drivers/net/igb/igb_main.c

   1 /*******************************************************************************
   2
   3   Intel(R) Gigabit Ethernet Linux driver
   4   Copyright(c) 2007 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   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/netdevice.h>
  34 #include <linux/ipv6.h>
  35 #include <net/checksum.h>
  36 #include <net/ip6_checksum.h>
  37 #include <linux/mii.h>
  38 #include <linux/ethtool.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/pci.h>
  41 #include <linux/delay.h>
  42 #include <linux/interrupt.h>
  43 #include <linux/if_ether.h>
  44 #ifdef CONFIG_DCA
  45 #include <linux/dca.h>
  46 #endif
  47 #include "igb.h"
  48
  49 #define DRV_VERSION "1.2.45-k2"
  50 char igb_driver_name[] = "igb";
  51 char igb_driver_version[] = DRV_VERSION;
  52 static const char igb_driver_string[] =
  53                                 "Intel(R) Gigabit Ethernet Network Driver";
  54 static const char igb_copyright[] = "Copyright (c) 2008 Intel Corporation.";
  55
  56 static const struct e1000_info *igb_info_tbl[] = {
  57         [board_82575] = &e1000_82575_info,
  58 };
  59
  60 static struct pci_device_id igb_pci_tbl[] = {
  61         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
  62         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
  63         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
  64         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
  65         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
  66         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
  67         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
  68         /* required last entry */
  69         {0, }
  70 };
  71
  72 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
  73
  74 void igb_reset(struct igb_adapter *);
  75 static int igb_setup_all_tx_resources(struct igb_adapter *);
  76 static int igb_setup_all_rx_resources(struct igb_adapter *);
  77 static void igb_free_all_tx_resources(struct igb_adapter *);
  78 static void igb_free_all_rx_resources(struct igb_adapter *);
  79 static void igb_free_tx_resources(struct igb_ring *);
  80 static void igb_free_rx_resources(struct igb_ring *);
  81 void igb_update_stats(struct igb_adapter *);
  82 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
  83 static void __devexit igb_remove(struct pci_dev *pdev);
  84 static int igb_sw_init(struct igb_adapter *);
  85 static int igb_open(struct net_device *);
  86 static int igb_close(struct net_device *);
  87 static void igb_configure_tx(struct igb_adapter *);
  88 static void igb_configure_rx(struct igb_adapter *);
  89 static void igb_setup_rctl(struct igb_adapter *);
  90 static void igb_clean_all_tx_rings(struct igb_adapter *);
  91 static void igb_clean_all_rx_rings(struct igb_adapter *);
  92 static void igb_clean_tx_ring(struct igb_ring *);
  93 static void igb_clean_rx_ring(struct igb_ring *);
  94 static void igb_set_multi(struct net_device *);
  95 static void igb_update_phy_info(unsigned long);
  96 static void igb_watchdog(unsigned long);
  97 static void igb_watchdog_task(struct work_struct *);
  98 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
  99                                   struct igb_ring *);
 100 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
 101 static struct net_device_stats *igb_get_stats(struct net_device *);
 102 static int igb_change_mtu(struct net_device *, int);
 103 static int igb_set_mac(struct net_device *, void *);
 104 static irqreturn_t igb_intr(int irq, void *);
 105 static irqreturn_t igb_intr_msi(int irq, void *);
 106 static irqreturn_t igb_msix_other(int irq, void *);
 107 static irqreturn_t igb_msix_rx(int irq, void *);
 108 static irqreturn_t igb_msix_tx(int irq, void *);
 109 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
 110 #ifdef CONFIG_DCA
 111 static void igb_update_rx_dca(struct igb_ring *);
 112 static void igb_update_tx_dca(struct igb_ring *);
 113 static void igb_setup_dca(struct igb_adapter *);
 114 #endif /* CONFIG_DCA */
 115 static bool igb_clean_tx_irq(struct igb_ring *);
 116 static int igb_poll(struct napi_struct *, int);
 117 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
 118 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
 119 #ifdef CONFIG_IGB_LRO
 120 static int igb_get_skb_hdr(struct sk_buff *skb, void **, void **, u64 *, void *);
 121 #endif
 122 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 123 static void igb_tx_timeout(struct net_device *);
 124 static void igb_reset_task(struct work_struct *);
 125 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
 126 static void igb_vlan_rx_add_vid(struct net_device *, u16);
 127 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
 128 static void igb_restore_vlan(struct igb_adapter *);
 129
 130 static int igb_suspend(struct pci_dev *, pm_message_t);
 131 #ifdef CONFIG_PM
 132 static int igb_resume(struct pci_dev *);
 133 #endif
 134 static void igb_shutdown(struct pci_dev *);
 135 #ifdef CONFIG_DCA
 136 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 137 static struct notifier_block dca_notifier = {
 138         .notifier_call  = igb_notify_dca,
 139         .next           = NULL,
 140         .priority       = 0
 141 };
 142 #endif
 143
 144 #ifdef CONFIG_NET_POLL_CONTROLLER
 145 /* for netdump / net console */
 146 static void igb_netpoll(struct net_device *);
 147 #endif
 148
 149 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 150                      pci_channel_state_t);
 151 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 152 static void igb_io_resume(struct pci_dev *);
 153
 154 static struct pci_error_handlers igb_err_handler = {
 155         .error_detected = igb_io_error_detected,
 156         .slot_reset = igb_io_slot_reset,
 157         .resume = igb_io_resume,
 158 };
 159
 160
 161 static struct pci_driver igb_driver = {
 162         .name     = igb_driver_name,
 163         .id_table = igb_pci_tbl,
 164         .probe    = igb_probe,
 165         .remove   = __devexit_p(igb_remove),
 166 #ifdef CONFIG_PM
 167         /* Power Managment Hooks */
 168         .suspend  = igb_suspend,
 169         .resume   = igb_resume,
 170 #endif
 171         .shutdown = igb_shutdown,
 172         .err_handler = &igb_err_handler
 173 };
 174
 175 static int global_quad_port_a; /* global quad port a indication */
 176
 177 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 178 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 179 MODULE_LICENSE("GPL");
 180 MODULE_VERSION(DRV_VERSION);
 181
 182 #ifdef DEBUG
 183 /**
 184  * igb_get_hw_dev_name - return device name string
 185  * used by hardware layer to print debugging information
 186  **/
 187 char *igb_get_hw_dev_name(struct e1000_hw *hw)
 188 {
 189         struct igb_adapter *adapter = hw->back;
 190         return adapter->netdev->name;
 191 }
 192 #endif
 193
 194 /**
 195  * igb_init_module - Driver Registration Routine
 196  *
 197  * igb_init_module is the first routine called when the driver is
 198  * loaded. All it does is register with the PCI subsystem.
 199  **/
 200 static int __init igb_init_module(void)
 201 {
 202         int ret;
 203         printk(KERN_INFO "%s - version %s\n",
 204                igb_driver_string, igb_driver_version);
 205
 206         printk(KERN_INFO "%s\n", igb_copyright);
 207
 208         global_quad_port_a = 0;
 209
 210         ret = pci_register_driver(&igb_driver);
 211 #ifdef CONFIG_DCA
 212         dca_register_notify(&dca_notifier);
 213 #endif
 214         return ret;
 215 }
 216
 217 module_init(igb_init_module);
 218
 219 /**
 220  * igb_exit_module - Driver Exit Cleanup Routine
 221  *
 222  * igb_exit_module is called just before the driver is removed
 223  * from memory.
 224  **/
 225 static void __exit igb_exit_module(void)
 226 {
 227 #ifdef CONFIG_DCA
 228         dca_unregister_notify(&dca_notifier);
 229 #endif
 230         pci_unregister_driver(&igb_driver);
 231 }
 232
 233 module_exit(igb_exit_module);
 234
 235 /**
 236  * igb_alloc_queues - Allocate memory for all rings
 237  * @adapter: board private structure to initialize
 238  *
 239  * We allocate one ring per queue at run-time since we don't know the
 240  * number of queues at compile-time.
 241  **/
 242 static int igb_alloc_queues(struct igb_adapter *adapter)
 243 {
 244         int i;
 245
 246         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
 247                                    sizeof(struct igb_ring), GFP_KERNEL);
 248         if (!adapter->tx_ring)
 249                 return -ENOMEM;
 250
 251         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
 252                                    sizeof(struct igb_ring), GFP_KERNEL);
 253         if (!adapter->rx_ring) {
 254                 kfree(adapter->tx_ring);
 255                 return -ENOMEM;
 256         }
 257
 258         adapter->rx_ring->buddy = adapter->tx_ring;
 259
 260         for (i = 0; i < adapter->num_tx_queues; i++) {
 261                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 262                 ring->adapter = adapter;
 263                 ring->queue_index = i;
 264         }
 265         for (i = 0; i < adapter->num_rx_queues; i++) {
 266                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 267                 ring->adapter = adapter;
 268                 ring->queue_index = i;
 269                 ring->itr_register = E1000_ITR;
 270
 271                 /* set a default napi handler for each rx_ring */
 272                 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
 273         }
 274         return 0;
 275 }
 276
 277 static void igb_free_queues(struct igb_adapter *adapter)
 278 {
 279         int i;
 280
 281         for (i = 0; i < adapter->num_rx_queues; i++)
 282                 netif_napi_del(&adapter->rx_ring[i].napi);
 283
 284         kfree(adapter->tx_ring);
 285         kfree(adapter->rx_ring);
 286 }
 287
 288 #define IGB_N0_QUEUE -1
 289 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
 290                               int tx_queue, int msix_vector)
 291 {
 292         u32 msixbm = 0;
 293         struct e1000_hw *hw = &adapter->hw;
 294         u32 ivar, index;
 295
 296         switch (hw->mac.type) {
 297         case e1000_82575:
 298                 /* The 82575 assigns vectors using a bitmask, which matches the
 299                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
 300                    or more queues to a vector, we write the appropriate bits
 301                    into the MSIXBM register for that vector. */
 302                 if (rx_queue > IGB_N0_QUEUE) {
 303                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 304                         adapter->rx_ring[rx_queue].eims_value = msixbm;
 305                 }
 306                 if (tx_queue > IGB_N0_QUEUE) {
 307                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 308                         adapter->tx_ring[tx_queue].eims_value =
 309                                   E1000_EICR_TX_QUEUE0 << tx_queue;
 310                 }
 311                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 312                 break;
 313         case e1000_82576:
 314                 /* Kawela uses a table-based method for assigning vectors.
 315                    Each queue has a single entry in the table to which we write
 316                    a vector number along with a "valid" bit.  Sadly, the layout
 317                    of the table is somewhat counterintuitive. */
 318                 if (rx_queue > IGB_N0_QUEUE) {
 319                         index = (rx_queue & 0x7);
 320                         ivar = array_rd32(E1000_IVAR0, index);
 321                         if (rx_queue < 8) {
 322                                 /* vector goes into low byte of register */
 323                                 ivar = ivar & 0xFFFFFF00;
 324                                 ivar |= msix_vector | E1000_IVAR_VALID;
 325                         } else {
 326                                 /* vector goes into third byte of register */
 327                                 ivar = ivar & 0xFF00FFFF;
 328                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 329                         }
 330                         adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
 331                         array_wr32(E1000_IVAR0, index, ivar);
 332                 }
 333                 if (tx_queue > IGB_N0_QUEUE) {
 334                         index = (tx_queue & 0x7);
 335                         ivar = array_rd32(E1000_IVAR0, index);
 336                         if (tx_queue < 8) {
 337                                 /* vector goes into second byte of register */
 338                                 ivar = ivar & 0xFFFF00FF;
 339                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 340                         } else {
 341                                 /* vector goes into high byte of register */
 342                                 ivar = ivar & 0x00FFFFFF;
 343                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 344                         }
 345                         adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
 346                         array_wr32(E1000_IVAR0, index, ivar);
 347                 }
 348                 break;
 349         default:
 350                 BUG();
 351                 break;
 352         }
 353 }
 354
 355 /**
 356  * igb_configure_msix - Configure MSI-X hardware
 357  *
 358  * igb_configure_msix sets up the hardware to properly
 359  * generate MSI-X interrupts.
 360  **/
 361 static void igb_configure_msix(struct igb_adapter *adapter)
 362 {
 363         u32 tmp;
 364         int i, vector = 0;
 365         struct e1000_hw *hw = &adapter->hw;
 366
 367         adapter->eims_enable_mask = 0;
 368         if (hw->mac.type == e1000_82576)
 369                 /* Turn on MSI-X capability first, or our settings
 370                  * won't stick.  And it will take days to debug. */
 371                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 372                                    E1000_GPIE_PBA | E1000_GPIE_EIAME |
 373                                    E1000_GPIE_NSICR);
 374
 375         for (i = 0; i < adapter->num_tx_queues; i++) {
 376                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
 377                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
 378                 adapter->eims_enable_mask |= tx_ring->eims_value;
 379                 if (tx_ring->itr_val)
 380                         writel(tx_ring->itr_val,
 381                                hw->hw_addr + tx_ring->itr_register);
 382                 else
 383                         writel(1, hw->hw_addr + tx_ring->itr_register);
 384         }
 385
 386         for (i = 0; i < adapter->num_rx_queues; i++) {
 387                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
 388                 rx_ring->buddy = 0;
 389                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
 390                 adapter->eims_enable_mask |= rx_ring->eims_value;
 391                 if (rx_ring->itr_val)
 392                         writel(rx_ring->itr_val,
 393                                hw->hw_addr + rx_ring->itr_register);
 394                 else
 395                         writel(1, hw->hw_addr + rx_ring->itr_register);
 396         }
 397
 398
 399         /* set vector for other causes, i.e. link changes */
 400         switch (hw->mac.type) {
 401         case e1000_82575:
 402                 array_wr32(E1000_MSIXBM(0), vector++,
 403                                       E1000_EIMS_OTHER);
 404
 405                 tmp = rd32(E1000_CTRL_EXT);
 406                 /* enable MSI-X PBA support*/
 407                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 408
 409                 /* Auto-Mask interrupts upon ICR read. */
 410                 tmp |= E1000_CTRL_EXT_EIAME;
 411                 tmp |= E1000_CTRL_EXT_IRCA;
 412
 413                 wr32(E1000_CTRL_EXT, tmp);
 414                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
 415                 adapter->eims_other = E1000_EIMS_OTHER;
 416
 417                 break;
 418
 419         case e1000_82576:
 420                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 421                 wr32(E1000_IVAR_MISC, tmp);
 422
 423                 adapter->eims_enable_mask = (1 << (vector)) - 1;
 424                 adapter->eims_other = 1 << (vector - 1);
 425                 break;
 426         default:
 427                 /* do nothing, since nothing else supports MSI-X */
 428                 break;
 429         } /* switch (hw->mac.type) */
 430         wrfl();
 431 }
 432
 433 /**
 434  * igb_request_msix - Initialize MSI-X interrupts
 435  *
 436  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
 437  * kernel.
 438  **/
 439 static int igb_request_msix(struct igb_adapter *adapter)
 440 {
 441         struct net_device *netdev = adapter->netdev;
 442         int i, err = 0, vector = 0;
 443
 444         vector = 0;
 445
 446         for (i = 0; i < adapter->num_tx_queues; i++) {
 447                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 448                 sprintf(ring->name, "%s-tx%d", netdev->name, i);
 449                 err = request_irq(adapter->msix_entries[vector].vector,
 450                                   &igb_msix_tx, 0, ring->name,
 451                                   &(adapter->tx_ring[i]));
 452                 if (err)
 453                         goto out;
 454                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 455                 ring->itr_val = 976; /* ~4000 ints/sec */
 456                 vector++;
 457         }
 458         for (i = 0; i < adapter->num_rx_queues; i++) {
 459                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 460                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
 461                         sprintf(ring->name, "%s-rx%d", netdev->name, i);
 462                 else
 463                         memcpy(ring->name, netdev->name, IFNAMSIZ);
 464                 err = request_irq(adapter->msix_entries[vector].vector,
 465                                   &igb_msix_rx, 0, ring->name,
 466                                   &(adapter->rx_ring[i]));
 467                 if (err)
 468                         goto out;
 469                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 470                 ring->itr_val = adapter->itr;
 471                 /* overwrite the poll routine for MSIX, we've already done
 472                  * netif_napi_add */
 473                 ring->napi.poll = &igb_clean_rx_ring_msix;
 474                 vector++;
 475         }
 476
 477         err = request_irq(adapter->msix_entries[vector].vector,
 478                           &igb_msix_other, 0, netdev->name, netdev);
 479         if (err)
 480                 goto out;
 481
 482         igb_configure_msix(adapter);
 483         return 0;
 484 out:
 485         return err;
 486 }
 487
 488 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 489 {
 490         if (adapter->msix_entries) {
 491                 pci_disable_msix(adapter->pdev);
 492                 kfree(adapter->msix_entries);
 493                 adapter->msix_entries = NULL;
 494         } else if (adapter->flags & IGB_FLAG_HAS_MSI)
 495                 pci_disable_msi(adapter->pdev);
 496         return;
 497 }
 498
 499
 500 /**
 501  * igb_set_interrupt_capability - set MSI or MSI-X if supported
 502  *
 503  * Attempt to configure interrupts using the best available
 504  * capabilities of the hardware and kernel.
 505  **/
 506 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
 507 {
 508         int err;
 509         int numvecs, i;
 510
 511         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
 512         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
 513                                         GFP_KERNEL);
 514         if (!adapter->msix_entries)
 515                 goto msi_only;
 516
 517         for (i = 0; i < numvecs; i++)
 518                 adapter->msix_entries[i].entry = i;
 519
 520         err = pci_enable_msix(adapter->pdev,
 521                               adapter->msix_entries,
 522                               numvecs);
 523         if (err == 0)
 524                 return;
 525
 526         igb_reset_interrupt_capability(adapter);
 527
 528         /* If we can't do MSI-X, try MSI */
 529 msi_only:
 530         adapter->num_rx_queues = 1;
 531         adapter->num_tx_queues = 1;
 532         if (!pci_enable_msi(adapter->pdev))
 533                 adapter->flags |= IGB_FLAG_HAS_MSI;
 534
 535         /* Notify the stack of the (possibly) reduced Tx Queue count. */
 536         adapter->netdev->egress_subqueue_count = adapter->num_tx_queues;
 537         return;
 538 }
 539
 540 /**
 541  * igb_request_irq - initialize interrupts
 542  *
 543  * Attempts to configure interrupts using the best available
 544  * capabilities of the hardware and kernel.
