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(0x1099),
 101         INTEL_E1000_ETHERNET_DEVICE(0x109A),
 102         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
 103         /* required last entry */
 104         {0,}
 105 };
 106
 107 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
 108
 109 int e1000_up(struct e1000_adapter *adapter);
 110 void e1000_down(struct e1000_adapter *adapter);
 111 void e1000_reset(struct e1000_adapter *adapter);
 112 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
 113 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
 114 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
 115 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
 116 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
 117 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
 118                                     struct e1000_tx_ring *txdr);
 119 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
 120                                     struct e1000_rx_ring *rxdr);
 121 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
 122                                     struct e1000_tx_ring *tx_ring);
 123 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
 124                                     struct e1000_rx_ring *rx_ring);
 125 void e1000_update_stats(struct e1000_adapter *adapter);
 126
 127 /* Local Function Prototypes */
 128
 129 static int e1000_init_module(void);
 130 static void e1000_exit_module(void);
 131 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
 132 static void __devexit e1000_remove(struct pci_dev *pdev);
 133 static int e1000_alloc_queues(struct e1000_adapter *adapter);
 134 #ifdef CONFIG_E1000_MQ
 135 static void e1000_setup_queue_mapping(struct e1000_adapter *adapter);
 136 #endif
 137 static int e1000_sw_init(struct e1000_adapter *adapter);
 138 static int e1000_open(struct net_device *netdev);
 139 static int e1000_close(struct net_device *netdev);
 140 static void e1000_configure_tx(struct e1000_adapter *adapter);
 141 static void e1000_configure_rx(struct e1000_adapter *adapter);
 142 static void e1000_setup_rctl(struct e1000_adapter *adapter);
 143 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
 144 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
 145 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
 146                                 struct e1000_tx_ring *tx_ring);
 147 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
 148                                 struct e1000_rx_ring *rx_ring);
 149 static void e1000_set_multi(struct net_device *netdev);
 150 static void e1000_update_phy_info(unsigned long data);
 151 static void e1000_watchdog(unsigned long data);
 152 static void e1000_watchdog_task(struct e1000_adapter *adapter);
 153 static void e1000_82547_tx_fifo_stall(unsigned long data);
 154 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
 155 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
 156 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
 157 static int e1000_set_mac(struct net_device *netdev, void *p);
 158 static irqreturn_t e1000_intr(int irq, void *data, struct pt_regs *regs);
 159 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
 160                                     struct e1000_tx_ring *tx_ring);
 161 #ifdef CONFIG_E1000_NAPI
 162 static int e1000_clean(struct net_device *poll_dev, int *budget);
 163 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
 164                                     struct e1000_rx_ring *rx_ring,
 165                                     int *work_done, int work_to_do);
 166 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 167                                        struct e1000_rx_ring *rx_ring,
 168                                        int *work_done, int work_to_do);
 169 #else
 170 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
 171                                     struct e1000_rx_ring *rx_ring);
 172 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 173                                        struct e1000_rx_ring *rx_ring);
 174 #endif
 175 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 176                                    struct e1000_rx_ring *rx_ring,
 177                                    int cleaned_count);
 178 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 179                                       struct e1000_rx_ring *rx_ring,
 180                                       int cleaned_count);
 181 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
 182 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
 183                            int cmd);
 184 void e1000_set_ethtool_ops(struct net_device *netdev);
 185 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
 186 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
 187 static void e1000_tx_timeout(struct net_device *dev);
 188 static void e1000_tx_timeout_task(struct net_device *dev);
 189 static void e1000_smartspeed(struct e1000_adapter *adapter);
 190 static inline int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
 191                                               struct sk_buff *skb);
 192
 193 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
 194 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
 195 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
 196 static void e1000_restore_vlan(struct e1000_adapter *adapter);
 197
 198 #ifdef CONFIG_PM
 199 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
 200 static int e1000_resume(struct pci_dev *pdev);
 201 #endif
 202
 203 #ifdef CONFIG_NET_POLL_CONTROLLER
 204 /* for netdump / net console */
 205 static void e1000_netpoll (struct net_device *netdev);
 206 #endif
 207
 208 #ifdef CONFIG_E1000_MQ
 209 /* for multiple Rx queues */
 210 void e1000_rx_schedule(void *data);
 211 #endif
 212
 213 /* Exported from other modules */
 214
 215 extern void e1000_check_options(struct e1000_adapter *adapter);
 216
 217 static struct pci_driver e1000_driver = {
 218         .name     = e1000_driver_name,
 219         .id_table = e1000_pci_tbl,
 220         .probe    = e1000_probe,
 221         .remove   = __devexit_p(e1000_remove),
 222         /* Power Managment Hooks */
 223 #ifdef CONFIG_PM
 224         .suspend  = e1000_suspend,
 225         .resume   = e1000_resume
 226 #endif
 227 };
 228
 229 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 230 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
 231 MODULE_LICENSE("GPL");
 232 MODULE_VERSION(DRV_VERSION);
 233
 234 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
 235 module_param(debug, int, 0);
 236 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 237
 238 /**
 239  * e1000_init_module - Driver Registration Routine
 240  *
 241  * e1000_init_module is the first routine called when the driver is
 242  * loaded. All it does is register with the PCI subsystem.
 243  **/
 244
 245 static int __init
 246 e1000_init_module(void)
 247 {
 248         int ret;
 249         printk(KERN_INFO "%s - version %s\n",
 250                e1000_driver_string, e1000_driver_version);
 251
 252         printk(KERN_INFO "%s\n", e1000_copyright);
 253
 254         ret = pci_module_init(&e1000_driver);
 255
 256         return ret;
 257 }
 258
 259 module_init(e1000_init_module);
 260
 261 /**
 262  * e1000_exit_module - Driver Exit Cleanup Routine
 263  *
 264  * e1000_exit_module is called just before the driver is removed
 265  * from memory.
 266  **/
 267
 268 static void __exit
 269 e1000_exit_module(void)
 270 {
 271         pci_unregister_driver(&e1000_driver);
 272 }
 273
 274 module_exit(e1000_exit_module);
 275
 276 /**
 277  * e1000_irq_disable - Mask off interrupt generation on the NIC
 278  * @adapter: board private structure
 279  **/
 280
 281 static inline void
 282 e1000_irq_disable(struct e1000_adapter *adapter)
 283 {
 284         atomic_inc(&adapter->irq_sem);
 285         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
 286         E1000_WRITE_FLUSH(&adapter->hw);
 287         synchronize_irq(adapter->pdev->irq);
 288 }
 289
 290 /**
 291  * e1000_irq_enable - Enable default interrupt generation settings
 292  * @adapter: board private structure
 293  **/
 294
 295 static inline void
 296 e1000_irq_enable(struct e1000_adapter *adapter)
 297 {
 298         if(likely(atomic_dec_and_test(&adapter->irq_sem))) {
 299                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
 300                 E1000_WRITE_FLUSH(&adapter->hw);
 301         }
 302 }
 303
 304 static void
 305 e1000_update_mng_vlan(struct e1000_adapter *adapter)
 306 {
 307         struct net_device *netdev = adapter->netdev;
 308         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
 309         uint16_t old_vid = adapter->mng_vlan_id;
 310         if(adapter->vlgrp) {
 311                 if(!adapter->vlgrp->vlan_devices[vid]) {
 312                         if(adapter->hw.mng_cookie.status &
 313                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
 314                                 e1000_vlan_rx_add_vid(netdev, vid);
 315                                 adapter->mng_vlan_id = vid;
 316                         } else
 317                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
 318
 319                         if((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
 320                                         (vid != old_vid) &&
 321                                         !adapter->vlgrp->vlan_devices[old_vid])
 322                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
 323                 }
 324         }
 325 }
 326
 327 /**
 328  * e1000_release_hw_control - release control of the h/w to f/w
 329  * @adapter: address of board private structure
 330  *
 331  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
 332  * For ASF and Pass Through versions of f/w this means that the
 333  * driver is no longer loaded. For AMT version (only with 82573) i
 334  * of the f/w this means that the netowrk i/f is closed.
 335  *
 336  **/
 337
 338 static inline void
 339 e1000_release_hw_control(struct e1000_adapter *adapter)
 340 {
 341         uint32_t ctrl_ext;
 342         uint32_t swsm;
 343
 344         /* Let firmware taken over control of h/w */
 345         switch (adapter->hw.mac_type) {
 346         case e1000_82571:
 347         case e1000_82572:
 348                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
 349                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
 350                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 351                 break;
 352         case e1000_82573:
 353                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
 354                 E1000_WRITE_REG(&adapter->hw, SWSM,
 355                                 swsm & ~E1000_SWSM_DRV_LOAD);
 356         default:
 357                 break;
 358         }
 359 }
 360
 361 /**
 362  * e1000_get_hw_control - get control of the h/w from f/w
 363  * @adapter: address of board private structure
 364  *
 365  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
 366  * For ASF and Pass Through versions of f/w this means that
 367  * the driver is loaded. For AMT version (only with 82573)
 368  * of the f/w this means that the netowrk i/f is open.
 369  *
 370  **/
 371
 372 static inline void
 373 e1000_get_hw_control(struct e1000_adapter *adapter)
 374 {
 375         uint32_t ctrl_ext;
 376         uint32_t swsm;
 377         /* Let firmware know the driver has taken over */
 378         switch (adapter->hw.mac_type) {
 379         case e1000_82571:
 380         case e1000_82572:
 381                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
 382                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
 383                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 384                 break;
 385         case e1000_82573:
 386                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
 387                 E1000_WRITE_REG(&adapter->hw, SWSM,
 388                                 swsm | E1000_SWSM_DRV_LOAD);
 389                 break;
 390         default:
 391                 break;
 392         }
 393 }
 394
 395 int
 396 e1000_up(struct e1000_adapter *adapter)
 397 {
 398         struct net_device *netdev = adapter->netdev;
 399         int i, err;
 400
 401         /* hardware has been reset, we need to reload some things */
 402
 403         /* Reset the PHY if it was previously powered down */
 404         if(adapter->hw.media_type == e1000_media_type_copper) {
 405                 uint16_t mii_reg;
 406                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
 407                 if(mii_reg & MII_CR_POWER_DOWN)
 408                         e1000_phy_reset(&adapter->hw);
 409         }
 410
 411         e1000_set_multi(netdev);
 412
 413         e1000_restore_vlan(adapter);
 414
 415         e1000_configure_tx(adapter);
 416         e1000_setup_rctl(adapter);
 417         e1000_configure_rx(adapter);
 418         /* call E1000_DESC_UNUSED which always leaves
 419          * at least 1 descriptor unused to make sure
 420          * next_to_use != next_to_clean */
 421         for (i = 0; i < adapter->num_rx_queues; i++) {
 422                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
 423                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
 424         }
 425
 426 #ifdef CONFIG_PCI_MSI
 427         if(adapter->hw.mac_type > e1000_82547_rev_2) {
 428                 adapter->have_msi = TRUE;
 429                 if((err = pci_enable_msi(adapter->pdev))) {
 430                         DPRINTK(PROBE, ERR,
 431                          "Unable to allocate MSI interrupt Error: %d\n", err);
 432                         adapter->have_msi = FALSE;
 433                 }
 434         }
 435 #endif
 436         if((err = request_irq(adapter->pdev->irq, &e1000_intr,
 437                               SA_SHIRQ | SA_SAMPLE_RANDOM,
 438                               netdev->name, netdev))) {
 439                 DPRINTK(PROBE, ERR,
 440                     "Unable to allocate interrupt Error: %d\n", err);
 441                 return err;
 442         }
 443
 444 #ifdef CONFIG_E1000_MQ
 445         e1000_setup_queue_mapping(adapter);
 446 #endif
 447
 448         adapter->tx_queue_len = netdev->tx_queue_len;
 449
 450         mod_timer(&adapter->watchdog_timer, jiffies);
 451
 452 #ifdef CONFIG_E1000_NAPI
 453         netif_poll_enable(netdev);
 454 #endif
 455         e1000_irq_enable(adapter);
 456
 457         return 0;
 458 }
 459
 460 void
 461 e1000_down(struct e1000_adapter *adapter)
 462 {
 463         struct net_device *netdev = adapter->netdev;
 464         boolean_t mng_mode_enabled = (adapter->hw.mac_type >= e1000_82571) &&
 465                                      e1000_check_mng_mode(&adapter->hw);
 466
 467         e1000_irq_disable(adapter);
 468 #ifdef CONFIG_E1000_MQ
 469         while (atomic_read(&adapter->rx_sched_call_data.count) != 0);
 470 #endif
 471         free_irq(adapter->pdev->irq, netdev);
 472 #ifdef CONFIG_PCI_MSI
 473         if(adapter->hw.mac_type > e1000_82547_rev_2 &&
 474            adapter->have_msi == TRUE)
 475                 pci_disable_msi(adapter->pdev);
 476 #endif
 477         del_timer_sync(&adapter->tx_fifo_stall_timer);
 478         del_timer_sync(&adapter->watchdog_timer);
 479         del_timer_sync(&adapter->phy_info_timer);
 480
 481 #ifdef CONFIG_E1000_NAPI
 482         netif_poll_disable(netdev);
 483 #endif
 484         netdev->tx_queue_len = adapter->tx_queue_len;
 485         adapter->link_speed = 0;
 486         adapter->link_duplex = 0;
 487         netif_carrier_off(netdev);
 488         netif_stop_queue(netdev);
 489
 490         e1000_reset(adapter);
 491         e1000_clean_all_tx_rings(adapter);
 492         e1000_clean_all_rx_rings(adapter);
 493
 494         /* Power down the PHY so no link is implied when interface is down *
 495          * The PHY cannot be powered down if any of the following is TRUE *
 496          * (a) WoL is enabled
 497          * (b) AMT is active
 498          * (c) SoL/IDER session is active */
 499         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
 500            adapter->hw.media_type == e1000_media_type_copper &&
 501            !(E1000_READ_REG(&adapter->hw, MANC) & E1000_MANC_SMBUS_EN) &&
 502            !mng_mode_enabled &&
 503            !e1000_check_phy_reset_block(&adapter->hw)) {
 504                 uint16_t mii_reg;
 505                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
 506                 mii_reg |= MII_CR_POWER_DOWN;
 507                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
 508                 mdelay(1);
 509         }
 510 }
 511
 512 void
 513 e1000_reset(struct e1000_adapter *adapter)
 514 {
 515         uint32_t pba, manc;
 516         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
 517
 518         /* Repartition Pba for greater than 9k mtu
 519          * To take effect CTRL.RST is required.
 520          */
 521
 522         switch (adapter->hw.mac_type) {
 523         case e1000_82547:
 524         case e1000_82547_rev_2:
 525                 pba = E1000_PBA_30K;
 526                 break;
 527         case e1000_82571:
 528         case e1000_82572:
 529                 pba = E1000_PBA_38K;
 530                 break;
 531         case e1000_82573:
 532                 pba = E1000_PBA_12K;
 533                 break;
 534         default:
 535                 pba = E1000_PBA_48K;
 536                 break;
 537         }
 538
 539         if((adapter->hw.mac_type != e1000_82573) &&
 540            (adapter->netdev->mtu > E1000_RXBUFFER_8192))
 541                 pba -= 8; /* allocate more FIFO for Tx */
 542
 543
 544         if(adapter->hw.mac_type == e1000_82547) {
 545                 adapter->tx_fifo_head = 0;
 546                 adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
 547                 adapter->tx_fifo_size =
 548                         (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
 549                 atomic_set(&adapter->tx_fifo_stall, 0);
 550         }
 551
 552         E1000_WRITE_REG(&adapter->hw, PBA, pba);
 553
 554         /* flow control settings */
 555         /* Set the FC high water mark to 90% of the FIFO size.
 556          * Required to clear last 3 LSB */
 557         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
 558
 559         adapter->hw.fc_high_water = fc_high_water_mark;
 560         adapter->hw.fc_low_water = fc_high_water_mark - 8;
 561         adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
 562         adapter->hw.fc_send_xon = 1;
 563         adapter->hw.fc = adapter->hw.original_fc;
 564
 565         /* Allow time for pending master requests to run */
 566         e1000_reset_hw(&adapter->hw);
 567         if(adapter->hw.mac_type >= e1000_82544)
 568                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
 569         if(e1000_init_hw(&adapter->hw))
 570                 DPRINTK(PROBE, ERR, "Hardware Error\n");
 571         e1000_update_mng_vlan(adapter);
 572         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 573         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
 574
 575         e1000_reset_adaptive(&adapter->hw);
 576         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
 577         if (adapter->en_mng_pt) {
 578                 manc = E1000_READ_REG(&adapter->hw, MANC);
 579                 manc |= (E1000_MANC_ARP_EN | E1000_MANC_EN_MNG2HOST);
 580                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
 581         }
 582 }
 583
 584 /**
 585  * e1000_probe - Device Initialization Routine
 586  * @pdev: PCI device information struct
 587  * @ent: entry in e1000_pci_tbl
 588  *
 589  * Returns 0 on success, negative on failure
 590  *
 591  * e1000_probe initializes an adapter identified by a pci_dev structure.
 592  * The OS initialization, configuring of the adapter private structure,
 593  * and a hardware reset occur.
