git.oblomov.eu Git - linux-2.6/blob - 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 struct pci_device_id 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 int ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
  85 static struct net_device_stats *ixgb_get_stats(struct net_device *netdev);
  86 static int ixgb_change_mtu(struct net_device *netdev, int new_mtu);
  87 static int ixgb_set_mac(struct net_device *netdev, void *p);
  88 static irqreturn_t ixgb_intr(int irq, void *data);
  89 static bool ixgb_clean_tx_irq(struct ixgb_adapter *adapter);
  90
  91 static int ixgb_clean(struct napi_struct *, int);
  92 static bool ixgb_clean_rx_irq(struct ixgb_adapter *, int *, int);
  93 static void ixgb_alloc_rx_buffers(struct ixgb_adapter *, int);
  94
  95 static void ixgb_tx_timeout(struct net_device *dev);
  96 static void ixgb_tx_timeout_task(struct work_struct *work);
  97
  98 static void ixgb_vlan_rx_register(struct net_device *netdev,
  99                                   struct vlan_group *grp);
 100 static void ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid);
 101 static void ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid);
 102 static void ixgb_restore_vlan(struct ixgb_adapter *adapter);
 103
 104 #ifdef CONFIG_NET_POLL_CONTROLLER
 105 /* for netdump / net console */
 106 static void ixgb_netpoll(struct net_device *dev);
 107 #endif
 108
 109 static pci_ers_result_t ixgb_io_error_detected (struct pci_dev *pdev,
 110                              enum pci_channel_state state);
 111 static pci_ers_result_t ixgb_io_slot_reset (struct pci_dev *pdev);
 112 static void ixgb_io_resume (struct pci_dev *pdev);
 113
 114 static struct pci_error_handlers ixgb_err_handler = {
 115         .error_detected = ixgb_io_error_detected,
 116         .slot_reset = ixgb_io_slot_reset,
 117         .resume = ixgb_io_resume,
 118 };
 119
 120 static struct pci_driver ixgb_driver = {
 121         .name     = ixgb_driver_name,
 122         .id_table = ixgb_pci_tbl,
 123         .probe    = ixgb_probe,
 124         .remove   = __devexit_p(ixgb_remove),
 125         .err_handler = &ixgb_err_handler
 126 };
 127
 128 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 129 MODULE_DESCRIPTION("Intel(R) PRO/10GbE Network Driver");
 130 MODULE_LICENSE("GPL");
 131 MODULE_VERSION(DRV_VERSION);
 132
 133 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
 134 static int debug = DEFAULT_DEBUG_LEVEL_SHIFT;
 135 module_param(debug, int, 0);
 136 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 137
 138 /**
 139  * ixgb_init_module - Driver Registration Routine
 140  *
 141  * ixgb_init_module is the first routine called when the driver is
 142  * loaded. All it does is register with the PCI subsystem.
 143  **/
 144
 145 static int __init
 146 ixgb_init_module(void)
 147 {
 148         printk(KERN_INFO "%s - version %s\n",
 149                ixgb_driver_string, ixgb_driver_version);
 150
 151         printk(KERN_INFO "%s\n", ixgb_copyright);
 152
 153         return pci_register_driver(&ixgb_driver);
 154 }
 155
 156 module_init(ixgb_init_module);
 157
 158 /**
 159  * ixgb_exit_module - Driver Exit Cleanup Routine
 160  *
 161  * ixgb_exit_module is called just before the driver is removed
 162  * from memory.
 163  **/
 164
 165 static void __exit
 166 ixgb_exit_module(void)
 167 {
 168         pci_unregister_driver(&ixgb_driver);
 169 }
 170
 171 module_exit(ixgb_exit_module);
 172
 173 /**
 174  * ixgb_irq_disable - Mask off interrupt generation on the NIC
 175  * @adapter: board private structure
 176  **/
 177
 178 static void
 179 ixgb_irq_disable(struct ixgb_adapter *adapter)
 180 {
 181         IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
 182         IXGB_WRITE_FLUSH(&adapter->hw);
 183         synchronize_irq(adapter->pdev->irq);
 184 }
 185
 186 /**
 187  * ixgb_irq_enable - Enable default interrupt generation settings
 188  * @adapter: board private structure
 189  **/
 190
 191 static void
 192 ixgb_irq_enable(struct ixgb_adapter *adapter)
 193 {
 194         u32 val = IXGB_INT_RXT0 | IXGB_INT_RXDMT0 |
 195                   IXGB_INT_TXDW | IXGB_INT_LSC;
 196         if (adapter->hw.subsystem_vendor_id == SUN_SUBVENDOR_ID)
 197                 val |= IXGB_INT_GPI0;
 198         IXGB_WRITE_REG(&adapter->hw, IMS, val);
 199         IXGB_WRITE_FLUSH(&adapter->hw);
 200 }
 201
 202 int
 203 ixgb_up(struct ixgb_adapter *adapter)
 204 {
 205         struct net_device *netdev = adapter->netdev;
 206         int err, irq_flags = IRQF_SHARED;
 207         int max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 208         struct ixgb_hw *hw = &adapter->hw;
 209
 210         /* hardware has been reset, we need to reload some things */
 211
 212         ixgb_rar_set(hw, netdev->dev_addr, 0);
 213         ixgb_set_multi(netdev);
 214
 215         ixgb_restore_vlan(adapter);
 216
 217         ixgb_configure_tx(adapter);
 218         ixgb_setup_rctl(adapter);
 219         ixgb_configure_rx(adapter);
 220         ixgb_alloc_rx_buffers(adapter, IXGB_DESC_UNUSED(&adapter->rx_ring));
 221
 222         /* disable interrupts and get the hardware into a known state */
 223         IXGB_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
 224
 225         /* only enable MSI if bus is in PCI-X mode */
 226         if (IXGB_READ_REG(&adapter->hw, STATUS) & IXGB_STATUS_PCIX_MODE) {
 227                 err = pci_enable_msi(adapter->pdev);
 228                 if (!err) {
 229                         adapter->have_msi = 1;
 230                         irq_flags = 0;
 231                 }
 232                 /* proceed to try to request regular interrupt */
 233         }
 234
 235         err = request_irq(adapter->pdev->irq, &ixgb_intr, irq_flags,
 236                           netdev->name, netdev);
 237         if (err) {
 238                 if (adapter->have_msi)
 239                         pci_disable_msi(adapter->pdev);
 240                 DPRINTK(PROBE, ERR,
 241                  "Unable to allocate interrupt Error: %d\n", err);
 242                 return err;
 243         }
 244
 245         if ((hw->max_frame_size != max_frame) ||
 246                 (hw->max_frame_size !=
 247                 (IXGB_READ_REG(hw, MFS) >> IXGB_MFS_SHIFT))) {
 248
 249                 hw->max_frame_size = max_frame;
 250
 251                 IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
 252
 253                 if (hw->max_frame_size >
 254                    IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
 255                         u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
 256
 257                         if (!(ctrl0 & IXGB_CTRL0_JFE)) {
 258                                 ctrl0 |= IXGB_CTRL0_JFE;
 259                                 IXGB_WRITE_REG(hw, CTRL0, ctrl0);
 260                         }
 261                 }
 262         }
 263
 264         clear_bit(__IXGB_DOWN, &adapter->flags);
 265
 266         napi_enable(&adapter->napi);
 267         ixgb_irq_enable(adapter);
 268
 269         mod_timer(&adapter->watchdog_timer, jiffies);
 270
 271         return 0;
 272 }
 273
 274 void
 275 ixgb_down(struct ixgb_adapter *adapter, bool kill_watchdog)
 276 {
 277         struct net_device *netdev = adapter->netdev;
 278
 279         /* prevent the interrupt handler from restarting watchdog */
 280         set_bit(__IXGB_DOWN, &adapter->flags);
 281
 282         napi_disable(&adapter->napi);
 283         /* waiting for NAPI to complete can re-enable interrupts */
 284         ixgb_irq_disable(adapter);
 285         free_irq(adapter->pdev->irq, netdev);
 286
 287         if (adapter->have_msi)
 288                 pci_disable_msi(adapter->pdev);
 289
 290         if (kill_watchdog)
 291                 del_timer_sync(&adapter->watchdog_timer);
 292
 293         adapter->link_speed = 0;
 294         adapter->link_duplex = 0;
 295         netif_carrier_off(netdev);
 296         netif_stop_queue(netdev);
 297
 298         ixgb_reset(adapter);
 299         ixgb_clean_tx_ring(adapter);
 300         ixgb_clean_rx_ring(adapter);
 301 }
 302
 303 void
 304 ixgb_reset(struct ixgb_adapter *adapter)
 305 {
 306         struct ixgb_hw *hw = &adapter->hw;
 307
 308         ixgb_adapter_stop(hw);
 309         if (!ixgb_init_hw(hw))
 310                 DPRINTK(PROBE, ERR, "ixgb_init_hw failed.\n");
 311
 312         /* restore frame size information */
 313         IXGB_WRITE_REG(hw, MFS, hw->max_frame_size << IXGB_MFS_SHIFT);
 314         if (hw->max_frame_size >
 315             IXGB_MAX_ENET_FRAME_SIZE_WITHOUT_FCS + ENET_FCS_LENGTH) {
 316                 u32 ctrl0 = IXGB_READ_REG(hw, CTRL0);
 317                 if (!(ctrl0 & IXGB_CTRL0_JFE)) {
 318                         ctrl0 |= IXGB_CTRL0_JFE;
 319                         IXGB_WRITE_REG(hw, CTRL0, ctrl0);
 320                 }
 321         }
 322 }
 323
 324 static const struct net_device_ops ixgb_netdev_ops = {
 325         .ndo_open               = ixgb_open,
 326         .ndo_stop               = ixgb_close,
 327         .ndo_start_xmit         = ixgb_xmit_frame,
 328         .ndo_get_stats          = ixgb_get_stats,
 329         .ndo_set_multicast_list = ixgb_set_multi,
 330         .ndo_validate_addr      = eth_validate_addr,
 331         .ndo_set_mac_address    = ixgb_set_mac,
 332         .ndo_change_mtu         = ixgb_change_mtu,
 333         .ndo_tx_timeout         = ixgb_tx_timeout,
 334         .ndo_vlan_rx_register   = ixgb_vlan_rx_register,
 335         .ndo_vlan_rx_add_vid    = ixgb_vlan_rx_add_vid,
 336         .ndo_vlan_rx_kill_vid   = ixgb_vlan_rx_kill_vid,
 337 #ifdef CONFIG_NET_POLL_CONTROLLER
 338         .ndo_poll_controller    = ixgb_netpoll,
 339 #endif
 340 };
 341
 342 /**
 343  * ixgb_probe - Device Initialization Routine
 344  * @pdev: PCI device information struct
 345  * @ent: entry in ixgb_pci_tbl
 346  *
 347  * Returns 0 on success, negative on failure
 348  *
 349  * ixgb_probe initializes an adapter identified by a pci_dev structure.
