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/tcp.h>
  35 #include <linux/ipv6.h>
  36 #include <net/checksum.h>
  37 #include <net/ip6_checksum.h>
  38 #include <linux/mii.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/if_vlan.h>
  41 #include <linux/pci.h>
  42 #include <linux/delay.h>
  43 #include <linux/interrupt.h>
  44 #include <linux/if_ether.h>
  45
  46 #include "igb.h"
  47
  48 #define DRV_VERSION "1.0.8-k2"
  49 char igb_driver_name[] = "igb";
  50 char igb_driver_version[] = DRV_VERSION;
  51 static const char igb_driver_string[] =
  52                                 "Intel(R) Gigabit Ethernet Network Driver";
  53 static const char igb_copyright[] = "Copyright (c) 2007 Intel Corporation.";
  54
  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_82575EB_COPPER), board_82575 },
  62         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
  63         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
  64         /* required last entry */
  65         {0, }
  66 };
  67
  68 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
  69
  70 void igb_reset(struct igb_adapter *);
  71 static int igb_setup_all_tx_resources(struct igb_adapter *);
  72 static int igb_setup_all_rx_resources(struct igb_adapter *);
  73 static void igb_free_all_tx_resources(struct igb_adapter *);
  74 static void igb_free_all_rx_resources(struct igb_adapter *);
  75 static void igb_free_tx_resources(struct igb_adapter *, struct igb_ring *);
  76 static void igb_free_rx_resources(struct igb_adapter *, struct igb_ring *);
  77 void igb_update_stats(struct igb_adapter *);
  78 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
  79 static void __devexit igb_remove(struct pci_dev *pdev);
  80 static int igb_sw_init(struct igb_adapter *);
  81 static int igb_open(struct net_device *);
  82 static int igb_close(struct net_device *);
  83 static void igb_configure_tx(struct igb_adapter *);
  84 static void igb_configure_rx(struct igb_adapter *);
  85 static void igb_setup_rctl(struct igb_adapter *);
  86 static void igb_clean_all_tx_rings(struct igb_adapter *);
  87 static void igb_clean_all_rx_rings(struct igb_adapter *);
  88 static void igb_clean_tx_ring(struct igb_adapter *, struct igb_ring *);
  89 static void igb_clean_rx_ring(struct igb_adapter *, struct igb_ring *);
  90 static void igb_set_multi(struct net_device *);
  91 static void igb_update_phy_info(unsigned long);
  92 static void igb_watchdog(unsigned long);
  93 static void igb_watchdog_task(struct work_struct *);
  94 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
  95                                   struct igb_ring *);
  96 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
  97 static struct net_device_stats *igb_get_stats(struct net_device *);
  98 static int igb_change_mtu(struct net_device *, int);
  99 static int igb_set_mac(struct net_device *, void *);
 100 static irqreturn_t igb_intr(int irq, void *);
 101 static irqreturn_t igb_intr_msi(int irq, void *);
 102 static irqreturn_t igb_msix_other(int irq, void *);
 103 static irqreturn_t igb_msix_rx(int irq, void *);
 104 static irqreturn_t igb_msix_tx(int irq, void *);
 105 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
 106 static bool igb_clean_tx_irq(struct igb_adapter *, struct igb_ring *);
 107 static int igb_clean(struct napi_struct *, int);
 108 static bool igb_clean_rx_irq_adv(struct igb_adapter *,
 109                                  struct igb_ring *, int *, int);
 110 static void igb_alloc_rx_buffers_adv(struct igb_adapter *,
 111                                      struct igb_ring *, int);
 112 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 113 static void igb_tx_timeout(struct net_device *);
 114 static void igb_reset_task(struct work_struct *);
 115 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
 116 static void igb_vlan_rx_add_vid(struct net_device *, u16);
 117 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
 118 static void igb_restore_vlan(struct igb_adapter *);
 119
 120 static int igb_suspend(struct pci_dev *, pm_message_t);
 121 #ifdef CONFIG_PM
 122 static int igb_resume(struct pci_dev *);
 123 #endif
 124 static void igb_shutdown(struct pci_dev *);
 125
 126 #ifdef CONFIG_NET_POLL_CONTROLLER
 127 /* for netdump / net console */
 128 static void igb_netpoll(struct net_device *);
 129 #endif
 130
 131 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 132                      pci_channel_state_t);
 133 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 134 static void igb_io_resume(struct pci_dev *);
 135
 136 static struct pci_error_handlers igb_err_handler = {
 137         .error_detected = igb_io_error_detected,
 138         .slot_reset = igb_io_slot_reset,
 139         .resume = igb_io_resume,
 140 };
 141
 142
 143 static struct pci_driver igb_driver = {
 144         .name     = igb_driver_name,
 145         .id_table = igb_pci_tbl,
 146         .probe    = igb_probe,
 147         .remove   = __devexit_p(igb_remove),
 148 #ifdef CONFIG_PM
 149         /* Power Managment Hooks */
 150         .suspend  = igb_suspend,
 151         .resume   = igb_resume,
 152 #endif
 153         .shutdown = igb_shutdown,
 154         .err_handler = &igb_err_handler
 155 };
 156
 157 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 158 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 159 MODULE_LICENSE("GPL");
 160 MODULE_VERSION(DRV_VERSION);
 161
 162 #ifdef DEBUG
 163 /**
 164  * igb_get_hw_dev_name - return device name string
 165  * used by hardware layer to print debugging information
 166  **/
 167 char *igb_get_hw_dev_name(struct e1000_hw *hw)
 168 {
 169         struct igb_adapter *adapter = hw->back;
 170         return adapter->netdev->name;
 171 }
 172 #endif
 173
 174 /**
 175  * igb_init_module - Driver Registration Routine
 176  *
 177  * igb_init_module is the first routine called when the driver is
 178  * loaded. All it does is register with the PCI subsystem.
 179  **/
 180 static int __init igb_init_module(void)
 181 {
 182         int ret;
 183         printk(KERN_INFO "%s - version %s\n",
 184                igb_driver_string, igb_driver_version);
 185
 186         printk(KERN_INFO "%s\n", igb_copyright);
 187
 188         ret = pci_register_driver(&igb_driver);
 189         return ret;
 190 }
 191
 192 module_init(igb_init_module);
 193
 194 /**
 195  * igb_exit_module - Driver Exit Cleanup Routine
 196  *
 197  * igb_exit_module is called just before the driver is removed
 198  * from memory.
 199  **/
 200 static void __exit igb_exit_module(void)
 201 {
 202         pci_unregister_driver(&igb_driver);
 203 }
 204
 205 module_exit(igb_exit_module);
 206
 207 /**
 208  * igb_alloc_queues - Allocate memory for all rings
 209  * @adapter: board private structure to initialize
 210  *
 211  * We allocate one ring per queue at run-time since we don't know the
 212  * number of queues at compile-time.
 213  **/
 214 static int igb_alloc_queues(struct igb_adapter *adapter)
 215 {
 216         int i;
 217
 218         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
 219                                    sizeof(struct igb_ring), GFP_KERNEL);
 220         if (!adapter->tx_ring)
 221                 return -ENOMEM;
 222
 223         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
 224                                    sizeof(struct igb_ring), GFP_KERNEL);
 225         if (!adapter->rx_ring) {
 226                 kfree(adapter->tx_ring);
 227                 return -ENOMEM;
 228         }
 229
 230         for (i = 0; i < adapter->num_rx_queues; i++) {
 231                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 232                 ring->adapter = adapter;
 233                 ring->itr_register = E1000_ITR;
 234
 235                 if (!ring->napi.poll)
 236                         netif_napi_add(adapter->netdev, &ring->napi, igb_clean,
 237                                        adapter->napi.weight /
 238                                        adapter->num_rx_queues);
 239         }
 240         return 0;
 241 }
 242
 243 #define IGB_N0_QUEUE -1
 244 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
 245                               int tx_queue, int msix_vector)
 246 {
 247         u32 msixbm = 0;
 248         struct e1000_hw *hw = &adapter->hw;
 249                 /* The 82575 assigns vectors using a bitmask, which matches the
 250                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
 251                    or more queues to a vector, we write the appropriate bits
 252                    into the MSIXBM register for that vector. */
 253                 if (rx_queue > IGB_N0_QUEUE) {
 254                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 255                         adapter->rx_ring[rx_queue].eims_value = msixbm;
 256                 }
 257                 if (tx_queue > IGB_N0_QUEUE) {
 258                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 259                         adapter->tx_ring[tx_queue].eims_value =
 260                                   E1000_EICR_TX_QUEUE0 << tx_queue;
 261                 }
 262                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 263 }
 264
 265 /**
 266  * igb_configure_msix - Configure MSI-X hardware
 267  *
 268  * igb_configure_msix sets up the hardware to properly
 269  * generate MSI-X interrupts.
 270  **/
 271 static void igb_configure_msix(struct igb_adapter *adapter)
 272 {
 273         u32 tmp;
 274         int i, vector = 0;
 275         struct e1000_hw *hw = &adapter->hw;
 276
 277         adapter->eims_enable_mask = 0;
 278
 279         for (i = 0; i < adapter->num_tx_queues; i++) {
 280                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
 281                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
 282                 adapter->eims_enable_mask |= tx_ring->eims_value;
 283                 if (tx_ring->itr_val)
 284                         writel(1000000000 / (tx_ring->itr_val * 256),
 285                                hw->hw_addr + tx_ring->itr_register);
 286                 else
 287                         writel(1, hw->hw_addr + tx_ring->itr_register);
 288         }
 289
 290         for (i = 0; i < adapter->num_rx_queues; i++) {
 291                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
 292                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
 293                 adapter->eims_enable_mask |= rx_ring->eims_value;
 294                 if (rx_ring->itr_val)
 295                         writel(1000000000 / (rx_ring->itr_val * 256),
 296                                hw->hw_addr + rx_ring->itr_register);
 297                 else
 298                         writel(1, hw->hw_addr + rx_ring->itr_register);
 299         }
 300
 301
 302         /* set vector for other causes, i.e. link changes */
 303                 array_wr32(E1000_MSIXBM(0), vector++,
 304                                       E1000_EIMS_OTHER);
 305
 306                 /* disable IAM for ICR interrupt bits */
 307                 wr32(E1000_IAM, 0);
 308
 309                 tmp = rd32(E1000_CTRL_EXT);
 310                 /* enable MSI-X PBA support*/
 311                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 312
 313                 /* Auto-Mask interrupts upon ICR read. */
 314                 tmp |= E1000_CTRL_EXT_EIAME;
 315                 tmp |= E1000_CTRL_EXT_IRCA;
 316
 317                 wr32(E1000_CTRL_EXT, tmp);
 318                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
 319
 320         wrfl();
 321 }
 322
 323 /**
 324  * igb_request_msix - Initialize MSI-X interrupts
 325  *
 326  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
 327  * kernel.
 328  **/
 329 static int igb_request_msix(struct igb_adapter *adapter)
 330 {
 331         struct net_device *netdev = adapter->netdev;
 332         int i, err = 0, vector = 0;
 333
 334         vector = 0;
 335
 336         for (i = 0; i < adapter->num_tx_queues; i++) {
 337                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 338                 sprintf(ring->name, "%s-tx%d", netdev->name, i);
 339                 err = request_irq(adapter->msix_entries[vector].vector,
 340                                   &igb_msix_tx, 0, ring->name,
 341                                   &(adapter->tx_ring[i]));
 342                 if (err)
 343                         goto out;
 344                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 345                 ring->itr_val = adapter->itr;
 346                 vector++;
 347         }
 348         for (i = 0; i < adapter->num_rx_queues; i++) {
 349                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 350                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
 351                         sprintf(ring->name, "%s-rx%d", netdev->name, i);
 352                 else
 353                         memcpy(ring->name, netdev->name, IFNAMSIZ);
 354                 err = request_irq(adapter->msix_entries[vector].vector,
 355                                   &igb_msix_rx, 0, ring->name,
 356                                   &(adapter->rx_ring[i]));
 357                 if (err)
 358                         goto out;
 359                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 360                 ring->itr_val = adapter->itr;
 361                 vector++;
 362         }
 363
 364         err = request_irq(adapter->msix_entries[vector].vector,
 365                           &igb_msix_other, 0, netdev->name, netdev);
 366         if (err)
 367                 goto out;
 368
 369         adapter->napi.poll = igb_clean_rx_ring_msix;
 370         for (i = 0; i < adapter->num_rx_queues; i++)
 371                 adapter->rx_ring[i].napi.poll = adapter->napi.poll;
 372         igb_configure_msix(adapter);
 373         return 0;
 374 out:
 375         return err;
 376 }
 377
 378 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 379 {
 380         if (adapter->msix_entries) {
 381                 pci_disable_msix(adapter->pdev);
 382                 kfree(adapter->msix_entries);
 383                 adapter->msix_entries = NULL;
 384         } else if (adapter->msi_enabled)
 385                 pci_disable_msi(adapter->pdev);
 386         return;
 387 }
 388
 389
 390 /**
 391  * igb_set_interrupt_capability - set MSI or MSI-X if supported
 392  *
 393  * Attempt to configure interrupts using the best available
 394  * capabilities of the hardware and kernel.
 395  **/
 396 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
 397 {
 398         int err;
 399         int numvecs, i;
 400
 401         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
 402         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
 403                                         GFP_KERNEL);
 404         if (!adapter->msix_entries)
 405                 goto msi_only;
 406
 407         for (i = 0; i < numvecs; i++)
 408                 adapter->msix_entries[i].entry = i;
 409
 410         err = pci_enable_msix(adapter->pdev,
 411                               adapter->msix_entries,
 412                               numvecs);
 413         if (err == 0)
 414                 return;
 415
 416         igb_reset_interrupt_capability(adapter);
 417
 418         /* If we can't do MSI-X, try MSI */
 419 msi_only:
 420         adapter->num_rx_queues = 1;
 421         if (!pci_enable_msi(adapter->pdev))
 422                 adapter->msi_enabled = 1;
 423         return;
 424 }
 425
 426 /**
 427  * igb_request_irq - initialize interrupts
 428  *
 429  * Attempts to configure interrupts using the best available
 430  * capabilities of the hardware and kernel.
 431  **/
 432 static int igb_request_irq(struct igb_adapter *adapter)
 433 {
 434         struct net_device *netdev = adapter->netdev;
 435         struct e1000_hw *hw = &adapter->hw;
 436         int err = 0;
 437
 438         if (adapter->msix_entries) {
 439                 err = igb_request_msix(adapter);
 440                 if (!err) {
 441                         /* enable IAM, auto-mask,
 442                          * DO NOT USE EIAME or IAME in legacy mode */
 443                         wr32(E1000_IAM, IMS_ENABLE_MASK);
 444                         goto request_done;
 445                 }
 446                 /* fall back to MSI */
 447                 igb_reset_interrupt_capability(adapter);
 448                 if (!pci_enable_msi(adapter->pdev))
 449                         adapter->msi_enabled = 1;
 450                 igb_free_all_tx_resources(adapter);
 451                 igb_free_all_rx_resources(adapter);
 452                 adapter->num_rx_queues = 1;
 453                 igb_alloc_queues(adapter);
 454         }
 455         if (adapter->msi_enabled) {
 456                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
 457                                   netdev->name, netdev);
 458                 if (!err)
 459                         goto request_done;
 460                 /* fall back to legacy interrupts */
 461                 igb_reset_interrupt_capability(adapter);
 462                 adapter->msi_enabled = 0;
 463         }
 464
 465         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
 466                           netdev->name, netdev);
 467
 468         if (err) {
 469                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
 470                         err);
 471                 goto request_done;
 472         }
 473
 474         /* enable IAM, auto-mask */
 475         wr32(E1000_IAM, IMS_ENABLE_MASK);
 476
 477 request_done:
 478         return err;
 479 }
 480
 481 static void igb_free_irq(struct igb_adapter *adapter)
 482 {
 483         struct net_device *netdev = adapter->netdev;
 484
 485         if (adapter->msix_entries) {
 486                 int vector = 0, i;
 487
 488                 for (i = 0; i < adapter->num_tx_queues; i++)
 489                         free_irq(adapter->msix_entries[vector++].vector,
 490                                 &(adapter->tx_ring[i]));
 491                 for (i = 0; i < adapter->num_rx_queues; i++)
 492                         free_irq(adapter->msix_entries[vector++].vector,
 493                                 &(adapter->rx_ring[i]));
 494
 495                 free_irq(adapter->msix_entries[vector++].vector, netdev);
 496                 return;
 497         }
 498
 499         free_irq(adapter->pdev->irq, netdev);
 500 }
 501
 502 /**
 503  * igb_irq_disable - Mask off interrupt generation on the NIC
 504  * @adapter: board private structure
 505  **/
 506 static void igb_irq_disable(struct igb_adapter *adapter)
 507 {
 508         struct e1000_hw *hw = &adapter->hw;
 509
 510         if (adapter->msix_entries) {
 511                 wr32(E1000_EIMC, ~0);
 512                 wr32(E1000_EIAC, 0);
 513         }
 514         wr32(E1000_IMC, ~0);
 515         wrfl();
 516         synchronize_irq(adapter->pdev->irq);
 517 }
 518
 519 /**
 520  * igb_irq_enable - Enable default interrupt generation settings
 521  * @adapter: board private structure
 522  **/
 523 static void igb_irq_enable(struct igb_adapter *adapter)
 524 {
 525         struct e1000_hw *hw = &adapter->hw;
 526
 527         if (adapter->msix_entries) {
 528                 wr32(E1000_EIMS,
 529                                 adapter->eims_enable_mask);
 530                 wr32(E1000_EIAC,
 531                                 adapter->eims_enable_mask);
 532                 wr32(E1000_IMS, E1000_IMS_LSC);
 533         } else
 534         wr32(E1000_IMS, IMS_ENABLE_MASK);
 535 }
 536
 537 static void igb_update_mng_vlan(struct igb_adapter *adapter)
 538 {
 539         struct net_device *netdev = adapter->netdev;
 540         u16 vid = adapter->hw.mng_cookie.vlan_id;
 541         u16 old_vid = adapter->mng_vlan_id;
 542         if (adapter->vlgrp) {
 543                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
 544                         if (adapter->hw.mng_cookie.status &
 545                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
 546                                 igb_vlan_rx_add_vid(netdev, vid);
 547                                 adapter->mng_vlan_id = vid;
 548                         } else
 549                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
 550
 551                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
 552                                         (vid != old_vid) &&
 553                             !vlan_group_get_device(adapter->vlgrp, old_vid))
 554                                 igb_vlan_rx_kill_vid(netdev, old_vid);
 555                 } else
 556                         adapter->mng_vlan_id = vid;
 557         }
 558 }
 559
 560 /**
 561  * igb_release_hw_control - release control of the h/w to f/w
 562  * @adapter: address of board private structure
 563  *
 564  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 565  * For ASF and Pass Through versions of f/w this means that the
 566  * driver is no longer loaded.
