kernel/drivers/net/ethernet/intel/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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  30
  31 #include <linux/prefetch.h>
  32 #include "ixgb.h"
  33
  34 char ixgb_driver_name[] = "ixgb";
  35 static char ixgb_driver_string[] = "Intel(R) PRO/10GbE Network Driver";
  36
  37 #define DRIVERNAPI "-NAPI"
  38 #define DRV_VERSION "1.0.135-k2" DRIVERNAPI
  39 const char ixgb_driver_version[] = DRV_VERSION;
  40 static const char ixgb_copyright[] = "Copyright (c) 1999-2008 Intel Corporation.";
  41
  42 #define IXGB_CB_LENGTH 256
  43 static unsigned int copybreak __read_mostly = IXGB_CB_LENGTH;
  44 module_param(copybreak, uint, 0644);
  45 MODULE_PARM_DESC(copybreak,
  46         "Maximum size of packet that is copied to a new buffer on receive");
  47
  48 /* ixgb_pci_tbl - PCI Device ID Table
  49  *
  50  * Wildcard entries (PCI_ANY_ID) should come last
  51  * Last entry must be all 0s
  52  *
  53  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  54  *   Class, Class Mask, private data (not used) }
  55  */
  56 static const struct pci_device_id ixgb_pci_tbl[] = {
  57         {PCI_VENDOR_ID_INTEL, IXGB_DEVICE_ID_82597EX,
  58          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  59         {PCI_VENDOR_ID_INTEL, IXGB_DEVICE_ID_82597EX_CX4,
  60          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  61         {PCI_VENDOR_ID_INTEL, IXGB_DEVICE_ID_82597EX_SR,
  62          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  63         {PCI_VENDOR_ID_INTEL, IXGB_DEVICE_ID_82597EX_LR,
  64          PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0},
  65
  66         /* required last entry */
  67         {0,}
  68 };
  69
  70 MODULE_DEVICE_TABLE(pci, ixgb_pci_tbl);
  71
  72 /* Local Function Prototypes */
  73 static int ixgb_init_module(void);
  74 static void ixgb_exit_module(void);
  75 static int ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
  76 static void ixgb_remove(struct pci_dev *pdev);
  77 static int ixgb_sw_init(struct ixgb_adapter *adapter);
  78 static int ixgb_open(struct net_device *netdev);
  79 static int ixgb_close(struct net_device *netdev);
  80 static void ixgb_configure_tx(struct ixgb_adapter *adapter);
  81 static void ixgb_configure_rx(struct ixgb_adapter *adapter);
  82 static void ixgb_setup_rctl(struct ixgb_adapter *adapter);
  83 static void ixgb_clean_tx_ring(struct ixgb_adapter *adapter);
  84 static void ixgb_clean_rx_ring(struct ixgb_adapter *adapter);
  85 static void ixgb_set_multi(struct net_device *netdev);
  86 static void ixgb_watchdog(unsigned long data);
  87 static netdev_tx_t ixgb_xmit_frame(struct sk_buff *skb,
  88                                    struct net_device *netdev);
  89 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
  90 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
  91 static int ixgb_set_mac(struct net_device *netdev, void *p);
  92 static irqreturn_t ixgb_intr(int irq, void *data);
  93 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
  94
  95 static int ixgb_clean(struct napi_struct *, int);
  96 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
  97 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
  98
  99 static void ixgb_tx_timeout(struct net_device *dev);
 100 static void ixgb_tx_timeout_task(struct work_struct *work);
 101
 102 static void ixgb_vlan_strip_enable(struct ixgb_adapter *adapter);
 103 static void ixgb_vlan_strip_disable(struct ixgb_adapter *adapter);
 104 static int ixgb_vlan_rx_add_vid(struct net_device *netdev,
 105                                 __be16 proto, u16 vid);
 106 static int ixgb_vlan_rx_kill_vid(struct net_device *netdev,
 107                                  __be16 proto, u16 vid);
 108 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
 109
 110 #ifdef CONFIG_NET_POLL_CONTROLLER
 111 /* for netdump / net console */
 112 static void ixgb_netpoll(struct net_device *dev);
 113 #endif
 114
 115 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
 116                              enum pci_channel_state state);
 117 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
 118 static void ixgb_io_resume (struct pci_dev *pdev);
 119
 120 static const struct pci_error_handlers ixgb_err_handler = {
 121         .error_detected = ixgb_io_error_detected,
 122         .slot_reset = ixgb_io_slot_reset,
 123         .resume = ixgb_io_resume,
 124 };
 125
 126 static struct pci_driver ixgb_driver = {
 127         .name     = ixgb_driver_name,
 128         .id_table = ixgb_pci_tbl,
 129         .probe    = ixgb_probe,
 130         .remove   = ixgb_remove,
 131         .err_handler = &ixgb_err_handler
 132 };
 133
 134 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 135 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
 136 MODULE_LICENSE("GPL");
 137 MODULE_VERSION(DRV_VERSION);
 138
 139 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 140 static int debug = -1;
 141 module_param(debug, int, 0);
 142 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 143
 144 /**
 145  * ixgb_init_module - Driver Registration Routine
 146  *
 147  * ixgb_init_module is the first routine called when the driver is
 148  * loaded. All it does is register with the PCI subsystem.
 149  **/
 150
 151 static int __init
 152 ixgb_init_module(void)
 153 {
 154         pr_info("%s - version %s\n", ixgb_driver_string, ixgb_driver_version);
 155         pr_info("%s\n", ixgb_copyright);
 156
 157         return pci_register_driver(&ixgb_driver);
 158 }
 159
 160 module_init(ixgb_init_module);
 161
 162 /**
 163  * ixgb_exit_module - Driver Exit Cleanup Routine
 164  *
 165  * ixgb_exit_module is called just before the driver is removed
 166  * from memory.
 167  **/
 168
 169 static void __exit
 170 ixgb_exit_module(void)
 171 {
 172         pci_unregister_driver(&ixgb_driver);
 173 }
 174
 175 module_exit(ixgb_exit_module);
 176
 177 /**
 178  * ixgb_irq_disable - Mask off interrupt generation on the NIC
 179  * @adapter: board private structure
 180  **/
 181
 182 static void
 183 ixgb_irq_disable(struct ixgb_adapter *adapter)
 184 {
 185         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
 186         IXGB_WRITE_FLUSH(&adapter->hw);
 187         synchronize_irq(adapter->pdev->irq);
 188 }
 189
 190 /**
 191  * ixgb_irq_enable - Enable default interrupt generation settings
 192  * @adapter: board private structure
 193  **/
 194
 195 static void
 196 ixgb_irq_enable(struct ixgb_adapter *adapter)
 197 {
 198         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
 199                   IXGB_INT_TXDW | IXGB_INT_LSC;
 200         if (adapter->hw.subsystem_vendor_id == PCI_VENDOR_ID_SUN)
 201                 val |= IXGB_INT_GPI0;
 202         IXGB_WRITE_REG(&adapter->hw, IMS, val);
 203         IXGB_WRITE_FLUSH(&adapter->hw);
 204 }
 205
 206 int
 207 ixgb_up(struct ixgb_adapter *adapter)
 208 {
 209         struct net_device *netdev = adapter->netdev;
 210         int err, irq_flags = IRQF_SHARED;
 211         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 212         struct ixgb_hw *hw = &adapter->hw;
 213
 214         /* hardware has been reset, we need to reload some things */
 215
 216         ixgb_rar_set(hw, netdev->dev_addr, 0);
 217         ixgb_set_multi(netdev);
 218
 219         ixgb_restore_vlan(adapter);
 220
 221         ixgb_configure_tx(adapter);
 222         ixgb_setup_rctl(adapter);
 223         ixgb_configure_rx(adapter);
 224         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
 225
 226         /* disable interrupts and get the hardware into a known state */
 227         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
 228
 229         /* only enable MSI if bus is in PCI-X mode */
 230         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
 231                 err = pci_enable_msi(adapter->pdev);
 232                 if (!err) {
 233                         adapter->have_msi = true;
 234                         irq_flags = 0;
 235                 }
 236                 /* proceed to try to request regular interrupt */
 237         }
 238
 239         err = request_irq(adapter->pdev->irq, ixgb_intr, irq_flags,
 240                           netdev->name, netdev);
 241         if (err) {
 242                 if (adapter->have_msi)
 243                         pci_disable_msi(adapter->pdev);
 244                 netif_err(adapter, probe, adapter->netdev,
 245                           "Unable to allocate interrupt Error: %d\n", err);
 246                 return err;
 247         }
 248
 249         if ((hw->max_frame_size != max_frame) ||
 250                 (hw->max_frame_size !=
 251                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
 252
 253                 hw->max_frame_size = max_frame;
 254
 255                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
 256
 257                 if (hw->max_frame_size >
 258                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
 259                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
 260
 261                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
 262                                 ctrl0 |= IXGB_CTRL0_JFE;
 263                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
 264                         }
 265                 }
 266         }
 267
 268         clear_bit(__IXGB_DOWN, &adapter->flags);
 269
 270         napi_enable(&adapter->napi);
 271         ixgb_irq_enable(adapter);
 272
 273         netif_wake_queue(netdev);
 274
 275         mod_timer(&adapter->watchdog_timer, jiffies);
 276
 277         return 0;
 278 }
 279
 280 void
 281 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
 282 {
 283         struct net_device *netdev = adapter->netdev;
 284
 285         /* prevent the interrupt handler from restarting watchdog */
 286         set_bit(__IXGB_DOWN, &adapter->flags);
 287
 288         netif_carrier_off(netdev);
 289
 290         napi_disable(&adapter->napi);
 291         /* waiting for NAPI to complete can re-enable interrupts */
 292         ixgb_irq_disable(adapter);
 293         free_irq(adapter->pdev->irq, netdev);
 294
 295         if (adapter->have_msi)
 296                 pci_disable_msi(adapter->pdev);
 297
 298         if (kill_watchdog)
 299                 del_timer_sync(&adapter->watchdog_timer);
 300
 301         adapter->link_speed = 0;
 302         adapter->link_duplex = 0;
 303         netif_stop_queue(netdev);
 304
 305         ixgb_reset(adapter);
 306         ixgb_clean_tx_ring(adapter);
 307         ixgb_clean_rx_ring(adapter);
 308 }
 309
 310 void
 311 ixgb_reset(struct ixgb_adapter *adapter)
 312 {
 313         struct ixgb_hw *hw = &adapter->hw;
 314
 315         ixgb_adapter_stop(hw);
 316         if (!ixgb_init_hw(hw))
 317                 netif_err(adapter, probe, adapter->netdev, "ixgb_init_hw failed\n");
 318
 319         /* restore frame size information */
 320         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
 321         if (hw->max_frame_size >
 322             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
 323                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
 324                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
 325                         ctrl0 |= IXGB_CTRL0_JFE;
 326                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
 327                 }
 328         }
 329 }
 330
 331 static netdev_features_t
 332 ixgb_fix_features(struct net_device *netdev, netdev_features_t features)
 333 {
 334         /*
 335          * Tx VLAN insertion does not work per HW design when Rx stripping is
 336          * disabled.