 350  * The OS initialization, configuring of the adapter private structure,
 351  * and a hardware reset occur.
 352  **/
 353
 354 static int __devinit
 355 ixgb_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 356 {
 357         struct net_device *netdev = NULL;
 358         struct ixgb_adapter *adapter;
 359         static int cards_found = 0;
 360         int pci_using_dac;
 361         int i;
 362         int err;
 363
 364         err = pci_enable_device(pdev);
 365         if (err)
 366                 return err;
 367
 368         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK)) &&
 369             !(err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK))) {
 370                 pci_using_dac = 1;
 371         } else {
 372                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK)) ||
 373                     (err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK))) {
 374                         printk(KERN_ERR
 375                          "ixgb: No usable DMA configuration, aborting\n");
 376                         goto err_dma_mask;
 377                 }
 378                 pci_using_dac = 0;
 379         }
 380
 381         err = pci_request_regions(pdev, ixgb_driver_name);
 382         if (err)
 383                 goto err_request_regions;
 384
 385         pci_set_master(pdev);
 386
 387         netdev = alloc_etherdev(sizeof(struct ixgb_adapter));
 388         if (!netdev) {
 389                 err = -ENOMEM;
 390                 goto err_alloc_etherdev;
 391         }
 392
 393         SET_NETDEV_DEV(netdev, &pdev->dev);
 394
 395         pci_set_drvdata(pdev, netdev);
 396         adapter = netdev_priv(netdev);
 397         adapter->netdev = netdev;
 398         adapter->pdev = pdev;
 399         adapter->hw.back = adapter;
 400         adapter->msg_enable = netif_msg_init(debug, DEFAULT_DEBUG_LEVEL_SHIFT);
 401
 402         adapter->hw.hw_addr = pci_ioremap_bar(pdev, BAR_0);
 403         if (!adapter->hw.hw_addr) {
 404                 err = -EIO;
 405                 goto err_ioremap;
 406         }
 407
 408         for (i = BAR_1; i <= BAR_5; i++) {
 409                 if (pci_resource_len(pdev, i) == 0)
 410                         continue;
 411                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
 412                         adapter->hw.io_base = pci_resource_start(pdev, i);
 413                         break;
 414                 }
 415         }
 416
 417         netdev->netdev_ops = &ixgb_netdev_ops;
 418         ixgb_set_ethtool_ops(netdev);
 419         netdev->watchdog_timeo = 5 * HZ;
 420         netif_napi_add(netdev, &adapter->napi, ixgb_clean, 64);
 421
 422         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 423
 424         adapter->bd_number = cards_found;
 425         adapter->link_speed = 0;
 426         adapter->link_duplex = 0;
 427
 428         /* setup the private structure */
 429
 430         err = ixgb_sw_init(adapter);
 431         if (err)
 432                 goto err_sw_init;
 433
 434         netdev->features = NETIF_F_SG |
 435                            NETIF_F_HW_CSUM |
 436                            NETIF_F_HW_VLAN_TX |
 437                            NETIF_F_HW_VLAN_RX |
 438                            NETIF_F_HW_VLAN_FILTER;
 439         netdev->features |= NETIF_F_TSO;
 440
 441         if (pci_using_dac)
 442                 netdev->features |= NETIF_F_HIGHDMA;
 443
 444         /* make sure the EEPROM is good */
 445
 446         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
 447                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
 448                 err = -EIO;
 449                 goto err_eeprom;
 450         }
 451
 452         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
 453         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
 454
 455         if (!is_valid_ether_addr(netdev->perm_addr)) {
 456                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
 457                 err = -EIO;
 458                 goto err_eeprom;
 459         }
 460
 461         adapter->part_num = ixgb_get_ee_pba_number(&adapter->hw);
 462
 463         init_timer(&adapter->watchdog_timer);
 464         adapter->watchdog_timer.function = &ixgb_watchdog;
 465         adapter->watchdog_timer.data = (unsigned long)adapter;
 466
 467         INIT_WORK(&adapter->tx_timeout_task, ixgb_tx_timeout_task);
 468
 469         strcpy(netdev->name, "eth%d");
 470         err = register_netdev(netdev);
 471         if (err)
 472                 goto err_register;
 473
 474         /* we're going to reset, so assume we have no link for now */
 475
 476         netif_carrier_off(netdev);
 477         netif_stop_queue(netdev);
 478
 479         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
 480         ixgb_check_options(adapter);
 481         /* reset the hardware with the new settings */
 482
 483         ixgb_reset(adapter);
 484
 485         cards_found++;
 486         return 0;
 487
 488 err_register:
 489 err_sw_init:
 490 err_eeprom:
 491         iounmap(adapter->hw.hw_addr);
 492 err_ioremap:
 493         free_netdev(netdev);
 494 err_alloc_etherdev:
 495         pci_release_regions(pdev);
 496 err_request_regions:
 497 err_dma_mask:
 498         pci_disable_device(pdev);
 499         return err;
 500 }
 501
 502 /**
 503  * ixgb_remove - Device Removal Routine
 504  * @pdev: PCI device information struct
 505  *
 506  * ixgb_remove is called by the PCI subsystem to alert the driver
 507  * that it should release a PCI device.  The could be caused by a
 508  * Hot-Plug event, or because the driver is going to be removed from
 509  * memory.
 510  **/
 511
 512 static void __devexit
 513 ixgb_remove(struct pci_dev *pdev)
 514 {
 515         struct net_device *netdev = pci_get_drvdata(pdev);
 516         struct ixgb_adapter *adapter = netdev_priv(netdev);
 517
 518         flush_scheduled_work();
 519
 520         unregister_netdev(netdev);
 521
 522         iounmap(adapter->hw.hw_addr);
 523         pci_release_regions(pdev);
 524
 525         free_netdev(netdev);
 526 }
 527
 528 /**
 529  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
 530  * @adapter: board private structure to initialize
 531  *
 532  * ixgb_sw_init initializes the Adapter private data structure.
 533  * Fields are initialized based on PCI device information and
 534  * OS network device settings (MTU size).
