SAFE public projects git trees. - safe/jmp/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_BIT_MASK(64))) &&
 369             !(err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)))) {
 370                 pci_using_dac = 1;
 371         } else {
 372                 if ((err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) ||
 373                     (err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)))) {
 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         /* carrier off reporting is important to ethtool even BEFORE open */
 475         netif_carrier_off(netdev);
 476
 477         DPRINTK(PROBE, INFO, "Intel(R) PRO/10GbE Network Connection\n");
 478         ixgb_check_options(adapter);
 479         /* reset the hardware with the new settings */
 480
 481         ixgb_reset(adapter);
 482
 483         cards_found++;
 484         return 0;
 485
 486 err_register:
 487 err_sw_init:
 488 err_eeprom:
 489         iounmap(adapter->hw.hw_addr);
 490 err_ioremap:
 491         free_netdev(netdev);
 492 err_alloc_etherdev:
 493         pci_release_regions(pdev);
 494 err_request_regions:
 495 err_dma_mask:
 496         pci_disable_device(pdev);
 497         return err;
 498 }
 499
 500 /**
 501  * ixgb_remove - Device Removal Routine
 502  * @pdev: PCI device information struct
 503  *
 504  * ixgb_remove is called by the PCI subsystem to alert the driver
 505  * that it should release a PCI device.  The could be caused by a
 506  * Hot-Plug event, or because the driver is going to be removed from
 507  * memory.
 508  **/
 509
 510 static void __devexit
 511 ixgb_remove(struct pci_dev *pdev)
 512 {
 513         struct net_device *netdev = pci_get_drvdata(pdev);
 514         struct ixgb_adapter *adapter = netdev_priv(netdev);
 515
 516         flush_scheduled_work();
 517
 518         unregister_netdev(netdev);
 519
 520         iounmap(adapter->hw.hw_addr);
 521         pci_release_regions(pdev);
 522
 523         free_netdev(netdev);
 524 }
 525
 526 /**
 527  * ixgb_sw_init - Initialize general software structures (struct ixgb_adapter)
 528  * @adapter: board private structure to initialize
 529  *
 530  * ixgb_sw_init initializes the Adapter private data structure.
 531  * Fields are initialized based on PCI device information and
 532  * OS network device settings (MTU size).
 533  **/
 534
 535 static int __devinit
 536 ixgb_sw_init(struct ixgb_adapter *adapter)
 537 {
 538         struct ixgb_hw *hw = &adapter->hw;
 539         struct net_device *netdev = adapter->netdev;
 540         struct pci_dev *pdev = adapter->pdev;
 541
 542         /* PCI config space info */
 543
 544         hw->vendor_id = pdev->vendor;
 545         hw->device_id = pdev->device;
 546         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 547         hw->subsystem_id = pdev->subsystem_device;
 548
 549         hw->max_frame_size = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
 550         adapter->rx_buffer_len = hw->max_frame_size + 8; /* + 8 for errata */
 551
 552         if ((hw->device_id == IXGB_DEVICE_ID_82597EX)
 553            || (hw->device_id == IXGB_DEVICE_ID_82597EX_CX4)
 554            || (hw->device_id == IXGB_DEVICE_ID_82597EX_LR)
 555            || (hw->device_id == IXGB_DEVICE_ID_82597EX_SR))
 556                 hw->mac_type = ixgb_82597;
 557         else {
 558                 /* should never have loaded on this device */
 559                 DPRINTK(PROBE, ERR, "unsupported device id\n");
 560         }
 561
 562         /* enable flow control to be programmed */
 563         hw->fc.send_xon = 1;
 564
 565         set_bit(__IXGB_DOWN, &adapter->flags);
 566         return 0;
 567 }
 568
 569 /**
 570  * ixgb_open - Called when a network interface is made active
 571  * @netdev: network interface device structure
 572  *
 573  * Returns 0 on success, negative value on failure
 574  *
 575  * The open entry point is called when a network interface is made
 576  * active by the system (IFF_UP).  At this point all resources needed
 577  * for transmit and receive operations are allocated, the interrupt
 578  * handler is registered with the OS, the watchdog timer is started,
 579  * and the stack is notified that the interface is ready.
 580  **/
 581
 582 static int
 583 ixgb_open(struct net_device *netdev)
 584 {
 585         struct ixgb_adapter *adapter = netdev_priv(netdev);
 586         int err;
 587
 588         /* allocate transmit descriptors */
 589         err = ixgb_setup_tx_resources(adapter);
 590         if (err)
 591                 goto err_setup_tx;
 592
 593         netif_carrier_off(netdev);
 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         netif_start_queue(netdev);
 606
 607         return 0;
 608
 609 err_up:
 610         ixgb_free_rx_resources(adapter);
 611 err_setup_rx:
 612         ixgb_free_tx_resources(adapter);
 613 err_setup_tx:
 614         ixgb_reset(adapter);
 615
 616         return err;
 617 }
 618
 619 /**
 620  * ixgb_close - Disables a network interface
 621  * @netdev: network interface device structure
 622  *
 623  * Returns 0, this is not allowed to fail
 624  *
 625  * The close entry point is called when an interface is de-activated
 626  * by the OS.  The hardware is still under the drivers control, but
 627  * needs to be disabled.  A global MAC reset is issued to stop the
 628  * hardware, and all transmit and receive resources are freed.
 629  **/
 630
 631 static int
 632 ixgb_close(struct net_device *netdev)
 633 {
 634         struct ixgb_adapter *adapter = netdev_priv(netdev);
 635
 636         ixgb_down(adapter, true);
 637
 638         ixgb_free_tx_resources(adapter);
 639         ixgb_free_rx_resources(adapter);
 640
 641         return 0;
 642 }
 643
 644 /**
 645  * ixgb_setup_tx_resources - allocate Tx resources (Descriptors)
 646  * @adapter: board private structure
 647  *
 648  * Return 0 on success, negative on failure
 649  **/
 650
 651 int
 652 ixgb_setup_tx_resources(struct ixgb_adapter *adapter)
 653 {
 654         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
 655         struct pci_dev *pdev = adapter->pdev;
 656         int size;
 657
 658         size = sizeof(struct ixgb_buffer) * txdr->count;
 659         txdr->buffer_info = vmalloc(size);
 660         if (!txdr->buffer_info) {
 661                 DPRINTK(PROBE, ERR,
 662                  "Unable to allocate transmit descriptor ring memory\n");
 663                 return -ENOMEM;
 664         }
 665         memset(txdr->buffer_info, 0, size);
 666
 667         /* round up to nearest 4K */
 668
 669         txdr->size = txdr->count * sizeof(struct ixgb_tx_desc);
 670         txdr->size = ALIGN(txdr->size, 4096);
 671
 672         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
 673         if (!txdr->desc) {
 674                 vfree(txdr->buffer_info);
 675                 DPRINTK(PROBE, ERR,
 676                  "Unable to allocate transmit descriptor memory\n");
 677                 return -ENOMEM;
 678         }
 679         memset(txdr->desc, 0, txdr->size);
 680
 681         txdr->next_to_use = 0;
 682         txdr->next_to_clean = 0;
 683
 684         return 0;
 685 }
 686
 687 /**
 688  * ixgb_configure_tx - Configure 82597 Transmit Unit after Reset.
 689  * @adapter: board private structure
 690  *
 691  * Configure the Tx unit of the MAC after a reset.
