drivers/net/igb/igb_main.c

   1 /*******************************************************************************
   2
   3   Intel(R) Gigabit Ethernet Linux driver
   4   Copyright(c) 2007 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/netdevice.h>
  34 #include <linux/ipv6.h>
  35 #include <net/checksum.h>
  36 #include <net/ip6_checksum.h>
  37 #include <linux/mii.h>
  38 #include <linux/ethtool.h>
  39 #include <linux/if_vlan.h>
  40 #include <linux/pci.h>
  41 #include <linux/delay.h>
  42 #include <linux/interrupt.h>
  43 #include <linux/if_ether.h>
  44
  45 #include "igb.h"
  46
  47 #define DRV_VERSION "1.0.8-k2"
  48 char igb_driver_name[] = "igb";
  49 char igb_driver_version[] = DRV_VERSION;
  50 static const char igb_driver_string[] =
  51                                 "Intel(R) Gigabit Ethernet Network Driver";
  52 static const char igb_copyright[] = "Copyright (c) 2007 Intel Corporation.";
  53
  54
  55 static const struct e1000_info *igb_info_tbl[] = {
  56         [board_82575] = &e1000_82575_info,
  57 };
  58
  59 static struct pci_device_id igb_pci_tbl[] = {
  60         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
  61         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
  62         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
  63         /* required last entry */
  64         {0, }
  65 };
  66
  67 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
  68
  69 void igb_reset(struct igb_adapter *);
  70 static int igb_setup_all_tx_resources(struct igb_adapter *);
  71 static int igb_setup_all_rx_resources(struct igb_adapter *);
  72 static void igb_free_all_tx_resources(struct igb_adapter *);
  73 static void igb_free_all_rx_resources(struct igb_adapter *);
  74 static void igb_free_tx_resources(struct igb_adapter *, struct igb_ring *);
  75 static void igb_free_rx_resources(struct igb_adapter *, struct igb_ring *);
  76 void igb_update_stats(struct igb_adapter *);
  77 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
  78 static void __devexit igb_remove(struct pci_dev *pdev);
  79 static int igb_sw_init(struct igb_adapter *);
  80 static int igb_open(struct net_device *);
  81 static int igb_close(struct net_device *);
  82 static void igb_configure_tx(struct igb_adapter *);
  83 static void igb_configure_rx(struct igb_adapter *);
  84 static void igb_setup_rctl(struct igb_adapter *);
  85 static void igb_clean_all_tx_rings(struct igb_adapter *);
  86 static void igb_clean_all_rx_rings(struct igb_adapter *);
  87 static void igb_clean_tx_ring(struct igb_adapter *, struct igb_ring *);
  88 static void igb_clean_rx_ring(struct igb_adapter *, struct igb_ring *);
  89 static void igb_set_multi(struct net_device *);
  90 static void igb_update_phy_info(unsigned long);
  91 static void igb_watchdog(unsigned long);
  92 static void igb_watchdog_task(struct work_struct *);
  93 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
  94                                   struct igb_ring *);
  95 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
  96 static struct net_device_stats *igb_get_stats(struct net_device *);
  97 static int igb_change_mtu(struct net_device *, int);
  98 static int igb_set_mac(struct net_device *, void *);
  99 static irqreturn_t igb_intr(int irq, void *);
 100 static irqreturn_t igb_intr_msi(int irq, void *);
 101 static irqreturn_t igb_msix_other(int irq, void *);
 102 static irqreturn_t igb_msix_rx(int irq, void *);
 103 static irqreturn_t igb_msix_tx(int irq, void *);
 104 static int igb_clean_rx_ring_msix(struct napi_struct *, int);
 105 static bool igb_clean_tx_irq(struct igb_adapter *, struct igb_ring *);
 106 static int igb_clean(struct napi_struct *, int);
 107 static bool igb_clean_rx_irq_adv(struct igb_adapter *,
 108                                  struct igb_ring *, int *, int);
 109 static void igb_alloc_rx_buffers_adv(struct igb_adapter *,
 110                                      struct igb_ring *, int);
 111 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 112 static void igb_tx_timeout(struct net_device *);
 113 static void igb_reset_task(struct work_struct *);
 114 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
 115 static void igb_vlan_rx_add_vid(struct net_device *, u16);
 116 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
 117 static void igb_restore_vlan(struct igb_adapter *);
 118
 119 static int igb_suspend(struct pci_dev *, pm_message_t);
 120 #ifdef CONFIG_PM
 121 static int igb_resume(struct pci_dev *);
 122 #endif
 123 static void igb_shutdown(struct pci_dev *);
 124
 125 #ifdef CONFIG_NET_POLL_CONTROLLER
 126 /* for netdump / net console */
 127 static void igb_netpoll(struct net_device *);
 128 #endif
 129
 130 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 131                      pci_channel_state_t);
 132 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 133 static void igb_io_resume(struct pci_dev *);
 134
 135 static struct pci_error_handlers igb_err_handler = {
 136         .error_detected = igb_io_error_detected,
 137         .slot_reset = igb_io_slot_reset,
 138         .resume = igb_io_resume,
 139 };
 140
 141
 142 static struct pci_driver igb_driver = {
 143         .name     = igb_driver_name,
 144         .id_table = igb_pci_tbl,
 145         .probe    = igb_probe,
 146         .remove   = __devexit_p(igb_remove),
 147 #ifdef CONFIG_PM
 148         /* Power Managment Hooks */
 149         .suspend  = igb_suspend,
 150         .resume   = igb_resume,
 151 #endif
 152         .shutdown = igb_shutdown,
 153         .err_handler = &igb_err_handler
 154 };
 155
 156 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 157 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 158 MODULE_LICENSE("GPL");
 159 MODULE_VERSION(DRV_VERSION);
 160
 161 #ifdef DEBUG
 162 /**
 163  * igb_get_hw_dev_name - return device name string
 164  * used by hardware layer to print debugging information
 165  **/
 166 char *igb_get_hw_dev_name(struct e1000_hw *hw)
 167 {
 168         struct igb_adapter *adapter = hw->back;
 169         return adapter->netdev->name;
 170 }
 171 #endif
 172
 173 /**
 174  * igb_init_module - Driver Registration Routine
 175  *
 176  * igb_init_module is the first routine called when the driver is
 177  * loaded. All it does is register with the PCI subsystem.
 178  **/
 179 static int __init igb_init_module(void)
 180 {
 181         int ret;
 182         printk(KERN_INFO "%s - version %s\n",
 183                igb_driver_string, igb_driver_version);
 184
 185         printk(KERN_INFO "%s\n", igb_copyright);
 186
 187         ret = pci_register_driver(&igb_driver);
 188         return ret;
 189 }
 190
 191 module_init(igb_init_module);
 192
 193 /**
 194  * igb_exit_module - Driver Exit Cleanup Routine
 195  *
 196  * igb_exit_module is called just before the driver is removed
 197  * from memory.
 198  **/
 199 static void __exit igb_exit_module(void)
 200 {
 201         pci_unregister_driver(&igb_driver);
 202 }
 203
 204 module_exit(igb_exit_module);
 205
 206 /**
 207  * igb_alloc_queues - Allocate memory for all rings
 208  * @adapter: board private structure to initialize
 209  *
 210  * We allocate one ring per queue at run-time since we don't know the
 211  * number of queues at compile-time.
 212  **/
 213 static int igb_alloc_queues(struct igb_adapter *adapter)
 214 {
 215         int i;
 216
 217         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
 218                                    sizeof(struct igb_ring), GFP_KERNEL);
 219         if (!adapter->tx_ring)
 220                 return -ENOMEM;
 221
 222         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
 223                                    sizeof(struct igb_ring), GFP_KERNEL);
 224         if (!adapter->rx_ring) {
 225                 kfree(adapter->tx_ring);
 226                 return -ENOMEM;
 227         }
 228
 229         for (i = 0; i < adapter->num_rx_queues; i++) {
 230                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 231                 ring->adapter = adapter;
 232                 ring->itr_register = E1000_ITR;
 233
 234                 if (!ring->napi.poll)
 235                         netif_napi_add(adapter->netdev, &ring->napi, igb_clean,
 236                                        adapter->napi.weight /
 237                                        adapter->num_rx_queues);
 238         }
 239         return 0;
 240 }
 241
 242 #define IGB_N0_QUEUE -1
 243 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
 244                               int tx_queue, int msix_vector)
 245 {
 246         u32 msixbm = 0;
 247         struct e1000_hw *hw = &adapter->hw;
 248                 /* The 82575 assigns vectors using a bitmask, which matches the
 249                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
 250                    or more queues to a vector, we write the appropriate bits
 251                    into the MSIXBM register for that vector. */
 252                 if (rx_queue > IGB_N0_QUEUE) {
 253                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 254                         adapter->rx_ring[rx_queue].eims_value = msixbm;
 255                 }
 256                 if (tx_queue > IGB_N0_QUEUE) {
 257                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 258                         adapter->tx_ring[tx_queue].eims_value =
 259                                   E1000_EICR_TX_QUEUE0 << tx_queue;
 260                 }
 261                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 262 }
 263
 264 /**
 265  * igb_configure_msix - Configure MSI-X hardware
 266  *
 267  * igb_configure_msix sets up the hardware to properly
 268  * generate MSI-X interrupts.
 269  **/
 270 static void igb_configure_msix(struct igb_adapter *adapter)
 271 {
 272         u32 tmp;
 273         int i, vector = 0;
 274         struct e1000_hw *hw = &adapter->hw;
 275
 276         adapter->eims_enable_mask = 0;
 277
 278         for (i = 0; i < adapter->num_tx_queues; i++) {
 279                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
 280                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
 281                 adapter->eims_enable_mask |= tx_ring->eims_value;
 282                 if (tx_ring->itr_val)
 283                         writel(1000000000 / (tx_ring->itr_val * 256),
 284                                hw->hw_addr + tx_ring->itr_register);
 285                 else
 286                         writel(1, hw->hw_addr + tx_ring->itr_register);
 287         }
 288
 289         for (i = 0; i < adapter->num_rx_queues; i++) {
 290                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
 291                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
 292                 adapter->eims_enable_mask |= rx_ring->eims_value;
 293                 if (rx_ring->itr_val)
 294                         writel(1000000000 / (rx_ring->itr_val * 256),
 295                                hw->hw_addr + rx_ring->itr_register);
 296                 else
 297                         writel(1, hw->hw_addr + rx_ring->itr_register);
 298         }
 299
 300
 301         /* set vector for other causes, i.e. link changes */
 302                 array_wr32(E1000_MSIXBM(0), vector++,
 303                                       E1000_EIMS_OTHER);
 304
 305                 /* disable IAM for ICR interrupt bits */
 306                 wr32(E1000_IAM, 0);
 307
 308                 tmp = rd32(E1000_CTRL_EXT);
 309                 /* enable MSI-X PBA support*/
 310                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 311
 312                 /* Auto-Mask interrupts upon ICR read. */
 313                 tmp |= E1000_CTRL_EXT_EIAME;
 314                 tmp |= E1000_CTRL_EXT_IRCA;
 315
 316                 wr32(E1000_CTRL_EXT, tmp);
 317                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
 318
 319         wrfl();
 320 }
 321
 322 /**
 323  * igb_request_msix - Initialize MSI-X interrupts
 324  *
 325  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
 326  * kernel.
 327  **/
 328 static int igb_request_msix(struct igb_adapter *adapter)
 329 {
 330         struct net_device *netdev = adapter->netdev;
 331         int i, err = 0, vector = 0;
 332
 333         vector = 0;
 334
 335         for (i = 0; i < adapter->num_tx_queues; i++) {
 336                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 337                 sprintf(ring->name, "%s-tx%d", netdev->name, i);
 338                 err = request_irq(adapter->msix_entries[vector].vector,
 339                                   &igb_msix_tx, 0, ring->name,
 340                                   &(adapter->tx_ring[i]));
 341                 if (err)
 342                         goto out;
 343                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 344                 ring->itr_val = adapter->itr;
 345                 vector++;
 346         }
 347         for (i = 0; i < adapter->num_rx_queues; i++) {
 348                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 349                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
 350                         sprintf(ring->name, "%s-rx%d", netdev->name, i);
 351                 else
 352                         memcpy(ring->name, netdev->name, IFNAMSIZ);
 353                 err = request_irq(adapter->msix_entries[vector].vector,
 354                                   &igb_msix_rx, 0, ring->name,
 355                                   &(adapter->rx_ring[i]));
 356                 if (err)
 357                         goto out;
 358                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 359                 ring->itr_val = adapter->itr;
 360                 vector++;
 361         }
 362
 363         err = request_irq(adapter->msix_entries[vector].vector,
 364                           &igb_msix_other, 0, netdev->name, netdev);
 365         if (err)
 366                 goto out;
 367
 368         adapter->napi.poll = igb_clean_rx_ring_msix;
 369         for (i = 0; i < adapter->num_rx_queues; i++)
 370                 adapter->rx_ring[i].napi.poll = adapter->napi.poll;
 371         igb_configure_msix(adapter);
 372         return 0;
 373 out:
 374         return err;
 375 }
 376
 377 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 378 {
 379         if (adapter->msix_entries) {
 380                 pci_disable_msix(adapter->pdev);
 381                 kfree(adapter->msix_entries);
 382                 adapter->msix_entries = NULL;
 383         } else if (adapter->msi_enabled)
 384                 pci_disable_msi(adapter->pdev);
 385         return;
 386 }
 387
 388
 389 /**
 390  * igb_set_interrupt_capability - set MSI or MSI-X if supported
 391  *
 392  * Attempt to configure interrupts using the best available
 393  * capabilities of the hardware and kernel.
 394  **/
 395 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
 396 {
 397         int err;
 398         int numvecs, i;
 399
 400         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
 401         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
 402                                         GFP_KERNEL);
 403         if (!adapter->msix_entries)
 404                 goto msi_only;
 405
 406         for (i = 0; i < numvecs; i++)
 407                 adapter->msix_entries[i].entry = i;
 408
 409         err = pci_enable_msix(adapter->pdev,
 410                               adapter->msix_entries,
 411                               numvecs);
 412         if (err == 0)
 413                 return;
 414
 415         igb_reset_interrupt_capability(adapter);
 416
 417         /* If we can't do MSI-X, try MSI */
 418 msi_only:
 419         adapter->num_rx_queues = 1;
 420         if (!pci_enable_msi(adapter->pdev))
 421                 adapter->msi_enabled = 1;
 422         return;
 423 }
 424
 425 /**
 426  * igb_request_irq - initialize interrupts
 427  *
 428  * Attempts to configure interrupts using the best available
 429  * capabilities of the hardware and kernel.
 430  **/
 431 static int igb_request_irq(struct igb_adapter *adapter)
 432 {
 433         struct net_device *netdev = adapter->netdev;
 434         struct e1000_hw *hw = &adapter->hw;
 435         int err = 0;
 436
 437         if (adapter->msix_entries) {
 438                 err = igb_request_msix(adapter);
 439                 if (!err) {
 440                         /* enable IAM, auto-mask,
 441                          * DO NOT USE EIAM or IAM in legacy mode */
 442                         wr32(E1000_IAM, IMS_ENABLE_MASK);
 443                         goto request_done;
 444                 }
 445                 /* fall back to MSI */
 446                 igb_reset_interrupt_capability(adapter);
 447                 if (!pci_enable_msi(adapter->pdev))
 448                         adapter->msi_enabled = 1;
 449                 igb_free_all_tx_resources(adapter);
 450                 igb_free_all_rx_resources(adapter);
 451                 adapter->num_rx_queues = 1;
 452                 igb_alloc_queues(adapter);
 453         }
 454         if (adapter->msi_enabled) {
 455                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
 456                                   netdev->name, netdev);
 457                 if (!err)
 458                         goto request_done;
 459                 /* fall back to legacy interrupts */
 460                 igb_reset_interrupt_capability(adapter);
 461                 adapter->msi_enabled = 0;
 462         }
 463
 464         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
 465                           netdev->name, netdev);
 466
 467         if (err)
 468                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
 469                         err);
 470
 471 request_done:
 472         return err;
 473 }
 474
 475 static void igb_free_irq(struct igb_adapter *adapter)
 476 {
 477         struct net_device *netdev = adapter->netdev;
 478
 479         if (adapter->msix_entries) {
 480                 int vector = 0, i;
 481
 482                 for (i = 0; i < adapter->num_tx_queues; i++)
 483                         free_irq(adapter->msix_entries[vector++].vector,
 484                                 &(adapter->tx_ring[i]));
 485                 for (i = 0; i < adapter->num_rx_queues; i++)
 486                         free_irq(adapter->msix_entries[vector++].vector,
 487                                 &(adapter->rx_ring[i]));
 488
 489                 free_irq(adapter->msix_entries[vector++].vector, netdev);
 490                 return;
 491         }
 492
 493         free_irq(adapter->pdev->irq, netdev);
 494 }
 495
 496 /**
 497  * igb_irq_disable - Mask off interrupt generation on the NIC
 498  * @adapter: board private structure
 499  **/
 500 static void igb_irq_disable(struct igb_adapter *adapter)
 501 {
 502         struct e1000_hw *hw = &adapter->hw;
 503
 504         if (adapter->msix_entries) {
 505                 wr32(E1000_EIMC, ~0);
 506                 wr32(E1000_EIAC, 0);
 507         }
 508         wr32(E1000_IMC, ~0);
 509         wrfl();
 510         synchronize_irq(adapter->pdev->irq);
 511 }
 512
 513 /**
 514  * igb_irq_enable - Enable default interrupt generation settings
 515  * @adapter: board private structure
 516  **/
 517 static void igb_irq_enable(struct igb_adapter *adapter)
 518 {
 519         struct e1000_hw *hw = &adapter->hw;
 520
 521         if (adapter->msix_entries) {
 522                 wr32(E1000_EIMS,
 523                                 adapter->eims_enable_mask);
 524                 wr32(E1000_EIAC,
 525                                 adapter->eims_enable_mask);
 526                 wr32(E1000_IMS, E1000_IMS_LSC);
 527         } else
 528         wr32(E1000_IMS, IMS_ENABLE_MASK);
 529 }
 530
 531 static void igb_update_mng_vlan(struct igb_adapter *adapter)
 532 {
 533         struct net_device *netdev = adapter->netdev;
 534         u16 vid = adapter->hw.mng_cookie.vlan_id;
 535         u16 old_vid = adapter->mng_vlan_id;
 536         if (adapter->vlgrp) {
 537                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
 538                         if (adapter->hw.mng_cookie.status &
 539                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
 540                                 igb_vlan_rx_add_vid(netdev, vid);
 541                                 adapter->mng_vlan_id = vid;
 542                         } else
 543                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
 544
 545                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
 546                                         (vid != old_vid) &&
 547                             !vlan_group_get_device(adapter->vlgrp, old_vid))
 548                                 igb_vlan_rx_kill_vid(netdev, old_vid);
 549                 } else
 550                         adapter->mng_vlan_id = vid;
 551         }
 552 }
 553
 554 /**
 555  * igb_release_hw_control - release control of the h/w to f/w
 556  * @adapter: address of board private structure
 557  *
 558  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 559  * For ASF and Pass Through versions of f/w this means that the
 560  * driver is no longer loaded.