 535  **/
 536
 537 static int __devinit
 538 ixgb_sw_init(struct ixgb_adapter *adapter)
 539 {
 540         struct ixgb_hw *hw = &adapter->hw;
 541         struct net_device *netdev = adapter->netdev;
 542         struct pci_dev *pdev = adapter->pdev;
 543
 544         /* PCI config space info */
 545
 546         hw->vendor_id = pdev->vendor;
 547         hw->device_id = pdev->device;
 548         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 549         hw->subsystem_id = pdev->subsystem_device;
 550
 551         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 552         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
 553
 554         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
 555            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
 556            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
 557            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
 558                 hw->mac_type = ixgb_82597;
 559         else {
 560                 /* should never have loaded on this device */
 561                 DPRINTK(PROBE, ERR, "unsupported device id\n");
 562         }
 563
 564         /* enable flow control to be programmed */
 565         hw->fc.send_xon = 1;
 566
 567         set_bit(__IXGB_DOWN, &adapter->flags);
 568         return 0;
 569 }
 570
 571 /**
 572  * ixgb_open - Called when a network interface is made active
 573  * @netdev: network interface device structure
 574  *
 575  * Returns 0 on success, negative value on failure
 576  *
 577  * The open entry point is called when a network interface is made
 578  * active by the system (IFF_UP).  At this point all resources needed
 579  * for transmit and receive operations are allocated, the interrupt
 580  * handler is registered with the OS, the watchdog timer is started,
 581  * and the stack is notified that the interface is ready.
 582  **/
 583
 584 static int
 585 ixgb_open(struct net_device *netdev)
 586 {
 587         struct ixgb_adapter *adapter = netdev_priv(netdev);
 588         int err;
 589
 590         /* allocate transmit descriptors */
 591         err = ixgb_setup_tx_resources(adapter);
 592         if (err)
 593                 goto err_setup_tx;
 594
 595         /* allocate receive descriptors */
 596
 597         err = ixgb_setup_rx_resources(adapter);
 598         if (err)
 599                 goto err_setup_rx;
 600
 601         err = ixgb_up(adapter);
 602         if (err)
 603                 goto err_up;
 604
 605         return 0;
 606
 607 err_up:
 608         ixgb_free_rx_resources(adapter);
 609 err_setup_rx:
 610         ixgb_free_tx_resources(adapter);
 611 err_setup_tx:
 612         ixgb_reset(adapter);
 613
 614         return err;
 615 }
 616
 617 /**
 618  * ixgb_close - Disables a network interface
 619  * @netdev: network interface device structure
 620  *
 621  * Returns 0, this is not allowed to fail
 622  *
 623  * The close entry point is called when an interface is de-activated
 624  * by the OS.  The hardware is still under the drivers control, but
 625  * needs to be disabled.  A global MAC reset is issued to stop the
 626  * hardware, and all transmit and receive resources are freed.
 627  **/
 628
 629 static int
 630 ixgb_close(struct net_device *netdev)
 631 {
 632         struct ixgb_adapter *adapter = netdev_priv(netdev);
 633
 634         ixgb_down(adapter, true);
 635
 636         ixgb_free_tx_resources(adapter);
 637         ixgb_free_rx_resources(adapter);
 638
 639         return 0;
 640 }
 641
 642 /**
 643  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
 644  * @adapter: board private structure
 645  *
 646  * Return 0 on success, negative on failure
 647  **/
 648
 649 int
 650 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
 651 {
 652         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
 653         struct pci_dev *pdev = adapter->pdev;
 654         int size;
 655
 656         size = sizeof(struct ixgb_buffer) * txdr->count;
 657         txdr->buffer_info = vmalloc(size);
 658         if (!txdr->buffer_info) {
 659                 DPRINTK(PROBE, ERR,
 660                  "Unable to allocate transmit descriptor ring memory\n");
 661                 return -ENOMEM;
 662         }
 663         memset(txdr->buffer_info, 0, size);
 664
 665         /* round up to nearest 4K */
 666
 667         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
 668         txdr->size = ALIGN(txdr->size, 4096);
 669
 670         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
 671         if (!txdr->desc) {
 672                 vfree(txdr->buffer_info);
 673                 DPRINTK(PROBE, ERR,
 674                  "Unable to allocate transmit descriptor memory\n");
 675                 return -ENOMEM;
 676         }
 677         memset(txdr->desc, 0, txdr->size);
 678
 679         txdr->next_to_use = 0;
 680         txdr->next_to_clean = 0;
 681
 682         return 0;
 683 }
 684
 685 /**
 686  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
 687  * @adapter: board private structure
 688  *
 689  * Configure the Tx unit of the MAC after a reset.
 690  **/
 691
 692 static void
 693 ixgb_configure_tx(struct ixgb_adapter *adapter)
 694 {
 695         u64 tdba = adapter->tx_ring.dma;
 696         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
 697         u32 tctl;
 698         struct ixgb_hw *hw = &adapter->hw;
 699
 700         /* Setup the Base and Length of the Tx Descriptor Ring
 701          * tx_ring.dma can be either a 32 or 64 bit value
 702          */
 703
 704         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
 705         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
 706
 707         IXGB_WRITE_REG(hw, TDLEN, tdlen);
 708
 709         /* Setup the HW Tx Head and Tail descriptor pointers */
 710
 711         IXGB_WRITE_REG(hw, TDH, 0);
 712         IXGB_WRITE_REG(hw, TDT, 0);
 713
 714         /* don't set up txdctl, it induces performance problems if configured
 715          * incorrectly */
 716         /* Set the Tx Interrupt Delay register */
 717
 718         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
 719
 720         /* Program the Transmit Control Register */
 721
 722         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
 723         IXGB_WRITE_REG(hw, TCTL, tctl);
 724
 725         /* Setup Transmit Descriptor Settings for this adapter */
 726         adapter->tx_cmd_type =
 727                 IXGB_TX_DESC_TYPE |
 728                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
 729 }
 730
 731 /**
 732  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
 733  * @adapter: board private structure
 734  *
 735  * Returns 0 on success, negative on failure
 736  **/
 737
 738 int
 739 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
 740 {
 741         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
 742         struct pci_dev *pdev = adapter->pdev;
 743         int size;
 744
 745         size = sizeof(struct ixgb_buffer) * rxdr->count;
 746         rxdr->buffer_info = vmalloc(size);
 747         if (!rxdr->buffer_info) {
 748                 DPRINTK(PROBE, ERR,
 749                  "Unable to allocate receive descriptor ring\n");
 750                 return -ENOMEM;
 751         }
 752         memset(rxdr->buffer_info, 0, size);
 753
 754         /* Round up to nearest 4K */
 755
 756         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
 757         rxdr->size = ALIGN(rxdr->size, 4096);
 758
 759         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
 760
 761         if (!rxdr->desc) {
 762                 vfree(rxdr->buffer_info);
 763                 DPRINTK(PROBE, ERR,
 764                  "Unable to allocate receive descriptors\n");
 765                 return -ENOMEM;
 766         }
 767         memset(rxdr->desc, 0, rxdr->size);
 768
 769         rxdr->next_to_clean = 0;
 770         rxdr->next_to_use = 0;
 771
 772         return 0;
 773 }
 774
 775 /**
 776  * ixgb_setup_rctl - configure the receive control register
 777  * @adapter: Board private structure
 778  **/
 779
 780 static void
 781 ixgb_setup_rctl(struct ixgb_adapter *adapter)
 782 {
 783         u32 rctl;
 784
 785         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
 786
 787         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
 788
 789         rctl |=
 790                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
 791                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
 792                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
 793
 794         rctl |= IXGB_RCTL_SECRC;
 795
 796         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
 797                 rctl |= IXGB_RCTL_BSIZE_2048;
 798         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
 799                 rctl |= IXGB_RCTL_BSIZE_4096;
 800         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
 801                 rctl |= IXGB_RCTL_BSIZE_8192;
 802         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
 803                 rctl |= IXGB_RCTL_BSIZE_16384;
 804
 805         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
 806 }
 807
 808 /**
 809  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
 810  * @adapter: board private structure
 811  *
 812  * Configure the Rx unit of the MAC after a reset.