 545  **/
 546 static int igb_request_irq(struct igb_adapter *adapter)
 547 {
 548         struct net_device *netdev = adapter->netdev;
 549         struct e1000_hw *hw = &adapter->hw;
 550         int err = 0;
 551
 552         if (adapter->msix_entries) {
 553                 err = igb_request_msix(adapter);
 554                 if (!err)
 555                         goto request_done;
 556                 /* fall back to MSI */
 557                 igb_reset_interrupt_capability(adapter);
 558                 if (!pci_enable_msi(adapter->pdev))
 559                         adapter->flags |= IGB_FLAG_HAS_MSI;
 560                 igb_free_all_tx_resources(adapter);
 561                 igb_free_all_rx_resources(adapter);
 562                 adapter->num_rx_queues = 1;
 563                 igb_alloc_queues(adapter);
 564         } else {
 565                 switch (hw->mac.type) {
 566                 case e1000_82575:
 567                         wr32(E1000_MSIXBM(0),
 568                              (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
 569                         break;
 570                 case e1000_82576:
 571                         wr32(E1000_IVAR0, E1000_IVAR_VALID);
 572                         break;
 573                 default:
 574                         break;
 575                 }
 576         }
 577
 578         if (adapter->flags & IGB_FLAG_HAS_MSI) {
 579                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
 580                                   netdev->name, netdev);
 581                 if (!err)
 582                         goto request_done;
 583                 /* fall back to legacy interrupts */
 584                 igb_reset_interrupt_capability(adapter);
 585                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
 586         }
 587
 588         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
 589                           netdev->name, netdev);
 590
 591         if (err)
 592                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
 593                         err);
 594
 595 request_done:
 596         return err;
 597 }
 598
 599 static void igb_free_irq(struct igb_adapter *adapter)
 600 {
 601         struct net_device *netdev = adapter->netdev;
 602
 603         if (adapter->msix_entries) {
 604                 int vector = 0, i;
 605
 606                 for (i = 0; i < adapter->num_tx_queues; i++)
 607                         free_irq(adapter->msix_entries[vector++].vector,
 608                                 &(adapter->tx_ring[i]));
 609                 for (i = 0; i < adapter->num_rx_queues; i++)
 610                         free_irq(adapter->msix_entries[vector++].vector,
 611                                 &(adapter->rx_ring[i]));
 612
 613                 free_irq(adapter->msix_entries[vector++].vector, netdev);
 614                 return;
 615         }
 616
 617         free_irq(adapter->pdev->irq, netdev);
 618 }
 619
 620 /**
 621  * igb_irq_disable - Mask off interrupt generation on the NIC
 622  * @adapter: board private structure
 623  **/
 624 static void igb_irq_disable(struct igb_adapter *adapter)
 625 {
 626         struct e1000_hw *hw = &adapter->hw;
 627
 628         if (adapter->msix_entries) {
 629                 wr32(E1000_EIAM, 0);
 630                 wr32(E1000_EIMC, ~0);
 631                 wr32(E1000_EIAC, 0);
 632         }
 633
 634         wr32(E1000_IAM, 0);
 635         wr32(E1000_IMC, ~0);
 636         wrfl();
 637         synchronize_irq(adapter->pdev->irq);
 638 }
 639
 640 /**
 641  * igb_irq_enable - Enable default interrupt generation settings
 642  * @adapter: board private structure
 643  **/
 644 static void igb_irq_enable(struct igb_adapter *adapter)
 645 {
 646         struct e1000_hw *hw = &adapter->hw;
 647
 648         if (adapter->msix_entries) {
 649                 wr32(E1000_EIAC, adapter->eims_enable_mask);
 650                 wr32(E1000_EIAM, adapter->eims_enable_mask);
 651                 wr32(E1000_EIMS, adapter->eims_enable_mask);
 652                 wr32(E1000_IMS, E1000_IMS_LSC);
 653         } else {
 654                 wr32(E1000_IMS, IMS_ENABLE_MASK);
 655                 wr32(E1000_IAM, IMS_ENABLE_MASK);
 656         }
 657 }
 658
 659 static void igb_update_mng_vlan(struct igb_adapter *adapter)
 660 {
 661         struct net_device *netdev = adapter->netdev;
 662         u16 vid = adapter->hw.mng_cookie.vlan_id;
 663         u16 old_vid = adapter->mng_vlan_id;
 664         if (adapter->vlgrp) {
 665                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
 666                         if (adapter->hw.mng_cookie.status &
 667                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
 668                                 igb_vlan_rx_add_vid(netdev, vid);
 669                                 adapter->mng_vlan_id = vid;
 670                         } else
 671                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
 672
 673                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
 674                                         (vid != old_vid) &&
 675                             !vlan_group_get_device(adapter->vlgrp, old_vid))
 676                                 igb_vlan_rx_kill_vid(netdev, old_vid);
 677                 } else
 678                         adapter->mng_vlan_id = vid;
 679         }
 680 }
 681
 682 /**
 683  * igb_release_hw_control - release control of the h/w to f/w
 684  * @adapter: address of board private structure
 685  *
 686  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 687  * For ASF and Pass Through versions of f/w this means that the
 688  * driver is no longer loaded.
 689  *
 690  **/
 691 static void igb_release_hw_control(struct igb_adapter *adapter)
 692 {
 693         struct e1000_hw *hw = &adapter->hw;
 694         u32 ctrl_ext;
 695
 696         /* Let firmware take over control of h/w */
 697         ctrl_ext = rd32(E1000_CTRL_EXT);
 698         wr32(E1000_CTRL_EXT,
 699                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 700 }
 701
 702
 703 /**
 704  * igb_get_hw_control - get control of the h/w from f/w
 705  * @adapter: address of board private structure
 706  *
 707  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 708  * For ASF and Pass Through versions of f/w this means that
 709  * the driver is loaded.
 710  *
 711  **/
 712 static void igb_get_hw_control(struct igb_adapter *adapter)
 713 {
 714         struct e1000_hw *hw = &adapter->hw;
 715         u32 ctrl_ext;
 716
 717         /* Let firmware know the driver has taken over */
 718         ctrl_ext = rd32(E1000_CTRL_EXT);
 719         wr32(E1000_CTRL_EXT,
 720                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 721 }
 722
 723 static void igb_init_manageability(struct igb_adapter *adapter)
 724 {
 725         struct e1000_hw *hw = &adapter->hw;
 726
 727         if (adapter->en_mng_pt) {
 728                 u32 manc2h = rd32(E1000_MANC2H);
 729                 u32 manc = rd32(E1000_MANC);
 730
 731                 /* enable receiving management packets to the host */
 732                 /* this will probably generate destination unreachable messages
 733                  * from the host OS, but the packets will be handled on SMBUS */
 734                 manc |= E1000_MANC_EN_MNG2HOST;
 735 #define E1000_MNG2HOST_PORT_623 (1 << 5)
 736 #define E1000_MNG2HOST_PORT_664 (1 << 6)
 737                 manc2h |= E1000_MNG2HOST_PORT_623;
 738                 manc2h |= E1000_MNG2HOST_PORT_664;
 739                 wr32(E1000_MANC2H, manc2h);
 740
 741                 wr32(E1000_MANC, manc);
 742         }
 743 }
 744
 745 /**
 746  * igb_configure - configure the hardware for RX and TX
 747  * @adapter: private board structure
 748  **/
 749 static void igb_configure(struct igb_adapter *adapter)
 750 {
 751         struct net_device *netdev = adapter->netdev;
 752         int i;
 753
 754         igb_get_hw_control(adapter);
 755         igb_set_multi(netdev);
 756
 757         igb_restore_vlan(adapter);
 758         igb_init_manageability(adapter);
 759
 760         igb_configure_tx(adapter);
 761         igb_setup_rctl(adapter);
 762         igb_configure_rx(adapter);
 763
 764         igb_rx_fifo_flush_82575(&adapter->hw);
 765
 766         /* call IGB_DESC_UNUSED which always leaves
 767          * at least 1 descriptor unused to make sure
 768          * next_to_use != next_to_clean */
 769         for (i = 0; i < adapter->num_rx_queues; i++) {
 770                 struct igb_ring *ring = &adapter->rx_ring[i];
 771                 igb_alloc_rx_buffers_adv(ring, IGB_DESC_UNUSED(ring));
 772         }
 773
 774
 775         adapter->tx_queue_len = netdev->tx_queue_len;
 776 }
 777
 778
 779 /**
 780  * igb_up - Open the interface and prepare it to handle traffic
 781  * @adapter: board private structure
 782  **/
 783
 784 int igb_up(struct igb_adapter *adapter)
 785 {
 786         struct e1000_hw *hw = &adapter->hw;
 787         int i;
 788
 789         /* hardware has been reset, we need to reload some things */
 790         igb_configure(adapter);
 791
 792         clear_bit(__IGB_DOWN, &adapter->state);
 793
 794         for (i = 0; i < adapter->num_rx_queues; i++)
 795                 napi_enable(&adapter->rx_ring[i].napi);
 796         if (adapter->msix_entries)
 797                 igb_configure_msix(adapter);
 798
 799         /* Clear any pending interrupts. */
 800         rd32(E1000_ICR);
 801         igb_irq_enable(adapter);
 802
 803         /* Fire a link change interrupt to start the watchdog. */
 804         wr32(E1000_ICS, E1000_ICS_LSC);
 805         return 0;
 806 }
 807
 808 void igb_down(struct igb_adapter *adapter)
 809 {
 810         struct e1000_hw *hw = &adapter->hw;
 811         struct net_device *netdev = adapter->netdev;
 812         u32 tctl, rctl;
 813         int i;
 814
 815         /* signal that we're down so the interrupt handler does not
 816          * reschedule our watchdog timer */
 817         set_bit(__IGB_DOWN, &adapter->state);
 818
 819         /* disable receives in the hardware */
 820         rctl = rd32(E1000_RCTL);
 821         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
 822         /* flush and sleep below */
 823
 824         netif_stop_queue(netdev);
 825         for (i = 0; i < adapter->num_tx_queues; i++)
 826                 netif_stop_subqueue(netdev, i);
 827
 828         /* disable transmits in the hardware */
 829         tctl = rd32(E1000_TCTL);
 830         tctl &= ~E1000_TCTL_EN;
 831         wr32(E1000_TCTL, tctl);
 832         /* flush both disables and wait for them to finish */
 833         wrfl();
 834         msleep(10);
 835
 836         for (i = 0; i < adapter->num_rx_queues; i++)
 837                 napi_disable(&adapter->rx_ring[i].napi);
 838
 839         igb_irq_disable(adapter);
 840
 841         del_timer_sync(&adapter->watchdog_timer);
 842         del_timer_sync(&adapter->phy_info_timer);
 843
 844         netdev->tx_queue_len = adapter->tx_queue_len;
 845         netif_carrier_off(netdev);
 846         adapter->link_speed = 0;
 847         adapter->link_duplex = 0;
 848
 849         if (!pci_channel_offline(adapter->pdev))
 850                 igb_reset(adapter);
 851         igb_clean_all_tx_rings(adapter);
 852         igb_clean_all_rx_rings(adapter);
 853 }
 854
 855 void igb_reinit_locked(struct igb_adapter *adapter)
 856 {
 857         WARN_ON(in_interrupt());
 858         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 859                 msleep(1);
 860         igb_down(adapter);
 861         igb_up(adapter);
 862         clear_bit(__IGB_RESETTING, &adapter->state);
 863 }
 864
 865 void igb_reset(struct igb_adapter *adapter)
 866 {
 867         struct e1000_hw *hw = &adapter->hw;
 868         struct e1000_mac_info *mac = &hw->mac;
 869         struct e1000_fc_info *fc = &hw->fc;
 870         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
 871         u16 hwm;
 872
 873         /* Repartition Pba for greater than 9k mtu
 874          * To take effect CTRL.RST is required.
 875          */
 876         if (mac->type != e1000_82576) {
 877         pba = E1000_PBA_34K;
 878         }
 879         else {
 880                 pba = E1000_PBA_64K;
 881         }
 882
 883         if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
 884             (mac->type < e1000_82576)) {
 885                 /* adjust PBA for jumbo frames */
 886                 wr32(E1000_PBA, pba);
 887
 888                 /* To maintain wire speed transmits, the Tx FIFO should be
 889                  * large enough to accommodate two full transmit packets,
 890                  * rounded up to the next 1KB and expressed in KB.  Likewise,
 891                  * the Rx FIFO should be large enough to accommodate at least
 892                  * one full receive packet and is similarly rounded up and
 893                  * expressed in KB. */
 894                 pba = rd32(E1000_PBA);
 895                 /* upper 16 bits has Tx packet buffer allocation size in KB */
 896                 tx_space = pba >> 16;
 897                 /* lower 16 bits has Rx packet buffer allocation size in KB */
 898                 pba &= 0xffff;
 899                 /* the tx fifo also stores 16 bytes of information about the tx
 900                  * but don't include ethernet FCS because hardware appends it */
 901                 min_tx_space = (adapter->max_frame_size +
 902                                 sizeof(struct e1000_tx_desc) -
 903                                 ETH_FCS_LEN) * 2;
 904                 min_tx_space = ALIGN(min_tx_space, 1024);
 905                 min_tx_space >>= 10;
 906                 /* software strips receive CRC, so leave room for it */
 907                 min_rx_space = adapter->max_frame_size;
 908                 min_rx_space = ALIGN(min_rx_space, 1024);
 909                 min_rx_space >>= 10;
 910
 911                 /* If current Tx allocation is less than the min Tx FIFO size,
 912                  * and the min Tx FIFO size is less than the current Rx FIFO
 913                  * allocation, take space away from current Rx allocation */
 914                 if (tx_space < min_tx_space &&
 915                     ((min_tx_space - tx_space) < pba)) {
 916                         pba = pba - (min_tx_space - tx_space);
 917
 918                         /* if short on rx space, rx wins and must trump tx
 919                          * adjustment */
 920                         if (pba < min_rx_space)
 921                                 pba = min_rx_space;
 922                 }
 923                 wr32(E1000_PBA, pba);
 924         }
 925
 926         /* flow control settings */
 927         /* The high water mark must be low enough to fit one full frame
 928          * (or the size used for early receive) above it in the Rx FIFO.
 929          * Set it to the lower of:
 930          * - 90% of the Rx FIFO size, or
 931          * - the full Rx FIFO size minus one full frame */
 932         hwm = min(((pba << 10) * 9 / 10),
 933                         ((pba << 10) - 2 * adapter->max_frame_size));
 934
 935         if (mac->type < e1000_82576) {
 936                 fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
 937                 fc->low_water = fc->high_water - 8;
 938         } else {
 939                 fc->high_water = hwm & 0xFFF0;  /* 16-byte granularity */
 940                 fc->low_water = fc->high_water - 16;
 941         }
 942         fc->pause_time = 0xFFFF;
 943         fc->send_xon = 1;
 944         fc->type = fc->original_type;
 945
 946         /* Allow time for pending master requests to run */
 947         adapter->hw.mac.ops.reset_hw(&adapter->hw);
 948         wr32(E1000_WUC, 0);
 949
 950         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
 951                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
 952
 953         igb_update_mng_vlan(adapter);
 954
 955         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 956         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
 957
 958         igb_reset_adaptive(&adapter->hw);
 959         if (adapter->hw.phy.ops.get_phy_info)
 960                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
 961 }
 962
 963 /**
 964  * igb_is_need_ioport - determine if an adapter needs ioport resources or not
 965  * @pdev: PCI device information struct
 966  *
 967  * Returns true if an adapter needs ioport resources
 968  **/
 969 static int igb_is_need_ioport(struct pci_dev *pdev)
 970 {
 971         switch (pdev->device) {
 972         /* Currently there are no adapters that need ioport resources */
 973         default:
 974                 return false;
 975         }
 976 }
 977
 978 /**
 979  * igb_probe - Device Initialization Routine
 980  * @pdev: PCI device information struct
 981  * @ent: entry in igb_pci_tbl
 982  *
 983  * Returns 0 on success, negative on failure
 984  *
 985  * igb_probe initializes an adapter identified by a pci_dev structure.
 986  * The OS initialization, configuring of the adapter private structure,
 987  * and a hardware reset occur.