 561  *
 562  **/
 563 static void igb_release_hw_control(struct igb_adapter *adapter)
 564 {
 565         struct e1000_hw *hw = &adapter->hw;
 566         u32 ctrl_ext;
 567
 568         /* Let firmware take over control of h/w */
 569         ctrl_ext = rd32(E1000_CTRL_EXT);
 570         wr32(E1000_CTRL_EXT,
 571                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 572 }
 573
 574
 575 /**
 576  * igb_get_hw_control - get control of the h/w from f/w
 577  * @adapter: address of board private structure
 578  *
 579  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 580  * For ASF and Pass Through versions of f/w this means that
 581  * the driver is loaded.
 582  *
 583  **/
 584 static void igb_get_hw_control(struct igb_adapter *adapter)
 585 {
 586         struct e1000_hw *hw = &adapter->hw;
 587         u32 ctrl_ext;
 588
 589         /* Let firmware know the driver has taken over */
 590         ctrl_ext = rd32(E1000_CTRL_EXT);
 591         wr32(E1000_CTRL_EXT,
 592                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 593 }
 594
 595 static void igb_init_manageability(struct igb_adapter *adapter)
 596 {
 597         struct e1000_hw *hw = &adapter->hw;
 598
 599         if (adapter->en_mng_pt) {
 600                 u32 manc2h = rd32(E1000_MANC2H);
 601                 u32 manc = rd32(E1000_MANC);
 602
 603                 /* enable receiving management packets to the host */
 604                 /* this will probably generate destination unreachable messages
 605                  * from the host OS, but the packets will be handled on SMBUS */
 606                 manc |= E1000_MANC_EN_MNG2HOST;
 607 #define E1000_MNG2HOST_PORT_623 (1 << 5)
 608 #define E1000_MNG2HOST_PORT_664 (1 << 6)
 609                 manc2h |= E1000_MNG2HOST_PORT_623;
 610                 manc2h |= E1000_MNG2HOST_PORT_664;
 611                 wr32(E1000_MANC2H, manc2h);
 612
 613                 wr32(E1000_MANC, manc);
 614         }
 615 }
 616
 617 /**
 618  * igb_configure - configure the hardware for RX and TX
 619  * @adapter: private board structure
 620  **/
 621 static void igb_configure(struct igb_adapter *adapter)
 622 {
 623         struct net_device *netdev = adapter->netdev;
 624         int i;
 625
 626         igb_get_hw_control(adapter);
 627         igb_set_multi(netdev);
 628
 629         igb_restore_vlan(adapter);
 630         igb_init_manageability(adapter);
 631
 632         igb_configure_tx(adapter);
 633         igb_setup_rctl(adapter);
 634         igb_configure_rx(adapter);
 635         /* call IGB_DESC_UNUSED which always leaves
 636          * at least 1 descriptor unused to make sure
 637          * next_to_use != next_to_clean */
 638         for (i = 0; i < adapter->num_rx_queues; i++) {
 639                 struct igb_ring *ring = &adapter->rx_ring[i];
 640                 igb_alloc_rx_buffers_adv(adapter, ring, IGB_DESC_UNUSED(ring));
 641         }
 642
 643
 644         adapter->tx_queue_len = netdev->tx_queue_len;
 645 }
 646
 647
 648 /**
 649  * igb_up - Open the interface and prepare it to handle traffic
 650  * @adapter: board private structure
 651  **/
 652
 653 int igb_up(struct igb_adapter *adapter)
 654 {
 655         struct e1000_hw *hw = &adapter->hw;
 656         int i;
 657
 658         /* hardware has been reset, we need to reload some things */
 659         igb_configure(adapter);
 660
 661         clear_bit(__IGB_DOWN, &adapter->state);
 662
 663         napi_enable(&adapter->napi);
 664
 665         if (adapter->msix_entries) {
 666                 for (i = 0; i < adapter->num_rx_queues; i++)
 667                         napi_enable(&adapter->rx_ring[i].napi);
 668                 igb_configure_msix(adapter);
 669         }
 670
 671         /* Clear any pending interrupts. */
 672         rd32(E1000_ICR);
 673         igb_irq_enable(adapter);
 674
 675         /* Fire a link change interrupt to start the watchdog. */
 676         wr32(E1000_ICS, E1000_ICS_LSC);
 677         return 0;
 678 }
 679
 680 void igb_down(struct igb_adapter *adapter)
 681 {
 682         struct e1000_hw *hw = &adapter->hw;
 683         struct net_device *netdev = adapter->netdev;
 684         u32 tctl, rctl;
 685         int i;
 686
 687         /* signal that we're down so the interrupt handler does not
 688          * reschedule our watchdog timer */
 689         set_bit(__IGB_DOWN, &adapter->state);
 690
 691         /* disable receives in the hardware */
 692         rctl = rd32(E1000_RCTL);
 693         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
 694         /* flush and sleep below */
 695
 696         netif_stop_queue(netdev);
 697
 698         /* disable transmits in the hardware */
 699         tctl = rd32(E1000_TCTL);
 700         tctl &= ~E1000_TCTL_EN;
 701         wr32(E1000_TCTL, tctl);
 702         /* flush both disables and wait for them to finish */
 703         wrfl();
 704         msleep(10);
 705
 706         napi_disable(&adapter->napi);
 707
 708         if (adapter->msix_entries)
 709                 for (i = 0; i < adapter->num_rx_queues; i++)
 710                         napi_disable(&adapter->rx_ring[i].napi);
 711         igb_irq_disable(adapter);
 712
 713         del_timer_sync(&adapter->watchdog_timer);
 714         del_timer_sync(&adapter->phy_info_timer);
 715
 716         netdev->tx_queue_len = adapter->tx_queue_len;
 717         netif_carrier_off(netdev);
 718         adapter->link_speed = 0;
 719         adapter->link_duplex = 0;
 720
 721         if (!pci_channel_offline(adapter->pdev))
 722                 igb_reset(adapter);
 723         igb_clean_all_tx_rings(adapter);
 724         igb_clean_all_rx_rings(adapter);
 725 }
 726
 727 void igb_reinit_locked(struct igb_adapter *adapter)
 728 {
 729         WARN_ON(in_interrupt());
 730         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 731                 msleep(1);
 732         igb_down(adapter);
 733         igb_up(adapter);
 734         clear_bit(__IGB_RESETTING, &adapter->state);
 735 }
 736
 737 void igb_reset(struct igb_adapter *adapter)
 738 {
 739         struct e1000_hw *hw = &adapter->hw;
 740         struct e1000_fc_info *fc = &adapter->hw.fc;
 741         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
 742         u16 hwm;
 743
 744         /* Repartition Pba for greater than 9k mtu
 745          * To take effect CTRL.RST is required.
 746          */
 747         pba = E1000_PBA_34K;
 748
 749         if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
 750                 /* adjust PBA for jumbo frames */
 751                 wr32(E1000_PBA, pba);
 752
 753                 /* To maintain wire speed transmits, the Tx FIFO should be
 754                  * large enough to accommodate two full transmit packets,
 755                  * rounded up to the next 1KB and expressed in KB.  Likewise,
 756                  * the Rx FIFO should be large enough to accommodate at least
 757                  * one full receive packet and is similarly rounded up and
 758                  * expressed in KB. */
 759                 pba = rd32(E1000_PBA);
 760                 /* upper 16 bits has Tx packet buffer allocation size in KB */
 761                 tx_space = pba >> 16;
 762                 /* lower 16 bits has Rx packet buffer allocation size in KB */
 763                 pba &= 0xffff;
 764                 /* the tx fifo also stores 16 bytes of information about the tx
 765                  * but don't include ethernet FCS because hardware appends it */
 766                 min_tx_space = (adapter->max_frame_size +
 767                                 sizeof(struct e1000_tx_desc) -
 768                                 ETH_FCS_LEN) * 2;
 769                 min_tx_space = ALIGN(min_tx_space, 1024);
 770                 min_tx_space >>= 10;
 771                 /* software strips receive CRC, so leave room for it */
 772                 min_rx_space = adapter->max_frame_size;
 773                 min_rx_space = ALIGN(min_rx_space, 1024);
 774                 min_rx_space >>= 10;
 775
 776                 /* If current Tx allocation is less than the min Tx FIFO size,
 777                  * and the min Tx FIFO size is less than the current Rx FIFO
 778                  * allocation, take space away from current Rx allocation */
 779                 if (tx_space < min_tx_space &&
 780                     ((min_tx_space - tx_space) < pba)) {
 781                         pba = pba - (min_tx_space - tx_space);
 782
 783                         /* if short on rx space, rx wins and must trump tx
 784                          * adjustment */
 785                         if (pba < min_rx_space)
 786                                 pba = min_rx_space;
 787                 }
 788         }
 789         wr32(E1000_PBA, pba);
 790
 791         /* flow control settings */
 792         /* The high water mark must be low enough to fit one full frame
 793          * (or the size used for early receive) above it in the Rx FIFO.
 794          * Set it to the lower of:
 795          * - 90% of the Rx FIFO size, or
 796          * - the full Rx FIFO size minus one full frame */
 797         hwm = min(((pba << 10) * 9 / 10),
 798                   ((pba << 10) - adapter->max_frame_size));
 799
 800         fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
 801         fc->low_water = fc->high_water - 8;
 802         fc->pause_time = 0xFFFF;
 803         fc->send_xon = 1;
 804         fc->type = fc->original_type;
 805
 806         /* Allow time for pending master requests to run */
 807         adapter->hw.mac.ops.reset_hw(&adapter->hw);
 808         wr32(E1000_WUC, 0);
 809
 810         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
 811                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
 812
 813         igb_update_mng_vlan(adapter);
 814
 815         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 816         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
 817
 818         igb_reset_adaptive(&adapter->hw);
 819         if (adapter->hw.phy.ops.get_phy_info)
 820                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
 821 }
 822
 823 /**
 824  * igb_probe - Device Initialization Routine
 825  * @pdev: PCI device information struct
 826  * @ent: entry in igb_pci_tbl
 827  *
 828  * Returns 0 on success, negative on failure
 829  *
 830  * igb_probe initializes an adapter identified by a pci_dev structure.
 831  * The OS initialization, configuring of the adapter private structure,
 832  * and a hardware reset occur.
 833  **/
 834 static int __devinit igb_probe(struct pci_dev *pdev,
 835                                const struct pci_device_id *ent)
 836 {
 837         struct net_device *netdev;
 838         struct igb_adapter *adapter;
 839         struct e1000_hw *hw;
 840         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
 841         unsigned long mmio_start, mmio_len;
 842         static int cards_found;
 843         int i, err, pci_using_dac;
 844         u16 eeprom_data = 0;
 845         u16 eeprom_apme_mask = IGB_EEPROM_APME;
 846         u32 part_num;
 847
 848         err = pci_enable_device(pdev);
 849         if (err)
 850                 return err;
 851
 852         pci_using_dac = 0;
 853         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
 854         if (!err) {
 855                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
 856                 if (!err)
 857                         pci_using_dac = 1;
 858         } else {
 859                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
 860                 if (err) {
 861                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
 862                         if (err) {
 863                                 dev_err(&pdev->dev, "No usable DMA "
 864                                         "configuration, aborting\n");
 865                                 goto err_dma;
 866                         }
 867                 }
 868         }
 869
 870         err = pci_request_regions(pdev, igb_driver_name);
 871         if (err)
 872                 goto err_pci_reg;
 873
 874         pci_set_master(pdev);
 875         pci_save_state(pdev);
 876
 877         err = -ENOMEM;
 878         netdev = alloc_etherdev(sizeof(struct igb_adapter));
 879         if (!netdev)
 880                 goto err_alloc_etherdev;
 881
 882         SET_NETDEV_DEV(netdev, &pdev->dev);
 883
 884         pci_set_drvdata(pdev, netdev);
 885         adapter = netdev_priv(netdev);
 886         adapter->netdev = netdev;
 887         adapter->pdev = pdev;
 888         hw = &adapter->hw;
 889         hw->back = adapter;
 890         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
 891
 892         mmio_start = pci_resource_start(pdev, 0);
 893         mmio_len = pci_resource_len(pdev, 0);
 894
 895         err = -EIO;
 896         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
 897         if (!adapter->hw.hw_addr)
 898                 goto err_ioremap;
 899
 900         netdev->open = &igb_open;
 901         netdev->stop = &igb_close;
 902         netdev->get_stats = &igb_get_stats;
 903         netdev->set_multicast_list = &igb_set_multi;
 904         netdev->set_mac_address = &igb_set_mac;
 905         netdev->change_mtu = &igb_change_mtu;
 906         netdev->do_ioctl = &igb_ioctl;
 907         igb_set_ethtool_ops(netdev);
 908         netdev->tx_timeout = &igb_tx_timeout;
 909         netdev->watchdog_timeo = 5 * HZ;
 910         netif_napi_add(netdev, &adapter->napi, igb_clean, 64);
 911         netdev->vlan_rx_register = igb_vlan_rx_register;
 912         netdev->vlan_rx_add_vid = igb_vlan_rx_add_vid;
 913         netdev->vlan_rx_kill_vid = igb_vlan_rx_kill_vid;
 914 #ifdef CONFIG_NET_POLL_CONTROLLER
 915         netdev->poll_controller = igb_netpoll;
 916 #endif
 917         netdev->hard_start_xmit = &igb_xmit_frame_adv;
 918
 919         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
 920
 921         netdev->mem_start = mmio_start;
 922         netdev->mem_end = mmio_start + mmio_len;
 923
 924         adapter->bd_number = cards_found;
 925
 926         /* PCI config space info */
 927         hw->vendor_id = pdev->vendor;
 928         hw->device_id = pdev->device;
 929         hw->revision_id = pdev->revision;
 930         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 931         hw->subsystem_device_id = pdev->subsystem_device;
 932
 933         /* setup the private structure */
 934         hw->back = adapter;
 935         /* Copy the default MAC, PHY and NVM function pointers */
 936         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
 937         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
 938         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
 939         /* Initialize skew-specific constants */
 940         err = ei->get_invariants(hw);
 941         if (err)
 942                 goto err_hw_init;
 943
 944         err = igb_sw_init(adapter);
 945         if (err)
 946                 goto err_sw_init;
 947
 948         igb_get_bus_info_pcie(hw);
 949
 950         hw->phy.autoneg_wait_to_complete = false;
 951         hw->mac.adaptive_ifs = true;
 952
 953         /* Copper options */
 954         if (hw->phy.media_type == e1000_media_type_copper) {
 955                 hw->phy.mdix = AUTO_ALL_MODES;
 956                 hw->phy.disable_polarity_correction = false;
 957                 hw->phy.ms_type = e1000_ms_hw_default;
 958         }
 959
 960         if (igb_check_reset_block(hw))
 961                 dev_info(&pdev->dev,
 962                         "PHY reset is blocked due to SOL/IDER session.\n");
 963
 964         netdev->features = NETIF_F_SG |
 965                            NETIF_F_HW_CSUM |
 966                            NETIF_F_HW_VLAN_TX |
 967                            NETIF_F_HW_VLAN_RX |
 968                            NETIF_F_HW_VLAN_FILTER;
 969
 970         netdev->features |= NETIF_F_TSO;
 971
 972         netdev->features |= NETIF_F_TSO6;
 973         if (pci_using_dac)
 974                 netdev->features |= NETIF_F_HIGHDMA;
 975
 976         netdev->features |= NETIF_F_LLTX;
 977         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
 978
 979         /* before reading the NVM, reset the controller to put the device in a
 980          * known good starting state */
 981         hw->mac.ops.reset_hw(hw);
 982
 983         /* make sure the NVM is good */
 984         if (igb_validate_nvm_checksum(hw) < 0) {
 985                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
 986                 err = -EIO;
 987                 goto err_eeprom;
 988         }
 989
 990         /* copy the MAC address out of the NVM */
 991         if (hw->mac.ops.read_mac_addr(hw))
 992                 dev_err(&pdev->dev, "NVM Read Error\n");
 993
 994         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
 995         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
 996
 997         if (!is_valid_ether_addr(netdev->perm_addr)) {
 998                 dev_err(&pdev->dev, "Invalid MAC Address\n");
 999                 err = -EIO;
1000                 goto err_eeprom;
1001         }
1002
1003         init_timer(&adapter->watchdog_timer);
1004         adapter->watchdog_timer.function = &igb_watchdog;
1005         adapter->watchdog_timer.data = (unsigned long) adapter;
1006
1007         init_timer(&adapter->phy_info_timer);
1008         adapter->phy_info_timer.function = &igb_update_phy_info;
1009         adapter->phy_info_timer.data = (unsigned long) adapter;
1010
1011         INIT_WORK(&adapter->reset_task, igb_reset_task);
1012         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1013
1014         /* Initialize link & ring properties that are user-changeable */
1015         adapter->tx_ring->count = 256;
1016         for (i = 0; i < adapter->num_tx_queues; i++)
1017                 adapter->tx_ring[i].count = adapter->tx_ring->count;
1018         adapter->rx_ring->count = 256;
1019         for (i = 0; i < adapter->num_rx_queues; i++)
1020                 adapter->rx_ring[i].count = adapter->rx_ring->count;
1021
1022         adapter->fc_autoneg = true;
1023         hw->mac.autoneg = true;
1024         hw->phy.autoneg_advertised = 0x2f;
1025
1026         hw->fc.original_type = e1000_fc_default;
1027         hw->fc.type = e1000_fc_default;
1028
1029         adapter->itr_setting = 3;
1030         adapter->itr = IGB_START_ITR;
1031
1032         igb_validate_mdi_setting(hw);
1033
1034         adapter->rx_csum = 1;
1035
1036         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1037          * enable the ACPI Magic Packet filter
1038          */
1039
1040         if (hw->bus.func == 0 ||
1041             hw->device_id == E1000_DEV_ID_82575EB_COPPER)
1042                 hw->nvm.ops.read_nvm(hw, NVM_INIT_CONTROL3_PORT_A, 1,
1043                                      &eeprom_data);
1044
1045         if (eeprom_data & eeprom_apme_mask)
1046                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1047
1048         /* now that we have the eeprom settings, apply the special cases where
1049          * the eeprom may be wrong or the board simply won't support wake on
1050          * lan on a particular port */
1051         switch (pdev->device) {
1052         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1053                 adapter->eeprom_wol = 0;
1054                 break;
1055         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1056                 /* Wake events only supported on port A for dual fiber
1057                  * regardless of eeprom setting */
1058                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1059                         adapter->eeprom_wol = 0;
1060                 break;
1061         }
1062
1063         /* initialize the wol settings based on the eeprom settings */
1064         adapter->wol = adapter->eeprom_wol;
1065
1066         /* reset the hardware with the new settings */
1067         igb_reset(adapter);
1068
1069         /* let the f/w know that the h/w is now under the control of the
1070          * driver. */
1071         igb_get_hw_control(adapter);
1072
1073         /* tell the stack to leave us alone until igb_open() is called */
1074         netif_carrier_off(netdev);
1075         netif_stop_queue(netdev);
1076
1077         strcpy(netdev->name, "eth%d");
1078         err = register_netdev(netdev);
1079         if (err)
1080                 goto err_register;
1081
1082         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1083         /* print bus type/speed/width info */
1084         dev_info(&pdev->dev,
1085                  "%s: (PCIe:%s:%s) %02x:%02x:%02x:%02x:%02x:%02x\n",
1086                  netdev->name,
1087                  ((hw->bus.speed == e1000_bus_speed_2500)
1088                   ? "2.5Gb/s" : "unknown"),
1089                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1090                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1091                   ? "Width x1" : "unknown"),
1092                  netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
1093                  netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
1094
1095         igb_read_part_num(hw, &part_num);
1096         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1097                 (part_num >> 8), (part_num & 0xff));
1098
1099         dev_info(&pdev->dev,
1100                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1101                 adapter->msix_entries ? "MSI-X" :
1102                 adapter->msi_enabled ? "MSI" : "legacy",
1103                 adapter->num_rx_queues, adapter->num_tx_queues);
1104
1105         cards_found++;
1106         return 0;
1107
1108 err_register:
1109         igb_release_hw_control(adapter);
1110 err_eeprom:
1111         if (!igb_check_reset_block(hw))
1112                 hw->phy.ops.reset_phy(hw);
1113
1114         if (hw->flash_address)
1115                 iounmap(hw->flash_address);
1116
1117         igb_remove_device(hw);
1118         kfree(adapter->tx_ring);
1119         kfree(adapter->rx_ring);
1120 err_sw_init:
1121 err_hw_init:
1122         iounmap(hw->hw_addr);
1123 err_ioremap:
1124         free_netdev(netdev);
1125 err_alloc_etherdev:
1126         pci_release_regions(pdev);
1127 err_pci_reg:
1128 err_dma:
1129         pci_disable_device(pdev);
1130         return err;
1131 }
1132
1133 /**
1134  * igb_remove - Device Removal Routine
1135  * @pdev: PCI device information struct
1136  *
1137  * igb_remove is called by the PCI subsystem to alert the driver
1138  * that it should release a PCI device.  The could be caused by a
1139  * Hot-Plug event, or because the driver is going to be removed from
1140  * memory.