 567  *
 568  **/
 569 static void igb_release_hw_control(struct igb_adapter *adapter)
 570 {
 571         struct e1000_hw *hw = &adapter->hw;
 572         u32 ctrl_ext;
 573
 574         /* Let firmware take over control of h/w */
 575         ctrl_ext = rd32(E1000_CTRL_EXT);
 576         wr32(E1000_CTRL_EXT,
 577                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 578 }
 579
 580
 581 /**
 582  * igb_get_hw_control - get control of the h/w from f/w
 583  * @adapter: address of board private structure
 584  *
 585  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 586  * For ASF and Pass Through versions of f/w this means that
 587  * the driver is loaded.
 588  *
 589  **/
 590 static void igb_get_hw_control(struct igb_adapter *adapter)
 591 {
 592         struct e1000_hw *hw = &adapter->hw;
 593         u32 ctrl_ext;
 594
 595         /* Let firmware know the driver has taken over */
 596         ctrl_ext = rd32(E1000_CTRL_EXT);
 597         wr32(E1000_CTRL_EXT,
 598                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 599 }
 600
 601 static void igb_init_manageability(struct igb_adapter *adapter)
 602 {
 603         struct e1000_hw *hw = &adapter->hw;
 604
 605         if (adapter->en_mng_pt) {
 606                 u32 manc2h = rd32(E1000_MANC2H);
 607                 u32 manc = rd32(E1000_MANC);
 608
 609                 /* disable hardware interception of ARP */
 610                 manc &= ~(E1000_MANC_ARP_EN);
 611
 612                 /* enable receiving management packets to the host */
 613                 /* this will probably generate destination unreachable messages
 614                  * from the host OS, but the packets will be handled on SMBUS */
 615                 manc |= E1000_MANC_EN_MNG2HOST;
 616 #define E1000_MNG2HOST_PORT_623 (1 << 5)
 617 #define E1000_MNG2HOST_PORT_664 (1 << 6)
 618                 manc2h |= E1000_MNG2HOST_PORT_623;
 619                 manc2h |= E1000_MNG2HOST_PORT_664;
 620                 wr32(E1000_MANC2H, manc2h);
 621
 622                 wr32(E1000_MANC, manc);
 623         }
 624 }
 625
 626 static void igb_release_manageability(struct igb_adapter *adapter)
 627 {
 628         struct e1000_hw *hw = &adapter->hw;
 629
 630         if (adapter->en_mng_pt) {
 631                 u32 manc = rd32(E1000_MANC);
 632
 633                 /* re-enable hardware interception of ARP */
 634                 manc |= E1000_MANC_ARP_EN;
 635                 manc &= ~E1000_MANC_EN_MNG2HOST;
 636
 637                 /* don't explicitly have to mess with MANC2H since
 638                  * MANC has an enable disable that gates MANC2H */
 639
 640                 /* XXX stop the hardware watchdog ? */
 641                 wr32(E1000_MANC, manc);
 642         }
 643 }
 644
 645 /**
 646  * igb_configure - configure the hardware for RX and TX
 647  * @adapter: private board structure
 648  **/
 649 static void igb_configure(struct igb_adapter *adapter)
 650 {
 651         struct net_device *netdev = adapter->netdev;
 652         int i;
 653
 654         igb_get_hw_control(adapter);
 655         igb_set_multi(netdev);
 656
 657         igb_restore_vlan(adapter);
 658         igb_init_manageability(adapter);
 659
 660         igb_configure_tx(adapter);
 661         igb_setup_rctl(adapter);
 662         igb_configure_rx(adapter);
 663         /* call IGB_DESC_UNUSED which always leaves
 664          * at least 1 descriptor unused to make sure
 665          * next_to_use != next_to_clean */
 666         for (i = 0; i < adapter->num_rx_queues; i++) {
 667                 struct igb_ring *ring = &adapter->rx_ring[i];
 668                 igb_alloc_rx_buffers_adv(adapter, ring, IGB_DESC_UNUSED(ring));
 669         }
 670
 671
 672         adapter->tx_queue_len = netdev->tx_queue_len;
 673 }
 674
 675
 676 /**
 677  * igb_up - Open the interface and prepare it to handle traffic
 678  * @adapter: board private structure
 679  **/
 680
 681 int igb_up(struct igb_adapter *adapter)
 682 {
 683         struct e1000_hw *hw = &adapter->hw;
 684         int i;
 685
 686         /* hardware has been reset, we need to reload some things */
 687         igb_configure(adapter);
 688
 689         clear_bit(__IGB_DOWN, &adapter->state);
 690
 691         napi_enable(&adapter->napi);
 692
 693         if (adapter->msix_entries) {
 694                 for (i = 0; i < adapter->num_rx_queues; i++)
 695                         napi_enable(&adapter->rx_ring[i].napi);
 696                 igb_configure_msix(adapter);
 697         }
 698
 699         /* Clear any pending interrupts. */
 700         rd32(E1000_ICR);
 701         igb_irq_enable(adapter);
 702
 703         /* Fire a link change interrupt to start the watchdog. */
 704         wr32(E1000_ICS, E1000_ICS_LSC);
 705         return 0;
 706 }
 707
 708 void igb_down(struct igb_adapter *adapter)
 709 {
 710         struct e1000_hw *hw = &adapter->hw;
 711         struct net_device *netdev = adapter->netdev;
 712         u32 tctl, rctl;
 713         int i;
 714
 715         /* signal that we're down so the interrupt handler does not
 716          * reschedule our watchdog timer */
 717         set_bit(__IGB_DOWN, &adapter->state);
 718
 719         /* disable receives in the hardware */
 720         rctl = rd32(E1000_RCTL);
 721         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
 722         /* flush and sleep below */
 723
 724         netif_stop_queue(netdev);
 725
 726         /* disable transmits in the hardware */
 727         tctl = rd32(E1000_TCTL);
 728         tctl &= ~E1000_TCTL_EN;
 729         wr32(E1000_TCTL, tctl);
 730         /* flush both disables and wait for them to finish */
 731         wrfl();
 732         msleep(10);
 733
 734         napi_disable(&adapter->napi);
 735
 736         if (adapter->msix_entries)
 737                 for (i = 0; i < adapter->num_rx_queues; i++)
 738                         napi_disable(&adapter->rx_ring[i].napi);
 739         igb_irq_disable(adapter);
 740
 741         del_timer_sync(&adapter->watchdog_timer);
 742         del_timer_sync(&adapter->phy_info_timer);
 743
 744         netdev->tx_queue_len = adapter->tx_queue_len;
 745         netif_carrier_off(netdev);
 746         adapter->link_speed = 0;
 747         adapter->link_duplex = 0;
 748
 749         igb_reset(adapter);
 750         igb_clean_all_tx_rings(adapter);
 751         igb_clean_all_rx_rings(adapter);
 752 }
 753
 754 void igb_reinit_locked(struct igb_adapter *adapter)
 755 {
 756         WARN_ON(in_interrupt());
 757         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 758                 msleep(1);
 759         igb_down(adapter);
 760         igb_up(adapter);
 761         clear_bit(__IGB_RESETTING, &adapter->state);
 762 }
 763
 764 void igb_reset(struct igb_adapter *adapter)
 765 {
 766         struct e1000_hw *hw = &adapter->hw;
 767         struct e1000_fc_info *fc = &adapter->hw.fc;
 768         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
 769         u16 hwm;
 770
 771         /* Repartition Pba for greater than 9k mtu
 772          * To take effect CTRL.RST is required.
 773          */
 774         pba = E1000_PBA_34K;
 775
 776         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
 777                 /* adjust PBA for jumbo frames */
 778                 wr32(E1000_PBA, pba);
 779
 780                 /* To maintain wire speed transmits, the Tx FIFO should be
 781                  * large enough to accommodate two full transmit packets,
 782                  * rounded up to the next 1KB and expressed in KB.  Likewise,
 783                  * the Rx FIFO should be large enough to accommodate at least
 784                  * one full receive packet and is similarly rounded up and
 785                  * expressed in KB. */
 786                 pba = rd32(E1000_PBA);
 787                 /* upper 16 bits has Tx packet buffer allocation size in KB */
 788                 tx_space = pba >> 16;
 789                 /* lower 16 bits has Rx packet buffer allocation size in KB */
 790                 pba &= 0xffff;
 791                 /* the tx fifo also stores 16 bytes of information about the tx
 792                  * but don't include ethernet FCS because hardware appends it */
 793                 min_tx_space = (adapter->max_frame_size +
 794                                 sizeof(struct e1000_tx_desc) -
 795                                 ETH_FCS_LEN) * 2;
 796                 min_tx_space = ALIGN(min_tx_space, 1024);
 797                 min_tx_space >>= 10;
 798                 /* software strips receive CRC, so leave room for it */
 799                 min_rx_space = adapter->max_frame_size;
 800                 min_rx_space = ALIGN(min_rx_space, 1024);
 801                 min_rx_space >>= 10;
 802
 803                 /* If current Tx allocation is less than the min Tx FIFO size,
 804                  * and the min Tx FIFO size is less than the current Rx FIFO
 805                  * allocation, take space away from current Rx allocation */
 806                 if (tx_space < min_tx_space &&
 807                     ((min_tx_space - tx_space) < pba)) {
 808                         pba = pba - (min_tx_space - tx_space);
 809
 810                         /* if short on rx space, rx wins and must trump tx
 811                          * adjustment */
 812                         if (pba < min_rx_space)
 813                                 pba = min_rx_space;
 814                 }
 815         }
 816         wr32(E1000_PBA, pba);
 817
 818         /* flow control settings */
 819         /* The high water mark must be low enough to fit one full frame
 820          * (or the size used for early receive) above it in the Rx FIFO.
 821          * Set it to the lower of:
 822          * - 90% of the Rx FIFO size, or
 823          * - the full Rx FIFO size minus one full frame */
 824         hwm = min(((pba << 10) * 9 / 10),
 825                   ((pba << 10) - adapter->max_frame_size));
 826
 827         fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
 828         fc->low_water = fc->high_water - 8;
 829         fc->pause_time = 0xFFFF;
 830         fc->send_xon = 1;
 831         fc->type = fc->original_type;
 832
 833         /* Allow time for pending master requests to run */
 834         adapter->hw.mac.ops.reset_hw(&adapter->hw);
 835         wr32(E1000_WUC, 0);
 836
 837         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
 838                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
 839
 840         igb_update_mng_vlan(adapter);
 841
 842         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 843         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
 844
 845         igb_reset_adaptive(&adapter->hw);
 846         adapter->hw.phy.ops.get_phy_info(&adapter->hw);
 847         igb_release_manageability(adapter);
 848 }
 849
 850 /**
 851  * igb_probe - Device Initialization Routine
 852  * @pdev: PCI device information struct
 853  * @ent: entry in igb_pci_tbl
 854  *
 855  * Returns 0 on success, negative on failure
 856  *
 857  * igb_probe initializes an adapter identified by a pci_dev structure.
 858  * The OS initialization, configuring of the adapter private structure,
 859  * and a hardware reset occur.
 860  **/
 861 static int __devinit igb_probe(struct pci_dev *pdev,
 862                                const struct pci_device_id *ent)
 863 {
 864         struct net_device *netdev;
 865         struct igb_adapter *adapter;
 866         struct e1000_hw *hw;
 867         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
 868         unsigned long mmio_start, mmio_len;
 869         static int cards_found;
 870         int i, err, pci_using_dac;
 871         u16 eeprom_data = 0;
 872         u16 eeprom_apme_mask = IGB_EEPROM_APME;
 873         u32 part_num;
 874
 875         err = pci_enable_device(pdev);
 876         if (err)
 877                 return err;
 878
 879         pci_using_dac = 0;
 880         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
 881         if (!err) {
 882                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
 883                 if (!err)
 884                         pci_using_dac = 1;
 885         } else {
 886                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
 887                 if (err) {
 888                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
 889                         if (err) {
 890                                 dev_err(&pdev->dev, "No usable DMA "
 891                                         "configuration, aborting\n");
 892                                 goto err_dma;
 893                         }
 894                 }
 895         }
 896
 897         err = pci_request_regions(pdev, igb_driver_name);
 898         if (err)
 899                 goto err_pci_reg;
 900
 901         pci_set_master(pdev);
 902
 903         err = -ENOMEM;
 904         netdev = alloc_etherdev(sizeof(struct igb_adapter));
 905         if (!netdev)
 906                 goto err_alloc_etherdev;
 907
 908         SET_NETDEV_DEV(netdev, &pdev->dev);
 909
 910         pci_set_drvdata(pdev, netdev);
 911         adapter = netdev_priv(netdev);
 912         adapter->netdev = netdev;
 913         adapter->pdev = pdev;
 914         hw = &adapter->hw;
 915         hw->back = adapter;
 916         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
 917
 918         mmio_start = pci_resource_start(pdev, 0);
 919         mmio_len = pci_resource_len(pdev, 0);
 920
 921         err = -EIO;
 922         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
 923         if (!adapter->hw.hw_addr)
 924                 goto err_ioremap;
 925
 926         netdev->open = &igb_open;
 927         netdev->stop = &igb_close;
 928         netdev->get_stats = &igb_get_stats;
 929         netdev->set_multicast_list = &igb_set_multi;
 930         netdev->set_mac_address = &igb_set_mac;
 931         netdev->change_mtu = &igb_change_mtu;
 932         netdev->do_ioctl = &igb_ioctl;
 933         igb_set_ethtool_ops(netdev);
 934         netdev->tx_timeout = &igb_tx_timeout;
 935         netdev->watchdog_timeo = 5 * HZ;
 936         netif_napi_add(netdev, &adapter->napi, igb_clean, 64);
 937         netdev->vlan_rx_register = igb_vlan_rx_register;
 938         netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
 939         netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
 940 #ifdef CONFIG_NET_POLL_CONTROLLER
 941         netdev->poll_controller = igb_netpoll;
 942 #endif
 943         netdev->hard_start_xmit = &igb_xmit_frame_adv;
 944
 945         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 946
 947         netdev->mem_start = mmio_start;
 948         netdev->mem_end = mmio_start + mmio_len;
 949
 950         adapter->bd_number = cards_found;
 951
 952         /* PCI config space info */
 953         hw->vendor_id = pdev->vendor;
 954         hw->device_id = pdev->device;
 955         hw->revision_id = pdev->revision;
 956         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 957         hw->subsystem_device_id = pdev->subsystem_device;
 958
 959         /* setup the private structure */
 960         hw->back = adapter;
 961         /* Copy the default MAC, PHY and NVM function pointers */
 962         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
 963         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
 964         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
 965         /* Initialize skew-specific constants */
 966         err = ei->get_invariants(hw);
 967         if (err)
 968                 goto err_hw_init;
 969
 970         err = igb_sw_init(adapter);
 971         if (err)
 972                 goto err_sw_init;
 973
 974         igb_get_bus_info_pcie(hw);
 975
 976         hw->phy.autoneg_wait_to_complete = false;
 977         hw->mac.adaptive_ifs = true;
 978
 979         /* Copper options */
 980         if (hw->phy.media_type == e1000_media_type_copper) {
 981                 hw->phy.mdix = AUTO_ALL_MODES;
 982                 hw->phy.disable_polarity_correction = false;
 983                 hw->phy.ms_type = e1000_ms_hw_default;
 984         }
 985
 986         if (igb_check_reset_block(hw))
 987                 dev_info(&pdev->dev,
 988                         "PHY reset is blocked due to SOL/IDER session.\n");
 989
 990         netdev->features = NETIF_F_SG |
 991                            NETIF_F_HW_CSUM |
 992                            NETIF_F_HW_VLAN_TX |
 993                            NETIF_F_HW_VLAN_RX |
 994                            NETIF_F_HW_VLAN_FILTER;
 995
 996         netdev->features |= NETIF_F_TSO;
 997
 998         netdev->features |= NETIF_F_TSO6;
 999         if (pci_using_dac)
1000                 netdev->features |= NETIF_F_HIGHDMA;
1001
1002         netdev->features |= NETIF_F_LLTX;
1003         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1004
1005         /* before reading the NVM, reset the controller to put the device in a
1006          * known good starting state */
1007         hw->mac.ops.reset_hw(hw);
1008
1009         /* make sure the NVM is good */
1010         if (igb_validate_nvm_checksum(hw) < 0) {
1011                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1012                 err = -EIO;
1013                 goto err_eeprom;
1014         }
1015
1016         /* copy the MAC address out of the NVM */
1017         if (hw->mac.ops.read_mac_addr(hw))
1018                 dev_err(&pdev->dev, "NVM Read Error\n");
1019
1020         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1021         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1022
1023         if (!is_valid_ether_addr(netdev->perm_addr)) {
1024                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1025                 err = -EIO;
1026                 goto err_eeprom;
1027         }
1028
1029         init_timer(&adapter->watchdog_timer);
1030         adapter->watchdog_timer.function = &igb_watchdog;
1031         adapter->watchdog_timer.data = (unsigned long) adapter;
1032
1033         init_timer(&adapter->phy_info_timer);
1034         adapter->phy_info_timer.function = &igb_update_phy_info;
1035         adapter->phy_info_timer.data = (unsigned long) adapter;
1036
1037         INIT_WORK(&adapter->reset_task, igb_reset_task);
1038         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1039
1040         /* Initialize link & ring properties that are user-changeable */
1041         adapter->tx_ring->count = 256;
1042         for (i = 0; i < adapter->num_tx_queues; i++)
1043                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1044         adapter->rx_ring->count = 256;
1045         for (i = 0; i < adapter->num_rx_queues; i++)
1046                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1047
1048         adapter->fc_autoneg = true;
1049         hw->mac.autoneg = true;
1050         hw->phy.autoneg_advertised = 0x2f;
1051
1052         hw->fc.original_type = e1000_fc_default;
1053         hw->fc.type = e1000_fc_default;
1054
1055         adapter->itr_setting = 3;
1056         adapter->itr = IGB_START_ITR;
1057
1058         igb_validate_mdi_setting(hw);
1059
1060         adapter->rx_csum = 1;
1061
1062         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1063          * enable the ACPI Magic Packet filter
1064          */
1065
1066         if (hw->bus.func == 0 ||
1067             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1068                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1069                                      &eeprom_data);
1070
1071         if (eeprom_data & eeprom_apme_mask)
1072                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1073
1074         /* now that we have the eeprom settings, apply the special cases where
1075          * the eeprom may be wrong or the board simply won't support wake on
1076          * lan on a particular port */
1077         switch (pdev->device) {
1078         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1079                 adapter->eeprom_wol = 0;
1080                 break;
1081         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1082                 /* Wake events only supported on port A for dual fiber
1083                  * regardless of eeprom setting */
1084                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1085                         adapter->eeprom_wol = 0;
1086                 break;
1087         }
1088
1089         /* initialize the wol settings based on the eeprom settings */
1090         adapter->wol = adapter->eeprom_wol;
1091
1092         /* reset the hardware with the new settings */
1093         igb_reset(adapter);
1094
1095         /* let the f/w know that the h/w is now under the control of the
1096          * driver. */
1097         igb_get_hw_control(adapter);
1098
1099         /* tell the stack to leave us alone until igb_open() is called */
1100         netif_carrier_off(netdev);
1101         netif_stop_queue(netdev);
1102
1103         strcpy(netdev->name, "eth%d");
1104         err = register_netdev(netdev);
1105         if (err)
1106                 goto err_register;
1107
1108         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1109         /* print bus type/speed/width info */
1110         dev_info(&pdev->dev,
1111                  "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1112                  netdev->name,
1113                  ((hw->bus.speed == e1000_bus_speed_2500)
1114                   ? "2.5Gb/s" : "unknown"),
1115                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1116                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1117                   ? "Width x1" : "unknown"),
1118                  netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1119                  netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1120
1121         igb_read_part_num(hw, &part_num);
1122         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1123                 (part_num >> 8), (part_num & 0xff));
1124
1125         dev_info(&pdev->dev,
1126                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1127                 adapter->msix_entries ? "MSI-X" :
1128                 adapter->msi_enabled ? "MSI" : "legacy",
1129                 adapter->num_rx_queues, adapter->num_tx_queues);
1130
1131         cards_found++;
1132         return 0;
1133
1134 err_register:
1135         igb_release_hw_control(adapter);
1136 err_eeprom:
1137         if (!igb_check_reset_block(hw))
1138                 hw->phy.ops.reset_phy(hw);
1139
1140         if (hw->flash_address)
1141                 iounmap(hw->flash_address);
1142
1143         igb_remove_device(hw);
1144         kfree(adapter->tx_ring);
1145         kfree(adapter->rx_ring);
1146 err_sw_init:
1147 err_hw_init:
1148         iounmap(hw->hw_addr);
1149 err_ioremap:
1150         free_netdev(netdev);
1151 err_alloc_etherdev:
1152         pci_release_regions(pdev);
1153 err_pci_reg:
1154 err_dma:
1155         pci_disable_device(pdev);
1156         return err;
1157 }
1158
1159 /**
1160  * igb_remove - Device Removal Routine
1161  * @pdev: PCI device information struct
1162  *
1163  * igb_remove is called by the PCI subsystem to alert the driver
1164  * that it should release a PCI device.  The could be caused by a
1165  * Hot-Plug event, or because the driver is going to be removed from
1166  * memory.