 813  **/
 814
 815 static void
 816 ixgb_configure_rx(struct ixgb_adapter *adapter)
 817 {
 818         u64 rdba = adapter->rx_ring.dma;
 819         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
 820         struct ixgb_hw *hw = &adapter->hw;
 821         u32 rctl;
 822         u32 rxcsum;
 823
 824         /* make sure receives are disabled while setting up the descriptors */
 825
 826         rctl = IXGB_READ_REG(hw, RCTL);
 827         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
 828
 829         /* set the Receive Delay Timer Register */
 830
 831         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
 832
 833         /* Setup the Base and Length of the Rx Descriptor Ring */
 834
 835         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
 836         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
 837
 838         IXGB_WRITE_REG(hw, RDLEN, rdlen);
 839
 840         /* Setup the HW Rx Head and Tail Descriptor Pointers */
 841         IXGB_WRITE_REG(hw, RDH, 0);
 842         IXGB_WRITE_REG(hw, RDT, 0);
 843
 844         /* due to the hardware errata with RXDCTL, we are unable to use any of
 845          * the performance enhancing features of it without causing other
 846          * subtle bugs, some of the bugs could include receive length
 847          * corruption at high data rates (WTHRESH > 0) and/or receive
 848          * descriptor ring irregularites (particularly in hardware cache) */
 849         IXGB_WRITE_REG(hw, RXDCTL, 0);
 850
 851         /* Enable Receive Checksum Offload for TCP and UDP */
 852         if (adapter->rx_csum) {
 853                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
 854                 rxcsum |= IXGB_RXCSUM_TUOFL;
 855                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
 856         }
 857
 858         /* Enable Receives */
 859
 860         IXGB_WRITE_REG(hw, RCTL, rctl);
 861 }
 862
 863 /**
 864  * ixgb_free_tx_resources - Free Tx Resources
 865  * @adapter: board private structure
 866  *
 867  * Free all transmit software resources
 868  **/
 869
 870 void
 871 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
 872 {
 873         struct pci_dev *pdev = adapter->pdev;
 874
 875         ixgb_clean_tx_ring(adapter);
 876
 877         vfree(adapter->tx_ring.buffer_info);
 878         adapter->tx_ring.buffer_info = NULL;
 879
 880         pci_free_consistent(pdev, adapter->tx_ring.size,
 881                             adapter->tx_ring.desc, adapter->tx_ring.dma);
 882
 883         adapter->tx_ring.desc = NULL;
 884 }
 885
 886 static void
 887 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
 888                                 struct ixgb_buffer *buffer_info)
 889 {
 890         struct pci_dev *pdev = adapter->pdev;
 891
 892         if (buffer_info->dma)
 893                 pci_unmap_page(pdev, buffer_info->dma, buffer_info->length,
 894                                PCI_DMA_TODEVICE);
 895
 896         /* okay to call kfree_skb here instead of kfree_skb_any because
 897          * this is never called in interrupt context */
 898         if (buffer_info->skb)
 899                 dev_kfree_skb(buffer_info->skb);
 900
 901         buffer_info->skb = NULL;
 902         buffer_info->dma = 0;
 903         buffer_info->time_stamp = 0;
 904         /* these fields must always be initialized in tx
 905          * buffer_info->length = 0;
 906          * buffer_info->next_to_watch = 0; */
 907 }
 908
 909 /**
 910  * ixgb_clean_tx_ring - Free Tx Buffers
 911  * @adapter: board private structure
 912  **/
 913
 914 static void
 915 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
 916 {
 917         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
 918         struct ixgb_buffer *buffer_info;
 919         unsigned long size;
 920         unsigned int i;
 921
 922         /* Free all the Tx ring sk_buffs */
 923
 924         for (i = 0; i < tx_ring->count; i++) {
 925                 buffer_info = &tx_ring->buffer_info[i];
 926                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
 927         }
 928
 929         size = sizeof(struct ixgb_buffer) * tx_ring->count;
 930         memset(tx_ring->buffer_info, 0, size);
 931
 932         /* Zero out the descriptor ring */
 933
 934         memset(tx_ring->desc, 0, tx_ring->size);
 935
 936         tx_ring->next_to_use = 0;
 937         tx_ring->next_to_clean = 0;
 938
 939         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
 940         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
 941 }
 942
 943 /**
 944  * ixgb_free_rx_resources - Free Rx Resources
 945  * @adapter: board private structure
 946  *
 947  * Free all receive software resources
 948  **/
 949
 950 void
 951 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
 952 {
 953         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 954         struct pci_dev *pdev = adapter->pdev;
 955
 956         ixgb_clean_rx_ring(adapter);
 957
 958         vfree(rx_ring->buffer_info);
 959         rx_ring->buffer_info = NULL;
 960
 961         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
 962
 963         rx_ring->desc = NULL;
 964 }
 965
 966 /**
 967  * ixgb_clean_rx_ring - Free Rx Buffers
 968  * @adapter: board private structure
 969  **/
 970
 971 static void
 972 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
 973 {
 974         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 975         struct ixgb_buffer *buffer_info;
 976         struct pci_dev *pdev = adapter->pdev;
 977         unsigned long size;
 978         unsigned int i;
 979
 980         /* Free all the Rx ring sk_buffs */
 981
 982         for (i = 0; i < rx_ring->count; i++) {
 983                 buffer_info = &rx_ring->buffer_info[i];
 984                 if (buffer_info->dma) {
 985                         pci_unmap_single(pdev,
 986                                          buffer_info->dma,
 987                                          buffer_info->length,
 988                                          PCI_DMA_FROMDEVICE);
 989                         buffer_info->dma = 0;
 990                         buffer_info->length = 0;
 991                 }
 992
 993                 if (buffer_info->skb) {
 994                         dev_kfree_skb(buffer_info->skb);
 995                         buffer_info->skb = NULL;
 996                 }
 997         }
 998
 999         size = sizeof(struct ixgb_buffer) * rx_ring->count;
1000         memset(rx_ring->buffer_info, 0, size);
1001
1002         /* Zero out the descriptor ring */
1003
1004         memset(rx_ring->desc, 0, rx_ring->size);
1005
1006         rx_ring->next_to_clean = 0;
1007         rx_ring->next_to_use = 0;
1008
1009         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1010         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1011 }
1012
1013 /**
1014  * ixgb_set_mac - Change the Ethernet Address of the NIC
1015  * @netdev: network interface device structure
1016  * @p: pointer to an address structure
1017  *
1018  * Returns 0 on success, negative on failure
1019  **/
1020
1021 static int
1022 ixgb_set_mac(struct net_device *netdev, void *p)
1023 {
1024         struct ixgb_adapter *adapter = netdev_priv(netdev);
1025         struct sockaddr *addr = p;
1026
1027         if (!is_valid_ether_addr(addr->sa_data))
1028                 return -EADDRNOTAVAIL;
1029
1030         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1031
1032         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1033
1034         return 0;
1035 }
1036
1037 /**
1038  * ixgb_set_multi - Multicast and Promiscuous mode set
1039  * @netdev: network interface device structure
1040  *
1041  * The set_multi entry point is called whenever the multicast address
1042  * list or the network interface flags are updated.  This routine is
1043  * responsible for configuring the hardware for proper multicast,
1044  * promiscuous mode, and all-multi behavior.
1045  **/
1046
1047 static void
1048 ixgb_set_multi(struct net_device *netdev)
1049 {
1050         struct ixgb_adapter *adapter = netdev_priv(netdev);
1051         struct ixgb_hw *hw = &adapter->hw;
1052         struct dev_mc_list *mc_ptr;
1053         u32 rctl;
1054         int i;
1055
1056         /* Check for Promiscuous and All Multicast modes */
1057
1058         rctl = IXGB_READ_REG(hw, RCTL);
1059
1060         if (netdev->flags & IFF_PROMISC) {
1061                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1062                 rctl &= ~IXGB_RCTL_VFE;
1063         } else {
1064                 if (netdev->flags & IFF_ALLMULTI) {
1065                         rctl |= IXGB_RCTL_MPE;
1066                         rctl &= ~IXGB_RCTL_UPE;
1067                 } else {
1068                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1069                 }
1070                 rctl |= IXGB_RCTL_VFE;
1071         }
1072
1073         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1074                 rctl |= IXGB_RCTL_MPE;
1075                 IXGB_WRITE_REG(hw, RCTL, rctl);
1076         } else {
1077                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1078                             IXGB_ETH_LENGTH_OF_ADDRESS];
1079
1080                 IXGB_WRITE_REG(hw, RCTL, rctl);
1081
1082                 for (i = 0, mc_ptr = netdev->mc_list;
1083                      mc_ptr;
1084                      i++, mc_ptr = mc_ptr->next)
1085                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1086                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1087
1088                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1089         }
1090 }
1091
1092 /**
1093  * ixgb_watchdog - Timer Call-back
1094  * @data: pointer to netdev cast into an unsigned long
1095  **/
1096
1097 static void
1098 ixgb_watchdog(unsigned long data)
1099 {
1100         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1101         struct net_device *netdev = adapter->netdev;
1102         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1103
1104         ixgb_check_for_link(&adapter->hw);
1105
1106         if (ixgb_check_for_bad_link(&adapter->hw)) {
1107                 /* force the reset path */
1108                 netif_stop_queue(netdev);
1109         }
1110
1111         if (adapter->hw.link_up) {
1112                 if (!netif_carrier_ok(netdev)) {
1113                         printk(KERN_INFO "ixgb: %s NIC Link is Up 10 Gbps "
1114                                "Full Duplex, Flow Control: %s\n",
1115                                netdev->name,
1116                                (adapter->hw.fc.type == ixgb_fc_full) ?
1117                                 "RX/TX" :
1118                                 ((adapter->hw.fc.type == ixgb_fc_rx_pause) ?
1119                                  "RX" :
1120                                  ((adapter->hw.fc.type == ixgb_fc_tx_pause) ?