 594  **/
 595
 596 static int __devinit
 597 e1000_probe(struct pci_dev *pdev,
 598             const struct pci_device_id *ent)
 599 {
 600         struct net_device *netdev;
 601         struct e1000_adapter *adapter;
 602         unsigned long mmio_start, mmio_len;
 603
 604         static int cards_found = 0;
 605         int i, err, pci_using_dac;
 606         uint16_t eeprom_data;
 607         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
 608         if((err = pci_enable_device(pdev)))
 609                 return err;
 610
 611         if(!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK))) {
 612                 pci_using_dac = 1;
 613         } else {
 614                 if((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK))) {
 615                         E1000_ERR("No usable DMA configuration, aborting\n");
 616                         return err;
 617                 }
 618                 pci_using_dac = 0;
 619         }
 620
 621         if((err = pci_request_regions(pdev, e1000_driver_name)))
 622                 return err;
 623
 624         pci_set_master(pdev);
 625
 626         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
 627         if(!netdev) {
 628                 err = -ENOMEM;
 629                 goto err_alloc_etherdev;
 630         }
 631
 632         SET_MODULE_OWNER(netdev);
 633         SET_NETDEV_DEV(netdev, &pdev->dev);
 634
 635         pci_set_drvdata(pdev, netdev);
 636         adapter = netdev_priv(netdev);
 637         adapter->netdev = netdev;
 638         adapter->pdev = pdev;
 639         adapter->hw.back = adapter;
 640         adapter->msg_enable = (1 << debug) - 1;
 641
 642         mmio_start = pci_resource_start(pdev, BAR_0);
 643         mmio_len = pci_resource_len(pdev, BAR_0);
 644
 645         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
 646         if(!adapter->hw.hw_addr) {
 647                 err = -EIO;
 648                 goto err_ioremap;
 649         }
 650
 651         for(i = BAR_1; i <= BAR_5; i++) {
 652                 if(pci_resource_len(pdev, i) == 0)
 653                         continue;
 654                 if(pci_resource_flags(pdev, i) & IORESOURCE_IO) {
 655                         adapter->hw.io_base = pci_resource_start(pdev, i);
 656                         break;
 657                 }
 658         }
 659
 660         netdev->open = &e1000_open;
 661         netdev->stop = &e1000_close;
 662         netdev->hard_start_xmit = &e1000_xmit_frame;
 663         netdev->get_stats = &e1000_get_stats;
 664         netdev->set_multicast_list = &e1000_set_multi;
 665         netdev->set_mac_address = &e1000_set_mac;
 666         netdev->change_mtu = &e1000_change_mtu;
 667         netdev->do_ioctl = &e1000_ioctl;
 668         e1000_set_ethtool_ops(netdev);
 669         netdev->tx_timeout = &e1000_tx_timeout;
 670         netdev->watchdog_timeo = 5 * HZ;
 671 #ifdef CONFIG_E1000_NAPI
 672         netdev->poll = &e1000_clean;
 673         netdev->weight = 64;
 674 #endif
 675         netdev->vlan_rx_register = e1000_vlan_rx_register;
 676         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
 677         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
 678 #ifdef CONFIG_NET_POLL_CONTROLLER
 679         netdev->poll_controller = e1000_netpoll;
 680 #endif
 681         strcpy(netdev->name, pci_name(pdev));
 682
 683         netdev->mem_start = mmio_start;
 684         netdev->mem_end = mmio_start + mmio_len;
 685         netdev->base_addr = adapter->hw.io_base;
 686
 687         adapter->bd_number = cards_found;
 688
 689         /* setup the private structure */
 690
 691         if((err = e1000_sw_init(adapter)))
 692                 goto err_sw_init;
 693
 694         if((err = e1000_check_phy_reset_block(&adapter->hw)))
 695                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
 696
 697         if(adapter->hw.mac_type >= e1000_82543) {
 698                 netdev->features = NETIF_F_SG |
 699                                    NETIF_F_HW_CSUM |
 700                                    NETIF_F_HW_VLAN_TX |
 701                                    NETIF_F_HW_VLAN_RX |
 702                                    NETIF_F_HW_VLAN_FILTER;
 703         }
 704
 705 #ifdef NETIF_F_TSO
 706         if((adapter->hw.mac_type >= e1000_82544) &&
 707            (adapter->hw.mac_type != e1000_82547))
 708                 netdev->features |= NETIF_F_TSO;
 709
 710 #ifdef NETIF_F_TSO_IPV6
 711         if(adapter->hw.mac_type > e1000_82547_rev_2)
 712                 netdev->features |= NETIF_F_TSO_IPV6;
 713 #endif
 714 #endif
 715         if(pci_using_dac)
 716                 netdev->features |= NETIF_F_HIGHDMA;
 717
 718         /* hard_start_xmit is safe against parallel locking */
 719         netdev->features |= NETIF_F_LLTX;
 720
 721         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
 722
 723         /* before reading the EEPROM, reset the controller to
 724          * put the device in a known good starting state */
 725
 726         e1000_reset_hw(&adapter->hw);
 727
 728         /* make sure the EEPROM is good */
 729
 730         if(e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
 731                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
 732                 err = -EIO;
 733                 goto err_eeprom;
 734         }
 735
 736         /* copy the MAC address out of the EEPROM */
 737
 738         if(e1000_read_mac_addr(&adapter->hw))
 739                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
 740         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
 741         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
 742
 743         if(!is_valid_ether_addr(netdev->perm_addr)) {
 744                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
 745                 err = -EIO;
 746                 goto err_eeprom;
 747         }
 748
 749         e1000_read_part_num(&adapter->hw, &(adapter->part_num));
 750
 751         e1000_get_bus_info(&adapter->hw);
 752
 753         init_timer(&adapter->tx_fifo_stall_timer);
 754         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
 755         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
 756
 757         init_timer(&adapter->watchdog_timer);
 758         adapter->watchdog_timer.function = &e1000_watchdog;
 759         adapter->watchdog_timer.data = (unsigned long) adapter;
 760
 761         INIT_WORK(&adapter->watchdog_task,
 762                 (void (*)(void *))e1000_watchdog_task, adapter);
 763
 764         init_timer(&adapter->phy_info_timer);
 765         adapter->phy_info_timer.function = &e1000_update_phy_info;
 766         adapter->phy_info_timer.data = (unsigned long) adapter;
 767
 768         INIT_WORK(&adapter->tx_timeout_task,
 769                 (void (*)(void *))e1000_tx_timeout_task, netdev);
 770
 771         /* we're going to reset, so assume we have no link for now */
 772
 773         netif_carrier_off(netdev);
 774         netif_stop_queue(netdev);
 775
 776         e1000_check_options(adapter);
 777
 778         /* Initial Wake on LAN setting
 779          * If APM wake is enabled in the EEPROM,
 780          * enable the ACPI Magic Packet filter
 781          */
 782
 783         switch(adapter->hw.mac_type) {
 784         case e1000_82542_rev2_0:
 785         case e1000_82542_rev2_1:
 786         case e1000_82543:
 787                 break;
 788         case e1000_82544:
 789                 e1000_read_eeprom(&adapter->hw,
 790                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
 791                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
 792                 break;
 793         case e1000_82546:
 794         case e1000_82546_rev_3:
 795         case e1000_82571:
 796                 if(E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
 797                         e1000_read_eeprom(&adapter->hw,
 798                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
 799                         break;
 800                 }
 801                 /* Fall Through */
 802         default:
 803                 e1000_read_eeprom(&adapter->hw,
 804                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
 805                 break;
 806         }
 807         if(eeprom_data & eeprom_apme_mask)
 808                 adapter->wol |= E1000_WUFC_MAG;
 809
 810         /* reset the hardware with the new settings */
 811         e1000_reset(adapter);
 812
 813         /* If the controller is 82573 and f/w is AMT, do not set
 814          * DRV_LOAD until the interface is up.  For all other cases,
 815          * let the f/w know that the h/w is now under the control
 816          * of the driver. */
 817         if (adapter->hw.mac_type != e1000_82573 ||
 818             !e1000_check_mng_mode(&adapter->hw))
 819                 e1000_get_hw_control(adapter);
 820
 821         strcpy(netdev->name, "eth%d");
 822         if((err = register_netdev(netdev)))
 823                 goto err_register;
 824
 825         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
 826
 827         cards_found++;
 828         return 0;
 829
 830 err_register:
 831 err_sw_init:
 832 err_eeprom:
 833         iounmap(adapter->hw.hw_addr);
 834 err_ioremap:
 835         free_netdev(netdev);
 836 err_alloc_etherdev:
 837         pci_release_regions(pdev);
 838         return err;
 839 }
 840
 841 /**
 842  * e1000_remove - Device Removal Routine
 843  * @pdev: PCI device information struct
 844  *
 845  * e1000_remove is called by the PCI subsystem to alert the driver
 846  * that it should release a PCI device.  The could be caused by a
 847  * Hot-Plug event, or because the driver is going to be removed from
 848  * memory.
 849  **/
 850
 851 static void __devexit
 852 e1000_remove(struct pci_dev *pdev)
 853 {
 854         struct net_device *netdev = pci_get_drvdata(pdev);
 855         struct e1000_adapter *adapter = netdev_priv(netdev);
 856         uint32_t manc;
 857 #ifdef CONFIG_E1000_NAPI
 858         int i;
 859 #endif
 860
 861         flush_scheduled_work();
 862
 863         if(adapter->hw.mac_type >= e1000_82540 &&
 864            adapter->hw.media_type == e1000_media_type_copper) {
 865                 manc = E1000_READ_REG(&adapter->hw, MANC);
 866                 if(manc & E1000_MANC_SMBUS_EN) {
 867                         manc |= E1000_MANC_ARP_EN;
 868                         E1000_WRITE_REG(&adapter->hw, MANC, manc);
 869                 }
 870         }
 871
 872         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
 873          * would have already happened in close and is redundant. */
 874         e1000_release_hw_control(adapter);
 875
 876         unregister_netdev(netdev);
 877 #ifdef CONFIG_E1000_NAPI
 878         for (i = 0; i < adapter->num_rx_queues; i++)
 879                 __dev_put(&adapter->polling_netdev[i]);
 880 #endif
 881
 882         if(!e1000_check_phy_reset_block(&adapter->hw))
 883                 e1000_phy_hw_reset(&adapter->hw);
 884
 885         kfree(adapter->tx_ring);
 886         kfree(adapter->rx_ring);
 887 #ifdef CONFIG_E1000_NAPI
 888         kfree(adapter->polling_netdev);
 889 #endif
 890
 891         iounmap(adapter->hw.hw_addr);
 892         pci_release_regions(pdev);
 893
 894 #ifdef CONFIG_E1000_MQ
 895         free_percpu(adapter->cpu_netdev);
 896         free_percpu(adapter->cpu_tx_ring);
 897 #endif
 898         free_netdev(netdev);
 899
 900         pci_disable_device(pdev);
 901 }
 902
 903 /**
 904  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
 905  * @adapter: board private structure to initialize
 906  *
 907  * e1000_sw_init initializes the Adapter private data structure.
 908  * Fields are initialized based on PCI device information and
 909  * OS network device settings (MTU size).
 910  **/
 911
 912 static int __devinit
 913 e1000_sw_init(struct e1000_adapter *adapter)
 914 {
 915         struct e1000_hw *hw = &adapter->hw;
 916         struct net_device *netdev = adapter->netdev;
 917         struct pci_dev *pdev = adapter->pdev;
 918 #ifdef CONFIG_E1000_NAPI
 919         int i;
 920 #endif
 921
 922         /* PCI config space info */
 923
 924         hw->vendor_id = pdev->vendor;
 925         hw->device_id = pdev->device;
 926         hw->subsystem_vendor_id = pdev->subsystem_vendor;
 927         hw->subsystem_id = pdev->subsystem_device;
 928
 929         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
 930
 931         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
 932
 933         adapter->rx_buffer_len = E1000_RXBUFFER_2048;
 934         adapter->rx_ps_bsize0 = E1000_RXBUFFER_256;
 935         hw->max_frame_size = netdev->mtu +
 936                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
 937         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
 938
 939         /* identify the MAC */
 940
 941         if(e1000_set_mac_type(hw)) {
 942                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
 943                 return -EIO;
 944         }
 945
 946         /* initialize eeprom parameters */
 947
 948         if(e1000_init_eeprom_params(hw)) {
 949                 E1000_ERR("EEPROM initialization failed\n");
 950                 return -EIO;
 951         }
 952
 953         switch(hw->mac_type) {
 954         default:
 955                 break;
 956         case e1000_82541:
 957         case e1000_82547:
 958         case e1000_82541_rev_2:
 959         case e1000_82547_rev_2:
 960                 hw->phy_init_script = 1;
 961                 break;
 962         }
 963
 964         e1000_set_media_type(hw);
 965
 966         hw->wait_autoneg_complete = FALSE;
 967         hw->tbi_compatibility_en = TRUE;
 968         hw->adaptive_ifs = TRUE;
 969
 970         /* Copper options */
 971
 972         if(hw->media_type == e1000_media_type_copper) {
 973                 hw->mdix = AUTO_ALL_MODES;
 974                 hw->disable_polarity_correction = FALSE;
 975                 hw->master_slave = E1000_MASTER_SLAVE;
 976         }
 977
 978 #ifdef CONFIG_E1000_MQ
 979         /* Number of supported queues */
 980         switch (hw->mac_type) {
 981         case e1000_82571:
 982         case e1000_82572:
 983                 /* These controllers support 2 tx queues, but with a single
 984                  * qdisc implementation, multiple tx queues aren't quite as
 985                  * interesting.  If we can find a logical way of mapping
 986                  * flows to a queue, then perhaps we can up the num_tx_queue
 987                  * count back to its default.  Until then, we run the risk of
 988                  * terrible performance due to SACK overload. */
 989                 adapter->num_tx_queues = 1;
 990                 adapter->num_rx_queues = 2;
 991                 break;
 992         default:
 993                 adapter->num_tx_queues = 1;
 994                 adapter->num_rx_queues = 1;
 995                 break;
 996         }
 997         adapter->num_rx_queues = min(adapter->num_rx_queues, num_online_cpus());
 998         adapter->num_tx_queues = min(adapter->num_tx_queues, num_online_cpus());
 999         DPRINTK(DRV, INFO, "Multiqueue Enabled: Rx Queue count = %u %s\n",
1000                 adapter->num_rx_queues,
1001                 ((adapter->num_rx_queues == 1)
1002                  ? ((num_online_cpus() > 1)
1003                         ? "(due to unsupported feature in current adapter)"
1004                         : "(due to unsupported system configuration)")
1005                  : ""));
1006         DPRINTK(DRV, INFO, "Multiqueue Enabled: Tx Queue count = %u\n",
1007                 adapter->num_tx_queues);
1008 #else
1009         adapter->num_tx_queues = 1;
1010         adapter->num_rx_queues = 1;
1011 #endif
1012
1013         if (e1000_alloc_queues(adapter)) {
1014                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1015                 return -ENOMEM;
1016         }
1017
1018 #ifdef CONFIG_E1000_NAPI
1019         for (i = 0; i < adapter->num_rx_queues; i++) {
1020                 adapter->polling_netdev[i].priv = adapter;
1021                 adapter->polling_netdev[i].poll = &e1000_clean;
1022                 adapter->polling_netdev[i].weight = 64;
1023                 dev_hold(&adapter->polling_netdev[i]);
1024                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1025         }
1026         spin_lock_init(&adapter->tx_queue_lock);
1027 #endif
1028
1029         atomic_set(&adapter->irq_sem, 1);
1030         spin_lock_init(&adapter->stats_lock);
1031
1032         return 0;
1033 }
1034
1035 /**
1036  * e1000_alloc_queues - Allocate memory for all rings
1037  * @adapter: board private structure to initialize
1038  *
1039  * We allocate one ring per queue at run-time since we don't know the
1040  * number of queues at compile-time.  The polling_netdev array is
1041  * intended for Multiqueue, but should work fine with a single queue.
1042  **/
1043
1044 static int __devinit
1045 e1000_alloc_queues(struct e1000_adapter *adapter)
1046 {
1047         int size;
1048
1049         size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
1050         adapter->tx_ring = kmalloc(size, GFP_KERNEL);
1051         if (!adapter->tx_ring)
1052                 return -ENOMEM;
1053         memset(adapter->tx_ring, 0, size);
1054
1055         size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
1056         adapter->rx_ring = kmalloc(size, GFP_KERNEL);
1057         if (!adapter->rx_ring) {
1058                 kfree(adapter->tx_ring);
1059                 return -ENOMEM;
1060         }
1061         memset(adapter->rx_ring, 0, size);
1062
1063 #ifdef CONFIG_E1000_NAPI
1064         size = sizeof(struct net_device) * adapter->num_rx_queues;
1065         adapter->polling_netdev = kmalloc(size, GFP_KERNEL);
1066         if (!adapter->polling_netdev) {
1067                 kfree(adapter->tx_ring);
1068                 kfree(adapter->rx_ring);
1069                 return -ENOMEM;
1070         }
1071         memset(adapter->polling_netdev, 0, size);
1072 #endif
1073
1074 #ifdef CONFIG_E1000_MQ
1075         adapter->rx_sched_call_data.func = e1000_rx_schedule;
1076         adapter->rx_sched_call_data.info = adapter->netdev;
1077
1078         adapter->cpu_netdev = alloc_percpu(struct net_device *);
1079         adapter->cpu_tx_ring = alloc_percpu(struct e1000_tx_ring *);
1080 #endif
1081
1082         return E1000_SUCCESS;
1083 }
1084
1085 #ifdef CONFIG_E1000_MQ
1086 static void __devinit
1087 e1000_setup_queue_mapping(struct e1000_adapter *adapter)
1088 {
1089         int i, cpu;
1090
1091         adapter->rx_sched_call_data.func = e1000_rx_schedule;
1092         adapter->rx_sched_call_data.info = adapter->netdev;
1093         cpus_clear(adapter->rx_sched_call_data.cpumask);
1094
1095         adapter->cpu_netdev = alloc_percpu(struct net_device *);
1096         adapter->cpu_tx_ring = alloc_percpu(struct e1000_tx_ring *);
1097
1098         lock_cpu_hotplug();
1099         i = 0;
1100         for_each_online_cpu(cpu) {
1101                 *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_tx_queues];
1102                 /* This is incomplete because we'd like to assign separate
1103                  * physical cpus to these netdev polling structures and
1104                  * avoid saturating a subset of cpus.
1105                  */
1106                 if (i < adapter->num_rx_queues) {
1107                         *per_cpu_ptr(adapter->cpu_netdev, cpu) = &adapter->polling_netdev[i];
1108                         adapter->rx_ring[i].cpu = cpu;
1109                         cpu_set(cpu, adapter->cpumask);
1110                 } else
1111                         *per_cpu_ptr(adapter->cpu_netdev, cpu) = NULL;
1112
1113                 i++;
1114         }
1115         unlock_cpu_hotplug();
1116 }
1117 #endif
1118
1119 /**
1120  * e1000_open - Called when a network interface is made active
1121  * @netdev: network interface device structure
1122  *
1123  * Returns 0 on success, negative value on failure
1124  *
1125  * The open entry point is called when a network interface is made
1126  * active by the system (IFF_UP).  At this point all resources needed
1127  * for transmit and receive operations are allocated, the interrupt
1128  * handler is registered with the OS, the watchdog timer is started,
1129  * and the stack is notified that the interface is ready.
