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  * 6.3.9        12/16/2005
  33  *   o incorporate fix for recycled skbs from IBM LTC
  34  * 6.3.7        11/18/2005
  35  *   o Honor eeprom setting for enabling/disabling Wake On Lan
  36  * 6.3.5        11/17/2005
  37  *   o Fix memory leak in rx ring handling for PCI Express adapters
  38  * 6.3.4        11/8/05
  39  *   o Patch from Jesper Juhl to remove redundant NULL checks for kfree
  40  * 6.3.2        9/20/05
  41  *   o Render logic that sets/resets DRV_LOAD as inline functions to
  42  *     avoid code replication. If f/w is AMT then set DRV_LOAD only when
  43  *     network interface is open.
  44  *   o Handle DRV_LOAD set/reset in cases where AMT uses VLANs.
  45  *   o Adjust PBA partioning for Jumbo frames using MTU size and not
  46  *     rx_buffer_len
  47  * 6.3.1        9/19/05
  48  *   o Use adapter->tx_timeout_factor in Tx Hung Detect logic
  49        (e1000_clean_tx_irq)
  50  *   o Support for 8086:10B5 device (Quad Port)
  51  * 6.2.14       9/15/05
  52  *   o In AMT enabled configurations, set/reset DRV_LOAD bit on interface
  53  *     open/close
  54  * 6.2.13       9/14/05
  55  *   o Invoke e1000_check_mng_mode only for 8257x controllers since it
  56  *     accesses the FWSM that is not supported in other controllers
  57  * 6.2.12       9/9/05
  58  *   o Add support for device id E1000_DEV_ID_82546GB_QUAD_COPPER
  59  *   o set RCTL:SECRC only for controllers newer than 82543.
  60  *   o When the n/w interface comes down reset DRV_LOAD bit to notify f/w.
  61  *     This code was moved from e1000_remove to e1000_close
  62  * 6.2.10       9/6/05
  63  *   o Fix error in updating RDT in el1000_alloc_rx_buffers[_ps] -- one off.
  64  *   o Enable fc by default on 82573 controllers (do not read eeprom)
  65  *   o Fix rx_errors statistic not to include missed_packet_count
  66  *   o Fix rx_dropped statistic not to include missed_packet_count
  67        (Padraig Brady)
  68  * 6.2.9        8/30/05
  69  *   o Remove call to update statistics from the controller ib e1000_get_stats
  70  * 6.2.8        8/30/05
  71  *   o Improved algorithm for rx buffer allocation/rdt update
  72  *   o Flow control watermarks relative to rx PBA size
  73  *   o Simplified 'Tx Hung' detect logic
  74  * 6.2.7        8/17/05
  75  *   o Report rx buffer allocation failures and tx timeout counts in stats
  76  * 6.2.6        8/16/05
  77  *   o Implement workaround for controller erratum -- linear non-tso packet
  78  *     following a TSO gets written back prematurely
  79  * 6.2.5        8/15/05
  80  *   o Set netdev->tx_queue_len based on link speed/duplex settings.
  81  *   o Fix net_stats.rx_fifo_errors <p@draigBrady.com>
  82  *   o Do not power off PHY if SoL/IDER session is active
  83  * 6.2.4        8/10/05
  84  *   o Fix loopback test setup/cleanup for 82571/3 controllers
  85  *   o Fix parsing of outgoing packets (e1000_transfer_dhcp_info) to treat
  86  *     all packets as raw
  87  *   o Prevent operations that will cause the PHY to be reset if SoL/IDER
  88  *     sessions are active and log a message
  89  * 6.2.2        7/21/05
  90  *   o used fixed size descriptors for all MTU sizes, reduces memory load
  91  * 6.1.2        4/13/05
  92  *   o Fixed ethtool diagnostics
  93  *   o Enabled flow control to take default eeprom settings
  94  *   o Added stats_lock around e1000_read_phy_reg commands to avoid concurrent
  95  *     calls, one from mii_ioctl and other from within update_stats while
  96  *     processing MIIREG ioctl.
  97  */
  98
  99 char e1000_driver_name[] = "e1000";
 100 static char e1000_driver_string[] = "Intel(R) PRO/1000 Network Driver";
 101 #ifndef CONFIG_E1000_NAPI
 102 #define DRIVERNAPI
 103 #else
 104 #define DRIVERNAPI "-NAPI"
 105 #endif
 106 #define DRV_VERSION "6.3.9-k4"DRIVERNAPI
 107 char e1000_driver_version[] = DRV_VERSION;
 108 static char e1000_copyright[] = "Copyright (c) 1999-2005 Intel Corporation.";
 109
 110 /* e1000_pci_tbl - PCI Device ID Table
 111  *
 112  * Last entry must be all 0s
 113  *
 114  * Macro expands to...
 115  *   {PCI_DEVICE(PCI_VENDOR_ID_INTEL, device_id)}
 116  */
 117 static struct pci_device_id e1000_pci_tbl[] = {
 118         INTEL_E1000_ETHERNET_DEVICE(0x1000),
 119         INTEL_E1000_ETHERNET_DEVICE(0x1001),
 120         INTEL_E1000_ETHERNET_DEVICE(0x1004),
 121         INTEL_E1000_ETHERNET_DEVICE(0x1008),
 122         INTEL_E1000_ETHERNET_DEVICE(0x1009),
 123         INTEL_E1000_ETHERNET_DEVICE(0x100C),
 124         INTEL_E1000_ETHERNET_DEVICE(0x100D),
 125         INTEL_E1000_ETHERNET_DEVICE(0x100E),
 126         INTEL_E1000_ETHERNET_DEVICE(0x100F),
 127         INTEL_E1000_ETHERNET_DEVICE(0x1010),
 128         INTEL_E1000_ETHERNET_DEVICE(0x1011),
 129         INTEL_E1000_ETHERNET_DEVICE(0x1012),
 130         INTEL_E1000_ETHERNET_DEVICE(0x1013),
 131         INTEL_E1000_ETHERNET_DEVICE(0x1014),
 132         INTEL_E1000_ETHERNET_DEVICE(0x1015),
 133         INTEL_E1000_ETHERNET_DEVICE(0x1016),
 134         INTEL_E1000_ETHERNET_DEVICE(0x1017),
 135         INTEL_E1000_ETHERNET_DEVICE(0x1018),
 136         INTEL_E1000_ETHERNET_DEVICE(0x1019),
 137         INTEL_E1000_ETHERNET_DEVICE(0x101A),
 138         INTEL_E1000_ETHERNET_DEVICE(0x101D),
 139         INTEL_E1000_ETHERNET_DEVICE(0x101E),
 140         INTEL_E1000_ETHERNET_DEVICE(0x1026),
 141         INTEL_E1000_ETHERNET_DEVICE(0x1027),
 142         INTEL_E1000_ETHERNET_DEVICE(0x1028),
 143         INTEL_E1000_ETHERNET_DEVICE(0x105E),
 144         INTEL_E1000_ETHERNET_DEVICE(0x105F),
 145         INTEL_E1000_ETHERNET_DEVICE(0x1060),
 146         INTEL_E1000_ETHERNET_DEVICE(0x1075),
 147         INTEL_E1000_ETHERNET_DEVICE(0x1076),
 148         INTEL_E1000_ETHERNET_DEVICE(0x1077),
 149         INTEL_E1000_ETHERNET_DEVICE(0x1078),
 150         INTEL_E1000_ETHERNET_DEVICE(0x1079),
 151         INTEL_E1000_ETHERNET_DEVICE(0x107A),
 152         INTEL_E1000_ETHERNET_DEVICE(0x107B),
 153         INTEL_E1000_ETHERNET_DEVICE(0x107C),
 154         INTEL_E1000_ETHERNET_DEVICE(0x107D),
 155         INTEL_E1000_ETHERNET_DEVICE(0x107E),
 156         INTEL_E1000_ETHERNET_DEVICE(0x107F),
 157         INTEL_E1000_ETHERNET_DEVICE(0x108A),
 158         INTEL_E1000_ETHERNET_DEVICE(0x108B),
 159         INTEL_E1000_ETHERNET_DEVICE(0x108C),
 160         INTEL_E1000_ETHERNET_DEVICE(0x1099),
 161         INTEL_E1000_ETHERNET_DEVICE(0x109A),
 162         INTEL_E1000_ETHERNET_DEVICE(0x10B5),
 163         /* required last entry */
 164         {0,}
 165 };
 166
 167 MODULE_DEVICE_TABLE(pci, e1000_pci_tbl);
 168
 169 int e1000_up(struct e1000_adapter *adapter);
 170 void e1000_down(struct e1000_adapter *adapter);
 171 void e1000_reset(struct e1000_adapter *adapter);
 172 int e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx);
 173 int e1000_setup_all_tx_resources(struct e1000_adapter *adapter);
 174 int e1000_setup_all_rx_resources(struct e1000_adapter *adapter);
 175 void e1000_free_all_tx_resources(struct e1000_adapter *adapter);
 176 void e1000_free_all_rx_resources(struct e1000_adapter *adapter);
 177 static int e1000_setup_tx_resources(struct e1000_adapter *adapter,
 178                                     struct e1000_tx_ring *txdr);
 179 static int e1000_setup_rx_resources(struct e1000_adapter *adapter,
 180                                     struct e1000_rx_ring *rxdr);
 181 static void e1000_free_tx_resources(struct e1000_adapter *adapter,
 182                                     struct e1000_tx_ring *tx_ring);
 183 static void e1000_free_rx_resources(struct e1000_adapter *adapter,
 184                                     struct e1000_rx_ring *rx_ring);
 185 void e1000_update_stats(struct e1000_adapter *adapter);
 186
 187 /* Local Function Prototypes */
 188
 189 static int e1000_init_module(void);
 190 static void e1000_exit_module(void);
 191 static int e1000_probe(struct pci_dev *pdev, const struct pci_device_id *ent);
 192 static void __devexit e1000_remove(struct pci_dev *pdev);
 193 static int e1000_alloc_queues(struct e1000_adapter *adapter);
 194 #ifdef CONFIG_E1000_MQ
 195 static void e1000_setup_queue_mapping(struct e1000_adapter *adapter);
 196 #endif
 197 static int e1000_sw_init(struct e1000_adapter *adapter);
 198 static int e1000_open(struct net_device *netdev);
 199 static int e1000_close(struct net_device *netdev);
 200 static void e1000_configure_tx(struct e1000_adapter *adapter);
 201 static void e1000_configure_rx(struct e1000_adapter *adapter);
 202 static void e1000_setup_rctl(struct e1000_adapter *adapter);
 203 static void e1000_clean_all_tx_rings(struct e1000_adapter *adapter);
 204 static void e1000_clean_all_rx_rings(struct e1000_adapter *adapter);
 205 static void e1000_clean_tx_ring(struct e1000_adapter *adapter,
 206                                 struct e1000_tx_ring *tx_ring);
 207 static void e1000_clean_rx_ring(struct e1000_adapter *adapter,
 208                                 struct e1000_rx_ring *rx_ring);
 209 static void e1000_set_multi(struct net_device *netdev);
 210 static void e1000_update_phy_info(unsigned long data);
 211 static void e1000_watchdog(unsigned long data);
 212 static void e1000_watchdog_task(struct e1000_adapter *adapter);
 213 static void e1000_82547_tx_fifo_stall(unsigned long data);
 214 static int e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev);
 215 static struct net_device_stats * e1000_get_stats(struct net_device *netdev);
 216 static int e1000_change_mtu(struct net_device *netdev, int new_mtu);
 217 static int e1000_set_mac(struct net_device *netdev, void *p);
 218 static irqreturn_t e1000_intr(int irq, void *data, struct pt_regs *regs);
 219 static boolean_t e1000_clean_tx_irq(struct e1000_adapter *adapter,
 220                                     struct e1000_tx_ring *tx_ring);
 221 #ifdef CONFIG_E1000_NAPI
 222 static int e1000_clean(struct net_device *poll_dev, int *budget);
 223 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
 224                                     struct e1000_rx_ring *rx_ring,
 225                                     int *work_done, int work_to_do);
 226 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 227                                        struct e1000_rx_ring *rx_ring,
 228                                        int *work_done, int work_to_do);
 229 #else
 230 static boolean_t e1000_clean_rx_irq(struct e1000_adapter *adapter,
 231                                     struct e1000_rx_ring *rx_ring);
 232 static boolean_t e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
 233                                        struct e1000_rx_ring *rx_ring);
 234 #endif
 235 static void e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
 236                                    struct e1000_rx_ring *rx_ring,
 237                                    int cleaned_count);
 238 static void e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
 239                                       struct e1000_rx_ring *rx_ring,
 240                                       int cleaned_count);
 241 static int e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd);
 242 static int e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr,
 243                            int cmd);
 244 void e1000_set_ethtool_ops(struct net_device *netdev);
 245 static void e1000_enter_82542_rst(struct e1000_adapter *adapter);
 246 static void e1000_leave_82542_rst(struct e1000_adapter *adapter);
 247 static void e1000_tx_timeout(struct net_device *dev);
 248 static void e1000_tx_timeout_task(struct net_device *dev);
 249 static void e1000_smartspeed(struct e1000_adapter *adapter);
 250 static inline int e1000_82547_fifo_workaround(struct e1000_adapter *adapter,
 251                                               struct sk_buff *skb);
 252
 253 static void e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp);
 254 static void e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid);
 255 static void e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid);
 256 static void e1000_restore_vlan(struct e1000_adapter *adapter);
 257
 258 #ifdef CONFIG_PM
 259 static int e1000_suspend(struct pci_dev *pdev, pm_message_t state);
 260 static int e1000_resume(struct pci_dev *pdev);
 261 #endif
 262
 263 #ifdef CONFIG_NET_POLL_CONTROLLER
 264 /* for netdump / net console */
 265 static void e1000_netpoll (struct net_device *netdev);
 266 #endif
 267
 268 #ifdef CONFIG_E1000_MQ
 269 /* for multiple Rx queues */
 270 void e1000_rx_schedule(void *data);
 271 #endif
 272
 273 /* Exported from other modules */
 274
 275 extern void e1000_check_options(struct e1000_adapter *adapter);
 276
 277 static struct pci_driver e1000_driver = {
 278         .name     = e1000_driver_name,
 279         .id_table = e1000_pci_tbl,
 280         .probe    = e1000_probe,
 281         .remove   = __devexit_p(e1000_remove),
 282         /* Power Managment Hooks */
 283 #ifdef CONFIG_PM
 284         .suspend  = e1000_suspend,
 285         .resume   = e1000_resume
 286 #endif
 287 };
 288
 289 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
 290 MODULE_DESCRIPTION("Intel(R) PRO/1000 Network Driver");
 291 MODULE_LICENSE("GPL");
 292 MODULE_VERSION(DRV_VERSION);
 293
 294 static int debug = NETIF_MSG_DRV | NETIF_MSG_PROBE;
 295 module_param(debug, int, 0);
 296 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
 297
 298 /**
 299  * e1000_init_module - Driver Registration Routine
 300  *
 301  * e1000_init_module is the first routine called when the driver is
 302  * loaded. All it does is register with the PCI subsystem.
 303  **/
 304
 305 static int __init
 306 e1000_init_module(void)
 307 {
 308         int ret;
 309         printk(KERN_INFO "%s - version %s\n",
 310                e1000_driver_string, e1000_driver_version);
 311
 312         printk(KERN_INFO "%s\n", e1000_copyright);
 313
 314         ret = pci_module_init(&e1000_driver);
 315
 316         return ret;
 317 }
 318
 319 module_init(e1000_init_module);
 320
 321 /**
 322  * e1000_exit_module - Driver Exit Cleanup Routine
 323  *
 324  * e1000_exit_module is called just before the driver is removed
 325  * from memory.
 326  **/
 327
 328 static void __exit
 329 e1000_exit_module(void)
 330 {
 331         pci_unregister_driver(&e1000_driver);
 332 }
 333
 334 module_exit(e1000_exit_module);
 335
 336 /**
 337  * e1000_irq_disable - Mask off interrupt generation on the NIC
 338  * @adapter: board private structure
 339  **/
 340
 341 static inline void
 342 e1000_irq_disable(struct e1000_adapter *adapter)
 343 {
 344         atomic_inc(&adapter->irq_sem);
 345         E1000_WRITE_REG(&adapter->hw, IMC, ~0);
 346         E1000_WRITE_FLUSH(&adapter->hw);
 347         synchronize_irq(adapter->pdev->irq);
 348 }
 349
 350 /**
 351  * e1000_irq_enable - Enable default interrupt generation settings
 352  * @adapter: board private structure
 353  **/
 354
 355 static inline void
 356 e1000_irq_enable(struct e1000_adapter *adapter)
 357 {
 358         if (likely(atomic_dec_and_test(&adapter->irq_sem))) {
 359                 E1000_WRITE_REG(&adapter->hw, IMS, IMS_ENABLE_MASK);
 360                 E1000_WRITE_FLUSH(&adapter->hw);
 361         }
 362 }
 363
 364 static void
 365 e1000_update_mng_vlan(struct e1000_adapter *adapter)
 366 {
 367         struct net_device *netdev = adapter->netdev;
 368         uint16_t vid = adapter->hw.mng_cookie.vlan_id;
 369         uint16_t old_vid = adapter->mng_vlan_id;
 370         if (adapter->vlgrp) {
 371                 if (!adapter->vlgrp->vlan_devices[vid]) {
 372                         if (adapter->hw.mng_cookie.status &
 373                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) {
 374                                 e1000_vlan_rx_add_vid(netdev, vid);
 375                                 adapter->mng_vlan_id = vid;
 376                         } else
 377                                 adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
 378
 379                         if ((old_vid != (uint16_t)E1000_MNG_VLAN_NONE) &&
 380                                         (vid != old_vid) &&
 381                                         !adapter->vlgrp->vlan_devices[old_vid])
 382                                 e1000_vlan_rx_kill_vid(netdev, old_vid);
 383                 }
 384         }
 385 }
 386
 387 /**
 388  * e1000_release_hw_control - release control of the h/w to f/w
 389  * @adapter: address of board private structure
 390  *
 391  * e1000_release_hw_control resets {CTRL_EXT|FWSM}:DRV_LOAD bit.
 392  * For ASF and Pass Through versions of f/w this means that the
 393  * driver is no longer loaded. For AMT version (only with 82573) i
 394  * of the f/w this means that the netowrk i/f is closed.
 395  *
 396  **/
 397
 398 static inline void
 399 e1000_release_hw_control(struct e1000_adapter *adapter)
 400 {
 401         uint32_t ctrl_ext;
 402         uint32_t swsm;
 403
 404         /* Let firmware taken over control of h/w */
 405         switch (adapter->hw.mac_type) {
 406         case e1000_82571:
 407         case e1000_82572:
 408                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
 409                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
 410                                 ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 411                 break;
 412         case e1000_82573:
 413                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
 414                 E1000_WRITE_REG(&adapter->hw, SWSM,
 415                                 swsm & ~E1000_SWSM_DRV_LOAD);
 416         default:
 417                 break;
 418         }
 419 }
 420
 421 /**
 422  * e1000_get_hw_control - get control of the h/w from f/w
 423  * @adapter: address of board private structure
 424  *
 425  * e1000_get_hw_control sets {CTRL_EXT|FWSM}:DRV_LOAD bit.