1121                                   "TX" : "None")));
1122                         adapter->link_speed = 10000;
1123                         adapter->link_duplex = FULL_DUPLEX;
1124                         netif_carrier_on(netdev);
1125                         netif_wake_queue(netdev);
1126                 }
1127         } else {
1128                 if (netif_carrier_ok(netdev)) {
1129                         adapter->link_speed = 0;
1130                         adapter->link_duplex = 0;
1131                         printk(KERN_INFO "ixgb: %s NIC Link is Down\n",
1132                                netdev->name);
1133                         netif_carrier_off(netdev);
1134                         netif_stop_queue(netdev);
1135
1136                 }
1137         }
1138
1139         ixgb_update_stats(adapter);
1140
1141         if (!netif_carrier_ok(netdev)) {
1142                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1143                         /* We've lost link, so the controller stops DMA,
1144                          * but we've got queued Tx work that's never going
1145                          * to get done, so reset controller to flush Tx.
1146                          * (Do the reset outside of interrupt context). */
1147                         schedule_work(&adapter->tx_timeout_task);
1148                 }
1149         }
1150
1151         /* Force detection of hung controller every watchdog period */
1152         adapter->detect_tx_hung = true;
1153
1154         /* generate an interrupt to force clean up of any stragglers */
1155         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1156
1157         /* Reset the timer */
1158         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1159 }
1160
1161 #define IXGB_TX_FLAGS_CSUM              0x00000001
1162 #define IXGB_TX_FLAGS_VLAN              0x00000002
1163 #define IXGB_TX_FLAGS_TSO               0x00000004
1164
1165 static int
1166 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1167 {
1168         struct ixgb_context_desc *context_desc;
1169         unsigned int i;
1170         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1171         u16 ipcse, tucse, mss;
1172         int err;
1173
1174         if (likely(skb_is_gso(skb))) {
1175                 struct ixgb_buffer *buffer_info;
1176                 struct iphdr *iph;
1177
1178                 if (skb_header_cloned(skb)) {
1179                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1180                         if (err)
1181                                 return err;
1182                 }
1183
1184                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1185                 mss = skb_shinfo(skb)->gso_size;
1186                 iph = ip_hdr(skb);
1187                 iph->tot_len = 0;
1188                 iph->check = 0;
1189                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1190                                                          iph->daddr, 0,
1191                                                          IPPROTO_TCP, 0);
1192                 ipcss = skb_network_offset(skb);
1193                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1194                 ipcse = skb_transport_offset(skb) - 1;
1195                 tucss = skb_transport_offset(skb);
1196                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1197                 tucse = 0;
1198
1199                 i = adapter->tx_ring.next_to_use;
1200                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1201                 buffer_info = &adapter->tx_ring.buffer_info[i];
1202                 WARN_ON(buffer_info->dma != 0);
1203
1204                 context_desc->ipcss = ipcss;
1205                 context_desc->ipcso = ipcso;
1206                 context_desc->ipcse = cpu_to_le16(ipcse);
1207                 context_desc->tucss = tucss;
1208                 context_desc->tucso = tucso;
1209                 context_desc->tucse = cpu_to_le16(tucse);
1210                 context_desc->mss = cpu_to_le16(mss);
1211                 context_desc->hdr_len = hdr_len;
1212                 context_desc->status = 0;
1213                 context_desc->cmd_type_len = cpu_to_le32(
1214                                                   IXGB_CONTEXT_DESC_TYPE
1215                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1216                                                 | IXGB_CONTEXT_DESC_CMD_IP
1217                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1218                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1219                                                 | (skb->len - (hdr_len)));
1220
1221
1222                 if (++i == adapter->tx_ring.count) i = 0;
1223                 adapter->tx_ring.next_to_use = i;
1224
1225                 return 1;
1226         }
1227
1228         return 0;
1229 }
1230
1231 static bool
1232 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1233 {
1234         struct ixgb_context_desc *context_desc;
1235         unsigned int i;
1236         u8 css, cso;
1237
1238         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1239                 struct ixgb_buffer *buffer_info;
1240                 css = skb_transport_offset(skb);
1241                 cso = css + skb->csum_offset;
1242
1243                 i = adapter->tx_ring.next_to_use;
1244                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1245                 buffer_info = &adapter->tx_ring.buffer_info[i];
1246                 WARN_ON(buffer_info->dma != 0);
1247
1248                 context_desc->tucss = css;
1249                 context_desc->tucso = cso;
1250                 context_desc->tucse = 0;
1251                 /* zero out any previously existing data in one instruction */
1252                 *(u32 *)&(context_desc->ipcss) = 0;
1253                 context_desc->status = 0;
1254                 context_desc->hdr_len = 0;
1255                 context_desc->mss = 0;
1256                 context_desc->cmd_type_len =
1257                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1258                                     | IXGB_TX_DESC_CMD_IDE);
1259
1260                 if (++i == adapter->tx_ring.count) i = 0;
1261                 adapter->tx_ring.next_to_use = i;
1262
1263                 return true;
1264         }
1265
1266         return false;
1267 }
1268
1269 #define IXGB_MAX_TXD_PWR        14
1270 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1271
1272 static int
1273 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1274             unsigned int first)
1275 {
1276         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1277         struct ixgb_buffer *buffer_info;
1278         int len = skb->len;
1279         unsigned int offset = 0, size, count = 0, i;
1280         unsigned int mss = skb_shinfo(skb)->gso_size;
1281
1282         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1283         unsigned int f;
1284
1285         len -= skb->data_len;
1286
1287         i = tx_ring->next_to_use;
1288
1289         while (len) {
1290                 buffer_info = &tx_ring->buffer_info[i];
1291                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1292                 /* Workaround for premature desc write-backs
1293                  * in TSO mode.  Append 4-byte sentinel desc */
1294                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1295                         size -= 4;
1296
1297                 buffer_info->length = size;
1298                 WARN_ON(buffer_info->dma != 0);
1299                 buffer_info->time_stamp = jiffies;
1300                 buffer_info->dma =
1301                         pci_map_single(adapter->pdev,
1302                                 skb->data + offset,
1303                                 size,
1304                                 PCI_DMA_TODEVICE);
1305                 buffer_info->next_to_watch = 0;
1306
1307                 len -= size;
1308                 offset += size;
1309                 count++;
1310                 if (++i == tx_ring->count) i = 0;
1311         }
1312
1313         for (f = 0; f < nr_frags; f++) {
1314                 struct skb_frag_struct *frag;
1315
1316                 frag = &skb_shinfo(skb)->frags[f];
1317                 len = frag->size;
1318                 offset = 0;
1319
1320                 while (len) {
1321                         buffer_info = &tx_ring->buffer_info[i];
1322                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1323
1324                         /* Workaround for premature desc write-backs
1325                          * in TSO mode.  Append 4-byte sentinel desc */
1326                         if (unlikely(mss && (f == (nr_frags - 1))
1327                                      && size == len && size > 8))
1328                                 size -= 4;
1329
1330                         buffer_info->length = size;
1331                         buffer_info->time_stamp = jiffies;
1332                         buffer_info->dma =
1333                                 pci_map_page(adapter->pdev,
1334                                         frag->page,
1335                                         frag->page_offset + offset,
1336                                         size,
1337                                         PCI_DMA_TODEVICE);
1338                         buffer_info->next_to_watch = 0;
1339
1340                         len -= size;
1341                         offset += size;
1342                         count++;
1343                         if (++i == tx_ring->count) i = 0;
1344                 }
1345         }
1346         i = (i == 0) ? tx_ring->count - 1 : i - 1;
1347         tx_ring->buffer_info[i].skb = skb;
1348         tx_ring->buffer_info[first].next_to_watch = i;
1349
1350         return count;
1351 }
1352
1353 static void
1354 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1355 {
1356         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1357         struct ixgb_tx_desc *tx_desc = NULL;
1358         struct ixgb_buffer *buffer_info;
1359         u32 cmd_type_len = adapter->tx_cmd_type;
1360         u8 status = 0;
1361         u8 popts = 0;
1362         unsigned int i;
1363
1364         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1365                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1366                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1367         }
1368
1369         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1370                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1371
1372         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1373                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1374
1375         i = tx_ring->next_to_use;
1376
1377         while (count--) {
1378                 buffer_info = &tx_ring->buffer_info[i];
1379                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1380                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1381                 tx_desc->cmd_type_len =
1382                         cpu_to_le32(cmd_type_len | buffer_info->length);
1383                 tx_desc->status = status;
1384                 tx_desc->popts = popts;
1385                 tx_desc->vlan = cpu_to_le16(vlan_id);
1386
1387                 if (++i == tx_ring->count) i = 0;
1388         }
1389
1390         tx_desc->cmd_type_len |=
1391                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1392
1393         /* Force memory writes to complete before letting h/w
1394          * know there are new descriptors to fetch.  (Only
1395          * applicable for weak-ordered memory model archs,
1396          * such as IA-64). */
1397         wmb();
1398
1399         tx_ring->next_to_use = i;
1400         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1401 }
1402
1403 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1404 {
1405         struct ixgb_adapter *adapter = netdev_priv(netdev);
1406         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1407
1408         netif_stop_queue(netdev);
1409         /* Herbert's original patch had:
1410          *  smp_mb__after_netif_stop_queue();
1411          * but since that doesn't exist yet, just open code it. */
1412         smp_mb();
1413
1414         /* We need to check again in a case another CPU has just
1415          * made room available. */
1416         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1417                 return -EBUSY;
1418
1419         /* A reprieve! */
1420         netif_start_queue(netdev);
1421         ++adapter->restart_queue;
1422         return 0;
1423 }
1424
1425 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1426                               struct ixgb_desc_ring *tx_ring, int size)
