drivers/net/e1000/e1000_main.c

   1 /*******************************************************************************
   2
   3
   4   Copyright(c) 1999 - 2005 Intel Corporation. All rights reserved.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU General Public License as published by the Free
   8   Software Foundation; either version 2 of the License, or (at your option)
   9   any later version.
  10
  11   This program is distributed in the hope that it will be useful, but WITHOUT
  12   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  13   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  14   more details.
  15
  16   You should have received a copy of the GNU General Public License along with
  17   this program; if not, write to the Free Software Foundation, Inc., 59
  18   Temple Place - Suite 330, Boston, MA  02111-1307, USA.
  19
  20   The full GNU General Public License is included in this distribution in the
  21   file called LICENSE.
  22
  23   Contact Information:
  24   Linux NICS <linux.nics@intel.com>
  25   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  26
  27 *******************************************************************************/
  28
  29 #include "e1000.h"
  30
  31 /* Change Log
  32  * 6.0.58       4/20/05
  33  *   o Accepted ethtool cleanup patch from Stephen Hemminger
  34  * 6.0.44+      2/15/05
  35  *   o applied Anton's patch to resolve tx hang in hardware
  36  *   o Applied Andrew Mortons patch - e1000 stops working after resume
  37  */
  38
  39 char e1000_driver_name[] = "e1000";
  40 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
  41 #ifndef CONFIG_E1000_NAPI
  42 #define DRIVERNAPI
  43 #else
  44 #define DRIVERNAPI "-NAPI"
  45 #endif
  46 #define DRV_VERSION "6.3.9-k2"DRIVERNAPI
  47 char e1000_driver_version[] = DRV_VERSION;
  48 static char e1000_copyright[] = "Copyright (c) 1999-2005 Intel Corporation.";
  49
  50 /* e1000_pci_tbl - PCI Device ID Table
  51  *
  52  * Last entry must be all 0s
  53  *
  54  * Macro expands to...
  55  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
  56  */
  57 static struct pci_device_id e1000_pci_tbl[] = {
  58         INTEL_E1000_ETHERNET_DEVICE(0x1000),
  59         INTEL_E1000_ETHERNET_DEVICE(0x1001),
  60         INTEL_E1000_ETHERNET_DEVICE(0x1004),
  61         INTEL_E1000_ETHERNET_DEVICE(0x1008),
  62         INTEL_E1000_ETHERNET_DEVICE(0x1009),
  63         INTEL_E1000_ETHERNET_DEVICE(0x100C),
  64         INTEL_E1000_ETHERNET_DEVICE(0x100D),
  65         INTEL_E1000_ETHERNET_DEVICE(0x100E),
  66         INTEL_E1000_ETHERNET_DEVICE(0x100F),
  67         INTEL_E1000_ETHERNET_DEVICE(0x1010),
  68         INTEL_E1000_ETHERNET_DEVICE(0x1011),
  69         INTEL_E1000_ETHERNET_DEVICE(0x1012),
  70         INTEL_E1000_ETHERNET_DEVICE(0x1013),
  71         INTEL_E1000_ETHERNET_DEVICE(0x1014),
  72         INTEL_E1000_ETHERNET_DEVICE(0x1015),
  73         INTEL_E1000_ETHERNET_DEVICE(0x1016),
  74         INTEL_E1000_ETHERNET_DEVICE(0x1017),
  75         INTEL_E1000_ETHERNET_DEVICE(0x1018),
  76         INTEL_E1000_ETHERNET_DEVICE(0x1019),
  77         INTEL_E1000_ETHERNET_DEVICE(0x101A),
  78         INTEL_E1000_ETHERNET_DEVICE(0x101D),
  79         INTEL_E1000_ETHERNET_DEVICE(0x101E),
  80         INTEL_E1000_ETHERNET_DEVICE(0x1026),
  81         INTEL_E1000_ETHERNET_DEVICE(0x1027),
  82         INTEL_E1000_ETHERNET_DEVICE(0x1028),
  83         INTEL_E1000_ETHERNET_DEVICE(0x105E),
  84         INTEL_E1000_ETHERNET_DEVICE(0x105F),
  85         INTEL_E1000_ETHERNET_DEVICE(0x1060),
  86         INTEL_E1000_ETHERNET_DEVICE(0x1075),
  87         INTEL_E1000_ETHERNET_DEVICE(0x1076),
  88         INTEL_E1000_ETHERNET_DEVICE(0x1077),
  89         INTEL_E1000_ETHERNET_DEVICE(0x1078),
  90         INTEL_E1000_ETHERNET_DEVICE(0x1079),
  91         INTEL_E1000_ETHERNET_DEVICE(0x107A),
  92         INTEL_E1000_ETHERNET_DEVICE(0x107B),
  93         INTEL_E1000_ETHERNET_DEVICE(0x107C),
  94         INTEL_E1000_ETHERNET_DEVICE(0x107D),
  95         INTEL_E1000_ETHERNET_DEVICE(0x107E),
  96         INTEL_E1000_ETHERNET_DEVICE(0x107F),
  97         INTEL_E1000_ETHERNET_DEVICE(0x108A),
  98         INTEL_E1000_ETHERNET_DEVICE(0x108B),
  99         INTEL_E1000_ETHERNET_DEVICE(0x108C),
 100         INTEL_E1000_ETHERNET_DEVICE(0x109A),
 101         /* required last entry */
 102         {0,}
 103 };
 104
 105 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
 106
 107 int e1000_up(struct e1000_adapter *adapter);
 108 void e1000_down(struct e1000_adapter *adapter);
 109 void e1000_reset(struct e1000_adapter *adapter);
 110 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
 111 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
 112 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
 113 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
 114 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
 115 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
 116                                     struct e1000_tx_ring *txdr);
 117 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
 118                                     struct e1000_rx_ring *rxdr);
 119 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
 120                                     struct e1000_tx_ring *tx_ring);
 121 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
 122                                     struct e1000_rx_ring *rx_ring);
 123 void e1000_update_stats(struct e1000_adapter *adapter);
 124
 125 /* Local Function Prototypes */
 126
 127 static int e1000_init_module(void);
 128 static void e1000_exit_module(void);
 129 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
 130 static void __devexit e1000_remove(struct pci_dev *pdev);
 131 static int e1000_alloc_queues(struct e1000_adapter *adapter);
 132 #ifdef CONFIG_E1000_MQ
 133 static void e1000_setup_queue_mapping(struct e1000_adapter *adapter);
 134 #endif
 135 static int e1000_sw_init(struct e1000_adapter *adapter);
 136 static int e1000_open(struct net_device *netdev);
 137 static int e1000_close(struct net_device *netdev);
 138 static void e1000_configure_tx(struct e1000_adapter *adapter);
 139 static void e1000_configure_rx(struct e1000_adapter *adapter);
 140 static void e1000_setup_rctl(struct e1000_adapter *adapter);
 141 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
 142 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
 143 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
 144                                 struct e1000_tx_ring *tx_ring);
 145 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
 146                                 struct e1000_rx_ring *rx_ring);
 147 static void e1000_set_multi(struct net_device *netdev);
 148 static void e1000_update_phy_info(unsigned long data);
 149 static void e1000_watchdog(unsigned long data);
 150 static void e1000_watchdog_task(struct e1000_adapter *adapter);
 151 static void e1000_82547_tx_fifo_stall(unsigned long data);
 152 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
 153 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
 154 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
 155 static int e1000_set_mac(struct net_device *netdev, void *p);
 156 static irqreturn_t e1000_intr(int irq, void *data, struct pt_regs *regs);
 157 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
 158                                     struct e1000_tx_ring *tx_ring);
 159 #ifdef CONFIG_E1000_NAPI
 160 static int e1000_clean(struct net_device *poll_dev, int *budget);
 161 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
 162                                     struct e1000_rx_ring *rx_ring,
 163                                     int *work_done, int work_to_do);
 164 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 165                                        struct e1000_rx_ring *rx_ring,
 166                                        int *work_done, int work_to_do);
 167 #else
 168 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
 169                                     struct e1000_rx_ring *rx_ring);
 170 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 171                                        struct e1000_rx_ring *rx_ring);
 172 #endif
 173 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 174                                    struct e1000_rx_ring *rx_ring);
 175 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 176                                       struct e1000_rx_ring *rx_ring);
 177 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
 178 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
 179                            int cmd);
 180 void e1000_set_ethtool_ops(struct net_device *netdev);
 181 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
 182 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
 183 static void e1000_tx_timeout(struct net_device *dev);
 184 static void e1000_tx_timeout_task(struct net_device *dev);
 185 static void e1000_smartspeed(struct e1000_adapter *adapter);
 186 static inline int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
 187                                               struct sk_buff *skb);
 188
 189 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
 190 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
 191 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
 192 static void e1000_restore_vlan(struct e1000_adapter *adapter);
 193
 194 #ifdef CONFIG_PM
 195 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
 196 static int e1000_resume(struct pci_dev *pdev);
 197 #endif
 198
 199 #ifdef CONFIG_NET_POLL_CONTROLLER
 200 /* for netdump / net console */
 201 static void e1000_netpoll (struct net_device *netdev);
 202 #endif
 203
 204 #ifdef CONFIG_E1000_MQ
 205 /* for multiple Rx queues */
 206 void e1000_rx_schedule(void *data);
 207 #endif
 208
 209 /* Exported from other modules */
 210
 211 extern void e1000_check_options(struct e1000_adapter *adapter);
 212
 213 static struct pci_driver e1000_driver = {
 214         .name     = e1000_driver_name,
 215         .id_table = e1000_pci_tbl,
 216         .probe    = e1000_probe,
 217         .remove   = __devexit_p(e1000_remove),
 218         /* Power Managment Hooks */
 219 #ifdef CONFIG_PM
 220         .suspend  = e1000_suspend,
 221         .resume   = e1000_resume
 222 #endif
 223 };
 224
 225 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 226 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
 227 MODULE_LICENSE("GPL");
 228 MODULE_VERSION(DRV_VERSION);
 229
 230 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
 231 module_param(debug, int, 0);
 232 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 233
 234 /**
 235  * e1000_init_module - Driver Registration Routine
 236  *
 237  * e1000_init_module is the first routine called when the driver is
 238  * loaded. All it does is register with the PCI subsystem.
 239  **/
 240
 241 static int __init
 242 e1000_init_module(void)
 243 {
 244         int ret;
 245         printk(KERN_INFO "%s - version %s\n",
 246                e1000_driver_string, e1000_driver_version);
 247
 248         printk(KERN_INFO "%s\n", e1000_copyright);
 249
 250         ret = pci_module_init(&e1000_driver);
 251
 252         return ret;
 253 }
 254
 255 module_init(e1000_init_module);
 256
 257 /**
 258  * e1000_exit_module - Driver Exit Cleanup Routine
 259  *
 260  * e1000_exit_module is called just before the driver is removed
 261  * from memory.
 262  **/
 263
 264 static void __exit
 265 e1000_exit_module(void)
 266 {
 267         pci_unregister_driver(&e1000_driver);
 268 }
 269
 270 module_exit(e1000_exit_module);
 271
 272 /**
 273  * e1000_irq_disable - Mask off interrupt generation on the NIC
 274  * @adapter: board private structure
 275  **/
 276
 277 static inline void
 278 e1000_irq_disable(struct e1000_adapter *adapter)
 279 {
 280         atomic_inc(&adapter->irq_sem);
 281         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
 282         E1000_WRITE_FLUSH(&adapter->hw);
 283         synchronize_irq(adapter->pdev->irq);
 284 }
 285
 286 /**
 287  * e1000_irq_enable - Enable default interrupt generation settings
 288  * @adapter: board private structure
 289  **/
 290
 291 static inline void
 292 e1000_irq_enable(struct e1000_adapter *adapter)
 293 {
 294         if(likely(atomic_dec_and_test(&adapter->irq_sem))) {
 295                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
 296                 E1000_WRITE_FLUSH(&adapter->hw);
 297         }
 298 }
 299
 300 static void
 301 e1000_update_mng_vlan(struct e1000_adapter *adapter)
 302 {
 303         struct net_device *netdev = adapter->netdev;
 304         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
 305         uint16_t old_vid = adapter->mng_vlan_id;
 306         if(adapter->vlgrp) {
 307                 if(!adapter->vlgrp->vlan_devices[vid]) {
 308                         if(adapter->hw.mng_cookie.status &
 309                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
 310                                 e1000_vlan_rx_add_vid(netdev, vid);
 311                                 adapter->mng_vlan_id = vid;
 312                         } else
 313                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
 314
 315                         if((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
 316                                         (vid != old_vid) &&
 317                                         !adapter->vlgrp->vlan_devices[old_vid])
 318                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
 319                 }
 320         }
 321 }
 322
 323 /**
 324  * e1000_release_hw_control - release control of the h/w to f/w
 325  * @adapter: address of board private structure
 326  *
 327  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
 328  * For ASF and Pass Through versions of f/w this means that the
 329  * driver is no longer loaded. For AMT version (only with 82573) i
 330  * of the f/w this means that the netowrk i/f is closed.
 331  *
 332  **/
 333
 334 static inline void
 335 e1000_release_hw_control(struct e1000_adapter *adapter)
 336 {
 337         uint32_t ctrl_ext;
 338         uint32_t swsm;
 339
 340         /* Let firmware taken over control of h/w */
 341         switch (adapter->hw.mac_type) {
 342         case e1000_82571:
 343         case e1000_82572:
 344                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
 345                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
 346                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 347                 break;
 348         case e1000_82573:
 349                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
 350                 E1000_WRITE_REG(&adapter->hw, SWSM,
 351                                 swsm & ~E1000_SWSM_DRV_LOAD);
 352         default:
 353                 break;
 354         }
 355 }
 356
 357 /**
 358  * e1000_get_hw_control - get control of the h/w from f/w
 359  * @adapter: address of board private structure
 360  *
 361  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
 362  * For ASF and Pass Through versions of f/w this means that
 363  * the driver is loaded. For AMT version (only with 82573)
 364  * of the f/w this means that the netowrk i/f is open.
 365  *
 366  **/
 367
 368 static inline void
 369 e1000_get_hw_control(struct e1000_adapter *adapter)
 370 {
 371         uint32_t ctrl_ext;
 372         uint32_t swsm;
 373         /* Let firmware know the driver has taken over */
 374         switch (adapter->hw.mac_type) {
 375         case e1000_82571:
 376         case e1000_82572:
 377                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
 378                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
 379                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 380                 break;
 381         case e1000_82573:
 382                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
 383                 E1000_WRITE_REG(&adapter->hw, SWSM,
 384                                 swsm | E1000_SWSM_DRV_LOAD);
 385                 break;
 386         default:
 387                 break;
 388         }
 389 }
 390
 391 int
 392 e1000_up(struct e1000_adapter *adapter)
 393 {
 394         struct net_device *netdev = adapter->netdev;
 395         int i, err;
 396
 397         /* hardware has been reset, we need to reload some things */
 398
 399         /* Reset the PHY if it was previously powered down */
 400         if(adapter->hw.media_type == e1000_media_type_copper) {
 401                 uint16_t mii_reg;
 402                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
 403                 if(mii_reg & MII_CR_POWER_DOWN)
 404                         e1000_phy_reset(&adapter->hw);
 405         }
 406
 407         e1000_set_multi(netdev);
 408
 409         e1000_restore_vlan(adapter);
 410
 411         e1000_configure_tx(adapter);
 412         e1000_setup_rctl(adapter);
 413         e1000_configure_rx(adapter);
 414         for (i = 0; i < adapter->num_rx_queues; i++) {
 415                 adapter->alloc_rx_buf(adapter, &adapter->rx_ring[i]);
 416         }
 417
 418 #ifdef CONFIG_PCI_MSI
 419         if(adapter->hw.mac_type > e1000_82547_rev_2) {
 420                 adapter->have_msi = TRUE;
 421                 if((err = pci_enable_msi(adapter->pdev))) {
 422                         DPRINTK(PROBE, ERR,
 423                          "Unable to allocate MSI interrupt Error: %d\n", err);
 424                         adapter->have_msi = FALSE;
 425                 }
 426         }
 427 #endif
 428         if((err = request_irq(adapter->pdev->irq, &e1000_intr,
 429                               SA_SHIRQ | SA_SAMPLE_RANDOM,
 430                               netdev->name, netdev))) {
 431                 DPRINTK(PROBE, ERR,
 432                     "Unable to allocate interrupt Error: %d\n", err);
 433                 return err;
 434         }
 435
 436 #ifdef CONFIG_E1000_MQ
 437         e1000_setup_queue_mapping(adapter);
 438 #endif
 439
 440         adapter->tx_queue_len = netdev->tx_queue_len;
 441
 442         mod_timer(&adapter->watchdog_timer, jiffies);
 443
 444 #ifdef CONFIG_E1000_NAPI
 445         netif_poll_enable(netdev);
 446 #endif
 447         e1000_irq_enable(adapter);
 448
 449         return 0;
 450 }
 451
 452 void
 453 e1000_down(struct e1000_adapter *adapter)
 454 {
 455         struct net_device *netdev = adapter->netdev;
 456         boolean_t mng_mode_enabled = (adapter->hw.mac_type >= e1000_82571) &&
 457                                      e1000_check_mng_mode(&adapter->hw);
 458
 459         e1000_irq_disable(adapter);
 460 #ifdef CONFIG_E1000_MQ
 461         while (atomic_read(&adapter->rx_sched_call_data.count) != 0);
 462 #endif
 463         free_irq(adapter->pdev->irq, netdev);
 464 #ifdef CONFIG_PCI_MSI
 465         if(adapter->hw.mac_type > e1000_82547_rev_2 &&
 466            adapter->have_msi == TRUE)
 467                 pci_disable_msi(adapter->pdev);
 468 #endif
 469         del_timer_sync(&adapter->tx_fifo_stall_timer);
 470         del_timer_sync(&adapter->watchdog_timer);
 471         del_timer_sync(&adapter->phy_info_timer);
 472
 473 #ifdef CONFIG_E1000_NAPI
 474         netif_poll_disable(netdev);
 475 #endif
 476         netdev->tx_queue_len = adapter->tx_queue_len;
 477         adapter->link_speed = 0;
 478         adapter->link_duplex = 0;
 479         netif_carrier_off(netdev);
 480         netif_stop_queue(netdev);
 481
 482         e1000_reset(adapter);
 483         e1000_clean_all_tx_rings(adapter);
 484         e1000_clean_all_rx_rings(adapter);
 485
 486         /* Power down the PHY so no link is implied when interface is down *
 487          * The PHY cannot be powered down if any of the following is TRUE *
 488          * (a) WoL is enabled
 489          * (b) AMT is active
 490          * (c) SoL/IDER session is active */
 491         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
 492            adapter->hw.media_type == e1000_media_type_copper &&
 493            !(E1000_READ_REG(&adapter->hw, MANC) & E1000_MANC_SMBUS_EN) &&
 494            !mng_mode_enabled &&
 495            !e1000_check_phy_reset_block(&adapter->hw)) {
 496                 uint16_t mii_reg;
 497                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
 498                 mii_reg |= MII_CR_POWER_DOWN;
 499                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
 500                 mdelay(1);
 501         }
 502 }
 503
 504 void
 505 e1000_reset(struct e1000_adapter *adapter)
 506 {
 507         uint32_t pba, manc;
 508         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
 509
 510         /* Repartition Pba for greater than 9k mtu
 511          * To take effect CTRL.RST is required.