 426  * For ASF and Pass Through versions of f/w this means that
 427  * the driver is loaded. For AMT version (only with 82573)
 428  * of the f/w this means that the netowrk i/f is open.
 429  *
 430  **/
 431
 432 static inline void
 433 e1000_get_hw_control(struct e1000_adapter *adapter)
 434 {
 435         uint32_t ctrl_ext;
 436         uint32_t swsm;
 437         /* Let firmware know the driver has taken over */
 438         switch (adapter->hw.mac_type) {
 439         case e1000_82571:
 440         case e1000_82572:
 441                 ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
 442                 E1000_WRITE_REG(&adapter->hw, CTRL_EXT,
 443                                 ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 444                 break;
 445         case e1000_82573:
 446                 swsm = E1000_READ_REG(&adapter->hw, SWSM);
 447                 E1000_WRITE_REG(&adapter->hw, SWSM,
 448                                 swsm | E1000_SWSM_DRV_LOAD);
 449                 break;
 450         default:
 451                 break;
 452         }
 453 }
 454
 455 int
 456 e1000_up(struct e1000_adapter *adapter)
 457 {
 458         struct net_device *netdev = adapter->netdev;
 459         int i, err;
 460
 461         /* hardware has been reset, we need to reload some things */
 462
 463         /* Reset the PHY if it was previously powered down */
 464         if (adapter->hw.media_type == e1000_media_type_copper) {
 465                 uint16_t mii_reg;
 466                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
 467                 if (mii_reg & MII_CR_POWER_DOWN)
 468                         e1000_phy_reset(&adapter->hw);
 469         }
 470
 471         e1000_set_multi(netdev);
 472
 473         e1000_restore_vlan(adapter);
 474
 475         e1000_configure_tx(adapter);
 476         e1000_setup_rctl(adapter);
 477         e1000_configure_rx(adapter);
 478         /* call E1000_DESC_UNUSED which always leaves
 479          * at least 1 descriptor unused to make sure
 480          * next_to_use != next_to_clean */
 481         for (i = 0; i < adapter->num_rx_queues; i++) {
 482                 struct e1000_rx_ring *ring = &adapter->rx_ring[i];
 483                 adapter->alloc_rx_buf(adapter, ring,
 484                                       E1000_DESC_UNUSED(ring));
 485         }
 486
 487 #ifdef CONFIG_PCI_MSI
 488         if (adapter->hw.mac_type > e1000_82547_rev_2) {
 489                 adapter->have_msi = TRUE;
 490                 if ((err = pci_enable_msi(adapter->pdev))) {
 491                         DPRINTK(PROBE, ERR,
 492                          "Unable to allocate MSI interrupt Error: %d\n", err);
 493                         adapter->have_msi = FALSE;
 494                 }
 495         }
 496 #endif
 497         if ((err = request_irq(adapter->pdev->irq, &e1000_intr,
 498                               SA_SHIRQ | SA_SAMPLE_RANDOM,
 499                               netdev->name, netdev))) {
 500                 DPRINTK(PROBE, ERR,
 501                     "Unable to allocate interrupt Error: %d\n", err);
 502                 return err;
 503         }
 504
 505 #ifdef CONFIG_E1000_MQ
 506         e1000_setup_queue_mapping(adapter);
 507 #endif
 508
 509         adapter->tx_queue_len = netdev->tx_queue_len;
 510
 511         mod_timer(&adapter->watchdog_timer, jiffies);
 512
 513 #ifdef CONFIG_E1000_NAPI
 514         netif_poll_enable(netdev);
 515 #endif
 516         e1000_irq_enable(adapter);
 517
 518         return 0;
 519 }
 520
 521 void
 522 e1000_down(struct e1000_adapter *adapter)
 523 {
 524         struct net_device *netdev = adapter->netdev;
 525         boolean_t mng_mode_enabled = (adapter->hw.mac_type >= e1000_82571) &&
 526                                      e1000_check_mng_mode(&adapter->hw);
 527
 528         e1000_irq_disable(adapter);
 529 #ifdef CONFIG_E1000_MQ
 530         while (atomic_read(&adapter->rx_sched_call_data.count) != 0);
 531 #endif
 532         free_irq(adapter->pdev->irq, netdev);
 533 #ifdef CONFIG_PCI_MSI
 534         if (adapter->hw.mac_type > e1000_82547_rev_2 &&
 535            adapter->have_msi == TRUE)
 536                 pci_disable_msi(adapter->pdev);
 537 #endif
 538         del_timer_sync(&adapter->tx_fifo_stall_timer);
 539         del_timer_sync(&adapter->watchdog_timer);
 540         del_timer_sync(&adapter->phy_info_timer);
 541
 542 #ifdef CONFIG_E1000_NAPI
 543         netif_poll_disable(netdev);
 544 #endif
 545         netdev->tx_queue_len = adapter->tx_queue_len;
 546         adapter->link_speed = 0;
 547         adapter->link_duplex = 0;
 548         netif_carrier_off(netdev);
 549         netif_stop_queue(netdev);
 550
 551         e1000_reset(adapter);
 552         e1000_clean_all_tx_rings(adapter);
 553         e1000_clean_all_rx_rings(adapter);
 554
 555         /* Power down the PHY so no link is implied when interface is down *
 556          * The PHY cannot be powered down if any of the following is TRUE *
 557          * (a) WoL is enabled
 558          * (b) AMT is active
 559          * (c) SoL/IDER session is active */
 560         if (!adapter->wol && adapter->hw.mac_type >= e1000_82540 &&
 561            adapter->hw.media_type == e1000_media_type_copper &&
 562            !(E1000_READ_REG(&adapter->hw, MANC) & E1000_MANC_SMBUS_EN) &&
 563            !mng_mode_enabled &&
 564            !e1000_check_phy_reset_block(&adapter->hw)) {
 565                 uint16_t mii_reg;
 566                 e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &mii_reg);
 567                 mii_reg |= MII_CR_POWER_DOWN;
 568                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL, mii_reg);
 569                 mdelay(1);
 570         }
 571 }
 572
 573 void
 574 e1000_reset(struct e1000_adapter *adapter)
 575 {
 576         uint32_t pba, manc;
 577         uint16_t fc_high_water_mark = E1000_FC_HIGH_DIFF;
 578
 579         /* Repartition Pba for greater than 9k mtu
 580          * To take effect CTRL.RST is required.
 581          */
 582
 583         switch (adapter->hw.mac_type) {
 584         case e1000_82547:
 585         case e1000_82547_rev_2:
 586                 pba = E1000_PBA_30K;
 587                 break;
 588         case e1000_82571:
 589         case e1000_82572:
 590                 pba = E1000_PBA_38K;
 591                 break;
 592         case e1000_82573:
 593                 pba = E1000_PBA_12K;
 594                 break;
 595         default:
 596                 pba = E1000_PBA_48K;
 597                 break;
 598         }
 599
 600         if ((adapter->hw.mac_type != e1000_82573) &&
 601            (adapter->netdev->mtu > E1000_RXBUFFER_8192))
 602                 pba -= 8; /* allocate more FIFO for Tx */
 603
 604
 605         if (adapter->hw.mac_type == e1000_82547) {
 606                 adapter->tx_fifo_head = 0;
 607                 adapter->tx_head_addr = pba << E1000_TX_HEAD_ADDR_SHIFT;
 608                 adapter->tx_fifo_size =
 609                         (E1000_PBA_40K - pba) << E1000_PBA_BYTES_SHIFT;
 610                 atomic_set(&adapter->tx_fifo_stall, 0);
 611         }
 612
 613         E1000_WRITE_REG(&adapter->hw, PBA, pba);
 614
 615         /* flow control settings */
 616         /* Set the FC high water mark to 90% of the FIFO size.
 617          * Required to clear last 3 LSB */
 618         fc_high_water_mark = ((pba * 9216)/10) & 0xFFF8;
 619
 620         adapter->hw.fc_high_water = fc_high_water_mark;
 621         adapter->hw.fc_low_water = fc_high_water_mark - 8;
 622         adapter->hw.fc_pause_time = E1000_FC_PAUSE_TIME;
 623         adapter->hw.fc_send_xon = 1;
 624         adapter->hw.fc = adapter->hw.original_fc;
 625
 626         /* Allow time for pending master requests to run */
 627         e1000_reset_hw(&adapter->hw);
 628         if (adapter->hw.mac_type >= e1000_82544)
 629                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
 630         if (e1000_init_hw(&adapter->hw))
 631                 DPRINTK(PROBE, ERR, "Hardware Error\n");
 632         e1000_update_mng_vlan(adapter);
 633         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
 634         E1000_WRITE_REG(&adapter->hw, VET, ETHERNET_IEEE_VLAN_TYPE);
 635
 636         e1000_reset_adaptive(&adapter->hw);
 637         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
 638         if (adapter->en_mng_pt) {
 639                 manc = E1000_READ_REG(&adapter->hw, MANC);
 640                 manc |= (E1000_MANC_ARP_EN | E1000_MANC_EN_MNG2HOST);
 641                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
 642         }
 643 }
 644
 645 /**
 646  * e1000_probe - Device Initialization Routine
 647  * @pdev: PCI device information struct
 648  * @ent: entry in e1000_pci_tbl
 649  *
 650  * Returns 0 on success, negative on failure
 651  *
 652  * e1000_probe initializes an adapter identified by a pci_dev structure.
 653  * The OS initialization, configuring of the adapter private structure,
 654  * and a hardware reset occur.
 655  **/
 656
 657 static int __devinit
 658 e1000_probe(struct pci_dev *pdev,
 659             const struct pci_device_id *ent)
 660 {
 661         struct net_device *netdev;
 662         struct e1000_adapter *adapter;
 663         unsigned long mmio_start, mmio_len;
 664
 665         static int cards_found = 0;
 666         int i, err, pci_using_dac;
 667         uint16_t eeprom_data;
 668         uint16_t eeprom_apme_mask = E1000_EEPROM_APME;
 669         if ((err = pci_enable_device(pdev)))
 670                 return err;
 671
 672         if (!(err = pci_set_dma_mask(pdev, DMA_64BIT_MASK))) {
 673                 pci_using_dac = 1;
 674         } else {
 675                 if ((err = pci_set_dma_mask(pdev, DMA_32BIT_MASK))) {
 676                         E1000_ERR("No usable DMA configuration, aborting\n");
 677                         return err;
 678                 }
 679                 pci_using_dac = 0;
 680         }
 681
 682         if ((err = pci_request_regions(pdev, e1000_driver_name)))
 683                 return err;
 684
 685         pci_set_master(pdev);
 686
 687         netdev = alloc_etherdev(sizeof(struct e1000_adapter));
 688         if (!netdev) {
 689                 err = -ENOMEM;
 690                 goto err_alloc_etherdev;
 691         }
 692
 693         SET_MODULE_OWNER(netdev);
 694         SET_NETDEV_DEV(netdev, &pdev->dev);
 695
 696         pci_set_drvdata(pdev, netdev);
 697         adapter = netdev_priv(netdev);
 698         adapter->netdev = netdev;
 699         adapter->pdev = pdev;
 700         adapter->hw.back = adapter;
 701         adapter->msg_enable = (1 << debug) - 1;
 702
 703         mmio_start = pci_resource_start(pdev, BAR_0);
 704         mmio_len = pci_resource_len(pdev, BAR_0);
 705
 706         adapter->hw.hw_addr = ioremap(mmio_start, mmio_len);
 707         if (!adapter->hw.hw_addr) {
 708                 err = -EIO;
 709                 goto err_ioremap;
 710         }
 711
 712         for (i = BAR_1; i <= BAR_5; i++) {
 713                 if (pci_resource_len(pdev, i) == 0)
 714                         continue;
 715                 if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
 716                         adapter->hw.io_base = pci_resource_start(pdev, i);
 717                         break;
 718                 }
 719         }
 720
 721         netdev->open = &e1000_open;
 722         netdev->stop = &e1000_close;
 723         netdev->hard_start_xmit = &e1000_xmit_frame;
 724         netdev->get_stats = &e1000_get_stats;
 725         netdev->set_multicast_list = &e1000_set_multi;
 726         netdev->set_mac_address = &e1000_set_mac;
 727         netdev->change_mtu = &e1000_change_mtu;
 728         netdev->do_ioctl = &e1000_ioctl;
 729         e1000_set_ethtool_ops(netdev);
 730         netdev->tx_timeout = &e1000_tx_timeout;
 731         netdev->watchdog_timeo = 5 * HZ;
 732 #ifdef CONFIG_E1000_NAPI
 733         netdev->poll = &e1000_clean;
 734         netdev->weight = 64;
 735 #endif
 736         netdev->vlan_rx_register = e1000_vlan_rx_register;
 737         netdev->vlan_rx_add_vid = e1000_vlan_rx_add_vid;
 738         netdev->vlan_rx_kill_vid = e1000_vlan_rx_kill_vid;
 739 #ifdef CONFIG_NET_POLL_CONTROLLER
 740         netdev->poll_controller = e1000_netpoll;
 741 #endif
 742         strcpy(netdev->name, pci_name(pdev));
 743
 744         netdev->mem_start = mmio_start;
 745         netdev->mem_end = mmio_start + mmio_len;
 746         netdev->base_addr = adapter->hw.io_base;
 747
 748         adapter->bd_number = cards_found;
 749
 750         /* setup the private structure */
 751
 752         if ((err = e1000_sw_init(adapter)))
 753                 goto err_sw_init;
 754
 755         if ((err = e1000_check_phy_reset_block(&adapter->hw)))
 756                 DPRINTK(PROBE, INFO, "PHY reset is blocked due to SOL/IDER session.\n");
 757
 758         if (adapter->hw.mac_type >= e1000_82543) {
 759                 netdev->features = NETIF_F_SG |
 760                                    NETIF_F_HW_CSUM |
 761                                    NETIF_F_HW_VLAN_TX |
 762                                    NETIF_F_HW_VLAN_RX |
 763                                    NETIF_F_HW_VLAN_FILTER;
 764         }
 765
 766 #ifdef NETIF_F_TSO
 767         if ((adapter->hw.mac_type >= e1000_82544) &&
 768            (adapter->hw.mac_type != e1000_82547))
 769                 netdev->features |= NETIF_F_TSO;
 770
 771 #ifdef NETIF_F_TSO_IPV6
 772         if (adapter->hw.mac_type > e1000_82547_rev_2)
 773                 netdev->features |= NETIF_F_TSO_IPV6;
 774 #endif
 775 #endif
 776         if (pci_using_dac)
 777                 netdev->features |= NETIF_F_HIGHDMA;
 778
 779         /* hard_start_xmit is safe against parallel locking */
 780         netdev->features |= NETIF_F_LLTX;
 781
 782         adapter->en_mng_pt = e1000_enable_mng_pass_thru(&adapter->hw);
 783
 784         /* before reading the EEPROM, reset the controller to
 785          * put the device in a known good starting state */
 786
 787         e1000_reset_hw(&adapter->hw);
 788
 789         /* make sure the EEPROM is good */
 790
 791         if (e1000_validate_eeprom_checksum(&adapter->hw) < 0) {
 792                 DPRINTK(PROBE, ERR, "The EEPROM Checksum Is Not Valid\n");
 793                 err = -EIO;
 794                 goto err_eeprom;
 795         }
 796
 797         /* copy the MAC address out of the EEPROM */
 798
 799         if (e1000_read_mac_addr(&adapter->hw))
 800                 DPRINTK(PROBE, ERR, "EEPROM Read Error\n");
 801         memcpy(netdev->dev_addr, adapter->hw.mac_addr, netdev->addr_len);
 802         memcpy(netdev->perm_addr, adapter->hw.mac_addr, netdev->addr_len);
 803
 804         if (!is_valid_ether_addr(netdev->perm_addr)) {
 805                 DPRINTK(PROBE, ERR, "Invalid MAC Address\n");
 806                 err = -EIO;
 807                 goto err_eeprom;
 808         }
 809
 810         e1000_read_part_num(&adapter->hw, &(adapter->part_num));
 811
 812         e1000_get_bus_info(&adapter->hw);
 813
 814         init_timer(&adapter->tx_fifo_stall_timer);
 815         adapter->tx_fifo_stall_timer.function = &e1000_82547_tx_fifo_stall;
 816         adapter->tx_fifo_stall_timer.data = (unsigned long) adapter;
 817
 818         init_timer(&adapter->watchdog_timer);
 819         adapter->watchdog_timer.function = &e1000_watchdog;
 820         adapter->watchdog_timer.data = (unsigned long) adapter;
 821
 822         INIT_WORK(&adapter->watchdog_task,
 823                 (void (*)(void *))e1000_watchdog_task, adapter);
 824
 825         init_timer(&adapter->phy_info_timer);
 826         adapter->phy_info_timer.function = &e1000_update_phy_info;
 827         adapter->phy_info_timer.data = (unsigned long) adapter;
 828
 829         INIT_WORK(&adapter->tx_timeout_task,
 830                 (void (*)(void *))e1000_tx_timeout_task, netdev);
 831
 832         /* we're going to reset, so assume we have no link for now */
 833
 834         netif_carrier_off(netdev);
 835         netif_stop_queue(netdev);
 836
 837         e1000_check_options(adapter);
 838
 839         /* Initial Wake on LAN setting
 840          * If APM wake is enabled in the EEPROM,
 841          * enable the ACPI Magic Packet filter
 842          */
 843
 844         switch (adapter->hw.mac_type) {
 845         case e1000_82542_rev2_0:
 846         case e1000_82542_rev2_1:
 847         case e1000_82543:
 848                 break;
 849         case e1000_82544:
 850                 e1000_read_eeprom(&adapter->hw,
 851                         EEPROM_INIT_CONTROL2_REG, 1, &eeprom_data);
 852                 eeprom_apme_mask = E1000_EEPROM_82544_APM;
 853                 break;
 854         case e1000_82546:
 855         case e1000_82546_rev_3:
 856         case e1000_82571:
 857                 if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_FUNC_1){
 858                         e1000_read_eeprom(&adapter->hw,
 859                                 EEPROM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
 860                         break;
 861                 }
 862                 /* Fall Through */
 863         default:
 864                 e1000_read_eeprom(&adapter->hw,
 865                         EEPROM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
 866                 break;
 867         }
 868         if (eeprom_data & eeprom_apme_mask)
 869                 adapter->wol |= E1000_WUFC_MAG;
 870
 871         /* print bus type/speed/width info */
 872         {
 873         struct e1000_hw *hw = &adapter->hw;
 874         DPRINTK(PROBE, INFO, "(PCI%s:%s:%s) ",
 875                 ((hw->bus_type == e1000_bus_type_pcix) ? "-X" :
 876                  (hw->bus_type == e1000_bus_type_pci_express ? " Express":"")),
 877                 ((hw->bus_speed == e1000_bus_speed_2500) ? "2.5Gb/s" :
 878                  (hw->bus_speed == e1000_bus_speed_133) ? "133MHz" :
 879                  (hw->bus_speed == e1000_bus_speed_120) ? "120MHz" :
 880                  (hw->bus_speed == e1000_bus_speed_100) ? "100MHz" :
 881                  (hw->bus_speed == e1000_bus_speed_66) ? "66MHz" : "33MHz"),
 882                 ((hw->bus_width == e1000_bus_width_64) ? "64-bit" :
 883                  (hw->bus_width == e1000_bus_width_pciex_4) ? "Width x4" :
 884                  (hw->bus_width == e1000_bus_width_pciex_1) ? "Width x1" :
 885                  "32-bit"));
 886         }
 887
 888         for (i = 0; i < 6; i++)
 889                 printk("%2.2x%c", netdev->dev_addr[i], i == 5 ? '\n' : ':');
 890
 891         /* reset the hardware with the new settings */
 892         e1000_reset(adapter);
 893
 894         /* If the controller is 82573 and f/w is AMT, do not set
 895          * DRV_LOAD until the interface is up.  For all other cases,
 896          * let the f/w know that the h/w is now under the control
 897          * of the driver. */
 898         if (adapter->hw.mac_type != e1000_82573 ||
 899             !e1000_check_mng_mode(&adapter->hw))
 900                 e1000_get_hw_control(adapter);
 901
 902         strcpy(netdev->name, "eth%d");
 903         if ((err = register_netdev(netdev)))
 904                 goto err_register;
 905
 906         DPRINTK(PROBE, INFO, "Intel(R) PRO/1000 Network Connection\n");
 907
 908         cards_found++;
 909         return 0;
 910
 911 err_register:
 912 err_sw_init:
 913 err_eeprom:
 914         iounmap(adapter->hw.hw_addr);
 915 err_ioremap:
 916         free_netdev(netdev);
 917 err_alloc_etherdev:
 918         pci_release_regions(pdev);
 919         return err;
 920 }
 921
 922 /**
 923  * e1000_remove - Device Removal Routine
 924  * @pdev: PCI device information struct
 925  *
 926  * e1000_remove is called by the PCI subsystem to alert the driver
 927  * that it should release a PCI device.  The could be caused by a
 928  * Hot-Plug event, or because the driver is going to be removed from
 929  * memory.
