drivers/net/ethernet/intel/e1000/e1000_main.c

   1 /*******************************************************************************
   2
   3   Intel PRO/1000 Linux driver
   4   Copyright(c) 1999 - 2006 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   Linux NICS <linux.nics@intel.com>
  24   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include "e1000.h"
  30 #include <net/ip6_checksum.h>
  31 #include <linux/io.h>
  32 #include <linux/prefetch.h>
  33 #include <linux/bitops.h>
  34 #include <linux/if_vlan.h>
  35
  36 char e1000_driver_name[] = "e1000";
  37 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
  38 #define DRV_VERSION "7.3.21-k8-NAPI"
  39 const char e1000_driver_version[] = DRV_VERSION;
  40 static const char e1000_copyright[] = "Copyright (c) 1999-2006 Intel Corporation.";
  41
  42 /* e1000_pci_tbl - PCI Device ID Table
  43  *
  44  * Last entry must be all 0s
  45  *
  46  * Macro expands to...
  47  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
  48  */
  49 static const struct pci_device_id e1000_pci_tbl[] = {
  50         INTEL_E1000_ETHERNET_DEVICE(0x1000),
  51         INTEL_E1000_ETHERNET_DEVICE(0x1001),
  52         INTEL_E1000_ETHERNET_DEVICE(0x1004),
  53         INTEL_E1000_ETHERNET_DEVICE(0x1008),
  54         INTEL_E1000_ETHERNET_DEVICE(0x1009),
  55         INTEL_E1000_ETHERNET_DEVICE(0x100C),
  56         INTEL_E1000_ETHERNET_DEVICE(0x100D),
  57         INTEL_E1000_ETHERNET_DEVICE(0x100E),
  58         INTEL_E1000_ETHERNET_DEVICE(0x100F),
  59         INTEL_E1000_ETHERNET_DEVICE(0x1010),
  60         INTEL_E1000_ETHERNET_DEVICE(0x1011),
  61         INTEL_E1000_ETHERNET_DEVICE(0x1012),
  62         INTEL_E1000_ETHERNET_DEVICE(0x1013),
  63         INTEL_E1000_ETHERNET_DEVICE(0x1014),
  64         INTEL_E1000_ETHERNET_DEVICE(0x1015),
  65         INTEL_E1000_ETHERNET_DEVICE(0x1016),
  66         INTEL_E1000_ETHERNET_DEVICE(0x1017),
  67         INTEL_E1000_ETHERNET_DEVICE(0x1018),
  68         INTEL_E1000_ETHERNET_DEVICE(0x1019),
  69         INTEL_E1000_ETHERNET_DEVICE(0x101A),
  70         INTEL_E1000_ETHERNET_DEVICE(0x101D),
  71         INTEL_E1000_ETHERNET_DEVICE(0x101E),
  72         INTEL_E1000_ETHERNET_DEVICE(0x1026),
  73         INTEL_E1000_ETHERNET_DEVICE(0x1027),
  74         INTEL_E1000_ETHERNET_DEVICE(0x1028),
  75         INTEL_E1000_ETHERNET_DEVICE(0x1075),
  76         INTEL_E1000_ETHERNET_DEVICE(0x1076),
  77         INTEL_E1000_ETHERNET_DEVICE(0x1077),
  78         INTEL_E1000_ETHERNET_DEVICE(0x1078),
  79         INTEL_E1000_ETHERNET_DEVICE(0x1079),
  80         INTEL_E1000_ETHERNET_DEVICE(0x107A),
  81         INTEL_E1000_ETHERNET_DEVICE(0x107B),
  82         INTEL_E1000_ETHERNET_DEVICE(0x107C),
  83         INTEL_E1000_ETHERNET_DEVICE(0x108A),
  84         INTEL_E1000_ETHERNET_DEVICE(0x1099),
  85         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
  86         INTEL_E1000_ETHERNET_DEVICE(0x2E6E),
  87         /* required last entry */
  88         {0,}
  89 };
  90
  91 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
  92
  93 int e1000_up(struct e1000_adapter *adapter);
  94 void e1000_down(struct e1000_adapter *adapter);
  95 void e1000_reinit_locked(struct e1000_adapter *adapter);
  96 void e1000_reset(struct e1000_adapter *adapter);
  97 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
  98 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
  99 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
 100 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
 101 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
 102                              struct e1000_tx_ring *txdr);
 103 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
 104                              struct e1000_rx_ring *rxdr);
 105 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
 106                              struct e1000_tx_ring *tx_ring);
 107 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
 108                              struct e1000_rx_ring *rx_ring);
 109 void e1000_update_stats(struct e1000_adapter *adapter);
 110
 111 static int e1000_init_module(void);
 112 static void e1000_exit_module(void);
 113 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
 114 static void e1000_remove(struct pci_dev *pdev);
 115 static int e1000_alloc_queues(struct e1000_adapter *adapter);
 116 static int e1000_sw_init(struct e1000_adapter *adapter);
 117 static int e1000_open(struct net_device *netdev);
 118 static int e1000_close(struct net_device *netdev);
 119 static void e1000_configure_tx(struct e1000_adapter *adapter);
 120 static void e1000_configure_rx(struct e1000_adapter *adapter);
 121 static void e1000_setup_rctl(struct e1000_adapter *adapter);
 122 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
 123 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
 124 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
 125                                 struct e1000_tx_ring *tx_ring);
 126 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
 127                                 struct e1000_rx_ring *rx_ring);
 128 static void e1000_set_rx_mode(struct net_device *netdev);
 129 static void e1000_update_phy_info_task(struct work_struct *work);
 130 static void e1000_watchdog(struct work_struct *work);
 131 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work);
 132 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
 133                                     struct net_device *netdev);
 134 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
 135 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
 136 static int e1000_set_mac(struct net_device *netdev, void *p);
 137 static irqreturn_t e1000_intr(int irq, void *data);
 138 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
 139                                struct e1000_tx_ring *tx_ring);
 140 static int e1000_clean(struct napi_struct *napi, int budget);
 141 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
 142                                struct e1000_rx_ring *rx_ring,
 143                                int *work_done, int work_to_do);
 144 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
 145                                      struct e1000_rx_ring *rx_ring,
 146                                      int *work_done, int work_to_do);
 147 static void e1000_alloc_dummy_rx_buffers(struct e1000_adapter *adapter,
 148                                          struct e1000_rx_ring *rx_ring,
 149                                          int cleaned_count)
 150 {
 151 }
 152 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 153                                    struct e1000_rx_ring *rx_ring,
 154                                    int cleaned_count);
 155 static void e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
 156                                          struct e1000_rx_ring *rx_ring,
 157                                          int cleaned_count);
 158 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
 159 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
 160                            int cmd);
 161 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
 162 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
 163 static void e1000_tx_timeout(struct net_device *dev);
 164 static void e1000_reset_task(struct work_struct *work);
 165 static void e1000_smartspeed(struct e1000_adapter *adapter);
 166 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
 167                                        struct sk_buff *skb);
 168
 169 static bool e1000_vlan_used(struct e1000_adapter *adapter);
 170 static void e1000_vlan_mode(struct net_device *netdev,
 171                             netdev_features_t features);
 172 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
 173                                      bool filter_on);
 174 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
 175                                  __be16 proto, u16 vid);
 176 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
 177                                   __be16 proto, u16 vid);
 178 static void e1000_restore_vlan(struct e1000_adapter *adapter);
 179
 180 #ifdef CONFIG_PM
 181 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
 182 static int e1000_resume(struct pci_dev *pdev);
 183 #endif
 184 static void e1000_shutdown(struct pci_dev *pdev);
 185
 186 #ifdef CONFIG_NET_POLL_CONTROLLER
 187 /* for netdump / net console */
 188 static void e1000_netpoll (struct net_device *netdev);
 189 #endif
 190
 191 #define COPYBREAK_DEFAULT 256
 192 static unsigned int copybreak __read_mostly = COPYBREAK_DEFAULT;
 193 module_param(copybreak, uint, 0644);
 194 MODULE_PARM_DESC(copybreak,
 195         "Maximum size of packet that is copied to a new buffer on receive");
 196
 197 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
 198                      pci_channel_state_t state);
 199 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev);
 200 static void e1000_io_resume(struct pci_dev *pdev);
 201
 202 static const struct pci_error_handlers e1000_err_handler = {
 203         .error_detected = e1000_io_error_detected,
 204         .slot_reset = e1000_io_slot_reset,
 205         .resume = e1000_io_resume,
 206 };
 207
 208 static struct pci_driver e1000_driver = {
 209         .name     = e1000_driver_name,
 210         .id_table = e1000_pci_tbl,
 211         .probe    = e1000_probe,
 212         .remove   = e1000_remove,
 213 #ifdef CONFIG_PM
 214         /* Power Management Hooks */
 215         .suspend  = e1000_suspend,
 216         .resume   = e1000_resume,
 217 #endif
 218         .shutdown = e1000_shutdown,
 219         .err_handler = &e1000_err_handler
 220 };
 221
 222 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 223 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
 224 MODULE_LICENSE("GPL");
 225 MODULE_VERSION(DRV_VERSION);
 226
 227 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
 228 static int debug = -1;
 229 module_param(debug, int, 0);
 230 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 231
 232 /**
 233  * e1000_get_hw_dev - return device
 234  * used by hardware layer to print debugging information
 235  *
 236  **/
 237 struct net_device *e1000_get_hw_dev(struct e1000_hw *hw)
 238 {
 239         struct e1000_adapter *adapter = hw->back;
 240         return adapter->netdev;
 241 }
 242
 243 /**
 244  * e1000_init_module - Driver Registration Routine
 245  *
 246  * e1000_init_module is the first routine called when the driver is
 247  * loaded. All it does is register with the PCI subsystem.
 248  **/
 249 static int __init e1000_init_module(void)
 250 {
 251         int ret;
 252         pr_info("%s - version %s\n", e1000_driver_string, e1000_driver_version);
 253
 254         pr_info("%s\n", e1000_copyright);
 255
 256         ret = pci_register_driver(&e1000_driver);
 257         if (copybreak != COPYBREAK_DEFAULT) {
 258                 if (copybreak == 0)
 259                         pr_info("copybreak disabled\n");
 260                 else
 261                         pr_info("copybreak enabled for "
 262                                    "packets <= %u bytes\n", copybreak);
 263         }
 264         return ret;
 265 }
 266
 267 module_init(e1000_init_module);
 268
 269 /**
 270  * e1000_exit_module - Driver Exit Cleanup Routine
 271  *
 272  * e1000_exit_module is called just before the driver is removed
 273  * from memory.
 274  **/
 275 static void __exit e1000_exit_module(void)
 276 {
 277         pci_unregister_driver(&e1000_driver);
 278 }
 279
 280 module_exit(e1000_exit_module);
 281
 282 static int e1000_request_irq(struct e1000_adapter *adapter)
 283 {
 284         struct net_device *netdev = adapter->netdev;
 285         irq_handler_t handler = e1000_intr;
 286         int irq_flags = IRQF_SHARED;
 287         int err;
 288
 289         err = request_irq(adapter->pdev->irq, handler, irq_flags, netdev->name,
 290                           netdev);
 291         if (err) {
 292                 e_err(probe, "Unable to allocate interrupt Error: %d\n", err);
 293         }
 294
 295         return err;
 296 }
 297
 298 static void e1000_free_irq(struct e1000_adapter *adapter)
 299 {
 300         struct net_device *netdev = adapter->netdev;
 301
 302         free_irq(adapter->pdev->irq, netdev);
 303 }
 304
 305 /**
 306  * e1000_irq_disable - Mask off interrupt generation on the NIC
 307  * @adapter: board private structure
 308  **/
 309 static void e1000_irq_disable(struct e1000_adapter *adapter)
 310 {
 311         struct e1000_hw *hw = &adapter->hw;
 312
 313         ew32(IMC, ~0);
 314         E1000_WRITE_FLUSH();
 315         synchronize_irq(adapter->pdev->irq);
 316 }
 317
 318 /**
 319  * e1000_irq_enable - Enable default interrupt generation settings
 320  * @adapter: board private structure
 321  **/
 322 static void e1000_irq_enable(struct e1000_adapter *adapter)
 323 {
 324         struct e1000_hw *hw = &adapter->hw;
 325
 326         ew32(IMS, IMS_ENABLE_MASK);
 327         E1000_WRITE_FLUSH();
 328 }
 329
 330 static void e1000_update_mng_vlan(struct e1000_adapter *adapter)
 331 {
 332         struct e1000_hw *hw = &adapter->hw;
 333         struct net_device *netdev = adapter->netdev;
 334         u16 vid = hw->mng_cookie.vlan_id;
 335         u16 old_vid = adapter->mng_vlan_id;
 336
 337         if (!e1000_vlan_used(adapter))
 338                 return;
 339
 340         if (!test_bit(vid, adapter->active_vlans)) {
 341                 if (hw->mng_cookie.status &
 342                     E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
 343                         e1000_vlan_rx_add_vid(netdev, htons(ETH_P_8021Q), vid);
 344                         adapter->mng_vlan_id = vid;
 345                 } else {
 346                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
 347                 }
 348                 if ((old_vid != (u16)E1000_MNG_VLAN_NONE) &&
 349                     (vid != old_vid) &&
 350                     !test_bit(old_vid, adapter->active_vlans))
 351                         e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
 352                                                old_vid);
 353         } else {
 354                 adapter->mng_vlan_id = vid;
 355         }
 356 }
 357
 358 static void e1000_init_manageability(struct e1000_adapter *adapter)
 359 {
 360         struct e1000_hw *hw = &adapter->hw;
 361
 362         if (adapter->en_mng_pt) {
 363                 u32 manc = er32(MANC);
 364
 365                 /* disable hardware interception of ARP */
 366                 manc &= ~(E1000_MANC_ARP_EN);
 367
 368                 ew32(MANC, manc);
 369         }
 370 }
 371
 372 static void e1000_release_manageability(struct e1000_adapter *adapter)
 373 {
 374         struct e1000_hw *hw = &adapter->hw;
 375
 376         if (adapter->en_mng_pt) {
 377                 u32 manc = er32(MANC);
 378
 379                 /* re-enable hardware interception of ARP */
 380                 manc |= E1000_MANC_ARP_EN;
 381
 382                 ew32(MANC, manc);
 383         }
 384 }
 385
 386 /**
 387  * e1000_configure - configure the hardware for RX and TX
 388  * @adapter = private board structure
 389  **/
 390 static void e1000_configure(struct e1000_adapter *adapter)
 391 {
 392         struct net_device *netdev = adapter->netdev;
 393         int i;
 394
 395         e1000_set_rx_mode(netdev);
 396
 397         e1000_restore_vlan(adapter);
 398         e1000_init_manageability(adapter);
 399
 400         e1000_configure_tx(adapter);
 401         e1000_setup_rctl(adapter);
 402         e1000_configure_rx(adapter);
 403         /* call E1000_DESC_UNUSED which always leaves
 404          * at least 1 descriptor unused to make sure
 405          * next_to_use != next_to_clean
 406          */
 407         for (i = 0; i < adapter->num_rx_queues; i++) {
 408                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
 409                 adapter->alloc_rx_buf(adapter, ring,
 410                                       E1000_DESC_UNUSED(ring));
 411         }
 412 }
 413
 414 int e1000_up(struct e1000_adapter *adapter)
 415 {
 416         struct e1000_hw *hw = &adapter->hw;
 417
 418         /* hardware has been reset, we need to reload some things */
 419         e1000_configure(adapter);
 420
 421         clear_bit(__E1000_DOWN, &adapter->flags);
 422
 423         napi_enable(&adapter->napi);
 424
 425         e1000_irq_enable(adapter);
 426
 427         netif_wake_queue(adapter->netdev);
 428
 429         /* fire a link change interrupt to start the watchdog */
 430         ew32(ICS, E1000_ICS_LSC);
 431         return 0;
 432 }
 433
 434 /**
 435  * e1000_power_up_phy - restore link in case the phy was powered down
 436  * @adapter: address of board private structure
 437  *
 438  * The phy may be powered down to save power and turn off link when the
 439  * driver is unloaded and wake on lan is not enabled (among others)
 440  * *** this routine MUST be followed by a call to e1000_reset ***
 441  **/
 442 void e1000_power_up_phy(struct e1000_adapter *adapter)
 443 {
 444         struct e1000_hw *hw = &adapter->hw;
 445         u16 mii_reg = 0;
 446
 447         /* Just clear the power down bit to wake the phy back up */
 448         if (hw->media_type == e1000_media_type_copper) {
 449                 /* according to the manual, the phy will retain its
 450                  * settings across a power-down/up cycle
 451                  */
 452                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
 453                 mii_reg &= ~MII_CR_POWER_DOWN;
 454                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
 455         }
 456 }
 457
 458 static void e1000_power_down_phy(struct e1000_adapter *adapter)
 459 {
 460         struct e1000_hw *hw = &adapter->hw;
 461
 462         /* Power down the PHY so no link is implied when interface is down *
 463          * The PHY cannot be powered down if any of the following is true *
 464          * (a) WoL is enabled
 465          * (b) AMT is active
 466          * (c) SoL/IDER session is active
 467          */
 468         if (!adapter->wol && hw->mac_type >= e1000_82540 &&
 469            hw->media_type == e1000_media_type_copper) {
 470                 u16 mii_reg = 0;
 471
 472                 switch (hw->mac_type) {
 473                 case e1000_82540:
 474                 case e1000_82545:
 475                 case e1000_82545_rev_3:
 476                 case e1000_82546:
 477                 case e1000_ce4100:
 478                 case e1000_82546_rev_3:
 479                 case e1000_82541:
 480                 case e1000_82541_rev_2:
 481                 case e1000_82547:
 482                 case e1000_82547_rev_2:
 483                         if (er32(MANC) & E1000_MANC_SMBUS_EN)
 484                                 goto out;
 485                         break;
 486                 default:
 487                         goto out;
 488                 }
 489                 e1000_read_phy_reg(hw, PHY_CTRL, &mii_reg);
 490                 mii_reg |= MII_CR_POWER_DOWN;
 491                 e1000_write_phy_reg(hw, PHY_CTRL, mii_reg);
 492                 msleep(1);
 493         }
 494 out:
 495         return;
 496 }
 497
 498 static void e1000_down_and_stop(struct e1000_adapter *adapter)
 499 {
 500         set_bit(__E1000_DOWN, &adapter->flags);
 501
 502         cancel_delayed_work_sync(&adapter->watchdog_task);
 503
 504         /*
 505          * Since the watchdog task can reschedule other tasks, we should cancel
 506          * it first, otherwise we can run into the situation when a work is
 507          * still running after the adapter has been turned down.
 508          */
 509
 510         cancel_delayed_work_sync(&adapter->phy_info_task);
 511         cancel_delayed_work_sync(&adapter->fifo_stall_task);
 512
 513         /* Only kill reset task if adapter is not resetting */
 514         if (!test_bit(__E1000_RESETTING, &adapter->flags))
 515                 cancel_work_sync(&adapter->reset_task);
 516 }
 517
 518 void e1000_down(struct e1000_adapter *adapter)
 519 {
 520         struct e1000_hw *hw = &adapter->hw;
 521         struct net_device *netdev = adapter->netdev;
 522         u32 rctl, tctl;
 523
 524         netif_carrier_off(netdev);
 525
 526         /* disable receives in the hardware */
 527         rctl = er32(RCTL);
 528         ew32(RCTL, rctl & ~E1000_RCTL_EN);
 529         /* flush and sleep below */
 530
 531         netif_tx_disable(netdev);
 532
 533         /* disable transmits in the hardware */
 534         tctl = er32(TCTL);
 535         tctl &= ~E1000_TCTL_EN;
 536         ew32(TCTL, tctl);
 537         /* flush both disables and wait for them to finish */
 538         E1000_WRITE_FLUSH();
 539         msleep(10);
 540
 541         napi_disable(&adapter->napi);
 542
 543         e1000_irq_disable(adapter);
 544
 545         /* Setting DOWN must be after irq_disable to prevent
 546          * a screaming interrupt.  Setting DOWN also prevents
 547          * tasks from rescheduling.
 548          */
 549         e1000_down_and_stop(adapter);
 550
 551         adapter->link_speed = 0;
 552         adapter->link_duplex = 0;
 553
 554         e1000_reset(adapter);
 555         e1000_clean_all_tx_rings(adapter);
 556         e1000_clean_all_rx_rings(adapter);
 557 }
 558
 559 void e1000_reinit_locked(struct e1000_adapter *adapter)
 560 {
 561         WARN_ON(in_interrupt());
 562         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
 563                 msleep(1);
 564         e1000_down(adapter);
 565         e1000_up(adapter);
 566         clear_bit(__E1000_RESETTING, &adapter->flags);
 567 }
 568
 569 void e1000_reset(struct e1000_adapter *adapter)
 570 {
 571         struct e1000_hw *hw = &adapter->hw;
 572         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
 573         bool legacy_pba_adjust = false;
 574         u16 hwm;
 575
 576         /* Repartition Pba for greater than 9k mtu
 577          * To take effect CTRL.RST is required.
 578          */
 579
 580         switch (hw->mac_type) {
 581         case e1000_82542_rev2_0:
 582         case e1000_82542_rev2_1:
 583         case e1000_82543:
 584         case e1000_82544:
 585         case e1000_82540:
 586         case e1000_82541:
 587         case e1000_82541_rev_2:
 588                 legacy_pba_adjust = true;
 589                 pba = E1000_PBA_48K;
 590                 break;
 591         case e1000_82545:
 592         case e1000_82545_rev_3:
 593         case e1000_82546:
 594         case e1000_ce4100:
 595         case e1000_82546_rev_3:
 596                 pba = E1000_PBA_48K;
 597                 break;
 598         case e1000_82547:
 599         case e1000_82547_rev_2:
 600                 legacy_pba_adjust = true;
 601                 pba = E1000_PBA_30K;
 602                 break;
 603         case e1000_undefined:
 604         case e1000_num_macs:
 605                 break;
 606         }
 607
 608         if (legacy_pba_adjust) {
 609                 if (hw->max_frame_size > E1000_RXBUFFER_8192)
 610                         pba -= 8; /* allocate more FIFO for Tx */
 611
 612                 if (hw->mac_type == e1000_82547) {
 613                         adapter->tx_fifo_head = 0;
 614                         adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
 615                         adapter->tx_fifo_size =
 616                                 (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
 617                         atomic_set(&adapter->tx_fifo_stall, 0);
 618                 }
 619         } else if (hw->max_frame_size >  ETH_FRAME_LEN + ETH_FCS_LEN) {
 620                 /* adjust PBA for jumbo frames */
 621                 ew32(PBA, pba);
 622
 623                 /* To maintain wire speed transmits, the Tx FIFO should be
 624                  * large enough to accommodate two full transmit packets,
 625                  * rounded up to the next 1KB and expressed in KB.  Likewise,
 626                  * the Rx FIFO should be large enough to accommodate at least
 627                  * one full receive packet and is similarly rounded up and
 628                  * expressed in KB.
 629                  */
 630                 pba = er32(PBA);
 631                 /* upper 16 bits has Tx packet buffer allocation size in KB */
 632                 tx_space = pba >> 16;
 633                 /* lower 16 bits has Rx packet buffer allocation size in KB */
 634                 pba &= 0xffff;
 635                 /* the Tx fifo also stores 16 bytes of information about the Tx
 636                  * but don't include ethernet FCS because hardware appends it
 637                  */
 638                 min_tx_space = (hw->max_frame_size +
 639                                 sizeof(struct e1000_tx_desc) -
 640                                 ETH_FCS_LEN) * 2;
 641                 min_tx_space = ALIGN(min_tx_space, 1024);
 642                 min_tx_space >>= 10;
 643                 /* software strips receive CRC, so leave room for it */
 644                 min_rx_space = hw->max_frame_size;
 645                 min_rx_space = ALIGN(min_rx_space, 1024);
 646                 min_rx_space >>= 10;
 647
 648                 /* If current Tx allocation is less than the min Tx FIFO size,
 649                  * and the min Tx FIFO size is less than the current Rx FIFO
 650                  * allocation, take space away from current Rx allocation
 651                  */
 652                 if (tx_space < min_tx_space &&
 653                     ((min_tx_space - tx_space) < pba)) {
 654                         pba = pba - (min_tx_space - tx_space);
 655
 656                         /* PCI/PCIx hardware has PBA alignment constraints */
 657                         switch (hw->mac_type) {
 658                         case e1000_82545 ... e1000_82546_rev_3:
 659                                 pba &= ~(E1000_PBA_8K - 1);
 660                                 break;
 661                         default:
 662                                 break;
 663                         }
 664
 665                         /* if short on Rx space, Rx wins and must trump Tx
 666                          * adjustment or use Early Receive if available
 667                          */
 668                         if (pba < min_rx_space)
 669                                 pba = min_rx_space;
 670                 }
 671         }
 672
 673         ew32(PBA, pba);
 674
 675         /* flow control settings:
 676          * The high water mark must be low enough to fit one full frame
 677          * (or the size used for early receive) above it in the Rx FIFO.