1167  **/
1168 static void __devexit igb_remove(struct pci_dev *pdev)
1169 {
1170         struct net_device *netdev = pci_get_drvdata(pdev);
1171         struct igb_adapter *adapter = netdev_priv(netdev);
1172
1173         /* flush_scheduled work may reschedule our watchdog task, so
1174          * explicitly disable watchdog tasks from being rescheduled  */
1175         set_bit(__IGB_DOWN, &adapter->state);
1176         del_timer_sync(&adapter->watchdog_timer);
1177         del_timer_sync(&adapter->phy_info_timer);
1178
1179         flush_scheduled_work();
1180
1181
1182         igb_release_manageability(adapter);
1183
1184         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1185          * would have already happened in close and is redundant. */
1186         igb_release_hw_control(adapter);
1187
1188         unregister_netdev(netdev);
1189
1190         if (!igb_check_reset_block(&adapter->hw))
1191                 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1192
1193         igb_remove_device(&adapter->hw);
1194         igb_reset_interrupt_capability(adapter);
1195
1196         kfree(adapter->tx_ring);
1197         kfree(adapter->rx_ring);
1198
1199         iounmap(adapter->hw.hw_addr);
1200         if (adapter->hw.flash_address)
1201                 iounmap(adapter->hw.flash_address);
1202         pci_release_regions(pdev);
1203
1204         free_netdev(netdev);
1205
1206         pci_disable_device(pdev);
1207 }
1208
1209 /**
1210  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1211  * @adapter: board private structure to initialize
1212  *
1213  * igb_sw_init initializes the Adapter private data structure.
1214  * Fields are initialized based on PCI device information and
1215  * OS network device settings (MTU size).
1216  **/
1217 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1218 {
1219         struct e1000_hw *hw = &adapter->hw;
1220         struct net_device *netdev = adapter->netdev;
1221         struct pci_dev *pdev = adapter->pdev;
1222
1223         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1224
1225         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1226         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1227         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1228         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1229
1230         /* Number of supported queues. */
1231         /* Having more queues than CPUs doesn't make sense. */
1232         adapter->num_tx_queues = 1;
1233         adapter->num_rx_queues = min(IGB_MAX_RX_QUEUES, num_online_cpus());
1234
1235         igb_set_interrupt_capability(adapter);
1236
1237         if (igb_alloc_queues(adapter)) {
1238                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1239                 return -ENOMEM;
1240         }
1241
1242         /* Explicitly disable IRQ since the NIC can be in any state. */
1243         igb_irq_disable(adapter);
1244
1245         set_bit(__IGB_DOWN, &adapter->state);
1246         return 0;
1247 }
1248
1249 /**
1250  * igb_open - Called when a network interface is made active
1251  * @netdev: network interface device structure
1252  *
1253  * Returns 0 on success, negative value on failure
1254  *
1255  * The open entry point is called when a network interface is made
1256  * active by the system (IFF_UP).  At this point all resources needed
1257  * for transmit and receive operations are allocated, the interrupt
1258  * handler is registered with the OS, the watchdog timer is started,
1259  * and the stack is notified that the interface is ready.
1260  **/
1261 static int igb_open(struct net_device *netdev)
1262 {
1263         struct igb_adapter *adapter = netdev_priv(netdev);
1264         struct e1000_hw *hw = &adapter->hw;
1265         int err;
1266         int i;
1267
1268         /* disallow open during test */
1269         if (test_bit(__IGB_TESTING, &adapter->state))
1270                 return -EBUSY;
1271
1272         /* allocate transmit descriptors */
1273         err = igb_setup_all_tx_resources(adapter);
1274         if (err)
1275                 goto err_setup_tx;
1276
1277         /* allocate receive descriptors */
1278         err = igb_setup_all_rx_resources(adapter);
1279         if (err)
1280                 goto err_setup_rx;
1281
1282         /* e1000_power_up_phy(adapter); */
1283
1284         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1285         if ((adapter->hw.mng_cookie.status &
1286              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1287                 igb_update_mng_vlan(adapter);
1288
1289         /* before we allocate an interrupt, we must be ready to handle it.
1290          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1291          * as soon as we call pci_request_irq, so we have to setup our
1292          * clean_rx handler before we do so.  */
1293         igb_configure(adapter);
1294
1295         err = igb_request_irq(adapter);
1296         if (err)
1297                 goto err_req_irq;
1298
1299         /* From here on the code is the same as igb_up() */
1300         clear_bit(__IGB_DOWN, &adapter->state);
1301
1302         napi_enable(&adapter->napi);
1303         if (adapter->msix_entries)
1304                 for (i = 0; i < adapter->num_rx_queues; i++)
1305                         napi_enable(&adapter->rx_ring[i].napi);
1306
1307         igb_irq_enable(adapter);
1308
1309         /* Clear any pending interrupts. */
1310         rd32(E1000_ICR);
1311         /* Fire a link status change interrupt to start the watchdog. */
1312         wr32(E1000_ICS, E1000_ICS_LSC);
1313
1314         return 0;
1315
1316 err_req_irq:
1317         igb_release_hw_control(adapter);
1318         /* e1000_power_down_phy(adapter); */
1319         igb_free_all_rx_resources(adapter);
1320 err_setup_rx:
1321         igb_free_all_tx_resources(adapter);
1322 err_setup_tx:
1323         igb_reset(adapter);
1324
1325         return err;
1326 }
1327
1328 /**
1329  * igb_close - Disables a network interface
1330  * @netdev: network interface device structure
1331  *
1332  * Returns 0, this is not allowed to fail
1333  *
1334  * The close entry point is called when an interface is de-activated
1335  * by the OS.  The hardware is still under the driver's control, but
1336  * needs to be disabled.  A global MAC reset is issued to stop the
1337  * hardware, and all transmit and receive resources are freed.
1338  **/
1339 static int igb_close(struct net_device *netdev)
1340 {
1341         struct igb_adapter *adapter = netdev_priv(netdev);
1342
1343         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1344         igb_down(adapter);
1345
1346         igb_free_irq(adapter);
1347
1348         igb_free_all_tx_resources(adapter);
1349         igb_free_all_rx_resources(adapter);
1350
1351         /* kill manageability vlan ID if supported, but not if a vlan with
1352          * the same ID is registered on the host OS (let 8021q kill it) */
1353         if ((adapter->hw.mng_cookie.status &
1354                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1355              !(adapter->vlgrp &&
1356                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1357                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1358
1359         return 0;
1360 }
1361
1362 /**
1363  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1364  * @adapter: board private structure
1365  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1366  *
1367  * Return 0 on success, negative on failure
1368  **/
1369
1370 int igb_setup_tx_resources(struct igb_adapter *adapter,
1371                            struct igb_ring *tx_ring)
1372 {
1373         struct pci_dev *pdev = adapter->pdev;
1374         int size;
1375
1376         size = sizeof(struct igb_buffer) * tx_ring->count;
1377         tx_ring->buffer_info = vmalloc(size);
1378         if (!tx_ring->buffer_info)
1379                 goto err;
1380         memset(tx_ring->buffer_info, 0, size);
1381
1382         /* round up to nearest 4K */
1383         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1384                         + sizeof(u32);
1385         tx_ring->size = ALIGN(tx_ring->size, 4096);
1386
1387         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1388                                              &tx_ring->dma);
1389
1390         if (!tx_ring->desc)
1391                 goto err;
1392
1393         tx_ring->adapter = adapter;
1394         tx_ring->next_to_use = 0;
1395         tx_ring->next_to_clean = 0;
1396         spin_lock_init(&tx_ring->tx_clean_lock);
1397         spin_lock_init(&tx_ring->tx_lock);
1398         return 0;
1399
1400 err:
1401         vfree(tx_ring->buffer_info);
1402         dev_err(&adapter->pdev->dev,
1403                 "Unable to allocate memory for the transmit descriptor ring\n");
1404         return -ENOMEM;
1405 }
1406
1407 /**
1408  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1409  *                                (Descriptors) for all queues
1410  * @adapter: board private structure
1411  *
1412  * Return 0 on success, negative on failure
1413  **/
1414 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1415 {
1416         int i, err = 0;
1417
1418         for (i = 0; i < adapter->num_tx_queues; i++) {
1419                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1420                 if (err) {
1421                         dev_err(&adapter->pdev->dev,
1422                                 "Allocation for Tx Queue %u failed\n", i);
1423                         for (i--; i >= 0; i--)
1424                                 igb_free_tx_resources(adapter,
1425                                                         &adapter->tx_ring[i]);
1426                         break;
1427                 }
1428         }
1429
1430         return err;
1431 }
1432
1433 /**
1434  * igb_configure_tx - Configure transmit Unit after Reset
1435  * @adapter: board private structure
1436  *
1437  * Configure the Tx unit of the MAC after a reset.
1438  **/
1439 static void igb_configure_tx(struct igb_adapter *adapter)
1440 {
1441         u64 tdba, tdwba;
1442         struct e1000_hw *hw = &adapter->hw;
1443         u32 tctl;
1444         u32 txdctl, txctrl;
1445         int i;
1446
1447         for (i = 0; i < adapter->num_tx_queues; i++) {
1448                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1449
1450                 wr32(E1000_TDLEN(i),
1451                                 ring->count * sizeof(struct e1000_tx_desc));
1452                 tdba = ring->dma;
1453                 wr32(E1000_TDBAL(i),
1454                                 tdba & 0x00000000ffffffffULL);
1455                 wr32(E1000_TDBAH(i), tdba >> 32);
1456
1457                 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1458                 tdwba |= 1; /* enable head wb */
1459                 wr32(E1000_TDWBAL(i),
1460                                 tdwba & 0x00000000ffffffffULL);
1461                 wr32(E1000_TDWBAH(i), tdwba >> 32);
1462
1463                 ring->head = E1000_TDH(i);
1464                 ring->tail = E1000_TDT(i);
1465                 writel(0, hw->hw_addr + ring->tail);
1466                 writel(0, hw->hw_addr + ring->head);
1467                 txdctl = rd32(E1000_TXDCTL(i));
1468                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1469                 wr32(E1000_TXDCTL(i), txdctl);
1470
1471                 /* Turn off Relaxed Ordering on head write-backs.  The
1472                  * writebacks MUST be delivered in order or it will
1473                  * completely screw up our bookeeping.