 692  **/
 693
 694 static void
 695 ixgb_configure_tx(struct ixgb_adapter *adapter)
 696 {
 697         u64 tdba = adapter->tx_ring.dma;
 698         u32 tdlen = adapter->tx_ring.count * sizeof(struct ixgb_tx_desc);
 699         u32 tctl;
 700         struct ixgb_hw *hw = &adapter->hw;
 701
 702         /* Setup the Base and Length of the Tx Descriptor Ring
 703          * tx_ring.dma can be either a 32 or 64 bit value
 704          */
 705
 706         IXGB_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
 707         IXGB_WRITE_REG(hw, TDBAH, (tdba >> 32));
 708
 709         IXGB_WRITE_REG(hw, TDLEN, tdlen);
 710
 711         /* Setup the HW Tx Head and Tail descriptor pointers */
 712
 713         IXGB_WRITE_REG(hw, TDH, 0);
 714         IXGB_WRITE_REG(hw, TDT, 0);
 715
 716         /* don't set up txdctl, it induces performance problems if configured
 717          * incorrectly */
 718         /* Set the Tx Interrupt Delay register */
 719
 720         IXGB_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
 721
 722         /* Program the Transmit Control Register */
 723
 724         tctl = IXGB_TCTL_TCE | IXGB_TCTL_TXEN | IXGB_TCTL_TPDE;
 725         IXGB_WRITE_REG(hw, TCTL, tctl);
 726
 727         /* Setup Transmit Descriptor Settings for this adapter */
 728         adapter->tx_cmd_type =
 729                 IXGB_TX_DESC_TYPE |
 730                 (adapter->tx_int_delay_enable ? IXGB_TX_DESC_CMD_IDE : 0);
 731 }
 732
 733 /**
 734  * ixgb_setup_rx_resources - allocate Rx resources (Descriptors)
 735  * @adapter: board private structure
 736  *
 737  * Returns 0 on success, negative on failure
 738  **/
 739
 740 int
 741 ixgb_setup_rx_resources(struct ixgb_adapter *adapter)
 742 {
 743         struct ixgb_desc_ring *rxdr = &adapter->rx_ring;
 744         struct pci_dev *pdev = adapter->pdev;
 745         int size;
 746
 747         size = sizeof(struct ixgb_buffer) * rxdr->count;
 748         rxdr->buffer_info = vmalloc(size);
 749         if (!rxdr->buffer_info) {
 750                 DPRINTK(PROBE, ERR,
 751                  "Unable to allocate receive descriptor ring\n");
 752                 return -ENOMEM;
 753         }
 754         memset(rxdr->buffer_info, 0, size);
 755
 756         /* Round up to nearest 4K */
 757
 758         rxdr->size = rxdr->count * sizeof(struct ixgb_rx_desc);
 759         rxdr->size = ALIGN(rxdr->size, 4096);
 760
 761         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
 762
 763         if (!rxdr->desc) {
 764                 vfree(rxdr->buffer_info);
 765                 DPRINTK(PROBE, ERR,
 766                  "Unable to allocate receive descriptors\n");
 767                 return -ENOMEM;
 768         }
 769         memset(rxdr->desc, 0, rxdr->size);
 770
 771         rxdr->next_to_clean = 0;
 772         rxdr->next_to_use = 0;
 773
 774         return 0;
 775 }
 776
 777 /**
 778  * ixgb_setup_rctl - configure the receive control register
 779  * @adapter: Board private structure
 780  **/
 781
 782 static void
 783 ixgb_setup_rctl(struct ixgb_adapter *adapter)
 784 {
 785         u32 rctl;
 786
 787         rctl = IXGB_READ_REG(&adapter->hw, RCTL);
 788
 789         rctl &= ~(3 << IXGB_RCTL_MO_SHIFT);
 790
 791         rctl |=
 792                 IXGB_RCTL_BAM | IXGB_RCTL_RDMTS_1_2 |
 793                 IXGB_RCTL_RXEN | IXGB_RCTL_CFF |
 794                 (adapter->hw.mc_filter_type << IXGB_RCTL_MO_SHIFT);
 795
 796         rctl |= IXGB_RCTL_SECRC;
 797
 798         if (adapter->rx_buffer_len <= IXGB_RXBUFFER_2048)
 799                 rctl |= IXGB_RCTL_BSIZE_2048;
 800         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_4096)
 801                 rctl |= IXGB_RCTL_BSIZE_4096;
 802         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_8192)
 803                 rctl |= IXGB_RCTL_BSIZE_8192;
 804         else if (adapter->rx_buffer_len <= IXGB_RXBUFFER_16384)
 805                 rctl |= IXGB_RCTL_BSIZE_16384;
 806
 807         IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
 808 }
 809
 810 /**
 811  * ixgb_configure_rx - Configure 82597 Receive Unit after Reset.
 812  * @adapter: board private structure
 813  *
 814  * Configure the Rx unit of the MAC after a reset.
 815  **/
 816
 817 static void
 818 ixgb_configure_rx(struct ixgb_adapter *adapter)
 819 {
 820         u64 rdba = adapter->rx_ring.dma;
 821         u32 rdlen = adapter->rx_ring.count * sizeof(struct ixgb_rx_desc);
 822         struct ixgb_hw *hw = &adapter->hw;
 823         u32 rctl;
 824         u32 rxcsum;
 825
 826         /* make sure receives are disabled while setting up the descriptors */
 827
 828         rctl = IXGB_READ_REG(hw, RCTL);
 829         IXGB_WRITE_REG(hw, RCTL, rctl & ~IXGB_RCTL_RXEN);
 830
 831         /* set the Receive Delay Timer Register */
 832
 833         IXGB_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
 834
 835         /* Setup the Base and Length of the Rx Descriptor Ring */
 836
 837         IXGB_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
 838         IXGB_WRITE_REG(hw, RDBAH, (rdba >> 32));
 839
 840         IXGB_WRITE_REG(hw, RDLEN, rdlen);
 841
 842         /* Setup the HW Rx Head and Tail Descriptor Pointers */
 843         IXGB_WRITE_REG(hw, RDH, 0);
 844         IXGB_WRITE_REG(hw, RDT, 0);
 845
 846         /* due to the hardware errata with RXDCTL, we are unable to use any of
 847          * the performance enhancing features of it without causing other
 848          * subtle bugs, some of the bugs could include receive length
 849          * corruption at high data rates (WTHRESH > 0) and/or receive
 850          * descriptor ring irregularites (particularly in hardware cache) */
 851         IXGB_WRITE_REG(hw, RXDCTL, 0);
 852
 853         /* Enable Receive Checksum Offload for TCP and UDP */
 854         if (adapter->rx_csum) {
 855                 rxcsum = IXGB_READ_REG(hw, RXCSUM);
 856                 rxcsum |= IXGB_RXCSUM_TUOFL;
 857                 IXGB_WRITE_REG(hw, RXCSUM, rxcsum);
 858         }
 859
 860         /* Enable Receives */
 861
 862         IXGB_WRITE_REG(hw, RCTL, rctl);
 863 }
 864
 865 /**
 866  * ixgb_free_tx_resources - Free Tx Resources
 867  * @adapter: board private structure
 868  *
 869  * Free all transmit software resources
 870  **/
 871
 872 void
 873 ixgb_free_tx_resources(struct ixgb_adapter *adapter)
 874 {
 875         struct pci_dev *pdev = adapter->pdev;
 876
 877         ixgb_clean_tx_ring(adapter);
 878
 879         vfree(adapter->tx_ring.buffer_info);
 880         adapter->tx_ring.buffer_info = NULL;
 881
 882         pci_free_consistent(pdev, adapter->tx_ring.size,
 883                             adapter->tx_ring.desc, adapter->tx_ring.dma);
 884
 885         adapter->tx_ring.desc = NULL;
 886 }
 887
 888 static void
 889 ixgb_unmap_and_free_tx_resource(struct ixgb_adapter *adapter,
 890                                 struct ixgb_buffer *buffer_info)
 891 {
 892         buffer_info->dma = 0;
 893         if (buffer_info->skb) {
 894                 skb_dma_unmap(&adapter->pdev->dev, buffer_info->skb,
 895                               DMA_TO_DEVICE);
 896                 dev_kfree_skb_any(buffer_info->skb);
 897                 buffer_info->skb = NULL;
 898         }
 899         buffer_info->time_stamp = 0;
 900         /* these fields must always be initialized in tx
 901          * buffer_info->length = 0;
 902          * buffer_info->next_to_watch = 0; */
 903 }
 904
 905 /**
 906  * ixgb_clean_tx_ring - Free Tx Buffers
 907  * @adapter: board private structure
 908  **/
 909
 910 static void
 911 ixgb_clean_tx_ring(struct ixgb_adapter *adapter)
 912 {
 913         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
 914         struct ixgb_buffer *buffer_info;
 915         unsigned long size;
 916         unsigned int i;
 917
 918         /* Free all the Tx ring sk_buffs */
 919
 920         for (i = 0; i < tx_ring->count; i++) {
 921                 buffer_info = &tx_ring->buffer_info[i];
 922                 ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
 923         }
 924
 925         size = sizeof(struct ixgb_buffer) * tx_ring->count;
 926         memset(tx_ring->buffer_info, 0, size);
 927
 928         /* Zero out the descriptor ring */
 929
 930         memset(tx_ring->desc, 0, tx_ring->size);
 931
 932         tx_ring->next_to_use = 0;
 933         tx_ring->next_to_clean = 0;
 934
 935         IXGB_WRITE_REG(&adapter->hw, TDH, 0);
 936         IXGB_WRITE_REG(&adapter->hw, TDT, 0);
 937 }
 938
 939 /**
 940  * ixgb_free_rx_resources - Free Rx Resources
 941  * @adapter: board private structure
 942  *
 943  * Free all receive software resources
 944  **/
 945
 946 void
 947 ixgb_free_rx_resources(struct ixgb_adapter *adapter)
 948 {
 949         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 950         struct pci_dev *pdev = adapter->pdev;
 951
 952         ixgb_clean_rx_ring(adapter);
 953
 954         vfree(rx_ring->buffer_info);
 955         rx_ring->buffer_info = NULL;
 956
 957         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
 958
 959         rx_ring->desc = NULL;
 960 }
 961
 962 /**
 963  * ixgb_clean_rx_ring - Free Rx Buffers
 964  * @adapter: board private structure
 965  **/
 966
 967 static void
 968 ixgb_clean_rx_ring(struct ixgb_adapter *adapter)
 969 {
 970         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
 971         struct ixgb_buffer *buffer_info;
 972         struct pci_dev *pdev = adapter->pdev;
 973         unsigned long size;
 974         unsigned int i;
 975
 976         /* Free all the Rx ring sk_buffs */
 977
 978         for (i = 0; i < rx_ring->count; i++) {
 979                 buffer_info = &rx_ring->buffer_info[i];
 980                 if (buffer_info->dma) {
 981                         pci_unmap_single(pdev,
 982                                          buffer_info->dma,
 983                                          buffer_info->length,
 984                                          PCI_DMA_FROMDEVICE);
 985                         buffer_info->dma = 0;
 986                         buffer_info->length = 0;
 987                 }
 988
 989                 if (buffer_info->skb) {
 990                         dev_kfree_skb(buffer_info->skb);
 991                         buffer_info->skb = NULL;
 992                 }
 993         }
 994
 995         size = sizeof(struct ixgb_buffer) * rx_ring->count;
 996         memset(rx_ring->buffer_info, 0, size);
 997
 998         /* Zero out the descriptor ring */
 999
1000         memset(rx_ring->desc, 0, rx_ring->size);
1001
1002         rx_ring->next_to_clean = 0;
1003         rx_ring->next_to_use = 0;
1004
1005         IXGB_WRITE_REG(&adapter->hw, RDH, 0);
1006         IXGB_WRITE_REG(&adapter->hw, RDT, 0);
1007 }
1008
1009 /**
1010  * ixgb_set_mac - Change the Ethernet Address of the NIC
1011  * @netdev: network interface device structure
1012  * @p: pointer to an address structure
1013  *
1014  * Returns 0 on success, negative on failure
1015  **/
1016
1017 static int
1018 ixgb_set_mac(struct net_device *netdev, void *p)
1019 {
1020         struct ixgb_adapter *adapter = netdev_priv(netdev);
1021         struct sockaddr *addr = p;
1022
1023         if (!is_valid_ether_addr(addr->sa_data))
1024                 return -EADDRNOTAVAIL;
1025
1026         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1027
1028         ixgb_rar_set(&adapter->hw, addr->sa_data, 0);
1029
1030         return 0;
1031 }
1032
1033 /**
1034  * ixgb_set_multi - Multicast and Promiscuous mode set
1035  * @netdev: network interface device structure
1036  *
1037  * The set_multi entry point is called whenever the multicast address
1038  * list or the network interface flags are updated.  This routine is
1039  * responsible for configuring the hardware for proper multicast,
1040  * promiscuous mode, and all-multi behavior.