 678          * Set it to the lower of:
 679          * - 90% of the Rx FIFO size, and
 680          * - the full Rx FIFO size minus the early receive size (for parts
 681          *   with ERT support assuming ERT set to E1000_ERT_2048), or
 682          * - the full Rx FIFO size minus one full frame
 683          */
 684         hwm = min(((pba << 10) * 9 / 10),
 685                   ((pba << 10) - hw->max_frame_size));
 686
 687         hw->fc_high_water = hwm & 0xFFF8;       /* 8-byte granularity */
 688         hw->fc_low_water = hw->fc_high_water - 8;
 689         hw->fc_pause_time = E1000_FC_PAUSE_TIME;
 690         hw->fc_send_xon = 1;
 691         hw->fc = hw->original_fc;
 692
 693         /* Allow time for pending master requests to run */
 694         e1000_reset_hw(hw);
 695         if (hw->mac_type >= e1000_82544)
 696                 ew32(WUC, 0);
 697
 698         if (e1000_init_hw(hw))
 699                 e_dev_err("Hardware Error\n");
 700         e1000_update_mng_vlan(adapter);
 701
 702         /* if (adapter->hwflags & HWFLAGS_PHY_PWR_BIT) { */
 703         if (hw->mac_type >= e1000_82544 &&
 704             hw->autoneg == 1 &&
 705             hw->autoneg_advertised == ADVERTISE_1000_FULL) {
 706                 u32 ctrl = er32(CTRL);
 707                 /* clear phy power management bit if we are in gig only mode,
 708                  * which if enabled will attempt negotiation to 100Mb, which
 709                  * can cause a loss of link at power off or driver unload
 710                  */
 711                 ctrl &= ~E1000_CTRL_SWDPIN3;
 712                 ew32(CTRL, ctrl);
 713         }
 714
 715         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 716         ew32(VET, ETHERNET_IEEE_VLAN_TYPE);
 717
 718         e1000_reset_adaptive(hw);
 719         e1000_phy_get_info(hw, &adapter->phy_info);
 720
 721         e1000_release_manageability(adapter);
 722 }
 723
 724 /* Dump the eeprom for users having checksum issues */
 725 static void e1000_dump_eeprom(struct e1000_adapter *adapter)
 726 {
 727         struct net_device *netdev = adapter->netdev;
 728         struct ethtool_eeprom eeprom;
 729         const struct ethtool_ops *ops = netdev->ethtool_ops;
 730         u8 *data;
 731         int i;
 732         u16 csum_old, csum_new = 0;
 733
 734         eeprom.len = ops->get_eeprom_len(netdev);
 735         eeprom.offset = 0;
 736
 737         data = kmalloc(eeprom.len, GFP_KERNEL);
 738         if (!data)
 739                 return;
 740
 741         ops->get_eeprom(netdev, &eeprom, data);
 742
 743         csum_old = (data[EEPROM_CHECKSUM_REG * 2]) +
 744                    (data[EEPROM_CHECKSUM_REG * 2 + 1] << 8);
 745         for (i = 0; i < EEPROM_CHECKSUM_REG * 2; i += 2)
 746                 csum_new += data[i] + (data[i + 1] << 8);
 747         csum_new = EEPROM_SUM - csum_new;
 748
 749         pr_err("/*********************/\n");
 750         pr_err("Current EEPROM Checksum : 0x%04x\n", csum_old);
 751         pr_err("Calculated              : 0x%04x\n", csum_new);
 752
 753         pr_err("Offset    Values\n");
 754         pr_err("========  ======\n");
 755         print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1, data, 128, 0);
 756
 757         pr_err("Include this output when contacting your support provider.\n");
 758         pr_err("This is not a software error! Something bad happened to\n");
 759         pr_err("your hardware or EEPROM image. Ignoring this problem could\n");
 760         pr_err("result in further problems, possibly loss of data,\n");
 761         pr_err("corruption or system hangs!\n");
 762         pr_err("The MAC Address will be reset to 00:00:00:00:00:00,\n");
 763         pr_err("which is invalid and requires you to set the proper MAC\n");
 764         pr_err("address manually before continuing to enable this network\n");
 765         pr_err("device. Please inspect the EEPROM dump and report the\n");
 766         pr_err("issue to your hardware vendor or Intel Customer Support.\n");
 767         pr_err("/*********************/\n");
 768
 769         kfree(data);
 770 }
 771
 772 /**
 773  * e1000_is_need_ioport - determine if an adapter needs ioport resources or not
 774  * @pdev: PCI device information struct
 775  *
 776  * Return true if an adapter needs ioport resources
 777  **/
 778 static int e1000_is_need_ioport(struct pci_dev *pdev)
 779 {
 780         switch (pdev->device) {
 781         case E1000_DEV_ID_82540EM:
 782         case E1000_DEV_ID_82540EM_LOM:
 783         case E1000_DEV_ID_82540EP:
 784         case E1000_DEV_ID_82540EP_LOM:
 785         case E1000_DEV_ID_82540EP_LP:
 786         case E1000_DEV_ID_82541EI:
 787         case E1000_DEV_ID_82541EI_MOBILE:
 788         case E1000_DEV_ID_82541ER:
 789         case E1000_DEV_ID_82541ER_LOM:
 790         case E1000_DEV_ID_82541GI:
 791         case E1000_DEV_ID_82541GI_LF:
 792         case E1000_DEV_ID_82541GI_MOBILE:
 793         case E1000_DEV_ID_82544EI_COPPER:
 794         case E1000_DEV_ID_82544EI_FIBER:
 795         case E1000_DEV_ID_82544GC_COPPER:
 796         case E1000_DEV_ID_82544GC_LOM:
 797         case E1000_DEV_ID_82545EM_COPPER:
 798         case E1000_DEV_ID_82545EM_FIBER:
 799         case E1000_DEV_ID_82546EB_COPPER:
 800         case E1000_DEV_ID_82546EB_FIBER:
 801         case E1000_DEV_ID_82546EB_QUAD_COPPER:
 802                 return true;
 803         default:
 804                 return false;
 805         }
 806 }
 807
 808 static netdev_features_t e1000_fix_features(struct net_device *netdev,
 809         netdev_features_t features)
 810 {
 811         /* Since there is no support for separate Rx/Tx vlan accel
 812          * enable/disable make sure Tx flag is always in same state as Rx.
 813          */
 814         if (features & NETIF_F_HW_VLAN_CTAG_RX)
 815                 features |= NETIF_F_HW_VLAN_CTAG_TX;
 816         else
 817                 features &= ~NETIF_F_HW_VLAN_CTAG_TX;
 818
 819         return features;
 820 }
 821
 822 static int e1000_set_features(struct net_device *netdev,
 823         netdev_features_t features)
 824 {
 825         struct e1000_adapter *adapter = netdev_priv(netdev);
 826         netdev_features_t changed = features ^ netdev->features;
 827
 828         if (changed & NETIF_F_HW_VLAN_CTAG_RX)
 829                 e1000_vlan_mode(netdev, features);
 830
 831         if (!(changed & (NETIF_F_RXCSUM | NETIF_F_RXALL)))
 832                 return 0;
 833
 834         netdev->features = features;
 835         adapter->rx_csum = !!(features & NETIF_F_RXCSUM);
 836
 837         if (netif_running(netdev))
 838                 e1000_reinit_locked(adapter);
 839         else
 840                 e1000_reset(adapter);
 841
 842         return 0;
 843 }
 844
 845 static const struct net_device_ops e1000_netdev_ops = {
 846         .ndo_open               = e1000_open,
 847         .ndo_stop               = e1000_close,
 848         .ndo_start_xmit         = e1000_xmit_frame,
 849         .ndo_get_stats          = e1000_get_stats,
 850         .ndo_set_rx_mode        = e1000_set_rx_mode,
 851         .ndo_set_mac_address    = e1000_set_mac,
 852         .ndo_tx_timeout         = e1000_tx_timeout,
 853         .ndo_change_mtu         = e1000_change_mtu,
 854         .ndo_do_ioctl           = e1000_ioctl,
 855         .ndo_validate_addr      = eth_validate_addr,
 856         .ndo_vlan_rx_add_vid    = e1000_vlan_rx_add_vid,
 857         .ndo_vlan_rx_kill_vid   = e1000_vlan_rx_kill_vid,
 858 #ifdef CONFIG_NET_POLL_CONTROLLER
 859         .ndo_poll_controller    = e1000_netpoll,
 860 #endif
 861         .ndo_fix_features       = e1000_fix_features,
 862         .ndo_set_features       = e1000_set_features,
 863 };
 864
 865 /**
 866  * e1000_init_hw_struct - initialize members of hw struct
 867  * @adapter: board private struct
 868  * @hw: structure used by e1000_hw.c
 869  *
 870  * Factors out initialization of the e1000_hw struct to its own function
 871  * that can be called very early at init (just after struct allocation).
 872  * Fields are initialized based on PCI device information and
 873  * OS network device settings (MTU size).
 874  * Returns negative error codes if MAC type setup fails.
 875  */
 876 static int e1000_init_hw_struct(struct e1000_adapter *adapter,
 877                                 struct e1000_hw *hw)
 878 {
 879         struct pci_dev *pdev = adapter->pdev;
 880
 881         /* PCI config space info */
 882         hw->vendor_id = pdev->vendor;
 883         hw->device_id = pdev->device;
 884         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 885         hw->subsystem_id = pdev->subsystem_device;
 886         hw->revision_id = pdev->revision;
 887
 888         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
 889
 890         hw->max_frame_size = adapter->netdev->mtu +
 891                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
 892         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
 893
 894         /* identify the MAC */
 895         if (e1000_set_mac_type(hw)) {
 896                 e_err(probe, "Unknown MAC Type\n");
 897                 return -EIO;
 898         }
 899
 900         switch (hw->mac_type) {
 901         default:
 902                 break;
 903         case e1000_82541:
 904         case e1000_82547:
 905         case e1000_82541_rev_2:
 906         case e1000_82547_rev_2:
 907                 hw->phy_init_script = 1;
 908                 break;
 909         }
 910
 911         e1000_set_media_type(hw);
 912         e1000_get_bus_info(hw);
 913
 914         hw->wait_autoneg_complete = false;
 915         hw->tbi_compatibility_en = true;
 916         hw->adaptive_ifs = true;
 917
 918         /* Copper options */
 919
 920         if (hw->media_type == e1000_media_type_copper) {
 921                 hw->mdix = AUTO_ALL_MODES;
 922                 hw->disable_polarity_correction = false;
 923                 hw->master_slave = E1000_MASTER_SLAVE;
 924         }
 925
 926         return 0;
 927 }
 928
 929 /**
 930  * e1000_probe - Device Initialization Routine
 931  * @pdev: PCI device information struct
 932  * @ent: entry in e1000_pci_tbl
 933  *
 934  * Returns 0 on success, negative on failure
 935  *
 936  * e1000_probe initializes an adapter identified by a pci_dev structure.
 937  * The OS initialization, configuring of the adapter private structure,
 938  * and a hardware reset occur.
 939  **/
 940 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
 941 {
 942         struct net_device *netdev;
 943         struct e1000_adapter *adapter;
 944         struct e1000_hw *hw;
 945
 946         static int cards_found = 0;
 947         static int global_quad_port_a = 0; /* global ksp3 port a indication */
 948         int i, err, pci_using_dac;
 949         u16 eeprom_data = 0;
 950         u16 tmp = 0;
 951         u16 eeprom_apme_mask = E1000_EEPROM_APME;
 952         int bars, need_ioport;
 953
 954         /* do not allocate ioport bars when not needed */
 955         need_ioport = e1000_is_need_ioport(pdev);
 956         if (need_ioport) {
 957                 bars = pci_select_bars(pdev, IORESOURCE_MEM | IORESOURCE_IO);
 958                 err = pci_enable_device(pdev);
 959         } else {
 960                 bars = pci_select_bars(pdev, IORESOURCE_MEM);
 961                 err = pci_enable_device_mem(pdev);
 962         }
 963         if (err)
 964                 return err;
 965
 966         err = pci_request_selected_regions(pdev, bars, e1000_driver_name);
 967         if (err)
 968                 goto err_pci_reg;
 969
 970         pci_set_master(pdev);
 971         err = pci_save_state(pdev);
 972         if (err)
 973                 goto err_alloc_etherdev;
 974
 975         err = -ENOMEM;
 976         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
 977         if (!netdev)
 978                 goto err_alloc_etherdev;
 979
 980         SET_NETDEV_DEV(netdev, &pdev->dev);
 981
 982         pci_set_drvdata(pdev, netdev);
 983         adapter = netdev_priv(netdev);
 984         adapter->netdev = netdev;
 985         adapter->pdev = pdev;
 986         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
 987         adapter->bars = bars;
 988         adapter->need_ioport = need_ioport;
 989
 990         hw = &adapter->hw;
 991         hw->back = adapter;
 992
 993         err = -EIO;
 994         hw->hw_addr = pci_ioremap_bar(pdev, BAR_0);
 995         if (!hw->hw_addr)
 996                 goto err_ioremap;
 997
 998         if (adapter->need_ioport) {
 999                 for (i = BAR_1; i <= BAR_5; i++) {
1000                         if (pci_resource_len(pdev, i) == 0)
1001                                 continue;
1002                         if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
1003                                 hw->io_base = pci_resource_start(pdev, i);
1004                                 break;
1005                         }
1006                 }
1007         }
1008
1009         /* make ready for any if (hw->...) below */
1010         err = e1000_init_hw_struct(adapter, hw);
1011         if (err)
1012                 goto err_sw_init;
1013
1014         /* there is a workaround being applied below that limits
1015          * 64-bit DMA addresses to 64-bit hardware.  There are some
1016          * 32-bit adapters that Tx hang when given 64-bit DMA addresses
1017          */
1018         pci_using_dac = 0;
1019         if ((hw->bus_type == e1000_bus_type_pcix) &&
1020             !dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
1021                 pci_using_dac = 1;
1022         } else {
1023                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
1024                 if (err) {
1025                         pr_err("No usable DMA config, aborting\n");
1026                         goto err_dma;
1027                 }
1028         }
1029
1030         netdev->netdev_ops = &e1000_netdev_ops;
1031         e1000_set_ethtool_ops(netdev);
1032         netdev->watchdog_timeo = 5 * HZ;
1033         netif_napi_add(netdev, &adapter->napi, e1000_clean, 64);
1034
1035         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1036
1037         adapter->bd_number = cards_found;
1038
1039         /* setup the private structure */
1040
1041         err = e1000_sw_init(adapter);
1042         if (err)
1043                 goto err_sw_init;
1044
1045         err = -EIO;
1046         if (hw->mac_type == e1000_ce4100) {
1047                 hw->ce4100_gbe_mdio_base_virt =
1048                                         ioremap(pci_resource_start(pdev, BAR_1),
1049                                                 pci_resource_len(pdev, BAR_1));
1050
1051                 if (!hw->ce4100_gbe_mdio_base_virt)
1052                         goto err_mdio_ioremap;
1053         }
1054
1055         if (hw->mac_type >= e1000_82543) {
1056                 netdev->hw_features = NETIF_F_SG |
1057                                    NETIF_F_HW_CSUM |
1058                                    NETIF_F_HW_VLAN_CTAG_RX;
1059                 netdev->features = NETIF_F_HW_VLAN_CTAG_TX |
1060                                    NETIF_F_HW_VLAN_CTAG_FILTER;
1061         }
1062
1063         if ((hw->mac_type >= e1000_82544) &&
1064            (hw->mac_type != e1000_82547))
1065                 netdev->hw_features |= NETIF_F_TSO;
1066
1067         netdev->priv_flags |= IFF_SUPP_NOFCS;
1068
1069         netdev->features |= netdev->hw_features;
1070         netdev->hw_features |= (NETIF_F_RXCSUM |
1071                                 NETIF_F_RXALL |
1072                                 NETIF_F_RXFCS);
1073
1074         if (pci_using_dac) {
1075                 netdev->features |= NETIF_F_HIGHDMA;
1076                 netdev->vlan_features |= NETIF_F_HIGHDMA;
1077         }
1078
1079         netdev->vlan_features |= (NETIF_F_TSO |
1080                                   NETIF_F_HW_CSUM |
1081                                   NETIF_F_SG);
1082
1083         /* Do not set IFF_UNICAST_FLT for VMWare's 82545EM */
1084         if (hw->device_id != E1000_DEV_ID_82545EM_COPPER ||
1085             hw->subsystem_vendor_id != PCI_VENDOR_ID_VMWARE)
1086                 netdev->priv_flags |= IFF_UNICAST_FLT;
1087
1088         adapter->en_mng_pt = e1000_enable_mng_pass_thru(hw);
1089
1090         /* initialize eeprom parameters */
1091         if (e1000_init_eeprom_params(hw)) {
1092                 e_err(probe, "EEPROM initialization failed\n");
1093                 goto err_eeprom;
1094         }
1095
1096         /* before reading the EEPROM, reset the controller to
1097          * put the device in a known good starting state
1098          */
1099
1100         e1000_reset_hw(hw);
1101
1102         /* make sure the EEPROM is good */
1103         if (e1000_validate_eeprom_checksum(hw) < 0) {
1104                 e_err(probe, "The EEPROM Checksum Is Not Valid\n");
1105                 e1000_dump_eeprom(adapter);
1106                 /* set MAC address to all zeroes to invalidate and temporary
1107                  * disable this device for the user. This blocks regular
1108                  * traffic while still permitting ethtool ioctls from reaching
1109                  * the hardware as well as allowing the user to run the
1110                  * interface after manually setting a hw addr using
1111                  * `ip set address`
1112                  */
1113                 memset(hw->mac_addr, 0, netdev->addr_len);
1114         } else {
1115                 /* copy the MAC address out of the EEPROM */
1116                 if (e1000_read_mac_addr(hw))
1117                         e_err(probe, "EEPROM Read Error\n");
1118         }
1119         /* don't block initialization here due to bad MAC address */
1120         memcpy(netdev->dev_addr, hw->mac_addr, netdev->addr_len);
1121
1122         if (!is_valid_ether_addr(netdev->dev_addr))
1123                 e_err(probe, "Invalid MAC Address\n");
1124
1125
1126         INIT_DELAYED_WORK(&adapter->watchdog_task, e1000_watchdog);
1127         INIT_DELAYED_WORK(&adapter->fifo_stall_task,
1128                           e1000_82547_tx_fifo_stall_task);
1129         INIT_DELAYED_WORK(&adapter->phy_info_task, e1000_update_phy_info_task);
1130         INIT_WORK(&adapter->reset_task, e1000_reset_task);
1131
1132         e1000_check_options(adapter);
1133
1134         /* Initial Wake on LAN setting
1135          * If APM wake is enabled in the EEPROM,
1136          * enable the ACPI Magic Packet filter
1137          */
1138
1139         switch (hw->mac_type) {
1140         case e1000_82542_rev2_0:
1141         case e1000_82542_rev2_1:
1142         case e1000_82543:
1143                 break;
1144         case e1000_82544:
1145                 e1000_read_eeprom(hw,
1146                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
1147                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
1148                 break;
1149         case e1000_82546:
1150         case e1000_82546_rev_3:
1151                 if (er32(STATUS) & E1000_STATUS_FUNC_1){
1152                         e1000_read_eeprom(hw,
1153                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1154                         break;
1155                 }
1156                 /* Fall Through */
1157         default:
1158                 e1000_read_eeprom(hw,
1159                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1160                 break;
1161         }
1162         if (eeprom_data & eeprom_apme_mask)
1163                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1164
1165         /* now that we have the eeprom settings, apply the special cases
1166          * where the eeprom may be wrong or the board simply won't support
1167          * wake on lan on a particular port
1168          */
1169         switch (pdev->device) {
1170         case E1000_DEV_ID_82546GB_PCIE:
1171                 adapter->eeprom_wol = 0;
1172                 break;
1173         case E1000_DEV_ID_82546EB_FIBER:
1174         case E1000_DEV_ID_82546GB_FIBER:
1175                 /* Wake events only supported on port A for dual fiber
1176                  * regardless of eeprom setting
1177                  */
1178                 if (er32(STATUS) & E1000_STATUS_FUNC_1)
1179                         adapter->eeprom_wol = 0;
1180                 break;
1181         case E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3:
1182                 /* if quad port adapter, disable WoL on all but port A */
1183                 if (global_quad_port_a != 0)
1184                         adapter->eeprom_wol = 0;
1185                 else
1186                         adapter->quad_port_a = true;
1187                 /* Reset for multiple quad port adapters */
1188                 if (++global_quad_port_a == 4)
1189                         global_quad_port_a = 0;
1190                 break;
1191         }
1192
1193         /* initialize the wol settings based on the eeprom settings */
1194         adapter->wol = adapter->eeprom_wol;
1195         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1196
1197         /* Auto detect PHY address */
1198         if (hw->mac_type == e1000_ce4100) {
1199                 for (i = 0; i < 32; i++) {
1200                         hw->phy_addr = i;
1201                         e1000_read_phy_reg(hw, PHY_ID2, &tmp);
1202                         if (tmp == 0 || tmp == 0xFF) {
1203                                 if (i == 31)
1204                                         goto err_eeprom;
1205                                 continue;
1206                         } else
1207                                 break;
1208                 }
1209         }
1210
1211         /* reset the hardware with the new settings */
1212         e1000_reset(adapter);
1213
1214         strcpy(netdev->name, "eth%d");
1215         err = register_netdev(netdev);
1216         if (err)
1217                 goto err_register;
1218
1219         e1000_vlan_filter_on_off(adapter, false);
1220
1221         /* print bus type/speed/width info */
1222         e_info(probe, "(PCI%s:%dMHz:%d-bit) %pM\n",
1223                ((hw->bus_type == e1000_bus_type_pcix) ? "-X" : ""),
1224                ((hw->bus_speed == e1000_bus_speed_133) ? 133 :
1225                 (hw->bus_speed == e1000_bus_speed_120) ? 120 :
1226                 (hw->bus_speed == e1000_bus_speed_100) ? 100 :
1227                 (hw->bus_speed == e1000_bus_speed_66) ? 66 : 33),
1228                ((hw->bus_width == e1000_bus_width_64) ? 64 : 32),
1229                netdev->dev_addr);
1230
1231         /* carrier off reporting is important to ethtool even BEFORE open */
1232         netif_carrier_off(netdev);
1233
1234         e_info(probe, "Intel(R) PRO/1000 Network Connection\n");
1235
1236         cards_found++;
1237         return 0;
1238
1239 err_register:
1240 err_eeprom:
1241         e1000_phy_hw_reset(hw);
1242
1243         if (hw->flash_address)
1244                 iounmap(hw->flash_address);
1245         kfree(adapter->tx_ring);
1246         kfree(adapter->rx_ring);
1247 err_dma:
1248 err_sw_init:
1249 err_mdio_ioremap:
1250         iounmap(hw->ce4100_gbe_mdio_base_virt);
1251         iounmap(hw->hw_addr);
1252 err_ioremap:
1253         free_netdev(netdev);
1254 err_alloc_etherdev:
1255         pci_release_selected_regions(pdev, bars);
1256 err_pci_reg:
1257         pci_disable_device(pdev);
1258         return err;
1259 }
1260
1261 /**
1262  * e1000_remove - Device Removal Routine
1263  * @pdev: PCI device information struct
1264  *
1265  * e1000_remove is called by the PCI subsystem to alert the driver
1266  * that it should release a PCI device.  The could be caused by a
1267  * Hot-Plug event, or because the driver is going to be removed from
1268  * memory.
1269  **/
1270 static void e1000_remove(struct pci_dev *pdev)
1271 {
1272         struct net_device *netdev = pci_get_drvdata(pdev);
1273         struct e1000_adapter *adapter = netdev_priv(netdev);
1274         struct e1000_hw *hw = &adapter->hw;
1275
1276         e1000_down_and_stop(adapter);
1277         e1000_release_manageability(adapter);
1278
1279         unregister_netdev(netdev);
1280
1281         e1000_phy_hw_reset(hw);
1282
1283         kfree(adapter->tx_ring);
1284         kfree(adapter->rx_ring);
1285
1286         if (hw->mac_type == e1000_ce4100)
1287                 iounmap(hw->ce4100_gbe_mdio_base_virt);
1288         iounmap(hw->hw_addr);
1289         if (hw->flash_address)
1290                 iounmap(hw->flash_address);
1291         pci_release_selected_regions(pdev, adapter->bars);
1292
1293         free_netdev(netdev);
1294
1295         pci_disable_device(pdev);
1296 }
1297
1298 /**
1299  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
1300  * @adapter: board private structure to initialize
1301  *
1302  * e1000_sw_init initializes the Adapter private data structure.