1474                  */
1475                 txctrl = rd32(E1000_DCA_TXCTRL(i));
1476                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1477                 wr32(E1000_DCA_TXCTRL(i), txctrl);
1478         }
1479
1480
1481
1482         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1483
1484         /* Program the Transmit Control Register */
1485
1486         tctl = rd32(E1000_TCTL);
1487         tctl &= ~E1000_TCTL_CT;
1488         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1489                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1490
1491         igb_config_collision_dist(hw);
1492
1493         /* Setup Transmit Descriptor Settings for eop descriptor */
1494         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1495
1496         /* Enable transmits */
1497         tctl |= E1000_TCTL_EN;
1498
1499         wr32(E1000_TCTL, tctl);
1500 }
1501
1502 /**
1503  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1504  * @adapter: board private structure
1505  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1506  *
1507  * Returns 0 on success, negative on failure
1508  **/
1509
1510 int igb_setup_rx_resources(struct igb_adapter *adapter,
1511                            struct igb_ring *rx_ring)
1512 {
1513         struct pci_dev *pdev = adapter->pdev;
1514         int size, desc_len;
1515
1516         size = sizeof(struct igb_buffer) * rx_ring->count;
1517         rx_ring->buffer_info = vmalloc(size);
1518         if (!rx_ring->buffer_info)
1519                 goto err;
1520         memset(rx_ring->buffer_info, 0, size);
1521
1522         desc_len = sizeof(union e1000_adv_rx_desc);
1523
1524         /* Round up to nearest 4K */
1525         rx_ring->size = rx_ring->count * desc_len;
1526         rx_ring->size = ALIGN(rx_ring->size, 4096);
1527
1528         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1529                                              &rx_ring->dma);
1530
1531         if (!rx_ring->desc)
1532                 goto err;
1533
1534         rx_ring->next_to_clean = 0;
1535         rx_ring->next_to_use = 0;
1536         rx_ring->pending_skb = NULL;
1537
1538         rx_ring->adapter = adapter;
1539         /* FIXME: do we want to setup ring->napi->poll here? */
1540         rx_ring->napi.poll = adapter->napi.poll;
1541
1542         return 0;
1543
1544 err:
1545         vfree(rx_ring->buffer_info);
1546         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1547                 "the receive descriptor ring\n");
1548         return -ENOMEM;
1549 }
1550
1551 /**
1552  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1553  *                                (Descriptors) for all queues
1554  * @adapter: board private structure
1555  *
1556  * Return 0 on success, negative on failure
1557  **/
1558 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1559 {
1560         int i, err = 0;
1561
1562         for (i = 0; i < adapter->num_rx_queues; i++) {
1563                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1564                 if (err) {
1565                         dev_err(&adapter->pdev->dev,
1566                                 "Allocation for Rx Queue %u failed\n", i);
1567                         for (i--; i >= 0; i--)
1568                                 igb_free_rx_resources(adapter,
1569                                                         &adapter->rx_ring[i]);
1570                         break;
1571                 }
1572         }
1573
1574         return err;
1575 }
1576
1577 /**
1578  * igb_setup_rctl - configure the receive control registers
1579  * @adapter: Board private structure
1580  **/
1581 static void igb_setup_rctl(struct igb_adapter *adapter)
1582 {
1583         struct e1000_hw *hw = &adapter->hw;
1584         u32 rctl;
1585         u32 srrctl = 0;
1586         int i;
1587
1588         rctl = rd32(E1000_RCTL);
1589
1590         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1591
1592         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1593                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1594                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1595
1596         /* disable the stripping of CRC because it breaks
1597          * BMC firmware connected over SMBUS
1598         rctl |= E1000_RCTL_SECRC;
1599         */
1600
1601         rctl &= ~E1000_RCTL_SBP;
1602
1603         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1604                 rctl &= ~E1000_RCTL_LPE;
1605         else
1606                 rctl |= E1000_RCTL_LPE;
1607         if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1608                 /* Setup buffer sizes */
1609                 rctl &= ~E1000_RCTL_SZ_4096;
1610                 rctl |= E1000_RCTL_BSEX;
1611                 switch (adapter->rx_buffer_len) {
1612                 case IGB_RXBUFFER_256:
1613                         rctl |= E1000_RCTL_SZ_256;
1614                         rctl &= ~E1000_RCTL_BSEX;
1615                         break;
1616                 case IGB_RXBUFFER_512:
1617                         rctl |= E1000_RCTL_SZ_512;
1618                         rctl &= ~E1000_RCTL_BSEX;
1619                         break;
1620                 case IGB_RXBUFFER_1024:
1621                         rctl |= E1000_RCTL_SZ_1024;
1622                         rctl &= ~E1000_RCTL_BSEX;
1623                         break;
1624                 case IGB_RXBUFFER_2048:
1625                 default:
1626                         rctl |= E1000_RCTL_SZ_2048;
1627                         rctl &= ~E1000_RCTL_BSEX;
1628                         break;
1629                 case IGB_RXBUFFER_4096:
1630                         rctl |= E1000_RCTL_SZ_4096;
1631                         break;
1632                 case IGB_RXBUFFER_8192:
1633                         rctl |= E1000_RCTL_SZ_8192;
1634                         break;
1635                 case IGB_RXBUFFER_16384:
1636                         rctl |= E1000_RCTL_SZ_16384;
1637                         break;
1638                 }
1639         } else {
1640                 rctl &= ~E1000_RCTL_BSEX;
1641                 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1642         }
1643
1644         /* 82575 and greater support packet-split where the protocol
1645          * header is placed in skb->data and the packet data is
1646          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1647          * In the case of a non-split, skb->data is linearly filled,
1648          * followed by the page buffers.  Therefore, skb->data is
1649          * sized to hold the largest protocol header.
1650          */
1651         /* allocations using alloc_page take too long for regular MTU
1652          * so only enable packet split for jumbo frames */
1653         if (rctl & E1000_RCTL_LPE) {
1654                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1655                 srrctl = adapter->rx_ps_hdr_size <<
1656                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1657                 /* buffer size is ALWAYS one page */
1658                 srrctl |= PAGE_SIZE >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1659                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1660         } else {
1661                 adapter->rx_ps_hdr_size = 0;
1662                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1663         }
1664
1665         for (i = 0; i < adapter->num_rx_queues; i++)
1666                 wr32(E1000_SRRCTL(i), srrctl);
1667
1668         wr32(E1000_RCTL, rctl);
1669 }
1670
1671 /**
1672  * igb_configure_rx - Configure receive Unit after Reset
1673  * @adapter: board private structure
1674  *
1675  * Configure the Rx unit of the MAC after a reset.
1676  **/
1677 static void igb_configure_rx(struct igb_adapter *adapter)
1678 {
1679         u64 rdba;
1680         struct e1000_hw *hw = &adapter->hw;
1681         u32 rctl, rxcsum;
1682         u32 rxdctl;
1683         int i;
1684
1685         /* disable receives while setting up the descriptors */
1686         rctl = rd32(E1000_RCTL);
1687         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1688         wrfl();
1689         mdelay(10);
1690
1691         if (adapter->itr_setting > 3)
1692                 wr32(E1000_ITR,
1693                                 1000000000 / (adapter->itr * 256));
1694
1695         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1696          * the Base and Length of the Rx Descriptor Ring */
1697         for (i = 0; i < adapter->num_rx_queues; i++) {
1698                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1699                 rdba = ring->dma;
1700                 wr32(E1000_RDBAL(i),
1701                                 rdba & 0x00000000ffffffffULL);
1702                 wr32(E1000_RDBAH(i), rdba >> 32);
1703                 wr32(E1000_RDLEN(i),
1704                                ring->count * sizeof(union e1000_adv_rx_desc));
1705
1706                 ring->head = E1000_RDH(i);
1707                 ring->tail = E1000_RDT(i);
1708                 writel(0, hw->hw_addr + ring->tail);
1709                 writel(0, hw->hw_addr + ring->head);
1710
1711                 rxdctl = rd32(E1000_RXDCTL(i));
1712                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1713                 rxdctl &= 0xFFF00000;
1714                 rxdctl |= IGB_RX_PTHRESH;
1715                 rxdctl |= IGB_RX_HTHRESH << 8;
1716                 rxdctl |= IGB_RX_WTHRESH << 16;
1717                 wr32(E1000_RXDCTL(i), rxdctl);
1718         }
1719
1720         if (adapter->num_rx_queues > 1) {
1721                 u32 random[10];
1722                 u32 mrqc;
1723                 u32 j, shift;
1724                 union e1000_reta {
1725                         u32 dword;
1726                         u8  bytes[4];
1727                 } reta;
1728
1729                 get_random_bytes(&random[0], 40);
1730
1731                 shift = 6;
1732                 for (j = 0; j < (32 * 4); j++) {
1733                         reta.bytes[j & 3] =
1734                                 (j % adapter->num_rx_queues) << shift;
1735                         if ((j & 3) == 3)
1736                                 writel(reta.dword,
1737                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1738                 }
1739                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1740
1741                 /* Fill out hash function seeds */
1742                 for (j = 0; j < 10; j++)
1743                         array_wr32(E1000_RSSRK(0), j, random[j]);
1744
1745                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1746                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1747                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1748                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1749                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1750                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1751                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1752                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1753
1754
1755                 wr32(E1000_MRQC, mrqc);
1756
1757                 /* Multiqueue and raw packet checksumming are mutually
1758                  * exclusive.  Note that this not the same as TCP/IP
1759                  * checksumming, which works fine. */
1760                 rxcsum = rd32(E1000_RXCSUM);
1761                 rxcsum |= E1000_RXCSUM_PCSD;
1762                 wr32(E1000_RXCSUM, rxcsum);
1763         } else {
1764                 /* Enable Receive Checksum Offload for TCP and UDP */
1765                 rxcsum = rd32(E1000_RXCSUM);
1766                 if (adapter->rx_csum) {
1767                         rxcsum |= E1000_RXCSUM_TUOFL;
1768
1769                         /* Enable IPv4 payload checksum for UDP fragments
1770                          * Must be used in conjunction with packet-split. */
1771                         if (adapter->rx_ps_hdr_size)
1772                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1773                 } else {
1774                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1775                         /* don't need to clear IPPCSE as it defaults to 0 */
1776                 }
1777                 wr32(E1000_RXCSUM, rxcsum);
1778         }
1779
1780         if (adapter->vlgrp)
1781                 wr32(E1000_RLPML,
1782                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1783         else
1784                 wr32(E1000_RLPML, adapter->max_frame_size);
1785
1786         /* Enable Receives */
1787         wr32(E1000_RCTL, rctl);
1788 }
1789
1790 /**
1791  * igb_free_tx_resources - Free Tx Resources per Queue
1792  * @adapter: board private structure
1793  * @tx_ring: Tx descriptor ring for a specific queue
1794  *
1795  * Free all transmit software resources
1796  **/
1797 static void igb_free_tx_resources(struct igb_adapter *adapter,
1798                                   struct igb_ring *tx_ring)
1799 {
1800         struct pci_dev *pdev = adapter->pdev;
1801
1802         igb_clean_tx_ring(adapter, tx_ring);
1803
1804         vfree(tx_ring->buffer_info);
1805         tx_ring->buffer_info = NULL;
1806
1807         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1808
1809         tx_ring->desc = NULL;
1810 }
1811
1812 /**
1813  * igb_free_all_tx_resources - Free Tx Resources for All Queues
1814  * @adapter: board private structure
1815  *
1816  * Free all transmit software resources
1817  **/
1818 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
1819 {
1820         int i;
1821
1822         for (i = 0; i < adapter->num_tx_queues; i++)
1823                 igb_free_tx_resources(adapter, &adapter->tx_ring[i]);
1824 }
1825
1826 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
1827                                            struct igb_buffer *buffer_info)
1828 {
1829         if (buffer_info->dma) {
1830                 pci_unmap_page(adapter->pdev,
1831                                 buffer_info->dma,
1832                                 buffer_info->length,
1833                                 PCI_DMA_TODEVICE);
1834                 buffer_info->dma = 0;
1835         }
1836         if (buffer_info->skb) {
1837                 dev_kfree_skb_any(buffer_info->skb);
1838                 buffer_info->skb = NULL;
1839         }
1840         buffer_info->time_stamp = 0;
1841         /* buffer_info must be completely set up in the transmit path */
1842 }
1843
1844 /**
1845  * igb_clean_tx_ring - Free Tx Buffers
1846  * @adapter: board private structure
1847  * @tx_ring: ring to be cleaned
1848  **/
1849 static void igb_clean_tx_ring(struct igb_adapter *adapter,
1850                               struct igb_ring *tx_ring)
1851 {
1852         struct igb_buffer *buffer_info;
1853         unsigned long size;
1854         unsigned int i;
1855
1856         if (!tx_ring->buffer_info)
1857                 return;
1858         /* Free all the Tx ring sk_buffs */
1859
1860         for (i = 0; i < tx_ring->count; i++) {
1861                 buffer_info = &tx_ring->buffer_info[i];
1862                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
1863         }
1864
1865         size = sizeof(struct igb_buffer) * tx_ring->count;
1866         memset(tx_ring->buffer_info, 0, size);
1867
1868         /* Zero out the descriptor ring */
1869
1870         memset(tx_ring->desc, 0, tx_ring->size);
1871
1872         tx_ring->next_to_use = 0;
1873         tx_ring->next_to_clean = 0;
1874
1875         writel(0, adapter->hw.hw_addr + tx_ring->head);
1876         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1877 }
1878
1879 /**
1880  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
1881  * @adapter: board private structure
1882  **/
1883 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
1884 {
1885         int i;
1886
1887         for (i = 0; i < adapter->num_tx_queues; i++)
1888                 igb_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1889 }
1890
1891 /**
1892  * igb_free_rx_resources - Free Rx Resources
1893  * @adapter: board private structure
1894  * @rx_ring: ring to clean the resources from
1895  *
1896  * Free all receive software resources
1897  **/
1898 static void igb_free_rx_resources(struct igb_adapter *adapter,
1899                                   struct igb_ring *rx_ring)
1900 {
1901         struct pci_dev *pdev = adapter->pdev;
1902
1903         igb_clean_rx_ring(adapter, rx_ring);
1904
1905         vfree(rx_ring->buffer_info);
1906         rx_ring->buffer_info = NULL;
1907
1908         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1909
1910         rx_ring->desc = NULL;
1911 }
1912
1913 /**
1914  * igb_free_all_rx_resources - Free Rx Resources for All Queues
1915  * @adapter: board private structure
1916  *
1917  * Free all receive software resources
1918  **/
1919 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
1920 {
1921         int i;
1922
1923         for (i = 0; i < adapter->num_rx_queues; i++)
1924                 igb_free_rx_resources(adapter, &adapter->rx_ring[i]);
1925 }
1926
1927 /**
1928  * igb_clean_rx_ring - Free Rx Buffers per Queue
1929  * @adapter: board private structure
1930  * @rx_ring: ring to free buffers from
1931  **/
1932 static void igb_clean_rx_ring(struct igb_adapter *adapter,
1933                               struct igb_ring *rx_ring)
1934 {
1935         struct igb_buffer *buffer_info;
1936         struct pci_dev *pdev = adapter->pdev;
1937         unsigned long size;
1938         unsigned int i;
1939
1940         if (!rx_ring->buffer_info)
1941                 return;
1942         /* Free all the Rx ring sk_buffs */
1943         for (i = 0; i < rx_ring->count; i++) {
1944                 buffer_info = &rx_ring->buffer_info[i];
1945                 if (buffer_info->dma) {
1946                         if (adapter->rx_ps_hdr_size)
1947                                 pci_unmap_single(pdev, buffer_info->dma,
1948                                                  adapter->rx_ps_hdr_size,
1949                                                  PCI_DMA_FROMDEVICE);
1950                         else
1951                                 pci_unmap_single(pdev, buffer_info->dma,
1952                                                  adapter->rx_buffer_len,
1953                                                  PCI_DMA_FROMDEVICE);
1954                         buffer_info->dma = 0;
1955                 }
1956
1957                 if (buffer_info->skb) {
1958                         dev_kfree_skb(buffer_info->skb);
1959                         buffer_info->skb = NULL;
1960                 }
1961                 if (buffer_info->page) {
1962                         pci_unmap_page(pdev, buffer_info->page_dma,
1963                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
1964                         put_page(buffer_info->page);
1965                         buffer_info->page = NULL;
1966                         buffer_info->page_dma = 0;
1967                 }
1968         }
1969
1970         /* there also may be some cached data from a chained receive */
1971         if (rx_ring->pending_skb) {
1972                 dev_kfree_skb(rx_ring->pending_skb);
1973                 rx_ring->pending_skb = NULL;
1974         }
1975
1976         size = sizeof(struct igb_buffer) * rx_ring->count;
1977         memset(rx_ring->buffer_info, 0, size);
1978
1979         /* Zero out the descriptor ring */
1980         memset(rx_ring->desc, 0, rx_ring->size);
1981
1982         rx_ring->next_to_clean = 0;
1983         rx_ring->next_to_use = 0;
1984
1985         writel(0, adapter->hw.hw_addr + rx_ring->head);
1986         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1987 }
1988
1989 /**
1990  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
1991  * @adapter: board private structure
1992  **/
1993 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
1994 {
1995         int i;
1996
1997         for (i = 0; i < adapter->num_rx_queues; i++)
1998                 igb_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1999 }
2000
2001 /**
2002  * igb_set_mac - Change the Ethernet Address of the NIC
2003  * @netdev: network interface device structure
2004  * @p: pointer to an address structure
2005  *
2006  * Returns 0 on success, negative on failure
2007  **/
2008 static int igb_set_mac(struct net_device *netdev, void *p)
2009 {
2010         struct igb_adapter *adapter = netdev_priv(netdev);
2011         struct sockaddr *addr = p;
2012
2013         if (!is_valid_ether_addr(addr->sa_data))
2014                 return -EADDRNOTAVAIL;
2015
2016         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2017         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
2018
2019         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
2020
2021         return 0;
2022 }
2023
2024 /**
2025  * igb_set_multi - Multicast and Promiscuous mode set
2026  * @netdev: network interface device structure
2027  *
2028  * The set_multi entry point is called whenever the multicast address
2029  * list or the network interface flags are updated.  This routine is
2030  * responsible for configuring the hardware for proper multicast,
2031  * promiscuous mode, and all-multi behavior.
2032  **/
2033 static void igb_set_multi(struct net_device *netdev)
2034 {
2035         struct igb_adapter *adapter = netdev_priv(netdev);
2036         struct e1000_hw *hw = &adapter->hw;
2037         struct e1000_mac_info *mac = &hw->mac;
2038         struct dev_mc_list *mc_ptr;
2039         u8  *mta_list;
2040         u32 rctl;
2041         int i;
2042
2043         /* Check for Promiscuous and All Multicast modes */
2044
2045         rctl = rd32(E1000_RCTL);
2046
2047         if (netdev->flags & IFF_PROMISC)
2048                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2049         else if (netdev->flags & IFF_ALLMULTI) {
2050                 rctl |= E1000_RCTL_MPE;
2051                 rctl &= ~E1000_RCTL_UPE;
2052         } else
2053                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2054
2055         wr32(E1000_RCTL, rctl);
2056
2057         if (!netdev->mc_count) {
2058                 /* nothing to program, so clear mc list */
2059                 igb_update_mc_addr_list(hw, NULL, 0, 1,
2060                                           mac->rar_entry_count);
2061                 return;
2062         }
2063
2064         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2065         if (!mta_list)
2066                 return;
2067
2068         /* The shared function expects a packed array of only addresses. */
2069         mc_ptr = netdev->mc_list;
2070
2071         for (i = 0; i < netdev->mc_count; i++) {
2072                 if (!mc_ptr)
2073                         break;
2074                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2075                 mc_ptr = mc_ptr->next;
2076         }
2077         igb_update_mc_addr_list(hw, mta_list, i, 1, mac->rar_entry_count);
2078         kfree(mta_list);
2079 }
2080
2081 /* Need to wait a few seconds after link up to get diagnostic information from
2082  * the phy */
2083 static void igb_update_phy_info(unsigned long data)
2084 {
2085         struct igb_adapter *adapter = (struct igb_adapter *) data;
2086         adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2087 }
2088
2089 /**
2090  * igb_watchdog - Timer Call-back
2091  * @data: pointer to adapter cast into an unsigned long
2092  **/
2093 static void igb_watchdog(unsigned long data)
2094 {
2095         struct igb_adapter *adapter = (struct igb_adapter *)data;
2096         /* Do the rest outside of interrupt context */
2097         schedule_work(&adapter->watchdog_task);
2098 }
2099
2100 static void igb_watchdog_task(struct work_struct *work)
2101 {
2102         struct igb_adapter *adapter = container_of(work,
2103                                         struct igb_adapter, watchdog_task);
2104         struct e1000_hw *hw = &adapter->hw;
2105
2106         struct net_device *netdev = adapter->netdev;
2107         struct igb_ring *tx_ring = adapter->tx_ring;
2108         struct e1000_mac_info *mac = &adapter->hw.mac;
2109         u32 link;
2110         s32 ret_val;
2111
2112         if ((netif_carrier_ok(netdev)) &&
2113             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2114                 goto link_up;
2115
2116         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2117         if ((ret_val == E1000_ERR_PHY) &&
2118             (hw->phy.type == e1000_phy_igp_3) &&
2119             (rd32(E1000_CTRL) &
2120              E1000_PHY_CTRL_GBE_DISABLE))
2121                 dev_info(&adapter->pdev->dev,
2122                          "Gigabit has been disabled, downgrading speed\n");
2123
2124         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2125             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2126                 link = mac->serdes_has_link;
2127         else
2128                 link = rd32(E1000_STATUS) &
2129                                       E1000_STATUS_LU;
2130
2131         if (link) {
2132                 if (!netif_carrier_ok(netdev)) {
2133                         u32 ctrl;
2134                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2135                                                    &adapter->link_speed,
2136                                                    &adapter->link_duplex);
2137
2138                         ctrl = rd32(E1000_CTRL);
2139                         dev_info(&adapter->pdev->dev,
2140                                  "NIC Link is Up %d Mbps %s, "
2141                                  "Flow Control: %s\n",
2142                                  adapter->link_speed,
2143                                  adapter->link_duplex == FULL_DUPLEX ?