1141  **/
1142 static void __devexit igb_remove(struct pci_dev *pdev)
1143 {
1144         struct net_device *netdev = pci_get_drvdata(pdev);
1145         struct igb_adapter *adapter = netdev_priv(netdev);
1146
1147         /* flush_scheduled work may reschedule our watchdog task, so
1148          * explicitly disable watchdog tasks from being rescheduled  */
1149         set_bit(__IGB_DOWN, &adapter->state);
1150         del_timer_sync(&adapter->watchdog_timer);
1151         del_timer_sync(&adapter->phy_info_timer);
1152
1153         flush_scheduled_work();
1154
1155         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1156          * would have already happened in close and is redundant. */
1157         igb_release_hw_control(adapter);
1158
1159         unregister_netdev(netdev);
1160
1161         if (!igb_check_reset_block(&adapter->hw))
1162                 adapter->hw.phy.ops.reset_phy(&adapter->hw);
1163
1164         igb_remove_device(&adapter->hw);
1165         igb_reset_interrupt_capability(adapter);
1166
1167         kfree(adapter->tx_ring);
1168         kfree(adapter->rx_ring);
1169
1170         iounmap(adapter->hw.hw_addr);
1171         if (adapter->hw.flash_address)
1172                 iounmap(adapter->hw.flash_address);
1173         pci_release_regions(pdev);
1174
1175         free_netdev(netdev);
1176
1177         pci_disable_device(pdev);
1178 }
1179
1180 /**
1181  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1182  * @adapter: board private structure to initialize
1183  *
1184  * igb_sw_init initializes the Adapter private data structure.
1185  * Fields are initialized based on PCI device information and
1186  * OS network device settings (MTU size).
1187  **/
1188 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1189 {
1190         struct e1000_hw *hw = &adapter->hw;
1191         struct net_device *netdev = adapter->netdev;
1192         struct pci_dev *pdev = adapter->pdev;
1193
1194         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1195
1196         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1197         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1198         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1199         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1200
1201         /* Number of supported queues. */
1202         /* Having more queues than CPUs doesn't make sense. */
1203         adapter->num_tx_queues = 1;
1204         adapter->num_rx_queues = min(IGB_MAX_RX_QUEUES, num_online_cpus());
1205
1206         igb_set_interrupt_capability(adapter);
1207
1208         if (igb_alloc_queues(adapter)) {
1209                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1210                 return -ENOMEM;
1211         }
1212
1213         /* Explicitly disable IRQ since the NIC can be in any state. */
1214         igb_irq_disable(adapter);
1215
1216         set_bit(__IGB_DOWN, &adapter->state);
1217         return 0;
1218 }
1219
1220 /**
1221  * igb_open - Called when a network interface is made active
1222  * @netdev: network interface device structure
1223  *
1224  * Returns 0 on success, negative value on failure
1225  *
1226  * The open entry point is called when a network interface is made
1227  * active by the system (IFF_UP).  At this point all resources needed
1228  * for transmit and receive operations are allocated, the interrupt
1229  * handler is registered with the OS, the watchdog timer is started,
1230  * and the stack is notified that the interface is ready.
1231  **/
1232 static int igb_open(struct net_device *netdev)
1233 {
1234         struct igb_adapter *adapter = netdev_priv(netdev);
1235         struct e1000_hw *hw = &adapter->hw;
1236         int err;
1237         int i;
1238
1239         /* disallow open during test */
1240         if (test_bit(__IGB_TESTING, &adapter->state))
1241                 return -EBUSY;
1242
1243         /* allocate transmit descriptors */
1244         err = igb_setup_all_tx_resources(adapter);
1245         if (err)
1246                 goto err_setup_tx;
1247
1248         /* allocate receive descriptors */
1249         err = igb_setup_all_rx_resources(adapter);
1250         if (err)
1251                 goto err_setup_rx;
1252
1253         /* e1000_power_up_phy(adapter); */
1254
1255         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1256         if ((adapter->hw.mng_cookie.status &
1257              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1258                 igb_update_mng_vlan(adapter);
1259
1260         /* before we allocate an interrupt, we must be ready to handle it.
1261          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1262          * as soon as we call pci_request_irq, so we have to setup our
1263          * clean_rx handler before we do so.  */
1264         igb_configure(adapter);
1265
1266         err = igb_request_irq(adapter);
1267         if (err)
1268                 goto err_req_irq;
1269
1270         /* From here on the code is the same as igb_up() */
1271         clear_bit(__IGB_DOWN, &adapter->state);
1272
1273         napi_enable(&adapter->napi);
1274         if (adapter->msix_entries)
1275                 for (i = 0; i < adapter->num_rx_queues; i++)
1276                         napi_enable(&adapter->rx_ring[i].napi);
1277
1278         igb_irq_enable(adapter);
1279
1280         /* Clear any pending interrupts. */
1281         rd32(E1000_ICR);
1282         /* Fire a link status change interrupt to start the watchdog. */
1283         wr32(E1000_ICS, E1000_ICS_LSC);
1284
1285         return 0;
1286
1287 err_req_irq:
1288         igb_release_hw_control(adapter);
1289         /* e1000_power_down_phy(adapter); */
1290         igb_free_all_rx_resources(adapter);
1291 err_setup_rx:
1292         igb_free_all_tx_resources(adapter);
1293 err_setup_tx:
1294         igb_reset(adapter);
1295
1296         return err;
1297 }
1298
1299 /**
1300  * igb_close - Disables a network interface
1301  * @netdev: network interface device structure
1302  *
1303  * Returns 0, this is not allowed to fail
1304  *
1305  * The close entry point is called when an interface is de-activated
1306  * by the OS.  The hardware is still under the driver's control, but
1307  * needs to be disabled.  A global MAC reset is issued to stop the
1308  * hardware, and all transmit and receive resources are freed.
1309  **/
1310 static int igb_close(struct net_device *netdev)
1311 {
1312         struct igb_adapter *adapter = netdev_priv(netdev);
1313
1314         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1315         igb_down(adapter);
1316
1317         igb_free_irq(adapter);
1318
1319         igb_free_all_tx_resources(adapter);
1320         igb_free_all_rx_resources(adapter);
1321
1322         /* kill manageability vlan ID if supported, but not if a vlan with
1323          * the same ID is registered on the host OS (let 8021q kill it) */
1324         if ((adapter->hw.mng_cookie.status &
1325                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1326              !(adapter->vlgrp &&
1327                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1328                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1329
1330         return 0;
1331 }
1332
1333 /**
1334  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1335  * @adapter: board private structure
1336  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1337  *
1338  * Return 0 on success, negative on failure
1339  **/
1340
1341 int igb_setup_tx_resources(struct igb_adapter *adapter,
1342                            struct igb_ring *tx_ring)
1343 {
1344         struct pci_dev *pdev = adapter->pdev;
1345         int size;
1346
1347         size = sizeof(struct igb_buffer) * tx_ring->count;
1348         tx_ring->buffer_info = vmalloc(size);
1349         if (!tx_ring->buffer_info)
1350                 goto err;
1351         memset(tx_ring->buffer_info, 0, size);
1352
1353         /* round up to nearest 4K */
1354         tx_ring->size = tx_ring->count * sizeof(struct e1000_tx_desc)
1355                         + sizeof(u32);
1356         tx_ring->size = ALIGN(tx_ring->size, 4096);
1357
1358         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1359                                              &tx_ring->dma);
1360
1361         if (!tx_ring->desc)
1362                 goto err;
1363
1364         tx_ring->adapter = adapter;
1365         tx_ring->next_to_use = 0;
1366         tx_ring->next_to_clean = 0;
1367         spin_lock_init(&tx_ring->tx_clean_lock);
1368         spin_lock_init(&tx_ring->tx_lock);
1369         return 0;
1370
1371 err:
1372         vfree(tx_ring->buffer_info);
1373         dev_err(&adapter->pdev->dev,
1374                 "Unable to allocate memory for the transmit descriptor ring\n");
1375         return -ENOMEM;
1376 }
1377
1378 /**
1379  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1380  *                                (Descriptors) for all queues
1381  * @adapter: board private structure
1382  *
1383  * Return 0 on success, negative on failure
1384  **/
1385 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1386 {
1387         int i, err = 0;
1388
1389         for (i = 0; i < adapter->num_tx_queues; i++) {
1390                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1391                 if (err) {
1392                         dev_err(&adapter->pdev->dev,
1393                                 "Allocation for Tx Queue %u failed\n", i);
1394                         for (i--; i >= 0; i--)
1395                                 igb_free_tx_resources(adapter,
1396                                                         &adapter->tx_ring[i]);
1397                         break;
1398                 }
1399         }
1400
1401         return err;
1402 }
1403
1404 /**
1405  * igb_configure_tx - Configure transmit Unit after Reset
1406  * @adapter: board private structure
1407  *
1408  * Configure the Tx unit of the MAC after a reset.
1409  **/
1410 static void igb_configure_tx(struct igb_adapter *adapter)
1411 {
1412         u64 tdba, tdwba;
1413         struct e1000_hw *hw = &adapter->hw;
1414         u32 tctl;
1415         u32 txdctl, txctrl;
1416         int i;
1417
1418         for (i = 0; i < adapter->num_tx_queues; i++) {
1419                 struct igb_ring *ring = &(adapter->tx_ring[i]);
1420
1421                 wr32(E1000_TDLEN(i),
1422                                 ring->count * sizeof(struct e1000_tx_desc));
1423                 tdba = ring->dma;
1424                 wr32(E1000_TDBAL(i),
1425                                 tdba & 0x00000000ffffffffULL);
1426                 wr32(E1000_TDBAH(i), tdba >> 32);
1427
1428                 tdwba = ring->dma + ring->count * sizeof(struct e1000_tx_desc);
1429                 tdwba |= 1; /* enable head wb */
1430                 wr32(E1000_TDWBAL(i),
1431                                 tdwba & 0x00000000ffffffffULL);
1432                 wr32(E1000_TDWBAH(i), tdwba >> 32);
1433
1434                 ring->head = E1000_TDH(i);
1435                 ring->tail = E1000_TDT(i);
1436                 writel(0, hw->hw_addr + ring->tail);
1437                 writel(0, hw->hw_addr + ring->head);
1438                 txdctl = rd32(E1000_TXDCTL(i));
1439                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1440                 wr32(E1000_TXDCTL(i), txdctl);
1441
1442                 /* Turn off Relaxed Ordering on head write-backs.  The
1443                  * writebacks MUST be delivered in order or it will
1444                  * completely screw up our bookeeping.
1445                  */
1446                 txctrl = rd32(E1000_DCA_TXCTRL(i));
1447                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1448                 wr32(E1000_DCA_TXCTRL(i), txctrl);
1449         }
1450
1451
1452
1453         /* Use the default values for the Tx Inter Packet Gap (IPG) timer */
1454
1455         /* Program the Transmit Control Register */
1456
1457         tctl = rd32(E1000_TCTL);
1458         tctl &= ~E1000_TCTL_CT;
1459         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1460                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1461
1462         igb_config_collision_dist(hw);
1463
1464         /* Setup Transmit Descriptor Settings for eop descriptor */
1465         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1466
1467         /* Enable transmits */
1468         tctl |= E1000_TCTL_EN;
1469
1470         wr32(E1000_TCTL, tctl);
1471 }
1472
1473 /**
1474  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1475  * @adapter: board private structure
1476  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1477  *
1478  * Returns 0 on success, negative on failure
1479  **/
1480
1481 int igb_setup_rx_resources(struct igb_adapter *adapter,
1482                            struct igb_ring *rx_ring)
1483 {
1484         struct pci_dev *pdev = adapter->pdev;
1485         int size, desc_len;
1486
1487         size = sizeof(struct igb_buffer) * rx_ring->count;
1488         rx_ring->buffer_info = vmalloc(size);
1489         if (!rx_ring->buffer_info)
1490                 goto err;
1491         memset(rx_ring->buffer_info, 0, size);
1492
1493         desc_len = sizeof(union e1000_adv_rx_desc);
1494
1495         /* Round up to nearest 4K */
1496         rx_ring->size = rx_ring->count * desc_len;
1497         rx_ring->size = ALIGN(rx_ring->size, 4096);
1498
1499         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1500                                              &rx_ring->dma);
1501
1502         if (!rx_ring->desc)
1503                 goto err;
1504
1505         rx_ring->next_to_clean = 0;
1506         rx_ring->next_to_use = 0;
1507         rx_ring->pending_skb = NULL;
1508
1509         rx_ring->adapter = adapter;
1510         /* FIXME: do we want to setup ring->napi->poll here? */
1511         rx_ring->napi.poll = adapter->napi.poll;
1512
1513         return 0;
1514
1515 err:
1516         vfree(rx_ring->buffer_info);
1517         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1518                 "the receive descriptor ring\n");
1519         return -ENOMEM;
1520 }
1521
1522 /**
1523  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1524  *                                (Descriptors) for all queues
1525  * @adapter: board private structure
1526  *
1527  * Return 0 on success, negative on failure
1528  **/
1529 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1530 {
1531         int i, err = 0;
1532
1533         for (i = 0; i < adapter->num_rx_queues; i++) {
1534                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1535                 if (err) {
1536                         dev_err(&adapter->pdev->dev,
1537                                 "Allocation for Rx Queue %u failed\n", i);
1538                         for (i--; i >= 0; i--)
1539                                 igb_free_rx_resources(adapter,
1540                                                         &adapter->rx_ring[i]);
1541                         break;
1542                 }
1543         }
1544
1545         return err;
1546 }
1547
1548 /**
1549  * igb_setup_rctl - configure the receive control registers
1550  * @adapter: Board private structure
1551  **/
1552 static void igb_setup_rctl(struct igb_adapter *adapter)
1553 {
1554         struct e1000_hw *hw = &adapter->hw;
1555         u32 rctl;
1556         u32 srrctl = 0;
1557         int i;
1558
1559         rctl = rd32(E1000_RCTL);
1560
1561         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1562
1563         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1564                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1565                 (adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1566
1567         /* disable the stripping of CRC because it breaks
1568          * BMC firmware connected over SMBUS
1569         rctl |= E1000_RCTL_SECRC;
1570         */
1571
1572         rctl &= ~E1000_RCTL_SBP;
1573
1574         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1575                 rctl &= ~E1000_RCTL_LPE;
1576         else
1577                 rctl |= E1000_RCTL_LPE;
1578         if (adapter->rx_buffer_len <= IGB_RXBUFFER_2048) {
1579                 /* Setup buffer sizes */
1580                 rctl &= ~E1000_RCTL_SZ_4096;
1581                 rctl |= E1000_RCTL_BSEX;
1582                 switch (adapter->rx_buffer_len) {
1583                 case IGB_RXBUFFER_256:
1584                         rctl |= E1000_RCTL_SZ_256;
1585                         rctl &= ~E1000_RCTL_BSEX;
1586                         break;
1587                 case IGB_RXBUFFER_512:
1588                         rctl |= E1000_RCTL_SZ_512;
1589                         rctl &= ~E1000_RCTL_BSEX;
1590                         break;
1591                 case IGB_RXBUFFER_1024:
1592                         rctl |= E1000_RCTL_SZ_1024;
1593                         rctl &= ~E1000_RCTL_BSEX;
1594                         break;
1595                 case IGB_RXBUFFER_2048:
1596                 default:
1597                         rctl |= E1000_RCTL_SZ_2048;
1598                         rctl &= ~E1000_RCTL_BSEX;
1599                         break;
1600                 case IGB_RXBUFFER_4096:
1601                         rctl |= E1000_RCTL_SZ_4096;
1602                         break;
1603                 case IGB_RXBUFFER_8192:
1604                         rctl |= E1000_RCTL_SZ_8192;
1605                         break;
1606                 case IGB_RXBUFFER_16384:
1607                         rctl |= E1000_RCTL_SZ_16384;
1608                         break;
1609                 }
1610         } else {
1611                 rctl &= ~E1000_RCTL_BSEX;
1612                 srrctl = adapter->rx_buffer_len >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1613         }
1614
1615         /* 82575 and greater support packet-split where the protocol
1616          * header is placed in skb->data and the packet data is
1617          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1618          * In the case of a non-split, skb->data is linearly filled,
1619          * followed by the page buffers.  Therefore, skb->data is
1620          * sized to hold the largest protocol header.