 337          */
 338         if (!(features & NETIF_F_HW_VLAN_CTAG_RX))
 339                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
 340
 341         return features;
 342 }
 343
 344 static int
 345 ixgb_set_features(struct net_device *netdev, netdev_features_t features)
 346 {
 347         struct ixgb_adapter *adapter = netdev_priv(netdev);
 348         netdev_features_t changed = features ^ netdev->features;
 349
 350         if (!(changed & (NETIF_F_RXCSUM|NETIF_F_HW_VLAN_CTAG_RX)))
 351                 return 0;
 352
 353         adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
 354
 355         if (netif_running(netdev)) {
 356                 ixgb_down(adapter, true);
 357                 ixgb_up(adapter);
 358                 ixgb_set_speed_duplex(netdev);
 359         } else
 360                 ixgb_reset(adapter);
 361
 362         return 0;
 363 }
 364
 365
 366 static const struct net_device_ops ixgb_netdev_ops = {
 367         .ndo_open               = ixgb_open,
 368         .ndo_stop               = ixgb_close,
 369         .ndo_start_xmit         = ixgb_xmit_frame,
 370         .ndo_get_stats          = ixgb_get_stats,
 371         .ndo_set_rx_mode        = ixgb_set_multi,
 372         .ndo_validate_addr      = eth_validate_addr,
 373         .ndo_set_mac_address    = ixgb_set_mac,
 374         .ndo_change_mtu         = ixgb_change_mtu,
 375         .ndo_tx_timeout         = ixgb_tx_timeout,
 376         .ndo_vlan_rx_add_vid    = ixgb_vlan_rx_add_vid,
 377         .ndo_vlan_rx_kill_vid   = ixgb_vlan_rx_kill_vid,
 378 #ifdef CONFIG_NET_POLL_CONTROLLER
 379         .ndo_poll_controller    = ixgb_netpoll,
 380 #endif
 381         .ndo_fix_features       = ixgb_fix_features,
 382         .ndo_set_features       = ixgb_set_features,
 383 };
 384
 385 /**
 386  * ixgb_probe - Device Initialization Routine
 387  * @pdev: PCI device information struct
 388  * @ent: entry in ixgb_pci_tbl
 389  *
 390  * Returns 0 on success, negative on failure
 391  *
 392  * ixgb_probe initializes an adapter identified by a pci_dev structure.
 393  * The OS initialization, configuring of the adapter private structure,
 394  * and a hardware reset occur.
 395  **/
 396
 397 static int
 398 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 399 {
 400         struct net_device *netdev = NULL;
 401         struct ixgb_adapter *adapter;
 402         static int cards_found = 0;
 403         int pci_using_dac;
 404         int i;
 405         int err;
 406
 407         err = pci_enable_device(pdev);
 408         if (err)
 409                 return err;
 410
 411         pci_using_dac = 0;
 412         err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
 413         if (!err) {
 414                 pci_using_dac = 1;
 415         } else {
 416                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
 417                 if (err) {
 418                         pr_err("No usable DMA configuration, aborting\n");
 419                         goto err_dma_mask;
 420                 }
 421         }
 422
 423         err = pci_request_regions(pdev, ixgb_driver_name);
 424         if (err)
 425                 goto err_request_regions;
 426
 427         pci_set_master(pdev);
 428
 429         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
 430         if (!netdev) {
 431                 err = -ENOMEM;
 432                 goto err_alloc_etherdev;
 433         }
 434
 435         SET_NETDEV_DEV(netdev, &pdev->dev);
 436
 437         pci_set_drvdata(pdev, netdev);
 438         adapter = netdev_priv(netdev);
 439         adapter->netdev = netdev;
 440         adapter->pdev = pdev;
 441         adapter->hw.back = adapter;
 442         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
 443
 444         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
 445         if (!adapter->hw.hw_addr) {
 446                 err = -EIO;
 447                 goto err_ioremap;
 448         }
 449
 450         for (i = BAR_1; i <= BAR_5; i++) {
 451                 if (pci_resource_len(pdev, i) == 0)
 452                         continue;
 453                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
 454                         adapter->hw.io_base = pci_resource_start(pdev, i);
 455                         break;
 456                 }
 457         }
 458
 459         netdev->netdev_ops = &ixgb_netdev_ops;
 460         ixgb_set_ethtool_ops(netdev);
 461         netdev->watchdog_timeo = 5 * HZ;
 462         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
 463
 464         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 465
 466         adapter->bd_number = cards_found;
 467         adapter->link_speed = 0;
 468         adapter->link_duplex = 0;
 469
 470         /* setup the private structure */
 471
 472         err = ixgb_sw_init(adapter);
 473         if (err)
 474                 goto err_sw_init;
 475
 476         netdev->hw_features = NETIF_F_SG |
 477                            NETIF_F_TSO |
 478                            NETIF_F_HW_CSUM |
 479                            NETIF_F_HW_VLAN_CTAG_TX |
 480                            NETIF_F_HW_VLAN_CTAG_RX;
 481         netdev->features = netdev->hw_features |
 482                            NETIF_F_HW_VLAN_CTAG_FILTER;
 483         netdev->hw_features |= NETIF_F_RXCSUM;
 484
 485         if (pci_using_dac) {
 486                 netdev->features |= NETIF_F_HIGHDMA;
 487                 netdev->vlan_features |= NETIF_F_HIGHDMA;
 488         }
 489
 490         /* make sure the EEPROM is good */
 491
 492         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
 493                 netif_err(adapter, probe, adapter->netdev,
 494                           "The EEPROM Checksum Is Not Valid\n");
 495                 err = -EIO;
 496                 goto err_eeprom;
 497         }
 498
 499         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
 500
 501         if (!is_valid_ether_addr(netdev->dev_addr)) {
 502                 netif_err(adapter, probe, adapter->netdev, "Invalid MAC Address\n");
 503                 err = -EIO;
 504                 goto err_eeprom;
 505         }
 506
 507         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
 508
 509         init_timer(&adapter->watchdog_timer);
 510         adapter->watchdog_timer.function = ixgb_watchdog;
 511         adapter->watchdog_timer.data = (unsigned long)adapter;
 512
 513         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
 514
 515         strcpy(netdev->name, "eth%d");
 516         err = register_netdev(netdev);
 517         if (err)
 518                 goto err_register;
 519
 520         /* carrier off reporting is important to ethtool even BEFORE open */
 521         netif_carrier_off(netdev);
 522
 523         netif_info(adapter, probe, adapter->netdev,
 524                    "Intel(R) PRO/10GbE Network Connection\n");
 525         ixgb_check_options(adapter);
 526         /* reset the hardware with the new settings */
 527
 528         ixgb_reset(adapter);
 529
 530         cards_found++;
 531         return 0;
 532
 533 err_register:
 534 err_sw_init:
 535 err_eeprom:
 536         iounmap(adapter->hw.hw_addr);
 537 err_ioremap:
 538         free_netdev(netdev);
 539 err_alloc_etherdev:
 540         pci_release_regions(pdev);
 541 err_request_regions:
 542 err_dma_mask:
 543         pci_disable_device(pdev);
 544         return err;
 545 }
 546
 547 /**
 548  * ixgb_remove - Device Removal Routine
 549  * @pdev: PCI device information struct
 550  *
 551  * ixgb_remove is called by the PCI subsystem to alert the driver
 552  * that it should release a PCI device.  The could be caused by a
 553  * Hot-Plug event, or because the driver is going to be removed from
 554  * memory.