2144                                  "Full Duplex" : "Half Duplex",
2145                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2146                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2147                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2148                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2149
2150                         /* tweak tx_queue_len according to speed/duplex and
2151                          * adjust the timeout factor */
2152                         netdev->tx_queue_len = adapter->tx_queue_len;
2153                         adapter->tx_timeout_factor = 1;
2154                         switch (adapter->link_speed) {
2155                         case SPEED_10:
2156                                 netdev->tx_queue_len = 10;
2157                                 adapter->tx_timeout_factor = 14;
2158                                 break;
2159                         case SPEED_100:
2160                                 netdev->tx_queue_len = 100;
2161                                 /* maybe add some timeout factor ? */
2162                                 break;
2163                         }
2164
2165                         netif_carrier_on(netdev);
2166                         netif_wake_queue(netdev);
2167
2168                         if (!test_bit(__IGB_DOWN, &adapter->state))
2169                                 mod_timer(&adapter->phy_info_timer,
2170                                           round_jiffies(jiffies + 2 * HZ));
2171                 }
2172         } else {
2173                 if (netif_carrier_ok(netdev)) {
2174                         adapter->link_speed = 0;
2175                         adapter->link_duplex = 0;
2176                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2177                         netif_carrier_off(netdev);
2178                         netif_stop_queue(netdev);
2179                         if (!test_bit(__IGB_DOWN, &adapter->state))
2180                                 mod_timer(&adapter->phy_info_timer,
2181                                           round_jiffies(jiffies + 2 * HZ));
2182                 }
2183         }
2184
2185 link_up:
2186         igb_update_stats(adapter);
2187
2188         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2189         adapter->tpt_old = adapter->stats.tpt;
2190         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2191         adapter->colc_old = adapter->stats.colc;
2192
2193         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2194         adapter->gorc_old = adapter->stats.gorc;
2195         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2196         adapter->gotc_old = adapter->stats.gotc;
2197
2198         igb_update_adaptive(&adapter->hw);
2199
2200         if (!netif_carrier_ok(netdev)) {
2201                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2202                         /* We've lost link, so the controller stops DMA,
2203                          * but we've got queued Tx work that's never going
2204                          * to get done, so reset controller to flush Tx.
2205                          * (Do the reset outside of interrupt context). */
2206                         adapter->tx_timeout_count++;
2207                         schedule_work(&adapter->reset_task);
2208                 }
2209         }
2210
2211         /* Cause software interrupt to ensure rx ring is cleaned */
2212         wr32(E1000_ICS, E1000_ICS_RXDMT0);
2213
2214         /* Force detection of hung controller every watchdog period */
2215         tx_ring->detect_tx_hung = true;
2216
2217         /* Reset the timer */
2218         if (!test_bit(__IGB_DOWN, &adapter->state))
2219                 mod_timer(&adapter->watchdog_timer,
2220                           round_jiffies(jiffies + 2 * HZ));
2221 }
2222
2223 enum latency_range {
2224         lowest_latency = 0,
2225         low_latency = 1,
2226         bulk_latency = 2,
2227         latency_invalid = 255
2228 };
2229
2230
2231 static void igb_lower_rx_eitr(struct igb_adapter *adapter,
2232                               struct igb_ring *rx_ring)
2233 {
2234         struct e1000_hw *hw = &adapter->hw;
2235         int new_val;
2236
2237         new_val = rx_ring->itr_val / 2;
2238         if (new_val < IGB_MIN_DYN_ITR)
2239                 new_val = IGB_MIN_DYN_ITR;
2240
2241         if (new_val != rx_ring->itr_val) {
2242                 rx_ring->itr_val = new_val;
2243                 wr32(rx_ring->itr_register,
2244                                 1000000000 / (new_val * 256));
2245         }
2246 }
2247
2248 static void igb_raise_rx_eitr(struct igb_adapter *adapter,
2249                               struct igb_ring *rx_ring)
2250 {
2251         struct e1000_hw *hw = &adapter->hw;
2252         int new_val;
2253
2254         new_val = rx_ring->itr_val * 2;
2255         if (new_val > IGB_MAX_DYN_ITR)
2256                 new_val = IGB_MAX_DYN_ITR;
2257
2258         if (new_val != rx_ring->itr_val) {
2259                 rx_ring->itr_val = new_val;
2260                 wr32(rx_ring->itr_register,
2261                                 1000000000 / (new_val * 256));
2262         }
2263 }
2264
2265 /**
2266  * igb_update_itr - update the dynamic ITR value based on statistics
2267  *      Stores a new ITR value based on packets and byte
2268  *      counts during the last interrupt.  The advantage of per interrupt
2269  *      computation is faster updates and more accurate ITR for the current
2270  *      traffic pattern.  Constants in this function were computed
2271  *      based on theoretical maximum wire speed and thresholds were set based
2272  *      on testing data as well as attempting to minimize response time
2273  *      while increasing bulk throughput.
2274  *      this functionality is controlled by the InterruptThrottleRate module
2275  *      parameter (see igb_param.c)
2276  *      NOTE:  These calculations are only valid when operating in a single-
2277  *             queue environment.
2278  * @adapter: pointer to adapter
2279  * @itr_setting: current adapter->itr
2280  * @packets: the number of packets during this measurement interval
2281  * @bytes: the number of bytes during this measurement interval
2282  **/
2283 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2284                                    int packets, int bytes)
2285 {
2286         unsigned int retval = itr_setting;
2287
2288         if (packets == 0)
2289                 goto update_itr_done;
2290
2291         switch (itr_setting) {
2292         case lowest_latency:
2293                 /* handle TSO and jumbo frames */
2294                 if (bytes/packets > 8000)
2295                         retval = bulk_latency;
2296                 else if ((packets < 5) && (bytes > 512))
2297                         retval = low_latency;
2298                 break;
2299         case low_latency:  /* 50 usec aka 20000 ints/s */
2300                 if (bytes > 10000) {
2301                         /* this if handles the TSO accounting */
2302                         if (bytes/packets > 8000) {
2303                                 retval = bulk_latency;
2304                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2305                                 retval = bulk_latency;
2306                         } else if ((packets > 35)) {
2307                                 retval = lowest_latency;
2308                         }
2309                 } else if (bytes/packets > 2000) {
2310                         retval = bulk_latency;
2311                 } else if (packets <= 2 && bytes < 512) {
2312                         retval = lowest_latency;
2313                 }
2314                 break;
2315         case bulk_latency: /* 250 usec aka 4000 ints/s */
2316                 if (bytes > 25000) {
2317                         if (packets > 35)
2318                                 retval = low_latency;
2319                 } else if (bytes < 6000) {
2320                         retval = low_latency;
2321                 }
2322                 break;
2323         }
2324
2325 update_itr_done:
2326         return retval;
2327 }
2328
2329 static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
2330                         int rx_only)
2331 {
2332         u16 current_itr;
2333         u32 new_itr = adapter->itr;
2334
2335         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2336         if (adapter->link_speed != SPEED_1000) {
2337                 current_itr = 0;
2338                 new_itr = 4000;
2339                 goto set_itr_now;
2340         }
2341
2342         adapter->rx_itr = igb_update_itr(adapter,
2343                                     adapter->rx_itr,
2344                                     adapter->rx_ring->total_packets,
2345                                     adapter->rx_ring->total_bytes);
2346         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2347         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2348                 adapter->rx_itr = low_latency;
2349
2350         if (!rx_only) {
2351                 adapter->tx_itr = igb_update_itr(adapter,
2352                                             adapter->tx_itr,
2353                                             adapter->tx_ring->total_packets,
2354                                             adapter->tx_ring->total_bytes);
2355                 /* conservative mode (itr 3) eliminates the
2356                  * lowest_latency setting */
2357                 if (adapter->itr_setting == 3 &&
2358                     adapter->tx_itr == lowest_latency)
2359                         adapter->tx_itr = low_latency;
2360
2361                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2362         } else {
2363                 current_itr = adapter->rx_itr;
2364         }
2365
2366         switch (current_itr) {
2367         /* counts and packets in update_itr are dependent on these numbers */
2368         case lowest_latency:
2369                 new_itr = 70000;
2370                 break;
2371         case low_latency:
2372                 new_itr = 20000; /* aka hwitr = ~200 */
2373                 break;
2374         case bulk_latency:
2375                 new_itr = 4000;
2376                 break;
2377         default:
2378                 break;
2379         }
2380
2381 set_itr_now:
2382         if (new_itr != adapter->itr) {
2383                 /* this attempts to bias the interrupt rate towards Bulk
2384                  * by adding intermediate steps when interrupt rate is
2385                  * increasing */
2386                 new_itr = new_itr > adapter->itr ?
2387                              min(adapter->itr + (new_itr >> 2), new_itr) :
2388                              new_itr;
2389                 /* Don't write the value here; it resets the adapter's
2390                  * internal timer, and causes us to delay far longer than
2391                  * we should between interrupts.  Instead, we write the ITR
2392                  * value at the beginning of the next interrupt so the timing
2393                  * ends up being correct.
2394                  */
2395                 adapter->itr = new_itr;
2396                 adapter->set_itr = 1;
2397         }
2398
2399         return;
2400 }
2401
2402
2403 #define IGB_TX_FLAGS_CSUM               0x00000001
2404 #define IGB_TX_FLAGS_VLAN               0x00000002
2405 #define IGB_TX_FLAGS_TSO                0x00000004
2406 #define IGB_TX_FLAGS_IPV4               0x00000008
2407 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2408 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2409
2410 static inline int igb_tso_adv(struct igb_adapter *adapter,
2411                               struct igb_ring *tx_ring,
2412                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2413 {
2414         struct e1000_adv_tx_context_desc *context_desc;
2415         unsigned int i;
2416         int err;
2417         struct igb_buffer *buffer_info;
2418         u32 info = 0, tu_cmd = 0;
2419         u32 mss_l4len_idx, l4len;
2420         *hdr_len = 0;
2421
2422         if (skb_header_cloned(skb)) {
2423                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2424                 if (err)
2425                         return err;
2426         }
2427
2428         l4len = tcp_hdrlen(skb);
2429         *hdr_len += l4len;
2430
2431         if (skb->protocol == htons(ETH_P_IP)) {
2432                 struct iphdr *iph = ip_hdr(skb);
2433                 iph->tot_len = 0;
2434                 iph->check = 0;
2435                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2436                                                          iph->daddr, 0,
2437                                                          IPPROTO_TCP,
2438                                                          0);
2439         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2440                 ipv6_hdr(skb)->payload_len = 0;
2441                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2442                                                        &ipv6_hdr(skb)->daddr,
2443                                                        0, IPPROTO_TCP, 0);
2444         }
2445
2446         i = tx_ring->next_to_use;
2447
2448         buffer_info = &tx_ring->buffer_info[i];
2449         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2450         /* VLAN MACLEN IPLEN */
2451         if (tx_flags & IGB_TX_FLAGS_VLAN)
2452                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2453         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2454         *hdr_len += skb_network_offset(skb);
2455         info |= skb_network_header_len(skb);
2456         *hdr_len += skb_network_header_len(skb);
2457         context_desc->vlan_macip_lens = cpu_to_le32(info);
2458
2459         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2460         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2461
2462         if (skb->protocol == htons(ETH_P_IP))
2463                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2464         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2465
2466         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2467
2468         /* MSS L4LEN IDX */
2469         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2470         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2471
2472         /* Context index must be unique per ring.  Luckily, so is the interrupt
2473          * mask value. */
2474         mss_l4len_idx |= tx_ring->eims_value >> 4;
2475
2476         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2477         context_desc->seqnum_seed = 0;
2478
2479         buffer_info->time_stamp = jiffies;
2480         buffer_info->dma = 0;
2481         i++;
2482         if (i == tx_ring->count)
2483                 i = 0;
2484
2485         tx_ring->next_to_use = i;
2486
2487         return true;
2488 }
2489
2490 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2491                                         struct igb_ring *tx_ring,
2492                                         struct sk_buff *skb, u32 tx_flags)
2493 {
2494         struct e1000_adv_tx_context_desc *context_desc;
2495         unsigned int i;
2496         struct igb_buffer *buffer_info;
2497         u32 info = 0, tu_cmd = 0;
2498
2499         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2500             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2501                 i = tx_ring->next_to_use;
2502                 buffer_info = &tx_ring->buffer_info[i];
2503                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2504
2505                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2506                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2507                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2508                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2509                         info |= skb_network_header_len(skb);
2510
2511                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2512
2513                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2514
2515                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2516                         if (skb->protocol == htons(ETH_P_IP))
2517                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2518                         if (skb->sk && (skb->sk->sk_protocol == IPPROTO_TCP))
2519                                 tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2520                 }
2521
2522                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2523                 context_desc->seqnum_seed = 0;
2524                 context_desc->mss_l4len_idx =
2525                                           cpu_to_le32(tx_ring->eims_value >> 4);
2526
2527                 buffer_info->time_stamp = jiffies;
2528                 buffer_info->dma = 0;
2529
2530                 i++;
2531                 if (i == tx_ring->count)
2532                         i = 0;
2533                 tx_ring->next_to_use = i;
2534
2535                 return true;
2536         }
2537
2538
2539         return false;
2540 }
2541
2542 #define IGB_MAX_TXD_PWR 16
2543 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2544
2545 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2546                                  struct igb_ring *tx_ring,
2547                                  struct sk_buff *skb)
2548 {
2549         struct igb_buffer *buffer_info;
2550         unsigned int len = skb_headlen(skb);
2551         unsigned int count = 0, i;
2552         unsigned int f;
2553
2554         i = tx_ring->next_to_use;
2555
2556         buffer_info = &tx_ring->buffer_info[i];
2557         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2558         buffer_info->length = len;
2559         /* set time_stamp *before* dma to help avoid a possible race */
2560         buffer_info->time_stamp = jiffies;
2561         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2562                                           PCI_DMA_TODEVICE);
2563         count++;
2564         i++;
2565         if (i == tx_ring->count)
2566                 i = 0;
2567
2568         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2569                 struct skb_frag_struct *frag;
2570
2571                 frag = &skb_shinfo(skb)->frags[f];
2572                 len = frag->size;
2573
2574                 buffer_info = &tx_ring->buffer_info[i];
2575                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2576                 buffer_info->length = len;
2577                 buffer_info->time_stamp = jiffies;
2578                 buffer_info->dma = pci_map_page(adapter->pdev,
2579                                                 frag->page,
2580                                                 frag->page_offset,
2581                                                 len,
2582                                                 PCI_DMA_TODEVICE);
2583
2584                 count++;
2585                 i++;
2586                 if (i == tx_ring->count)
2587                         i = 0;
2588         }
2589
2590         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2591         tx_ring->buffer_info[i].skb = skb;
2592
2593         return count;
2594 }
2595
2596 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2597                                     struct igb_ring *tx_ring,
2598                                     int tx_flags, int count, u32 paylen,
2599                                     u8 hdr_len)
2600 {
2601         union e1000_adv_tx_desc *tx_desc = NULL;
2602         struct igb_buffer *buffer_info;
2603         u32 olinfo_status = 0, cmd_type_len;
2604         unsigned int i;
2605
2606         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2607                         E1000_ADVTXD_DCMD_DEXT);
2608
2609         if (tx_flags & IGB_TX_FLAGS_VLAN)
2610                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2611
2612         if (tx_flags & IGB_TX_FLAGS_TSO) {
2613                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2614
2615                 /* insert tcp checksum */
2616                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2617
2618                 /* insert ip checksum */
2619                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2620                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2621
2622         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2623                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2624         }
2625
2626         if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2627                         IGB_TX_FLAGS_VLAN))
2628                 olinfo_status |= tx_ring->eims_value >> 4;
2629
2630         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2631
2632         i = tx_ring->next_to_use;
2633         while (count--) {
2634                 buffer_info = &tx_ring->buffer_info[i];
2635                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2636                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2637                 tx_desc->read.cmd_type_len =
2638                         cpu_to_le32(cmd_type_len | buffer_info->length);