1621          */
1622         /* allocations using alloc_page take too long for regular MTU
1623          * so only enable packet split for jumbo frames */
1624         if (rctl & E1000_RCTL_LPE) {
1625                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
1626                 srrctl = adapter->rx_ps_hdr_size <<
1627                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
1628                 /* buffer size is ALWAYS one page */
1629                 srrctl |= PAGE_SIZE >> E1000_SRRCTL_BSIZEPKT_SHIFT;
1630                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1631         } else {
1632                 adapter->rx_ps_hdr_size = 0;
1633                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
1634         }
1635
1636         for (i = 0; i < adapter->num_rx_queues; i++)
1637                 wr32(E1000_SRRCTL(i), srrctl);
1638
1639         wr32(E1000_RCTL, rctl);
1640 }
1641
1642 /**
1643  * igb_configure_rx - Configure receive Unit after Reset
1644  * @adapter: board private structure
1645  *
1646  * Configure the Rx unit of the MAC after a reset.
1647  **/
1648 static void igb_configure_rx(struct igb_adapter *adapter)
1649 {
1650         u64 rdba;
1651         struct e1000_hw *hw = &adapter->hw;
1652         u32 rctl, rxcsum;
1653         u32 rxdctl;
1654         int i;
1655
1656         /* disable receives while setting up the descriptors */
1657         rctl = rd32(E1000_RCTL);
1658         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
1659         wrfl();
1660         mdelay(10);
1661
1662         if (adapter->itr_setting > 3)
1663                 wr32(E1000_ITR,
1664                                 1000000000 / (adapter->itr * 256));
1665
1666         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1667          * the Base and Length of the Rx Descriptor Ring */
1668         for (i = 0; i < adapter->num_rx_queues; i++) {
1669                 struct igb_ring *ring = &(adapter->rx_ring[i]);
1670                 rdba = ring->dma;
1671                 wr32(E1000_RDBAL(i),
1672                                 rdba & 0x00000000ffffffffULL);
1673                 wr32(E1000_RDBAH(i), rdba >> 32);
1674                 wr32(E1000_RDLEN(i),
1675                                ring->count * sizeof(union e1000_adv_rx_desc));
1676
1677                 ring->head = E1000_RDH(i);
1678                 ring->tail = E1000_RDT(i);
1679                 writel(0, hw->hw_addr + ring->tail);
1680                 writel(0, hw->hw_addr + ring->head);
1681
1682                 rxdctl = rd32(E1000_RXDCTL(i));
1683                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
1684                 rxdctl &= 0xFFF00000;
1685                 rxdctl |= IGB_RX_PTHRESH;
1686                 rxdctl |= IGB_RX_HTHRESH << 8;
1687                 rxdctl |= IGB_RX_WTHRESH << 16;
1688                 wr32(E1000_RXDCTL(i), rxdctl);
1689         }
1690
1691         if (adapter->num_rx_queues > 1) {
1692                 u32 random[10];
1693                 u32 mrqc;
1694                 u32 j, shift;
1695                 union e1000_reta {
1696                         u32 dword;
1697                         u8  bytes[4];
1698                 } reta;
1699
1700                 get_random_bytes(&random[0], 40);
1701
1702                 shift = 6;
1703                 for (j = 0; j < (32 * 4); j++) {
1704                         reta.bytes[j & 3] =
1705                                 (j % adapter->num_rx_queues) << shift;
1706                         if ((j & 3) == 3)
1707                                 writel(reta.dword,
1708                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
1709                 }
1710                 mrqc = E1000_MRQC_ENABLE_RSS_4Q;
1711
1712                 /* Fill out hash function seeds */
1713                 for (j = 0; j < 10; j++)
1714                         array_wr32(E1000_RSSRK(0), j, random[j]);
1715
1716                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1717                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1718                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
1719                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
1720                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
1721                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
1722                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
1723                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
1724
1725
1726                 wr32(E1000_MRQC, mrqc);
1727
1728                 /* Multiqueue and raw packet checksumming are mutually
1729                  * exclusive.  Note that this not the same as TCP/IP
1730                  * checksumming, which works fine. */
1731                 rxcsum = rd32(E1000_RXCSUM);
1732                 rxcsum |= E1000_RXCSUM_PCSD;
1733                 wr32(E1000_RXCSUM, rxcsum);
1734         } else {
1735                 /* Enable Receive Checksum Offload for TCP and UDP */
1736                 rxcsum = rd32(E1000_RXCSUM);
1737                 if (adapter->rx_csum) {
1738                         rxcsum |= E1000_RXCSUM_TUOFL;
1739
1740                         /* Enable IPv4 payload checksum for UDP fragments
1741                          * Must be used in conjunction with packet-split. */
1742                         if (adapter->rx_ps_hdr_size)
1743                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1744                 } else {
1745                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1746                         /* don't need to clear IPPCSE as it defaults to 0 */
1747                 }
1748                 wr32(E1000_RXCSUM, rxcsum);
1749         }
1750
1751         if (adapter->vlgrp)
1752                 wr32(E1000_RLPML,
1753                                 adapter->max_frame_size + VLAN_TAG_SIZE);
1754         else
1755                 wr32(E1000_RLPML, adapter->max_frame_size);
1756
1757         /* Enable Receives */
1758         wr32(E1000_RCTL, rctl);
1759 }
1760
1761 /**
1762  * igb_free_tx_resources - Free Tx Resources per Queue
1763  * @adapter: board private structure
1764  * @tx_ring: Tx descriptor ring for a specific queue
1765  *
1766  * Free all transmit software resources
1767  **/
1768 static void igb_free_tx_resources(struct igb_adapter *adapter,
1769                                   struct igb_ring *tx_ring)
1770 {
1771         struct pci_dev *pdev = adapter->pdev;
1772
1773         igb_clean_tx_ring(adapter, tx_ring);
1774
1775         vfree(tx_ring->buffer_info);
1776         tx_ring->buffer_info = NULL;
1777
1778         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1779
1780         tx_ring->desc = NULL;
1781 }
1782
1783 /**
1784  * igb_free_all_tx_resources - Free Tx Resources for All Queues
1785  * @adapter: board private structure
1786  *
1787  * Free all transmit software resources
1788  **/
1789 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
1790 {
1791         int i;
1792
1793         for (i = 0; i < adapter->num_tx_queues; i++)
1794                 igb_free_tx_resources(adapter, &adapter->tx_ring[i]);
1795 }
1796
1797 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
1798                                            struct igb_buffer *buffer_info)
1799 {
1800         if (buffer_info->dma) {
1801                 pci_unmap_page(adapter->pdev,
1802                                 buffer_info->dma,
1803                                 buffer_info->length,
1804                                 PCI_DMA_TODEVICE);
1805                 buffer_info->dma = 0;
1806         }
1807         if (buffer_info->skb) {
1808                 dev_kfree_skb_any(buffer_info->skb);
1809                 buffer_info->skb = NULL;
1810         }
1811         buffer_info->time_stamp = 0;
1812         /* buffer_info must be completely set up in the transmit path */
1813 }
1814
1815 /**
1816  * igb_clean_tx_ring - Free Tx Buffers
1817  * @adapter: board private structure
1818  * @tx_ring: ring to be cleaned
1819  **/
1820 static void igb_clean_tx_ring(struct igb_adapter *adapter,
1821                               struct igb_ring *tx_ring)
1822 {
1823         struct igb_buffer *buffer_info;
1824         unsigned long size;
1825         unsigned int i;
1826
1827         if (!tx_ring->buffer_info)
1828                 return;
1829         /* Free all the Tx ring sk_buffs */
1830
1831         for (i = 0; i < tx_ring->count; i++) {
1832                 buffer_info = &tx_ring->buffer_info[i];
1833                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
1834         }
1835
1836         size = sizeof(struct igb_buffer) * tx_ring->count;
1837         memset(tx_ring->buffer_info, 0, size);
1838
1839         /* Zero out the descriptor ring */
1840
1841         memset(tx_ring->desc, 0, tx_ring->size);
1842
1843         tx_ring->next_to_use = 0;
1844         tx_ring->next_to_clean = 0;
1845
1846         writel(0, adapter->hw.hw_addr + tx_ring->head);
1847         writel(0, adapter->hw.hw_addr + tx_ring->tail);
1848 }
1849
1850 /**
1851  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
1852  * @adapter: board private structure
1853  **/
1854 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
1855 {
1856         int i;
1857
1858         for (i = 0; i < adapter->num_tx_queues; i++)
1859                 igb_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1860 }
1861
1862 /**
1863  * igb_free_rx_resources - Free Rx Resources
1864  * @adapter: board private structure
1865  * @rx_ring: ring to clean the resources from
1866  *
1867  * Free all receive software resources
1868  **/
1869 static void igb_free_rx_resources(struct igb_adapter *adapter,
1870                                   struct igb_ring *rx_ring)
1871 {
1872         struct pci_dev *pdev = adapter->pdev;
1873
1874         igb_clean_rx_ring(adapter, rx_ring);
1875
1876         vfree(rx_ring->buffer_info);
1877         rx_ring->buffer_info = NULL;
1878
1879         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1880
1881         rx_ring->desc = NULL;
1882 }
1883
1884 /**
1885  * igb_free_all_rx_resources - Free Rx Resources for All Queues
1886  * @adapter: board private structure
1887  *
1888  * Free all receive software resources
1889  **/
1890 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
1891 {
1892         int i;
1893
1894         for (i = 0; i < adapter->num_rx_queues; i++)
1895                 igb_free_rx_resources(adapter, &adapter->rx_ring[i]);
1896 }
1897
1898 /**
1899  * igb_clean_rx_ring - Free Rx Buffers per Queue
1900  * @adapter: board private structure
1901  * @rx_ring: ring to free buffers from
1902  **/
1903 static void igb_clean_rx_ring(struct igb_adapter *adapter,
1904                               struct igb_ring *rx_ring)
1905 {
1906         struct igb_buffer *buffer_info;
1907         struct pci_dev *pdev = adapter->pdev;
1908         unsigned long size;
1909         unsigned int i;
1910
1911         if (!rx_ring->buffer_info)
1912                 return;
1913         /* Free all the Rx ring sk_buffs */
1914         for (i = 0; i < rx_ring->count; i++) {
1915                 buffer_info = &rx_ring->buffer_info[i];
1916                 if (buffer_info->dma) {
1917                         if (adapter->rx_ps_hdr_size)
1918                                 pci_unmap_single(pdev, buffer_info->dma,
1919                                                  adapter->rx_ps_hdr_size,
1920                                                  PCI_DMA_FROMDEVICE);
1921                         else
1922                                 pci_unmap_single(pdev, buffer_info->dma,
1923                                                  adapter->rx_buffer_len,
1924                                                  PCI_DMA_FROMDEVICE);
1925                         buffer_info->dma = 0;
1926                 }
1927
1928                 if (buffer_info->skb) {
1929                         dev_kfree_skb(buffer_info->skb);
1930                         buffer_info->skb = NULL;
1931                 }
1932                 if (buffer_info->page) {
1933                         pci_unmap_page(pdev, buffer_info->page_dma,
1934                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
1935                         put_page(buffer_info->page);
1936                         buffer_info->page = NULL;
1937                         buffer_info->page_dma = 0;
1938                 }
1939         }
1940
1941         /* there also may be some cached data from a chained receive */
1942         if (rx_ring->pending_skb) {
1943                 dev_kfree_skb(rx_ring->pending_skb);
1944                 rx_ring->pending_skb = NULL;
1945         }
1946
1947         size = sizeof(struct igb_buffer) * rx_ring->count;
1948         memset(rx_ring->buffer_info, 0, size);
1949
1950         /* Zero out the descriptor ring */
1951         memset(rx_ring->desc, 0, rx_ring->size);
1952
1953         rx_ring->next_to_clean = 0;
1954         rx_ring->next_to_use = 0;
1955
1956         writel(0, adapter->hw.hw_addr + rx_ring->head);
1957         writel(0, adapter->hw.hw_addr + rx_ring->tail);
1958 }
1959
1960 /**
1961  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
1962  * @adapter: board private structure
1963  **/
1964 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
1965 {
1966         int i;
1967
1968         for (i = 0; i < adapter->num_rx_queues; i++)
1969                 igb_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1970 }
1971
1972 /**
1973  * igb_set_mac - Change the Ethernet Address of the NIC
1974  * @netdev: network interface device structure
1975  * @p: pointer to an address structure
1976  *
1977  * Returns 0 on success, negative on failure
1978  **/
1979 static int igb_set_mac(struct net_device *netdev, void *p)
1980 {
1981         struct igb_adapter *adapter = netdev_priv(netdev);
1982         struct sockaddr *addr = p;
1983
1984         if (!is_valid_ether_addr(addr->sa_data))
1985                 return -EADDRNOTAVAIL;
1986
1987         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1988         memcpy(adapter->hw.mac.addr, addr->sa_data, netdev->addr_len);
1989
1990         adapter->hw.mac.ops.rar_set(&adapter->hw, adapter->hw.mac.addr, 0);
1991
1992         return 0;
1993 }
1994
1995 /**
1996  * igb_set_multi - Multicast and Promiscuous mode set
1997  * @netdev: network interface device structure
1998  *
1999  * The set_multi entry point is called whenever the multicast address
2000  * list or the network interface flags are updated.  This routine is
2001  * responsible for configuring the hardware for proper multicast,
2002  * promiscuous mode, and all-multi behavior.
2003  **/
2004 static void igb_set_multi(struct net_device *netdev)
2005 {
2006         struct igb_adapter *adapter = netdev_priv(netdev);
2007         struct e1000_hw *hw = &adapter->hw;
2008         struct e1000_mac_info *mac = &hw->mac;
2009         struct dev_mc_list *mc_ptr;
2010         u8  *mta_list;
2011         u32 rctl;
2012         int i;
2013
2014         /* Check for Promiscuous and All Multicast modes */
2015
2016         rctl = rd32(E1000_RCTL);
2017
2018         if (netdev->flags & IFF_PROMISC)
2019                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2020         else if (netdev->flags & IFF_ALLMULTI) {
2021                 rctl |= E1000_RCTL_MPE;
2022                 rctl &= ~E1000_RCTL_UPE;
2023         } else
2024                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2025
2026         wr32(E1000_RCTL, rctl);
2027
2028         if (!netdev->mc_count) {
2029                 /* nothing to program, so clear mc list */
2030                 igb_update_mc_addr_list(hw, NULL, 0, 1,
2031                                           mac->rar_entry_count);
2032                 return;
2033         }
2034
2035         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2036         if (!mta_list)
2037                 return;
2038
2039         /* The shared function expects a packed array of only addresses. */
2040         mc_ptr = netdev->mc_list;
2041
2042         for (i = 0; i < netdev->mc_count; i++) {
2043                 if (!mc_ptr)
2044                         break;
2045                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2046                 mc_ptr = mc_ptr->next;
2047         }
2048         igb_update_mc_addr_list(hw, mta_list, i, 1, mac->rar_entry_count);
2049         kfree(mta_list);
2050 }
2051
2052 /* Need to wait a few seconds after link up to get diagnostic information from
2053  * the phy */
2054 static void igb_update_phy_info(unsigned long data)
2055 {
2056         struct igb_adapter *adapter = (struct igb_adapter *) data;
2057         if (adapter->hw.phy.ops.get_phy_info)
2058                 adapter->hw.phy.ops.get_phy_info(&adapter->hw);
2059 }
2060
2061 /**
2062  * igb_watchdog - Timer Call-back
2063  * @data: pointer to adapter cast into an unsigned long
2064  **/
2065 static void igb_watchdog(unsigned long data)
2066 {
2067         struct igb_adapter *adapter = (struct igb_adapter *)data;
2068         /* Do the rest outside of interrupt context */
2069         schedule_work(&adapter->watchdog_task);
2070 }
2071
2072 static void igb_watchdog_task(struct work_struct *work)
2073 {
2074         struct igb_adapter *adapter = container_of(work,
2075                                         struct igb_adapter, watchdog_task);
2076         struct e1000_hw *hw = &adapter->hw;
2077
2078         struct net_device *netdev = adapter->netdev;
2079         struct igb_ring *tx_ring = adapter->tx_ring;
2080         struct e1000_mac_info *mac = &adapter->hw.mac;
2081         u32 link;
2082         s32 ret_val;
2083
2084         if ((netif_carrier_ok(netdev)) &&
2085             (rd32(E1000_STATUS) & E1000_STATUS_LU))
2086                 goto link_up;
2087
2088         ret_val = hw->mac.ops.check_for_link(&adapter->hw);
2089         if ((ret_val == E1000_ERR_PHY) &&
2090             (hw->phy.type == e1000_phy_igp_3) &&
2091             (rd32(E1000_CTRL) &
2092              E1000_PHY_CTRL_GBE_DISABLE))
2093                 dev_info(&adapter->pdev->dev,
2094                          "Gigabit has been disabled, downgrading speed\n");
2095
2096         if ((hw->phy.media_type == e1000_media_type_internal_serdes) &&
2097             !(rd32(E1000_TXCW) & E1000_TXCW_ANE))
2098                 link = mac->serdes_has_link;
2099         else
2100                 link = rd32(E1000_STATUS) &
2101                                       E1000_STATUS_LU;
2102
2103         if (link) {
2104                 if (!netif_carrier_ok(netdev)) {
2105                         u32 ctrl;
2106                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2107                                                    &adapter->link_speed,
2108                                                    &adapter->link_duplex);
2109
2110                         ctrl = rd32(E1000_CTRL);
2111                         dev_info(&adapter->pdev->dev,
2112                                  "NIC Link is Up %d Mbps %s, "
2113                                  "Flow Control: %s\n",
2114                                  adapter->link_speed,
2115                                  adapter->link_duplex == FULL_DUPLEX ?