1130  **/
1131
1132 static int
1133 e1000_open(struct net_device *netdev)
1134 {
1135         struct e1000_adapter *adapter = netdev_priv(netdev);
1136         int err;
1137
1138         /* allocate transmit descriptors */
1139
1140         if ((err = e1000_setup_all_tx_resources(adapter)))
1141                 goto err_setup_tx;
1142
1143         /* allocate receive descriptors */
1144
1145         if ((err = e1000_setup_all_rx_resources(adapter)))
1146                 goto err_setup_rx;
1147
1148         if((err = e1000_up(adapter)))
1149                 goto err_up;
1150         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1151         if((adapter->hw.mng_cookie.status &
1152                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1153                 e1000_update_mng_vlan(adapter);
1154         }
1155
1156         /* If AMT is enabled, let the firmware know that the network
1157          * interface is now open */
1158         if (adapter->hw.mac_type == e1000_82573 &&
1159             e1000_check_mng_mode(&adapter->hw))
1160                 e1000_get_hw_control(adapter);
1161
1162         return E1000_SUCCESS;
1163
1164 err_up:
1165         e1000_free_all_rx_resources(adapter);
1166 err_setup_rx:
1167         e1000_free_all_tx_resources(adapter);
1168 err_setup_tx:
1169         e1000_reset(adapter);
1170
1171         return err;
1172 }
1173
1174 /**
1175  * e1000_close - Disables a network interface
1176  * @netdev: network interface device structure
1177  *
1178  * Returns 0, this is not allowed to fail
1179  *
1180  * The close entry point is called when an interface is de-activated
1181  * by the OS.  The hardware is still under the drivers control, but
1182  * needs to be disabled.  A global MAC reset is issued to stop the
1183  * hardware, and all transmit and receive resources are freed.
1184  **/
1185
1186 static int
1187 e1000_close(struct net_device *netdev)
1188 {
1189         struct e1000_adapter *adapter = netdev_priv(netdev);
1190
1191         e1000_down(adapter);
1192
1193         e1000_free_all_tx_resources(adapter);
1194         e1000_free_all_rx_resources(adapter);
1195
1196         if((adapter->hw.mng_cookie.status &
1197                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1198                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1199         }
1200
1201         /* If AMT is enabled, let the firmware know that the network
1202          * interface is now closed */
1203         if (adapter->hw.mac_type == e1000_82573 &&
1204             e1000_check_mng_mode(&adapter->hw))
1205                 e1000_release_hw_control(adapter);
1206
1207         return 0;
1208 }
1209
1210 /**
1211  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1212  * @adapter: address of board private structure
1213  * @start: address of beginning of memory
1214  * @len: length of memory
1215  **/
1216 static inline boolean_t
1217 e1000_check_64k_bound(struct e1000_adapter *adapter,
1218                       void *start, unsigned long len)
1219 {
1220         unsigned long begin = (unsigned long) start;
1221         unsigned long end = begin + len;
1222
1223         /* First rev 82545 and 82546 need to not allow any memory
1224          * write location to cross 64k boundary due to errata 23 */
1225         if (adapter->hw.mac_type == e1000_82545 ||
1226             adapter->hw.mac_type == e1000_82546) {
1227                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1228         }
1229
1230         return TRUE;
1231 }
1232
1233 /**
1234  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1235  * @adapter: board private structure
1236  * @txdr:    tx descriptor ring (for a specific queue) to setup
1237  *
1238  * Return 0 on success, negative on failure
1239  **/
1240
1241 static int
1242 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1243                          struct e1000_tx_ring *txdr)
1244 {
1245         struct pci_dev *pdev = adapter->pdev;
1246         int size;
1247
1248         size = sizeof(struct e1000_buffer) * txdr->count;
1249
1250         txdr->buffer_info = vmalloc_node(size, pcibus_to_node(pdev->bus));
1251         if(!txdr->buffer_info) {
1252                 DPRINTK(PROBE, ERR,
1253                 "Unable to allocate memory for the transmit descriptor ring\n");
1254                 return -ENOMEM;
1255         }
1256         memset(txdr->buffer_info, 0, size);
1257
1258         /* round up to nearest 4K */
1259
1260         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1261         E1000_ROUNDUP(txdr->size, 4096);
1262
1263         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1264         if(!txdr->desc) {
1265 setup_tx_desc_die:
1266                 vfree(txdr->buffer_info);
1267                 DPRINTK(PROBE, ERR,
1268                 "Unable to allocate memory for the transmit descriptor ring\n");
1269                 return -ENOMEM;
1270         }
1271
1272         /* Fix for errata 23, can't cross 64kB boundary */
1273         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1274                 void *olddesc = txdr->desc;
1275                 dma_addr_t olddma = txdr->dma;
1276                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1277                                      "at %p\n", txdr->size, txdr->desc);
1278                 /* Try again, without freeing the previous */
1279                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1280                 if(!txdr->desc) {
1281                 /* Failed allocation, critical failure */
1282                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1283                         goto setup_tx_desc_die;
1284                 }
1285
1286                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1287                         /* give up */
1288                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1289                                             txdr->dma);
1290                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1291                         DPRINTK(PROBE, ERR,
1292                                 "Unable to allocate aligned memory "
1293                                 "for the transmit descriptor ring\n");
1294                         vfree(txdr->buffer_info);
1295                         return -ENOMEM;
1296                 } else {
1297                         /* Free old allocation, new allocation was successful */
1298                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1299                 }
1300         }
1301         memset(txdr->desc, 0, txdr->size);
1302
1303         txdr->next_to_use = 0;
1304         txdr->next_to_clean = 0;
1305         spin_lock_init(&txdr->tx_lock);
1306
1307         return 0;
1308 }
1309
1310 /**
1311  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1312  *                                (Descriptors) for all queues
1313  * @adapter: board private structure
1314  *
1315  * If this function returns with an error, then it's possible one or
1316  * more of the rings is populated (while the rest are not).  It is the
1317  * callers duty to clean those orphaned rings.
1318  *
1319  * Return 0 on success, negative on failure
1320  **/
1321
1322 int
1323 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1324 {
1325         int i, err = 0;
1326
1327         for (i = 0; i < adapter->num_tx_queues; i++) {
1328                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1329                 if (err) {
1330                         DPRINTK(PROBE, ERR,
1331                                 "Allocation for Tx Queue %u failed\n", i);
1332                         break;
1333                 }
1334         }
1335
1336         return err;
1337 }
1338
1339 /**
1340  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1341  * @adapter: board private structure
1342  *
1343  * Configure the Tx unit of the MAC after a reset.
1344  **/
1345
1346 static void
1347 e1000_configure_tx(struct e1000_adapter *adapter)
1348 {
1349         uint64_t tdba;
1350         struct e1000_hw *hw = &adapter->hw;
1351         uint32_t tdlen, tctl, tipg, tarc;
1352         uint32_t ipgr1, ipgr2;
1353
1354         /* Setup the HW Tx Head and Tail descriptor pointers */
1355
1356         switch (adapter->num_tx_queues) {
1357         case 2:
1358                 tdba = adapter->tx_ring[1].dma;
1359                 tdlen = adapter->tx_ring[1].count *
1360                         sizeof(struct e1000_tx_desc);
1361                 E1000_WRITE_REG(hw, TDBAL1, (tdba & 0x00000000ffffffffULL));
1362                 E1000_WRITE_REG(hw, TDBAH1, (tdba >> 32));
1363                 E1000_WRITE_REG(hw, TDLEN1, tdlen);
1364                 E1000_WRITE_REG(hw, TDH1, 0);
1365                 E1000_WRITE_REG(hw, TDT1, 0);
1366                 adapter->tx_ring[1].tdh = E1000_TDH1;
1367                 adapter->tx_ring[1].tdt = E1000_TDT1;
1368                 /* Fall Through */
1369         case 1:
1370         default:
1371                 tdba = adapter->tx_ring[0].dma;
1372                 tdlen = adapter->tx_ring[0].count *
1373                         sizeof(struct e1000_tx_desc);
1374                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1375                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1376                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1377                 E1000_WRITE_REG(hw, TDH, 0);
1378                 E1000_WRITE_REG(hw, TDT, 0);
1379                 adapter->tx_ring[0].tdh = E1000_TDH;
1380                 adapter->tx_ring[0].tdt = E1000_TDT;
1381                 break;
1382         }
1383
1384         /* Set the default values for the Tx Inter Packet Gap timer */
1385
1386         if (hw->media_type == e1000_media_type_fiber ||
1387             hw->media_type == e1000_media_type_internal_serdes)
1388                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1389         else
1390                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1391
1392         switch (hw->mac_type) {
1393         case e1000_82542_rev2_0:
1394         case e1000_82542_rev2_1:
1395                 tipg = DEFAULT_82542_TIPG_IPGT;
1396                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1397                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1398                 break;
1399         default:
1400                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1401                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1402                 break;
1403         }
1404         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1405         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1406         E1000_WRITE_REG(hw, TIPG, tipg);
1407
1408         /* Set the Tx Interrupt Delay register */
1409
1410         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1411         if (hw->mac_type >= e1000_82540)
1412                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1413
1414         /* Program the Transmit Control Register */
1415
1416         tctl = E1000_READ_REG(hw, TCTL);
1417
1418         tctl &= ~E1000_TCTL_CT;
1419         tctl |= E1000_TCTL_EN | E1000_TCTL_PSP | E1000_TCTL_RTLC |
1420                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1421
1422         E1000_WRITE_REG(hw, TCTL, tctl);
1423
1424         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1425                 tarc = E1000_READ_REG(hw, TARC0);
1426                 tarc |= ((1 << 25) | (1 << 21));
1427                 E1000_WRITE_REG(hw, TARC0, tarc);
1428                 tarc = E1000_READ_REG(hw, TARC1);
1429                 tarc |= (1 << 25);
1430                 if (tctl & E1000_TCTL_MULR)
1431                         tarc &= ~(1 << 28);
1432                 else
1433                         tarc |= (1 << 28);
1434                 E1000_WRITE_REG(hw, TARC1, tarc);
1435         }
1436
1437         e1000_config_collision_dist(hw);
1438
1439         /* Setup Transmit Descriptor Settings for eop descriptor */
1440         adapter->txd_cmd = E1000_TXD_CMD_IDE | E1000_TXD_CMD_EOP |
1441                 E1000_TXD_CMD_IFCS;
1442
1443         if (hw->mac_type < e1000_82543)
1444                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1445         else
1446                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1447
1448         /* Cache if we're 82544 running in PCI-X because we'll
1449          * need this to apply a workaround later in the send path. */
1450         if (hw->mac_type == e1000_82544 &&
1451             hw->bus_type == e1000_bus_type_pcix)
1452                 adapter->pcix_82544 = 1;
1453 }
1454
1455 /**
1456  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1457  * @adapter: board private structure
1458  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1459  *
1460  * Returns 0 on success, negative on failure
1461  **/
1462
1463 static int
1464 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1465                          struct e1000_rx_ring *rxdr)
1466 {
1467         struct pci_dev *pdev = adapter->pdev;
1468         int size, desc_len;
1469
1470         size = sizeof(struct e1000_buffer) * rxdr->count;
1471         rxdr->buffer_info = vmalloc_node(size, pcibus_to_node(pdev->bus));
1472         if (!rxdr->buffer_info) {
1473                 DPRINTK(PROBE, ERR,
1474                 "Unable to allocate memory for the receive descriptor ring\n");
1475                 return -ENOMEM;
1476         }
1477         memset(rxdr->buffer_info, 0, size);
1478
1479         size = sizeof(struct e1000_ps_page) * rxdr->count;
1480         rxdr->ps_page = kmalloc(size, GFP_KERNEL);
1481         if(!rxdr->ps_page) {
1482                 vfree(rxdr->buffer_info);
1483                 DPRINTK(PROBE, ERR,
1484                 "Unable to allocate memory for the receive descriptor ring\n");
1485                 return -ENOMEM;
1486         }
1487         memset(rxdr->ps_page, 0, size);
1488
1489         size = sizeof(struct e1000_ps_page_dma) * rxdr->count;
1490         rxdr->ps_page_dma = kmalloc(size, GFP_KERNEL);
1491         if(!rxdr->ps_page_dma) {
1492                 vfree(rxdr->buffer_info);
1493                 kfree(rxdr->ps_page);
1494                 DPRINTK(PROBE, ERR,
1495                 "Unable to allocate memory for the receive descriptor ring\n");
1496                 return -ENOMEM;
1497         }
1498         memset(rxdr->ps_page_dma, 0, size);
1499
1500         if(adapter->hw.mac_type <= e1000_82547_rev_2)
1501                 desc_len = sizeof(struct e1000_rx_desc);
1502         else
1503                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1504
1505         /* Round up to nearest 4K */
1506
1507         rxdr->size = rxdr->count * desc_len;
1508         E1000_ROUNDUP(rxdr->size, 4096);
1509
1510         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1511
1512         if (!rxdr->desc) {
1513                 DPRINTK(PROBE, ERR,
1514                 "Unable to allocate memory for the receive descriptor ring\n");
1515 setup_rx_desc_die:
1516                 vfree(rxdr->buffer_info);
1517                 kfree(rxdr->ps_page);
1518                 kfree(rxdr->ps_page_dma);
1519                 return -ENOMEM;
1520         }
1521
1522         /* Fix for errata 23, can't cross 64kB boundary */
1523         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1524                 void *olddesc = rxdr->desc;
1525                 dma_addr_t olddma = rxdr->dma;
1526                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1527                                      "at %p\n", rxdr->size, rxdr->desc);
1528                 /* Try again, without freeing the previous */
1529                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1530                 /* Failed allocation, critical failure */
1531                 if (!rxdr->desc) {
1532                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1533                         DPRINTK(PROBE, ERR,
1534                                 "Unable to allocate memory "
1535                                 "for the receive descriptor ring\n");
1536                         goto setup_rx_desc_die;
1537                 }
1538
1539                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1540                         /* give up */
1541                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1542                                             rxdr->dma);
1543                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1544                         DPRINTK(PROBE, ERR,
1545                                 "Unable to allocate aligned memory "
1546                                 "for the receive descriptor ring\n");
1547                         goto setup_rx_desc_die;
1548                 } else {
1549                         /* Free old allocation, new allocation was successful */
1550                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1551                 }
1552         }
1553         memset(rxdr->desc, 0, rxdr->size);
1554
1555         rxdr->next_to_clean = 0;
1556         rxdr->next_to_use = 0;
1557         rxdr->rx_skb_top = NULL;
1558         rxdr->rx_skb_prev = NULL;
1559
1560         return 0;
1561 }
1562
1563 /**
1564  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1565  *                                (Descriptors) for all queues
1566  * @adapter: board private structure
1567  *
1568  * If this function returns with an error, then it's possible one or
1569  * more of the rings is populated (while the rest are not).  It is the
1570  * callers duty to clean those orphaned rings.
1571  *
1572  * Return 0 on success, negative on failure
1573  **/
1574
1575 int
1576 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1577 {
1578         int i, err = 0;
1579
1580         for (i = 0; i < adapter->num_rx_queues; i++) {
1581                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1582                 if (err) {
1583                         DPRINTK(PROBE, ERR,
1584                                 "Allocation for Rx Queue %u failed\n", i);
1585                         break;
1586                 }
1587         }
1588
1589         return err;
1590 }
1591
1592 /**
1593  * e1000_setup_rctl - configure the receive control registers
1594  * @adapter: Board private structure
1595  **/
1596 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1597                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1598 static void
1599 e1000_setup_rctl(struct e1000_adapter *adapter)
1600 {
1601         uint32_t rctl, rfctl;
1602         uint32_t psrctl = 0;
1603 #ifdef CONFIG_E1000_PACKET_SPLIT
1604         uint32_t pages = 0;
1605 #endif
1606
1607         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1608
1609         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1610
1611         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1612                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1613                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1614
1615         if (adapter->hw.mac_type > e1000_82543)
1616                 rctl |= E1000_RCTL_SECRC;
1617
1618         if (adapter->hw.tbi_compatibility_on == 1)
1619                 rctl |= E1000_RCTL_SBP;
1620         else
1621                 rctl &= ~E1000_RCTL_SBP;
1622
1623         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1624                 rctl &= ~E1000_RCTL_LPE;
1625         else
1626                 rctl |= E1000_RCTL_LPE;
1627
1628         /* Setup buffer sizes */
1629         if(adapter->hw.mac_type >= e1000_82571) {
1630                 /* We can now specify buffers in 1K increments.
1631                  * BSIZE and BSEX are ignored in this case. */
1632                 rctl |= adapter->rx_buffer_len << 0x11;
1633         } else {
1634                 rctl &= ~E1000_RCTL_SZ_4096;
1635                 rctl |= E1000_RCTL_BSEX;
1636                 switch (adapter->rx_buffer_len) {
1637                 case E1000_RXBUFFER_2048:
1638                 default:
1639                         rctl |= E1000_RCTL_SZ_2048;
1640                         rctl &= ~E1000_RCTL_BSEX;
1641                         break;
1642                 case E1000_RXBUFFER_4096:
1643                         rctl |= E1000_RCTL_SZ_4096;
1644                         break;
1645                 case E1000_RXBUFFER_8192:
1646                         rctl |= E1000_RCTL_SZ_8192;
1647                         break;
1648                 case E1000_RXBUFFER_16384:
1649                         rctl |= E1000_RCTL_SZ_16384;
1650                         break;
1651                 }
1652         }
1653
1654 #ifdef CONFIG_E1000_PACKET_SPLIT
1655         /* 82571 and greater support packet-split where the protocol
1656          * header is placed in skb->data and the packet data is
1657          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1658          * In the case of a non-split, skb->data is linearly filled,
1659          * followed by the page buffers.  Therefore, skb->data is
1660          * sized to hold the largest protocol header.
1661          */
1662         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1663         if ((adapter->hw.mac_type > e1000_82547_rev_2) && (pages <= 3) &&
1664             PAGE_SIZE <= 16384)
1665                 adapter->rx_ps_pages = pages;
1666         else
1667                 adapter->rx_ps_pages = 0;
1668 #endif
1669         if (adapter->rx_ps_pages) {
1670                 /* Configure extra packet-split registers */
1671                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1672                 rfctl |= E1000_RFCTL_EXTEN;
1673                 /* disable IPv6 packet split support */
1674                 rfctl |= E1000_RFCTL_IPV6_DIS;
1675                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1676
1677                 rctl |= E1000_RCTL_DTYP_PS | E1000_RCTL_SECRC;
1678
1679                 psrctl |= adapter->rx_ps_bsize0 >>
1680                         E1000_PSRCTL_BSIZE0_SHIFT;
1681
1682                 switch (adapter->rx_ps_pages) {
1683                 case 3:
1684                         psrctl |= PAGE_SIZE <<
1685                                 E1000_PSRCTL_BSIZE3_SHIFT;
1686                 case 2:
1687                         psrctl |= PAGE_SIZE <<
1688                                 E1000_PSRCTL_BSIZE2_SHIFT;
1689                 case 1:
1690                         psrctl |= PAGE_SIZE >>
1691                                 E1000_PSRCTL_BSIZE1_SHIFT;
1692                         break;
1693                 }
1694
1695                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1696         }
1697
1698         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1699 }
1700
1701 /**
1702  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1703  * @adapter: board private structure
1704  *
1705  * Configure the Rx unit of the MAC after a reset.
