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