1303  * e1000_init_hw_struct MUST be called before this function
1304  **/
1305 static int e1000_sw_init(struct e1000_adapter *adapter)
1306 {
1307         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1308
1309         adapter->num_tx_queues = 1;
1310         adapter->num_rx_queues = 1;
1311
1312         if (e1000_alloc_queues(adapter)) {
1313                 e_err(probe, "Unable to allocate memory for queues\n");
1314                 return -ENOMEM;
1315         }
1316
1317         /* Explicitly disable IRQ since the NIC can be in any state. */
1318         e1000_irq_disable(adapter);
1319
1320         spin_lock_init(&adapter->stats_lock);
1321
1322         set_bit(__E1000_DOWN, &adapter->flags);
1323
1324         return 0;
1325 }
1326
1327 /**
1328  * e1000_alloc_queues - Allocate memory for all rings
1329  * @adapter: board private structure to initialize
1330  *
1331  * We allocate one ring per queue at run-time since we don't know the
1332  * number of queues at compile-time.
1333  **/
1334 static int e1000_alloc_queues(struct e1000_adapter *adapter)
1335 {
1336         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1337                                    sizeof(struct e1000_tx_ring), GFP_KERNEL);
1338         if (!adapter->tx_ring)
1339                 return -ENOMEM;
1340
1341         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1342                                    sizeof(struct e1000_rx_ring), GFP_KERNEL);
1343         if (!adapter->rx_ring) {
1344                 kfree(adapter->tx_ring);
1345                 return -ENOMEM;
1346         }
1347
1348         return E1000_SUCCESS;
1349 }
1350
1351 /**
1352  * e1000_open - Called when a network interface is made active
1353  * @netdev: network interface device structure
1354  *
1355  * Returns 0 on success, negative value on failure
1356  *
1357  * The open entry point is called when a network interface is made
1358  * active by the system (IFF_UP).  At this point all resources needed
1359  * for transmit and receive operations are allocated, the interrupt
1360  * handler is registered with the OS, the watchdog task is started,
1361  * and the stack is notified that the interface is ready.
1362  **/
1363 static int e1000_open(struct net_device *netdev)
1364 {
1365         struct e1000_adapter *adapter = netdev_priv(netdev);
1366         struct e1000_hw *hw = &adapter->hw;
1367         int err;
1368
1369         /* disallow open during test */
1370         if (test_bit(__E1000_TESTING, &adapter->flags))
1371                 return -EBUSY;
1372
1373         netif_carrier_off(netdev);
1374
1375         /* allocate transmit descriptors */
1376         err = e1000_setup_all_tx_resources(adapter);
1377         if (err)
1378                 goto err_setup_tx;
1379
1380         /* allocate receive descriptors */
1381         err = e1000_setup_all_rx_resources(adapter);
1382         if (err)
1383                 goto err_setup_rx;
1384
1385         e1000_power_up_phy(adapter);
1386
1387         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1388         if ((hw->mng_cookie.status &
1389                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1390                 e1000_update_mng_vlan(adapter);
1391         }
1392
1393         /* before we allocate an interrupt, we must be ready to handle it.
1394          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1395          * as soon as we call pci_request_irq, so we have to setup our
1396          * clean_rx handler before we do so.
1397          */
1398         e1000_configure(adapter);
1399
1400         err = e1000_request_irq(adapter);
1401         if (err)
1402                 goto err_req_irq;
1403
1404         /* From here on the code is the same as e1000_up() */
1405         clear_bit(__E1000_DOWN, &adapter->flags);
1406
1407         napi_enable(&adapter->napi);
1408
1409         e1000_irq_enable(adapter);
1410
1411         netif_start_queue(netdev);
1412
1413         /* fire a link status change interrupt to start the watchdog */
1414         ew32(ICS, E1000_ICS_LSC);
1415
1416         return E1000_SUCCESS;
1417
1418 err_req_irq:
1419         e1000_power_down_phy(adapter);
1420         e1000_free_all_rx_resources(adapter);
1421 err_setup_rx:
1422         e1000_free_all_tx_resources(adapter);
1423 err_setup_tx:
1424         e1000_reset(adapter);
1425
1426         return err;
1427 }
1428
1429 /**
1430  * e1000_close - Disables a network interface
1431  * @netdev: network interface device structure
1432  *
1433  * Returns 0, this is not allowed to fail
1434  *
1435  * The close entry point is called when an interface is de-activated
1436  * by the OS.  The hardware is still under the drivers control, but
1437  * needs to be disabled.  A global MAC reset is issued to stop the
1438  * hardware, and all transmit and receive resources are freed.
1439  **/
1440 static int e1000_close(struct net_device *netdev)
1441 {
1442         struct e1000_adapter *adapter = netdev_priv(netdev);
1443         struct e1000_hw *hw = &adapter->hw;
1444         int count = E1000_CHECK_RESET_COUNT;
1445
1446         while (test_bit(__E1000_RESETTING, &adapter->flags) && count--)
1447                 usleep_range(10000, 20000);
1448
1449         WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
1450         e1000_down(adapter);
1451         e1000_power_down_phy(adapter);
1452         e1000_free_irq(adapter);
1453
1454         e1000_free_all_tx_resources(adapter);
1455         e1000_free_all_rx_resources(adapter);
1456
1457         /* kill manageability vlan ID if supported, but not if a vlan with
1458          * the same ID is registered on the host OS (let 8021q kill it)
1459          */
1460         if ((hw->mng_cookie.status &
1461              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
1462             !test_bit(adapter->mng_vlan_id, adapter->active_vlans)) {
1463                 e1000_vlan_rx_kill_vid(netdev, htons(ETH_P_8021Q),
1464                                        adapter->mng_vlan_id);
1465         }
1466
1467         return 0;
1468 }
1469
1470 /**
1471  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1472  * @adapter: address of board private structure
1473  * @start: address of beginning of memory
1474  * @len: length of memory
1475  **/
1476 static bool e1000_check_64k_bound(struct e1000_adapter *adapter, void *start,
1477                                   unsigned long len)
1478 {
1479         struct e1000_hw *hw = &adapter->hw;
1480         unsigned long begin = (unsigned long)start;
1481         unsigned long end = begin + len;
1482
1483         /* First rev 82545 and 82546 need to not allow any memory
1484          * write location to cross 64k boundary due to errata 23
1485          */
1486         if (hw->mac_type == e1000_82545 ||
1487             hw->mac_type == e1000_ce4100 ||
1488             hw->mac_type == e1000_82546) {
1489                 return ((begin ^ (end - 1)) >> 16) != 0 ? false : true;
1490         }
1491
1492         return true;
1493 }
1494
1495 /**
1496  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1497  * @adapter: board private structure
1498  * @txdr:    tx descriptor ring (for a specific queue) to setup
1499  *
1500  * Return 0 on success, negative on failure
1501  **/
1502 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
1503                                     struct e1000_tx_ring *txdr)
1504 {
1505         struct pci_dev *pdev = adapter->pdev;
1506         int size;
1507
1508         size = sizeof(struct e1000_tx_buffer) * txdr->count;
1509         txdr->buffer_info = vzalloc(size);
1510         if (!txdr->buffer_info)
1511                 return -ENOMEM;
1512
1513         /* round up to nearest 4K */
1514
1515         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1516         txdr->size = ALIGN(txdr->size, 4096);
1517
1518         txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size, &txdr->dma,
1519                                         GFP_KERNEL);
1520         if (!txdr->desc) {
1521 setup_tx_desc_die:
1522                 vfree(txdr->buffer_info);
1523                 return -ENOMEM;
1524         }
1525
1526         /* Fix for errata 23, can't cross 64kB boundary */
1527         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1528                 void *olddesc = txdr->desc;
1529                 dma_addr_t olddma = txdr->dma;
1530                 e_err(tx_err, "txdr align check failed: %u bytes at %p\n",
1531                       txdr->size, txdr->desc);
1532                 /* Try again, without freeing the previous */
1533                 txdr->desc = dma_alloc_coherent(&pdev->dev, txdr->size,
1534                                                 &txdr->dma, GFP_KERNEL);
1535                 /* Failed allocation, critical failure */
1536                 if (!txdr->desc) {
1537                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1538                                           olddma);
1539                         goto setup_tx_desc_die;
1540                 }
1541
1542                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1543                         /* give up */
1544                         dma_free_coherent(&pdev->dev, txdr->size, txdr->desc,
1545                                           txdr->dma);
1546                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1547                                           olddma);
1548                         e_err(probe, "Unable to allocate aligned memory "
1549                               "for the transmit descriptor ring\n");
1550                         vfree(txdr->buffer_info);
1551                         return -ENOMEM;
1552                 } else {
1553                         /* Free old allocation, new allocation was successful */
1554                         dma_free_coherent(&pdev->dev, txdr->size, olddesc,
1555                                           olddma);
1556                 }
1557         }
1558         memset(txdr->desc, 0, txdr->size);
1559
1560         txdr->next_to_use = 0;
1561         txdr->next_to_clean = 0;
1562
1563         return 0;
1564 }
1565
1566 /**
1567  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1568  *                                (Descriptors) for all queues
1569  * @adapter: board private structure
1570  *
1571  * Return 0 on success, negative on failure
1572  **/
1573 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1574 {
1575         int i, err = 0;
1576
1577         for (i = 0; i < adapter->num_tx_queues; i++) {
1578                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1579                 if (err) {
1580                         e_err(probe, "Allocation for Tx Queue %u failed\n", i);
1581                         for (i-- ; i >= 0; i--)
1582                                 e1000_free_tx_resources(adapter,
1583                                                         &adapter->tx_ring[i]);
1584                         break;
1585                 }
1586         }
1587
1588         return err;
1589 }
1590
1591 /**
1592  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1593  * @adapter: board private structure
1594  *
1595  * Configure the Tx unit of the MAC after a reset.
1596  **/
1597 static void e1000_configure_tx(struct e1000_adapter *adapter)
1598 {
1599         u64 tdba;
1600         struct e1000_hw *hw = &adapter->hw;
1601         u32 tdlen, tctl, tipg;
1602         u32 ipgr1, ipgr2;
1603
1604         /* Setup the HW Tx Head and Tail descriptor pointers */
1605
1606         switch (adapter->num_tx_queues) {
1607         case 1:
1608         default:
1609                 tdba = adapter->tx_ring[0].dma;
1610                 tdlen = adapter->tx_ring[0].count *
1611                         sizeof(struct e1000_tx_desc);
1612                 ew32(TDLEN, tdlen);
1613                 ew32(TDBAH, (tdba >> 32));
1614                 ew32(TDBAL, (tdba & 0x00000000ffffffffULL));
1615                 ew32(TDT, 0);
1616                 ew32(TDH, 0);
1617                 adapter->tx_ring[0].tdh = ((hw->mac_type >= e1000_82543) ?
1618                                            E1000_TDH : E1000_82542_TDH);
1619                 adapter->tx_ring[0].tdt = ((hw->mac_type >= e1000_82543) ?
1620                                            E1000_TDT : E1000_82542_TDT);
1621                 break;
1622         }
1623
1624         /* Set the default values for the Tx Inter Packet Gap timer */
1625         if ((hw->media_type == e1000_media_type_fiber ||
1626              hw->media_type == e1000_media_type_internal_serdes))
1627                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1628         else
1629                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1630
1631         switch (hw->mac_type) {
1632         case e1000_82542_rev2_0:
1633         case e1000_82542_rev2_1:
1634                 tipg = DEFAULT_82542_TIPG_IPGT;
1635                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1636                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1637                 break;
1638         default:
1639                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1640                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1641                 break;
1642         }
1643         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1644         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1645         ew32(TIPG, tipg);
1646
1647         /* Set the Tx Interrupt Delay register */
1648
1649         ew32(TIDV, adapter->tx_int_delay);
1650         if (hw->mac_type >= e1000_82540)
1651                 ew32(TADV, adapter->tx_abs_int_delay);
1652
1653         /* Program the Transmit Control Register */
1654
1655         tctl = er32(TCTL);
1656         tctl &= ~E1000_TCTL_CT;
1657         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1658                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1659
1660         e1000_config_collision_dist(hw);
1661
1662         /* Setup Transmit Descriptor Settings for eop descriptor */
1663         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_IFCS;
1664
1665         /* only set IDE if we are delaying interrupts using the timers */
1666         if (adapter->tx_int_delay)
1667                 adapter->txd_cmd |= E1000_TXD_CMD_IDE;
1668
1669         if (hw->mac_type < e1000_82543)
1670                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1671         else
1672                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1673
1674         /* Cache if we're 82544 running in PCI-X because we'll
1675          * need this to apply a workaround later in the send path.
1676          */
1677         if (hw->mac_type == e1000_82544 &&
1678             hw->bus_type == e1000_bus_type_pcix)
1679                 adapter->pcix_82544 = true;
1680
1681         ew32(TCTL, tctl);
1682
1683 }
1684
1685 /**
1686  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1687  * @adapter: board private structure
1688  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1689  *
1690  * Returns 0 on success, negative on failure
1691  **/
1692 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
1693                                     struct e1000_rx_ring *rxdr)
1694 {
1695         struct pci_dev *pdev = adapter->pdev;
1696         int size, desc_len;
1697
1698         size = sizeof(struct e1000_rx_buffer) * rxdr->count;
1699         rxdr->buffer_info = vzalloc(size);
1700         if (!rxdr->buffer_info)
1701                 return -ENOMEM;
1702
1703         desc_len = sizeof(struct e1000_rx_desc);
1704
1705         /* Round up to nearest 4K */
1706
1707         rxdr->size = rxdr->count * desc_len;
1708         rxdr->size = ALIGN(rxdr->size, 4096);
1709
1710         rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size, &rxdr->dma,
1711                                         GFP_KERNEL);
1712         if (!rxdr->desc) {
1713 setup_rx_desc_die:
1714                 vfree(rxdr->buffer_info);
1715                 return -ENOMEM;
1716         }
1717
1718         /* Fix for errata 23, can't cross 64kB boundary */
1719         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1720                 void *olddesc = rxdr->desc;
1721                 dma_addr_t olddma = rxdr->dma;
1722                 e_err(rx_err, "rxdr align check failed: %u bytes at %p\n",
1723                       rxdr->size, rxdr->desc);
1724                 /* Try again, without freeing the previous */
1725                 rxdr->desc = dma_alloc_coherent(&pdev->dev, rxdr->size,
1726                                                 &rxdr->dma, GFP_KERNEL);
1727                 /* Failed allocation, critical failure */
1728                 if (!rxdr->desc) {
1729                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1730                                           olddma);
1731                         goto setup_rx_desc_die;
1732                 }
1733
1734                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1735                         /* give up */
1736                         dma_free_coherent(&pdev->dev, rxdr->size, rxdr->desc,
1737                                           rxdr->dma);
1738                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1739                                           olddma);
1740                         e_err(probe, "Unable to allocate aligned memory for "
1741                               "the Rx descriptor ring\n");
1742                         goto setup_rx_desc_die;
1743                 } else {
1744                         /* Free old allocation, new allocation was successful */
1745                         dma_free_coherent(&pdev->dev, rxdr->size, olddesc,
1746                                           olddma);
1747                 }
1748         }
1749         memset(rxdr->desc, 0, rxdr->size);
1750
1751         rxdr->next_to_clean = 0;
1752         rxdr->next_to_use = 0;
1753         rxdr->rx_skb_top = NULL;
1754
1755         return 0;
1756 }
1757
1758 /**
1759  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1760  *                                (Descriptors) for all queues
1761  * @adapter: board private structure
1762  *
1763  * Return 0 on success, negative on failure
1764  **/
1765 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1766 {
1767         int i, err = 0;
1768
1769         for (i = 0; i < adapter->num_rx_queues; i++) {
1770                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1771                 if (err) {
1772                         e_err(probe, "Allocation for Rx Queue %u failed\n", i);
1773                         for (i-- ; i >= 0; i--)
1774                                 e1000_free_rx_resources(adapter,
1775                                                         &adapter->rx_ring[i]);
1776                         break;
1777                 }
1778         }
1779
1780         return err;
1781 }
1782
1783 /**
1784  * e1000_setup_rctl - configure the receive control registers
1785  * @adapter: Board private structure
1786  **/
1787 static void e1000_setup_rctl(struct e1000_adapter *adapter)
1788 {
1789         struct e1000_hw *hw = &adapter->hw;
1790         u32 rctl;
1791
1792         rctl = er32(RCTL);
1793
1794         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1795
1796         rctl |= E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
1797                 E1000_RCTL_RDMTS_HALF |
1798                 (hw->mc_filter_type << E1000_RCTL_MO_SHIFT);
1799
1800         if (hw->tbi_compatibility_on == 1)
1801                 rctl |= E1000_RCTL_SBP;
1802         else
1803                 rctl &= ~E1000_RCTL_SBP;
1804
1805         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1806                 rctl &= ~E1000_RCTL_LPE;
1807         else
1808                 rctl |= E1000_RCTL_LPE;
1809
1810         /* Setup buffer sizes */
1811         rctl &= ~E1000_RCTL_SZ_4096;
1812         rctl |= E1000_RCTL_BSEX;
1813         switch (adapter->rx_buffer_len) {
1814                 case E1000_RXBUFFER_2048:
1815                 default:
1816                         rctl |= E1000_RCTL_SZ_2048;
1817                         rctl &= ~E1000_RCTL_BSEX;
1818                         break;
1819                 case E1000_RXBUFFER_4096:
1820                         rctl |= E1000_RCTL_SZ_4096;
1821                         break;
1822                 case E1000_RXBUFFER_8192:
1823                         rctl |= E1000_RCTL_SZ_8192;
1824                         break;
1825                 case E1000_RXBUFFER_16384:
1826                         rctl |= E1000_RCTL_SZ_16384;
1827                         break;
1828         }
1829
1830         /* This is useful for sniffing bad packets. */
1831         if (adapter->netdev->features & NETIF_F_RXALL) {
1832                 /* UPE and MPE will be handled by normal PROMISC logic
1833                  * in e1000e_set_rx_mode
1834                  */
1835                 rctl |= (E1000_RCTL_SBP | /* Receive bad packets */
1836                          E1000_RCTL_BAM | /* RX All Bcast Pkts */
1837                          E1000_RCTL_PMCF); /* RX All MAC Ctrl Pkts */
1838
1839                 rctl &= ~(E1000_RCTL_VFE | /* Disable VLAN filter */
1840                           E1000_RCTL_DPF | /* Allow filtered pause */
1841                           E1000_RCTL_CFIEN); /* Dis VLAN CFIEN Filter */
1842                 /* Do not mess with E1000_CTRL_VME, it affects transmit as well,
1843                  * and that breaks VLANs.
1844                  */
1845         }
1846
1847         ew32(RCTL, rctl);
1848 }
1849
1850 /**
1851  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1852  * @adapter: board private structure
1853  *
1854  * Configure the Rx unit of the MAC after a reset.
1855  **/
1856 static void e1000_configure_rx(struct e1000_adapter *adapter)
1857 {
1858         u64 rdba;
1859         struct e1000_hw *hw = &adapter->hw;
1860         u32 rdlen, rctl, rxcsum;
1861
1862         if (adapter->netdev->mtu > ETH_DATA_LEN) {
1863                 rdlen = adapter->rx_ring[0].count *
1864                         sizeof(struct e1000_rx_desc);
1865                 adapter->clean_rx = e1000_clean_jumbo_rx_irq;
1866                 adapter->alloc_rx_buf = e1000_alloc_jumbo_rx_buffers;
1867         } else {
1868                 rdlen = adapter->rx_ring[0].count *
1869                         sizeof(struct e1000_rx_desc);
1870                 adapter->clean_rx = e1000_clean_rx_irq;
1871                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1872         }
1873
1874         /* disable receives while setting up the descriptors */
1875         rctl = er32(RCTL);
1876         ew32(RCTL, rctl & ~E1000_RCTL_EN);
1877
1878         /* set the Receive Delay Timer Register */
1879         ew32(RDTR, adapter->rx_int_delay);
1880
1881         if (hw->mac_type >= e1000_82540) {
1882                 ew32(RADV, adapter->rx_abs_int_delay);
1883                 if (adapter->itr_setting != 0)
1884                         ew32(ITR, 1000000000 / (adapter->itr * 256));
1885         }
1886
1887         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1888          * the Base and Length of the Rx Descriptor Ring
1889          */
1890         switch (adapter->num_rx_queues) {
1891         case 1:
1892         default:
1893                 rdba = adapter->rx_ring[0].dma;
1894                 ew32(RDLEN, rdlen);
1895                 ew32(RDBAH, (rdba >> 32));
1896                 ew32(RDBAL, (rdba & 0x00000000ffffffffULL));
1897                 ew32(RDT, 0);
1898                 ew32(RDH, 0);
1899                 adapter->rx_ring[0].rdh = ((hw->mac_type >= e1000_82543) ?
1900                                            E1000_RDH : E1000_82542_RDH);
1901                 adapter->rx_ring[0].rdt = ((hw->mac_type >= e1000_82543) ?