1706  **/
1707
1708 static void
1709 e1000_configure_rx(struct e1000_adapter *adapter)
1710 {
1711         uint64_t rdba;
1712         struct e1000_hw *hw = &adapter->hw;
1713         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
1714 #ifdef CONFIG_E1000_MQ
1715         uint32_t reta, mrqc;
1716         int i;
1717 #endif
1718
1719         if (adapter->rx_ps_pages) {
1720                 rdlen = adapter->rx_ring[0].count *
1721                         sizeof(union e1000_rx_desc_packet_split);
1722                 adapter->clean_rx = e1000_clean_rx_irq_ps;
1723                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
1724         } else {
1725                 rdlen = adapter->rx_ring[0].count *
1726                         sizeof(struct e1000_rx_desc);
1727                 adapter->clean_rx = e1000_clean_rx_irq;
1728                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1729         }
1730
1731         /* disable receives while setting up the descriptors */
1732         rctl = E1000_READ_REG(hw, RCTL);
1733         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
1734
1735         /* set the Receive Delay Timer Register */
1736         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
1737
1738         if (hw->mac_type >= e1000_82540) {
1739                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
1740                 if(adapter->itr > 1)
1741                         E1000_WRITE_REG(hw, ITR,
1742                                 1000000000 / (adapter->itr * 256));
1743         }
1744
1745         if (hw->mac_type >= e1000_82571) {
1746                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
1747                 /* Reset delay timers after every interrupt */
1748                 ctrl_ext |= E1000_CTRL_EXT_CANC;
1749 #ifdef CONFIG_E1000_NAPI
1750                 /* Auto-Mask interrupts upon ICR read. */
1751                 ctrl_ext |= E1000_CTRL_EXT_IAME;
1752 #endif
1753                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
1754                 E1000_WRITE_REG(hw, IAM, ~0);
1755                 E1000_WRITE_FLUSH(hw);
1756         }
1757
1758         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1759          * the Base and Length of the Rx Descriptor Ring */
1760         switch (adapter->num_rx_queues) {
1761 #ifdef CONFIG_E1000_MQ
1762         case 2:
1763                 rdba = adapter->rx_ring[1].dma;
1764                 E1000_WRITE_REG(hw, RDBAL1, (rdba & 0x00000000ffffffffULL));
1765                 E1000_WRITE_REG(hw, RDBAH1, (rdba >> 32));
1766                 E1000_WRITE_REG(hw, RDLEN1, rdlen);
1767                 E1000_WRITE_REG(hw, RDH1, 0);
1768                 E1000_WRITE_REG(hw, RDT1, 0);
1769                 adapter->rx_ring[1].rdh = E1000_RDH1;
1770                 adapter->rx_ring[1].rdt = E1000_RDT1;
1771                 /* Fall Through */
1772 #endif
1773         case 1:
1774         default:
1775                 rdba = adapter->rx_ring[0].dma;
1776                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
1777                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
1778                 E1000_WRITE_REG(hw, RDLEN, rdlen);
1779                 E1000_WRITE_REG(hw, RDH, 0);
1780                 E1000_WRITE_REG(hw, RDT, 0);
1781                 adapter->rx_ring[0].rdh = E1000_RDH;
1782                 adapter->rx_ring[0].rdt = E1000_RDT;
1783                 break;
1784         }
1785
1786 #ifdef CONFIG_E1000_MQ
1787         if (adapter->num_rx_queues > 1) {
1788                 uint32_t random[10];
1789
1790                 get_random_bytes(&random[0], 40);
1791
1792                 if (hw->mac_type <= e1000_82572) {
1793                         E1000_WRITE_REG(hw, RSSIR, 0);
1794                         E1000_WRITE_REG(hw, RSSIM, 0);
1795                 }
1796
1797                 switch (adapter->num_rx_queues) {
1798                 case 2:
1799                 default:
1800                         reta = 0x00800080;
1801                         mrqc = E1000_MRQC_ENABLE_RSS_2Q;
1802                         break;
1803                 }
1804
1805                 /* Fill out redirection table */
1806                 for (i = 0; i < 32; i++)
1807                         E1000_WRITE_REG_ARRAY(hw, RETA, i, reta);
1808                 /* Fill out hash function seeds */
1809                 for (i = 0; i < 10; i++)
1810                         E1000_WRITE_REG_ARRAY(hw, RSSRK, i, random[i]);
1811
1812                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1813                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1814                 E1000_WRITE_REG(hw, MRQC, mrqc);
1815         }
1816
1817         /* Multiqueue and packet checksumming are mutually exclusive. */
1818         if (hw->mac_type >= e1000_82571) {
1819                 rxcsum = E1000_READ_REG(hw, RXCSUM);
1820                 rxcsum |= E1000_RXCSUM_PCSD;
1821                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
1822         }
1823
1824 #else
1825
1826         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1827         if (hw->mac_type >= e1000_82543) {
1828                 rxcsum = E1000_READ_REG(hw, RXCSUM);
1829                 if(adapter->rx_csum == TRUE) {
1830                         rxcsum |= E1000_RXCSUM_TUOFL;
1831
1832                         /* Enable 82571 IPv4 payload checksum for UDP fragments
1833                          * Must be used in conjunction with packet-split. */
1834                         if ((hw->mac_type >= e1000_82571) &&
1835                            (adapter->rx_ps_pages)) {
1836                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1837                         }
1838                 } else {
1839                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1840                         /* don't need to clear IPPCSE as it defaults to 0 */
1841                 }
1842                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
1843         }
1844 #endif /* CONFIG_E1000_MQ */
1845
1846         if (hw->mac_type == e1000_82573)
1847                 E1000_WRITE_REG(hw, ERT, 0x0100);
1848
1849         /* Enable Receives */
1850         E1000_WRITE_REG(hw, RCTL, rctl);
1851 }
1852
1853 /**
1854  * e1000_free_tx_resources - Free Tx Resources per Queue
1855  * @adapter: board private structure
1856  * @tx_ring: Tx descriptor ring for a specific queue
1857  *
1858  * Free all transmit software resources
1859  **/
1860
1861 static void
1862 e1000_free_tx_resources(struct e1000_adapter *adapter,
1863                         struct e1000_tx_ring *tx_ring)
1864 {
1865         struct pci_dev *pdev = adapter->pdev;
1866
1867         e1000_clean_tx_ring(adapter, tx_ring);
1868
1869         vfree(tx_ring->buffer_info);
1870         tx_ring->buffer_info = NULL;
1871
1872         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1873
1874         tx_ring->desc = NULL;
1875 }
1876
1877 /**
1878  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1879  * @adapter: board private structure
1880  *
1881  * Free all transmit software resources
1882  **/
1883
1884 void
1885 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1886 {
1887         int i;
1888
1889         for (i = 0; i < adapter->num_tx_queues; i++)
1890                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1891 }
1892
1893 static inline void
1894 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1895                         struct e1000_buffer *buffer_info)
1896 {
1897         if(buffer_info->dma) {
1898                 pci_unmap_page(adapter->pdev,
1899                                 buffer_info->dma,
1900                                 buffer_info->length,
1901                                 PCI_DMA_TODEVICE);
1902                 buffer_info->dma = 0;
1903         }
1904         if(buffer_info->skb) {
1905                 dev_kfree_skb_any(buffer_info->skb);
1906                 buffer_info->skb = NULL;
1907         }
1908 }
1909
1910 /**
1911  * e1000_clean_tx_ring - Free Tx Buffers
1912  * @adapter: board private structure
1913  * @tx_ring: ring to be cleaned
1914  **/
1915
1916 static void
1917 e1000_clean_tx_ring(struct e1000_adapter *adapter,
1918                     struct e1000_tx_ring *tx_ring)
1919 {
1920         struct e1000_buffer *buffer_info;
1921         unsigned long size;
1922         unsigned int i;
1923
1924         /* Free all the Tx ring sk_buffs */
1925
1926         for(i = 0; i < tx_ring->count; i++) {
1927                 buffer_info = &tx_ring->buffer_info[i];
1928                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
1929         }
1930
1931         size = sizeof(struct e1000_buffer) * tx_ring->count;
1932         memset(tx_ring->buffer_info, 0, size);
1933
1934         /* Zero out the descriptor ring */
1935
1936         memset(tx_ring->desc, 0, tx_ring->size);
1937
1938         tx_ring->next_to_use = 0;
1939         tx_ring->next_to_clean = 0;
1940         tx_ring->last_tx_tso = 0;
1941
1942         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
1943         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
1944 }
1945
1946 /**
1947  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
1948  * @adapter: board private structure
1949  **/
1950
1951 static void
1952 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
1953 {
1954         int i;
1955
1956         for (i = 0; i < adapter->num_tx_queues; i++)
1957                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1958 }
1959
1960 /**
1961  * e1000_free_rx_resources - Free Rx Resources
1962  * @adapter: board private structure
1963  * @rx_ring: ring to clean the resources from
1964  *
1965  * Free all receive software resources
1966  **/
1967
1968 static void
1969 e1000_free_rx_resources(struct e1000_adapter *adapter,
1970                         struct e1000_rx_ring *rx_ring)
1971 {
1972         struct pci_dev *pdev = adapter->pdev;
1973
1974         e1000_clean_rx_ring(adapter, rx_ring);
1975
1976         vfree(rx_ring->buffer_info);
1977         rx_ring->buffer_info = NULL;
1978         kfree(rx_ring->ps_page);
1979         rx_ring->ps_page = NULL;
1980         kfree(rx_ring->ps_page_dma);
1981         rx_ring->ps_page_dma = NULL;
1982
1983         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
1984
1985         rx_ring->desc = NULL;
1986 }
1987
1988 /**
1989  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
1990  * @adapter: board private structure
1991  *
1992  * Free all receive software resources
1993  **/
1994
1995 void
1996 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
1997 {
1998         int i;
1999
2000         for (i = 0; i < adapter->num_rx_queues; i++)
2001                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2002 }
2003
2004 /**
2005  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2006  * @adapter: board private structure
2007  * @rx_ring: ring to free buffers from
2008  **/
2009
2010 static void
2011 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2012                     struct e1000_rx_ring *rx_ring)
2013 {
2014         struct e1000_buffer *buffer_info;
2015         struct e1000_ps_page *ps_page;
2016         struct e1000_ps_page_dma *ps_page_dma;
2017         struct pci_dev *pdev = adapter->pdev;
2018         unsigned long size;
2019         unsigned int i, j;
2020
2021         /* Free all the Rx ring sk_buffs */
2022
2023         for(i = 0; i < rx_ring->count; i++) {
2024                 buffer_info = &rx_ring->buffer_info[i];
2025                 if(buffer_info->skb) {
2026                         ps_page = &rx_ring->ps_page[i];
2027                         ps_page_dma = &rx_ring->ps_page_dma[i];
2028                         pci_unmap_single(pdev,
2029                                          buffer_info->dma,
2030                                          buffer_info->length,
2031                                          PCI_DMA_FROMDEVICE);
2032
2033                         dev_kfree_skb(buffer_info->skb);
2034                         buffer_info->skb = NULL;
2035                 }
2036                 ps_page = &rx_ring->ps_page[i];
2037                 ps_page_dma = &rx_ring->ps_page_dma[i];
2038                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2039                         if (!ps_page->ps_page[j]) break;
2040                         pci_unmap_page(pdev,
2041                                        ps_page_dma->ps_page_dma[j],
2042                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2043                         ps_page_dma->ps_page_dma[j] = 0;
2044                         put_page(ps_page->ps_page[j]);
2045                         ps_page->ps_page[j] = NULL;
2046                 }
2047         }
2048
2049         /* there also may be some cached data in our adapter */
2050         if (rx_ring->rx_skb_top) {
2051                 dev_kfree_skb(rx_ring->rx_skb_top);
2052
2053                 /* rx_skb_prev will be wiped out by rx_skb_top */
2054                 rx_ring->rx_skb_top = NULL;
2055                 rx_ring->rx_skb_prev = NULL;
2056         }
2057
2058
2059         size = sizeof(struct e1000_buffer) * rx_ring->count;
2060         memset(rx_ring->buffer_info, 0, size);
2061         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2062         memset(rx_ring->ps_page, 0, size);
2063         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2064         memset(rx_ring->ps_page_dma, 0, size);
2065
2066         /* Zero out the descriptor ring */
2067
2068         memset(rx_ring->desc, 0, rx_ring->size);
2069
2070         rx_ring->next_to_clean = 0;
2071         rx_ring->next_to_use = 0;
2072
2073         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2074         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2075 }
2076
2077 /**
2078  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2079  * @adapter: board private structure
2080  **/
2081
2082 static void
2083 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2084 {
2085         int i;
2086
2087         for (i = 0; i < adapter->num_rx_queues; i++)
2088                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2089 }
2090
2091 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2092  * and memory write and invalidate disabled for certain operations
2093  */
2094 static void
2095 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2096 {
2097         struct net_device *netdev = adapter->netdev;
2098         uint32_t rctl;
2099
2100         e1000_pci_clear_mwi(&adapter->hw);
2101
2102         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2103         rctl |= E1000_RCTL_RST;
2104         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2105         E1000_WRITE_FLUSH(&adapter->hw);
2106         mdelay(5);
2107
2108         if(netif_running(netdev))
2109                 e1000_clean_all_rx_rings(adapter);
2110 }
2111
2112 static void
2113 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2114 {
2115         struct net_device *netdev = adapter->netdev;
2116         uint32_t rctl;
2117
2118         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2119         rctl &= ~E1000_RCTL_RST;
2120         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2121         E1000_WRITE_FLUSH(&adapter->hw);
2122         mdelay(5);
2123
2124         if(adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2125                 e1000_pci_set_mwi(&adapter->hw);
2126
2127         if(netif_running(netdev)) {
2128                 e1000_configure_rx(adapter);
2129                 /* No need to loop, because 82542 supports only 1 queue */
2130                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2131                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2132         }
2133 }
2134
2135 /**
2136  * e1000_set_mac - Change the Ethernet Address of the NIC
2137  * @netdev: network interface device structure
2138  * @p: pointer to an address structure
2139  *
2140  * Returns 0 on success, negative on failure
2141  **/
2142
2143 static int
2144 e1000_set_mac(struct net_device *netdev, void *p)
2145 {
2146         struct e1000_adapter *adapter = netdev_priv(netdev);
2147         struct sockaddr *addr = p;
2148
2149         if(!is_valid_ether_addr(addr->sa_data))
2150                 return -EADDRNOTAVAIL;
2151
2152         /* 82542 2.0 needs to be in reset to write receive address registers */
2153
2154         if(adapter->hw.mac_type == e1000_82542_rev2_0)
2155                 e1000_enter_82542_rst(adapter);
2156
2157         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2158         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2159
2160         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2161
2162         /* With 82571 controllers, LAA may be overwritten (with the default)
2163          * due to controller reset from the other port. */
2164         if (adapter->hw.mac_type == e1000_82571) {
2165                 /* activate the work around */
2166                 adapter->hw.laa_is_present = 1;
2167
2168                 /* Hold a copy of the LAA in RAR[14] This is done so that
2169                  * between the time RAR[0] gets clobbered  and the time it
2170                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2171                  * of the RARs and no incoming packets directed to this port
2172                  * are dropped. Eventaully the LAA will be in RAR[0] and
2173                  * RAR[14] */
2174                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2175                                         E1000_RAR_ENTRIES - 1);
2176         }
2177
2178         if(adapter->hw.mac_type == e1000_82542_rev2_0)
2179                 e1000_leave_82542_rst(adapter);
2180
2181         return 0;
2182 }
2183
2184 /**
2185  * e1000_set_multi - Multicast and Promiscuous mode set
2186  * @netdev: network interface device structure
2187  *
2188  * The set_multi entry point is called whenever the multicast address
2189  * list or the network interface flags are updated.  This routine is
2190  * responsible for configuring the hardware for proper multicast,
2191  * promiscuous mode, and all-multi behavior.
2192  **/
2193
2194 static void
2195 e1000_set_multi(struct net_device *netdev)
2196 {
2197         struct e1000_adapter *adapter = netdev_priv(netdev);
2198         struct e1000_hw *hw = &adapter->hw;
2199         struct dev_mc_list *mc_ptr;
2200         uint32_t rctl;
2201         uint32_t hash_value;
2202         int i, rar_entries = E1000_RAR_ENTRIES;
2203
2204         /* reserve RAR[14] for LAA over-write work-around */
2205         if (adapter->hw.mac_type == e1000_82571)
2206                 rar_entries--;
2207
2208         /* Check for Promiscuous and All Multicast modes */
2209
2210         rctl = E1000_READ_REG(hw, RCTL);
2211
2212         if(netdev->flags & IFF_PROMISC) {
2213                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2214         } else if(netdev->flags & IFF_ALLMULTI) {
2215                 rctl |= E1000_RCTL_MPE;
2216                 rctl &= ~E1000_RCTL_UPE;
2217         } else {
2218                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2219         }
2220
2221         E1000_WRITE_REG(hw, RCTL, rctl);
2222
2223         /* 82542 2.0 needs to be in reset to write receive address registers */
2224
2225         if(hw->mac_type == e1000_82542_rev2_0)
2226                 e1000_enter_82542_rst(adapter);
2227
2228         /* load the first 14 multicast address into the exact filters 1-14
2229          * RAR 0 is used for the station MAC adddress
2230          * if there are not 14 addresses, go ahead and clear the filters
2231          * -- with 82571 controllers only 0-13 entries are filled here
2232          */
2233         mc_ptr = netdev->mc_list;
2234
2235         for(i = 1; i < rar_entries; i++) {
2236                 if (mc_ptr) {
2237                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2238                         mc_ptr = mc_ptr->next;
2239                 } else {
2240                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2241                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2242                 }
2243         }
2244
2245         /* clear the old settings from the multicast hash table */
2246
2247         for(i = 0; i < E1000_NUM_MTA_REGISTERS; i++)
2248                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2249
2250         /* load any remaining addresses into the hash table */
2251
2252         for(; mc_ptr; mc_ptr = mc_ptr->next) {
2253                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2254                 e1000_mta_set(hw, hash_value);
2255         }
2256
2257         if(hw->mac_type == e1000_82542_rev2_0)
2258                 e1000_leave_82542_rst(adapter);
2259 }
2260
2261 /* Need to wait a few seconds after link up to get diagnostic information from
2262  * the phy */
2263
2264 static void
2265 e1000_update_phy_info(unsigned long data)
2266 {
2267         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2268         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2269 }
2270
2271 /**
2272  * e1000_82547_tx_fifo_stall - Timer Call-back
2273  * @data: pointer to adapter cast into an unsigned long
2274  **/
2275
2276 static void
2277 e1000_82547_tx_fifo_stall(unsigned long data)
2278 {
2279         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2280         struct net_device *netdev = adapter->netdev;
2281         uint32_t tctl;
2282
2283         if(atomic_read(&adapter->tx_fifo_stall)) {
2284                 if((E1000_READ_REG(&adapter->hw, TDT) ==
2285                     E1000_READ_REG(&adapter->hw, TDH)) &&
2286                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2287                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2288                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2289                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2290                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2291                         E1000_WRITE_REG(&adapter->hw, TCTL,
2292                                         tctl & ~E1000_TCTL_EN);
2293                         E1000_WRITE_REG(&adapter->hw, TDFT,
2294                                         adapter->tx_head_addr);
2295                         E1000_WRITE_REG(&adapter->hw, TDFH,
2296                                         adapter->tx_head_addr);
2297                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2298                                         adapter->tx_head_addr);
2299                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2300                                         adapter->tx_head_addr);
2301                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2302                         E1000_WRITE_FLUSH(&adapter->hw);
2303
2304                         adapter->tx_fifo_head = 0;
2305                         atomic_set(&adapter->tx_fifo_stall, 0);
2306                         netif_wake_queue(netdev);
2307                 } else {
2308                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2309                 }
2310         }
2311 }
2312
2313 /**
2314  * e1000_watchdog - Timer Call-back
2315  * @data: pointer to adapter cast into an unsigned long
2316  **/
2317 static void
2318 e1000_watchdog(unsigned long data)
2319 {
2320         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2321
2322         /* Do the rest outside of interrupt context */
2323         schedule_work(&adapter->watchdog_task);
2324 }
2325
2326 static void
2327 e1000_watchdog_task(struct e1000_adapter *adapter)
2328 {
2329         struct net_device *netdev = adapter->netdev;
2330         struct e1000_tx_ring *txdr = adapter->tx_ring;
2331         uint32_t link;
2332
2333         e1000_check_for_link(&adapter->hw);
2334         if (adapter->hw.mac_type == e1000_82573) {
2335                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2336                 if(adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2337                         e1000_update_mng_vlan(adapter);
2338         }
2339
2340         if((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2341            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2342                 link = !adapter->hw.serdes_link_down;
2343         else
2344                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2345
2346         if(link) {
2347                 if(!netif_carrier_ok(netdev)) {
2348                         e1000_get_speed_and_duplex(&adapter->hw,
2349                                                    &adapter->link_speed,
2350                                                    &adapter->link_duplex);
2351
2352                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s\n",
2353                                adapter->link_speed,
2354                                adapter->link_duplex == FULL_DUPLEX ?