 555  **/
 556
 557 static void
 558 ixgb_remove(struct pci_dev *pdev)
 559 {
 560         struct net_device *netdev = pci_get_drvdata(pdev);
 561         struct ixgb_adapter *adapter = netdev_priv(netdev);
 562
 563         cancel_work_sync(&adapter->tx_timeout_task);
 564
 565         unregister_netdev(netdev);
 566
 567         iounmap(adapter->hw.hw_addr);
 568         pci_release_regions(pdev);
 569
 570         free_netdev(netdev);
 571         pci_disable_device(pdev);
 572 }
 573
 574 /**
 575  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
 576  * @adapter: board private structure to initialize
 577  *
 578  * ixgb_sw_init initializes the Adapter private data structure.
 579  * Fields are initialized based on PCI device information and
 580  * OS network device settings (MTU size).
 581  **/
 582
 583 static int
 584 ixgb_sw_init(struct ixgb_adapter *adapter)
 585 {
 586         struct ixgb_hw *hw = &adapter->hw;
 587         struct net_device *netdev = adapter->netdev;
 588         struct pci_dev *pdev = adapter->pdev;
 589
 590         /* PCI config space info */
 591
 592         hw->vendor_id = pdev->vendor;
 593         hw->device_id = pdev->device;
 594         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 595         hw->subsystem_id = pdev->subsystem_device;
 596
 597         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 598         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
 599
 600         if ((hw->device_id == IXGB_DEVICE_ID_82597EX) ||
 601             (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4) ||
 602             (hw->device_id == IXGB_DEVICE_ID_82597EX_LR) ||
 603             (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
 604                 hw->mac_type = ixgb_82597;
 605         else {
 606                 /* should never have loaded on this device */
 607                 netif_err(adapter, probe, adapter->netdev, "unsupported device id\n");
 608         }
 609
 610         /* enable flow control to be programmed */
 611         hw->fc.send_xon = 1;
 612
 613         set_bit(__IXGB_DOWN, &adapter->flags);
 614         return 0;
 615 }
 616
 617 /**
 618  * ixgb_open - Called when a network interface is made active
 619  * @netdev: network interface device structure
 620  *
 621  * Returns 0 on success, negative value on failure
 622  *
 623  * The open entry point is called when a network interface is made
 624  * active by the system (IFF_UP).  At this point all resources needed
 625  * for transmit and receive operations are allocated, the interrupt
 626  * handler is registered with the OS, the watchdog timer is started,
 627  * and the stack is notified that the interface is ready.
 628  **/
 629
 630 static int
 631 ixgb_open(struct net_device *netdev)
 632 {
 633         struct ixgb_adapter *adapter = netdev_priv(netdev);
 634         int err;
 635
 636         /* allocate transmit descriptors */
 637         err = ixgb_setup_tx_resources(adapter);
 638         if (err)
 639                 goto err_setup_tx;
 640
 641         netif_carrier_off(netdev);
 642
 643         /* allocate receive descriptors */
 644
 645         err = ixgb_setup_rx_resources(adapter);
 646         if (err)
 647                 goto err_setup_rx;
 648
 649         err = ixgb_up(adapter);
 650         if (err)
 651                 goto err_up;
 652
 653         netif_start_queue(netdev);
 654
 655         return 0;
 656
 657 err_up:
 658         ixgb_free_rx_resources(adapter);
 659 err_setup_rx:
 660         ixgb_free_tx_resources(adapter);
 661 err_setup_tx:
 662         ixgb_reset(adapter);
 663
 664         return err;
 665 }
 666
 667 /**
 668  * ixgb_close - Disables a network interface
 669  * @netdev: network interface device structure
 670  *
 671  * Returns 0, this is not allowed to fail
 672  *
 673  * The close entry point is called when an interface is de-activated
 674  * by the OS.  The hardware is still under the drivers control, but
 675  * needs to be disabled.  A global MAC reset is issued to stop the
 676  * hardware, and all transmit and receive resources are freed.
 677  **/
 678
 679 static int
 680 ixgb_close(struct net_device *netdev)
 681 {
 682         struct ixgb_adapter *adapter = netdev_priv(netdev);
 683
 684         ixgb_down(adapter, true);
 685
 686         ixgb_free_tx_resources(adapter);
 687         ixgb_free_rx_resources(adapter);
 688
 689         return 0;
 690 }
 691
 692 /**
 693  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
 694  * @adapter: board private structure
 695  *
 696  * Return 0 on success, negative on failure
 697  **/
 698
 699 int
 700 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
 701 {
 702         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
 703         struct pci_dev *pdev = adapter->pdev;
 704         int size;
 705
 706         size = sizeof(struct ixgb_buffer) * txdr->count;
 707         txdr->buffer_info = vzalloc(size);
 708         if (!txdr->buffer_info)
 709                 return -ENOMEM;
 710
 711         /* round up to nearest 4K */
 712
 713         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
 714         txdr->size = ALIGN(txdr->size, 4096);
 715
 716         txdr->desc = dma_zalloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
 717                                          GFP_KERNEL);
 718         if (!txdr->desc) {
 719                 vfree(txdr->buffer_info);
 720                 return -ENOMEM;
 721         }
 722
 723         txdr->next_to_use = 0;
 724         txdr->next_to_clean = 0;
 725
 726         return 0;
 727 }
 728
 729 /**
 730  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
 731  * @adapter: board private structure
 732  *
 733  * Configure the Tx unit of the MAC after a reset.
 734  **/
 735
 736 static void
 737 ixgb_configure_tx(struct ixgb_adapter *adapter)
 738 {
 739         u64 tdba = adapter->tx_ring.dma;
 740         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
 741         u32 tctl;
 742         struct ixgb_hw *hw = &adapter->hw;
 743
 744         /* Setup the Base and Length of the Tx Descriptor Ring
 745          * tx_ring.dma can be either a 32 or 64 bit value
 746          */
 747
 748         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
 749         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
 750
 751         IXGB_WRITE_REG(hw, TDLEN, tdlen);
 752
 753         /* Setup the HW Tx Head and Tail descriptor pointers */
 754
 755         IXGB_WRITE_REG(hw, TDH, 0);
 756         IXGB_WRITE_REG(hw, TDT, 0);
 757
 758         /* don't set up txdctl, it induces performance problems if configured
 759          * incorrectly */
 760         /* Set the Tx Interrupt Delay register */
 761
 762         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
 763
 764         /* Program the Transmit Control Register */
 765
 766         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
 767         IXGB_WRITE_REG(hw, TCTL, tctl);
 768
 769         /* Setup Transmit Descriptor Settings for this adapter */
 770         adapter->tx_cmd_type =
 771                 IXGB_TX_DESC_TYPE |
 772                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
 773 }
 774
 775 /**
 776  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
 777  * @adapter: board private structure
 778  *
 779  * Returns 0 on success, negative on failure
 780  **/
 781
 782 int
 783 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
 784 {
 785         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
 786         struct pci_dev *pdev = adapter->pdev;
 787         int size;
 788
 789         size = sizeof(struct ixgb_buffer) * rxdr->count;
 790         rxdr->buffer_info = vzalloc(size);
 791         if (!rxdr->buffer_info)
 792                 return -ENOMEM;
 793
 794         /* Round up to nearest 4K */
 795
 796         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
 797         rxdr->size = ALIGN(rxdr->size, 4096);
 798
 799         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
 800                                         GFP_KERNEL);
 801
 802         if (!rxdr->desc) {
 803                 vfree(rxdr->buffer_info);
 804                 return -ENOMEM;
 805         }
 806         memset(rxdr->desc, 0, rxdr->size);
 807
 808         rxdr->next_to_clean = 0;
 809         rxdr->next_to_use = 0;
 810
 811         return 0;
 812 }
 813
 814 /**
 815  * ixgb_setup_rctl - configure the receive control register
 816  * @adapter: Board private structure
 817  **/
 818
 819 static void
 820 ixgb_setup_rctl(struct ixgb_adapter *adapter)
 821 {
 822         u32 rctl;
 823
 824         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
 825
 826         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
 827
 828         rctl |=
 829                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
 830                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
 831                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
 832
 833         rctl |= IXGB_RCTL_SECRC;
 834
 835         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
 836                 rctl |= IXGB_RCTL_BSIZE_2048;
 837         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
 838                 rctl |= IXGB_RCTL_BSIZE_4096;
 839         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
 840                 rctl |= IXGB_RCTL_BSIZE_8192;
 841         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
 842                 rctl |= IXGB_RCTL_BSIZE_16384;
 843
 844         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
 845 }
 846
 847 /**
 848  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
 849  * @adapter: board private structure
 850  *
 851  * Configure the Rx unit of the MAC after a reset.