 930  **/
 931
 932 static void __devexit
 933 e1000_remove(struct pci_dev *pdev)
 934 {
 935         struct net_device *netdev = pci_get_drvdata(pdev);
 936         struct e1000_adapter *adapter = netdev_priv(netdev);
 937         uint32_t manc;
 938 #ifdef CONFIG_E1000_NAPI
 939         int i;
 940 #endif
 941
 942         flush_scheduled_work();
 943
 944         if (adapter->hw.mac_type >= e1000_82540 &&
 945            adapter->hw.media_type == e1000_media_type_copper) {
 946                 manc = E1000_READ_REG(&adapter->hw, MANC);
 947                 if (manc & E1000_MANC_SMBUS_EN) {
 948                         manc |= E1000_MANC_ARP_EN;
 949                         E1000_WRITE_REG(&adapter->hw, MANC, manc);
 950                 }
 951         }
 952
 953         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
 954          * would have already happened in close and is redundant. */
 955         e1000_release_hw_control(adapter);
 956
 957         unregister_netdev(netdev);
 958 #ifdef CONFIG_E1000_NAPI
 959         for (i = 0; i < adapter->num_rx_queues; i++)
 960                 __dev_put(&adapter->polling_netdev[i]);
 961 #endif
 962
 963         if (!e1000_check_phy_reset_block(&adapter->hw))
 964                 e1000_phy_hw_reset(&adapter->hw);
 965
 966         kfree(adapter->tx_ring);
 967         kfree(adapter->rx_ring);
 968 #ifdef CONFIG_E1000_NAPI
 969         kfree(adapter->polling_netdev);
 970 #endif
 971
 972         iounmap(adapter->hw.hw_addr);
 973         pci_release_regions(pdev);
 974
 975 #ifdef CONFIG_E1000_MQ
 976         free_percpu(adapter->cpu_netdev);
 977         free_percpu(adapter->cpu_tx_ring);
 978 #endif
 979         free_netdev(netdev);
 980
 981         pci_disable_device(pdev);
 982 }
 983
 984 /**
 985  * e1000_sw_init - Initialize general software structures (struct e1000_adapter)
 986  * @adapter: board private structure to initialize
 987  *
 988  * e1000_sw_init initializes the Adapter private data structure.
 989  * Fields are initialized based on PCI device information and
 990  * OS network device settings (MTU size).
 991  **/
 992
 993 static int __devinit
 994 e1000_sw_init(struct e1000_adapter *adapter)
 995 {
 996         struct e1000_hw *hw = &adapter->hw;
 997         struct net_device *netdev = adapter->netdev;
 998         struct pci_dev *pdev = adapter->pdev;
 999 #ifdef CONFIG_E1000_NAPI
1000         int i;
1001 #endif
1002
1003         /* PCI config space info */
1004
1005         hw->vendor_id = pdev->vendor;
1006         hw->device_id = pdev->device;
1007         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1008         hw->subsystem_id = pdev->subsystem_device;
1009
1010         pci_read_config_byte(pdev, PCI_REVISION_ID, &hw->revision_id);
1011
1012         pci_read_config_word(pdev, PCI_COMMAND, &hw->pci_cmd_word);
1013
1014         adapter->rx_buffer_len = E1000_RXBUFFER_2048;
1015         adapter->rx_ps_bsize0 = E1000_RXBUFFER_256;
1016         hw->max_frame_size = netdev->mtu +
1017                              ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
1018         hw->min_frame_size = MINIMUM_ETHERNET_FRAME_SIZE;
1019
1020         /* identify the MAC */
1021
1022         if (e1000_set_mac_type(hw)) {
1023                 DPRINTK(PROBE, ERR, "Unknown MAC Type\n");
1024                 return -EIO;
1025         }
1026
1027         /* initialize eeprom parameters */
1028
1029         if (e1000_init_eeprom_params(hw)) {
1030                 E1000_ERR("EEPROM initialization failed\n");
1031                 return -EIO;
1032         }
1033
1034         switch (hw->mac_type) {
1035         default:
1036                 break;
1037         case e1000_82541:
1038         case e1000_82547:
1039         case e1000_82541_rev_2:
1040         case e1000_82547_rev_2:
1041                 hw->phy_init_script = 1;
1042                 break;
1043         }
1044
1045         e1000_set_media_type(hw);
1046
1047         hw->wait_autoneg_complete = FALSE;
1048         hw->tbi_compatibility_en = TRUE;
1049         hw->adaptive_ifs = TRUE;
1050
1051         /* Copper options */
1052
1053         if (hw->media_type == e1000_media_type_copper) {
1054                 hw->mdix = AUTO_ALL_MODES;
1055                 hw->disable_polarity_correction = FALSE;
1056                 hw->master_slave = E1000_MASTER_SLAVE;
1057         }
1058
1059 #ifdef CONFIG_E1000_MQ
1060         /* Number of supported queues */
1061         switch (hw->mac_type) {
1062         case e1000_82571:
1063         case e1000_82572:
1064                 /* These controllers support 2 tx queues, but with a single
1065                  * qdisc implementation, multiple tx queues aren't quite as
1066                  * interesting.  If we can find a logical way of mapping
1067                  * flows to a queue, then perhaps we can up the num_tx_queue
1068                  * count back to its default.  Until then, we run the risk of
1069                  * terrible performance due to SACK overload. */
1070                 adapter->num_tx_queues = 1;
1071                 adapter->num_rx_queues = 2;
1072                 break;
1073         default:
1074                 adapter->num_tx_queues = 1;
1075                 adapter->num_rx_queues = 1;
1076                 break;
1077         }
1078         adapter->num_rx_queues = min(adapter->num_rx_queues, num_online_cpus());
1079         adapter->num_tx_queues = min(adapter->num_tx_queues, num_online_cpus());
1080         DPRINTK(DRV, INFO, "Multiqueue Enabled: Rx Queue count = %u %s\n",
1081                 adapter->num_rx_queues,
1082                 ((adapter->num_rx_queues == 1)
1083                  ? ((num_online_cpus() > 1)
1084                         ? "(due to unsupported feature in current adapter)"
1085                         : "(due to unsupported system configuration)")
1086                  : ""));
1087         DPRINTK(DRV, INFO, "Multiqueue Enabled: Tx Queue count = %u\n",
1088                 adapter->num_tx_queues);
1089 #else
1090         adapter->num_tx_queues = 1;
1091         adapter->num_rx_queues = 1;
1092 #endif
1093
1094         if (e1000_alloc_queues(adapter)) {
1095                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
1096                 return -ENOMEM;
1097         }
1098
1099 #ifdef CONFIG_E1000_NAPI
1100         for (i = 0; i < adapter->num_rx_queues; i++) {
1101                 adapter->polling_netdev[i].priv = adapter;
1102                 adapter->polling_netdev[i].poll = &e1000_clean;
1103                 adapter->polling_netdev[i].weight = 64;
1104                 dev_hold(&adapter->polling_netdev[i]);
1105                 set_bit(__LINK_STATE_START, &adapter->polling_netdev[i].state);
1106         }
1107         spin_lock_init(&adapter->tx_queue_lock);
1108 #endif
1109
1110         atomic_set(&adapter->irq_sem, 1);
1111         spin_lock_init(&adapter->stats_lock);
1112
1113         return 0;
1114 }
1115
1116 /**
1117  * e1000_alloc_queues - Allocate memory for all rings
1118  * @adapter: board private structure to initialize
1119  *
1120  * We allocate one ring per queue at run-time since we don't know the
1121  * number of queues at compile-time.  The polling_netdev array is
1122  * intended for Multiqueue, but should work fine with a single queue.
1123  **/
1124
1125 static int __devinit
1126 e1000_alloc_queues(struct e1000_adapter *adapter)
1127 {
1128         int size;
1129
1130         size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
1131         adapter->tx_ring = kmalloc(size, GFP_KERNEL);
1132         if (!adapter->tx_ring)
1133                 return -ENOMEM;
1134         memset(adapter->tx_ring, 0, size);
1135
1136         size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
1137         adapter->rx_ring = kmalloc(size, GFP_KERNEL);
1138         if (!adapter->rx_ring) {
1139                 kfree(adapter->tx_ring);
1140                 return -ENOMEM;
1141         }
1142         memset(adapter->rx_ring, 0, size);
1143
1144 #ifdef CONFIG_E1000_NAPI
1145         size = sizeof(struct net_device) * adapter->num_rx_queues;
1146         adapter->polling_netdev = kmalloc(size, GFP_KERNEL);
1147         if (!adapter->polling_netdev) {
1148                 kfree(adapter->tx_ring);
1149                 kfree(adapter->rx_ring);
1150                 return -ENOMEM;
1151         }
1152         memset(adapter->polling_netdev, 0, size);
1153 #endif
1154
1155 #ifdef CONFIG_E1000_MQ
1156         adapter->rx_sched_call_data.func = e1000_rx_schedule;
1157         adapter->rx_sched_call_data.info = adapter->netdev;
1158
1159         adapter->cpu_netdev = alloc_percpu(struct net_device *);
1160         adapter->cpu_tx_ring = alloc_percpu(struct e1000_tx_ring *);
1161 #endif
1162
1163         return E1000_SUCCESS;
1164 }
1165
1166 #ifdef CONFIG_E1000_MQ
1167 static void __devinit
1168 e1000_setup_queue_mapping(struct e1000_adapter *adapter)
1169 {
1170         int i, cpu;
1171
1172         adapter->rx_sched_call_data.func = e1000_rx_schedule;
1173         adapter->rx_sched_call_data.info = adapter->netdev;
1174         cpus_clear(adapter->rx_sched_call_data.cpumask);
1175
1176         adapter->cpu_netdev = alloc_percpu(struct net_device *);
1177         adapter->cpu_tx_ring = alloc_percpu(struct e1000_tx_ring *);
1178
1179         lock_cpu_hotplug();
1180         i = 0;
1181         for_each_online_cpu(cpu) {
1182                 *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_tx_queues];
1183                 /* This is incomplete because we'd like to assign separate
1184                  * physical cpus to these netdev polling structures and
1185                  * avoid saturating a subset of cpus.
1186                  */
1187                 if (i < adapter->num_rx_queues) {
1188                         *per_cpu_ptr(adapter->cpu_netdev, cpu) = &adapter->polling_netdev[i];
1189                         adapter->rx_ring[i].cpu = cpu;
1190                         cpu_set(cpu, adapter->cpumask);
1191                 } else
1192                         *per_cpu_ptr(adapter->cpu_netdev, cpu) = NULL;
1193
1194                 i++;
1195         }
1196         unlock_cpu_hotplug();
1197 }
1198 #endif
1199
1200 /**
1201  * e1000_open - Called when a network interface is made active
1202  * @netdev: network interface device structure
1203  *
1204  * Returns 0 on success, negative value on failure
1205  *
1206  * The open entry point is called when a network interface is made
1207  * active by the system (IFF_UP).  At this point all resources needed
1208  * for transmit and receive operations are allocated, the interrupt
1209  * handler is registered with the OS, the watchdog timer is started,
1210  * and the stack is notified that the interface is ready.
1211  **/
1212
1213 static int
1214 e1000_open(struct net_device *netdev)
1215 {
1216         struct e1000_adapter *adapter = netdev_priv(netdev);
1217         int err;
1218
1219         /* allocate transmit descriptors */
1220
1221         if ((err = e1000_setup_all_tx_resources(adapter)))
1222                 goto err_setup_tx;
1223
1224         /* allocate receive descriptors */
1225
1226         if ((err = e1000_setup_all_rx_resources(adapter)))
1227                 goto err_setup_rx;
1228
1229         if ((err = e1000_up(adapter)))
1230                 goto err_up;
1231         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
1232         if ((adapter->hw.mng_cookie.status &
1233                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1234                 e1000_update_mng_vlan(adapter);
1235         }
1236
1237         /* If AMT is enabled, let the firmware know that the network
1238          * interface is now open */
1239         if (adapter->hw.mac_type == e1000_82573 &&
1240             e1000_check_mng_mode(&adapter->hw))
1241                 e1000_get_hw_control(adapter);
1242
1243         return E1000_SUCCESS;
1244
1245 err_up:
1246         e1000_free_all_rx_resources(adapter);
1247 err_setup_rx:
1248         e1000_free_all_tx_resources(adapter);
1249 err_setup_tx:
1250         e1000_reset(adapter);
1251
1252         return err;
1253 }
1254
1255 /**
1256  * e1000_close - Disables a network interface
1257  * @netdev: network interface device structure
1258  *
1259  * Returns 0, this is not allowed to fail
1260  *
1261  * The close entry point is called when an interface is de-activated
1262  * by the OS.  The hardware is still under the drivers control, but
1263  * needs to be disabled.  A global MAC reset is issued to stop the
1264  * hardware, and all transmit and receive resources are freed.
1265  **/
1266
1267 static int
1268 e1000_close(struct net_device *netdev)
1269 {
1270         struct e1000_adapter *adapter = netdev_priv(netdev);
1271
1272         e1000_down(adapter);
1273
1274         e1000_free_all_tx_resources(adapter);
1275         e1000_free_all_rx_resources(adapter);
1276
1277         if ((adapter->hw.mng_cookie.status &
1278                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) {
1279                 e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1280         }
1281
1282         /* If AMT is enabled, let the firmware know that the network
1283          * interface is now closed */
1284         if (adapter->hw.mac_type == e1000_82573 &&
1285             e1000_check_mng_mode(&adapter->hw))
1286                 e1000_release_hw_control(adapter);
1287
1288         return 0;
1289 }
1290
1291 /**
1292  * e1000_check_64k_bound - check that memory doesn't cross 64kB boundary
1293  * @adapter: address of board private structure
1294  * @start: address of beginning of memory
1295  * @len: length of memory
1296  **/
1297 static inline boolean_t
1298 e1000_check_64k_bound(struct e1000_adapter *adapter,
1299                       void *start, unsigned long len)
1300 {
1301         unsigned long begin = (unsigned long) start;
1302         unsigned long end = begin + len;
1303
1304         /* First rev 82545 and 82546 need to not allow any memory
1305          * write location to cross 64k boundary due to errata 23 */
1306         if (adapter->hw.mac_type == e1000_82545 ||
1307             adapter->hw.mac_type == e1000_82546) {
1308                 return ((begin ^ (end - 1)) >> 16) != 0 ? FALSE : TRUE;
1309         }
1310
1311         return TRUE;
1312 }
1313
1314 /**
1315  * e1000_setup_tx_resources - allocate Tx resources (Descriptors)
1316  * @adapter: board private structure
1317  * @txdr:    tx descriptor ring (for a specific queue) to setup
1318  *
1319  * Return 0 on success, negative on failure
1320  **/
1321
1322 static int
1323 e1000_setup_tx_resources(struct e1000_adapter *adapter,
1324                          struct e1000_tx_ring *txdr)
1325 {
1326         struct pci_dev *pdev = adapter->pdev;
1327         int size;
1328
1329         size = sizeof(struct e1000_buffer) * txdr->count;
1330
1331         txdr->buffer_info = vmalloc_node(size, pcibus_to_node(pdev->bus));
1332         if (!txdr->buffer_info) {
1333                 DPRINTK(PROBE, ERR,
1334                 "Unable to allocate memory for the transmit descriptor ring\n");
1335                 return -ENOMEM;
1336         }
1337         memset(txdr->buffer_info, 0, size);
1338
1339         /* round up to nearest 4K */
1340
1341         txdr->size = txdr->count * sizeof(struct e1000_tx_desc);
1342         E1000_ROUNDUP(txdr->size, 4096);
1343
1344         txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1345         if (!txdr->desc) {
1346 setup_tx_desc_die:
1347                 vfree(txdr->buffer_info);
1348                 DPRINTK(PROBE, ERR,
1349                 "Unable to allocate memory for the transmit descriptor ring\n");
1350                 return -ENOMEM;
1351         }
1352
1353         /* Fix for errata 23, can't cross 64kB boundary */
1354         if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1355                 void *olddesc = txdr->desc;
1356                 dma_addr_t olddma = txdr->dma;
1357                 DPRINTK(TX_ERR, ERR, "txdr align check failed: %u bytes "
1358                                      "at %p\n", txdr->size, txdr->desc);
1359                 /* Try again, without freeing the previous */
1360                 txdr->desc = pci_alloc_consistent(pdev, txdr->size, &txdr->dma);
1361                 /* Failed allocation, critical failure */
1362                 if (!txdr->desc) {
1363                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1364                         goto setup_tx_desc_die;
1365                 }
1366
1367                 if (!e1000_check_64k_bound(adapter, txdr->desc, txdr->size)) {
1368                         /* give up */
1369                         pci_free_consistent(pdev, txdr->size, txdr->desc,
1370                                             txdr->dma);
1371                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1372                         DPRINTK(PROBE, ERR,
1373                                 "Unable to allocate aligned memory "
1374                                 "for the transmit descriptor ring\n");
1375                         vfree(txdr->buffer_info);
1376                         return -ENOMEM;
1377                 } else {
1378                         /* Free old allocation, new allocation was successful */
1379                         pci_free_consistent(pdev, txdr->size, olddesc, olddma);
1380                 }
1381         }
1382         memset(txdr->desc, 0, txdr->size);
1383
1384         txdr->next_to_use = 0;
1385         txdr->next_to_clean = 0;
1386         spin_lock_init(&txdr->tx_lock);
1387
1388         return 0;
1389 }
1390
1391 /**
1392  * e1000_setup_all_tx_resources - wrapper to allocate Tx resources
1393  *                                (Descriptors) for all queues
1394  * @adapter: board private structure
1395  *
1396  * If this function returns with an error, then it's possible one or
1397  * more of the rings is populated (while the rest are not).  It is the
1398  * callers duty to clean those orphaned rings.