1427 {
1428         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1429                 return 0;
1430         return __ixgb_maybe_stop_tx(netdev, size);
1431 }
1432
1433
1434 /* Tx Descriptors needed, worst case */
1435 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1436                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1437 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1438         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1439         + 1 /* one more needed for sentinel TSO workaround */
1440
1441 static int
1442 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1443 {
1444         struct ixgb_adapter *adapter = netdev_priv(netdev);
1445         unsigned int first;
1446         unsigned int tx_flags = 0;
1447         int vlan_id = 0;
1448         int tso;
1449
1450         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1451                 dev_kfree_skb(skb);
1452                 return NETDEV_TX_OK;
1453         }
1454
1455         if (skb->len <= 0) {
1456                 dev_kfree_skb(skb);
1457                 return 0;
1458         }
1459
1460         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1461                      DESC_NEEDED)))
1462                 return NETDEV_TX_BUSY;
1463
1464         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1465                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1466                 vlan_id = vlan_tx_tag_get(skb);
1467         }
1468
1469         first = adapter->tx_ring.next_to_use;
1470
1471         tso = ixgb_tso(adapter, skb);
1472         if (tso < 0) {
1473                 dev_kfree_skb(skb);
1474                 return NETDEV_TX_OK;
1475         }
1476
1477         if (likely(tso))
1478                 tx_flags |= IXGB_TX_FLAGS_TSO;
1479         else if (ixgb_tx_csum(adapter, skb))
1480                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1481
1482         ixgb_tx_queue(adapter, ixgb_tx_map(adapter, skb, first), vlan_id,
1483                         tx_flags);
1484
1485         netdev->trans_start = jiffies;
1486
1487         /* Make sure there is space in the ring for the next send. */
1488         ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1489
1490         return NETDEV_TX_OK;
1491 }
1492
1493 /**
1494  * ixgb_tx_timeout - Respond to a Tx Hang
1495  * @netdev: network interface device structure
1496  **/
1497
1498 static void
1499 ixgb_tx_timeout(struct net_device *netdev)
1500 {
1501         struct ixgb_adapter *adapter = netdev_priv(netdev);
1502
1503         /* Do the reset outside of interrupt context */
1504         schedule_work(&adapter->tx_timeout_task);
1505 }
1506
1507 static void
1508 ixgb_tx_timeout_task(struct work_struct *work)
1509 {
1510         struct ixgb_adapter *adapter =
1511                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1512
1513         adapter->tx_timeout_count++;
1514         ixgb_down(adapter, true);
1515         ixgb_up(adapter);
1516 }
1517
1518 /**
1519  * ixgb_get_stats - Get System Network Statistics
1520  * @netdev: network interface device structure
1521  *
1522  * Returns the address of the device statistics structure.
1523  * The statistics are actually updated from the timer callback.
1524  **/
1525
1526 static struct net_device_stats *
1527 ixgb_get_stats(struct net_device *netdev)
1528 {
1529         struct ixgb_adapter *adapter = netdev_priv(netdev);
1530
1531         return &adapter->net_stats;
1532 }
1533
1534 /**
1535  * ixgb_change_mtu - Change the Maximum Transfer Unit
1536  * @netdev: network interface device structure
1537  * @new_mtu: new value for maximum frame size
1538  *
1539  * Returns 0 on success, negative on failure
1540  **/
1541
1542 static int
1543 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1544 {
1545         struct ixgb_adapter *adapter = netdev_priv(netdev);
1546         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1547         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1548
1549         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1550         if ((new_mtu < 68) ||
1551             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1552                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1553                 return -EINVAL;
1554         }
1555
1556         if (old_max_frame == max_frame)
1557                 return 0;
1558
1559         if (netif_running(netdev))
1560                 ixgb_down(adapter, true);
1561
1562         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1563
1564         netdev->mtu = new_mtu;
1565
1566         if (netif_running(netdev))
1567                 ixgb_up(adapter);
1568
1569         return 0;
1570 }
1571
1572 /**
1573  * ixgb_update_stats - Update the board statistics counters.
1574  * @adapter: board private structure
1575  **/
1576
1577 void
1578 ixgb_update_stats(struct ixgb_adapter *adapter)
1579 {
1580         struct net_device *netdev = adapter->netdev;
1581         struct pci_dev *pdev = adapter->pdev;
1582
1583         /* Prevent stats update while adapter is being reset */
1584         if (pci_channel_offline(pdev))
1585                 return;
1586
1587         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1588            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1589                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1590                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1591                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1592                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1593
1594                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1595                 /* fix up multicast stats by removing broadcasts */
1596                 if (multi >= bcast)
1597                         multi -= bcast;
1598
1599                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1600                 adapter->stats.mprch += (multi >> 32);
1601                 adapter->stats.bprcl += bcast_l;
1602                 adapter->stats.bprch += bcast_h;
1603         } else {
1604                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1605                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1606                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1607                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1608         }
1609         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1610         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1611         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1612         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1613         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1614         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1615         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1616         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1617         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1618         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1619         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1620         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1621         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1622         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1623         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1624         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1625         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1626         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1627         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1628         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1629         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1630         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1631         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1632         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1633         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1634         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1635         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1636         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1637         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1638         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1639         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1640         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1641         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1642         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1643         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1644         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1645         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1646         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1647         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1648         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1649         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1650         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1651         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1652         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1653         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1654         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1655         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1656         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1657         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1658         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1659         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1660         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1661         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1662         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1663         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1664         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1665
1666         /* Fill out the OS statistics structure */
1667
1668         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1669         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1670         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1671         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1672         adapter->net_stats.multicast = adapter->stats.mprcl;
1673         adapter->net_stats.collisions = 0;
1674
1675         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1676          * with a length in the type/len field */
1677         adapter->net_stats.rx_errors =
1678             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1679             adapter->stats.ruc +
1680             adapter->stats.roc /*+ adapter->stats.rlec */  +
1681             adapter->stats.icbc +
1682             adapter->stats.ecbc + adapter->stats.mpc;
1683
1684         /* see above
1685          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1686          */
1687
1688         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1689         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1690         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1691         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1692
1693         adapter->net_stats.tx_errors = 0;
1694         adapter->net_stats.rx_frame_errors = 0;
1695         adapter->net_stats.tx_aborted_errors = 0;
1696         adapter->net_stats.tx_carrier_errors = 0;
1697         adapter->net_stats.tx_fifo_errors = 0;
1698         adapter->net_stats.tx_heartbeat_errors = 0;
1699         adapter->net_stats.tx_window_errors = 0;
1700 }
1701
1702 #define IXGB_MAX_INTR 10
1703 /**
1704  * ixgb_intr - Interrupt Handler
1705  * @irq: interrupt number
1706  * @data: pointer to a network interface device structure
1707  **/
1708
1709 static irqreturn_t
1710 ixgb_intr(int irq, void *data)
1711 {
1712         struct net_device *netdev = data;
1713         struct ixgb_adapter *adapter = netdev_priv(netdev);
1714         struct ixgb_hw *hw = &adapter->hw;
1715         u32 icr = IXGB_READ_REG(hw, ICR);
1716
1717         if (unlikely(!icr))
1718                 return IRQ_NONE;  /* Not our interrupt */
1719
1720         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1721                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1722                         mod_timer(&adapter->watchdog_timer, jiffies);
1723
1724         if (netif_rx_schedule_prep(&adapter->napi)) {
1725
1726                 /* Disable interrupts and register for poll. The flush
1727                   of the posted write is intentionally left out.