1902                                            E1000_RDT : E1000_82542_RDT);
1903                 break;
1904         }
1905
1906         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1907         if (hw->mac_type >= e1000_82543) {
1908                 rxcsum = er32(RXCSUM);
1909                 if (adapter->rx_csum)
1910                         rxcsum |= E1000_RXCSUM_TUOFL;
1911                 else
1912                         /* don't need to clear IPPCSE as it defaults to 0 */
1913                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1914                 ew32(RXCSUM, rxcsum);
1915         }
1916
1917         /* Enable Receives */
1918         ew32(RCTL, rctl | E1000_RCTL_EN);
1919 }
1920
1921 /**
1922  * e1000_free_tx_resources - Free Tx Resources per Queue
1923  * @adapter: board private structure
1924  * @tx_ring: Tx descriptor ring for a specific queue
1925  *
1926  * Free all transmit software resources
1927  **/
1928 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
1929                                     struct e1000_tx_ring *tx_ring)
1930 {
1931         struct pci_dev *pdev = adapter->pdev;
1932
1933         e1000_clean_tx_ring(adapter, tx_ring);
1934
1935         vfree(tx_ring->buffer_info);
1936         tx_ring->buffer_info = NULL;
1937
1938         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
1939                           tx_ring->dma);
1940
1941         tx_ring->desc = NULL;
1942 }
1943
1944 /**
1945  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1946  * @adapter: board private structure
1947  *
1948  * Free all transmit software resources
1949  **/
1950 void e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1951 {
1952         int i;
1953
1954         for (i = 0; i < adapter->num_tx_queues; i++)
1955                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1956 }
1957
1958 static void
1959 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1960                                  struct e1000_tx_buffer *buffer_info)
1961 {
1962         if (buffer_info->dma) {
1963                 if (buffer_info->mapped_as_page)
1964                         dma_unmap_page(&adapter->pdev->dev, buffer_info->dma,
1965                                        buffer_info->length, DMA_TO_DEVICE);
1966                 else
1967                         dma_unmap_single(&adapter->pdev->dev, buffer_info->dma,
1968                                          buffer_info->length,
1969                                          DMA_TO_DEVICE);
1970                 buffer_info->dma = 0;
1971         }
1972         if (buffer_info->skb) {
1973                 dev_kfree_skb_any(buffer_info->skb);
1974                 buffer_info->skb = NULL;
1975         }
1976         buffer_info->time_stamp = 0;
1977         /* buffer_info must be completely set up in the transmit path */
1978 }
1979
1980 /**
1981  * e1000_clean_tx_ring - Free Tx Buffers
1982  * @adapter: board private structure
1983  * @tx_ring: ring to be cleaned
1984  **/
1985 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
1986                                 struct e1000_tx_ring *tx_ring)
1987 {
1988         struct e1000_hw *hw = &adapter->hw;
1989         struct e1000_tx_buffer *buffer_info;
1990         unsigned long size;
1991         unsigned int i;
1992
1993         /* Free all the Tx ring sk_buffs */
1994
1995         for (i = 0; i < tx_ring->count; i++) {
1996                 buffer_info = &tx_ring->buffer_info[i];
1997                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1998         }
1999
2000         netdev_reset_queue(adapter->netdev);
2001         size = sizeof(struct e1000_tx_buffer) * tx_ring->count;
2002         memset(tx_ring->buffer_info, 0, size);
2003
2004         /* Zero out the descriptor ring */
2005
2006         memset(tx_ring->desc, 0, tx_ring->size);
2007
2008         tx_ring->next_to_use = 0;
2009         tx_ring->next_to_clean = 0;
2010         tx_ring->last_tx_tso = false;
2011
2012         writel(0, hw->hw_addr + tx_ring->tdh);
2013         writel(0, hw->hw_addr + tx_ring->tdt);
2014 }
2015
2016 /**
2017  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2018  * @adapter: board private structure
2019  **/
2020 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2021 {
2022         int i;
2023
2024         for (i = 0; i < adapter->num_tx_queues; i++)
2025                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2026 }
2027
2028 /**
2029  * e1000_free_rx_resources - Free Rx Resources
2030  * @adapter: board private structure
2031  * @rx_ring: ring to clean the resources from
2032  *
2033  * Free all receive software resources
2034  **/
2035 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
2036                                     struct e1000_rx_ring *rx_ring)
2037 {
2038         struct pci_dev *pdev = adapter->pdev;
2039
2040         e1000_clean_rx_ring(adapter, rx_ring);
2041
2042         vfree(rx_ring->buffer_info);
2043         rx_ring->buffer_info = NULL;
2044
2045         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2046                           rx_ring->dma);
2047
2048         rx_ring->desc = NULL;
2049 }
2050
2051 /**
2052  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2053  * @adapter: board private structure
2054  *
2055  * Free all receive software resources
2056  **/
2057 void e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2058 {
2059         int i;
2060
2061         for (i = 0; i < adapter->num_rx_queues; i++)
2062                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2063 }
2064
2065 #define E1000_HEADROOM (NET_SKB_PAD + NET_IP_ALIGN)
2066 static unsigned int e1000_frag_len(const struct e1000_adapter *a)
2067 {
2068         return SKB_DATA_ALIGN(a->rx_buffer_len + E1000_HEADROOM) +
2069                 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
2070 }
2071
2072 static void *e1000_alloc_frag(const struct e1000_adapter *a)
2073 {
2074         unsigned int len = e1000_frag_len(a);
2075         u8 *data = netdev_alloc_frag(len);
2076
2077         if (likely(data))
2078                 data += E1000_HEADROOM;
2079         return data;
2080 }
2081
2082 /**
2083  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2084  * @adapter: board private structure
2085  * @rx_ring: ring to free buffers from
2086  **/
2087 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
2088                                 struct e1000_rx_ring *rx_ring)
2089 {
2090         struct e1000_hw *hw = &adapter->hw;
2091         struct e1000_rx_buffer *buffer_info;
2092         struct pci_dev *pdev = adapter->pdev;
2093         unsigned long size;
2094         unsigned int i;
2095
2096         /* Free all the Rx netfrags */
2097         for (i = 0; i < rx_ring->count; i++) {
2098                 buffer_info = &rx_ring->buffer_info[i];
2099                 if (adapter->clean_rx == e1000_clean_rx_irq) {
2100                         if (buffer_info->dma)
2101                                 dma_unmap_single(&pdev->dev, buffer_info->dma,
2102                                                  adapter->rx_buffer_len,
2103                                                  DMA_FROM_DEVICE);
2104                         if (buffer_info->rxbuf.data) {
2105                                 skb_free_frag(buffer_info->rxbuf.data);
2106                                 buffer_info->rxbuf.data = NULL;
2107                         }
2108                 } else if (adapter->clean_rx == e1000_clean_jumbo_rx_irq) {
2109                         if (buffer_info->dma)
2110                                 dma_unmap_page(&pdev->dev, buffer_info->dma,
2111                                                adapter->rx_buffer_len,
2112                                                DMA_FROM_DEVICE);
2113                         if (buffer_info->rxbuf.page) {
2114                                 put_page(buffer_info->rxbuf.page);
2115                                 buffer_info->rxbuf.page = NULL;
2116                         }
2117                 }
2118
2119                 buffer_info->dma = 0;
2120         }
2121
2122         /* there also may be some cached data from a chained receive */
2123         napi_free_frags(&adapter->napi);
2124         rx_ring->rx_skb_top = NULL;
2125
2126         size = sizeof(struct e1000_rx_buffer) * rx_ring->count;
2127         memset(rx_ring->buffer_info, 0, size);
2128
2129         /* Zero out the descriptor ring */
2130         memset(rx_ring->desc, 0, rx_ring->size);
2131
2132         rx_ring->next_to_clean = 0;
2133         rx_ring->next_to_use = 0;
2134
2135         writel(0, hw->hw_addr + rx_ring->rdh);
2136         writel(0, hw->hw_addr + rx_ring->rdt);
2137 }
2138
2139 /**
2140  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2141  * @adapter: board private structure
2142  **/
2143 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2144 {
2145         int i;
2146
2147         for (i = 0; i < adapter->num_rx_queues; i++)
2148                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2149 }
2150
2151 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2152  * and memory write and invalidate disabled for certain operations
2153  */
2154 static void e1000_enter_82542_rst(struct e1000_adapter *adapter)
2155 {
2156         struct e1000_hw *hw = &adapter->hw;
2157         struct net_device *netdev = adapter->netdev;
2158         u32 rctl;
2159
2160         e1000_pci_clear_mwi(hw);
2161
2162         rctl = er32(RCTL);
2163         rctl |= E1000_RCTL_RST;
2164         ew32(RCTL, rctl);
2165         E1000_WRITE_FLUSH();
2166         mdelay(5);
2167
2168         if (netif_running(netdev))
2169                 e1000_clean_all_rx_rings(adapter);
2170 }
2171
2172 static void e1000_leave_82542_rst(struct e1000_adapter *adapter)
2173 {
2174         struct e1000_hw *hw = &adapter->hw;
2175         struct net_device *netdev = adapter->netdev;
2176         u32 rctl;
2177
2178         rctl = er32(RCTL);
2179         rctl &= ~E1000_RCTL_RST;
2180         ew32(RCTL, rctl);
2181         E1000_WRITE_FLUSH();
2182         mdelay(5);
2183
2184         if (hw->pci_cmd_word & PCI_COMMAND_INVALIDATE)
2185                 e1000_pci_set_mwi(hw);
2186
2187         if (netif_running(netdev)) {
2188                 /* No need to loop, because 82542 supports only 1 queue */
2189                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2190                 e1000_configure_rx(adapter);
2191                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2192         }
2193 }
2194
2195 /**
2196  * e1000_set_mac - Change the Ethernet Address of the NIC
2197  * @netdev: network interface device structure
2198  * @p: pointer to an address structure
2199  *
2200  * Returns 0 on success, negative on failure
2201  **/
2202 static int e1000_set_mac(struct net_device *netdev, void *p)
2203 {
2204         struct e1000_adapter *adapter = netdev_priv(netdev);
2205         struct e1000_hw *hw = &adapter->hw;
2206         struct sockaddr *addr = p;
2207
2208         if (!is_valid_ether_addr(addr->sa_data))
2209                 return -EADDRNOTAVAIL;
2210
2211         /* 82542 2.0 needs to be in reset to write receive address registers */
2212
2213         if (hw->mac_type == e1000_82542_rev2_0)
2214                 e1000_enter_82542_rst(adapter);
2215
2216         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2217         memcpy(hw->mac_addr, addr->sa_data, netdev->addr_len);
2218
2219         e1000_rar_set(hw, hw->mac_addr, 0);
2220
2221         if (hw->mac_type == e1000_82542_rev2_0)
2222                 e1000_leave_82542_rst(adapter);
2223
2224         return 0;
2225 }
2226
2227 /**
2228  * e1000_set_rx_mode - Secondary Unicast, Multicast and Promiscuous mode set
2229  * @netdev: network interface device structure
2230  *
2231  * The set_rx_mode entry point is called whenever the unicast or multicast
2232  * address lists or the network interface flags are updated. This routine is
2233  * responsible for configuring the hardware for proper unicast, multicast,
2234  * promiscuous mode, and all-multi behavior.
2235  **/
2236 static void e1000_set_rx_mode(struct net_device *netdev)
2237 {
2238         struct e1000_adapter *adapter = netdev_priv(netdev);
2239         struct e1000_hw *hw = &adapter->hw;
2240         struct netdev_hw_addr *ha;
2241         bool use_uc = false;
2242         u32 rctl;
2243         u32 hash_value;
2244         int i, rar_entries = E1000_RAR_ENTRIES;
2245         int mta_reg_count = E1000_NUM_MTA_REGISTERS;
2246         u32 *mcarray = kcalloc(mta_reg_count, sizeof(u32), GFP_ATOMIC);
2247
2248         if (!mcarray)
2249                 return;
2250
2251         /* Check for Promiscuous and All Multicast modes */
2252
2253         rctl = er32(RCTL);
2254
2255         if (netdev->flags & IFF_PROMISC) {
2256                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2257                 rctl &= ~E1000_RCTL_VFE;
2258         } else {
2259                 if (netdev->flags & IFF_ALLMULTI)
2260                         rctl |= E1000_RCTL_MPE;
2261                 else
2262                         rctl &= ~E1000_RCTL_MPE;
2263                 /* Enable VLAN filter if there is a VLAN */
2264                 if (e1000_vlan_used(adapter))
2265                         rctl |= E1000_RCTL_VFE;
2266         }
2267
2268         if (netdev_uc_count(netdev) > rar_entries - 1) {
2269                 rctl |= E1000_RCTL_UPE;
2270         } else if (!(netdev->flags & IFF_PROMISC)) {
2271                 rctl &= ~E1000_RCTL_UPE;
2272                 use_uc = true;
2273         }
2274
2275         ew32(RCTL, rctl);
2276
2277         /* 82542 2.0 needs to be in reset to write receive address registers */
2278
2279         if (hw->mac_type == e1000_82542_rev2_0)
2280                 e1000_enter_82542_rst(adapter);
2281
2282         /* load the first 14 addresses into the exact filters 1-14. Unicast
2283          * addresses take precedence to avoid disabling unicast filtering
2284          * when possible.
2285          *
2286          * RAR 0 is used for the station MAC address
2287          * if there are not 14 addresses, go ahead and clear the filters
2288          */
2289         i = 1;
2290         if (use_uc)
2291                 netdev_for_each_uc_addr(ha, netdev) {
2292                         if (i == rar_entries)
2293                                 break;
2294                         e1000_rar_set(hw, ha->addr, i++);
2295                 }
2296
2297         netdev_for_each_mc_addr(ha, netdev) {
2298                 if (i == rar_entries) {
2299                         /* load any remaining addresses into the hash table */
2300                         u32 hash_reg, hash_bit, mta;
2301                         hash_value = e1000_hash_mc_addr(hw, ha->addr);
2302                         hash_reg = (hash_value >> 5) & 0x7F;
2303                         hash_bit = hash_value & 0x1F;
2304                         mta = (1 << hash_bit);
2305                         mcarray[hash_reg] |= mta;
2306                 } else {
2307                         e1000_rar_set(hw, ha->addr, i++);
2308                 }
2309         }
2310
2311         for (; i < rar_entries; i++) {
2312                 E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2313                 E1000_WRITE_FLUSH();
2314                 E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2315                 E1000_WRITE_FLUSH();
2316         }
2317
2318         /* write the hash table completely, write from bottom to avoid
2319          * both stupid write combining chipsets, and flushing each write
2320          */
2321         for (i = mta_reg_count - 1; i >= 0 ; i--) {
2322                 /* If we are on an 82544 has an errata where writing odd
2323                  * offsets overwrites the previous even offset, but writing
2324                  * backwards over the range solves the issue by always
2325                  * writing the odd offset first
2326                  */
2327                 E1000_WRITE_REG_ARRAY(hw, MTA, i, mcarray[i]);
2328         }
2329         E1000_WRITE_FLUSH();
2330
2331         if (hw->mac_type == e1000_82542_rev2_0)
2332                 e1000_leave_82542_rst(adapter);
2333
2334         kfree(mcarray);
2335 }
2336
2337 /**
2338  * e1000_update_phy_info_task - get phy info
2339  * @work: work struct contained inside adapter struct
2340  *
2341  * Need to wait a few seconds after link up to get diagnostic information from
2342  * the phy
2343  */
2344 static void e1000_update_phy_info_task(struct work_struct *work)
2345 {
2346         struct e1000_adapter *adapter = container_of(work,
2347                                                      struct e1000_adapter,
2348                                                      phy_info_task.work);
2349
2350         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2351 }
2352
2353 /**
2354  * e1000_82547_tx_fifo_stall_task - task to complete work
2355  * @work: work struct contained inside adapter struct
2356  **/
2357 static void e1000_82547_tx_fifo_stall_task(struct work_struct *work)
2358 {
2359         struct e1000_adapter *adapter = container_of(work,
2360                                                      struct e1000_adapter,
2361                                                      fifo_stall_task.work);
2362         struct e1000_hw *hw = &adapter->hw;
2363         struct net_device *netdev = adapter->netdev;
2364         u32 tctl;
2365
2366         if (atomic_read(&adapter->tx_fifo_stall)) {
2367                 if ((er32(TDT) == er32(TDH)) &&
2368                    (er32(TDFT) == er32(TDFH)) &&
2369                    (er32(TDFTS) == er32(TDFHS))) {
2370                         tctl = er32(TCTL);
2371                         ew32(TCTL, tctl & ~E1000_TCTL_EN);
2372                         ew32(TDFT, adapter->tx_head_addr);
2373                         ew32(TDFH, adapter->tx_head_addr);
2374                         ew32(TDFTS, adapter->tx_head_addr);
2375                         ew32(TDFHS, adapter->tx_head_addr);
2376                         ew32(TCTL, tctl);
2377                         E1000_WRITE_FLUSH();
2378
2379                         adapter->tx_fifo_head = 0;
2380                         atomic_set(&adapter->tx_fifo_stall, 0);
2381                         netif_wake_queue(netdev);
2382                 } else if (!test_bit(__E1000_DOWN, &adapter->flags)) {
2383                         schedule_delayed_work(&adapter->fifo_stall_task, 1);
2384                 }
2385         }
2386 }
2387
2388 bool e1000_has_link(struct e1000_adapter *adapter)
2389 {
2390         struct e1000_hw *hw = &adapter->hw;
2391         bool link_active = false;
2392
2393         /* get_link_status is set on LSC (link status) interrupt or rx
2394          * sequence error interrupt (except on intel ce4100).
2395          * get_link_status will stay false until the
2396          * e1000_check_for_link establishes link for copper adapters
2397          * ONLY
2398          */
2399         switch (hw->media_type) {
2400         case e1000_media_type_copper:
2401                 if (hw->mac_type == e1000_ce4100)
2402                         hw->get_link_status = 1;
2403                 if (hw->get_link_status) {
2404                         e1000_check_for_link(hw);
2405                         link_active = !hw->get_link_status;
2406                 } else {
2407                         link_active = true;
2408                 }
2409                 break;
2410         case e1000_media_type_fiber:
2411                 e1000_check_for_link(hw);
2412                 link_active = !!(er32(STATUS) & E1000_STATUS_LU);
2413                 break;
2414         case e1000_media_type_internal_serdes:
2415                 e1000_check_for_link(hw);
2416                 link_active = hw->serdes_has_link;
2417                 break;
2418         default:
2419                 break;
2420         }
2421
2422         return link_active;
2423 }
2424
2425 /**
2426  * e1000_watchdog - work function
2427  * @work: work struct contained inside adapter struct
2428  **/
2429 static void e1000_watchdog(struct work_struct *work)
2430 {
2431         struct e1000_adapter *adapter = container_of(work,
2432                                                      struct e1000_adapter,
2433                                                      watchdog_task.work);
2434         struct e1000_hw *hw = &adapter->hw;
2435         struct net_device *netdev = adapter->netdev;
2436         struct e1000_tx_ring *txdr = adapter->tx_ring;
2437         u32 link, tctl;
2438
2439         link = e1000_has_link(adapter);
2440         if ((netif_carrier_ok(netdev)) && link)
2441                 goto link_up;
2442
2443         if (link) {
2444                 if (!netif_carrier_ok(netdev)) {
2445                         u32 ctrl;
2446                         bool txb2b = true;
2447                         /* update snapshot of PHY registers on LSC */
2448                         e1000_get_speed_and_duplex(hw,
2449                                                    &adapter->link_speed,
2450                                                    &adapter->link_duplex);
2451
2452                         ctrl = er32(CTRL);
2453                         pr_info("%s NIC Link is Up %d Mbps %s, "
2454                                 "Flow Control: %s\n",
2455                                 netdev->name,
2456                                 adapter->link_speed,
2457                                 adapter->link_duplex == FULL_DUPLEX ?
2458                                 "Full Duplex" : "Half Duplex",
2459                                 ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2460                                 E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2461                                 E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2462                                 E1000_CTRL_TFCE) ? "TX" : "None")));
2463
2464                         /* adjust timeout factor according to speed/duplex */
2465                         adapter->tx_timeout_factor = 1;
2466                         switch (adapter->link_speed) {
2467                         case SPEED_10:
2468                                 txb2b = false;
2469                                 adapter->tx_timeout_factor = 16;
2470                                 break;
2471                         case SPEED_100:
2472                                 txb2b = false;
2473                                 /* maybe add some timeout factor ? */
2474                                 break;
2475                         }
2476
2477                         /* enable transmits in the hardware */
2478                         tctl = er32(TCTL);
2479                         tctl |= E1000_TCTL_EN;
2480                         ew32(TCTL, tctl);
2481
2482                         netif_carrier_on(netdev);
2483                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2484                                 schedule_delayed_work(&adapter->phy_info_task,
2485                                                       2 * HZ);
2486                         adapter->smartspeed = 0;
2487                 }
2488         } else {
2489                 if (netif_carrier_ok(netdev)) {
2490                         adapter->link_speed = 0;
2491                         adapter->link_duplex = 0;
2492                         pr_info("%s NIC Link is Down\n",
2493                                 netdev->name);
2494                         netif_carrier_off(netdev);
2495
2496                         if (!test_bit(__E1000_DOWN, &adapter->flags))
2497                                 schedule_delayed_work(&adapter->phy_info_task,
2498                                                       2 * HZ);
2499                 }
2500
2501                 e1000_smartspeed(adapter);
2502         }
2503
2504 link_up:
2505         e1000_update_stats(adapter);
2506
2507         hw->tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2508         adapter->tpt_old = adapter->stats.tpt;
2509         hw->collision_delta = adapter->stats.colc - adapter->colc_old;
2510         adapter->colc_old = adapter->stats.colc;
2511
2512         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2513         adapter->gorcl_old = adapter->stats.gorcl;
2514         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2515         adapter->gotcl_old = adapter->stats.gotcl;
2516
2517         e1000_update_adaptive(hw);
2518
2519         if (!netif_carrier_ok(netdev)) {
2520                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2521                         /* We've lost link, so the controller stops DMA,
2522                          * but we've got queued Tx work that's never going
2523                          * to get done, so reset controller to flush Tx.
2524                          * (Do the reset outside of interrupt context).
2525                          */
2526                         adapter->tx_timeout_count++;
2527                         schedule_work(&adapter->reset_task);
2528                         /* exit immediately since reset is imminent */
2529                         return;
2530                 }
2531         }
2532
2533         /* Simple mode for Interrupt Throttle Rate (ITR) */
2534         if (hw->mac_type >= e1000_82540 && adapter->itr_setting == 4) {
2535                 /* Symmetric Tx/Rx gets a reduced ITR=2000;
2536                  * Total asymmetrical Tx or Rx gets ITR=8000;
2537                  * everyone else is between 2000-8000.
2538                  */
2539                 u32 goc = (adapter->gotcl + adapter->gorcl) / 10000;
2540                 u32 dif = (adapter->gotcl > adapter->gorcl ?
2541                             adapter->gotcl - adapter->gorcl :
2542                             adapter->gorcl - adapter->gotcl) / 10000;
2543                 u32 itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2544
2545                 ew32(ITR, 1000000000 / (itr * 256));
2546         }
2547
2548         /* Cause software interrupt to ensure rx ring is cleaned */
2549         ew32(ICS, E1000_ICS_RXDMT0);
2550
2551         /* Force detection of hung controller every watchdog period */
2552         adapter->detect_tx_hung = true;
2553
2554         /* Reschedule the task */
2555         if (!test_bit(__E1000_DOWN, &adapter->flags))
2556                 schedule_delayed_work(&adapter->watchdog_task, 2 * HZ);
2557 }
2558
2559 enum latency_range {
2560         lowest_latency = 0,
2561         low_latency = 1,
2562         bulk_latency = 2,
2563         latency_invalid = 255
2564 };
2565
2566 /**
2567  * e1000_update_itr - update the dynamic ITR value based on statistics
2568  * @adapter: pointer to adapter
2569  * @itr_setting: current adapter->itr
2570  * @packets: the number of packets during this measurement interval
2571  * @bytes: the number of bytes during this measurement interval
2572  *
2573  *      Stores a new ITR value based on packets and byte
2574  *      counts during the last interrupt.  The advantage of per interrupt
2575  *      computation is faster updates and more accurate ITR for the current
2576  *      traffic pattern.  Constants in this function were computed
2577  *      based on theoretical maximum wire speed and thresholds were set based
2578  *      on testing data as well as attempting to minimize response time
2579  *      while increasing bulk throughput.
2580  *      this functionality is controlled by the InterruptThrottleRate module
2581  *      parameter (see e1000_param.c)
2582  **/
2583 static unsigned int e1000_update_itr(struct e1000_adapter *adapter,
2584                                      u16 itr_setting, int packets, int bytes)
2585 {
2586         unsigned int retval = itr_setting;
2587         struct e1000_hw *hw = &adapter->hw;
2588
2589         if (unlikely(hw->mac_type < e1000_82540))
2590                 goto update_itr_done;
2591
2592         if (packets == 0)
2593                 goto update_itr_done;
2594
2595         switch (itr_setting) {
2596         case lowest_latency:
2597                 /* jumbo frames get bulk treatment*/
2598                 if (bytes/packets > 8000)
2599                         retval = bulk_latency;
2600                 else if ((packets < 5) && (bytes > 512))
2601                         retval = low_latency;
2602                 break;
2603         case low_latency:  /* 50 usec aka 20000 ints/s */
2604                 if (bytes > 10000) {
2605                         /* jumbo frames need bulk latency setting */
2606                         if (bytes/packets > 8000)
2607                                 retval = bulk_latency;
2608                         else if ((packets < 10) || ((bytes/packets) > 1200))
2609                                 retval = bulk_latency;
2610                         else if ((packets > 35))
2611                                 retval = lowest_latency;
2612                 } else if (bytes/packets > 2000)
2613                         retval = bulk_latency;
2614                 else if (packets <= 2 && bytes < 512)
2615                         retval = lowest_latency;
2616                 break;
2617         case bulk_latency: /* 250 usec aka 4000 ints/s */
2618                 if (bytes > 25000) {
2619                         if (packets > 35)
2620                                 retval = low_latency;
2621                 } else if (bytes < 6000) {
2622                         retval = low_latency;
2623                 }
2624                 break;
2625         }
2626
2627 update_itr_done:
2628         return retval;
2629 }
2630
2631 static void e1000_set_itr(struct e1000_adapter *adapter)
2632 {
2633         struct e1000_hw *hw = &adapter->hw;
2634         u16 current_itr;
2635         u32 new_itr = adapter->itr;
2636
2637         if (unlikely(hw->mac_type < e1000_82540))
2638                 return;
2639
2640         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2641         if (unlikely(adapter->link_speed != SPEED_1000)) {
2642                 current_itr = 0;
2643                 new_itr = 4000;
2644                 goto set_itr_now;
2645         }
2646
2647         adapter->tx_itr = e1000_update_itr(adapter, adapter->tx_itr,
2648                                            adapter->total_tx_packets,
2649                                            adapter->total_tx_bytes);
2650         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2651         if (adapter->itr_setting == 3 && adapter->tx_itr == lowest_latency)
2652                 adapter->tx_itr = low_latency;
2653
2654         adapter->rx_itr = e1000_update_itr(adapter, adapter->rx_itr,
2655                                            adapter->total_rx_packets,
2656                                            adapter->total_rx_bytes);
2657         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2658         if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
2659                 adapter->rx_itr = low_latency;
2660
2661         current_itr = max(adapter->rx_itr, adapter->tx_itr);
2662
2663         switch (current_itr) {
2664         /* counts and packets in update_itr are dependent on these numbers */
2665         case lowest_latency:
2666                 new_itr = 70000;
2667                 break;
2668         case low_latency:
2669                 new_itr = 20000; /* aka hwitr = ~200 */
2670                 break;
2671         case bulk_latency:
2672                 new_itr = 4000;
2673                 break;
2674         default:
2675                 break;
2676         }
2677
2678 set_itr_now:
2679         if (new_itr != adapter->itr) {
2680                 /* this attempts to bias the interrupt rate towards Bulk
2681                  * by adding intermediate steps when interrupt rate is
2682                  * increasing
2683                  */
2684                 new_itr = new_itr > adapter->itr ?