 512          */
 513
 514         switch (adapter->hw.mac_type) {
 515         case e1000_82547:
 516         case e1000_82547_rev_2:
 517                 pba = E1000_PBA_30K;
 518                 break;
 519         case e1000_82571:
 520         case e1000_82572:
 521                 pba = E1000_PBA_38K;
 522                 break;
 523         case e1000_82573:
 524                 pba = E1000_PBA_12K;
 525                 break;
 526         default:
 527                 pba = E1000_PBA_48K;
 528                 break;
 529         }
 530
 531         if((adapter->hw.mac_type != e1000_82573) &&
 532            (adapter->netdev->mtu > E1000_RXBUFFER_8192))
 533                 pba -= 8; /* allocate more FIFO for Tx */
 534
 535
 536         if(adapter->hw.mac_type == e1000_82547) {
 537                 adapter->tx_fifo_head = 0;
 538                 adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
 539                 adapter->tx_fifo_size =
 540                         (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
 541                 atomic_set(&adapter->tx_fifo_stall, 0);
 542         }
 543
 544         E1000_WRITE_REG(&adapter->hw, PBA, pba);
 545
 546         /* flow control settings */
 547         /* Set the FC high water mark to 90% of the FIFO size.
 548          * Required to clear last 3 LSB */
 549         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
 550
 551         adapter->hw.fc_high_water = fc_high_water_mark;
 552         adapter->hw.fc_low_water = fc_high_water_mark - 8;
 553         adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
 554         adapter->hw.fc_send_xon = 1;
 555         adapter->hw.fc = adapter->hw.original_fc;
 556
 557         /* Allow time for pending master requests to run */
 558         e1000_reset_hw(&adapter->hw);
 559         if(adapter->hw.mac_type >= e1000_82544)
 560                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
 561         if(e1000_init_hw(&adapter->hw))
 562                 DPRINTK(PROBE, ERR, "Hardware Error\n");
 563         e1000_update_mng_vlan(adapter);
 564         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 565         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
 566
 567         e1000_reset_adaptive(&adapter->hw);
 568         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
 569         if (adapter->en_mng_pt) {
 570                 manc = E1000_READ_REG(&adapter->hw, MANC);
 571                 manc |= (E1000_MANC_ARP_EN | E1000_MANC_EN_MNG2HOST);
 572                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
 573         }
 574 }
 575
 576 /**
 577  * e1000_probe - Device Initialization Routine
 578  * @pdev: PCI device information struct
 579  * @ent: entry in e1000_pci_tbl
 580  *
 581  * Returns 0 on success, negative on failure
 582  *
 583  * e1000_probe initializes an adapter identified by a pci_dev structure.
 584  * The OS initialization, configuring of the adapter private structure,
 585  * and a hardware reset occur.
 586  **/
 587
 588 static int __devinit
 589 e1000_probe(struct pci_dev *pdev,
 590             const struct pci_device_id *ent)
 591 {
 592         struct net_device *netdev;
 593         struct e1000_adapter *adapter;
 594         unsigned long mmio_start, mmio_len;
 595
 596         static int cards_found = 0;
 597         int i, err, pci_using_dac;
 598         uint16_t eeprom_data;
 599         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
 600         if((err = pci_enable_device(pdev)))
 601                 return err;
 602
 603         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK))) {
 604                 pci_using_dac = 1;
 605         } else {
 606                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK))) {
 607                         E1000_ERR("No usable DMA configuration, aborting\n");
 608                         return err;
 609                 }
 610                 pci_using_dac = 0;
 611         }
 612
 613         if((err = pci_request_regions(pdev, e1000_driver_name)))
 614                 return err;
 615
 616         pci_set_master(pdev);
 617
 618         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
 619         if(!netdev) {
 620                 err = -ENOMEM;
 621                 goto err_alloc_etherdev;
 622         }
 623
 624         SET_MODULE_OWNER(netdev);
 625         SET_NETDEV_DEV(netdev, &pdev->dev);
 626
 627         pci_set_drvdata(pdev, netdev);
 628         adapter = netdev_priv(netdev);
 629         adapter->netdev = netdev;
 630         adapter->pdev = pdev;
 631         adapter->hw.back = adapter;
 632         adapter->msg_enable = (1 << debug) - 1;
 633
 634         mmio_start = pci_resource_start(pdev, BAR_0);
 635         mmio_len = pci_resource_len(pdev, BAR_0);
 636
 637         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
 638         if(!adapter->hw.hw_addr) {
 639                 err = -EIO;
 640                 goto err_ioremap;
 641         }
 642
 643         for(i = BAR_1; i <= BAR_5; i++) {
 644                 if(pci_resource_len(pdev, i) == 0)
 645                         continue;
 646                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
 647                         adapter->hw.io_base = pci_resource_start(pdev, i);
 648                         break;
 649                 }
 650         }
 651
 652         netdev->open = &e1000_open;
 653         netdev->stop = &e1000_close;
 654         netdev->hard_start_xmit = &e1000_xmit_frame;
 655         netdev->get_stats = &e1000_get_stats;
 656         netdev->set_multicast_list = &e1000_set_multi;
 657         netdev->set_mac_address = &e1000_set_mac;
 658         netdev->change_mtu = &e1000_change_mtu;
 659         netdev->do_ioctl = &e1000_ioctl;
 660         e1000_set_ethtool_ops(netdev);
 661         netdev->tx_timeout = &e1000_tx_timeout;
 662         netdev->watchdog_timeo = 5 * HZ;
 663 #ifdef CONFIG_E1000_NAPI
 664         netdev->poll = &e1000_clean;
 665         netdev->weight = 64;
 666 #endif
 667         netdev->vlan_rx_register = e1000_vlan_rx_register;
 668         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
 669         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
 670 #ifdef CONFIG_NET_POLL_CONTROLLER
 671         netdev->poll_controller = e1000_netpoll;
 672 #endif
 673         strcpy(netdev->name, pci_name(pdev));
 674
 675         netdev->mem_start = mmio_start;
 676         netdev->mem_end = mmio_start + mmio_len;
 677         netdev->base_addr = adapter->hw.io_base;
 678
 679         adapter->bd_number = cards_found;
 680
 681         /* setup the private structure */
 682
 683         if((err = e1000_sw_init(adapter)))
 684                 goto err_sw_init;
 685
 686         if((err = e1000_check_phy_reset_block(&adapter->hw)))
 687                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
 688
 689         if(adapter->hw.mac_type >= e1000_82543) {
 690                 netdev->features = NETIF_F_SG |
 691                                    NETIF_F_HW_CSUM |
 692                                    NETIF_F_HW_VLAN_TX |
 693                                    NETIF_F_HW_VLAN_RX |
 694                                    NETIF_F_HW_VLAN_FILTER;
 695         }
 696
 697 #ifdef NETIF_F_TSO
 698         if((adapter->hw.mac_type >= e1000_82544) &&
 699            (adapter->hw.mac_type != e1000_82547))
 700                 netdev->features |= NETIF_F_TSO;
 701
 702 #ifdef NETIF_F_TSO_IPV6
 703         if(adapter->hw.mac_type > e1000_82547_rev_2)
 704                 netdev->features |= NETIF_F_TSO_IPV6;
 705 #endif
 706 #endif
 707         if(pci_using_dac)
 708                 netdev->features |= NETIF_F_HIGHDMA;
 709
 710         /* hard_start_xmit is safe against parallel locking */
 711         netdev->features |= NETIF_F_LLTX;
 712
 713         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
 714
 715         /* before reading the EEPROM, reset the controller to
 716          * put the device in a known good starting state */
 717
 718         e1000_reset_hw(&adapter->hw);
 719
 720         /* make sure the EEPROM is good */
 721
 722         if(e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
 723                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
 724                 err = -EIO;
 725                 goto err_eeprom;
 726         }
 727
 728         /* copy the MAC address out of the EEPROM */
 729
 730         if(e1000_read_mac_addr(&adapter->hw))
 731                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
 732         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
 733         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
 734
 735         if(!is_valid_ether_addr(netdev->perm_addr)) {
 736                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
 737                 err = -EIO;
 738                 goto err_eeprom;
 739         }
 740
 741         e1000_read_part_num(&adapter->hw, &(adapter->part_num));
 742
 743         e1000_get_bus_info(&adapter->hw);
 744
 745         init_timer(&adapter->tx_fifo_stall_timer);
 746         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
 747         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
 748
 749         init_timer(&adapter->watchdog_timer);
 750         adapter->watchdog_timer.function = &e1000_watchdog;
 751         adapter->watchdog_timer.data = (unsigned long) adapter;
 752
 753         INIT_WORK(&adapter->watchdog_task,
 754                 (void (*)(void *))e1000_watchdog_task, adapter);
 755
 756         init_timer(&adapter->phy_info_timer);
 757         adapter->phy_info_timer.function = &e1000_update_phy_info;
 758         adapter->phy_info_timer.data = (unsigned long) adapter;
 759
 760         INIT_WORK(&adapter->tx_timeout_task,
 761                 (void (*)(void *))e1000_tx_timeout_task, netdev);
 762
 763         /* we're going to reset, so assume we have no link for now */
 764
 765         netif_carrier_off(netdev);
 766         netif_stop_queue(netdev);
 767
 768         e1000_check_options(adapter);
 769
 770         /* Initial Wake on LAN setting
 771          * If APM wake is enabled in the EEPROM,
 772          * enable the ACPI Magic Packet filter
 773          */
 774
 775         switch(adapter->hw.mac_type) {
 776         case e1000_82542_rev2_0:
 777         case e1000_82542_rev2_1:
 778         case e1000_82543:
 779                 break;
 780         case e1000_82544:
 781                 e1000_read_eeprom(&adapter->hw,
 782                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
 783                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
 784                 break;
 785         case e1000_82546:
 786         case e1000_82546_rev_3:
 787         case e1000_82571:
 788                 if((E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1)
 789                    && (adapter->hw.media_type == e1000_media_type_copper)) {
 790                         e1000_read_eeprom(&adapter->hw,
 791                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
 792                         break;
 793                 }
 794                 /* Fall Through */
 795         default:
 796                 e1000_read_eeprom(&adapter->hw,
 797                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
 798                 break;
 799         }
 800         if(eeprom_data & eeprom_apme_mask)
 801                 adapter->wol |= E1000_WUFC_MAG;
 802
 803         /* reset the hardware with the new settings */
 804         e1000_reset(adapter);
 805
 806         /* If the controller is 82573 and f/w is AMT, do not set
 807          * DRV_LOAD until the interface is up.  For all other cases,
 808          * let the f/w know that the h/w is now under the control
 809          * of the driver. */
 810         if (adapter->hw.mac_type != e1000_82573 ||
 811             !e1000_check_mng_mode(&adapter->hw))
 812                 e1000_get_hw_control(adapter);
 813
 814         strcpy(netdev->name, "eth%d");
 815         if((err = register_netdev(netdev)))
 816                 goto err_register;
 817
 818         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
 819
 820         cards_found++;
 821         return 0;
 822
 823 err_register:
 824 err_sw_init:
 825 err_eeprom:
 826         iounmap(adapter->hw.hw_addr);
 827 err_ioremap:
 828         free_netdev(netdev);
 829 err_alloc_etherdev:
 830         pci_release_regions(pdev);
 831         return err;
 832 }
 833
 834 /**
 835  * e1000_remove - Device Removal Routine
 836  * @pdev: PCI device information struct
 837  *
 838  * e1000_remove is called by the PCI subsystem to alert the driver
 839  * that it should release a PCI device.  The could be caused by a
 840  * Hot-Plug event, or because the driver is going to be removed from
 841  * memory.
 842  **/
 843
 844 static void __devexit
 845 e1000_remove(struct pci_dev *pdev)
 846 {
 847         struct net_device *netdev = pci_get_drvdata(pdev);
 848         struct e1000_adapter *adapter = netdev_priv(netdev);
 849         uint32_t manc;
 850 #ifdef CONFIG_E1000_NAPI
 851         int i;
 852 #endif
 853
 854         flush_scheduled_work();
 855
 856         if(adapter->hw.mac_type >= e1000_82540 &&
 857            adapter->hw.media_type == e1000_media_type_copper) {
 858                 manc = E1000_READ_REG(&adapter->hw, MANC);
 859                 if(manc & E1000_MANC_SMBUS_EN) {
 860                         manc |= E1000_MANC_ARP_EN;
 861                         E1000_WRITE_REG(&adapter->hw, MANC, manc);
 862                 }
 863         }
 864
 865         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
 866          * would have already happened in close and is redundant. */
 867         e1000_release_hw_control(adapter);
 868
 869         unregister_netdev(netdev);
 870 #ifdef CONFIG_E1000_NAPI
 871         for (i = 0; i < adapter->num_rx_queues; i++)
 872                 __dev_put(&adapter->polling_netdev[i]);
 873 #endif
 874
 875         if(!e1000_check_phy_reset_block(&adapter->hw))
 876                 e1000_phy_hw_reset(&adapter->hw);
 877
 878         kfree(adapter->tx_ring);
 879         kfree(adapter->rx_ring);
 880 #ifdef CONFIG_E1000_NAPI
 881         kfree(adapter->polling_netdev);
 882 #endif
 883
 884         iounmap(adapter->hw.hw_addr);
 885         pci_release_regions(pdev);
 886
 887 #ifdef CONFIG_E1000_MQ
 888         free_percpu(adapter->cpu_netdev);
 889         free_percpu(adapter->cpu_tx_ring);
 890 #endif
 891         free_netdev(netdev);
 892
 893         pci_disable_device(pdev);
 894 }
 895
 896 /**
 897  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
 898  * @adapter: board private structure to initialize
 899  *
 900  * e1000_sw_init initializes the Adapter private data structure.
 901  * Fields are initialized based on PCI device information and
 902  * OS network device settings (MTU size).
 903  **/
 904
 905 static int __devinit
 906 e1000_sw_init(struct e1000_adapter *adapter)
 907 {
 908         struct e1000_hw *hw = &adapter->hw;
 909         struct net_device *netdev = adapter->netdev;
 910         struct pci_dev *pdev = adapter->pdev;
 911 #ifdef CONFIG_E1000_NAPI
 912         int i;
 913 #endif
 914
 915         /* PCI config space info */
 916
 917         hw->vendor_id = pdev->vendor;
 918         hw->device_id = pdev->device;
 919         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 920         hw->subsystem_id = pdev->subsystem_device;
 921
 922         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
 923
 924         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
 925
 926         adapter->rx_buffer_len = E1000_RXBUFFER_2048;
 927         adapter->rx_ps_bsize0 = E1000_RXBUFFER_256;
 928         hw->max_frame_size = netdev->mtu +
 929                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
 930         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
 931
 932         /* identify the MAC */
 933
 934         if(e1000_set_mac_type(hw)) {
 935                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
 936                 return -EIO;
 937         }
 938
 939         /* initialize eeprom parameters */
 940
 941         if(e1000_init_eeprom_params(hw)) {
 942                 E1000_ERR("EEPROM initialization failed\n");
 943                 return -EIO;
 944         }
 945
 946         switch(hw->mac_type) {
 947         default:
 948                 break;
 949         case e1000_82541:
 950         case e1000_82547:
 951         case e1000_82541_rev_2:
 952         case e1000_82547_rev_2:
 953                 hw->phy_init_script = 1;
 954                 break;
 955         }
 956
 957         e1000_set_media_type(hw);
 958
 959         hw->wait_autoneg_complete = FALSE;
 960         hw->tbi_compatibility_en = TRUE;
 961         hw->adaptive_ifs = TRUE;
 962
 963         /* Copper options */
 964
 965         if(hw->media_type == e1000_media_type_copper) {
 966                 hw->mdix = AUTO_ALL_MODES;
 967                 hw->disable_polarity_correction = FALSE;
 968                 hw->master_slave = E1000_MASTER_SLAVE;
 969         }
 970
 971 #ifdef CONFIG_E1000_MQ
 972         /* Number of supported queues */
 973         switch (hw->mac_type) {
 974         case e1000_82571:
 975         case e1000_82572:
 976                 /* These controllers support 2 tx queues, but with a single
 977                  * qdisc implementation, multiple tx queues aren't quite as
 978                  * interesting.  If we can find a logical way of mapping
 979                  * flows to a queue, then perhaps we can up the num_tx_queue
 980                  * count back to its default.  Until then, we run the risk of
 981                  * terrible performance due to SACK overload. */
 982                 adapter->num_tx_queues = 1;
 983                 adapter->num_rx_queues = 2;
 984                 break;
 985         default:
 986                 adapter->num_tx_queues = 1;
 987                 adapter->num_rx_queues = 1;
 988                 break;
 989         }
 990         adapter->num_rx_queues = min(adapter->num_rx_queues, num_online_cpus());
 991         adapter->num_tx_queues = min(adapter->num_tx_queues, num_online_cpus());
 992         DPRINTK(DRV, INFO, "Multiqueue Enabled: Rx Queue count = %u %s\n",
 993                 adapter->num_rx_queues,
 994                 ((adapter->num_rx_queues == 1)
 995                  ? ((num_online_cpus() > 1)
 996                         ? "(due to unsupported feature in current adapter)"
 997                         : "(due to unsupported system configuration)")
 998                  : ""));
 999         DPRINTK(DRV, INFO, "Multiqueue Enabled: Tx Queue count = %u\n",
1000                 adapter->num_tx_queues);
1001 #else
1002         adapter->num_tx_queues = 1;
1003         adapter->num_rx_queues = 1;
1004 #endif
1005
1006         if (e1000_alloc_queues(adapter)) {
1007                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1008                 return -ENOMEM;
1009         }
1010
1011 #ifdef CONFIG_E1000_NAPI
1012         for (i = 0; i < adapter->num_rx_queues; i++) {
1013                 adapter->polling_netdev[i].priv = adapter;
1014                 adapter->polling_netdev[i].poll = &e1000_clean;
1015                 adapter->polling_netdev[i].weight = 64;
1016                 dev_hold(&adapter->polling_netdev[i]);
1017                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1018         }
1019         spin_lock_init(&adapter->tx_queue_lock);
1020 #endif
1021
1022         atomic_set(&adapter->irq_sem, 1);
1023         spin_lock_init(&adapter->stats_lock);
1024
1025         return 0;
1026 }
1027
1028 /**
1029  * e1000_alloc_queues - Allocate memory for all rings
1030  * @adapter: board private structure to initialize
1031  *
1032  * We allocate one ring per queue at run-time since we don't know the
1033  * number of queues at compile-time.  The polling_netdev array is
1034  * intended for Multiqueue, but should work fine with a single queue.
