errors = 0;
1732         netdev->stats.tx_heartbeat_errors = 0;
1733         netdev->stats.tx_window_errors = 0;
1734 }
1735
1736 #define IXGB_MAX_INTR 10
1737 /**
1738  * ixgb_intr - Interrupt Handler
1739  * @irq: interrupt number
1740  * @data: pointer to a network interface device structure
1741  **/
1742
1743 static irqreturn_t
1744 ixgb_intr(int irq, void *data)
1745 {
1746         struct net_device *netdev = data;
1747         struct ixgb_adapter *adapter = netdev_priv(netdev);
1748         struct ixgb_hw *hw = &adapter->hw;
1749         u32 icr = IXGB_READ_REG(hw, ICR);
1750
1751         if (unlikely(!icr))
1752                 return IRQ_NONE;  /* Not our interrupt */
1753
1754         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1755                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1756                         mod_timer(&adapter->watchdog_timer, jiffies);
1757
1758         if (napi_schedule_prep(&adapter->napi)) {
1759
1760                 /* Disable interrupts and register for poll. The flush
1761                   of the posted write is intentionally left out.
1762                 */
1763
1764                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1765                 __napi_schedule(&adapter->napi);
1766         }
1767         return IRQ_HANDLED;
1768 }
1769
1770 /**
1771  * ixgb_clean - NAPI Rx polling callback
1772  * @adapter: board private structure
1773  **/
1774
1775 static int
1776 ixgb_clean(struct napi_struct *napi, int budget)
1777 {
1778         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1779         int work_done = 0;
1780
1781         ixgb_clean_tx_irq(adapter);
1782         ixgb_clean_rx_irq(adapter, &work_done, budget);
1783
1784         /* If budget not fully consumed, exit the polling mode */
1785         if (work_done < budget) {
1786                 napi_complete(napi);
1787                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1788                         ixgb_irq_enable(adapter);
1789         }
1790
1791         return work_done;
1792 }
1793
1794 /**
1795  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1796  * @adapter: board private structure
1797  **/
1798
1799 static bool
1800 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1801 {
1802         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1803         struct net_device *netdev = adapter->netdev;
1804         struct ixgb_tx_desc *tx_desc, *eop_desc;
1805         struct ixgb_buffer *buffer_info;
1806         unsigned int i, eop;
1807         bool cleaned = false;
1808
1809         i = tx_ring->next_to_clean;
1810         eop = tx_ring->buffer_info[i].next_to_watch;
1811         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1812
1813         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1814
1815                 for (cleaned = false; !cleaned; ) {
1816                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1817                         buffer_info = &tx_ring->buffer_info[i];
1818
1819                         if (tx_desc->popts &
1820                            (IXGB_TX_DESC_POPTS_TXSM |
1821                             IXGB_TX_DESC_POPTS_IXSM))
1822                                 adapter->hw_csum_tx_good++;
1823
1824                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1825
1826                         *(u32 *)&(tx_desc->status) = 0;
1827
1828                         cleaned = (i == eop);
1829                         if (++i == tx_ring->count) i = 0;
1830                 }
1831
1832                 eop = tx_ring->buffer_info[i].next_to_watch;
1833                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1834         }
1835
1836         tx_ring->next_to_clean = i;
1837
1838         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1839                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1840                 /* Make sure that anybody stopping the queue after this
1841                  * sees the new next_to_clean. */
1842                 smp_mb();
1843
1844                 if (netif_queue_stopped(netdev) &&
1845                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1846                         netif_wake_queue(netdev);
1847                         ++adapter->restart_queue;
1848                 }
1849         }
1850
1851         if (adapter->detect_tx_hung) {
1852                 /* detect a transmit hang in hardware, this serializes the
1853                  * check with the clearing of time_stamp and movement of i */
1854                 adapter->detect_tx_hung = false;
1855                 if (tx_ring->buffer_info[eop].time_stamp &&
1856                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1857                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1858                         IXGB_STATUS_TXOFF)) {
1859                         /* detected Tx unit hang */