 988  **/
 989 static int __devinit igb_probe(struct pci_dev *pdev,
 990                                const struct pci_device_id *ent)
 991 {
 992         struct net_device *netdev;
 993         struct igb_adapter *adapter;
 994         struct e1000_hw *hw;
 995         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
 996         unsigned long mmio_start, mmio_len;
 997         int i, err, pci_using_dac;
 998         u16 eeprom_data = 0;
 999         u16 eeprom_apme_mask = IGB_EEPROM_APME;
1000         u32 part_num;
1001         int bars, need_ioport;
1002
1003         /* do not allocate ioport bars when not needed */
1004         need_ioport = igb_is_need_ioport(pdev);
1005         if (need_ioport) {
1006                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
1007                 err = pci_enable_device(pdev);
1008         } else {
1009                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
1010                 err = pci_enable_device_mem(pdev);
1011         }
1012         if (err)
1013                 return err;
1014
1015         pci_using_dac = 0;
1016         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1017         if (!err) {
1018                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1019                 if (!err)
1020                         pci_using_dac = 1;
1021         } else {
1022                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1023                 if (err) {
1024                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1025                         if (err) {
1026                                 dev_err(&pdev->dev, "No usable DMA "
1027                                         "configuration, aborting\n");
1028                                 goto err_dma;
1029                         }
1030                 }
1031         }
1032
1033         err = pci_request_selected_regions(pdev, bars, igb_driver_name);
1034         if (err)
1035                 goto err_pci_reg;
1036
1037         pci_set_master(pdev);
1038         pci_save_state(pdev);
1039
1040         err = -ENOMEM;
1041         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter), IGB_MAX_TX_QUEUES);
1042         if (!netdev)
1043                 goto err_alloc_etherdev;
1044
1045         SET_NETDEV_DEV(netdev, &pdev->dev);
1046
1047         pci_set_drvdata(pdev, netdev);
1048         adapter = netdev_priv(netdev);
1049         adapter->netdev = netdev;
1050         adapter->pdev = pdev;
1051         hw = &adapter->hw;
1052         hw->back = adapter;
1053         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1054         adapter->bars = bars;
1055         adapter->need_ioport = need_ioport;
1056
1057         mmio_start = pci_resource_start(pdev, 0);
1058         mmio_len = pci_resource_len(pdev, 0);
1059
1060         err = -EIO;
1061         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
1062         if (!adapter->hw.hw_addr)
1063                 goto err_ioremap;
1064
1065         netdev->open = &igb_open;
1066         netdev->stop = &igb_close;
1067         netdev->get_stats = &igb_get_stats;
1068         netdev->set_multicast_list = &igb_set_multi;
1069         netdev->set_mac_address = &igb_set_mac;
1070         netdev->change_mtu = &igb_change_mtu;
1071         netdev->do_ioctl = &igb_ioctl;
1072         igb_set_ethtool_ops(netdev);
1073         netdev->tx_timeout = &igb_tx_timeout;
1074         netdev->watchdog_timeo = 5 * HZ;
1075         netdev->vlan_rx_register = igb_vlan_rx_register;
1076         netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
1077         netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
1078 #ifdef CONFIG_NET_POLL_CONTROLLER
1079         netdev->poll_controller = igb_netpoll;
1080 #endif
1081         netdev->hard_start_xmit = &igb_xmit_frame_adv;
1082
1083         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1084
1085         netdev->mem_start = mmio_start;
1086         netdev->mem_end = mmio_start + mmio_len;
1087
1088         /* PCI config space info */
1089         hw->vendor_id = pdev->vendor;
1090         hw->device_id = pdev->device;
1091         hw->revision_id = pdev->revision;
1092         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1093         hw->subsystem_device_id = pdev->subsystem_device;
1094
1095         /* setup the private structure */
1096         hw->back = adapter;
1097         /* Copy the default MAC, PHY and NVM function pointers */
1098         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1099         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1100         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1101         /* Initialize skew-specific constants */
1102         err = ei->get_invariants(hw);
1103         if (err)
1104                 goto err_hw_init;
1105
1106         err = igb_sw_init(adapter);
1107         if (err)
1108                 goto err_sw_init;
1109
1110         igb_get_bus_info_pcie(hw);
1111
1112         /* set flags */
1113         switch (hw->mac.type) {
1114         case e1000_82576:
1115         case e1000_82575:
1116                 adapter->flags |= IGB_FLAG_HAS_DCA;
1117                 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1118                 break;
1119         default:
1120                 break;
1121         }
1122
1123         hw->phy.autoneg_wait_to_complete = false;
1124         hw->mac.adaptive_ifs = true;
1125
1126         /* Copper options */
1127         if (hw->phy.media_type == e1000_media_type_copper) {
1128                 hw->phy.mdix = AUTO_ALL_MODES;
1129                 hw->phy.disable_polarity_correction = false;
1130                 hw->phy.ms_type = e1000_ms_hw_default;
1131         }
1132
1133         if (igb_check_reset_block(hw))
1134                 dev_info(&pdev->dev,
1135                         "PHY reset is blocked due to SOL/IDER session.\n");
1136
1137         netdev->features = NETIF_F_SG |
1138                            NETIF_F_HW_CSUM |
1139                            NETIF_F_HW_VLAN_TX |
1140                            NETIF_F_HW_VLAN_RX |
1141                            NETIF_F_HW_VLAN_FILTER;
1142
1143         netdev->features |= NETIF_F_TSO;
1144         netdev->features |= NETIF_F_TSO6;
1145
1146 #ifdef CONFIG_IGB_LRO
1147         netdev->features |= NETIF_F_LRO;
1148 #endif
1149
1150         netdev->vlan_features |= NETIF_F_TSO;
1151         netdev->vlan_features |= NETIF_F_TSO6;
1152         netdev->vlan_features |= NETIF_F_HW_CSUM;
1153         netdev->vlan_features |= NETIF_F_SG;
1154
1155         if (pci_using_dac)
1156                 netdev->features |= NETIF_F_HIGHDMA;
1157
1158         netdev->features |= NETIF_F_MULTI_QUEUE;
1159
1160         netdev->features |= NETIF_F_LLTX;
1161         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1162
1163         /* before reading the NVM, reset the controller to put the device in a
1164          * known good starting state */
1165         hw->mac.ops.reset_hw(hw);
1166
1167         /* make sure the NVM is good */
1168         if (igb_validate_nvm_checksum(hw) < 0) {
1169                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1170                 err = -EIO;
1171                 goto err_eeprom;
1172         }
1173
1174         /* copy the MAC address out of the NVM */
1175         if (hw->mac.ops.read_mac_addr(hw))
1176                 dev_err(&pdev->dev, "NVM Read Error\n");
1177
1178         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1179         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1180
1181         if (!is_valid_ether_addr(netdev->perm_addr)) {
1182                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1183                 err = -EIO;
1184                 goto err_eeprom;
1185         }
1186
1187         init_timer(&adapter->watchdog_timer);
1188         adapter->watchdog_timer.function = &igb_watchdog;
1189         adapter->watchdog_timer.data = (unsigned long) adapter;
1190
1191         init_timer(&adapter->phy_info_timer);
1192         adapter->phy_info_timer.function = &igb_update_phy_info;
1193         adapter->phy_info_timer.data = (unsigned long) adapter;
1194
1195         INIT_WORK(&adapter->reset_task, igb_reset_task);
1196         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1197
1198         /* Initialize link & ring properties that are user-changeable */
1199         adapter->tx_ring->count = 256;
1200         for (i = 0; i < adapter->num_tx_queues; i++)
1201                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1202         adapter->rx_ring->count = 256;
1203         for (i = 0; i < adapter->num_rx_queues; i++)
1204                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1205
1206         adapter->fc_autoneg = true;
1207         hw->mac.autoneg = true;
1208         hw->phy.autoneg_advertised = 0x2f;
1209
1210         hw->fc.original_type = e1000_fc_default;
1211         hw->fc.type = e1000_fc_default;
1212
1213         adapter->itr_setting = 3;
1214         adapter->itr = IGB_START_ITR;
1215
1216         igb_validate_mdi_setting(hw);
1217
1218         adapter->rx_csum = 1;
1219
1220         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1221          * enable the ACPI Magic Packet filter
1222          */
1223
1224         if (hw->bus.func == 0 ||
1225             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1226                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1227                                      &eeprom_data);
1228
1229         if (eeprom_data & eeprom_apme_mask)
1230                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1231
1232         /* now that we have the eeprom settings, apply the special cases where
1233          * the eeprom may be wrong or the board simply won't support wake on
1234          * lan on a particular port */
1235         switch (pdev->device) {
1236         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1237                 adapter->eeprom_wol = 0;
1238                 break;
1239         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1240         case E1000_DEV_ID_82576_FIBER:
1241         case E1000_DEV_ID_82576_SERDES:
1242                 /* Wake events only supported on port A for dual fiber
1243                  * regardless of eeprom setting */
1244                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1245                         adapter->eeprom_wol = 0;
1246                 break;
1247         case E1000_DEV_ID_82576_QUAD_COPPER:
1248                 /* if quad port adapter, disable WoL on all but port A */
1249                 if (global_quad_port_a != 0)
1250                         adapter->eeprom_wol = 0;
1251                 else
1252                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1253                 /* Reset for multiple quad port adapters */
1254                 if (++global_quad_port_a == 4)
1255                         global_quad_port_a = 0;
1256                 break;
1257         }
1258
1259         /* initialize the wol settings based on the eeprom settings */
1260         adapter->wol = adapter->eeprom_wol;
1261
1262         /* reset the hardware with the new settings */
1263         igb_reset(adapter);
1264
1265         /* let the f/w know that the h/w is now under the control of the
1266          * driver. */
1267         igb_get_hw_control(adapter);
1268
1269         /* tell the stack to leave us alone until igb_open() is called */
1270         netif_carrier_off(netdev);
1271         netif_stop_queue(netdev);
1272         for (i = 0; i < adapter->num_tx_queues; i++)
1273                 netif_stop_subqueue(netdev, i);
1274
1275         strcpy(netdev->name, "eth%d");
1276         err = register_netdev(netdev);
1277         if (err)
1278                 goto err_register;
1279
1280 #ifdef CONFIG_DCA
1281         if ((adapter->flags & IGB_FLAG_HAS_DCA) &&
1282             (dca_add_requester(&pdev->dev) == 0)) {
1283                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1284                 dev_info(&pdev->dev, "DCA enabled\n");
1285                 /* Always use CB2 mode, difference is masked
1286                  * in the CB driver. */
1287                 wr32(E1000_DCA_CTRL, 2);
1288                 igb_setup_dca(adapter);
1289         }
1290 #endif
1291
1292         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1293         /* print bus type/speed/width info */
1294         dev_info(&pdev->dev,
1295                  "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1296                  netdev->name,
1297                  ((hw->bus.speed == e1000_bus_speed_2500)
1298                   ? "2.5Gb/s" : "unknown"),
1299                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1300                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1301                   ? "Width x1" : "unknown"),
1302                  netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1303                  netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1304
1305         igb_read_part_num(hw, &part_num);
1306         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1307                 (part_num >> 8), (part_num & 0xff));
1308
1309         dev_info(&pdev->dev,
1310                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1311                 adapter->msix_entries ? "MSI-X" :
1312                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1313                 adapter->num_rx_queues, adapter->num_tx_queues);
1314
1315         return 0;
1316
1317 err_register:
1318         igb_release_hw_control(adapter);
1319 err_eeprom:
1320         if (!igb_check_reset_block(hw))
1321                 hw->phy.ops.reset_phy(hw);
1322
1323         if (hw->flash_address)
1324                 iounmap(hw->flash_address);
1325
1326         igb_remove_device(hw);
1327         igb_free_queues(adapter);
1328 err_sw_init:
1329 err_hw_init:
1330         iounmap(hw->hw_addr);
1331 err_ioremap:
1332         free_netdev(netdev);
1333 err_alloc_etherdev:
1334         pci_release_selected_regions(pdev, bars);
1335 err_pci_reg:
1336 err_dma:
1337         pci_disable_device(pdev);
1338         return err;
1339 }
1340
1341 /**
1342  * igb_remove - Device Removal Routine
1343  * @pdev: PCI device information struct
1344  *
1345  * igb_remove is called by the PCI subsystem to alert the driver
1346  * that it should release a PCI device.  The could be caused by a
1347  * Hot-Plug event, or because the driver is going to be removed from
1348  * memory.
1349  **/
1350 static void __devexit igb_remove(struct pci_dev *pdev)
1351 {
1352         struct net_device *netdev = pci_get_drvdata(pdev);
1353         struct igb_adapter *adapter = netdev_priv(netdev);
1354 #ifdef CONFIG_DCA
1355         struct e1000_hw *hw = &adapter->hw;
1356 #endif
1357
1358         /* flush_scheduled work may reschedule our watchdog task, so
1359          * explicitly disable watchdog tasks from being rescheduled  */
1360         set_bit(__IGB_DOWN, &adapter->state);
1361         del_timer_sync(&adapter->watchdog_timer);
1362         del_timer_sync(&adapter->phy_info_timer);
1363
1364         flush_scheduled_work();
1365
1366 #ifdef CONFIG_DCA
1367         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1368                 dev_info(&pdev->dev, "DCA disabled\n");
1369                 dca_remove_requester(&pdev->dev);
1370                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1371                 wr32(E1000_DCA_CTRL, 1);
1372         }
1373 #endif
1374
1375         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1376          * would have already happened in close and is redundant. */
1377         igb_release_hw_control(adapter);
1378
1379         unregister_netdev(netdev);
1380
1381         if (!igb_check_reset_block(&adapter->hw))
1382                 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1383
1384         igb_remove_device(&adapter->hw);
1385         igb_reset_interrupt_capability(adapter);
1386
1387         igb_free_queues(adapter);
1388
1389         iounmap(adapter->hw.hw_addr);
1390         if (adapter->hw.flash_address)
1391                 iounmap(adapter->hw.flash_address);
1392         pci_release_selected_regions(pdev, adapter->bars);
1393
1394         free_netdev(netdev);
1395
1396         pci_disable_device(pdev);
1397 }
1398
1399 /**
1400  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1401  * @adapter: board private structure to initialize
1402  *
1403  * igb_sw_init initializes the Adapter private data structure.
1404  * Fields are initialized based on PCI device information and
1405  * OS network device settings (MTU size).
1406  **/
1407 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1408 {
1409         struct e1000_hw *hw = &adapter->hw;
1410         struct net_device *netdev = adapter->netdev;
1411         struct pci_dev *pdev = adapter->pdev;
1412
1413         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1414
1415         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1416         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1417         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1418         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1419
1420         /* Number of supported queues. */
1421         /* Having more queues than CPUs doesn't make sense. */
1422         adapter->num_rx_queues = min((u32)IGB_MAX_RX_QUEUES, (u32)num_online_cpus());
1423         adapter->num_tx_queues = min(IGB_MAX_TX_QUEUES, num_online_cpus());
1424
1425         /* This call may decrease the number of queues depending on
1426          * interrupt mode. */
1427         igb_set_interrupt_capability(adapter);
1428
1429         if (igb_alloc_queues(adapter)) {
1430                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1431                 return -ENOMEM;
1432         }
1433
1434         /* Explicitly disable IRQ since the NIC can be in any state. */
1435         igb_irq_disable(adapter);
1436
1437         set_bit(__IGB_DOWN, &adapter->state);
1438         return 0;
1439 }
1440
1441 /**
1442  * igb_open - Called when a network interface is made active
1443  * @netdev: network interface device structure
1444  *
1445  * Returns 0 on success, negative value on failure
1446  *
1447  * The open entry point is called when a network interface is made
1448  * active by the system (IFF_UP).  At this point all resources needed
1449  * for transmit and receive operations are allocated, the interrupt
1450  * handler is registered with the OS, the watchdog timer is started,
1451  * and the stack is notified that the interface is ready.
1452  **/
1453 static int igb_open(struct net_device *netdev)
1454 {
1455         struct igb_adapter *adapter = netdev_priv(netdev);
1456         struct e1000_hw *hw = &adapter->hw;
1457         int err;
1458         int i;
1459
1460         /* disallow open during test */
1461         if (test_bit(__IGB_TESTING, &adapter->state))
1462                 return -EBUSY;
1463
1464         /* allocate transmit descriptors */
1465         err = igb_setup_all_tx_resources(adapter);
1466         if (err)
1467                 goto err_setup_tx;
1468
1469         /* allocate receive descriptors */
1470         err = igb_setup_all_rx_resources(adapter);
1471         if (err)
1472                 goto err_setup_rx;
1473
1474         /* e1000_power_up_phy(adapter); */
1475
1476         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1477         if ((adapter->hw.mng_cookie.status &
1478              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1479                 igb_update_mng_vlan(adapter);
1480
1481         /* before we allocate an interrupt, we must be ready to handle it.
1482          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1483          * as soon as we call pci_request_irq, so we have to setup our
1484          * clean_rx handler before we do so.  */
1485         igb_configure(adapter);
1486
1487         err = igb_request_irq(adapter);
1488         if (err)
1489                 goto err_req_irq;
1490
1491         /* From here on the code is the same as igb_up() */
1492         clear_bit(__IGB_DOWN, &adapter->state);
1493
1494         for (i = 0; i < adapter->num_rx_queues; i++)
1495                 napi_enable(&adapter->rx_ring[i].napi);
1496
1497         /* Clear any pending interrupts. */
1498         rd32(E1000_ICR);
1499
1500         igb_irq_enable(adapter);
1501
1502         /* Fire a link status change interrupt to start the watchdog. */
1503         wr32(E1000_ICS, E1000_ICS_LSC);
1504
1505         return 0;
1506
1507 err_req_irq:
1508         igb_release_hw_control(adapter);
1509         /* e1000_power_down_phy(adapter); */
1510         igb_free_all_rx_resources(adapter);
1511 err_setup_rx:
1512         igb_free_all_tx_resources(adapter);
1513 err_setup_tx:
1514         igb_reset(adapter);
1515
1516         return err;
1517 }
1518
1519 /**
1520  * igb_close - Disables a network interface
1521  * @netdev: network interface device structure
1522  *
1523  * Returns 0, this is not allowed to fail
1524  *
1525  * The close entry point is called when an interface is de-activated
1526  * by the OS.  The hardware is still under the driver's control, but
1527  * needs to be disabled.  A global MAC reset is issued to stop the
1528  * hardware, and all transmit and receive resources are freed.
1529  **/
1530 static int igb_close(struct net_device *netdev)
1531 {
1532         struct igb_adapter *adapter = netdev_priv(netdev);
1533
1534         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1535         igb_down(adapter);
1536
1537         igb_free_irq(adapter);
1538
1539         igb_free_all_tx_resources(adapter);
1540         igb_free_all_rx_resources(adapter);
1541
1542         /* kill manageability vlan ID if supported, but not if a vlan with
1543          * the same ID is registered on the host OS (let 8021q kill it) */
1544         if ((adapter->hw.mng_cookie.status &
1545                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1546              !(adapter->vlgrp &&
1547                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1548                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1549
1550         return 0;
1551 }
1552
1553 /**
1554  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1555  * @adapter: board private structure
1556  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1557  *
1558  * Return 0 on success, negative on failure
1559  **/
1560
1561 int igb_setup_tx_resources(struct igb_adapter *adapter,
1562                            struct igb_ring *tx_ring)
1563 {
1564         struct pci_dev *pdev = adapter->pdev;
1565         int size;
1566
1567         size = sizeof(struct igb_buffer) * tx_ring->count;
1568         tx_ring->buffer_info = vmalloc(size);
1569         if (!tx_ring->buffer_info)
1570                 goto err;
1571         memset(tx_ring->buffer_info, 0, size);
1572
1573         /* round up to nearest 4K */
1574         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1575                         + sizeof(u32);
1576         tx_ring->size = ALIGN(tx_ring->size, 4096);
1577
1578         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1579                                              &tx_ring->dma);
1580
1581         if (!tx_ring->desc)
1582                 goto err;
1583
1584         tx_ring->adapter = adapter;
1585         tx_ring->next_to_use = 0;
1586         tx_ring->next_to_clean = 0;
1587         return 0;
1588
1589 err:
1590         vfree(tx_ring->buffer_info);
1591         dev_err(&adapter->pdev->dev,
1592                 "Unable to allocate memory for the transmit descriptor ring\n");
1593         return -ENOMEM;
1594 }
1595
1596 /**
1597  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1598  *                                (Descriptors) for all queues
1599  * @adapter: board private structure
1600  *
1601  * Return 0 on success, negative on failure
1602  **/
1603 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1604 {
1605         int i, err = 0;
1606         int r_idx;
1607
1608         for (i = 0; i < adapter->num_tx_queues; i++) {
1609                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1610                 if (err) {
1611                         dev_err(&adapter->pdev->dev,
1612                                 "Allocation for Tx Queue %u failed\n", i);
1613                         for (i--; i >= 0; i--)
1614                                 igb_free_tx_resources(&adapter->tx_ring[i]);
1615                         break;
1616                 }
1617         }
1618
1619         for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1620                 r_idx = i % adapter->num_tx_queues;
1621                 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1622         }
1623         return err;
1624 }
1625
1626 /**
1627  * igb_configure_tx - Configure transmit Unit after Reset
1628  * @adapter: board private structure
1629  *
1630  * Configure the Tx unit of the MAC after a reset.
1631  **/
1632 static void igb_configure_tx(struct igb_adapter *adapter)
1633 {
1634         u64 tdba, tdwba;
1635         struct e1000_hw *hw = &adapter->hw;
1636         u32 tctl;
1637         u32 txdctl, txctrl;
1638         int i;
1639
1640         for (i = 0; i < adapter->num_tx_queues; i++) {
1641                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1642
1643                 wr32(E1000_TDLEN(i),
1644                                 ring->count * sizeof(struct e1000_tx_desc));
1645                 tdba = ring->dma;
1646                 wr32(E1000_TDBAL(i),
1647                                 tdba & 0x00000000ffffffffULL);
1648                 wr32(E1000_TDBAH(i), tdba >> 32);
1649
1650                 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1651                 tdwba |= 1; /* enable head wb */
1652                 wr32(E1000_TDWBAL(i),
1653                                 tdwba & 0x00000000ffffffffULL);
1654                 wr32(E1000_TDWBAH(i), tdwba >> 32);
1655
1656                 ring->head = E1000_TDH(i);
1657                 ring->tail = E1000_TDT(i);
1658                 writel(0, hw->hw_addr + ring->tail);
1659                 writel(0, hw->hw_addr + ring->head);
1660                 txdctl = rd32(E1000_TXDCTL(i));
1661                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1662                 wr32(E1000_TXDCTL(i), txdctl);
1663
1664                 /* Turn off Relaxed Ordering on head write-backs.  The
1665                  * writebacks MUST be delivered in order or it will
1666                  * completely screw up our bookeeping.