 852  **/
 853
 854 static void
 855 ixgb_configure_rx(struct ixgb_adapter *adapter)
 856 {
 857         u64 rdba = adapter->rx_ring.dma;
 858         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
 859         struct ixgb_hw *hw = &adapter->hw;
 860         u32 rctl;
 861         u32 rxcsum;
 862
 863         /* make sure receives are disabled while setting up the descriptors */
 864
 865         rctl = IXGB_READ_REG(hw, RCTL);
 866         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
 867
 868         /* set the Receive Delay Timer Register */
 869
 870         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
 871
 872         /* Setup the Base and Length of the Rx Descriptor Ring */
 873
 874         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
 875         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
 876
 877         IXGB_WRITE_REG(hw, RDLEN, rdlen);
 878
 879         /* Setup the HW Rx Head and Tail Descriptor Pointers */
 880         IXGB_WRITE_REG(hw, RDH, 0);
 881         IXGB_WRITE_REG(hw, RDT, 0);
 882
 883         /* due to the hardware errata with RXDCTL, we are unable to use any of
 884          * the performance enhancing features of it without causing other
 885          * subtle bugs, some of the bugs could include receive length
 886          * corruption at high data rates (WTHRESH > 0) and/or receive
 887          * descriptor ring irregularites (particularly in hardware cache) */
 888         IXGB_WRITE_REG(hw, RXDCTL, 0);
 889
 890         /* Enable Receive Checksum Offload for TCP and UDP */
 891         if (adapter->rx_csum) {
 892                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
 893                 rxcsum |= IXGB_RXCSUM_TUOFL;
 894                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
 895         }
 896
 897         /* Enable Receives */
 898
 899         IXGB_WRITE_REG(hw, RCTL, rctl);
 900 }
 901
 902 /**
 903  * ixgb_free_tx_resources - Free Tx Resources
 904  * @adapter: board private structure
 905  *
 906  * Free all transmit software resources
 907  **/
 908
 909 void
 910 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
 911 {
 912         struct pci_dev *pdev = adapter->pdev;
 913
 914         ixgb_clean_tx_ring(adapter);
 915
 916         vfree(adapter->tx_ring.buffer_info);
 917         adapter->tx_ring.buffer_info = NULL;
 918
 919         dma_free_coherent(&pdev->dev, adapter->tx_ring.size,
 920                           adapter->tx_ring.desc, adapter->tx_ring.dma);
 921
 922         adapter->tx_ring.desc = NULL;
 923 }
 924
 925 static void
 926 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
 927                                 struct ixgb_buffer *buffer_info)
 928 {
 929         if (buffer_info->dma) {
 930                 if (buffer_info->mapped_as_page)
 931                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
 932                                        buffer_info->length, DMA_TO_DEVICE);
 933                 else
 934                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
 935                                          buffer_info->length, DMA_TO_DEVICE);
 936                 buffer_info->dma = 0;
 937         }
 938
 939         if (buffer_info->skb) {
 940                 dev_kfree_skb_any(buffer_info->skb);
 941                 buffer_info->skb = NULL;
 942         }
 943         buffer_info->time_stamp = 0;
 944         /* these fields must always be initialized in tx
 945          * buffer_info->length = 0;
 946          * buffer_info->next_to_watch = 0; */
 947 }
 948
 949 /**
 950  * ixgb_clean_tx_ring - Free Tx Buffers
 951  * @adapter: board private structure
 952  **/
 953
 954 static void
 955 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
 956 {
 957         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
 958         struct ixgb_buffer *buffer_info;
 959         unsigned long size;
 960         unsigned int i;
 961
 962         /* Free all the Tx ring sk_buffs */
 963
 964         for (i = 0; i < tx_ring->count; i++) {
 965                 buffer_info = &tx_ring->buffer_info[i];
 966                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
 967         }
 968
 969         size = sizeof(struct ixgb_buffer) * tx_ring->count;
 970         memset(tx_ring->buffer_info, 0, size);
 971
 972         /* Zero out the descriptor ring */
 973
 974         memset(tx_ring->desc, 0, tx_ring->size);
 975
 976         tx_ring->next_to_use = 0;
 977         tx_ring->next_to_clean = 0;
 978
 979         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
 980         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
 981 }
 982
 983 /**
 984  * ixgb_free_rx_resources - Free Rx Resources
 985  * @adapter: board private structure
 986  *
 987  * Free all receive software resources
 988  **/
 989
 990 void
 991 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
 992 {
 993         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 994         struct pci_dev *pdev = adapter->pdev;
 995
 996         ixgb_clean_rx_ring(adapter);
 997
 998         vfree(rx_ring->buffer_info);
 999         rx_ring->buffer_info = NULL;
1000
1001         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
1002                           rx_ring->dma);
1003
1004         rx_ring->desc = NULL;
1005 }
1006
1007 /**
1008  * ixgb_clean_rx_ring - Free Rx Buffers
1009  * @adapter: board private structure
1010  **/
1011
1012 static void
1013 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
1014 {
1015         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1016         struct ixgb_buffer *buffer_info;
1017         struct pci_dev *pdev = adapter->pdev;
1018         unsigned long size;
1019         unsigned int i;
1020
1021         /* Free all the Rx ring sk_buffs */
1022
1023         for (i = 0; i < rx_ring->count; i++) {
1024                 buffer_info = &rx_ring->buffer_info[i];
1025                 if (buffer_info->dma) {
1026                         dma_unmap_single(&pdev->dev,
1027                                          buffer_info->dma,
1028                                          buffer_info->length,
1029                                          DMA_FROM_DEVICE);
1030                         buffer_info->dma = 0;
1031                         buffer_info->length = 0;
1032                 }
1033
1034                 if (buffer_info->skb) {
1035                         dev_kfree_skb(buffer_info->skb);
1036                         buffer_info->skb = NULL;
1037                 }
1038         }
1039
1040         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1041         memset(rx_ring->buffer_info, 0, size);
1042
1043         /* Zero out the descriptor ring */
1044
1045         memset(rx_ring->desc, 0, rx_ring->size);
1046
1047         rx_ring->next_to_clean = 0;
1048         rx_ring->next_to_use = 0;
1049
1050         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1051         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1052 }
1053
1054 /**
1055  * ixgb_set_mac - Change the Ethernet Address of the NIC
1056  * @netdev: network interface device structure
1057  * @p: pointer to an address structure
1058  *
1059  * Returns 0 on success, negative on failure
1060  **/
1061
1062 static int
1063 ixgb_set_mac(struct net_device *netdev, void *p)
1064 {
1065         struct ixgb_adapter *adapter = netdev_priv(netdev);
1066         struct sockaddr *addr = p;
1067
1068         if (!is_valid_ether_addr(addr->sa_data))
1069                 return -EADDRNOTAVAIL;
1070
1071         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1072
1073         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1074
1075         return 0;
1076 }
1077
1078 /**
1079  * ixgb_set_multi - Multicast and Promiscuous mode set
1080  * @netdev: network interface device structure
1081  *
1082  * The set_multi entry point is called whenever the multicast address
1083  * list or the network interface flags are updated.  This routine is
1084  * responsible for configuring the hardware for proper multicast,
1085  * promiscuous mode, and all-multi behavior.
1086  **/
1087
1088 static void
1089 ixgb_set_multi(struct net_device *netdev)
1090 {
1091         struct ixgb_adapter *adapter = netdev_priv(netdev);
1092         struct ixgb_hw *hw = &adapter->hw;
1093         struct netdev_hw_addr *ha;
1094         u32 rctl;
1095
1096         /* Check for Promiscuous and All Multicast modes */
1097
1098         rctl = IXGB_READ_REG(hw, RCTL);
1099
1100         if (netdev->flags & IFF_PROMISC) {
1101                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1102                 /* disable VLAN filtering */
1103                 rctl &= ~IXGB_RCTL_CFIEN;
1104                 rctl &= ~IXGB_RCTL_VFE;
1105         } else {
1106                 if (netdev->flags & IFF_ALLMULTI) {
1107                         rctl |= IXGB_RCTL_MPE;
1108                         rctl &= ~IXGB_RCTL_UPE;
1109                 } else {
1110                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1111                 }
1112                 /* enable VLAN filtering */
1113                 rctl |= IXGB_RCTL_VFE;
1114                 rctl &= ~IXGB_RCTL_CFIEN;
1115         }
1116
1117         if (netdev_mc_count(netdev) > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1118                 rctl |= IXGB_RCTL_MPE;
1119                 IXGB_WRITE_REG(hw, RCTL, rctl);
1120         } else {
1121                 u8 *mta = kmalloc(IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1122                               ETH_ALEN, GFP_ATOMIC);
1123                 u8 *addr;
1124                 if (!mta)
1125                         goto alloc_failed;
1126
1127                 IXGB_WRITE_REG(hw, RCTL, rctl);
1128
1129                 addr = mta;
1130                 netdev_for_each_mc_addr(ha, netdev) {
1131                         memcpy(addr, ha->addr, ETH_ALEN);
1132                         addr += ETH_ALEN;
1133                 }
1134
1135                 ixgb_mc_addr_list_update(hw, mta, netdev_mc_count(netdev), 0);
1136                 kfree(mta);
1137         }
1138
1139 alloc_failed:
1140         if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
1141                 ixgb_vlan_strip_enable(adapter);
1142         else
1143                 ixgb_vlan_strip_disable(adapter);
1144
1145 }
1146
1147 /**
1148  * ixgb_watchdog - Timer Call-back
1149  * @data: pointer to netdev cast into an unsigned long
1150  **/
1151
1152 static void
1153 ixgb_watchdog(unsigned long data)
1154 {
1155         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1156         struct net_device *netdev = adapter->netdev;
1157         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1158
1159         ixgb_check_for_link(&adapter->hw);
1160
1161         if (ixgb_check_for_bad_link(&adapter->hw)) {
1162                 /* force the reset path */
1163                 netif_stop_queue(netdev);
1164         }
1165
1166         if (adapter->hw.link_up) {
1167                 if (!netif_carrier_ok(netdev)) {
1168                         netdev_info(netdev,
1169                                     "NIC Link is Up 10 Gbps Full Duplex, Flow Control: %s\n",
1170                                     (adapter->hw.fc.type == ixgb_fc_full) ?