2685                           min(adapter->itr + (new_itr >> 2), new_itr) :
2686                           new_itr;
2687                 adapter->itr = new_itr;
2688                 ew32(ITR, 1000000000 / (new_itr * 256));
2689         }
2690 }
2691
2692 #define E1000_TX_FLAGS_CSUM             0x00000001
2693 #define E1000_TX_FLAGS_VLAN             0x00000002
2694 #define E1000_TX_FLAGS_TSO              0x00000004
2695 #define E1000_TX_FLAGS_IPV4             0x00000008
2696 #define E1000_TX_FLAGS_NO_FCS           0x00000010
2697 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2698 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2699
2700 static int e1000_tso(struct e1000_adapter *adapter,
2701                      struct e1000_tx_ring *tx_ring, struct sk_buff *skb,
2702                      __be16 protocol)
2703 {
2704         struct e1000_context_desc *context_desc;
2705         struct e1000_tx_buffer *buffer_info;
2706         unsigned int i;
2707         u32 cmd_length = 0;
2708         u16 ipcse = 0, tucse, mss;
2709         u8 ipcss, ipcso, tucss, tucso, hdr_len;
2710
2711         if (skb_is_gso(skb)) {
2712                 int err;
2713
2714                 err = skb_cow_head(skb, 0);
2715                 if (err < 0)
2716                         return err;
2717
2718                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
2719                 mss = skb_shinfo(skb)->gso_size;
2720                 if (protocol == htons(ETH_P_IP)) {
2721                         struct iphdr *iph = ip_hdr(skb);
2722                         iph->tot_len = 0;
2723                         iph->check = 0;
2724                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2725                                                                  iph->daddr, 0,
2726                                                                  IPPROTO_TCP,
2727                                                                  0);
2728                         cmd_length = E1000_TXD_CMD_IP;
2729                         ipcse = skb_transport_offset(skb) - 1;
2730                 } else if (skb_is_gso_v6(skb)) {
2731                         ipv6_hdr(skb)->payload_len = 0;
2732                         tcp_hdr(skb)->check =
2733                                 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2734                                                  &ipv6_hdr(skb)->daddr,
2735                                                  0, IPPROTO_TCP, 0);
2736                         ipcse = 0;
2737                 }
2738                 ipcss = skb_network_offset(skb);
2739                 ipcso = (void *)&(ip_hdr(skb)->check) - (void *)skb->data;
2740                 tucss = skb_transport_offset(skb);
2741                 tucso = (void *)&(tcp_hdr(skb)->check) - (void *)skb->data;
2742                 tucse = 0;
2743
2744                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2745                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2746
2747                 i = tx_ring->next_to_use;
2748                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2749                 buffer_info = &tx_ring->buffer_info[i];
2750
2751                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2752                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2753                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2754                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2755                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2756                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2757                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2758                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2759                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2760
2761                 buffer_info->time_stamp = jiffies;
2762                 buffer_info->next_to_watch = i;
2763
2764                 if (++i == tx_ring->count) i = 0;
2765                 tx_ring->next_to_use = i;
2766
2767                 return true;
2768         }
2769         return false;
2770 }
2771
2772 static bool e1000_tx_csum(struct e1000_adapter *adapter,
2773                           struct e1000_tx_ring *tx_ring, struct sk_buff *skb,
2774                           __be16 protocol)
2775 {
2776         struct e1000_context_desc *context_desc;
2777         struct e1000_tx_buffer *buffer_info;
2778         unsigned int i;
2779         u8 css;
2780         u32 cmd_len = E1000_TXD_CMD_DEXT;
2781
2782         if (skb->ip_summed != CHECKSUM_PARTIAL)
2783                 return false;
2784
2785         switch (protocol) {
2786         case cpu_to_be16(ETH_P_IP):
2787                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2788                         cmd_len |= E1000_TXD_CMD_TCP;
2789                 break;
2790         case cpu_to_be16(ETH_P_IPV6):
2791                 /* XXX not handling all IPV6 headers */
2792                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2793                         cmd_len |= E1000_TXD_CMD_TCP;
2794                 break;
2795         default:
2796                 if (unlikely(net_ratelimit()))
2797                         e_warn(drv, "checksum_partial proto=%x!\n",
2798                                skb->protocol);
2799                 break;
2800         }
2801
2802         css = skb_checksum_start_offset(skb);
2803
2804         i = tx_ring->next_to_use;
2805         buffer_info = &tx_ring->buffer_info[i];
2806         context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2807
2808         context_desc->lower_setup.ip_config = 0;
2809         context_desc->upper_setup.tcp_fields.tucss = css;
2810         context_desc->upper_setup.tcp_fields.tucso =
2811                 css + skb->csum_offset;
2812         context_desc->upper_setup.tcp_fields.tucse = 0;
2813         context_desc->tcp_seg_setup.data = 0;
2814         context_desc->cmd_and_length = cpu_to_le32(cmd_len);
2815
2816         buffer_info->time_stamp = jiffies;
2817         buffer_info->next_to_watch = i;
2818
2819         if (unlikely(++i == tx_ring->count)) i = 0;
2820         tx_ring->next_to_use = i;
2821
2822         return true;
2823 }
2824
2825 #define E1000_MAX_TXD_PWR       12
2826 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2827
2828 static int e1000_tx_map(struct e1000_adapter *adapter,
2829                         struct e1000_tx_ring *tx_ring,
2830                         struct sk_buff *skb, unsigned int first,
2831                         unsigned int max_per_txd, unsigned int nr_frags,
2832                         unsigned int mss)
2833 {
2834         struct e1000_hw *hw = &adapter->hw;
2835         struct pci_dev *pdev = adapter->pdev;
2836         struct e1000_tx_buffer *buffer_info;
2837         unsigned int len = skb_headlen(skb);
2838         unsigned int offset = 0, size, count = 0, i;
2839         unsigned int f, bytecount, segs;
2840
2841         i = tx_ring->next_to_use;
2842
2843         while (len) {
2844                 buffer_info = &tx_ring->buffer_info[i];
2845                 size = min(len, max_per_txd);
2846                 /* Workaround for Controller erratum --
2847                  * descriptor for non-tso packet in a linear SKB that follows a
2848                  * tso gets written back prematurely before the data is fully
2849                  * DMA'd to the controller
2850                  */
2851                 if (!skb->data_len && tx_ring->last_tx_tso &&
2852                     !skb_is_gso(skb)) {
2853                         tx_ring->last_tx_tso = false;
2854                         size -= 4;
2855                 }
2856
2857                 /* Workaround for premature desc write-backs
2858                  * in TSO mode.  Append 4-byte sentinel desc
2859                  */
2860                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2861                         size -= 4;
2862                 /* work-around for errata 10 and it applies
2863                  * to all controllers in PCI-X mode
2864                  * The fix is to make sure that the first descriptor of a
2865                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2866                  */
2867                 if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
2868                                 (size > 2015) && count == 0))
2869                         size = 2015;
2870
2871                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2872                  * terminating buffers within evenly-aligned dwords.
2873                  */
2874                 if (unlikely(adapter->pcix_82544 &&
2875                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2876                    size > 4))
2877                         size -= 4;
2878
2879                 buffer_info->length = size;
2880                 /* set time_stamp *before* dma to help avoid a possible race */
2881                 buffer_info->time_stamp = jiffies;
2882                 buffer_info->mapped_as_page = false;
2883                 buffer_info->dma = dma_map_single(&pdev->dev,
2884                                                   skb->data + offset,
2885                                                   size, DMA_TO_DEVICE);
2886                 if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2887                         goto dma_error;
2888                 buffer_info->next_to_watch = i;
2889
2890                 len -= size;
2891                 offset += size;
2892                 count++;
2893                 if (len) {
2894                         i++;
2895                         if (unlikely(i == tx_ring->count))
2896                                 i = 0;
2897                 }
2898         }
2899
2900         for (f = 0; f < nr_frags; f++) {
2901                 const struct skb_frag_struct *frag;
2902
2903                 frag = &skb_shinfo(skb)->frags[f];
2904                 len = skb_frag_size(frag);
2905                 offset = 0;
2906
2907                 while (len) {
2908                         unsigned long bufend;
2909                         i++;
2910                         if (unlikely(i == tx_ring->count))
2911                                 i = 0;
2912
2913                         buffer_info = &tx_ring->buffer_info[i];
2914                         size = min(len, max_per_txd);
2915                         /* Workaround for premature desc write-backs
2916                          * in TSO mode.  Append 4-byte sentinel desc
2917                          */
2918                         if (unlikely(mss && f == (nr_frags-1) &&
2919                             size == len && size > 8))
2920                                 size -= 4;
2921                         /* Workaround for potential 82544 hang in PCI-X.
2922                          * Avoid terminating buffers within evenly-aligned
2923                          * dwords.
2924                          */
2925                         bufend = (unsigned long)
2926                                 page_to_phys(skb_frag_page(frag));
2927                         bufend += offset + size - 1;
2928                         if (unlikely(adapter->pcix_82544 &&
2929                                      !(bufend & 4) &&
2930                                      size > 4))
2931                                 size -= 4;
2932
2933                         buffer_info->length = size;
2934                         buffer_info->time_stamp = jiffies;
2935                         buffer_info->mapped_as_page = true;
2936                         buffer_info->dma = skb_frag_dma_map(&pdev->dev, frag,
2937                                                 offset, size, DMA_TO_DEVICE);
2938                         if (dma_mapping_error(&pdev->dev, buffer_info->dma))
2939                                 goto dma_error;
2940                         buffer_info->next_to_watch = i;
2941
2942                         len -= size;
2943                         offset += size;
2944                         count++;
2945                 }
2946         }
2947
2948         segs = skb_shinfo(skb)->gso_segs ?: 1;
2949         /* multiply data chunks by size of headers */
2950         bytecount = ((segs - 1) * skb_headlen(skb)) + skb->len;
2951
2952         tx_ring->buffer_info[i].skb = skb;
2953         tx_ring->buffer_info[i].segs = segs;
2954         tx_ring->buffer_info[i].bytecount = bytecount;
2955         tx_ring->buffer_info[first].next_to_watch = i;
2956
2957         return count;
2958
2959 dma_error:
2960         dev_err(&pdev->dev, "TX DMA map failed\n");
2961         buffer_info->dma = 0;
2962         if (count)
2963                 count--;
2964
2965         while (count--) {
2966                 if (i==0)
2967                         i += tx_ring->count;
2968                 i--;
2969                 buffer_info = &tx_ring->buffer_info[i];
2970                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2971         }
2972
2973         return 0;
2974 }
2975
2976 static void e1000_tx_queue(struct e1000_adapter *adapter,
2977                            struct e1000_tx_ring *tx_ring, int tx_flags,
2978                            int count)
2979 {
2980         struct e1000_tx_desc *tx_desc = NULL;
2981         struct e1000_tx_buffer *buffer_info;
2982         u32 txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2983         unsigned int i;
2984
2985         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2986                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2987                              E1000_TXD_CMD_TSE;
2988                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2989
2990                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2991                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2992         }
2993
2994         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2995                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2996                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2997         }
2998
2999         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
3000                 txd_lower |= E1000_TXD_CMD_VLE;
3001                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
3002         }
3003
3004         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3005                 txd_lower &= ~(E1000_TXD_CMD_IFCS);
3006
3007         i = tx_ring->next_to_use;
3008
3009         while (count--) {
3010                 buffer_info = &tx_ring->buffer_info[i];
3011                 tx_desc = E1000_TX_DESC(*tx_ring, i);
3012                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3013                 tx_desc->lower.data =
3014                         cpu_to_le32(txd_lower | buffer_info->length);
3015                 tx_desc->upper.data = cpu_to_le32(txd_upper);
3016                 if (unlikely(++i == tx_ring->count)) i = 0;
3017         }
3018
3019         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
3020
3021         /* txd_cmd re-enables FCS, so we'll re-disable it here as desired. */
3022         if (unlikely(tx_flags & E1000_TX_FLAGS_NO_FCS))
3023                 tx_desc->lower.data &= ~(cpu_to_le32(E1000_TXD_CMD_IFCS));
3024
3025         /* Force memory writes to complete before letting h/w
3026          * know there are new descriptors to fetch.  (Only
3027          * applicable for weak-ordered memory model archs,
3028          * such as IA-64).
3029          */
3030         wmb();
3031
3032         tx_ring->next_to_use = i;
3033 }
3034
3035 /* 82547 workaround to avoid controller hang in half-duplex environment.
3036  * The workaround is to avoid queuing a large packet that would span
3037  * the internal Tx FIFO ring boundary by notifying the stack to resend
3038  * the packet at a later time.  This gives the Tx FIFO an opportunity to
3039  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
3040  * to the beginning of the Tx FIFO.
3041  */
3042
3043 #define E1000_FIFO_HDR                  0x10
3044 #define E1000_82547_PAD_LEN             0x3E0
3045
3046 static int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
3047                                        struct sk_buff *skb)
3048 {
3049         u32 fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
3050         u32 skb_fifo_len = skb->len + E1000_FIFO_HDR;
3051
3052         skb_fifo_len = ALIGN(skb_fifo_len, E1000_FIFO_HDR);
3053
3054         if (adapter->link_duplex != HALF_DUPLEX)
3055                 goto no_fifo_stall_required;
3056
3057         if (atomic_read(&adapter->tx_fifo_stall))
3058                 return 1;
3059
3060         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
3061                 atomic_set(&adapter->tx_fifo_stall, 1);
3062                 return 1;
3063         }
3064
3065 no_fifo_stall_required:
3066         adapter->tx_fifo_head += skb_fifo_len;
3067         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
3068                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
3069         return 0;
3070 }
3071
3072 static int __e1000_maybe_stop_tx(struct net_device *netdev, int size)
3073 {
3074         struct e1000_adapter *adapter = netdev_priv(netdev);
3075         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3076
3077         netif_stop_queue(netdev);
3078         /* Herbert's original patch had:
3079          *  smp_mb__after_netif_stop_queue();
3080          * but since that doesn't exist yet, just open code it.
3081          */
3082         smp_mb();
3083
3084         /* We need to check again in a case another CPU has just
3085          * made room available.
3086          */
3087         if (likely(E1000_DESC_UNUSED(tx_ring) < size))
3088                 return -EBUSY;
3089
3090         /* A reprieve! */
3091         netif_start_queue(netdev);
3092         ++adapter->restart_queue;
3093         return 0;
3094 }
3095
3096 static int e1000_maybe_stop_tx(struct net_device *netdev,
3097                                struct e1000_tx_ring *tx_ring, int size)
3098 {
3099         if (likely(E1000_DESC_UNUSED(tx_ring) >= size))
3100                 return 0;
3101         return __e1000_maybe_stop_tx(netdev, size);
3102 }
3103
3104 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
3105 static netdev_tx_t e1000_xmit_frame(struct sk_buff *skb,
3106                                     struct net_device *netdev)
3107 {
3108         struct e1000_adapter *adapter = netdev_priv(netdev);
3109         struct e1000_hw *hw = &adapter->hw;
3110         struct e1000_tx_ring *tx_ring;
3111         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
3112         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
3113         unsigned int tx_flags = 0;
3114         unsigned int len = skb_headlen(skb);
3115         unsigned int nr_frags;
3116         unsigned int mss;
3117         int count = 0;
3118         int tso;
3119         unsigned int f;
3120         __be16 protocol = vlan_get_protocol(skb);
3121
3122         /* This goes back to the question of how to logically map a Tx queue
3123          * to a flow.  Right now, performance is impacted slightly negatively
3124          * if using multiple Tx queues.  If the stack breaks away from a
3125          * single qdisc implementation, we can look at this again.
3126          */
3127         tx_ring = adapter->tx_ring;
3128
3129         /* On PCI/PCI-X HW, if packet size is less than ETH_ZLEN,
3130          * packets may get corrupted during padding by HW.
3131          * To WA this issue, pad all small packets manually.
3132          */
3133         if (eth_skb_pad(skb))
3134                 return NETDEV_TX_OK;
3135
3136         mss = skb_shinfo(skb)->gso_size;
3137         /* The controller does a simple calculation to
3138          * make sure there is enough room in the FIFO before
3139          * initiating the DMA for each buffer.  The calc is:
3140          * 4 = ceil(buffer len/mss).  To make sure we don't
3141          * overrun the FIFO, adjust the max buffer len if mss
3142          * drops.
3143          */
3144         if (mss) {
3145                 u8 hdr_len;
3146                 max_per_txd = min(mss << 2, max_per_txd);
3147                 max_txd_pwr = fls(max_per_txd) - 1;
3148
3149                 hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
3150                 if (skb->data_len && hdr_len == len) {
3151                         switch (hw->mac_type) {
3152                                 unsigned int pull_size;
3153                         case e1000_82544:
3154                                 /* Make sure we have room to chop off 4 bytes,
3155                                  * and that the end alignment will work out to
3156                                  * this hardware's requirements
3157                                  * NOTE: this is a TSO only workaround
3158                                  * if end byte alignment not correct move us
3159                                  * into the next dword
3160                                  */
3161                                 if ((unsigned long)(skb_tail_pointer(skb) - 1)
3162                                     & 4)
3163                                         break;
3164                                 /* fall through */
3165                                 pull_size = min((unsigned int)4, skb->data_len);
3166                                 if (!__pskb_pull_tail(skb, pull_size)) {
3167                                         e_err(drv, "__pskb_pull_tail "
3168                                               "failed.\n");
3169                                         dev_kfree_skb_any(skb);
3170                                         return NETDEV_TX_OK;
3171                                 }
3172                                 len = skb_headlen(skb);
3173                                 break;
3174                         default:
3175                                 /* do nothing */
3176                                 break;
3177                         }
3178                 }
3179         }
3180
3181         /* reserve a descriptor for the offload context */
3182         if ((mss) || (skb->ip_summed == CHECKSUM_PARTIAL))
3183                 count++;
3184         count++;
3185
3186         /* Controller Erratum workaround */
3187         if (!skb->data_len && tx_ring->last_tx_tso && !skb_is_gso(skb))
3188                 count++;
3189
3190         count += TXD_USE_COUNT(len, max_txd_pwr);
3191
3192         if (adapter->pcix_82544)
3193                 count++;
3194
3195         /* work-around for errata 10 and it applies to all controllers
3196          * in PCI-X mode, so add one more descriptor to the count
3197          */
3198         if (unlikely((hw->bus_type == e1000_bus_type_pcix) &&
3199                         (len > 2015)))
3200                 count++;
3201
3202         nr_frags = skb_shinfo(skb)->nr_frags;
3203         for (f = 0; f < nr_frags; f++)
3204                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]),
3205                                        max_txd_pwr);
3206         if (adapter->pcix_82544)
3207                 count += nr_frags;
3208
3209         /* need: count + 2 desc gap to keep tail from touching
3210          * head, otherwise try next time
3211          */
3212         if (unlikely(e1000_maybe_stop_tx(netdev, tx_ring, count + 2)))
3213                 return NETDEV_TX_BUSY;
3214
3215         if (unlikely((hw->mac_type == e1000_82547) &&
3216                      (e1000_82547_fifo_workaround(adapter, skb)))) {
3217                 netif_stop_queue(netdev);
3218                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3219                         schedule_delayed_work(&adapter->fifo_stall_task, 1);
3220                 return NETDEV_TX_BUSY;
3221         }
3222
3223         if (skb_vlan_tag_present(skb)) {
3224                 tx_flags |= E1000_TX_FLAGS_VLAN;
3225                 tx_flags |= (skb_vlan_tag_get(skb) <<
3226                              E1000_TX_FLAGS_VLAN_SHIFT);
3227         }
3228
3229         first = tx_ring->next_to_use;
3230
3231         tso = e1000_tso(adapter, tx_ring, skb, protocol);
3232         if (tso < 0) {
3233                 dev_kfree_skb_any(skb);
3234                 return NETDEV_TX_OK;
3235         }
3236
3237         if (likely(tso)) {
3238                 if (likely(hw->mac_type != e1000_82544))
3239                         tx_ring->last_tx_tso = true;
3240                 tx_flags |= E1000_TX_FLAGS_TSO;
3241         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb, protocol)))
3242                 tx_flags |= E1000_TX_FLAGS_CSUM;
3243
3244         if (protocol == htons(ETH_P_IP))
3245                 tx_flags |= E1000_TX_FLAGS_IPV4;
3246
3247         if (unlikely(skb->no_fcs))
3248                 tx_flags |= E1000_TX_FLAGS_NO_FCS;
3249
3250         count = e1000_tx_map(adapter, tx_ring, skb, first, max_per_txd,
3251                              nr_frags, mss);
3252
3253         if (count) {
3254                 netdev_sent_queue(netdev, skb->len);
3255                 skb_tx_timestamp(skb);
3256
3257                 e1000_tx_queue(adapter, tx_ring, tx_flags, count);
3258                 /* Make sure there is space in the ring for the next send. */
3259                 e1000_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 2);
3260
3261                 if (!skb->xmit_more ||
3262                     netif_xmit_stopped(netdev_get_tx_queue(netdev, 0))) {
3263                         writel(tx_ring->next_to_use, hw->hw_addr + tx_ring->tdt);
3264                         /* we need this if more than one processor can write to
3265                          * our tail at a time, it synchronizes IO on IA64/Altix
3266                          * systems
3267                          */
3268                         mmiowb();
3269                 }
3270         } else {
3271                 dev_kfree_skb_any(skb);
3272                 tx_ring->buffer_info[first].time_stamp = 0;
3273                 tx_ring->next_to_use = first;
3274         }
3275
3276         return NETDEV_TX_OK;
3277 }
3278
3279 #define NUM_REGS 38 /* 1 based count */
3280 static void e1000_regdump(struct e1000_adapter *adapter)
3281 {
3282         struct e1000_hw *hw = &adapter->hw;
3283         u32 regs[NUM_REGS];
3284         u32 *regs_buff = regs;
3285         int i = 0;
3286
3287         static const char * const reg_name[] = {
3288                 "CTRL",  "STATUS",
3289                 "RCTL", "RDLEN", "RDH", "RDT", "RDTR",
3290                 "TCTL", "TDBAL", "TDBAH", "TDLEN", "TDH", "TDT",
3291                 "TIDV", "TXDCTL", "TADV", "TARC0",
3292                 "TDBAL1", "TDBAH1", "TDLEN1", "TDH1", "TDT1",
3293                 "TXDCTL1", "TARC1",
3294                 "CTRL_EXT", "ERT", "RDBAL", "RDBAH",
3295                 "TDFH", "TDFT", "TDFHS", "TDFTS", "TDFPC",
3296                 "RDFH", "RDFT", "RDFHS", "RDFTS", "RDFPC"
3297         };
3298
3299         regs_buff[0]  = er32(CTRL);
3300         regs_buff[1]  = er32(STATUS);
3301
3302         regs_buff[2]  = er32(RCTL);
3303         regs_buff[3]  = er32(RDLEN);
3304         regs_buff[4]  = er32(RDH);
3305         regs_buff[5]  = er32(RDT);
3306         regs_buff[6]  = er32(RDTR);
3307
3308         regs_buff[7]  = er32(TCTL);
3309         regs_buff[8]  = er32(TDBAL);
3310         regs_buff[9]  = er32(TDBAH);
3311         regs_buff[10] = er32(TDLEN);
3312         regs_buff[11] = er32(TDH);
3313         regs_buff[12] = er32(TDT);
3314         regs_buff[13] = er32(TIDV);
3315         regs_buff[14] = er32(TXDCTL);
3316         regs_buff[15] = er32(TADV);
3317         regs_buff[16] = er32(TARC0);
3318
3319         regs_buff[17] = er32(TDBAL1);
3320         regs_buff[18] = er32(TDBAH1);
3321         regs_buff[19] = er32(TDLEN1);
3322         regs_buff[20] = er32(TDH1);
3323         regs_buff[21] = er32(TDT1);
3324         regs_buff[22] = er32(TXDCTL1);
3325         regs_buff[23] = er32(TARC1);
3326         regs_buff[24] = er32(CTRL_EXT);
3327         regs_buff[25] = er32(ERT);
3328         regs_buff[26] = er32(RDBAL0);
3329         regs_buff[27] = er32(RDBAH0);
3330         regs_buff[28] = er32(TDFH);
3331         regs_buff[29] = er32(TDFT);
3332         regs_buff[30] = er32(TDFHS);
3333         regs_buff[31] = er32(TDFTS);
3334         regs_buff[32] = er32(TDFPC);
3335         regs_buff[33] = er32(RDFH);
3336         regs_buff[34] = er32(RDFT);
3337         regs_buff[35] = er32(RDFHS);
3338         regs_buff[36] = er32(RDFTS);
3339         regs_buff[37] = er32(RDFPC);
3340
3341         pr_info("Register dump\n");
3342         for (i = 0; i < NUM_REGS; i++)
3343                 pr_info("%-15s  %08x\n", reg_name[i], regs_buff[i]);
3344 }
3345
3346 /*
3347  * e1000_dump: Print registers, tx ring and rx ring
3348  */
3349 static void e1000_dump(struct e1000_adapter *adapter)
3350 {
3351         /* this code doesn't handle multiple rings */
3352         struct e1000_tx_ring *tx_ring = adapter->tx_ring;
3353         struct e1000_rx_ring *rx_ring = adapter->rx_ring;
3354         int i;
3355
3356         if (!netif_msg_hw(adapter))
3357                 return;
3358
3359         /* Print Registers */
3360         e1000_regdump(adapter);
3361
3362         /* transmit dump */
3363         pr_info("TX Desc ring0 dump\n");
3364
3365         /* Transmit Descriptor Formats - DEXT[29] is 0 (Legacy) or 1 (Extended)
3366          *
3367          * Legacy Transmit Descriptor
3368          *   +--------------------------------------------------------------+
3369          * 0 |         Buffer Address [63:0] (Reserved on Write Back)       |
3370          *   +--------------------------------------------------------------+
3371          * 8 | Special  |    CSS     | Status |  CMD    |  CSO   |  Length  |
3372          *   +--------------------------------------------------------------+
3373          *   63       48 47        36 35    32 31     24 23    16 15        0
3374          *
3375          * Extended Context Descriptor (DTYP=0x0) for TSO or checksum offload
3376          *   63      48 47    40 39       32 31             16 15    8 7      0
3377          *   +----------------------------------------------------------------+
3378          * 0 |  TUCSE  | TUCS0  |   TUCSS   |     IPCSE       | IPCS0 | IPCSS |
3379          *   +----------------------------------------------------------------+
3380          * 8 |   MSS   | HDRLEN | RSV | STA | TUCMD | DTYP |      PAYLEN      |
3381          *   +----------------------------------------------------------------+
3382          *   63      48 47    40 39 36 35 32 31   24 23  20 19                0
3383          *
3384          * Extended Data Descriptor (DTYP=0x1)
3385          *   +----------------------------------------------------------------+
3386          * 0 |                     Buffer Address [63:0]                      |
3387          *   +----------------------------------------------------------------+
3388          * 8 | VLAN tag |  POPTS  | Rsvd | Status | Command | DTYP |  DTALEN  |
3389          *   +----------------------------------------------------------------+
3390          *   63       48 47     40 39  36 35    32 31     24 23  20 19        0
3391          */
3392         pr_info("Tc[desc]     [Ce CoCsIpceCoS] [MssHlRSCm0Plen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3393         pr_info("Td[desc]     [address 63:0  ] [VlaPoRSCm1Dlen] [bi->dma       ] leng  ntw timestmp         bi->skb\n");
3394
3395         if (!netif_msg_tx_done(adapter))
3396                 goto rx_ring_summary;
3397
3398         for (i = 0; tx_ring->desc && (i < tx_ring->count); i++) {
3399                 struct e1000_tx_desc *tx_desc = E1000_TX_DESC(*tx_ring, i);
3400                 struct e1000_tx_buffer *buffer_info = &tx_ring->buffer_info[i];
3401                 struct my_u { __le64 a; __le64 b; };
3402                 struct my_u *u = (struct my_u *)tx_desc;
3403                 const char *type;
3404
3405                 if (i == tx_ring->next_to_use && i == tx_ring->next_to_clean)
3406                         type = "NTC/U";
3407                 else if (i == tx_ring->next_to_use)
3408                         type = "NTU";
3409                 else if (i == tx_ring->next_to_clean)
3410                         type = "NTC";
3411                 else
3412                         type = "";
3413
3414                 pr_info("T%c[0x%03X]    %016llX %016llX %016llX %04X  %3X %016llX %p %s\n",
3415                         ((le64_to_cpu(u->b) & (1<<20)) ? 'd' : 'c'), i,
3416                         le64_to_cpu(u->a), le64_to_cpu(u->b),
3417                         (u64)buffer_info->dma, buffer_info->length,
3418                         buffer_info->next_to_watch,
3419                         (u64)buffer_info->time_stamp, buffer_info->skb, type);
3420         }
3421
3422 rx_ring_summary:
3423         /* receive dump */
3424         pr_info("\nRX Desc ring dump\n");
3425
3426         /* Legacy Receive Descriptor Format
3427          *
3428          * +-----------------------------------------------------+
3429          * |                Buffer Address [63:0]                |
3430          * +-----------------------------------------------------+
3431          * | VLAN Tag | Errors | Status 0 | Packet csum | Length |
3432          * +-----------------------------------------------------+
3433          * 63       48 47    40 39      32 31         16 15      0
3434          */
3435         pr_info("R[desc]      [address 63:0  ] [vl er S cks ln] [bi->dma       ] [bi->skb]\n");
3436
3437         if (!netif_msg_rx_status(adapter))
3438                 goto exit;
3439
3440         for (i = 0; rx_ring->desc && (i < rx_ring->count); i++) {
3441                 struct e1000_rx_desc *rx_desc = E1000_RX_DESC(*rx_ring, i);
3442                 struct e1000_rx_buffer *buffer_info = &rx_ring->buffer_info[i];
3443                 struct my_u { __le64 a; __le64 b; };
3444                 struct my_u *u = (struct my_u *)rx_desc;
3445                 const char *type;
3446
3447                 if (i == rx_ring->next_to_use)
3448                         type = "NTU";
3449                 else if (i == rx_ring->next_to_clean)
3450                         type = "NTC";
3451                 else
3452                         type = "";
3453
3454                 pr_info("R[0x%03X]     %016llX %016llX %016llX %p %s\n",
3455                         i, le64_to_cpu(u->a), le64_to_cpu(u->b),
3456                         (u64)buffer_info->dma, buffer_info->rxbuf.data, type);
3457         } /* for */
3458
3459         /* dump the descriptor caches */
3460         /* rx */
3461         pr_info("Rx descriptor cache in 64bit format\n");
3462         for (i = 0x6000; i <= 0x63FF ; i += 0x10) {
3463                 pr_info("R%04X: %08X|%08X %08X|%08X\n",
3464                         i,
3465                         readl(adapter->hw.hw_addr + i+4),
3466                         readl(adapter->hw.hw_addr + i),
3467                         readl(adapter->hw.hw_addr + i+12),
3468                         readl(adapter->hw.hw_addr + i+8));
3469         }
3470         /* tx */
3471         pr_info("Tx descriptor cache in 64bit format\n");
3472         for (i = 0x7000; i <= 0x73FF ; i += 0x10) {
3473                 pr_info("T%04X: %08X|%08X %08X|%08X\n",
3474                         i,
3475                         readl(adapter->hw.hw_addr + i+4),
3476                         readl(adapter->hw.hw_addr + i),
3477                         readl(adapter->hw.hw_addr + i+12),
3478                         readl(adapter->hw.hw_addr + i+8));
3479         }
3480 exit:
3481         return;
3482 }
3483
3484 /**
3485  * e1000_tx_timeout - Respond to a Tx Hang
3486  * @netdev: network interface device structure
3487  **/
3488 static void e1000_tx_timeout(struct net_device *netdev)
3489 {
3490         struct e1000_adapter *adapter = netdev_priv(netdev);
3491
3492         /* Do the reset outside of interrupt context */
3493         adapter->tx_timeout_count++;
3494         schedule_work(&adapter->reset_task);
3495 }
3496
3497 static void e1000_reset_task(struct work_struct *work)
3498 {
3499         struct e1000_adapter *adapter =
3500                 container_of(work, struct e1000_adapter, reset_task);
3501
3502         e_err(drv, "Reset adapter\n");
3503         e1000_reinit_locked(adapter);
3504 }
3505
3506 /**
3507  * e1000_get_stats - Get System Network Statistics
3508  * @netdev: network interface device structure
3509  *
3510  * Returns the address of the device statistics structure.
