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_82575EB_COPPER), board_82575 },
  65         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
  66         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
  67         /* required last entry */
  68         {0, }
  69 };
  70
  71 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
  72
  73 void igb_reset(struct igb_adapter *);
  74 static int igb_setup_all_tx_resources(struct igb_adapter *);
  75 static int igb_setup_all_rx_resources(struct igb_adapter *);
  76 static void igb_free_all_tx_resources(struct igb_adapter *);
  77 static void igb_free_all_rx_resources(struct igb_adapter *);
  78 static void igb_free_tx_resources(struct igb_ring *);
  79 static void igb_free_rx_resources(struct igb_ring *);
  80 void igb_update_stats(struct igb_adapter *);
  81 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
  82 static void __devexit igb_remove(struct pci_dev *pdev);
  83 static int igb_sw_init(struct igb_adapter *);
  84 static int igb_open(struct net_device *);
  85 static int igb_close(struct net_device *);
  86 static void igb_configure_tx(struct igb_adapter *);
  87 static void igb_configure_rx(struct igb_adapter *);
  88 static void igb_setup_rctl(struct igb_adapter *);
  89 static void igb_clean_all_tx_rings(struct igb_adapter *);
  90 static void igb_clean_all_rx_rings(struct igb_adapter *);
  91 static void igb_clean_tx_ring(struct igb_ring *);
  92 static void igb_clean_rx_ring(struct igb_ring *);
  93 static void igb_set_multi(struct net_device *);
  94 static void igb_update_phy_info(unsigned long);
  95 static void igb_watchdog(unsigned long);
  96 static void igb_watchdog_task(struct work_struct *);
  97 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
  98                                   struct igb_ring *);
  99 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
 100 static struct net_device_stats *igb_get_stats(struct net_device *);
 101 static int igb_change_mtu(struct net_device *, int);
 102 static int igb_set_mac(struct net_device *, void *);
 103 static irqreturn_t igb_intr(int irq, void *);
 104 static irqreturn_t igb_intr_msi(int irq, void *);
 105 static irqreturn_t igb_msix_other(int irq, void *);
 106 static irqreturn_t igb_msix_rx(int irq, void *);
 107 static irqreturn_t igb_msix_tx(int irq, void *);
 108 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
 109 #ifdef CONFIG_DCA
 110 static void igb_update_rx_dca(struct igb_ring *);
 111 static void igb_update_tx_dca(struct igb_ring *);
 112 static void igb_setup_dca(struct igb_adapter *);
 113 #endif /* CONFIG_DCA */
 114 static bool igb_clean_tx_irq(struct igb_ring *);
 115 static int igb_poll(struct napi_struct *, int);
 116 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
 117 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
 118 #ifdef CONFIG_IGB_LRO
 119 static int igb_get_skb_hdr(struct sk_buff *skb, void **, void **, u64 *, void *);
 120 #endif
 121 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 122 static void igb_tx_timeout(struct net_device *);
 123 static void igb_reset_task(struct work_struct *);
 124 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
 125 static void igb_vlan_rx_add_vid(struct net_device *, u16);
 126 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
 127 static void igb_restore_vlan(struct igb_adapter *);
 128
 129 static int igb_suspend(struct pci_dev *, pm_message_t);
 130 #ifdef CONFIG_PM
 131 static int igb_resume(struct pci_dev *);
 132 #endif
 133 static void igb_shutdown(struct pci_dev *);
 134 #ifdef CONFIG_DCA
 135 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 136 static struct notifier_block dca_notifier = {
 137         .notifier_call  = igb_notify_dca,
 138         .next           = NULL,
 139         .priority       = 0
 140 };
 141 #endif
 142
 143 #ifdef CONFIG_NET_POLL_CONTROLLER
 144 /* for netdump / net console */
 145 static void igb_netpoll(struct net_device *);
 146 #endif
 147
 148 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 149                      pci_channel_state_t);
 150 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 151 static void igb_io_resume(struct pci_dev *);
 152
 153 static struct pci_error_handlers igb_err_handler = {
 154         .error_detected = igb_io_error_detected,
 155         .slot_reset = igb_io_slot_reset,
 156         .resume = igb_io_resume,
 157 };
 158
 159
 160 static struct pci_driver igb_driver = {
 161         .name     = igb_driver_name,
 162         .id_table = igb_pci_tbl,
 163         .probe    = igb_probe,
 164         .remove   = __devexit_p(igb_remove),
 165 #ifdef CONFIG_PM
 166         /* Power Managment Hooks */
 167         .suspend  = igb_suspend,
 168         .resume   = igb_resume,
 169 #endif
 170         .shutdown = igb_shutdown,
 171         .err_handler = &igb_err_handler
 172 };
 173
 174 static int global_quad_port_a; /* global quad port a indication */
 175
 176 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 177 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 178 MODULE_LICENSE("GPL");
 179 MODULE_VERSION(DRV_VERSION);
 180
 181 #ifdef DEBUG
 182 /**
 183  * igb_get_hw_dev_name - return device name string
 184  * used by hardware layer to print debugging information
 185  **/
 186 char *igb_get_hw_dev_name(struct e1000_hw *hw)
 187 {
 188         struct igb_adapter *adapter = hw->back;
 189         return adapter->netdev->name;
 190 }
 191 #endif
 192
 193 /**
 194  * igb_init_module - Driver Registration Routine
 195  *
 196  * igb_init_module is the first routine called when the driver is
 197  * loaded. All it does is register with the PCI subsystem.
 198  **/
 199 static int __init igb_init_module(void)
 200 {
 201         int ret;
 202         printk(KERN_INFO "%s - version %s\n",
 203                igb_driver_string, igb_driver_version);
 204
 205         printk(KERN_INFO "%s\n", igb_copyright);
 206
 207         global_quad_port_a = 0;
 208
 209         ret = pci_register_driver(&igb_driver);
 210 #ifdef CONFIG_DCA
 211         dca_register_notify(&dca_notifier);
 212 #endif
 213         return ret;
 214 }
 215
 216 module_init(igb_init_module);
 217
 218 /**
 219  * igb_exit_module - Driver Exit Cleanup Routine
 220  *
 221  * igb_exit_module is called just before the driver is removed
 222  * from memory.
 223  **/
 224 static void __exit igb_exit_module(void)
 225 {
 226 #ifdef CONFIG_DCA
 227         dca_unregister_notify(&dca_notifier);
 228 #endif
 229         pci_unregister_driver(&igb_driver);
 230 }
 231
 232 module_exit(igb_exit_module);
 233
 234 /**
 235  * igb_alloc_queues - Allocate memory for all rings
 236  * @adapter: board private structure to initialize
 237  *
 238  * We allocate one ring per queue at run-time since we don't know the
 239  * number of queues at compile-time.
 240  **/
 241 static int igb_alloc_queues(struct igb_adapter *adapter)
 242 {
 243         int i;
 244
 245         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
 246                                    sizeof(struct igb_ring), GFP_KERNEL);
 247         if (!adapter->tx_ring)
 248                 return -ENOMEM;
 249
 250         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
 251                                    sizeof(struct igb_ring), GFP_KERNEL);
 252         if (!adapter->rx_ring) {
 253                 kfree(adapter->tx_ring);
 254                 return -ENOMEM;
 255         }
 256
 257         adapter->rx_ring->buddy = adapter->tx_ring;
 258
 259         for (i = 0; i < adapter->num_tx_queues; i++) {
 260                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 261                 ring->adapter = adapter;
 262                 ring->queue_index = i;
 263         }
 264         for (i = 0; i < adapter->num_rx_queues; i++) {
 265                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 266                 ring->adapter = adapter;
 267                 ring->queue_index = i;
 268                 ring->itr_register = E1000_ITR;
 269
 270                 /* set a default napi handler for each rx_ring */
 271                 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
 272         }
 273         return 0;
 274 }
 275
 276 static void igb_free_queues(struct igb_adapter *adapter)
 277 {
 278         int i;
 279
 280         for (i = 0; i < adapter->num_rx_queues; i++)
 281                 netif_napi_del(&adapter->rx_ring[i].napi);
 282
 283         kfree(adapter->tx_ring);
 284         kfree(adapter->rx_ring);
 285 }
 286
 287 #define IGB_N0_QUEUE -1
 288 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
 289                               int tx_queue, int msix_vector)
 290 {
 291         u32 msixbm = 0;
 292         struct e1000_hw *hw = &adapter->hw;
 293         u32 ivar, index;
 294
 295         switch (hw->mac.type) {
 296         case e1000_82575:
 297                 /* The 82575 assigns vectors using a bitmask, which matches the
 298                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
 299                    or more queues to a vector, we write the appropriate bits
 300                    into the MSIXBM register for that vector. */
 301                 if (rx_queue > IGB_N0_QUEUE) {
 302                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 303                         adapter->rx_ring[rx_queue].eims_value = msixbm;
 304                 }
 305                 if (tx_queue > IGB_N0_QUEUE) {
 306                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 307                         adapter->tx_ring[tx_queue].eims_value =
 308                                   E1000_EICR_TX_QUEUE0 << tx_queue;
 309                 }
 310                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 311                 break;
 312         case e1000_82576:
 313                 /* The 82576 uses a table-based method for assigning vectors.
 314                    Each queue has a single entry in the table to which we write
 315                    a vector number along with a "valid" bit.  Sadly, the layout
 316                    of the table is somewhat counterintuitive. */
 317                 if (rx_queue > IGB_N0_QUEUE) {
 318                         index = (rx_queue & 0x7);
 319                         ivar = array_rd32(E1000_IVAR0, index);
 320                         if (rx_queue < 8) {
 321                                 /* vector goes into low byte of register */
 322                                 ivar = ivar & 0xFFFFFF00;
 323                                 ivar |= msix_vector | E1000_IVAR_VALID;
 324                         } else {
 325                                 /* vector goes into third byte of register */
 326                                 ivar = ivar & 0xFF00FFFF;
 327                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 328                         }
 329                         adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
 330                         array_wr32(E1000_IVAR0, index, ivar);
 331                 }
 332                 if (tx_queue > IGB_N0_QUEUE) {
 333                         index = (tx_queue & 0x7);
 334                         ivar = array_rd32(E1000_IVAR0, index);
 335                         if (tx_queue < 8) {
 336                                 /* vector goes into second byte of register */
 337                                 ivar = ivar & 0xFFFF00FF;
 338                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 339                         } else {
 340                                 /* vector goes into high byte of register */
 341                                 ivar = ivar & 0x00FFFFFF;
 342                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 343                         }
 344                         adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
 345                         array_wr32(E1000_IVAR0, index, ivar);
 346                 }
 347                 break;
 348         default:
 349                 BUG();
 350                 break;
 351         }
 352 }
 353
 354 /**
 355  * igb_configure_msix - Configure MSI-X hardware
 356  *
 357  * igb_configure_msix sets up the hardware to properly
 358  * generate MSI-X interrupts.
 359  **/
 360 static void igb_configure_msix(struct igb_adapter *adapter)
 361 {
 362         u32 tmp;
 363         int i, vector = 0;
 364         struct e1000_hw *hw = &adapter->hw;
 365
 366         adapter->eims_enable_mask = 0;
 367         if (hw->mac.type == e1000_82576)
 368                 /* Turn on MSI-X capability first, or our settings
 369                  * won't stick.  And it will take days to debug. */
 370                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 371                                    E1000_GPIE_PBA | E1000_GPIE_EIAME |
 372                                    E1000_GPIE_NSICR);
 373
 374         for (i = 0; i < adapter->num_tx_queues; i++) {
 375                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
 376                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
 377                 adapter->eims_enable_mask |= tx_ring->eims_value;
 378                 if (tx_ring->itr_val)
 379                         writel(tx_ring->itr_val,
 380                                hw->hw_addr + tx_ring->itr_register);
 381                 else
 382                         writel(1, hw->hw_addr + tx_ring->itr_register);
 383         }
 384
 385         for (i = 0; i < adapter->num_rx_queues; i++) {
 386                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
 387                 rx_ring->buddy = NULL;
 388                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
 389                 adapter->eims_enable_mask |= rx_ring->eims_value;
 390                 if (rx_ring->itr_val)
 391                         writel(rx_ring->itr_val,
 392                                hw->hw_addr + rx_ring->itr_register);
 393                 else
 394                         writel(1, hw->hw_addr + rx_ring->itr_register);
 395         }
 396
 397
 398         /* set vector for other causes, i.e. link changes */
 399         switch (hw->mac.type) {
 400         case e1000_82575:
 401                 array_wr32(E1000_MSIXBM(0), vector++,
 402                                       E1000_EIMS_OTHER);
 403
 404                 tmp = rd32(E1000_CTRL_EXT);
 405                 /* enable MSI-X PBA support*/
 406                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 407
 408                 /* Auto-Mask interrupts upon ICR read. */
 409                 tmp |= E1000_CTRL_EXT_EIAME;
 410                 tmp |= E1000_CTRL_EXT_IRCA;
 411
 412                 wr32(E1000_CTRL_EXT, tmp);
 413                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
 414                 adapter->eims_other = E1000_EIMS_OTHER;
 415
 416                 break;
 417
 418         case e1000_82576:
 419                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 420                 wr32(E1000_IVAR_MISC, tmp);
 421
 422                 adapter->eims_enable_mask = (1 << (vector)) - 1;
 423                 adapter->eims_other = 1 << (vector - 1);
 424                 break;
 425         default:
 426                 /* do nothing, since nothing else supports MSI-X */
 427                 break;
 428         } /* switch (hw->mac.type) */
 429         wrfl();
 430 }
 431
 432 /**
 433  * igb_request_msix - Initialize MSI-X interrupts
 434  *
 435  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
 436  * kernel.
 437  **/
 438 static int igb_request_msix(struct igb_adapter *adapter)
 439 {
 440         struct net_device *netdev = adapter->netdev;
 441         int i, err = 0, vector = 0;
 442
 443         vector = 0;
 444
 445         for (i = 0; i < adapter->num_tx_queues; i++) {
 446                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 447                 sprintf(ring->name, "%s-tx%d", netdev->name, i);
 448                 err = request_irq(adapter->msix_entries[vector].vector,
 449                                   &igb_msix_tx, 0, ring->name,
 450                                   &(adapter->tx_ring[i]));
 451                 if (err)
 452                         goto out;
 453                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 454                 ring->itr_val = 976; /* ~4000 ints/sec */
 455                 vector++;
 456         }
 457         for (i = 0; i < adapter->num_rx_queues; i++) {
 458                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 459                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
 460                         sprintf(ring->name, "%s-rx%d", netdev->name, i);
 461                 else
 462                         memcpy(ring->name, netdev->name, IFNAMSIZ);
 463                 err = request_irq(adapter->msix_entries[vector].vector,
 464                                   &igb_msix_rx, 0, ring->name,
 465                                   &(adapter->rx_ring[i]));
 466                 if (err)
 467                         goto out;
 468                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 469                 ring->itr_val = adapter->itr;
 470                 /* overwrite the poll routine for MSIX, we've already done
 471                  * netif_napi_add */
 472                 ring->napi.poll = &igb_clean_rx_ring_msix;
 473                 vector++;
 474         }
 475
 476         err = request_irq(adapter->msix_entries[vector].vector,
 477                           &igb_msix_other, 0, netdev->name, netdev);
 478         if (err)
 479                 goto out;
 480
 481         igb_configure_msix(adapter);
 482         return 0;
 483 out:
 484         return err;
 485 }
 486
 487 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 488 {
 489         if (adapter->msix_entries) {
 490                 pci_disable_msix(adapter->pdev);
 491                 kfree(adapter->msix_entries);
 492                 adapter->msix_entries = NULL;
 493         } else if (adapter->flags & IGB_FLAG_HAS_MSI)
 494                 pci_disable_msi(adapter->pdev);
 495         return;
 496 }
 497
 498
 499 /**
 500  * igb_set_interrupt_capability - set MSI or MSI-X if supported
 501  *
 502  * Attempt to configure interrupts using the best available
 503  * capabilities of the hardware and kernel.
 504  **/
 505 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
 506 {
 507         int err;
 508         int numvecs, i;
 509
 510         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
 511         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
 512                                         GFP_KERNEL);
 513         if (!adapter->msix_entries)
 514                 goto msi_only;
 515
 516         for (i = 0; i < numvecs; i++)
 517                 adapter->msix_entries[i].entry = i;
 518
 519         err = pci_enable_msix(adapter->pdev,
 520                               adapter->msix_entries,
 521                               numvecs);
 522         if (err == 0)
 523                 goto out;
 524
 525         igb_reset_interrupt_capability(adapter);
 526
 527         /* If we can't do MSI-X, try MSI */
 528 msi_only:
 529         adapter->num_rx_queues = 1;
 530         adapter->num_tx_queues = 1;
 531         if (!pci_enable_msi(adapter->pdev))
 532                 adapter->flags |= IGB_FLAG_HAS_MSI;
 533 out:
 534         /* Notify the stack of the (possibly) reduced Tx Queue count. */
 535         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
 536         return;
 537 }
 538
 539 /**
 540  * igb_request_irq - initialize interrupts
 541  *
 542  * Attempts to configure interrupts using the best available
 543  * capabilities of the hardware and kernel.