2639                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2640                 i++;
2641                 if (i == tx_ring->count)
2642                         i = 0;
2643         }
2644
2645         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2646         /* Force memory writes to complete before letting h/w
2647          * know there are new descriptors to fetch.  (Only
2648          * applicable for weak-ordered memory model archs,
2649          * such as IA-64). */
2650         wmb();
2651
2652         tx_ring->next_to_use = i;
2653         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2654         /* we need this if more than one processor can write to our tail
2655          * at a time, it syncronizes IO on IA64/Altix systems */
2656         mmiowb();
2657 }
2658
2659 static int __igb_maybe_stop_tx(struct net_device *netdev,
2660                                struct igb_ring *tx_ring, int size)
2661 {
2662         struct igb_adapter *adapter = netdev_priv(netdev);
2663
2664         netif_stop_queue(netdev);
2665         /* Herbert's original patch had:
2666          *  smp_mb__after_netif_stop_queue();
2667          * but since that doesn't exist yet, just open code it. */
2668         smp_mb();
2669
2670         /* We need to check again in a case another CPU has just
2671          * made room available. */
2672         if (IGB_DESC_UNUSED(tx_ring) < size)
2673                 return -EBUSY;
2674
2675         /* A reprieve! */
2676         netif_start_queue(netdev);
2677         ++adapter->restart_queue;
2678         return 0;
2679 }
2680
2681 static int igb_maybe_stop_tx(struct net_device *netdev,
2682                              struct igb_ring *tx_ring, int size)
2683 {
2684         if (IGB_DESC_UNUSED(tx_ring) >= size)
2685                 return 0;
2686         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2687 }
2688
2689 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2690
2691 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2692                                    struct net_device *netdev,
2693                                    struct igb_ring *tx_ring)
2694 {
2695         struct igb_adapter *adapter = netdev_priv(netdev);
2696         unsigned int tx_flags = 0;
2697         unsigned int len;
2698         unsigned long irq_flags;
2699         u8 hdr_len = 0;
2700         int tso = 0;
2701
2702         len = skb_headlen(skb);
2703
2704         if (test_bit(__IGB_DOWN, &adapter->state)) {
2705                 dev_kfree_skb_any(skb);
2706                 return NETDEV_TX_OK;
2707         }
2708
2709         if (skb->len <= 0) {
2710                 dev_kfree_skb_any(skb);
2711                 return NETDEV_TX_OK;
2712         }
2713
2714         if (!spin_trylock_irqsave(&tx_ring->tx_lock, irq_flags))
2715                 /* Collision - tell upper layer to requeue */
2716                 return NETDEV_TX_LOCKED;
2717
2718         /* need: 1 descriptor per page,
2719          *       + 2 desc gap to keep tail from touching head,
2720          *       + 1 desc for skb->data,
2721          *       + 1 desc for context descriptor,
2722          * otherwise try next time */
2723         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2724                 /* this is a hard error */
2725                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2726                 return NETDEV_TX_BUSY;
2727         }
2728
2729         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2730                 tx_flags |= IGB_TX_FLAGS_VLAN;
2731                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2732         }
2733
2734         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2735                                               &hdr_len) : 0;
2736
2737         if (tso < 0) {
2738                 dev_kfree_skb_any(skb);
2739                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2740                 return NETDEV_TX_OK;
2741         }
2742
2743         if (tso)
2744                 tx_flags |= IGB_TX_FLAGS_TSO;
2745         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
2746                         if (skb->ip_summed == CHECKSUM_PARTIAL)
2747                                 tx_flags |= IGB_TX_FLAGS_CSUM;
2748
2749         if (skb->protocol == htons(ETH_P_IP))
2750                 tx_flags |= IGB_TX_FLAGS_IPV4;
2751
2752         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
2753                          igb_tx_map_adv(adapter, tx_ring, skb),
2754                          skb->len, hdr_len);
2755
2756         netdev->trans_start = jiffies;
2757
2758         /* Make sure there is space in the ring for the next send. */
2759         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
2760
2761         spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2762         return NETDEV_TX_OK;
2763 }
2764
2765 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
2766 {
2767         struct igb_adapter *adapter = netdev_priv(netdev);
2768         struct igb_ring *tx_ring = &adapter->tx_ring[0];
2769
2770         /* This goes back to the question of how to logically map a tx queue
2771          * to a flow.  Right now, performance is impacted slightly negatively
2772          * if using multiple tx queues.  If the stack breaks away from a
2773          * single qdisc implementation, we can look at this again. */
2774         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
2775 }
2776
2777 /**
2778  * igb_tx_timeout - Respond to a Tx Hang
2779  * @netdev: network interface device structure
2780  **/
2781 static void igb_tx_timeout(struct net_device *netdev)
2782 {
2783         struct igb_adapter *adapter = netdev_priv(netdev);
2784         struct e1000_hw *hw = &adapter->hw;
2785
2786         /* Do the reset outside of interrupt context */
2787         adapter->tx_timeout_count++;
2788         schedule_work(&adapter->reset_task);
2789         wr32(E1000_EICS, adapter->eims_enable_mask &
2790                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
2791 }
2792
2793 static void igb_reset_task(struct work_struct *work)
2794 {
2795         struct igb_adapter *adapter;
2796         adapter = container_of(work, struct igb_adapter, reset_task);
2797
2798         igb_reinit_locked(adapter);
2799 }
2800
2801 /**
2802  * igb_get_stats - Get System Network Statistics
2803  * @netdev: network interface device structure
2804  *
2805  * Returns the address of the device statistics structure.
2806  * The statistics are actually updated from the timer callback.
2807  **/
2808 static struct net_device_stats *
2809 igb_get_stats(struct net_device *netdev)
2810 {
2811         struct igb_adapter *adapter = netdev_priv(netdev);
2812
2813         /* only return the current stats */
2814         return &adapter->net_stats;
2815 }
2816
2817 /**
2818  * igb_change_mtu - Change the Maximum Transfer Unit
2819  * @netdev: network interface device structure
2820  * @new_mtu: new value for maximum frame size
2821  *
2822  * Returns 0 on success, negative on failure
2823  **/
2824 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
2825 {
2826         struct igb_adapter *adapter = netdev_priv(netdev);
2827         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2828
2829         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
2830             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2831                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2832                 return -EINVAL;
2833         }
2834
2835 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2836         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2837                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2838                 return -EINVAL;
2839         }
2840
2841         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2842                 msleep(1);
2843         /* igb_down has a dependency on max_frame_size */
2844         adapter->max_frame_size = max_frame;
2845         if (netif_running(netdev))
2846                 igb_down(adapter);
2847
2848         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2849          * means we reserve 2 more, this pushes us to allocate from the next
2850          * larger slab size.
2851          * i.e. RXBUFFER_2048 --> size-4096 slab
2852          */
2853
2854         if (max_frame <= IGB_RXBUFFER_256)
2855                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
2856         else if (max_frame <= IGB_RXBUFFER_512)
2857                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
2858         else if (max_frame <= IGB_RXBUFFER_1024)
2859                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
2860         else if (max_frame <= IGB_RXBUFFER_2048)
2861                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
2862         else
2863                 adapter->rx_buffer_len = IGB_RXBUFFER_4096;
2864         /* adjust allocation if LPE protects us, and we aren't using SBP */
2865         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2866              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
2867                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2868
2869         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2870                  netdev->mtu, new_mtu);
2871         netdev->mtu = new_mtu;
2872
2873         if (netif_running(netdev))
2874                 igb_up(adapter);
2875         else
2876                 igb_reset(adapter);
2877
2878         clear_bit(__IGB_RESETTING, &adapter->state);
2879
2880         return 0;
2881 }
2882
2883 /**
2884  * igb_update_stats - Update the board statistics counters
2885  * @adapter: board private structure
2886  **/
2887
2888 void igb_update_stats(struct igb_adapter *adapter)
2889 {
2890         struct e1000_hw *hw = &adapter->hw;
2891         struct pci_dev *pdev = adapter->pdev;
2892         u16 phy_tmp;
2893
2894 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2895
2896         /*
2897          * Prevent stats update while adapter is being reset, or if the pci
2898          * connection is down.
2899          */
2900         if (adapter->link_speed == 0)
2901                 return;
2902         if (pci_channel_offline(pdev))
2903                 return;
2904
2905         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
2906         adapter->stats.gprc += rd32(E1000_GPRC);
2907         adapter->stats.gorc += rd32(E1000_GORCL);
2908         rd32(E1000_GORCH); /* clear GORCL */
2909         adapter->stats.bprc += rd32(E1000_BPRC);
2910         adapter->stats.mprc += rd32(E1000_MPRC);
2911         adapter->stats.roc += rd32(E1000_ROC);
2912
2913         adapter->stats.prc64 += rd32(E1000_PRC64);
2914         adapter->stats.prc127 += rd32(E1000_PRC127);
2915         adapter->stats.prc255 += rd32(E1000_PRC255);
2916         adapter->stats.prc511 += rd32(E1000_PRC511);
2917         adapter->stats.prc1023 += rd32(E1000_PRC1023);
2918         adapter->stats.prc1522 += rd32(E1000_PRC1522);
2919         adapter->stats.symerrs += rd32(E1000_SYMERRS);
2920         adapter->stats.sec += rd32(E1000_SEC);
2921
2922         adapter->stats.mpc += rd32(E1000_MPC);
2923         adapter->stats.scc += rd32(E1000_SCC);
2924         adapter->stats.ecol += rd32(E1000_ECOL);
2925         adapter->stats.mcc += rd32(E1000_MCC);
2926         adapter->stats.latecol += rd32(E1000_LATECOL);
2927         adapter->stats.dc += rd32(E1000_DC);
2928         adapter->stats.rlec += rd32(E1000_RLEC);
2929         adapter->stats.xonrxc += rd32(E1000_XONRXC);
2930         adapter->stats.xontxc += rd32(E1000_XONTXC);
2931         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
2932         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
2933         adapter->stats.fcruc += rd32(E1000_FCRUC);
2934         adapter->stats.gptc += rd32(E1000_GPTC);
2935         adapter->stats.gotc += rd32(E1000_GOTCL);
2936         rd32(E1000_GOTCH); /* clear GOTCL */
2937         adapter->stats.rnbc += rd32(E1000_RNBC);
2938         adapter->stats.ruc += rd32(E1000_RUC);
2939         adapter->stats.rfc += rd32(E1000_RFC);
2940         adapter->stats.rjc += rd32(E1000_RJC);
2941         adapter->stats.tor += rd32(E1000_TORH);
2942         adapter->stats.tot += rd32(E1000_TOTH);
2943         adapter->stats.tpr += rd32(E1000_TPR);
2944
2945         adapter->stats.ptc64 += rd32(E1000_PTC64);
2946         adapter->stats.ptc127 += rd32(E1000_PTC127);
2947         adapter->stats.ptc255 += rd32(E1000_PTC255);
2948         adapter->stats.ptc511 += rd32(E1000_PTC511);
2949         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
2950         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
2951
2952         adapter->stats.mptc += rd32(E1000_MPTC);
2953         adapter->stats.bptc += rd32(E1000_BPTC);
2954
2955         /* used for adaptive IFS */
2956
2957         hw->mac.tx_packet_delta = rd32(E1000_TPT);
2958         adapter->stats.tpt += hw->mac.tx_packet_delta;
2959         hw->mac.collision_delta = rd32(E1000_COLC);
2960         adapter->stats.colc += hw->mac.collision_delta;
2961
2962         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
2963         adapter->stats.rxerrc += rd32(E1000_RXERRC);
2964         adapter->stats.tncrs += rd32(E1000_TNCRS);
2965         adapter->stats.tsctc += rd32(E1000_TSCTC);
2966         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
2967
2968         adapter->stats.iac += rd32(E1000_IAC);
2969         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
2970         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
2971         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
2972         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
2973         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
2974         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
2975         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
2976         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
2977
2978         /* Fill out the OS statistics structure */
2979         adapter->net_stats.multicast = adapter->stats.mprc;
2980         adapter->net_stats.collisions = adapter->stats.colc;
2981
2982         /* Rx Errors */
2983
2984         /* RLEC on some newer hardware can be incorrect so build
2985         * our own version based on RUC and ROC */
2986         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2987                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2988                 adapter->stats.ruc + adapter->stats.roc +
2989                 adapter->stats.cexterr;
2990         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2991                                               adapter->stats.roc;
2992         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2993         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2994         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2995
2996         /* Tx Errors */
2997         adapter->net_stats.tx_errors = adapter->stats.ecol +
2998                                        adapter->stats.latecol;
2999         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3000         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3001         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3002
3003         /* Tx Dropped needs to be maintained elsewhere */
3004
3005         /* Phy Stats */
3006         if (hw->phy.media_type == e1000_media_type_copper) {
3007                 if ((adapter->link_speed == SPEED_1000) &&
3008                    (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
3009                                               &phy_tmp))) {
3010                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3011                         adapter->phy_stats.idle_errors += phy_tmp;
3012                 }
3013         }
3014
3015         /* Management Stats */
3016         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3017         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3018         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3019 }
3020
3021
3022 static irqreturn_t igb_msix_other(int irq, void *data)
3023 {
3024         struct net_device *netdev = data;
3025         struct igb_adapter *adapter = netdev_priv(netdev);
3026         struct e1000_hw *hw = &adapter->hw;
3027         u32 eicr;
3028         /* disable interrupts from the "other" bit, avoid re-entry */
3029         wr32(E1000_EIMC, E1000_EIMS_OTHER);
3030
3031         eicr = rd32(E1000_EICR);
3032
3033         if (eicr & E1000_EIMS_OTHER) {
3034                 u32 icr = rd32(E1000_ICR);
3035                 /* reading ICR causes bit 31 of EICR to be cleared */
3036                 if (!(icr & E1000_ICR_LSC))
3037                         goto no_link_interrupt;
3038                 hw->mac.get_link_status = 1;
3039                 /* guard against interrupt when we're going down */
3040                 if (!test_bit(__IGB_DOWN, &adapter->state))
3041                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3042         }
3043
3044 no_link_interrupt:
3045         wr32(E1000_IMS, E1000_IMS_LSC);
3046         wr32(E1000_EIMS, E1000_EIMS_OTHER);
3047
3048         return IRQ_HANDLED;
3049 }
3050
3051 static irqreturn_t igb_msix_tx(int irq, void *data)
3052 {
3053         struct igb_ring *tx_ring = data;
3054         struct igb_adapter *adapter = tx_ring->adapter;
3055         struct e1000_hw *hw = &adapter->hw;
3056
3057         if (!tx_ring->itr_val)
3058                 wr32(E1000_EIMC, tx_ring->eims_value);
3059
3060         tx_ring->total_bytes = 0;
3061         tx_ring->total_packets = 0;
3062         if (!igb_clean_tx_irq(adapter, tx_ring))
3063                 /* Ring was not completely cleaned, so fire another interrupt */
3064                 wr32(E1000_EICS, tx_ring->eims_value);
3065
3066         if (!tx_ring->itr_val)
3067                 wr32(E1000_EIMS, tx_ring->eims_value);
3068         return IRQ_HANDLED;
3069 }
3070
3071 static irqreturn_t igb_msix_rx(int irq, void *data)
3072 {
3073         struct igb_ring *rx_ring = data;
3074         struct igb_adapter *adapter = rx_ring->adapter;
3075         struct e1000_hw *hw = &adapter->hw;
3076
3077         if (!rx_ring->itr_val)
3078                 wr32(E1000_EIMC, rx_ring->eims_value);
3079
3080         if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi)) {
3081                 rx_ring->total_bytes = 0;
3082                 rx_ring->total_packets = 0;
3083                 rx_ring->no_itr_adjust = 0;
3084                 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3085         } else {
3086                 if (!rx_ring->no_itr_adjust) {
3087                         igb_lower_rx_eitr(adapter, rx_ring);
3088                         rx_ring->no_itr_adjust = 1;
3089                 }
3090         }
3091
3092         return IRQ_HANDLED;
3093 }
3094
3095
3096 /**
3097  * igb_intr_msi - Interrupt Handler
3098  * @irq: interrupt number
3099  * @data: pointer to a network interface device structure
3100  **/
3101 static irqreturn_t igb_intr_msi(int irq, void *data)
3102 {
3103         struct net_device *netdev = data;
3104         struct igb_adapter *adapter = netdev_priv(netdev);
3105         struct napi_struct *napi = &adapter->napi;
3106         struct e1000_hw *hw = &adapter->hw;
3107         /* read ICR disables interrupts using IAM */
3108         u32 icr = rd32(E1000_ICR);
3109
3110         /* Write the ITR value calculated at the end of the
3111          * previous interrupt.
3112          */
3113         if (adapter->set_itr) {
3114                 wr32(E1000_ITR,
3115                         1000000000 / (adapter->itr * 256));
3116                 adapter->set_itr = 0;
3117         }
3118
3119         /* read ICR disables interrupts using IAM */
3120         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3121                 hw->mac.get_link_status = 1;
3122                 if (!test_bit(__IGB_DOWN, &adapter->state))
3123                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3124         }
3125
3126         if (netif_rx_schedule_prep(netdev, napi)) {
3127                 adapter->tx_ring->total_bytes = 0;
3128                 adapter->tx_ring->total_packets = 0;
3129                 adapter->rx_ring->total_bytes = 0;
3130                 adapter->rx_ring->total_packets = 0;
3131                 __netif_rx_schedule(netdev, napi);
3132         }
3133
3134         return IRQ_HANDLED;
3135 }
3136
3137 /**
3138  * igb_intr - Interrupt Handler
3139  * @irq: interrupt number
3140  * @data: pointer to a network interface device structure
3141  **/
3142 static irqreturn_t igb_intr(int irq, void *data)
3143 {
3144         struct net_device *netdev = data;
3145         struct igb_adapter *adapter = netdev_priv(netdev);
3146         struct napi_struct *napi = &adapter->napi;
3147         struct e1000_hw *hw = &adapter->hw;
3148         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3149          * need for the IMC write */
3150         u32 icr = rd32(E1000_ICR);
3151         u32 eicr = 0;
3152         if (!icr)
3153                 return IRQ_NONE;  /* Not our interrupt */
3154
3155         /* Write the ITR value calculated at the end of the
3156          * previous interrupt.