1041  **/
1042
1043 static void
1044 ixgb_set_multi(struct net_device *netdev)
1045 {
1046         struct ixgb_adapter *adapter = netdev_priv(netdev);
1047         struct ixgb_hw *hw = &adapter->hw;
1048         struct dev_mc_list *mc_ptr;
1049         u32 rctl;
1050         int i;
1051
1052         /* Check for Promiscuous and All Multicast modes */
1053
1054         rctl = IXGB_READ_REG(hw, RCTL);
1055
1056         if (netdev->flags & IFF_PROMISC) {
1057                 rctl |= (IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1058                 rctl &= ~IXGB_RCTL_VFE;
1059         } else {
1060                 if (netdev->flags & IFF_ALLMULTI) {
1061                         rctl |= IXGB_RCTL_MPE;
1062                         rctl &= ~IXGB_RCTL_UPE;
1063                 } else {
1064                         rctl &= ~(IXGB_RCTL_UPE | IXGB_RCTL_MPE);
1065                 }
1066                 rctl |= IXGB_RCTL_VFE;
1067         }
1068
1069         if (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES) {
1070                 rctl |= IXGB_RCTL_MPE;
1071                 IXGB_WRITE_REG(hw, RCTL, rctl);
1072         } else {
1073                 u8 mta[IXGB_MAX_NUM_MULTICAST_ADDRESSES *
1074                             IXGB_ETH_LENGTH_OF_ADDRESS];
1075
1076                 IXGB_WRITE_REG(hw, RCTL, rctl);
1077
1078                 for (i = 0, mc_ptr = netdev->mc_list;
1079                      mc_ptr;
1080                      i++, mc_ptr = mc_ptr->next)
1081                         memcpy(&mta[i * IXGB_ETH_LENGTH_OF_ADDRESS],
1082                                mc_ptr->dmi_addr, IXGB_ETH_LENGTH_OF_ADDRESS);
1083
1084                 ixgb_mc_addr_list_update(hw, mta, netdev->mc_count, 0);
1085         }
1086 }
1087
1088 /**
1089  * ixgb_watchdog - Timer Call-back
1090  * @data: pointer to netdev cast into an unsigned long
1091  **/
1092
1093 static void
1094 ixgb_watchdog(unsigned long data)
1095 {
1096         struct ixgb_adapter *adapter = (struct ixgb_adapter *)data;
1097         struct net_device *netdev = adapter->netdev;
1098         struct ixgb_desc_ring *txdr = &adapter->tx_ring;
1099
1100         ixgb_check_for_link(&adapter->hw);
1101
1102         if (ixgb_check_for_bad_link(&adapter->hw)) {
1103                 /* force the reset path */
1104                 netif_stop_queue(netdev);
1105         }
1106
1107         if (adapter->hw.link_up) {
1108                 if (!netif_carrier_ok(netdev)) {
1109                         printk(KERN_INFO "ixgb: %s NIC Link is Up 10 Gbps "
1110                                "Full Duplex, Flow Control: %s\n",
1111                                netdev->name,
1112                                (adapter->hw.fc.type == ixgb_fc_full) ?
1113                                 "RX/TX" :
1114                                 ((adapter->hw.fc.type == ixgb_fc_rx_pause) ?
1115                                  "RX" :
1116                                  ((adapter->hw.fc.type == ixgb_fc_tx_pause) ?
1117                                   "TX" : "None")));
1118                         adapter->link_speed = 10000;
1119                         adapter->link_duplex = FULL_DUPLEX;
1120                         netif_carrier_on(netdev);
1121                         netif_wake_queue(netdev);
1122                 }
1123         } else {
1124                 if (netif_carrier_ok(netdev)) {
1125                         adapter->link_speed = 0;
1126                         adapter->link_duplex = 0;
1127                         printk(KERN_INFO "ixgb: %s NIC Link is Down\n",
1128                                netdev->name);
1129                         netif_carrier_off(netdev);
1130                         netif_stop_queue(netdev);
1131
1132                 }
1133         }
1134
1135         ixgb_update_stats(adapter);
1136
1137         if (!netif_carrier_ok(netdev)) {
1138                 if (IXGB_DESC_UNUSED(txdr) + 1 < txdr->count) {
1139                         /* We've lost link, so the controller stops DMA,
1140                          * but we've got queued Tx work that's never going
1141                          * to get done, so reset controller to flush Tx.
1142                          * (Do the reset outside of interrupt context). */
1143                         schedule_work(&adapter->tx_timeout_task);
1144                 }
1145         }
1146
1147         /* Force detection of hung controller every watchdog period */
1148         adapter->detect_tx_hung = true;
1149
1150         /* generate an interrupt to force clean up of any stragglers */
1151         IXGB_WRITE_REG(&adapter->hw, ICS, IXGB_INT_TXDW);
1152
1153         /* Reset the timer */
1154         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
1155 }
1156
1157 #define IXGB_TX_FLAGS_CSUM              0x00000001
1158 #define IXGB_TX_FLAGS_VLAN              0x00000002
1159 #define IXGB_TX_FLAGS_TSO               0x00000004
1160
1161 static int
1162 ixgb_tso(struct ixgb_adapter *adapter, struct sk_buff *skb)
1163 {
1164         struct ixgb_context_desc *context_desc;
1165         unsigned int i;
1166         u8 ipcss, ipcso, tucss, tucso, hdr_len;
1167         u16 ipcse, tucse, mss;
1168         int err;
1169
1170         if (likely(skb_is_gso(skb))) {
1171                 struct ixgb_buffer *buffer_info;
1172                 struct iphdr *iph;
1173
1174                 if (skb_header_cloned(skb)) {
1175                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
1176                         if (err)
1177                                 return err;
1178                 }
1179
1180                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
1181                 mss = skb_shinfo(skb)->gso_size;
1182                 iph = ip_hdr(skb);
1183                 iph->tot_len = 0;
1184                 iph->check = 0;
1185                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
1186                                                          iph->daddr, 0,
1187                                                          IPPROTO_TCP, 0);
1188                 ipcss = skb_network_offset(skb);
1189                 ipcso = (void *)&(iph->check) - (void *)skb->data;
1190                 ipcse = skb_transport_offset(skb) - 1;
1191                 tucss = skb_transport_offset(skb);
1192                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
1193                 tucse = 0;
1194
1195                 i = adapter->tx_ring.next_to_use;
1196                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1197                 buffer_info = &adapter->tx_ring.buffer_info[i];
1198                 WARN_ON(buffer_info->dma != 0);
1199
1200                 context_desc->ipcss = ipcss;
1201                 context_desc->ipcso = ipcso;
1202                 context_desc->ipcse = cpu_to_le16(ipcse);
1203                 context_desc->tucss = tucss;
1204                 context_desc->tucso = tucso;
1205                 context_desc->tucse = cpu_to_le16(tucse);
1206                 context_desc->mss = cpu_to_le16(mss);
1207                 context_desc->hdr_len = hdr_len;
1208                 context_desc->status = 0;
1209                 context_desc->cmd_type_len = cpu_to_le32(
1210                                                   IXGB_CONTEXT_DESC_TYPE
1211                                                 | IXGB_CONTEXT_DESC_CMD_TSE
1212                                                 | IXGB_CONTEXT_DESC_CMD_IP
1213                                                 | IXGB_CONTEXT_DESC_CMD_TCP
1214                                                 | IXGB_CONTEXT_DESC_CMD_IDE
1215                                                 | (skb->len - (hdr_len)));
1216
1217
1218                 if (++i == adapter->tx_ring.count) i = 0;
1219                 adapter->tx_ring.next_to_use = i;
1220
1221                 return 1;
1222         }
1223
1224         return 0;
1225 }
1226
1227 static bool
1228 ixgb_tx_csum(struct ixgb_adapter *adapter, struct sk_buff *skb)
1229 {
1230         struct ixgb_context_desc *context_desc;
1231         unsigned int i;
1232         u8 css, cso;
1233
1234         if (likely(skb->ip_summed == CHECKSUM_PARTIAL)) {
1235                 struct ixgb_buffer *buffer_info;
1236                 css = skb_transport_offset(skb);
1237                 cso = css + skb->csum_offset;
1238
1239                 i = adapter->tx_ring.next_to_use;
1240                 context_desc = IXGB_CONTEXT_DESC(adapter->tx_ring, i);
1241                 buffer_info = &adapter->tx_ring.buffer_info[i];
1242                 WARN_ON(buffer_info->dma != 0);
1243
1244                 context_desc->tucss = css;
1245                 context_desc->tucso = cso;
1246                 context_desc->tucse = 0;
1247                 /* zero out any previously existing data in one instruction */
1248                 *(u32 *)&(context_desc->ipcss) = 0;
1249                 context_desc->status = 0;
1250                 context_desc->hdr_len = 0;
1251                 context_desc->mss = 0;
1252                 context_desc->cmd_type_len =
1253                         cpu_to_le32(IXGB_CONTEXT_DESC_TYPE
1254                                     | IXGB_TX_DESC_CMD_IDE);
1255
1256                 if (++i == adapter->tx_ring.count) i = 0;
1257                 adapter->tx_ring.next_to_use = i;
1258
1259                 return true;
1260         }
1261
1262         return false;
1263 }
1264
1265 #define IXGB_MAX_TXD_PWR        14
1266 #define IXGB_MAX_DATA_PER_TXD   (1<<IXGB_MAX_TXD_PWR)
1267
1268 static int
1269 ixgb_tx_map(struct ixgb_adapter *adapter, struct sk_buff *skb,
1270             unsigned int first)
1271 {
1272         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1273         struct ixgb_buffer *buffer_info;