1399  *
1400  * Return 0 on success, negative on failure
1401  **/
1402
1403 int
1404 e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
1405 {
1406         int i, err = 0;
1407
1408         for (i = 0; i < adapter->num_tx_queues; i++) {
1409                 err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1410                 if (err) {
1411                         DPRINTK(PROBE, ERR,
1412                                 "Allocation for Tx Queue %u failed\n", i);
1413                         break;
1414                 }
1415         }
1416
1417         return err;
1418 }
1419
1420 /**
1421  * e1000_configure_tx - Configure 8254x Transmit Unit after Reset
1422  * @adapter: board private structure
1423  *
1424  * Configure the Tx unit of the MAC after a reset.
1425  **/
1426
1427 static void
1428 e1000_configure_tx(struct e1000_adapter *adapter)
1429 {
1430         uint64_t tdba;
1431         struct e1000_hw *hw = &adapter->hw;
1432         uint32_t tdlen, tctl, tipg, tarc;
1433         uint32_t ipgr1, ipgr2;
1434
1435         /* Setup the HW Tx Head and Tail descriptor pointers */
1436
1437         switch (adapter->num_tx_queues) {
1438         case 2:
1439                 tdba = adapter->tx_ring[1].dma;
1440                 tdlen = adapter->tx_ring[1].count *
1441                         sizeof(struct e1000_tx_desc);
1442                 E1000_WRITE_REG(hw, TDBAL1, (tdba & 0x00000000ffffffffULL));
1443                 E1000_WRITE_REG(hw, TDBAH1, (tdba >> 32));
1444                 E1000_WRITE_REG(hw, TDLEN1, tdlen);
1445                 E1000_WRITE_REG(hw, TDH1, 0);
1446                 E1000_WRITE_REG(hw, TDT1, 0);
1447                 adapter->tx_ring[1].tdh = E1000_TDH1;
1448                 adapter->tx_ring[1].tdt = E1000_TDT1;
1449                 /* Fall Through */
1450         case 1:
1451         default:
1452                 tdba = adapter->tx_ring[0].dma;
1453                 tdlen = adapter->tx_ring[0].count *
1454                         sizeof(struct e1000_tx_desc);
1455                 E1000_WRITE_REG(hw, TDBAL, (tdba & 0x00000000ffffffffULL));
1456                 E1000_WRITE_REG(hw, TDBAH, (tdba >> 32));
1457                 E1000_WRITE_REG(hw, TDLEN, tdlen);
1458                 E1000_WRITE_REG(hw, TDH, 0);
1459                 E1000_WRITE_REG(hw, TDT, 0);
1460                 adapter->tx_ring[0].tdh = E1000_TDH;
1461                 adapter->tx_ring[0].tdt = E1000_TDT;
1462                 break;
1463         }
1464
1465         /* Set the default values for the Tx Inter Packet Gap timer */
1466
1467         if (hw->media_type == e1000_media_type_fiber ||
1468             hw->media_type == e1000_media_type_internal_serdes)
1469                 tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
1470         else
1471                 tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
1472
1473         switch (hw->mac_type) {
1474         case e1000_82542_rev2_0:
1475         case e1000_82542_rev2_1:
1476                 tipg = DEFAULT_82542_TIPG_IPGT;
1477                 ipgr1 = DEFAULT_82542_TIPG_IPGR1;
1478                 ipgr2 = DEFAULT_82542_TIPG_IPGR2;
1479                 break;
1480         default:
1481                 ipgr1 = DEFAULT_82543_TIPG_IPGR1;
1482                 ipgr2 = DEFAULT_82543_TIPG_IPGR2;
1483                 break;
1484         }
1485         tipg |= ipgr1 << E1000_TIPG_IPGR1_SHIFT;
1486         tipg |= ipgr2 << E1000_TIPG_IPGR2_SHIFT;
1487         E1000_WRITE_REG(hw, TIPG, tipg);
1488
1489         /* Set the Tx Interrupt Delay register */
1490
1491         E1000_WRITE_REG(hw, TIDV, adapter->tx_int_delay);
1492         if (hw->mac_type >= e1000_82540)
1493                 E1000_WRITE_REG(hw, TADV, adapter->tx_abs_int_delay);
1494
1495         /* Program the Transmit Control Register */
1496
1497         tctl = E1000_READ_REG(hw, TCTL);
1498
1499         tctl &= ~E1000_TCTL_CT;
1500         tctl |= E1000_TCTL_EN | E1000_TCTL_PSP | E1000_TCTL_RTLC |
1501                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1502
1503         E1000_WRITE_REG(hw, TCTL, tctl);
1504
1505         if (hw->mac_type == e1000_82571 || hw->mac_type == e1000_82572) {
1506                 tarc = E1000_READ_REG(hw, TARC0);
1507                 tarc |= ((1 << 25) | (1 << 21));
1508                 E1000_WRITE_REG(hw, TARC0, tarc);
1509                 tarc = E1000_READ_REG(hw, TARC1);
1510                 tarc |= (1 << 25);
1511                 if (tctl & E1000_TCTL_MULR)
1512                         tarc &= ~(1 << 28);
1513                 else
1514                         tarc |= (1 << 28);
1515                 E1000_WRITE_REG(hw, TARC1, tarc);
1516         }
1517
1518         e1000_config_collision_dist(hw);
1519
1520         /* Setup Transmit Descriptor Settings for eop descriptor */
1521         adapter->txd_cmd = E1000_TXD_CMD_IDE | E1000_TXD_CMD_EOP |
1522                 E1000_TXD_CMD_IFCS;
1523
1524         if (hw->mac_type < e1000_82543)
1525                 adapter->txd_cmd |= E1000_TXD_CMD_RPS;
1526         else
1527                 adapter->txd_cmd |= E1000_TXD_CMD_RS;
1528
1529         /* Cache if we're 82544 running in PCI-X because we'll
1530          * need this to apply a workaround later in the send path. */
1531         if (hw->mac_type == e1000_82544 &&
1532             hw->bus_type == e1000_bus_type_pcix)
1533                 adapter->pcix_82544 = 1;
1534 }
1535
1536 /**
1537  * e1000_setup_rx_resources - allocate Rx resources (Descriptors)
1538  * @adapter: board private structure
1539  * @rxdr:    rx descriptor ring (for a specific queue) to setup
1540  *
1541  * Returns 0 on success, negative on failure
1542  **/
1543
1544 static int
1545 e1000_setup_rx_resources(struct e1000_adapter *adapter,
1546                          struct e1000_rx_ring *rxdr)
1547 {
1548         struct pci_dev *pdev = adapter->pdev;
1549         int size, desc_len;
1550
1551         size = sizeof(struct e1000_buffer) * rxdr->count;
1552         rxdr->buffer_info = vmalloc_node(size, pcibus_to_node(pdev->bus));
1553         if (!rxdr->buffer_info) {
1554                 DPRINTK(PROBE, ERR,
1555                 "Unable to allocate memory for the receive descriptor ring\n");
1556                 return -ENOMEM;
1557         }
1558         memset(rxdr->buffer_info, 0, size);
1559
1560         size = sizeof(struct e1000_ps_page) * rxdr->count;
1561         rxdr->ps_page = kmalloc(size, GFP_KERNEL);
1562         if (!rxdr->ps_page) {
1563                 vfree(rxdr->buffer_info);
1564                 DPRINTK(PROBE, ERR,
1565                 "Unable to allocate memory for the receive descriptor ring\n");
1566                 return -ENOMEM;
1567         }
1568         memset(rxdr->ps_page, 0, size);
1569
1570         size = sizeof(struct e1000_ps_page_dma) * rxdr->count;
1571         rxdr->ps_page_dma = kmalloc(size, GFP_KERNEL);
1572         if (!rxdr->ps_page_dma) {
1573                 vfree(rxdr->buffer_info);
1574                 kfree(rxdr->ps_page);
1575                 DPRINTK(PROBE, ERR,
1576                 "Unable to allocate memory for the receive descriptor ring\n");
1577                 return -ENOMEM;
1578         }
1579         memset(rxdr->ps_page_dma, 0, size);
1580
1581         if (adapter->hw.mac_type <= e1000_82547_rev_2)
1582                 desc_len = sizeof(struct e1000_rx_desc);
1583         else
1584                 desc_len = sizeof(union e1000_rx_desc_packet_split);
1585
1586         /* Round up to nearest 4K */
1587
1588         rxdr->size = rxdr->count * desc_len;
1589         E1000_ROUNDUP(rxdr->size, 4096);
1590
1591         rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1592
1593         if (!rxdr->desc) {
1594                 DPRINTK(PROBE, ERR,
1595                 "Unable to allocate memory for the receive descriptor ring\n");
1596 setup_rx_desc_die:
1597                 vfree(rxdr->buffer_info);
1598                 kfree(rxdr->ps_page);
1599                 kfree(rxdr->ps_page_dma);
1600                 return -ENOMEM;
1601         }
1602
1603         /* Fix for errata 23, can't cross 64kB boundary */
1604         if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1605                 void *olddesc = rxdr->desc;
1606                 dma_addr_t olddma = rxdr->dma;
1607                 DPRINTK(RX_ERR, ERR, "rxdr align check failed: %u bytes "
1608                                      "at %p\n", rxdr->size, rxdr->desc);
1609                 /* Try again, without freeing the previous */
1610                 rxdr->desc = pci_alloc_consistent(pdev, rxdr->size, &rxdr->dma);
1611                 /* Failed allocation, critical failure */
1612                 if (!rxdr->desc) {
1613                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1614                         DPRINTK(PROBE, ERR,
1615                                 "Unable to allocate memory "
1616                                 "for the receive descriptor ring\n");
1617                         goto setup_rx_desc_die;
1618                 }
1619
1620                 if (!e1000_check_64k_bound(adapter, rxdr->desc, rxdr->size)) {
1621                         /* give up */
1622                         pci_free_consistent(pdev, rxdr->size, rxdr->desc,
1623                                             rxdr->dma);
1624                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1625                         DPRINTK(PROBE, ERR,
1626                                 "Unable to allocate aligned memory "
1627                                 "for the receive descriptor ring\n");
1628                         goto setup_rx_desc_die;
1629                 } else {
1630                         /* Free old allocation, new allocation was successful */
1631                         pci_free_consistent(pdev, rxdr->size, olddesc, olddma);
1632                 }
1633         }
1634         memset(rxdr->desc, 0, rxdr->size);
1635
1636         rxdr->next_to_clean = 0;
1637         rxdr->next_to_use = 0;
1638
1639         return 0;
1640 }
1641
1642 /**
1643  * e1000_setup_all_rx_resources - wrapper to allocate Rx resources
1644  *                                (Descriptors) for all queues
1645  * @adapter: board private structure
1646  *
1647  * If this function returns with an error, then it's possible one or
1648  * more of the rings is populated (while the rest are not).  It is the
1649  * callers duty to clean those orphaned rings.
1650  *
1651  * Return 0 on success, negative on failure
1652  **/
1653
1654 int
1655 e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
1656 {
1657         int i, err = 0;
1658
1659         for (i = 0; i < adapter->num_rx_queues; i++) {
1660                 err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1661                 if (err) {
1662                         DPRINTK(PROBE, ERR,
1663                                 "Allocation for Rx Queue %u failed\n", i);
1664                         break;
1665                 }
1666         }
1667
1668         return err;
1669 }
1670
1671 /**
1672  * e1000_setup_rctl - configure the receive control registers
1673  * @adapter: Board private structure
1674  **/
1675 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1676                         (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1677 static void
1678 e1000_setup_rctl(struct e1000_adapter *adapter)
1679 {
1680         uint32_t rctl, rfctl;
1681         uint32_t psrctl = 0;
1682 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1683         uint32_t pages = 0;
1684 #endif
1685
1686         rctl = E1000_READ_REG(&adapter->hw, RCTL);
1687
1688         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1689
1690         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM |
1691                 E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
1692                 (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
1693
1694         if (adapter->hw.mac_type > e1000_82543)
1695                 rctl |= E1000_RCTL_SECRC;
1696
1697         if (adapter->hw.tbi_compatibility_on == 1)
1698                 rctl |= E1000_RCTL_SBP;
1699         else
1700                 rctl &= ~E1000_RCTL_SBP;
1701
1702         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1703                 rctl &= ~E1000_RCTL_LPE;
1704         else
1705                 rctl |= E1000_RCTL_LPE;
1706
1707         /* Setup buffer sizes */
1708         if (adapter->hw.mac_type >= e1000_82571) {
1709                 /* We can now specify buffers in 1K increments.
1710                  * BSIZE and BSEX are ignored in this case. */
1711                 rctl |= adapter->rx_buffer_len << 0x11;
1712         } else {
1713                 rctl &= ~E1000_RCTL_SZ_4096;
1714                 rctl |= E1000_RCTL_BSEX;
1715                 switch (adapter->rx_buffer_len) {
1716                 case E1000_RXBUFFER_2048:
1717                 default:
1718                         rctl |= E1000_RCTL_SZ_2048;
1719                         rctl &= ~E1000_RCTL_BSEX;
1720                         break;
1721                 case E1000_RXBUFFER_4096:
1722                         rctl |= E1000_RCTL_SZ_4096;
1723                         break;
1724                 case E1000_RXBUFFER_8192:
1725                         rctl |= E1000_RCTL_SZ_8192;
1726                         break;
1727                 case E1000_RXBUFFER_16384:
1728                         rctl |= E1000_RCTL_SZ_16384;
1729                         break;
1730                 }
1731         }
1732
1733 #ifndef CONFIG_E1000_DISABLE_PACKET_SPLIT
1734         /* 82571 and greater support packet-split where the protocol
1735          * header is placed in skb->data and the packet data is
1736          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
1737          * In the case of a non-split, skb->data is linearly filled,
1738          * followed by the page buffers.  Therefore, skb->data is
1739          * sized to hold the largest protocol header.
1740          */
1741         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1742         if ((adapter->hw.mac_type > e1000_82547_rev_2) && (pages <= 3) &&
1743             PAGE_SIZE <= 16384)
1744                 adapter->rx_ps_pages = pages;
1745         else
1746                 adapter->rx_ps_pages = 0;
1747 #endif
1748         if (adapter->rx_ps_pages) {
1749                 /* Configure extra packet-split registers */
1750                 rfctl = E1000_READ_REG(&adapter->hw, RFCTL);
1751                 rfctl |= E1000_RFCTL_EXTEN;
1752                 /* disable IPv6 packet split support */
1753                 rfctl |= E1000_RFCTL_IPV6_DIS;
1754                 E1000_WRITE_REG(&adapter->hw, RFCTL, rfctl);
1755
1756                 rctl |= E1000_RCTL_DTYP_PS | E1000_RCTL_SECRC;
1757
1758                 psrctl |= adapter->rx_ps_bsize0 >>
1759                         E1000_PSRCTL_BSIZE0_SHIFT;
1760
1761                 switch (adapter->rx_ps_pages) {
1762                 case 3:
1763                         psrctl |= PAGE_SIZE <<
1764                                 E1000_PSRCTL_BSIZE3_SHIFT;
1765                 case 2:
1766                         psrctl |= PAGE_SIZE <<
1767                                 E1000_PSRCTL_BSIZE2_SHIFT;
1768                 case 1:
1769                         psrctl |= PAGE_SIZE >>
1770                                 E1000_PSRCTL_BSIZE1_SHIFT;
1771                         break;
1772                 }
1773
1774                 E1000_WRITE_REG(&adapter->hw, PSRCTL, psrctl);
1775         }
1776
1777         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
1778 }
1779
1780 /**
1781  * e1000_configure_rx - Configure 8254x Receive Unit after Reset
1782  * @adapter: board private structure
1783  *
1784  * Configure the Rx unit of the MAC after a reset.