3157          */
3158         if (adapter->set_itr) {
3159                 wr32(E1000_ITR,
3160                         1000000000 / (adapter->itr * 256));
3161                 adapter->set_itr = 0;
3162         }
3163
3164         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3165          * not set, then the adapter didn't send an interrupt */
3166         if (!(icr & E1000_ICR_INT_ASSERTED))
3167                 return IRQ_NONE;
3168
3169         eicr = rd32(E1000_EICR);
3170
3171         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3172                 hw->mac.get_link_status = 1;
3173                 /* guard against interrupt when we're going down */
3174                 if (!test_bit(__IGB_DOWN, &adapter->state))
3175                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3176         }
3177
3178         if (netif_rx_schedule_prep(netdev, napi)) {
3179                 adapter->tx_ring->total_bytes = 0;
3180                 adapter->rx_ring->total_bytes = 0;
3181                 adapter->tx_ring->total_packets = 0;
3182                 adapter->rx_ring->total_packets = 0;
3183                 __netif_rx_schedule(netdev, napi);
3184         }
3185
3186         return IRQ_HANDLED;
3187 }
3188
3189 /**
3190  * igb_clean - NAPI Rx polling callback
3191  * @adapter: board private structure
3192  **/
3193 static int igb_clean(struct napi_struct *napi, int budget)
3194 {
3195         struct igb_adapter *adapter = container_of(napi, struct igb_adapter,
3196                                                    napi);
3197         struct net_device *netdev = adapter->netdev;
3198         int tx_clean_complete = 1, work_done = 0;
3199         int i;
3200
3201         /* Must NOT use netdev_priv macro here. */
3202         adapter = netdev->priv;
3203
3204         /* Keep link state information with original netdev */
3205         if (!netif_carrier_ok(netdev))
3206                 goto quit_polling;
3207
3208         /* igb_clean is called per-cpu.  This lock protects tx_ring[i] from
3209          * being cleaned by multiple cpus simultaneously.  A failure obtaining
3210          * the lock means tx_ring[i] is currently being cleaned anyway. */
3211         for (i = 0; i < adapter->num_tx_queues; i++) {
3212                 if (spin_trylock(&adapter->tx_ring[i].tx_clean_lock)) {
3213                         tx_clean_complete &= igb_clean_tx_irq(adapter,
3214                                                         &adapter->tx_ring[i]);
3215                         spin_unlock(&adapter->tx_ring[i].tx_clean_lock);
3216                 }
3217         }
3218
3219         for (i = 0; i < adapter->num_rx_queues; i++)
3220                 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i], &work_done,
3221                                      adapter->rx_ring[i].napi.weight);
3222
3223         /* If no Tx and not enough Rx work done, exit the polling mode */
3224         if ((tx_clean_complete && (work_done < budget)) ||
3225             !netif_running(netdev)) {
3226 quit_polling:
3227                 if (adapter->itr_setting & 3)
3228                         igb_set_itr(adapter, E1000_ITR, false);
3229                 netif_rx_complete(netdev, napi);
3230                 if (!test_bit(__IGB_DOWN, &adapter->state))
3231                         igb_irq_enable(adapter);
3232                 return 0;
3233         }
3234
3235         return 1;
3236 }
3237
3238 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3239 {
3240         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3241         struct igb_adapter *adapter = rx_ring->adapter;
3242         struct e1000_hw *hw = &adapter->hw;
3243         struct net_device *netdev = adapter->netdev;
3244         int work_done = 0;
3245
3246         /* Keep link state information with original netdev */
3247         if (!netif_carrier_ok(netdev))
3248                 goto quit_polling;
3249
3250         igb_clean_rx_irq_adv(adapter, rx_ring, &work_done, budget);
3251
3252
3253         /* If not enough Rx work done, exit the polling mode */
3254         if ((work_done == 0) || !netif_running(netdev)) {
3255 quit_polling:
3256                 netif_rx_complete(netdev, napi);
3257
3258                 wr32(E1000_EIMS, rx_ring->eims_value);
3259                 if ((adapter->itr_setting & 3) && !rx_ring->no_itr_adjust &&
3260                     (rx_ring->total_packets > IGB_DYN_ITR_PACKET_THRESHOLD)) {
3261                         int mean_size = rx_ring->total_bytes /
3262                                         rx_ring->total_packets;
3263                         if (mean_size < IGB_DYN_ITR_LENGTH_LOW)
3264                                 igb_raise_rx_eitr(adapter, rx_ring);
3265                         else if (mean_size > IGB_DYN_ITR_LENGTH_HIGH)
3266                                 igb_lower_rx_eitr(adapter, rx_ring);
3267                 }
3268                 return 0;
3269         }
3270
3271         return 1;
3272 }
3273 /**
3274  * igb_clean_tx_irq - Reclaim resources after transmit completes
3275  * @adapter: board private structure
3276  * returns true if ring is completely cleaned
3277  **/
3278 static bool igb_clean_tx_irq(struct igb_adapter *adapter,
3279                                   struct igb_ring *tx_ring)
3280 {
3281         struct net_device *netdev = adapter->netdev;
3282         struct e1000_hw *hw = &adapter->hw;
3283         struct e1000_tx_desc *tx_desc;
3284         struct igb_buffer *buffer_info;
3285         struct sk_buff *skb;
3286         unsigned int i;
3287         u32 head, oldhead;
3288         unsigned int count = 0;
3289         bool cleaned = false;
3290         bool retval = true;
3291         unsigned int total_bytes = 0, total_packets = 0;
3292
3293         rmb();
3294         head = *(volatile u32 *)((struct e1000_tx_desc *)tx_ring->desc
3295                                  + tx_ring->count);
3296         head = le32_to_cpu(head);
3297         i = tx_ring->next_to_clean;
3298         while (1) {
3299                 while (i != head) {
3300                         cleaned = true;
3301                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3302                         buffer_info = &tx_ring->buffer_info[i];
3303                         skb = buffer_info->skb;
3304
3305                         if (skb) {
3306                                 unsigned int segs, bytecount;
3307                                 /* gso_segs is currently only valid for tcp */
3308                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3309                                 /* multiply data chunks by size of headers */
3310                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3311                                             skb->len;
3312                                 total_packets += segs;
3313                                 total_bytes += bytecount;
3314                         }
3315
3316                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3317                         tx_desc->upper.data = 0;
3318
3319                         i++;
3320                         if (i == tx_ring->count)
3321                                 i = 0;
3322
3323                         count++;
3324                         if (count == IGB_MAX_TX_CLEAN) {
3325                                 retval = false;
3326                                 goto done_cleaning;
3327                         }
3328                 }
3329                 oldhead = head;
3330                 rmb();
3331                 head = *(volatile u32 *)((struct e1000_tx_desc *)tx_ring->desc
3332                                          + tx_ring->count);
3333                 head = le32_to_cpu(head);
3334                 if (head == oldhead)
3335                         goto done_cleaning;
3336         }  /* while (1) */
3337
3338 done_cleaning:
3339         tx_ring->next_to_clean = i;
3340
3341         if (unlikely(cleaned &&
3342                      netif_carrier_ok(netdev) &&
3343                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3344                 /* Make sure that anybody stopping the queue after this
3345                  * sees the new next_to_clean.
3346                  */
3347                 smp_mb();
3348                 if (netif_queue_stopped(netdev) &&
3349                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3350                         netif_wake_queue(netdev);
3351                         ++adapter->restart_queue;
3352                 }
3353         }
3354
3355         if (tx_ring->detect_tx_hung) {
3356                 /* Detect a transmit hang in hardware, this serializes the
3357                  * check with the clearing of time_stamp and movement of i */
3358                 tx_ring->detect_tx_hung = false;
3359                 if (tx_ring->buffer_info[i].time_stamp &&
3360                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3361                                (adapter->tx_timeout_factor * HZ))
3362                     && !(rd32(E1000_STATUS) &
3363                          E1000_STATUS_TXOFF)) {
3364
3365                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3366                         /* detected Tx unit hang */
3367                         dev_err(&adapter->pdev->dev,
3368                                 "Detected Tx Unit Hang\n"
3369                                 "  Tx Queue             <%lu>\n"
3370                                 "  TDH                  <%x>\n"
3371                                 "  TDT                  <%x>\n"
3372                                 "  next_to_use          <%x>\n"
3373                                 "  next_to_clean        <%x>\n"
3374                                 "  head (WB)            <%x>\n"
3375                                 "buffer_info[next_to_clean]\n"
3376                                 "  time_stamp           <%lx>\n"
3377                                 "  jiffies              <%lx>\n"
3378                                 "  desc.status          <%x>\n",
3379                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3380                                         sizeof(struct igb_ring)),
3381                                 readl(adapter->hw.hw_addr + tx_ring->head),
3382                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3383                                 tx_ring->next_to_use,
3384                                 tx_ring->next_to_clean,
3385                                 head,
3386                                 tx_ring->buffer_info[i].time_stamp,
3387                                 jiffies,
3388                                 tx_desc->upper.fields.status);
3389                         netif_stop_queue(netdev);
3390                 }
3391         }
3392         tx_ring->total_bytes += total_bytes;
3393         tx_ring->total_packets += total_packets;
3394         adapter->net_stats.tx_bytes += total_bytes;
3395         adapter->net_stats.tx_packets += total_packets;
3396         return retval;
3397 }
3398
3399
3400 /**
3401  * igb_receive_skb - helper function to handle rx indications
3402  * @adapter: board private structure
3403  * @status: descriptor status field as written by hardware
3404  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3405  * @skb: pointer to sk_buff to be indicated to stack
3406  **/
3407 static void igb_receive_skb(struct igb_adapter *adapter, u8 status, u16 vlan,
3408                             struct sk_buff *skb)
3409 {
3410         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
3411                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3412                                          le16_to_cpu(vlan) &
3413                                          E1000_RXD_SPC_VLAN_MASK);
3414         else
3415                 netif_receive_skb(skb);
3416 }
3417
3418
3419 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3420                                        u32 status_err, struct sk_buff *skb)
3421 {
3422         skb->ip_summed = CHECKSUM_NONE;
3423
3424         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3425         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3426                 return;
3427         /* TCP/UDP checksum error bit is set */
3428         if (status_err &
3429             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3430                 /* let the stack verify checksum errors */
3431                 adapter->hw_csum_err++;
3432                 return;
3433         }
3434         /* It must be a TCP or UDP packet with a valid checksum */
3435         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3436                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3437
3438         adapter->hw_csum_good++;
3439 }
3440
3441 static bool igb_clean_rx_irq_adv(struct igb_adapter *adapter,
3442                                       struct igb_ring *rx_ring,
3443                                       int *work_done, int budget)
3444 {
3445         struct net_device *netdev = adapter->netdev;
3446         struct pci_dev *pdev = adapter->pdev;
3447         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3448         struct igb_buffer *buffer_info , *next_buffer;
3449         struct sk_buff *skb;
3450         unsigned int i, j;
3451         u32 length, hlen, staterr;
3452         bool cleaned = false;
3453         int cleaned_count = 0;
3454         unsigned int total_bytes = 0, total_packets = 0;
3455
3456         i = rx_ring->next_to_clean;
3457         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3458         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3459
3460         while (staterr & E1000_RXD_STAT_DD) {
3461                 if (*work_done >= budget)
3462                         break;
3463                 (*work_done)++;
3464                 buffer_info = &rx_ring->buffer_info[i];
3465
3466                 /* HW will not DMA in data larger than the given buffer, even
3467                  * if it parses the (NFS, of course) header to be larger.  In
3468                  * that case, it fills the header buffer and spills the rest
3469                  * into the page.
3470                  */
3471                 hlen = le16_to_cpu((rx_desc->wb.lower.lo_dword.hdr_info &
3472                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT);
3473                 if (hlen > adapter->rx_ps_hdr_size)
3474                         hlen = adapter->rx_ps_hdr_size;
3475
3476                 length = le16_to_cpu(rx_desc->wb.upper.length);
3477                 cleaned = true;
3478                 cleaned_count++;
3479
3480                 if (rx_ring->pending_skb != NULL) {
3481                         skb = rx_ring->pending_skb;
3482                         rx_ring->pending_skb = NULL;
3483                         j = rx_ring->pending_skb_page;
3484                 } else {
3485                         skb = buffer_info->skb;
3486                         prefetch(skb->data - NET_IP_ALIGN);
3487                         buffer_info->skb = NULL;
3488                         if (hlen) {
3489                                 pci_unmap_single(pdev, buffer_info->dma,
3490                                                  adapter->rx_ps_hdr_size +
3491                                                    NET_IP_ALIGN,
3492                                                  PCI_DMA_FROMDEVICE);
3493                                 skb_put(skb, hlen);
3494                         } else {
3495                                 pci_unmap_single(pdev, buffer_info->dma,
3496                                                  adapter->rx_buffer_len +
3497                                                    NET_IP_ALIGN,
3498                                                  PCI_DMA_FROMDEVICE);
3499                                 skb_put(skb, length);
3500                                 goto send_up;
3501                         }
3502                         j = 0;
3503                 }
3504
3505                 while (length) {
3506                         pci_unmap_page(pdev, buffer_info->page_dma,
3507                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
3508                         buffer_info->page_dma = 0;
3509                         skb_fill_page_desc(skb, j, buffer_info->page,
3510                                                 0, length);
3511                         buffer_info->page = NULL;
3512
3513                         skb->len += length;
3514                         skb->data_len += length;
3515                         skb->truesize += length;
3516                         rx_desc->wb.upper.status_error = 0;
3517                         if (staterr & E1000_RXD_STAT_EOP)
3518                                 break;
3519
3520                         j++;
3521                         cleaned_count++;
3522                         i++;
3523                         if (i == rx_ring->count)
3524                                 i = 0;
3525
3526                         buffer_info = &rx_ring->buffer_info[i];
3527                         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3528                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3529                         length = le16_to_cpu(rx_desc->wb.upper.length);
3530                         if (!(staterr & E1000_RXD_STAT_DD)) {
3531                                 rx_ring->pending_skb = skb;
3532                                 rx_ring->pending_skb_page = j;
3533                                 goto out;
3534                         }
3535                 }
3536 send_up:
3537                 pskb_trim(skb, skb->len - 4);
3538                 i++;
3539                 if (i == rx_ring->count)
3540                         i = 0;
3541                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3542                 prefetch(next_rxd);
3543                 next_buffer = &rx_ring->buffer_info[i];
3544
3545                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3546                         dev_kfree_skb_irq(skb);
3547                         goto next_desc;
3548                 }
3549                 rx_ring->no_itr_adjust |= (staterr & E1000_RXD_STAT_DYNINT);
3550
3551                 total_bytes += skb->len;
3552                 total_packets++;
3553
3554                 igb_rx_checksum_adv(adapter, staterr, skb);
3555
3556                 skb->protocol = eth_type_trans(skb, netdev);
3557
3558                 igb_receive_skb(adapter, staterr, rx_desc->wb.upper.vlan, skb);
3559
3560                 netdev->last_rx = jiffies;
3561
3562 next_desc:
3563                 rx_desc->wb.upper.status_error = 0;
3564
3565                 /* return some buffers to hardware, one at a time is too slow */
3566                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3567                         igb_alloc_rx_buffers_adv(adapter, rx_ring,
3568                                                  cleaned_count);
3569                         cleaned_count = 0;
3570                 }
3571
3572                 /* use prefetched values */
3573                 rx_desc = next_rxd;
3574                 buffer_info = next_buffer;
3575
3576                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3577         }
3578 out:
3579         rx_ring->next_to_clean = i;
3580         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3581
3582         if (cleaned_count)
3583                 igb_alloc_rx_buffers_adv(adapter, rx_ring, cleaned_count);
3584
3585         rx_ring->total_packets += total_packets;
3586         rx_ring->total_bytes += total_bytes;
3587         rx_ring->rx_stats.packets += total_packets;
3588         rx_ring->rx_stats.bytes += total_bytes;
3589         adapter->net_stats.rx_bytes += total_bytes;
3590         adapter->net_stats.rx_packets += total_packets;
3591         return cleaned;
3592 }
3593
3594
3595 /**
3596  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3597  * @adapter: address of board private structure
3598  **/
3599 static void igb_alloc_rx_buffers_adv(struct igb_adapter *adapter,
3600                                      struct igb_ring *rx_ring,