1667                  */
1668                 txctrl = rd32(E1000_DCA_TXCTRL(i));
1669                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1670                 wr32(E1000_DCA_TXCTRL(i), txctrl);
1671         }
1672
1673
1674
1675         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1676
1677         /* Program the Transmit Control Register */
1678
1679         tctl = rd32(E1000_TCTL);
1680         tctl &= ~E1000_TCTL_CT;
1681         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1682                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1683
1684         igb_config_collision_dist(hw);
1685
1686         /* Setup Transmit Descriptor Settings for eop descriptor */
1687         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1688
1689         /* Enable transmits */
1690         tctl |= E1000_TCTL_EN;
1691
1692         wr32(E1000_TCTL, tctl);
1693 }
1694
1695 /**
1696  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1697  * @adapter: board private structure
1698  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1699  *
1700  * Returns 0 on success, negative on failure
1701  **/
1702
1703 int igb_setup_rx_resources(struct igb_adapter *adapter,
1704                            struct igb_ring *rx_ring)
1705 {
1706         struct pci_dev *pdev = adapter->pdev;
1707         int size, desc_len;
1708
1709 #ifdef CONFIG_IGB_LRO
1710         size = sizeof(struct net_lro_desc) * MAX_LRO_DESCRIPTORS;
1711         rx_ring->lro_mgr.lro_arr = vmalloc(size);
1712         if (!rx_ring->lro_mgr.lro_arr)
1713                 goto err;
1714         memset(rx_ring->lro_mgr.lro_arr, 0, size);
1715 #endif
1716
1717         size = sizeof(struct igb_buffer) * rx_ring->count;
1718         rx_ring->buffer_info = vmalloc(size);
1719         if (!rx_ring->buffer_info)
1720                 goto err;
1721         memset(rx_ring->buffer_info, 0, size);
1722
1723         desc_len = sizeof(union e1000_adv_rx_desc);
1724
1725         /* Round up to nearest 4K */
1726         rx_ring->size = rx_ring->count * desc_len;
1727         rx_ring->size = ALIGN(rx_ring->size, 4096);
1728
1729         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1730                                              &rx_ring->dma);
1731
1732         if (!rx_ring->desc)
1733                 goto err;
1734
1735         rx_ring->next_to_clean = 0;
1736         rx_ring->next_to_use = 0;
1737
1738         rx_ring->adapter = adapter;
1739
1740         return 0;
1741
1742 err:
1743 #ifdef CONFIG_IGB_LRO
1744         vfree(rx_ring->lro_mgr.lro_arr);
1745         rx_ring->lro_mgr.lro_arr = NULL;
1746 #endif
1747         vfree(rx_ring->buffer_info);
1748         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1749                 "the receive descriptor ring\n");
1750         return -ENOMEM;
1751 }
1752
1753 /**
1754  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1755  *                                (Descriptors) for all queues
1756  * @adapter: board private structure
1757  *
1758  * Return 0 on success, negative on failure
1759  **/
1760 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1761 {
1762         int i, err = 0;
1763
1764         for (i = 0; i < adapter->num_rx_queues; i++) {
1765                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1766                 if (err) {
1767                         dev_err(&adapter->pdev->dev,
1768                                 "Allocation for Rx Queue %u failed\n", i);
1769                         for (i--; i >= 0; i--)
1770                                 igb_free_rx_resources(&adapter->rx_ring[i]);
1771                         break;
1772                 }
1773         }
1774
1775         return err;
1776 }
1777
1778 /**
1779  * igb_setup_rctl - configure the receive control registers
1780  * @adapter: Board private structure
1781  **/
1782 static void igb_setup_rctl(struct igb_adapter *adapter)
1783 {
1784         struct e1000_hw *hw = &adapter->hw;
1785         u32 rctl;
1786         u32 srrctl = 0;
1787         int i;
1788
1789         rctl = rd32(E1000_RCTL);
1790
1791         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1792
1793         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1794                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1795                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1796
1797         /*
1798          * enable stripping of CRC. It's unlikely this will break BMC
1799          * redirection as it did with e1000. Newer features require
1800          * that the HW strips the CRC.
1801         */
1802         rctl |= E1000_RCTL_SECRC;
1803
1804         rctl &= ~E1000_RCTL_SBP;
1805
1806         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1807                 rctl &= ~E1000_RCTL_LPE;
1808         else
1809                 rctl |= E1000_RCTL_LPE;
1810         if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1811                 /* Setup buffer sizes */
1812                 rctl &= ~E1000_RCTL_SZ_4096;
1813                 rctl |= E1000_RCTL_BSEX;
1814                 switch (adapter->rx_buffer_len) {
1815                 case IGB_RXBUFFER_256:
1816                         rctl |= E1000_RCTL_SZ_256;
1817                         rctl &= ~E1000_RCTL_BSEX;
1818                         break;
1819                 case IGB_RXBUFFER_512:
1820                         rctl |= E1000_RCTL_SZ_512;
1821                         rctl &= ~E1000_RCTL_BSEX;
1822                         break;
1823                 case IGB_RXBUFFER_1024:
1824                         rctl |= E1000_RCTL_SZ_1024;
1825                         rctl &= ~E1000_RCTL_BSEX;
1826                         break;
1827                 case IGB_RXBUFFER_2048:
1828                 default:
1829                         rctl |= E1000_RCTL_SZ_2048;
1830                         rctl &= ~E1000_RCTL_BSEX;
1831                         break;
1832                 }
1833         } else {
1834                 rctl &= ~E1000_RCTL_BSEX;
1835                 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1836         }
1837
1838         /* 82575 and greater support packet-split where the protocol
1839          * header is placed in skb->data and the packet data is
1840          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1841          * In the case of a non-split, skb->data is linearly filled,
1842          * followed by the page buffers.  Therefore, skb->data is
1843          * sized to hold the largest protocol header.
1844          */
1845         /* allocations using alloc_page take too long for regular MTU
1846          * so only enable packet split for jumbo frames */
1847         if (rctl & E1000_RCTL_LPE) {
1848                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1849                 srrctl |= adapter->rx_ps_hdr_size <<
1850                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1851                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1852         } else {
1853                 adapter->rx_ps_hdr_size = 0;
1854                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1855         }
1856
1857         for (i = 0; i < adapter->num_rx_queues; i++)
1858                 wr32(E1000_SRRCTL(i), srrctl);
1859
1860         wr32(E1000_RCTL, rctl);
1861 }
1862
1863 /**
1864  * igb_configure_rx - Configure receive Unit after Reset
1865  * @adapter: board private structure
1866  *
1867  * Configure the Rx unit of the MAC after a reset.
1868  **/
1869 static void igb_configure_rx(struct igb_adapter *adapter)
1870 {
1871         u64 rdba;
1872         struct e1000_hw *hw = &adapter->hw;
1873         u32 rctl, rxcsum;
1874         u32 rxdctl;
1875         int i;
1876
1877         /* disable receives while setting up the descriptors */
1878         rctl = rd32(E1000_RCTL);
1879         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1880         wrfl();
1881         mdelay(10);
1882
1883         if (adapter->itr_setting > 3)
1884                 wr32(E1000_ITR, adapter->itr);
1885
1886         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1887          * the Base and Length of the Rx Descriptor Ring */
1888         for (i = 0; i < adapter->num_rx_queues; i++) {
1889                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1890                 rdba = ring->dma;
1891                 wr32(E1000_RDBAL(i),
1892                                 rdba & 0x00000000ffffffffULL);
1893                 wr32(E1000_RDBAH(i), rdba >> 32);
1894                 wr32(E1000_RDLEN(i),
1895                                ring->count * sizeof(union e1000_adv_rx_desc));
1896
1897                 ring->head = E1000_RDH(i);
1898                 ring->tail = E1000_RDT(i);
1899                 writel(0, hw->hw_addr + ring->tail);
1900                 writel(0, hw->hw_addr + ring->head);
1901
1902                 rxdctl = rd32(E1000_RXDCTL(i));
1903                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1904                 rxdctl &= 0xFFF00000;
1905                 rxdctl |= IGB_RX_PTHRESH;
1906                 rxdctl |= IGB_RX_HTHRESH << 8;
1907                 rxdctl |= IGB_RX_WTHRESH << 16;
1908                 wr32(E1000_RXDCTL(i), rxdctl);
1909 #ifdef CONFIG_IGB_LRO
1910                 /* Intitial LRO Settings */
1911                 ring->lro_mgr.max_aggr = MAX_LRO_AGGR;
1912                 ring->lro_mgr.max_desc = MAX_LRO_DESCRIPTORS;
1913                 ring->lro_mgr.get_skb_header = igb_get_skb_hdr;
1914                 ring->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1915                 ring->lro_mgr.dev = adapter->netdev;
1916                 ring->lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1917                 ring->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1918 #endif
1919         }
1920
1921         if (adapter->num_rx_queues > 1) {
1922                 u32 random[10];
1923                 u32 mrqc;
1924                 u32 j, shift;
1925                 union e1000_reta {
1926                         u32 dword;
1927                         u8  bytes[4];
1928                 } reta;
1929
1930                 get_random_bytes(&random[0], 40);
1931
1932                 if (hw->mac.type >= e1000_82576)
1933                         shift = 0;
1934                 else
1935                         shift = 6;
1936                 for (j = 0; j < (32 * 4); j++) {
1937                         reta.bytes[j & 3] =
1938                                 (j % adapter->num_rx_queues) << shift;
1939                         if ((j & 3) == 3)
1940                                 writel(reta.dword,
1941                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1942                 }
1943                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1944
1945                 /* Fill out hash function seeds */
1946                 for (j = 0; j < 10; j++)
1947                         array_wr32(E1000_RSSRK(0), j, random[j]);
1948
1949                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1950                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1951                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1952                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1953                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1954                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1955                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1956                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1957
1958
1959                 wr32(E1000_MRQC, mrqc);
1960
1961                 /* Multiqueue and raw packet checksumming are mutually
1962                  * exclusive.  Note that this not the same as TCP/IP
1963                  * checksumming, which works fine. */
1964                 rxcsum = rd32(E1000_RXCSUM);
1965                 rxcsum |= E1000_RXCSUM_PCSD;
1966                 wr32(E1000_RXCSUM, rxcsum);
1967         } else {
1968                 /* Enable Receive Checksum Offload for TCP and UDP */
1969                 rxcsum = rd32(E1000_RXCSUM);
1970                 if (adapter->rx_csum) {
1971                         rxcsum |= E1000_RXCSUM_TUOFL;
1972
1973                         /* Enable IPv4 payload checksum for UDP fragments
1974                          * Must be used in conjunction with packet-split. */
1975                         if (adapter->rx_ps_hdr_size)
1976                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1977                 } else {
1978                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1979                         /* don't need to clear IPPCSE as it defaults to 0 */
1980                 }
1981                 wr32(E1000_RXCSUM, rxcsum);
1982         }
1983
1984         if (adapter->vlgrp)
1985                 wr32(E1000_RLPML,
1986                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1987         else
1988                 wr32(E1000_RLPML, adapter->max_frame_size);
1989
1990         /* Enable Receives */
1991         wr32(E1000_RCTL, rctl);
1992 }
1993
1994 /**
1995  * igb_free_tx_resources - Free Tx Resources per Queue
1996  * @adapter: board private structure
1997  * @tx_ring: Tx descriptor ring for a specific queue
1998  *
1999  * Free all transmit software resources
2000  **/
2001 static void igb_free_tx_resources(struct igb_ring *tx_ring)
2002 {
2003         struct pci_dev *pdev = tx_ring->adapter->pdev;
2004
2005         igb_clean_tx_ring(tx_ring);
2006
2007         vfree(tx_ring->buffer_info);
2008         tx_ring->buffer_info = NULL;
2009
2010         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2011
2012         tx_ring->desc = NULL;
2013 }
2014
2015 /**
2016  * igb_free_all_tx_resources - Free Tx Resources for All Queues
2017  * @adapter: board private structure
2018  *
2019  * Free all transmit software resources
2020  **/
2021 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2022 {
2023         int i;
2024
2025         for (i = 0; i < adapter->num_tx_queues; i++)
2026                 igb_free_tx_resources(&adapter->tx_ring[i]);
2027 }
2028
2029 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2030                                            struct igb_buffer *buffer_info)
2031 {
2032         if (buffer_info->dma) {
2033                 pci_unmap_page(adapter->pdev,
2034                                 buffer_info->dma,
2035                                 buffer_info->length,
2036                                 PCI_DMA_TODEVICE);
2037                 buffer_info->dma = 0;
2038         }
2039         if (buffer_info->skb) {
2040                 dev_kfree_skb_any(buffer_info->skb);
2041                 buffer_info->skb = NULL;
2042         }
2043         buffer_info->time_stamp = 0;
2044         /* buffer_info must be completely set up in the transmit path */
2045 }
2046
2047 /**
2048  * igb_clean_tx_ring - Free Tx Buffers
2049  * @adapter: board private structure
2050  * @tx_ring: ring to be cleaned
2051  **/
2052 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2053 {
2054         struct igb_adapter *adapter = tx_ring->adapter;
2055         struct igb_buffer *buffer_info;
2056         unsigned long size;
2057         unsigned int i;
2058
2059         if (!tx_ring->buffer_info)
2060                 return;
2061         /* Free all the Tx ring sk_buffs */
2062
2063         for (i = 0; i < tx_ring->count; i++) {
2064                 buffer_info = &tx_ring->buffer_info[i];
2065                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2066         }
2067
2068         size = sizeof(struct igb_buffer) * tx_ring->count;
2069         memset(tx_ring->buffer_info, 0, size);
2070
2071         /* Zero out the descriptor ring */
2072
2073         memset(tx_ring->desc, 0, tx_ring->size);
2074
2075         tx_ring->next_to_use = 0;
2076         tx_ring->next_to_clean = 0;
2077
2078         writel(0, adapter->hw.hw_addr + tx_ring->head);
2079         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2080 }
2081
2082 /**
2083  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2084  * @adapter: board private structure
2085  **/
2086 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2087 {
2088         int i;
2089
2090         for (i = 0; i < adapter->num_tx_queues; i++)
2091                 igb_clean_tx_ring(&adapter->tx_ring[i]);
2092 }
2093
2094 /**
2095  * igb_free_rx_resources - Free Rx Resources
2096  * @adapter: board private structure
2097  * @rx_ring: ring to clean the resources from
2098  *
2099  * Free all receive software resources
2100  **/
2101 static void igb_free_rx_resources(struct igb_ring *rx_ring)
2102 {
2103         struct pci_dev *pdev = rx_ring->adapter->pdev;
2104
2105         igb_clean_rx_ring(rx_ring);
2106
2107         vfree(rx_ring->buffer_info);
2108         rx_ring->buffer_info = NULL;
2109
2110 #ifdef CONFIG_IGB_LRO
2111         vfree(rx_ring->lro_mgr.lro_arr);
2112         rx_ring->lro_mgr.lro_arr = NULL;
2113 #endif
2114
2115         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2116
2117         rx_ring->desc = NULL;
2118 }
2119
2120 /**
2121  * igb_free_all_rx_resources - Free Rx Resources for All Queues
2122  * @adapter: board private structure
2123  *
2124  * Free all receive software resources
2125  **/
2126 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2127 {
2128         int i;
2129
2130         for (i = 0; i < adapter->num_rx_queues; i++)
2131                 igb_free_rx_resources(&adapter->rx_ring[i]);
2132 }
2133
2134 /**
2135  * igb_clean_rx_ring - Free Rx Buffers per Queue
2136  * @adapter: board private structure
2137  * @rx_ring: ring to free buffers from
2138  **/
2139 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2140 {
2141         struct igb_adapter *adapter = rx_ring->adapter;
2142         struct igb_buffer *buffer_info;
2143         struct pci_dev *pdev = adapter->pdev;
2144         unsigned long size;
2145         unsigned int i;
2146
2147         if (!rx_ring->buffer_info)
2148                 return;
2149         /* Free all the Rx ring sk_buffs */
2150         for (i = 0; i < rx_ring->count; i++) {
2151                 buffer_info = &rx_ring->buffer_info[i];
2152                 if (buffer_info->dma) {
2153                         if (adapter->rx_ps_hdr_size)
2154                                 pci_unmap_single(pdev, buffer_info->dma,
2155                                                  adapter->rx_ps_hdr_size,
2156                                                  PCI_DMA_FROMDEVICE);
2157                         else
2158                                 pci_unmap_single(pdev, buffer_info->dma,
2159                                                  adapter->rx_buffer_len,
2160                                                  PCI_DMA_FROMDEVICE);
2161                         buffer_info->dma = 0;
2162                 }
2163
2164                 if (buffer_info->skb) {
2165                         dev_kfree_skb(buffer_info->skb);
2166                         buffer_info->skb = NULL;
2167                 }
2168                 if (buffer_info->page) {
2169                         if (buffer_info->page_dma)
2170                                 pci_unmap_page(pdev, buffer_info->page_dma,
2171                                                PAGE_SIZE / 2,
2172                                                PCI_DMA_FROMDEVICE);
2173                         put_page(buffer_info->page);
2174                         buffer_info->page = NULL;
2175                         buffer_info->page_dma = 0;
2176                         buffer_info->page_offset = 0;
2177                 }
2178         }
2179
2180         size = sizeof(struct igb_buffer) * rx_ring->count;
2181         memset(rx_ring->buffer_info, 0, size);
2182
2183         /* Zero out the descriptor ring */
2184         memset(rx_ring->desc, 0, rx_ring->size);
2185
2186         rx_ring->next_to_clean = 0;
2187         rx_ring->next_to_use = 0;
2188
2189         writel(0, adapter->hw.hw_addr + rx_ring->head);
2190         writel(0, adapter->hw.hw_addr + rx_ring->tail);
2191 }
2192
2193 /**
2194  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2195  * @adapter: board private structure
2196  **/
2197 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2198 {
2199         int i;
2200
2201         for (i = 0; i < adapter->num_rx_queues; i++)
2202                 igb_clean_rx_ring(&adapter->rx_ring[i]);
2203 }
2204
2205 /**
2206  * igb_set_mac - Change the Ethernet Address of the NIC
2207  * @netdev: network interface device structure
2208  * @p: pointer to an address structure
2209  *
2210  * Returns 0 on success, negative on failure
2211  **/
2212 static int igb_set_mac(struct net_device *netdev, void *p)
2213 {
2214         struct igb_adapter *adapter = netdev_priv(netdev);
2215         struct sockaddr *addr = p;
2216
2217         if (!is_valid_ether_addr(addr->sa_data))
2218                 return -EADDRNOTAVAIL;
2219
2220         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2221         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2222
2223         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2224
2225         return 0;
2226 }
2227
2228 /**
2229  * igb_set_multi - Multicast and Promiscuous mode set
2230  * @netdev: network interface device structure
2231  *
2232  * The set_multi entry point is called whenever the multicast address
2233  * list or the network interface flags are updated.  This routine is
2234  * responsible for configuring the hardware for proper multicast,
2235  * promiscuous mode, and all-multi behavior.