1274         int len = skb_headlen(skb);
1275         unsigned int offset = 0, size, count = 0, i;
1276         unsigned int mss = skb_shinfo(skb)->gso_size;
1277
1278         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
1279         unsigned int f;
1280         dma_addr_t *map;
1281
1282         i = tx_ring->next_to_use;
1283
1284         if (skb_dma_map(&adapter->pdev->dev, skb, DMA_TO_DEVICE)) {
1285                 dev_err(&adapter->pdev->dev, "TX DMA map failed\n");
1286                 return 0;
1287         }
1288
1289         map = skb_shinfo(skb)->dma_maps;
1290
1291         while (len) {
1292                 buffer_info = &tx_ring->buffer_info[i];
1293                 size = min(len, IXGB_MAX_DATA_PER_TXD);
1294                 /* Workaround for premature desc write-backs
1295                  * in TSO mode.  Append 4-byte sentinel desc */
1296                 if (unlikely(mss && !nr_frags && size == len && size > 8))
1297                         size -= 4;
1298
1299                 buffer_info->length = size;
1300                 WARN_ON(buffer_info->dma != 0);
1301                 buffer_info->time_stamp = jiffies;
1302                 buffer_info->dma = map[0] + offset;
1303                         pci_map_single(adapter->pdev,
1304                                 skb->data + offset,
1305                                 size,
1306                                 PCI_DMA_TODEVICE);
1307                 buffer_info->next_to_watch = 0;
1308
1309                 len -= size;
1310                 offset += size;
1311                 count++;
1312                 if (len) {
1313                         i++;
1314                         if (i == tx_ring->count)
1315                                 i = 0;
1316                 }
1317         }
1318
1319         for (f = 0; f < nr_frags; f++) {
1320                 struct skb_frag_struct *frag;
1321
1322                 frag = &skb_shinfo(skb)->frags[f];
1323                 len = frag->size;
1324                 offset = 0;
1325
1326                 while (len) {
1327                         i++;
1328                         if (i == tx_ring->count)
1329                                 i = 0;
1330
1331                         buffer_info = &tx_ring->buffer_info[i];
1332                         size = min(len, IXGB_MAX_DATA_PER_TXD);
1333
1334                         /* Workaround for premature desc write-backs
1335                          * in TSO mode.  Append 4-byte sentinel desc */
1336                         if (unlikely(mss && (f == (nr_frags - 1))
1337                                      && size == len && size > 8))
1338                                 size -= 4;
1339
1340                         buffer_info->length = size;
1341                         buffer_info->time_stamp = jiffies;
1342                         buffer_info->dma = map[f + 1] + offset;
1343                         buffer_info->next_to_watch = 0;
1344
1345                         len -= size;
1346                         offset += size;
1347                         count++;
1348                 }
1349         }
1350         tx_ring->buffer_info[i].skb = skb;
1351         tx_ring->buffer_info[first].next_to_watch = i;
1352
1353         return count;
1354 }
1355
1356 static void
1357 ixgb_tx_queue(struct ixgb_adapter *adapter, int count, int vlan_id,int tx_flags)
1358 {
1359         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1360         struct ixgb_tx_desc *tx_desc = NULL;
1361         struct ixgb_buffer *buffer_info;
1362         u32 cmd_type_len = adapter->tx_cmd_type;
1363         u8 status = 0;
1364         u8 popts = 0;
1365         unsigned int i;
1366
1367         if (tx_flags & IXGB_TX_FLAGS_TSO) {
1368                 cmd_type_len |= IXGB_TX_DESC_CMD_TSE;
1369                 popts |= (IXGB_TX_DESC_POPTS_IXSM | IXGB_TX_DESC_POPTS_TXSM);
1370         }
1371
1372         if (tx_flags & IXGB_TX_FLAGS_CSUM)
1373                 popts |= IXGB_TX_DESC_POPTS_TXSM;
1374
1375         if (tx_flags & IXGB_TX_FLAGS_VLAN)
1376                 cmd_type_len |= IXGB_TX_DESC_CMD_VLE;
1377
1378         i = tx_ring->next_to_use;
1379
1380         while (count--) {
1381                 buffer_info = &tx_ring->buffer_info[i];
1382                 tx_desc = IXGB_TX_DESC(*tx_ring, i);
1383                 tx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
1384                 tx_desc->cmd_type_len =
1385                         cpu_to_le32(cmd_type_len | buffer_info->length);
1386                 tx_desc->status = status;
1387                 tx_desc->popts = popts;
1388                 tx_desc->vlan = cpu_to_le16(vlan_id);
1389
1390                 if (++i == tx_ring->count) i = 0;
1391         }
1392
1393         tx_desc->cmd_type_len |=
1394                 cpu_to_le32(IXGB_TX_DESC_CMD_EOP | IXGB_TX_DESC_CMD_RS);
1395
1396         /* Force memory writes to complete before letting h/w
1397          * know there are new descriptors to fetch.  (Only
1398          * applicable for weak-ordered memory model archs,
1399          * such as IA-64). */
1400         wmb();
1401
1402         tx_ring->next_to_use = i;
1403         IXGB_WRITE_REG(&adapter->hw, TDT, i);
1404 }
1405
1406 static int __ixgb_maybe_stop_tx(struct net_device *netdev, int size)
1407 {
1408         struct ixgb_adapter *adapter = netdev_priv(netdev);
1409         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1410
1411         netif_stop_queue(netdev);
1412         /* Herbert's original patch had:
1413          *  smp_mb__after_netif_stop_queue();
1414          * but since that doesn't exist yet, just open code it. */
1415         smp_mb();
1416
1417         /* We need to check again in a case another CPU has just
1418          * made room available. */
1419         if (likely(IXGB_DESC_UNUSED(tx_ring) < size))
1420                 return -EBUSY;
1421
1422         /* A reprieve! */
1423         netif_start_queue(netdev);
1424         ++adapter->restart_queue;
1425         return 0;
1426 }
1427
1428 static int ixgb_maybe_stop_tx(struct net_device *netdev,
1429                               struct ixgb_desc_ring *tx_ring, int size)
1430 {
1431         if (likely(IXGB_DESC_UNUSED(tx_ring) >= size))
1432                 return 0;
1433         return __ixgb_maybe_stop_tx(netdev, size);
1434 }
1435
1436
1437 /* Tx Descriptors needed, worst case */
1438 #define TXD_USE_COUNT(S) (((S) >> IXGB_MAX_TXD_PWR) + \
1439                          (((S) & (IXGB_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
1440 #define DESC_NEEDED TXD_USE_COUNT(IXGB_MAX_DATA_PER_TXD) /* skb->date */ + \
1441         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1 /* for context */ \
1442         + 1 /* one more needed for sentinel TSO workaround */
1443
1444 static int
1445 ixgb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1446 {
1447         struct ixgb_adapter *adapter = netdev_priv(netdev);
1448         unsigned int first;
1449         unsigned int tx_flags = 0;
1450         int vlan_id = 0;
1451         int count = 0;
1452         int tso;
1453
1454         if (test_bit(__IXGB_DOWN, &adapter->flags)) {
1455                 dev_kfree_skb(skb);
1456                 return NETDEV_TX_OK;
1457         }
1458
1459         if (skb->len <= 0) {
1460                 dev_kfree_skb(skb);
1461                 return 0;
1462         }
1463
1464         if (unlikely(ixgb_maybe_stop_tx(netdev, &adapter->tx_ring,
1465                      DESC_NEEDED)))
1466                 return NETDEV_TX_BUSY;
1467
1468         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
1469                 tx_flags |= IXGB_TX_FLAGS_VLAN;
1470                 vlan_id = vlan_tx_tag_get(skb);
1471         }
1472
1473         first = adapter->tx_ring.next_to_use;
1474
1475         tso = ixgb_tso(adapter, skb);
1476         if (tso < 0) {
1477                 dev_kfree_skb(skb);
1478                 return NETDEV_TX_OK;
1479         }
1480
1481         if (likely(tso))
1482                 tx_flags |= IXGB_TX_FLAGS_TSO;
1483         else if (ixgb_tx_csum(adapter, skb))
1484                 tx_flags |= IXGB_TX_FLAGS_CSUM;
1485
1486         count = ixgb_tx_map(adapter, skb, first);
1487
1488         if (count) {
1489                 ixgb_tx_queue(adapter, count, vlan_id, tx_flags);
1490                 netdev->trans_start = jiffies;
1491                 /* Make sure there is space in the ring for the next send. */
1492                 ixgb_maybe_stop_tx(netdev, &adapter->tx_ring, DESC_NEEDED);
1493
1494         } else {
1495                 dev_kfree_skb_any(skb);
1496                 adapter->tx_ring.buffer_info[first].time_stamp = 0;
1497                 adapter->tx_ring.next_to_use = first;
1498         }
1499
1500         return NETDEV_TX_OK;
1501 }
1502
1503 /**
1504  * ixgb_tx_timeout - Respond to a Tx Hang
1505  * @netdev: network interface device structure
1506  **/
1507
1508 static void
1509 ixgb_tx_timeout(struct net_device *netdev)
1510 {
1511         struct ixgb_adapter *adapter = netdev_priv(netdev);
1512
1513         /* Do the reset outside of interrupt context */
1514         schedule_work(&adapter->tx_timeout_task);
1515 }
1516
1517 static void
1518 ixgb_tx_timeout_task(struct work_struct *work)
1519 {
1520         struct ixgb_adapter *adapter =
1521                 container_of(work, struct ixgb_adapter, tx_timeout_task);
1522
1523         adapter->tx_timeout_count++;
1524         ixgb_down(adapter, true);
1525         ixgb_up(adapter);
1526 }
1527
1528 /**
1529  * ixgb_get_stats - Get System Network Statistics
1530  * @netdev: network interface device structure
1531  *
1532  * Returns the address of the device statistics structure.