1035  **/
1036
1037 static int __devinit
1038 e1000_alloc_queues(struct e1000_adapter *adapter)
1039 {
1040         int size;
1041
1042         size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
1043         adapter->tx_ring = kmalloc(size, GFP_KERNEL);
1044         if (!adapter->tx_ring)
1045                 return -ENOMEM;
1046         memset(adapter->tx_ring, 0, size);
1047
1048         size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
1049         adapter->rx_ring = kmalloc(size, GFP_KERNEL);
1050         if (!adapter->rx_ring) {
1051                 kfree(adapter->tx_ring);
1052                 return -ENOMEM;
1053         }
1054         memset(adapter->rx_ring, 0, size);
1055
1056 #ifdef CONFIG_E1000_NAPI
1057         size = sizeof(struct net_device) * adapter->num_rx_queues;
1058         adapter->polling_netdev = kmalloc(size, GFP_KERNEL);
1059         if (!adapter->polling_netdev) {
1060                 kfree(adapter->tx_ring);
1061                 kfree(adapter->rx_ring);
1062                 return -ENOMEM;
1063         }
1064         memset(adapter->polling_netdev, 0, size);
1065 #endif
1066
1067 #ifdef CONFIG_E1000_MQ
1068         adapter->rx_sched_call_data.func = e1000_rx_schedule;
1069         adapter->rx_sched_call_data.info = adapter->netdev;
1070
1071         adapter->cpu_netdev = alloc_percpu(struct net_device *);
1072         adapter->cpu_tx_ring = alloc_percpu(struct e1000_tx_ring *);
1073 #endif
1074
1075         return E1000_SUCCESS;
1076 }
1077
1078 #ifdef CONFIG_E1000_MQ
1079 static void __devinit
1080 e1000_setup_queue_mapping(struct e1000_adapter *adapter)
1081 {
1082         int i, cpu;
1083
1084         adapter->rx_sched_call_data.func = e1000_rx_schedule;
1085         adapter->rx_sched_call_data.info = adapter->netdev;
1086         cpus_clear(adapter->rx_sched_call_data.cpumask);
1087
1088         adapter->cpu_netdev = alloc_percpu(struct net_device *);
1089         adapter->cpu_tx_ring = alloc_percpu(struct e1000_tx_ring *);
1090
1091         lock_cpu_hotplug();
1092         i = 0;
1093         for_each_online_cpu(cpu) {
1094                 *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_tx_queues];
1095                 /* This is incomplete because we'd like to assign separate
1096                  * physical cpus to these netdev polling structures and
1097                  * avoid saturating a subset of cpus.
1098                  */
1099                 if (i < adapter->num_rx_queues) {
1100                         *per_cpu_ptr(adapter->cpu_netdev, cpu) = &adapter->polling_netdev[i];
1101                         adapter->rx_ring[i].cpu = cpu;
1102                         cpu_set(cpu, adapter->cpumask);
1103                 } else
1104                         *per_cpu_ptr(adapter->cpu_netdev, cpu) = NULL;
1105
1106                 i++;
1107         }
1108         unlock_cpu_hotplug();
1109 }
1110 #endif
1111
1112 /**
1113  * e1000_open - Called when a network interface is made active
1114  * @netdev: network interface device structure
1115  *
1116  * Returns 0 on success, negative value on failure
1117  *
1118  * The open entry point is called when a network interface is made
1119  * active by the system (IFF_UP).  At this point all resources needed
1120  * for transmit and receive operations are allocated, the interrupt
1121  * handler is registered with the OS, the watchdog timer is started,
1122  * and the stack is notified that the interface is ready.
1123  **/
1124
1125 static int
1126 e1000_open(struct net_device *netdev)
1127 {
1128         struct e1000_adapter *adapter = netdev_priv(netdev);
1129         int err;
1130
1131         /* allocate transmit descriptors */
1132
1133         if ((err = e1000_setup_all_tx_resources(adapter)))
1134                 goto err_setup_tx;
1135
1136         /* allocate receive descriptors */
1137
1138         if ((err = e1000_setup_all_rx_resources(adapter)))
1139                 goto err_setup_rx;
1140
1141         if((err = e1000_up(adapter)))
1142                 goto err_up;
1143         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1144         if((adapter->hw.mng_cookie.status &
1145                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1146                 e1000_update_mng_vlan(adapter);
1147         }
1148
1149         /* If AMT is enabled, let the firmware know that the network
1150          * interface is now open */
1151         if (adapter->hw.mac_type == e1000_82573 &&
1152             e1000_check_mng_mode(&adapter->hw))
1153                 e1000_get_hw_control(adapter);
1154
1155         return E1000_SUCCESS;
1156
1157 err_up:
1158         e1000_free_all_rx_resources(adapter);
1159 err_setup_rx:
1160         e1000_free_all_tx_resources(adapter);
1161 err_setup_tx:
1162         e1000_reset(adapter);
1163
1164         return err;
1165 }
1166
1167 /**
1168  * e1000_close - Disables a network interface
1169  * @netdev: network interface device structure
1170  *
1171  * Returns 0, this is not allowed to fail
1172  *
1173  * The close entry point is called when an interface is de-activated
1174  * by the OS.  The hardware is still under the drivers control, but
1175  * needs to be disabled.  A global MAC reset is issued to stop the
1176  * hardware, and all transmit and receive resources are freed.
1177  **/
1178
1179 static int
1180 e1000_close(struct net_device *netdev)
1181 {
1182         struct e1000_adapter *adapter = netdev_priv(netdev);
1183
1184         e1000_down(adapter);
1185
1186         e1000_free_all_tx_resources(adapter);
1187         e1000_free_all_rx_resources(adapter);
1188
1189         if((adapter->hw.mng_cookie.status &
1190                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1191                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1192         }
1193
1194         /* If AMT is enabled, let the firmware know that the network
1195          * interface is now closed */
1196         if (adapter->hw.mac_type == e1000_82573 &&
1197             e1000_check_mng_mode(&adapter->hw))
1198                 e1000_release_hw_control(adapter);
1199
1200         return 0;
1201 }
1202
1203 /**
1204  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1205  * @adapter: address of board private structure
1206  * @start: address of beginning of memory
1207  * @len: length of memory
1208  **/
1209 static inline boolean_t
1210 e1000_check_64k_bound(struct e1000_adapter *adapter,
1211                       void *start, unsigned long len)
1212 {
1213         unsigned long begin = (unsigned long) start;
1214         unsigned long end = begin + len;
1215
1216         /* First rev 82545 and 82546 need to not allow any memory
1217          * write location to cross 64k boundary due to errata 23 */
1218         if (adapter->hw.mac_type == e1000_82545 ||
1219             adapter->hw.mac_type == e1000_82546) {
1220                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1221         }
1222
1223         return TRUE;
1224 }
1225
1226 /**
1227  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1228  * @adapter: board private structure
1229  * @txdr:    tx descriptor ring (for a specific queue) to setup
1230  *
1231  * Return 0 on success, negative on failure
1232  **/
1233
1234 static int
1235 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1236                          struct e1000_tx_ring *txdr)
1237 {
1238         struct pci_dev *pdev = adapter->pdev;
1239         int size;
1240
1241         size = sizeof(struct e1000_buffer) * txdr->count;
1242
1243         txdr->buffer_info = vmalloc_node(size, pcibus_to_node(pdev->bus));
1244         if(!txdr->buffer_info) {
1245                 DPRINTK(PROBE, ERR,
1246                 "Unable to allocate memory for the transmit descriptor ring\n");
1247                 return -ENOMEM;
1248         }
1249         memset(txdr->buffer_info, 0, size);
1250
1251         /* round up to nearest 4K */
1252
1253         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1254         E1000_ROUNDUP(txdr->size, 4096);
1255
1256         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1257         if(!txdr->desc) {
1258 setup_tx_desc_die:
1259                 vfree(txdr->buffer_info);
1260                 DPRINTK(PROBE, ERR,
1261                 "Unable to allocate memory for the transmit descriptor ring\n");
1262                 return -ENOMEM;
1263         }
1264
1265         /* Fix for errata 23, can't cross 64kB boundary */
1266         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1267                 void *olddesc = txdr->desc;
1268                 dma_addr_t olddma = txdr->dma;
1269                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1270                                      "at %p\n", txdr->size, txdr->desc);
1271                 /* Try again, without freeing the previous */
1272                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1273                 if(!txdr->desc) {
1274                 /* Failed allocation, critical failure */
1275                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1276                         goto setup_tx_desc_die;
1277                 }
1278
1279                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1280                         /* give up */
1281                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1282                                             txdr->dma);
1283                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1284                         DPRINTK(PROBE, ERR,
1285                                 "Unable to allocate aligned memory "
1286                                 "for the transmit descriptor ring\n");
1287                         vfree(txdr->buffer_info);
1288                         return -ENOMEM;
1289                 } else {
1290                         /* Free old allocation, new allocation was successful */
1291                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1292                 }
1293         }
1294         memset(txdr->desc, 0, txdr->size);
1295
1296         txdr->next_to_use = 0;
1297         txdr->next_to_clean = 0;
1298         spin_lock_init(&txdr->tx_lock);
1299
1300         return 0;
1301 }
1302
1303 /**
1304  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1305  *                                (Descriptors) for all queues
1306  * @adapter: board private structure
1307  *
1308  * If this function returns with an error, then it's possible one or
1309  * more of the rings is populated (while the rest are not).  It is the
1310  * callers duty to clean those orphaned rings.
1311  *
1312  * Return 0 on success, negative on failure
1313  **/
1314
1315 int
1316 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1317 {
1318         int i, err = 0;
1319
1320         for (i = 0; i < adapter->num_tx_queues; i++) {
1321                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1322                 if (err) {
1323                         DPRINTK(PROBE, ERR,
1324                                 "Allocation for Tx Queue %u failed\n", i);
1325                         break;
1326                 }
1327         }
1328
1329         return err;
1330 }
1331
1332 /**
1333  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1334  * @adapter: board private structure
1335  *
1336  * Configure the Tx unit of the MAC after a reset.
1337  **/
1338
1339 static void
1340 e1000_configure_tx(struct e1000_adapter *adapter)
1341 {
1342         uint64_t tdba;
1343         struct e1000_hw *hw = &adapter->hw;
1344         uint32_t tdlen, tctl, tipg, tarc;
1345         uint32_t ipgr1, ipgr2;
1346
1347         /* Setup the HW Tx Head and Tail descriptor pointers */
1348
1349         switch (adapter->num_tx_queues) {
1350         case 2:
1351                 tdba = adapter->tx_ring[1].dma;
1352                 tdlen = adapter->tx_ring[1].count *
1353                         sizeof(struct e1000_tx_desc);
1354                 E1000_WRITE_REG(hw, TDBAL1, (tdba & 0x00000000ffffffffULL));
1355                 E1000_WRITE_REG(hw, TDBAH1, (tdba >> 32));
1356                 E1000_WRITE_REG(hw, TDLEN1, tdlen);
1357                 E1000_WRITE_REG(hw, TDH1, 0);
1358                 E1000_WRITE_REG(hw, TDT1, 0);
1359                 adapter->tx_ring[1].tdh = E1000_TDH1;
1360                 adapter->tx_ring[1].tdt = E1000_TDT1;
1361                 /* Fall Through */
1362         case 1:
1363         default:
1364                 tdba = adapter->tx_ring[0].dma;
1365                 tdlen = adapter->tx_ring[0].count *
1366                         sizeof(struct e1000_tx_desc);
1367                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1368                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1369                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1370                 E1000_WRITE_REG(hw, TDH, 0);
1371                 E1000_WRITE_REG(hw, TDT, 0);
1372                 adapter->tx_ring[0].tdh = E1000_TDH;
1373                 adapter->tx_ring[0].tdt = E1000_TDT;
1374                 break;
1375         }
1376
1377         /* Set the default values for the Tx Inter Packet Gap timer */
1378
1379         if (hw->media_type == e1000_media_type_fiber ||
1380             hw->media_type == e1000_media_type_internal_serdes)
1381                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1382         else
1383                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1384
1385         switch (hw->mac_type) {
1386         case e1000_82542_rev2_0:
1387         case e1000_82542_rev2_1:
1388                 tipg = DEFAULT_82542_TIPG_IPGT;
1389                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1390                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1391                 break;
1392         default:
1393                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1394                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1395                 break;
1396         }
1397         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1398         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1399         E1000_WRITE_REG(hw, TIPG, tipg);
1400
1401         /* Set the Tx Interrupt Delay register */
1402
1403         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1404         if (hw->mac_type >= e1000_82540)
1405                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1406
1407         /* Program the Transmit Control Register */
1408
1409         tctl = E1000_READ_REG(hw, TCTL);
1410
1411         tctl &= ~E1000_TCTL_CT;
1412         tctl |= E1000_TCTL_EN | E1000_TCTL_PSP | E1000_TCTL_RTLC |
1413                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1414
1415         E1000_WRITE_REG(hw, TCTL, tctl);
1416
1417         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1418                 tarc = E1000_READ_REG(hw, TARC0);
1419                 tarc |= ((1 << 25) | (1 << 21));
1420                 E1000_WRITE_REG(hw, TARC0, tarc);
1421                 tarc = E1000_READ_REG(hw, TARC1);
1422                 tarc |= (1 << 25);
1423                 if (tctl & E1000_TCTL_MULR)
1424                         tarc &= ~(1 << 28);
1425                 else
1426                         tarc |= (1 << 28);
1427                 E1000_WRITE_REG(hw, TARC1, tarc);
1428         }
1429
1430         e1000_config_collision_dist(hw);
1431
1432         /* Setup Transmit Descriptor Settings for eop descriptor */
1433         adapter->txd_cmd = E1000_TXD_CMD_IDE | E1000_TXD_CMD_EOP |
1434                 E1000_TXD_CMD_IFCS;
1435
1436         if (hw->mac_type < e1000_82543)
1437                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1438         else
1439                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1440
1441         /* Cache if we're 82544 running in PCI-X because we'll
1442          * need this to apply a workaround later in the send path. */
1443         if (hw->mac_type == e1000_82544 &&
1444             hw->bus_type == e1000_bus_type_pcix)
1445                 adapter->pcix_82544 = 1;
1446 }
1447
1448 /**
1449  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1450  * @adapter: board private structure
1451  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1452  *
1453  * Returns 0 on success, negative on failure
1454  **/
1455
1456 static int
1457 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1458                          struct e1000_rx_ring *rxdr)
1459 {
1460         struct pci_dev *pdev = adapter->pdev;
1461         int size, desc_len;
1462
1463         size = sizeof(struct e1000_buffer) * rxdr->count;
1464         rxdr->buffer_info = vmalloc_node(size, pcibus_to_node(pdev->bus));
1465         if (!rxdr->buffer_info) {
1466                 DPRINTK(PROBE, ERR,
1467                 "Unable to allocate memory for the receive descriptor ring\n");
1468                 return -ENOMEM;
1469         }
1470         memset(rxdr->buffer_info, 0, size);
1471
1472         size = sizeof(struct e1000_ps_page) * rxdr->count;
1473         rxdr->ps_page = kmalloc(size, GFP_KERNEL);
1474         if(!rxdr->ps_page) {
1475                 vfree(rxdr->buffer_info);
1476                 DPRINTK(PROBE, ERR,
1477                 "Unable to allocate memory for the receive descriptor ring\n");
1478                 return -ENOMEM;
1479         }
1480         memset(rxdr->ps_page, 0, size);
1481
1482         size = sizeof(struct e1000_ps_page_dma) * rxdr->count;
1483         rxdr->ps_page_dma = kmalloc(size, GFP_KERNEL);
1484         if(!rxdr->ps_page_dma) {
1485                 vfree(rxdr->buffer_info);
1486                 kfree(rxdr->ps_page);
1487                 DPRINTK(PROBE, ERR,
1488                 "Unable to allocate memory for the receive descriptor ring\n");
1489                 return -ENOMEM;
1490         }
1491         memset(rxdr->ps_page_dma, 0, size);
1492
1493         if(adapter->hw.mac_type <= e1000_82547_rev_2)
1494                 desc_len = sizeof(struct e1000_rx_desc);
1495         else
1496                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1497
1498         /* Round up to nearest 4K */
1499
1500         rxdr->size = rxdr->count * desc_len;
1501         E1000_ROUNDUP(rxdr->size, 4096);
1502
1503         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1504
1505         if (!rxdr->desc) {
1506                 DPRINTK(PROBE, ERR,
1507                 "Unable to allocate memory for the receive descriptor ring\n");
1508 setup_rx_desc_die:
1509                 vfree(rxdr->buffer_info);
1510                 kfree(rxdr->ps_page);
1511                 kfree(rxdr->ps_page_dma);
1512                 return -ENOMEM;
1513         }
1514
1515         /* Fix for errata 23, can't cross 64kB boundary */
1516         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1517                 void *olddesc = rxdr->desc;
1518                 dma_addr_t olddma = rxdr->dma;
1519                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1520                                      "at %p\n", rxdr->size, rxdr->desc);
1521                 /* Try again, without freeing the previous */
1522                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1523                 /* Failed allocation, critical failure */
1524                 if (!rxdr->desc) {
1525                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1526                         DPRINTK(PROBE, ERR,
1527                                 "Unable to allocate memory "
1528                                 "for the receive descriptor ring\n");
1529                         goto setup_rx_desc_die;
1530                 }
1531
1532                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1533                         /* give up */
1534                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1535                                             rxdr->dma);
1536                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1537                         DPRINTK(PROBE, ERR,
1538                                 "Unable to allocate aligned memory "
1539                                 "for the receive descriptor ring\n");
1540                         goto setup_rx_desc_die;
1541                 } else {
1542                         /* Free old allocation, new allocation was successful */
1543                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1544                 }
1545         }
1546         memset(rxdr->desc, 0, rxdr->size);
1547
1548         rxdr->next_to_clean = 0;
1549         rxdr->next_to_use = 0;
1550         rxdr->rx_skb_top = NULL;
1551         rxdr->rx_skb_prev = NULL;
1552
1553         return 0;
1554 }
1555
1556 /**
1557  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1558  *                                (Descriptors) for all queues
1559  * @adapter: board private structure
1560  *
1561  * If this function returns with an error, then it's possible one or
1562  * more of the rings is populated (while the rest are not).  It is the
1563  * callers duty to clean those orphaned rings.
1564  *
1565  * Return 0 on success, negative on failure
1566  **/
1567
1568 int
1569 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1570 {
1571         int i, err = 0;
1572
1573         for (i = 0; i < adapter->num_rx_queues; i++) {
1574                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1575                 if (err) {
1576                         DPRINTK(PROBE, ERR,
1577                                 "Allocation for Rx Queue %u failed\n", i);
1578                         break;
1579                 }
1580         }
1581
1582         return err;
1583 }
1584
1585 /**
1586  * e1000_setup_rctl - configure the receive control registers
1587  * @adapter: Board private structure
1588  **/
1589 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1590                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1591 static void
1592 e1000_setup_rctl(struct e1000_adapter *adapter)
1593 {
1594         uint32_t rctl, rfctl;
1595         uint32_t psrctl = 0;
1596 #ifdef CONFIG_E1000_PACKET_SPLIT
1597         uint32_t pages = 0;
1598 #endif
1599
1600         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1601
1602         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1603
1604         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1605                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1606                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1607
1608         if (adapter->hw.mac_type > e1000_82543)
1609                 rctl |= E1000_RCTL_SECRC;
1610
1611         if (adapter->hw.tbi_compatibility_on == 1)
1612                 rctl |= E1000_RCTL_SBP;
1613         else
1614                 rctl &= ~E1000_RCTL_SBP;
1615
1616         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1617                 rctl &= ~E1000_RCTL_LPE;
1618         else
1619                 rctl |= E1000_RCTL_LPE;
1620
1621         /* Setup buffer sizes */
1622         if(adapter->hw.mac_type >= e1000_82571) {
1623                 /* We can now specify buffers in 1K increments.