 544  **/
 545 static int igb_request_irq(struct igb_adapter *adapter)
 546 {
 547         struct net_device *netdev = adapter->netdev;
 548         struct e1000_hw *hw = &adapter->hw;
 549         int err = 0;
 550
 551         if (adapter->msix_entries) {
 552                 err = igb_request_msix(adapter);
 553                 if (!err)
 554                         goto request_done;
 555                 /* fall back to MSI */
 556                 igb_reset_interrupt_capability(adapter);
 557                 if (!pci_enable_msi(adapter->pdev))
 558                         adapter->flags |= IGB_FLAG_HAS_MSI;
 559                 igb_free_all_tx_resources(adapter);
 560                 igb_free_all_rx_resources(adapter);
 561                 adapter->num_rx_queues = 1;
 562                 igb_alloc_queues(adapter);
 563         } else {
 564                 switch (hw->mac.type) {
 565                 case e1000_82575:
 566                         wr32(E1000_MSIXBM(0),
 567                              (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
 568                         break;
 569                 case e1000_82576:
 570                         wr32(E1000_IVAR0, E1000_IVAR_VALID);
 571                         break;
 572                 default:
 573                         break;
 574                 }
 575         }
 576
 577         if (adapter->flags & IGB_FLAG_HAS_MSI) {
 578                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
 579                                   netdev->name, netdev);
 580                 if (!err)
 581                         goto request_done;
 582                 /* fall back to legacy interrupts */
 583                 igb_reset_interrupt_capability(adapter);
 584                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
 585         }
 586
 587         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
 588                           netdev->name, netdev);
 589
 590         if (err)
 591                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
 592                         err);
 593
 594 request_done:
 595         return err;
 596 }
 597
 598 static void igb_free_irq(struct igb_adapter *adapter)
 599 {
 600         struct net_device *netdev = adapter->netdev;
 601
 602         if (adapter->msix_entries) {
 603                 int vector = 0, i;
 604
 605                 for (i = 0; i < adapter->num_tx_queues; i++)
 606                         free_irq(adapter->msix_entries[vector++].vector,
 607                                 &(adapter->tx_ring[i]));
 608                 for (i = 0; i < adapter->num_rx_queues; i++)
 609                         free_irq(adapter->msix_entries[vector++].vector,
 610                                 &(adapter->rx_ring[i]));
 611
 612                 free_irq(adapter->msix_entries[vector++].vector, netdev);
 613                 return;
 614         }
 615
 616         free_irq(adapter->pdev->irq, netdev);
 617 }
 618
 619 /**
 620  * igb_irq_disable - Mask off interrupt generation on the NIC
 621  * @adapter: board private structure
 622  **/
 623 static void igb_irq_disable(struct igb_adapter *adapter)
 624 {
 625         struct e1000_hw *hw = &adapter->hw;
 626
 627         if (adapter->msix_entries) {
 628                 wr32(E1000_EIAM, 0);
 629                 wr32(E1000_EIMC, ~0);
 630                 wr32(E1000_EIAC, 0);
 631         }
 632
 633         wr32(E1000_IAM, 0);
 634         wr32(E1000_IMC, ~0);
 635         wrfl();
 636         synchronize_irq(adapter->pdev->irq);
 637 }
 638
 639 /**
 640  * igb_irq_enable - Enable default interrupt generation settings
 641  * @adapter: board private structure
 642  **/
 643 static void igb_irq_enable(struct igb_adapter *adapter)
 644 {
 645         struct e1000_hw *hw = &adapter->hw;
 646
 647         if (adapter->msix_entries) {
 648                 wr32(E1000_EIAC, adapter->eims_enable_mask);
 649                 wr32(E1000_EIAM, adapter->eims_enable_mask);
 650                 wr32(E1000_EIMS, adapter->eims_enable_mask);
 651                 wr32(E1000_IMS, E1000_IMS_LSC);
 652         } else {
 653                 wr32(E1000_IMS, IMS_ENABLE_MASK);
 654                 wr32(E1000_IAM, IMS_ENABLE_MASK);
 655         }
 656 }
 657
 658 static void igb_update_mng_vlan(struct igb_adapter *adapter)
 659 {
 660         struct net_device *netdev = adapter->netdev;
 661         u16 vid = adapter->hw.mng_cookie.vlan_id;
 662         u16 old_vid = adapter->mng_vlan_id;
 663         if (adapter->vlgrp) {
 664                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
 665                         if (adapter->hw.mng_cookie.status &
 666                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
 667                                 igb_vlan_rx_add_vid(netdev, vid);
 668                                 adapter->mng_vlan_id = vid;
 669                         } else
 670                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
 671
 672                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
 673                                         (vid != old_vid) &&
 674                             !vlan_group_get_device(adapter->vlgrp, old_vid))
 675                                 igb_vlan_rx_kill_vid(netdev, old_vid);
 676                 } else
 677                         adapter->mng_vlan_id = vid;
 678         }
 679 }
 680
 681 /**
 682  * igb_release_hw_control - release control of the h/w to f/w
 683  * @adapter: address of board private structure
 684  *
 685  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 686  * For ASF and Pass Through versions of f/w this means that the
 687  * driver is no longer loaded.
 688  *
 689  **/
 690 static void igb_release_hw_control(struct igb_adapter *adapter)
 691 {
 692         struct e1000_hw *hw = &adapter->hw;
 693         u32 ctrl_ext;
 694
 695         /* Let firmware take over control of h/w */
 696         ctrl_ext = rd32(E1000_CTRL_EXT);
 697         wr32(E1000_CTRL_EXT,
 698                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 699 }
 700
 701
 702 /**
 703  * igb_get_hw_control - get control of the h/w from f/w
 704  * @adapter: address of board private structure
 705  *
 706  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 707  * For ASF and Pass Through versions of f/w this means that
 708  * the driver is loaded.
 709  *
 710  **/
 711 static void igb_get_hw_control(struct igb_adapter *adapter)
 712 {
 713         struct e1000_hw *hw = &adapter->hw;
 714         u32 ctrl_ext;
 715
 716         /* Let firmware know the driver has taken over */
 717         ctrl_ext = rd32(E1000_CTRL_EXT);
 718         wr32(E1000_CTRL_EXT,
 719                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 720 }
 721
 722 /**
 723  * igb_configure - configure the hardware for RX and TX
 724  * @adapter: private board structure
 725  **/
 726 static void igb_configure(struct igb_adapter *adapter)
 727 {
 728         struct net_device *netdev = adapter->netdev;
 729         int i;
 730
 731         igb_get_hw_control(adapter);
 732         igb_set_multi(netdev);
 733
 734         igb_restore_vlan(adapter);
 735
 736         igb_configure_tx(adapter);
 737         igb_setup_rctl(adapter);
 738         igb_configure_rx(adapter);
 739
 740         igb_rx_fifo_flush_82575(&adapter->hw);
 741
 742         /* call IGB_DESC_UNUSED which always leaves
 743          * at least 1 descriptor unused to make sure
 744          * next_to_use != next_to_clean */
 745         for (i = 0; i < adapter->num_rx_queues; i++) {
 746                 struct igb_ring *ring = &adapter->rx_ring[i];
 747                 igb_alloc_rx_buffers_adv(ring, IGB_DESC_UNUSED(ring));
 748         }
 749
 750
 751         adapter->tx_queue_len = netdev->tx_queue_len;
 752 }
 753
 754
 755 /**
 756  * igb_up - Open the interface and prepare it to handle traffic
 757  * @adapter: board private structure
 758  **/
 759
 760 int igb_up(struct igb_adapter *adapter)
 761 {
 762         struct e1000_hw *hw = &adapter->hw;
 763         int i;
 764
 765         /* hardware has been reset, we need to reload some things */
 766         igb_configure(adapter);
 767
 768         clear_bit(__IGB_DOWN, &adapter->state);
 769
 770         for (i = 0; i < adapter->num_rx_queues; i++)
 771                 napi_enable(&adapter->rx_ring[i].napi);
 772         if (adapter->msix_entries)
 773                 igb_configure_msix(adapter);
 774
 775         /* Clear any pending interrupts. */
 776         rd32(E1000_ICR);
 777         igb_irq_enable(adapter);
 778
 779         /* Fire a link change interrupt to start the watchdog. */
 780         wr32(E1000_ICS, E1000_ICS_LSC);
 781         return 0;
 782 }
 783
 784 void igb_down(struct igb_adapter *adapter)
 785 {
 786         struct e1000_hw *hw = &adapter->hw;
 787         struct net_device *netdev = adapter->netdev;
 788         u32 tctl, rctl;
 789         int i;
 790
 791         /* signal that we're down so the interrupt handler does not
 792          * reschedule our watchdog timer */
 793         set_bit(__IGB_DOWN, &adapter->state);
 794
 795         /* disable receives in the hardware */
 796         rctl = rd32(E1000_RCTL);
 797         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
 798         /* flush and sleep below */
 799
 800         netif_tx_stop_all_queues(netdev);
 801
 802         /* disable transmits in the hardware */
 803         tctl = rd32(E1000_TCTL);
 804         tctl &= ~E1000_TCTL_EN;
 805         wr32(E1000_TCTL, tctl);
 806         /* flush both disables and wait for them to finish */
 807         wrfl();
 808         msleep(10);
 809
 810         for (i = 0; i < adapter->num_rx_queues; i++)
 811                 napi_disable(&adapter->rx_ring[i].napi);
 812
 813         igb_irq_disable(adapter);
 814
 815         del_timer_sync(&adapter->watchdog_timer);
 816         del_timer_sync(&adapter->phy_info_timer);
 817
 818         netdev->tx_queue_len = adapter->tx_queue_len;
 819         netif_carrier_off(netdev);
 820         adapter->link_speed = 0;
 821         adapter->link_duplex = 0;
 822
 823         if (!pci_channel_offline(adapter->pdev))
 824                 igb_reset(adapter);
 825         igb_clean_all_tx_rings(adapter);
 826         igb_clean_all_rx_rings(adapter);
 827 }
 828
 829 void igb_reinit_locked(struct igb_adapter *adapter)
 830 {
 831         WARN_ON(in_interrupt());
 832         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 833                 msleep(1);
 834         igb_down(adapter);
 835         igb_up(adapter);
 836         clear_bit(__IGB_RESETTING, &adapter->state);
 837 }
 838
 839 void igb_reset(struct igb_adapter *adapter)
 840 {
 841         struct e1000_hw *hw = &adapter->hw;
 842         struct e1000_mac_info *mac = &hw->mac;
 843         struct e1000_fc_info *fc = &hw->fc;
 844         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
 845         u16 hwm;
 846
 847         /* Repartition Pba for greater than 9k mtu
 848          * To take effect CTRL.RST is required.
 849          */
 850         if (mac->type != e1000_82576) {
 851         pba = E1000_PBA_34K;
 852         }
 853         else {
 854                 pba = E1000_PBA_64K;
 855         }
 856
 857         if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
 858             (mac->type < e1000_82576)) {
 859                 /* adjust PBA for jumbo frames */
 860                 wr32(E1000_PBA, pba);
 861
 862                 /* To maintain wire speed transmits, the Tx FIFO should be
 863                  * large enough to accommodate two full transmit packets,
 864                  * rounded up to the next 1KB and expressed in KB.  Likewise,
 865                  * the Rx FIFO should be large enough to accommodate at least
 866                  * one full receive packet and is similarly rounded up and
 867                  * expressed in KB. */
 868                 pba = rd32(E1000_PBA);
 869                 /* upper 16 bits has Tx packet buffer allocation size in KB */
 870                 tx_space = pba >> 16;
 871                 /* lower 16 bits has Rx packet buffer allocation size in KB */
 872                 pba &= 0xffff;
 873                 /* the tx fifo also stores 16 bytes of information about the tx
 874                  * but don't include ethernet FCS because hardware appends it */
 875                 min_tx_space = (adapter->max_frame_size +
 876                                 sizeof(struct e1000_tx_desc) -
 877                                 ETH_FCS_LEN) * 2;
 878                 min_tx_space = ALIGN(min_tx_space, 1024);
 879                 min_tx_space >>= 10;
 880                 /* software strips receive CRC, so leave room for it */
 881                 min_rx_space = adapter->max_frame_size;
 882                 min_rx_space = ALIGN(min_rx_space, 1024);
 883                 min_rx_space >>= 10;
 884
 885                 /* If current Tx allocation is less than the min Tx FIFO size,
 886                  * and the min Tx FIFO size is less than the current Rx FIFO
 887                  * allocation, take space away from current Rx allocation */
 888                 if (tx_space < min_tx_space &&
 889                     ((min_tx_space - tx_space) < pba)) {
 890                         pba = pba - (min_tx_space - tx_space);
 891
 892                         /* if short on rx space, rx wins and must trump tx
 893                          * adjustment */
 894                         if (pba < min_rx_space)
 895                                 pba = min_rx_space;
 896                 }
 897                 wr32(E1000_PBA, pba);
 898         }
 899
 900         /* flow control settings */
 901         /* The high water mark must be low enough to fit one full frame
 902          * (or the size used for early receive) above it in the Rx FIFO.
 903          * Set it to the lower of:
 904          * - 90% of the Rx FIFO size, or
 905          * - the full Rx FIFO size minus one full frame */
 906         hwm = min(((pba << 10) * 9 / 10),
 907                         ((pba << 10) - 2 * adapter->max_frame_size));
 908
 909         if (mac->type < e1000_82576) {
 910                 fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
 911                 fc->low_water = fc->high_water - 8;
 912         } else {
 913                 fc->high_water = hwm & 0xFFF0;  /* 16-byte granularity */
 914                 fc->low_water = fc->high_water - 16;
 915         }
 916         fc->pause_time = 0xFFFF;
 917         fc->send_xon = 1;
 918         fc->type = fc->original_type;
 919
 920         /* Allow time for pending master requests to run */
 921         adapter->hw.mac.ops.reset_hw(&adapter->hw);
 922         wr32(E1000_WUC, 0);
 923
 924         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
 925                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
 926
 927         igb_update_mng_vlan(adapter);
 928
 929         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 930         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
 931
 932         igb_reset_adaptive(&adapter->hw);
 933         if (adapter->hw.phy.ops.get_phy_info)
 934                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
 935 }
 936
 937 /**
 938  * igb_is_need_ioport - determine if an adapter needs ioport resources or not
 939  * @pdev: PCI device information struct
 940  *
 941  * Returns true if an adapter needs ioport resources
 942  **/
 943 static int igb_is_need_ioport(struct pci_dev *pdev)
 944 {
 945         switch (pdev->device) {
 946         /* Currently there are no adapters that need ioport resources */
 947         default:
 948                 return false;
 949         }
 950 }
 951
 952 /**
 953  * igb_probe - Device Initialization Routine
 954  * @pdev: PCI device information struct
 955  * @ent: entry in igb_pci_tbl
 956  *
 957  * Returns 0 on success, negative on failure
 958  *
 959  * igb_probe initializes an adapter identified by a pci_dev structure.
 960  * The OS initialization, configuring of the adapter private structure,
 961  * and a hardware reset occur.
 962  **/
 963 static int __devinit igb_probe(struct pci_dev *pdev,
 964                                const struct pci_device_id *ent)
 965 {
 966         struct net_device *netdev;
 967         struct igb_adapter *adapter;
 968         struct e1000_hw *hw;
 969         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
 970         unsigned long mmio_start, mmio_len;
 971         int i, err, pci_using_dac;
 972         u16 eeprom_data = 0;
 973         u16 eeprom_apme_mask = IGB_EEPROM_APME;
 974         u32 part_num;
 975         int bars, need_ioport;
 976
 977         /* do not allocate ioport bars when not needed */
 978         need_ioport = igb_is_need_ioport(pdev);
 979         if (need_ioport) {
 980                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
 981                 err = pci_enable_device(pdev);
 982         } else {
 983                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
 984                 err = pci_enable_device_mem(pdev);
 985         }
 986         if (err)
 987                 return err;
 988
 989         pci_using_dac = 0;
 990         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
 991         if (!err) {
 992                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
 993                 if (!err)
 994                         pci_using_dac = 1;
 995         } else {
 996                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
 997                 if (err) {
 998                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
 999                         if (err) {
1000                                 dev_err(&pdev->dev, "No usable DMA "
1001                                         "configuration, aborting\n");
1002                                 goto err_dma;
1003                         }
1004                 }
1005         }
1006
1007         err = pci_request_selected_regions(pdev, bars, igb_driver_name);
1008         if (err)
1009                 goto err_pci_reg;
1010
1011         pci_set_master(pdev);
1012         pci_save_state(pdev);
1013
1014         err = -ENOMEM;
1015         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter), IGB_MAX_TX_QUEUES);
1016         if (!netdev)
1017                 goto err_alloc_etherdev;
1018
1019         SET_NETDEV_DEV(netdev, &pdev->dev);
1020
1021         pci_set_drvdata(pdev, netdev);
1022         adapter = netdev_priv(netdev);
1023         adapter->netdev = netdev;
1024         adapter->pdev = pdev;
1025         hw = &adapter->hw;
1026         hw->back = adapter;
1027         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1028         adapter->bars = bars;
1029         adapter->need_ioport = need_ioport;
1030
1031         mmio_start = pci_resource_start(pdev, 0);
1032         mmio_len = pci_resource_len(pdev, 0);
1033
1034         err = -EIO;
1035         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
1036         if (!adapter->hw.hw_addr)
1037                 goto err_ioremap;
1038
1039         netdev->open = &igb_open;
1040         netdev->stop = &igb_close;
1041         netdev->get_stats = &igb_get_stats;
1042         netdev->set_multicast_list = &igb_set_multi;
1043         netdev->set_mac_address = &igb_set_mac;
1044         netdev->change_mtu = &igb_change_mtu;
1045         netdev->do_ioctl = &igb_ioctl;
1046         igb_set_ethtool_ops(netdev);
1047         netdev->tx_timeout = &igb_tx_timeout;
1048         netdev->watchdog_timeo = 5 * HZ;
1049         netdev->vlan_rx_register = igb_vlan_rx_register;
1050         netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
1051         netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
1052 #ifdef CONFIG_NET_POLL_CONTROLLER
1053         netdev->poll_controller = igb_netpoll;
1054 #endif
1055         netdev->hard_start_xmit = &igb_xmit_frame_adv;
1056
1057         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1058
1059         netdev->mem_start = mmio_start;
1060         netdev->mem_end = mmio_start + mmio_len;
1061
1062         /* PCI config space info */
1063         hw->vendor_id = pdev->vendor;
1064         hw->device_id = pdev->device;
1065         hw->revision_id = pdev->revision;
1066         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1067         hw->subsystem_device_id = pdev->subsystem_device;
1068
1069         /* setup the private structure */
1070         hw->back = adapter;
1071         /* Copy the default MAC, PHY and NVM function pointers */
1072         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1073         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1074         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1075         /* Initialize skew-specific constants */
1076         err = ei->get_invariants(hw);
1077         if (err)
1078                 goto err_hw_init;
1079
1080         err = igb_sw_init(adapter);
1081         if (err)
1082                 goto err_sw_init;
1083
1084         igb_get_bus_info_pcie(hw);
1085
1086         /* set flags */
1087         switch (hw->mac.type) {
1088         case e1000_82576:
1089         case e1000_82575:
1090                 adapter->flags |= IGB_FLAG_HAS_DCA;
1091                 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1092                 break;
1093         default:
1094                 break;
1095         }
1096
1097         hw->phy.autoneg_wait_to_complete = false;
1098         hw->mac.adaptive_ifs = true;
1099
1100         /* Copper options */
1101         if (hw->phy.media_type == e1000_media_type_copper) {
1102                 hw->phy.mdix = AUTO_ALL_MODES;
1103                 hw->phy.disable_polarity_correction = false;
1104                 hw->phy.ms_type = e1000_ms_hw_default;
1105         }
1106
1107         if (igb_check_reset_block(hw))
1108                 dev_info(&pdev->dev,
1109                         "PHY reset is blocked due to SOL/IDER session.\n");
1110
1111         netdev->features = NETIF_F_SG |
1112                            NETIF_F_HW_CSUM |
1113                            NETIF_F_HW_VLAN_TX |
1114                            NETIF_F_HW_VLAN_RX |
1115                            NETIF_F_HW_VLAN_FILTER;
1116
1117         netdev->features |= NETIF_F_TSO;
1118         netdev->features |= NETIF_F_TSO6;
1119
1120 #ifdef CONFIG_IGB_LRO
1121         netdev->features |= NETIF_F_LRO;
1122 #endif
1123
1124         netdev->vlan_features |= NETIF_F_TSO;
1125         netdev->vlan_features |= NETIF_F_TSO6;
1126         netdev->vlan_features |= NETIF_F_HW_CSUM;
1127         netdev->vlan_features |= NETIF_F_SG;
1128
1129         if (pci_using_dac)
1130                 netdev->features |= NETIF_F_HIGHDMA;
1131
1132         netdev->features |= NETIF_F_LLTX;
1133         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1134
1135         /* before reading the NVM, reset the controller to put the device in a
1136          * known good starting state */
1137         hw->mac.ops.reset_hw(hw);
1138
1139         /* make sure the NVM is good */
1140         if (igb_validate_nvm_checksum(hw) < 0) {
1141                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1142                 err = -EIO;
1143                 goto err_eeprom;
1144         }
1145
1146         /* copy the MAC address out of the NVM */