1860                         netif_err(adapter, drv, adapter->netdev,
1861                                   "Detected Tx Unit Hang\n"
1862                                   "  TDH                  <%x>\n"
1863                                   "  TDT                  <%x>\n"
1864                                   "  next_to_use          <%x>\n"
1865                                   "  next_to_clean        <%x>\n"
1866                                   "buffer_info[next_to_clean]\n"
1867                                   "  time_stamp           <%lx>\n"
1868                                   "  next_to_watch        <%x>\n"
1869                                   "  jiffies              <%lx>\n"
1870                                   "  next_to_watch.status <%x>\n",
1871                                   IXGB_READ_REG(&adapter->hw, TDH),
1872                                   IXGB_READ_REG(&adapter->hw, TDT),
1873                                   tx_ring->next_to_use,
1874                                   tx_ring->next_to_clean,
1875                                   tx_ring->buffer_info[eop].time_stamp,
1876                                   eop,
1877                                   jiffies,
1878                                   eop_desc->status);
1879                         netif_stop_queue(netdev);
1880                 }
1881         }
1882
1883         return cleaned;
1884 }
1885
1886 /**
1887  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1888  * @adapter: board private structure
1889  * @rx_desc: receive descriptor
1890  * @sk_buff: socket buffer with received data
1891  **/
1892
1893 static void
1894 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1895                  struct ixgb_rx_desc *rx_desc,
1896                  struct sk_buff *skb)
1897 {
1898         /* Ignore Checksum bit is set OR
1899          * TCP Checksum has not been calculated
1900          */
1901         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1902            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1903                 skb->ip_summed = CHECKSUM_NONE;
1904                 return;
1905         }
1906
1907         /* At this point we know the hardware did the TCP checksum */
1908         /* now look at the TCP checksum error bit */
1909         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1910                 /* let the stack verify checksum errors */
1911                 skb->ip_summed = CHECKSUM_NONE;
1912                 adapter->hw_csum_rx_error++;
1913         } else {
1914                 /* TCP checksum is good */
1915                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1916                 adapter->hw_csum_rx_good++;
1917         }
1918 }
1919
1920 /**
1921  * ixgb_clean_rx_irq - Send received data up the network stack,
1922  * @adapter: board private structure
1923  **/
1924
1925 static bool
1926 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1927 {
1928         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1929         struct net_device *netdev = adapter->netdev;
1930         struct pci_dev *pdev = adapter->pdev;
1931         struct ixgb_rx_desc *rx_desc, *next_rxd;
1932         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1933         u32 length;
1934         unsigned int i, j;
1935         int cleaned_count = 0;
1936         bool cleaned = false;
1937
1938         i = rx_ring->next_to_clean;
1939         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1940         buffer_info = &rx_ring->buffer_info[i];
1941
1942         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1943                 struct sk_buff *skb;
1944                 u8 status;
1945
1946                 if (*work_done >= work_to_do)
1947                         break;
1948
1949                 (*work_done)++;
1950                 status = rx_desc->status;
1951                 skb = buffer_info->skb;
1952                 buffer_info->skb = NULL;
1953
1954                 prefetch(skb->data - NET_IP_ALIGN);
1955
1956                 if (++i == rx_ring->count) i = 0;
1957                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1958                 prefetch(next_rxd);
1959
1960                 if ((j = i + 1) == rx_ring->count) j = 0;
1961                 next2_buffer = &rx_ring->buffer_info[j];
1962                 prefetch(next2_buffer);
1963
1964                 next_buffer = &rx_ring->buffer_info[i];
1965
1966                 cleaned = true;
1967                 cleaned_count++;
1968
1969                 pci_unmap_single(pdev,
1970                                  buffer_info->dma,
1971                                  buffer_info->length,
1972                                  PCI_DMA_FROMDEVICE);
1973                 buffer_info->dma = 0;
1974
1975                 length = le16_to_cpu(rx_desc->length);
1976                 rx_desc->length = 0;
1977