1624                  * BSIZE and BSEX are ignored in this case. */
1625                 rctl |= adapter->rx_buffer_len << 0x11;
1626         } else {
1627                 rctl &= ~E1000_RCTL_SZ_4096;
1628                 rctl |= E1000_RCTL_BSEX;
1629                 switch (adapter->rx_buffer_len) {
1630                 case E1000_RXBUFFER_2048:
1631                 default:
1632                         rctl |= E1000_RCTL_SZ_2048;
1633                         rctl &= ~E1000_RCTL_BSEX;
1634                         break;
1635                 case E1000_RXBUFFER_4096:
1636                         rctl |= E1000_RCTL_SZ_4096;
1637                         break;
1638                 case E1000_RXBUFFER_8192:
1639                         rctl |= E1000_RCTL_SZ_8192;
1640                         break;
1641                 case E1000_RXBUFFER_16384:
1642                         rctl |= E1000_RCTL_SZ_16384;
1643                         break;
1644                 }
1645         }
1646
1647 #ifdef CONFIG_E1000_PACKET_SPLIT
1648         /* 82571 and greater support packet-split where the protocol
1649          * header is placed in skb->data and the packet data is
1650          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1651          * In the case of a non-split, skb->data is linearly filled,
1652          * followed by the page buffers.  Therefore, skb->data is
1653          * sized to hold the largest protocol header.
1654          */
1655         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1656         if ((adapter->hw.mac_type > e1000_82547_rev_2) && (pages <= 3) &&
1657             PAGE_SIZE <= 16384)
1658                 adapter->rx_ps_pages = pages;
1659         else
1660                 adapter->rx_ps_pages = 0;
1661 #endif
1662         if (adapter->rx_ps_pages) {
1663                 /* Configure extra packet-split registers */
1664                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1665                 rfctl |= E1000_RFCTL_EXTEN;
1666                 /* disable IPv6 packet split support */
1667                 rfctl |= E1000_RFCTL_IPV6_DIS;
1668                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1669
1670                 rctl |= E1000_RCTL_DTYP_PS | E1000_RCTL_SECRC;
1671
1672                 psrctl |= adapter->rx_ps_bsize0 >>
1673                         E1000_PSRCTL_BSIZE0_SHIFT;
1674
1675                 switch (adapter->rx_ps_pages) {
1676                 case 3:
1677                         psrctl |= PAGE_SIZE <<
1678                                 E1000_PSRCTL_BSIZE3_SHIFT;
1679                 case 2:
1680                         psrctl |= PAGE_SIZE <<
1681                                 E1000_PSRCTL_BSIZE2_SHIFT;
1682                 case 1:
1683                         psrctl |= PAGE_SIZE >>
1684                                 E1000_PSRCTL_BSIZE1_SHIFT;
1685                         break;
1686                 }
1687
1688                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1689         }
1690
1691         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1692 }
1693
1694 /**
1695  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1696  * @adapter: board private structure
1697  *
1698  * Configure the Rx unit of the MAC after a reset.
1699  **/
1700
1701 static void
1702 e1000_configure_rx(struct e1000_adapter *adapter)
1703 {
1704         uint64_t rdba;
1705         struct e1000_hw *hw = &adapter->hw;
1706         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
1707 #ifdef CONFIG_E1000_MQ
1708         uint32_t reta, mrqc;
1709         int i;
1710 #endif
1711
1712         if (adapter->rx_ps_pages) {
1713                 rdlen = adapter->rx_ring[0].count *
1714                         sizeof(union e1000_rx_desc_packet_split);
1715                 adapter->clean_rx = e1000_clean_rx_irq_ps;
1716                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
1717         } else {
1718                 rdlen = adapter->rx_ring[0].count *
1719                         sizeof(struct e1000_rx_desc);
1720                 adapter->clean_rx = e1000_clean_rx_irq;
1721                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1722         }
1723
1724         /* disable receives while setting up the descriptors */
1725         rctl = E1000_READ_REG(hw, RCTL);
1726         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
1727
1728         /* set the Receive Delay Timer Register */
1729         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
1730
1731         if (hw->mac_type >= e1000_82540) {
1732                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
1733                 if(adapter->itr > 1)
1734                         E1000_WRITE_REG(hw, ITR,
1735                                 1000000000 / (adapter->itr * 256));
1736         }
1737
1738         if (hw->mac_type >= e1000_82571) {
1739                 /* Reset delay timers after every interrupt */
1740                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
1741                 ctrl_ext |= E1000_CTRL_EXT_CANC;
1742                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
1743                 E1000_WRITE_FLUSH(hw);
1744         }
1745
1746         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1747          * the Base and Length of the Rx Descriptor Ring */
1748         switch (adapter->num_rx_queues) {
1749 #ifdef CONFIG_E1000_MQ
1750         case 2:
1751                 rdba = adapter->rx_ring[1].dma;
1752                 E1000_WRITE_REG(hw, RDBAL1, (rdba & 0x00000000ffffffffULL));
1753                 E1000_WRITE_REG(hw, RDBAH1, (rdba >> 32));
1754                 E1000_WRITE_REG(hw, RDLEN1, rdlen);
1755                 E1000_WRITE_REG(hw, RDH1, 0);
1756                 E1000_WRITE_REG(hw, RDT1, 0);
1757                 adapter->rx_ring[1].rdh = E1000_RDH1;
1758                 adapter->rx_ring[1].rdt = E1000_RDT1;
1759                 /* Fall Through */
1760 #endif
1761         case 1:
1762         default:
1763                 rdba = adapter->rx_ring[0].dma;
1764                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
1765                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
1766                 E1000_WRITE_REG(hw, RDLEN, rdlen);
1767                 E1000_WRITE_REG(hw, RDH, 0);
1768                 E1000_WRITE_REG(hw, RDT, 0);
1769                 adapter->rx_ring[0].rdh = E1000_RDH;
1770                 adapter->rx_ring[0].rdt = E1000_RDT;
1771                 break;
1772         }
1773
1774 #ifdef CONFIG_E1000_MQ
1775         if (adapter->num_rx_queues > 1) {
1776                 uint32_t random[10];
1777
1778                 get_random_bytes(&random[0], 40);
1779
1780                 if (hw->mac_type <= e1000_82572) {
1781                         E1000_WRITE_REG(hw, RSSIR, 0);
1782                         E1000_WRITE_REG(hw, RSSIM, 0);
1783                 }
1784
1785                 switch (adapter->num_rx_queues) {
1786                 case 2:
1787                 default:
1788                         reta = 0x00800080;
1789                         mrqc = E1000_MRQC_ENABLE_RSS_2Q;
1790                         break;
1791                 }
1792
1793                 /* Fill out redirection table */
1794                 for (i = 0; i < 32; i++)
1795                         E1000_WRITE_REG_ARRAY(hw, RETA, i, reta);
1796                 /* Fill out hash function seeds */
1797                 for (i = 0; i < 10; i++)
1798                         E1000_WRITE_REG_ARRAY(hw, RSSRK, i, random[i]);
1799
1800                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1801                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1802                 E1000_WRITE_REG(hw, MRQC, mrqc);
1803         }
1804
1805         /* Multiqueue and packet checksumming are mutually exclusive. */
1806         if (hw->mac_type >= e1000_82571) {
1807                 rxcsum = E1000_READ_REG(hw, RXCSUM);
1808                 rxcsum |= E1000_RXCSUM_PCSD;
1809                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
1810         }
1811
1812 #else
1813
1814         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1815         if (hw->mac_type >= e1000_82543) {
1816                 rxcsum = E1000_READ_REG(hw, RXCSUM);
1817                 if(adapter->rx_csum == TRUE) {
1818                         rxcsum |= E1000_RXCSUM_TUOFL;
1819
1820                         /* Enable 82571 IPv4 payload checksum for UDP fragments
1821                          * Must be used in conjunction with packet-split. */
1822                         if ((hw->mac_type >= e1000_82571) &&
1823                            (adapter->rx_ps_pages)) {
1824                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1825                         }
1826                 } else {
1827                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1828                         /* don't need to clear IPPCSE as it defaults to 0 */
1829                 }
1830                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
1831         }
1832 #endif /* CONFIG_E1000_MQ */
1833
1834         if (hw->mac_type == e1000_82573)
1835                 E1000_WRITE_REG(hw, ERT, 0x0100);
1836
1837         /* Enable Receives */
1838         E1000_WRITE_REG(hw, RCTL, rctl);
1839 }
1840
1841 /**
1842  * e1000_free_tx_resources - Free Tx Resources per Queue
1843  * @adapter: board private structure
1844  * @tx_ring: Tx descriptor ring for a specific queue
1845  *
1846  * Free all transmit software resources
1847  **/
1848
1849 static void
1850 e1000_free_tx_resources(struct e1000_adapter *adapter,
1851                         struct e1000_tx_ring *tx_ring)
1852 {
1853         struct pci_dev *pdev = adapter->pdev;
1854
1855         e1000_clean_tx_ring(adapter, tx_ring);
1856
1857         vfree(tx_ring->buffer_info);
1858         tx_ring->buffer_info = NULL;
1859
1860         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1861
1862         tx_ring->desc = NULL;
1863 }
1864
1865 /**
1866  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1867  * @adapter: board private structure
1868  *
1869  * Free all transmit software resources
1870  **/
1871
1872 void
1873 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1874 {
1875         int i;
1876
1877         for (i = 0; i < adapter->num_tx_queues; i++)
1878                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1879 }
1880
1881 static inline void
1882 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1883                         struct e1000_buffer *buffer_info)
1884 {
1885         if(buffer_info->dma) {
1886                 pci_unmap_page(adapter->pdev,
1887                                 buffer_info->dma,
1888                                 buffer_info->length,
1889                                 PCI_DMA_TODEVICE);
1890                 buffer_info->dma = 0;
1891         }
1892         if(buffer_info->skb) {
1893                 dev_kfree_skb_any(buffer_info->skb);
1894                 buffer_info->skb = NULL;
1895         }
1896 }
1897
1898 /**
1899  * e1000_clean_tx_ring - Free Tx Buffers
1900  * @adapter: board private structure
1901  * @tx_ring: ring to be cleaned
1902  **/
1903
1904 static void
1905 e1000_clean_tx_ring(struct e1000_adapter *adapter,
1906                     struct e1000_tx_ring *tx_ring)
1907 {
1908         struct e1000_buffer *buffer_info;
1909         unsigned long size;
1910         unsigned int i;
1911
1912         /* Free all the Tx ring sk_buffs */
1913
1914         for(i = 0; i < tx_ring->count; i++) {
1915                 buffer_info = &tx_ring->buffer_info[i];
1916                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1917         }
1918
1919         size = sizeof(struct e1000_buffer) * tx_ring->count;
1920         memset(tx_ring->buffer_info, 0, size);
1921
1922         /* Zero out the descriptor ring */
1923
1924         memset(tx_ring->desc, 0, tx_ring->size);
1925
1926         tx_ring->next_to_use = 0;
1927         tx_ring->next_to_clean = 0;
1928         tx_ring->last_tx_tso = 0;
1929
1930         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
1931         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
1932 }
1933
1934 /**
1935  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1936  * @adapter: board private structure
1937  **/
1938
1939 static void
1940 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1941 {
1942         int i;
1943
1944         for (i = 0; i < adapter->num_tx_queues; i++)
1945                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1946 }
1947
1948 /**
1949  * e1000_free_rx_resources - Free Rx Resources
1950  * @adapter: board private structure
1951  * @rx_ring: ring to clean the resources from
1952  *
1953  * Free all receive software resources
1954  **/
1955
1956 static void
1957 e1000_free_rx_resources(struct e1000_adapter *adapter,
1958                         struct e1000_rx_ring *rx_ring)
1959 {
1960         struct pci_dev *pdev = adapter->pdev;
1961
1962         e1000_clean_rx_ring(adapter, rx_ring);
1963
1964         vfree(rx_ring->buffer_info);
1965         rx_ring->buffer_info = NULL;
1966         kfree(rx_ring->ps_page);
1967         rx_ring->ps_page = NULL;
1968         kfree(rx_ring->ps_page_dma);
1969         rx_ring->ps_page_dma = NULL;
1970
1971         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1972
1973         rx_ring->desc = NULL;
1974 }
1975
1976 /**
1977  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1978  * @adapter: board private structure
1979  *
1980  * Free all receive software resources
1981  **/
1982
1983 void
1984 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1985 {
1986         int i;
1987
1988         for (i = 0; i < adapter->num_rx_queues; i++)
1989                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
1990 }
1991
1992 /**
1993  * e1000_clean_rx_ring - Free Rx Buffers per Queue
1994  * @adapter: board private structure
1995  * @rx_ring: ring to free buffers from
1996  **/
1997
1998 static void
1999 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2000                     struct e1000_rx_ring *rx_ring)
2001 {
2002         struct e1000_buffer *buffer_info;
2003         struct e1000_ps_page *ps_page;
2004         struct e1000_ps_page_dma *ps_page_dma;
2005         struct pci_dev *pdev = adapter->pdev;
2006         unsigned long size;
2007         unsigned int i, j;
2008
2009         /* Free all the Rx ring sk_buffs */
2010
2011         for(i = 0; i < rx_ring->count; i++) {
2012                 buffer_info = &rx_ring->buffer_info[i];
2013                 if(buffer_info->skb) {
2014                         ps_page = &rx_ring->ps_page[i];
2015                         ps_page_dma = &rx_ring->ps_page_dma[i];
2016                         pci_unmap_single(pdev,
2017                                          buffer_info->dma,
2018                                          buffer_info->length,
2019                                          PCI_DMA_FROMDEVICE);
2020
2021                         dev_kfree_skb(buffer_info->skb);
2022                         buffer_info->skb = NULL;
2023                 }
2024                 ps_page = &rx_ring->ps_page[i];
2025                 ps_page_dma = &rx_ring->ps_page_dma[i];
2026                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2027                         if (!ps_page->ps_page[j]) break;
2028                         pci_unmap_page(pdev,
2029                                        ps_page_dma->ps_page_dma[j],
2030                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2031                         ps_page_dma->ps_page_dma[j] = 0;
2032                         put_page(ps_page->ps_page[j]);
2033                         ps_page->ps_page[j] = NULL;
2034                 }
2035         }
2036
2037         /* there also may be some cached data in our adapter */
2038         if (rx_ring->rx_skb_top) {
2039                 dev_kfree_skb(rx_ring->rx_skb_top);
2040
2041                 /* rx_skb_prev will be wiped out by rx_skb_top */
2042                 rx_ring->rx_skb_top = NULL;
2043                 rx_ring->rx_skb_prev = NULL;
2044         }
2045
2046
2047         size = sizeof(struct e1000_buffer) * rx_ring->count;
2048         memset(rx_ring->buffer_info, 0, size);
2049         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2050         memset(rx_ring->ps_page, 0, size);
2051         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2052         memset(rx_ring->ps_page_dma, 0, size);
2053
2054         /* Zero out the descriptor ring */
2055
2056         memset(rx_ring->desc, 0, rx_ring->size);
2057
2058         rx_ring->next_to_clean = 0;
2059         rx_ring->next_to_use = 0;
2060
2061         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2062         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2063 }
2064
2065 /**
2066  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2067  * @adapter: board private structure
2068  **/
2069
2070 static void
2071 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2072 {
2073         int i;
2074
2075         for (i = 0; i < adapter->num_rx_queues; i++)
2076                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2077 }
2078
2079 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2080  * and memory write and invalidate disabled for certain operations
2081  */
2082 static void
2083 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2084 {
2085         struct net_device *netdev = adapter->netdev;
2086         uint32_t rctl;
2087
2088         e1000_pci_clear_mwi(&adapter->hw);
2089
2090         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2091         rctl |= E1000_RCTL_RST;
2092         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2093         E1000_WRITE_FLUSH(&adapter->hw);
2094         mdelay(5);
2095
2096         if(netif_running(netdev))
2097                 e1000_clean_all_rx_rings(adapter);
2098 }
2099
2100 static void
2101 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2102 {
2103         struct net_device *netdev = adapter->netdev;
2104         uint32_t rctl;
2105
2106         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2107         rctl &= ~E1000_RCTL_RST;
2108         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2109         E1000_WRITE_FLUSH(&adapter->hw);
2110         mdelay(5);
2111
2112         if(adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2113                 e1000_pci_set_mwi(&adapter->hw);
2114
2115         if(netif_running(netdev)) {
2116                 e1000_configure_rx(adapter);
2117                 e1000_alloc_rx_buffers(adapter, &adapter->rx_ring[0]);
2118         }
2119 }
2120
2121 /**
2122  * e1000_set_mac - Change the Ethernet Address of the NIC
2123  * @netdev: network interface device structure
2124  * @p: pointer to an address structure
2125  *
2126  * Returns 0 on success, negative on failure
2127  **/
2128
2129 static int
2130 e1000_set_mac(struct net_device *netdev, void *p)
2131 {
2132         struct e1000_adapter *adapter = netdev_priv(netdev);
2133         struct sockaddr *addr = p;
2134
2135         if(!is_valid_ether_addr(addr->sa_data))
2136                 return -EADDRNOTAVAIL;
2137
2138         /* 82542 2.0 needs to be in reset to write receive address registers */
2139
2140         if(adapter->hw.mac_type == e1000_82542_rev2_0)
2141                 e1000_enter_82542_rst(adapter);
2142
2143         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2144         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2145
2146         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2147
2148         /* With 82571 controllers, LAA may be overwritten (with the default)
2149          * due to controller reset from the other port. */
2150         if (adapter->hw.mac_type == e1000_82571) {
2151                 /* activate the work around */
2152                 adapter->hw.laa_is_present = 1;
2153
2154                 /* Hold a copy of the LAA in RAR[14] This is done so that
2155                  * between the time RAR[0] gets clobbered  and the time it
2156                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2157                  * of the RARs and no incoming packets directed to this port
2158                  * are dropped. Eventaully the LAA will be in RAR[0] and
2159                  * RAR[14] */
2160                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2161                                         E1000_RAR_ENTRIES - 1);
2162         }
2163
2164         if(adapter->hw.mac_type == e1000_82542_rev2_0)
2165                 e1000_leave_82542_rst(adapter);
2166
2167         return 0;
2168 }
2169
2170 /**
2171  * e1000_set_multi - Multicast and Promiscuous mode set
2172  * @netdev: network interface device structure
2173  *
2174  * The set_multi entry point is called whenever the multicast address
2175  * list or the network interface flags are updated.  This routine is
2176  * responsible for configuring the hardware for proper multicast,
2177  * promiscuous mode, and all-multi behavior.