1147         if (hw->mac.ops.read_mac_addr(hw))
1148                 dev_err(&pdev->dev, "NVM Read Error\n");
1149
1150         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1151         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1152
1153         if (!is_valid_ether_addr(netdev->perm_addr)) {
1154                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1155                 err = -EIO;
1156                 goto err_eeprom;
1157         }
1158
1159         init_timer(&adapter->watchdog_timer);
1160         adapter->watchdog_timer.function = &igb_watchdog;
1161         adapter->watchdog_timer.data = (unsigned long) adapter;
1162
1163         init_timer(&adapter->phy_info_timer);
1164         adapter->phy_info_timer.function = &igb_update_phy_info;
1165         adapter->phy_info_timer.data = (unsigned long) adapter;
1166
1167         INIT_WORK(&adapter->reset_task, igb_reset_task);
1168         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1169
1170         /* Initialize link & ring properties that are user-changeable */
1171         adapter->tx_ring->count = 256;
1172         for (i = 0; i < adapter->num_tx_queues; i++)
1173                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1174         adapter->rx_ring->count = 256;
1175         for (i = 0; i < adapter->num_rx_queues; i++)
1176                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1177
1178         adapter->fc_autoneg = true;
1179         hw->mac.autoneg = true;
1180         hw->phy.autoneg_advertised = 0x2f;
1181
1182         hw->fc.original_type = e1000_fc_default;
1183         hw->fc.type = e1000_fc_default;
1184
1185         adapter->itr_setting = 3;
1186         adapter->itr = IGB_START_ITR;
1187
1188         igb_validate_mdi_setting(hw);
1189
1190         adapter->rx_csum = 1;
1191
1192         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1193          * enable the ACPI Magic Packet filter
1194          */
1195
1196         if (hw->bus.func == 0 ||
1197             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1198                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1199                                      &eeprom_data);
1200
1201         if (eeprom_data & eeprom_apme_mask)
1202                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1203
1204         /* now that we have the eeprom settings, apply the special cases where
1205          * the eeprom may be wrong or the board simply won't support wake on
1206          * lan on a particular port */
1207         switch (pdev->device) {
1208         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1209                 adapter->eeprom_wol = 0;
1210                 break;
1211         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1212         case E1000_DEV_ID_82576_FIBER:
1213         case E1000_DEV_ID_82576_SERDES:
1214                 /* Wake events only supported on port A for dual fiber
1215                  * regardless of eeprom setting */
1216                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1217                         adapter->eeprom_wol = 0;
1218                 break;
1219         }
1220
1221         /* initialize the wol settings based on the eeprom settings */
1222         adapter->wol = adapter->eeprom_wol;
1223
1224         /* reset the hardware with the new settings */
1225         igb_reset(adapter);
1226
1227         /* let the f/w know that the h/w is now under the control of the
1228          * driver. */
1229         igb_get_hw_control(adapter);
1230
1231         /* tell the stack to leave us alone until igb_open() is called */
1232         netif_carrier_off(netdev);
1233         netif_tx_stop_all_queues(netdev);
1234
1235         strcpy(netdev->name, "eth%d");
1236         err = register_netdev(netdev);
1237         if (err)
1238                 goto err_register;
1239
1240 #ifdef CONFIG_DCA
1241         if ((adapter->flags & IGB_FLAG_HAS_DCA) &&
1242             (dca_add_requester(&pdev->dev) == 0)) {
1243                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1244                 dev_info(&pdev->dev, "DCA enabled\n");
1245                 /* Always use CB2 mode, difference is masked
1246                  * in the CB driver. */
1247                 wr32(E1000_DCA_CTRL, 2);
1248                 igb_setup_dca(adapter);
1249         }
1250 #endif
1251
1252         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1253         /* print bus type/speed/width info */
1254         dev_info(&pdev->dev,
1255                  "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1256                  netdev->name,
1257                  ((hw->bus.speed == e1000_bus_speed_2500)
1258                   ? "2.5Gb/s" : "unknown"),
1259                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1260                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1261                   ? "Width x1" : "unknown"),
1262                  netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1263                  netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1264
1265         igb_read_part_num(hw, &part_num);
1266         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1267                 (part_num >> 8), (part_num & 0xff));
1268
1269         dev_info(&pdev->dev,
1270                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1271                 adapter->msix_entries ? "MSI-X" :
1272                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1273                 adapter->num_rx_queues, adapter->num_tx_queues);
1274
1275         return 0;
1276
1277 err_register:
1278         igb_release_hw_control(adapter);
1279 err_eeprom:
1280         if (!igb_check_reset_block(hw))
1281                 hw->phy.ops.reset_phy(hw);
1282
1283         if (hw->flash_address)
1284                 iounmap(hw->flash_address);
1285
1286         igb_remove_device(hw);
1287         igb_free_queues(adapter);
1288 err_sw_init:
1289 err_hw_init:
1290         iounmap(hw->hw_addr);
1291 err_ioremap:
1292         free_netdev(netdev);
1293 err_alloc_etherdev:
1294         pci_release_selected_regions(pdev, bars);
1295 err_pci_reg:
1296 err_dma:
1297         pci_disable_device(pdev);
1298         return err;
1299 }
1300
1301 /**
1302  * igb_remove - Device Removal Routine
1303  * @pdev: PCI device information struct
1304  *
1305  * igb_remove is called by the PCI subsystem to alert the driver
1306  * that it should release a PCI device.  The could be caused by a
1307  * Hot-Plug event, or because the driver is going to be removed from
1308  * memory.
1309  **/
1310 static void __devexit igb_remove(struct pci_dev *pdev)
1311 {
1312         struct net_device *netdev = pci_get_drvdata(pdev);
1313         struct igb_adapter *adapter = netdev_priv(netdev);
1314 #ifdef CONFIG_DCA
1315         struct e1000_hw *hw = &adapter->hw;
1316 #endif
1317
1318         /* flush_scheduled work may reschedule our watchdog task, so
1319          * explicitly disable watchdog tasks from being rescheduled  */
1320         set_bit(__IGB_DOWN, &adapter->state);
1321         del_timer_sync(&adapter->watchdog_timer);
1322         del_timer_sync(&adapter->phy_info_timer);
1323
1324         flush_scheduled_work();
1325
1326 #ifdef CONFIG_DCA
1327         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1328                 dev_info(&pdev->dev, "DCA disabled\n");
1329                 dca_remove_requester(&pdev->dev);
1330                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1331                 wr32(E1000_DCA_CTRL, 1);
1332         }
1333 #endif
1334
1335         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1336          * would have already happened in close and is redundant. */
1337         igb_release_hw_control(adapter);
1338
1339         unregister_netdev(netdev);
1340
1341         if (adapter->hw.phy.ops.reset_phy &&
1342             !igb_check_reset_block(&adapter->hw))
1343                 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1344
1345         igb_remove_device(&adapter->hw);
1346         igb_reset_interrupt_capability(adapter);
1347
1348         igb_free_queues(adapter);
1349
1350         iounmap(adapter->hw.hw_addr);
1351         if (adapter->hw.flash_address)
1352                 iounmap(adapter->hw.flash_address);
1353         pci_release_selected_regions(pdev, adapter->bars);
1354
1355         free_netdev(netdev);
1356
1357         pci_disable_device(pdev);
1358 }
1359
1360 /**
1361  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1362  * @adapter: board private structure to initialize
1363  *
1364  * igb_sw_init initializes the Adapter private data structure.
1365  * Fields are initialized based on PCI device information and
1366  * OS network device settings (MTU size).
1367  **/
1368 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1369 {
1370         struct e1000_hw *hw = &adapter->hw;
1371         struct net_device *netdev = adapter->netdev;
1372         struct pci_dev *pdev = adapter->pdev;
1373
1374         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1375
1376         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1377         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1378         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1379         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1380
1381         /* Number of supported queues. */
1382         /* Having more queues than CPUs doesn't make sense. */
1383         adapter->num_rx_queues = min((u32)IGB_MAX_RX_QUEUES, (u32)num_online_cpus());
1384         adapter->num_tx_queues = min(IGB_MAX_TX_QUEUES, num_online_cpus());
1385
1386         /* This call may decrease the number of queues depending on
1387          * interrupt mode. */
1388         igb_set_interrupt_capability(adapter);
1389
1390         if (igb_alloc_queues(adapter)) {
1391                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1392                 return -ENOMEM;
1393         }
1394
1395         /* Explicitly disable IRQ since the NIC can be in any state. */
1396         igb_irq_disable(adapter);
1397
1398         set_bit(__IGB_DOWN, &adapter->state);
1399         return 0;
1400 }
1401
1402 /**
1403  * igb_open - Called when a network interface is made active
1404  * @netdev: network interface device structure
1405  *
1406  * Returns 0 on success, negative value on failure
1407  *
1408  * The open entry point is called when a network interface is made
1409  * active by the system (IFF_UP).  At this point all resources needed
1410  * for transmit and receive operations are allocated, the interrupt
1411  * handler is registered with the OS, the watchdog timer is started,
1412  * and the stack is notified that the interface is ready.
1413  **/
1414 static int igb_open(struct net_device *netdev)
1415 {
1416         struct igb_adapter *adapter = netdev_priv(netdev);
1417         struct e1000_hw *hw = &adapter->hw;
1418         int err;
1419         int i;
1420
1421         /* disallow open during test */
1422         if (test_bit(__IGB_TESTING, &adapter->state))
1423                 return -EBUSY;
1424
1425         /* allocate transmit descriptors */
1426         err = igb_setup_all_tx_resources(adapter);
1427         if (err)
1428                 goto err_setup_tx;
1429
1430         /* allocate receive descriptors */
1431         err = igb_setup_all_rx_resources(adapter);
1432         if (err)
1433                 goto err_setup_rx;
1434
1435         /* e1000_power_up_phy(adapter); */
1436
1437         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1438         if ((adapter->hw.mng_cookie.status &
1439              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1440                 igb_update_mng_vlan(adapter);
1441
1442         /* before we allocate an interrupt, we must be ready to handle it.
1443          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1444          * as soon as we call pci_request_irq, so we have to setup our
1445          * clean_rx handler before we do so.  */
1446         igb_configure(adapter);
1447
1448         err = igb_request_irq(adapter);
1449         if (err)
1450                 goto err_req_irq;
1451
1452         /* From here on the code is the same as igb_up() */
1453         clear_bit(__IGB_DOWN, &adapter->state);
1454
1455         for (i = 0; i < adapter->num_rx_queues; i++)
1456                 napi_enable(&adapter->rx_ring[i].napi);
1457
1458         /* Clear any pending interrupts. */
1459         rd32(E1000_ICR);
1460
1461         igb_irq_enable(adapter);
1462
1463         netif_tx_start_all_queues(netdev);
1464
1465         /* Fire a link status change interrupt to start the watchdog. */
1466         wr32(E1000_ICS, E1000_ICS_LSC);
1467
1468         return 0;
1469
1470 err_req_irq:
1471         igb_release_hw_control(adapter);
1472         /* e1000_power_down_phy(adapter); */
1473         igb_free_all_rx_resources(adapter);
1474 err_setup_rx:
1475         igb_free_all_tx_resources(adapter);
1476 err_setup_tx:
1477         igb_reset(adapter);
1478
1479         return err;
1480 }
1481
1482 /**
1483  * igb_close - Disables a network interface
1484  * @netdev: network interface device structure
1485  *
1486  * Returns 0, this is not allowed to fail
1487  *
1488  * The close entry point is called when an interface is de-activated
1489  * by the OS.  The hardware is still under the driver's control, but
1490  * needs to be disabled.  A global MAC reset is issued to stop the
1491  * hardware, and all transmit and receive resources are freed.
1492  **/
1493 static int igb_close(struct net_device *netdev)
1494 {
1495         struct igb_adapter *adapter = netdev_priv(netdev);
1496
1497         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1498         igb_down(adapter);
1499
1500         igb_free_irq(adapter);
1501
1502         igb_free_all_tx_resources(adapter);
1503         igb_free_all_rx_resources(adapter);
1504
1505         /* kill manageability vlan ID if supported, but not if a vlan with
1506          * the same ID is registered on the host OS (let 8021q kill it) */
1507         if ((adapter->hw.mng_cookie.status &
1508                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1509              !(adapter->vlgrp &&
1510                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1511                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1512
1513         return 0;
1514 }
1515
1516 /**
1517  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1518  * @adapter: board private structure
1519  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1520  *
1521  * Return 0 on success, negative on failure
1522  **/
1523
1524 int igb_setup_tx_resources(struct igb_adapter *adapter,
1525                            struct igb_ring *tx_ring)
1526 {
1527         struct pci_dev *pdev = adapter->pdev;
1528         int size;
1529
1530         size = sizeof(struct igb_buffer) * tx_ring->count;
1531         tx_ring->buffer_info = vmalloc(size);
1532         if (!tx_ring->buffer_info)
1533                 goto err;
1534         memset(tx_ring->buffer_info, 0, size);
1535
1536         /* round up to nearest 4K */
1537         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1538                         + sizeof(u32);
1539         tx_ring->size = ALIGN(tx_ring->size, 4096);
1540
1541         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1542                                              &tx_ring->dma);
1543
1544         if (!tx_ring->desc)
1545                 goto err;
1546
1547         tx_ring->adapter = adapter;
1548         tx_ring->next_to_use = 0;
1549         tx_ring->next_to_clean = 0;
1550         return 0;
1551
1552 err:
1553         vfree(tx_ring->buffer_info);
1554         dev_err(&adapter->pdev->dev,
1555                 "Unable to allocate memory for the transmit descriptor ring\n");
1556         return -ENOMEM;
1557 }
1558
1559 /**
1560  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1561  *                                (Descriptors) for all queues
1562  * @adapter: board private structure
1563  *
1564  * Return 0 on success, negative on failure
1565  **/
1566 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1567 {
1568         int i, err = 0;
1569         int r_idx;
1570
1571         for (i = 0; i < adapter->num_tx_queues; i++) {
1572                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1573                 if (err) {
1574                         dev_err(&adapter->pdev->dev,
1575                                 "Allocation for Tx Queue %u failed\n", i);
1576                         for (i--; i >= 0; i--)
1577                                 igb_free_tx_resources(&adapter->tx_ring[i]);
1578                         break;
1579                 }
1580         }
1581
1582         for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1583                 r_idx = i % adapter->num_tx_queues;
1584                 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1585         }
1586         return err;
1587 }
1588
1589 /**
1590  * igb_configure_tx - Configure transmit Unit after Reset
1591  * @adapter: board private structure
1592  *
1593  * Configure the Tx unit of the MAC after a reset.
1594  **/
1595 static void igb_configure_tx(struct igb_adapter *adapter)
1596 {
1597         u64 tdba, tdwba;
1598         struct e1000_hw *hw = &adapter->hw;
1599         u32 tctl;
1600         u32 txdctl, txctrl;
1601         int i;
1602
1603         for (i = 0; i < adapter->num_tx_queues; i++) {
1604                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1605
1606                 wr32(E1000_TDLEN(i),
1607                                 ring->count * sizeof(struct e1000_tx_desc));
1608                 tdba = ring->dma;
1609                 wr32(E1000_TDBAL(i),
1610                                 tdba & 0x00000000ffffffffULL);
1611                 wr32(E1000_TDBAH(i), tdba >> 32);
1612
1613                 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1614                 tdwba |= 1; /* enable head wb */
1615                 wr32(E1000_TDWBAL(i),
1616                                 tdwba & 0x00000000ffffffffULL);
1617                 wr32(E1000_TDWBAH(i), tdwba >> 32);
1618
1619                 ring->head = E1000_TDH(i);
1620                 ring->tail = E1000_TDT(i);
1621                 writel(0, hw->hw_addr + ring->tail);
1622                 writel(0, hw->hw_addr + ring->head);
1623                 txdctl = rd32(E1000_TXDCTL(i));
1624                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1625                 wr32(E1000_TXDCTL(i), txdctl);
1626
1627                 /* Turn off Relaxed Ordering on head write-backs.  The
1628                  * writebacks MUST be delivered in order or it will
1629                  * completely screw up our bookeeping.
1630                  */
1631                 txctrl = rd32(E1000_DCA_TXCTRL(i));
1632                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1633                 wr32(E1000_DCA_TXCTRL(i), txctrl);
1634         }
1635
1636
1637
1638         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1639
1640         /* Program the Transmit Control Register */
1641
1642         tctl = rd32(E1000_TCTL);
1643         tctl &= ~E1000_TCTL_CT;
1644         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1645                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1646
1647         igb_config_collision_dist(hw);
1648
1649         /* Setup Transmit Descriptor Settings for eop descriptor */
1650         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1651
1652         /* Enable transmits */
1653         tctl |= E1000_TCTL_EN;
1654
1655         wr32(E1000_TCTL, tctl);
1656 }
1657
1658 /**
1659  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1660  * @adapter: board private structure
1661  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1662  *
1663  * Returns 0 on success, negative on failure
1664  **/
1665
1666 int igb_setup_rx_resources(struct igb_adapter *adapter,
1667                            struct igb_ring *rx_ring)
1668 {
1669         struct pci_dev *pdev = adapter->pdev;
1670         int size, desc_len;
1671
1672 #ifdef CONFIG_IGB_LRO
1673         size = sizeof(struct net_lro_desc) * MAX_LRO_DESCRIPTORS;
1674         rx_ring->lro_mgr.lro_arr = vmalloc(size);
1675         if (!rx_ring->lro_mgr.lro_arr)
1676                 goto err;
1677         memset(rx_ring->lro_mgr.lro_arr, 0, size);
1678 #endif
1679
1680         size = sizeof(struct igb_buffer) * rx_ring->count;
1681         rx_ring->buffer_info = vmalloc(size);
1682         if (!rx_ring->buffer_info)
1683                 goto err;
1684         memset(rx_ring->buffer_info, 0, size);
1685
1686         desc_len = sizeof(union e1000_adv_rx_desc);
1687
1688         /* Round up to nearest 4K */
1689         rx_ring->size = rx_ring->count * desc_len;
1690         rx_ring->size = ALIGN(rx_ring->size, 4096);
1691
1692         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1693                                              &rx_ring->dma);
1694
1695         if (!rx_ring->desc)
1696                 goto err;
1697
1698         rx_ring->next_to_clean = 0;
1699         rx_ring->next_to_use = 0;
1700
1701         rx_ring->adapter = adapter;
1702
1703         return 0;
1704
1705 err:
1706 #ifdef CONFIG_IGB_LRO
1707         vfree(rx_ring->lro_mgr.lro_arr);
1708         rx_ring->lro_mgr.lro_arr = NULL;
1709 #endif
1710         vfree(rx_ring->buffer_info);
1711         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1712                 "the receive descriptor ring\n");
1713         return -ENOMEM;
1714 }
1715
1716 /**
1717  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1718  *                                (Descriptors) for all queues
1719  * @adapter: board private structure
1720  *
1721  * Return 0 on success, negative on failure
1722  **/
1723 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1724 {
1725         int i, err = 0;
1726
1727         for (i = 0; i < adapter->num_rx_queues; i++) {
1728                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1729                 if (err) {
1730                         dev_err(&adapter->pdev->dev,
1731                                 "Allocation for Rx Queue %u failed\n", i);
1732                         for (i--; i >= 0; i--)
1733                                 igb_free_rx_resources(&adapter->rx_ring[i]);
1734                         break;
1735                 }
1736         }
1737
1738         return err;
1739 }
1740
1741 /**
1742  * igb_setup_rctl - configure the receive control registers
1743  * @adapter: Board private structure
1744  **/
1745 static void igb_setup_rctl(struct igb_adapter *adapter)
1746 {
1747         struct e1000_hw *hw = &adapter->hw;
1748         u32 rctl;
1749         u32 srrctl = 0;
1750         int i;
1751
1752         rctl = rd32(E1000_RCTL);
1753
1754         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1755
1756         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1757                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1758                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1759
1760         /*
1761          * enable stripping of CRC. It's unlikely this will break BMC
1762          * redirection as it did with e1000. Newer features require
1763          * that the HW strips the CRC.