1978                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1979
1980                         /* All receives must fit into a single buffer */
1981
1982                         IXGB_DBG("Receive packet consumed multiple buffers "
1983                                          "length<%x>\n", length);
1984
1985                         dev_kfree_skb_irq(skb);
1986                         goto rxdesc_done;
1987                 }
1988
1989                 if (unlikely(rx_desc->errors &
1990                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1991                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1992                         dev_kfree_skb_irq(skb);
1993                         goto rxdesc_done;
1994                 }
1995
1996                 /* code added for copybreak, this should improve
1997                  * performance for small packets with large amounts
1998                  * of reassembly being done in the stack */
1999                 if (length < copybreak) {
2000                         struct sk_buff *new_skb =
2001                             netdev_alloc_skb_ip_align(netdev, length);
2002                         if (new_skb) {
2003                                 skb_copy_to_linear_data_offset(new_skb,
2004                                                                -NET_IP_ALIGN,
2005                                                                (skb->data -
2006                                                                 NET_IP_ALIGN),
2007                                                                (length +
2008                                                                 NET_IP_ALIGN));
2009                                 /* save the skb in buffer_info as good */
2010                                 buffer_info->skb = skb;
2011                                 skb = new_skb;
2012                         }
2013                 }
2014                 /* end copybreak code */
2015
2016                 /* Good Receive */
2017                 skb_put(skb, length);
2018
2019                 /* Receive Checksum Offload */
2020                 ixgb_rx_checksum(adapter, rx_desc, skb);
2021
2022                 skb->protocol = eth_type_trans(skb, netdev);
2023                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2024                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2025                                                 le16_to_cpu(rx_desc->special));
2026                 } else {
2027                         netif_receive_skb(skb);
2028                 }
2029
2030 rxdesc_done:
2031                 /* clean up descriptor, might be written over by hw */
2032                 rx_desc->status = 0;
2033
2034                 /* return some buffers to hardware, one at a time is too slow */
2035                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2036                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2037                         cleaned_count = 0;
2038                 }
2039
2040                 /* use prefetched values */
2041                 rx_desc = next_rxd;
2042                 buffer_info = next_buffer;
2043         }
2044
2045         rx_ring->next_to_clean = i;
2046
2047         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2048         if (cleaned_count)
2049                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2050
2051         return cleaned;
2052 }
2053
2054 /**
2055  * ixgb_alloc_rx_buffers - Replace used receive buffers
2056  * @adapter: address of board private structure
2057  **/
2058
2059 static void
2060 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2061 {
2062         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2063         struct net_device *netdev = adapter->netdev;
2064         struct pci_dev *pdev = adapter->pdev;
2065         struct ixgb_rx_desc *rx_desc;
2066         struct ixgb_buffer *buffer_info;
2067         struct sk_buff *skb;
2068         unsigned int i;
2069         long cleancount;
2070
2071         i = rx_ring->next_to_use;
2072         buffer_info = &rx_ring->buffer_info[i];
2073         cleancount = IXGB_DESC_UNUSED(rx_ring);
2074
2075
2076         /* leave three descriptors unused */
2077         while (--cleancount > 2 && cleaned_count--) {
2078                 /* recycle! its good for you */
2079                 skb = buffer_info->skb;
2080                 if (skb) {
2081                         skb_trim(skb, 0);
2082                         goto map_skb;
2083                 }
2084
2085                 skb = netdev_alloc_skb_ip_align(netdev, adapter->rx_buffer_len);
2086                 if (unlikely(!skb)) {
2087                         /* Better luck next round */
2088                         adapter->alloc_rx_buff_failed++;
2089                         break;
2090                 }
2091
2092                 buffer_info->skb = skb;