1171                                     "RX/TX" :
1172                                     (adapter->hw.fc.type == ixgb_fc_rx_pause) ?
1173                                      "RX" :
1174                                     (adapter->hw.fc.type == ixgb_fc_tx_pause) ?
1175                                     "TX" : "None");
1176                         adapter->link_speed = 10000;
1177                         adapter->link_duplex = FULL_DUPLEX;
1178                         netif_carrier_on(netdev);
1179                 }
1180         } else {
1181                 if (netif_carrier_ok(netdev)) {
1182                         adapter->link_speed = 0;
1183                         adapter->link_duplex = 0;
1184                         netdev_info(netdev, "NIC Link is Down\n");
1185                         netif_carrier_off(netdev);
1186                 }
1187         }
1188
1189         ixgb_update_stats(adapter);
1190
1191         if (!netif_carrier_ok(netdev)) {
1192                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1193                         /* We've lost link, so the controller stops DMA,
1194                          * but we've got queued Tx work that's never going
1195                          * to get done, so reset controller to flush Tx.
1196                          * (Do the reset outside of interrupt context). */
1197                         schedule_work(&adapter->tx_timeout_task);
1198                         /* return immediately since reset is imminent */
1199                         return;
1200                 }
1201         }
1202
1203         /* Force detection of hung controller every watchdog period */
1204         adapter->detect_tx_hung = true;
1205
1206         /* generate an interrupt to force clean up of any stragglers */
1207         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1208
1209         /* Reset the timer */
1210         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1211 }
1212
1213 #define IXGB_TX_FLAGS_CSUM              0x00000001
1214 #define IXGB_TX_FLAGS_VLAN              0x00000002
1215 #define IXGB_TX_FLAGS_TSO               0x00000004
1216
1217 static int
1218 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1219 {
1220         struct ixgb_context_desc *context_desc;
1221         unsigned int i;
1222         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1223         u16 ipcse, tucse, mss;
1224
1225         if (likely(skb_is_gso(skb))) {
1226                 struct ixgb_buffer *buffer_info;
1227                 struct iphdr *iph;
1228                 int err;
1229
1230                 err = skb_cow_head(skb, 0);
1231                 if (err < 0)
1232                         return err;
1233
1234                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1235                 mss = skb_shinfo(skb)->gso_size;
1236                 iph = ip_hdr(skb);
1237                 iph->tot_len = 0;
1238                 iph->check = 0;
1239                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1240                                                          iph->daddr, 0,
1241                                                          IPPROTO_TCP, 0);
1242                 ipcss = skb_network_offset(skb);
1243                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1244                 ipcse = skb_transport_offset(skb) - 1;
1245                 tucss = skb_transport_offset(skb);
1246                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1247                 tucse = 0;
1248
1249                 i = adapter->tx_ring.next_to_use;
1250                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1251                 buffer_info = &adapter->tx_ring.buffer_info[i];
1252                 WARN_ON(buffer_info->dma != 0);
1253
1254                 context_desc->ipcss = ipcss;
1255                 context_desc->ipcso = ipcso;
1256                 context_desc->ipcse = cpu_to_le16(ipcse);
1257                 context_desc->tucss = tucss;
1258                 context_desc->tucso = tucso;
1259                 context_desc->tucse = cpu_to_le16(tucse);
1260                 context_desc->mss = cpu_to_le16(mss);
1261                 context_desc->hdr_len = hdr_len;
1262                 context_desc->status = 0;
1263                 context_desc->cmd_type_len = cpu_to_le32(
1264                                                   IXGB_CONTEXT_DESC_TYPE
1265                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1266                                                 | IXGB_CONTEXT_DESC_CMD_IP
1267                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1268                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1269                                                 | (skb->len - (hdr_len)));
1270
1271
1272                 if (++i == adapter->tx_ring.count) i = 0;
1273                 adapter->tx_ring.next_to_use = i;
1274
1275                 return 1;
1276         }
1277
1278         return 0;
1279 }
1280
1281 static bool
1282 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1283 {
1284         struct ixgb_context_desc *context_desc;
1285         unsigned int i;
1286         u8 css, cso;
1287
1288         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1289                 struct ixgb_buffer *buffer_info;
1290                 css = skb_checksum_start_offset(skb);
1291                 cso = css + skb->csum_offset;
1292
1293                 i = adapter->tx_ring.next_to_use;
1294                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1295                 buffer_info = &adapter->tx_ring.buffer_info[i];
1296                 WARN_ON(buffer_info->dma != 0);
1297
1298                 context_desc->tucss = css;
1299                 context_desc->tucso = cso;
1300                 context_desc->tucse = 0;
1301                 /* zero out any previously existing data in one instruction */
1302                 *(u32 *)&(context_desc->ipcss) = 0;
1303                 context_desc->status = 0;
1304                 context_desc->hdr_len = 0;
1305                 context_desc->mss = 0;
1306                 context_desc->cmd_type_len =
1307                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1308                                     | IXGB_TX_DESC_CMD_IDE);
1309
1310                 if (++i == adapter->tx_ring.count) i = 0;
1311                 adapter->tx_ring.next_to_use = i;
1312
1313                 return true;
1314         }
1315
1316         return false;
1317 }
1318
1319 #define IXGB_MAX_TXD_PWR        14
1320 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1321
1322 static int
1323 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1324             unsigned int first)
1325 {
1326         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1327         struct pci_dev *pdev = adapter->pdev;
1328         struct ixgb_buffer *buffer_info;
1329         int len = skb_headlen(skb);
1330         unsigned int offset = 0, size, count = 0, i;
1331         unsigned int mss = skb_shinfo(skb)->gso_size;
1332         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1333         unsigned int f;
1334
1335         i = tx_ring->next_to_use;
1336
1337         while (len) {
1338                 buffer_info = &tx_ring->buffer_info[i];
1339                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1340                 /* Workaround for premature desc write-backs
1341                  * in TSO mode.  Append 4-byte sentinel desc */
1342                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1343                         size -= 4;
1344
1345                 buffer_info->length = size;
1346                 WARN_ON(buffer_info->dma != 0);
1347                 buffer_info->time_stamp = jiffies;
1348                 buffer_info->mapped_as_page = false;
1349                 buffer_info->dma = dma_map_single(&pdev->dev,
1350                                                   skb->data + offset,
1351                                                   size, DMA_TO_DEVICE);
1352                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
1353                         goto dma_error;
1354                 buffer_info->next_to_watch = 0;
1355
1356                 len -= size;
1357                 offset += size;
1358                 count++;
1359                 if (len) {
1360                         i++;
1361                         if (i == tx_ring->count)
1362                                 i = 0;
1363                 }
1364         }
1365
1366         for (f = 0; f < nr_frags; f++) {
1367                 const struct skb_frag_struct *frag;
1368
1369                 frag = &skb_shinfo(skb)->frags[f];
1370                 len = skb_frag_size(frag);
1371                 offset = 0;
1372
1373                 while (len) {
1374                         i++;
1375                         if (i == tx_ring->count)
1376                                 i = 0;
1377
1378                         buffer_info = &tx_ring->buffer_info[i];
1379                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1380
1381                         /* Workaround for premature desc write-backs
1382                          * in TSO mode.  Append 4-byte sentinel desc */
1383                         if (unlikely(mss && (f == (nr_frags - 1))
1384                                      && size == len && size > 8))
1385                                 size -= 4;
1386
1387                         buffer_info->length = size;
1388                         buffer_info->time_stamp = jiffies;
1389                         buffer_info->mapped_as_page = true;
1390                         buffer_info->dma =
1391                                 skb_frag_dma_map(&pdev->dev, frag, offset, size,
1392                                                  DMA_TO_DEVICE);
1393                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
1394                                 goto dma_error;
1395                         buffer_info->next_to_watch = 0;
1396
1397                         len -= size;
1398                         offset += size;
1399                         count++;
1400                 }
1401         }
1402         tx_ring->buffer_info[i].skb = skb;
1403         tx_ring->buffer_info[first].next_to_watch = i;
1404
1405         return count;
1406
1407 dma_error:
1408         dev_err(&pdev->dev, "TX DMA map failed\n");
1409         buffer_info->dma = 0;
1410         if (count)
1411                 count--;
1412
1413         while (count--) {
1414                 if (i==0)
1415                         i += tx_ring->count;
1416                 i--;
1417                 buffer_info = &tx_ring->buffer_info[i];
1418                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1419         }
1420
1421         return 0;
1422 }
1423
1424 static void
1425 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1426 {
1427         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1428         struct ixgb_tx_desc *tx_desc = NULL;
1429         struct ixgb_buffer *buffer_info;
1430         u32 cmd_type_len = adapter->tx_cmd_type;
1431         u8 status = 0;
1432         u8 popts = 0;
1433         unsigned int i;
1434
1435         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1436                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1437                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1438         }
1439
1440         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1441                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1442
1443         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1444                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1445
1446         i = tx_ring->next_to_use;
1447
1448         while (count--) {
1449                 buffer_info = &tx_ring->buffer_info[i];
1450                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1451                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1452                 tx_desc->cmd_type_len =
1453                         cpu_to_le32(cmd_type_len | buffer_info->length);
1454                 tx_desc->status = status;
1455                 tx_desc->popts = popts;
1456                 tx_desc->vlan = cpu_to_le16(vlan_id);
1457
1458                 if (++i == tx_ring->count) i = 0;
1459         }
1460
1461         tx_desc->cmd_type_len |=
1462                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1463
1464         /* Force memory writes to complete before letting h/w
1465          * know there are new descriptors to fetch.  (Only
1466          * applicable for weak-ordered memory model archs,
1467          * such as IA-64). */
1468         wmb();
1469
1470         tx_ring->next_to_use = i;
1471         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1472 }
1473
1474 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1475 {
1476         struct ixgb_adapter *adapter = netdev_priv(netdev);
1477         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1478
1479         netif_stop_queue(netdev);
1480         /* Herbert's original patch had:
1481          *  smp_mb__after_netif_stop_queue();
1482          * but since that doesn't exist yet, just open code it. */
1483         smp_mb();
1484
1485         /* We need to check again in a case another CPU has just