1764         */
1765         rctl |= E1000_RCTL_SECRC;
1766
1767         rctl &= ~E1000_RCTL_SBP;
1768
1769         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1770                 rctl &= ~E1000_RCTL_LPE;
1771         else
1772                 rctl |= E1000_RCTL_LPE;
1773         if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1774                 /* Setup buffer sizes */
1775                 rctl &= ~E1000_RCTL_SZ_4096;
1776                 rctl |= E1000_RCTL_BSEX;
1777                 switch (adapter->rx_buffer_len) {
1778                 case IGB_RXBUFFER_256:
1779                         rctl |= E1000_RCTL_SZ_256;
1780                         rctl &= ~E1000_RCTL_BSEX;
1781                         break;
1782                 case IGB_RXBUFFER_512:
1783                         rctl |= E1000_RCTL_SZ_512;
1784                         rctl &= ~E1000_RCTL_BSEX;
1785                         break;
1786                 case IGB_RXBUFFER_1024:
1787                         rctl |= E1000_RCTL_SZ_1024;
1788                         rctl &= ~E1000_RCTL_BSEX;
1789                         break;
1790                 case IGB_RXBUFFER_2048:
1791                 default:
1792                         rctl |= E1000_RCTL_SZ_2048;
1793                         rctl &= ~E1000_RCTL_BSEX;
1794                         break;
1795                 }
1796         } else {
1797                 rctl &= ~E1000_RCTL_BSEX;
1798                 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1799         }
1800
1801         /* 82575 and greater support packet-split where the protocol
1802          * header is placed in skb->data and the packet data is
1803          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1804          * In the case of a non-split, skb->data is linearly filled,
1805          * followed by the page buffers.  Therefore, skb->data is
1806          * sized to hold the largest protocol header.
1807          */
1808         /* allocations using alloc_page take too long for regular MTU
1809          * so only enable packet split for jumbo frames */
1810         if (rctl & E1000_RCTL_LPE) {
1811                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1812                 srrctl |= adapter->rx_ps_hdr_size <<
1813                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1814                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1815         } else {
1816                 adapter->rx_ps_hdr_size = 0;
1817                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1818         }
1819
1820         for (i = 0; i < adapter->num_rx_queues; i++)
1821                 wr32(E1000_SRRCTL(i), srrctl);
1822
1823         wr32(E1000_RCTL, rctl);
1824 }
1825
1826 /**
1827  * igb_configure_rx - Configure receive Unit after Reset
1828  * @adapter: board private structure
1829  *
1830  * Configure the Rx unit of the MAC after a reset.
1831  **/
1832 static void igb_configure_rx(struct igb_adapter *adapter)
1833 {
1834         u64 rdba;
1835         struct e1000_hw *hw = &adapter->hw;
1836         u32 rctl, rxcsum;
1837         u32 rxdctl;
1838         int i;
1839
1840         /* disable receives while setting up the descriptors */
1841         rctl = rd32(E1000_RCTL);
1842         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1843         wrfl();
1844         mdelay(10);
1845
1846         if (adapter->itr_setting > 3)
1847                 wr32(E1000_ITR, adapter->itr);
1848
1849         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1850          * the Base and Length of the Rx Descriptor Ring */
1851         for (i = 0; i < adapter->num_rx_queues; i++) {
1852                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1853                 rdba = ring->dma;
1854                 wr32(E1000_RDBAL(i),
1855                                 rdba & 0x00000000ffffffffULL);
1856                 wr32(E1000_RDBAH(i), rdba >> 32);
1857                 wr32(E1000_RDLEN(i),
1858                                ring->count * sizeof(union e1000_adv_rx_desc));
1859
1860                 ring->head = E1000_RDH(i);
1861                 ring->tail = E1000_RDT(i);
1862                 writel(0, hw->hw_addr + ring->tail);
1863                 writel(0, hw->hw_addr + ring->head);
1864
1865                 rxdctl = rd32(E1000_RXDCTL(i));
1866                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1867                 rxdctl &= 0xFFF00000;
1868                 rxdctl |= IGB_RX_PTHRESH;
1869                 rxdctl |= IGB_RX_HTHRESH << 8;
1870                 rxdctl |= IGB_RX_WTHRESH << 16;
1871                 wr32(E1000_RXDCTL(i), rxdctl);
1872 #ifdef CONFIG_IGB_LRO
1873                 /* Intitial LRO Settings */
1874                 ring->lro_mgr.max_aggr = MAX_LRO_AGGR;
1875                 ring->lro_mgr.max_desc = MAX_LRO_DESCRIPTORS;
1876                 ring->lro_mgr.get_skb_header = igb_get_skb_hdr;
1877                 ring->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1878                 ring->lro_mgr.dev = adapter->netdev;
1879                 ring->lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1880                 ring->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1881 #endif
1882         }
1883
1884         if (adapter->num_rx_queues > 1) {
1885                 u32 random[10];
1886                 u32 mrqc;
1887                 u32 j, shift;
1888                 union e1000_reta {
1889                         u32 dword;
1890                         u8  bytes[4];
1891                 } reta;
1892
1893                 get_random_bytes(&random[0], 40);
1894
1895                 if (hw->mac.type >= e1000_82576)
1896                         shift = 0;
1897                 else
1898                         shift = 6;
1899                 for (j = 0; j < (32 * 4); j++) {
1900                         reta.bytes[j & 3] =
1901                                 (j % adapter->num_rx_queues) << shift;
1902                         if ((j & 3) == 3)
1903                                 writel(reta.dword,
1904                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1905                 }
1906                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1907
1908                 /* Fill out hash function seeds */
1909                 for (j = 0; j < 10; j++)
1910                         array_wr32(E1000_RSSRK(0), j, random[j]);
1911
1912                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1913                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1914                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1915                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1916                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1917                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1918                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1919                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1920
1921
1922                 wr32(E1000_MRQC, mrqc);
1923
1924                 /* Multiqueue and raw packet checksumming are mutually
1925                  * exclusive.  Note that this not the same as TCP/IP
1926                  * checksumming, which works fine. */
1927                 rxcsum = rd32(E1000_RXCSUM);
1928                 rxcsum |= E1000_RXCSUM_PCSD;
1929                 wr32(E1000_RXCSUM, rxcsum);
1930         } else {
1931                 /* Enable Receive Checksum Offload for TCP and UDP */
1932                 rxcsum = rd32(E1000_RXCSUM);
1933                 if (adapter->rx_csum) {
1934                         rxcsum |= E1000_RXCSUM_TUOFL;
1935
1936                         /* Enable IPv4 payload checksum for UDP fragments
1937                          * Must be used in conjunction with packet-split. */
1938                         if (adapter->rx_ps_hdr_size)
1939                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1940                 } else {
1941                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1942                         /* don't need to clear IPPCSE as it defaults to 0 */
1943                 }
1944                 wr32(E1000_RXCSUM, rxcsum);
1945         }
1946
1947         if (adapter->vlgrp)
1948                 wr32(E1000_RLPML,
1949                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1950         else
1951                 wr32(E1000_RLPML, adapter->max_frame_size);
1952
1953         /* Enable Receives */
1954         wr32(E1000_RCTL, rctl);
1955 }
1956
1957 /**
1958  * igb_free_tx_resources - Free Tx Resources per Queue
1959  * @adapter: board private structure
1960  * @tx_ring: Tx descriptor ring for a specific queue
1961  *
1962  * Free all transmit software resources
1963  **/
1964 static void igb_free_tx_resources(struct igb_ring *tx_ring)
1965 {
1966         struct pci_dev *pdev = tx_ring->adapter->pdev;
1967
1968         igb_clean_tx_ring(tx_ring);
1969
1970         vfree(tx_ring->buffer_info);
1971         tx_ring->buffer_info = NULL;
1972
1973         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1974
1975         tx_ring->desc = NULL;
1976 }
1977
1978 /**
1979  * igb_free_all_tx_resources - Free Tx Resources for All Queues
1980  * @adapter: board private structure
1981  *
1982  * Free all transmit software resources
1983  **/
1984 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
1985 {
1986         int i;
1987
1988         for (i = 0; i < adapter->num_tx_queues; i++)
1989                 igb_free_tx_resources(&adapter->tx_ring[i]);
1990 }
1991
1992 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
1993                                            struct igb_buffer *buffer_info)
1994 {
1995         if (buffer_info->dma) {
1996                 pci_unmap_page(adapter->pdev,
1997                                 buffer_info->dma,
1998                                 buffer_info->length,
1999                                 PCI_DMA_TODEVICE);
2000                 buffer_info->dma = 0;
2001         }
2002         if (buffer_info->skb) {
2003                 dev_kfree_skb_any(buffer_info->skb);
2004                 buffer_info->skb = NULL;
2005         }
2006         buffer_info->time_stamp = 0;
2007         /* buffer_info must be completely set up in the transmit path */
2008 }
2009
2010 /**
2011  * igb_clean_tx_ring - Free Tx Buffers
2012  * @adapter: board private structure
2013  * @tx_ring: ring to be cleaned
2014  **/
2015 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2016 {
2017         struct igb_adapter *adapter = tx_ring->adapter;
2018         struct igb_buffer *buffer_info;
2019         unsigned long size;
2020         unsigned int i;
2021
2022         if (!tx_ring->buffer_info)
2023                 return;
2024         /* Free all the Tx ring sk_buffs */
2025
2026         for (i = 0; i < tx_ring->count; i++) {
2027                 buffer_info = &tx_ring->buffer_info[i];
2028                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2029         }
2030
2031         size = sizeof(struct igb_buffer) * tx_ring->count;
2032         memset(tx_ring->buffer_info, 0, size);
2033
2034         /* Zero out the descriptor ring */
2035
2036         memset(tx_ring->desc, 0, tx_ring->size);
2037
2038         tx_ring->next_to_use = 0;
2039         tx_ring->next_to_clean = 0;
2040
2041         writel(0, adapter->hw.hw_addr + tx_ring->head);
2042         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2043 }
2044
2045 /**
2046  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2047  * @adapter: board private structure
2048  **/
2049 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2050 {
2051         int i;
2052
2053         for (i = 0; i < adapter->num_tx_queues; i++)
2054                 igb_clean_tx_ring(&adapter->tx_ring[i]);
2055 }
2056
2057 /**
2058  * igb_free_rx_resources - Free Rx Resources
2059  * @adapter: board private structure
2060  * @rx_ring: ring to clean the resources from
2061  *
2062  * Free all receive software resources
2063  **/
2064 static void igb_free_rx_resources(struct igb_ring *rx_ring)
2065 {
2066         struct pci_dev *pdev = rx_ring->adapter->pdev;
2067
2068         igb_clean_rx_ring(rx_ring);
2069
2070         vfree(rx_ring->buffer_info);
2071         rx_ring->buffer_info = NULL;
2072
2073 #ifdef CONFIG_IGB_LRO
2074         vfree(rx_ring->lro_mgr.lro_arr);
2075         rx_ring->lro_mgr.lro_arr = NULL;
2076 #endif
2077
2078         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2079
2080         rx_ring->desc = NULL;
2081 }
2082
2083 /**
2084  * igb_free_all_rx_resources - Free Rx Resources for All Queues
2085  * @adapter: board private structure
2086  *
2087  * Free all receive software resources
2088  **/
2089 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2090 {
2091         int i;
2092
2093         for (i = 0; i < adapter->num_rx_queues; i++)
2094                 igb_free_rx_resources(&adapter->rx_ring[i]);
2095 }
2096
2097 /**
2098  * igb_clean_rx_ring - Free Rx Buffers per Queue
2099  * @adapter: board private structure
2100  * @rx_ring: ring to free buffers from
2101  **/
2102 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2103 {
2104         struct igb_adapter *adapter = rx_ring->adapter;
2105         struct igb_buffer *buffer_info;
2106         struct pci_dev *pdev = adapter->pdev;
2107         unsigned long size;
2108         unsigned int i;
2109
2110         if (!rx_ring->buffer_info)
2111                 return;
2112         /* Free all the Rx ring sk_buffs */
2113         for (i = 0; i < rx_ring->count; i++) {
2114                 buffer_info = &rx_ring->buffer_info[i];
2115                 if (buffer_info->dma) {
2116                         if (adapter->rx_ps_hdr_size)
2117                                 pci_unmap_single(pdev, buffer_info->dma,
2118                                                  adapter->rx_ps_hdr_size,
2119                                                  PCI_DMA_FROMDEVICE);
2120                         else
2121                                 pci_unmap_single(pdev, buffer_info->dma,
2122                                                  adapter->rx_buffer_len,
2123                                                  PCI_DMA_FROMDEVICE);
2124                         buffer_info->dma = 0;
2125                 }
2126
2127                 if (buffer_info->skb) {
2128                         dev_kfree_skb(buffer_info->skb);
2129                         buffer_info->skb = NULL;
2130                 }
2131                 if (buffer_info->page) {
2132                         if (buffer_info->page_dma)
2133                                 pci_unmap_page(pdev, buffer_info->page_dma,
2134                                                PAGE_SIZE / 2,
2135                                                PCI_DMA_FROMDEVICE);
2136                         put_page(buffer_info->page);
2137                         buffer_info->page = NULL;
2138                         buffer_info->page_dma = 0;
2139                         buffer_info->page_offset = 0;
2140                 }
2141         }
2142
2143         size = sizeof(struct igb_buffer) * rx_ring->count;
2144         memset(rx_ring->buffer_info, 0, size);
2145
2146         /* Zero out the descriptor ring */
2147         memset(rx_ring->desc, 0, rx_ring->size);
2148
2149         rx_ring->next_to_clean = 0;
2150         rx_ring->next_to_use = 0;
2151
2152         writel(0, adapter->hw.hw_addr + rx_ring->head);
2153         writel(0, adapter->hw.hw_addr + rx_ring->tail);
2154 }
2155
2156 /**
2157  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2158  * @adapter: board private structure
2159  **/
2160 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2161 {
2162         int i;
2163
2164         for (i = 0; i < adapter->num_rx_queues; i++)
2165                 igb_clean_rx_ring(&adapter->rx_ring[i]);
2166 }
2167
2168 /**
2169  * igb_set_mac - Change the Ethernet Address of the NIC
2170  * @netdev: network interface device structure
2171  * @p: pointer to an address structure
2172  *
2173  * Returns 0 on success, negative on failure
2174  **/
2175 static int igb_set_mac(struct net_device *netdev, void *p)
2176 {
2177         struct igb_adapter *adapter = netdev_priv(netdev);
2178         struct sockaddr *addr = p;
2179
2180         if (!is_valid_ether_addr(addr->sa_data))
2181                 return -EADDRNOTAVAIL;
2182
2183         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2184         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2185
2186         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2187
2188         return 0;
2189 }
2190
2191 /**
2192  * igb_set_multi - Multicast and Promiscuous mode set
2193  * @netdev: network interface device structure
2194  *
2195  * The set_multi entry point is called whenever the multicast address
2196  * list or the network interface flags are updated.  This routine is
2197  * responsible for configuring the hardware for proper multicast,
2198  * promiscuous mode, and all-multi behavior.