2236  **/
2237 static void igb_set_multi(struct net_device *netdev)
2238 {
2239         struct igb_adapter *adapter = netdev_priv(netdev);
2240         struct e1000_hw *hw = &adapter->hw;
2241         struct e1000_mac_info *mac = &hw->mac;
2242         struct dev_mc_list *mc_ptr;
2243         u8  *mta_list;
2244         u32 rctl;
2245         int i;
2246
2247         /* Check for Promiscuous and All Multicast modes */
2248
2249         rctl = rd32(E1000_RCTL);
2250
2251         if (netdev->flags & IFF_PROMISC) {
2252                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2253                 rctl &= ~E1000_RCTL_VFE;
2254         } else {
2255                 if (netdev->flags & IFF_ALLMULTI) {
2256                         rctl |= E1000_RCTL_MPE;
2257                         rctl &= ~E1000_RCTL_UPE;
2258                 } else
2259                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2260                 rctl |= E1000_RCTL_VFE;
2261         }
2262         wr32(E1000_RCTL, rctl);
2263
2264         if (!netdev->mc_count) {
2265                 /* nothing to program, so clear mc list */
2266                 igb_update_mc_addr_list_82575(hw, NULL, 0, 1,
2267                                           mac->rar_entry_count);
2268                 return;
2269         }
2270
2271         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2272         if (!mta_list)
2273                 return;
2274
2275         /* The shared function expects a packed array of only addresses. */
2276         mc_ptr = netdev->mc_list;
2277
2278         for (i = 0; i < netdev->mc_count; i++) {
2279                 if (!mc_ptr)
2280                         break;
2281                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2282                 mc_ptr = mc_ptr->next;
2283         }
2284         igb_update_mc_addr_list_82575(hw, mta_list, i, 1,
2285                                       mac->rar_entry_count);
2286         kfree(mta_list);
2287 }
2288
2289 /* Need to wait a few seconds after link up to get diagnostic information from
2290  * the phy */
2291 static void igb_update_phy_info(unsigned long data)
2292 {
2293         struct igb_adapter *adapter = (struct igb_adapter *) data;
2294         if (adapter->hw.phy.ops.get_phy_info)
2295                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2296 }
2297
2298 /**
2299  * igb_watchdog - Timer Call-back
2300  * @data: pointer to adapter cast into an unsigned long
2301  **/
2302 static void igb_watchdog(unsigned long data)
2303 {
2304         struct igb_adapter *adapter = (struct igb_adapter *)data;
2305         /* Do the rest outside of interrupt context */
2306         schedule_work(&adapter->watchdog_task);
2307 }
2308
2309 static void igb_watchdog_task(struct work_struct *work)
2310 {
2311         struct igb_adapter *adapter = container_of(work,
2312                                         struct igb_adapter, watchdog_task);
2313         struct e1000_hw *hw = &adapter->hw;
2314
2315         struct net_device *netdev = adapter->netdev;
2316         struct igb_ring *tx_ring = adapter->tx_ring;
2317         struct e1000_mac_info *mac = &adapter->hw.mac;
2318         u32 link;
2319         s32 ret_val;
2320         int i;
2321
2322         if ((netif_carrier_ok(netdev)) &&
2323             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2324                 goto link_up;
2325
2326         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2327         if ((ret_val == E1000_ERR_PHY) &&
2328             (hw->phy.type == e1000_phy_igp_3) &&
2329             (rd32(E1000_CTRL) &
2330              E1000_PHY_CTRL_GBE_DISABLE))
2331                 dev_info(&adapter->pdev->dev,
2332                          "Gigabit has been disabled, downgrading speed\n");
2333
2334         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2335             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2336                 link = mac->serdes_has_link;
2337         else
2338                 link = rd32(E1000_STATUS) &
2339                                       E1000_STATUS_LU;
2340
2341         if (link) {
2342                 if (!netif_carrier_ok(netdev)) {
2343                         u32 ctrl;
2344                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2345                                                    &adapter->link_speed,
2346                                                    &adapter->link_duplex);
2347
2348                         ctrl = rd32(E1000_CTRL);
2349                         dev_info(&adapter->pdev->dev,
2350                                  "NIC Link is Up %d Mbps %s, "
2351                                  "Flow Control: %s\n",
2352                                  adapter->link_speed,
2353                                  adapter->link_duplex == FULL_DUPLEX ?
2354                                  "Full Duplex" : "Half Duplex",
2355                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2356                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2357                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2358                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2359
2360                         /* tweak tx_queue_len according to speed/duplex and
2361                          * adjust the timeout factor */
2362                         netdev->tx_queue_len = adapter->tx_queue_len;
2363                         adapter->tx_timeout_factor = 1;
2364                         switch (adapter->link_speed) {
2365                         case SPEED_10:
2366                                 netdev->tx_queue_len = 10;
2367                                 adapter->tx_timeout_factor = 14;
2368                                 break;
2369                         case SPEED_100:
2370                                 netdev->tx_queue_len = 100;
2371                                 /* maybe add some timeout factor ? */
2372                                 break;
2373                         }
2374
2375                         netif_carrier_on(netdev);
2376                         netif_wake_queue(netdev);
2377                         for (i = 0; i < adapter->num_tx_queues; i++)
2378                                 netif_wake_subqueue(netdev, i);
2379
2380                         if (!test_bit(__IGB_DOWN, &adapter->state))
2381                                 mod_timer(&adapter->phy_info_timer,
2382                                           round_jiffies(jiffies + 2 * HZ));
2383                 }
2384         } else {
2385                 if (netif_carrier_ok(netdev)) {
2386                         adapter->link_speed = 0;
2387                         adapter->link_duplex = 0;
2388                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2389                         netif_carrier_off(netdev);
2390                         netif_stop_queue(netdev);
2391                         for (i = 0; i < adapter->num_tx_queues; i++)
2392                                 netif_stop_subqueue(netdev, i);
2393                         if (!test_bit(__IGB_DOWN, &adapter->state))
2394                                 mod_timer(&adapter->phy_info_timer,
2395                                           round_jiffies(jiffies + 2 * HZ));
2396                 }
2397         }
2398
2399 link_up:
2400         igb_update_stats(adapter);
2401
2402         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2403         adapter->tpt_old = adapter->stats.tpt;
2404         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2405         adapter->colc_old = adapter->stats.colc;
2406
2407         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2408         adapter->gorc_old = adapter->stats.gorc;
2409         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2410         adapter->gotc_old = adapter->stats.gotc;
2411
2412         igb_update_adaptive(&adapter->hw);
2413
2414         if (!netif_carrier_ok(netdev)) {
2415                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2416                         /* We've lost link, so the controller stops DMA,
2417                          * but we've got queued Tx work that's never going
2418                          * to get done, so reset controller to flush Tx.
2419                          * (Do the reset outside of interrupt context). */
2420                         adapter->tx_timeout_count++;
2421                         schedule_work(&adapter->reset_task);
2422                 }
2423         }
2424
2425         /* Cause software interrupt to ensure rx ring is cleaned */
2426         wr32(E1000_ICS, E1000_ICS_RXDMT0);
2427
2428         /* Force detection of hung controller every watchdog period */
2429         tx_ring->detect_tx_hung = true;
2430
2431         /* Reset the timer */
2432         if (!test_bit(__IGB_DOWN, &adapter->state))
2433                 mod_timer(&adapter->watchdog_timer,
2434                           round_jiffies(jiffies + 2 * HZ));
2435 }
2436
2437 enum latency_range {
2438         lowest_latency = 0,
2439         low_latency = 1,
2440         bulk_latency = 2,
2441         latency_invalid = 255
2442 };
2443
2444
2445 /**
2446  * igb_update_ring_itr - update the dynamic ITR value based on packet size
2447  *
2448  *      Stores a new ITR value based on strictly on packet size.  This
2449  *      algorithm is less sophisticated than that used in igb_update_itr,
2450  *      due to the difficulty of synchronizing statistics across multiple
2451  *      receive rings.  The divisors and thresholds used by this fuction
2452  *      were determined based on theoretical maximum wire speed and testing
2453  *      data, in order to minimize response time while increasing bulk
2454  *      throughput.
2455  *      This functionality is controlled by the InterruptThrottleRate module
2456  *      parameter (see igb_param.c)
2457  *      NOTE:  This function is called only when operating in a multiqueue
2458  *             receive environment.
2459  * @rx_ring: pointer to ring
2460  **/
2461 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2462 {
2463         int new_val = rx_ring->itr_val;
2464         int avg_wire_size = 0;
2465         struct igb_adapter *adapter = rx_ring->adapter;
2466
2467         if (!rx_ring->total_packets)
2468                 goto clear_counts; /* no packets, so don't do anything */
2469
2470         /* For non-gigabit speeds, just fix the interrupt rate at 4000
2471          * ints/sec - ITR timer value of 120 ticks.
2472          */
2473         if (adapter->link_speed != SPEED_1000) {
2474                 new_val = 120;
2475                 goto set_itr_val;
2476         }
2477         avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2478
2479         /* Add 24 bytes to size to account for CRC, preamble, and gap */
2480         avg_wire_size += 24;
2481
2482         /* Don't starve jumbo frames */
2483         avg_wire_size = min(avg_wire_size, 3000);
2484
2485         /* Give a little boost to mid-size frames */
2486         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2487                 new_val = avg_wire_size / 3;
2488         else
2489                 new_val = avg_wire_size / 2;
2490
2491 set_itr_val:
2492         if (new_val != rx_ring->itr_val) {
2493                 rx_ring->itr_val = new_val;
2494                 rx_ring->set_itr = 1;
2495         }
2496 clear_counts:
2497         rx_ring->total_bytes = 0;
2498         rx_ring->total_packets = 0;
2499 }
2500
2501 /**
2502  * igb_update_itr - update the dynamic ITR value based on statistics
2503  *      Stores a new ITR value based on packets and byte
2504  *      counts during the last interrupt.  The advantage of per interrupt
2505  *      computation is faster updates and more accurate ITR for the current
2506  *      traffic pattern.  Constants in this function were computed
2507  *      based on theoretical maximum wire speed and thresholds were set based
2508  *      on testing data as well as attempting to minimize response time
2509  *      while increasing bulk throughput.
2510  *      this functionality is controlled by the InterruptThrottleRate module
2511  *      parameter (see igb_param.c)
2512  *      NOTE:  These calculations are only valid when operating in a single-
2513  *             queue environment.
2514  * @adapter: pointer to adapter
2515  * @itr_setting: current adapter->itr
2516  * @packets: the number of packets during this measurement interval
2517  * @bytes: the number of bytes during this measurement interval
2518  **/
2519 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2520                                    int packets, int bytes)
2521 {
2522         unsigned int retval = itr_setting;
2523
2524         if (packets == 0)
2525                 goto update_itr_done;
2526
2527         switch (itr_setting) {
2528         case lowest_latency:
2529                 /* handle TSO and jumbo frames */
2530                 if (bytes/packets > 8000)
2531                         retval = bulk_latency;
2532                 else if ((packets < 5) && (bytes > 512))
2533                         retval = low_latency;
2534                 break;
2535         case low_latency:  /* 50 usec aka 20000 ints/s */
2536                 if (bytes > 10000) {
2537                         /* this if handles the TSO accounting */
2538                         if (bytes/packets > 8000) {
2539                                 retval = bulk_latency;
2540                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2541                                 retval = bulk_latency;
2542                         } else if ((packets > 35)) {
2543                                 retval = lowest_latency;
2544                         }
2545                 } else if (bytes/packets > 2000) {
2546                         retval = bulk_latency;
2547                 } else if (packets <= 2 && bytes < 512) {
2548                         retval = lowest_latency;
2549                 }
2550                 break;
2551         case bulk_latency: /* 250 usec aka 4000 ints/s */
2552                 if (bytes > 25000) {
2553                         if (packets > 35)
2554                                 retval = low_latency;
2555                 } else if (bytes < 6000) {
2556                         retval = low_latency;
2557                 }
2558                 break;
2559         }
2560
2561 update_itr_done:
2562         return retval;
2563 }
2564
2565 static void igb_set_itr(struct igb_adapter *adapter)
2566 {
2567         u16 current_itr;
2568         u32 new_itr = adapter->itr;
2569
2570         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2571         if (adapter->link_speed != SPEED_1000) {
2572                 current_itr = 0;
2573                 new_itr = 4000;
2574                 goto set_itr_now;
2575         }
2576
2577         adapter->rx_itr = igb_update_itr(adapter,
2578                                     adapter->rx_itr,
2579                                     adapter->rx_ring->total_packets,
2580                                     adapter->rx_ring->total_bytes);
2581
2582         if (adapter->rx_ring->buddy) {
2583                 adapter->tx_itr = igb_update_itr(adapter,
2584                                             adapter->tx_itr,
2585                                             adapter->tx_ring->total_packets,
2586                                             adapter->tx_ring->total_bytes);
2587
2588                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2589         } else {
2590                 current_itr = adapter->rx_itr;
2591         }
2592
2593         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2594         if (adapter->itr_setting == 3 &&
2595             current_itr == lowest_latency)
2596                 current_itr = low_latency;
2597
2598         switch (current_itr) {
2599         /* counts and packets in update_itr are dependent on these numbers */
2600         case lowest_latency:
2601                 new_itr = 70000;
2602                 break;
2603         case low_latency:
2604                 new_itr = 20000; /* aka hwitr = ~200 */
2605                 break;
2606         case bulk_latency:
2607                 new_itr = 4000;
2608                 break;
2609         default:
2610                 break;
2611         }
2612
2613 set_itr_now:
2614         adapter->rx_ring->total_bytes = 0;
2615         adapter->rx_ring->total_packets = 0;
2616         if (adapter->rx_ring->buddy) {
2617                 adapter->rx_ring->buddy->total_bytes = 0;
2618                 adapter->rx_ring->buddy->total_packets = 0;
2619         }
2620
2621         if (new_itr != adapter->itr) {
2622                 /* this attempts to bias the interrupt rate towards Bulk
2623                  * by adding intermediate steps when interrupt rate is
2624                  * increasing */
2625                 new_itr = new_itr > adapter->itr ?
2626                              min(adapter->itr + (new_itr >> 2), new_itr) :
2627                              new_itr;
2628                 /* Don't write the value here; it resets the adapter's
2629                  * internal timer, and causes us to delay far longer than
2630                  * we should between interrupts.  Instead, we write the ITR
2631                  * value at the beginning of the next interrupt so the timing
2632                  * ends up being correct.