1533  * The statistics are actually updated from the timer callback.
1534  **/
1535
1536 static struct net_device_stats *
1537 ixgb_get_stats(struct net_device *netdev)
1538 {
1539         struct ixgb_adapter *adapter = netdev_priv(netdev);
1540
1541         return &adapter->net_stats;
1542 }
1543
1544 /**
1545  * ixgb_change_mtu - Change the Maximum Transfer Unit
1546  * @netdev: network interface device structure
1547  * @new_mtu: new value for maximum frame size
1548  *
1549  * Returns 0 on success, negative on failure
1550  **/
1551
1552 static int
1553 ixgb_change_mtu(struct net_device *netdev, int new_mtu)
1554 {
1555         struct ixgb_adapter *adapter = netdev_priv(netdev);
1556         int max_frame = new_mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1557         int old_max_frame = netdev->mtu + ENET_HEADER_SIZE + ENET_FCS_LENGTH;
1558
1559         /* MTU < 68 is an error for IPv4 traffic, just don't allow it */
1560         if ((new_mtu < 68) ||
1561             (max_frame > IXGB_MAX_JUMBO_FRAME_SIZE + ENET_FCS_LENGTH)) {
1562                 DPRINTK(PROBE, ERR, "Invalid MTU setting %d\n", new_mtu);
1563                 return -EINVAL;
1564         }
1565
1566         if (old_max_frame == max_frame)
1567                 return 0;
1568
1569         if (netif_running(netdev))
1570                 ixgb_down(adapter, true);
1571
1572         adapter->rx_buffer_len = max_frame + 8; /* + 8 for errata */
1573
1574         netdev->mtu = new_mtu;
1575
1576         if (netif_running(netdev))
1577                 ixgb_up(adapter);
1578
1579         return 0;
1580 }
1581
1582 /**
1583  * ixgb_update_stats - Update the board statistics counters.
1584  * @adapter: board private structure
1585  **/
1586
1587 void
1588 ixgb_update_stats(struct ixgb_adapter *adapter)
1589 {
1590         struct net_device *netdev = adapter->netdev;
1591         struct pci_dev *pdev = adapter->pdev;
1592
1593         /* Prevent stats update while adapter is being reset */
1594         if (pci_channel_offline(pdev))
1595                 return;
1596
1597         if ((netdev->flags & IFF_PROMISC) || (netdev->flags & IFF_ALLMULTI) ||
1598            (netdev->mc_count > IXGB_MAX_NUM_MULTICAST_ADDRESSES)) {
1599                 u64 multi = IXGB_READ_REG(&adapter->hw, MPRCL);
1600                 u32 bcast_l = IXGB_READ_REG(&adapter->hw, BPRCL);
1601                 u32 bcast_h = IXGB_READ_REG(&adapter->hw, BPRCH);
1602                 u64 bcast = ((u64)bcast_h << 32) | bcast_l;
1603
1604                 multi |= ((u64)IXGB_READ_REG(&adapter->hw, MPRCH) << 32);
1605                 /* fix up multicast stats by removing broadcasts */
1606                 if (multi >= bcast)
1607                         multi -= bcast;
1608
1609                 adapter->stats.mprcl += (multi & 0xFFFFFFFF);
1610                 adapter->stats.mprch += (multi >> 32);
1611                 adapter->stats.bprcl += bcast_l;
1612                 adapter->stats.bprch += bcast_h;
1613         } else {
1614                 adapter->stats.mprcl += IXGB_READ_REG(&adapter->hw, MPRCL);
1615                 adapter->stats.mprch += IXGB_READ_REG(&adapter->hw, MPRCH);
1616                 adapter->stats.bprcl += IXGB_READ_REG(&adapter->hw, BPRCL);
1617                 adapter->stats.bprch += IXGB_READ_REG(&adapter->hw, BPRCH);
1618         }
1619         adapter->stats.tprl += IXGB_READ_REG(&adapter->hw, TPRL);
1620         adapter->stats.tprh += IXGB_READ_REG(&adapter->hw, TPRH);
1621         adapter->stats.gprcl += IXGB_READ_REG(&adapter->hw, GPRCL);
1622         adapter->stats.gprch += IXGB_READ_REG(&adapter->hw, GPRCH);
1623         adapter->stats.uprcl += IXGB_READ_REG(&adapter->hw, UPRCL);
1624         adapter->stats.uprch += IXGB_READ_REG(&adapter->hw, UPRCH);
1625         adapter->stats.vprcl += IXGB_READ_REG(&adapter->hw, VPRCL);
1626         adapter->stats.vprch += IXGB_READ_REG(&adapter->hw, VPRCH);
1627         adapter->stats.jprcl += IXGB_READ_REG(&adapter->hw, JPRCL);
1628         adapter->stats.jprch += IXGB_READ_REG(&adapter->hw, JPRCH);
1629         adapter->stats.gorcl += IXGB_READ_REG(&adapter->hw, GORCL);
1630         adapter->stats.gorch += IXGB_READ_REG(&adapter->hw, GORCH);
1631         adapter->stats.torl += IXGB_READ_REG(&adapter->hw, TORL);
1632         adapter->stats.torh += IXGB_READ_REG(&adapter->hw, TORH);
1633         adapter->stats.rnbc += IXGB_READ_REG(&adapter->hw, RNBC);
1634         adapter->stats.ruc += IXGB_READ_REG(&adapter->hw, RUC);
1635         adapter->stats.roc += IXGB_READ_REG(&adapter->hw, ROC);
1636         adapter->stats.rlec += IXGB_READ_REG(&adapter->hw, RLEC);
1637         adapter->stats.crcerrs += IXGB_READ_REG(&adapter->hw, CRCERRS);
1638         adapter->stats.icbc += IXGB_READ_REG(&adapter->hw, ICBC);
1639         adapter->stats.ecbc += IXGB_READ_REG(&adapter->hw, ECBC);
1640         adapter->stats.mpc += IXGB_READ_REG(&adapter->hw, MPC);
1641         adapter->stats.tptl += IXGB_READ_REG(&adapter->hw, TPTL);
1642         adapter->stats.tpth += IXGB_READ_REG(&adapter->hw, TPTH);
1643         adapter->stats.gptcl += IXGB_READ_REG(&adapter->hw, GPTCL);
1644         adapter->stats.gptch += IXGB_READ_REG(&adapter->hw, GPTCH);
1645         adapter->stats.bptcl += IXGB_READ_REG(&adapter->hw, BPTCL);
1646         adapter->stats.bptch += IXGB_READ_REG(&adapter->hw, BPTCH);
1647         adapter->stats.mptcl += IXGB_READ_REG(&adapter->hw, MPTCL);
1648         adapter->stats.mptch += IXGB_READ_REG(&adapter->hw, MPTCH);
1649         adapter->stats.uptcl += IXGB_READ_REG(&adapter->hw, UPTCL);
1650         adapter->stats.uptch += IXGB_READ_REG(&adapter->hw, UPTCH);
1651         adapter->stats.vptcl += IXGB_READ_REG(&adapter->hw, VPTCL);
1652         adapter->stats.vptch += IXGB_READ_REG(&adapter->hw, VPTCH);
1653         adapter->stats.jptcl += IXGB_READ_REG(&adapter->hw, JPTCL);
1654         adapter->stats.jptch += IXGB_READ_REG(&adapter->hw, JPTCH);
1655         adapter->stats.gotcl += IXGB_READ_REG(&adapter->hw, GOTCL);
1656         adapter->stats.gotch += IXGB_READ_REG(&adapter->hw, GOTCH);
1657         adapter->stats.totl += IXGB_READ_REG(&adapter->hw, TOTL);
1658         adapter->stats.toth += IXGB_READ_REG(&adapter->hw, TOTH);
1659         adapter->stats.dc += IXGB_READ_REG(&adapter->hw, DC);
1660         adapter->stats.plt64c += IXGB_READ_REG(&adapter->hw, PLT64C);
1661         adapter->stats.tsctc += IXGB_READ_REG(&adapter->hw, TSCTC);
1662         adapter->stats.tsctfc += IXGB_READ_REG(&adapter->hw, TSCTFC);
1663         adapter->stats.ibic += IXGB_READ_REG(&adapter->hw, IBIC);
1664         adapter->stats.rfc += IXGB_READ_REG(&adapter->hw, RFC);
1665         adapter->stats.lfc += IXGB_READ_REG(&adapter->hw, LFC);
1666         adapter->stats.pfrc += IXGB_READ_REG(&adapter->hw, PFRC);
1667         adapter->stats.pftc += IXGB_READ_REG(&adapter->hw, PFTC);
1668         adapter->stats.mcfrc += IXGB_READ_REG(&adapter->hw, MCFRC);
1669         adapter->stats.mcftc += IXGB_READ_REG(&adapter->hw, MCFTC);
1670         adapter->stats.xonrxc += IXGB_READ_REG(&adapter->hw, XONRXC);
1671         adapter->stats.xontxc += IXGB_READ_REG(&adapter->hw, XONTXC);
1672         adapter->stats.xoffrxc += IXGB_READ_REG(&adapter->hw, XOFFRXC);
1673         adapter->stats.xofftxc += IXGB_READ_REG(&adapter->hw, XOFFTXC);
1674         adapter->stats.rjc += IXGB_READ_REG(&adapter->hw, RJC);
1675
1676         /* Fill out the OS statistics structure */
1677
1678         adapter->net_stats.rx_packets = adapter->stats.gprcl;
1679         adapter->net_stats.tx_packets = adapter->stats.gptcl;
1680         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