2355                                "Full Duplex" : "Half Duplex");
2356
2357                         /* tweak tx_queue_len according to speed/duplex */
2358                         netdev->tx_queue_len = adapter->tx_queue_len;
2359                         adapter->tx_timeout_factor = 1;
2360                         if (adapter->link_duplex == HALF_DUPLEX) {
2361                                 switch (adapter->link_speed) {
2362                                 case SPEED_10:
2363                                         netdev->tx_queue_len = 10;
2364                                         adapter->tx_timeout_factor = 8;
2365                                         break;
2366                                 case SPEED_100:
2367                                         netdev->tx_queue_len = 100;
2368                                         break;
2369                                 }
2370                         }
2371
2372                         netif_carrier_on(netdev);
2373                         netif_wake_queue(netdev);
2374                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2375                         adapter->smartspeed = 0;
2376                 }
2377         } else {
2378                 if(netif_carrier_ok(netdev)) {
2379                         adapter->link_speed = 0;
2380                         adapter->link_duplex = 0;
2381                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2382                         netif_carrier_off(netdev);
2383                         netif_stop_queue(netdev);
2384                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2385                 }
2386
2387                 e1000_smartspeed(adapter);
2388         }
2389
2390         e1000_update_stats(adapter);
2391
2392         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2393         adapter->tpt_old = adapter->stats.tpt;
2394         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2395         adapter->colc_old = adapter->stats.colc;
2396
2397         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2398         adapter->gorcl_old = adapter->stats.gorcl;
2399         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2400         adapter->gotcl_old = adapter->stats.gotcl;
2401
2402         e1000_update_adaptive(&adapter->hw);
2403
2404 #ifdef CONFIG_E1000_MQ
2405         txdr = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
2406 #endif
2407         if (!netif_carrier_ok(netdev)) {
2408                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2409                         /* We've lost link, so the controller stops DMA,
2410                          * but we've got queued Tx work that's never going
2411                          * to get done, so reset controller to flush Tx.
2412                          * (Do the reset outside of interrupt context). */
2413                         schedule_work(&adapter->tx_timeout_task);
2414                 }
2415         }
2416
2417         /* Dynamic mode for Interrupt Throttle Rate (ITR) */
2418         if(adapter->hw.mac_type >= e1000_82540 && adapter->itr == 1) {
2419                 /* Symmetric Tx/Rx gets a reduced ITR=2000; Total
2420                  * asymmetrical Tx or Rx gets ITR=8000; everyone
2421                  * else is between 2000-8000. */
2422                 uint32_t goc = (adapter->gotcl + adapter->gorcl) / 10000;
2423                 uint32_t dif = (adapter->gotcl > adapter->gorcl ?
2424                         adapter->gotcl - adapter->gorcl :
2425                         adapter->gorcl - adapter->gotcl) / 10000;
2426                 uint32_t itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2427                 E1000_WRITE_REG(&adapter->hw, ITR, 1000000000 / (itr * 256));
2428         }
2429
2430         /* Cause software interrupt to ensure rx ring is cleaned */
2431         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2432
2433         /* Force detection of hung controller every watchdog period */
2434         adapter->detect_tx_hung = TRUE;
2435
2436         /* With 82571 controllers, LAA may be overwritten due to controller
2437          * reset from the other port. Set the appropriate LAA in RAR[0] */
2438         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2439                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2440
2441         /* Reset the timer */
2442         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
2443 }
2444
2445 #define E1000_TX_FLAGS_CSUM             0x00000001
2446 #define E1000_TX_FLAGS_VLAN             0x00000002
2447 #define E1000_TX_FLAGS_TSO              0x00000004
2448 #define E1000_TX_FLAGS_IPV4             0x00000008
2449 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2450 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2451
2452 static inline int
2453 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2454           struct sk_buff *skb)
2455 {
2456 #ifdef NETIF_F_TSO
2457         struct e1000_context_desc *context_desc;
2458         struct e1000_buffer *buffer_info;
2459         unsigned int i;
2460         uint32_t cmd_length = 0;
2461         uint16_t ipcse = 0, tucse, mss;
2462         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2463         int err;
2464
2465         if(skb_shinfo(skb)->tso_size) {
2466                 if (skb_header_cloned(skb)) {
2467                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2468                         if (err)
2469                                 return err;
2470                 }
2471
2472                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2473                 mss = skb_shinfo(skb)->tso_size;
2474                 if(skb->protocol == ntohs(ETH_P_IP)) {
2475                         skb->nh.iph->tot_len = 0;
2476                         skb->nh.iph->check = 0;
2477                         skb->h.th->check =
2478                                 ~csum_tcpudp_magic(skb->nh.iph->saddr,
2479                                                    skb->nh.iph->daddr,
2480                                                    0,
2481                                                    IPPROTO_TCP,
2482                                                    0);
2483                         cmd_length = E1000_TXD_CMD_IP;
2484                         ipcse = skb->h.raw - skb->data - 1;
2485 #ifdef NETIF_F_TSO_IPV6
2486                 } else if(skb->protocol == ntohs(ETH_P_IPV6)) {
2487                         skb->nh.ipv6h->payload_len = 0;
2488                         skb->h.th->check =
2489                                 ~csum_ipv6_magic(&skb->nh.ipv6h->saddr,
2490                                                  &skb->nh.ipv6h->daddr,
2491                                                  0,
2492                                                  IPPROTO_TCP,
2493                                                  0);
2494                         ipcse = 0;
2495 #endif
2496                 }
2497                 ipcss = skb->nh.raw - skb->data;
2498                 ipcso = (void *)&(skb->nh.iph->check) - (void *)skb->data;
2499                 tucss = skb->h.raw - skb->data;
2500                 tucso = (void *)&(skb->h.th->check) - (void *)skb->data;
2501                 tucse = 0;
2502
2503                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2504                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2505
2506                 i = tx_ring->next_to_use;
2507                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2508                 buffer_info = &tx_ring->buffer_info[i];
2509
2510                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2511                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2512                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2513                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2514                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2515                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2516                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2517                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2518                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2519
2520                 buffer_info->time_stamp = jiffies;
2521
2522                 if (++i == tx_ring->count) i = 0;
2523                 tx_ring->next_to_use = i;
2524
2525                 return 1;
2526         }
2527 #endif
2528
2529         return 0;
2530 }
2531
2532 static inline boolean_t
2533 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2534               struct sk_buff *skb)
2535 {
2536         struct e1000_context_desc *context_desc;
2537         struct e1000_buffer *buffer_info;
2538         unsigned int i;
2539         uint8_t css;
2540
2541         if(likely(skb->ip_summed == CHECKSUM_HW)) {
2542                 css = skb->h.raw - skb->data;
2543
2544                 i = tx_ring->next_to_use;
2545                 buffer_info = &tx_ring->buffer_info[i];
2546                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2547
2548                 context_desc->upper_setup.tcp_fields.tucss = css;
2549                 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum;
2550                 context_desc->upper_setup.tcp_fields.tucse = 0;
2551                 context_desc->tcp_seg_setup.data = 0;
2552                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2553
2554                 buffer_info->time_stamp = jiffies;
2555
2556                 if (unlikely(++i == tx_ring->count)) i = 0;
2557                 tx_ring->next_to_use = i;
2558
2559                 return TRUE;
2560         }
2561
2562         return FALSE;
2563 }
2564
2565 #define E1000_MAX_TXD_PWR       12
2566 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2567
2568 static inline int
2569 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2570              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2571              unsigned int nr_frags, unsigned int mss)
2572 {
2573         struct e1000_buffer *buffer_info;
2574         unsigned int len = skb->len;
2575         unsigned int offset = 0, size, count = 0, i;
2576         unsigned int f;
2577         len -= skb->data_len;
2578
2579         i = tx_ring->next_to_use;
2580
2581         while(len) {
2582                 buffer_info = &tx_ring->buffer_info[i];
2583                 size = min(len, max_per_txd);
2584 #ifdef NETIF_F_TSO
2585                 /* Workaround for Controller erratum --
2586                  * descriptor for non-tso packet in a linear SKB that follows a
2587                  * tso gets written back prematurely before the data is fully
2588                  * DMAd to the controller */
2589                 if (!skb->data_len && tx_ring->last_tx_tso &&
2590                                 !skb_shinfo(skb)->tso_size) {
2591                         tx_ring->last_tx_tso = 0;
2592                         size -= 4;
2593                 }
2594
2595                 /* Workaround for premature desc write-backs
2596                  * in TSO mode.  Append 4-byte sentinel desc */
2597                 if(unlikely(mss && !nr_frags && size == len && size > 8))
2598                         size -= 4;
2599 #endif
2600                 /* work-around for errata 10 and it applies
2601                  * to all controllers in PCI-X mode
2602                  * The fix is to make sure that the first descriptor of a
2603                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2604                  */
2605                 if(unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
2606                                 (size > 2015) && count == 0))
2607                         size = 2015;
2608
2609                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2610                  * terminating buffers within evenly-aligned dwords. */
2611                 if(unlikely(adapter->pcix_82544 &&
2612                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2613                    size > 4))
2614                         size -= 4;
2615
2616                 buffer_info->length = size;
2617                 buffer_info->dma =
2618                         pci_map_single(adapter->pdev,
2619                                 skb->data + offset,
2620                                 size,
2621                                 PCI_DMA_TODEVICE);
2622                 buffer_info->time_stamp = jiffies;
2623
2624                 len -= size;
2625                 offset += size;
2626                 count++;
2627                 if(unlikely(++i == tx_ring->count)) i = 0;
2628         }
2629
2630         for(f = 0; f < nr_frags; f++) {
2631                 struct skb_frag_struct *frag;
2632
2633                 frag = &skb_shinfo(skb)->frags[f];
2634                 len = frag->size;
2635                 offset = frag->page_offset;
2636
2637                 while(len) {
2638                         buffer_info = &tx_ring->buffer_info[i];
2639                         size = min(len, max_per_txd);
2640 #ifdef NETIF_F_TSO
2641                         /* Workaround for premature desc write-backs
2642                          * in TSO mode.  Append 4-byte sentinel desc */
2643                         if(unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2644                                 size -= 4;
2645 #endif
2646                         /* Workaround for potential 82544 hang in PCI-X.
2647                          * Avoid terminating buffers within evenly-aligned
2648                          * dwords. */
2649                         if(unlikely(adapter->pcix_82544 &&
2650                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
2651                            size > 4))
2652                                 size -= 4;
2653
2654                         buffer_info->length = size;
2655                         buffer_info->dma =
2656                                 pci_map_page(adapter->pdev,
2657                                         frag->page,
2658                                         offset,
2659                                         size,
2660                                         PCI_DMA_TODEVICE);
2661                         buffer_info->time_stamp = jiffies;
2662
2663                         len -= size;
2664                         offset += size;
2665                         count++;
2666                         if(unlikely(++i == tx_ring->count)) i = 0;
2667                 }
2668         }
2669
2670         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2671         tx_ring->buffer_info[i].skb = skb;
2672         tx_ring->buffer_info[first].next_to_watch = i;
2673
2674         return count;
2675 }
2676
2677 static inline void
2678 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2679                int tx_flags, int count)
2680 {
2681         struct e1000_tx_desc *tx_desc = NULL;
2682         struct e1000_buffer *buffer_info;
2683         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2684         unsigned int i;
2685
2686         if(likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2687                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2688                              E1000_TXD_CMD_TSE;
2689                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2690
2691                 if(likely(tx_flags & E1000_TX_FLAGS_IPV4))
2692                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2693         }
2694
2695         if(likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2696                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2697                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2698         }
2699
2700         if(unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2701                 txd_lower |= E1000_TXD_CMD_VLE;
2702                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2703         }
2704
2705         i = tx_ring->next_to_use;
2706
2707         while(count--) {
2708                 buffer_info = &tx_ring->buffer_info[i];
2709                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2710                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2711                 tx_desc->lower.data =
2712                         cpu_to_le32(txd_lower | buffer_info->length);
2713                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2714                 if(unlikely(++i == tx_ring->count)) i = 0;
2715         }
2716
2717         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2718
2719         /* Force memory writes to complete before letting h/w
2720          * know there are new descriptors to fetch.  (Only
2721          * applicable for weak-ordered memory model archs,
2722          * such as IA-64). */
2723         wmb();
2724
2725         tx_ring->next_to_use = i;
2726         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
2727 }
2728
2729 /**
2730  * 82547 workaround to avoid controller hang in half-duplex environment.
2731  * The workaround is to avoid queuing a large packet that would span
2732  * the internal Tx FIFO ring boundary by notifying the stack to resend
2733  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2734  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2735  * to the beginning of the Tx FIFO.