3511  * The statistics are actually updated from the watchdog.
3512  **/
3513 static struct net_device_stats *e1000_get_stats(struct net_device *netdev)
3514 {
3515         /* only return the current stats */
3516         return &netdev->stats;
3517 }
3518
3519 /**
3520  * e1000_change_mtu - Change the Maximum Transfer Unit
3521  * @netdev: network interface device structure
3522  * @new_mtu: new value for maximum frame size
3523  *
3524  * Returns 0 on success, negative on failure
3525  **/
3526 static int e1000_change_mtu(struct net_device *netdev, int new_mtu)
3527 {
3528         struct e1000_adapter *adapter = netdev_priv(netdev);
3529         struct e1000_hw *hw = &adapter->hw;
3530         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3531
3532         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3533             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3534                 e_err(probe, "Invalid MTU setting\n");
3535                 return -EINVAL;
3536         }
3537
3538         /* Adapter-specific max frame size limits. */
3539         switch (hw->mac_type) {
3540         case e1000_undefined ... e1000_82542_rev2_1:
3541                 if (max_frame > (ETH_FRAME_LEN + ETH_FCS_LEN)) {
3542                         e_err(probe, "Jumbo Frames not supported.\n");
3543                         return -EINVAL;
3544                 }
3545                 break;
3546         default:
3547                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3548                 break;
3549         }
3550
3551         while (test_and_set_bit(__E1000_RESETTING, &adapter->flags))
3552                 msleep(1);
3553         /* e1000_down has a dependency on max_frame_size */
3554         hw->max_frame_size = max_frame;
3555         if (netif_running(netdev)) {
3556                 /* prevent buffers from being reallocated */
3557                 adapter->alloc_rx_buf = e1000_alloc_dummy_rx_buffers;
3558                 e1000_down(adapter);
3559         }
3560
3561         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3562          * means we reserve 2 more, this pushes us to allocate from the next
3563          * larger slab size.
3564          * i.e. RXBUFFER_2048 --> size-4096 slab
3565          * however with the new *_jumbo_rx* routines, jumbo receives will use
3566          * fragmented skbs
3567          */
3568
3569         if (max_frame <= E1000_RXBUFFER_2048)
3570                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3571         else
3572 #if (PAGE_SIZE >= E1000_RXBUFFER_16384)
3573                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3574 #elif (PAGE_SIZE >= E1000_RXBUFFER_4096)
3575                 adapter->rx_buffer_len = PAGE_SIZE;
3576 #endif
3577
3578         /* adjust allocation if LPE protects us, and we aren't using SBP */
3579         if (!hw->tbi_compatibility_on &&
3580             ((max_frame == (ETH_FRAME_LEN + ETH_FCS_LEN)) ||
3581              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE)))
3582                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3583
3584         pr_info("%s changing MTU from %d to %d\n",
3585                 netdev->name, netdev->mtu, new_mtu);
3586         netdev->mtu = new_mtu;
3587
3588         if (netif_running(netdev))
3589                 e1000_up(adapter);
3590         else
3591                 e1000_reset(adapter);
3592
3593         clear_bit(__E1000_RESETTING, &adapter->flags);
3594
3595         return 0;
3596 }
3597
3598 /**
3599  * e1000_update_stats - Update the board statistics counters
3600  * @adapter: board private structure
3601  **/
3602 void e1000_update_stats(struct e1000_adapter *adapter)
3603 {
3604         struct net_device *netdev = adapter->netdev;
3605         struct e1000_hw *hw = &adapter->hw;
3606         struct pci_dev *pdev = adapter->pdev;
3607         unsigned long flags;
3608         u16 phy_tmp;
3609
3610 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3611
3612         /* Prevent stats update while adapter is being reset, or if the pci
3613          * connection is down.
3614          */
3615         if (adapter->link_speed == 0)
3616                 return;
3617         if (pci_channel_offline(pdev))
3618                 return;
3619
3620         spin_lock_irqsave(&adapter->stats_lock, flags);
3621
3622         /* these counters are modified from e1000_tbi_adjust_stats,
3623          * called from the interrupt context, so they must only
3624          * be written while holding adapter->stats_lock
3625          */
3626
3627         adapter->stats.crcerrs += er32(CRCERRS);
3628         adapter->stats.gprc += er32(GPRC);
3629         adapter->stats.gorcl += er32(GORCL);
3630         adapter->stats.gorch += er32(GORCH);
3631         adapter->stats.bprc += er32(BPRC);
3632         adapter->stats.mprc += er32(MPRC);
3633         adapter->stats.roc += er32(ROC);
3634
3635         adapter->stats.prc64 += er32(PRC64);
3636         adapter->stats.prc127 += er32(PRC127);
3637         adapter->stats.prc255 += er32(PRC255);
3638         adapter->stats.prc511 += er32(PRC511);
3639         adapter->stats.prc1023 += er32(PRC1023);
3640         adapter->stats.prc1522 += er32(PRC1522);
3641
3642         adapter->stats.symerrs += er32(SYMERRS);
3643         adapter->stats.mpc += er32(MPC);
3644         adapter->stats.scc += er32(SCC);
3645         adapter->stats.ecol += er32(ECOL);
3646         adapter->stats.mcc += er32(MCC);
3647         adapter->stats.latecol += er32(LATECOL);
3648         adapter->stats.dc += er32(DC);
3649         adapter->stats.sec += er32(SEC);
3650         adapter->stats.rlec += er32(RLEC);
3651         adapter->stats.xonrxc += er32(XONRXC);
3652         adapter->stats.xontxc += er32(XONTXC);
3653         adapter->stats.xoffrxc += er32(XOFFRXC);
3654         adapter->stats.xofftxc += er32(XOFFTXC);
3655         adapter->stats.fcruc += er32(FCRUC);
3656         adapter->stats.gptc += er32(GPTC);
3657         adapter->stats.gotcl += er32(GOTCL);
3658         adapter->stats.gotch += er32(GOTCH);
3659         adapter->stats.rnbc += er32(RNBC);
3660         adapter->stats.ruc += er32(RUC);
3661         adapter->stats.rfc += er32(RFC);
3662         adapter->stats.rjc += er32(RJC);
3663         adapter->stats.torl += er32(TORL);
3664         adapter->stats.torh += er32(TORH);
3665         adapter->stats.totl += er32(TOTL);
3666         adapter->stats.toth += er32(TOTH);
3667         adapter->stats.tpr += er32(TPR);
3668
3669         adapter->stats.ptc64 += er32(PTC64);
3670         adapter->stats.ptc127 += er32(PTC127);
3671         adapter->stats.ptc255 += er32(PTC255);
3672         adapter->stats.ptc511 += er32(PTC511);
3673         adapter->stats.ptc1023 += er32(PTC1023);
3674         adapter->stats.ptc1522 += er32(PTC1522);
3675
3676         adapter->stats.mptc += er32(MPTC);
3677         adapter->stats.bptc += er32(BPTC);
3678
3679         /* used for adaptive IFS */
3680
3681         hw->tx_packet_delta = er32(TPT);
3682         adapter->stats.tpt += hw->tx_packet_delta;
3683         hw->collision_delta = er32(COLC);
3684         adapter->stats.colc += hw->collision_delta;
3685
3686         if (hw->mac_type >= e1000_82543) {
3687                 adapter->stats.algnerrc += er32(ALGNERRC);
3688                 adapter->stats.rxerrc += er32(RXERRC);
3689                 adapter->stats.tncrs += er32(TNCRS);
3690                 adapter->stats.cexterr += er32(CEXTERR);
3691                 adapter->stats.tsctc += er32(TSCTC);
3692                 adapter->stats.tsctfc += er32(TSCTFC);
3693         }
3694
3695         /* Fill out the OS statistics structure */
3696         netdev->stats.multicast = adapter->stats.mprc;
3697         netdev->stats.collisions = adapter->stats.colc;
3698
3699         /* Rx Errors */
3700
3701         /* RLEC on some newer hardware can be incorrect so build
3702          * our own version based on RUC and ROC
3703          */
3704         netdev->stats.rx_errors = adapter->stats.rxerrc +
3705                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3706                 adapter->stats.ruc + adapter->stats.roc +
3707                 adapter->stats.cexterr;
3708         adapter->stats.rlerrc = adapter->stats.ruc + adapter->stats.roc;
3709         netdev->stats.rx_length_errors = adapter->stats.rlerrc;
3710         netdev->stats.rx_crc_errors = adapter->stats.crcerrs;
3711         netdev->stats.rx_frame_errors = adapter->stats.algnerrc;
3712         netdev->stats.rx_missed_errors = adapter->stats.mpc;
3713
3714         /* Tx Errors */
3715         adapter->stats.txerrc = adapter->stats.ecol + adapter->stats.latecol;
3716         netdev->stats.tx_errors = adapter->stats.txerrc;
3717         netdev->stats.tx_aborted_errors = adapter->stats.ecol;
3718         netdev->stats.tx_window_errors = adapter->stats.latecol;
3719         netdev->stats.tx_carrier_errors = adapter->stats.tncrs;
3720         if (hw->bad_tx_carr_stats_fd &&
3721             adapter->link_duplex == FULL_DUPLEX) {
3722                 netdev->stats.tx_carrier_errors = 0;
3723                 adapter->stats.tncrs = 0;
3724         }
3725
3726         /* Tx Dropped needs to be maintained elsewhere */
3727
3728         /* Phy Stats */
3729         if (hw->media_type == e1000_media_type_copper) {
3730                 if ((adapter->link_speed == SPEED_1000) &&
3731                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3732                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3733                         adapter->phy_stats.idle_errors += phy_tmp;
3734                 }
3735
3736                 if ((hw->mac_type <= e1000_82546) &&
3737                    (hw->phy_type == e1000_phy_m88) &&
3738                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3739                         adapter->phy_stats.receive_errors += phy_tmp;
3740         }
3741
3742         /* Management Stats */
3743         if (hw->has_smbus) {
3744                 adapter->stats.mgptc += er32(MGTPTC);
3745                 adapter->stats.mgprc += er32(MGTPRC);
3746                 adapter->stats.mgpdc += er32(MGTPDC);
3747         }
3748
3749         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3750 }
3751
3752 /**
3753  * e1000_intr - Interrupt Handler
3754  * @irq: interrupt number
3755  * @data: pointer to a network interface device structure
3756  **/
3757 static irqreturn_t e1000_intr(int irq, void *data)
3758 {
3759         struct net_device *netdev = data;
3760         struct e1000_adapter *adapter = netdev_priv(netdev);
3761         struct e1000_hw *hw = &adapter->hw;
3762         u32 icr = er32(ICR);
3763
3764         if (unlikely((!icr)))
3765                 return IRQ_NONE;  /* Not our interrupt */
3766
3767         /* we might have caused the interrupt, but the above
3768          * read cleared it, and just in case the driver is
3769          * down there is nothing to do so return handled
3770          */
3771         if (unlikely(test_bit(__E1000_DOWN, &adapter->flags)))
3772                 return IRQ_HANDLED;
3773
3774         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3775                 hw->get_link_status = 1;
3776                 /* guard against interrupt when we're going down */
3777                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3778                         schedule_delayed_work(&adapter->watchdog_task, 1);
3779         }
3780
3781         /* disable interrupts, without the synchronize_irq bit */
3782         ew32(IMC, ~0);
3783         E1000_WRITE_FLUSH();
3784
3785         if (likely(napi_schedule_prep(&adapter->napi))) {
3786                 adapter->total_tx_bytes = 0;
3787                 adapter->total_tx_packets = 0;
3788                 adapter->total_rx_bytes = 0;
3789                 adapter->total_rx_packets = 0;
3790                 __napi_schedule(&adapter->napi);
3791         } else {
3792                 /* this really should not happen! if it does it is basically a
3793                  * bug, but not a hard error, so enable ints and continue
3794                  */
3795                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3796                         e1000_irq_enable(adapter);
3797         }
3798
3799         return IRQ_HANDLED;
3800 }
3801
3802 /**
3803  * e1000_clean - NAPI Rx polling callback
3804  * @adapter: board private structure
3805  **/
3806 static int e1000_clean(struct napi_struct *napi, int budget)
3807 {
3808         struct e1000_adapter *adapter = container_of(napi, struct e1000_adapter,
3809                                                      napi);
3810         int tx_clean_complete = 0, work_done = 0;
3811
3812         tx_clean_complete = e1000_clean_tx_irq(adapter, &adapter->tx_ring[0]);
3813
3814         adapter->clean_rx(adapter, &adapter->rx_ring[0], &work_done, budget);
3815
3816         if (!tx_clean_complete)
3817                 work_done = budget;
3818
3819         /* If budget not fully consumed, exit the polling mode */
3820         if (work_done < budget) {
3821                 if (likely(adapter->itr_setting & 3))
3822                         e1000_set_itr(adapter);
3823                 napi_complete_done(napi, work_done);
3824                 if (!test_bit(__E1000_DOWN, &adapter->flags))
3825                         e1000_irq_enable(adapter);
3826         }
3827
3828         return work_done;
3829 }
3830
3831 /**
3832  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3833  * @adapter: board private structure
3834  **/
3835 static bool e1000_clean_tx_irq(struct e1000_adapter *adapter,
3836                                struct e1000_tx_ring *tx_ring)
3837 {
3838         struct e1000_hw *hw = &adapter->hw;
3839         struct net_device *netdev = adapter->netdev;
3840         struct e1000_tx_desc *tx_desc, *eop_desc;
3841         struct e1000_tx_buffer *buffer_info;
3842         unsigned int i, eop;
3843         unsigned int count = 0;
3844         unsigned int total_tx_bytes=0, total_tx_packets=0;
3845         unsigned int bytes_compl = 0, pkts_compl = 0;
3846
3847         i = tx_ring->next_to_clean;
3848         eop = tx_ring->buffer_info[i].next_to_watch;
3849         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3850
3851         while ((eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) &&
3852                (count < tx_ring->count)) {
3853                 bool cleaned = false;
3854                 dma_rmb();      /* read buffer_info after eop_desc */
3855                 for ( ; !cleaned; count++) {
3856                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3857                         buffer_info = &tx_ring->buffer_info[i];
3858                         cleaned = (i == eop);
3859
3860                         if (cleaned) {
3861                                 total_tx_packets += buffer_info->segs;
3862                                 total_tx_bytes += buffer_info->bytecount;
3863                                 if (buffer_info->skb) {
3864                                         bytes_compl += buffer_info->skb->len;
3865                                         pkts_compl++;
3866                                 }
3867
3868                         }
3869                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3870                         tx_desc->upper.data = 0;
3871
3872                         if (unlikely(++i == tx_ring->count)) i = 0;
3873                 }
3874
3875                 eop = tx_ring->buffer_info[i].next_to_watch;
3876                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3877         }
3878
3879         tx_ring->next_to_clean = i;
3880
3881         netdev_completed_queue(netdev, pkts_compl, bytes_compl);
3882
3883 #define TX_WAKE_THRESHOLD 32
3884         if (unlikely(count && netif_carrier_ok(netdev) &&
3885                      E1000_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD)) {
3886                 /* Make sure that anybody stopping the queue after this
3887                  * sees the new next_to_clean.