2199  **/
2200 static void igb_set_multi(struct net_device *netdev)
2201 {
2202         struct igb_adapter *adapter = netdev_priv(netdev);
2203         struct e1000_hw *hw = &adapter->hw;
2204         struct e1000_mac_info *mac = &hw->mac;
2205         struct dev_mc_list *mc_ptr;
2206         u8  *mta_list;
2207         u32 rctl;
2208         int i;
2209
2210         /* Check for Promiscuous and All Multicast modes */
2211
2212         rctl = rd32(E1000_RCTL);
2213
2214         if (netdev->flags & IFF_PROMISC) {
2215                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2216                 rctl &= ~E1000_RCTL_VFE;
2217         } else {
2218                 if (netdev->flags & IFF_ALLMULTI) {
2219                         rctl |= E1000_RCTL_MPE;
2220                         rctl &= ~E1000_RCTL_UPE;
2221                 } else
2222                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2223                 rctl |= E1000_RCTL_VFE;
2224         }
2225         wr32(E1000_RCTL, rctl);
2226
2227         if (!netdev->mc_count) {
2228                 /* nothing to program, so clear mc list */
2229                 igb_update_mc_addr_list_82575(hw, NULL, 0, 1,
2230                                           mac->rar_entry_count);
2231                 return;
2232         }
2233
2234         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2235         if (!mta_list)
2236                 return;
2237
2238         /* The shared function expects a packed array of only addresses. */
2239         mc_ptr = netdev->mc_list;
2240
2241         for (i = 0; i < netdev->mc_count; i++) {
2242                 if (!mc_ptr)
2243                         break;
2244                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2245                 mc_ptr = mc_ptr->next;
2246         }
2247         igb_update_mc_addr_list_82575(hw, mta_list, i, 1,
2248                                       mac->rar_entry_count);
2249         kfree(mta_list);
2250 }
2251
2252 /* Need to wait a few seconds after link up to get diagnostic information from
2253  * the phy */
2254 static void igb_update_phy_info(unsigned long data)
2255 {
2256         struct igb_adapter *adapter = (struct igb_adapter *) data;
2257         if (adapter->hw.phy.ops.get_phy_info)
2258                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2259 }
2260
2261 /**
2262  * igb_watchdog - Timer Call-back
2263  * @data: pointer to adapter cast into an unsigned long
2264  **/
2265 static void igb_watchdog(unsigned long data)
2266 {
2267         struct igb_adapter *adapter = (struct igb_adapter *)data;
2268         /* Do the rest outside of interrupt context */
2269         schedule_work(&adapter->watchdog_task);
2270 }
2271
2272 static void igb_watchdog_task(struct work_struct *work)
2273 {
2274         struct igb_adapter *adapter = container_of(work,
2275                                         struct igb_adapter, watchdog_task);
2276         struct e1000_hw *hw = &adapter->hw;
2277
2278         struct net_device *netdev = adapter->netdev;
2279         struct igb_ring *tx_ring = adapter->tx_ring;
2280         struct e1000_mac_info *mac = &adapter->hw.mac;
2281         u32 link;
2282         u32 eics = 0;
2283         s32 ret_val;
2284         int i;
2285
2286         if ((netif_carrier_ok(netdev)) &&
2287             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2288                 goto link_up;
2289
2290         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2291         if ((ret_val == E1000_ERR_PHY) &&
2292             (hw->phy.type == e1000_phy_igp_3) &&
2293             (rd32(E1000_CTRL) &
2294              E1000_PHY_CTRL_GBE_DISABLE))
2295                 dev_info(&adapter->pdev->dev,
2296                          "Gigabit has been disabled, downgrading speed\n");
2297
2298         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2299             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2300                 link = mac->serdes_has_link;
2301         else
2302                 link = rd32(E1000_STATUS) &
2303                                       E1000_STATUS_LU;
2304
2305         if (link) {
2306                 if (!netif_carrier_ok(netdev)) {
2307                         u32 ctrl;
2308                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2309                                                    &adapter->link_speed,
2310                                                    &adapter->link_duplex);
2311
2312                         ctrl = rd32(E1000_CTRL);
2313                         dev_info(&adapter->pdev->dev,
2314                                  "NIC Link is Up %d Mbps %s, "
2315                                  "Flow Control: %s\n",
2316                                  adapter->link_speed,
2317                                  adapter->link_duplex == FULL_DUPLEX ?
2318                                  "Full Duplex" : "Half Duplex",
2319                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2320                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2321                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2322                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2323
2324                         /* tweak tx_queue_len according to speed/duplex and
2325                          * adjust the timeout factor */
2326                         netdev->tx_queue_len = adapter->tx_queue_len;
2327                         adapter->tx_timeout_factor = 1;
2328                         switch (adapter->link_speed) {
2329                         case SPEED_10:
2330                                 netdev->tx_queue_len = 10;
2331                                 adapter->tx_timeout_factor = 14;
2332                                 break;
2333                         case SPEED_100:
2334                                 netdev->tx_queue_len = 100;
2335                                 /* maybe add some timeout factor ? */
2336                                 break;
2337                         }
2338
2339                         netif_carrier_on(netdev);
2340                         netif_tx_wake_all_queues(netdev);
2341
2342                         if (!test_bit(__IGB_DOWN, &adapter->state))
2343                                 mod_timer(&adapter->phy_info_timer,
2344                                           round_jiffies(jiffies + 2 * HZ));
2345                 }
2346         } else {
2347                 if (netif_carrier_ok(netdev)) {
2348                         adapter->link_speed = 0;
2349                         adapter->link_duplex = 0;
2350                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2351                         netif_carrier_off(netdev);
2352                         netif_tx_stop_all_queues(netdev);
2353                         if (!test_bit(__IGB_DOWN, &adapter->state))
2354                                 mod_timer(&adapter->phy_info_timer,
2355                                           round_jiffies(jiffies + 2 * HZ));
2356                 }
2357         }
2358
2359 link_up:
2360         igb_update_stats(adapter);
2361
2362         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2363         adapter->tpt_old = adapter->stats.tpt;
2364         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2365         adapter->colc_old = adapter->stats.colc;
2366
2367         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2368         adapter->gorc_old = adapter->stats.gorc;
2369         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2370         adapter->gotc_old = adapter->stats.gotc;
2371
2372         igb_update_adaptive(&adapter->hw);
2373
2374         if (!netif_carrier_ok(netdev)) {
2375                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2376                         /* We've lost link, so the controller stops DMA,
2377                          * but we've got queued Tx work that's never going
2378                          * to get done, so reset controller to flush Tx.
2379                          * (Do the reset outside of interrupt context). */
2380                         adapter->tx_timeout_count++;
2381                         schedule_work(&adapter->reset_task);
2382                 }
2383         }
2384
2385         /* Cause software interrupt to ensure rx ring is cleaned */
2386         if (adapter->msix_entries) {
2387                 for (i = 0; i < adapter->num_rx_queues; i++)
2388                         eics |= adapter->rx_ring[i].eims_value;
2389                 wr32(E1000_EICS, eics);
2390         } else {
2391                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2392         }
2393
2394         /* Force detection of hung controller every watchdog period */
2395         tx_ring->detect_tx_hung = true;
2396
2397         /* Reset the timer */
2398         if (!test_bit(__IGB_DOWN, &adapter->state))
2399                 mod_timer(&adapter->watchdog_timer,
2400                           round_jiffies(jiffies + 2 * HZ));
2401 }
2402
2403 enum latency_range {
2404         lowest_latency = 0,
2405         low_latency = 1,
2406         bulk_latency = 2,
2407         latency_invalid = 255
2408 };
2409
2410
2411 /**
2412  * igb_update_ring_itr - update the dynamic ITR value based on packet size
2413  *
2414  *      Stores a new ITR value based on strictly on packet size.  This
2415  *      algorithm is less sophisticated than that used in igb_update_itr,
2416  *      due to the difficulty of synchronizing statistics across multiple
2417  *      receive rings.  The divisors and thresholds used by this fuction
2418  *      were determined based on theoretical maximum wire speed and testing
2419  *      data, in order to minimize response time while increasing bulk
2420  *      throughput.
2421  *      This functionality is controlled by the InterruptThrottleRate module
2422  *      parameter (see igb_param.c)
2423  *      NOTE:  This function is called only when operating in a multiqueue
2424  *             receive environment.
2425  * @rx_ring: pointer to ring
2426  **/
2427 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2428 {
2429         int new_val = rx_ring->itr_val;
2430         int avg_wire_size = 0;
2431         struct igb_adapter *adapter = rx_ring->adapter;
2432
2433         if (!rx_ring->total_packets)
2434                 goto clear_counts; /* no packets, so don't do anything */
2435
2436         /* For non-gigabit speeds, just fix the interrupt rate at 4000
2437          * ints/sec - ITR timer value of 120 ticks.
2438          */
2439         if (adapter->link_speed != SPEED_1000) {
2440                 new_val = 120;
2441                 goto set_itr_val;
2442         }
2443         avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2444
2445         /* Add 24 bytes to size to account for CRC, preamble, and gap */
2446         avg_wire_size += 24;
2447
2448         /* Don't starve jumbo frames */
2449         avg_wire_size = min(avg_wire_size, 3000);
2450
2451         /* Give a little boost to mid-size frames */
2452         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2453                 new_val = avg_wire_size / 3;
2454         else
2455                 new_val = avg_wire_size / 2;
2456
2457 set_itr_val:
2458         if (new_val != rx_ring->itr_val) {
2459                 rx_ring->itr_val = new_val;
2460                 rx_ring->set_itr = 1;
2461         }
2462 clear_counts:
2463         rx_ring->total_bytes = 0;
2464         rx_ring->total_packets = 0;
2465 }
2466
2467 /**
2468  * igb_update_itr - update the dynamic ITR value based on statistics
2469  *      Stores a new ITR value based on packets and byte
2470  *      counts during the last interrupt.  The advantage of per interrupt
2471  *      computation is faster updates and more accurate ITR for the current
2472  *      traffic pattern.  Constants in this function were computed
2473  *      based on theoretical maximum wire speed and thresholds were set based
2474  *      on testing data as well as attempting to minimize response time
2475  *      while increasing bulk throughput.
2476  *      this functionality is controlled by the InterruptThrottleRate module
2477  *      parameter (see igb_param.c)
2478  *      NOTE:  These calculations are only valid when operating in a single-
2479  *             queue environment.
2480  * @adapter: pointer to adapter
2481  * @itr_setting: current adapter->itr
2482  * @packets: the number of packets during this measurement interval
2483  * @bytes: the number of bytes during this measurement interval
2484  **/
2485 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2486                                    int packets, int bytes)
2487 {
2488         unsigned int retval = itr_setting;
2489
2490         if (packets == 0)
2491                 goto update_itr_done;
2492
2493         switch (itr_setting) {
2494         case lowest_latency:
2495                 /* handle TSO and jumbo frames */
2496                 if (bytes/packets > 8000)
2497                         retval = bulk_latency;
2498                 else if ((packets < 5) && (bytes > 512))
2499                         retval = low_latency;
2500                 break;
2501         case low_latency:  /* 50 usec aka 20000 ints/s */
2502                 if (bytes > 10000) {
2503                         /* this if handles the TSO accounting */
2504                         if (bytes/packets > 8000) {
2505                                 retval = bulk_latency;
2506                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2507                                 retval = bulk_latency;
2508                         } else if ((packets > 35)) {
2509                                 retval = lowest_latency;
2510                         }
2511                 } else if (bytes/packets > 2000) {
2512                         retval = bulk_latency;
2513                 } else if (packets <= 2 && bytes < 512) {
2514                         retval = lowest_latency;
2515                 }
2516                 break;
2517         case bulk_latency: /* 250 usec aka 4000 ints/s */
2518                 if (bytes > 25000) {
2519                         if (packets > 35)
2520                                 retval = low_latency;
2521                 } else if (bytes < 6000) {
2522                         retval = low_latency;
2523                 }
2524                 break;
2525         }
2526
2527 update_itr_done:
2528         return retval;
2529 }
2530
2531 static void igb_set_itr(struct igb_adapter *adapter)
2532 {
2533         u16 current_itr;
2534         u32 new_itr = adapter->itr;
2535
2536         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2537         if (adapter->link_speed != SPEED_1000) {
2538                 current_itr = 0;
2539                 new_itr = 4000;
2540                 goto set_itr_now;
2541         }
2542
2543         adapter->rx_itr = igb_update_itr(adapter,
2544                                     adapter->rx_itr,
2545                                     adapter->rx_ring->total_packets,
2546                                     adapter->rx_ring->total_bytes);
2547
2548         if (adapter->rx_ring->buddy) {
2549                 adapter->tx_itr = igb_update_itr(adapter,
2550                                             adapter->tx_itr,
2551                                             adapter->tx_ring->total_packets,
2552                                             adapter->tx_ring->total_bytes);
2553
2554                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2555         } else {
2556                 current_itr = adapter->rx_itr;
2557         }
2558
2559         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2560         if (adapter->itr_setting == 3 &&
2561             current_itr == lowest_latency)
2562                 current_itr = low_latency;
2563
2564         switch (current_itr) {
2565         /* counts and packets in update_itr are dependent on these numbers */
2566         case lowest_latency:
2567                 new_itr = 70000;
2568                 break;
2569         case low_latency:
2570                 new_itr = 20000; /* aka hwitr = ~200 */
2571                 break;
2572         case bulk_latency:
2573                 new_itr = 4000;
2574                 break;
2575         default:
2576                 break;
2577         }
2578
2579 set_itr_now:
2580         adapter->rx_ring->total_bytes = 0;
2581         adapter->rx_ring->total_packets = 0;
2582         if (adapter->rx_ring->buddy) {
2583                 adapter->rx_ring->buddy->total_bytes = 0;
2584                 adapter->rx_ring->buddy->total_packets = 0;
2585         }
2586
2587         if (new_itr != adapter->itr) {
2588                 /* this attempts to bias the interrupt rate towards Bulk
2589                  * by adding intermediate steps when interrupt rate is
2590                  * increasing */
2591                 new_itr = new_itr > adapter->itr ?
2592                              min(adapter->itr + (new_itr >> 2), new_itr) :
2593                              new_itr;
2594                 /* Don't write the value here; it resets the adapter's
2595                  * internal timer, and causes us to delay far longer than
2596                  * we should between interrupts.  Instead, we write the ITR
2597                  * value at the beginning of the next interrupt so the timing
2598                  * ends up being correct.
2599                  */
2600                 adapter->itr = new_itr;
2601                 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2602                 adapter->rx_ring->set_itr = 1;
2603         }
2604
2605         return;
2606 }
2607
2608
2609 #define IGB_TX_FLAGS_CSUM               0x00000001
2610 #define IGB_TX_FLAGS_VLAN               0x00000002
2611 #define IGB_TX_FLAGS_TSO                0x00000004
2612 #define IGB_TX_FLAGS_IPV4               0x00000008
2613 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2614 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2615
2616 static inline int igb_tso_adv(struct igb_adapter *adapter,
2617                               struct igb_ring *tx_ring,
2618                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2619 {
2620         struct e1000_adv_tx_context_desc *context_desc;
2621         unsigned int i;
2622         int err;
2623         struct igb_buffer *buffer_info;
2624         u32 info = 0, tu_cmd = 0;
2625         u32 mss_l4len_idx, l4len;
2626         *hdr_len = 0;
2627
2628         if (skb_header_cloned(skb)) {
2629                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2630                 if (err)
2631                         return err;
2632         }
2633
2634         l4len = tcp_hdrlen(skb);
2635         *hdr_len += l4len;
2636
2637         if (skb->protocol == htons(ETH_P_IP)) {
2638                 struct iphdr *iph = ip_hdr(skb);
2639                 iph->tot_len = 0;
2640                 iph->check = 0;
2641                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2642                                                          iph->daddr, 0,
2643                                                          IPPROTO_TCP,
2644                                                          0);
2645         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2646                 ipv6_hdr(skb)->payload_len = 0;
2647                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2648                                                        &ipv6_hdr(skb)->daddr,
2649                                                        0, IPPROTO_TCP, 0);
2650         }
2651
2652         i = tx_ring->next_to_use;
2653
2654         buffer_info = &tx_ring->buffer_info[i];
2655         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2656         /* VLAN MACLEN IPLEN */
2657         if (tx_flags & IGB_TX_FLAGS_VLAN)
2658                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2659         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2660         *hdr_len += skb_network_offset(skb);
2661         info |= skb_network_header_len(skb);
2662         *hdr_len += skb_network_header_len(skb);
2663         context_desc->vlan_macip_lens = cpu_to_le32(info);
2664
2665         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2666         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2667
2668         if (skb->protocol == htons(ETH_P_IP))
2669                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2670         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2671
2672         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2673
2674         /* MSS L4LEN IDX */
2675         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2676         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2677
2678         /* Context index must be unique per ring. */
2679         if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2680                 mss_l4len_idx |= tx_ring->queue_index << 4;
2681
2682         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2683         context_desc->seqnum_seed = 0;
2684
2685         buffer_info->time_stamp = jiffies;
2686         buffer_info->dma = 0;
2687         i++;
2688         if (i == tx_ring->count)
2689                 i = 0;
2690
2691         tx_ring->next_to_use = i;
2692
2693         return true;
2694 }
2695
2696 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2697                                         struct igb_ring *tx_ring,
2698                                         struct sk_buff *skb, u32 tx_flags)
2699 {
2700         struct e1000_adv_tx_context_desc *context_desc;
2701         unsigned int i;
2702         struct igb_buffer *buffer_info;
2703         u32 info = 0, tu_cmd = 0;
2704
2705         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2706             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2707                 i = tx_ring->next_to_use;
2708                 buffer_info = &tx_ring->buffer_info[i];
2709                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2710
2711                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2712                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2713                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2714                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2715                         info |= skb_network_header_len(skb);
2716
2717                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2718
2719                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2720
2721                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2722                         switch (skb->protocol) {
2723                         case __constant_htons(ETH_P_IP):
2724                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2725                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2726                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2727                                 break;
2728                         case __constant_htons(ETH_P_IPV6):
2729                                 /* XXX what about other V6 headers?? */
2730                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2731                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2732                                 break;
2733                         default:
2734                                 if (unlikely(net_ratelimit()))
2735                                         dev_warn(&adapter->pdev->dev,
2736                                             "partial checksum but proto=%x!\n",
2737                                             skb->protocol);