2116                                  "Full Duplex" : "Half Duplex",
2117                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2118                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2119                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2120                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2121
2122                         /* tweak tx_queue_len according to speed/duplex and
2123                          * adjust the timeout factor */
2124                         netdev->tx_queue_len = adapter->tx_queue_len;
2125                         adapter->tx_timeout_factor = 1;
2126                         switch (adapter->link_speed) {
2127                         case SPEED_10:
2128                                 netdev->tx_queue_len = 10;
2129                                 adapter->tx_timeout_factor = 14;
2130                                 break;
2131                         case SPEED_100:
2132                                 netdev->tx_queue_len = 100;
2133                                 /* maybe add some timeout factor ? */
2134                                 break;
2135                         }
2136
2137                         netif_carrier_on(netdev);
2138                         netif_wake_queue(netdev);
2139
2140                         if (!test_bit(__IGB_DOWN, &adapter->state))
2141                                 mod_timer(&adapter->phy_info_timer,
2142                                           round_jiffies(jiffies + 2 * HZ));
2143                 }
2144         } else {
2145                 if (netif_carrier_ok(netdev)) {
2146                         adapter->link_speed = 0;
2147                         adapter->link_duplex = 0;
2148                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2149                         netif_carrier_off(netdev);
2150                         netif_stop_queue(netdev);
2151                         if (!test_bit(__IGB_DOWN, &adapter->state))
2152                                 mod_timer(&adapter->phy_info_timer,
2153                                           round_jiffies(jiffies + 2 * HZ));
2154                 }
2155         }
2156
2157 link_up:
2158         igb_update_stats(adapter);
2159
2160         mac->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2161         adapter->tpt_old = adapter->stats.tpt;
2162         mac->collision_delta = adapter->stats.colc - adapter->colc_old;
2163         adapter->colc_old = adapter->stats.colc;
2164
2165         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2166         adapter->gorc_old = adapter->stats.gorc;
2167         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2168         adapter->gotc_old = adapter->stats.gotc;
2169
2170         igb_update_adaptive(&adapter->hw);
2171
2172         if (!netif_carrier_ok(netdev)) {
2173                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2174                         /* We've lost link, so the controller stops DMA,
2175                          * but we've got queued Tx work that's never going
2176                          * to get done, so reset controller to flush Tx.
2177                          * (Do the reset outside of interrupt context). */
2178                         adapter->tx_timeout_count++;
2179                         schedule_work(&adapter->reset_task);
2180                 }
2181         }
2182
2183         /* Cause software interrupt to ensure rx ring is cleaned */
2184         wr32(E1000_ICS, E1000_ICS_RXDMT0);
2185
2186         /* Force detection of hung controller every watchdog period */
2187         tx_ring->detect_tx_hung = true;
2188
2189         /* Reset the timer */
2190         if (!test_bit(__IGB_DOWN, &adapter->state))
2191                 mod_timer(&adapter->watchdog_timer,
2192                           round_jiffies(jiffies + 2 * HZ));
2193 }
2194
2195 enum latency_range {
2196         lowest_latency = 0,
2197         low_latency = 1,
2198         bulk_latency = 2,
2199         latency_invalid = 255
2200 };
2201
2202
2203 static void igb_lower_rx_eitr(struct igb_adapter *adapter,
2204                               struct igb_ring *rx_ring)
2205 {
2206         struct e1000_hw *hw = &adapter->hw;
2207         int new_val;
2208
2209         new_val = rx_ring->itr_val / 2;
2210         if (new_val < IGB_MIN_DYN_ITR)
2211                 new_val = IGB_MIN_DYN_ITR;
2212
2213         if (new_val != rx_ring->itr_val) {
2214                 rx_ring->itr_val = new_val;
2215                 wr32(rx_ring->itr_register,
2216                                 1000000000 / (new_val * 256));
2217         }
2218 }
2219
2220 static void igb_raise_rx_eitr(struct igb_adapter *adapter,
2221                               struct igb_ring *rx_ring)
2222 {
2223         struct e1000_hw *hw = &adapter->hw;
2224         int new_val;
2225
2226         new_val = rx_ring->itr_val * 2;
2227         if (new_val > IGB_MAX_DYN_ITR)
2228                 new_val = IGB_MAX_DYN_ITR;
2229
2230         if (new_val != rx_ring->itr_val) {
2231                 rx_ring->itr_val = new_val;
2232                 wr32(rx_ring->itr_register,
2233                                 1000000000 / (new_val * 256));
2234         }
2235 }
2236
2237 /**
2238  * igb_update_itr - update the dynamic ITR value based on statistics
2239  *      Stores a new ITR value based on packets and byte
2240  *      counts during the last interrupt.  The advantage of per interrupt
2241  *      computation is faster updates and more accurate ITR for the current
2242  *      traffic pattern.  Constants in this function were computed
2243  *      based on theoretical maximum wire speed and thresholds were set based
2244  *      on testing data as well as attempting to minimize response time
2245  *      while increasing bulk throughput.
2246  *      this functionality is controlled by the InterruptThrottleRate module
2247  *      parameter (see igb_param.c)
2248  *      NOTE:  These calculations are only valid when operating in a single-
2249  *             queue environment.
2250  * @adapter: pointer to adapter
2251  * @itr_setting: current adapter->itr
2252  * @packets: the number of packets during this measurement interval
2253  * @bytes: the number of bytes during this measurement interval
2254  **/
2255 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2256                                    int packets, int bytes)
2257 {
2258         unsigned int retval = itr_setting;
2259
2260         if (packets == 0)
2261                 goto update_itr_done;
2262
2263         switch (itr_setting) {
2264         case lowest_latency:
2265                 /* handle TSO and jumbo frames */
2266                 if (bytes/packets > 8000)
2267                         retval = bulk_latency;
2268                 else if ((packets < 5) && (bytes > 512))
2269                         retval = low_latency;
2270                 break;
2271         case low_latency:  /* 50 usec aka 20000 ints/s */
2272                 if (bytes > 10000) {
2273                         /* this if handles the TSO accounting */
2274                         if (bytes/packets > 8000) {
2275                                 retval = bulk_latency;
2276                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2277                                 retval = bulk_latency;
2278                         } else if ((packets > 35)) {
2279                                 retval = lowest_latency;
2280                         }
2281                 } else if (bytes/packets > 2000) {
2282                         retval = bulk_latency;
2283                 } else if (packets <= 2 && bytes < 512) {
2284                         retval = lowest_latency;
2285                 }
2286                 break;
2287         case bulk_latency: /* 250 usec aka 4000 ints/s */
2288                 if (bytes > 25000) {
2289                         if (packets > 35)
2290                                 retval = low_latency;
2291                 } else if (bytes < 6000) {
2292                         retval = low_latency;
2293                 }
2294                 break;
2295         }
2296
2297 update_itr_done:
2298         return retval;
2299 }
2300
2301 static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
2302                         int rx_only)
2303 {
2304         u16 current_itr;
2305         u32 new_itr = adapter->itr;
2306
2307         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2308         if (adapter->link_speed != SPEED_1000) {
2309                 current_itr = 0;
2310                 new_itr = 4000;
2311                 goto set_itr_now;
2312         }
2313
2314         adapter->rx_itr = igb_update_itr(adapter,
2315                                     adapter->rx_itr,
2316                                     adapter->rx_ring->total_packets,
2317                                     adapter->rx_ring->total_bytes);
2318         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2319         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2320                 adapter->rx_itr = low_latency;
2321
2322         if (!rx_only) {
2323                 adapter->tx_itr = igb_update_itr(adapter,
2324                                             adapter->tx_itr,
2325                                             adapter->tx_ring->total_packets,
2326                                             adapter->tx_ring->total_bytes);
2327                 /* conservative mode (itr 3) eliminates the
2328                  * lowest_latency setting */
2329                 if (adapter->itr_setting == 3 &&
2330                     adapter->tx_itr == lowest_latency)
2331                         adapter->tx_itr = low_latency;
2332
2333                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2334         } else {
2335                 current_itr = adapter->rx_itr;
2336         }
2337
2338         switch (current_itr) {
2339         /* counts and packets in update_itr are dependent on these numbers */
2340         case lowest_latency:
2341                 new_itr = 70000;
2342                 break;
2343         case low_latency:
2344                 new_itr = 20000; /* aka hwitr = ~200 */
2345                 break;
2346         case bulk_latency:
2347                 new_itr = 4000;
2348                 break;
2349         default:
2350                 break;
2351         }
2352
2353 set_itr_now:
2354         if (new_itr != adapter->itr) {
2355                 /* this attempts to bias the interrupt rate towards Bulk
2356                  * by adding intermediate steps when interrupt rate is
2357                  * increasing */
2358                 new_itr = new_itr > adapter->itr ?
2359                              min(adapter->itr + (new_itr >> 2), new_itr) :
2360                              new_itr;
2361                 /* Don't write the value here; it resets the adapter's
2362                  * internal timer, and causes us to delay far longer than
2363                  * we should between interrupts.  Instead, we write the ITR
2364                  * value at the beginning of the next interrupt so the timing
2365                  * ends up being correct.
2366                  */
2367                 adapter->itr = new_itr;
2368                 adapter->set_itr = 1;
2369         }
2370
2371         return;
2372 }
2373
2374
2375 #define IGB_TX_FLAGS_CSUM               0x00000001
2376 #define IGB_TX_FLAGS_VLAN               0x00000002
2377 #define IGB_TX_FLAGS_TSO                0x00000004
2378 #define IGB_TX_FLAGS_IPV4               0x00000008
2379 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2380 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2381
2382 static inline int igb_tso_adv(struct igb_adapter *adapter,
2383                               struct igb_ring *tx_ring,
2384                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2385 {
2386         struct e1000_adv_tx_context_desc *context_desc;
2387         unsigned int i;
2388         int err;
2389         struct igb_buffer *buffer_info;
2390         u32 info = 0, tu_cmd = 0;
2391         u32 mss_l4len_idx, l4len;
2392         *hdr_len = 0;
2393
2394         if (skb_header_cloned(skb)) {
2395                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2396                 if (err)
2397                         return err;
2398         }
2399
2400         l4len = tcp_hdrlen(skb);
2401         *hdr_len += l4len;
2402
2403         if (skb->protocol == htons(ETH_P_IP)) {
2404                 struct iphdr *iph = ip_hdr(skb);
2405                 iph->tot_len = 0;
2406                 iph->check = 0;
2407                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2408                                                          iph->daddr, 0,
2409                                                          IPPROTO_TCP,
2410                                                          0);
2411         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2412                 ipv6_hdr(skb)->payload_len = 0;
2413                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2414                                                        &ipv6_hdr(skb)->daddr,
2415                                                        0, IPPROTO_TCP, 0);
2416         }
2417
2418         i = tx_ring->next_to_use;
2419
2420         buffer_info = &tx_ring->buffer_info[i];
2421         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2422         /* VLAN MACLEN IPLEN */
2423         if (tx_flags & IGB_TX_FLAGS_VLAN)
2424                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2425         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2426         *hdr_len += skb_network_offset(skb);
2427         info |= skb_network_header_len(skb);
2428         *hdr_len += skb_network_header_len(skb);
2429         context_desc->vlan_macip_lens = cpu_to_le32(info);
2430
2431         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2432         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2433
2434         if (skb->protocol == htons(ETH_P_IP))
2435                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2436         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2437
2438         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2439
2440         /* MSS L4LEN IDX */
2441         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2442         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2443
2444         /* Context index must be unique per ring.  Luckily, so is the interrupt
2445          * mask value. */
2446         mss_l4len_idx |= tx_ring->eims_value >> 4;
2447
2448         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2449         context_desc->seqnum_seed = 0;
2450
2451         buffer_info->time_stamp = jiffies;
2452         buffer_info->dma = 0;
2453         i++;
2454         if (i == tx_ring->count)
2455                 i = 0;
2456
2457         tx_ring->next_to_use = i;
2458
2459         return true;
2460 }
2461
2462 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
2463                                         struct igb_ring *tx_ring,
2464                                         struct sk_buff *skb, u32 tx_flags)
2465 {
2466         struct e1000_adv_tx_context_desc *context_desc;
2467         unsigned int i;
2468         struct igb_buffer *buffer_info;
2469         u32 info = 0, tu_cmd = 0;
2470
2471         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
2472             (tx_flags & IGB_TX_FLAGS_VLAN)) {
2473                 i = tx_ring->next_to_use;
2474                 buffer_info = &tx_ring->buffer_info[i];
2475                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2476
2477                 if (tx_flags & IGB_TX_FLAGS_VLAN)
2478                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2479                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2480                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2481                         info |= skb_network_header_len(skb);
2482
2483                 context_desc->vlan_macip_lens = cpu_to_le32(info);
2484
2485                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2486
2487                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2488                         switch (skb->protocol) {
2489                         case __constant_htons(ETH_P_IP):
2490                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2491                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2492                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2493                                 break;
2494                         case __constant_htons(ETH_P_IPV6):
2495                                 /* XXX what about other V6 headers?? */
2496                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2497                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2498                                 break;
2499                         default:
2500                                 if (unlikely(net_ratelimit()))
2501                                         dev_warn(&adapter->pdev->dev,
2502                                             "partial checksum but proto=%x!\n",
2503                                             skb->protocol);
2504                                 break;
2505                         }
2506                 }
2507
2508                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2509                 context_desc->seqnum_seed = 0;
2510                 context_desc->mss_l4len_idx =
2511                                           cpu_to_le32(tx_ring->eims_value >> 4);
2512
2513                 buffer_info->time_stamp = jiffies;
2514                 buffer_info->dma = 0;
2515
2516                 i++;
2517                 if (i == tx_ring->count)
2518                         i = 0;
2519                 tx_ring->next_to_use = i;
2520
2521                 return true;
2522         }
2523
2524
2525         return false;
2526 }
2527
2528 #define IGB_MAX_TXD_PWR 16
2529 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
2530
2531 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
2532                                  struct igb_ring *tx_ring,
2533                                  struct sk_buff *skb)
2534 {
2535         struct igb_buffer *buffer_info;
2536         unsigned int len = skb_headlen(skb);
2537         unsigned int count = 0, i;
2538         unsigned int f;
2539
2540         i = tx_ring->next_to_use;
2541
2542         buffer_info = &tx_ring->buffer_info[i];
2543         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2544         buffer_info->length = len;
2545         /* set time_stamp *before* dma to help avoid a possible race */
2546         buffer_info->time_stamp = jiffies;
2547         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
2548                                           PCI_DMA_TODEVICE);
2549         count++;
2550         i++;
2551         if (i == tx_ring->count)
2552                 i = 0;
2553
2554         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
2555                 struct skb_frag_struct *frag;
2556
2557                 frag = &skb_shinfo(skb)->frags[f];
2558                 len = frag->size;
2559
2560                 buffer_info = &tx_ring->buffer_info[i];
2561                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
2562                 buffer_info->length = len;
2563                 buffer_info->time_stamp = jiffies;
2564                 buffer_info->dma = pci_map_page(adapter->pdev,
2565                                                 frag->page,
2566                                                 frag->page_offset,
2567                                                 len,
2568                                                 PCI_DMA_TODEVICE);
2569
2570                 count++;
2571                 i++;
2572                 if (i == tx_ring->count)
2573                         i = 0;
2574         }
2575
2576         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2577         tx_ring->buffer_info[i].skb = skb;
2578
2579         return count;
2580 }
2581
2582 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
2583                                     struct igb_ring *tx_ring,
2584                                     int tx_flags, int count, u32 paylen,
2585                                     u8 hdr_len)
2586 {
2587         union e1000_adv_tx_desc *tx_desc = NULL;
2588         struct igb_buffer *buffer_info;
2589         u32 olinfo_status = 0, cmd_type_len;
2590         unsigned int i;
2591
2592         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
2593                         E1000_ADVTXD_DCMD_DEXT);
2594
2595         if (tx_flags & IGB_TX_FLAGS_VLAN)
2596                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