1681         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
1682         adapter->net_stats.multicast = adapter->stats.mprcl;
1683         adapter->net_stats.collisions = 0;
1684
1685         /* ignore RLEC as it reports errors for padded (<64bytes) frames
1686          * with a length in the type/len field */
1687         adapter->net_stats.rx_errors =
1688             /* adapter->stats.rnbc + */ adapter->stats.crcerrs +
1689             adapter->stats.ruc +
1690             adapter->stats.roc /*+ adapter->stats.rlec */  +
1691             adapter->stats.icbc +
1692             adapter->stats.ecbc + adapter->stats.mpc;
1693
1694         /* see above
1695          * adapter->net_stats.rx_length_errors = adapter->stats.rlec;
1696          */
1697
1698         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
1699         adapter->net_stats.rx_fifo_errors = adapter->stats.mpc;
1700         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
1701         adapter->net_stats.rx_over_errors = adapter->stats.mpc;
1702
1703         adapter->net_stats.tx_errors = 0;
1704         adapter->net_stats.rx_frame_errors = 0;
1705         adapter->net_stats.tx_aborted_errors = 0;
1706         adapter->net_stats.tx_carrier_errors = 0;
1707         adapter->net_stats.tx_fifo_errors = 0;
1708         adapter->net_stats.tx_heartbeat_errors = 0;
1709         adapter->net_stats.tx_window_errors = 0;
1710 }
1711
1712 #define IXGB_MAX_INTR 10
1713 /**
1714  * ixgb_intr - Interrupt Handler
1715  * @irq: interrupt number
1716  * @data: pointer to a network interface device structure
1717  **/
1718
1719 static irqreturn_t
1720 ixgb_intr(int irq, void *data)
1721 {
1722         struct net_device *netdev = data;
1723         struct ixgb_adapter *adapter = netdev_priv(netdev);
1724         struct ixgb_hw *hw = &adapter->hw;
1725         u32 icr = IXGB_READ_REG(hw, ICR);
1726
1727         if (unlikely(!icr))
1728                 return IRQ_NONE;  /* Not our interrupt */
1729
1730         if (unlikely(icr & (IXGB_INT_RXSEQ | IXGB_INT_LSC)))
1731                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1732                         mod_timer(&adapter->watchdog_timer, jiffies);
1733
1734         if (napi_schedule_prep(&adapter->napi)) {
1735
1736                 /* Disable interrupts and register for poll. The flush
1737                   of the posted write is intentionally left out.
1738                 */
1739
1740                 IXGB_WRITE_REG(&adapter->hw, IMC, ~0);
1741                 __napi_schedule(&adapter->napi);
1742         }
1743         return IRQ_HANDLED;
1744 }
1745
1746 /**
1747  * ixgb_clean - NAPI Rx polling callback
1748  * @adapter: board private structure
1749  **/
1750
1751 static int
1752 ixgb_clean(struct napi_struct *napi, int budget)
1753 {
1754         struct ixgb_adapter *adapter = container_of(napi, struct ixgb_adapter, napi);
1755         int work_done = 0;
1756
1757         ixgb_clean_tx_irq(adapter);
1758         ixgb_clean_rx_irq(adapter, &work_done, budget);
1759
1760         /* If budget not fully consumed, exit the polling mode */
1761         if (work_done < budget) {
1762                 napi_complete(napi);
1763                 if (!test_bit(__IXGB_DOWN, &adapter->flags))
1764                         ixgb_irq_enable(adapter);
1765         }
1766
1767         return work_done;
1768 }
1769
1770 /**
1771  * ixgb_clean_tx_irq - Reclaim resources after transmit completes
1772  * @adapter: board private structure
1773  **/
1774
1775 static bool
1776 ixgb_clean_tx_irq(struct ixgb_adapter *adapter)
1777 {
1778         struct ixgb_desc_ring *tx_ring = &adapter->tx_ring;
1779         struct net_device *netdev = adapter->netdev;
1780         struct ixgb_tx_desc *tx_desc, *eop_desc;
1781         struct ixgb_buffer *buffer_info;
1782         unsigned int i, eop;
1783         bool cleaned = false;
1784
1785         i = tx_ring->next_to_clean;
1786         eop = tx_ring->buffer_info[i].next_to_watch;
1787         eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1788
1789         while (eop_desc->status & IXGB_TX_DESC_STATUS_DD) {
1790
1791                 for (cleaned = false; !cleaned; ) {
1792                         tx_desc = IXGB_TX_DESC(*tx_ring, i);
1793                         buffer_info = &tx_ring->buffer_info[i];
1794
1795                         if (tx_desc->popts &
1796                            (IXGB_TX_DESC_POPTS_TXSM |
1797                             IXGB_TX_DESC_POPTS_IXSM))
1798                                 adapter->hw_csum_tx_good++;
1799
1800                         ixgb_unmap_and_free_tx_resource(adapter, buffer_info);
1801
1802                         *(u32 *)&(tx_desc->status) = 0;
1803
1804                         cleaned = (i == eop);
1805                         if (++i == tx_ring->count) i = 0;
1806                 }
1807
1808                 eop = tx_ring->buffer_info[i].next_to_watch;
1809                 eop_desc = IXGB_TX_DESC(*tx_ring, eop);
1810         }
1811
1812         tx_ring->next_to_clean = i;
1813
1814         if (unlikely(cleaned && netif_carrier_ok(netdev) &&
1815                      IXGB_DESC_UNUSED(tx_ring) >= DESC_NEEDED)) {
1816                 /* Make sure that anybody stopping the queue after this
1817                  * sees the new next_to_clean. */
1818                 smp_mb();
1819
1820                 if (netif_queue_stopped(netdev) &&
1821                     !(test_bit(__IXGB_DOWN, &adapter->flags))) {
1822                         netif_wake_queue(netdev);
1823                         ++adapter->restart_queue;
1824                 }
1825         }
1826
1827         if (adapter->detect_tx_hung) {
1828                 /* detect a transmit hang in hardware, this serializes the
1829                  * check with the clearing of time_stamp and movement of i */
1830                 adapter->detect_tx_hung = false;
1831                 if (tx_ring->buffer_info[eop].time_stamp &&
1832                    time_after(jiffies, tx_ring->buffer_info[eop].time_stamp + HZ)
1833                    && !(IXGB_READ_REG(&adapter->hw, STATUS) &
1834                         IXGB_STATUS_TXOFF)) {
1835                         /* detected Tx unit hang */
1836                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
1837                                         "  TDH                  <%x>\n"
1838                                         "  TDT                  <%x>\n"
1839                                         "  next_to_use          <%x>\n"
1840                                         "  next_to_clean        <%x>\n"
1841                                         "buffer_info[next_to_clean]\n"
1842                                         "  time_stamp           <%lx>\n"
1843                                         "  next_to_watch        <%x>\n"
1844                                         "  jiffies              <%lx>\n"
1845                                         "  next_to_watch.status <%x>\n",
1846                                 IXGB_READ_REG(&adapter->hw, TDH),
1847                                 IXGB_READ_REG(&adapter->hw, TDT),
1848                                 tx_ring->next_to_use,
1849                                 tx_ring->next_to_clean,
1850                                 tx_ring->buffer_info[eop].time_stamp,
1851                                 eop,
1852                                 jiffies,
1853                                 eop_desc->status);
1854                         netif_stop_queue(netdev);
1855                 }
1856         }
1857
1858         return cleaned;
1859 }
1860
1861 /**
1862  * ixgb_rx_checksum - Receive Checksum Offload for 82597.