2178  **/
2179
2180 static void
2181 e1000_set_multi(struct net_device *netdev)
2182 {
2183         struct e1000_adapter *adapter = netdev_priv(netdev);
2184         struct e1000_hw *hw = &adapter->hw;
2185         struct dev_mc_list *mc_ptr;
2186         uint32_t rctl;
2187         uint32_t hash_value;
2188         int i, rar_entries = E1000_RAR_ENTRIES;
2189
2190         /* reserve RAR[14] for LAA over-write work-around */
2191         if (adapter->hw.mac_type == e1000_82571)
2192                 rar_entries--;
2193
2194         /* Check for Promiscuous and All Multicast modes */
2195
2196         rctl = E1000_READ_REG(hw, RCTL);
2197
2198         if(netdev->flags & IFF_PROMISC) {
2199                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2200         } else if(netdev->flags & IFF_ALLMULTI) {
2201                 rctl |= E1000_RCTL_MPE;
2202                 rctl &= ~E1000_RCTL_UPE;
2203         } else {
2204                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2205         }
2206
2207         E1000_WRITE_REG(hw, RCTL, rctl);
2208
2209         /* 82542 2.0 needs to be in reset to write receive address registers */
2210
2211         if(hw->mac_type == e1000_82542_rev2_0)
2212                 e1000_enter_82542_rst(adapter);
2213
2214         /* load the first 14 multicast address into the exact filters 1-14
2215          * RAR 0 is used for the station MAC adddress
2216          * if there are not 14 addresses, go ahead and clear the filters
2217          * -- with 82571 controllers only 0-13 entries are filled here
2218          */
2219         mc_ptr = netdev->mc_list;
2220
2221         for(i = 1; i < rar_entries; i++) {
2222                 if (mc_ptr) {
2223                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2224                         mc_ptr = mc_ptr->next;
2225                 } else {
2226                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2227                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2228                 }
2229         }
2230
2231         /* clear the old settings from the multicast hash table */
2232
2233         for(i = 0; i < E1000_NUM_MTA_REGISTERS; i++)
2234                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2235
2236         /* load any remaining addresses into the hash table */
2237
2238         for(; mc_ptr; mc_ptr = mc_ptr->next) {
2239                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2240                 e1000_mta_set(hw, hash_value);
2241         }
2242
2243         if(hw->mac_type == e1000_82542_rev2_0)
2244                 e1000_leave_82542_rst(adapter);
2245 }
2246
2247 /* Need to wait a few seconds after link up to get diagnostic information from
2248  * the phy */
2249
2250 static void
2251 e1000_update_phy_info(unsigned long data)
2252 {
2253         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2254         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2255 }
2256
2257 /**
2258  * e1000_82547_tx_fifo_stall - Timer Call-back
2259  * @data: pointer to adapter cast into an unsigned long
2260  **/
2261
2262 static void
2263 e1000_82547_tx_fifo_stall(unsigned long data)
2264 {
2265         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2266         struct net_device *netdev = adapter->netdev;
2267         uint32_t tctl;
2268
2269         if(atomic_read(&adapter->tx_fifo_stall)) {
2270                 if((E1000_READ_REG(&adapter->hw, TDT) ==
2271                     E1000_READ_REG(&adapter->hw, TDH)) &&
2272                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2273                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2274                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2275                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2276                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2277                         E1000_WRITE_REG(&adapter->hw, TCTL,
2278                                         tctl & ~E1000_TCTL_EN);
2279                         E1000_WRITE_REG(&adapter->hw, TDFT,
2280                                         adapter->tx_head_addr);
2281                         E1000_WRITE_REG(&adapter->hw, TDFH,
2282                                         adapter->tx_head_addr);
2283                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2284                                         adapter->tx_head_addr);
2285                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2286                                         adapter->tx_head_addr);
2287                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2288                         E1000_WRITE_FLUSH(&adapter->hw);
2289
2290                         adapter->tx_fifo_head = 0;
2291                         atomic_set(&adapter->tx_fifo_stall, 0);
2292                         netif_wake_queue(netdev);
2293                 } else {
2294                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2295                 }
2296         }
2297 }
2298
2299 /**
2300  * e1000_watchdog - Timer Call-back
2301  * @data: pointer to adapter cast into an unsigned long
2302  **/
2303 static void
2304 e1000_watchdog(unsigned long data)
2305 {
2306         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2307
2308         /* Do the rest outside of interrupt context */
2309         schedule_work(&adapter->watchdog_task);
2310 }
2311
2312 static void
2313 e1000_watchdog_task(struct e1000_adapter *adapter)
2314 {
2315         struct net_device *netdev = adapter->netdev;
2316         struct e1000_tx_ring *txdr = adapter->tx_ring;
2317         uint32_t link;
2318
2319         e1000_check_for_link(&adapter->hw);
2320         if (adapter->hw.mac_type == e1000_82573) {
2321                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2322                 if(adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2323                         e1000_update_mng_vlan(adapter);
2324         }
2325
2326         if((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2327            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2328                 link = !adapter->hw.serdes_link_down;
2329         else
2330                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2331
2332         if(link) {
2333                 if(!netif_carrier_ok(netdev)) {
2334                         e1000_get_speed_and_duplex(&adapter->hw,
2335                                                    &adapter->link_speed,
2336                                                    &adapter->link_duplex);
2337
2338                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s\n",
2339                                adapter->link_speed,
2340                                adapter->link_duplex == FULL_DUPLEX ?
2341                                "Full Duplex" : "Half Duplex");
2342
2343                         /* tweak tx_queue_len according to speed/duplex */
2344                         netdev->tx_queue_len = adapter->tx_queue_len;
2345                         adapter->tx_timeout_factor = 1;
2346                         if (adapter->link_duplex == HALF_DUPLEX) {
2347                                 switch (adapter->link_speed) {
2348                                 case SPEED_10:
2349                                         netdev->tx_queue_len = 10;
2350                                         adapter->tx_timeout_factor = 8;
2351                                         break;
2352                                 case SPEED_100:
2353                                         netdev->tx_queue_len = 100;
2354                                         break;
2355                                 }
2356                         }
2357
2358                         netif_carrier_on(netdev);
2359                         netif_wake_queue(netdev);
2360                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2361                         adapter->smartspeed = 0;
2362                 }
2363         } else {
2364                 if(netif_carrier_ok(netdev)) {
2365                         adapter->link_speed = 0;
2366                         adapter->link_duplex = 0;
2367                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2368                         netif_carrier_off(netdev);
2369                         netif_stop_queue(netdev);
2370                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2371                 }
2372
2373                 e1000_smartspeed(adapter);
2374         }
2375
2376         e1000_update_stats(adapter);
2377
2378         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2379         adapter->tpt_old = adapter->stats.tpt;
2380         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2381         adapter->colc_old = adapter->stats.colc;
2382
2383         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2384         adapter->gorcl_old = adapter->stats.gorcl;
2385         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2386         adapter->gotcl_old = adapter->stats.gotcl;
2387
2388         e1000_update_adaptive(&adapter->hw);
2389
2390 #ifdef CONFIG_E1000_MQ
2391         txdr = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
2392 #endif
2393         if (!netif_carrier_ok(netdev)) {
2394                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2395                         /* We've lost link, so the controller stops DMA,
2396                          * but we've got queued Tx work that's never going
2397                          * to get done, so reset controller to flush Tx.
2398                          * (Do the reset outside of interrupt context). */
2399                         schedule_work(&adapter->tx_timeout_task);
2400                 }
2401         }
2402
2403         /* Dynamic mode for Interrupt Throttle Rate (ITR) */
2404         if(adapter->hw.mac_type >= e1000_82540 && adapter->itr == 1) {
2405                 /* Symmetric Tx/Rx gets a reduced ITR=2000; Total
2406                  * asymmetrical Tx or Rx gets ITR=8000; everyone
2407                  * else is between 2000-8000. */
2408                 uint32_t goc = (adapter->gotcl + adapter->gorcl) / 10000;
2409                 uint32_t dif = (adapter->gotcl > adapter->gorcl ?
2410                         adapter->gotcl - adapter->gorcl :
2411                         adapter->gorcl - adapter->gotcl) / 10000;
2412                 uint32_t itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2413                 E1000_WRITE_REG(&adapter->hw, ITR, 1000000000 / (itr * 256));
2414         }
2415
2416         /* Cause software interrupt to ensure rx ring is cleaned */
2417         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2418
2419         /* Force detection of hung controller every watchdog period */
2420         adapter->detect_tx_hung = TRUE;
2421
2422         /* With 82571 controllers, LAA may be overwritten due to controller
2423          * reset from the other port. Set the appropriate LAA in RAR[0] */
2424         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2425                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2426
2427         /* Reset the timer */
2428         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
2429 }
2430
2431 #define E1000_TX_FLAGS_CSUM             0x00000001
2432 #define E1000_TX_FLAGS_VLAN             0x00000002
2433 #define E1000_TX_FLAGS_TSO              0x00000004
2434 #define E1000_TX_FLAGS_IPV4             0x00000008
2435 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2436 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2437
2438 static inline int
2439 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2440           struct sk_buff *skb)
2441 {
2442 #ifdef NETIF_F_TSO
2443         struct e1000_context_desc *context_desc;
2444         struct e1000_buffer *buffer_info;
2445         unsigned int i;
2446         uint32_t cmd_length = 0;
2447         uint16_t ipcse = 0, tucse, mss;
2448         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2449         int err;
2450
2451         if(skb_shinfo(skb)->tso_size) {
2452                 if (skb_header_cloned(skb)) {
2453                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2454                         if (err)
2455                                 return err;
2456                 }
2457
2458                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2459                 mss = skb_shinfo(skb)->tso_size;
2460                 if(skb->protocol == ntohs(ETH_P_IP)) {
2461                         skb->nh.iph->tot_len = 0;
2462                         skb->nh.iph->check = 0;
2463                         skb->h.th->check =
2464                                 ~csum_tcpudp_magic(skb->nh.iph->saddr,
2465                                                    skb->nh.iph->daddr,
2466                                                    0,
2467                                                    IPPROTO_TCP,
2468                                                    0);
2469                         cmd_length = E1000_TXD_CMD_IP;
2470                         ipcse = skb->h.raw - skb->data - 1;
2471 #ifdef NETIF_F_TSO_IPV6
2472                 } else if(skb->protocol == ntohs(ETH_P_IPV6)) {
2473                         skb->nh.ipv6h->payload_len = 0;
2474                         skb->h.th->check =
2475                                 ~csum_ipv6_magic(&skb->nh.ipv6h->saddr,
2476                                                  &skb->nh.ipv6h->daddr,
2477                                                  0,
2478                                                  IPPROTO_TCP,
2479                                                  0);
2480                         ipcse = 0;
2481 #endif
2482                 }
2483                 ipcss = skb->nh.raw - skb->data;
2484                 ipcso = (void *)&(skb->nh.iph->check) - (void *)skb->data;
2485                 tucss = skb->h.raw - skb->data;
2486                 tucso = (void *)&(skb->h.th->check) - (void *)skb->data;
2487                 tucse = 0;
2488
2489                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2490                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2491
2492                 i = tx_ring->next_to_use;
2493                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2494                 buffer_info = &tx_ring->buffer_info[i];
2495
2496                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2497                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2498                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2499                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2500                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2501                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2502                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2503                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2504                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2505
2506                 buffer_info->time_stamp = jiffies;
2507
2508                 if (++i == tx_ring->count) i = 0;
2509                 tx_ring->next_to_use = i;
2510
2511                 return 1;
2512         }
2513 #endif
2514
2515         return 0;
2516 }
2517
2518 static inline boolean_t
2519 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2520               struct sk_buff *skb)
2521 {
2522         struct e1000_context_desc *context_desc;
2523         struct e1000_buffer *buffer_info;
2524         unsigned int i;
2525         uint8_t css;
2526
2527         if(likely(skb->ip_summed == CHECKSUM_HW)) {
2528                 css = skb->h.raw - skb->data;
2529
2530                 i = tx_ring->next_to_use;
2531                 buffer_info = &tx_ring->buffer_info[i];
2532                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2533
2534                 context_desc->upper_setup.tcp_fields.tucss = css;
2535                 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum;
2536                 context_desc->upper_setup.tcp_fields.tucse = 0;
2537                 context_desc->tcp_seg_setup.data = 0;
2538                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2539
2540                 buffer_info->time_stamp = jiffies;
2541
2542                 if (unlikely(++i == tx_ring->count)) i = 0;
2543                 tx_ring->next_to_use = i;
2544
2545                 return TRUE;
2546         }
2547
2548         return FALSE;
2549 }
2550
2551 #define E1000_MAX_TXD_PWR       12
2552 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2553
2554 static inline int
2555 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2556              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2557              unsigned int nr_frags, unsigned int mss)
2558 {
2559         struct e1000_buffer *buffer_info;
2560         unsigned int len = skb->len;
2561         unsigned int offset = 0, size, count = 0, i;
2562         unsigned int f;
2563         len -= skb->data_len;
2564
2565         i = tx_ring->next_to_use;
2566
2567         while(len) {
2568                 buffer_info = &tx_ring->buffer_info[i];
2569                 size = min(len, max_per_txd);
2570 #ifdef NETIF_F_TSO
2571                 /* Workaround for Controller erratum --
2572                  * descriptor for non-tso packet in a linear SKB that follows a
2573                  * tso gets written back prematurely before the data is fully
2574                  * DMAd to the controller */
2575                 if (!skb->data_len && tx_ring->last_tx_tso &&
2576                                 !skb_shinfo(skb)->tso_size) {
2577                         tx_ring->last_tx_tso = 0;
2578                         size -= 4;
2579                 }
2580
2581                 /* Workaround for premature desc write-backs
2582                  * in TSO mode.  Append 4-byte sentinel desc */
2583                 if(unlikely(mss && !nr_frags && size == len && size > 8))
2584                         size -= 4;
2585 #endif
2586                 /* work-around for errata 10 and it applies
2587                  * to all controllers in PCI-X mode
2588                  * The fix is to make sure that the first descriptor of a
2589                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2590                  */
2591                 if(unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
2592                                 (size > 2015) && count == 0))
2593                         size = 2015;
2594
2595                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2596                  * terminating buffers within evenly-aligned dwords. */
2597                 if(unlikely(adapter->pcix_82544 &&
2598                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2599                    size > 4))
2600                         size -= 4;
2601
2602                 buffer_info->length = size;
2603                 buffer_info->dma =
2604                         pci_map_single(adapter->pdev,
2605                                 skb->data + offset,
2606                                 size,
2607                                 PCI_DMA_TODEVICE);
2608                 buffer_info->time_stamp = jiffies;
2609
2610                 len -= size;
2611                 offset += size;
2612                 count++;
2613                 if(unlikely(++i == tx_ring->count)) i = 0;
2614         }
2615
2616         for(f = 0; f < nr_frags; f++) {
2617                 struct skb_frag_struct *frag;
2618
2619                 frag = &skb_shinfo(skb)->frags[f];
2620                 len = frag->size;
2621                 offset = frag->page_offset;
2622
2623                 while(len) {
2624                         buffer_info = &tx_ring->buffer_info[i];
2625                         size = min(len, max_per_txd);
2626 #ifdef NETIF_F_TSO
2627                         /* Workaround for premature desc write-backs
2628                          * in TSO mode.  Append 4-byte sentinel desc */
2629                         if(unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2630                                 size -= 4;
2631 #endif
2632                         /* Workaround for potential 82544 hang in PCI-X.
2633                          * Avoid terminating buffers within evenly-aligned
2634                          * dwords. */
2635                         if(unlikely(adapter->pcix_82544 &&
2636                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
2637                            size > 4))
2638                                 size -= 4;
2639
2640                         buffer_info->length = size;
2641                         buffer_info->dma =
2642                                 pci_map_page(adapter->pdev,
2643                                         frag->page,
2644                                         offset,
2645                                         size,
2646                                         PCI_DMA_TODEVICE);
2647                         buffer_info->time_stamp = jiffies;
2648
2649                         len -= size;
2650                         offset += size;
2651                         count++;
2652                         if(unlikely(++i == tx_ring->count)) i = 0;
2653                 }
2654         }
2655
2656         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2657         tx_ring->buffer_info[i].skb = skb;
2658         tx_ring->buffer_info[first].next_to_watch = i;
2659
2660         return count;
2661 }
2662
2663 static inline void
2664 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2665                int tx_flags, int count)
2666 {
2667         struct e1000_tx_desc *tx_desc = NULL;
2668         struct e1000_buffer *buffer_info;
2669         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2670         unsigned int i;
2671
2672         if(likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2673                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2674                              E1000_TXD_CMD_TSE;
2675                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2676
2677                 if(likely(tx_flags & E1000_TX_FLAGS_IPV4))
2678                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2679         }
2680
2681         if(likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2682                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2683                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2684         }
2685
2686         if(unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2687                 txd_lower |= E1000_TXD_CMD_VLE;
2688                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2689         }
2690
2691         i = tx_ring->next_to_use;
2692
2693         while(count--) {
2694                 buffer_info = &tx_ring->buffer_info[i];
2695                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2696                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2697                 tx_desc->lower.data =
2698                         cpu_to_le32(txd_lower | buffer_info->length);
2699                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2700                 if(unlikely(++i == tx_ring->count)) i = 0;
2701         }
2702
2703         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2704
2705         /* Force memory writes to complete before letting h/w
2706          * know there are new descriptors to fetch.  (Only
2707          * applicable for weak-ordered memory model archs,
2708          * such as IA-64). */
2709         wmb();
2710
2711         tx_ring->next_to_use = i;
2712         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
2713 }
2714
2715 /**
2716  * 82547 workaround to avoid controller hang in half-duplex environment.
2717  * The workaround is to avoid queuing a large packet that would span
2718  * the internal Tx FIFO ring boundary by notifying the stack to resend
2719  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2720  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2721  * to the beginning of the Tx FIFO.