2597
2598         if (tx_flags & IGB_TX_FLAGS_TSO) {
2599                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
2600
2601                 /* insert tcp checksum */
2602                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2603
2604                 /* insert ip checksum */
2605                 if (tx_flags & IGB_TX_FLAGS_IPV4)
2606                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
2607
2608         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
2609                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
2610         }
2611
2612         if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
2613                         IGB_TX_FLAGS_VLAN))
2614                 olinfo_status |= tx_ring->eims_value >> 4;
2615
2616         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
2617
2618         i = tx_ring->next_to_use;
2619         while (count--) {
2620                 buffer_info = &tx_ring->buffer_info[i];
2621                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
2622                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
2623                 tx_desc->read.cmd_type_len =
2624                         cpu_to_le32(cmd_type_len | buffer_info->length);
2625                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2626                 i++;
2627                 if (i == tx_ring->count)
2628                         i = 0;
2629         }
2630
2631         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
2632         /* Force memory writes to complete before letting h/w
2633          * know there are new descriptors to fetch.  (Only
2634          * applicable for weak-ordered memory model archs,
2635          * such as IA-64). */
2636         wmb();
2637
2638         tx_ring->next_to_use = i;
2639         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2640         /* we need this if more than one processor can write to our tail
2641          * at a time, it syncronizes IO on IA64/Altix systems */
2642         mmiowb();
2643 }
2644
2645 static int __igb_maybe_stop_tx(struct net_device *netdev,
2646                                struct igb_ring *tx_ring, int size)
2647 {
2648         struct igb_adapter *adapter = netdev_priv(netdev);
2649
2650         netif_stop_queue(netdev);
2651         /* Herbert's original patch had:
2652          *  smp_mb__after_netif_stop_queue();
2653          * but since that doesn't exist yet, just open code it. */
2654         smp_mb();
2655
2656         /* We need to check again in a case another CPU has just
2657          * made room available. */
2658         if (IGB_DESC_UNUSED(tx_ring) < size)
2659                 return -EBUSY;
2660
2661         /* A reprieve! */
2662         netif_start_queue(netdev);
2663         ++adapter->restart_queue;
2664         return 0;
2665 }
2666
2667 static int igb_maybe_stop_tx(struct net_device *netdev,
2668                              struct igb_ring *tx_ring, int size)
2669 {
2670         if (IGB_DESC_UNUSED(tx_ring) >= size)
2671                 return 0;
2672         return __igb_maybe_stop_tx(netdev, tx_ring, size);
2673 }
2674
2675 #define TXD_USE_COUNT(S) (((S) >> (IGB_MAX_TXD_PWR)) + 1)
2676
2677 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
2678                                    struct net_device *netdev,
2679                                    struct igb_ring *tx_ring)
2680 {
2681         struct igb_adapter *adapter = netdev_priv(netdev);
2682         unsigned int tx_flags = 0;
2683         unsigned int len;
2684         unsigned long irq_flags;
2685         u8 hdr_len = 0;
2686         int tso = 0;
2687
2688         len = skb_headlen(skb);
2689
2690         if (test_bit(__IGB_DOWN, &adapter->state)) {
2691                 dev_kfree_skb_any(skb);
2692                 return NETDEV_TX_OK;
2693         }
2694
2695         if (skb->len <= 0) {
2696                 dev_kfree_skb_any(skb);
2697                 return NETDEV_TX_OK;
2698         }
2699
2700         if (!spin_trylock_irqsave(&tx_ring->tx_lock, irq_flags))
2701                 /* Collision - tell upper layer to requeue */
2702                 return NETDEV_TX_LOCKED;
2703
2704         /* need: 1 descriptor per page,
2705          *       + 2 desc gap to keep tail from touching head,
2706          *       + 1 desc for skb->data,
2707          *       + 1 desc for context descriptor,
2708          * otherwise try next time */
2709         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
2710                 /* this is a hard error */
2711                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2712                 return NETDEV_TX_BUSY;
2713         }
2714
2715         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
2716                 tx_flags |= IGB_TX_FLAGS_VLAN;
2717                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
2718         }
2719
2720         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
2721                                               &hdr_len) : 0;
2722
2723         if (tso < 0) {
2724                 dev_kfree_skb_any(skb);
2725                 spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2726                 return NETDEV_TX_OK;
2727         }
2728
2729         if (tso)
2730                 tx_flags |= IGB_TX_FLAGS_TSO;
2731         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags))
2732                         if (skb->ip_summed == CHECKSUM_PARTIAL)
2733                                 tx_flags |= IGB_TX_FLAGS_CSUM;
2734
2735         if (skb->protocol == htons(ETH_P_IP))
2736                 tx_flags |= IGB_TX_FLAGS_IPV4;
2737
2738         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
2739                          igb_tx_map_adv(adapter, tx_ring, skb),
2740                          skb->len, hdr_len);
2741
2742         netdev->trans_start = jiffies;
2743
2744         /* Make sure there is space in the ring for the next send. */
2745         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
2746
2747         spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
2748         return NETDEV_TX_OK;
2749 }
2750
2751 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
2752 {
2753         struct igb_adapter *adapter = netdev_priv(netdev);
2754         struct igb_ring *tx_ring = &adapter->tx_ring[0];
2755
2756         /* This goes back to the question of how to logically map a tx queue
2757          * to a flow.  Right now, performance is impacted slightly negatively
2758          * if using multiple tx queues.  If the stack breaks away from a
2759          * single qdisc implementation, we can look at this again. */
2760         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
2761 }
2762
2763 /**
2764  * igb_tx_timeout - Respond to a Tx Hang
2765  * @netdev: network interface device structure
2766  **/
2767 static void igb_tx_timeout(struct net_device *netdev)
2768 {
2769         struct igb_adapter *adapter = netdev_priv(netdev);
2770         struct e1000_hw *hw = &adapter->hw;
2771
2772         /* Do the reset outside of interrupt context */
2773         adapter->tx_timeout_count++;
2774         schedule_work(&adapter->reset_task);
2775         wr32(E1000_EICS, adapter->eims_enable_mask &
2776                 ~(E1000_EIMS_TCP_TIMER | E1000_EIMS_OTHER));
2777 }
2778
2779 static void igb_reset_task(struct work_struct *work)
2780 {
2781         struct igb_adapter *adapter;
2782         adapter = container_of(work, struct igb_adapter, reset_task);
2783
2784         igb_reinit_locked(adapter);
2785 }
2786
2787 /**
2788  * igb_get_stats - Get System Network Statistics
2789  * @netdev: network interface device structure
2790  *
2791  * Returns the address of the device statistics structure.
2792  * The statistics are actually updated from the timer callback.
2793  **/
2794 static struct net_device_stats *
2795 igb_get_stats(struct net_device *netdev)
2796 {
2797         struct igb_adapter *adapter = netdev_priv(netdev);
2798
2799         /* only return the current stats */
2800         return &adapter->net_stats;
2801 }
2802
2803 /**
2804  * igb_change_mtu - Change the Maximum Transfer Unit
2805  * @netdev: network interface device structure
2806  * @new_mtu: new value for maximum frame size
2807  *
2808  * Returns 0 on success, negative on failure
2809  **/
2810 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
2811 {
2812         struct igb_adapter *adapter = netdev_priv(netdev);
2813         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2814
2815         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
2816             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
2817                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
2818                 return -EINVAL;
2819         }
2820
2821 #define MAX_STD_JUMBO_FRAME_SIZE 9234
2822         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
2823                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
2824                 return -EINVAL;
2825         }
2826
2827         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
2828                 msleep(1);
2829         /* igb_down has a dependency on max_frame_size */
2830         adapter->max_frame_size = max_frame;
2831         if (netif_running(netdev))
2832                 igb_down(adapter);
2833
2834         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
2835          * means we reserve 2 more, this pushes us to allocate from the next
2836          * larger slab size.
2837          * i.e. RXBUFFER_2048 --> size-4096 slab
2838          */
2839
2840         if (max_frame <= IGB_RXBUFFER_256)
2841                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
2842         else if (max_frame <= IGB_RXBUFFER_512)
2843                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
2844         else if (max_frame <= IGB_RXBUFFER_1024)
2845                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
2846         else if (max_frame <= IGB_RXBUFFER_2048)
2847                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
2848         else
2849                 adapter->rx_buffer_len = IGB_RXBUFFER_4096;
2850         /* adjust allocation if LPE protects us, and we aren't using SBP */
2851         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
2852              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
2853                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
2854
2855         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
2856                  netdev->mtu, new_mtu);
2857         netdev->mtu = new_mtu;
2858
2859         if (netif_running(netdev))
2860                 igb_up(adapter);
2861         else
2862                 igb_reset(adapter);
2863
2864         clear_bit(__IGB_RESETTING, &adapter->state);
2865
2866         return 0;
2867 }
2868
2869 /**
2870  * igb_update_stats - Update the board statistics counters
2871  * @adapter: board private structure
2872  **/
2873
2874 void igb_update_stats(struct igb_adapter *adapter)
2875 {
2876         struct e1000_hw *hw = &adapter->hw;
2877         struct pci_dev *pdev = adapter->pdev;
2878         u16 phy_tmp;
2879
2880 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
2881
2882         /*
2883          * Prevent stats update while adapter is being reset, or if the pci
2884          * connection is down.
2885          */
2886         if (adapter->link_speed == 0)
2887                 return;
2888         if (pci_channel_offline(pdev))
2889                 return;
2890
2891         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
2892         adapter->stats.gprc += rd32(E1000_GPRC);
2893         adapter->stats.gorc += rd32(E1000_GORCL);
2894         rd32(E1000_GORCH); /* clear GORCL */
2895         adapter->stats.bprc += rd32(E1000_BPRC);
2896         adapter->stats.mprc += rd32(E1000_MPRC);
2897         adapter->stats.roc += rd32(E1000_ROC);
2898
2899         adapter->stats.prc64 += rd32(E1000_PRC64);
2900         adapter->stats.prc127 += rd32(E1000_PRC127);
2901         adapter->stats.prc255 += rd32(E1000_PRC255);
2902         adapter->stats.prc511 += rd32(E1000_PRC511);
2903         adapter->stats.prc1023 += rd32(E1000_PRC1023);
2904         adapter->stats.prc1522 += rd32(E1000_PRC1522);
2905         adapter->stats.symerrs += rd32(E1000_SYMERRS);
2906         adapter->stats.sec += rd32(E1000_SEC);
2907
2908         adapter->stats.mpc += rd32(E1000_MPC);
2909         adapter->stats.scc += rd32(E1000_SCC);
2910         adapter->stats.ecol += rd32(E1000_ECOL);
2911         adapter->stats.mcc += rd32(E1000_MCC);
2912         adapter->stats.latecol += rd32(E1000_LATECOL);
2913         adapter->stats.dc += rd32(E1000_DC);
2914         adapter->stats.rlec += rd32(E1000_RLEC);
2915         adapter->stats.xonrxc += rd32(E1000_XONRXC);
2916         adapter->stats.xontxc += rd32(E1000_XONTXC);
2917         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
2918         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
2919         adapter->stats.fcruc += rd32(E1000_FCRUC);
2920         adapter->stats.gptc += rd32(E1000_GPTC);
2921         adapter->stats.gotc += rd32(E1000_GOTCL);
2922         rd32(E1000_GOTCH); /* clear GOTCL */
2923         adapter->stats.rnbc += rd32(E1000_RNBC);
2924         adapter->stats.ruc += rd32(E1000_RUC);
2925         adapter->stats.rfc += rd32(E1000_RFC);
2926         adapter->stats.rjc += rd32(E1000_RJC);
2927         adapter->stats.tor += rd32(E1000_TORH);
2928         adapter->stats.tot += rd32(E1000_TOTH);
2929         adapter->stats.tpr += rd32(E1000_TPR);
2930
2931         adapter->stats.ptc64 += rd32(E1000_PTC64);
2932         adapter->stats.ptc127 += rd32(E1000_PTC127);
2933         adapter->stats.ptc255 += rd32(E1000_PTC255);
2934         adapter->stats.ptc511 += rd32(E1000_PTC511);
2935         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
2936         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
2937
2938         adapter->stats.mptc += rd32(E1000_MPTC);
2939         adapter->stats.bptc += rd32(E1000_BPTC);
2940
2941         /* used for adaptive IFS */
2942
2943         hw->mac.tx_packet_delta = rd32(E1000_TPT);
2944         adapter->stats.tpt += hw->mac.tx_packet_delta;
2945         hw->mac.collision_delta = rd32(E1000_COLC);
2946         adapter->stats.colc += hw->mac.collision_delta;
2947
2948         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
2949         adapter->stats.rxerrc += rd32(E1000_RXERRC);
2950         adapter->stats.tncrs += rd32(E1000_TNCRS);
2951         adapter->stats.tsctc += rd32(E1000_TSCTC);
2952         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
2953
2954         adapter->stats.iac += rd32(E1000_IAC);
2955         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
2956         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
2957         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
2958         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
2959         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
2960         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
2961         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
2962         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
2963
2964         /* Fill out the OS statistics structure */
2965         adapter->net_stats.multicast = adapter->stats.mprc;
2966         adapter->net_stats.collisions = adapter->stats.colc;
2967
2968         /* Rx Errors */
2969
2970         /* RLEC on some newer hardware can be incorrect so build
2971         * our own version based on RUC and ROC */
2972         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
2973                 adapter->stats.crcerrs + adapter->stats.algnerrc +
2974                 adapter->stats.ruc + adapter->stats.roc +
2975                 adapter->stats.cexterr;
2976         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
2977                                               adapter->stats.roc;
2978         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
2979         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
2980         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
2981
2982         /* Tx Errors */
2983         adapter->net_stats.tx_errors = adapter->stats.ecol +
2984                                        adapter->stats.latecol;
2985         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
2986         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
2987         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
2988
2989         /* Tx Dropped needs to be maintained elsewhere */
2990
2991         /* Phy Stats */
2992         if (hw->phy.media_type == e1000_media_type_copper) {
2993                 if ((adapter->link_speed == SPEED_1000) &&
2994                    (!hw->phy.ops.read_phy_reg(hw, PHY_1000T_STATUS,
2995                                               &phy_tmp))) {
2996                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
2997                         adapter->phy_stats.idle_errors += phy_tmp;
2998                 }
2999         }
3000
3001         /* Management Stats */
3002         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3003         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3004         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3005 }
3006
3007
3008 static irqreturn_t igb_msix_other(int irq, void *data)
3009 {
3010         struct net_device *netdev = data;
3011         struct igb_adapter *adapter = netdev_priv(netdev);
3012         struct e1000_hw *hw = &adapter->hw;
3013         u32 eicr;
3014         /* disable interrupts from the "other" bit, avoid re-entry */
3015         wr32(E1000_EIMC, E1000_EIMS_OTHER);
3016
3017         eicr = rd32(E1000_EICR);
3018
3019         if (eicr & E1000_EIMS_OTHER) {
3020                 u32 icr = rd32(E1000_ICR);
3021                 /* reading ICR causes bit 31 of EICR to be cleared */
3022                 if (!(icr & E1000_ICR_LSC))
3023                         goto no_link_interrupt;
3024                 hw->mac.get_link_status = 1;
3025                 /* guard against interrupt when we're going down */
3026                 if (!test_bit(__IGB_DOWN, &adapter->state))
3027                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3028         }
3029
3030 no_link_interrupt:
3031         wr32(E1000_IMS, E1000_IMS_LSC);
3032         wr32(E1000_EIMS, E1000_EIMS_OTHER);
3033
3034         return IRQ_HANDLED;
3035 }
3036
3037 static irqreturn_t igb_msix_tx(int irq, void *data)
3038 {
3039         struct igb_ring *tx_ring = data;
3040         struct igb_adapter *adapter = tx_ring->adapter;
3041         struct e1000_hw *hw = &adapter->hw;
3042
3043         if (!tx_ring->itr_val)
3044                 wr32(E1000_EIMC, tx_ring->eims_value);
3045
3046         tx_ring->total_bytes = 0;
3047         tx_ring->total_packets = 0;
3048         if (!igb_clean_tx_irq(adapter, tx_ring))
3049                 /* Ring was not completely cleaned, so fire another interrupt */
3050                 wr32(E1000_EICS, tx_ring->eims_value);
3051
3052         if (!tx_ring->itr_val)
3053                 wr32(E1000_EIMS, tx_ring->eims_value);
3054         return IRQ_HANDLED;
3055 }
3056
3057 static irqreturn_t igb_msix_rx(int irq, void *data)
3058 {
3059         struct igb_ring *rx_ring = data;
3060         struct igb_adapter *adapter = rx_ring->adapter;
3061         struct e1000_hw *hw = &adapter->hw;
3062
3063         if (!rx_ring->itr_val)
3064                 wr32(E1000_EIMC, rx_ring->eims_value);
3065
3066         if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi)) {
3067                 rx_ring->total_bytes = 0;
3068                 rx_ring->total_packets = 0;
3069                 rx_ring->no_itr_adjust = 0;
3070                 __netif_rx_schedule(adapter->netdev, &rx_ring->napi);
3071         } else {
3072                 if (!rx_ring->no_itr_adjust) {
3073                         igb_lower_rx_eitr(adapter, rx_ring);
3074                         rx_ring->no_itr_adjust = 1;
3075                 }
3076         }
3077
3078         return IRQ_HANDLED;
3079 }
3080
3081
3082 /**
3083  * igb_intr_msi - Interrupt Handler
3084  * @irq: interrupt number
3085  * @data: pointer to a network interface device structure
3086  **/
3087 static irqreturn_t igb_intr_msi(int irq, void *data)
3088 {
3089         struct net_device *netdev = data;
3090         struct igb_adapter *adapter = netdev_priv(netdev);
3091         struct napi_struct *napi = &adapter->napi;
3092         struct e1000_hw *hw = &adapter->hw;
3093         /* read ICR disables interrupts using IAM */
3094         u32 icr = rd32(E1000_ICR);
3095
3096         /* Write the ITR value calculated at the end of the
3097          * previous interrupt.