3601                                      int cleaned_count)
3602 {
3603         struct net_device *netdev = adapter->netdev;
3604         struct pci_dev *pdev = adapter->pdev;
3605         union e1000_adv_rx_desc *rx_desc;
3606         struct igb_buffer *buffer_info;
3607         struct sk_buff *skb;
3608         unsigned int i;
3609
3610         i = rx_ring->next_to_use;
3611         buffer_info = &rx_ring->buffer_info[i];
3612
3613         while (cleaned_count--) {
3614                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3615
3616                 if (adapter->rx_ps_hdr_size && !buffer_info->page) {
3617                         buffer_info->page = alloc_page(GFP_ATOMIC);
3618                         if (!buffer_info->page) {
3619                                 adapter->alloc_rx_buff_failed++;
3620                                 goto no_buffers;
3621                         }
3622                         buffer_info->page_dma =
3623                                 pci_map_page(pdev,
3624                                              buffer_info->page,
3625                                              0, PAGE_SIZE,
3626                                              PCI_DMA_FROMDEVICE);
3627                 }
3628
3629                 if (!buffer_info->skb) {
3630                         int bufsz;
3631
3632                         if (adapter->rx_ps_hdr_size)
3633                                 bufsz = adapter->rx_ps_hdr_size;
3634                         else
3635                                 bufsz = adapter->rx_buffer_len;
3636                         bufsz += NET_IP_ALIGN;
3637                         skb = netdev_alloc_skb(netdev, bufsz);
3638
3639                         if (!skb) {
3640                                 adapter->alloc_rx_buff_failed++;
3641                                 goto no_buffers;
3642                         }
3643
3644                         /* Make buffer alignment 2 beyond a 16 byte boundary
3645                          * this will result in a 16 byte aligned IP header after
3646                          * the 14 byte MAC header is removed
3647                          */
3648                         skb_reserve(skb, NET_IP_ALIGN);
3649
3650                         buffer_info->skb = skb;
3651                         buffer_info->dma = pci_map_single(pdev, skb->data,
3652                                                           bufsz,
3653                                                           PCI_DMA_FROMDEVICE);
3654
3655                 }
3656                 /* Refresh the desc even if buffer_addrs didn't change because
3657                  * each write-back erases this info. */
3658                 if (adapter->rx_ps_hdr_size) {
3659                         rx_desc->read.pkt_addr =
3660                              cpu_to_le64(buffer_info->page_dma);
3661                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
3662                 } else {
3663                         rx_desc->read.pkt_addr =
3664                              cpu_to_le64(buffer_info->dma);
3665                         rx_desc->read.hdr_addr = 0;
3666                 }
3667
3668                 i++;
3669                 if (i == rx_ring->count)
3670                         i = 0;
3671                 buffer_info = &rx_ring->buffer_info[i];
3672         }
3673
3674 no_buffers:
3675         if (rx_ring->next_to_use != i) {
3676                 rx_ring->next_to_use = i;
3677                 if (i == 0)
3678                         i = (rx_ring->count - 1);
3679                 else
3680                         i--;
3681
3682                 /* Force memory writes to complete before letting h/w
3683                  * know there are new descriptors to fetch.  (Only
3684                  * applicable for weak-ordered memory model archs,
3685                  * such as IA-64). */
3686                 wmb();
3687                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
3688         }
3689 }
3690
3691 /**
3692  * igb_mii_ioctl -
3693  * @netdev:
3694  * @ifreq:
3695  * @cmd:
3696  **/
3697 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3698 {
3699         struct igb_adapter *adapter = netdev_priv(netdev);
3700         struct mii_ioctl_data *data = if_mii(ifr);
3701
3702         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3703                 return -EOPNOTSUPP;
3704
3705         switch (cmd) {
3706         case SIOCGMIIPHY:
3707                 data->phy_id = adapter->hw.phy.addr;
3708                 break;
3709         case SIOCGMIIREG:
3710                 if (!capable(CAP_NET_ADMIN))
3711                         return -EPERM;
3712                 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
3713                                                      data->reg_num
3714                                                      & 0x1F, &data->val_out))
3715                         return -EIO;
3716                 break;
3717         case SIOCSMIIREG:
3718         default:
3719                 return -EOPNOTSUPP;
3720         }
3721         return 0;
3722 }
3723
3724 /**
3725  * igb_ioctl -
3726  * @netdev:
3727  * @ifreq:
3728  * @cmd:
3729  **/
3730 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3731 {
3732         switch (cmd) {
3733         case SIOCGMIIPHY:
3734         case SIOCGMIIREG:
3735         case SIOCSMIIREG:
3736                 return igb_mii_ioctl(netdev, ifr, cmd);
3737         default:
3738                 return -EOPNOTSUPP;
3739         }
3740 }
3741
3742 static void igb_vlan_rx_register(struct net_device *netdev,
3743                                  struct vlan_group *grp)
3744 {
3745         struct igb_adapter *adapter = netdev_priv(netdev);
3746         struct e1000_hw *hw = &adapter->hw;
3747         u32 ctrl, rctl;
3748
3749         igb_irq_disable(adapter);
3750         adapter->vlgrp = grp;
3751
3752         if (grp) {
3753                 /* enable VLAN tag insert/strip */
3754                 ctrl = rd32(E1000_CTRL);
3755                 ctrl |= E1000_CTRL_VME;
3756                 wr32(E1000_CTRL, ctrl);
3757
3758                 /* enable VLAN receive filtering */
3759                 rctl = rd32(E1000_RCTL);
3760                 rctl |= E1000_RCTL_VFE;
3761                 rctl &= ~E1000_RCTL_CFIEN;
3762                 wr32(E1000_RCTL, rctl);
3763                 igb_update_mng_vlan(adapter);
3764                 wr32(E1000_RLPML,
3765                                 adapter->max_frame_size + VLAN_TAG_SIZE);
3766         } else {
3767                 /* disable VLAN tag insert/strip */
3768                 ctrl = rd32(E1000_CTRL);
3769                 ctrl &= ~E1000_CTRL_VME;
3770                 wr32(E1000_CTRL, ctrl);
3771
3772                 /* disable VLAN filtering */
3773                 rctl = rd32(E1000_RCTL);
3774                 rctl &= ~E1000_RCTL_VFE;
3775                 wr32(E1000_RCTL, rctl);
3776                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
3777                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3778                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
3779                 }
3780                 wr32(E1000_RLPML,
3781                                 adapter->max_frame_size);
3782         }
3783
3784         if (!test_bit(__IGB_DOWN, &adapter->state))
3785                 igb_irq_enable(adapter);
3786 }
3787
3788 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3789 {
3790         struct igb_adapter *adapter = netdev_priv(netdev);
3791         struct e1000_hw *hw = &adapter->hw;
3792         u32 vfta, index;
3793
3794         if ((adapter->hw.mng_cookie.status &
3795              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3796             (vid == adapter->mng_vlan_id))
3797                 return;
3798         /* add VID to filter table */
3799         index = (vid >> 5) & 0x7F;
3800         vfta = array_rd32(E1000_VFTA, index);
3801         vfta |= (1 << (vid & 0x1F));
3802         igb_write_vfta(&adapter->hw, index, vfta);
3803 }
3804
3805 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3806 {
3807         struct igb_adapter *adapter = netdev_priv(netdev);
3808         struct e1000_hw *hw = &adapter->hw;
3809         u32 vfta, index;
3810
3811         igb_irq_disable(adapter);
3812         vlan_group_set_device(adapter->vlgrp, vid, NULL);
3813
3814         if (!test_bit(__IGB_DOWN, &adapter->state))
3815                 igb_irq_enable(adapter);
3816
3817         if ((adapter->hw.mng_cookie.status &
3818              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3819             (vid == adapter->mng_vlan_id)) {
3820                 /* release control to f/w */
3821                 igb_release_hw_control(adapter);
3822                 return;
3823         }
3824
3825         /* remove VID from filter table */
3826         index = (vid >> 5) & 0x7F;
3827         vfta = array_rd32(E1000_VFTA, index);
3828         vfta &= ~(1 << (vid & 0x1F));
3829         igb_write_vfta(&adapter->hw, index, vfta);
3830 }
3831
3832 static void igb_restore_vlan(struct igb_adapter *adapter)
3833 {
3834         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3835
3836         if (adapter->vlgrp) {
3837                 u16 vid;
3838                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3839                         if (!vlan_group_get_device(adapter->vlgrp, vid))
3840                                 continue;
3841                         igb_vlan_rx_add_vid(adapter->netdev, vid);
3842                 }
3843         }
3844 }
3845
3846 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
3847 {
3848         struct e1000_mac_info *mac = &adapter->hw.mac;
3849
3850         mac->autoneg = 0;
3851
3852         /* Fiber NICs only allow 1000 gbps Full duplex */
3853         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
3854                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
3855                 dev_err(&adapter->pdev->dev,
3856                         "Unsupported Speed/Duplex configuration\n");
3857                 return -EINVAL;
3858         }
3859
3860         switch (spddplx) {
3861         case SPEED_10 + DUPLEX_HALF:
3862                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
3863                 break;
3864         case SPEED_10 + DUPLEX_FULL:
3865                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
3866                 break;
3867         case SPEED_100 + DUPLEX_HALF:
3868                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
3869                 break;
3870         case SPEED_100 + DUPLEX_FULL:
3871                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
3872                 break;
3873         case SPEED_1000 + DUPLEX_FULL:
3874                 mac->autoneg = 1;
3875                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
3876                 break;
3877         case SPEED_1000 + DUPLEX_HALF: /* not supported */
3878         default:
3879                 dev_err(&adapter->pdev->dev,
3880                         "Unsupported Speed/Duplex configuration\n");
3881                 return -EINVAL;
3882         }
3883         return 0;
3884 }
3885
3886
3887 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
3888 {
3889         struct net_device *netdev = pci_get_drvdata(pdev);
3890         struct igb_adapter *adapter = netdev_priv(netdev);
3891         struct e1000_hw *hw = &adapter->hw;
3892         u32 ctrl, ctrl_ext, rctl, status;
3893         u32 wufc = adapter->wol;
3894 #ifdef CONFIG_PM
3895         int retval = 0;
3896 #endif
3897
3898         netif_device_detach(netdev);
3899
3900         if (netif_running(netdev)) {
3901                 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3902                 igb_down(adapter);
3903                 igb_free_irq(adapter);
3904         }
3905
3906 #ifdef CONFIG_PM
3907         retval = pci_save_state(pdev);
3908         if (retval)
3909                 return retval;
3910 #endif
3911
3912         status = rd32(E1000_STATUS);
3913         if (status & E1000_STATUS_LU)
3914                 wufc &= ~E1000_WUFC_LNKC;
3915
3916         if (wufc) {
3917                 igb_setup_rctl(adapter);
3918                 igb_set_multi(netdev);
3919
3920                 /* turn on all-multi mode if wake on multicast is enabled */
3921                 if (wufc & E1000_WUFC_MC) {
3922                         rctl = rd32(E1000_RCTL);
3923                         rctl |= E1000_RCTL_MPE;
3924                         wr32(E1000_RCTL, rctl);
3925                 }
3926
3927                 ctrl = rd32(E1000_CTRL);
3928                 /* advertise wake from D3Cold */
3929                 #define E1000_CTRL_ADVD3WUC 0x00100000
3930                 /* phy power management enable */
3931                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3932                 ctrl |= E1000_CTRL_ADVD3WUC;
3933                 wr32(E1000_CTRL, ctrl);
3934
3935                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3936                    adapter->hw.phy.media_type ==
3937                                         e1000_media_type_internal_serdes) {
3938                         /* keep the laser running in D3 */
3939                         ctrl_ext = rd32(E1000_CTRL_EXT);
3940                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3941                         wr32(E1000_CTRL_EXT, ctrl_ext);
3942                 }
3943
3944                 /* Allow time for pending master requests to run */
3945                 igb_disable_pcie_master(&adapter->hw);
3946
3947                 wr32(E1000_WUC, E1000_WUC_PME_EN);
3948                 wr32(E1000_WUFC, wufc);
3949                 pci_enable_wake(pdev, PCI_D3hot, 1);
3950                 pci_enable_wake(pdev, PCI_D3cold, 1);
3951         } else {
3952                 wr32(E1000_WUC, 0);
3953                 wr32(E1000_WUFC, 0);
3954                 pci_enable_wake(pdev, PCI_D3hot, 0);
3955                 pci_enable_wake(pdev, PCI_D3cold, 0);
3956         }
3957
3958         igb_release_manageability(adapter);
3959
3960         /* make sure adapter isn't asleep if manageability is enabled */
3961         if (adapter->en_mng_pt) {
3962                 pci_enable_wake(pdev, PCI_D3hot, 1);
3963                 pci_enable_wake(pdev, PCI_D3cold, 1);
3964         }
3965
3966         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3967          * would have already happened in close and is redundant. */
3968         igb_release_hw_control(adapter);
3969
3970         pci_disable_device(pdev);
3971
3972         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3973
3974         return 0;
3975 }
3976
3977 #ifdef CONFIG_PM
3978 static int igb_resume(struct pci_dev *pdev)
3979 {
3980         struct net_device *netdev = pci_get_drvdata(pdev);
3981         struct igb_adapter *adapter = netdev_priv(netdev);
3982         struct e1000_hw *hw = &adapter->hw;
3983         u32 err;
3984
3985         pci_set_power_state(pdev, PCI_D0);
3986         pci_restore_state(pdev);
3987         err = pci_enable_device(pdev);
3988         if (err) {
3989                 dev_err(&pdev->dev,
3990                         "igb: Cannot enable PCI device from suspend\n");
3991                 return err;
3992         }
3993         pci_set_master(pdev);
3994
3995         pci_enable_wake(pdev, PCI_D3hot, 0);
3996         pci_enable_wake(pdev, PCI_D3cold, 0);
3997
3998         if (netif_running(netdev)) {
3999                 err = igb_request_irq(adapter);
4000                 if (err)
4001                         return err;
4002         }
4003
4004         /* e1000_power_up_phy(adapter); */
4005
4006         igb_reset(adapter);
4007         wr32(E1000_WUS, ~0);
4008
4009         igb_init_manageability(adapter);
4010
4011         if (netif_running(netdev))
4012                 igb_up(adapter);
4013
4014         netif_device_attach(netdev);
4015
4016         /* let the f/w know that the h/w is now under the control of the
4017          * driver. */
4018         igb_get_hw_control(adapter);
4019
4020         return 0;
4021 }
4022 #endif
4023
4024 static void igb_shutdown(struct pci_dev *pdev)
4025 {
4026         igb_suspend(pdev, PMSG_SUSPEND);
4027 }
4028
4029 #ifdef CONFIG_NET_POLL_CONTROLLER
4030 /*
4031  * Polling 'interrupt' - used by things like netconsole to send skbs
4032  * without having to re-enable interrupts. It's not called while
4033  * the interrupt routine is executing.
4034  */
4035 static void igb_netpoll(struct net_device *netdev)
4036 {
4037         struct igb_adapter *adapter = netdev_priv(netdev);
4038         int i;
4039         int work_done = 0;
4040
4041         igb_irq_disable(adapter);
4042         for (i = 0; i < adapter->num_tx_queues; i++)
4043                 igb_clean_tx_irq(adapter, &adapter->tx_ring[i]);
4044
4045         for (i = 0; i < adapter->num_rx_queues; i++)
4046                 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i],
4047                                      &work_done,
4048                                      adapter->rx_ring[i].napi.weight);
4049
4050         igb_irq_enable(adapter);
4051 }
4052 #endif /* CONFIG_NET_POLL_CONTROLLER */
4053
4054 /**
4055  * igb_io_error_detected - called when PCI error is detected
4056  * @pdev: Pointer to PCI device
4057  * @state: The current pci connection state
4058  *
4059  * This function is called after a PCI bus error affecting
4060  * this device has been detected.
4061  */
4062 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4063                                               pci_channel_state_t state)
4064 {
4065         struct net_device *netdev = pci_get_drvdata(pdev);
4066         struct igb_adapter *adapter = netdev_priv(netdev);
4067
4068         netif_device_detach(netdev);
4069
4070         if (netif_running(netdev))
4071                 igb_down(adapter);
4072         pci_disable_device(pdev);
4073
4074         /* Request a slot slot reset. */
4075         return PCI_ERS_RESULT_NEED_RESET;
4076 }
4077
4078 /**
4079  * igb_io_slot_reset - called after the pci bus has been reset.
4080  * @pdev: Pointer to PCI device
4081  *
4082  * Restart the card from scratch, as if from a cold-boot. Implementation
4083  * resembles the first-half of the igb_resume routine.
4084  */
4085 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4086 {
4087         struct net_device *netdev = pci_get_drvdata(pdev);
4088         struct igb_adapter *adapter = netdev_priv(netdev);
4089         struct e1000_hw *hw = &adapter->hw;
4090
4091         if (pci_enable_device(pdev)) {
4092                 dev_err(&pdev->dev,
4093                         "Cannot re-enable PCI device after reset.\n");
4094                 return PCI_ERS_RESULT_DISCONNECT;
4095         }
4096         pci_set_master(pdev);
4097
4098         pci_enable_wake(pdev, PCI_D3hot, 0);
4099         pci_enable_wake(pdev, PCI_D3cold, 0);
4100
4101         igb_reset(adapter);
4102         wr32(E1000_WUS, ~0);
4103
4104         return PCI_ERS_RESULT_RECOVERED;
4105 }
4106
4107 /**
4108  * igb_io_resume - called when traffic can start flowing again.
4109  * @pdev: Pointer to PCI device
4110  *
4111  * This callback is called when the error recovery driver tells us that
4112  * its OK to resume normal operation. Implementation resembles the
4113  * second-half of the igb_resume routine.
4114  */
4115 static void igb_io_resume(struct pci_dev *pdev)
4116 {
4117         struct net_device *netdev = pci_get_drvdata(pdev);
4118         struct igb_adapter *adapter = netdev_priv(netdev);
4119
4120         igb_init_manageability(adapter);
4121
4122         if (netif_running(netdev)) {
4123                 if (igb_up(adapter)) {
4124                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4125                         return;
4126                 }
4127         }
4128
4129         netif_device_attach(netdev);
4130
4131         /* let the f/w know that the h/w is now under the control of the
4132          * driver. */
4133         igb_get_hw_control(adapter);
4134
4135 }
4136
4137 /* igb_main.c */