2736  **/
2737
2738 #define E1000_FIFO_HDR                  0x10
2739 #define E1000_82547_PAD_LEN             0x3E0
2740
2741 static inline int
2742 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
2743 {
2744         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2745         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
2746
2747         E1000_ROUNDUP(skb_fifo_len, E1000_FIFO_HDR);
2748
2749         if(adapter->link_duplex != HALF_DUPLEX)
2750                 goto no_fifo_stall_required;
2751
2752         if(atomic_read(&adapter->tx_fifo_stall))
2753                 return 1;
2754
2755         if(skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2756                 atomic_set(&adapter->tx_fifo_stall, 1);
2757                 return 1;
2758         }
2759
2760 no_fifo_stall_required:
2761         adapter->tx_fifo_head += skb_fifo_len;
2762         if(adapter->tx_fifo_head >= adapter->tx_fifo_size)
2763                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2764         return 0;
2765 }
2766
2767 #define MINIMUM_DHCP_PACKET_SIZE 282
2768 static inline int
2769 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
2770 {
2771         struct e1000_hw *hw =  &adapter->hw;
2772         uint16_t length, offset;
2773         if(vlan_tx_tag_present(skb)) {
2774                 if(!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
2775                         ( adapter->hw.mng_cookie.status &
2776                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
2777                         return 0;
2778         }
2779         if ((skb->len > MINIMUM_DHCP_PACKET_SIZE) && (!skb->protocol)) {
2780                 struct ethhdr *eth = (struct ethhdr *) skb->data;
2781                 if((htons(ETH_P_IP) == eth->h_proto)) {
2782                         const struct iphdr *ip =
2783                                 (struct iphdr *)((uint8_t *)skb->data+14);
2784                         if(IPPROTO_UDP == ip->protocol) {
2785                                 struct udphdr *udp =
2786                                         (struct udphdr *)((uint8_t *)ip +
2787                                                 (ip->ihl << 2));
2788                                 if(ntohs(udp->dest) == 67) {
2789                                         offset = (uint8_t *)udp + 8 - skb->data;
2790                                         length = skb->len - offset;
2791
2792                                         return e1000_mng_write_dhcp_info(hw,
2793                                                         (uint8_t *)udp + 8,
2794                                                         length);
2795                                 }
2796                         }
2797                 }
2798         }
2799         return 0;
2800 }
2801
2802 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2803 static int
2804 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2805 {
2806         struct e1000_adapter *adapter = netdev_priv(netdev);
2807         struct e1000_tx_ring *tx_ring;
2808         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2809         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2810         unsigned int tx_flags = 0;
2811         unsigned int len = skb->len;
2812         unsigned long flags;
2813         unsigned int nr_frags = 0;
2814         unsigned int mss = 0;
2815         int count = 0;
2816         int tso;
2817         unsigned int f;
2818         len -= skb->data_len;
2819
2820 #ifdef CONFIG_E1000_MQ
2821         tx_ring = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
2822 #else
2823         tx_ring = adapter->tx_ring;
2824 #endif
2825
2826         if (unlikely(skb->len <= 0)) {
2827                 dev_kfree_skb_any(skb);
2828                 return NETDEV_TX_OK;
2829         }
2830
2831 #ifdef NETIF_F_TSO
2832         mss = skb_shinfo(skb)->tso_size;
2833         /* The controller does a simple calculation to
2834          * make sure there is enough room in the FIFO before
2835          * initiating the DMA for each buffer.  The calc is:
2836          * 4 = ceil(buffer len/mss).  To make sure we don't
2837          * overrun the FIFO, adjust the max buffer len if mss
2838          * drops. */
2839         if(mss) {
2840                 uint8_t hdr_len;
2841                 max_per_txd = min(mss << 2, max_per_txd);
2842                 max_txd_pwr = fls(max_per_txd) - 1;
2843
2844         /* TSO Workaround for 82571/2 Controllers -- if skb->data
2845          * points to just header, pull a few bytes of payload from
2846          * frags into skb->data */
2847                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2848                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len)) &&
2849                         (adapter->hw.mac_type == e1000_82571 ||
2850                         adapter->hw.mac_type == e1000_82572)) {
2851                         unsigned int pull_size;
2852                         pull_size = min((unsigned int)4, skb->data_len);
2853                         if (!__pskb_pull_tail(skb, pull_size)) {
2854                                 printk(KERN_ERR "__pskb_pull_tail failed.\n");
2855                                 dev_kfree_skb_any(skb);
2856                                 return -EFAULT;
2857                         }
2858                         len = skb->len - skb->data_len;
2859                 }
2860         }
2861
2862         if((mss) || (skb->ip_summed == CHECKSUM_HW))
2863         /* reserve a descriptor for the offload context */
2864                 count++;
2865         count++;
2866 #else
2867         if(skb->ip_summed == CHECKSUM_HW)
2868                 count++;
2869 #endif
2870
2871 #ifdef NETIF_F_TSO
2872         /* Controller Erratum workaround */
2873         if (!skb->data_len && tx_ring->last_tx_tso &&
2874                 !skb_shinfo(skb)->tso_size)
2875                 count++;
2876 #endif
2877
2878         count += TXD_USE_COUNT(len, max_txd_pwr);
2879
2880         if(adapter->pcix_82544)
2881                 count++;
2882
2883         /* work-around for errata 10 and it applies to all controllers
2884          * in PCI-X mode, so add one more descriptor to the count
2885          */
2886         if(unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
2887                         (len > 2015)))
2888                 count++;
2889
2890         nr_frags = skb_shinfo(skb)->nr_frags;
2891         for(f = 0; f < nr_frags; f++)
2892                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
2893                                        max_txd_pwr);
2894         if(adapter->pcix_82544)
2895                 count += nr_frags;
2896
2897         if(adapter->hw.tx_pkt_filtering && (adapter->hw.mac_type == e1000_82573) )
2898                 e1000_transfer_dhcp_info(adapter, skb);
2899
2900         local_irq_save(flags);
2901         if (!spin_trylock(&tx_ring->tx_lock)) {
2902                 /* Collision - tell upper layer to requeue */
2903                 local_irq_restore(flags);
2904                 return NETDEV_TX_LOCKED;
2905         }
2906
2907         /* need: count + 2 desc gap to keep tail from touching
2908          * head, otherwise try next time */
2909         if (unlikely(E1000_DESC_UNUSED(tx_ring) < count + 2)) {
2910                 netif_stop_queue(netdev);
2911                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2912                 return NETDEV_TX_BUSY;
2913         }
2914
2915         if(unlikely(adapter->hw.mac_type == e1000_82547)) {
2916                 if(unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
2917                         netif_stop_queue(netdev);
2918                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies);
2919                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2920                         return NETDEV_TX_BUSY;
2921                 }
2922         }
2923
2924         if(unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
2925                 tx_flags |= E1000_TX_FLAGS_VLAN;
2926                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
2927         }
2928
2929         first = tx_ring->next_to_use;
2930
2931         tso = e1000_tso(adapter, tx_ring, skb);
2932         if (tso < 0) {
2933                 dev_kfree_skb_any(skb);
2934                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2935                 return NETDEV_TX_OK;
2936         }
2937
2938         if (likely(tso)) {
2939                 tx_ring->last_tx_tso = 1;
2940                 tx_flags |= E1000_TX_FLAGS_TSO;
2941         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
2942                 tx_flags |= E1000_TX_FLAGS_CSUM;
2943
2944         /* Old method was to assume IPv4 packet by default if TSO was enabled.
2945          * 82571 hardware supports TSO capabilities for IPv6 as well...
2946          * no longer assume, we must. */
2947         if (likely(skb->protocol == ntohs(ETH_P_IP)))
2948                 tx_flags |= E1000_TX_FLAGS_IPV4;
2949
2950         e1000_tx_queue(adapter, tx_ring, tx_flags,
2951                        e1000_tx_map(adapter, tx_ring, skb, first,
2952                                     max_per_txd, nr_frags, mss));
2953
2954         netdev->trans_start = jiffies;
2955
2956         /* Make sure there is space in the ring for the next send. */
2957         if (unlikely(E1000_DESC_UNUSED(tx_ring) < MAX_SKB_FRAGS + 2))
2958                 netif_stop_queue(netdev);
2959
2960         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2961         return NETDEV_TX_OK;
2962 }
2963
2964 /**
2965  * e1000_tx_timeout - Respond to a Tx Hang
2966  * @netdev: network interface device structure
2967  **/
2968
2969 static void
2970 e1000_tx_timeout(struct net_device *netdev)
2971 {
2972         struct e1000_adapter *adapter = netdev_priv(netdev);
2973
2974         /* Do the reset outside of interrupt context */
2975         schedule_work(&adapter->tx_timeout_task);
2976 }
2977
2978 static void
2979 e1000_tx_timeout_task(struct net_device *netdev)
2980 {
2981         struct e1000_adapter *adapter = netdev_priv(netdev);
2982
2983         adapter->tx_timeout_count++;
2984         e1000_down(adapter);
2985         e1000_up(adapter);
2986 }
2987
2988 /**
2989  * e1000_get_stats - Get System Network Statistics
2990  * @netdev: network interface device structure
2991  *
2992  * Returns the address of the device statistics structure.
2993  * The statistics are actually updated from the timer callback.
2994  **/
2995
2996 static struct net_device_stats *
2997 e1000_get_stats(struct net_device *netdev)
2998 {
2999         struct e1000_adapter *adapter = netdev_priv(netdev);
3000
3001         /* only return the current stats */
3002         return &adapter->net_stats;
3003 }
3004
3005 /**
3006  * e1000_change_mtu - Change the Maximum Transfer Unit
3007  * @netdev: network interface device structure
3008  * @new_mtu: new value for maximum frame size
3009  *
3010  * Returns 0 on success, negative on failure
3011  **/
3012
3013 static int
3014 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3015 {
3016         struct e1000_adapter *adapter = netdev_priv(netdev);
3017         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3018
3019         if((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3020                 (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3021                         DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3022                 return -EINVAL;
3023         }
3024
3025         /* Adapter-specific max frame size limits. */
3026         switch (adapter->hw.mac_type) {
3027         case e1000_82542_rev2_0:
3028         case e1000_82542_rev2_1:
3029         case e1000_82573:
3030                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3031                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3032                         return -EINVAL;
3033                 }
3034                 break;
3035         case e1000_82571:
3036         case e1000_82572:
3037 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3038                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3039                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3040                         return -EINVAL;
3041                 }
3042                 break;
3043         default:
3044                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3045                 break;
3046         }
3047
3048         /* since the driver code now supports splitting a packet across
3049          * multiple descriptors, most of the fifo related limitations on
3050          * jumbo frame traffic have gone away.
3051          * simply use 2k descriptors for everything.
3052          *
3053          * NOTE: dev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3054          * means we reserve 2 more, this pushes us to allocate from the next
3055          * larger slab size
3056          * i.e. RXBUFFER_2048 --> size-4096 slab */
3057
3058         /* recent hardware supports 1KB granularity */
3059         if (adapter->hw.mac_type > e1000_82547_rev_2) {
3060                 adapter->rx_buffer_len =
3061                     ((max_frame < E1000_RXBUFFER_2048) ?
3062                         max_frame : E1000_RXBUFFER_2048);
3063                 E1000_ROUNDUP(adapter->rx_buffer_len, 1024);
3064         } else
3065                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3066
3067         netdev->mtu = new_mtu;
3068
3069         if(netif_running(netdev)) {
3070                 e1000_down(adapter);
3071                 e1000_up(adapter);
3072         }
3073
3074         adapter->hw.max_frame_size = max_frame;
3075
3076         return 0;
3077 }
3078
3079 /**
3080  * e1000_update_stats - Update the board statistics counters
3081  * @adapter: board private structure
3082  **/
3083
3084 void
3085 e1000_update_stats(struct e1000_adapter *adapter)
3086 {
3087         struct e1000_hw *hw = &adapter->hw;
3088         unsigned long flags;
3089         uint16_t phy_tmp;
3090
3091 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3092
3093         spin_lock_irqsave(&adapter->stats_lock, flags);
3094
3095         /* these counters are modified from e1000_adjust_tbi_stats,
3096          * called from the interrupt context, so they must only
3097          * be written while holding adapter->stats_lock
3098          */
3099
3100         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3101         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3102         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3103         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3104         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3105         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3106         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3107         adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3108         adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3109         adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3110         adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3111         adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3112         adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3113
3114         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3115         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3116         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3117         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3118         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3119         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3120         adapter->stats.dc += E1000_READ_REG(hw, DC);
3121         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3122         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3123         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3124         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3125         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3126         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3127         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3128         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3129         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3130         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3131         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3132         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3133         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3134         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3135         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3136         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3137         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3138         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3139         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3140         adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3141         adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3142         adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3143         adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3144         adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3145         adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3146         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3147         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3148
3149         /* used for adaptive IFS */
3150
3151         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3152         adapter->stats.tpt += hw->tx_packet_delta;
3153         hw->collision_delta = E1000_READ_REG(hw, COLC);
3154         adapter->stats.colc += hw->collision_delta;
3155
3156         if(hw->mac_type >= e1000_82543) {
3157                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3158                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3159                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3160                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3161                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3162                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3163         }
3164         if(hw->mac_type > e1000_82547_rev_2) {
3165                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3166                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3167                 adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3168                 adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3169                 adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3170                 adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3171                 adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3172                 adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3173                 adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3174         }
3175
3176         /* Fill out the OS statistics structure */
3177
3178         adapter->net_stats.rx_packets = adapter->stats.gprc;
3179         adapter->net_stats.tx_packets = adapter->stats.gptc;
3180         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3181         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3182         adapter->net_stats.multicast = adapter->stats.mprc;
3183         adapter->net_stats.collisions = adapter->stats.colc;
3184
3185         /* Rx Errors */
3186
3187         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3188                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3189                 adapter->stats.rlec + adapter->stats.cexterr;
3190         adapter->net_stats.rx_dropped = 0;
3191         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3192         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3193         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3194         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3195
3196         /* Tx Errors */
3197
3198         adapter->net_stats.tx_errors = adapter->stats.ecol +
3199                                        adapter->stats.latecol;
3200         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3201         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3202         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3203
3204         /* Tx Dropped needs to be maintained elsewhere */
3205
3206         /* Phy Stats */
3207
3208         if(hw->media_type == e1000_media_type_copper) {
3209                 if((adapter->link_speed == SPEED_1000) &&
3210                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3211                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3212                         adapter->phy_stats.idle_errors += phy_tmp;
3213                 }
3214
3215                 if((hw->mac_type <= e1000_82546) &&
3216                    (hw->phy_type == e1000_phy_m88) &&
3217                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3218                         adapter->phy_stats.receive_errors += phy_tmp;
3219         }
3220
3221         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3222 }
3223
3224 #ifdef CONFIG_E1000_MQ
3225 void
3226 e1000_rx_schedule(void *data)
3227 {
3228         struct net_device *poll_dev, *netdev = data;
3229         struct e1000_adapter *adapter = netdev->priv;
3230         int this_cpu = get_cpu();
3231
3232         poll_dev = *per_cpu_ptr(adapter->cpu_netdev, this_cpu);
3233         if (poll_dev == NULL) {
3234                 put_cpu();
3235                 return;
3236         }
3237
3238         if (likely(netif_rx_schedule_prep(poll_dev)))
3239                 __netif_rx_schedule(poll_dev);
3240         else
3241                 e1000_irq_enable(adapter);
3242
3243         put_cpu();
3244 }
3245 #endif
3246
3247 /**
3248  * e1000_intr - Interrupt Handler
3249  * @irq: interrupt number
3250  * @data: pointer to a network interface device structure
3251  * @pt_regs: CPU registers structure
3252  **/
3253
3254 static irqreturn_t
3255 e1000_intr(int irq, void *data, struct pt_regs *regs)
3256 {
3257         struct net_device *netdev = data;
3258         struct e1000_adapter *adapter = netdev_priv(netdev);
3259         struct e1000_hw *hw = &adapter->hw;
3260         uint32_t icr = E1000_READ_REG(hw, ICR);
3261 #ifndef CONFIG_E1000_NAPI
3262         int i;
3263 #else
3264         /* Interrupt Auto-Mask...upon reading ICR,
3265          * interrupts are masked.  No need for the
3266          * IMC write, but it does mean we should
3267          * account for it ASAP. */
3268         if (likely(hw->mac_type >= e1000_82571))
3269                 atomic_inc(&adapter->irq_sem);
3270 #endif
3271
3272         if (unlikely(!icr)) {
3273 #ifdef CONFIG_E1000_NAPI
3274                 if (hw->mac_type >= e1000_82571)
3275                         e1000_irq_enable(adapter);
3276 #endif
3277                 return IRQ_NONE;  /* Not our interrupt */
3278         }
3279
3280         if(unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3281                 hw->get_link_status = 1;
3282                 mod_timer(&adapter->watchdog_timer, jiffies);
3283         }
3284
3285 #ifdef CONFIG_E1000_NAPI
3286         if (unlikely(hw->mac_type < e1000_82571)) {
3287                 atomic_inc(&adapter->irq_sem);
3288                 E1000_WRITE_REG(hw, IMC, ~0);
3289                 E1000_WRITE_FLUSH(hw);
3290         }
3291 #ifdef CONFIG_E1000_MQ
3292         if (atomic_read(&adapter->rx_sched_call_data.count) == 0) {
3293                 /* We must setup the cpumask once count == 0 since
3294                  * each cpu bit is cleared when the work is done. */
3295                 adapter->rx_sched_call_data.cpumask = adapter->cpumask;
3296                 atomic_add(adapter->num_rx_queues - 1, &adapter->irq_sem);
3297                 atomic_set(&adapter->rx_sched_call_data.count,
3298                            adapter->num_rx_queues);
3299                 smp_call_async_mask(&adapter->rx_sched_call_data);
3300         } else {
3301                 printk("call_data.count == %u\n", atomic_read(&adapter->rx_sched_call_data.count));
3302         }
3303 #else /* if !CONFIG_E1000_MQ */
3304         if (likely(netif_rx_schedule_prep(&adapter->polling_netdev[0])))
3305                 __netif_rx_schedule(&adapter->polling_netdev[0]);
3306         else
3307                 e1000_irq_enable(adapter);
3308 #endif /* CONFIG_E1000_MQ */
3309
3310 #else /* if !CONFIG_E1000_NAPI */
3311         /* Writing IMC and IMS is needed for 82547.
3312            Due to Hub Link bus being occupied, an interrupt
3313            de-assertion message is not able to be sent.
3314            When an interrupt assertion message is generated later,
3315            two messages are re-ordered and sent out.
3316            That causes APIC to think 82547 is in de-assertion
3317            state, while 82547 is in assertion state, resulting
3318            in dead lock. Writing IMC forces 82547 into
3319            de-assertion state.
3320         */
3321         if(hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2){
3322                 atomic_inc(&adapter->irq_sem);
3323                 E1000_WRITE_REG(hw, IMC, ~0);
3324         }
3325
3326         for(i = 0; i < E1000_MAX_INTR; i++)
3327                 if(unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3328                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3329                         break;
3330
3331         if(hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3332                 e1000_irq_enable(adapter);
3333
3334 #endif /* CONFIG_E1000_NAPI */
3335
3336         return IRQ_HANDLED;
3337 }
3338
3339 #ifdef CONFIG_E1000_NAPI
3340 /**
3341  * e1000_clean - NAPI Rx polling callback
3342  * @adapter: board private structure
3343  **/
3344
3345 static int
3346 e1000_clean(struct net_device *poll_dev, int *budget)
3347 {
3348         struct e1000_adapter *adapter;
3349         int work_to_do = min(*budget, poll_dev->quota);
3350         int tx_cleaned, i = 0, work_done = 0;
3351
3352         /* Must NOT use netdev_priv macro here. */
3353         adapter = poll_dev->priv;
3354
3355         /* Keep link state information with original netdev */
3356         if (!netif_carrier_ok(adapter->netdev))
3357                 goto quit_polling;
3358
3359         while (poll_dev != &adapter->polling_netdev[i]) {
3360                 i++;
3361                 if (unlikely(i == adapter->num_rx_queues))
3362                         BUG();
3363         }
3364
3365         tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[i]);
3366         adapter->clean_rx(adapter, &adapter->rx_ring[i],
3367                           &work_done, work_to_do);
3368
3369         *budget -= work_done;
3370         poll_dev->quota -= work_done;
3371
3372         /* If no Tx and not enough Rx work done, exit the polling mode */
3373         if((!tx_cleaned && (work_done == 0)) ||
3374            !netif_running(adapter->netdev)) {
3375 quit_polling:
3376                 netif_rx_complete(poll_dev);
3377                 e1000_irq_enable(adapter);
3378                 return 0;
3379         }
3380
3381         return 1;
3382 }
3383
3384 #endif
3385 /**
3386  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3387  * @adapter: board private structure
3388  **/
3389
3390 static boolean_t
3391 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3392                    struct e1000_tx_ring *tx_ring)
3393 {
3394         struct net_device *netdev = adapter->netdev;
3395         struct e1000_tx_desc *tx_desc, *eop_desc;
3396         struct e1000_buffer *buffer_info;
3397         unsigned int i, eop;
3398         boolean_t cleaned = FALSE;
3399
3400         i = tx_ring->next_to_clean;
3401         eop = tx_ring->buffer_info[i].next_to_watch;
3402         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3403
3404         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3405                 for(cleaned = FALSE; !cleaned; ) {
3406                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3407                         buffer_info = &tx_ring->buffer_info[i];
3408                         cleaned = (i == eop);
3409
3410                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3411
3412                         tx_desc->buffer_addr = 0;
3413                         tx_desc->lower.data = 0;
3414                         tx_desc->upper.data = 0;
3415
3416                         if(unlikely(++i == tx_ring->count)) i = 0;
3417                 }
3418
3419 #ifdef CONFIG_E1000_MQ
3420                 tx_ring->tx_stats.packets++;
3421 #endif
3422
3423                 eop = tx_ring->buffer_info[i].next_to_watch;
3424                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3425         }
3426
3427         tx_ring->next_to_clean = i;
3428
3429         spin_lock(&tx_ring->tx_lock);
3430
3431         if(unlikely(cleaned && netif_queue_stopped(netdev) &&
3432                     netif_carrier_ok(netdev)))
3433                 netif_wake_queue(netdev);
3434
3435         spin_unlock(&tx_ring->tx_lock);
3436
3437         if (adapter->detect_tx_hung) {
3438                 /* Detect a transmit hang in hardware, this serializes the
3439                  * check with the clearing of time_stamp and movement of i */
3440                 adapter->detect_tx_hung = FALSE;
3441                 if (tx_ring->buffer_info[eop].dma &&
3442                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3443                                adapter->tx_timeout_factor * HZ)
3444                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
3445                          E1000_STATUS_TXOFF)) {
3446
3447                         /* detected Tx unit hang */
3448                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3449                                         "  Tx Queue             <%lu>\n"
3450                                         "  TDH                  <%x>\n"
3451                                         "  TDT                  <%x>\n"
3452                                         "  next_to_use          <%x>\n"
3453                                         "  next_to_clean        <%x>\n"
3454                                         "buffer_info[next_to_clean]\n"
3455                                         "  time_stamp           <%lx>\n"
3456                                         "  next_to_watch        <%x>\n"
3457                                         "  jiffies              <%lx>\n"
3458                                         "  next_to_watch.status <%x>\n",
3459                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3460                                         sizeof(struct e1000_tx_ring)),
3461                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
3462                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
3463                                 tx_ring->next_to_use,
3464                                 tx_ring->next_to_clean,
3465                                 tx_ring->buffer_info[eop].time_stamp,
3466                                 eop,
3467                                 jiffies,
3468                                 eop_desc->upper.fields.status);
3469                         netif_stop_queue(netdev);
3470                 }
3471         }
3472         return cleaned;
3473 }
3474
3475 /**
3476  * e1000_rx_checksum - Receive Checksum Offload for 82543
3477  * @adapter:     board private structure
3478  * @status_err:  receive descriptor status and error fields
3479  * @csum:        receive descriptor csum field
3480  * @sk_buff:     socket buffer with received data
3481  **/
3482
3483 static inline void
3484 e1000_rx_checksum(struct e1000_adapter *adapter,
3485                   uint32_t status_err, uint32_t csum,
3486                   struct sk_buff *skb)
3487 {
3488         uint16_t status = (uint16_t)status_err;
3489         uint8_t errors = (uint8_t)(status_err >> 24);
3490         skb->ip_summed = CHECKSUM_NONE;
3491
3492         /* 82543 or newer only */
3493         if(unlikely(adapter->hw.mac_type < e1000_82543)) return;
3494         /* Ignore Checksum bit is set */
3495         if(unlikely(status & E1000_RXD_STAT_IXSM)) return;
3496         /* TCP/UDP checksum error bit is set */
3497         if(unlikely(errors & E1000_RXD_ERR_TCPE)) {
3498                 /* let the stack verify checksum errors */
3499                 adapter->hw_csum_err++;
3500                 return;
3501         }
3502         /* TCP/UDP Checksum has not been calculated */
3503         if(adapter->hw.mac_type <= e1000_82547_rev_2) {
3504                 if(!(status & E1000_RXD_STAT_TCPCS))
3505                         return;
3506         } else {
3507                 if(!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
3508                         return;
3509         }
3510         /* It must be a TCP or UDP packet with a valid checksum */
3511         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3512                 /* TCP checksum is good */
3513                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3514         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
3515                 /* IP fragment with UDP payload */
3516                 /* Hardware complements the payload checksum, so we undo it
3517                  * and then put the value in host order for further stack use.