2722  **/
2723
2724 #define E1000_FIFO_HDR                  0x10
2725 #define E1000_82547_PAD_LEN             0x3E0
2726
2727 static inline int
2728 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
2729 {
2730         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2731         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
2732
2733         E1000_ROUNDUP(skb_fifo_len, E1000_FIFO_HDR);
2734
2735         if(adapter->link_duplex != HALF_DUPLEX)
2736                 goto no_fifo_stall_required;
2737
2738         if(atomic_read(&adapter->tx_fifo_stall))
2739                 return 1;
2740
2741         if(skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2742                 atomic_set(&adapter->tx_fifo_stall, 1);
2743                 return 1;
2744         }
2745
2746 no_fifo_stall_required:
2747         adapter->tx_fifo_head += skb_fifo_len;
2748         if(adapter->tx_fifo_head >= adapter->tx_fifo_size)
2749                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2750         return 0;
2751 }
2752
2753 #define MINIMUM_DHCP_PACKET_SIZE 282
2754 static inline int
2755 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
2756 {
2757         struct e1000_hw *hw =  &adapter->hw;
2758         uint16_t length, offset;
2759         if(vlan_tx_tag_present(skb)) {
2760                 if(!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
2761                         ( adapter->hw.mng_cookie.status &
2762                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
2763                         return 0;
2764         }
2765         if ((skb->len > MINIMUM_DHCP_PACKET_SIZE) && (!skb->protocol)) {
2766                 struct ethhdr *eth = (struct ethhdr *) skb->data;
2767                 if((htons(ETH_P_IP) == eth->h_proto)) {
2768                         const struct iphdr *ip =
2769                                 (struct iphdr *)((uint8_t *)skb->data+14);
2770                         if(IPPROTO_UDP == ip->protocol) {
2771                                 struct udphdr *udp =
2772                                         (struct udphdr *)((uint8_t *)ip +
2773                                                 (ip->ihl << 2));
2774                                 if(ntohs(udp->dest) == 67) {
2775                                         offset = (uint8_t *)udp + 8 - skb->data;
2776                                         length = skb->len - offset;
2777
2778                                         return e1000_mng_write_dhcp_info(hw,
2779                                                         (uint8_t *)udp + 8,
2780                                                         length);
2781                                 }
2782                         }
2783                 }
2784         }
2785         return 0;
2786 }
2787
2788 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2789 static int
2790 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2791 {
2792         struct e1000_adapter *adapter = netdev_priv(netdev);
2793         struct e1000_tx_ring *tx_ring;
2794         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2795         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2796         unsigned int tx_flags = 0;
2797         unsigned int len = skb->len;
2798         unsigned long flags;
2799         unsigned int nr_frags = 0;
2800         unsigned int mss = 0;
2801         int count = 0;
2802         int tso;
2803         unsigned int f;
2804         len -= skb->data_len;
2805
2806 #ifdef CONFIG_E1000_MQ
2807         tx_ring = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
2808 #else
2809         tx_ring = adapter->tx_ring;
2810 #endif
2811
2812         if (unlikely(skb->len <= 0)) {
2813                 dev_kfree_skb_any(skb);
2814                 return NETDEV_TX_OK;
2815         }
2816
2817 #ifdef NETIF_F_TSO
2818         mss = skb_shinfo(skb)->tso_size;
2819         /* The controller does a simple calculation to
2820          * make sure there is enough room in the FIFO before
2821          * initiating the DMA for each buffer.  The calc is:
2822          * 4 = ceil(buffer len/mss).  To make sure we don't
2823          * overrun the FIFO, adjust the max buffer len if mss
2824          * drops. */
2825         if(mss) {
2826                 uint8_t hdr_len;
2827                 max_per_txd = min(mss << 2, max_per_txd);
2828                 max_txd_pwr = fls(max_per_txd) - 1;
2829
2830         /* TSO Workaround for 82571/2 Controllers -- if skb->data
2831          * points to just header, pull a few bytes of payload from
2832          * frags into skb->data */
2833                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2834                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len)) &&
2835                         (adapter->hw.mac_type == e1000_82571 ||
2836                         adapter->hw.mac_type == e1000_82572)) {
2837                         unsigned int pull_size;
2838                         pull_size = min((unsigned int)4, skb->data_len);
2839                         if (!__pskb_pull_tail(skb, pull_size)) {
2840                                 printk(KERN_ERR "__pskb_pull_tail failed.\n");
2841                                 dev_kfree_skb_any(skb);
2842                                 return -EFAULT;
2843                         }
2844                         len = skb->len - skb->data_len;
2845                 }
2846         }
2847
2848         if((mss) || (skb->ip_summed == CHECKSUM_HW))
2849         /* reserve a descriptor for the offload context */
2850                 count++;
2851         count++;
2852 #else
2853         if(skb->ip_summed == CHECKSUM_HW)
2854                 count++;
2855 #endif
2856
2857 #ifdef NETIF_F_TSO
2858         /* Controller Erratum workaround */
2859         if (!skb->data_len && tx_ring->last_tx_tso &&
2860                 !skb_shinfo(skb)->tso_size)
2861                 count++;
2862 #endif
2863
2864         count += TXD_USE_COUNT(len, max_txd_pwr);
2865
2866         if(adapter->pcix_82544)
2867                 count++;
2868
2869         /* work-around for errata 10 and it applies to all controllers
2870          * in PCI-X mode, so add one more descriptor to the count
2871          */
2872         if(unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
2873                         (len > 2015)))
2874                 count++;
2875
2876         nr_frags = skb_shinfo(skb)->nr_frags;
2877         for(f = 0; f < nr_frags; f++)
2878                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
2879                                        max_txd_pwr);
2880         if(adapter->pcix_82544)
2881                 count += nr_frags;
2882
2883         if(adapter->hw.tx_pkt_filtering && (adapter->hw.mac_type == e1000_82573) )
2884                 e1000_transfer_dhcp_info(adapter, skb);
2885
2886         local_irq_save(flags);
2887         if (!spin_trylock(&tx_ring->tx_lock)) {
2888                 /* Collision - tell upper layer to requeue */
2889                 local_irq_restore(flags);
2890                 return NETDEV_TX_LOCKED;
2891         }
2892
2893         /* need: count + 2 desc gap to keep tail from touching
2894          * head, otherwise try next time */
2895         if (unlikely(E1000_DESC_UNUSED(tx_ring) < count + 2)) {
2896                 netif_stop_queue(netdev);
2897                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2898                 return NETDEV_TX_BUSY;
2899         }
2900
2901         if(unlikely(adapter->hw.mac_type == e1000_82547)) {
2902                 if(unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
2903                         netif_stop_queue(netdev);
2904                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies);
2905                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2906                         return NETDEV_TX_BUSY;
2907                 }
2908         }
2909
2910         if(unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
2911                 tx_flags |= E1000_TX_FLAGS_VLAN;
2912                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
2913         }
2914
2915         first = tx_ring->next_to_use;
2916
2917         tso = e1000_tso(adapter, tx_ring, skb);
2918         if (tso < 0) {
2919                 dev_kfree_skb_any(skb);
2920                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2921                 return NETDEV_TX_OK;
2922         }
2923
2924         if (likely(tso)) {
2925                 tx_ring->last_tx_tso = 1;
2926                 tx_flags |= E1000_TX_FLAGS_TSO;
2927         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
2928                 tx_flags |= E1000_TX_FLAGS_CSUM;
2929
2930         /* Old method was to assume IPv4 packet by default if TSO was enabled.
2931          * 82571 hardware supports TSO capabilities for IPv6 as well...
2932          * no longer assume, we must. */
2933         if (likely(skb->protocol == ntohs(ETH_P_IP)))
2934                 tx_flags |= E1000_TX_FLAGS_IPV4;
2935
2936         e1000_tx_queue(adapter, tx_ring, tx_flags,
2937                        e1000_tx_map(adapter, tx_ring, skb, first,
2938                                     max_per_txd, nr_frags, mss));
2939
2940         netdev->trans_start = jiffies;
2941
2942         /* Make sure there is space in the ring for the next send. */
2943         if (unlikely(E1000_DESC_UNUSED(tx_ring) < MAX_SKB_FRAGS + 2))
2944                 netif_stop_queue(netdev);
2945
2946         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2947         return NETDEV_TX_OK;
2948 }
2949
2950 /**
2951  * e1000_tx_timeout - Respond to a Tx Hang
2952  * @netdev: network interface device structure
2953  **/
2954
2955 static void
2956 e1000_tx_timeout(struct net_device *netdev)
2957 {
2958         struct e1000_adapter *adapter = netdev_priv(netdev);
2959
2960         /* Do the reset outside of interrupt context */
2961         schedule_work(&adapter->tx_timeout_task);
2962 }
2963
2964 static void
2965 e1000_tx_timeout_task(struct net_device *netdev)
2966 {
2967         struct e1000_adapter *adapter = netdev_priv(netdev);
2968
2969         adapter->tx_timeout_count++;
2970         e1000_down(adapter);
2971         e1000_up(adapter);
2972 }
2973
2974 /**
2975  * e1000_get_stats - Get System Network Statistics
2976  * @netdev: network interface device structure
2977  *
2978  * Returns the address of the device statistics structure.
2979  * The statistics are actually updated from the timer callback.
2980  **/
2981
2982 static struct net_device_stats *
2983 e1000_get_stats(struct net_device *netdev)
2984 {
2985         struct e1000_adapter *adapter = netdev_priv(netdev);
2986
2987         /* only return the current stats */
2988         return &adapter->net_stats;
2989 }
2990
2991 /**
2992  * e1000_change_mtu - Change the Maximum Transfer Unit
2993  * @netdev: network interface device structure
2994  * @new_mtu: new value for maximum frame size
2995  *
2996  * Returns 0 on success, negative on failure
2997  **/
2998
2999 static int
3000 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3001 {
3002         struct e1000_adapter *adapter = netdev_priv(netdev);
3003         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3004
3005         if((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3006                 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3007                         DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3008                 return -EINVAL;
3009         }
3010
3011         /* Adapter-specific max frame size limits. */
3012         switch (adapter->hw.mac_type) {
3013         case e1000_82542_rev2_0:
3014         case e1000_82542_rev2_1:
3015         case e1000_82573:
3016                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3017                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3018                         return -EINVAL;
3019                 }
3020                 break;
3021         case e1000_82571:
3022         case e1000_82572:
3023 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3024                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3025                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3026                         return -EINVAL;
3027                 }
3028                 break;
3029         default:
3030                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3031                 break;
3032         }
3033
3034         /* since the driver code now supports splitting a packet across
3035          * multiple descriptors, most of the fifo related limitations on
3036          * jumbo frame traffic have gone away.
3037          * simply use 2k descriptors for everything.
3038          *
3039          * NOTE: dev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3040          * means we reserve 2 more, this pushes us to allocate from the next
3041          * larger slab size
3042          * i.e. RXBUFFER_2048 --> size-4096 slab */
3043
3044         /* recent hardware supports 1KB granularity */
3045         if (adapter->hw.mac_type > e1000_82547_rev_2) {
3046                 adapter->rx_buffer_len =
3047                     ((max_frame < E1000_RXBUFFER_2048) ?
3048                         max_frame : E1000_RXBUFFER_2048);
3049                 E1000_ROUNDUP(adapter->rx_buffer_len, 1024);
3050         } else
3051                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3052
3053         netdev->mtu = new_mtu;
3054
3055         if(netif_running(netdev)) {
3056                 e1000_down(adapter);
3057                 e1000_up(adapter);
3058         }
3059
3060         adapter->hw.max_frame_size = max_frame;
3061
3062         return 0;
3063 }
3064
3065 /**
3066  * e1000_update_stats - Update the board statistics counters
3067  * @adapter: board private structure
3068  **/
3069
3070 void
3071 e1000_update_stats(struct e1000_adapter *adapter)
3072 {
3073         struct e1000_hw *hw = &adapter->hw;
3074         unsigned long flags;
3075         uint16_t phy_tmp;
3076
3077 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3078
3079         spin_lock_irqsave(&adapter->stats_lock, flags);
3080
3081         /* these counters are modified from e1000_adjust_tbi_stats,
3082          * called from the interrupt context, so they must only
3083          * be written while holding adapter->stats_lock
3084          */
3085
3086         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3087         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3088         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3089         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3090         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3091         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3092         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3093         adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3094         adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3095         adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3096         adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3097         adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3098         adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3099
3100         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3101         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3102         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3103         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3104         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3105         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3106         adapter->stats.dc += E1000_READ_REG(hw, DC);
3107         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3108         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3109         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3110         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3111         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3112         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3113         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3114         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3115         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3116         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3117         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3118         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3119         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3120         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3121         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3122         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3123         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3124         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3125         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3126         adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3127         adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3128         adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3129         adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3130         adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3131         adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3132         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3133         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3134
3135         /* used for adaptive IFS */
3136
3137         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3138         adapter->stats.tpt += hw->tx_packet_delta;
3139         hw->collision_delta = E1000_READ_REG(hw, COLC);
3140         adapter->stats.colc += hw->collision_delta;
3141
3142         if(hw->mac_type >= e1000_82543) {
3143                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3144                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3145                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3146                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3147                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3148                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3149         }
3150         if(hw->mac_type > e1000_82547_rev_2) {
3151                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3152                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3153                 adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3154                 adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3155                 adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3156                 adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3157                 adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3158                 adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3159                 adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3160         }
3161
3162         /* Fill out the OS statistics structure */
3163
3164         adapter->net_stats.rx_packets = adapter->stats.gprc;
3165         adapter->net_stats.tx_packets = adapter->stats.gptc;
3166         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3167         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3168         adapter->net_stats.multicast = adapter->stats.mprc;
3169         adapter->net_stats.collisions = adapter->stats.colc;
3170
3171         /* Rx Errors */
3172
3173         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3174                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3175                 adapter->stats.rlec + adapter->stats.cexterr;
3176         adapter->net_stats.rx_dropped = 0;
3177         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3178         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3179         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3180         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3181
3182         /* Tx Errors */
3183
3184         adapter->net_stats.tx_errors = adapter->stats.ecol +
3185                                        adapter->stats.latecol;
3186         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3187         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3188         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3189
3190         /* Tx Dropped needs to be maintained elsewhere */
3191
3192         /* Phy Stats */
3193
3194         if(hw->media_type == e1000_media_type_copper) {
3195                 if((adapter->link_speed == SPEED_1000) &&
3196                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3197                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3198                         adapter->phy_stats.idle_errors += phy_tmp;
3199                 }
3200
3201                 if((hw->mac_type <= e1000_82546) &&
3202                    (hw->phy_type == e1000_phy_m88) &&
3203                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3204                         adapter->phy_stats.receive_errors += phy_tmp;
3205         }
3206
3207         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3208 }
3209
3210 #ifdef CONFIG_E1000_MQ
3211 void
3212 e1000_rx_schedule(void *data)
3213 {
3214         struct net_device *poll_dev, *netdev = data;
3215         struct e1000_adapter *adapter = netdev->priv;
3216         int this_cpu = get_cpu();
3217
3218         poll_dev = *per_cpu_ptr(adapter->cpu_netdev, this_cpu);
3219         if (poll_dev == NULL) {
3220                 put_cpu();
3221                 return;
3222         }
3223
3224         if (likely(netif_rx_schedule_prep(poll_dev)))
3225                 __netif_rx_schedule(poll_dev);
3226         else
3227                 e1000_irq_enable(adapter);
3228
3229         put_cpu();
3230 }
3231 #endif
3232
3233 /**
3234  * e1000_intr - Interrupt Handler
3235  * @irq: interrupt number
3236  * @data: pointer to a network interface device structure
3237  * @pt_regs: CPU registers structure
3238  **/
3239
3240 static irqreturn_t
3241 e1000_intr(int irq, void *data, struct pt_regs *regs)
3242 {
3243         struct net_device *netdev = data;
3244         struct e1000_adapter *adapter = netdev_priv(netdev);
3245         struct e1000_hw *hw = &adapter->hw;
3246         uint32_t icr = E1000_READ_REG(hw, ICR);
3247 #if defined(CONFIG_E1000_NAPI) && defined(CONFIG_E1000_MQ) || !defined(CONFIG_E1000_NAPI)
3248         int i;
3249 #endif
3250
3251         if(unlikely(!icr))
3252                 return IRQ_NONE;  /* Not our interrupt */
3253
3254         if(unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3255                 hw->get_link_status = 1;
3256                 mod_timer(&adapter->watchdog_timer, jiffies);
3257         }
3258
3259 #ifdef CONFIG_E1000_NAPI
3260         atomic_inc(&adapter->irq_sem);
3261         E1000_WRITE_REG(hw, IMC, ~0);
3262         E1000_WRITE_FLUSH(hw);
3263 #ifdef CONFIG_E1000_MQ
3264         if (atomic_read(&adapter->rx_sched_call_data.count) == 0) {
3265                 /* We must setup the cpumask once count == 0 since
3266                  * each cpu bit is cleared when the work is done. */
3267                 adapter->rx_sched_call_data.cpumask = adapter->cpumask;
3268                 atomic_add(adapter->num_rx_queues - 1, &adapter->irq_sem);
3269                 atomic_set(&adapter->rx_sched_call_data.count,
3270                            adapter->num_rx_queues);
3271                 smp_call_async_mask(&adapter->rx_sched_call_data);
3272         } else {
3273                 printk("call_data.count == %u\n", atomic_read(&adapter->rx_sched_call_data.count));
3274         }
3275 #else /* if !CONFIG_E1000_MQ */
3276         if (likely(netif_rx_schedule_prep(&adapter->polling_netdev[0])))
3277                 __netif_rx_schedule(&adapter->polling_netdev[0]);
3278         else
3279                 e1000_irq_enable(adapter);
3280 #endif /* CONFIG_E1000_MQ */
3281
3282 #else /* if !CONFIG_E1000_NAPI */
3283         /* Writing IMC and IMS is needed for 82547.
3284            Due to Hub Link bus being occupied, an interrupt
3285            de-assertion message is not able to be sent.
3286            When an interrupt assertion message is generated later,
3287            two messages are re-ordered and sent out.
3288            That causes APIC to think 82547 is in de-assertion
3289            state, while 82547 is in assertion state, resulting
3290            in dead lock. Writing IMC forces 82547 into
3291            de-assertion state.