2633                  */
2634                 adapter->itr = new_itr;
2635                 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2636                 adapter->rx_ring->set_itr = 1;
2637         }
2638
2639         return;
2640 }
2641
2642
2643 #define IGB_TX_FLAGS_CSUM               0x00000001
2644 #define IGB_TX_FLAGS_VLAN               0x00000002
2645 #define IGB_TX_FLAGS_TSO                0x00000004
2646 #define IGB_TX_FLAGS_IPV4               0x00000008
2647 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2648 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2649
2650 static inline int igb_tso_adv(struct igb_adapter *adapter,
2651                               struct igb_ring *tx_ring,
2652                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2653 {
2654         struct e1000_adv_tx_context_desc *context_desc;
2655         unsigned int i;
2656         int err;
2657         struct igb_buffer *buffer_info;
2658         u32 info = 0, tu_cmd = 0;
2659         u32 mss_l4len_idx, l4len;
2660         *hdr_len = 0;
2661
2662         if (skb_header_cloned(skb)) {
2663                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2664                 if (err)
2665                         return err;
2666         }
2667
2668         l4len = tcp_hdrlen(skb);
2669         *hdr_len += l4len;
2670
2671         if (skb->protocol == htons(ETH_P_IP)) {
2672                 struct iphdr *iph = ip_hdr(skb);
2673                 iph->tot_len = 0;
2674                 iph->check = 0;
2675                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2676                                                          iph->daddr, 0,
2677                                                          IPPROTO_TCP,
2678                                                          0);
2679         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2680                 ipv6_hdr(skb)->payload_len = 0;
2681                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2682                                                        &ipv6_hdr(skb)->daddr,
2683                                                        0, IPPROTO_TCP, 0);
2684         }
2685
2686         i = tx_ring->next_to_use;
2687
2688         buffer_info = &tx_ring->buffer_info[i];
2689         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2690         /* VLAN MACLEN IPLEN */
2691         if (tx_flags & IGB_TX_FLAGS_VLAN)
2692                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2693         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2694         *hdr_len += skb_network_offset(skb);
2695         info |= skb_network_header_len(skb);
2696         *hdr_len += skb_network_header_len(skb);
2697         context_desc->vlan_macip_lens = cpu_to_le32(info);
2698
2699         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2700         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2701
2702         if (skb->protocol == htons(ETH_P_IP))
2703                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2704         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2705
2706         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2707
2708         /* MSS L4LEN IDX */
2709         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2710         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2711
2712         /* Context index must be unique per ring. */
2713         if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2714                 mss_l4len_idx |= tx_ring->queue_index << 4;
2715
2716         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2717         context_desc->seqnum_seed = 0;
2718
2719         buffer_info->time_stamp = jiffies;
2720         buffer_info->dma = 0;
2721         i++;
2722         if (i == tx_ring->count)
2723                 i = 0;
2724
2725         tx_ring->next_to_use = i;
2726
2727         return true;
2728 }
2729
2730 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2731                                         struct igb_ring *tx_ring,
2732                                         struct sk_buff *skb, u32 tx_flags)
2733 {
2734         struct e1000_adv_tx_context_desc *context_desc;
2735         unsigned int i;
2736         struct igb_buffer *buffer_info;
2737         u32 info = 0, tu_cmd = 0;
2738
2739         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2740             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2741                 i = tx_ring->next_to_use;
2742                 buffer_info = &tx_ring->buffer_info[i];
2743                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2744
2745                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2746                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2747                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2748                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2749                         info |= skb_network_header_len(skb);
2750
2751                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2752
2753                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2754
2755                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2756                         switch (skb->protocol) {
2757                         case __constant_htons(ETH_P_IP):
2758                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2759                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2760                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2761                                 break;
2762                         case __constant_htons(ETH_P_IPV6):
2763                                 /* XXX what about other V6 headers?? */
2764                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2765                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2766                                 break;
2767                         default:
2768                                 if (unlikely(net_ratelimit()))
2769                                         dev_warn(&adapter->pdev->dev,
2770                                             "partial checksum but proto=%x!\n",
2771                                             skb->protocol);
2772                                 break;
2773                         }
2774                 }
2775
2776                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2777                 context_desc->seqnum_seed = 0;
2778                 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2779                         context_desc->mss_l4len_idx =
2780                                 cpu_to_le32(tx_ring->queue_index << 4);
2781
2782                 buffer_info->time_stamp = jiffies;
2783                 buffer_info->dma = 0;
2784
2785                 i++;
2786                 if (i == tx_ring->count)
2787                         i = 0;
2788                 tx_ring->next_to_use = i;
2789
2790                 return true;
2791         }
2792
2793
2794         return false;
2795 }
2796
2797 #define IGB_MAX_TXD_PWR 16
2798 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2799
2800 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2801                                  struct igb_ring *tx_ring,
2802                                  struct sk_buff *skb)
2803 {
2804         struct igb_buffer *buffer_info;
2805         unsigned int len = skb_headlen(skb);
2806         unsigned int count = 0, i;
2807         unsigned int f;
2808
2809         i = tx_ring->next_to_use;
2810
2811         buffer_info = &tx_ring->buffer_info[i];
2812         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2813         buffer_info->length = len;
2814         /* set time_stamp *before* dma to help avoid a possible race */
2815         buffer_info->time_stamp = jiffies;
2816         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2817                                           PCI_DMA_TODEVICE);
2818         count++;
2819         i++;
2820         if (i == tx_ring->count)
2821                 i = 0;
2822
2823         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2824                 struct skb_frag_struct *frag;
2825
2826                 frag = &skb_shinfo(skb)->frags[f];
2827                 len = frag->size;
2828
2829                 buffer_info = &tx_ring->buffer_info[i];
2830                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2831                 buffer_info->length = len;
2832                 buffer_info->time_stamp = jiffies;
2833                 buffer_info->dma = pci_map_page(adapter->pdev,
2834                                                 frag->page,
2835                                                 frag->page_offset,
2836                                                 len,
2837                                                 PCI_DMA_TODEVICE);
2838
2839                 count++;
2840                 i++;
2841                 if (i == tx_ring->count)
2842                         i = 0;
2843         }
2844
2845         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2846         tx_ring->buffer_info[i].skb = skb;
2847
2848         return count;
2849 }
2850
2851 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2852                                     struct igb_ring *tx_ring,
2853                                     int tx_flags, int count, u32 paylen,
2854                                     u8 hdr_len)
2855 {
2856         union e1000_adv_tx_desc *tx_desc = NULL;
2857         struct igb_buffer *buffer_info;
2858         u32 olinfo_status = 0, cmd_type_len;
2859         unsigned int i;
2860
2861         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2862                         E1000_ADVTXD_DCMD_DEXT);
2863
2864         if (tx_flags & IGB_TX_FLAGS_VLAN)
2865                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2866
2867         if (tx_flags & IGB_TX_FLAGS_TSO) {
2868                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2869
2870                 /* insert tcp checksum */
2871                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2872
2873                 /* insert ip checksum */
2874                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2875                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2876
2877         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2878                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2879         }
2880
2881         if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
2882             (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2883                          IGB_TX_FLAGS_VLAN)))
2884                 olinfo_status |= tx_ring->queue_index << 4;
2885
2886         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2887
2888         i = tx_ring->next_to_use;
2889         while (count--) {
2890                 buffer_info = &tx_ring->buffer_info[i];
2891                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2892                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2893                 tx_desc->read.cmd_type_len =
2894                         cpu_to_le32(cmd_type_len | buffer_info->length);
2895                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2896                 i++;
2897                 if (i == tx_ring->count)
2898                         i = 0;
2899         }
2900
2901         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2902         /* Force memory writes to complete before letting h/w
2903          * know there are new descriptors to fetch.  (Only
2904          * applicable for weak-ordered memory model archs,
2905          * such as IA-64). */
2906         wmb();
2907
2908         tx_ring->next_to_use = i;
2909         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2910         /* we need this if more than one processor can write to our tail
2911          * at a time, it syncronizes IO on IA64/Altix systems */
2912         mmiowb();
2913 }
2914
2915 static int __igb_maybe_stop_tx(struct net_device *netdev,
2916                                struct igb_ring *tx_ring, int size)
2917 {
2918         struct igb_adapter *adapter = netdev_priv(netdev);
2919
2920         netif_stop_subqueue(netdev, tx_ring->queue_index);
2921
2922         /* Herbert's original patch had:
2923          *  smp_mb__after_netif_stop_queue();
2924          * but since that doesn't exist yet, just open code it. */
2925         smp_mb();
2926
2927         /* We need to check again in a case another CPU has just
2928          * made room available. */
2929         if (IGB_DESC_UNUSED(tx_ring) < size)
2930                 return -EBUSY;
2931
2932         /* A reprieve! */
2933         netif_wake_subqueue(netdev, tx_ring->queue_index);
2934         ++adapter->restart_queue;
2935         return 0;
2936 }
2937
2938 static int igb_maybe_stop_tx(struct net_device *netdev,
2939                              struct igb_ring *tx_ring, int size)
2940 {
2941         if (IGB_DESC_UNUSED(tx_ring) >= size)
2942                 return 0;
2943         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2944 }
2945
2946 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2947
2948 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2949                                    struct net_device *netdev,
2950                                    struct igb_ring *tx_ring)
2951 {
2952         struct igb_adapter *adapter = netdev_priv(netdev);
2953         unsigned int tx_flags = 0;
2954         unsigned int len;
2955         u8 hdr_len = 0;
2956         int tso = 0;
2957
2958         len = skb_headlen(skb);
2959
2960         if (test_bit(__IGB_DOWN, &adapter->state)) {
2961                 dev_kfree_skb_any(skb);
2962                 return NETDEV_TX_OK;
2963         }
2964
2965         if (skb->len <= 0) {
2966                 dev_kfree_skb_any(skb);
2967                 return NETDEV_TX_OK;
2968         }
2969
2970         /* need: 1 descriptor per page,
2971          *       + 2 desc gap to keep tail from touching head,
2972          *       + 1 desc for skb->data,
2973          *       + 1 desc for context descriptor,
2974          * otherwise try next time */
2975         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2976                 /* this is a hard error */
2977                 return NETDEV_TX_BUSY;
2978         }
2979         skb_orphan(skb);
2980
2981         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2982                 tx_flags |= IGB_TX_FLAGS_VLAN;
2983                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2984         }
2985
2986         if (skb->protocol == htons(ETH_P_IP))
2987                 tx_flags |= IGB_TX_FLAGS_IPV4;
2988
2989         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2990                                               &hdr_len) : 0;
2991
2992         if (tso < 0) {
2993                 dev_kfree_skb_any(skb);
2994                 return NETDEV_TX_OK;
2995         }
2996
2997         if (tso)
2998                 tx_flags |= IGB_TX_FLAGS_TSO;
2999         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
3000                         if (skb->ip_summed == CHECKSUM_PARTIAL)
3001                                 tx_flags |= IGB_TX_FLAGS_CSUM;
3002
3003         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
3004                          igb_tx_map_adv(adapter, tx_ring, skb),
3005                          skb->len, hdr_len);
3006
3007         netdev->trans_start = jiffies;
3008
3009         /* Make sure there is space in the ring for the next send. */
3010         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3011
3012         return NETDEV_TX_OK;
3013 }
3014
3015 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3016 {
3017         struct igb_adapter *adapter = netdev_priv(netdev);
3018         struct igb_ring *tx_ring;
3019
3020         int r_idx = 0;
3021         r_idx = skb->queue_mapping & (IGB_MAX_TX_QUEUES - 1);
3022         tx_ring = adapter->multi_tx_table[r_idx];
3023
3024         /* This goes back to the question of how to logically map a tx queue
3025          * to a flow.  Right now, performance is impacted slightly negatively
3026          * if using multiple tx queues.  If the stack breaks away from a
3027          * single qdisc implementation, we can look at this again. */
3028         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3029 }
3030
3031 /**
3032  * igb_tx_timeout - Respond to a Tx Hang
3033  * @netdev: network interface device structure
3034  **/
3035 static void igb_tx_timeout(struct net_device *netdev)
3036 {
3037         struct igb_adapter *adapter = netdev_priv(netdev);
3038         struct e1000_hw *hw = &adapter->hw;
3039
3040         /* Do the reset outside of interrupt context */
3041         adapter->tx_timeout_count++;
3042         schedule_work(&adapter->reset_task);
3043         wr32(E1000_EICS, adapter->eims_enable_mask &
3044                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
3045 }
3046
3047 static void igb_reset_task(struct work_struct *work)
3048 {
3049         struct igb_adapter *adapter;
3050         adapter = container_of(work, struct igb_adapter, reset_task);
3051
3052         igb_reinit_locked(adapter);
3053 }
3054
3055 /**
3056  * igb_get_stats - Get System Network Statistics
3057  * @netdev: network interface device structure
3058  *
3059  * Returns the address of the device statistics structure.
3060  * The statistics are actually updated from the timer callback.
3061  **/
3062 static struct net_device_stats *
3063 igb_get_stats(struct net_device *netdev)
3064 {
3065         struct igb_adapter *adapter = netdev_priv(netdev);
3066
3067         /* only return the current stats */
3068         return &adapter->net_stats;
3069 }
3070
3071 /**
3072  * igb_change_mtu - Change the Maximum Transfer Unit
3073  * @netdev: network interface device structure
3074  * @new_mtu: new value for maximum frame size
3075  *
3076  * Returns 0 on success, negative on failure
3077  **/
3078 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3079 {
3080         struct igb_adapter *adapter = netdev_priv(netdev);
3081         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3082
3083         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3084             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3085                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3086                 return -EINVAL;
3087         }
3088
3089 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3090         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3091                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3092                 return -EINVAL;
3093         }
3094
3095         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3096                 msleep(1);
3097         /* igb_down has a dependency on max_frame_size */
3098         adapter->max_frame_size = max_frame;
3099         if (netif_running(netdev))
3100                 igb_down(adapter);
3101
3102         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3103          * means we reserve 2 more, this pushes us to allocate from the next
3104          * larger slab size.
3105          * i.e. RXBUFFER_2048 --> size-4096 slab
3106          */
3107
3108         if (max_frame <= IGB_RXBUFFER_256)
3109                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3110         else if (max_frame <= IGB_RXBUFFER_512)
3111                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3112         else if (max_frame <= IGB_RXBUFFER_1024)
3113                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3114         else if (max_frame <= IGB_RXBUFFER_2048)
3115                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3116         else
3117 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3118                 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3119 #else
3120                 adapter->rx_buffer_len = PAGE_SIZE / 2;
3121 #endif
3122         /* adjust allocation if LPE protects us, and we aren't using SBP */
3123         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3124              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3125                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3126
3127         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3128                  netdev->mtu, new_mtu);
3129         netdev->mtu = new_mtu;
3130
3131         if (netif_running(netdev))
3132                 igb_up(adapter);
3133         else
3134                 igb_reset(adapter);
3135
3136         clear_bit(__IGB_RESETTING, &adapter->state);
3137
3138         return 0;
3139 }
3140
3141 /**
3142  * igb_update_stats - Update the board statistics counters
3143  * @adapter: board private structure
3144  **/
3145
3146 void igb_update_stats(struct igb_adapter *adapter)
3147 {
3148         struct e1000_hw *hw = &adapter->hw;
3149         struct pci_dev *pdev = adapter->pdev;
3150         u16 phy_tmp;
3151
3152 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3153
3154         /*
3155          * Prevent stats update while adapter is being reset, or if the pci
3156          * connection is down.
3157          */
3158         if (adapter->link_speed == 0)
3159                 return;
3160         if (pci_channel_offline(pdev))
3161                 return;
3162
3163         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3164         adapter->stats.gprc += rd32(E1000_GPRC);
3165         adapter->stats.gorc += rd32(E1000_GORCL);
3166         rd32(E1000_GORCH); /* clear GORCL */
3167         adapter->stats.bprc += rd32(E1000_BPRC);
3168         adapter->stats.mprc += rd32(E1000_MPRC);
3169         adapter->stats.roc += rd32(E1000_ROC);
3170
3171         adapter->stats.prc64 += rd32(E1000_PRC64);
3172         adapter->stats.prc127 += rd32(E1000_PRC127);
3173         adapter->stats.prc255 += rd32(E1000_PRC255);
3174         adapter->stats.prc511 += rd32(E1000_PRC511);
3175         adapter->stats.prc1023 += rd32(E1000_PRC1023);
3176         adapter->stats.prc1522 += rd32(E1000_PRC1522);
3177         adapter->stats.symerrs += rd32(E1000_SYMERRS);
3178         adapter->stats.sec += rd32(E1000_SEC);
3179
3180         adapter->stats.mpc += rd32(E1000_MPC);
3181         adapter->stats.scc += rd32(E1000_SCC);
3182         adapter->stats.ecol += rd32(E1000_ECOL);
3183         adapter->stats.mcc += rd32(E1000_MCC);
3184         adapter->stats.latecol += rd32(E1000_LATECOL);
3185         adapter->stats.dc += rd32(E1000_DC);
3186         adapter->stats.rlec += rd32(E1000_RLEC);
3187         adapter->stats.xonrxc += rd32(E1000_XONRXC);
3188         adapter->stats.xontxc += rd32(E1000_XONTXC);
3189         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3190         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3191         adapter->stats.fcruc += rd32(E1000_FCRUC);
3192         adapter->stats.gptc += rd32(E1000_GPTC);
3193         adapter->stats.gotc += rd32(E1000_GOTCL);
3194         rd32(E1000_GOTCH); /* clear GOTCL */
3195         adapter->stats.rnbc += rd32(E1000_RNBC);
3196         adapter->stats.ruc += rd32(E1000_RUC);
3197         adapter->stats.rfc += rd32(E1000_RFC);
3198         adapter->stats.rjc += rd32(E1000_RJC);
3199         adapter->stats.tor += rd32(E1000_TORH);
3200         adapter->stats.tot += rd32(E1000_TOTH);
3201         adapter->stats.tpr += rd32(E1000_TPR);
3202
3203         adapter->stats.ptc64 += rd32(E1000_PTC64);
3204         adapter->stats.ptc127 += rd32(E1000_PTC127);
3205         adapter->stats.ptc255 += rd32(E1000_PTC255);
3206         adapter->stats.ptc511 += rd32(E1000_PTC511);
3207         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3208         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3209
3210         adapter->stats.mptc += rd32(E1000_MPTC);
3211         adapter->stats.bptc += rd32(E1000_BPTC);
3212
3213         /* used for adaptive IFS */
3214
3215         hw->mac.tx_packet_delta = rd32(E1000_TPT);
3216         adapter->stats.tpt += hw->mac.tx_packet_delta;
3217         hw->mac.collision_delta = rd32(E1000_COLC);
3218         adapter->stats.colc += hw->mac.collision_delta;
3219
3220         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3221         adapter->stats.rxerrc += rd32(E1000_RXERRC);
3222         adapter->stats.tncrs += rd32(E1000_TNCRS);
3223         adapter->stats.tsctc += rd32(E1000_TSCTC);
3224         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3225
3226         adapter->stats.iac += rd32(E1000_IAC);
3227         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3228         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3229         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3230         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3231         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3232         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3233         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3234         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3235
3236         /* Fill out the OS statistics structure */
3237         adapter->net_stats.multicast = adapter->stats.mprc;
3238         adapter->net_stats.collisions = adapter->stats.colc;
3239
3240         /* Rx Errors */
3241
3242         /* RLEC on some newer hardware can be incorrect so build
3243         * our own version based on RUC and ROC */
3244         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3245                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3246                 adapter->stats.ruc + adapter->stats.roc +
3247                 adapter->stats.cexterr;
3248         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3249                                               adapter->stats.roc;
3250         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3251         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3252         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3253
3254         /* Tx Errors */
3255         adapter->net_stats.tx_errors = adapter->stats.ecol +
3256                                        adapter->stats.latecol;
3257         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3258         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3259         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3260
3261         /* Tx Dropped needs to be maintained elsewhere */
3262
3263         /* Phy Stats */
3264         if (hw->phy.media_type == e1000_media_type_copper) {
3265                 if ((adapter->link_speed == SPEED_1000) &&
3266                    (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
3267                                               &phy_tmp))) {
3268                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3269                         adapter->phy_stats.idle_errors += phy_tmp;
3270                 }
3271         }
3272
3273         /* Management Stats */
3274         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3275         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3276         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3277 }
3278
3279
3280 static irqreturn_t igb_msix_other(int irq, void *data)
3281 {
3282         struct net_device *netdev = data;
3283         struct igb_adapter *adapter = netdev_priv(netdev);
3284         struct e1000_hw *hw = &adapter->hw;
3285         u32 icr = rd32(E1000_ICR);
3286
3287         /* reading ICR causes bit 31 of EICR to be cleared */
3288         if (!(icr & E1000_ICR_LSC))
3289                 goto no_link_interrupt;
3290         hw->mac.get_link_status = 1;
3291         /* guard against interrupt when we're going down */
3292         if (!test_bit(__IGB_DOWN, &adapter->state))
3293                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3294
3295 no_link_interrupt:
3296         wr32(E1000_IMS, E1000_IMS_LSC);
3297         wr32(E1000_EIMS, adapter->eims_other);
3298
3299         return IRQ_HANDLED;
3300 }
3301
3302 static irqreturn_t igb_msix_tx(int irq, void *data)
3303 {
3304         struct igb_ring *tx_ring = data;
3305         struct igb_adapter *adapter = tx_ring->adapter;
3306         struct e1000_hw *hw = &adapter->hw;
3307
3308 #ifdef CONFIG_DCA
3309         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3310                 igb_update_tx_dca(tx_ring);
3311 #endif
3312         tx_ring->total_bytes = 0;
3313         tx_ring->total_packets = 0;
3314
3315         /* auto mask will automatically reenable the interrupt when we write
3316          * EICS */
3317         if (!igb_clean_tx_irq(tx_ring))
3318                 /* Ring was not completely cleaned, so fire another interrupt */
3319                 wr32(E1000_EICS, tx_ring->eims_value);
3320         else
3321                 wr32(E1000_EIMS, tx_ring->eims_value);
3322
3323         return IRQ_HANDLED;
3324 }
3325
3326 static void igb_write_itr(struct igb_ring *ring)
3327 {
3328         struct e1000_hw *hw = &ring->adapter->hw;
3329         if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3330                 switch (hw->mac.type) {
3331                 case e1000_82576:
3332                         wr32(ring->itr_register,
3333                              ring->itr_val |
3334                              0x80000000);
3335                         break;
3336                 default:
3337                         wr32(ring->itr_register,
3338                              ring->itr_val |
3339                              (ring->itr_val << 16));
3340                         break;
3341                 }
3342                 ring->set_itr = 0;
3343         }
3344 }
3345
3346 static irqreturn_t igb_msix_rx(int irq, void *data)
3347 {
3348         struct igb_ring *rx_ring = data;
3349         struct igb_adapter *adapter = rx_ring->adapter;
3350
3351         /* Write the ITR value calculated at the end of the
3352          * previous interrupt.