3518                  */
3519                 csum = ntohl(csum ^ 0xFFFF);
3520                 skb->csum = csum;
3521                 skb->ip_summed = CHECKSUM_HW;
3522         }
3523         adapter->hw_csum_good++;
3524 }
3525
3526 /**
3527  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3528  * @adapter: board private structure
3529  **/
3530
3531 static boolean_t
3532 #ifdef CONFIG_E1000_NAPI
3533 e1000_clean_rx_irq(struct e1000_adapter *adapter,
3534                    struct e1000_rx_ring *rx_ring,
3535                    int *work_done, int work_to_do)
3536 #else
3537 e1000_clean_rx_irq(struct e1000_adapter *adapter,
3538                    struct e1000_rx_ring *rx_ring)
3539 #endif
3540 {
3541         struct net_device *netdev = adapter->netdev;
3542         struct pci_dev *pdev = adapter->pdev;
3543         struct e1000_rx_desc *rx_desc;
3544         struct e1000_buffer *buffer_info;
3545         struct sk_buff *skb;
3546         unsigned long flags;
3547         uint32_t length;
3548         uint8_t last_byte;
3549         unsigned int i;
3550         boolean_t cleaned = FALSE;
3551         int cleaned_count = 0;
3552
3553         i = rx_ring->next_to_clean;
3554         rx_desc = E1000_RX_DESC(*rx_ring, i);
3555
3556         while(rx_desc->status & E1000_RXD_STAT_DD) {
3557                 buffer_info = &rx_ring->buffer_info[i];
3558 #ifdef CONFIG_E1000_NAPI
3559                 if(*work_done >= work_to_do)
3560                         break;
3561                 (*work_done)++;
3562 #endif
3563
3564                 cleaned = TRUE;
3565                 cleaned_count++;
3566                 pci_unmap_single(pdev, buffer_info->dma, buffer_info->length,
3567                                  PCI_DMA_FROMDEVICE);
3568
3569                 skb = buffer_info->skb;
3570                 length = le16_to_cpu(rx_desc->length);
3571
3572                 if(unlikely(!(rx_desc->status & E1000_RXD_STAT_EOP))) {
3573                         /* All receives must fit into a single buffer */
3574                         E1000_DBG("%s: Receive packet consumed multiple"
3575                                   " buffers\n", netdev->name);
3576                         dev_kfree_skb_irq(skb);
3577                         goto next_desc;
3578                 }
3579
3580                 if(unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3581                         last_byte = *(skb->data + length - 1);
3582                         if(TBI_ACCEPT(&adapter->hw, rx_desc->status,
3583                                       rx_desc->errors, length, last_byte)) {
3584                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3585                                 e1000_tbi_adjust_stats(&adapter->hw, &adapter->stats,
3586                                                        length, skb->data);
3587                                 spin_unlock_irqrestore(&adapter->stats_lock,
3588                                                        flags);
3589                                 length--;
3590                         } else {
3591                                 dev_kfree_skb_irq(skb);
3592                                 goto next_desc;
3593                         }
3594                 }
3595
3596                 /* Good Receive */
3597                 skb_put(skb, length - ETHERNET_FCS_SIZE);
3598
3599                 /* Receive Checksum Offload */
3600                 e1000_rx_checksum(adapter, (uint32_t)(rx_desc->status) |
3601                                   ((uint32_t)(rx_desc->errors) << 24),
3602                                   rx_desc->csum, skb);
3603                 skb->protocol = eth_type_trans(skb, netdev);
3604 #ifdef CONFIG_E1000_NAPI
3605                 if(unlikely(adapter->vlgrp &&
3606                             (rx_desc->status & E1000_RXD_STAT_VP))) {
3607                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3608                                                  le16_to_cpu(rx_desc->special) &
3609                                                  E1000_RXD_SPC_VLAN_MASK);
3610                 } else {
3611                         netif_receive_skb(skb);
3612                 }
3613 #else /* CONFIG_E1000_NAPI */
3614                 if(unlikely(adapter->vlgrp &&
3615                             (rx_desc->status & E1000_RXD_STAT_VP))) {
3616                         vlan_hwaccel_rx(skb, adapter->vlgrp,
3617                                         le16_to_cpu(rx_desc->special) &
3618                                         E1000_RXD_SPC_VLAN_MASK);
3619                 } else {
3620                         netif_rx(skb);
3621                 }
3622 #endif /* CONFIG_E1000_NAPI */
3623                 netdev->last_rx = jiffies;
3624 #ifdef CONFIG_E1000_MQ
3625                 rx_ring->rx_stats.packets++;
3626                 rx_ring->rx_stats.bytes += length;
3627 #endif
3628
3629 next_desc:
3630                 rx_desc->status = 0;
3631
3632                 /* return some buffers to hardware, one at a time is too slow */
3633                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3634                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3635                         cleaned_count = 0;
3636                 }
3637
3638         }
3639         rx_ring->next_to_clean = i;
3640
3641         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3642         if (cleaned_count)
3643                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3644
3645         return cleaned;
3646 }
3647
3648 /**
3649  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
3650  * @adapter: board private structure
3651  **/
3652
3653 static boolean_t
3654 #ifdef CONFIG_E1000_NAPI
3655 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
3656                       struct e1000_rx_ring *rx_ring,
3657                       int *work_done, int work_to_do)
3658 #else
3659 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
3660                       struct e1000_rx_ring *rx_ring)
3661 #endif
3662 {
3663         union e1000_rx_desc_packet_split *rx_desc;
3664         struct net_device *netdev = adapter->netdev;
3665         struct pci_dev *pdev = adapter->pdev;
3666         struct e1000_buffer *buffer_info;
3667         struct e1000_ps_page *ps_page;
3668         struct e1000_ps_page_dma *ps_page_dma;
3669         struct sk_buff *skb;
3670         unsigned int i, j;
3671         uint32_t length, staterr;
3672         int cleaned_count = 0;
3673         boolean_t cleaned = FALSE;
3674
3675         i = rx_ring->next_to_clean;
3676         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
3677         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
3678
3679         while(staterr & E1000_RXD_STAT_DD) {
3680                 buffer_info = &rx_ring->buffer_info[i];
3681                 ps_page = &rx_ring->ps_page[i];
3682                 ps_page_dma = &rx_ring->ps_page_dma[i];
3683 #ifdef CONFIG_E1000_NAPI
3684                 if(unlikely(*work_done >= work_to_do))
3685                         break;
3686                 (*work_done)++;
3687 #endif
3688                 cleaned = TRUE;
3689                 cleaned_count++;
3690                 pci_unmap_single(pdev, buffer_info->dma,
3691                                  buffer_info->length,
3692                                  PCI_DMA_FROMDEVICE);
3693
3694                 skb = buffer_info->skb;
3695
3696                 if(unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
3697                         E1000_DBG("%s: Packet Split buffers didn't pick up"
3698                                   " the full packet\n", netdev->name);
3699                         dev_kfree_skb_irq(skb);
3700                         goto next_desc;
3701                 }
3702
3703                 if(unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
3704                         dev_kfree_skb_irq(skb);
3705                         goto next_desc;
3706                 }
3707
3708                 length = le16_to_cpu(rx_desc->wb.middle.length0);
3709
3710                 if(unlikely(!length)) {
3711                         E1000_DBG("%s: Last part of the packet spanning"
3712                                   " multiple descriptors\n", netdev->name);
3713                         dev_kfree_skb_irq(skb);
3714                         goto next_desc;
3715                 }
3716
3717                 /* Good Receive */
3718                 skb_put(skb, length);
3719
3720                 for(j = 0; j < adapter->rx_ps_pages; j++) {
3721                         if(!(length = le16_to_cpu(rx_desc->wb.upper.length[j])))
3722                                 break;
3723
3724                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
3725                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
3726                         ps_page_dma->ps_page_dma[j] = 0;
3727                         skb_shinfo(skb)->frags[j].page =
3728                                 ps_page->ps_page[j];
3729                         ps_page->ps_page[j] = NULL;
3730                         skb_shinfo(skb)->frags[j].page_offset = 0;
3731                         skb_shinfo(skb)->frags[j].size = length;
3732                         skb_shinfo(skb)->nr_frags++;
3733                         skb->len += length;
3734                         skb->data_len += length;
3735                 }
3736
3737                 e1000_rx_checksum(adapter, staterr,
3738                                   rx_desc->wb.lower.hi_dword.csum_ip.csum, skb);
3739                 skb->protocol = eth_type_trans(skb, netdev);
3740
3741                 if(likely(rx_desc->wb.upper.header_status &
3742                           E1000_RXDPS_HDRSTAT_HDRSP)) {
3743                         adapter->rx_hdr_split++;
3744 #ifdef HAVE_RX_ZERO_COPY
3745                         skb_shinfo(skb)->zero_copy = TRUE;
3746 #endif
3747                 }
3748 #ifdef CONFIG_E1000_NAPI
3749                 if(unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
3750                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3751                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
3752                                 E1000_RXD_SPC_VLAN_MASK);
3753                 } else {
3754                         netif_receive_skb(skb);
3755                 }
3756 #else /* CONFIG_E1000_NAPI */
3757                 if(unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
3758                         vlan_hwaccel_rx(skb, adapter->vlgrp,
3759                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
3760                                 E1000_RXD_SPC_VLAN_MASK);
3761                 } else {
3762                         netif_rx(skb);
3763                 }
3764 #endif /* CONFIG_E1000_NAPI */
3765                 netdev->last_rx = jiffies;
3766 #ifdef CONFIG_E1000_MQ
3767                 rx_ring->rx_stats.packets++;
3768                 rx_ring->rx_stats.bytes += length;
3769 #endif
3770
3771 next_desc:
3772                 rx_desc->wb.middle.status_error &= ~0xFF;
3773                 buffer_info->skb = NULL;
3774
3775                 /* return some buffers to hardware, one at a time is too slow */
3776                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3777                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3778                         cleaned_count = 0;
3779                 }
3780
3781                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
3782         }
3783         rx_ring->next_to_clean = i;
3784
3785         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3786         if (cleaned_count)
3787                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3788
3789         return cleaned;
3790 }
3791
3792 /**
3793  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
3794  * @adapter: address of board private structure
3795  **/
3796
3797 static void
3798 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
3799                        struct e1000_rx_ring *rx_ring,
3800                        int cleaned_count)
3801 {
3802         struct net_device *netdev = adapter->netdev;
3803         struct pci_dev *pdev = adapter->pdev;
3804         struct e1000_rx_desc *rx_desc;
3805         struct e1000_buffer *buffer_info;
3806         struct sk_buff *skb;
3807         unsigned int i;
3808         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
3809
3810         i = rx_ring->next_to_use;
3811         buffer_info = &rx_ring->buffer_info[i];
3812
3813         while(!buffer_info->skb) {
3814                 skb = dev_alloc_skb(bufsz);
3815
3816                 if(unlikely(!skb)) {
3817                         /* Better luck next round */
3818                         adapter->alloc_rx_buff_failed++;
3819                         break;
3820                 }
3821
3822                 /* Fix for errata 23, can't cross 64kB boundary */
3823                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3824                         struct sk_buff *oldskb = skb;
3825                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
3826                                              "at %p\n", bufsz, skb->data);
3827                         /* Try again, without freeing the previous */
3828                         skb = dev_alloc_skb(bufsz);
3829                         /* Failed allocation, critical failure */
3830                         if (!skb) {
3831                                 dev_kfree_skb(oldskb);
3832                                 break;
3833                         }
3834
3835                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3836                                 /* give up */
3837                                 dev_kfree_skb(skb);
3838                                 dev_kfree_skb(oldskb);
3839                                 break; /* while !buffer_info->skb */
3840                         } else {
3841                                 /* Use new allocation */
3842                                 dev_kfree_skb(oldskb);
3843                         }
3844                 }
3845                 /* Make buffer alignment 2 beyond a 16 byte boundary
3846                  * this will result in a 16 byte aligned IP header after
3847                  * the 14 byte MAC header is removed
3848                  */
3849                 skb_reserve(skb, NET_IP_ALIGN);
3850
3851                 skb->dev = netdev;
3852
3853                 buffer_info->skb = skb;
3854                 buffer_info->length = adapter->rx_buffer_len;
3855                 buffer_info->dma = pci_map_single(pdev,
3856                                                   skb->data,
3857                                                   adapter->rx_buffer_len,
3858                                                   PCI_DMA_FROMDEVICE);
3859
3860                 /* Fix for errata 23, can't cross 64kB boundary */
3861                 if (!e1000_check_64k_bound(adapter,
3862                                         (void *)(unsigned long)buffer_info->dma,
3863                                         adapter->rx_buffer_len)) {
3864                         DPRINTK(RX_ERR, ERR,
3865                                 "dma align check failed: %u bytes at %p\n",
3866                                 adapter->rx_buffer_len,
3867                                 (void *)(unsigned long)buffer_info->dma);
3868                         dev_kfree_skb(skb);
3869                         buffer_info->skb = NULL;
3870
3871                         pci_unmap_single(pdev, buffer_info->dma,
3872                                          adapter->rx_buffer_len,
3873                                          PCI_DMA_FROMDEVICE);
3874
3875                         break; /* while !buffer_info->skb */
3876                 }
3877                 rx_desc = E1000_RX_DESC(*rx_ring, i);
3878                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
3879
3880                 if(unlikely((i & ~(E1000_RX_BUFFER_WRITE - 1)) == i)) {
3881                         /* Force memory writes to complete before letting h/w
3882                          * know there are new descriptors to fetch.  (Only
3883                          * applicable for weak-ordered memory model archs,
3884                          * such as IA-64). */
3885                         wmb();
3886                         writel(i, adapter->hw.hw_addr + rx_ring->rdt);
3887                 }
3888
3889                 if(unlikely(++i == rx_ring->count)) i = 0;
3890                 buffer_info = &rx_ring->buffer_info[i];
3891         }
3892
3893         rx_ring->next_to_use = i;
3894 }
3895
3896 /**
3897  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
3898  * @adapter: address of board private structure
3899  **/
3900
3901 static void
3902 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
3903                           struct e1000_rx_ring *rx_ring,
3904                           int cleaned_count)
3905 {
3906         struct net_device *netdev = adapter->netdev;
3907         struct pci_dev *pdev = adapter->pdev;
3908         union e1000_rx_desc_packet_split *rx_desc;
3909         struct e1000_buffer *buffer_info;
3910         struct e1000_ps_page *ps_page;
3911         struct e1000_ps_page_dma *ps_page_dma;
3912         struct sk_buff *skb;
3913         unsigned int i, j;
3914
3915         i = rx_ring->next_to_use;
3916         buffer_info = &rx_ring->buffer_info[i];
3917         ps_page = &rx_ring->ps_page[i];
3918         ps_page_dma = &rx_ring->ps_page_dma[i];
3919
3920         while (cleaned_count--) {
3921                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
3922
3923                 for(j = 0; j < PS_PAGE_BUFFERS; j++) {
3924                         if (j < adapter->rx_ps_pages) {
3925                                 if (likely(!ps_page->ps_page[j])) {
3926                                         ps_page->ps_page[j] =
3927                                                 alloc_page(GFP_ATOMIC);
3928                                         if (unlikely(!ps_page->ps_page[j]))
3929                                                 goto no_buffers;
3930                                         ps_page_dma->ps_page_dma[j] =
3931                                                 pci_map_page(pdev,
3932                                                             ps_page->ps_page[j],
3933                                                             0, PAGE_SIZE,
3934                                                             PCI_DMA_FROMDEVICE);
3935                                 }
3936                                 /* Refresh the desc even if buffer_addrs didn't
3937                                  * change because each write-back erases
3938                                  * this info.