1785  **/
1786
1787 static void
1788 e1000_configure_rx(struct e1000_adapter *adapter)
1789 {
1790         uint64_t rdba;
1791         struct e1000_hw *hw = &adapter->hw;
1792         uint32_t rdlen, rctl, rxcsum, ctrl_ext;
1793 #ifdef CONFIG_E1000_MQ
1794         uint32_t reta, mrqc;
1795         int i;
1796 #endif
1797
1798         if (adapter->rx_ps_pages) {
1799                 rdlen = adapter->rx_ring[0].count *
1800                         sizeof(union e1000_rx_desc_packet_split);
1801                 adapter->clean_rx = e1000_clean_rx_irq_ps;
1802                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers_ps;
1803         } else {
1804                 rdlen = adapter->rx_ring[0].count *
1805                         sizeof(struct e1000_rx_desc);
1806                 adapter->clean_rx = e1000_clean_rx_irq;
1807                 adapter->alloc_rx_buf = e1000_alloc_rx_buffers;
1808         }
1809
1810         /* disable receives while setting up the descriptors */
1811         rctl = E1000_READ_REG(hw, RCTL);
1812         E1000_WRITE_REG(hw, RCTL, rctl & ~E1000_RCTL_EN);
1813
1814         /* set the Receive Delay Timer Register */
1815         E1000_WRITE_REG(hw, RDTR, adapter->rx_int_delay);
1816
1817         if (hw->mac_type >= e1000_82540) {
1818                 E1000_WRITE_REG(hw, RADV, adapter->rx_abs_int_delay);
1819                 if (adapter->itr > 1)
1820                         E1000_WRITE_REG(hw, ITR,
1821                                 1000000000 / (adapter->itr * 256));
1822         }
1823
1824         if (hw->mac_type >= e1000_82571) {
1825                 ctrl_ext = E1000_READ_REG(hw, CTRL_EXT);
1826                 /* Reset delay timers after every interrupt */
1827                 ctrl_ext |= E1000_CTRL_EXT_CANC;
1828 #ifdef CONFIG_E1000_NAPI
1829                 /* Auto-Mask interrupts upon ICR read. */
1830                 ctrl_ext |= E1000_CTRL_EXT_IAME;
1831 #endif
1832                 E1000_WRITE_REG(hw, CTRL_EXT, ctrl_ext);
1833                 E1000_WRITE_REG(hw, IAM, ~0);
1834                 E1000_WRITE_FLUSH(hw);
1835         }
1836
1837         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1838          * the Base and Length of the Rx Descriptor Ring */
1839         switch (adapter->num_rx_queues) {
1840 #ifdef CONFIG_E1000_MQ
1841         case 2:
1842                 rdba = adapter->rx_ring[1].dma;
1843                 E1000_WRITE_REG(hw, RDBAL1, (rdba & 0x00000000ffffffffULL));
1844                 E1000_WRITE_REG(hw, RDBAH1, (rdba >> 32));
1845                 E1000_WRITE_REG(hw, RDLEN1, rdlen);
1846                 E1000_WRITE_REG(hw, RDH1, 0);
1847                 E1000_WRITE_REG(hw, RDT1, 0);
1848                 adapter->rx_ring[1].rdh = E1000_RDH1;
1849                 adapter->rx_ring[1].rdt = E1000_RDT1;
1850                 /* Fall Through */
1851 #endif
1852         case 1:
1853         default:
1854                 rdba = adapter->rx_ring[0].dma;
1855                 E1000_WRITE_REG(hw, RDBAL, (rdba & 0x00000000ffffffffULL));
1856                 E1000_WRITE_REG(hw, RDBAH, (rdba >> 32));
1857                 E1000_WRITE_REG(hw, RDLEN, rdlen);
1858                 E1000_WRITE_REG(hw, RDH, 0);
1859                 E1000_WRITE_REG(hw, RDT, 0);
1860                 adapter->rx_ring[0].rdh = E1000_RDH;
1861                 adapter->rx_ring[0].rdt = E1000_RDT;
1862                 break;
1863         }
1864
1865 #ifdef CONFIG_E1000_MQ
1866         if (adapter->num_rx_queues > 1) {
1867                 uint32_t random[10];
1868
1869                 get_random_bytes(&random[0], 40);
1870
1871                 if (hw->mac_type <= e1000_82572) {
1872                         E1000_WRITE_REG(hw, RSSIR, 0);
1873                         E1000_WRITE_REG(hw, RSSIM, 0);
1874                 }
1875
1876                 switch (adapter->num_rx_queues) {
1877                 case 2:
1878                 default:
1879                         reta = 0x00800080;
1880                         mrqc = E1000_MRQC_ENABLE_RSS_2Q;
1881                         break;
1882                 }
1883
1884                 /* Fill out redirection table */
1885                 for (i = 0; i < 32; i++)
1886                         E1000_WRITE_REG_ARRAY(hw, RETA, i, reta);
1887                 /* Fill out hash function seeds */
1888                 for (i = 0; i < 10; i++)
1889                         E1000_WRITE_REG_ARRAY(hw, RSSRK, i, random[i]);
1890
1891                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
1892                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
1893                 E1000_WRITE_REG(hw, MRQC, mrqc);
1894         }
1895
1896         /* Multiqueue and packet checksumming are mutually exclusive. */
1897         if (hw->mac_type >= e1000_82571) {
1898                 rxcsum = E1000_READ_REG(hw, RXCSUM);
1899                 rxcsum |= E1000_RXCSUM_PCSD;
1900                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
1901         }
1902
1903 #else
1904
1905         /* Enable 82543 Receive Checksum Offload for TCP and UDP */
1906         if (hw->mac_type >= e1000_82543) {
1907                 rxcsum = E1000_READ_REG(hw, RXCSUM);
1908                 if (adapter->rx_csum == TRUE) {
1909                         rxcsum |= E1000_RXCSUM_TUOFL;
1910
1911                         /* Enable 82571 IPv4 payload checksum for UDP fragments
1912                          * Must be used in conjunction with packet-split. */
1913                         if ((hw->mac_type >= e1000_82571) &&
1914                             (adapter->rx_ps_pages)) {
1915                                 rxcsum |= E1000_RXCSUM_IPPCSE;
1916                         }
1917                 } else {
1918                         rxcsum &= ~E1000_RXCSUM_TUOFL;
1919                         /* don't need to clear IPPCSE as it defaults to 0 */
1920                 }
1921                 E1000_WRITE_REG(hw, RXCSUM, rxcsum);
1922         }
1923 #endif /* CONFIG_E1000_MQ */
1924
1925         if (hw->mac_type == e1000_82573)
1926                 E1000_WRITE_REG(hw, ERT, 0x0100);
1927
1928         /* Enable Receives */
1929         E1000_WRITE_REG(hw, RCTL, rctl);
1930 }
1931
1932 /**
1933  * e1000_free_tx_resources - Free Tx Resources per Queue
1934  * @adapter: board private structure
1935  * @tx_ring: Tx descriptor ring for a specific queue
1936  *
1937  * Free all transmit software resources
1938  **/
1939
1940 static void
1941 e1000_free_tx_resources(struct e1000_adapter *adapter,
1942                         struct e1000_tx_ring *tx_ring)
1943 {
1944         struct pci_dev *pdev = adapter->pdev;
1945
1946         e1000_clean_tx_ring(adapter, tx_ring);
1947
1948         vfree(tx_ring->buffer_info);
1949         tx_ring->buffer_info = NULL;
1950
1951         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
1952
1953         tx_ring->desc = NULL;
1954 }
1955
1956 /**
1957  * e1000_free_all_tx_resources - Free Tx Resources for All Queues
1958  * @adapter: board private structure
1959  *
1960  * Free all transmit software resources
1961  **/
1962
1963 void
1964 e1000_free_all_tx_resources(struct e1000_adapter *adapter)
1965 {
1966         int i;
1967
1968         for (i = 0; i < adapter->num_tx_queues; i++)
1969                 e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
1970 }
1971
1972 static inline void
1973 e1000_unmap_and_free_tx_resource(struct e1000_adapter *adapter,
1974                         struct e1000_buffer *buffer_info)
1975 {
1976         if (buffer_info->dma) {
1977                 pci_unmap_page(adapter->pdev,
1978                                 buffer_info->dma,
1979                                 buffer_info->length,
1980                                 PCI_DMA_TODEVICE);
1981         }
1982         if (buffer_info->skb)
1983                 dev_kfree_skb_any(buffer_info->skb);
1984         memset(buffer_info, 0, sizeof(struct e1000_buffer));
1985 }
1986
1987 /**
1988  * e1000_clean_tx_ring - Free Tx Buffers
1989  * @adapter: board private structure
1990  * @tx_ring: ring to be cleaned
1991  **/
1992
1993 static void
1994 e1000_clean_tx_ring(struct e1000_adapter *adapter,
1995                     struct e1000_tx_ring *tx_ring)
1996 {
1997         struct e1000_buffer *buffer_info;
1998         unsigned long size;
1999         unsigned int i;
2000
2001         /* Free all the Tx ring sk_buffs */
2002
2003         for (i = 0; i < tx_ring->count; i++) {
2004                 buffer_info = &tx_ring->buffer_info[i];
2005                 e1000_unmap_and_free_tx_resource(adapter, buffer_info);
2006         }
2007
2008         size = sizeof(struct e1000_buffer) * tx_ring->count;
2009         memset(tx_ring->buffer_info, 0, size);
2010
2011         /* Zero out the descriptor ring */
2012
2013         memset(tx_ring->desc, 0, tx_ring->size);
2014
2015         tx_ring->next_to_use = 0;
2016         tx_ring->next_to_clean = 0;
2017         tx_ring->last_tx_tso = 0;
2018
2019         writel(0, adapter->hw.hw_addr + tx_ring->tdh);
2020         writel(0, adapter->hw.hw_addr + tx_ring->tdt);
2021 }
2022
2023 /**
2024  * e1000_clean_all_tx_rings - Free Tx Buffers for all queues
2025  * @adapter: board private structure
2026  **/
2027
2028 static void
2029 e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
2030 {
2031         int i;
2032
2033         for (i = 0; i < adapter->num_tx_queues; i++)
2034                 e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2035 }
2036
2037 /**
2038  * e1000_free_rx_resources - Free Rx Resources
2039  * @adapter: board private structure
2040  * @rx_ring: ring to clean the resources from
2041  *
2042  * Free all receive software resources
2043  **/
2044
2045 static void
2046 e1000_free_rx_resources(struct e1000_adapter *adapter,
2047                         struct e1000_rx_ring *rx_ring)
2048 {
2049         struct pci_dev *pdev = adapter->pdev;
2050
2051         e1000_clean_rx_ring(adapter, rx_ring);
2052
2053         vfree(rx_ring->buffer_info);
2054         rx_ring->buffer_info = NULL;
2055         kfree(rx_ring->ps_page);
2056         rx_ring->ps_page = NULL;
2057         kfree(rx_ring->ps_page_dma);
2058         rx_ring->ps_page_dma = NULL;
2059
2060         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2061
2062         rx_ring->desc = NULL;
2063 }
2064
2065 /**
2066  * e1000_free_all_rx_resources - Free Rx Resources for All Queues
2067  * @adapter: board private structure
2068  *
2069  * Free all receive software resources
2070  **/
2071
2072 void
2073 e1000_free_all_rx_resources(struct e1000_adapter *adapter)
2074 {
2075         int i;
2076
2077         for (i = 0; i < adapter->num_rx_queues; i++)
2078                 e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
2079 }
2080
2081 /**
2082  * e1000_clean_rx_ring - Free Rx Buffers per Queue
2083  * @adapter: board private structure
2084  * @rx_ring: ring to free buffers from
2085  **/
2086
2087 static void
2088 e1000_clean_rx_ring(struct e1000_adapter *adapter,
2089                     struct e1000_rx_ring *rx_ring)
2090 {
2091         struct e1000_buffer *buffer_info;
2092         struct e1000_ps_page *ps_page;
2093         struct e1000_ps_page_dma *ps_page_dma;
2094         struct pci_dev *pdev = adapter->pdev;
2095         unsigned long size;
2096         unsigned int i, j;
2097
2098         /* Free all the Rx ring sk_buffs */
2099         for (i = 0; i < rx_ring->count; i++) {
2100                 buffer_info = &rx_ring->buffer_info[i];
2101                 if (buffer_info->skb) {
2102                         pci_unmap_single(pdev,
2103                                          buffer_info->dma,
2104                                          buffer_info->length,
2105                                          PCI_DMA_FROMDEVICE);
2106
2107                         dev_kfree_skb(buffer_info->skb);
2108                         buffer_info->skb = NULL;
2109                 }
2110                 ps_page = &rx_ring->ps_page[i];
2111                 ps_page_dma = &rx_ring->ps_page_dma[i];
2112                 for (j = 0; j < adapter->rx_ps_pages; j++) {
2113                         if (!ps_page->ps_page[j]) break;
2114                         pci_unmap_page(pdev,
2115                                        ps_page_dma->ps_page_dma[j],
2116                                        PAGE_SIZE, PCI_DMA_FROMDEVICE);
2117                         ps_page_dma->ps_page_dma[j] = 0;
2118                         put_page(ps_page->ps_page[j]);
2119                         ps_page->ps_page[j] = NULL;
2120                 }
2121         }
2122
2123         size = sizeof(struct e1000_buffer) * rx_ring->count;
2124         memset(rx_ring->buffer_info, 0, size);
2125         size = sizeof(struct e1000_ps_page) * rx_ring->count;
2126         memset(rx_ring->ps_page, 0, size);
2127         size = sizeof(struct e1000_ps_page_dma) * rx_ring->count;
2128         memset(rx_ring->ps_page_dma, 0, size);
2129
2130         /* Zero out the descriptor ring */
2131
2132         memset(rx_ring->desc, 0, rx_ring->size);
2133
2134         rx_ring->next_to_clean = 0;
2135         rx_ring->next_to_use = 0;
2136
2137         writel(0, adapter->hw.hw_addr + rx_ring->rdh);
2138         writel(0, adapter->hw.hw_addr + rx_ring->rdt);
2139 }
2140
2141 /**
2142  * e1000_clean_all_rx_rings - Free Rx Buffers for all queues
2143  * @adapter: board private structure
2144  **/
2145
2146 static void
2147 e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
2148 {
2149         int i;
2150
2151         for (i = 0; i < adapter->num_rx_queues; i++)
2152                 e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2153 }
2154
2155 /* The 82542 2.0 (revision 2) needs to have the receive unit in reset
2156  * and memory write and invalidate disabled for certain operations
2157  */
2158 static void
2159 e1000_enter_82542_rst(struct e1000_adapter *adapter)
2160 {
2161         struct net_device *netdev = adapter->netdev;
2162         uint32_t rctl;
2163
2164         e1000_pci_clear_mwi(&adapter->hw);
2165
2166         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2167         rctl |= E1000_RCTL_RST;
2168         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2169         E1000_WRITE_FLUSH(&adapter->hw);
2170         mdelay(5);
2171
2172         if (netif_running(netdev))
2173                 e1000_clean_all_rx_rings(adapter);
2174 }
2175
2176 static void
2177 e1000_leave_82542_rst(struct e1000_adapter *adapter)
2178 {
2179         struct net_device *netdev = adapter->netdev;
2180         uint32_t rctl;
2181
2182         rctl = E1000_READ_REG(&adapter->hw, RCTL);
2183         rctl &= ~E1000_RCTL_RST;
2184         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
2185         E1000_WRITE_FLUSH(&adapter->hw);
2186         mdelay(5);
2187
2188         if (adapter->hw.pci_cmd_word & PCI_COMMAND_INVALIDATE)
2189                 e1000_pci_set_mwi(&adapter->hw);
2190
2191         if (netif_running(netdev)) {
2192                 /* No need to loop, because 82542 supports only 1 queue */
2193                 struct e1000_rx_ring *ring = &adapter->rx_ring[0];
2194                 e1000_configure_rx(adapter);
2195                 adapter->alloc_rx_buf(adapter, ring, E1000_DESC_UNUSED(ring));
2196         }
2197 }
2198
2199 /**
2200  * e1000_set_mac - Change the Ethernet Address of the NIC
2201  * @netdev: network interface device structure
2202  * @p: pointer to an address structure
2203  *
2204  * Returns 0 on success, negative on failure
2205  **/
2206
2207 static int
2208 e1000_set_mac(struct net_device *netdev, void *p)
2209 {
2210         struct e1000_adapter *adapter = netdev_priv(netdev);
2211         struct sockaddr *addr = p;
2212
2213         if (!is_valid_ether_addr(addr->sa_data))
2214                 return -EADDRNOTAVAIL;
2215
2216         /* 82542 2.0 needs to be in reset to write receive address registers */
2217
2218         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2219                 e1000_enter_82542_rst(adapter);
2220
2221         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2222         memcpy(adapter->hw.mac_addr, addr->sa_data, netdev->addr_len);
2223
2224         e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2225
2226         /* With 82571 controllers, LAA may be overwritten (with the default)
2227          * due to controller reset from the other port. */
2228         if (adapter->hw.mac_type == e1000_82571) {
2229                 /* activate the work around */
2230                 adapter->hw.laa_is_present = 1;
2231
2232                 /* Hold a copy of the LAA in RAR[14] This is done so that
2233                  * between the time RAR[0] gets clobbered  and the time it
2234                  * gets fixed (in e1000_watchdog), the actual LAA is in one
2235                  * of the RARs and no incoming packets directed to this port
2236                  * are dropped. Eventaully the LAA will be in RAR[0] and
2237                  * RAR[14] */
2238                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr,
2239                                         E1000_RAR_ENTRIES - 1);
2240         }
2241
2242         if (adapter->hw.mac_type == e1000_82542_rev2_0)
2243                 e1000_leave_82542_rst(adapter);
2244
2245         return 0;
2246 }
2247
2248 /**
2249  * e1000_set_multi - Multicast and Promiscuous mode set
2250  * @netdev: network interface device structure
2251  *
2252  * The set_multi entry point is called whenever the multicast address
2253  * list or the network interface flags are updated.  This routine is
2254  * responsible for configuring the hardware for proper multicast,
2255  * promiscuous mode, and all-multi behavior.
2256  **/
2257
2258 static void
2259 e1000_set_multi(struct net_device *netdev)
2260 {
2261         struct e1000_adapter *adapter = netdev_priv(netdev);
2262         struct e1000_hw *hw = &adapter->hw;
2263         struct dev_mc_list *mc_ptr;
2264         uint32_t rctl;
2265         uint32_t hash_value;
2266         int i, rar_entries = E1000_RAR_ENTRIES;
2267
2268         /* reserve RAR[14] for LAA over-write work-around */
2269         if (adapter->hw.mac_type == e1000_82571)
2270                 rar_entries--;
2271
2272         /* Check for Promiscuous and All Multicast modes */
2273
2274         rctl = E1000_READ_REG(hw, RCTL);
2275
2276         if (netdev->flags & IFF_PROMISC) {
2277                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2278         } else if (netdev->flags & IFF_ALLMULTI) {
2279                 rctl |= E1000_RCTL_MPE;
2280                 rctl &= ~E1000_RCTL_UPE;
2281         } else {
2282                 rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2283         }
2284
2285         E1000_WRITE_REG(hw, RCTL, rctl);
2286
2287         /* 82542 2.0 needs to be in reset to write receive address registers */
2288
2289         if (hw->mac_type == e1000_82542_rev2_0)
2290                 e1000_enter_82542_rst(adapter);
2291
2292         /* load the first 14 multicast address into the exact filters 1-14
2293          * RAR 0 is used for the station MAC adddress
2294          * if there are not 14 addresses, go ahead and clear the filters
2295          * -- with 82571 controllers only 0-13 entries are filled here
2296          */
2297         mc_ptr = netdev->mc_list;
2298
2299         for (i = 1; i < rar_entries; i++) {
2300                 if (mc_ptr) {
2301                         e1000_rar_set(hw, mc_ptr->dmi_addr, i);
2302                         mc_ptr = mc_ptr->next;
2303                 } else {
2304                         E1000_WRITE_REG_ARRAY(hw, RA, i << 1, 0);
2305                         E1000_WRITE_REG_ARRAY(hw, RA, (i << 1) + 1, 0);
2306                 }
2307         }
2308
2309         /* clear the old settings from the multicast hash table */
2310
2311         for (i = 0; i < E1000_NUM_MTA_REGISTERS; i++)
2312                 E1000_WRITE_REG_ARRAY(hw, MTA, i, 0);
2313
2314         /* load any remaining addresses into the hash table */
2315
2316         for (; mc_ptr; mc_ptr = mc_ptr->next) {
2317                 hash_value = e1000_hash_mc_addr(hw, mc_ptr->dmi_addr);
2318                 e1000_mta_set(hw, hash_value);
2319         }
2320
2321         if (hw->mac_type == e1000_82542_rev2_0)
2322                 e1000_leave_82542_rst(adapter);
2323 }
2324
2325 /* Need to wait a few seconds after link up to get diagnostic information from
2326  * the phy */
2327
2328 static void
2329 e1000_update_phy_info(unsigned long data)
2330 {
2331         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2332         e1000_phy_get_info(&adapter->hw, &adapter->phy_info);
2333 }
2334
2335 /**
2336  * e1000_82547_tx_fifo_stall - Timer Call-back
2337  * @data: pointer to adapter cast into an unsigned long
2338  **/
2339
2340 static void
2341 e1000_82547_tx_fifo_stall(unsigned long data)
2342 {
2343         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2344         struct net_device *netdev = adapter->netdev;
2345         uint32_t tctl;
2346
2347         if (atomic_read(&adapter->tx_fifo_stall)) {
2348                 if ((E1000_READ_REG(&adapter->hw, TDT) ==
2349                     E1000_READ_REG(&adapter->hw, TDH)) &&
2350                    (E1000_READ_REG(&adapter->hw, TDFT) ==
2351                     E1000_READ_REG(&adapter->hw, TDFH)) &&
2352                    (E1000_READ_REG(&adapter->hw, TDFTS) ==
2353                     E1000_READ_REG(&adapter->hw, TDFHS))) {
2354                         tctl = E1000_READ_REG(&adapter->hw, TCTL);
2355                         E1000_WRITE_REG(&adapter->hw, TCTL,
2356                                         tctl & ~E1000_TCTL_EN);
2357                         E1000_WRITE_REG(&adapter->hw, TDFT,
2358                                         adapter->tx_head_addr);
2359                         E1000_WRITE_REG(&adapter->hw, TDFH,
2360                                         adapter->tx_head_addr);
2361                         E1000_WRITE_REG(&adapter->hw, TDFTS,
2362                                         adapter->tx_head_addr);
2363                         E1000_WRITE_REG(&adapter->hw, TDFHS,
2364                                         adapter->tx_head_addr);
2365                         E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
2366                         E1000_WRITE_FLUSH(&adapter->hw);
2367
2368                         adapter->tx_fifo_head = 0;
2369                         atomic_set(&adapter->tx_fifo_stall, 0);
2370                         netif_wake_queue(netdev);
2371                 } else {
2372                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies + 1);
2373                 }
2374         }
2375 }
2376
2377 /**
2378  * e1000_watchdog - Timer Call-back
2379  * @data: pointer to adapter cast into an unsigned long
2380  **/
2381 static void
2382 e1000_watchdog(unsigned long data)
2383 {
2384         struct e1000_adapter *adapter = (struct e1000_adapter *) data;
2385
2386         /* Do the rest outside of interrupt context */
2387         schedule_work(&adapter->watchdog_task);
2388 }
2389
2390 static void
2391 e1000_watchdog_task(struct e1000_adapter *adapter)
2392 {
2393         struct net_device *netdev = adapter->netdev;
2394         struct e1000_tx_ring *txdr = adapter->tx_ring;
2395         uint32_t link;
2396
2397         e1000_check_for_link(&adapter->hw);
2398         if (adapter->hw.mac_type == e1000_82573) {
2399                 e1000_enable_tx_pkt_filtering(&adapter->hw);
2400                 if (adapter->mng_vlan_id != adapter->hw.mng_cookie.vlan_id)
2401                         e1000_update_mng_vlan(adapter);
2402         }
2403
2404         if ((adapter->hw.media_type == e1000_media_type_internal_serdes) &&
2405            !(E1000_READ_REG(&adapter->hw, TXCW) & E1000_TXCW_ANE))
2406                 link = !adapter->hw.serdes_link_down;
2407         else
2408                 link = E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU;
2409
2410         if (link) {
2411                 if (!netif_carrier_ok(netdev)) {
2412                         e1000_get_speed_and_duplex(&adapter->hw,
2413                                                    &adapter->link_speed,
2414                                                    &adapter->link_duplex);
2415
2416                         DPRINTK(LINK, INFO, "NIC Link is Up %d Mbps %s\n",
2417                                adapter->link_speed,
2418                                adapter->link_duplex == FULL_DUPLEX ?