1863  * @adapter: board private structure
1864  * @rx_desc: receive descriptor
1865  * @sk_buff: socket buffer with received data
1866  **/
1867
1868 static void
1869 ixgb_rx_checksum(struct ixgb_adapter *adapter,
1870                  struct ixgb_rx_desc *rx_desc,
1871                  struct sk_buff *skb)
1872 {
1873         /* Ignore Checksum bit is set OR
1874          * TCP Checksum has not been calculated
1875          */
1876         if ((rx_desc->status & IXGB_RX_DESC_STATUS_IXSM) ||
1877            (!(rx_desc->status & IXGB_RX_DESC_STATUS_TCPCS))) {
1878                 skb->ip_summed = CHECKSUM_NONE;
1879                 return;
1880         }
1881
1882         /* At this point we know the hardware did the TCP checksum */
1883         /* now look at the TCP checksum error bit */
1884         if (rx_desc->errors & IXGB_RX_DESC_ERRORS_TCPE) {
1885                 /* let the stack verify checksum errors */
1886                 skb->ip_summed = CHECKSUM_NONE;
1887                 adapter->hw_csum_rx_error++;
1888         } else {
1889                 /* TCP checksum is good */
1890                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1891                 adapter->hw_csum_rx_good++;
1892         }
1893 }
1894
1895 /**
1896  * ixgb_clean_rx_irq - Send received data up the network stack,
1897  * @adapter: board private structure
1898  **/
1899
1900 static bool
1901 ixgb_clean_rx_irq(struct ixgb_adapter *adapter, int *work_done, int work_to_do)
1902 {
1903         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
1904         struct net_device *netdev = adapter->netdev;
1905         struct pci_dev *pdev = adapter->pdev;
1906         struct ixgb_rx_desc *rx_desc, *next_rxd;
1907         struct ixgb_buffer *buffer_info, *next_buffer, *next2_buffer;
1908         u32 length;
1909         unsigned int i, j;
1910         int cleaned_count = 0;
1911         bool cleaned = false;
1912
1913         i = rx_ring->next_to_clean;
1914         rx_desc = IXGB_RX_DESC(*rx_ring, i);
1915         buffer_info = &rx_ring->buffer_info[i];
1916
1917         while (rx_desc->status & IXGB_RX_DESC_STATUS_DD) {
1918                 struct sk_buff *skb;
1919                 u8 status;
1920
1921                 if (*work_done >= work_to_do)
1922                         break;
1923
1924                 (*work_done)++;
1925                 status = rx_desc->status;
1926                 skb = buffer_info->skb;
1927                 buffer_info->skb = NULL;
1928
1929                 prefetch(skb->data - NET_IP_ALIGN);
1930
1931                 if (++i == rx_ring->count) i = 0;
1932                 next_rxd = IXGB_RX_DESC(*rx_ring, i);
1933                 prefetch(next_rxd);
1934
1935                 if ((j = i + 1) == rx_ring->count) j = 0;
1936                 next2_buffer = &rx_ring->buffer_info[j];
1937                 prefetch(next2_buffer);
1938
1939                 next_buffer = &rx_ring->buffer_info[i];
1940
1941                 cleaned = true;
1942                 cleaned_count++;
1943
1944                 pci_unmap_single(pdev,
1945                                  buffer_info->dma,
1946                                  buffer_info->length,
1947                                  PCI_DMA_FROMDEVICE);
1948                 buffer_info->dma = 0;
1949
1950                 length = le16_to_cpu(rx_desc->length);
1951                 rx_desc->length = 0;
1952
1953                 if (unlikely(!(status & IXGB_RX_DESC_STATUS_EOP))) {
1954
1955                         /* All receives must fit into a single buffer */
1956
1957                         IXGB_DBG("Receive packet consumed multiple buffers "
1958                                          "length<%x>\n", length);
1959
1960                         dev_kfree_skb_irq(skb);
1961                         goto rxdesc_done;
1962                 }
1963
1964                 if (unlikely(rx_desc->errors &
1965                     (IXGB_RX_DESC_ERRORS_CE | IXGB_RX_DESC_ERRORS_SE |
1966                      IXGB_RX_DESC_ERRORS_P | IXGB_RX_DESC_ERRORS_RXE))) {
1967                         dev_kfree_skb_irq(skb);
1968                         goto rxdesc_done;
1969                 }
1970
1971                 /* code added for copybreak, this should improve
1972                  * performance for small packets with large amounts
1973                  * of reassembly being done in the stack */
1974                 if (length < copybreak) {
1975                         struct sk_buff *new_skb =
1976                             netdev_alloc_skb(netdev, length + NET_IP_ALIGN);
1977                         if (new_skb) {
1978                                 skb_reserve(new_skb, NET_IP_ALIGN);
1979                                 skb_copy_to_linear_data_offset(new_skb,
1980                                                                -NET_IP_ALIGN,
1981                                                                (skb->data -
1982                                                                 NET_IP_ALIGN),
1983                                                                (length +
1984                                                                 NET_IP_ALIGN));
1985                                 /* save the skb in buffer_info as good */
1986                                 buffer_info->skb = skb;
1987                                 skb = new_skb;
1988                         }
1989                 }
1990                 /* end copybreak code */
1991
1992                 /* Good Receive */
1993                 skb_put(skb, length);
1994
1995                 /* Receive Checksum Offload */
1996                 ixgb_rx_checksum(adapter, rx_desc, skb);
1997
1998                 skb->protocol = eth_type_trans(skb, netdev);
1999                 if (adapter->vlgrp && (status & IXGB_RX_DESC_STATUS_VP)) {
2000                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
2001                                                 le16_to_cpu(rx_desc->special));
2002                 } else {
2003                         netif_receive_skb(skb);
2004                 }
2005
2006 rxdesc_done:
2007                 /* clean up descriptor, might be written over by hw */
2008                 rx_desc->status = 0;
2009
2010                 /* return some buffers to hardware, one at a time is too slow */
2011                 if (unlikely(cleaned_count >= IXGB_RX_BUFFER_WRITE)) {
2012                         ixgb_alloc_rx_buffers(adapter, cleaned_count);
2013                         cleaned_count = 0;
2014                 }
2015
2016                 /* use prefetched values */
2017                 rx_desc = next_rxd;
2018                 buffer_info = next_buffer;
2019         }
2020
2021         rx_ring->next_to_clean = i;
2022
2023         cleaned_count = IXGB_DESC_UNUSED(rx_ring);
2024         if (cleaned_count)
2025                 ixgb_alloc_rx_buffers(adapter, cleaned_count);
2026
2027         return cleaned;
2028 }
2029
2030 /**
2031  * ixgb_alloc_rx_buffers - Replace used receive buffers
2032  * @adapter: address of board private structure
2033  **/
2034
2035 static void
2036 ixgb_alloc_rx_buffers(struct ixgb_adapter *adapter, int cleaned_count)
2037 {
2038         struct ixgb_desc_ring *rx_ring = &adapter->rx_ring;
2039         struct net_device *netdev = adapter->netdev;
2040         struct pci_dev *pdev = adapter->pdev;
2041         struct ixgb_rx_desc *rx_desc;
2042         struct ixgb_buffer *buffer_info;
2043         struct sk_buff *skb;
2044         unsigned int i;
2045         long cleancount;
2046
2047         i = rx_ring->next_to_use;
2048         buffer_info = &rx_ring->buffer_info[i];
2049         cleancount = IXGB_DESC_UNUSED(rx_ring);
2050
2051
2052         /* leave three descriptors unused */
2053         while (--cleancount > 2 && cleaned_count--) {
2054                 /* recycle! its good for you */
2055                 skb = buffer_info->skb;
2056                 if (skb) {
2057                         skb_trim(skb, 0);
2058                         goto map_skb;
2059                 }
2060
2061                 skb = netdev_alloc_skb(netdev, adapter->rx_buffer_len
2062                                        + NET_IP_ALIGN);
2063                 if (unlikely(!skb)) {
2064                         /* Better luck next round */
2065                         adapter->alloc_rx_buff_failed++;
2066                         break;
2067                 }
2068
2069                 /* Make buffer alignment 2 beyond a 16 byte boundary
2070                  * this will result in a 16 byte aligned IP header after
2071                  * the 14 byte MAC header is removed
2072                  */
2073                 skb_reserve(skb, NET_IP_ALIGN);
2074
2075                 buffer_info->skb = skb;
2076                 buffer_info->length = adapter->rx_buffer_len;
2077 map_skb:
2078                 buffer_info->dma = pci_map_single(pdev,
2079                                                   skb->data,
2080                                                   adapter->rx_buffer_len,
2081                                                   PCI_DMA_FROMDEVICE);
2082
2083                 rx_desc = IXGB_RX_DESC(*rx_ring, i);
2084                 rx_desc->buff_addr = cpu_to_le64(buffer_info->dma);
2085                 /* guarantee DD bit not set now before h/w gets descriptor
2086                  * this is the rest of the workaround for h/w double
2087                  * writeback. */
2088                 rx_desc->status = 0;
2089
2090
2091                 if (++i == rx_ring->count) i = 0;
2092                 buffer_info = &rx_ring->buffer_info[i];
2093         }
2094
2095         if (likely(rx_ring->next_to_use != i)) {
2096                 rx_ring->next_to_use = i;
2097                 if (unlikely(i-- == 0))
2098                         i = (rx_ring->count - 1);
2099
2100                 /* Force memory writes to complete before letting h/w
2101                  * know there are new descriptors to fetch.  (Only
2102                  * applicable for weak-ordered memory model archs, such
2103                  * as IA-64). */
2104                 wmb();
2105                 IXGB_WRITE_REG(&adapter->hw, RDT, i);
2106         }
2107 }
2108
2109 /**
2110  * ixgb_vlan_rx_register - enables or disables vlan tagging/stripping.