3939                                  */
3940                                 rx_desc->read.buffer_addr[j+1] =
3941                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
3942                         } else
3943                                 rx_desc->read.buffer_addr[j+1] = ~0;
3944                 }
3945
3946                 skb = dev_alloc_skb(adapter->rx_ps_bsize0 + NET_IP_ALIGN);
3947
3948                 if(unlikely(!skb))
3949                         break;
3950
3951                 /* Make buffer alignment 2 beyond a 16 byte boundary
3952                  * this will result in a 16 byte aligned IP header after
3953                  * the 14 byte MAC header is removed
3954                  */
3955                 skb_reserve(skb, NET_IP_ALIGN);
3956
3957                 skb->dev = netdev;
3958
3959                 buffer_info->skb = skb;
3960                 buffer_info->length = adapter->rx_ps_bsize0;
3961                 buffer_info->dma = pci_map_single(pdev, skb->data,
3962                                                   adapter->rx_ps_bsize0,
3963                                                   PCI_DMA_FROMDEVICE);
3964
3965                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
3966
3967                 if(unlikely((i & ~(E1000_RX_BUFFER_WRITE - 1)) == i)) {
3968                         /* Force memory writes to complete before letting h/w
3969                          * know there are new descriptors to fetch.  (Only
3970                          * applicable for weak-ordered memory model archs,
3971                          * such as IA-64). */
3972                         wmb();
3973                         /* Hardware increments by 16 bytes, but packet split
3974                          * descriptors are 32 bytes...so we increment tail
3975                          * twice as much.
3976                          */
3977                         writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
3978                 }
3979
3980                 if(unlikely(++i == rx_ring->count)) i = 0;
3981                 buffer_info = &rx_ring->buffer_info[i];
3982                 ps_page = &rx_ring->ps_page[i];
3983                 ps_page_dma = &rx_ring->ps_page_dma[i];
3984         }
3985
3986 no_buffers:
3987         rx_ring->next_to_use = i;
3988 }
3989
3990 /**
3991  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
3992  * @adapter:
3993  **/
3994
3995 static void
3996 e1000_smartspeed(struct e1000_adapter *adapter)
3997 {
3998         uint16_t phy_status;
3999         uint16_t phy_ctrl;
4000
4001         if((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4002            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4003                 return;
4004
4005         if(adapter->smartspeed == 0) {
4006                 /* If Master/Slave config fault is asserted twice,
4007                  * we assume back-to-back */
4008                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4009                 if(!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4010                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4011                 if(!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4012                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4013                 if(phy_ctrl & CR_1000T_MS_ENABLE) {
4014                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4015                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4016                                             phy_ctrl);
4017                         adapter->smartspeed++;
4018                         if(!e1000_phy_setup_autoneg(&adapter->hw) &&
4019                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4020                                                &phy_ctrl)) {
4021                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4022                                              MII_CR_RESTART_AUTO_NEG);
4023                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4024                                                     phy_ctrl);
4025                         }
4026                 }
4027                 return;
4028         } else if(adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4029                 /* If still no link, perhaps using 2/3 pair cable */
4030                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4031                 phy_ctrl |= CR_1000T_MS_ENABLE;
4032                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4033                 if(!e1000_phy_setup_autoneg(&adapter->hw) &&
4034                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4035                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4036                                      MII_CR_RESTART_AUTO_NEG);
4037                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4038                 }
4039         }
4040         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4041         if(adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4042                 adapter->smartspeed = 0;
4043 }
4044
4045 /**
4046  * e1000_ioctl -
4047  * @netdev:
4048  * @ifreq:
4049  * @cmd:
4050  **/
4051
4052 static int
4053 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4054 {
4055         switch (cmd) {
4056         case SIOCGMIIPHY:
4057         case SIOCGMIIREG:
4058         case SIOCSMIIREG:
4059                 return e1000_mii_ioctl(netdev, ifr, cmd);
4060         default:
4061                 return -EOPNOTSUPP;
4062         }
4063 }
4064
4065 /**
4066  * e1000_mii_ioctl -
4067  * @netdev:
4068  * @ifreq:
4069  * @cmd:
4070  **/
4071
4072 static int
4073 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4074 {
4075         struct e1000_adapter *adapter = netdev_priv(netdev);
4076         struct mii_ioctl_data *data = if_mii(ifr);
4077         int retval;
4078         uint16_t mii_reg;
4079         uint16_t spddplx;
4080         unsigned long flags;
4081
4082         if(adapter->hw.media_type != e1000_media_type_copper)
4083                 return -EOPNOTSUPP;
4084
4085         switch (cmd) {
4086         case SIOCGMIIPHY:
4087                 data->phy_id = adapter->hw.phy_addr;
4088                 break;
4089         case SIOCGMIIREG:
4090                 if(!capable(CAP_NET_ADMIN))
4091                         return -EPERM;
4092                 spin_lock_irqsave(&adapter->stats_lock, flags);
4093                 if(e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4094                                    &data->val_out)) {
4095                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4096                         return -EIO;
4097                 }
4098                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4099                 break;
4100         case SIOCSMIIREG:
4101                 if(!capable(CAP_NET_ADMIN))
4102                         return -EPERM;
4103                 if(data->reg_num & ~(0x1F))
4104                         return -EFAULT;
4105                 mii_reg = data->val_in;
4106                 spin_lock_irqsave(&adapter->stats_lock, flags);
4107                 if(e1000_write_phy_reg(&adapter->hw, data->reg_num,
4108                                         mii_reg)) {
4109                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4110                         return -EIO;
4111                 }
4112                 if(adapter->hw.phy_type == e1000_phy_m88) {
4113                         switch (data->reg_num) {
4114                         case PHY_CTRL:
4115                                 if(mii_reg & MII_CR_POWER_DOWN)
4116                                         break;
4117                                 if(mii_reg & MII_CR_AUTO_NEG_EN) {
4118                                         adapter->hw.autoneg = 1;
4119                                         adapter->hw.autoneg_advertised = 0x2F;
4120                                 } else {
4121                                         if (mii_reg & 0x40)
4122                                                 spddplx = SPEED_1000;
4123                                         else if (mii_reg & 0x2000)
4124                                                 spddplx = SPEED_100;
4125                                         else
4126                                                 spddplx = SPEED_10;
4127                                         spddplx += (mii_reg & 0x100)
4128                                                    ? FULL_DUPLEX :
4129                                                    HALF_DUPLEX;
4130                                         retval = e1000_set_spd_dplx(adapter,
4131                                                                     spddplx);
4132                                         if(retval) {
4133                                                 spin_unlock_irqrestore(
4134                                                         &adapter->stats_lock,
4135                                                         flags);
4136                                                 return retval;
4137                                         }
4138                                 }
4139                                 if(netif_running(adapter->netdev)) {
4140                                         e1000_down(adapter);
4141                                         e1000_up(adapter);
4142                                 } else
4143                                         e1000_reset(adapter);
4144                                 break;
4145                         case M88E1000_PHY_SPEC_CTRL:
4146                         case M88E1000_EXT_PHY_SPEC_CTRL:
4147                                 if(e1000_phy_reset(&adapter->hw)) {
4148                                         spin_unlock_irqrestore(
4149                                                 &adapter->stats_lock, flags);
4150                                         return -EIO;
4151                                 }
4152                                 break;
4153                         }
4154                 } else {
4155                         switch (data->reg_num) {
4156                         case PHY_CTRL:
4157                                 if(mii_reg & MII_CR_POWER_DOWN)
4158                                         break;
4159                                 if(netif_running(adapter->netdev)) {
4160                                         e1000_down(adapter);
4161                                         e1000_up(adapter);
4162                                 } else
4163                                         e1000_reset(adapter);
4164                                 break;
4165                         }
4166                 }
4167                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4168                 break;
4169         default:
4170                 return -EOPNOTSUPP;
4171         }
4172         return E1000_SUCCESS;
4173 }
4174
4175 void
4176 e1000_pci_set_mwi(struct e1000_hw *hw)
4177 {
4178         struct e1000_adapter *adapter = hw->back;
4179         int ret_val = pci_set_mwi(adapter->pdev);
4180
4181         if(ret_val)
4182                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4183 }
4184
4185 void
4186 e1000_pci_clear_mwi(struct e1000_hw *hw)
4187 {
4188         struct e1000_adapter *adapter = hw->back;
4189
4190         pci_clear_mwi(adapter->pdev);
4191 }
4192
4193 void
4194 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4195 {
4196         struct e1000_adapter *adapter = hw->back;
4197
4198         pci_read_config_word(adapter->pdev, reg, value);
4199 }
4200
4201 void
4202 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4203 {
4204         struct e1000_adapter *adapter = hw->back;
4205
4206         pci_write_config_word(adapter->pdev, reg, *value);
4207 }
4208
4209 uint32_t
4210 e1000_io_read(struct e1000_hw *hw, unsigned long port)
4211 {
4212         return inl(port);
4213 }
4214
4215 void
4216 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4217 {
4218         outl(value, port);
4219 }
4220
4221 static void
4222 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4223 {
4224         struct e1000_adapter *adapter = netdev_priv(netdev);
4225         uint32_t ctrl, rctl;
4226
4227         e1000_irq_disable(adapter);
4228         adapter->vlgrp = grp;
4229
4230         if(grp) {
4231                 /* enable VLAN tag insert/strip */
4232                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4233                 ctrl |= E1000_CTRL_VME;
4234                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4235
4236                 /* enable VLAN receive filtering */
4237                 rctl = E1000_READ_REG(&adapter->hw, RCTL);
4238                 rctl |= E1000_RCTL_VFE;
4239                 rctl &= ~E1000_RCTL_CFIEN;
4240                 E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4241                 e1000_update_mng_vlan(adapter);
4242         } else {
4243                 /* disable VLAN tag insert/strip */
4244                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4245                 ctrl &= ~E1000_CTRL_VME;
4246                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4247
4248                 /* disable VLAN filtering */
4249                 rctl = E1000_READ_REG(&adapter->hw, RCTL);
4250                 rctl &= ~E1000_RCTL_VFE;
4251                 E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4252                 if(adapter->mng_vlan_id != (uint16_t)E1000_MNG_VLAN_NONE) {
4253                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4254                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4255                 }
4256         }
4257
4258         e1000_irq_enable(adapter);
4259 }
4260
4261 static void
4262 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4263 {
4264         struct e1000_adapter *adapter = netdev_priv(netdev);
4265         uint32_t vfta, index;
4266         if((adapter->hw.mng_cookie.status &
4267                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4268                 (vid == adapter->mng_vlan_id))
4269                 return;
4270         /* add VID to filter table */
4271         index = (vid >> 5) & 0x7F;
4272         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4273         vfta |= (1 << (vid & 0x1F));
4274         e1000_write_vfta(&adapter->hw, index, vfta);
4275 }
4276
4277 static void
4278 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4279 {
4280         struct e1000_adapter *adapter = netdev_priv(netdev);
4281         uint32_t vfta, index;
4282
4283         e1000_irq_disable(adapter);
4284
4285         if(adapter->vlgrp)
4286                 adapter->vlgrp->vlan_devices[vid] = NULL;
4287
4288         e1000_irq_enable(adapter);
4289
4290         if((adapter->hw.mng_cookie.status &
4291                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4292             (vid == adapter->mng_vlan_id)) {
4293                 /* release control to f/w */
4294                 e1000_release_hw_control(adapter);
4295                 return;
4296         }
4297
4298         /* remove VID from filter table */
4299         index = (vid >> 5) & 0x7F;
4300         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4301         vfta &= ~(1 << (vid & 0x1F));
4302         e1000_write_vfta(&adapter->hw, index, vfta);
4303 }
4304
4305 static void
4306 e1000_restore_vlan(struct e1000_adapter *adapter)
4307 {
4308         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4309
4310         if(adapter->vlgrp) {
4311                 uint16_t vid;
4312                 for(vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4313                         if(!adapter->vlgrp->vlan_devices[vid])
4314                                 continue;
4315                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4316                 }
4317         }
4318 }
4319
4320 int
4321 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
4322 {
4323         adapter->hw.autoneg = 0;
4324
4325         /* Fiber NICs only allow 1000 gbps Full duplex */
4326         if((adapter->hw.media_type == e1000_media_type_fiber) &&
4327                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4328                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4329                 return -EINVAL;
4330         }
4331
4332         switch(spddplx) {
4333         case SPEED_10 + DUPLEX_HALF:
4334                 adapter->hw.forced_speed_duplex = e1000_10_half;
4335                 break;
4336         case SPEED_10 + DUPLEX_FULL:
4337                 adapter->hw.forced_speed_duplex = e1000_10_full;
4338                 break;
4339         case SPEED_100 + DUPLEX_HALF:
4340                 adapter->hw.forced_speed_duplex = e1000_100_half;
4341                 break;
4342         case SPEED_100 + DUPLEX_FULL:
4343                 adapter->hw.forced_speed_duplex = e1000_100_full;
4344                 break;
4345         case SPEED_1000 + DUPLEX_FULL:
4346                 adapter->hw.autoneg = 1;
4347                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
4348                 break;
4349         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4350         default:
4351                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4352                 return -EINVAL;
4353         }
4354         return 0;
4355 }
4356
4357 #ifdef CONFIG_PM
4358 static int
4359 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4360 {
4361         struct net_device *netdev = pci_get_drvdata(pdev);
4362         struct e1000_adapter *adapter = netdev_priv(netdev);
4363         uint32_t ctrl, ctrl_ext, rctl, manc, status;
4364         uint32_t wufc = adapter->wol;
4365
4366         netif_device_detach(netdev);
4367
4368         if(netif_running(netdev))
4369                 e1000_down(adapter);
4370
4371         status = E1000_READ_REG(&adapter->hw, STATUS);
4372         if(status & E1000_STATUS_LU)
4373                 wufc &= ~E1000_WUFC_LNKC;
4374
4375         if(wufc) {
4376                 e1000_setup_rctl(adapter);
4377                 e1000_set_multi(netdev);
4378
4379                 /* turn on all-multi mode if wake on multicast is enabled */
4380                 if(adapter->wol & E1000_WUFC_MC) {
4381                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4382                         rctl |= E1000_RCTL_MPE;
4383                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4384                 }
4385
4386                 if(adapter->hw.mac_type >= e1000_82540) {
4387                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4388                         /* advertise wake from D3Cold */
4389                         #define E1000_CTRL_ADVD3WUC 0x00100000
4390                         /* phy power management enable */
4391                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4392                         ctrl |= E1000_CTRL_ADVD3WUC |
4393                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4394                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4395                 }
4396
4397                 if(adapter->hw.media_type == e1000_media_type_fiber ||
4398                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
4399                         /* keep the laser running in D3 */
4400                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
4401                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4402                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
4403                 }
4404
4405                 /* Allow time for pending master requests to run */
4406                 e1000_disable_pciex_master(&adapter->hw);
4407
4408                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
4409                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
4410                 pci_enable_wake(pdev, 3, 1);
4411                 pci_enable_wake(pdev, 4, 1); /* 4 == D3 cold */
4412         } else {
4413                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
4414                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
4415                 pci_enable_wake(pdev, 3, 0);
4416                 pci_enable_wake(pdev, 4, 0); /* 4 == D3 cold */
4417         }
4418
4419         pci_save_state(pdev);
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                 if(manc & E1000_MANC_SMBUS_EN) {
4425                         manc |= E1000_MANC_ARP_EN;
4426                         E1000_WRITE_REG(&adapter->hw, MANC, manc);
4427                         pci_enable_wake(pdev, 3, 1);
4428                         pci_enable_wake(pdev, 4, 1); /* 4 == D3 cold */
4429                 }
4430         }
4431
4432         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4433          * would have already happened in close and is redundant. */
4434         e1000_release_hw_control(adapter);
4435
4436         pci_disable_device(pdev);
4437         pci_set_power_state(pdev, pci_choose_state(pdev, state));
4438
4439         return 0;
4440 }
4441
4442 static int
4443 e1000_resume(struct pci_dev *pdev)
4444 {
4445         struct net_device *netdev = pci_get_drvdata(pdev);
4446         struct e1000_adapter *adapter = netdev_priv(netdev);
4447         uint32_t manc, ret_val;
4448
4449         pci_set_power_state(pdev, PCI_D0);
4450         pci_restore_state(pdev);
4451         ret_val = pci_enable_device(pdev);
4452         pci_set_master(pdev);
4453
4454         pci_enable_wake(pdev, PCI_D3hot, 0);
4455         pci_enable_wake(pdev, PCI_D3cold, 0);
4456
4457         e1000_reset(adapter);
4458         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
4459
4460         if(netif_running(netdev))
4461                 e1000_up(adapter);
4462
4463         netif_device_attach(netdev);
4464
4465         if(adapter->hw.mac_type >= e1000_82540 &&
4466            adapter->hw.media_type == e1000_media_type_copper) {
4467                 manc = E1000_READ_REG(&adapter->hw, MANC);
4468                 manc &= ~(E1000_MANC_ARP_EN);
4469                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
4470         }
4471
4472         /* If the controller is 82573 and f/w is AMT, do not set
4473          * DRV_LOAD until the interface is up.  For all other cases,
4474          * let the f/w know that the h/w is now under the control
4475          * of the driver. */
4476         if (adapter->hw.mac_type != e1000_82573 ||
4477             !e1000_check_mng_mode(&adapter->hw))
4478                 e1000_get_hw_control(adapter);
4479
4480         return 0;
4481 }
4482 #endif
4483 #ifdef CONFIG_NET_POLL_CONTROLLER
4484 /*
4485  * Polling 'interrupt' - used by things like netconsole to send skbs
4486  * without having to re-enable interrupts. It's not called while
4487  * the interrupt routine is executing.
4488  */
4489 static void
4490 e1000_netpoll(struct net_device *netdev)
4491 {
4492         struct e1000_adapter *adapter = netdev_priv(netdev);
4493         disable_irq(adapter->pdev->irq);
4494         e1000_intr(adapter->pdev->irq, netdev, NULL);
4495         e1000_clean_tx_irq(adapter, adapter->tx_ring);
4496 #ifndef CONFIG_E1000_NAPI
4497         adapter->clean_rx(adapter, adapter->rx_ring);
4498 #endif
4499         enable_irq(adapter->pdev->irq);
4500 }
4501 #endif
4502
4503 /* e1000_main.c */