2093                 buffer_info->length = adapter->rx_buffer_len;
2094 map_skb:
2095                 buffer_info->dma = pci_map_single(pdev,
2096                                                   skb->data,
2097                                                   adapter->rx_buffer_len,
2098                                                   PCI_DMA_FROMDEVICE);
2099
2100                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2101                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2102                 /* guarantee DD bit not set now before h/w gets descriptor
2103                  * this is the rest of the workaround for h/w double
2104                  * writeback. */
2105                 rx_desc->status = 0;
2106
2107
2108                 if (++i == rx_ring->count) i = 0;
2109                 buffer_info = &rx_ring->buffer_info[i];
2110         }
2111
2112         if (likely(rx_ring->next_to_use != i)) {
2113                 rx_ring->next_to_use = i;
2114                 if (unlikely(i-- == 0))
2115                         i = (rx_ring->count - 1);
2116
2117                 /* Force memory writes to complete before letting h/w
2118                  * know there are new descriptors to fetch.  (Only
2119                  * applicable for weak-ordered memory model archs, such
2120                  * as IA-64). */
2121                 wmb();
2122                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2123         }
2124 }
2125
2126 /**
2127  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2128  *
2129  * @param netdev network interface device structure
2130  * @param grp indicates to enable or disable tagging/stripping
2131  **/
2132 static void
2133 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2134 {
2135         struct ixgb_adapter *adapter = netdev_priv(netdev);
2136         u32 ctrl, rctl;
2137
2138         ixgb_irq_disable(adapter);
2139         adapter->vlgrp = grp;
2140
2141         if (grp) {
2142                 /* enable VLAN tag insert/strip */
2143                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2144                 ctrl |= IXGB_CTRL0_VME;
2145                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2146
2147                 /* enable VLAN receive filtering */
2148
2149                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2150                 rctl &= ~IXGB_RCTL_CFIEN;
2151                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2152         } else {
2153                 /* disable VLAN tag insert/strip */
2154
2155                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2156                 ctrl &= ~IXGB_CTRL0_VME;
2157                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2158         }
2159
2160         /* don't enable interrupts unless we are UP */
2161         if (adapter->netdev->flags & IFF_UP)
2162                 ixgb_irq_enable(adapter);
2163 }
2164
2165 static void
2166 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2167 {
2168         struct ixgb_adapter *adapter = netdev_priv(netdev);
2169         u32 vfta, index;
2170
2171         /* add VID to filter table */
2172
2173         index = (vid >> 5) & 0x7F;
2174         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2175         vfta |= (1 << (vid & 0x1F));
2176         ixgb_write_vfta(&adapter->hw, index, vfta);
2177 }
2178
2179 static void
2180 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2181 {
2182         struct ixgb_adapter *adapter = netdev_priv(netdev);
2183         u32 vfta, index;
2184
2185         ixgb_irq_disable(adapter);
2186
2187         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2188
2189         /* don't enable interrupts unless we are UP */
2190         if (adapter->netdev->flags & IFF_UP)
2191                 ixgb_irq_enable(adapter);
2192
2193         /* remove VID from filter table */
2194
2195         index = (vid >> 5) & 0x7F;
2196         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2197         vfta &= ~(1 << (vid & 0x1F));
2198         ixgb_write_vfta(&adapter->hw, index, vfta);
2199 }
2200
2201 static void
2202 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2203 {
2204         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2205
2206         if (adapter->vlgrp) {
2207                 u16 vid;
2208                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2209                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2210                                 continue;
2211                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2212                 }
2213         }
2214 }
2215
2216 #ifdef CONFIG_NET_POLL_CONTROLLER
2217 /*
2218  * Polling 'interrupt' - used by things like netconsole to send skbs
2219  * without having to re-enable interrupts. It's not called while
2220  * the interrupt routine is executing.