3292         */
3293         if(hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2){
3294                 atomic_inc(&adapter->irq_sem);
3295                 E1000_WRITE_REG(hw, IMC, ~0);
3296         }
3297
3298         for(i = 0; i < E1000_MAX_INTR; i++)
3299                 if(unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3300                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3301                         break;
3302
3303         if(hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3304                 e1000_irq_enable(adapter);
3305
3306 #endif /* CONFIG_E1000_NAPI */
3307
3308         return IRQ_HANDLED;
3309 }
3310
3311 #ifdef CONFIG_E1000_NAPI
3312 /**
3313  * e1000_clean - NAPI Rx polling callback
3314  * @adapter: board private structure
3315  **/
3316
3317 static int
3318 e1000_clean(struct net_device *poll_dev, int *budget)
3319 {
3320         struct e1000_adapter *adapter;
3321         int work_to_do = min(*budget, poll_dev->quota);
3322         int tx_cleaned, i = 0, work_done = 0;
3323
3324         /* Must NOT use netdev_priv macro here. */
3325         adapter = poll_dev->priv;
3326
3327         /* Keep link state information with original netdev */
3328         if (!netif_carrier_ok(adapter->netdev))
3329                 goto quit_polling;
3330
3331         while (poll_dev != &adapter->polling_netdev[i]) {
3332                 i++;
3333                 if (unlikely(i == adapter->num_rx_queues))
3334                         BUG();
3335         }
3336
3337         tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[i]);
3338         adapter->clean_rx(adapter, &adapter->rx_ring[i],
3339                           &work_done, work_to_do);
3340
3341         *budget -= work_done;
3342         poll_dev->quota -= work_done;
3343
3344         /* If no Tx and not enough Rx work done, exit the polling mode */
3345         if((!tx_cleaned && (work_done == 0)) ||
3346            !netif_running(adapter->netdev)) {
3347 quit_polling:
3348                 netif_rx_complete(poll_dev);
3349                 e1000_irq_enable(adapter);
3350                 return 0;
3351         }
3352
3353         return 1;
3354 }
3355
3356 #endif
3357 /**
3358  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3359  * @adapter: board private structure
3360  **/
3361
3362 static boolean_t
3363 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3364                    struct e1000_tx_ring *tx_ring)
3365 {
3366         struct net_device *netdev = adapter->netdev;
3367         struct e1000_tx_desc *tx_desc, *eop_desc;
3368         struct e1000_buffer *buffer_info;
3369         unsigned int i, eop;
3370         boolean_t cleaned = FALSE;
3371
3372         i = tx_ring->next_to_clean;
3373         eop = tx_ring->buffer_info[i].next_to_watch;
3374         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3375
3376         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3377                 for(cleaned = FALSE; !cleaned; ) {
3378                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3379                         buffer_info = &tx_ring->buffer_info[i];
3380                         cleaned = (i == eop);
3381
3382                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3383
3384                         tx_desc->buffer_addr = 0;
3385                         tx_desc->lower.data = 0;
3386                         tx_desc->upper.data = 0;
3387
3388                         if(unlikely(++i == tx_ring->count)) i = 0;
3389                 }
3390
3391 #ifdef CONFIG_E1000_MQ
3392                 tx_ring->tx_stats.packets++;
3393 #endif
3394
3395                 eop = tx_ring->buffer_info[i].next_to_watch;
3396                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3397         }
3398
3399         tx_ring->next_to_clean = i;
3400
3401         spin_lock(&tx_ring->tx_lock);
3402
3403         if(unlikely(cleaned && netif_queue_stopped(netdev) &&
3404                     netif_carrier_ok(netdev)))
3405                 netif_wake_queue(netdev);
3406
3407         spin_unlock(&tx_ring->tx_lock);
3408
3409         if (adapter->detect_tx_hung) {
3410                 /* Detect a transmit hang in hardware, this serializes the
3411                  * check with the clearing of time_stamp and movement of i */
3412                 adapter->detect_tx_hung = FALSE;
3413                 if (tx_ring->buffer_info[eop].dma &&
3414                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3415                                adapter->tx_timeout_factor * HZ)
3416                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
3417                          E1000_STATUS_TXOFF)) {
3418
3419                         /* detected Tx unit hang */
3420                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3421                                         "  Tx Queue             <%lu>\n"
3422                                         "  TDH                  <%x>\n"
3423                                         "  TDT                  <%x>\n"
3424                                         "  next_to_use          <%x>\n"
3425                                         "  next_to_clean        <%x>\n"
3426                                         "buffer_info[next_to_clean]\n"
3427                                         "  time_stamp           <%lx>\n"
3428                                         "  next_to_watch        <%x>\n"
3429                                         "  jiffies              <%lx>\n"
3430                                         "  next_to_watch.status <%x>\n",
3431                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3432                                         sizeof(struct e1000_tx_ring)),
3433                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
3434                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
3435                                 tx_ring->next_to_use,
3436                                 tx_ring->next_to_clean,
3437                                 tx_ring->buffer_info[eop].time_stamp,
3438                                 eop,
3439                                 jiffies,
3440                                 eop_desc->upper.fields.status);
3441                         netif_stop_queue(netdev);
3442                 }
3443         }
3444         return cleaned;
3445 }
3446
3447 /**
3448  * e1000_rx_checksum - Receive Checksum Offload for 82543
3449  * @adapter:     board private structure
3450  * @status_err:  receive descriptor status and error fields
3451  * @csum:        receive descriptor csum field
3452  * @sk_buff:     socket buffer with received data
3453  **/
3454
3455 static inline void
3456 e1000_rx_checksum(struct e1000_adapter *adapter,
3457                   uint32_t status_err, uint32_t csum,
3458                   struct sk_buff *skb)
3459 {
3460         uint16_t status = (uint16_t)status_err;
3461         uint8_t errors = (uint8_t)(status_err >> 24);
3462         skb->ip_summed = CHECKSUM_NONE;
3463
3464         /* 82543 or newer only */
3465         if(unlikely(adapter->hw.mac_type < e1000_82543)) return;
3466         /* Ignore Checksum bit is set */
3467         if(unlikely(status & E1000_RXD_STAT_IXSM)) return;
3468         /* TCP/UDP checksum error bit is set */
3469         if(unlikely(errors & E1000_RXD_ERR_TCPE)) {
3470                 /* let the stack verify checksum errors */
3471                 adapter->hw_csum_err++;
3472                 return;
3473         }
3474         /* TCP/UDP Checksum has not been calculated */
3475         if(adapter->hw.mac_type <= e1000_82547_rev_2) {
3476                 if(!(status & E1000_RXD_STAT_TCPCS))
3477                         return;
3478         } else {
3479                 if(!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
3480                         return;
3481         }
3482         /* It must be a TCP or UDP packet with a valid checksum */
3483         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3484                 /* TCP checksum is good */
3485                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3486         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
3487                 /* IP fragment with UDP payload */
3488                 /* Hardware complements the payload checksum, so we undo it
3489                  * and then put the value in host order for further stack use.
3490                  */
3491                 csum = ntohl(csum ^ 0xFFFF);
3492                 skb->csum = csum;
3493                 skb->ip_summed = CHECKSUM_HW;
3494         }
3495         adapter->hw_csum_good++;
3496 }
3497
3498 /**
3499  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3500  * @adapter: board private structure
3501  **/
3502
3503 static boolean_t
3504 #ifdef CONFIG_E1000_NAPI
3505 e1000_clean_rx_irq(struct e1000_adapter *adapter,
3506                    struct e1000_rx_ring *rx_ring,
3507                    int *work_done, int work_to_do)
3508 #else
3509 e1000_clean_rx_irq(struct e1000_adapter *adapter,
3510                    struct e1000_rx_ring *rx_ring)
3511 #endif
3512 {
3513         struct net_device *netdev = adapter->netdev;
3514         struct pci_dev *pdev = adapter->pdev;
3515         struct e1000_rx_desc *rx_desc;
3516         struct e1000_buffer *buffer_info;
3517         struct sk_buff *skb;
3518         unsigned long flags;
3519         uint32_t length;
3520         uint8_t last_byte;
3521         unsigned int i;
3522         boolean_t cleaned = FALSE;
3523
3524         i = rx_ring->next_to_clean;
3525         rx_desc = E1000_RX_DESC(*rx_ring, i);
3526
3527         while(rx_desc->status & E1000_RXD_STAT_DD) {
3528                 buffer_info = &rx_ring->buffer_info[i];
3529 #ifdef CONFIG_E1000_NAPI
3530                 if(*work_done >= work_to_do)
3531                         break;
3532                 (*work_done)++;
3533 #endif
3534                 cleaned = TRUE;
3535
3536                 pci_unmap_single(pdev,
3537                                  buffer_info->dma,
3538                                  buffer_info->length,
3539                                  PCI_DMA_FROMDEVICE);
3540
3541                 skb = buffer_info->skb;
3542                 length = le16_to_cpu(rx_desc->length);
3543
3544                 if(unlikely(!(rx_desc->status & E1000_RXD_STAT_EOP))) {
3545                         /* All receives must fit into a single buffer */
3546                         E1000_DBG("%s: Receive packet consumed multiple"
3547                                   " buffers\n", netdev->name);
3548                         dev_kfree_skb_irq(skb);
3549                         goto next_desc;
3550                 }
3551
3552                 if(unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3553                         last_byte = *(skb->data + length - 1);
3554                         if(TBI_ACCEPT(&adapter->hw, rx_desc->status,
3555                                       rx_desc->errors, length, last_byte)) {
3556                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3557                                 e1000_tbi_adjust_stats(&adapter->hw,
3558                                                        &adapter->stats,
3559                                                        length, skb->data);
3560                                 spin_unlock_irqrestore(&adapter->stats_lock,
3561                                                        flags);
3562                                 length--;
3563                         } else {
3564                                 dev_kfree_skb_irq(skb);
3565                                 goto next_desc;
3566                         }
3567                 }
3568
3569                 /* Good Receive */
3570                 skb_put(skb, length - ETHERNET_FCS_SIZE);
3571
3572                 /* Receive Checksum Offload */
3573                 e1000_rx_checksum(adapter,
3574                                   (uint32_t)(rx_desc->status) |
3575                                   ((uint32_t)(rx_desc->errors) << 24),
3576                                   rx_desc->csum, skb);
3577                 skb->protocol = eth_type_trans(skb, netdev);
3578 #ifdef CONFIG_E1000_NAPI
3579                 if(unlikely(adapter->vlgrp &&
3580                             (rx_desc->status & E1000_RXD_STAT_VP))) {
3581                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3582                                                  le16_to_cpu(rx_desc->special) &
3583                                                  E1000_RXD_SPC_VLAN_MASK);
3584                 } else {
3585                         netif_receive_skb(skb);
3586                 }
3587 #else /* CONFIG_E1000_NAPI */
3588                 if(unlikely(adapter->vlgrp &&
3589                             (rx_desc->status & E1000_RXD_STAT_VP))) {
3590                         vlan_hwaccel_rx(skb, adapter->vlgrp,
3591                                         le16_to_cpu(rx_desc->special) &
3592                                         E1000_RXD_SPC_VLAN_MASK);
3593                 } else {
3594                         netif_rx(skb);
3595                 }
3596 #endif /* CONFIG_E1000_NAPI */
3597                 netdev->last_rx = jiffies;
3598 #ifdef CONFIG_E1000_MQ
3599                 rx_ring->rx_stats.packets++;
3600                 rx_ring->rx_stats.bytes += length;
3601 #endif
3602
3603 next_desc:
3604                 rx_desc->status = 0;
3605                 buffer_info->skb = NULL;
3606                 if(unlikely(++i == rx_ring->count)) i = 0;
3607
3608                 rx_desc = E1000_RX_DESC(*rx_ring, i);
3609         }
3610         rx_ring->next_to_clean = i;
3611         adapter->alloc_rx_buf(adapter, rx_ring);
3612
3613         return cleaned;
3614 }
3615
3616 /**
3617  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
3618  * @adapter: board private structure
3619  **/
3620
3621 static boolean_t
3622 #ifdef CONFIG_E1000_NAPI
3623 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
3624                       struct e1000_rx_ring *rx_ring,
3625                       int *work_done, int work_to_do)
3626 #else
3627 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
3628                       struct e1000_rx_ring *rx_ring)
3629 #endif
3630 {
3631         union e1000_rx_desc_packet_split *rx_desc;
3632         struct net_device *netdev = adapter->netdev;
3633         struct pci_dev *pdev = adapter->pdev;
3634         struct e1000_buffer *buffer_info;
3635         struct e1000_ps_page *ps_page;
3636         struct e1000_ps_page_dma *ps_page_dma;
3637         struct sk_buff *skb;
3638         unsigned int i, j;
3639         uint32_t length, staterr;
3640         boolean_t cleaned = FALSE;
3641
3642         i = rx_ring->next_to_clean;
3643         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
3644         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
3645
3646         while(staterr & E1000_RXD_STAT_DD) {
3647                 buffer_info = &rx_ring->buffer_info[i];
3648                 ps_page = &rx_ring->ps_page[i];
3649                 ps_page_dma = &rx_ring->ps_page_dma[i];
3650 #ifdef CONFIG_E1000_NAPI
3651                 if(unlikely(*work_done >= work_to_do))
3652                         break;
3653                 (*work_done)++;
3654 #endif
3655                 cleaned = TRUE;
3656                 pci_unmap_single(pdev, buffer_info->dma,
3657                                  buffer_info->length,
3658                                  PCI_DMA_FROMDEVICE);
3659
3660                 skb = buffer_info->skb;
3661
3662                 if(unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
3663                         E1000_DBG("%s: Packet Split buffers didn't pick up"
3664                                   " the full packet\n", netdev->name);
3665                         dev_kfree_skb_irq(skb);
3666                         goto next_desc;
3667                 }
3668
3669                 if(unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
3670                         dev_kfree_skb_irq(skb);
3671                         goto next_desc;
3672                 }
3673
3674                 length = le16_to_cpu(rx_desc->wb.middle.length0);
3675
3676                 if(unlikely(!length)) {
3677                         E1000_DBG("%s: Last part of the packet spanning"
3678                                   " multiple descriptors\n", netdev->name);
3679                         dev_kfree_skb_irq(skb);
3680                         goto next_desc;
3681                 }
3682
3683                 /* Good Receive */
3684                 skb_put(skb, length);
3685
3686                 for(j = 0; j < adapter->rx_ps_pages; j++) {
3687                         if(!(length = le16_to_cpu(rx_desc->wb.upper.length[j])))
3688                                 break;
3689
3690                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
3691                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
3692                         ps_page_dma->ps_page_dma[j] = 0;
3693                         skb_shinfo(skb)->frags[j].page =
3694                                 ps_page->ps_page[j];
3695                         ps_page->ps_page[j] = NULL;
3696                         skb_shinfo(skb)->frags[j].page_offset = 0;
3697                         skb_shinfo(skb)->frags[j].size = length;
3698                         skb_shinfo(skb)->nr_frags++;
3699                         skb->len += length;
3700                         skb->data_len += length;
3701                 }
3702
3703                 e1000_rx_checksum(adapter, staterr,
3704                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
3705                 skb->protocol = eth_type_trans(skb, netdev);
3706
3707                 if(likely(rx_desc->wb.upper.header_status &
3708                           E1000_RXDPS_HDRSTAT_HDRSP)) {
3709                         adapter->rx_hdr_split++;
3710 #ifdef HAVE_RX_ZERO_COPY
3711                         skb_shinfo(skb)->zero_copy = TRUE;
3712 #endif
3713                 }
3714 #ifdef CONFIG_E1000_NAPI
3715                 if(unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
3716                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3717                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
3718                                 E1000_RXD_SPC_VLAN_MASK);
3719                 } else {
3720                         netif_receive_skb(skb);
3721                 }
3722 #else /* CONFIG_E1000_NAPI */
3723                 if(unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
3724                         vlan_hwaccel_rx(skb, adapter->vlgrp,
3725                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
3726                                 E1000_RXD_SPC_VLAN_MASK);
3727                 } else {
3728                         netif_rx(skb);
3729                 }
3730 #endif /* CONFIG_E1000_NAPI */
3731                 netdev->last_rx = jiffies;
3732 #ifdef CONFIG_E1000_MQ
3733                 rx_ring->rx_stats.packets++;
3734                 rx_ring->rx_stats.bytes += length;
3735 #endif
3736
3737 next_desc:
3738                 rx_desc->wb.middle.status_error &= ~0xFF;
3739                 buffer_info->skb = NULL;
3740                 if(unlikely(++i == rx_ring->count)) i = 0;
3741
3742                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
3743                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
3744         }
3745         rx_ring->next_to_clean = i;
3746         adapter->alloc_rx_buf(adapter, rx_ring);
3747
3748         return cleaned;
3749 }
3750
3751 /**
3752  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
3753  * @adapter: address of board private structure
3754  **/
3755
3756 static void
3757 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
3758                        struct e1000_rx_ring *rx_ring)
3759 {
3760         struct net_device *netdev = adapter->netdev;
3761         struct pci_dev *pdev = adapter->pdev;
3762         struct e1000_rx_desc *rx_desc;
3763         struct e1000_buffer *buffer_info;
3764         struct sk_buff *skb;
3765         unsigned int i;
3766         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
3767
3768         i = rx_ring->next_to_use;
3769         buffer_info = &rx_ring->buffer_info[i];
3770
3771         while(!buffer_info->skb) {
3772                 skb = dev_alloc_skb(bufsz);
3773
3774                 if(unlikely(!skb)) {
3775                         /* Better luck next round */
3776                         break;
3777                 }
3778
3779                 /* Fix for errata 23, can't cross 64kB boundary */
3780                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3781                         struct sk_buff *oldskb = skb;
3782                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
3783                                              "at %p\n", bufsz, skb->data);
3784                         /* Try again, without freeing the previous */
3785                         skb = dev_alloc_skb(bufsz);
3786                         /* Failed allocation, critical failure */
3787                         if (!skb) {
3788                                 dev_kfree_skb(oldskb);
3789                                 break;
3790                         }
3791
3792                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3793                                 /* give up */
3794                                 dev_kfree_skb(skb);
3795                                 dev_kfree_skb(oldskb);
3796                                 break; /* while !buffer_info->skb */
3797                         } else {
3798                                 /* Use new allocation */
3799                                 dev_kfree_skb(oldskb);
3800                         }
3801                 }
3802                 /* Make buffer alignment 2 beyond a 16 byte boundary
3803                  * this will result in a 16 byte aligned IP header after
3804                  * the 14 byte MAC header is removed
3805                  */
3806                 skb_reserve(skb, NET_IP_ALIGN);
3807
3808                 skb->dev = netdev;
3809
3810                 buffer_info->skb = skb;
3811                 buffer_info->length = adapter->rx_buffer_len;
3812                 buffer_info->dma = pci_map_single(pdev,
3813                                                   skb->data,
3814                                                   adapter->rx_buffer_len,
3815                                                   PCI_DMA_FROMDEVICE);
3816
3817                 /* Fix for errata 23, can't cross 64kB boundary */
3818                 if (!e1000_check_64k_bound(adapter,
3819                                         (void *)(unsigned long)buffer_info->dma,
3820                                         adapter->rx_buffer_len)) {
3821                         DPRINTK(RX_ERR, ERR,
3822                                 "dma align check failed: %u bytes at %p\n",
3823                                 adapter->rx_buffer_len,
3824                                 (void *)(unsigned long)buffer_info->dma);
3825                         dev_kfree_skb(skb);
3826                         buffer_info->skb = NULL;
3827
3828                         pci_unmap_single(pdev, buffer_info->dma,
3829                                          adapter->rx_buffer_len,
3830                                          PCI_DMA_FROMDEVICE);
3831
3832                         break; /* while !buffer_info->skb */
3833                 }
3834                 rx_desc = E1000_RX_DESC(*rx_ring, i);
3835                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3836
3837                 if(unlikely((i & ~(E1000_RX_BUFFER_WRITE - 1)) == i)) {
3838                         /* Force memory writes to complete before letting h/w
3839                          * know there are new descriptors to fetch.  (Only
3840                          * applicable for weak-ordered memory model archs,
3841                          * such as IA-64). */
3842                         wmb();
3843                         writel(i, adapter->hw.hw_addr + rx_ring->rdt);
3844                 }
3845
3846                 if(unlikely(++i == rx_ring->count)) i = 0;
3847                 buffer_info = &rx_ring->buffer_info[i];
3848         }
3849
3850         rx_ring->next_to_use = i;
3851 }
3852
3853 /**
3854  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
3855  * @adapter: address of board private structure
3856  **/
3857
3858 static void
3859 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
3860                           struct e1000_rx_ring *rx_ring)
3861 {
3862         struct net_device *netdev = adapter->netdev;
3863         struct pci_dev *pdev = adapter->pdev;
3864         union e1000_rx_desc_packet_split *rx_desc;
3865         struct e1000_buffer *buffer_info;
3866         struct e1000_ps_page *ps_page;
3867         struct e1000_ps_page_dma *ps_page_dma;
3868         struct sk_buff *skb;
3869         unsigned int i, j;
3870
3871         i = rx_ring->next_to_use;
3872         buffer_info = &rx_ring->buffer_info[i];
3873         ps_page = &rx_ring->ps_page[i];
3874         ps_page_dma = &rx_ring->ps_page_dma[i];
3875
3876         while(!buffer_info->skb) {
3877                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
3878
3879                 for(j = 0; j < PS_PAGE_BUFFERS; j++) {
3880                         if (j < adapter->rx_ps_pages) {
3881                                 if (likely(!ps_page->ps_page[j])) {
3882                                         ps_page->ps_page[j] =
3883                                                 alloc_page(GFP_ATOMIC);
3884                                         if (unlikely(!ps_page->ps_page[j]))
3885                                                 goto no_buffers;
3886                                         ps_page_dma->ps_page_dma[j] =
3887                                                 pci_map_page(pdev,
3888                                                             ps_page->ps_page[j],
3889                                                             0, PAGE_SIZE,
3890                                                             PCI_DMA_FROMDEVICE);
3891                                 }
3892                                 /* Refresh the desc even if buffer_addrs didn't
3893                                  * change because each write-back erases
3894                                  * this info.