2419                                "Full Duplex" : "Half Duplex");
2420
2421                         /* tweak tx_queue_len according to speed/duplex */
2422                         netdev->tx_queue_len = adapter->tx_queue_len;
2423                         adapter->tx_timeout_factor = 1;
2424                         if (adapter->link_duplex == HALF_DUPLEX) {
2425                                 switch (adapter->link_speed) {
2426                                 case SPEED_10:
2427                                         netdev->tx_queue_len = 10;
2428                                         adapter->tx_timeout_factor = 8;
2429                                         break;
2430                                 case SPEED_100:
2431                                         netdev->tx_queue_len = 100;
2432                                         break;
2433                                 }
2434                         }
2435
2436                         netif_carrier_on(netdev);
2437                         netif_wake_queue(netdev);
2438                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2439                         adapter->smartspeed = 0;
2440                 }
2441         } else {
2442                 if (netif_carrier_ok(netdev)) {
2443                         adapter->link_speed = 0;
2444                         adapter->link_duplex = 0;
2445                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
2446                         netif_carrier_off(netdev);
2447                         netif_stop_queue(netdev);
2448                         mod_timer(&adapter->phy_info_timer, jiffies + 2 * HZ);
2449                 }
2450
2451                 e1000_smartspeed(adapter);
2452         }
2453
2454         e1000_update_stats(adapter);
2455
2456         adapter->hw.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2457         adapter->tpt_old = adapter->stats.tpt;
2458         adapter->hw.collision_delta = adapter->stats.colc - adapter->colc_old;
2459         adapter->colc_old = adapter->stats.colc;
2460
2461         adapter->gorcl = adapter->stats.gorcl - adapter->gorcl_old;
2462         adapter->gorcl_old = adapter->stats.gorcl;
2463         adapter->gotcl = adapter->stats.gotcl - adapter->gotcl_old;
2464         adapter->gotcl_old = adapter->stats.gotcl;
2465
2466         e1000_update_adaptive(&adapter->hw);
2467
2468 #ifdef CONFIG_E1000_MQ
2469         txdr = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
2470 #endif
2471         if (!netif_carrier_ok(netdev)) {
2472                 if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
2473                         /* We've lost link, so the controller stops DMA,
2474                          * but we've got queued Tx work that's never going
2475                          * to get done, so reset controller to flush Tx.
2476                          * (Do the reset outside of interrupt context). */
2477                         schedule_work(&adapter->tx_timeout_task);
2478                 }
2479         }
2480
2481         /* Dynamic mode for Interrupt Throttle Rate (ITR) */
2482         if (adapter->hw.mac_type >= e1000_82540 && adapter->itr == 1) {
2483                 /* Symmetric Tx/Rx gets a reduced ITR=2000; Total
2484                  * asymmetrical Tx or Rx gets ITR=8000; everyone
2485                  * else is between 2000-8000. */
2486                 uint32_t goc = (adapter->gotcl + adapter->gorcl) / 10000;
2487                 uint32_t dif = (adapter->gotcl > adapter->gorcl ?
2488                         adapter->gotcl - adapter->gorcl :
2489                         adapter->gorcl - adapter->gotcl) / 10000;
2490                 uint32_t itr = goc > 0 ? (dif * 6000 / goc + 2000) : 8000;
2491                 E1000_WRITE_REG(&adapter->hw, ITR, 1000000000 / (itr * 256));
2492         }
2493
2494         /* Cause software interrupt to ensure rx ring is cleaned */
2495         E1000_WRITE_REG(&adapter->hw, ICS, E1000_ICS_RXDMT0);
2496
2497         /* Force detection of hung controller every watchdog period */
2498         adapter->detect_tx_hung = TRUE;
2499
2500         /* With 82571 controllers, LAA may be overwritten due to controller
2501          * reset from the other port. Set the appropriate LAA in RAR[0] */
2502         if (adapter->hw.mac_type == e1000_82571 && adapter->hw.laa_is_present)
2503                 e1000_rar_set(&adapter->hw, adapter->hw.mac_addr, 0);
2504
2505         /* Reset the timer */
2506         mod_timer(&adapter->watchdog_timer, jiffies + 2 * HZ);
2507 }
2508
2509 #define E1000_TX_FLAGS_CSUM             0x00000001
2510 #define E1000_TX_FLAGS_VLAN             0x00000002
2511 #define E1000_TX_FLAGS_TSO              0x00000004
2512 #define E1000_TX_FLAGS_IPV4             0x00000008
2513 #define E1000_TX_FLAGS_VLAN_MASK        0xffff0000
2514 #define E1000_TX_FLAGS_VLAN_SHIFT       16
2515
2516 static inline int
2517 e1000_tso(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2518           struct sk_buff *skb)
2519 {
2520 #ifdef NETIF_F_TSO
2521         struct e1000_context_desc *context_desc;
2522         struct e1000_buffer *buffer_info;
2523         unsigned int i;
2524         uint32_t cmd_length = 0;
2525         uint16_t ipcse = 0, tucse, mss;
2526         uint8_t ipcss, ipcso, tucss, tucso, hdr_len;
2527         int err;
2528
2529         if (skb_shinfo(skb)->tso_size) {
2530                 if (skb_header_cloned(skb)) {
2531                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2532                         if (err)
2533                                 return err;
2534                 }
2535
2536                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2537                 mss = skb_shinfo(skb)->tso_size;
2538                 if (skb->protocol == ntohs(ETH_P_IP)) {
2539                         skb->nh.iph->tot_len = 0;
2540                         skb->nh.iph->check = 0;
2541                         skb->h.th->check =
2542                                 ~csum_tcpudp_magic(skb->nh.iph->saddr,
2543                                                    skb->nh.iph->daddr,
2544                                                    0,
2545                                                    IPPROTO_TCP,
2546                                                    0);
2547                         cmd_length = E1000_TXD_CMD_IP;
2548                         ipcse = skb->h.raw - skb->data - 1;
2549 #ifdef NETIF_F_TSO_IPV6
2550                 } else if (skb->protocol == ntohs(ETH_P_IPV6)) {
2551                         skb->nh.ipv6h->payload_len = 0;
2552                         skb->h.th->check =
2553                                 ~csum_ipv6_magic(&skb->nh.ipv6h->saddr,
2554                                                  &skb->nh.ipv6h->daddr,
2555                                                  0,
2556                                                  IPPROTO_TCP,
2557                                                  0);
2558                         ipcse = 0;
2559 #endif
2560                 }
2561                 ipcss = skb->nh.raw - skb->data;
2562                 ipcso = (void *)&(skb->nh.iph->check) - (void *)skb->data;
2563                 tucss = skb->h.raw - skb->data;
2564                 tucso = (void *)&(skb->h.th->check) - (void *)skb->data;
2565                 tucse = 0;
2566
2567                 cmd_length |= (E1000_TXD_CMD_DEXT | E1000_TXD_CMD_TSE |
2568                                E1000_TXD_CMD_TCP | (skb->len - (hdr_len)));
2569
2570                 i = tx_ring->next_to_use;
2571                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2572                 buffer_info = &tx_ring->buffer_info[i];
2573
2574                 context_desc->lower_setup.ip_fields.ipcss  = ipcss;
2575                 context_desc->lower_setup.ip_fields.ipcso  = ipcso;
2576                 context_desc->lower_setup.ip_fields.ipcse  = cpu_to_le16(ipcse);
2577                 context_desc->upper_setup.tcp_fields.tucss = tucss;
2578                 context_desc->upper_setup.tcp_fields.tucso = tucso;
2579                 context_desc->upper_setup.tcp_fields.tucse = cpu_to_le16(tucse);
2580                 context_desc->tcp_seg_setup.fields.mss     = cpu_to_le16(mss);
2581                 context_desc->tcp_seg_setup.fields.hdr_len = hdr_len;
2582                 context_desc->cmd_and_length = cpu_to_le32(cmd_length);
2583
2584                 buffer_info->time_stamp = jiffies;
2585
2586                 if (++i == tx_ring->count) i = 0;
2587                 tx_ring->next_to_use = i;
2588
2589                 return TRUE;
2590         }
2591 #endif
2592
2593         return FALSE;
2594 }
2595
2596 static inline boolean_t
2597 e1000_tx_csum(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2598               struct sk_buff *skb)
2599 {
2600         struct e1000_context_desc *context_desc;
2601         struct e1000_buffer *buffer_info;
2602         unsigned int i;
2603         uint8_t css;
2604
2605         if (likely(skb->ip_summed == CHECKSUM_HW)) {
2606                 css = skb->h.raw - skb->data;
2607
2608                 i = tx_ring->next_to_use;
2609                 buffer_info = &tx_ring->buffer_info[i];
2610                 context_desc = E1000_CONTEXT_DESC(*tx_ring, i);
2611
2612                 context_desc->upper_setup.tcp_fields.tucss = css;
2613                 context_desc->upper_setup.tcp_fields.tucso = css + skb->csum;
2614                 context_desc->upper_setup.tcp_fields.tucse = 0;
2615                 context_desc->tcp_seg_setup.data = 0;
2616                 context_desc->cmd_and_length = cpu_to_le32(E1000_TXD_CMD_DEXT);
2617
2618                 buffer_info->time_stamp = jiffies;
2619
2620                 if (unlikely(++i == tx_ring->count)) i = 0;
2621                 tx_ring->next_to_use = i;
2622
2623                 return TRUE;
2624         }
2625
2626         return FALSE;
2627 }
2628
2629 #define E1000_MAX_TXD_PWR       12
2630 #define E1000_MAX_DATA_PER_TXD  (1<<E1000_MAX_TXD_PWR)
2631
2632 static inline int
2633 e1000_tx_map(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2634              struct sk_buff *skb, unsigned int first, unsigned int max_per_txd,
2635              unsigned int nr_frags, unsigned int mss)
2636 {
2637         struct e1000_buffer *buffer_info;
2638         unsigned int len = skb->len;
2639         unsigned int offset = 0, size, count = 0, i;
2640         unsigned int f;
2641         len -= skb->data_len;
2642
2643         i = tx_ring->next_to_use;
2644
2645         while (len) {
2646                 buffer_info = &tx_ring->buffer_info[i];
2647                 size = min(len, max_per_txd);
2648 #ifdef NETIF_F_TSO
2649                 /* Workaround for Controller erratum --
2650                  * descriptor for non-tso packet in a linear SKB that follows a
2651                  * tso gets written back prematurely before the data is fully
2652                  * DMAd to the controller */
2653                 if (!skb->data_len && tx_ring->last_tx_tso &&
2654                                 !skb_shinfo(skb)->tso_size) {
2655                         tx_ring->last_tx_tso = 0;
2656                         size -= 4;
2657                 }
2658
2659                 /* Workaround for premature desc write-backs
2660                  * in TSO mode.  Append 4-byte sentinel desc */
2661                 if (unlikely(mss && !nr_frags && size == len && size > 8))
2662                         size -= 4;
2663 #endif
2664                 /* work-around for errata 10 and it applies
2665                  * to all controllers in PCI-X mode
2666                  * The fix is to make sure that the first descriptor of a
2667                  * packet is smaller than 2048 - 16 - 16 (or 2016) bytes
2668                  */
2669                 if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
2670                                 (size > 2015) && count == 0))
2671                         size = 2015;
2672
2673                 /* Workaround for potential 82544 hang in PCI-X.  Avoid
2674                  * terminating buffers within evenly-aligned dwords. */
2675                 if (unlikely(adapter->pcix_82544 &&
2676                    !((unsigned long)(skb->data + offset + size - 1) & 4) &&
2677                    size > 4))
2678                         size -= 4;
2679
2680                 buffer_info->length = size;
2681                 buffer_info->dma =
2682                         pci_map_single(adapter->pdev,
2683                                 skb->data + offset,
2684                                 size,
2685                                 PCI_DMA_TODEVICE);
2686                 buffer_info->time_stamp = jiffies;
2687
2688                 len -= size;
2689                 offset += size;
2690                 count++;
2691                 if (unlikely(++i == tx_ring->count)) i = 0;
2692         }
2693
2694         for (f = 0; f < nr_frags; f++) {
2695                 struct skb_frag_struct *frag;
2696
2697                 frag = &skb_shinfo(skb)->frags[f];
2698                 len = frag->size;
2699                 offset = frag->page_offset;
2700
2701                 while (len) {
2702                         buffer_info = &tx_ring->buffer_info[i];
2703                         size = min(len, max_per_txd);
2704 #ifdef NETIF_F_TSO
2705                         /* Workaround for premature desc write-backs
2706                          * in TSO mode.  Append 4-byte sentinel desc */
2707                         if (unlikely(mss && f == (nr_frags-1) && size == len && size > 8))
2708                                 size -= 4;
2709 #endif
2710                         /* Workaround for potential 82544 hang in PCI-X.
2711                          * Avoid terminating buffers within evenly-aligned
2712                          * dwords. */
2713                         if (unlikely(adapter->pcix_82544 &&
2714                            !((unsigned long)(frag->page+offset+size-1) & 4) &&
2715                            size > 4))
2716                                 size -= 4;
2717
2718                         buffer_info->length = size;
2719                         buffer_info->dma =
2720                                 pci_map_page(adapter->pdev,
2721                                         frag->page,
2722                                         offset,
2723                                         size,
2724                                         PCI_DMA_TODEVICE);
2725                         buffer_info->time_stamp = jiffies;
2726
2727                         len -= size;
2728                         offset += size;
2729                         count++;
2730                         if (unlikely(++i == tx_ring->count)) i = 0;
2731                 }
2732         }
2733
2734         i = (i == 0) ? tx_ring->count - 1 : i - 1;
2735         tx_ring->buffer_info[i].skb = skb;
2736         tx_ring->buffer_info[first].next_to_watch = i;
2737
2738         return count;
2739 }
2740
2741 static inline void
2742 e1000_tx_queue(struct e1000_adapter *adapter, struct e1000_tx_ring *tx_ring,
2743                int tx_flags, int count)
2744 {
2745         struct e1000_tx_desc *tx_desc = NULL;
2746         struct e1000_buffer *buffer_info;
2747         uint32_t txd_upper = 0, txd_lower = E1000_TXD_CMD_IFCS;
2748         unsigned int i;
2749
2750         if (likely(tx_flags & E1000_TX_FLAGS_TSO)) {
2751                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D |
2752                              E1000_TXD_CMD_TSE;
2753                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2754
2755                 if (likely(tx_flags & E1000_TX_FLAGS_IPV4))
2756                         txd_upper |= E1000_TXD_POPTS_IXSM << 8;
2757         }
2758
2759         if (likely(tx_flags & E1000_TX_FLAGS_CSUM)) {
2760                 txd_lower |= E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2761                 txd_upper |= E1000_TXD_POPTS_TXSM << 8;
2762         }
2763
2764         if (unlikely(tx_flags & E1000_TX_FLAGS_VLAN)) {
2765                 txd_lower |= E1000_TXD_CMD_VLE;
2766                 txd_upper |= (tx_flags & E1000_TX_FLAGS_VLAN_MASK);
2767         }
2768
2769         i = tx_ring->next_to_use;
2770
2771         while (count--) {
2772                 buffer_info = &tx_ring->buffer_info[i];
2773                 tx_desc = E1000_TX_DESC(*tx_ring, i);
2774                 tx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
2775                 tx_desc->lower.data =
2776                         cpu_to_le32(txd_lower | buffer_info->length);
2777                 tx_desc->upper.data = cpu_to_le32(txd_upper);
2778                 if (unlikely(++i == tx_ring->count)) i = 0;
2779         }
2780
2781         tx_desc->lower.data |= cpu_to_le32(adapter->txd_cmd);
2782
2783         /* Force memory writes to complete before letting h/w
2784          * know there are new descriptors to fetch.  (Only
2785          * applicable for weak-ordered memory model archs,
2786          * such as IA-64). */
2787         wmb();
2788
2789         tx_ring->next_to_use = i;
2790         writel(i, adapter->hw.hw_addr + tx_ring->tdt);
2791 }
2792
2793 /**
2794  * 82547 workaround to avoid controller hang in half-duplex environment.
2795  * The workaround is to avoid queuing a large packet that would span
2796  * the internal Tx FIFO ring boundary by notifying the stack to resend
2797  * the packet at a later time.  This gives the Tx FIFO an opportunity to
2798  * flush all packets.  When that occurs, we reset the Tx FIFO pointers
2799  * to the beginning of the Tx FIFO.