2221  */
2222
2223 static void ixgb_netpoll(struct net_device *dev)
2224 {
2225         struct ixgb_adapter *adapter = netdev_priv(dev);
2226
2227         disable_irq(adapter->pdev->irq);
2228         ixgb_intr(adapter->pdev->irq, dev);
2229         enable_irq(adapter->pdev->irq);
2230 }
2231 #endif
2232
2233 /**
2234  * ixgb_io_error_detected() - called when PCI error is detected
2235  * @pdev    pointer to pci device with error
2236  * @state   pci channel state after error
2237  *
2238  * This callback is called by the PCI subsystem whenever
2239  * a PCI bus error is detected.
2240  */
2241 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2242                                                enum pci_channel_state state)
2243 {
2244         struct net_device *netdev = pci_get_drvdata(pdev);
2245         struct ixgb_adapter *adapter = netdev_priv(netdev);
2246
2247         netif_device_detach(netdev);
2248
2249         if (state == pci_channel_io_perm_failure)
2250                 return PCI_ERS_RESULT_DISCONNECT;
2251
2252         if (netif_running(netdev))
2253                 ixgb_down(adapter, true);
2254
2255         pci_disable_device(pdev);
2256
2257         /* Request a slot reset. */
2258         return PCI_ERS_RESULT_NEED_RESET;
2259 }
2260
2261 /**
2262  * ixgb_io_slot_reset - called after the pci bus has been reset.
2263  * @pdev    pointer to pci device with error
2264  *
2265  * This callback is called after the PCI bus has been reset.
2266  * Basically, this tries to restart the card from scratch.
2267  * This is a shortened version of the device probe/discovery code,
2268  * it resembles the first-half of the ixgb_probe() routine.
2269  */
2270 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2271 {
2272         struct net_device *netdev = pci_get_drvdata(pdev);
2273         struct ixgb_adapter *adapter = netdev_priv(netdev);
2274
2275         if (pci_enable_device(pdev)) {
2276                 netif_err(adapter, probe, adapter->netdev,
2277                           "Cannot re-enable PCI device after reset\n");
2278                 return PCI_ERS_RESULT_DISCONNECT;
2279         }
2280
2281         /* Perform card reset only on one instance of the card */
2282         if (0 != PCI_FUNC (pdev->devfn))
2283                 return PCI_ERS_RESULT_RECOVERED;
2284
2285         pci_set_master(pdev);
2286
2287         netif_carrier_off(netdev);
2288         netif_stop_queue(netdev);
2289         ixgb_reset(adapter);
2290
2291         /* Make sure the EEPROM is good */
2292         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2293                 netif_err(adapter, probe, adapter->netdev,
2294                           "After reset, the EEPROM checksum is not valid\n");
2295                 return PCI_ERS_RESULT_DISCONNECT;
2296         }
2297         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2298         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2299
2300         if (!is_valid_ether_addr(netdev->perm_addr)) {
2301                 netif_err(adapter, probe, adapter->netdev,
2302                           "After reset, invalid MAC address\n");
2303                 return PCI_ERS_RESULT_DISCONNECT;
2304         }
2305
2306         return PCI_ERS_RESULT_RECOVERED;
2307 }
2308
2309 /**
2310  * ixgb_io_resume - called when its OK to resume normal operations
2311  * @pdev    pointer to pci device with error
2312  *
2313  * The error recovery driver tells us that its OK to resume
2314  * normal operation. Implementation resembles the second-half
2315  * of the ixgb_probe() routine.
2316  */
2317 static void ixgb_io_resume(struct pci_dev *pdev)
2318 {
2319         struct net_device *netdev = pci_get_drvdata(pdev);
2320         struct ixgb_adapter *adapter = netdev_priv(netdev);
2321
2322         pci_set_master(pdev);
2323
2324         if (netif_running(netdev)) {
2325                 if (ixgb_up(adapter)) {
2326                         printk ("ixgb: can't bring device back up after reset\n");
2327                         return;
2328                 }
2329         }
2330
2331         netif_device_attach(netdev);
2332         mod_timer(&adapter->watchdog_timer, jiffies);
2333 }
2334
2335 /* ixgb_main.c */