2111  *
2112  * @param netdev network interface device structure
2113  * @param grp indicates to enable or disable tagging/stripping
2114  **/
2115 static void
2116 ixgb_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
2117 {
2118         struct ixgb_adapter *adapter = netdev_priv(netdev);
2119         u32 ctrl, rctl;
2120
2121         ixgb_irq_disable(adapter);
2122         adapter->vlgrp = grp;
2123
2124         if (grp) {
2125                 /* enable VLAN tag insert/strip */
2126                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2127                 ctrl |= IXGB_CTRL0_VME;
2128                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2129
2130                 /* enable VLAN receive filtering */
2131
2132                 rctl = IXGB_READ_REG(&adapter->hw, RCTL);
2133                 rctl &= ~IXGB_RCTL_CFIEN;
2134                 IXGB_WRITE_REG(&adapter->hw, RCTL, rctl);
2135         } else {
2136                 /* disable VLAN tag insert/strip */
2137
2138                 ctrl = IXGB_READ_REG(&adapter->hw, CTRL0);
2139                 ctrl &= ~IXGB_CTRL0_VME;
2140                 IXGB_WRITE_REG(&adapter->hw, CTRL0, ctrl);
2141         }
2142
2143         /* don't enable interrupts unless we are UP */
2144         if (adapter->netdev->flags & IFF_UP)
2145                 ixgb_irq_enable(adapter);
2146 }
2147
2148 static void
2149 ixgb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
2150 {
2151         struct ixgb_adapter *adapter = netdev_priv(netdev);
2152         u32 vfta, index;
2153
2154         /* add VID to filter table */
2155
2156         index = (vid >> 5) & 0x7F;
2157         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2158         vfta |= (1 << (vid & 0x1F));
2159         ixgb_write_vfta(&adapter->hw, index, vfta);
2160 }
2161
2162 static void
2163 ixgb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2164 {
2165         struct ixgb_adapter *adapter = netdev_priv(netdev);
2166         u32 vfta, index;
2167
2168         ixgb_irq_disable(adapter);
2169
2170         vlan_group_set_device(adapter->vlgrp, vid, NULL);
2171
2172         /* don't enable interrupts unless we are UP */
2173         if (adapter->netdev->flags & IFF_UP)
2174                 ixgb_irq_enable(adapter);
2175
2176         /* remove VID from filter table */
2177
2178         index = (vid >> 5) & 0x7F;
2179         vfta = IXGB_READ_REG_ARRAY(&adapter->hw, VFTA, index);
2180         vfta &= ~(1 << (vid & 0x1F));
2181         ixgb_write_vfta(&adapter->hw, index, vfta);
2182 }
2183
2184 static void
2185 ixgb_restore_vlan(struct ixgb_adapter *adapter)
2186 {
2187         ixgb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
2188
2189         if (adapter->vlgrp) {
2190                 u16 vid;
2191                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
2192                         if (!vlan_group_get_device(adapter->vlgrp, vid))
2193                                 continue;
2194                         ixgb_vlan_rx_add_vid(adapter->netdev, vid);
2195                 }
2196         }
2197 }
2198
2199 #ifdef CONFIG_NET_POLL_CONTROLLER
2200 /*
2201  * Polling 'interrupt' - used by things like netconsole to send skbs
2202  * without having to re-enable interrupts. It's not called while
2203  * the interrupt routine is executing.
2204  */
2205
2206 static void ixgb_netpoll(struct net_device *dev)
2207 {
2208         struct ixgb_adapter *adapter = netdev_priv(dev);
2209
2210         disable_irq(adapter->pdev->irq);
2211         ixgb_intr(adapter->pdev->irq, dev);
2212         enable_irq(adapter->pdev->irq);
2213 }
2214 #endif
2215
2216 /**
2217  * ixgb_io_error_detected() - called when PCI error is detected
2218  * @pdev    pointer to pci device with error
2219  * @state   pci channel state after error
2220  *
2221  * This callback is called by the PCI subsystem whenever
2222  * a PCI bus error is detected.
2223  */
2224 static pci_ers_result_t ixgb_io_error_detected(struct pci_dev *pdev,
2225                                                enum pci_channel_state state)
2226 {
2227         struct net_device *netdev = pci_get_drvdata(pdev);
2228         struct ixgb_adapter *adapter = netdev_priv(netdev);
2229
2230         if (netif_running(netdev))
2231                 ixgb_down(adapter, true);
2232
2233         pci_disable_device(pdev);
2234
2235         /* Request a slot reset. */
2236         return PCI_ERS_RESULT_NEED_RESET;
2237 }
2238
2239 /**
2240  * ixgb_io_slot_reset - called after the pci bus has been reset.
2241  * @pdev    pointer to pci device with error
2242  *
2243  * This callback is called after the PCI bus has been reset.
2244  * Basically, this tries to restart the card from scratch.
2245  * This is a shortened version of the device probe/discovery code,
2246  * it resembles the first-half of the ixgb_probe() routine.
2247  */
2248 static pci_ers_result_t ixgb_io_slot_reset(struct pci_dev *pdev)
2249 {
2250         struct net_device *netdev = pci_get_drvdata(pdev);
2251         struct ixgb_adapter *adapter = netdev_priv(netdev);
2252
2253         if (pci_enable_device(pdev)) {
2254                 DPRINTK(PROBE, ERR, "Cannot re-enable PCI device after reset.\n");
2255                 return PCI_ERS_RESULT_DISCONNECT;
2256         }
2257
2258         /* Perform card reset only on one instance of the card */
2259         if (0 != PCI_FUNC (pdev->devfn))
2260                 return PCI_ERS_RESULT_RECOVERED;
2261
2262         pci_set_master(pdev);
2263
2264         netif_carrier_off(netdev);
2265         netif_stop_queue(netdev);
2266         ixgb_reset(adapter);
2267
2268         /* Make sure the EEPROM is good */
2269         if (!ixgb_validate_eeprom_checksum(&adapter->hw)) {
2270                 DPRINTK(PROBE, ERR, "After reset, the EEPROM checksum is not valid.\n");
2271                 return PCI_ERS_RESULT_DISCONNECT;
2272         }
2273         ixgb_get_ee_mac_addr(&adapter->hw, netdev->dev_addr);
2274         memcpy(netdev->perm_addr, netdev->dev_addr, netdev->addr_len);
2275
2276         if (!is_valid_ether_addr(netdev->perm_addr)) {
2277                 DPRINTK(PROBE, ERR, "After reset, invalid MAC address.\n");
2278                 return PCI_ERS_RESULT_DISCONNECT;
2279         }
2280
2281         return PCI_ERS_RESULT_RECOVERED;
2282 }
2283
2284 /**
2285  * ixgb_io_resume - called when its OK to resume normal operations
2286  * @pdev    pointer to pci device with error
2287  *
2288  * The error recovery driver tells us that its OK to resume
2289  * normal operation. Implementation resembles the second-half
2290  * of the ixgb_probe() routine.
2291  */
2292 static void ixgb_io_resume(struct pci_dev *pdev)
2293 {
2294         struct net_device *netdev = pci_get_drvdata(pdev);
2295         struct ixgb_adapter *adapter = netdev_priv(netdev);
2296
2297         pci_set_master(pdev);
2298
2299         if (netif_running(netdev)) {
2300                 if (ixgb_up(adapter)) {
2301                         printk ("ixgb: can't bring device back up after reset\n");
2302                         return;
2303                 }
2304         }
2305
2306         netif_device_attach(netdev);
2307         mod_timer(&adapter->watchdog_timer, jiffies);
2308 }
2309
2310 /* ixgb_main.c */