2800  **/
2801
2802 #define E1000_FIFO_HDR                  0x10
2803 #define E1000_82547_PAD_LEN             0x3E0
2804
2805 static inline int
2806 e1000_82547_fifo_workaround(struct e1000_adapter *adapter, struct sk_buff *skb)
2807 {
2808         uint32_t fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
2809         uint32_t skb_fifo_len = skb->len + E1000_FIFO_HDR;
2810
2811         E1000_ROUNDUP(skb_fifo_len, E1000_FIFO_HDR);
2812
2813         if (adapter->link_duplex != HALF_DUPLEX)
2814                 goto no_fifo_stall_required;
2815
2816         if (atomic_read(&adapter->tx_fifo_stall))
2817                 return 1;
2818
2819         if (skb_fifo_len >= (E1000_82547_PAD_LEN + fifo_space)) {
2820                 atomic_set(&adapter->tx_fifo_stall, 1);
2821                 return 1;
2822         }
2823
2824 no_fifo_stall_required:
2825         adapter->tx_fifo_head += skb_fifo_len;
2826         if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
2827                 adapter->tx_fifo_head -= adapter->tx_fifo_size;
2828         return 0;
2829 }
2830
2831 #define MINIMUM_DHCP_PACKET_SIZE 282
2832 static inline int
2833 e1000_transfer_dhcp_info(struct e1000_adapter *adapter, struct sk_buff *skb)
2834 {
2835         struct e1000_hw *hw =  &adapter->hw;
2836         uint16_t length, offset;
2837         if (vlan_tx_tag_present(skb)) {
2838                 if (!((vlan_tx_tag_get(skb) == adapter->hw.mng_cookie.vlan_id) &&
2839                         ( adapter->hw.mng_cookie.status &
2840                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT)) )
2841                         return 0;
2842         }
2843         if ((skb->len > MINIMUM_DHCP_PACKET_SIZE) && (!skb->protocol)) {
2844                 struct ethhdr *eth = (struct ethhdr *) skb->data;
2845                 if ((htons(ETH_P_IP) == eth->h_proto)) {
2846                         const struct iphdr *ip =
2847                                 (struct iphdr *)((uint8_t *)skb->data+14);
2848                         if (IPPROTO_UDP == ip->protocol) {
2849                                 struct udphdr *udp =
2850                                         (struct udphdr *)((uint8_t *)ip +
2851                                                 (ip->ihl << 2));
2852                                 if (ntohs(udp->dest) == 67) {
2853                                         offset = (uint8_t *)udp + 8 - skb->data;
2854                                         length = skb->len - offset;
2855
2856                                         return e1000_mng_write_dhcp_info(hw,
2857                                                         (uint8_t *)udp + 8,
2858                                                         length);
2859                                 }
2860                         }
2861                 }
2862         }
2863         return 0;
2864 }
2865
2866 #define TXD_USE_COUNT(S, X) (((S) >> (X)) + 1 )
2867 static int
2868 e1000_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2869 {
2870         struct e1000_adapter *adapter = netdev_priv(netdev);
2871         struct e1000_tx_ring *tx_ring;
2872         unsigned int first, max_per_txd = E1000_MAX_DATA_PER_TXD;
2873         unsigned int max_txd_pwr = E1000_MAX_TXD_PWR;
2874         unsigned int tx_flags = 0;
2875         unsigned int len = skb->len;
2876         unsigned long flags;
2877         unsigned int nr_frags = 0;
2878         unsigned int mss = 0;
2879         int count = 0;
2880         int tso;
2881         unsigned int f;
2882         len -= skb->data_len;
2883
2884 #ifdef CONFIG_E1000_MQ
2885         tx_ring = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
2886 #else
2887         tx_ring = adapter->tx_ring;
2888 #endif
2889
2890         if (unlikely(skb->len <= 0)) {
2891                 dev_kfree_skb_any(skb);
2892                 return NETDEV_TX_OK;
2893         }
2894
2895 #ifdef NETIF_F_TSO
2896         mss = skb_shinfo(skb)->tso_size;
2897         /* The controller does a simple calculation to
2898          * make sure there is enough room in the FIFO before
2899          * initiating the DMA for each buffer.  The calc is:
2900          * 4 = ceil(buffer len/mss).  To make sure we don't
2901          * overrun the FIFO, adjust the max buffer len if mss
2902          * drops. */
2903         if (mss) {
2904                 uint8_t hdr_len;
2905                 max_per_txd = min(mss << 2, max_per_txd);
2906                 max_txd_pwr = fls(max_per_txd) - 1;
2907
2908         /* TSO Workaround for 82571/2 Controllers -- if skb->data
2909          * points to just header, pull a few bytes of payload from
2910          * frags into skb->data */
2911                 hdr_len = ((skb->h.raw - skb->data) + (skb->h.th->doff << 2));
2912                 if (skb->data_len && (hdr_len == (skb->len - skb->data_len)) &&
2913                         (adapter->hw.mac_type == e1000_82571 ||
2914                         adapter->hw.mac_type == e1000_82572)) {
2915                         unsigned int pull_size;
2916                         pull_size = min((unsigned int)4, skb->data_len);
2917                         if (!__pskb_pull_tail(skb, pull_size)) {
2918                                 printk(KERN_ERR "__pskb_pull_tail failed.\n");
2919                                 dev_kfree_skb_any(skb);
2920                                 return NETDEV_TX_OK;
2921                         }
2922                         len = skb->len - skb->data_len;
2923                 }
2924         }
2925
2926         /* reserve a descriptor for the offload context */
2927         if ((mss) || (skb->ip_summed == CHECKSUM_HW))
2928                 count++;
2929         count++;
2930 #else
2931         if (skb->ip_summed == CHECKSUM_HW)
2932                 count++;
2933 #endif
2934
2935 #ifdef NETIF_F_TSO
2936         /* Controller Erratum workaround */
2937         if (!skb->data_len && tx_ring->last_tx_tso &&
2938                 !skb_shinfo(skb)->tso_size)
2939                 count++;
2940 #endif
2941
2942         count += TXD_USE_COUNT(len, max_txd_pwr);
2943
2944         if (adapter->pcix_82544)
2945                 count++;
2946
2947         /* work-around for errata 10 and it applies to all controllers
2948          * in PCI-X mode, so add one more descriptor to the count
2949          */
2950         if (unlikely((adapter->hw.bus_type == e1000_bus_type_pcix) &&
2951                         (len > 2015)))
2952                 count++;
2953
2954         nr_frags = skb_shinfo(skb)->nr_frags;
2955         for (f = 0; f < nr_frags; f++)
2956                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size,
2957                                        max_txd_pwr);
2958         if (adapter->pcix_82544)
2959                 count += nr_frags;
2960
2961         if (adapter->hw.tx_pkt_filtering && (adapter->hw.mac_type == e1000_82573) )
2962                 e1000_transfer_dhcp_info(adapter, skb);
2963
2964         local_irq_save(flags);
2965         if (!spin_trylock(&tx_ring->tx_lock)) {
2966                 /* Collision - tell upper layer to requeue */
2967                 local_irq_restore(flags);
2968                 return NETDEV_TX_LOCKED;
2969         }
2970
2971         /* need: count + 2 desc gap to keep tail from touching
2972          * head, otherwise try next time */
2973         if (unlikely(E1000_DESC_UNUSED(tx_ring) < count + 2)) {
2974                 netif_stop_queue(netdev);
2975                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2976                 return NETDEV_TX_BUSY;
2977         }
2978
2979         if (unlikely(adapter->hw.mac_type == e1000_82547)) {
2980                 if (unlikely(e1000_82547_fifo_workaround(adapter, skb))) {
2981                         netif_stop_queue(netdev);
2982                         mod_timer(&adapter->tx_fifo_stall_timer, jiffies);
2983                         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2984                         return NETDEV_TX_BUSY;
2985                 }
2986         }
2987
2988         if (unlikely(adapter->vlgrp && vlan_tx_tag_present(skb))) {
2989                 tx_flags |= E1000_TX_FLAGS_VLAN;
2990                 tx_flags |= (vlan_tx_tag_get(skb) << E1000_TX_FLAGS_VLAN_SHIFT);
2991         }
2992
2993         first = tx_ring->next_to_use;
2994
2995         tso = e1000_tso(adapter, tx_ring, skb);
2996         if (tso < 0) {
2997                 dev_kfree_skb_any(skb);
2998                 spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
2999                 return NETDEV_TX_OK;
3000         }
3001
3002         if (likely(tso)) {
3003                 tx_ring->last_tx_tso = 1;
3004                 tx_flags |= E1000_TX_FLAGS_TSO;
3005         } else if (likely(e1000_tx_csum(adapter, tx_ring, skb)))
3006                 tx_flags |= E1000_TX_FLAGS_CSUM;
3007
3008         /* Old method was to assume IPv4 packet by default if TSO was enabled.
3009          * 82571 hardware supports TSO capabilities for IPv6 as well...
3010          * no longer assume, we must. */
3011         if (likely(skb->protocol == ntohs(ETH_P_IP)))
3012                 tx_flags |= E1000_TX_FLAGS_IPV4;
3013
3014         e1000_tx_queue(adapter, tx_ring, tx_flags,
3015                        e1000_tx_map(adapter, tx_ring, skb, first,
3016                                     max_per_txd, nr_frags, mss));
3017
3018         netdev->trans_start = jiffies;
3019
3020         /* Make sure there is space in the ring for the next send. */
3021         if (unlikely(E1000_DESC_UNUSED(tx_ring) < MAX_SKB_FRAGS + 2))
3022                 netif_stop_queue(netdev);
3023
3024         spin_unlock_irqrestore(&tx_ring->tx_lock, flags);
3025         return NETDEV_TX_OK;
3026 }
3027
3028 /**
3029  * e1000_tx_timeout - Respond to a Tx Hang
3030  * @netdev: network interface device structure
3031  **/
3032
3033 static void
3034 e1000_tx_timeout(struct net_device *netdev)
3035 {
3036         struct e1000_adapter *adapter = netdev_priv(netdev);
3037
3038         /* Do the reset outside of interrupt context */
3039         schedule_work(&adapter->tx_timeout_task);
3040 }
3041
3042 static void
3043 e1000_tx_timeout_task(struct net_device *netdev)
3044 {
3045         struct e1000_adapter *adapter = netdev_priv(netdev);
3046
3047         adapter->tx_timeout_count++;
3048         e1000_down(adapter);
3049         e1000_up(adapter);
3050 }
3051
3052 /**
3053  * e1000_get_stats - Get System Network Statistics
3054  * @netdev: network interface device structure
3055  *
3056  * Returns the address of the device statistics structure.
3057  * The statistics are actually updated from the timer callback.
3058  **/
3059
3060 static struct net_device_stats *
3061 e1000_get_stats(struct net_device *netdev)
3062 {
3063         struct e1000_adapter *adapter = netdev_priv(netdev);
3064
3065         /* only return the current stats */
3066         return &adapter->net_stats;
3067 }
3068
3069 /**
3070  * e1000_change_mtu - Change the Maximum Transfer Unit
3071  * @netdev: network interface device structure
3072  * @new_mtu: new value for maximum frame size
3073  *
3074  * Returns 0 on success, negative on failure
3075  **/
3076
3077 static int
3078 e1000_change_mtu(struct net_device *netdev, int new_mtu)
3079 {
3080         struct e1000_adapter *adapter = netdev_priv(netdev);
3081         int max_frame = new_mtu + ENET_HEADER_SIZE + ETHERNET_FCS_SIZE;
3082
3083         if ((max_frame < MINIMUM_ETHERNET_FRAME_SIZE) ||
3084             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3085                 DPRINTK(PROBE, ERR, "Invalid MTU setting\n");
3086                 return -EINVAL;
3087         }
3088
3089         /* Adapter-specific max frame size limits. */
3090         switch (adapter->hw.mac_type) {
3091         case e1000_82542_rev2_0:
3092         case e1000_82542_rev2_1:
3093         case e1000_82573:
3094                 if (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE) {
3095                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported.\n");
3096                         return -EINVAL;
3097                 }
3098                 break;
3099         case e1000_82571:
3100         case e1000_82572:
3101 #define MAX_STD_JUMBO_FRAME_SIZE 9234
3102                 if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3103                         DPRINTK(PROBE, ERR, "MTU > 9216 not supported.\n");
3104                         return -EINVAL;
3105                 }
3106                 break;
3107         default:
3108                 /* Capable of supporting up to MAX_JUMBO_FRAME_SIZE limit. */
3109                 break;
3110         }
3111
3112
3113         if (adapter->hw.mac_type > e1000_82547_rev_2) {
3114                 adapter->rx_buffer_len = max_frame;
3115                 E1000_ROUNDUP(adapter->rx_buffer_len, 1024);
3116         } else {
3117                 if(unlikely((adapter->hw.mac_type < e1000_82543) &&
3118                    (max_frame > MAXIMUM_ETHERNET_FRAME_SIZE))) {
3119                         DPRINTK(PROBE, ERR, "Jumbo Frames not supported "
3120                                             "on 82542\n");
3121                         return -EINVAL;
3122                 } else {
3123                         if(max_frame <= E1000_RXBUFFER_2048)
3124                                 adapter->rx_buffer_len = E1000_RXBUFFER_2048;
3125                         else if(max_frame <= E1000_RXBUFFER_4096)
3126                                 adapter->rx_buffer_len = E1000_RXBUFFER_4096;
3127                         else if(max_frame <= E1000_RXBUFFER_8192)
3128                                 adapter->rx_buffer_len = E1000_RXBUFFER_8192;
3129                         else if(max_frame <= E1000_RXBUFFER_16384)
3130                                 adapter->rx_buffer_len = E1000_RXBUFFER_16384;
3131                 }
3132         }
3133
3134         netdev->mtu = new_mtu;
3135
3136         if (netif_running(netdev)) {
3137                 e1000_down(adapter);
3138                 e1000_up(adapter);
3139         }
3140
3141         adapter->hw.max_frame_size = max_frame;
3142
3143         return 0;
3144 }
3145
3146 /**
3147  * e1000_update_stats - Update the board statistics counters
3148  * @adapter: board private structure
3149  **/
3150
3151 void
3152 e1000_update_stats(struct e1000_adapter *adapter)
3153 {
3154         struct e1000_hw *hw = &adapter->hw;
3155         unsigned long flags;
3156         uint16_t phy_tmp;
3157
3158 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3159
3160         spin_lock_irqsave(&adapter->stats_lock, flags);
3161
3162         /* these counters are modified from e1000_adjust_tbi_stats,
3163          * called from the interrupt context, so they must only
3164          * be written while holding adapter->stats_lock
3165          */
3166
3167         adapter->stats.crcerrs += E1000_READ_REG(hw, CRCERRS);
3168         adapter->stats.gprc += E1000_READ_REG(hw, GPRC);
3169         adapter->stats.gorcl += E1000_READ_REG(hw, GORCL);
3170         adapter->stats.gorch += E1000_READ_REG(hw, GORCH);
3171         adapter->stats.bprc += E1000_READ_REG(hw, BPRC);
3172         adapter->stats.mprc += E1000_READ_REG(hw, MPRC);
3173         adapter->stats.roc += E1000_READ_REG(hw, ROC);
3174         adapter->stats.prc64 += E1000_READ_REG(hw, PRC64);
3175         adapter->stats.prc127 += E1000_READ_REG(hw, PRC127);
3176         adapter->stats.prc255 += E1000_READ_REG(hw, PRC255);
3177         adapter->stats.prc511 += E1000_READ_REG(hw, PRC511);
3178         adapter->stats.prc1023 += E1000_READ_REG(hw, PRC1023);
3179         adapter->stats.prc1522 += E1000_READ_REG(hw, PRC1522);
3180
3181         adapter->stats.symerrs += E1000_READ_REG(hw, SYMERRS);
3182         adapter->stats.mpc += E1000_READ_REG(hw, MPC);
3183         adapter->stats.scc += E1000_READ_REG(hw, SCC);
3184         adapter->stats.ecol += E1000_READ_REG(hw, ECOL);
3185         adapter->stats.mcc += E1000_READ_REG(hw, MCC);
3186         adapter->stats.latecol += E1000_READ_REG(hw, LATECOL);
3187         adapter->stats.dc += E1000_READ_REG(hw, DC);
3188         adapter->stats.sec += E1000_READ_REG(hw, SEC);
3189         adapter->stats.rlec += E1000_READ_REG(hw, RLEC);
3190         adapter->stats.xonrxc += E1000_READ_REG(hw, XONRXC);
3191         adapter->stats.xontxc += E1000_READ_REG(hw, XONTXC);
3192         adapter->stats.xoffrxc += E1000_READ_REG(hw, XOFFRXC);
3193         adapter->stats.xofftxc += E1000_READ_REG(hw, XOFFTXC);
3194         adapter->stats.fcruc += E1000_READ_REG(hw, FCRUC);
3195         adapter->stats.gptc += E1000_READ_REG(hw, GPTC);
3196         adapter->stats.gotcl += E1000_READ_REG(hw, GOTCL);
3197         adapter->stats.gotch += E1000_READ_REG(hw, GOTCH);
3198         adapter->stats.rnbc += E1000_READ_REG(hw, RNBC);
3199         adapter->stats.ruc += E1000_READ_REG(hw, RUC);
3200         adapter->stats.rfc += E1000_READ_REG(hw, RFC);
3201         adapter->stats.rjc += E1000_READ_REG(hw, RJC);
3202         adapter->stats.torl += E1000_READ_REG(hw, TORL);
3203         adapter->stats.torh += E1000_READ_REG(hw, TORH);
3204         adapter->stats.totl += E1000_READ_REG(hw, TOTL);
3205         adapter->stats.toth += E1000_READ_REG(hw, TOTH);
3206         adapter->stats.tpr += E1000_READ_REG(hw, TPR);
3207         adapter->stats.ptc64 += E1000_READ_REG(hw, PTC64);
3208         adapter->stats.ptc127 += E1000_READ_REG(hw, PTC127);
3209         adapter->stats.ptc255 += E1000_READ_REG(hw, PTC255);
3210         adapter->stats.ptc511 += E1000_READ_REG(hw, PTC511);
3211         adapter->stats.ptc1023 += E1000_READ_REG(hw, PTC1023);
3212         adapter->stats.ptc1522 += E1000_READ_REG(hw, PTC1522);
3213         adapter->stats.mptc += E1000_READ_REG(hw, MPTC);
3214         adapter->stats.bptc += E1000_READ_REG(hw, BPTC);
3215
3216         /* used for adaptive IFS */
3217
3218         hw->tx_packet_delta = E1000_READ_REG(hw, TPT);
3219         adapter->stats.tpt += hw->tx_packet_delta;
3220         hw->collision_delta = E1000_READ_REG(hw, COLC);
3221         adapter->stats.colc += hw->collision_delta;
3222
3223         if (hw->mac_type >= e1000_82543) {
3224                 adapter->stats.algnerrc += E1000_READ_REG(hw, ALGNERRC);
3225                 adapter->stats.rxerrc += E1000_READ_REG(hw, RXERRC);
3226                 adapter->stats.tncrs += E1000_READ_REG(hw, TNCRS);
3227                 adapter->stats.cexterr += E1000_READ_REG(hw, CEXTERR);
3228                 adapter->stats.tsctc += E1000_READ_REG(hw, TSCTC);
3229                 adapter->stats.tsctfc += E1000_READ_REG(hw, TSCTFC);
3230         }
3231         if (hw->mac_type > e1000_82547_rev_2) {
3232                 adapter->stats.iac += E1000_READ_REG(hw, IAC);
3233                 adapter->stats.icrxoc += E1000_READ_REG(hw, ICRXOC);
3234                 adapter->stats.icrxptc += E1000_READ_REG(hw, ICRXPTC);
3235                 adapter->stats.icrxatc += E1000_READ_REG(hw, ICRXATC);
3236                 adapter->stats.ictxptc += E1000_READ_REG(hw, ICTXPTC);
3237                 adapter->stats.ictxatc += E1000_READ_REG(hw, ICTXATC);
3238                 adapter->stats.ictxqec += E1000_READ_REG(hw, ICTXQEC);
3239                 adapter->stats.ictxqmtc += E1000_READ_REG(hw, ICTXQMTC);
3240                 adapter->stats.icrxdmtc += E1000_READ_REG(hw, ICRXDMTC);
3241         }
3242
3243         /* Fill out the OS statistics structure */
3244
3245         adapter->net_stats.rx_packets = adapter->stats.gprc;
3246         adapter->net_stats.tx_packets = adapter->stats.gptc;