1486          * made room available. */
1487         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1488                 return -EBUSY;
1489
1490         /* A reprieve! */
1491         netif_start_queue(netdev);
1492         ++adapter->restart_queue;
1493         return 0;
1494 }
1495
1496 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1497                               struct ixgb_desc_ring *tx_ring, int size)
1498 {
1499         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1500                 return 0;
1501         return __ixgb_maybe_stop_tx(netdev, size);
1502 }
1503
1504
1505 /* Tx Descriptors needed, worst case */
1506 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1507                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1508 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1509         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1510         + 1 /* one more needed for sentinel TSO workaround */
1511
1512 static netdev_tx_t
1513 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1514 {
1515         struct ixgb_adapter *adapter = netdev_priv(netdev);
1516         unsigned int first;
1517         unsigned int tx_flags = 0;
1518         int vlan_id = 0;
1519         int count = 0;
1520         int tso;
1521
1522         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1523                 dev_kfree_skb_any(skb);
1524                 return NETDEV_TX_OK;
1525         }
1526
1527         if (skb->len <= 0) {
1528                 dev_kfree_skb_any(skb);
1529                 return NETDEV_TX_OK;
1530         }
1531
1532         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1533                      DESC_NEEDED)))
1534                 return NETDEV_TX_BUSY;
1535
1536         if (skb_vlan_tag_present(skb)) {
1537                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1538                 vlan_id = skb_vlan_tag_get(skb);
1539         }
1540
1541         first = adapter->tx_ring.next_to_use;
1542
1543         tso = ixgb_tso(adapter, skb);
1544         if (tso < 0) {
1545                 dev_kfree_skb_any(skb);
1546                 return NETDEV_TX_OK;
1547         }
1548
1549         if (likely(tso))
1550                 tx_flags |= IXGB_TX_FLAGS_TSO;
1551         else if (ixgb_tx_csum(adapter, skb))
1552                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1553
1554         count = ixgb_tx_map(adapter, skb, first);
1555
1556         if (count) {
1557                 ixgb_tx_queue(adapter, count, vlan_id, tx_flags);
1558                 /* Make sure there is space in the ring for the next send. */
1559                 ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1560
1561         } else {
1562                 dev_kfree_skb_any(skb);
1563                 adapter->tx_ring.buffer_info[first].time_stamp = 0;
1564                 adapter->tx_ring.next_to_use = first;
1565         }
1566
1567         return NETDEV_TX_OK;
1568 }
1569
1570 /**
1571  * ixgb_tx_timeout - Respond to a Tx Hang
1572  * @netdev: network interface device structure
1573  **/
1574
1575 static void
1576 ixgb_tx_timeout(struct net_device *netdev)
1577 {
1578         struct ixgb_adapter *adapter = netdev_priv(netdev);
1579
1580         /* Do the reset outside of interrupt context */
1581         schedule_work(&adapter->tx_timeout_task);
1582 }
1583
1584 static void
1585 ixgb_tx_timeout_task(struct work_struct *work)
1586 {
1587         struct ixgb_adapter *adapter =
1588                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1589
1590         adapter->tx_timeout_count++;
1591         ixgb_down(adapter, true);
1592         ixgb_up(adapter);
1593 }
1594
1595 /**
1596  * ixgb_get_stats - Get System Network Statistics
1597  * @netdev: network interface device structure
1598  *
1599  * Returns the address of the device statistics structure.
1600  * The statistics are actually updated from the timer callback.
1601  **/
1602
1603 static struct net_device_stats *
1604 ixgb_get_stats(struct net_device *netdev)
1605 {
1606         return &netdev->stats;
1607 }
1608
1609 /**
1610  * ixgb_change_mtu - Change the Maximum Transfer Unit
1611  * @netdev: network interface device structure
1612  * @new_mtu: new value for maximum frame size
1613  *
1614  * Returns 0 on success, negative on failure
1615  **/
1616
1617 static int
1618 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1619 {
1620         struct ixgb_adapter *adapter = netdev_priv(netdev);
1621         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1622         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1623
1624         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1625         if ((new_mtu < 68) ||
1626             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1627                 netif_err(adapter, probe, adapter->netdev,
1628                           "Invalid MTU setting %d\n", new_mtu);
1629                 return -EINVAL;
1630         }
1631
1632         if (old_max_frame == max_frame)
1633                 return 0;
1634
1635         if (netif_running(netdev))
1636                 ixgb_down(adapter, true);
1637
1638         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1639
1640         netdev->mtu = new_mtu;
1641
1642         if (netif_running(netdev))
1643                 ixgb_up(adapter);
1644
1645         return 0;
1646 }
1647
1648 /**
1649  * ixgb_update_stats - Update the board statistics counters.
1650  * @adapter: board private structure
1651  **/
1652
1653 void
1654 ixgb_update_stats(struct ixgb_adapter *adapter)
1655 {
1656         struct net_device *netdev = adapter->netdev;
1657         struct pci_dev *pdev = adapter->pdev;
1658
1659         /* Prevent stats update while adapter is being reset */
1660         if (pci_channel_offline(pdev))
1661                 return;
1662
1663         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1664            (netdev_mc_count(netdev) > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1665                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1666                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1667                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1668                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1669
1670                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1671                 /* fix up multicast stats by removing broadcasts */
1672                 if (multi >= bcast)
1673                         multi -= bcast;
1674
1675                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1676                 adapter->stats.mprch += (multi >> 32);
1677                 adapter->stats.bprcl += bcast_l;
1678                 adapter->stats.bprch += bcast_h;
1679         } else {
1680                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1681                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1682                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1683                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1684         }
1685         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1686         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1687         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1688         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1689         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1690         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1691         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1692         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1693         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1694         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1695         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1696         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1697         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1698         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1699         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1700         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1701         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1702         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1703         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1704         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1705         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1706         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1707         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1708         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1709         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1710         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1711         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1712         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1713         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1714         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1715         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1716         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1717         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1718         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1719         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1720         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1721         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1722         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1723         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1724         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1725         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1726         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1727         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1728         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1729         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1730         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1731         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1732         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1733         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1734         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1735         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1736         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1737         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1738         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1739         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1740         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1741
1742         /* Fill out the OS statistics structure */
1743
1744         netdev->stats.rx_packets = adapter->stats.gprcl;
1745         netdev->stats.tx_packets = adapter->stats.gptcl;
1746         netdev->stats.rx_bytes = adapter->stats.gorcl;
1747         netdev->stats.tx_bytes = adapter->stats.gotcl;
1748         netdev->stats.multicast = adapter->stats.mprcl;
1749         netdev->stats.collisions = 0;
1750
1751         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1752          * with a length in the type/len field */
1753         netdev->stats.rx_errors =
1754             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1755             adapter->stats.ruc +
1756             adapter->stats.roc /*+ adapter->stats.rlec */  +
1757             adapter->stats.icbc +
1758             adapter->stats.ecbc + adapter->stats.mpc;
1759
1760         /* see above
1761          * netdev->stats.rx_length_errors = adapter->stats.rlec;
1762          */
1763
1764         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
1765         netdev->stats.rx_fifo_errors = adapter->stats.mpc;
1766         netdev->stats.rx_missed_errors = adapter->stats.mpc;
1767         netdev->stats.rx_over_errors = adapter->stats.mpc;
1768
1769         netdev->stats.tx_errors = 0;
1770         netdev->stats.rx_frame_errors = 0;
1771         netdev->stats.tx_aborted_errors = 0;
1772         netdev->stats.tx_carrier_errors = 0;
1773         netdev->stats.tx_fifo_errors = 0;
1774         netdev->stats.tx_heartbeat_errors = 0;
1775         netdev->stats.tx_window_errors = 0;
1776 }
1777
1778 #define IXGB_MAX_INTR 10
1779 /**
1780  * ixgb_intr - Interrupt Handler
1781  * @irq: interrupt number
1782  * @data: pointer to a network interface device structure
1783  **/
1784
1785 static irqreturn_t
1786 ixgb_intr(int irq, void *data)
1787 {
1788         struct net_device *netdev = data;
1789         struct ixgb_adapter *adapter = netdev_priv(netdev);
1790         struct ixgb_hw *hw = &adapter->hw;
1791         u32 icr = IXGB_READ_REG(hw, ICR);
1792
1793         if (unlikely(!icr))
1794                 return IRQ_NONE;  /* Not our interrupt */
1795
1796         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1797                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1798                         mod_timer(&adapter->watchdog_timer, jiffies);
1799
1800         if (napi_schedule_prep(&adapter->napi)) {
1801
1802                 /* Disable interrupts and register for poll. The flush
1803                   of the posted write is intentionally left out.