2738                                 break;
2739                         }
2740                 }
2741
2742                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2743                 context_desc->seqnum_seed = 0;
2744                 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2745                         context_desc->mss_l4len_idx =
2746                                 cpu_to_le32(tx_ring->queue_index << 4);
2747
2748                 buffer_info->time_stamp = jiffies;
2749                 buffer_info->dma = 0;
2750
2751                 i++;
2752                 if (i == tx_ring->count)
2753                         i = 0;
2754                 tx_ring->next_to_use = i;
2755
2756                 return true;
2757         }
2758
2759
2760         return false;
2761 }
2762
2763 #define IGB_MAX_TXD_PWR 16
2764 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2765
2766 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2767                                  struct igb_ring *tx_ring,
2768                                  struct sk_buff *skb)
2769 {
2770         struct igb_buffer *buffer_info;
2771         unsigned int len = skb_headlen(skb);
2772         unsigned int count = 0, i;
2773         unsigned int f;
2774
2775         i = tx_ring->next_to_use;
2776
2777         buffer_info = &tx_ring->buffer_info[i];
2778         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2779         buffer_info->length = len;
2780         /* set time_stamp *before* dma to help avoid a possible race */
2781         buffer_info->time_stamp = jiffies;
2782         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2783                                           PCI_DMA_TODEVICE);
2784         count++;
2785         i++;
2786         if (i == tx_ring->count)
2787                 i = 0;
2788
2789         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2790                 struct skb_frag_struct *frag;
2791
2792                 frag = &skb_shinfo(skb)->frags[f];
2793                 len = frag->size;
2794
2795                 buffer_info = &tx_ring->buffer_info[i];
2796                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2797                 buffer_info->length = len;
2798                 buffer_info->time_stamp = jiffies;
2799                 buffer_info->dma = pci_map_page(adapter->pdev,
2800                                                 frag->page,
2801                                                 frag->page_offset,
2802                                                 len,
2803                                                 PCI_DMA_TODEVICE);
2804
2805                 count++;
2806                 i++;
2807                 if (i == tx_ring->count)
2808                         i = 0;
2809         }
2810
2811         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2812         tx_ring->buffer_info[i].skb = skb;
2813
2814         return count;
2815 }
2816
2817 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2818                                     struct igb_ring *tx_ring,
2819                                     int tx_flags, int count, u32 paylen,
2820                                     u8 hdr_len)
2821 {
2822         union e1000_adv_tx_desc *tx_desc = NULL;
2823         struct igb_buffer *buffer_info;
2824         u32 olinfo_status = 0, cmd_type_len;
2825         unsigned int i;
2826
2827         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2828                         E1000_ADVTXD_DCMD_DEXT);
2829
2830         if (tx_flags & IGB_TX_FLAGS_VLAN)
2831                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2832
2833         if (tx_flags & IGB_TX_FLAGS_TSO) {
2834                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2835
2836                 /* insert tcp checksum */
2837                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2838
2839                 /* insert ip checksum */
2840                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2841                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2842
2843         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2844                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2845         }
2846
2847         if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
2848             (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2849                          IGB_TX_FLAGS_VLAN)))
2850                 olinfo_status |= tx_ring->queue_index << 4;
2851
2852         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2853
2854         i = tx_ring->next_to_use;
2855         while (count--) {
2856                 buffer_info = &tx_ring->buffer_info[i];
2857                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2858                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2859                 tx_desc->read.cmd_type_len =
2860                         cpu_to_le32(cmd_type_len | buffer_info->length);
2861                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2862                 i++;
2863                 if (i == tx_ring->count)
2864                         i = 0;
2865         }
2866
2867         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2868         /* Force memory writes to complete before letting h/w
2869          * know there are new descriptors to fetch.  (Only
2870          * applicable for weak-ordered memory model archs,
2871          * such as IA-64). */
2872         wmb();
2873
2874         tx_ring->next_to_use = i;
2875         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2876         /* we need this if more than one processor can write to our tail
2877          * at a time, it syncronizes IO on IA64/Altix systems */
2878         mmiowb();
2879 }
2880
2881 static int __igb_maybe_stop_tx(struct net_device *netdev,
2882                                struct igb_ring *tx_ring, int size)
2883 {
2884         struct igb_adapter *adapter = netdev_priv(netdev);
2885
2886         netif_stop_subqueue(netdev, tx_ring->queue_index);
2887
2888         /* Herbert's original patch had:
2889          *  smp_mb__after_netif_stop_queue();
2890          * but since that doesn't exist yet, just open code it. */
2891         smp_mb();
2892
2893         /* We need to check again in a case another CPU has just
2894          * made room available. */
2895         if (IGB_DESC_UNUSED(tx_ring) < size)
2896                 return -EBUSY;
2897
2898         /* A reprieve! */
2899         netif_wake_subqueue(netdev, tx_ring->queue_index);
2900         ++adapter->restart_queue;
2901         return 0;
2902 }
2903
2904 static int igb_maybe_stop_tx(struct net_device *netdev,
2905                              struct igb_ring *tx_ring, int size)
2906 {
2907         if (IGB_DESC_UNUSED(tx_ring) >= size)
2908                 return 0;
2909         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2910 }
2911
2912 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2913
2914 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2915                                    struct net_device *netdev,
2916                                    struct igb_ring *tx_ring)
2917 {
2918         struct igb_adapter *adapter = netdev_priv(netdev);
2919         unsigned int tx_flags = 0;
2920         unsigned int len;
2921         u8 hdr_len = 0;
2922         int tso = 0;
2923
2924         len = skb_headlen(skb);
2925
2926         if (test_bit(__IGB_DOWN, &adapter->state)) {
2927                 dev_kfree_skb_any(skb);
2928                 return NETDEV_TX_OK;
2929         }
2930
2931         if (skb->len <= 0) {
2932                 dev_kfree_skb_any(skb);
2933                 return NETDEV_TX_OK;
2934         }
2935
2936         /* need: 1 descriptor per page,
2937          *       + 2 desc gap to keep tail from touching head,
2938          *       + 1 desc for skb->data,
2939          *       + 1 desc for context descriptor,
2940          * otherwise try next time */
2941         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2942                 /* this is a hard error */
2943                 return NETDEV_TX_BUSY;
2944         }
2945         skb_orphan(skb);
2946
2947         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2948                 tx_flags |= IGB_TX_FLAGS_VLAN;
2949                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2950         }
2951
2952         if (skb->protocol == htons(ETH_P_IP))
2953                 tx_flags |= IGB_TX_FLAGS_IPV4;
2954
2955         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2956                                               &hdr_len) : 0;
2957
2958         if (tso < 0) {
2959                 dev_kfree_skb_any(skb);
2960                 return NETDEV_TX_OK;
2961         }
2962
2963         if (tso)
2964                 tx_flags |= IGB_TX_FLAGS_TSO;
2965         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
2966                         if (skb->ip_summed == CHECKSUM_PARTIAL)
2967                                 tx_flags |= IGB_TX_FLAGS_CSUM;
2968
2969         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
2970                          igb_tx_map_adv(adapter, tx_ring, skb),
2971                          skb->len, hdr_len);
2972
2973         netdev->trans_start = jiffies;
2974
2975         /* Make sure there is space in the ring for the next send. */
2976         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
2977
2978         return NETDEV_TX_OK;
2979 }
2980
2981 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
2982 {
2983         struct igb_adapter *adapter = netdev_priv(netdev);
2984         struct igb_ring *tx_ring;
2985
2986         int r_idx = 0;
2987         r_idx = skb->queue_mapping & (IGB_MAX_TX_QUEUES - 1);
2988         tx_ring = adapter->multi_tx_table[r_idx];
2989
2990         /* This goes back to the question of how to logically map a tx queue
2991          * to a flow.  Right now, performance is impacted slightly negatively
2992          * if using multiple tx queues.  If the stack breaks away from a
2993          * single qdisc implementation, we can look at this again. */
2994         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
2995 }
2996
2997 /**
2998  * igb_tx_timeout - Respond to a Tx Hang
2999  * @netdev: network interface device structure
3000  **/
3001 static void igb_tx_timeout(struct net_device *netdev)
3002 {
3003         struct igb_adapter *adapter = netdev_priv(netdev);
3004         struct e1000_hw *hw = &adapter->hw;
3005
3006         /* Do the reset outside of interrupt context */
3007         adapter->tx_timeout_count++;
3008         schedule_work(&adapter->reset_task);
3009         wr32(E1000_EICS, adapter->eims_enable_mask &
3010                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
3011 }
3012
3013 static void igb_reset_task(struct work_struct *work)
3014 {
3015         struct igb_adapter *adapter;
3016         adapter = container_of(work, struct igb_adapter, reset_task);
3017
3018         igb_reinit_locked(adapter);
3019 }
3020
3021 /**
3022  * igb_get_stats - Get System Network Statistics
3023  * @netdev: network interface device structure
3024  *
3025  * Returns the address of the device statistics structure.
3026  * The statistics are actually updated from the timer callback.
3027  **/
3028 static struct net_device_stats *
3029 igb_get_stats(struct net_device *netdev)
3030 {
3031         struct igb_adapter *adapter = netdev_priv(netdev);
3032
3033         /* only return the current stats */
3034         return &adapter->net_stats;
3035 }
3036
3037 /**
3038  * igb_change_mtu - Change the Maximum Transfer Unit
3039  * @netdev: network interface device structure
3040  * @new_mtu: new value for maximum frame size
3041  *
3042  * Returns 0 on success, negative on failure
3043  **/
3044 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3045 {
3046         struct igb_adapter *adapter = netdev_priv(netdev);
3047         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3048
3049         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3050             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3051                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3052                 return -EINVAL;
3053         }
3054
3055 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3056         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3057                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3058                 return -EINVAL;
3059         }
3060
3061         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3062                 msleep(1);
3063         /* igb_down has a dependency on max_frame_size */
3064         adapter->max_frame_size = max_frame;
3065         if (netif_running(netdev))
3066                 igb_down(adapter);
3067
3068         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3069          * means we reserve 2 more, this pushes us to allocate from the next
3070          * larger slab size.
3071          * i.e. RXBUFFER_2048 --> size-4096 slab
3072          */
3073
3074         if (max_frame <= IGB_RXBUFFER_256)
3075                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3076         else if (max_frame <= IGB_RXBUFFER_512)
3077                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3078         else if (max_frame <= IGB_RXBUFFER_1024)
3079                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3080         else if (max_frame <= IGB_RXBUFFER_2048)
3081                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3082         else
3083 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3084                 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3085 #else
3086                 adapter->rx_buffer_len = PAGE_SIZE / 2;
3087 #endif
3088         /* adjust allocation if LPE protects us, and we aren't using SBP */
3089         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3090              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3091                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3092
3093         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3094                  netdev->mtu, new_mtu);
3095         netdev->mtu = new_mtu;
3096
3097         if (netif_running(netdev))
3098                 igb_up(adapter);
3099         else
3100                 igb_reset(adapter);
3101
3102         clear_bit(__IGB_RESETTING, &adapter->state);
3103
3104         return 0;
3105 }
3106
3107 /**
3108  * igb_update_stats - Update the board statistics counters
3109  * @adapter: board private structure
3110  **/
3111
3112 void igb_update_stats(struct igb_adapter *adapter)
3113 {
3114         struct e1000_hw *hw = &adapter->hw;
3115         struct pci_dev *pdev = adapter->pdev;
3116         u16 phy_tmp;
3117
3118 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3119
3120         /*
3121          * Prevent stats update while adapter is being reset, or if the pci
3122          * connection is down.
3123          */
3124         if (adapter->link_speed == 0)
3125                 return;
3126         if (pci_channel_offline(pdev))
3127                 return;
3128
3129         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3130         adapter->stats.gprc += rd32(E1000_GPRC);
3131         adapter->stats.gorc += rd32(E1000_GORCL);
3132         rd32(E1000_GORCH); /* clear GORCL */
3133         adapter->stats.bprc += rd32(E1000_BPRC);
3134         adapter->stats.mprc += rd32(E1000_MPRC);
3135         adapter->stats.roc += rd32(E1000_ROC);
3136
3137         adapter->stats.prc64 += rd32(E1000_PRC64);
3138         adapter->stats.prc127 += rd32(E1000_PRC127);
3139         adapter->stats.prc255 += rd32(E1000_PRC255);
3140         adapter->stats.prc511 += rd32(E1000_PRC511);
3141         adapter->stats.prc1023 += rd32(E1000_PRC1023);
3142         adapter->stats.prc1522 += rd32(E1000_PRC1522);
3143         adapter->stats.symerrs += rd32(E1000_SYMERRS);
3144         adapter->stats.sec += rd32(E1000_SEC);
3145
3146         adapter->stats.mpc += rd32(E1000_MPC);
3147         adapter->stats.scc += rd32(E1000_SCC);
3148         adapter->stats.ecol += rd32(E1000_ECOL);
3149         adapter->stats.mcc += rd32(E1000_MCC);
3150         adapter->stats.latecol += rd32(E1000_LATECOL);
3151         adapter->stats.dc += rd32(E1000_DC);
3152         adapter->stats.rlec += rd32(E1000_RLEC);
3153         adapter->stats.xonrxc += rd32(E1000_XONRXC);
3154         adapter->stats.xontxc += rd32(E1000_XONTXC);
3155         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3156         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3157         adapter->stats.fcruc += rd32(E1000_FCRUC);
3158         adapter->stats.gptc += rd32(E1000_GPTC);
3159         adapter->stats.gotc += rd32(E1000_GOTCL);
3160         rd32(E1000_GOTCH); /* clear GOTCL */
3161         adapter->stats.rnbc += rd32(E1000_RNBC);
3162         adapter->stats.ruc += rd32(E1000_RUC);
3163         adapter->stats.rfc += rd32(E1000_RFC);
3164         adapter->stats.rjc += rd32(E1000_RJC);
3165         adapter->stats.tor += rd32(E1000_TORH);
3166         adapter->stats.tot += rd32(E1000_TOTH);
3167         adapter->stats.tpr += rd32(E1000_TPR);
3168
3169         adapter->stats.ptc64 += rd32(E1000_PTC64);
3170         adapter->stats.ptc127 += rd32(E1000_PTC127);
3171         adapter->stats.ptc255 += rd32(E1000_PTC255);
3172         adapter->stats.ptc511 += rd32(E1000_PTC511);
3173         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3174         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3175
3176         adapter->stats.mptc += rd32(E1000_MPTC);
3177         adapter->stats.bptc += rd32(E1000_BPTC);
3178
3179         /* used for adaptive IFS */
3180
3181         hw->mac.tx_packet_delta = rd32(E1000_TPT);
3182         adapter->stats.tpt += hw->mac.tx_packet_delta;
3183         hw->mac.collision_delta = rd32(E1000_COLC);
3184         adapter->stats.colc += hw->mac.collision_delta;
3185
3186         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3187         adapter->stats.rxerrc += rd32(E1000_RXERRC);
3188         adapter->stats.tncrs += rd32(E1000_TNCRS);
3189         adapter->stats.tsctc += rd32(E1000_TSCTC);
3190         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3191
3192         adapter->stats.iac += rd32(E1000_IAC);
3193         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3194         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3195         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3196         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3197         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3198         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3199         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3200         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3201
3202         /* Fill out the OS statistics structure */
3203         adapter->net_stats.multicast = adapter->stats.mprc;
3204         adapter->net_stats.collisions = adapter->stats.colc;
3205
3206         /* Rx Errors */
3207
3208         /* RLEC on some newer hardware can be incorrect so build
3209         * our own version based on RUC and ROC */
3210         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3211                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3212                 adapter->stats.ruc + adapter->stats.roc +
3213                 adapter->stats.cexterr;
3214         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3215                                               adapter->stats.roc;
3216         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3217         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3218         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3219
3220         /* Tx Errors */
3221         adapter->net_stats.tx_errors = adapter->stats.ecol +
3222                                        adapter->stats.latecol;
3223         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3224         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3225         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3226
3227         /* Tx Dropped needs to be maintained elsewhere */
3228
3229         /* Phy Stats */
3230         if (hw->phy.media_type == e1000_media_type_copper) {
3231                 if ((adapter->link_speed == SPEED_1000) &&
3232                    (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
3233                                               &phy_tmp))) {
3234                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3235                         adapter->phy_stats.idle_errors += phy_tmp;
3236                 }
3237         }
3238
3239         /* Management Stats */
3240         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3241         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3242         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3243 }
3244
3245
3246 static irqreturn_t igb_msix_other(int irq, void *data)
3247 {
3248         struct net_device *netdev = data;
3249         struct igb_adapter *adapter = netdev_priv(netdev);
3250         struct e1000_hw *hw = &adapter->hw;
3251         u32 icr = rd32(E1000_ICR);
3252
3253         /* reading ICR causes bit 31 of EICR to be cleared */
3254         if (!(icr & E1000_ICR_LSC))
3255                 goto no_link_interrupt;
3256         hw->mac.get_link_status = 1;
3257         /* guard against interrupt when we're going down */
3258         if (!test_bit(__IGB_DOWN, &adapter->state))
3259                 mod_timer(&adapter->watchdog_timer, jiffies + 1);
3260
3261 no_link_interrupt:
3262         wr32(E1000_IMS, E1000_IMS_LSC);
3263         wr32(E1000_EIMS, adapter->eims_other);
3264
3265         return IRQ_HANDLED;
3266 }
3267
3268 static irqreturn_t igb_msix_tx(int irq, void *data)
3269 {
3270         struct igb_ring *tx_ring = data;
3271         struct igb_adapter *adapter = tx_ring->adapter;
3272         struct e1000_hw *hw = &adapter->hw;
3273
3274 #ifdef CONFIG_DCA
3275         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3276                 igb_update_tx_dca(tx_ring);
3277 #endif
3278         tx_ring->total_bytes = 0;
3279         tx_ring->total_packets = 0;
3280
3281         /* auto mask will automatically reenable the interrupt when we write
3282          * EICS */
3283         if (!igb_clean_tx_irq(tx_ring))
3284                 /* Ring was not completely cleaned, so fire another interrupt */
3285                 wr32(E1000_EICS, tx_ring->eims_value);
3286         else
3287                 wr32(E1000_EIMS, tx_ring->eims_value);
3288
3289         return IRQ_HANDLED;
3290 }
3291
3292 static void igb_write_itr(struct igb_ring *ring)
3293 {
3294         struct e1000_hw *hw = &ring->adapter->hw;
3295         if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3296                 switch (hw->mac.type) {
3297                 case e1000_82576:
3298                         wr32(ring->itr_register,
3299                              ring->itr_val |
3300                              0x80000000);
3301                         break;
3302                 default:
3303                         wr32(ring->itr_register,
3304                              ring->itr_val |
3305                              (ring->itr_val << 16));
3306                         break;
3307                 }
3308                 ring->set_itr = 0;
3309         }
3310 }
3311
3312 static irqreturn_t igb_msix_rx(int irq, void *data)
3313 {
3314         struct igb_ring *rx_ring = data;
3315         struct igb_adapter *adapter = rx_ring->adapter;
3316
3317         /* Write the ITR value calculated at the end of the
3318          * previous interrupt.