1728                 */
1729
1730                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1731                 __netif_rx_schedule(&adapter->napi);
1732         }
1733         return IRQ_HANDLED;
1734 }
1735
1736 /**
1737  * ixgb_clean - NAPI Rx polling callback
1738  * @adapter: board private structure
1739  **/
1740
1741 static int
1742 ixgb_clean(struct napi_struct *napi, int budget)
1743 {
1744         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1745         int work_done = 0;
1746
1747         ixgb_clean_tx_irq(adapter);
1748         ixgb_clean_rx_irq(adapter, &work_done, budget);
1749
1750         /* If budget not fully consumed, exit the polling mode */
1751         if (work_done < budget) {
1752                 netif_rx_complete(napi);
1753                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1754                         ixgb_irq_enable(adapter);
1755         }
1756
1757         return work_done;
1758 }
1759
1760 /**
1761  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1762  * @adapter: board private structure
1763  **/
1764
1765 static bool
1766 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1767 {
1768         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1769         struct net_device *netdev = adapter->netdev;
1770         struct ixgb_tx_desc *tx_desc, *eop_desc;
1771         struct ixgb_buffer *buffer_info;
1772         unsigned int i, eop;
1773         bool cleaned = false;
1774
1775         i = tx_ring->next_to_clean;
1776         eop = tx_ring->buffer_info[i].next_to_watch;
1777         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1778
1779         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1780
1781                 for (cleaned = false; !cleaned; ) {
1782                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1783                         buffer_info = &tx_ring->buffer_info[i];
1784
1785                         if (tx_desc->popts &
1786                            (IXGB_TX_DESC_POPTS_TXSM |
1787                             IXGB_TX_DESC_POPTS_IXSM))
1788                                 adapter->hw_csum_tx_good++;
1789
1790                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1791
1792                         *(u32 *)&(tx_desc->status) = 0;
1793
1794                         cleaned = (i == eop);
1795                         if (++i == tx_ring->count) i = 0;
1796                 }
1797
1798                 eop = tx_ring->buffer_info[i].next_to_watch;
1799                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1800         }
1801
1802         tx_ring->next_to_clean = i;
1803
1804         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1805                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1806                 /* Make sure that anybody stopping the queue after this
1807                  * sees the new next_to_clean. */
1808                 smp_mb();
1809
1810                 if (netif_queue_stopped(netdev) &&
1811                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1812                         netif_wake_queue(netdev);
1813                         ++adapter->restart_queue;
1814                 }
1815         }
1816
1817         if (adapter->detect_tx_hung) {
1818                 /* detect a transmit hang in hardware, this serializes the
1819                  * check with the clearing of time_stamp and movement of i */
1820                 adapter->detect_tx_hung = false;
1821                 if (tx_ring->buffer_info[eop].dma &&
1822                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1823                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1824                         IXGB_STATUS_TXOFF)) {
1825                         /* detected Tx unit hang */
1826                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1827                                         "  TDH                  <%x>\n"
1828                                         "  TDT                  <%x>\n"
1829                                         "  next_to_use          <%x>\n"
1830                                         "  next_to_clean        <%x>\n"
1831                                         "buffer_info[next_to_clean]\n"
1832                                         "  time_stamp           <%lx>\n"
1833                                         "  next_to_watch        <%x>\n"
1834                                         "  jiffies              <%lx>\n"
1835                                         "  next_to_watch.status <%x>\n",
1836                                 IXGB_READ_REG(&adapter->hw, TDH),
1837                                 IXGB_READ_REG(&adapter->hw, TDT),
1838                                 tx_ring->next_to_use,
1839                                 tx_ring->next_to_clean,
1840                                 tx_ring->buffer_info[eop].time_stamp,
1841                                 eop,
1842                                 jiffies,
1843                                 eop_desc->status);
1844                         netif_stop_queue(netdev);
1845                 }
1846         }
1847
1848         return cleaned;
1849 }
1850
1851 /**
1852  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1853  * @adapter: board private structure
1854  * @rx_desc: receive descriptor
1855  * @sk_buff: socket buffer with received data
1856  **/
1857
1858 static void
1859 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1860                  struct ixgb_rx_desc *rx_desc,
1861                  struct sk_buff *skb)
1862 {
1863         /* Ignore Checksum bit is set OR
1864          * TCP Checksum has not been calculated
1865          */
1866         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1867            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1868                 skb->ip_summed = CHECKSUM_NONE;
1869                 return;
1870         }
1871
1872         /* At this point we know the hardware did the TCP checksum */
1873         /* now look at the TCP checksum error bit */
1874         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1875                 /* let the stack verify checksum errors */
1876                 skb->ip_summed = CHECKSUM_NONE;
1877                 adapter->hw_csum_rx_error++;
1878         } else {
1879                 /* TCP checksum is good */
1880                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1881                 adapter->hw_csum_rx_good++;
1882         }
1883 }
1884
1885 /**
1886  * ixgb_clean_rx_irq - Send received data up the network stack,
1887  * @adapter: board private structure
1888  **/
1889
1890 static bool
1891 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1892 {
1893         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1894         struct net_device *netdev = adapter->netdev;
1895         struct pci_dev *pdev = adapter->pdev;
1896         struct ixgb_rx_desc *rx_desc, *next_rxd;
1897         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1898         u32 length;
1899         unsigned int i, j;
1900         int cleaned_count = 0;
1901         bool cleaned = false;
1902
1903         i = rx_ring->next_to_clean;
1904         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1905         buffer_info = &rx_ring->buffer_info[i];
1906
1907         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1908                 struct sk_buff *skb;
1909                 u8 status;
1910
1911                 if (*work_done >= work_to_do)
1912                         break;
1913
1914                 (*work_done)++;
1915                 status = rx_desc->status;
1916                 skb = buffer_info->skb;
1917                 buffer_info->skb = NULL;
1918
1919                 prefetch(skb->data - NET_IP_ALIGN);
1920
1921                 if (++i == rx_ring->count) i = 0;
1922                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1923                 prefetch(next_rxd);
1924
1925                 if ((j = i + 1) == rx_ring->count) j = 0;
1926                 next2_buffer = &rx_ring->buffer_info[j];
1927                 prefetch(next2_buffer);
1928
1929                 next_buffer = &rx_ring->buffer_info[i];
1930
1931                 cleaned = true;
1932                 cleaned_count++;
1933
1934                 pci_unmap_single(pdev,
1935                                  buffer_info->dma,
1936                                  buffer_info->length,
1937                                  PCI_DMA_FROMDEVICE);
1938                 buffer_info->dma = 0;
1939
1940                 length = le16_to_cpu(rx_desc->length);
1941                 rx_desc->length = 0;
1942
1943                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1944
1945                         /* All receives must fit into a single buffer */
1946
1947                         IXGB_DBG("Receive packet consumed multiple buffers "
1948                                          "length<%x>\n", length);
1949
1950                         dev_kfree_skb_irq(skb);
1951                         goto rxdesc_done;
1952                 }
1953
1954                 if (unlikely(rx_desc->errors &
1955                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1956                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1957                         dev_kfree_skb_irq(skb);
1958                         goto rxdesc_done;
1959                 }
1960
1961                 /* code added for copybreak, this should improve
1962                  * performance for small packets with large amounts
1963                  * of reassembly being done in the stack */
1964                 if (length < copybreak) {
1965                         struct sk_buff *new_skb =
1966                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
1967                         if (new_skb) {
1968                                 skb_reserve(new_skb, NET_IP_ALIGN);
1969                                 skb_copy_to_linear_data_offset(new_skb,
1970                                                                -NET_IP_ALIGN,
1971                                                                (skb->data -
1972                                                                 NET_IP_ALIGN),
1973                                                                (length +
1974                                                                 NET_IP_ALIGN));
1975                                 /* save the skb in buffer_info as good */
1976                                 buffer_info->skb = skb;
1977                                 skb = new_skb;
1978                         }
1979                 }
1980                 /* end copybreak code */
1981
1982                 /* Good Receive */
1983                 skb_put(skb, length);
1984
1985                 /* Receive Checksum Offload */
1986                 ixgb_rx_checksum(adapter, rx_desc, skb);
1987
1988                 skb->protocol = eth_type_trans(skb, netdev);
1989                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
1990                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
1991                                                 le16_to_cpu(rx_desc->special));
1992                 } else {
1993                         netif_receive_skb(skb);
1994                 }
1995
1996 rxdesc_done:
1997                 /* clean up descriptor, might be written over by hw */
1998                 rx_desc->status = 0;
1999
2000                 /* return some buffers to hardware, one at a time is too slow */
2001                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2002                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2003                         cleaned_count = 0;
2004                 }
2005
2006                 /* use prefetched values */
2007                 rx_desc = next_rxd;
2008                 buffer_info = next_buffer;
2009         }
2010
2011         rx_ring->next_to_clean = i;
2012
2013         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2014         if (cleaned_count)
2015                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2016
2017         return cleaned;
2018 }
2019
2020 /**
2021  * ixgb_alloc_rx_buffers - Replace used receive buffers
2022  * @adapter: address of board private structure
2023  **/
2024
2025 static void
2026 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2027 {
2028         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2029         struct net_device *netdev = adapter->netdev;
2030         struct pci_dev *pdev = adapter->pdev;
2031         struct ixgb_rx_desc *rx_desc;
2032         struct ixgb_buffer *buffer_info;
2033         struct sk_buff *skb;
2034         unsigned int i;
2035         long cleancount;
2036
2037         i = rx_ring->next_to_use;
2038         buffer_info = &rx_ring->buffer_info[i];
2039         cleancount = IXGB_DESC_UNUSED(rx_ring);
2040
2041
2042         /* leave three descriptors unused */
2043         while (--cleancount > 2 && cleaned_count--) {
2044                 /* recycle! its good for you */
2045                 skb = buffer_info->skb;
2046                 if (skb) {
2047                         skb_trim(skb, 0);
2048                         goto map_skb;
2049                 }
2050
2051                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2052                                        + NET_IP_ALIGN);
2053                 if (unlikely(!skb)) {
2054                         /* Better luck next round */
2055                         adapter->alloc_rx_buff_failed++;
2056                         break;
2057                 }
2058
2059                 /* Make buffer alignment 2 beyond a 16 byte boundary
2060                  * this will result in a 16 byte aligned IP header after
2061                  * the 14 byte MAC header is removed
2062                  */
2063                 skb_reserve(skb, NET_IP_ALIGN);
2064
2065                 buffer_info->skb = skb;
2066                 buffer_info->length = adapter->rx_buffer_len;
2067 map_skb:
2068                 buffer_info->dma = pci_map_single(pdev,
2069                                                   skb->data,
2070                                                   adapter->rx_buffer_len,
2071                                                   PCI_DMA_FROMDEVICE);
2072
2073                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2074                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2075                 /* guarantee DD bit not set now before h/w gets descriptor
2076                  * this is the rest of the workaround for h/w double
2077                  * writeback. */
2078                 rx_desc->status = 0;
2079
2080
2081                 if (++i == rx_ring->count) i = 0;
2082                 buffer_info = &rx_ring->buffer_info[i];
2083         }
2084
2085         if (likely(rx_ring->next_to_use != i)) {
2086                 rx_ring->next_to_use = i;
2087                 if (unlikely(i-- == 0))
2088                         i = (rx_ring->count - 1);
2089
2090                 /* Force memory writes to complete before letting h/w
2091                  * know there are new descriptors to fetch.  (Only
2092                  * applicable for weak-ordered memory model archs, such
2093                  * as IA-64). */
2094                 wmb();
2095                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2096         }
2097 }
2098
2099 /**
2100  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2101  *
2102  * @param netdev network interface device structure
2103  * @param grp indicates to enable or disable tagging/stripping
2104  **/
2105 static void
2106 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2107 {
2108         struct ixgb_adapter *adapter = netdev_priv(netdev);
2109         u32 ctrl, rctl;
2110
2111         ixgb_irq_disable(adapter);
2112         adapter->vlgrp = grp;
2113
2114         if (grp) {
2115                 /* enable VLAN tag insert/strip */
2116                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2117                 ctrl |= IXGB_CTRL0_VME;
2118                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2119
2120                 /* enable VLAN receive filtering */
2121
2122                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2123                 rctl &= ~IXGB_RCTL_CFIEN;
2124                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2125         } else {
2126                 /* disable VLAN tag insert/strip */
2127
2128                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2129                 ctrl &= ~IXGB_CTRL0_VME;
2130                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2131         }
2132
2133         /* don't enable interrupts unless we are UP */
2134         if (adapter->netdev->flags & IFF_UP)
2135                 ixgb_irq_enable(adapter);
2136 }
2137
2138 static void
2139 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2140 {
2141         struct ixgb_adapter *adapter = netdev_priv(netdev);
2142         u32 vfta, index;
2143
2144         /* add VID to filter table */
2145
2146         index = (vid >> 5) & 0x7F;
2147         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2148         vfta |= (1 << (vid & 0x1F));
2149         ixgb_write_vfta(&adapter->hw, index, vfta);
2150 }
2151
2152 static void
2153 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2154 {
2155         struct ixgb_adapter *adapter = netdev_priv(netdev);
2156         u32 vfta, index;
2157
2158         ixgb_irq_disable(adapter);
2159
2160         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2161
2162         /* don't enable interrupts unless we are UP */
2163         if (adapter->netdev->flags & IFF_UP)
2164                 ixgb_irq_enable(adapter);
2165
2166         /* remove VID from filter table */
2167
2168         index = (vid >> 5) & 0x7F;
2169         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2170         vfta &= ~(1 << (vid & 0x1F));
2171         ixgb_write_vfta(&adapter->hw, index, vfta);
2172 }
2173
2174 static void
2175 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2176 {
2177         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2178
2179         if (adapter->vlgrp) {
2180                 u16 vid;
2181                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2182                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2183                                 continue;
2184                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2185                 }
2186         }
2187 }
2188
2189 #ifdef CONFIG_NET_POLL_CONTROLLER
2190 /*
2191  * Polling 'interrupt' - used by things like netconsole to send skbs
2192  * without having to re-enable interrupts. It's not called while
2193  * the interrupt routine is executing.
2194  */
2195
2196 static void ixgb_netpoll(struct net_device *dev)
2197 {
2198         struct ixgb_adapter *adapter = netdev_priv(dev);
2199
2200         disable_irq(adapter->pdev->irq);
2201         ixgb_intr(adapter->pdev->irq, dev);
2202         enable_irq(adapter->pdev->irq);
2203 }
2204 #endif
2205
2206 /**
2207  * ixgb_io_error_detected() - called when PCI error is detected
2208  * @pdev    pointer to pci device with error
2209  * @state   pci channel state after error
2210  *
2211  * This callback is called by the PCI subsystem whenever
2212  * a PCI bus error is detected.
2213  */
2214 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2215                                                enum pci_channel_state state)
2216 {
2217         struct net_device *netdev = pci_get_drvdata(pdev);
2218         struct ixgb_adapter *adapter = netdev_priv(netdev);
2219
2220         if (netif_running(netdev))
2221                 ixgb_down(adapter, true);
2222
2223         pci_disable_device(pdev);
2224
2225         /* Request a slot reset. */
2226         return PCI_ERS_RESULT_NEED_RESET;
2227 }
2228
2229 /**
2230  * ixgb_io_slot_reset - called after the pci bus has been reset.
2231  * @pdev    pointer to pci device with error
2232  *
2233  * This callback is called after the PCI bus has been reset.
2234  * Basically, this tries to restart the card from scratch.
2235  * This is a shortened version of the device probe/discovery code,
2236  * it resembles the first-half of the ixgb_probe() routine.
2237  */
2238 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2239 {
2240         struct net_device *netdev = pci_get_drvdata(pdev);
2241         struct ixgb_adapter *adapter = netdev_priv(netdev);
2242
2243         if (pci_enable_device(pdev)) {
2244                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2245                 return PCI_ERS_RESULT_DISCONNECT;
2246         }
2247
2248         /* Perform card reset only on one instance of the card */
2249         if (0 != PCI_FUNC (pdev->devfn))
2250                 return PCI_ERS_RESULT_RECOVERED;
2251
2252         pci_set_master(pdev);
2253
2254         netif_carrier_off(netdev);
2255         netif_stop_queue(netdev);
2256         ixgb_reset(adapter);
2257
2258         /* Make sure the EEPROM is good */
2259         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2260                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2261                 return PCI_ERS_RESULT_DISCONNECT;
2262         }
2263         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2264         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2265
2266         if (!is_valid_ether_addr(netdev->perm_addr)) {
2267                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2268                 return PCI_ERS_RESULT_DISCONNECT;
2269         }
2270
2271         return PCI_ERS_RESULT_RECOVERED;
2272 }
2273
2274 /**
2275  * ixgb_io_resume - called when its OK to resume normal operations
2276  * @pdev    pointer to pci device with error
2277  *
2278  * The error recovery driver tells us that its OK to resume
2279  * normal operation. Implementation resembles the second-half
2280  * of the ixgb_probe() routine.
2281  */
2282 static void ixgb_io_resume(struct pci_dev *pdev)
2283 {
2284         struct net_device *netdev = pci_get_drvdata(pdev);
2285         struct ixgb_adapter *adapter = netdev_priv(netdev);
2286
2287         pci_set_master(pdev);
2288
2289         if (netif_running(netdev)) {
2290                 if (ixgb_up(adapter)) {
2291                         printk ("ixgb: can't bring device back up after reset\n");
2292                         return;
2293                 }
2294         }
2295
2296         netif_device_attach(netdev);
2297         mod_timer(&adapter->watchdog_timer, jiffies);
2298 }
2299
2300 /* ixgb_main.c */