3098          */
3099         if (adapter->set_itr) {
3100                 wr32(E1000_ITR,
3101                         1000000000 / (adapter->itr * 256));
3102                 adapter->set_itr = 0;
3103         }
3104
3105         /* read ICR disables interrupts using IAM */
3106         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3107                 hw->mac.get_link_status = 1;
3108                 if (!test_bit(__IGB_DOWN, &adapter->state))
3109                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3110         }
3111
3112         if (netif_rx_schedule_prep(netdev, napi)) {
3113                 adapter->tx_ring->total_bytes = 0;
3114                 adapter->tx_ring->total_packets = 0;
3115                 adapter->rx_ring->total_bytes = 0;
3116                 adapter->rx_ring->total_packets = 0;
3117                 __netif_rx_schedule(netdev, napi);
3118         }
3119
3120         return IRQ_HANDLED;
3121 }
3122
3123 /**
3124  * igb_intr - Interrupt Handler
3125  * @irq: interrupt number
3126  * @data: pointer to a network interface device structure
3127  **/
3128 static irqreturn_t igb_intr(int irq, void *data)
3129 {
3130         struct net_device *netdev = data;
3131         struct igb_adapter *adapter = netdev_priv(netdev);
3132         struct napi_struct *napi = &adapter->napi;
3133         struct e1000_hw *hw = &adapter->hw;
3134         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
3135          * need for the IMC write */
3136         u32 icr = rd32(E1000_ICR);
3137         u32 eicr = 0;
3138         if (!icr)
3139                 return IRQ_NONE;  /* Not our interrupt */
3140
3141         /* Write the ITR value calculated at the end of the
3142          * previous interrupt.
3143          */
3144         if (adapter->set_itr) {
3145                 wr32(E1000_ITR,
3146                         1000000000 / (adapter->itr * 256));
3147                 adapter->set_itr = 0;
3148         }
3149
3150         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
3151          * not set, then the adapter didn't send an interrupt */
3152         if (!(icr & E1000_ICR_INT_ASSERTED))
3153                 return IRQ_NONE;
3154
3155         eicr = rd32(E1000_EICR);
3156
3157         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
3158                 hw->mac.get_link_status = 1;
3159                 /* guard against interrupt when we're going down */
3160                 if (!test_bit(__IGB_DOWN, &adapter->state))
3161                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3162         }
3163
3164         if (netif_rx_schedule_prep(netdev, napi)) {
3165                 adapter->tx_ring->total_bytes = 0;
3166                 adapter->rx_ring->total_bytes = 0;
3167                 adapter->tx_ring->total_packets = 0;
3168                 adapter->rx_ring->total_packets = 0;
3169                 __netif_rx_schedule(netdev, napi);
3170         }
3171
3172         return IRQ_HANDLED;
3173 }
3174
3175 /**
3176  * igb_clean - NAPI Rx polling callback
3177  * @adapter: board private structure
3178  **/
3179 static int igb_clean(struct napi_struct *napi, int budget)
3180 {
3181         struct igb_adapter *adapter = container_of(napi, struct igb_adapter,
3182                                                    napi);
3183         struct net_device *netdev = adapter->netdev;
3184         int tx_clean_complete = 1, work_done = 0;
3185         int i;
3186
3187         /* Must NOT use netdev_priv macro here. */
3188         adapter = netdev->priv;
3189
3190         /* Keep link state information with original netdev */
3191         if (!netif_carrier_ok(netdev))
3192                 goto quit_polling;
3193
3194         /* igb_clean is called per-cpu.  This lock protects tx_ring[i] from
3195          * being cleaned by multiple cpus simultaneously.  A failure obtaining
3196          * the lock means tx_ring[i] is currently being cleaned anyway. */
3197         for (i = 0; i < adapter->num_tx_queues; i++) {
3198                 if (spin_trylock(&adapter->tx_ring[i].tx_clean_lock)) {
3199                         tx_clean_complete &= igb_clean_tx_irq(adapter,
3200                                                         &adapter->tx_ring[i]);
3201                         spin_unlock(&adapter->tx_ring[i].tx_clean_lock);
3202                 }
3203         }
3204
3205         for (i = 0; i < adapter->num_rx_queues; i++)
3206                 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i], &work_done,
3207                                      adapter->rx_ring[i].napi.weight);
3208
3209         /* If no Tx and not enough Rx work done, exit the polling mode */
3210         if ((tx_clean_complete && (work_done < budget)) ||
3211             !netif_running(netdev)) {
3212 quit_polling:
3213                 if (adapter->itr_setting & 3)
3214                         igb_set_itr(adapter, E1000_ITR, false);
3215                 netif_rx_complete(netdev, napi);
3216                 if (!test_bit(__IGB_DOWN, &adapter->state))
3217                         igb_irq_enable(adapter);
3218                 return 0;
3219         }
3220
3221         return 1;
3222 }
3223
3224 static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
3225 {
3226         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
3227         struct igb_adapter *adapter = rx_ring->adapter;
3228         struct e1000_hw *hw = &adapter->hw;
3229         struct net_device *netdev = adapter->netdev;
3230         int work_done = 0;
3231
3232         /* Keep link state information with original netdev */
3233         if (!netif_carrier_ok(netdev))
3234                 goto quit_polling;
3235
3236         igb_clean_rx_irq_adv(adapter, rx_ring, &work_done, budget);
3237
3238
3239         /* If not enough Rx work done, exit the polling mode */
3240         if ((work_done == 0) || !netif_running(netdev)) {
3241 quit_polling:
3242                 netif_rx_complete(netdev, napi);
3243
3244                 wr32(E1000_EIMS, rx_ring->eims_value);
3245                 if ((adapter->itr_setting & 3) && !rx_ring->no_itr_adjust &&
3246                     (rx_ring->total_packets > IGB_DYN_ITR_PACKET_THRESHOLD)) {
3247                         int mean_size = rx_ring->total_bytes /
3248                                         rx_ring->total_packets;
3249                         if (mean_size < IGB_DYN_ITR_LENGTH_LOW)
3250                                 igb_raise_rx_eitr(adapter, rx_ring);
3251                         else if (mean_size > IGB_DYN_ITR_LENGTH_HIGH)
3252                                 igb_lower_rx_eitr(adapter, rx_ring);
3253                 }
3254                 return 0;
3255         }
3256
3257         return 1;
3258 }
3259
3260 static inline u32 get_head(struct igb_ring *tx_ring)
3261 {
3262         void *end = (struct e1000_tx_desc *)tx_ring->desc + tx_ring->count;
3263         return le32_to_cpu(*(volatile __le32 *)end);
3264 }
3265
3266 /**
3267  * igb_clean_tx_irq - Reclaim resources after transmit completes
3268  * @adapter: board private structure
3269  * returns true if ring is completely cleaned
3270  **/
3271 static bool igb_clean_tx_irq(struct igb_adapter *adapter,
3272                                   struct igb_ring *tx_ring)
3273 {
3274         struct net_device *netdev = adapter->netdev;
3275         struct e1000_hw *hw = &adapter->hw;
3276         struct e1000_tx_desc *tx_desc;
3277         struct igb_buffer *buffer_info;
3278         struct sk_buff *skb;
3279         unsigned int i;
3280         u32 head, oldhead;
3281         unsigned int count = 0;
3282         bool cleaned = false;
3283         bool retval = true;
3284         unsigned int total_bytes = 0, total_packets = 0;
3285
3286         rmb();
3287         head = get_head(tx_ring);
3288         i = tx_ring->next_to_clean;
3289         while (1) {
3290                 while (i != head) {
3291                         cleaned = true;
3292                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3293                         buffer_info = &tx_ring->buffer_info[i];
3294                         skb = buffer_info->skb;
3295
3296                         if (skb) {
3297                                 unsigned int segs, bytecount;
3298                                 /* gso_segs is currently only valid for tcp */
3299                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
3300                                 /* multiply data chunks by size of headers */
3301                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
3302                                             skb->len;
3303                                 total_packets += segs;
3304                                 total_bytes += bytecount;
3305                         }
3306
3307                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
3308                         tx_desc->upper.data = 0;
3309
3310                         i++;
3311                         if (i == tx_ring->count)
3312                                 i = 0;
3313
3314                         count++;
3315                         if (count == IGB_MAX_TX_CLEAN) {
3316                                 retval = false;
3317                                 goto done_cleaning;
3318                         }
3319                 }
3320                 oldhead = head;
3321                 rmb();
3322                 head = get_head(tx_ring);
3323                 if (head == oldhead)
3324                         goto done_cleaning;
3325         }  /* while (1) */
3326
3327 done_cleaning:
3328         tx_ring->next_to_clean = i;
3329
3330         if (unlikely(cleaned &&
3331                      netif_carrier_ok(netdev) &&
3332                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
3333                 /* Make sure that anybody stopping the queue after this
3334                  * sees the new next_to_clean.
3335                  */
3336                 smp_mb();
3337                 if (netif_queue_stopped(netdev) &&
3338                     !(test_bit(__IGB_DOWN, &adapter->state))) {
3339                         netif_wake_queue(netdev);
3340                         ++adapter->restart_queue;
3341                 }
3342         }
3343
3344         if (tx_ring->detect_tx_hung) {
3345                 /* Detect a transmit hang in hardware, this serializes the
3346                  * check with the clearing of time_stamp and movement of i */
3347                 tx_ring->detect_tx_hung = false;
3348                 if (tx_ring->buffer_info[i].time_stamp &&
3349                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
3350                                (adapter->tx_timeout_factor * HZ))
3351                     && !(rd32(E1000_STATUS) &
3352                          E1000_STATUS_TXOFF)) {
3353
3354                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3355                         /* detected Tx unit hang */
3356                         dev_err(&adapter->pdev->dev,
3357                                 "Detected Tx Unit Hang\n"
3358                                 "  Tx Queue             <%lu>\n"
3359                                 "  TDH                  <%x>\n"
3360                                 "  TDT                  <%x>\n"
3361                                 "  next_to_use          <%x>\n"
3362                                 "  next_to_clean        <%x>\n"
3363                                 "  head (WB)            <%x>\n"
3364                                 "buffer_info[next_to_clean]\n"
3365                                 "  time_stamp           <%lx>\n"
3366                                 "  jiffies              <%lx>\n"
3367                                 "  desc.status          <%x>\n",
3368                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3369                                         sizeof(struct igb_ring)),
3370                                 readl(adapter->hw.hw_addr + tx_ring->head),
3371                                 readl(adapter->hw.hw_addr + tx_ring->tail),
3372                                 tx_ring->next_to_use,
3373                                 tx_ring->next_to_clean,
3374                                 head,
3375                                 tx_ring->buffer_info[i].time_stamp,
3376                                 jiffies,
3377                                 tx_desc->upper.fields.status);
3378                         netif_stop_queue(netdev);
3379                 }
3380         }
3381         tx_ring->total_bytes += total_bytes;
3382         tx_ring->total_packets += total_packets;
3383         adapter->net_stats.tx_bytes += total_bytes;
3384         adapter->net_stats.tx_packets += total_packets;
3385         return retval;
3386 }
3387
3388
3389 /**
3390  * igb_receive_skb - helper function to handle rx indications
3391  * @adapter: board private structure
3392  * @status: descriptor status field as written by hardware
3393  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3394  * @skb: pointer to sk_buff to be indicated to stack
3395  **/
3396 static void igb_receive_skb(struct igb_adapter *adapter, u8 status, __le16 vlan,
3397                             struct sk_buff *skb)
3398 {
3399         if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
3400                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3401                                          le16_to_cpu(vlan) &
3402                                          E1000_RXD_SPC_VLAN_MASK);
3403         else
3404                 netif_receive_skb(skb);
3405 }
3406
3407
3408 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
3409                                        u32 status_err, struct sk_buff *skb)
3410 {
3411         skb->ip_summed = CHECKSUM_NONE;
3412
3413         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
3414         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
3415                 return;
3416         /* TCP/UDP checksum error bit is set */
3417         if (status_err &
3418             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
3419                 /* let the stack verify checksum errors */
3420                 adapter->hw_csum_err++;
3421                 return;
3422         }
3423         /* It must be a TCP or UDP packet with a valid checksum */
3424         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
3425                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3426
3427         adapter->hw_csum_good++;
3428 }
3429
3430 static bool igb_clean_rx_irq_adv(struct igb_adapter *adapter,
3431                                       struct igb_ring *rx_ring,
3432                                       int *work_done, int budget)
3433 {
3434         struct net_device *netdev = adapter->netdev;
3435         struct pci_dev *pdev = adapter->pdev;
3436         union e1000_adv_rx_desc *rx_desc , *next_rxd;
3437         struct igb_buffer *buffer_info , *next_buffer;
3438         struct sk_buff *skb;
3439         unsigned int i, j;
3440         u32 length, hlen, staterr;
3441         bool cleaned = false;
3442         int cleaned_count = 0;
3443         unsigned int total_bytes = 0, total_packets = 0;
3444
3445         i = rx_ring->next_to_clean;
3446         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3447         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3448
3449         while (staterr & E1000_RXD_STAT_DD) {
3450                 if (*work_done >= budget)
3451                         break;
3452                 (*work_done)++;
3453                 buffer_info = &rx_ring->buffer_info[i];
3454
3455                 /* HW will not DMA in data larger than the given buffer, even
3456                  * if it parses the (NFS, of course) header to be larger.  In
3457                  * that case, it fills the header buffer and spills the rest
3458                  * into the page.