3888                  */
3889                 smp_mb();
3890
3891                 if (netif_queue_stopped(netdev) &&
3892                     !(test_bit(__E1000_DOWN, &adapter->flags))) {
3893                         netif_wake_queue(netdev);
3894                         ++adapter->restart_queue;
3895                 }
3896         }
3897
3898         if (adapter->detect_tx_hung) {
3899                 /* Detect a transmit hang in hardware, this serializes the
3900                  * check with the clearing of time_stamp and movement of i
3901                  */
3902                 adapter->detect_tx_hung = false;
3903                 if (tx_ring->buffer_info[eop].time_stamp &&
3904                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3905                                (adapter->tx_timeout_factor * HZ)) &&
3906                     !(er32(STATUS) & E1000_STATUS_TXOFF)) {
3907
3908                         /* detected Tx unit hang */
3909                         e_err(drv, "Detected Tx Unit Hang\n"
3910                               "  Tx Queue             <%lu>\n"
3911                               "  TDH                  <%x>\n"
3912                               "  TDT                  <%x>\n"
3913                               "  next_to_use          <%x>\n"
3914                               "  next_to_clean        <%x>\n"
3915                               "buffer_info[next_to_clean]\n"
3916                               "  time_stamp           <%lx>\n"
3917                               "  next_to_watch        <%x>\n"
3918                               "  jiffies              <%lx>\n"
3919                               "  next_to_watch.status <%x>\n",
3920                                 (unsigned long)(tx_ring - adapter->tx_ring),
3921                                 readl(hw->hw_addr + tx_ring->tdh),
3922                                 readl(hw->hw_addr + tx_ring->tdt),
3923                                 tx_ring->next_to_use,
3924                                 tx_ring->next_to_clean,
3925                                 tx_ring->buffer_info[eop].time_stamp,
3926                                 eop,
3927                                 jiffies,
3928                                 eop_desc->upper.fields.status);
3929                         e1000_dump(adapter);
3930                         netif_stop_queue(netdev);
3931                 }
3932         }
3933         adapter->total_tx_bytes += total_tx_bytes;
3934         adapter->total_tx_packets += total_tx_packets;
3935         netdev->stats.tx_bytes += total_tx_bytes;
3936         netdev->stats.tx_packets += total_tx_packets;
3937         return count < tx_ring->count;
3938 }
3939
3940 /**
3941  * e1000_rx_checksum - Receive Checksum Offload for 82543
3942  * @adapter:     board private structure
3943  * @status_err:  receive descriptor status and error fields
3944  * @csum:        receive descriptor csum field
3945  * @sk_buff:     socket buffer with received data
3946  **/
3947 static void e1000_rx_checksum(struct e1000_adapter *adapter, u32 status_err,
3948                               u32 csum, struct sk_buff *skb)
3949 {
3950         struct e1000_hw *hw = &adapter->hw;
3951         u16 status = (u16)status_err;
3952         u8 errors = (u8)(status_err >> 24);
3953
3954         skb_checksum_none_assert(skb);
3955
3956         /* 82543 or newer only */
3957         if (unlikely(hw->mac_type < e1000_82543)) return;
3958         /* Ignore Checksum bit is set */
3959         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3960         /* TCP/UDP checksum error bit is set */
3961         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3962                 /* let the stack verify checksum errors */
3963                 adapter->hw_csum_err++;
3964                 return;
3965         }
3966         /* TCP/UDP Checksum has not been calculated */
3967         if (!(status & E1000_RXD_STAT_TCPCS))
3968                 return;
3969
3970         /* It must be a TCP or UDP packet with a valid checksum */
3971         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3972                 /* TCP checksum is good */
3973                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3974         }
3975         adapter->hw_csum_good++;
3976 }
3977
3978 /**
3979  * e1000_consume_page - helper function for jumbo Rx path
3980  **/
3981 static void e1000_consume_page(struct e1000_rx_buffer *bi, struct sk_buff *skb,
3982                                u16 length)
3983 {
3984         bi->rxbuf.page = NULL;
3985         skb->len += length;
3986         skb->data_len += length;
3987         skb->truesize += PAGE_SIZE;
3988 }
3989
3990 /**
3991  * e1000_receive_skb - helper function to handle rx indications
3992  * @adapter: board private structure
3993  * @status: descriptor status field as written by hardware
3994  * @vlan: descriptor vlan field as written by hardware (no le/be conversion)
3995  * @skb: pointer to sk_buff to be indicated to stack
3996  */
3997 static void e1000_receive_skb(struct e1000_adapter *adapter, u8 status,
3998                               __le16 vlan, struct sk_buff *skb)
3999 {
4000         skb->protocol = eth_type_trans(skb, adapter->netdev);
4001
4002         if (status & E1000_RXD_STAT_VP) {
4003                 u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4004
4005                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4006         }
4007         napi_gro_receive(&adapter->napi, skb);
4008 }
4009
4010 /**
4011  * e1000_tbi_adjust_stats
4012  * @hw: Struct containing variables accessed by shared code
4013  * @frame_len: The length of the frame in question
4014  * @mac_addr: The Ethernet destination address of the frame in question
4015  *
4016  * Adjusts the statistic counters when a frame is accepted by TBI_ACCEPT
4017  */
4018 static void e1000_tbi_adjust_stats(struct e1000_hw *hw,
4019                                    struct e1000_hw_stats *stats,
4020                                    u32 frame_len, const u8 *mac_addr)
4021 {
4022         u64 carry_bit;
4023
4024         /* First adjust the frame length. */
4025         frame_len--;
4026         /* We need to adjust the statistics counters, since the hardware
4027          * counters overcount this packet as a CRC error and undercount
4028          * the packet as a good packet
4029          */
4030         /* This packet should not be counted as a CRC error. */
4031         stats->crcerrs--;
4032         /* This packet does count as a Good Packet Received. */
4033         stats->gprc++;
4034
4035         /* Adjust the Good Octets received counters */
4036         carry_bit = 0x80000000 & stats->gorcl;
4037         stats->gorcl += frame_len;
4038         /* If the high bit of Gorcl (the low 32 bits of the Good Octets
4039          * Received Count) was one before the addition,
4040          * AND it is zero after, then we lost the carry out,
4041          * need to add one to Gorch (Good Octets Received Count High).
4042          * This could be simplified if all environments supported
4043          * 64-bit integers.
4044          */
4045         if (carry_bit && ((stats->gorcl & 0x80000000) == 0))
4046                 stats->gorch++;
4047         /* Is this a broadcast or multicast?  Check broadcast first,
4048          * since the test for a multicast frame will test positive on
4049          * a broadcast frame.
4050          */
4051         if (is_broadcast_ether_addr(mac_addr))
4052                 stats->bprc++;
4053         else if (is_multicast_ether_addr(mac_addr))
4054                 stats->mprc++;
4055
4056         if (frame_len == hw->max_frame_size) {
4057                 /* In this case, the hardware has overcounted the number of
4058                  * oversize frames.
4059                  */
4060                 if (stats->roc > 0)
4061                         stats->roc--;
4062         }
4063
4064         /* Adjust the bin counters when the extra byte put the frame in the
4065          * wrong bin. Remember that the frame_len was adjusted above.
4066          */
4067         if (frame_len == 64) {
4068                 stats->prc64++;
4069                 stats->prc127--;
4070         } else if (frame_len == 127) {
4071                 stats->prc127++;
4072                 stats->prc255--;
4073         } else if (frame_len == 255) {
4074                 stats->prc255++;
4075                 stats->prc511--;
4076         } else if (frame_len == 511) {
4077                 stats->prc511++;
4078                 stats->prc1023--;
4079         } else if (frame_len == 1023) {
4080                 stats->prc1023++;
4081                 stats->prc1522--;
4082         } else if (frame_len == 1522) {
4083                 stats->prc1522++;
4084         }
4085 }
4086
4087 static bool e1000_tbi_should_accept(struct e1000_adapter *adapter,
4088                                     u8 status, u8 errors,
4089                                     u32 length, const u8 *data)
4090 {
4091         struct e1000_hw *hw = &adapter->hw;
4092         u8 last_byte = *(data + length - 1);
4093
4094         if (TBI_ACCEPT(hw, status, errors, length, last_byte)) {
4095                 unsigned long irq_flags;
4096
4097                 spin_lock_irqsave(&adapter->stats_lock, irq_flags);
4098                 e1000_tbi_adjust_stats(hw, &adapter->stats, length, data);
4099                 spin_unlock_irqrestore(&adapter->stats_lock, irq_flags);
4100
4101                 return true;
4102         }
4103
4104         return false;
4105 }
4106
4107 static struct sk_buff *e1000_alloc_rx_skb(struct e1000_adapter *adapter,
4108                                           unsigned int bufsz)
4109 {
4110         struct sk_buff *skb = napi_alloc_skb(&adapter->napi, bufsz);
4111
4112         if (unlikely(!skb))
4113                 adapter->alloc_rx_buff_failed++;
4114         return skb;
4115 }
4116
4117 /**
4118  * e1000_clean_jumbo_rx_irq - Send received data up the network stack; legacy
4119  * @adapter: board private structure
4120  * @rx_ring: ring to clean
4121  * @work_done: amount of napi work completed this call
4122  * @work_to_do: max amount of work allowed for this call to do
4123  *
4124  * the return value indicates whether actual cleaning was done, there
4125  * is no guarantee that everything was cleaned
4126  */
4127 static bool e1000_clean_jumbo_rx_irq(struct e1000_adapter *adapter,
4128                                      struct e1000_rx_ring *rx_ring,
4129                                      int *work_done, int work_to_do)
4130 {
4131         struct net_device *netdev = adapter->netdev;
4132         struct pci_dev *pdev = adapter->pdev;
4133         struct e1000_rx_desc *rx_desc, *next_rxd;
4134         struct e1000_rx_buffer *buffer_info, *next_buffer;
4135         u32 length;
4136         unsigned int i;
4137         int cleaned_count = 0;
4138         bool cleaned = false;
4139         unsigned int total_rx_bytes=0, total_rx_packets=0;
4140
4141         i = rx_ring->next_to_clean;
4142         rx_desc = E1000_RX_DESC(*rx_ring, i);
4143         buffer_info = &rx_ring->buffer_info[i];
4144
4145         while (rx_desc->status & E1000_RXD_STAT_DD) {
4146                 struct sk_buff *skb;
4147                 u8 status;
4148
4149                 if (*work_done >= work_to_do)
4150                         break;
4151                 (*work_done)++;
4152                 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
4153
4154                 status = rx_desc->status;
4155
4156                 if (++i == rx_ring->count) i = 0;
4157                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4158                 prefetch(next_rxd);
4159
4160                 next_buffer = &rx_ring->buffer_info[i];
4161
4162                 cleaned = true;
4163                 cleaned_count++;
4164                 dma_unmap_page(&pdev->dev, buffer_info->dma,
4165                                adapter->rx_buffer_len, DMA_FROM_DEVICE);
4166                 buffer_info->dma = 0;
4167
4168                 length = le16_to_cpu(rx_desc->length);
4169
4170                 /* errors is only valid for DD + EOP descriptors */
4171                 if (unlikely((status & E1000_RXD_STAT_EOP) &&
4172                     (rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK))) {
4173                         u8 *mapped = page_address(buffer_info->rxbuf.page);
4174
4175                         if (e1000_tbi_should_accept(adapter, status,
4176                                                     rx_desc->errors,
4177                                                     length, mapped)) {
4178                                 length--;
4179                         } else if (netdev->features & NETIF_F_RXALL) {
4180                                 goto process_skb;
4181                         } else {
4182                                 /* an error means any chain goes out the window
4183                                  * too
4184                                  */
4185                                 if (rx_ring->rx_skb_top)
4186                                         dev_kfree_skb(rx_ring->rx_skb_top);
4187                                 rx_ring->rx_skb_top = NULL;
4188                                 goto next_desc;
4189                         }
4190                 }
4191
4192 #define rxtop rx_ring->rx_skb_top
4193 process_skb:
4194                 if (!(status & E1000_RXD_STAT_EOP)) {
4195                         /* this descriptor is only the beginning (or middle) */
4196                         if (!rxtop) {
4197                                 /* this is the beginning of a chain */
4198                                 rxtop = napi_get_frags(&adapter->napi);
4199                                 if (!rxtop)
4200                                         break;
4201
4202                                 skb_fill_page_desc(rxtop, 0,
4203                                                    buffer_info->rxbuf.page,
4204                                                    0, length);
4205                         } else {
4206                                 /* this is the middle of a chain */
4207                                 skb_fill_page_desc(rxtop,
4208                                     skb_shinfo(rxtop)->nr_frags,
4209                                     buffer_info->rxbuf.page, 0, length);
4210                         }
4211                         e1000_consume_page(buffer_info, rxtop, length);
4212                         goto next_desc;
4213                 } else {
4214                         if (rxtop) {
4215                                 /* end of the chain */
4216                                 skb_fill_page_desc(rxtop,
4217                                     skb_shinfo(rxtop)->nr_frags,
4218                                     buffer_info->rxbuf.page, 0, length);
4219                                 skb = rxtop;
4220                                 rxtop = NULL;
4221                                 e1000_consume_page(buffer_info, skb, length);
4222                         } else {
4223                                 struct page *p;
4224                                 /* no chain, got EOP, this buf is the packet
4225                                  * copybreak to save the put_page/alloc_page
4226                                  */
4227                                 p = buffer_info->rxbuf.page;
4228                                 if (length <= copybreak) {
4229                                         u8 *vaddr;
4230
4231                                         if (likely(!(netdev->features & NETIF_F_RXFCS)))
4232                                                 length -= 4;
4233                                         skb = e1000_alloc_rx_skb(adapter,
4234                                                                  length);
4235                                         if (!skb)
4236                                                 break;
4237
4238                                         vaddr = kmap_atomic(p);
4239                                         memcpy(skb_tail_pointer(skb), vaddr,
4240                                                length);
4241                                         kunmap_atomic(vaddr);
4242                                         /* re-use the page, so don't erase
4243                                          * buffer_info->rxbuf.page
4244                                          */
4245                                         skb_put(skb, length);
4246                                         e1000_rx_checksum(adapter,
4247                                                           status | rx_desc->errors << 24,
4248                                                           le16_to_cpu(rx_desc->csum), skb);
4249
4250                                         total_rx_bytes += skb->len;
4251                                         total_rx_packets++;
4252
4253                                         e1000_receive_skb(adapter, status,
4254                                                           rx_desc->special, skb);
4255                                         goto next_desc;
4256                                 } else {
4257                                         skb = napi_get_frags(&adapter->napi);
4258                                         if (!skb) {
4259                                                 adapter->alloc_rx_buff_failed++;
4260                                                 break;
4261                                         }
4262                                         skb_fill_page_desc(skb, 0, p, 0,
4263                                                            length);
4264                                         e1000_consume_page(buffer_info, skb,
4265                                                            length);
4266                                 }
4267                         }
4268                 }
4269
4270                 /* Receive Checksum Offload XXX recompute due to CRC strip? */
4271                 e1000_rx_checksum(adapter,
4272                                   (u32)(status) |
4273                                   ((u32)(rx_desc->errors) << 24),
4274                                   le16_to_cpu(rx_desc->csum), skb);
4275
4276                 total_rx_bytes += (skb->len - 4); /* don't count FCS */
4277                 if (likely(!(netdev->features & NETIF_F_RXFCS)))
4278                         pskb_trim(skb, skb->len - 4);
4279                 total_rx_packets++;
4280
4281                 if (status & E1000_RXD_STAT_VP) {
4282                         __le16 vlan = rx_desc->special;
4283                         u16 vid = le16_to_cpu(vlan) & E1000_RXD_SPC_VLAN_MASK;
4284
4285                         __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
4286                 }
4287
4288                 napi_gro_frags(&adapter->napi);
4289
4290 next_desc:
4291                 rx_desc->status = 0;
4292
4293                 /* return some buffers to hardware, one at a time is too slow */
4294                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4295                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4296                         cleaned_count = 0;
4297                 }
4298
4299                 /* use prefetched values */
4300                 rx_desc = next_rxd;
4301                 buffer_info = next_buffer;
4302         }
4303         rx_ring->next_to_clean = i;
4304
4305         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4306         if (cleaned_count)
4307                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4308
4309         adapter->total_rx_packets += total_rx_packets;
4310         adapter->total_rx_bytes += total_rx_bytes;
4311         netdev->stats.rx_bytes += total_rx_bytes;
4312         netdev->stats.rx_packets += total_rx_packets;
4313         return cleaned;
4314 }
4315
4316 /* this should improve performance for small packets with large amounts
4317  * of reassembly being done in the stack
4318  */
4319 static struct sk_buff *e1000_copybreak(struct e1000_adapter *adapter,
4320                                        struct e1000_rx_buffer *buffer_info,
4321                                        u32 length, const void *data)
4322 {
4323         struct sk_buff *skb;
4324
4325         if (length > copybreak)
4326                 return NULL;
4327
4328         skb = e1000_alloc_rx_skb(adapter, length);
4329         if (!skb)
4330                 return NULL;
4331
4332         dma_sync_single_for_cpu(&adapter->pdev->dev, buffer_info->dma,
4333                                 length, DMA_FROM_DEVICE);
4334
4335         memcpy(skb_put(skb, length), data, length);
4336
4337         return skb;
4338 }
4339
4340 /**
4341  * e1000_clean_rx_irq - Send received data up the network stack; legacy
4342  * @adapter: board private structure
4343  * @rx_ring: ring to clean
4344  * @work_done: amount of napi work completed this call
4345  * @work_to_do: max amount of work allowed for this call to do
4346  */
4347 static bool e1000_clean_rx_irq(struct e1000_adapter *adapter,
4348                                struct e1000_rx_ring *rx_ring,
4349                                int *work_done, int work_to_do)
4350 {
4351         struct net_device *netdev = adapter->netdev;
4352         struct pci_dev *pdev = adapter->pdev;
4353         struct e1000_rx_desc *rx_desc, *next_rxd;
4354         struct e1000_rx_buffer *buffer_info, *next_buffer;
4355         u32 length;
4356         unsigned int i;
4357         int cleaned_count = 0;
4358         bool cleaned = false;
4359         unsigned int total_rx_bytes=0, total_rx_packets=0;
4360
4361         i = rx_ring->next_to_clean;
4362         rx_desc = E1000_RX_DESC(*rx_ring, i);
4363         buffer_info = &rx_ring->buffer_info[i];
4364
4365         while (rx_desc->status & E1000_RXD_STAT_DD) {
4366                 struct sk_buff *skb;
4367                 u8 *data;
4368                 u8 status;
4369
4370                 if (*work_done >= work_to_do)
4371                         break;
4372                 (*work_done)++;
4373                 dma_rmb(); /* read descriptor and rx_buffer_info after status DD */
4374
4375                 status = rx_desc->status;
4376                 length = le16_to_cpu(rx_desc->length);
4377
4378                 data = buffer_info->rxbuf.data;
4379                 prefetch(data);
4380                 skb = e1000_copybreak(adapter, buffer_info, length, data);
4381                 if (!skb) {
4382                         unsigned int frag_len = e1000_frag_len(adapter);
4383
4384                         skb = build_skb(data - E1000_HEADROOM, frag_len);
4385                         if (!skb) {
4386                                 adapter->alloc_rx_buff_failed++;
4387                                 break;
4388                         }
4389
4390                         skb_reserve(skb, E1000_HEADROOM);
4391                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4392                                          adapter->rx_buffer_len,
4393                                          DMA_FROM_DEVICE);
4394                         buffer_info->dma = 0;
4395                         buffer_info->rxbuf.data = NULL;
4396                 }
4397
4398                 if (++i == rx_ring->count) i = 0;
4399                 next_rxd = E1000_RX_DESC(*rx_ring, i);
4400                 prefetch(next_rxd);
4401
4402                 next_buffer = &rx_ring->buffer_info[i];
4403
4404                 cleaned = true;
4405                 cleaned_count++;
4406
4407                 /* !EOP means multiple descriptors were used to store a single
4408                  * packet, if thats the case we need to toss it.  In fact, we
4409                  * to toss every packet with the EOP bit clear and the next
4410                  * frame that _does_ have the EOP bit set, as it is by
4411                  * definition only a frame fragment
4412                  */
4413                 if (unlikely(!(status & E1000_RXD_STAT_EOP)))
4414                         adapter->discarding = true;
4415
4416                 if (adapter->discarding) {
4417                         /* All receives must fit into a single buffer */
4418                         netdev_dbg(netdev, "Receive packet consumed multiple buffers\n");
4419                         dev_kfree_skb(skb);
4420                         if (status & E1000_RXD_STAT_EOP)
4421                                 adapter->discarding = false;
4422                         goto next_desc;
4423                 }
4424
4425                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
4426                         if (e1000_tbi_should_accept(adapter, status,
4427                                                     rx_desc->errors,
4428                                                     length, data)) {
4429                                 length--;
4430                         } else if (netdev->features & NETIF_F_RXALL) {
4431                                 goto process_skb;
4432                         } else {
4433                                 dev_kfree_skb(skb);
4434                                 goto next_desc;
4435                         }
4436                 }
4437
4438 process_skb:
4439                 total_rx_bytes += (length - 4); /* don't count FCS */
4440                 total_rx_packets++;
4441
4442                 if (likely(!(netdev->features & NETIF_F_RXFCS)))
4443                         /* adjust length to remove Ethernet CRC, this must be
4444                          * done after the TBI_ACCEPT workaround above
4445                          */
4446                         length -= 4;
4447
4448                 if (buffer_info->rxbuf.data == NULL)
4449                         skb_put(skb, length);
4450                 else /* copybreak skb */
4451                         skb_trim(skb, length);
4452
4453                 /* Receive Checksum Offload */
4454                 e1000_rx_checksum(adapter,
4455                                   (u32)(status) |
4456                                   ((u32)(rx_desc->errors) << 24),
4457                                   le16_to_cpu(rx_desc->csum), skb);
4458
4459                 e1000_receive_skb(adapter, status, rx_desc->special, skb);
4460
4461 next_desc:
4462                 rx_desc->status = 0;
4463
4464                 /* return some buffers to hardware, one at a time is too slow */
4465                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
4466                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4467                         cleaned_count = 0;
4468                 }
4469
4470                 /* use prefetched values */
4471                 rx_desc = next_rxd;
4472                 buffer_info = next_buffer;
4473         }
4474         rx_ring->next_to_clean = i;
4475
4476         cleaned_count = E1000_DESC_UNUSED(rx_ring);
4477         if (cleaned_count)
4478                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
4479
4480         adapter->total_rx_packets += total_rx_packets;
4481         adapter->total_rx_bytes += total_rx_bytes;
4482         netdev->stats.rx_bytes += total_rx_bytes;
4483         netdev->stats.rx_packets += total_rx_packets;
4484         return cleaned;
4485 }
4486
4487 /**
4488  * e1000_alloc_jumbo_rx_buffers - Replace used jumbo receive buffers
4489  * @adapter: address of board private structure
4490  * @rx_ring: pointer to receive ring structure
4491  * @cleaned_count: number of buffers to allocate this pass
4492  **/
4493 static void
4494 e1000_alloc_jumbo_rx_buffers(struct e1000_adapter *adapter,
4495                              struct e1000_rx_ring *rx_ring, int cleaned_count)
4496 {
4497         struct pci_dev *pdev = adapter->pdev;
4498         struct e1000_rx_desc *rx_desc;
4499         struct e1000_rx_buffer *buffer_info;
4500         unsigned int i;
4501
4502         i = rx_ring->next_to_use;
4503         buffer_info = &rx_ring->buffer_info[i];
4504
4505         while (cleaned_count--) {
4506                 /* allocate a new page if necessary */
4507                 if (!buffer_info->rxbuf.page) {
4508                         buffer_info->rxbuf.page = alloc_page(GFP_ATOMIC);
4509                         if (unlikely(!buffer_info->rxbuf.page)) {
4510                                 adapter->alloc_rx_buff_failed++;
4511                                 break;
4512                         }
4513                 }
4514
4515                 if (!buffer_info->dma) {
4516                         buffer_info->dma = dma_map_page(&pdev->dev,
4517                                                         buffer_info->rxbuf.page, 0,
4518                                                         adapter->rx_buffer_len,
4519                                                         DMA_FROM_DEVICE);
4520                         if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4521                                 put_page(buffer_info->rxbuf.page);
4522                                 buffer_info->rxbuf.page = NULL;
4523                                 buffer_info->dma = 0;
4524                                 adapter->alloc_rx_buff_failed++;
4525                                 break;
4526                         }
4527                 }
4528
4529                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4530                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4531
4532                 if (unlikely(++i == rx_ring->count))
4533                         i = 0;
4534                 buffer_info = &rx_ring->buffer_info[i];
4535         }
4536
4537         if (likely(rx_ring->next_to_use != i)) {
4538                 rx_ring->next_to_use = i;
4539                 if (unlikely(i-- == 0))
4540                         i = (rx_ring->count - 1);
4541
4542                 /* Force memory writes to complete before letting h/w
4543                  * know there are new descriptors to fetch.  (Only
4544                  * applicable for weak-ordered memory model archs,
4545                  * such as IA-64).
4546                  */
4547                 wmb();
4548                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4549         }
4550 }
4551
4552 /**
4553  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
4554  * @adapter: address of board private structure
4555  **/
4556 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
4557                                    struct e1000_rx_ring *rx_ring,
4558                                    int cleaned_count)
4559 {
4560         struct e1000_hw *hw = &adapter->hw;
4561         struct pci_dev *pdev = adapter->pdev;
4562         struct e1000_rx_desc *rx_desc;
4563         struct e1000_rx_buffer *buffer_info;
4564         unsigned int i;
4565         unsigned int bufsz = adapter->rx_buffer_len;
4566
4567         i = rx_ring->next_to_use;
4568         buffer_info = &rx_ring->buffer_info[i];
4569
4570         while (cleaned_count--) {
4571                 void *data;
4572
4573                 if (buffer_info->rxbuf.data)
4574                         goto skip;
4575
4576                 data = e1000_alloc_frag(adapter);
4577                 if (!data) {
4578                         /* Better luck next round */
4579                         adapter->alloc_rx_buff_failed++;
4580                         break;
4581                 }
4582
4583                 /* Fix for errata 23, can't cross 64kB boundary */
4584                 if (!e1000_check_64k_bound(adapter, data, bufsz)) {
4585                         void *olddata = data;
4586                         e_err(rx_err, "skb align check failed: %u bytes at "
4587                               "%p\n", bufsz, data);
4588                         /* Try again, without freeing the previous */
4589                         data = e1000_alloc_frag(adapter);
4590                         /* Failed allocation, critical failure */
4591                         if (!data) {
4592                                 skb_free_frag(olddata);
4593                                 adapter->alloc_rx_buff_failed++;
4594                                 break;
4595                         }
4596
4597                         if (!e1000_check_64k_bound(adapter, data, bufsz)) {
4598                                 /* give up */
4599                                 skb_free_frag(data);
4600                                 skb_free_frag(olddata);
4601                                 adapter->alloc_rx_buff_failed++;
4602                                 break;
4603                         }
4604
4605                         /* Use new allocation */
4606                         skb_free_frag(olddata);
4607                 }
4608                 buffer_info->dma = dma_map_single(&pdev->dev,
4609                                                   data,
4610                                                   adapter->rx_buffer_len,
4611                                                   DMA_FROM_DEVICE);
4612                 if (dma_mapping_error(&pdev->dev, buffer_info->dma)) {
4613                         skb_free_frag(data);
4614                         buffer_info->dma = 0;
4615                         adapter->alloc_rx_buff_failed++;
4616                         break;
4617                 }
4618
4619                 /* XXX if it was allocated cleanly it will never map to a
4620                  * boundary crossing
4621                  */
4622
4623                 /* Fix for errata 23, can't cross 64kB boundary */
4624                 if (!e1000_check_64k_bound(adapter,
4625                                         (void *)(unsigned long)buffer_info->dma,
4626                                         adapter->rx_buffer_len)) {
4627                         e_err(rx_err, "dma align check failed: %u bytes at "
4628                               "%p\n", adapter->rx_buffer_len,
4629                               (void *)(unsigned long)buffer_info->dma);
4630
4631                         dma_unmap_single(&pdev->dev, buffer_info->dma,
4632                                          adapter->rx_buffer_len,
4633                                          DMA_FROM_DEVICE);
4634
4635                         skb_free_frag(data);
4636                         buffer_info->rxbuf.data = NULL;
4637                         buffer_info->dma = 0;
4638
4639                         adapter->alloc_rx_buff_failed++;
4640                         break;
4641                 }
4642                 buffer_info->rxbuf.data = data;
4643  skip:
4644                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4645                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4646
4647                 if (unlikely(++i == rx_ring->count))
4648                         i = 0;
4649                 buffer_info = &rx_ring->buffer_info[i];
4650         }
4651
4652         if (likely(rx_ring->next_to_use != i)) {
4653                 rx_ring->next_to_use = i;
4654                 if (unlikely(i-- == 0))
4655                         i = (rx_ring->count - 1);
4656
4657                 /* Force memory writes to complete before letting h/w
4658                  * know there are new descriptors to fetch.  (Only
4659                  * applicable for weak-ordered memory model archs,
4660                  * such as IA-64).