3895                                  */
3896                                 rx_desc->read.buffer_addr[j+1] =
3897                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
3898                         } else
3899                                 rx_desc->read.buffer_addr[j+1] = ~0;
3900                 }
3901
3902                 skb = dev_alloc_skb(adapter->rx_ps_bsize0 + NET_IP_ALIGN);
3903
3904                 if(unlikely(!skb))
3905                         break;
3906
3907                 /* Make buffer alignment 2 beyond a 16 byte boundary
3908                  * this will result in a 16 byte aligned IP header after
3909                  * the 14 byte MAC header is removed
3910                  */
3911                 skb_reserve(skb, NET_IP_ALIGN);
3912
3913                 skb->dev = netdev;
3914
3915                 buffer_info->skb = skb;
3916                 buffer_info->length = adapter->rx_ps_bsize0;
3917                 buffer_info->dma = pci_map_single(pdev, skb->data,
3918                                                   adapter->rx_ps_bsize0,
3919                                                   PCI_DMA_FROMDEVICE);
3920
3921                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
3922
3923                 if(unlikely((i & ~(E1000_RX_BUFFER_WRITE - 1)) == i)) {
3924                         /* Force memory writes to complete before letting h/w
3925                          * know there are new descriptors to fetch.  (Only
3926                          * applicable for weak-ordered memory model archs,
3927                          * such as IA-64). */
3928                         wmb();
3929                         /* Hardware increments by 16 bytes, but packet split
3930                          * descriptors are 32 bytes...so we increment tail
3931                          * twice as much.
3932                          */
3933                         writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
3934                 }
3935
3936                 if(unlikely(++i == rx_ring->count)) i = 0;
3937                 buffer_info = &rx_ring->buffer_info[i];
3938                 ps_page = &rx_ring->ps_page[i];
3939                 ps_page_dma = &rx_ring->ps_page_dma[i];
3940         }
3941
3942 no_buffers:
3943         rx_ring->next_to_use = i;
3944 }
3945
3946 /**
3947  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
3948  * @adapter:
3949  **/
3950
3951 static void
3952 e1000_smartspeed(struct e1000_adapter *adapter)
3953 {
3954         uint16_t phy_status;
3955         uint16_t phy_ctrl;
3956
3957         if((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
3958            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
3959                 return;
3960
3961         if(adapter->smartspeed == 0) {
3962                 /* If Master/Slave config fault is asserted twice,
3963                  * we assume back-to-back */
3964                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
3965                 if(!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
3966                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
3967                 if(!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
3968                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
3969                 if(phy_ctrl & CR_1000T_MS_ENABLE) {
3970                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
3971                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
3972                                             phy_ctrl);
3973                         adapter->smartspeed++;
3974                         if(!e1000_phy_setup_autoneg(&adapter->hw) &&
3975                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
3976                                                &phy_ctrl)) {
3977                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
3978                                              MII_CR_RESTART_AUTO_NEG);
3979                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
3980                                                     phy_ctrl);
3981                         }
3982                 }
3983                 return;
3984         } else if(adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
3985                 /* If still no link, perhaps using 2/3 pair cable */
3986                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
3987                 phy_ctrl |= CR_1000T_MS_ENABLE;
3988                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
3989                 if(!e1000_phy_setup_autoneg(&adapter->hw) &&
3990                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
3991                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
3992                                      MII_CR_RESTART_AUTO_NEG);
3993                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
3994                 }
3995         }
3996         /* Restart process after E1000_SMARTSPEED_MAX iterations */
3997         if(adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
3998                 adapter->smartspeed = 0;
3999 }
4000
4001 /**
4002  * e1000_ioctl -
4003  * @netdev:
4004  * @ifreq:
4005  * @cmd:
4006  **/
4007
4008 static int
4009 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4010 {
4011         switch (cmd) {
4012         case SIOCGMIIPHY:
4013         case SIOCGMIIREG:
4014         case SIOCSMIIREG:
4015                 return e1000_mii_ioctl(netdev, ifr, cmd);
4016         default:
4017                 return -EOPNOTSUPP;
4018         }
4019 }
4020
4021 /**
4022  * e1000_mii_ioctl -
4023  * @netdev:
4024  * @ifreq:
4025  * @cmd:
4026  **/
4027
4028 static int
4029 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4030 {
4031         struct e1000_adapter *adapter = netdev_priv(netdev);
4032         struct mii_ioctl_data *data = if_mii(ifr);
4033         int retval;
4034         uint16_t mii_reg;
4035         uint16_t spddplx;
4036         unsigned long flags;
4037
4038         if(adapter->hw.media_type != e1000_media_type_copper)
4039                 return -EOPNOTSUPP;
4040
4041         switch (cmd) {
4042         case SIOCGMIIPHY:
4043                 data->phy_id = adapter->hw.phy_addr;
4044                 break;
4045         case SIOCGMIIREG:
4046                 if(!capable(CAP_NET_ADMIN))
4047                         return -EPERM;
4048                 spin_lock_irqsave(&adapter->stats_lock, flags);
4049                 if(e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4050                                    &data->val_out)) {
4051                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4052                         return -EIO;
4053                 }
4054                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4055                 break;
4056         case SIOCSMIIREG:
4057                 if(!capable(CAP_NET_ADMIN))
4058                         return -EPERM;
4059                 if(data->reg_num & ~(0x1F))
4060                         return -EFAULT;
4061                 mii_reg = data->val_in;
4062                 spin_lock_irqsave(&adapter->stats_lock, flags);
4063                 if(e1000_write_phy_reg(&adapter->hw, data->reg_num,
4064                                         mii_reg)) {
4065                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4066                         return -EIO;
4067                 }
4068                 if(adapter->hw.phy_type == e1000_phy_m88) {
4069                         switch (data->reg_num) {
4070                         case PHY_CTRL:
4071                                 if(mii_reg & MII_CR_POWER_DOWN)
4072                                         break;
4073                                 if(mii_reg & MII_CR_AUTO_NEG_EN) {
4074                                         adapter->hw.autoneg = 1;
4075                                         adapter->hw.autoneg_advertised = 0x2F;
4076                                 } else {
4077                                         if (mii_reg & 0x40)
4078                                                 spddplx = SPEED_1000;
4079                                         else if (mii_reg & 0x2000)
4080                                                 spddplx = SPEED_100;
4081                                         else
4082                                                 spddplx = SPEED_10;
4083                                         spddplx += (mii_reg & 0x100)
4084                                                    ? FULL_DUPLEX :
4085                                                    HALF_DUPLEX;
4086                                         retval = e1000_set_spd_dplx(adapter,
4087                                                                     spddplx);
4088                                         if(retval) {
4089                                                 spin_unlock_irqrestore(
4090                                                         &adapter->stats_lock,
4091                                                         flags);
4092                                                 return retval;
4093                                         }
4094                                 }
4095                                 if(netif_running(adapter->netdev)) {
4096                                         e1000_down(adapter);
4097                                         e1000_up(adapter);
4098                                 } else
4099                                         e1000_reset(adapter);
4100                                 break;
4101                         case M88E1000_PHY_SPEC_CTRL:
4102                         case M88E1000_EXT_PHY_SPEC_CTRL:
4103                                 if(e1000_phy_reset(&adapter->hw)) {
4104                                         spin_unlock_irqrestore(
4105                                                 &adapter->stats_lock, flags);
4106                                         return -EIO;
4107                                 }
4108                                 break;
4109                         }
4110                 } else {
4111                         switch (data->reg_num) {
4112                         case PHY_CTRL:
4113                                 if(mii_reg & MII_CR_POWER_DOWN)
4114                                         break;
4115                                 if(netif_running(adapter->netdev)) {
4116                                         e1000_down(adapter);
4117                                         e1000_up(adapter);
4118                                 } else
4119                                         e1000_reset(adapter);
4120                                 break;
4121                         }
4122                 }
4123                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4124                 break;
4125         default:
4126                 return -EOPNOTSUPP;
4127         }
4128         return E1000_SUCCESS;
4129 }
4130
4131 void
4132 e1000_pci_set_mwi(struct e1000_hw *hw)
4133 {
4134         struct e1000_adapter *adapter = hw->back;
4135         int ret_val = pci_set_mwi(adapter->pdev);
4136
4137         if(ret_val)
4138                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4139 }
4140
4141 void
4142 e1000_pci_clear_mwi(struct e1000_hw *hw)
4143 {
4144         struct e1000_adapter *adapter = hw->back;
4145
4146         pci_clear_mwi(adapter->pdev);
4147 }
4148
4149 void
4150 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4151 {
4152         struct e1000_adapter *adapter = hw->back;
4153
4154         pci_read_config_word(adapter->pdev, reg, value);
4155 }
4156
4157 void
4158 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4159 {
4160         struct e1000_adapter *adapter = hw->back;
4161
4162         pci_write_config_word(adapter->pdev, reg, *value);
4163 }
4164
4165 uint32_t
4166 e1000_io_read(struct e1000_hw *hw, unsigned long port)
4167 {
4168         return inl(port);
4169 }
4170
4171 void
4172 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4173 {
4174         outl(value, port);
4175 }
4176
4177 static void
4178 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4179 {
4180         struct e1000_adapter *adapter = netdev_priv(netdev);
4181         uint32_t ctrl, rctl;
4182
4183         e1000_irq_disable(adapter);
4184         adapter->vlgrp = grp;
4185
4186         if(grp) {
4187                 /* enable VLAN tag insert/strip */
4188                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4189                 ctrl |= E1000_CTRL_VME;
4190                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4191
4192                 /* enable VLAN receive filtering */
4193                 rctl = E1000_READ_REG(&adapter->hw, RCTL);
4194                 rctl |= E1000_RCTL_VFE;
4195                 rctl &= ~E1000_RCTL_CFIEN;
4196                 E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4197                 e1000_update_mng_vlan(adapter);
4198         } else {
4199                 /* disable VLAN tag insert/strip */
4200                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4201                 ctrl &= ~E1000_CTRL_VME;
4202                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4203
4204                 /* disable VLAN filtering */
4205                 rctl = E1000_READ_REG(&adapter->hw, RCTL);
4206                 rctl &= ~E1000_RCTL_VFE;
4207                 E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4208                 if(adapter->mng_vlan_id != (uint16_t)E1000_MNG_VLAN_NONE) {
4209                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4210                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4211                 }
4212         }
4213
4214         e1000_irq_enable(adapter);
4215 }
4216
4217 static void
4218 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4219 {
4220         struct e1000_adapter *adapter = netdev_priv(netdev);
4221         uint32_t vfta, index;
4222         if((adapter->hw.mng_cookie.status &
4223                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4224                 (vid == adapter->mng_vlan_id))
4225                 return;
4226         /* add VID to filter table */
4227         index = (vid >> 5) & 0x7F;
4228         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4229         vfta |= (1 << (vid & 0x1F));
4230         e1000_write_vfta(&adapter->hw, index, vfta);
4231 }
4232
4233 static void
4234 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4235 {
4236         struct e1000_adapter *adapter = netdev_priv(netdev);
4237         uint32_t vfta, index;
4238
4239         e1000_irq_disable(adapter);
4240
4241         if(adapter->vlgrp)
4242                 adapter->vlgrp->vlan_devices[vid] = NULL;
4243
4244         e1000_irq_enable(adapter);
4245
4246         if((adapter->hw.mng_cookie.status &
4247                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4248             (vid == adapter->mng_vlan_id)) {
4249                 /* release control to f/w */
4250                 e1000_release_hw_control(adapter);
4251                 return;
4252         }
4253
4254         /* remove VID from filter table */
4255         index = (vid >> 5) & 0x7F;
4256         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4257         vfta &= ~(1 << (vid & 0x1F));
4258         e1000_write_vfta(&adapter->hw, index, vfta);
4259 }
4260
4261 static void
4262 e1000_restore_vlan(struct e1000_adapter *adapter)
4263 {
4264         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4265
4266         if(adapter->vlgrp) {
4267                 uint16_t vid;
4268                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4269                         if(!adapter->vlgrp->vlan_devices[vid])
4270                                 continue;
4271                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4272                 }
4273         }
4274 }
4275
4276 int
4277 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
4278 {
4279         adapter->hw.autoneg = 0;
4280
4281         /* Fiber NICs only allow 1000 gbps Full duplex */
4282         if((adapter->hw.media_type == e1000_media_type_fiber) &&
4283                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4284                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4285                 return -EINVAL;
4286         }
4287
4288         switch(spddplx) {
4289         case SPEED_10 + DUPLEX_HALF:
4290                 adapter->hw.forced_speed_duplex = e1000_10_half;
4291                 break;
4292         case SPEED_10 + DUPLEX_FULL:
4293                 adapter->hw.forced_speed_duplex = e1000_10_full;
4294                 break;
4295         case SPEED_100 + DUPLEX_HALF:
4296                 adapter->hw.forced_speed_duplex = e1000_100_half;
4297                 break;
4298         case SPEED_100 + DUPLEX_FULL:
4299                 adapter->hw.forced_speed_duplex = e1000_100_full;
4300                 break;
4301         case SPEED_1000 + DUPLEX_FULL:
4302                 adapter->hw.autoneg = 1;
4303                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
4304                 break;
4305         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4306         default:
4307                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4308                 return -EINVAL;
4309         }
4310         return 0;
4311 }
4312
4313 #ifdef CONFIG_PM
4314 static int
4315 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4316 {
4317         struct net_device *netdev = pci_get_drvdata(pdev);
4318         struct e1000_adapter *adapter = netdev_priv(netdev);
4319         uint32_t ctrl, ctrl_ext, rctl, manc, status;
4320         uint32_t wufc = adapter->wol;
4321
4322         netif_device_detach(netdev);
4323
4324         if(netif_running(netdev))
4325                 e1000_down(adapter);
4326
4327         status = E1000_READ_REG(&adapter->hw, STATUS);
4328         if(status & E1000_STATUS_LU)
4329                 wufc &= ~E1000_WUFC_LNKC;
4330
4331         if(wufc) {
4332                 e1000_setup_rctl(adapter);
4333                 e1000_set_multi(netdev);
4334
4335                 /* turn on all-multi mode if wake on multicast is enabled */
4336                 if(adapter->wol & E1000_WUFC_MC) {
4337                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4338                         rctl |= E1000_RCTL_MPE;
4339                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4340                 }
4341
4342                 if(adapter->hw.mac_type >= e1000_82540) {
4343                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4344                         /* advertise wake from D3Cold */
4345                         #define E1000_CTRL_ADVD3WUC 0x00100000
4346                         /* phy power management enable */
4347                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4348                         ctrl |= E1000_CTRL_ADVD3WUC |
4349                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4350                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4351                 }
4352
4353                 if(adapter->hw.media_type == e1000_media_type_fiber ||
4354                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
4355                         /* keep the laser running in D3 */
4356                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
4357                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4358                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
4359                 }
4360
4361                 /* Allow time for pending master requests to run */
4362                 e1000_disable_pciex_master(&adapter->hw);
4363
4364                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
4365                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
4366                 pci_enable_wake(pdev, 3, 1);
4367                 pci_enable_wake(pdev, 4, 1); /* 4 == D3 cold */
4368         } else {
4369                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
4370                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
4371                 pci_enable_wake(pdev, 3, 0);
4372                 pci_enable_wake(pdev, 4, 0); /* 4 == D3 cold */
4373         }
4374
4375         pci_save_state(pdev);
4376
4377         if(adapter->hw.mac_type >= e1000_82540 &&
4378            adapter->hw.media_type == e1000_media_type_copper) {
4379                 manc = E1000_READ_REG(&adapter->hw, MANC);
4380                 if(manc & E1000_MANC_SMBUS_EN) {
4381                         manc |= E1000_MANC_ARP_EN;
4382                         E1000_WRITE_REG(&adapter->hw, MANC, manc);
4383                         pci_enable_wake(pdev, 3, 1);
4384                         pci_enable_wake(pdev, 4, 1); /* 4 == D3 cold */
4385                 }
4386         }
4387
4388         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4389          * would have already happened in close and is redundant. */
4390         e1000_release_hw_control(adapter);
4391
4392         pci_disable_device(pdev);
4393         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4394
4395         return 0;
4396 }
4397
4398 static int
4399 e1000_resume(struct pci_dev *pdev)
4400 {
4401         struct net_device *netdev = pci_get_drvdata(pdev);
4402         struct e1000_adapter *adapter = netdev_priv(netdev);
4403         uint32_t manc, ret_val;
4404
4405         pci_set_power_state(pdev, PCI_D0);
4406         pci_restore_state(pdev);
4407         ret_val = pci_enable_device(pdev);
4408         pci_set_master(pdev);
4409
4410         pci_enable_wake(pdev, PCI_D3hot, 0);
4411         pci_enable_wake(pdev, PCI_D3cold, 0);
4412
4413         e1000_reset(adapter);
4414         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
4415
4416         if(netif_running(netdev))
4417                 e1000_up(adapter);
4418
4419         netif_device_attach(netdev);
4420
4421         if(adapter->hw.mac_type >= e1000_82540 &&
4422            adapter->hw.media_type == e1000_media_type_copper) {
4423                 manc = E1000_READ_REG(&adapter->hw, MANC);
4424                 manc &= ~(E1000_MANC_ARP_EN);
4425                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
4426         }
4427
4428         /* If the controller is 82573 and f/w is AMT, do not set
4429          * DRV_LOAD until the interface is up.  For all other cases,
4430          * let the f/w know that the h/w is now under the control
4431          * of the driver. */
4432         if (adapter->hw.mac_type != e1000_82573 ||
4433             !e1000_check_mng_mode(&adapter->hw))
4434                 e1000_get_hw_control(adapter);
4435
4436         return 0;
4437 }
4438 #endif
4439 #ifdef CONFIG_NET_POLL_CONTROLLER
4440 /*
4441  * Polling 'interrupt' - used by things like netconsole to send skbs
4442  * without having to re-enable interrupts. It's not called while
4443  * the interrupt routine is executing.
4444  */
4445 static void
4446 e1000_netpoll(struct net_device *netdev)
4447 {
4448         struct e1000_adapter *adapter = netdev_priv(netdev);
4449         disable_irq(adapter->pdev->irq);
4450         e1000_intr(adapter->pdev->irq, netdev, NULL);
4451         e1000_clean_tx_irq(adapter, adapter->tx_ring);
4452 #ifndef CONFIG_E1000_NAPI
4453         adapter->clean_rx(adapter, adapter->rx_ring);
4454 #endif
4455         enable_irq(adapter->pdev->irq);
4456 }
4457 #endif
4458
4459 /* e1000_main.c */