1804                 */
1805
1806                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1807                 __napi_schedule(&adapter->napi);
1808         }
1809         return IRQ_HANDLED;
1810 }
1811
1812 /**
1813  * ixgb_clean - NAPI Rx polling callback
1814  * @adapter: board private structure
1815  **/
1816
1817 static int
1818 ixgb_clean(struct napi_struct *napi, int budget)
1819 {
1820         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1821         int work_done = 0;
1822
1823         ixgb_clean_tx_irq(adapter);
1824         ixgb_clean_rx_irq(adapter, &work_done, budget);
1825
1826         /* If budget not fully consumed, exit the polling mode */
1827         if (work_done < budget) {
1828                 napi_complete(napi);
1829                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1830                         ixgb_irq_enable(adapter);
1831         }
1832
1833         return work_done;
1834 }
1835
1836 /**
1837  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1838  * @adapter: board private structure
1839  **/
1840
1841 static bool
1842 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1843 {
1844         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1845         struct net_device *netdev = adapter->netdev;
1846         struct ixgb_tx_desc *tx_desc, *eop_desc;
1847         struct ixgb_buffer *buffer_info;
1848         unsigned int i, eop;
1849         bool cleaned = false;
1850
1851         i = tx_ring->next_to_clean;
1852         eop = tx_ring->buffer_info[i].next_to_watch;
1853         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1854
1855         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1856
1857                 rmb(); /* read buffer_info after eop_desc */
1858                 for (cleaned = false; !cleaned; ) {
1859                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1860                         buffer_info = &tx_ring->buffer_info[i];
1861
1862                         if (tx_desc->popts &
1863                            (IXGB_TX_DESC_POPTS_TXSM |
1864                             IXGB_TX_DESC_POPTS_IXSM))
1865                                 adapter->hw_csum_tx_good++;
1866
1867                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1868
1869                         *(u32 *)&(tx_desc->status) = 0;
1870
1871                         cleaned = (i == eop);
1872                         if (++i == tx_ring->count) i = 0;
1873                 }
1874
1875                 eop = tx_ring->buffer_info[i].next_to_watch;
1876                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1877         }
1878
1879         tx_ring->next_to_clean = i;
1880
1881         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1882                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1883                 /* Make sure that anybody stopping the queue after this
1884                  * sees the new next_to_clean. */
1885                 smp_mb();
1886
1887                 if (netif_queue_stopped(netdev) &&
1888                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1889                         netif_wake_queue(netdev);
1890                         ++adapter->restart_queue;
1891                 }
1892         }
1893
1894         if (adapter->detect_tx_hung) {
1895                 /* detect a transmit hang in hardware, this serializes the
1896                  * check with the clearing of time_stamp and movement of i */
1897                 adapter->detect_tx_hung = false;
1898                 if (tx_ring->buffer_info[eop].time_stamp &&
1899                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1900                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1901                         IXGB_STATUS_TXOFF)) {
1902                         /* detected Tx unit hang */
1903                         netif_err(adapter, drv, adapter->netdev,
1904                                   "Detected Tx Unit Hang\n"
1905                                   "  TDH                  <%x>\n"
1906                                   "  TDT                  <%x>\n"
1907                                   "  next_to_use          <%x>\n"
1908                                   "  next_to_clean        <%x>\n"
1909                                   "buffer_info[next_to_clean]\n"
1910                                   "  time_stamp           <%lx>\n"
1911                                   "  next_to_watch        <%x>\n"
1912                                   "  jiffies              <%lx>\n"
1913                                   "  next_to_watch.status <%x>\n",
1914                                   IXGB_READ_REG(&adapter->hw, TDH),
1915                                   IXGB_READ_REG(&adapter->hw, TDT),
1916                                   tx_ring->next_to_use,
1917                                   tx_ring->next_to_clean,
1918                                   tx_ring->buffer_info[eop].time_stamp,
1919                                   eop,
1920                                   jiffies,
1921                                   eop_desc->status);
1922                         netif_stop_queue(netdev);
1923                 }
1924         }
1925
1926         return cleaned;
1927 }
1928
1929 /**
1930  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1931  * @adapter: board private structure
1932  * @rx_desc: receive descriptor
1933  * @sk_buff: socket buffer with received data
1934  **/
1935
1936 static void
1937 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1938                  struct ixgb_rx_desc *rx_desc,
1939                  struct sk_buff *skb)
1940 {
1941         /* Ignore Checksum bit is set OR
1942          * TCP Checksum has not been calculated
1943          */
1944         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1945            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1946                 skb_checksum_none_assert(skb);
1947                 return;
1948         }
1949
1950         /* At this point we know the hardware did the TCP checksum */
1951         /* now look at the TCP checksum error bit */
1952         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1953                 /* let the stack verify checksum errors */
1954                 skb_checksum_none_assert(skb);
1955                 adapter->hw_csum_rx_error++;
1956         } else {
1957                 /* TCP checksum is good */
1958                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1959                 adapter->hw_csum_rx_good++;
1960         }
1961 }
1962
1963 /*
1964  * this should improve performance for small packets with large amounts
1965  * of reassembly being done in the stack
1966  */
1967 static void ixgb_check_copybreak(struct napi_struct *napi,
1968                                  struct ixgb_buffer *buffer_info,
1969                                  u32 length, struct sk_buff **skb)
1970 {
1971         struct sk_buff *new_skb;
1972
1973         if (length > copybreak)
1974                 return;
1975
1976         new_skb = napi_alloc_skb(napi, length);
1977         if (!new_skb)
1978                 return;
1979
1980         skb_copy_to_linear_data_offset(new_skb, -NET_IP_ALIGN,
1981                                        (*skb)->data - NET_IP_ALIGN,
1982                                        length + NET_IP_ALIGN);
1983         /* save the skb in buffer_info as good */
1984         buffer_info->skb = *skb;
1985         *skb = new_skb;
1986 }
1987
1988 /**
1989  * ixgb_clean_rx_irq - Send received data up the network stack,
1990  * @adapter: board private structure
1991  **/
1992
1993 static bool
1994 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1995 {
1996         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1997         struct net_device *netdev = adapter->netdev;
1998         struct pci_dev *pdev = adapter->pdev;
1999         struct ixgb_rx_desc *rx_desc, *next_rxd;
2000         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
2001         u32 length;
2002         unsigned int i, j;
2003         int cleaned_count = 0;
2004         bool cleaned = false;
2005
2006         i = rx_ring->next_to_clean;
2007         rx_desc = IXGB_RX_DESC(*rx_ring, i);
2008         buffer_info = &rx_ring->buffer_info[i];
2009
2010         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
2011                 struct sk_buff *skb;
2012                 u8 status;
2013
2014                 if (*work_done >= work_to_do)
2015                         break;
2016
2017                 (*work_done)++;
2018                 rmb();  /* read descriptor and rx_buffer_info after status DD */
2019                 status = rx_desc->status;
2020                 skb = buffer_info->skb;
2021                 buffer_info->skb = NULL;
2022
2023                 prefetch(skb->data - NET_IP_ALIGN);
2024
2025                 if (++i == rx_ring->count)
2026                         i = 0;
2027                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
2028                 prefetch(next_rxd);
2029
2030                 j = i + 1;
2031                 if (j == rx_ring->count)
2032                         j = 0;
2033                 next2_buffer = &rx_ring->buffer_info[j];
2034                 prefetch(next2_buffer);
2035
2036                 next_buffer = &rx_ring->buffer_info[i];
2037
2038                 cleaned = true;
2039                 cleaned_count++;
2040
2041                 dma_unmap_single(&pdev->dev,
2042                                  buffer_info->dma,
2043                                  buffer_info->length,
2044                                  DMA_FROM_DEVICE);
2045                 buffer_info->dma = 0;
2046
2047                 length = le16_to_cpu(rx_desc->length);
2048                 rx_desc->length = 0;
2049
2050                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
2051
2052                         /* All receives must fit into a single buffer */
2053
2054                         pr_debug("Receive packet consumed multiple buffers length<%x>\n",
2055                                  length);
2056
2057                         dev_kfree_skb_irq(skb);
2058                         goto rxdesc_done;
2059                 }
2060
2061                 if (unlikely(rx_desc->errors &
2062                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
2063                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
2064                         dev_kfree_skb_irq(skb);
2065                         goto rxdesc_done;
2066                 }
2067
2068                 ixgb_check_copybreak(&adapter->napi, buffer_info, length, &skb);
2069
2070                 /* Good Receive */
2071                 skb_put(skb, length);
2072
2073                 /* Receive Checksum Offload */