3459                  */
3460                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
3461                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
3462                 if (hlen > adapter->rx_ps_hdr_size)
3463                         hlen = adapter->rx_ps_hdr_size;
3464
3465                 length = le16_to_cpu(rx_desc->wb.upper.length);
3466                 cleaned = true;
3467                 cleaned_count++;
3468
3469                 if (rx_ring->pending_skb != NULL) {
3470                         skb = rx_ring->pending_skb;
3471                         rx_ring->pending_skb = NULL;
3472                         j = rx_ring->pending_skb_page;
3473                 } else {
3474                         skb = buffer_info->skb;
3475                         prefetch(skb->data - NET_IP_ALIGN);
3476                         buffer_info->skb = NULL;
3477                         if (hlen) {
3478                                 pci_unmap_single(pdev, buffer_info->dma,
3479                                                  adapter->rx_ps_hdr_size +
3480                                                    NET_IP_ALIGN,
3481                                                  PCI_DMA_FROMDEVICE);
3482                                 skb_put(skb, hlen);
3483                         } else {
3484                                 pci_unmap_single(pdev, buffer_info->dma,
3485                                                  adapter->rx_buffer_len +
3486                                                    NET_IP_ALIGN,
3487                                                  PCI_DMA_FROMDEVICE);
3488                                 skb_put(skb, length);
3489                                 goto send_up;
3490                         }
3491                         j = 0;
3492                 }
3493
3494                 while (length) {
3495                         pci_unmap_page(pdev, buffer_info->page_dma,
3496                                 PAGE_SIZE, PCI_DMA_FROMDEVICE);
3497                         buffer_info->page_dma = 0;
3498                         skb_fill_page_desc(skb, j, buffer_info->page,
3499                                                 0, length);
3500                         buffer_info->page = NULL;
3501
3502                         skb->len += length;
3503                         skb->data_len += length;
3504                         skb->truesize += length;
3505                         rx_desc->wb.upper.status_error = 0;
3506                         if (staterr & E1000_RXD_STAT_EOP)
3507                                 break;
3508
3509                         j++;
3510                         cleaned_count++;
3511                         i++;
3512                         if (i == rx_ring->count)
3513                                 i = 0;
3514
3515                         buffer_info = &rx_ring->buffer_info[i];
3516                         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3517                         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3518                         length = le16_to_cpu(rx_desc->wb.upper.length);
3519                         if (!(staterr & E1000_RXD_STAT_DD)) {
3520                                 rx_ring->pending_skb = skb;
3521                                 rx_ring->pending_skb_page = j;
3522                                 goto out;
3523                         }
3524                 }
3525 send_up:
3526                 pskb_trim(skb, skb->len - 4);
3527                 i++;
3528                 if (i == rx_ring->count)
3529                         i = 0;
3530                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
3531                 prefetch(next_rxd);
3532                 next_buffer = &rx_ring->buffer_info[i];
3533
3534                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
3535                         dev_kfree_skb_irq(skb);
3536                         goto next_desc;
3537                 }
3538                 rx_ring->no_itr_adjust |= (staterr & E1000_RXD_STAT_DYNINT);
3539
3540                 total_bytes += skb->len;
3541                 total_packets++;
3542
3543                 igb_rx_checksum_adv(adapter, staterr, skb);
3544
3545                 skb->protocol = eth_type_trans(skb, netdev);
3546
3547                 igb_receive_skb(adapter, staterr, rx_desc->wb.upper.vlan, skb);
3548
3549                 netdev->last_rx = jiffies;
3550
3551 next_desc:
3552                 rx_desc->wb.upper.status_error = 0;
3553
3554                 /* return some buffers to hardware, one at a time is too slow */
3555                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
3556                         igb_alloc_rx_buffers_adv(adapter, rx_ring,
3557                                                  cleaned_count);
3558                         cleaned_count = 0;
3559                 }
3560
3561                 /* use prefetched values */
3562                 rx_desc = next_rxd;
3563                 buffer_info = next_buffer;
3564
3565                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
3566         }
3567 out:
3568         rx_ring->next_to_clean = i;
3569         cleaned_count = IGB_DESC_UNUSED(rx_ring);
3570
3571         if (cleaned_count)
3572                 igb_alloc_rx_buffers_adv(adapter, rx_ring, cleaned_count);
3573
3574         rx_ring->total_packets += total_packets;
3575         rx_ring->total_bytes += total_bytes;
3576         rx_ring->rx_stats.packets += total_packets;
3577         rx_ring->rx_stats.bytes += total_bytes;
3578         adapter->net_stats.rx_bytes += total_bytes;
3579         adapter->net_stats.rx_packets += total_packets;
3580         return cleaned;
3581 }
3582
3583
3584 /**
3585  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
3586  * @adapter: address of board private structure
3587  **/
3588 static void igb_alloc_rx_buffers_adv(struct igb_adapter *adapter,
3589                                      struct igb_ring *rx_ring,
3590                                      int cleaned_count)
3591 {
3592         struct net_device *netdev = adapter->netdev;
3593         struct pci_dev *pdev = adapter->pdev;
3594         union e1000_adv_rx_desc *rx_desc;
3595         struct igb_buffer *buffer_info;
3596         struct sk_buff *skb;
3597         unsigned int i;
3598
3599         i = rx_ring->next_to_use;
3600         buffer_info = &rx_ring->buffer_info[i];
3601
3602         while (cleaned_count--) {
3603                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
3604
3605                 if (adapter->rx_ps_hdr_size && !buffer_info->page) {
3606                         buffer_info->page = alloc_page(GFP_ATOMIC);
3607                         if (!buffer_info->page) {
3608                                 adapter->alloc_rx_buff_failed++;
3609                                 goto no_buffers;
3610                         }
3611                         buffer_info->page_dma =
3612                                 pci_map_page(pdev,
3613                                              buffer_info->page,
3614                                              0, PAGE_SIZE,
3615                                              PCI_DMA_FROMDEVICE);
3616                 }
3617
3618                 if (!buffer_info->skb) {
3619                         int bufsz;
3620
3621                         if (adapter->rx_ps_hdr_size)
3622                                 bufsz = adapter->rx_ps_hdr_size;
3623                         else
3624                                 bufsz = adapter->rx_buffer_len;
3625                         bufsz += NET_IP_ALIGN;
3626                         skb = netdev_alloc_skb(netdev, bufsz);
3627
3628                         if (!skb) {
3629                                 adapter->alloc_rx_buff_failed++;
3630                                 goto no_buffers;
3631                         }
3632
3633                         /* Make buffer alignment 2 beyond a 16 byte boundary
3634                          * this will result in a 16 byte aligned IP header after
3635                          * the 14 byte MAC header is removed
3636                          */
3637                         skb_reserve(skb, NET_IP_ALIGN);
3638
3639                         buffer_info->skb = skb;
3640                         buffer_info->dma = pci_map_single(pdev, skb->data,
3641                                                           bufsz,
3642                                                           PCI_DMA_FROMDEVICE);
3643
3644                 }
3645                 /* Refresh the desc even if buffer_addrs didn't change because
3646                  * each write-back erases this info. */
3647                 if (adapter->rx_ps_hdr_size) {
3648                         rx_desc->read.pkt_addr =
3649                              cpu_to_le64(buffer_info->page_dma);
3650                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
3651                 } else {
3652                         rx_desc->read.pkt_addr =
3653                              cpu_to_le64(buffer_info->dma);
3654                         rx_desc->read.hdr_addr = 0;
3655                 }
3656
3657                 i++;
3658                 if (i == rx_ring->count)
3659                         i = 0;
3660                 buffer_info = &rx_ring->buffer_info[i];
3661         }
3662
3663 no_buffers:
3664         if (rx_ring->next_to_use != i) {
3665                 rx_ring->next_to_use = i;
3666                 if (i == 0)
3667                         i = (rx_ring->count - 1);
3668                 else
3669                         i--;
3670
3671                 /* Force memory writes to complete before letting h/w
3672                  * know there are new descriptors to fetch.  (Only
3673                  * applicable for weak-ordered memory model archs,
3674                  * such as IA-64). */
3675                 wmb();
3676                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
3677         }
3678 }
3679
3680 /**
3681  * igb_mii_ioctl -
3682  * @netdev:
3683  * @ifreq:
3684  * @cmd:
3685  **/
3686 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3687 {
3688         struct igb_adapter *adapter = netdev_priv(netdev);
3689         struct mii_ioctl_data *data = if_mii(ifr);
3690
3691         if (adapter->hw.phy.media_type != e1000_media_type_copper)
3692                 return -EOPNOTSUPP;
3693
3694         switch (cmd) {
3695         case SIOCGMIIPHY:
3696                 data->phy_id = adapter->hw.phy.addr;
3697                 break;
3698         case SIOCGMIIREG:
3699                 if (!capable(CAP_NET_ADMIN))
3700                         return -EPERM;
3701                 if (adapter->hw.phy.ops.read_phy_reg(&adapter->hw,
3702                                                      data->reg_num
3703                                                      & 0x1F, &data->val_out))
3704                         return -EIO;
3705                 break;
3706         case SIOCSMIIREG:
3707         default:
3708                 return -EOPNOTSUPP;
3709         }
3710         return 0;
3711 }
3712
3713 /**
3714  * igb_ioctl -
3715  * @netdev:
3716  * @ifreq:
3717  * @cmd:
3718  **/
3719 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
3720 {
3721         switch (cmd) {
3722         case SIOCGMIIPHY:
3723         case SIOCGMIIREG:
3724         case SIOCSMIIREG:
3725                 return igb_mii_ioctl(netdev, ifr, cmd);
3726         default:
3727                 return -EOPNOTSUPP;
3728         }
3729 }
3730
3731 static void igb_vlan_rx_register(struct net_device *netdev,
3732                                  struct vlan_group *grp)
3733 {
3734         struct igb_adapter *adapter = netdev_priv(netdev);
3735         struct e1000_hw *hw = &adapter->hw;
3736         u32 ctrl, rctl;
3737
3738         igb_irq_disable(adapter);
3739         adapter->vlgrp = grp;
3740
3741         if (grp) {
3742                 /* enable VLAN tag insert/strip */
3743                 ctrl = rd32(E1000_CTRL);
3744                 ctrl |= E1000_CTRL_VME;
3745                 wr32(E1000_CTRL, ctrl);
3746
3747                 /* enable VLAN receive filtering */
3748                 rctl = rd32(E1000_RCTL);
3749                 rctl |= E1000_RCTL_VFE;
3750                 rctl &= ~E1000_RCTL_CFIEN;
3751                 wr32(E1000_RCTL, rctl);
3752                 igb_update_mng_vlan(adapter);
3753                 wr32(E1000_RLPML,
3754                                 adapter->max_frame_size + VLAN_TAG_SIZE);
3755         } else {
3756                 /* disable VLAN tag insert/strip */
3757                 ctrl = rd32(E1000_CTRL);
3758                 ctrl &= ~E1000_CTRL_VME;
3759                 wr32(E1000_CTRL, ctrl);
3760
3761                 /* disable VLAN filtering */
3762                 rctl = rd32(E1000_RCTL);
3763                 rctl &= ~E1000_RCTL_VFE;
3764                 wr32(E1000_RCTL, rctl);
3765                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
3766                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
3767                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
3768                 }
3769                 wr32(E1000_RLPML,
3770                                 adapter->max_frame_size);
3771         }
3772
3773         if (!test_bit(__IGB_DOWN, &adapter->state))
3774                 igb_irq_enable(adapter);
3775 }
3776
3777 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
3778 {
3779         struct igb_adapter *adapter = netdev_priv(netdev);
3780         struct e1000_hw *hw = &adapter->hw;
3781         u32 vfta, index;
3782
3783         if ((adapter->hw.mng_cookie.status &
3784              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3785             (vid == adapter->mng_vlan_id))
3786                 return;
3787         /* add VID to filter table */
3788         index = (vid >> 5) & 0x7F;
3789         vfta = array_rd32(E1000_VFTA, index);
3790         vfta |= (1 << (vid & 0x1F));
3791         igb_write_vfta(&adapter->hw, index, vfta);
3792 }
3793
3794 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
3795 {
3796         struct igb_adapter *adapter = netdev_priv(netdev);
3797         struct e1000_hw *hw = &adapter->hw;
3798         u32 vfta, index;
3799
3800         igb_irq_disable(adapter);
3801         vlan_group_set_device(adapter->vlgrp, vid, NULL);
3802
3803         if (!test_bit(__IGB_DOWN, &adapter->state))
3804                 igb_irq_enable(adapter);
3805
3806         if ((adapter->hw.mng_cookie.status &
3807              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
3808             (vid == adapter->mng_vlan_id)) {
3809                 /* release control to f/w */
3810                 igb_release_hw_control(adapter);
3811                 return;
3812         }
3813
3814         /* remove VID from filter table */
3815         index = (vid >> 5) & 0x7F;
3816         vfta = array_rd32(E1000_VFTA, index);
3817         vfta &= ~(1 << (vid & 0x1F));
3818         igb_write_vfta(&adapter->hw, index, vfta);
3819 }
3820
3821 static void igb_restore_vlan(struct igb_adapter *adapter)
3822 {
3823         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
3824
3825         if (adapter->vlgrp) {
3826                 u16 vid;
3827                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
3828                         if (!vlan_group_get_device(adapter->vlgrp, vid))
3829                                 continue;
3830                         igb_vlan_rx_add_vid(adapter->netdev, vid);
3831                 }
3832         }
3833 }
3834
3835 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
3836 {
3837         struct e1000_mac_info *mac = &adapter->hw.mac;
3838
3839         mac->autoneg = 0;
3840
3841         /* Fiber NICs only allow 1000 gbps Full duplex */
3842         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
3843                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
3844                 dev_err(&adapter->pdev->dev,
3845                         "Unsupported Speed/Duplex configuration\n");
3846                 return -EINVAL;
3847         }
3848
3849         switch (spddplx) {
3850         case SPEED_10 + DUPLEX_HALF:
3851                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
3852                 break;
3853         case SPEED_10 + DUPLEX_FULL:
3854                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
3855                 break;
3856         case SPEED_100 + DUPLEX_HALF:
3857                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
3858                 break;
3859         case SPEED_100 + DUPLEX_FULL:
3860                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
3861                 break;
3862         case SPEED_1000 + DUPLEX_FULL:
3863                 mac->autoneg = 1;
3864                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
3865                 break;
3866         case SPEED_1000 + DUPLEX_HALF: /* not supported */
3867         default:
3868                 dev_err(&adapter->pdev->dev,
3869                         "Unsupported Speed/Duplex configuration\n");
3870                 return -EINVAL;
3871         }
3872         return 0;
3873 }
3874
3875
3876 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
3877 {
3878         struct net_device *netdev = pci_get_drvdata(pdev);
3879         struct igb_adapter *adapter = netdev_priv(netdev);
3880         struct e1000_hw *hw = &adapter->hw;
3881         u32 ctrl, ctrl_ext, rctl, status;
3882         u32 wufc = adapter->wol;
3883 #ifdef CONFIG_PM
3884         int retval = 0;
3885 #endif
3886
3887         netif_device_detach(netdev);
3888
3889         if (netif_running(netdev)) {
3890                 WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
3891                 igb_down(adapter);
3892                 igb_free_irq(adapter);
3893         }
3894
3895 #ifdef CONFIG_PM
3896         retval = pci_save_state(pdev);
3897         if (retval)
3898                 return retval;
3899 #endif
3900
3901         status = rd32(E1000_STATUS);
3902         if (status & E1000_STATUS_LU)
3903                 wufc &= ~E1000_WUFC_LNKC;
3904
3905         if (wufc) {
3906                 igb_setup_rctl(adapter);
3907                 igb_set_multi(netdev);
3908
3909                 /* turn on all-multi mode if wake on multicast is enabled */
3910                 if (wufc & E1000_WUFC_MC) {
3911                         rctl = rd32(E1000_RCTL);
3912                         rctl |= E1000_RCTL_MPE;
3913                         wr32(E1000_RCTL, rctl);
3914                 }
3915
3916                 ctrl = rd32(E1000_CTRL);
3917                 /* advertise wake from D3Cold */
3918                 #define E1000_CTRL_ADVD3WUC 0x00100000
3919                 /* phy power management enable */
3920                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
3921                 ctrl |= E1000_CTRL_ADVD3WUC;
3922                 wr32(E1000_CTRL, ctrl);
3923
3924                 if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
3925                    adapter->hw.phy.media_type ==
3926                                         e1000_media_type_internal_serdes) {
3927                         /* keep the laser running in D3 */
3928                         ctrl_ext = rd32(E1000_CTRL_EXT);
3929                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
3930                         wr32(E1000_CTRL_EXT, ctrl_ext);
3931                 }
3932
3933                 /* Allow time for pending master requests to run */
3934                 igb_disable_pcie_master(&adapter->hw);
3935
3936                 wr32(E1000_WUC, E1000_WUC_PME_EN);
3937                 wr32(E1000_WUFC, wufc);
3938                 pci_enable_wake(pdev, PCI_D3hot, 1);
3939                 pci_enable_wake(pdev, PCI_D3cold, 1);
3940         } else {
3941                 wr32(E1000_WUC, 0);
3942                 wr32(E1000_WUFC, 0);
3943                 pci_enable_wake(pdev, PCI_D3hot, 0);
3944                 pci_enable_wake(pdev, PCI_D3cold, 0);
3945         }
3946
3947         /* make sure adapter isn't asleep if manageability is enabled */
3948         if (adapter->en_mng_pt) {
3949                 pci_enable_wake(pdev, PCI_D3hot, 1);
3950                 pci_enable_wake(pdev, PCI_D3cold, 1);
3951         }
3952
3953         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
3954          * would have already happened in close and is redundant. */
3955         igb_release_hw_control(adapter);
3956
3957         pci_disable_device(pdev);
3958
3959         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3960
3961         return 0;
3962 }
3963
3964 #ifdef CONFIG_PM
3965 static int igb_resume(struct pci_dev *pdev)
3966 {
3967         struct net_device *netdev = pci_get_drvdata(pdev);
3968         struct igb_adapter *adapter = netdev_priv(netdev);
3969         struct e1000_hw *hw = &adapter->hw;
3970         u32 err;
3971
3972         pci_set_power_state(pdev, PCI_D0);
3973         pci_restore_state(pdev);
3974         err = pci_enable_device(pdev);
3975         if (err) {
3976                 dev_err(&pdev->dev,
3977                         "igb: Cannot enable PCI device from suspend\n");
3978                 return err;
3979         }
3980         pci_set_master(pdev);
3981
3982         pci_enable_wake(pdev, PCI_D3hot, 0);
3983         pci_enable_wake(pdev, PCI_D3cold, 0);
3984
3985         if (netif_running(netdev)) {
3986                 err = igb_request_irq(adapter);
3987                 if (err)
3988                         return err;
3989         }
3990
3991         /* e1000_power_up_phy(adapter); */
3992
3993         igb_reset(adapter);
3994         wr32(E1000_WUS, ~0);
3995
3996         igb_init_manageability(adapter);
3997
3998         if (netif_running(netdev))
3999                 igb_up(adapter);
4000
4001         netif_device_attach(netdev);
4002
4003         /* let the f/w know that the h/w is now under the control of the
4004          * driver. */
4005         igb_get_hw_control(adapter);
4006
4007         return 0;
4008 }
4009 #endif
4010
4011 static void igb_shutdown(struct pci_dev *pdev)
4012 {
4013         igb_suspend(pdev, PMSG_SUSPEND);
4014 }
4015
4016 #ifdef CONFIG_NET_POLL_CONTROLLER
4017 /*
4018  * Polling 'interrupt' - used by things like netconsole to send skbs
4019  * without having to re-enable interrupts. It's not called while
4020  * the interrupt routine is executing.
4021  */
4022 static void igb_netpoll(struct net_device *netdev)
4023 {
4024         struct igb_adapter *adapter = netdev_priv(netdev);
4025         int i;
4026         int work_done = 0;
4027
4028         igb_irq_disable(adapter);
4029         for (i = 0; i < adapter->num_tx_queues; i++)
4030                 igb_clean_tx_irq(adapter, &adapter->tx_ring[i]);
4031
4032         for (i = 0; i < adapter->num_rx_queues; i++)
4033                 igb_clean_rx_irq_adv(adapter, &adapter->rx_ring[i],
4034                                      &work_done,
4035                                      adapter->rx_ring[i].napi.weight);
4036
4037         igb_irq_enable(adapter);
4038 }
4039 #endif /* CONFIG_NET_POLL_CONTROLLER */
4040
4041 /**
4042  * igb_io_error_detected - called when PCI error is detected
4043  * @pdev: Pointer to PCI device
4044  * @state: The current pci connection state
4045  *
4046  * This function is called after a PCI bus error affecting
4047  * this device has been detected.
4048  */
4049 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
4050                                               pci_channel_state_t state)
4051 {
4052         struct net_device *netdev = pci_get_drvdata(pdev);
4053         struct igb_adapter *adapter = netdev_priv(netdev);
4054
4055         netif_device_detach(netdev);
4056
4057         if (netif_running(netdev))
4058                 igb_down(adapter);
4059         pci_disable_device(pdev);
4060
4061         /* Request a slot slot reset. */
4062         return PCI_ERS_RESULT_NEED_RESET;
4063 }
4064
4065 /**
4066  * igb_io_slot_reset - called after the pci bus has been reset.
4067  * @pdev: Pointer to PCI device
4068  *
4069  * Restart the card from scratch, as if from a cold-boot. Implementation
4070  * resembles the first-half of the igb_resume routine.
4071  */
4072 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
4073 {
4074         struct net_device *netdev = pci_get_drvdata(pdev);
4075         struct igb_adapter *adapter = netdev_priv(netdev);
4076         struct e1000_hw *hw = &adapter->hw;
4077
4078         if (pci_enable_device(pdev)) {
4079                 dev_err(&pdev->dev,
4080                         "Cannot re-enable PCI device after reset.\n");
4081                 return PCI_ERS_RESULT_DISCONNECT;
4082         }
4083         pci_set_master(pdev);
4084         pci_restore_state(pdev);
4085
4086         pci_enable_wake(pdev, PCI_D3hot, 0);
4087         pci_enable_wake(pdev, PCI_D3cold, 0);
4088
4089         igb_reset(adapter);
4090         wr32(E1000_WUS, ~0);
4091
4092         return PCI_ERS_RESULT_RECOVERED;
4093 }
4094
4095 /**
4096  * igb_io_resume - called when traffic can start flowing again.
4097  * @pdev: Pointer to PCI device
4098  *
4099  * This callback is called when the error recovery driver tells us that
4100  * its OK to resume normal operation. Implementation resembles the
4101  * second-half of the igb_resume routine.
4102  */
4103 static void igb_io_resume(struct pci_dev *pdev)
4104 {
4105         struct net_device *netdev = pci_get_drvdata(pdev);
4106         struct igb_adapter *adapter = netdev_priv(netdev);
4107
4108         igb_init_manageability(adapter);
4109
4110         if (netif_running(netdev)) {
4111                 if (igb_up(adapter)) {
4112                         dev_err(&pdev->dev, "igb_up failed after reset\n");
4113                         return;
4114                 }
4115         }
4116
4117         netif_device_attach(netdev);
4118
4119         /* let the f/w know that the h/w is now under the control of the
4120          * driver. */
4121         igb_get_hw_control(adapter);
4122
4123 }
4124
4125 /* igb_main.c */