4661                  */
4662                 wmb();
4663                 writel(i, hw->hw_addr + rx_ring->rdt);
4664         }
4665 }
4666
4667 /**
4668  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4669  * @adapter:
4670  **/
4671 static void e1000_smartspeed(struct e1000_adapter *adapter)
4672 {
4673         struct e1000_hw *hw = &adapter->hw;
4674         u16 phy_status;
4675         u16 phy_ctrl;
4676
4677         if ((hw->phy_type != e1000_phy_igp) || !hw->autoneg ||
4678            !(hw->autoneg_advertised & ADVERTISE_1000_FULL))
4679                 return;
4680
4681         if (adapter->smartspeed == 0) {
4682                 /* If Master/Slave config fault is asserted twice,
4683                  * we assume back-to-back
4684                  */
4685                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4686                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4687                 e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_status);
4688                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4689                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4690                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4691                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4692                         e1000_write_phy_reg(hw, PHY_1000T_CTRL,
4693                                             phy_ctrl);
4694                         adapter->smartspeed++;
4695                         if (!e1000_phy_setup_autoneg(hw) &&
4696                            !e1000_read_phy_reg(hw, PHY_CTRL,
4697                                                &phy_ctrl)) {
4698                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4699                                              MII_CR_RESTART_AUTO_NEG);
4700                                 e1000_write_phy_reg(hw, PHY_CTRL,
4701                                                     phy_ctrl);
4702                         }
4703                 }
4704                 return;
4705         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4706                 /* If still no link, perhaps using 2/3 pair cable */
4707                 e1000_read_phy_reg(hw, PHY_1000T_CTRL, &phy_ctrl);
4708                 phy_ctrl |= CR_1000T_MS_ENABLE;
4709                 e1000_write_phy_reg(hw, PHY_1000T_CTRL, phy_ctrl);
4710                 if (!e1000_phy_setup_autoneg(hw) &&
4711                    !e1000_read_phy_reg(hw, PHY_CTRL, &phy_ctrl)) {
4712                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4713                                      MII_CR_RESTART_AUTO_NEG);
4714                         e1000_write_phy_reg(hw, PHY_CTRL, phy_ctrl);
4715                 }
4716         }
4717         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4718         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4719                 adapter->smartspeed = 0;
4720 }
4721
4722 /**
4723  * e1000_ioctl -
4724  * @netdev:
4725  * @ifreq:
4726  * @cmd:
4727  **/
4728 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4729 {
4730         switch (cmd) {
4731         case SIOCGMIIPHY:
4732         case SIOCGMIIREG:
4733         case SIOCSMIIREG:
4734                 return e1000_mii_ioctl(netdev, ifr, cmd);
4735         default:
4736                 return -EOPNOTSUPP;
4737         }
4738 }
4739
4740 /**
4741  * e1000_mii_ioctl -
4742  * @netdev:
4743  * @ifreq:
4744  * @cmd:
4745  **/
4746 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
4747                            int cmd)
4748 {
4749         struct e1000_adapter *adapter = netdev_priv(netdev);
4750         struct e1000_hw *hw = &adapter->hw;
4751         struct mii_ioctl_data *data = if_mii(ifr);
4752         int retval;
4753         u16 mii_reg;
4754         unsigned long flags;
4755
4756         if (hw->media_type != e1000_media_type_copper)
4757                 return -EOPNOTSUPP;
4758
4759         switch (cmd) {
4760         case SIOCGMIIPHY:
4761                 data->phy_id = hw->phy_addr;
4762                 break;
4763         case SIOCGMIIREG:
4764                 spin_lock_irqsave(&adapter->stats_lock, flags);
4765                 if (e1000_read_phy_reg(hw, data->reg_num & 0x1F,
4766                                    &data->val_out)) {
4767                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4768                         return -EIO;
4769                 }
4770                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4771                 break;
4772         case SIOCSMIIREG:
4773                 if (data->reg_num & ~(0x1F))
4774                         return -EFAULT;
4775                 mii_reg = data->val_in;
4776                 spin_lock_irqsave(&adapter->stats_lock, flags);
4777                 if (e1000_write_phy_reg(hw, data->reg_num,
4778                                         mii_reg)) {
4779                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4780                         return -EIO;
4781                 }
4782                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4783                 if (hw->media_type == e1000_media_type_copper) {
4784                         switch (data->reg_num) {
4785                         case PHY_CTRL:
4786                                 if (mii_reg & MII_CR_POWER_DOWN)
4787                                         break;
4788                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4789                                         hw->autoneg = 1;
4790                                         hw->autoneg_advertised = 0x2F;
4791                                 } else {
4792                                         u32 speed;
4793                                         if (mii_reg & 0x40)
4794                                                 speed = SPEED_1000;
4795                                         else if (mii_reg & 0x2000)
4796                                                 speed = SPEED_100;
4797                                         else
4798                                                 speed = SPEED_10;
4799                                         retval = e1000_set_spd_dplx(
4800                                                 adapter, speed,
4801                                                 ((mii_reg & 0x100)
4802                                                  ? DUPLEX_FULL :
4803                                                  DUPLEX_HALF));
4804                                         if (retval)
4805                                                 return retval;
4806                                 }
4807                                 if (netif_running(adapter->netdev))
4808                                         e1000_reinit_locked(adapter);
4809                                 else
4810                                         e1000_reset(adapter);
4811                                 break;
4812                         case M88E1000_PHY_SPEC_CTRL:
4813                         case M88E1000_EXT_PHY_SPEC_CTRL:
4814                                 if (e1000_phy_reset(hw))
4815                                         return -EIO;
4816                                 break;
4817                         }
4818                 } else {
4819                         switch (data->reg_num) {
4820                         case PHY_CTRL:
4821                                 if (mii_reg & MII_CR_POWER_DOWN)
4822                                         break;
4823                                 if (netif_running(adapter->netdev))
4824                                         e1000_reinit_locked(adapter);
4825                                 else
4826                                         e1000_reset(adapter);
4827                                 break;
4828                         }
4829                 }
4830                 break;
4831         default:
4832                 return -EOPNOTSUPP;
4833         }
4834         return E1000_SUCCESS;
4835 }
4836
4837 void e1000_pci_set_mwi(struct e1000_hw *hw)
4838 {
4839         struct e1000_adapter *adapter = hw->back;
4840         int ret_val = pci_set_mwi(adapter->pdev);
4841
4842         if (ret_val)
4843                 e_err(probe, "Error in setting MWI\n");
4844 }
4845
4846 void e1000_pci_clear_mwi(struct e1000_hw *hw)
4847 {
4848         struct e1000_adapter *adapter = hw->back;
4849
4850         pci_clear_mwi(adapter->pdev);
4851 }
4852
4853 int e1000_pcix_get_mmrbc(struct e1000_hw *hw)
4854 {
4855         struct e1000_adapter *adapter = hw->back;
4856         return pcix_get_mmrbc(adapter->pdev);
4857 }
4858
4859 void e1000_pcix_set_mmrbc(struct e1000_hw *hw, int mmrbc)
4860 {
4861         struct e1000_adapter *adapter = hw->back;
4862         pcix_set_mmrbc(adapter->pdev, mmrbc);
4863 }
4864
4865 void e1000_io_write(struct e1000_hw *hw, unsigned long port, u32 value)
4866 {
4867         outl(value, port);
4868 }
4869
4870 static bool e1000_vlan_used(struct e1000_adapter *adapter)
4871 {
4872         u16 vid;
4873
4874         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4875                 return true;
4876         return false;
4877 }
4878
4879 static void __e1000_vlan_mode(struct e1000_adapter *adapter,
4880                               netdev_features_t features)
4881 {
4882         struct e1000_hw *hw = &adapter->hw;
4883         u32 ctrl;
4884
4885         ctrl = er32(CTRL);
4886         if (features & NETIF_F_HW_VLAN_CTAG_RX) {
4887                 /* enable VLAN tag insert/strip */
4888                 ctrl |= E1000_CTRL_VME;
4889         } else {
4890                 /* disable VLAN tag insert/strip */
4891                 ctrl &= ~E1000_CTRL_VME;
4892         }
4893         ew32(CTRL, ctrl);
4894 }
4895 static void e1000_vlan_filter_on_off(struct e1000_adapter *adapter,
4896                                      bool filter_on)
4897 {
4898         struct e1000_hw *hw = &adapter->hw;
4899         u32 rctl;
4900
4901         if (!test_bit(__E1000_DOWN, &adapter->flags))
4902                 e1000_irq_disable(adapter);
4903
4904         __e1000_vlan_mode(adapter, adapter->netdev->features);
4905         if (filter_on) {
4906                 /* enable VLAN receive filtering */
4907                 rctl = er32(RCTL);
4908                 rctl &= ~E1000_RCTL_CFIEN;
4909                 if (!(adapter->netdev->flags & IFF_PROMISC))
4910                         rctl |= E1000_RCTL_VFE;
4911                 ew32(RCTL, rctl);
4912                 e1000_update_mng_vlan(adapter);
4913         } else {
4914                 /* disable VLAN receive filtering */
4915                 rctl = er32(RCTL);
4916                 rctl &= ~E1000_RCTL_VFE;
4917                 ew32(RCTL, rctl);
4918         }
4919
4920         if (!test_bit(__E1000_DOWN, &adapter->flags))
4921                 e1000_irq_enable(adapter);
4922 }
4923
4924 static void e1000_vlan_mode(struct net_device *netdev,
4925                             netdev_features_t features)
4926 {
4927         struct e1000_adapter *adapter = netdev_priv(netdev);
4928
4929         if (!test_bit(__E1000_DOWN, &adapter->flags))
4930                 e1000_irq_disable(adapter);
4931
4932         __e1000_vlan_mode(adapter, features);
4933
4934         if (!test_bit(__E1000_DOWN, &adapter->flags))
4935                 e1000_irq_enable(adapter);
4936 }
4937
4938 static int e1000_vlan_rx_add_vid(struct net_device *netdev,
4939                                  __be16 proto, u16 vid)
4940 {
4941         struct e1000_adapter *adapter = netdev_priv(netdev);
4942         struct e1000_hw *hw = &adapter->hw;
4943         u32 vfta, index;
4944
4945         if ((hw->mng_cookie.status &
4946              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4947             (vid == adapter->mng_vlan_id))
4948                 return 0;
4949
4950         if (!e1000_vlan_used(adapter))
4951                 e1000_vlan_filter_on_off(adapter, true);
4952
4953         /* add VID to filter table */
4954         index = (vid >> 5) & 0x7F;
4955         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4956         vfta |= (1 << (vid & 0x1F));
4957         e1000_write_vfta(hw, index, vfta);
4958
4959         set_bit(vid, adapter->active_vlans);
4960
4961         return 0;
4962 }
4963
4964 static int e1000_vlan_rx_kill_vid(struct net_device *netdev,
4965                                   __be16 proto, u16 vid)
4966 {
4967         struct e1000_adapter *adapter = netdev_priv(netdev);
4968         struct e1000_hw *hw = &adapter->hw;
4969         u32 vfta, index;
4970
4971         if (!test_bit(__E1000_DOWN, &adapter->flags))
4972                 e1000_irq_disable(adapter);
4973         if (!test_bit(__E1000_DOWN, &adapter->flags))
4974                 e1000_irq_enable(adapter);
4975
4976         /* remove VID from filter table */
4977         index = (vid >> 5) & 0x7F;
4978         vfta = E1000_READ_REG_ARRAY(hw, VFTA, index);
4979         vfta &= ~(1 << (vid & 0x1F));
4980         e1000_write_vfta(hw, index, vfta);
4981
4982         clear_bit(vid, adapter->active_vlans);
4983
4984         if (!e1000_vlan_used(adapter))
4985                 e1000_vlan_filter_on_off(adapter, false);
4986
4987         return 0;
4988 }
4989
4990 static void e1000_restore_vlan(struct e1000_adapter *adapter)
4991 {
4992         u16 vid;
4993
4994         if (!e1000_vlan_used(adapter))
4995                 return;
4996
4997         e1000_vlan_filter_on_off(adapter, true);
4998         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
4999                 e1000_vlan_rx_add_vid(adapter->netdev, htons(ETH_P_8021Q), vid);
5000 }
5001
5002 int e1000_set_spd_dplx(struct e1000_adapter *adapter, u32 spd, u8 dplx)
5003 {
5004         struct e1000_hw *hw = &adapter->hw;
5005
5006         hw->autoneg = 0;
5007
5008         /* Make sure dplx is at most 1 bit and lsb of speed is not set
5009          * for the switch() below to work
5010          */
5011         if ((spd & 1) || (dplx & ~1))
5012                 goto err_inval;
5013
5014         /* Fiber NICs only allow 1000 gbps Full duplex */
5015         if ((hw->media_type == e1000_media_type_fiber) &&
5016             spd != SPEED_1000 &&
5017             dplx != DUPLEX_FULL)
5018                 goto err_inval;
5019
5020         switch (spd + dplx) {
5021         case SPEED_10 + DUPLEX_HALF:
5022                 hw->forced_speed_duplex = e1000_10_half;
5023                 break;
5024         case SPEED_10 + DUPLEX_FULL:
5025                 hw->forced_speed_duplex = e1000_10_full;
5026                 break;
5027         case SPEED_100 + DUPLEX_HALF:
5028                 hw->forced_speed_duplex = e1000_100_half;
5029                 break;
5030         case SPEED_100 + DUPLEX_FULL:
5031                 hw->forced_speed_duplex = e1000_100_full;
5032                 break;
5033         case SPEED_1000 + DUPLEX_FULL:
5034                 hw->autoneg = 1;
5035                 hw->autoneg_advertised = ADVERTISE_1000_FULL;
5036                 break;
5037         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5038         default:
5039                 goto err_inval;
5040         }
5041
5042         /* clear MDI, MDI(-X) override is only allowed when autoneg enabled */
5043         hw->mdix = AUTO_ALL_MODES;
5044
5045         return 0;
5046
5047 err_inval:
5048         e_err(probe, "Unsupported Speed/Duplex configuration\n");
5049         return -EINVAL;
5050 }
5051
5052 static int __e1000_shutdown(struct pci_dev *pdev, bool *enable_wake)
5053 {
5054         struct net_device *netdev = pci_get_drvdata(pdev);
5055         struct e1000_adapter *adapter = netdev_priv(netdev);
5056         struct e1000_hw *hw = &adapter->hw;
5057         u32 ctrl, ctrl_ext, rctl, status;
5058         u32 wufc = adapter->wol;
5059 #ifdef CONFIG_PM
5060         int retval = 0;
5061 #endif
5062
5063         netif_device_detach(netdev);
5064
5065         if (netif_running(netdev)) {
5066                 int count = E1000_CHECK_RESET_COUNT;
5067
5068                 while (test_bit(__E1000_RESETTING, &adapter->flags) && count--)
5069                         usleep_range(10000, 20000);
5070
5071                 WARN_ON(test_bit(__E1000_RESETTING, &adapter->flags));
5072                 e1000_down(adapter);
5073         }
5074
5075 #ifdef CONFIG_PM
5076         retval = pci_save_state(pdev);
5077         if (retval)
5078                 return retval;
5079 #endif
5080
5081         status = er32(STATUS);
5082         if (status & E1000_STATUS_LU)
5083                 wufc &= ~E1000_WUFC_LNKC;
5084
5085         if (wufc) {
5086                 e1000_setup_rctl(adapter);
5087                 e1000_set_rx_mode(netdev);
5088
5089                 rctl = er32(RCTL);
5090
5091                 /* turn on all-multi mode if wake on multicast is enabled */
5092                 if (wufc & E1000_WUFC_MC)
5093                         rctl |= E1000_RCTL_MPE;
5094
5095                 /* enable receives in the hardware */
5096                 ew32(RCTL, rctl | E1000_RCTL_EN);
5097
5098                 if (hw->mac_type >= e1000_82540) {
5099                         ctrl = er32(CTRL);
5100                         /* advertise wake from D3Cold */
5101                         #define E1000_CTRL_ADVD3WUC 0x00100000
5102                         /* phy power management enable */
5103                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5104                         ctrl |= E1000_CTRL_ADVD3WUC |
5105                                 E1000_CTRL_EN_PHY_PWR_MGMT;
5106                         ew32(CTRL, ctrl);
5107                 }
5108
5109                 if (hw->media_type == e1000_media_type_fiber ||
5110                     hw->media_type == e1000_media_type_internal_serdes) {
5111                         /* keep the laser running in D3 */
5112                         ctrl_ext = er32(CTRL_EXT);
5113                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
5114                         ew32(CTRL_EXT, ctrl_ext);
5115                 }
5116
5117                 ew32(WUC, E1000_WUC_PME_EN);
5118                 ew32(WUFC, wufc);
5119         } else {
5120                 ew32(WUC, 0);
5121                 ew32(WUFC, 0);
5122         }
5123
5124         e1000_release_manageability(adapter);
5125
5126         *enable_wake = !!wufc;
5127
5128         /* make sure adapter isn't asleep if manageability is enabled */
5129         if (adapter->en_mng_pt)
5130                 *enable_wake = true;
5131
5132         if (netif_running(netdev))
5133                 e1000_free_irq(adapter);
5134
5135         pci_disable_device(pdev);
5136
5137         return 0;
5138 }
5139
5140 #ifdef CONFIG_PM
5141 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state)
5142 {
5143         int retval;
5144         bool wake;
5145
5146         retval = __e1000_shutdown(pdev, &wake);
5147         if (retval)
5148                 return retval;
5149
5150         if (wake) {
5151                 pci_prepare_to_sleep(pdev);
5152         } else {
5153                 pci_wake_from_d3(pdev, false);
5154                 pci_set_power_state(pdev, PCI_D3hot);
5155         }
5156
5157         return 0;
5158 }
5159
5160 static int e1000_resume(struct pci_dev *pdev)
5161 {
5162         struct net_device *netdev = pci_get_drvdata(pdev);
5163         struct e1000_adapter *adapter = netdev_priv(netdev);
5164         struct e1000_hw *hw = &adapter->hw;
5165         u32 err;
5166
5167         pci_set_power_state(pdev, PCI_D0);
5168         pci_restore_state(pdev);
5169         pci_save_state(pdev);
5170
5171         if (adapter->need_ioport)
5172                 err = pci_enable_device(pdev);
5173         else
5174                 err = pci_enable_device_mem(pdev);
5175         if (err) {
5176                 pr_err("Cannot enable PCI device from suspend\n");
5177                 return err;
5178         }
5179         pci_set_master(pdev);
5180
5181         pci_enable_wake(pdev, PCI_D3hot, 0);
5182         pci_enable_wake(pdev, PCI_D3cold, 0);
5183
5184         if (netif_running(netdev)) {
5185                 err = e1000_request_irq(adapter);
5186                 if (err)
5187                         return err;
5188         }
5189
5190         e1000_power_up_phy(adapter);
5191         e1000_reset(adapter);
5192         ew32(WUS, ~0);
5193
5194         e1000_init_manageability(adapter);
5195
5196         if (netif_running(netdev))
5197                 e1000_up(adapter);
5198
5199         netif_device_attach(netdev);
5200
5201         return 0;
5202 }
5203 #endif
5204
5205 static void e1000_shutdown(struct pci_dev *pdev)
5206 {
5207         bool wake;
5208
5209         __e1000_shutdown(pdev, &wake);
5210
5211         if (system_state == SYSTEM_POWER_OFF) {
5212                 pci_wake_from_d3(pdev, wake);
5213                 pci_set_power_state(pdev, PCI_D3hot);
5214         }
5215 }
5216
5217 #ifdef CONFIG_NET_POLL_CONTROLLER
5218 /* Polling 'interrupt' - used by things like netconsole to send skbs
5219  * without having to re-enable interrupts. It's not called while
5220  * the interrupt routine is executing.
5221  */
5222 static void e1000_netpoll(struct net_device *netdev)
5223 {
5224         struct e1000_adapter *adapter = netdev_priv(netdev);
5225
5226         disable_irq(adapter->pdev->irq);
5227         e1000_intr(adapter->pdev->irq, netdev);
5228         enable_irq(adapter->pdev->irq);
5229 }
5230 #endif
5231
5232 /**
5233  * e1000_io_error_detected - called when PCI error is detected
5234  * @pdev: Pointer to PCI device
5235  * @state: The current pci connection state
5236  *
5237  * This function is called after a PCI bus error affecting
5238  * this device has been detected.
5239  */
5240 static pci_ers_result_t e1000_io_error_detected(struct pci_dev *pdev,
5241                                                 pci_channel_state_t state)
5242 {
5243         struct net_device *netdev = pci_get_drvdata(pdev);
5244         struct e1000_adapter *adapter = netdev_priv(netdev);
5245
5246         netif_device_detach(netdev);
5247
5248         if (state == pci_channel_io_perm_failure)
5249                 return PCI_ERS_RESULT_DISCONNECT;
5250
5251         if (netif_running(netdev))
5252                 e1000_down(adapter);
5253         pci_disable_device(pdev);
5254
5255         /* Request a slot slot reset. */
5256         return PCI_ERS_RESULT_NEED_RESET;
5257 }
5258
5259 /**
5260  * e1000_io_slot_reset - called after the pci bus has been reset.
5261  * @pdev: Pointer to PCI device
5262  *
5263  * Restart the card from scratch, as if from a cold-boot. Implementation
5264  * resembles the first-half of the e1000_resume routine.
5265  */
5266 static pci_ers_result_t e1000_io_slot_reset(struct pci_dev *pdev)
5267 {
5268         struct net_device *netdev = pci_get_drvdata(pdev);
5269         struct e1000_adapter *adapter = netdev_priv(netdev);
5270         struct e1000_hw *hw = &adapter->hw;
5271         int err;
5272
5273         if (adapter->need_ioport)
5274                 err = pci_enable_device(pdev);
5275         else
5276                 err = pci_enable_device_mem(pdev);
5277         if (err) {
5278                 pr_err("Cannot re-enable PCI device after reset.\n");
5279                 return PCI_ERS_RESULT_DISCONNECT;
5280         }
5281         pci_set_master(pdev);
5282
5283         pci_enable_wake(pdev, PCI_D3hot, 0);
5284         pci_enable_wake(pdev, PCI_D3cold, 0);
5285
5286         e1000_reset(adapter);
5287         ew32(WUS, ~0);
5288
5289         return PCI_ERS_RESULT_RECOVERED;
5290 }
5291
5292 /**
5293  * e1000_io_resume - called when traffic can start flowing again.
5294  * @pdev: Pointer to PCI device
5295  *
5296  * This callback is called when the error recovery driver tells us that
5297  * its OK to resume normal operation. Implementation resembles the
5298  * second-half of the e1000_resume routine.
5299  */
5300 static void e1000_io_resume(struct pci_dev *pdev)
5301 {
5302         struct net_device *netdev = pci_get_drvdata(pdev);
5303         struct e1000_adapter *adapter = netdev_priv(netdev);
5304
5305         e1000_init_manageability(adapter);
5306
5307         if (netif_running(netdev)) {
5308                 if (e1000_up(adapter)) {
5309                         pr_info("can't bring device back up after reset\n");
5310                         return;
5311                 }
5312         }
5313
5314         netif_device_attach(netdev);
5315 }
5316
5317 /* e1000_main.c */