3353          */
3354
3355         igb_write_itr(rx_ring);
3356
3357         if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi))
3358                 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3359
3360 #ifdef CONFIG_DCA
3361         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3362                 igb_update_rx_dca(rx_ring);
3363 #endif
3364                 return IRQ_HANDLED;
3365 }
3366
3367 #ifdef CONFIG_DCA
3368 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3369 {
3370         u32 dca_rxctrl;
3371         struct igb_adapter *adapter = rx_ring->adapter;
3372         struct e1000_hw *hw = &adapter->hw;
3373         int cpu = get_cpu();
3374         int q = rx_ring - adapter->rx_ring;
3375
3376         if (rx_ring->cpu != cpu) {
3377                 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3378                 if (hw->mac.type == e1000_82576) {
3379                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3380                         dca_rxctrl |= dca_get_tag(cpu) <<
3381                                       E1000_DCA_RXCTRL_CPUID_SHIFT;
3382                 } else {
3383                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3384                         dca_rxctrl |= dca_get_tag(cpu);
3385                 }
3386                 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3387                 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3388                 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3389                 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3390                 rx_ring->cpu = cpu;
3391         }
3392         put_cpu();
3393 }
3394
3395 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3396 {
3397         u32 dca_txctrl;
3398         struct igb_adapter *adapter = tx_ring->adapter;
3399         struct e1000_hw *hw = &adapter->hw;
3400         int cpu = get_cpu();
3401         int q = tx_ring - adapter->tx_ring;
3402
3403         if (tx_ring->cpu != cpu) {
3404                 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3405                 if (hw->mac.type == e1000_82576) {
3406                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3407                         dca_txctrl |= dca_get_tag(cpu) <<
3408                                       E1000_DCA_TXCTRL_CPUID_SHIFT;
3409                 } else {
3410                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3411                         dca_txctrl |= dca_get_tag(cpu);
3412                 }
3413                 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3414                 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3415                 tx_ring->cpu = cpu;
3416         }
3417         put_cpu();
3418 }
3419
3420 static void igb_setup_dca(struct igb_adapter *adapter)
3421 {
3422         int i;
3423
3424         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3425                 return;
3426
3427         for (i = 0; i < adapter->num_tx_queues; i++) {
3428                 adapter->tx_ring[i].cpu = -1;
3429                 igb_update_tx_dca(&adapter->tx_ring[i]);
3430         }
3431         for (i = 0; i < adapter->num_rx_queues; i++) {
3432                 adapter->rx_ring[i].cpu = -1;
3433                 igb_update_rx_dca(&adapter->rx_ring[i]);
3434         }
3435 }
3436
3437 static int __igb_notify_dca(struct device *dev, void *data)
3438 {
3439         struct net_device *netdev = dev_get_drvdata(dev);
3440         struct igb_adapter *adapter = netdev_priv(netdev);
3441         struct e1000_hw *hw = &adapter->hw;
3442         unsigned long event = *(unsigned long *)data;
3443
3444         if (!(adapter->flags & IGB_FLAG_HAS_DCA))
3445                 goto out;
3446
3447         switch (event) {
3448         case DCA_PROVIDER_ADD:
3449                 /* if already enabled, don't do it again */
3450                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3451                         break;
3452                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3453                 /* Always use CB2 mode, difference is masked
3454                  * in the CB driver. */
3455                 wr32(E1000_DCA_CTRL, 2);
3456                 if (dca_add_requester(dev) == 0) {
3457                         dev_info(&adapter->pdev->dev, "DCA enabled\n");
3458                         igb_setup_dca(adapter);
3459                         break;
3460                 }
3461                 /* Fall Through since DCA is disabled. */
3462         case DCA_PROVIDER_REMOVE:
3463                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3464                         /* without this a class_device is left
3465                          * hanging around in the sysfs model */
3466                         dca_remove_requester(dev);
3467                         dev_info(&adapter->pdev->dev, "DCA disabled\n");
3468                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3469                         wr32(E1000_DCA_CTRL, 1);
3470                 }
3471                 break;
3472         }
3473 out:
3474         return 0;
3475 }
3476
3477 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3478                           void *p)
3479 {
3480         int ret_val;
3481
3482         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3483                                          __igb_notify_dca);
3484
3485         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3486 }
3487 #endif /* CONFIG_DCA */
3488
3489 /**
3490  * igb_intr_msi - Interrupt Handler
3491  * @irq: interrupt number
3492  * @data: pointer to a network interface device structure
3493  **/
3494 static irqreturn_t igb_intr_msi(int irq, void *data)
3495 {
3496         struct net_device *netdev = data;
3497         struct igb_adapter *adapter = netdev_priv(netdev);
3498         struct e1000_hw *hw = &adapter->hw;
3499         /* read ICR disables interrupts using IAM */
3500         u32 icr = rd32(E1000_ICR);
3501
3502         igb_write_itr(adapter->rx_ring);
3503
3504         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3505                 hw->mac.get_link_status = 1;
3506                 if (!test_bit(__IGB_DOWN, &adapter->state))
3507                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3508         }
3509
3510         netif_rx_schedule(netdev, &adapter->rx_ring[0].napi);
3511
3512         return IRQ_HANDLED;
3513 }
3514
3515 /**
3516  * igb_intr - Interrupt Handler
3517  * @irq: interrupt number
3518  * @data: pointer to a network interface device structure
3519  **/
3520 static irqreturn_t igb_intr(int irq, void *data)
3521 {
3522         struct net_device *netdev = data;
3523         struct igb_adapter *adapter = netdev_priv(netdev);
3524         struct e1000_hw *hw = &adapter->hw;
3525         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3526          * need for the IMC write */
3527         u32 icr = rd32(E1000_ICR);
3528         u32 eicr = 0;
3529         if (!icr)
3530                 return IRQ_NONE;  /* Not our interrupt */
3531
3532         igb_write_itr(adapter->rx_ring);
3533
3534         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3535          * not set, then the adapter didn't send an interrupt */
3536         if (!(icr & E1000_ICR_INT_ASSERTED))
3537                 return IRQ_NONE;
3538
3539         eicr = rd32(E1000_EICR);
3540
3541         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3542                 hw->mac.get_link_status = 1;
3543                 /* guard against interrupt when we're going down */
3544                 if (!test_bit(__IGB_DOWN, &adapter->state))
3545                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3546         }
3547
3548         netif_rx_schedule(netdev, &adapter->rx_ring[0].napi);
3549
3550         return IRQ_HANDLED;
3551 }
3552
3553 /**
3554  * igb_poll - NAPI Rx polling callback
3555  * @napi: napi polling structure
3556  * @budget: count of how many packets we should handle
3557  **/
3558 static int igb_poll(struct napi_struct *napi, int budget)
3559 {
3560         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3561         struct igb_adapter *adapter = rx_ring->adapter;
3562         struct net_device *netdev = adapter->netdev;
3563         int tx_clean_complete, work_done = 0;
3564
3565         /* this poll routine only supports one tx and one rx queue */
3566 #ifdef CONFIG_DCA
3567         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3568                 igb_update_tx_dca(&adapter->tx_ring[0]);
3569 #endif
3570         tx_clean_complete = igb_clean_tx_irq(&adapter->tx_ring[0]);
3571
3572 #ifdef CONFIG_DCA
3573         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3574                 igb_update_rx_dca(&adapter->rx_ring[0]);
3575 #endif
3576         igb_clean_rx_irq_adv(&adapter->rx_ring[0], &work_done, budget);
3577
3578         /* If no Tx and not enough Rx work done, exit the polling mode */
3579         if ((tx_clean_complete && (work_done < budget)) ||
3580             !netif_running(netdev)) {
3581                 if (adapter->itr_setting & 3)
3582                         igb_set_itr(adapter);
3583                 netif_rx_complete(netdev, napi);
3584                 if (!test_bit(__IGB_DOWN, &adapter->state))
3585                         igb_irq_enable(adapter);
3586                 return 0;
3587         }
3588
3589         return 1;
3590 }
3591
3592 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3593 {
3594         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3595         struct igb_adapter *adapter = rx_ring->adapter;
3596         struct e1000_hw *hw = &adapter->hw;
3597         struct net_device *netdev = adapter->netdev;
3598         int work_done = 0;
3599
3600         /* Keep link state information with original netdev */
3601         if (!netif_carrier_ok(netdev))
3602                 goto quit_polling;
3603
3604 #ifdef CONFIG_DCA
3605         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3606                 igb_update_rx_dca(rx_ring);
3607 #endif
3608         igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
3609
3610
3611         /* If not enough Rx work done, exit the polling mode */
3612         if ((work_done == 0) || !netif_running(netdev)) {
3613 quit_polling:
3614                 netif_rx_complete(netdev, napi);
3615
3616                 if (adapter->itr_setting & 3) {
3617                         if (adapter->num_rx_queues == 1)
3618                                 igb_set_itr(adapter);
3619                         else
3620                                 igb_update_ring_itr(rx_ring);
3621                 }
3622
3623                 if (!test_bit(__IGB_DOWN, &adapter->state))
3624                         wr32(E1000_EIMS, rx_ring->eims_value);
3625
3626                 return 0;
3627         }
3628
3629         return 1;
3630 }
3631
3632 static inline u32 get_head(struct igb_ring *tx_ring)
3633 {
3634         void *end = (struct e1000_tx_desc *)tx_ring->desc + tx_ring->count;
3635         return le32_to_cpu(*(volatile __le32 *)end);
3636 }
3637
3638 /**
3639  * igb_clean_tx_irq - Reclaim resources after transmit completes
3640  * @adapter: board private structure
3641  * returns true if ring is completely cleaned
3642  **/
3643 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
3644 {
3645         struct igb_adapter *adapter = tx_ring->adapter;
3646         struct e1000_hw *hw = &adapter->hw;
3647         struct net_device *netdev = adapter->netdev;
3648         struct e1000_tx_desc *tx_desc;
3649         struct igb_buffer *buffer_info;
3650         struct sk_buff *skb;
3651         unsigned int i;
3652         u32 head, oldhead;
3653         unsigned int count = 0;
3654         bool cleaned = false;
3655         bool retval = true;
3656         unsigned int total_bytes = 0, total_packets = 0;
3657
3658         rmb();
3659         head = get_head(tx_ring);
3660         i = tx_ring->next_to_clean;
3661         while (1) {
3662                 while (i != head) {
3663                         cleaned = true;
3664                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3665                         buffer_info = &tx_ring->buffer_info[i];
3666                         skb = buffer_info->skb;
3667
3668                         if (skb) {
3669                                 unsigned int segs, bytecount;
3670                                 /* gso_segs is currently only valid for tcp */
3671                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3672                                 /* multiply data chunks by size of headers */
3673                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3674                                             skb->len;
3675                                 total_packets += segs;
3676                                 total_bytes += bytecount;
3677                         }
3678
3679                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3680                         tx_desc->upper.data = 0;
3681
3682                         i++;
3683                         if (i == tx_ring->count)
3684                                 i = 0;
3685
3686                         count++;
3687                         if (count == IGB_MAX_TX_CLEAN) {
3688                                 retval = false;
3689                                 goto done_cleaning;
3690                         }
3691                 }
3692                 oldhead = head;
3693                 rmb();
3694                 head = get_head(tx_ring);
3695                 if (head == oldhead)
3696                         goto done_cleaning;
3697         }  /* while (1) */
3698
3699 done_cleaning:
3700         tx_ring->next_to_clean = i;
3701
3702         if (unlikely(cleaned &&
3703                      netif_carrier_ok(netdev) &&
3704                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3705                 /* Make sure that anybody stopping the queue after this
3706                  * sees the new next_to_clean.
3707                  */
3708                 smp_mb();
3709                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
3710                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3711                         netif_wake_subqueue(netdev, tx_ring->queue_index);
3712                         ++adapter->restart_queue;
3713                 }
3714         }
3715
3716         if (tx_ring->detect_tx_hung) {
3717                 /* Detect a transmit hang in hardware, this serializes the
3718                  * check with the clearing of time_stamp and movement of i */
3719                 tx_ring->detect_tx_hung = false;
3720                 if (tx_ring->buffer_info[i].time_stamp &&
3721                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3722                                (adapter->tx_timeout_factor * HZ))
3723                     && !(rd32(E1000_STATUS) &
3724                          E1000_STATUS_TXOFF)) {
3725
3726                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3727                         /* detected Tx unit hang */
3728                         dev_err(&adapter->pdev->dev,
3729                                 "Detected Tx Unit Hang\n"
3730                                 "  Tx Queue             <%d>\n"
3731                                 "  TDH                  <%x>\n"
3732                                 "  TDT                  <%x>\n"
3733                                 "  next_to_use          <%x>\n"
3734                                 "  next_to_clean        <%x>\n"
3735                                 "  head (WB)            <%x>\n"
3736                                 "buffer_info[next_to_clean]\n"
3737                                 "  time_stamp           <%lx>\n"
3738                                 "  jiffies              <%lx>\n"
3739                                 "  desc.status          <%x>\n",
3740                                 tx_ring->queue_index,
3741                                 readl(adapter->hw.hw_addr + tx_ring->head),
3742                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3743                                 tx_ring->next_to_use,
3744                                 tx_ring->next_to_clean,
3745                                 head,
3746                                 tx_ring->buffer_info[i].time_stamp,
3747                                 jiffies,
3748                                 tx_desc->upper.fields.status);
3749                         netif_stop_subqueue(netdev, tx_ring->queue_index);
3750                 }
3751         }
3752         tx_ring->total_bytes += total_bytes;
3753         tx_ring->total_packets += total_packets;
3754         tx_ring->tx_stats.bytes += total_bytes;
3755         tx_ring->tx_stats.packets += total_packets;
3756         adapter->net_stats.tx_bytes += total_bytes;
3757         adapter->net_stats.tx_packets += total_packets;
3758         return retval;
3759 }
3760
3761 #ifdef CONFIG_IGB_LRO
3762  /**
3763  * igb_get_skb_hdr - helper function for LRO header processing
3764  * @skb: pointer to sk_buff to be added to LRO packet
3765  * @iphdr: pointer to ip header structure
3766  * @tcph: pointer to tcp header structure
3767  * @hdr_flags: pointer to header flags
3768  * @priv: pointer to the receive descriptor for the current sk_buff
3769  **/
3770 static int igb_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
3771                            u64 *hdr_flags, void *priv)
3772 {
3773         union e1000_adv_rx_desc *rx_desc = priv;
3774         u16 pkt_type = rx_desc->wb.lower.lo_dword.pkt_info &
3775                        (E1000_RXDADV_PKTTYPE_IPV4 | E1000_RXDADV_PKTTYPE_TCP);
3776
3777         /* Verify that this is a valid IPv4 TCP packet */
3778         if (pkt_type != (E1000_RXDADV_PKTTYPE_IPV4 |
3779                           E1000_RXDADV_PKTTYPE_TCP))
3780                 return -1;
3781
3782         /* Set network headers */
3783         skb_reset_network_header(skb);
3784         skb_set_transport_header(skb, ip_hdrlen(skb));
3785         *iphdr = ip_hdr(skb);
3786         *tcph = tcp_hdr(skb);
3787         *hdr_flags = LRO_IPV4 | LRO_TCP;
3788
3789         return 0;
3790
3791 }
3792 #endif /* CONFIG_IGB_LRO */
3793
3794 /**
3795  * igb_receive_skb - helper function to handle rx indications
3796  * @ring: pointer to receive ring receving this packet
3797  * @status: descriptor status field as written by hardware
3798  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3799  * @skb: pointer to sk_buff to be indicated to stack
3800  **/
3801 static void igb_receive_skb(struct igb_ring *ring, u8 status,
3802                             union e1000_adv_rx_desc * rx_desc,
3803                             struct sk_buff *skb)
3804 {
3805         struct igb_adapter * adapter = ring->adapter;
3806         bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
3807
3808 #ifdef CONFIG_IGB_LRO
3809         if (adapter->netdev->features & NETIF_F_LRO &&
3810             skb->ip_summed == CHECKSUM_UNNECESSARY) {
3811                 if (vlan_extracted)
3812                         lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
3813                                            adapter->vlgrp,
3814                                            le16_to_cpu(rx_desc->wb.upper.vlan),
3815                                            rx_desc);
3816                 else
3817                         lro_receive_skb(&ring->lro_mgr,skb, rx_desc);
3818                 ring->lro_used = 1;
3819         } else {
3820 #endif
3821                 if (vlan_extracted)
3822                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3823                                           le16_to_cpu(rx_desc->wb.upper.vlan));
3824                 else
3825
3826                         netif_receive_skb(skb);
3827 #ifdef CONFIG_IGB_LRO
3828         }
3829 #endif
3830 }
3831
3832
3833 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3834                                        u32 status_err, struct sk_buff *skb)
3835 {
3836         skb->ip_summed = CHECKSUM_NONE;
3837
3838         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3839         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3840                 return;
3841         /* TCP/UDP checksum error bit is set */
3842         if (status_err &
3843             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3844                 /* let the stack verify checksum errors */
3845                 adapter->hw_csum_err++;
3846                 return;
3847         }
3848         /* It must be a TCP or UDP packet with a valid checksum */
3849         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3850                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3851
3852         adapter->hw_csum_good++;
3853 }
3854
3855 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
3856                                  int *work_done, int budget)
3857 {
3858         struct igb_adapter *adapter = rx_ring->adapter;
3859         struct net_device *netdev = adapter->netdev;
3860         struct pci_dev *pdev = adapter->pdev;
3861         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3862         struct igb_buffer *buffer_info , *next_buffer;
3863         struct sk_buff *skb;
3864         unsigned int i;
3865         u32 length, hlen, staterr;
3866         bool cleaned = false;
3867         int cleaned_count = 0;
3868         unsigned int total_bytes = 0, total_packets = 0;
3869
3870         i = rx_ring->next_to_clean;
3871         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3872         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3873
3874         while (staterr & E1000_RXD_STAT_DD) {
3875                 if (*work_done >= budget)
3876                         break;
3877                 (*work_done)++;
3878                 buffer_info = &rx_ring->buffer_info[i];
3879
3880                 /* HW will not DMA in data larger than the given buffer, even
3881                  * if it parses the (NFS, of course) header to be larger.  In
3882                  * that case, it fills the header buffer and spills the rest
3883                  * into the page.