3319          */
3320
3321         igb_write_itr(rx_ring);
3322
3323         if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi))
3324                 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3325
3326 #ifdef CONFIG_DCA
3327         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3328                 igb_update_rx_dca(rx_ring);
3329 #endif
3330                 return IRQ_HANDLED;
3331 }
3332
3333 #ifdef CONFIG_DCA
3334 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3335 {
3336         u32 dca_rxctrl;
3337         struct igb_adapter *adapter = rx_ring->adapter;
3338         struct e1000_hw *hw = &adapter->hw;
3339         int cpu = get_cpu();
3340         int q = rx_ring - adapter->rx_ring;
3341
3342         if (rx_ring->cpu != cpu) {
3343                 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3344                 if (hw->mac.type == e1000_82576) {
3345                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3346                         dca_rxctrl |= dca_get_tag(cpu) <<
3347                                       E1000_DCA_RXCTRL_CPUID_SHIFT;
3348                 } else {
3349                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3350                         dca_rxctrl |= dca_get_tag(cpu);
3351                 }
3352                 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3353                 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3354                 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3355                 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3356                 rx_ring->cpu = cpu;
3357         }
3358         put_cpu();
3359 }
3360
3361 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3362 {
3363         u32 dca_txctrl;
3364         struct igb_adapter *adapter = tx_ring->adapter;
3365         struct e1000_hw *hw = &adapter->hw;
3366         int cpu = get_cpu();
3367         int q = tx_ring - adapter->tx_ring;
3368
3369         if (tx_ring->cpu != cpu) {
3370                 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3371                 if (hw->mac.type == e1000_82576) {
3372                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3373                         dca_txctrl |= dca_get_tag(cpu) <<
3374                                       E1000_DCA_TXCTRL_CPUID_SHIFT;
3375                 } else {
3376                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3377                         dca_txctrl |= dca_get_tag(cpu);
3378                 }
3379                 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3380                 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3381                 tx_ring->cpu = cpu;
3382         }
3383         put_cpu();
3384 }
3385
3386 static void igb_setup_dca(struct igb_adapter *adapter)
3387 {
3388         int i;
3389
3390         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3391                 return;
3392
3393         for (i = 0; i < adapter->num_tx_queues; i++) {
3394                 adapter->tx_ring[i].cpu = -1;
3395                 igb_update_tx_dca(&adapter->tx_ring[i]);
3396         }
3397         for (i = 0; i < adapter->num_rx_queues; i++) {
3398                 adapter->rx_ring[i].cpu = -1;
3399                 igb_update_rx_dca(&adapter->rx_ring[i]);
3400         }
3401 }
3402
3403 static int __igb_notify_dca(struct device *dev, void *data)
3404 {
3405         struct net_device *netdev = dev_get_drvdata(dev);
3406         struct igb_adapter *adapter = netdev_priv(netdev);
3407         struct e1000_hw *hw = &adapter->hw;
3408         unsigned long event = *(unsigned long *)data;
3409
3410         if (!(adapter->flags & IGB_FLAG_HAS_DCA))
3411                 goto out;
3412
3413         switch (event) {
3414         case DCA_PROVIDER_ADD:
3415                 /* if already enabled, don't do it again */
3416                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3417                         break;
3418                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
3419                 /* Always use CB2 mode, difference is masked
3420                  * in the CB driver. */
3421                 wr32(E1000_DCA_CTRL, 2);
3422                 if (dca_add_requester(dev) == 0) {
3423                         dev_info(&adapter->pdev->dev, "DCA enabled\n");
3424                         igb_setup_dca(adapter);
3425                         break;
3426                 }
3427                 /* Fall Through since DCA is disabled. */
3428         case DCA_PROVIDER_REMOVE:
3429                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3430                         /* without this a class_device is left
3431                          * hanging around in the sysfs model */
3432                         dca_remove_requester(dev);
3433                         dev_info(&adapter->pdev->dev, "DCA disabled\n");
3434                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3435                         wr32(E1000_DCA_CTRL, 1);
3436                 }
3437                 break;
3438         }
3439 out:
3440         return 0;
3441 }
3442
3443 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3444                           void *p)
3445 {
3446         int ret_val;
3447
3448         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3449                                          __igb_notify_dca);
3450
3451         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3452 }
3453 #endif /* CONFIG_DCA */
3454
3455 /**
3456  * igb_intr_msi - Interrupt Handler
3457  * @irq: interrupt number
3458  * @data: pointer to a network interface device structure
3459  **/
3460 static irqreturn_t igb_intr_msi(int irq, void *data)
3461 {
3462         struct net_device *netdev = data;
3463         struct igb_adapter *adapter = netdev_priv(netdev);
3464         struct e1000_hw *hw = &adapter->hw;
3465         /* read ICR disables interrupts using IAM */
3466         u32 icr = rd32(E1000_ICR);
3467
3468         igb_write_itr(adapter->rx_ring);
3469
3470         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3471                 hw->mac.get_link_status = 1;
3472                 if (!test_bit(__IGB_DOWN, &adapter->state))
3473                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3474         }
3475
3476         netif_rx_schedule(netdev, &adapter->rx_ring[0].napi);
3477
3478         return IRQ_HANDLED;
3479 }
3480
3481 /**
3482  * igb_intr - Interrupt Handler
3483  * @irq: interrupt number
3484  * @data: pointer to a network interface device structure
3485  **/
3486 static irqreturn_t igb_intr(int irq, void *data)
3487 {
3488         struct net_device *netdev = data;
3489         struct igb_adapter *adapter = netdev_priv(netdev);
3490         struct e1000_hw *hw = &adapter->hw;
3491         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3492          * need for the IMC write */
3493         u32 icr = rd32(E1000_ICR);
3494         u32 eicr = 0;
3495         if (!icr)
3496                 return IRQ_NONE;  /* Not our interrupt */
3497
3498         igb_write_itr(adapter->rx_ring);
3499
3500         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3501          * not set, then the adapter didn't send an interrupt */
3502         if (!(icr & E1000_ICR_INT_ASSERTED))
3503                 return IRQ_NONE;
3504
3505         eicr = rd32(E1000_EICR);
3506
3507         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3508                 hw->mac.get_link_status = 1;
3509                 /* guard against interrupt when we're going down */
3510                 if (!test_bit(__IGB_DOWN, &adapter->state))
3511                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3512         }
3513
3514         netif_rx_schedule(netdev, &adapter->rx_ring[0].napi);
3515
3516         return IRQ_HANDLED;
3517 }
3518
3519 /**
3520  * igb_poll - NAPI Rx polling callback
3521  * @napi: napi polling structure
3522  * @budget: count of how many packets we should handle
3523  **/
3524 static int igb_poll(struct napi_struct *napi, int budget)
3525 {
3526         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3527         struct igb_adapter *adapter = rx_ring->adapter;
3528         struct net_device *netdev = adapter->netdev;
3529         int tx_clean_complete, work_done = 0;
3530
3531         /* this poll routine only supports one tx and one rx queue */
3532 #ifdef CONFIG_DCA
3533         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3534                 igb_update_tx_dca(&adapter->tx_ring[0]);
3535 #endif
3536         tx_clean_complete = igb_clean_tx_irq(&adapter->tx_ring[0]);
3537
3538 #ifdef CONFIG_DCA
3539         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3540                 igb_update_rx_dca(&adapter->rx_ring[0]);
3541 #endif
3542         igb_clean_rx_irq_adv(&adapter->rx_ring[0], &work_done, budget);
3543
3544         /* If no Tx and not enough Rx work done, exit the polling mode */
3545         if ((tx_clean_complete && (work_done < budget)) ||
3546             !netif_running(netdev)) {
3547                 if (adapter->itr_setting & 3)
3548                         igb_set_itr(adapter);
3549                 netif_rx_complete(netdev, napi);
3550                 if (!test_bit(__IGB_DOWN, &adapter->state))
3551                         igb_irq_enable(adapter);
3552                 return 0;
3553         }
3554
3555         return 1;
3556 }
3557
3558 static int igb_clean_rx_ring_msix(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 e1000_hw *hw = &adapter->hw;
3563         struct net_device *netdev = adapter->netdev;
3564         int work_done = 0;
3565
3566         /* Keep link state information with original netdev */
3567         if (!netif_carrier_ok(netdev))
3568                 goto quit_polling;
3569
3570 #ifdef CONFIG_DCA
3571         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3572                 igb_update_rx_dca(rx_ring);
3573 #endif
3574         igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
3575
3576
3577         /* If not enough Rx work done, exit the polling mode */
3578         if ((work_done == 0) || !netif_running(netdev)) {
3579 quit_polling:
3580                 netif_rx_complete(netdev, napi);
3581
3582                 if (adapter->itr_setting & 3) {
3583                         if (adapter->num_rx_queues == 1)
3584                                 igb_set_itr(adapter);
3585                         else
3586                                 igb_update_ring_itr(rx_ring);
3587                 }
3588
3589                 if (!test_bit(__IGB_DOWN, &adapter->state))
3590                         wr32(E1000_EIMS, rx_ring->eims_value);
3591
3592                 return 0;
3593         }
3594
3595         return 1;
3596 }
3597
3598 static inline u32 get_head(struct igb_ring *tx_ring)
3599 {
3600         void *end = (struct e1000_tx_desc *)tx_ring->desc + tx_ring->count;
3601         return le32_to_cpu(*(volatile __le32 *)end);
3602 }
3603
3604 /**
3605  * igb_clean_tx_irq - Reclaim resources after transmit completes
3606  * @adapter: board private structure
3607  * returns true if ring is completely cleaned
3608  **/
3609 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
3610 {
3611         struct igb_adapter *adapter = tx_ring->adapter;
3612         struct e1000_hw *hw = &adapter->hw;
3613         struct net_device *netdev = adapter->netdev;
3614         struct e1000_tx_desc *tx_desc;
3615         struct igb_buffer *buffer_info;
3616         struct sk_buff *skb;
3617         unsigned int i;
3618         u32 head, oldhead;
3619         unsigned int count = 0;
3620         bool cleaned = false;
3621         bool retval = true;
3622         unsigned int total_bytes = 0, total_packets = 0;
3623
3624         rmb();
3625         head = get_head(tx_ring);
3626         i = tx_ring->next_to_clean;
3627         while (1) {
3628                 while (i != head) {
3629                         cleaned = true;
3630                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3631                         buffer_info = &tx_ring->buffer_info[i];
3632                         skb = buffer_info->skb;
3633
3634                         if (skb) {
3635                                 unsigned int segs, bytecount;
3636                                 /* gso_segs is currently only valid for tcp */
3637                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3638                                 /* multiply data chunks by size of headers */
3639                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3640                                             skb->len;
3641                                 total_packets += segs;
3642                                 total_bytes += bytecount;
3643                         }
3644
3645                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3646                         tx_desc->upper.data = 0;
3647
3648                         i++;
3649                         if (i == tx_ring->count)
3650                                 i = 0;
3651
3652                         count++;
3653                         if (count == IGB_MAX_TX_CLEAN) {
3654                                 retval = false;
3655                                 goto done_cleaning;
3656                         }
3657                 }
3658                 oldhead = head;
3659                 rmb();
3660                 head = get_head(tx_ring);
3661                 if (head == oldhead)
3662                         goto done_cleaning;
3663         }  /* while (1) */
3664
3665 done_cleaning:
3666         tx_ring->next_to_clean = i;
3667
3668         if (unlikely(cleaned &&
3669                      netif_carrier_ok(netdev) &&
3670                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3671                 /* Make sure that anybody stopping the queue after this
3672                  * sees the new next_to_clean.
3673                  */
3674                 smp_mb();
3675                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
3676                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3677                         netif_wake_subqueue(netdev, tx_ring->queue_index);
3678                         ++adapter->restart_queue;
3679                 }
3680         }
3681
3682         if (tx_ring->detect_tx_hung) {
3683                 /* Detect a transmit hang in hardware, this serializes the
3684                  * check with the clearing of time_stamp and movement of i */
3685                 tx_ring->detect_tx_hung = false;
3686                 if (tx_ring->buffer_info[i].time_stamp &&
3687                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3688                                (adapter->tx_timeout_factor * HZ))
3689                     && !(rd32(E1000_STATUS) &
3690                          E1000_STATUS_TXOFF)) {
3691
3692                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3693                         /* detected Tx unit hang */
3694                         dev_err(&adapter->pdev->dev,
3695                                 "Detected Tx Unit Hang\n"
3696                                 "  Tx Queue             <%d>\n"
3697                                 "  TDH                  <%x>\n"
3698                                 "  TDT                  <%x>\n"
3699                                 "  next_to_use          <%x>\n"
3700                                 "  next_to_clean        <%x>\n"
3701                                 "  head (WB)            <%x>\n"
3702                                 "buffer_info[next_to_clean]\n"
3703                                 "  time_stamp           <%lx>\n"
3704                                 "  jiffies              <%lx>\n"
3705                                 "  desc.status          <%x>\n",
3706                                 tx_ring->queue_index,
3707                                 readl(adapter->hw.hw_addr + tx_ring->head),
3708                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3709                                 tx_ring->next_to_use,
3710                                 tx_ring->next_to_clean,
3711                                 head,
3712                                 tx_ring->buffer_info[i].time_stamp,
3713                                 jiffies,
3714                                 tx_desc->upper.fields.status);
3715                         netif_stop_subqueue(netdev, tx_ring->queue_index);
3716                 }
3717         }
3718         tx_ring->total_bytes += total_bytes;
3719         tx_ring->total_packets += total_packets;
3720         tx_ring->tx_stats.bytes += total_bytes;
3721         tx_ring->tx_stats.packets += total_packets;
3722         adapter->net_stats.tx_bytes += total_bytes;
3723         adapter->net_stats.tx_packets += total_packets;
3724         return retval;
3725 }
3726
3727 #ifdef CONFIG_IGB_LRO
3728  /**
3729  * igb_get_skb_hdr - helper function for LRO header processing
3730  * @skb: pointer to sk_buff to be added to LRO packet
3731  * @iphdr: pointer to ip header structure
3732  * @tcph: pointer to tcp header structure
3733  * @hdr_flags: pointer to header flags
3734  * @priv: pointer to the receive descriptor for the current sk_buff
3735  **/
3736 static int igb_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
3737                            u64 *hdr_flags, void *priv)
3738 {
3739         union e1000_adv_rx_desc *rx_desc = priv;
3740         u16 pkt_type = rx_desc->wb.lower.lo_dword.pkt_info &
3741                        (E1000_RXDADV_PKTTYPE_IPV4 | E1000_RXDADV_PKTTYPE_TCP);
3742
3743         /* Verify that this is a valid IPv4 TCP packet */
3744         if (pkt_type != (E1000_RXDADV_PKTTYPE_IPV4 |
3745                           E1000_RXDADV_PKTTYPE_TCP))
3746                 return -1;
3747
3748         /* Set network headers */
3749         skb_reset_network_header(skb);
3750         skb_set_transport_header(skb, ip_hdrlen(skb));
3751         *iphdr = ip_hdr(skb);
3752         *tcph = tcp_hdr(skb);
3753         *hdr_flags = LRO_IPV4 | LRO_TCP;
3754
3755         return 0;
3756
3757 }
3758 #endif /* CONFIG_IGB_LRO */
3759
3760 /**
3761  * igb_receive_skb - helper function to handle rx indications
3762  * @ring: pointer to receive ring receving this packet
3763  * @status: descriptor status field as written by hardware
3764  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3765  * @skb: pointer to sk_buff to be indicated to stack
3766  **/
3767 static void igb_receive_skb(struct igb_ring *ring, u8 status,
3768                             union e1000_adv_rx_desc * rx_desc,
3769                             struct sk_buff *skb)
3770 {
3771         struct igb_adapter * adapter = ring->adapter;
3772         bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
3773
3774 #ifdef CONFIG_IGB_LRO
3775         if (adapter->netdev->features & NETIF_F_LRO &&
3776             skb->ip_summed == CHECKSUM_UNNECESSARY) {
3777                 if (vlan_extracted)
3778                         lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
3779                                            adapter->vlgrp,
3780                                            le16_to_cpu(rx_desc->wb.upper.vlan),
3781                                            rx_desc);
3782                 else
3783                         lro_receive_skb(&ring->lro_mgr,skb, rx_desc);
3784                 ring->lro_used = 1;
3785         } else {
3786 #endif
3787                 if (vlan_extracted)
3788                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3789                                           le16_to_cpu(rx_desc->wb.upper.vlan));
3790                 else
3791
3792                         netif_receive_skb(skb);
3793 #ifdef CONFIG_IGB_LRO
3794         }
3795 #endif
3796 }
3797
3798
3799 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3800                                        u32 status_err, struct sk_buff *skb)
3801 {
3802         skb->ip_summed = CHECKSUM_NONE;
3803
3804         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3805         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3806                 return;
3807         /* TCP/UDP checksum error bit is set */
3808         if (status_err &
3809             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3810                 /* let the stack verify checksum errors */
3811                 adapter->hw_csum_err++;
3812                 return;
3813         }
3814         /* It must be a TCP or UDP packet with a valid checksum */
3815         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3816                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3817
3818         adapter->hw_csum_good++;
3819 }
3820
3821 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
3822                                  int *work_done, int budget)
3823 {
3824         struct igb_adapter *adapter = rx_ring->adapter;
3825         struct net_device *netdev = adapter->netdev;
3826         struct pci_dev *pdev = adapter->pdev;
3827         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3828         struct igb_buffer *buffer_info , *next_buffer;
3829         struct sk_buff *skb;
3830         unsigned int i;
3831         u32 length, hlen, staterr;
3832         bool cleaned = false;
3833         int cleaned_count = 0;
3834         unsigned int total_bytes = 0, total_packets = 0;
3835
3836         i = rx_ring->next_to_clean;
3837         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3838         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3839
3840         while (staterr & E1000_RXD_STAT_DD) {
3841                 if (*work_done >= budget)
3842                         break;
3843                 (*work_done)++;
3844                 buffer_info = &rx_ring->buffer_info[i];
3845
3846                 /* HW will not DMA in data larger than the given buffer, even
3847                  * if it parses the (NFS, of course) header to be larger.  In
3848                  * that case, it fills the header buffer and spills the rest
3849                  * into the page.