2074                 ixgb_rx_checksum(adapter, rx_desc, skb);
2075
2076                 skb->protocol = eth_type_trans(skb, netdev);
2077                 if (status & IXGB_RX_DESC_STATUS_VP)
2078                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2079                                        le16_to_cpu(rx_desc->special));
2080
2081                 netif_receive_skb(skb);
2082
2083 rxdesc_done:
2084                 /* clean up descriptor, might be written over by hw */
2085                 rx_desc->status = 0;
2086
2087                 /* return some buffers to hardware, one at a time is too slow */
2088                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2089                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2090                         cleaned_count = 0;
2091                 }
2092
2093                 /* use prefetched values */
2094                 rx_desc = next_rxd;
2095                 buffer_info = next_buffer;
2096         }
2097
2098         rx_ring->next_to_clean = i;
2099
2100         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2101         if (cleaned_count)
2102                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2103
2104         return cleaned;
2105 }
2106
2107 /**
2108  * ixgb_alloc_rx_buffers - Replace used receive buffers
2109  * @adapter: address of board private structure
2110  **/
2111
2112 static void
2113 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2114 {
2115         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2116         struct net_device *netdev = adapter->netdev;
2117         struct pci_dev *pdev = adapter->pdev;
2118         struct ixgb_rx_desc *rx_desc;
2119         struct ixgb_buffer *buffer_info;
2120         struct sk_buff *skb;
2121         unsigned int i;
2122         long cleancount;
2123
2124         i = rx_ring->next_to_use;
2125         buffer_info = &rx_ring->buffer_info[i];
2126         cleancount = IXGB_DESC_UNUSED(rx_ring);
2127
2128
2129         /* leave three descriptors unused */
2130         while (--cleancount > 2 && cleaned_count--) {
2131                 /* recycle! its good for you */
2132                 skb = buffer_info->skb;
2133                 if (skb) {
2134                         skb_trim(skb, 0);
2135                         goto map_skb;
2136                 }
2137
2138                 skb = netdev_alloc_skb_ip_align(netdev, adapter->rx_buffer_len);
2139                 if (unlikely(!skb)) {
2140                         /* Better luck next round */
2141                         adapter->alloc_rx_buff_failed++;
2142                         break;
2143                 }
2144
2145                 buffer_info->skb = skb;
2146                 buffer_info->length = adapter->rx_buffer_len;
2147 map_skb:
2148                 buffer_info->dma = dma_map_single(&pdev->dev,
2149                                                   skb->data,
2150                                                   adapter->rx_buffer_len,
2151                                                   DMA_FROM_DEVICE);
2152                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
2153                         adapter->alloc_rx_buff_failed++;
2154                         break;
2155                 }
2156
2157                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2158                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2159                 /* guarantee DD bit not set now before h/w gets descriptor
2160                  * this is the rest of the workaround for h/w double
2161                  * writeback. */
2162                 rx_desc->status = 0;
2163
2164
2165                 if (++i == rx_ring->count)
2166                         i = 0;
2167                 buffer_info = &rx_ring->buffer_info[i];
2168         }
2169
2170         if (likely(rx_ring->next_to_use != i)) {
2171                 rx_ring->next_to_use = i;
2172                 if (unlikely(i-- == 0))
2173                         i = (rx_ring->count - 1);
2174
2175                 /* Force memory writes to complete before letting h/w
2176                  * know there are new descriptors to fetch.  (Only
2177                  * applicable for weak-ordered memory model archs, such
2178                  * as IA-64). */
2179                 wmb();
2180                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2181         }
2182 }
2183
2184 static void
2185 ixgb_vlan_strip_enable(struct ixgb_adapter *adapter)
2186 {
2187         u32 ctrl;
2188
2189         /* enable VLAN tag insert/strip */
2190         ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2191         ctrl |= IXGB_CTRL0_VME;
2192         IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2193 }
2194
2195 static void
2196 ixgb_vlan_strip_disable(struct ixgb_adapter *adapter)
2197 {
2198         u32 ctrl;
2199
2200         /* disable VLAN tag insert/strip */
2201         ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2202         ctrl &= ~IXGB_CTRL0_VME;
2203         IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2204 }
2205
2206 static int
2207 ixgb_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2208 {
2209         struct ixgb_adapter *adapter = netdev_priv(netdev);
2210         u32 vfta, index;
2211
2212         /* add VID to filter table */
2213
2214         index = (vid >> 5) & 0x7F;
2215         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2216         vfta |= (1 << (vid & 0x1F));
2217         ixgb_write_vfta(&adapter->hw, index, vfta);
2218         set_bit(vid, adapter->active_vlans);
2219
2220         return 0;
2221 }
2222
2223 static int
2224 ixgb_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2225 {
2226         struct ixgb_adapter *adapter = netdev_priv(netdev);
2227         u32 vfta, index;
2228
2229         /* remove VID from filter table */
2230
2231         index = (vid >> 5) & 0x7F;
2232         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2233         vfta &= ~(1 << (vid & 0x1F));
2234         ixgb_write_vfta(&adapter->hw, index, vfta);
2235         clear_bit(vid, adapter->active_vlans);
2236
2237         return 0;
2238 }
2239
2240 static void
2241 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2242 {
2243         u16 vid;
2244
2245         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2246                 ixgb_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
2247 }
2248
2249 #ifdef CONFIG_NET_POLL_CONTROLLER
2250 /*
2251  * Polling 'interrupt' - used by things like netconsole to send skbs
2252  * without having to re-enable interrupts. It's not called while
2253  * the interrupt routine is executing.
2254  */
2255
2256 static void ixgb_netpoll(struct net_device *dev)
2257 {
2258         struct ixgb_adapter *adapter = netdev_priv(dev);
2259
2260         disable_irq(adapter->pdev->irq);
2261         ixgb_intr(adapter->pdev->irq, dev);
2262         enable_irq(adapter->pdev->irq);
2263 }
2264 #endif
2265
2266 /**
2267  * ixgb_io_error_detected - called when PCI error is detected
2268  * @pdev:    pointer to pci device with error
2269  * @state:   pci channel state after error
2270  *
2271  * This callback is called by the PCI subsystem whenever
2272  * a PCI bus error is detected.
2273  */
2274 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2275                                                enum pci_channel_state state)
2276 {
2277         struct net_device *netdev = pci_get_drvdata(pdev);
2278         struct ixgb_adapter *adapter = netdev_priv(netdev);
2279
2280         netif_device_detach(netdev);
2281
2282         if (state == pci_channel_io_perm_failure)
2283                 return PCI_ERS_RESULT_DISCONNECT;
2284
2285         if (netif_running(netdev))
2286                 ixgb_down(adapter, true);
2287
2288         pci_disable_device(pdev);
2289
2290         /* Request a slot reset. */
2291         return PCI_ERS_RESULT_NEED_RESET;
2292 }
2293
2294 /**
2295  * ixgb_io_slot_reset - called after the pci bus has been reset.
2296  * @pdev    pointer to pci device with error
2297  *
2298  * This callback is called after the PCI bus has been reset.
2299  * Basically, this tries to restart the card from scratch.
2300  * This is a shortened version of the device probe/discovery code,
2301  * it resembles the first-half of the ixgb_probe() routine.
2302  */
2303 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2304 {
2305         struct net_device *netdev = pci_get_drvdata(pdev);
2306         struct ixgb_adapter *adapter = netdev_priv(netdev);
2307
2308         if (pci_enable_device(pdev)) {
2309                 netif_err(adapter, probe, adapter->netdev,
2310                           "Cannot re-enable PCI device after reset\n");
2311                 return PCI_ERS_RESULT_DISCONNECT;
2312         }
2313
2314         /* Perform card reset only on one instance of the card */
2315         if (0 != PCI_FUNC (pdev->devfn))
2316                 return PCI_ERS_RESULT_RECOVERED;
2317
2318         pci_set_master(pdev);
2319
2320         netif_carrier_off(netdev);
2321         netif_stop_queue(netdev);
2322         ixgb_reset(adapter);
2323
2324         /* Make sure the EEPROM is good */
2325         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2326                 netif_err(adapter, probe, adapter->netdev,
2327                           "After reset, the EEPROM checksum is not valid\n");
2328                 return PCI_ERS_RESULT_DISCONNECT;
2329         }
2330         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2331         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2332
2333         if (!is_valid_ether_addr(netdev->perm_addr)) {
2334                 netif_err(adapter, probe, adapter->netdev,
2335                           "After reset, invalid MAC address\n");
2336                 return PCI_ERS_RESULT_DISCONNECT;
2337         }
2338
2339         return PCI_ERS_RESULT_RECOVERED;
2340 }
2341
2342 /**
2343  * ixgb_io_resume - called when its OK to resume normal operations
2344  * @pdev    pointer to pci device with error
2345  *
2346  * The error recovery driver tells us that its OK to resume
2347  * normal operation. Implementation resembles the second-half
2348  * of the ixgb_probe() routine.
2349  */
2350 static void ixgb_io_resume(struct pci_dev *pdev)
2351 {
2352         struct net_device *netdev = pci_get_drvdata(pdev);
2353         struct ixgb_adapter *adapter = netdev_priv(netdev);
2354
2355         pci_set_master(pdev);
2356
2357         if (netif_running(netdev)) {
2358                 if (ixgb_up(adapter)) {
2359                         pr_err("can't bring device back up after reset\n");
2360                         return;
2361                 }
2362         }
2363
2364         netif_device_attach(netdev);
2365         mod_timer(&adapter->watchdog_timer, jiffies);
2366 }
2367
2368 /* ixgb_main.c */