3884                  */
3885                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
3886                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
3887                 if (hlen > adapter->rx_ps_hdr_size)
3888                         hlen = adapter->rx_ps_hdr_size;
3889
3890                 length = le16_to_cpu(rx_desc->wb.upper.length);
3891                 cleaned = true;
3892                 cleaned_count++;
3893
3894                 skb = buffer_info->skb;
3895                 prefetch(skb->data - NET_IP_ALIGN);
3896                 buffer_info->skb = NULL;
3897                 if (!adapter->rx_ps_hdr_size) {
3898                         pci_unmap_single(pdev, buffer_info->dma,
3899                                          adapter->rx_buffer_len +
3900                                            NET_IP_ALIGN,
3901                                          PCI_DMA_FROMDEVICE);
3902                         skb_put(skb, length);
3903                         goto send_up;
3904                 }
3905
3906                 if (!skb_shinfo(skb)->nr_frags) {
3907                         pci_unmap_single(pdev, buffer_info->dma,
3908                                          adapter->rx_ps_hdr_size +
3909                                            NET_IP_ALIGN,
3910                                          PCI_DMA_FROMDEVICE);
3911                         skb_put(skb, hlen);
3912                 }
3913
3914                 if (length) {
3915                         pci_unmap_page(pdev, buffer_info->page_dma,
3916                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3917                         buffer_info->page_dma = 0;
3918
3919                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
3920                                                 buffer_info->page,
3921                                                 buffer_info->page_offset,
3922                                                 length);
3923
3924                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
3925                             (page_count(buffer_info->page) != 1))
3926                                 buffer_info->page = NULL;
3927                         else
3928                                 get_page(buffer_info->page);
3929
3930                         skb->len += length;
3931                         skb->data_len += length;
3932
3933                         skb->truesize += length;
3934                 }
3935 send_up:
3936                 i++;
3937                 if (i == rx_ring->count)
3938                         i = 0;
3939                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3940                 prefetch(next_rxd);
3941                 next_buffer = &rx_ring->buffer_info[i];
3942
3943                 if (!(staterr & E1000_RXD_STAT_EOP)) {
3944                         buffer_info->skb = xchg(&next_buffer->skb, skb);
3945                         buffer_info->dma = xchg(&next_buffer->dma, 0);
3946                         goto next_desc;
3947                 }
3948
3949                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3950                         dev_kfree_skb_irq(skb);
3951                         goto next_desc;
3952                 }
3953
3954                 total_bytes += skb->len;
3955                 total_packets++;
3956
3957                 igb_rx_checksum_adv(adapter, staterr, skb);
3958
3959                 skb->protocol = eth_type_trans(skb, netdev);
3960
3961                 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
3962
3963                 netdev->last_rx = jiffies;
3964
3965 next_desc:
3966                 rx_desc->wb.upper.status_error = 0;
3967
3968                 /* return some buffers to hardware, one at a time is too slow */
3969                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3970                         igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3971                         cleaned_count = 0;
3972                 }
3973
3974                 /* use prefetched values */
3975                 rx_desc = next_rxd;
3976                 buffer_info = next_buffer;
3977
3978                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3979         }
3980
3981         rx_ring->next_to_clean = i;
3982         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3983
3984 #ifdef CONFIG_IGB_LRO
3985         if (rx_ring->lro_used) {
3986                 lro_flush_all(&rx_ring->lro_mgr);
3987                 rx_ring->lro_used = 0;
3988         }
3989 #endif
3990
3991         if (cleaned_count)
3992                 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3993
3994         rx_ring->total_packets += total_packets;
3995         rx_ring->total_bytes += total_bytes;
3996         rx_ring->rx_stats.packets += total_packets;
3997         rx_ring->rx_stats.bytes += total_bytes;
3998         adapter->net_stats.rx_bytes += total_bytes;
3999         adapter->net_stats.rx_packets += total_packets;
4000         return cleaned;
4001 }
4002
4003
4004 /**
4005  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4006  * @adapter: address of board private structure
4007  **/
4008 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4009                                      int cleaned_count)
4010 {
4011         struct igb_adapter *adapter = rx_ring->adapter;
4012         struct net_device *netdev = adapter->netdev;
4013         struct pci_dev *pdev = adapter->pdev;
4014         union e1000_adv_rx_desc *rx_desc;
4015         struct igb_buffer *buffer_info;
4016         struct sk_buff *skb;
4017         unsigned int i;
4018
4019         i = rx_ring->next_to_use;
4020         buffer_info = &rx_ring->buffer_info[i];
4021
4022         while (cleaned_count--) {
4023                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4024
4025                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4026                         if (!buffer_info->page) {
4027                                 buffer_info->page = alloc_page(GFP_ATOMIC);
4028                                 if (!buffer_info->page) {
4029                                         adapter->alloc_rx_buff_failed++;
4030                                         goto no_buffers;
4031                                 }
4032                                 buffer_info->page_offset = 0;
4033                         } else {
4034                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
4035                         }
4036                         buffer_info->page_dma =
4037                                 pci_map_page(pdev,
4038                                              buffer_info->page,
4039                                              buffer_info->page_offset,
4040                                              PAGE_SIZE / 2,
4041                                              PCI_DMA_FROMDEVICE);
4042                 }
4043
4044                 if (!buffer_info->skb) {
4045                         int bufsz;
4046
4047                         if (adapter->rx_ps_hdr_size)
4048                                 bufsz = adapter->rx_ps_hdr_size;
4049                         else
4050                                 bufsz = adapter->rx_buffer_len;
4051                         bufsz += NET_IP_ALIGN;
4052                         skb = netdev_alloc_skb(netdev, bufsz);
4053
4054                         if (!skb) {
4055                                 adapter->alloc_rx_buff_failed++;
4056                                 goto no_buffers;
4057                         }
4058
4059                         /* Make buffer alignment 2 beyond a 16 byte boundary
4060                          * this will result in a 16 byte aligned IP header after
4061                          * the 14 byte MAC header is removed
4062                          */
4063                         skb_reserve(skb, NET_IP_ALIGN);
4064
4065                         buffer_info->skb = skb;
4066                         buffer_info->dma = pci_map_single(pdev, skb->data,
4067                                                           bufsz,
4068                                                           PCI_DMA_FROMDEVICE);
4069
4070                 }
4071                 /* Refresh the desc even if buffer_addrs didn't change because
4072                  * each write-back erases this info. */
4073                 if (adapter->rx_ps_hdr_size) {
4074                         rx_desc->read.pkt_addr =
4075                              cpu_to_le64(buffer_info->page_dma);
4076                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4077                 } else {
4078                         rx_desc->read.pkt_addr =
4079                              cpu_to_le64(buffer_info->dma);
4080                         rx_desc->read.hdr_addr = 0;
4081                 }
4082
4083                 i++;
4084                 if (i == rx_ring->count)
4085                         i = 0;
4086                 buffer_info = &rx_ring->buffer_info[i];
4087         }
4088
4089 no_buffers:
4090         if (rx_ring->next_to_use != i) {
4091                 rx_ring->next_to_use = i;
4092                 if (i == 0)
4093                         i = (rx_ring->count - 1);
4094                 else
4095                         i--;
4096
4097                 /* Force memory writes to complete before letting h/w
4098                  * know there are new descriptors to fetch.  (Only
4099                  * applicable for weak-ordered memory model archs,
4100                  * such as IA-64). */
4101                 wmb();
4102                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4103         }
4104 }
4105
4106 /**
4107  * igb_mii_ioctl -
4108  * @netdev:
4109  * @ifreq:
4110  * @cmd:
4111  **/
4112 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4113 {
4114         struct igb_adapter *adapter = netdev_priv(netdev);
4115         struct mii_ioctl_data *data = if_mii(ifr);
4116
4117         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4118                 return -EOPNOTSUPP;
4119
4120         switch (cmd) {
4121         case SIOCGMIIPHY:
4122                 data->phy_id = adapter->hw.phy.addr;
4123                 break;
4124         case SIOCGMIIREG:
4125                 if (!capable(CAP_NET_ADMIN))
4126                         return -EPERM;
4127                 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
4128                                                      data->reg_num
4129                                                      & 0x1F, &data->val_out))
4130                         return -EIO;
4131                 break;
4132         case SIOCSMIIREG:
4133         default:
4134                 return -EOPNOTSUPP;
4135         }
4136         return 0;
4137 }
4138
4139 /**
4140  * igb_ioctl -
4141  * @netdev:
4142  * @ifreq:
4143  * @cmd:
4144  **/
4145 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4146 {
4147         switch (cmd) {
4148         case SIOCGMIIPHY:
4149         case SIOCGMIIREG:
4150         case SIOCSMIIREG:
4151                 return igb_mii_ioctl(netdev, ifr, cmd);
4152         default:
4153                 return -EOPNOTSUPP;
4154         }
4155 }
4156
4157 static void igb_vlan_rx_register(struct net_device *netdev,
4158                                  struct vlan_group *grp)
4159 {
4160         struct igb_adapter *adapter = netdev_priv(netdev);
4161         struct e1000_hw *hw = &adapter->hw;
4162         u32 ctrl, rctl;
4163
4164         igb_irq_disable(adapter);
4165         adapter->vlgrp = grp;
4166
4167         if (grp) {
4168                 /* enable VLAN tag insert/strip */
4169                 ctrl = rd32(E1000_CTRL);
4170                 ctrl |= E1000_CTRL_VME;
4171                 wr32(E1000_CTRL, ctrl);
4172
4173                 /* enable VLAN receive filtering */
4174                 rctl = rd32(E1000_RCTL);
4175                 rctl &= ~E1000_RCTL_CFIEN;
4176                 wr32(E1000_RCTL, rctl);
4177                 igb_update_mng_vlan(adapter);
4178                 wr32(E1000_RLPML,
4179                                 adapter->max_frame_size + VLAN_TAG_SIZE);
4180         } else {
4181                 /* disable VLAN tag insert/strip */
4182                 ctrl = rd32(E1000_CTRL);
4183                 ctrl &= ~E1000_CTRL_VME;
4184                 wr32(E1000_CTRL, ctrl);
4185
4186                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4187                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4188                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4189                 }
4190                 wr32(E1000_RLPML,
4191                                 adapter->max_frame_size);
4192         }
4193
4194         if (!test_bit(__IGB_DOWN, &adapter->state))
4195                 igb_irq_enable(adapter);
4196 }
4197
4198 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4199 {
4200         struct igb_adapter *adapter = netdev_priv(netdev);
4201         struct e1000_hw *hw = &adapter->hw;
4202         u32 vfta, index;
4203
4204         if ((adapter->hw.mng_cookie.status &
4205              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4206             (vid == adapter->mng_vlan_id))
4207                 return;
4208         /* add VID to filter table */
4209         index = (vid >> 5) & 0x7F;
4210         vfta = array_rd32(E1000_VFTA, index);
4211         vfta |= (1 << (vid & 0x1F));
4212         igb_write_vfta(&adapter->hw, index, vfta);
4213 }
4214
4215 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4216 {
4217         struct igb_adapter *adapter = netdev_priv(netdev);
4218         struct e1000_hw *hw = &adapter->hw;
4219         u32 vfta, index;
4220
4221         igb_irq_disable(adapter);
4222         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4223
4224         if (!test_bit(__IGB_DOWN, &adapter->state))
4225                 igb_irq_enable(adapter);
4226
4227         if ((adapter->hw.mng_cookie.status &
4228              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4229             (vid == adapter->mng_vlan_id)) {
4230                 /* release control to f/w */
4231                 igb_release_hw_control(adapter);
4232                 return;
4233         }
4234
4235         /* remove VID from filter table */
4236         index = (vid >> 5) & 0x7F;
4237         vfta = array_rd32(E1000_VFTA, index);
4238         vfta &= ~(1 << (vid & 0x1F));
4239         igb_write_vfta(&adapter->hw, index, vfta);
4240 }
4241
4242 static void igb_restore_vlan(struct igb_adapter *adapter)
4243 {
4244         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4245
4246         if (adapter->vlgrp) {
4247                 u16 vid;
4248                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4249                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4250                                 continue;
4251                         igb_vlan_rx_add_vid(adapter->netdev, vid);
4252                 }
4253         }
4254 }
4255
4256 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
4257 {
4258         struct e1000_mac_info *mac = &adapter->hw.mac;
4259
4260         mac->autoneg = 0;
4261
4262         /* Fiber NICs only allow 1000 gbps Full duplex */
4263         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
4264                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4265                 dev_err(&adapter->pdev->dev,
4266                         "Unsupported Speed/Duplex configuration\n");
4267                 return -EINVAL;
4268         }
4269
4270         switch (spddplx) {
4271         case SPEED_10 + DUPLEX_HALF:
4272                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
4273                 break;
4274         case SPEED_10 + DUPLEX_FULL:
4275                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
4276                 break;
4277         case SPEED_100 + DUPLEX_HALF:
4278                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
4279                 break;
4280         case SPEED_100 + DUPLEX_FULL:
4281                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
4282                 break;
4283         case SPEED_1000 + DUPLEX_FULL:
4284                 mac->autoneg = 1;
4285                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
4286                 break;
4287         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4288         default:
4289                 dev_err(&adapter->pdev->dev,
4290                         "Unsupported Speed/Duplex configuration\n");
4291                 return -EINVAL;
4292         }
4293         return 0;
4294 }
4295
4296
4297 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
4298 {
4299         struct net_device *netdev = pci_get_drvdata(pdev);
4300         struct igb_adapter *adapter = netdev_priv(netdev);
4301         struct e1000_hw *hw = &adapter->hw;
4302         u32 ctrl, rctl, status;
4303         u32 wufc = adapter->wol;
4304 #ifdef CONFIG_PM
4305         int retval = 0;
4306 #endif
4307
4308         netif_device_detach(netdev);
4309
4310         if (netif_running(netdev))
4311                 igb_close(netdev);
4312
4313         igb_reset_interrupt_capability(adapter);
4314
4315         igb_free_queues(adapter);
4316
4317 #ifdef CONFIG_PM
4318         retval = pci_save_state(pdev);
4319         if (retval)
4320                 return retval;
4321 #endif
4322
4323         status = rd32(E1000_STATUS);
4324         if (status & E1000_STATUS_LU)
4325                 wufc &= ~E1000_WUFC_LNKC;
4326
4327         if (wufc) {
4328                 igb_setup_rctl(adapter);
4329                 igb_set_multi(netdev);
4330
4331                 /* turn on all-multi mode if wake on multicast is enabled */
4332                 if (wufc & E1000_WUFC_MC) {
4333                         rctl = rd32(E1000_RCTL);
4334                         rctl |= E1000_RCTL_MPE;
4335                         wr32(E1000_RCTL, rctl);
4336                 }
4337
4338                 ctrl = rd32(E1000_CTRL);
4339                 /* advertise wake from D3Cold */
4340                 #define E1000_CTRL_ADVD3WUC 0x00100000
4341                 /* phy power management enable */
4342                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4343                 ctrl |= E1000_CTRL_ADVD3WUC;
4344                 wr32(E1000_CTRL, ctrl);
4345
4346                 /* Allow time for pending master requests to run */
4347                 igb_disable_pcie_master(&adapter->hw);
4348
4349                 wr32(E1000_WUC, E1000_WUC_PME_EN);
4350                 wr32(E1000_WUFC, wufc);
4351         } else {
4352                 wr32(E1000_WUC, 0);
4353                 wr32(E1000_WUFC, 0);
4354         }
4355
4356         /* make sure adapter isn't asleep if manageability/wol is enabled */
4357         if (wufc || adapter->en_mng_pt) {
4358                 pci_enable_wake(pdev, PCI_D3hot, 1);
4359                 pci_enable_wake(pdev, PCI_D3cold, 1);
4360         } else {
4361                 igb_shutdown_fiber_serdes_link_82575(hw);
4362                 pci_enable_wake(pdev, PCI_D3hot, 0);
4363                 pci_enable_wake(pdev, PCI_D3cold, 0);
4364         }
4365
4366         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4367          * would have already happened in close and is redundant. */
4368         igb_release_hw_control(adapter);
4369
4370         pci_disable_device(pdev);
4371
4372         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4373
4374         return 0;
4375 }
4376
4377 #ifdef CONFIG_PM
4378 static int igb_resume(struct pci_dev *pdev)
4379 {
4380         struct net_device *netdev = pci_get_drvdata(pdev);
4381         struct igb_adapter *adapter = netdev_priv(netdev);
4382         struct e1000_hw *hw = &adapter->hw;
4383         u32 err;
4384
4385         pci_set_power_state(pdev, PCI_D0);
4386         pci_restore_state(pdev);
4387
4388         if (adapter->need_ioport)
4389                 err = pci_enable_device(pdev);
4390         else
4391                 err = pci_enable_device_mem(pdev);
4392         if (err) {
4393                 dev_err(&pdev->dev,
4394                         "igb: Cannot enable PCI device from suspend\n");
4395                 return err;
4396         }
4397         pci_set_master(pdev);
4398
4399         pci_enable_wake(pdev, PCI_D3hot, 0);
4400         pci_enable_wake(pdev, PCI_D3cold, 0);
4401
4402         igb_set_interrupt_capability(adapter);
4403
4404         if (igb_alloc_queues(adapter)) {
4405                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4406                 return -ENOMEM;
4407         }
4408
4409         /* e1000_power_up_phy(adapter); */
4410
4411         igb_reset(adapter);
4412         wr32(E1000_WUS, ~0);
4413
4414         if (netif_running(netdev)) {
4415                 err = igb_open(netdev);
4416                 if (err)
4417                         return err;
4418         }
4419
4420         netif_device_attach(netdev);
4421
4422         /* let the f/w know that the h/w is now under the control of the
4423          * driver. */
4424         igb_get_hw_control(adapter);
4425
4426         return 0;
4427 }
4428 #endif
4429
4430 static void igb_shutdown(struct pci_dev *pdev)
4431 {
4432         igb_suspend(pdev, PMSG_SUSPEND);
4433 }
4434
4435 #ifdef CONFIG_NET_POLL_CONTROLLER
4436 /*
4437  * Polling 'interrupt' - used by things like netconsole to send skbs
4438  * without having to re-enable interrupts. It's not called while
4439  * the interrupt routine is executing.
4440  */
4441 static void igb_netpoll(struct net_device *netdev)
4442 {
4443         struct igb_adapter *adapter = netdev_priv(netdev);
4444         int i;
4445         int work_done = 0;
4446
4447         igb_irq_disable(adapter);
4448         adapter->flags |= IGB_FLAG_IN_NETPOLL;
4449
4450         for (i = 0; i < adapter->num_tx_queues; i++)
4451                 igb_clean_tx_irq(&adapter->tx_ring[i]);
4452
4453         for (i = 0; i < adapter->num_rx_queues; i++)
4454                 igb_clean_rx_irq_adv(&adapter->rx_ring[i],
4455                                      &work_done,
4456                                      adapter->rx_ring[i].napi.weight);
4457
4458         adapter->flags &= ~IGB_FLAG_IN_NETPOLL;
4459         igb_irq_enable(adapter);
4460 }
4461 #endif /* CONFIG_NET_POLL_CONTROLLER */
4462
4463 /**
4464  * igb_io_error_detected - called when PCI error is detected
4465  * @pdev: Pointer to PCI device
4466  * @state: The current pci connection state
4467  *
4468  * This function is called after a PCI bus error affecting
4469  * this device has been detected.
4470  */
4471 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4472                                               pci_channel_state_t state)
4473 {
4474         struct net_device *netdev = pci_get_drvdata(pdev);
4475         struct igb_adapter *adapter = netdev_priv(netdev);
4476
4477         netif_device_detach(netdev);
4478
4479         if (netif_running(netdev))
4480                 igb_down(adapter);
4481         pci_disable_device(pdev);
4482
4483         /* Request a slot slot reset. */
4484         return PCI_ERS_RESULT_NEED_RESET;
4485 }
4486
4487 /**
4488  * igb_io_slot_reset - called after the pci bus has been reset.
4489  * @pdev: Pointer to PCI device
4490  *
4491  * Restart the card from scratch, as if from a cold-boot. Implementation
4492  * resembles the first-half of the igb_resume routine.
4493  */
4494 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4495 {
4496         struct net_device *netdev = pci_get_drvdata(pdev);
4497         struct igb_adapter *adapter = netdev_priv(netdev);
4498         struct e1000_hw *hw = &adapter->hw;
4499         int err;
4500
4501         if (adapter->need_ioport)
4502                 err = pci_enable_device(pdev);
4503         else
4504                 err = pci_enable_device_mem(pdev);
4505         if (err) {
4506                 dev_err(&pdev->dev,
4507                         "Cannot re-enable PCI device after reset.\n");
4508                 return PCI_ERS_RESULT_DISCONNECT;
4509         }
4510         pci_set_master(pdev);
4511         pci_restore_state(pdev);
4512
4513         pci_enable_wake(pdev, PCI_D3hot, 0);
4514         pci_enable_wake(pdev, PCI_D3cold, 0);
4515
4516         igb_reset(adapter);
4517         wr32(E1000_WUS, ~0);
4518
4519         return PCI_ERS_RESULT_RECOVERED;
4520 }
4521
4522 /**
4523  * igb_io_resume - called when traffic can start flowing again.
4524  * @pdev: Pointer to PCI device
4525  *
4526  * This callback is called when the error recovery driver tells us that
4527  * its OK to resume normal operation. Implementation resembles the
4528  * second-half of the igb_resume routine.
4529  */
4530 static void igb_io_resume(struct pci_dev *pdev)
4531 {
4532         struct net_device *netdev = pci_get_drvdata(pdev);
4533         struct igb_adapter *adapter = netdev_priv(netdev);
4534
4535         igb_init_manageability(adapter);
4536
4537         if (netif_running(netdev)) {
4538                 if (igb_up(adapter)) {
4539                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4540                         return;
4541                 }
4542         }
4543
4544         netif_device_attach(netdev);
4545
4546         /* let the f/w know that the h/w is now under the control of the
4547          * driver. */
4548         igb_get_hw_control(adapter);
4549
4550 }
4551
4552 /* igb_main.c */