3850                  */
3851                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
3852                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
3853                 if (hlen > adapter->rx_ps_hdr_size)
3854                         hlen = adapter->rx_ps_hdr_size;
3855
3856                 length = le16_to_cpu(rx_desc->wb.upper.length);
3857                 cleaned = true;
3858                 cleaned_count++;
3859
3860                 skb = buffer_info->skb;
3861                 prefetch(skb->data - NET_IP_ALIGN);
3862                 buffer_info->skb = NULL;
3863                 if (!adapter->rx_ps_hdr_size) {
3864                         pci_unmap_single(pdev, buffer_info->dma,
3865                                          adapter->rx_buffer_len +
3866                                            NET_IP_ALIGN,
3867                                          PCI_DMA_FROMDEVICE);
3868                         skb_put(skb, length);
3869                         goto send_up;
3870                 }
3871
3872                 if (!skb_shinfo(skb)->nr_frags) {
3873                         pci_unmap_single(pdev, buffer_info->dma,
3874                                          adapter->rx_ps_hdr_size +
3875                                            NET_IP_ALIGN,
3876                                          PCI_DMA_FROMDEVICE);
3877                         skb_put(skb, hlen);
3878                 }
3879
3880                 if (length) {
3881                         pci_unmap_page(pdev, buffer_info->page_dma,
3882                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
3883                         buffer_info->page_dma = 0;
3884
3885                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
3886                                                 buffer_info->page,
3887                                                 buffer_info->page_offset,
3888                                                 length);
3889
3890                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
3891                             (page_count(buffer_info->page) != 1))
3892                                 buffer_info->page = NULL;
3893                         else
3894                                 get_page(buffer_info->page);
3895
3896                         skb->len += length;
3897                         skb->data_len += length;
3898
3899                         skb->truesize += length;
3900                 }
3901 send_up:
3902                 i++;
3903                 if (i == rx_ring->count)
3904                         i = 0;
3905                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3906                 prefetch(next_rxd);
3907                 next_buffer = &rx_ring->buffer_info[i];
3908
3909                 if (!(staterr & E1000_RXD_STAT_EOP)) {
3910                         buffer_info->skb = xchg(&next_buffer->skb, skb);
3911                         buffer_info->dma = xchg(&next_buffer->dma, 0);
3912                         goto next_desc;
3913                 }
3914
3915                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3916                         dev_kfree_skb_irq(skb);
3917                         goto next_desc;
3918                 }
3919
3920                 total_bytes += skb->len;
3921                 total_packets++;
3922
3923                 igb_rx_checksum_adv(adapter, staterr, skb);
3924
3925                 skb->protocol = eth_type_trans(skb, netdev);
3926
3927                 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
3928
3929                 netdev->last_rx = jiffies;
3930
3931 next_desc:
3932                 rx_desc->wb.upper.status_error = 0;
3933
3934                 /* return some buffers to hardware, one at a time is too slow */
3935                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3936                         igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3937                         cleaned_count = 0;
3938                 }
3939
3940                 /* use prefetched values */
3941                 rx_desc = next_rxd;
3942                 buffer_info = next_buffer;
3943
3944                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3945         }
3946
3947         rx_ring->next_to_clean = i;
3948         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3949
3950 #ifdef CONFIG_IGB_LRO
3951         if (rx_ring->lro_used) {
3952                 lro_flush_all(&rx_ring->lro_mgr);
3953                 rx_ring->lro_used = 0;
3954         }
3955 #endif
3956
3957         if (cleaned_count)
3958                 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
3959
3960         rx_ring->total_packets += total_packets;
3961         rx_ring->total_bytes += total_bytes;
3962         rx_ring->rx_stats.packets += total_packets;
3963         rx_ring->rx_stats.bytes += total_bytes;
3964         adapter->net_stats.rx_bytes += total_bytes;
3965         adapter->net_stats.rx_packets += total_packets;
3966         return cleaned;
3967 }
3968
3969
3970 /**
3971  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3972  * @adapter: address of board private structure
3973  **/
3974 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
3975                                      int cleaned_count)
3976 {
3977         struct igb_adapter *adapter = rx_ring->adapter;
3978         struct net_device *netdev = adapter->netdev;
3979         struct pci_dev *pdev = adapter->pdev;
3980         union e1000_adv_rx_desc *rx_desc;
3981         struct igb_buffer *buffer_info;
3982         struct sk_buff *skb;
3983         unsigned int i;
3984
3985         i = rx_ring->next_to_use;
3986         buffer_info = &rx_ring->buffer_info[i];
3987
3988         while (cleaned_count--) {
3989                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3990
3991                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
3992                         if (!buffer_info->page) {
3993                                 buffer_info->page = alloc_page(GFP_ATOMIC);
3994                                 if (!buffer_info->page) {
3995                                         adapter->alloc_rx_buff_failed++;
3996                                         goto no_buffers;
3997                                 }
3998                                 buffer_info->page_offset = 0;
3999                         } else {
4000                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
4001                         }
4002                         buffer_info->page_dma =
4003                                 pci_map_page(pdev,
4004                                              buffer_info->page,
4005                                              buffer_info->page_offset,
4006                                              PAGE_SIZE / 2,
4007                                              PCI_DMA_FROMDEVICE);
4008                 }
4009
4010                 if (!buffer_info->skb) {
4011                         int bufsz;
4012
4013                         if (adapter->rx_ps_hdr_size)
4014                                 bufsz = adapter->rx_ps_hdr_size;
4015                         else
4016                                 bufsz = adapter->rx_buffer_len;
4017                         bufsz += NET_IP_ALIGN;
4018                         skb = netdev_alloc_skb(netdev, bufsz);
4019
4020                         if (!skb) {
4021                                 adapter->alloc_rx_buff_failed++;
4022                                 goto no_buffers;
4023                         }
4024
4025                         /* Make buffer alignment 2 beyond a 16 byte boundary
4026                          * this will result in a 16 byte aligned IP header after
4027                          * the 14 byte MAC header is removed
4028                          */
4029                         skb_reserve(skb, NET_IP_ALIGN);
4030
4031                         buffer_info->skb = skb;
4032                         buffer_info->dma = pci_map_single(pdev, skb->data,
4033                                                           bufsz,
4034                                                           PCI_DMA_FROMDEVICE);
4035
4036                 }
4037                 /* Refresh the desc even if buffer_addrs didn't change because
4038                  * each write-back erases this info. */
4039                 if (adapter->rx_ps_hdr_size) {
4040                         rx_desc->read.pkt_addr =
4041                              cpu_to_le64(buffer_info->page_dma);
4042                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4043                 } else {
4044                         rx_desc->read.pkt_addr =
4045                              cpu_to_le64(buffer_info->dma);
4046                         rx_desc->read.hdr_addr = 0;
4047                 }
4048
4049                 i++;
4050                 if (i == rx_ring->count)
4051                         i = 0;
4052                 buffer_info = &rx_ring->buffer_info[i];
4053         }
4054
4055 no_buffers:
4056         if (rx_ring->next_to_use != i) {
4057                 rx_ring->next_to_use = i;
4058                 if (i == 0)
4059                         i = (rx_ring->count - 1);
4060                 else
4061                         i--;
4062
4063                 /* Force memory writes to complete before letting h/w
4064                  * know there are new descriptors to fetch.  (Only
4065                  * applicable for weak-ordered memory model archs,
4066                  * such as IA-64). */
4067                 wmb();
4068                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4069         }
4070 }
4071
4072 /**
4073  * igb_mii_ioctl -
4074  * @netdev:
4075  * @ifreq:
4076  * @cmd:
4077  **/
4078 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4079 {
4080         struct igb_adapter *adapter = netdev_priv(netdev);
4081         struct mii_ioctl_data *data = if_mii(ifr);
4082
4083         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4084                 return -EOPNOTSUPP;
4085
4086         switch (cmd) {
4087         case SIOCGMIIPHY:
4088                 data->phy_id = adapter->hw.phy.addr;
4089                 break;
4090         case SIOCGMIIREG:
4091                 if (!capable(CAP_NET_ADMIN))
4092                         return -EPERM;
4093                 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
4094                                                      data->reg_num
4095                                                      & 0x1F, &data->val_out))
4096                         return -EIO;
4097                 break;
4098         case SIOCSMIIREG:
4099         default:
4100                 return -EOPNOTSUPP;
4101         }
4102         return 0;
4103 }
4104
4105 /**
4106  * igb_ioctl -
4107  * @netdev:
4108  * @ifreq:
4109  * @cmd:
4110  **/
4111 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4112 {
4113         switch (cmd) {
4114         case SIOCGMIIPHY:
4115         case SIOCGMIIREG:
4116         case SIOCSMIIREG:
4117                 return igb_mii_ioctl(netdev, ifr, cmd);
4118         default:
4119                 return -EOPNOTSUPP;
4120         }
4121 }
4122
4123 static void igb_vlan_rx_register(struct net_device *netdev,
4124                                  struct vlan_group *grp)
4125 {
4126         struct igb_adapter *adapter = netdev_priv(netdev);
4127         struct e1000_hw *hw = &adapter->hw;
4128         u32 ctrl, rctl;
4129
4130         igb_irq_disable(adapter);
4131         adapter->vlgrp = grp;
4132
4133         if (grp) {
4134                 /* enable VLAN tag insert/strip */
4135                 ctrl = rd32(E1000_CTRL);
4136                 ctrl |= E1000_CTRL_VME;
4137                 wr32(E1000_CTRL, ctrl);
4138
4139                 /* enable VLAN receive filtering */
4140                 rctl = rd32(E1000_RCTL);
4141                 rctl &= ~E1000_RCTL_CFIEN;
4142                 wr32(E1000_RCTL, rctl);
4143                 igb_update_mng_vlan(adapter);
4144                 wr32(E1000_RLPML,
4145                                 adapter->max_frame_size + VLAN_TAG_SIZE);
4146         } else {
4147                 /* disable VLAN tag insert/strip */
4148                 ctrl = rd32(E1000_CTRL);
4149                 ctrl &= ~E1000_CTRL_VME;
4150                 wr32(E1000_CTRL, ctrl);
4151
4152                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4153                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4154                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4155                 }
4156                 wr32(E1000_RLPML,
4157                                 adapter->max_frame_size);
4158         }
4159
4160         if (!test_bit(__IGB_DOWN, &adapter->state))
4161                 igb_irq_enable(adapter);
4162 }
4163
4164 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4165 {
4166         struct igb_adapter *adapter = netdev_priv(netdev);
4167         struct e1000_hw *hw = &adapter->hw;
4168         u32 vfta, index;
4169
4170         if ((adapter->hw.mng_cookie.status &
4171              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4172             (vid == adapter->mng_vlan_id))
4173                 return;
4174         /* add VID to filter table */
4175         index = (vid >> 5) & 0x7F;
4176         vfta = array_rd32(E1000_VFTA, index);
4177         vfta |= (1 << (vid & 0x1F));
4178         igb_write_vfta(&adapter->hw, index, vfta);
4179 }
4180
4181 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4182 {
4183         struct igb_adapter *adapter = netdev_priv(netdev);
4184         struct e1000_hw *hw = &adapter->hw;
4185         u32 vfta, index;
4186
4187         igb_irq_disable(adapter);
4188         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4189
4190         if (!test_bit(__IGB_DOWN, &adapter->state))
4191                 igb_irq_enable(adapter);
4192
4193         if ((adapter->hw.mng_cookie.status &
4194              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4195             (vid == adapter->mng_vlan_id)) {
4196                 /* release control to f/w */
4197                 igb_release_hw_control(adapter);
4198                 return;
4199         }
4200
4201         /* remove VID from filter table */
4202         index = (vid >> 5) & 0x7F;
4203         vfta = array_rd32(E1000_VFTA, index);
4204         vfta &= ~(1 << (vid & 0x1F));
4205         igb_write_vfta(&adapter->hw, index, vfta);
4206 }
4207
4208 static void igb_restore_vlan(struct igb_adapter *adapter)
4209 {
4210         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4211
4212         if (adapter->vlgrp) {
4213                 u16 vid;
4214                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4215                         if (!vlan_group_get_device(adapter->vlgrp, vid))
4216                                 continue;
4217                         igb_vlan_rx_add_vid(adapter->netdev, vid);
4218                 }
4219         }
4220 }
4221
4222 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
4223 {
4224         struct e1000_mac_info *mac = &adapter->hw.mac;
4225
4226         mac->autoneg = 0;
4227
4228         /* Fiber NICs only allow 1000 gbps Full duplex */
4229         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
4230                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4231                 dev_err(&adapter->pdev->dev,
4232                         "Unsupported Speed/Duplex configuration\n");
4233                 return -EINVAL;
4234         }
4235
4236         switch (spddplx) {
4237         case SPEED_10 + DUPLEX_HALF:
4238                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
4239                 break;
4240         case SPEED_10 + DUPLEX_FULL:
4241                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
4242                 break;
4243         case SPEED_100 + DUPLEX_HALF:
4244                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
4245                 break;
4246         case SPEED_100 + DUPLEX_FULL:
4247                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
4248                 break;
4249         case SPEED_1000 + DUPLEX_FULL:
4250                 mac->autoneg = 1;
4251                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
4252                 break;
4253         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4254         default:
4255                 dev_err(&adapter->pdev->dev,
4256                         "Unsupported Speed/Duplex configuration\n");
4257                 return -EINVAL;
4258         }
4259         return 0;
4260 }
4261
4262
4263 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
4264 {
4265         struct net_device *netdev = pci_get_drvdata(pdev);
4266         struct igb_adapter *adapter = netdev_priv(netdev);
4267         struct e1000_hw *hw = &adapter->hw;
4268         u32 ctrl, rctl, status;
4269         u32 wufc = adapter->wol;
4270 #ifdef CONFIG_PM
4271         int retval = 0;
4272 #endif
4273
4274         netif_device_detach(netdev);
4275
4276         if (netif_running(netdev))
4277                 igb_close(netdev);
4278
4279         igb_reset_interrupt_capability(adapter);
4280
4281         igb_free_queues(adapter);
4282
4283 #ifdef CONFIG_PM
4284         retval = pci_save_state(pdev);
4285         if (retval)
4286                 return retval;
4287 #endif
4288
4289         status = rd32(E1000_STATUS);
4290         if (status & E1000_STATUS_LU)
4291                 wufc &= ~E1000_WUFC_LNKC;
4292
4293         if (wufc) {
4294                 igb_setup_rctl(adapter);
4295                 igb_set_multi(netdev);
4296
4297                 /* turn on all-multi mode if wake on multicast is enabled */
4298                 if (wufc & E1000_WUFC_MC) {
4299                         rctl = rd32(E1000_RCTL);
4300                         rctl |= E1000_RCTL_MPE;
4301                         wr32(E1000_RCTL, rctl);
4302                 }
4303
4304                 ctrl = rd32(E1000_CTRL);
4305                 /* advertise wake from D3Cold */
4306                 #define E1000_CTRL_ADVD3WUC 0x00100000
4307                 /* phy power management enable */
4308                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4309                 ctrl |= E1000_CTRL_ADVD3WUC;
4310                 wr32(E1000_CTRL, ctrl);
4311
4312                 /* Allow time for pending master requests to run */
4313                 igb_disable_pcie_master(&adapter->hw);
4314
4315                 wr32(E1000_WUC, E1000_WUC_PME_EN);
4316                 wr32(E1000_WUFC, wufc);
4317         } else {
4318                 wr32(E1000_WUC, 0);
4319                 wr32(E1000_WUFC, 0);
4320         }
4321
4322         /* make sure adapter isn't asleep if manageability/wol is enabled */
4323         if (wufc || adapter->en_mng_pt) {
4324                 pci_enable_wake(pdev, PCI_D3hot, 1);
4325                 pci_enable_wake(pdev, PCI_D3cold, 1);
4326         } else {
4327                 igb_shutdown_fiber_serdes_link_82575(hw);
4328                 pci_enable_wake(pdev, PCI_D3hot, 0);
4329                 pci_enable_wake(pdev, PCI_D3cold, 0);
4330         }
4331
4332         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4333          * would have already happened in close and is redundant. */
4334         igb_release_hw_control(adapter);
4335
4336         pci_disable_device(pdev);
4337
4338         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4339
4340         return 0;
4341 }
4342
4343 #ifdef CONFIG_PM
4344 static int igb_resume(struct pci_dev *pdev)
4345 {
4346         struct net_device *netdev = pci_get_drvdata(pdev);
4347         struct igb_adapter *adapter = netdev_priv(netdev);
4348         struct e1000_hw *hw = &adapter->hw;
4349         u32 err;
4350
4351         pci_set_power_state(pdev, PCI_D0);
4352         pci_restore_state(pdev);
4353
4354         if (adapter->need_ioport)
4355                 err = pci_enable_device(pdev);
4356         else
4357                 err = pci_enable_device_mem(pdev);
4358         if (err) {
4359                 dev_err(&pdev->dev,
4360                         "igb: Cannot enable PCI device from suspend\n");
4361                 return err;
4362         }
4363         pci_set_master(pdev);
4364
4365         pci_enable_wake(pdev, PCI_D3hot, 0);
4366         pci_enable_wake(pdev, PCI_D3cold, 0);
4367
4368         igb_set_interrupt_capability(adapter);
4369
4370         if (igb_alloc_queues(adapter)) {
4371                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
4372                 return -ENOMEM;
4373         }
4374
4375         /* e1000_power_up_phy(adapter); */
4376
4377         igb_reset(adapter);
4378         wr32(E1000_WUS, ~0);
4379
4380         if (netif_running(netdev)) {
4381                 err = igb_open(netdev);
4382                 if (err)
4383                         return err;
4384         }
4385
4386         netif_device_attach(netdev);
4387
4388         /* let the f/w know that the h/w is now under the control of the
4389          * driver. */
4390         igb_get_hw_control(adapter);
4391
4392         return 0;
4393 }
4394 #endif
4395
4396 static void igb_shutdown(struct pci_dev *pdev)
4397 {
4398         igb_suspend(pdev, PMSG_SUSPEND);
4399 }
4400
4401 #ifdef CONFIG_NET_POLL_CONTROLLER
4402 /*
4403  * Polling 'interrupt' - used by things like netconsole to send skbs
4404  * without having to re-enable interrupts. It's not called while
4405  * the interrupt routine is executing.
4406  */
4407 static void igb_netpoll(struct net_device *netdev)
4408 {
4409         struct igb_adapter *adapter = netdev_priv(netdev);
4410         int i;
4411         int work_done = 0;
4412
4413         igb_irq_disable(adapter);
4414         adapter->flags |= IGB_FLAG_IN_NETPOLL;
4415
4416         for (i = 0; i < adapter->num_tx_queues; i++)
4417                 igb_clean_tx_irq(&adapter->tx_ring[i]);
4418
4419         for (i = 0; i < adapter->num_rx_queues; i++)
4420                 igb_clean_rx_irq_adv(&adapter->rx_ring[i],
4421                                      &work_done,
4422                                      adapter->rx_ring[i].napi.weight);
4423
4424         adapter->flags &= ~IGB_FLAG_IN_NETPOLL;
4425         igb_irq_enable(adapter);
4426 }
4427 #endif /* CONFIG_NET_POLL_CONTROLLER */
4428
4429 /**
4430  * igb_io_error_detected - called when PCI error is detected
4431  * @pdev: Pointer to PCI device
4432  * @state: The current pci connection state
4433  *
4434  * This function is called after a PCI bus error affecting
4435  * this device has been detected.
4436  */
4437 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4438                                               pci_channel_state_t state)
4439 {
4440         struct net_device *netdev = pci_get_drvdata(pdev);
4441         struct igb_adapter *adapter = netdev_priv(netdev);
4442
4443         netif_device_detach(netdev);
4444
4445         if (netif_running(netdev))
4446                 igb_down(adapter);
4447         pci_disable_device(pdev);
4448
4449         /* Request a slot slot reset. */
4450         return PCI_ERS_RESULT_NEED_RESET;
4451 }
4452
4453 /**
4454  * igb_io_slot_reset - called after the pci bus has been reset.
4455  * @pdev: Pointer to PCI device
4456  *
4457  * Restart the card from scratch, as if from a cold-boot. Implementation
4458  * resembles the first-half of the igb_resume routine.
4459  */
4460 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4461 {
4462         struct net_device *netdev = pci_get_drvdata(pdev);
4463         struct igb_adapter *adapter = netdev_priv(netdev);
4464         struct e1000_hw *hw = &adapter->hw;
4465         int err;
4466
4467         if (adapter->need_ioport)
4468                 err = pci_enable_device(pdev);
4469         else
4470                 err = pci_enable_device_mem(pdev);
4471         if (err) {
4472                 dev_err(&pdev->dev,
4473                         "Cannot re-enable PCI device after reset.\n");
4474                 return PCI_ERS_RESULT_DISCONNECT;
4475         }
4476         pci_set_master(pdev);
4477         pci_restore_state(pdev);
4478
4479         pci_enable_wake(pdev, PCI_D3hot, 0);
4480         pci_enable_wake(pdev, PCI_D3cold, 0);
4481
4482         igb_reset(adapter);
4483         wr32(E1000_WUS, ~0);
4484
4485         return PCI_ERS_RESULT_RECOVERED;
4486 }
4487
4488 /**
4489  * igb_io_resume - called when traffic can start flowing again.
4490  * @pdev: Pointer to PCI device
4491  *
4492  * This callback is called when the error recovery driver tells us that
4493  * its OK to resume normal operation. Implementation resembles the
4494  * second-half of the igb_resume routine.
4495  */
4496 static void igb_io_resume(struct pci_dev *pdev)
4497 {
4498         struct net_device *netdev = pci_get_drvdata(pdev);
4499         struct igb_adapter *adapter = netdev_priv(netdev);
4500
4501         if (netif_running(netdev)) {
4502                 if (igb_up(adapter)) {
4503                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4504                         return;
4505                 }
4506         }
4507
4508         netif_device_attach(netdev);
4509
4510         /* let the f/w know that the h/w is now under the control of the
4511          * driver. */
4512         igb_get_hw_control(adapter);
4513
4514 }
4515
4516 /* igb_main.c */