3247         adapter->net_stats.rx_bytes = adapter->stats.gorcl;
3248         adapter->net_stats.tx_bytes = adapter->stats.gotcl;
3249         adapter->net_stats.multicast = adapter->stats.mprc;
3250         adapter->net_stats.collisions = adapter->stats.colc;
3251
3252         /* Rx Errors */
3253
3254         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3255                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3256                 adapter->stats.rlec + adapter->stats.cexterr;
3257         adapter->net_stats.rx_dropped = 0;
3258         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3259         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3260         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3261         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3262
3263         /* Tx Errors */
3264
3265         adapter->net_stats.tx_errors = adapter->stats.ecol +
3266                                        adapter->stats.latecol;
3267         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3268         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3269         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3270
3271         /* Tx Dropped needs to be maintained elsewhere */
3272
3273         /* Phy Stats */
3274
3275         if (hw->media_type == e1000_media_type_copper) {
3276                 if ((adapter->link_speed == SPEED_1000) &&
3277                    (!e1000_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3278                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3279                         adapter->phy_stats.idle_errors += phy_tmp;
3280                 }
3281
3282                 if ((hw->mac_type <= e1000_82546) &&
3283                    (hw->phy_type == e1000_phy_m88) &&
3284                    !e1000_read_phy_reg(hw, M88E1000_RX_ERR_CNTR, &phy_tmp))
3285                         adapter->phy_stats.receive_errors += phy_tmp;
3286         }
3287
3288         spin_unlock_irqrestore(&adapter->stats_lock, flags);
3289 }
3290
3291 #ifdef CONFIG_E1000_MQ
3292 void
3293 e1000_rx_schedule(void *data)
3294 {
3295         struct net_device *poll_dev, *netdev = data;
3296         struct e1000_adapter *adapter = netdev->priv;
3297         int this_cpu = get_cpu();
3298
3299         poll_dev = *per_cpu_ptr(adapter->cpu_netdev, this_cpu);
3300         if (poll_dev == NULL) {
3301                 put_cpu();
3302                 return;
3303         }
3304
3305         if (likely(netif_rx_schedule_prep(poll_dev)))
3306                 __netif_rx_schedule(poll_dev);
3307         else
3308                 e1000_irq_enable(adapter);
3309
3310         put_cpu();
3311 }
3312 #endif
3313
3314 /**
3315  * e1000_intr - Interrupt Handler
3316  * @irq: interrupt number
3317  * @data: pointer to a network interface device structure
3318  * @pt_regs: CPU registers structure
3319  **/
3320
3321 static irqreturn_t
3322 e1000_intr(int irq, void *data, struct pt_regs *regs)
3323 {
3324         struct net_device *netdev = data;
3325         struct e1000_adapter *adapter = netdev_priv(netdev);
3326         struct e1000_hw *hw = &adapter->hw;
3327         uint32_t icr = E1000_READ_REG(hw, ICR);
3328 #ifndef CONFIG_E1000_NAPI
3329         int i;
3330 #else
3331         /* Interrupt Auto-Mask...upon reading ICR,
3332          * interrupts are masked.  No need for the
3333          * IMC write, but it does mean we should
3334          * account for it ASAP. */
3335         if (likely(hw->mac_type >= e1000_82571))
3336                 atomic_inc(&adapter->irq_sem);
3337 #endif
3338
3339         if (unlikely(!icr)) {
3340 #ifdef CONFIG_E1000_NAPI
3341                 if (hw->mac_type >= e1000_82571)
3342                         e1000_irq_enable(adapter);
3343 #endif
3344                 return IRQ_NONE;  /* Not our interrupt */
3345         }
3346
3347         if (unlikely(icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC))) {
3348                 hw->get_link_status = 1;
3349                 mod_timer(&adapter->watchdog_timer, jiffies);
3350         }
3351
3352 #ifdef CONFIG_E1000_NAPI
3353         if (unlikely(hw->mac_type < e1000_82571)) {
3354                 atomic_inc(&adapter->irq_sem);
3355                 E1000_WRITE_REG(hw, IMC, ~0);
3356                 E1000_WRITE_FLUSH(hw);
3357         }
3358 #ifdef CONFIG_E1000_MQ
3359         if (atomic_read(&adapter->rx_sched_call_data.count) == 0) {
3360                 /* We must setup the cpumask once count == 0 since
3361                  * each cpu bit is cleared when the work is done. */
3362                 adapter->rx_sched_call_data.cpumask = adapter->cpumask;
3363                 atomic_add(adapter->num_rx_queues - 1, &adapter->irq_sem);
3364                 atomic_set(&adapter->rx_sched_call_data.count,
3365                            adapter->num_rx_queues);
3366                 smp_call_async_mask(&adapter->rx_sched_call_data);
3367         } else {
3368                 printk("call_data.count == %u\n", atomic_read(&adapter->rx_sched_call_data.count));
3369         }
3370 #else /* if !CONFIG_E1000_MQ */
3371         if (likely(netif_rx_schedule_prep(&adapter->polling_netdev[0])))
3372                 __netif_rx_schedule(&adapter->polling_netdev[0]);
3373         else
3374                 e1000_irq_enable(adapter);
3375 #endif /* CONFIG_E1000_MQ */
3376
3377 #else /* if !CONFIG_E1000_NAPI */
3378         /* Writing IMC and IMS is needed for 82547.
3379          * Due to Hub Link bus being occupied, an interrupt
3380          * de-assertion message is not able to be sent.
3381          * When an interrupt assertion message is generated later,
3382          * two messages are re-ordered and sent out.
3383          * That causes APIC to think 82547 is in de-assertion
3384          * state, while 82547 is in assertion state, resulting
3385          * in dead lock. Writing IMC forces 82547 into
3386          * de-assertion state.
3387          */
3388         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2) {
3389                 atomic_inc(&adapter->irq_sem);
3390                 E1000_WRITE_REG(hw, IMC, ~0);
3391         }
3392
3393         for (i = 0; i < E1000_MAX_INTR; i++)
3394                 if (unlikely(!adapter->clean_rx(adapter, adapter->rx_ring) &
3395                    !e1000_clean_tx_irq(adapter, adapter->tx_ring)))
3396                         break;
3397
3398         if (hw->mac_type == e1000_82547 || hw->mac_type == e1000_82547_rev_2)
3399                 e1000_irq_enable(adapter);
3400
3401 #endif /* CONFIG_E1000_NAPI */
3402
3403         return IRQ_HANDLED;
3404 }
3405
3406 #ifdef CONFIG_E1000_NAPI
3407 /**
3408  * e1000_clean - NAPI Rx polling callback
3409  * @adapter: board private structure
3410  **/
3411
3412 static int
3413 e1000_clean(struct net_device *poll_dev, int *budget)
3414 {
3415         struct e1000_adapter *adapter;
3416         int work_to_do = min(*budget, poll_dev->quota);
3417         int tx_cleaned = 0, i = 0, work_done = 0;
3418
3419         /* Must NOT use netdev_priv macro here. */
3420         adapter = poll_dev->priv;
3421
3422         /* Keep link state information with original netdev */
3423         if (!netif_carrier_ok(adapter->netdev))
3424                 goto quit_polling;
3425
3426         while (poll_dev != &adapter->polling_netdev[i]) {
3427                 i++;
3428                 if (unlikely(i == adapter->num_rx_queues))
3429                         BUG();
3430         }
3431
3432         if (likely(adapter->num_tx_queues == 1)) {
3433                 /* e1000_clean is called per-cpu.  This lock protects
3434                  * tx_ring[0] from being cleaned by multiple cpus
3435                  * simultaneously.  A failure obtaining the lock means
3436                  * tx_ring[0] is currently being cleaned anyway. */
3437                 if (spin_trylock(&adapter->tx_queue_lock)) {
3438                         tx_cleaned = e1000_clean_tx_irq(adapter,
3439                                                         &adapter->tx_ring[0]);
3440                         spin_unlock(&adapter->tx_queue_lock);
3441                 }
3442         } else
3443                 tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[i]);
3444
3445         adapter->clean_rx(adapter, &adapter->rx_ring[i],
3446                           &work_done, work_to_do);
3447
3448         *budget -= work_done;
3449         poll_dev->quota -= work_done;
3450
3451         /* If no Tx and not enough Rx work done, exit the polling mode */
3452         if ((!tx_cleaned && (work_done == 0)) ||
3453            !netif_running(adapter->netdev)) {
3454 quit_polling:
3455                 netif_rx_complete(poll_dev);
3456                 e1000_irq_enable(adapter);
3457                 return 0;
3458         }
3459
3460         return 1;
3461 }
3462
3463 #endif
3464 /**
3465  * e1000_clean_tx_irq - Reclaim resources after transmit completes
3466  * @adapter: board private structure
3467  **/
3468
3469 static boolean_t
3470 e1000_clean_tx_irq(struct e1000_adapter *adapter,
3471                    struct e1000_tx_ring *tx_ring)
3472 {
3473         struct net_device *netdev = adapter->netdev;
3474         struct e1000_tx_desc *tx_desc, *eop_desc;
3475         struct e1000_buffer *buffer_info;
3476         unsigned int i, eop;
3477         boolean_t cleaned = FALSE;
3478
3479         i = tx_ring->next_to_clean;
3480         eop = tx_ring->buffer_info[i].next_to_watch;
3481         eop_desc = E1000_TX_DESC(*tx_ring, eop);
3482
3483         while (eop_desc->upper.data & cpu_to_le32(E1000_TXD_STAT_DD)) {
3484                 for (cleaned = FALSE; !cleaned; ) {
3485                         tx_desc = E1000_TX_DESC(*tx_ring, i);
3486                         buffer_info = &tx_ring->buffer_info[i];
3487                         cleaned = (i == eop);
3488
3489 #ifdef CONFIG_E1000_MQ
3490                         tx_ring->tx_stats.bytes += buffer_info->length;
3491 #endif
3492                         e1000_unmap_and_free_tx_resource(adapter, buffer_info);
3493                         memset(tx_desc, 0, sizeof(struct e1000_tx_desc));
3494
3495                         if (unlikely(++i == tx_ring->count)) i = 0;
3496                 }
3497
3498 #ifdef CONFIG_E1000_MQ
3499                 tx_ring->tx_stats.packets++;
3500 #endif
3501
3502                 eop = tx_ring->buffer_info[i].next_to_watch;
3503                 eop_desc = E1000_TX_DESC(*tx_ring, eop);
3504         }
3505
3506         tx_ring->next_to_clean = i;
3507
3508         spin_lock(&tx_ring->tx_lock);
3509
3510         if (unlikely(cleaned && netif_queue_stopped(netdev) &&
3511                     netif_carrier_ok(netdev)))
3512                 netif_wake_queue(netdev);
3513
3514         spin_unlock(&tx_ring->tx_lock);
3515
3516         if (adapter->detect_tx_hung) {
3517                 /* Detect a transmit hang in hardware, this serializes the
3518                  * check with the clearing of time_stamp and movement of i */
3519                 adapter->detect_tx_hung = FALSE;
3520                 if (tx_ring->buffer_info[eop].dma &&
3521                     time_after(jiffies, tx_ring->buffer_info[eop].time_stamp +
3522                                adapter->tx_timeout_factor * HZ)
3523                     && !(E1000_READ_REG(&adapter->hw, STATUS) &
3524                          E1000_STATUS_TXOFF)) {
3525
3526                         /* detected Tx unit hang */
3527                         DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
3528                                         "  Tx Queue             <%lu>\n"
3529                                         "  TDH                  <%x>\n"
3530                                         "  TDT                  <%x>\n"
3531                                         "  next_to_use          <%x>\n"
3532                                         "  next_to_clean        <%x>\n"
3533                                         "buffer_info[next_to_clean]\n"
3534                                         "  time_stamp           <%lx>\n"
3535                                         "  next_to_watch        <%x>\n"
3536                                         "  jiffies              <%lx>\n"
3537                                         "  next_to_watch.status <%x>\n",
3538                                 (unsigned long)((tx_ring - adapter->tx_ring) /
3539                                         sizeof(struct e1000_tx_ring)),
3540                                 readl(adapter->hw.hw_addr + tx_ring->tdh),
3541                                 readl(adapter->hw.hw_addr + tx_ring->tdt),
3542                                 tx_ring->next_to_use,
3543                                 tx_ring->next_to_clean,
3544                                 tx_ring->buffer_info[eop].time_stamp,
3545                                 eop,
3546                                 jiffies,
3547                                 eop_desc->upper.fields.status);
3548                         netif_stop_queue(netdev);
3549                 }
3550         }
3551         return cleaned;
3552 }
3553
3554 /**
3555  * e1000_rx_checksum - Receive Checksum Offload for 82543
3556  * @adapter:     board private structure
3557  * @status_err:  receive descriptor status and error fields
3558  * @csum:        receive descriptor csum field
3559  * @sk_buff:     socket buffer with received data
3560  **/
3561
3562 static inline void
3563 e1000_rx_checksum(struct e1000_adapter *adapter,
3564                   uint32_t status_err, uint32_t csum,
3565                   struct sk_buff *skb)
3566 {
3567         uint16_t status = (uint16_t)status_err;
3568         uint8_t errors = (uint8_t)(status_err >> 24);
3569         skb->ip_summed = CHECKSUM_NONE;
3570
3571         /* 82543 or newer only */
3572         if (unlikely(adapter->hw.mac_type < e1000_82543)) return;
3573         /* Ignore Checksum bit is set */
3574         if (unlikely(status & E1000_RXD_STAT_IXSM)) return;
3575         /* TCP/UDP checksum error bit is set */
3576         if (unlikely(errors & E1000_RXD_ERR_TCPE)) {
3577                 /* let the stack verify checksum errors */
3578                 adapter->hw_csum_err++;
3579                 return;
3580         }
3581         /* TCP/UDP Checksum has not been calculated */
3582         if (adapter->hw.mac_type <= e1000_82547_rev_2) {
3583                 if (!(status & E1000_RXD_STAT_TCPCS))
3584                         return;
3585         } else {
3586                 if (!(status & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS)))
3587                         return;
3588         }
3589         /* It must be a TCP or UDP packet with a valid checksum */
3590         if (likely(status & E1000_RXD_STAT_TCPCS)) {
3591                 /* TCP checksum is good */
3592                 skb->ip_summed = CHECKSUM_UNNECESSARY;
3593         } else if (adapter->hw.mac_type > e1000_82547_rev_2) {
3594                 /* IP fragment with UDP payload */
3595                 /* Hardware complements the payload checksum, so we undo it
3596                  * and then put the value in host order for further stack use.
3597                  */
3598                 csum = ntohl(csum ^ 0xFFFF);
3599                 skb->csum = csum;
3600                 skb->ip_summed = CHECKSUM_HW;
3601         }
3602         adapter->hw_csum_good++;
3603 }
3604
3605 /**
3606  * e1000_clean_rx_irq - Send received data up the network stack; legacy
3607  * @adapter: board private structure
3608  **/
3609
3610 static boolean_t
3611 #ifdef CONFIG_E1000_NAPI
3612 e1000_clean_rx_irq(struct e1000_adapter *adapter,
3613                    struct e1000_rx_ring *rx_ring,
3614                    int *work_done, int work_to_do)
3615 #else
3616 e1000_clean_rx_irq(struct e1000_adapter *adapter,
3617                    struct e1000_rx_ring *rx_ring)
3618 #endif
3619 {
3620         struct net_device *netdev = adapter->netdev;
3621         struct pci_dev *pdev = adapter->pdev;
3622         struct e1000_rx_desc *rx_desc, *next_rxd;
3623         struct e1000_buffer *buffer_info, *next_buffer;
3624         unsigned long flags;
3625         uint32_t length;
3626         uint8_t last_byte;
3627         unsigned int i;
3628         int cleaned_count = 0;
3629         boolean_t cleaned = FALSE;
3630
3631         i = rx_ring->next_to_clean;
3632         rx_desc = E1000_RX_DESC(*rx_ring, i);
3633         buffer_info = &rx_ring->buffer_info[i];
3634
3635         while (rx_desc->status & E1000_RXD_STAT_DD) {
3636                 struct sk_buff *skb, *next_skb;
3637                 u8 status;
3638 #ifdef CONFIG_E1000_NAPI
3639                 if (*work_done >= work_to_do)
3640                         break;
3641                 (*work_done)++;
3642 #endif
3643                 status = rx_desc->status;
3644                 skb = buffer_info->skb;
3645                 buffer_info->skb = NULL;
3646
3647                 if (++i == rx_ring->count) i = 0;
3648                 next_rxd = E1000_RX_DESC(*rx_ring, i);
3649                 next_buffer = &rx_ring->buffer_info[i];
3650                 next_skb = next_buffer->skb;
3651
3652                 cleaned = TRUE;
3653                 cleaned_count++;
3654                 pci_unmap_single(pdev,
3655                                  buffer_info->dma,
3656                                  buffer_info->length,
3657                                  PCI_DMA_FROMDEVICE);
3658
3659                 length = le16_to_cpu(rx_desc->length);
3660
3661                 if (unlikely(!(status & E1000_RXD_STAT_EOP))) {
3662                         /* All receives must fit into a single buffer */
3663                         E1000_DBG("%s: Receive packet consumed multiple"
3664                                   " buffers\n", netdev->name);
3665                         dev_kfree_skb_irq(skb);
3666                         goto next_desc;
3667                 }
3668
3669                 if (unlikely(rx_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK)) {
3670                         last_byte = *(skb->data + length - 1);
3671                         if (TBI_ACCEPT(&adapter->hw, status,
3672                                       rx_desc->errors, length, last_byte)) {
3673                                 spin_lock_irqsave(&adapter->stats_lock, flags);
3674                                 e1000_tbi_adjust_stats(&adapter->hw,
3675                                                        &adapter->stats,
3676                                                        length, skb->data);
3677                                 spin_unlock_irqrestore(&adapter->stats_lock,
3678                                                        flags);
3679                                 length--;
3680                         } else {
3681                                 dev_kfree_skb_irq(skb);
3682                                 goto next_desc;
3683                         }
3684                 }
3685
3686                 /* code added for copybreak, this should improve
3687                  * performance for small packets with large amounts
3688                  * of reassembly being done in the stack */
3689 #define E1000_CB_LENGTH 256
3690                 if (length < E1000_CB_LENGTH) {
3691                         struct sk_buff *new_skb =
3692                             dev_alloc_skb(length + NET_IP_ALIGN);
3693                         if (new_skb) {
3694                                 skb_reserve(new_skb, NET_IP_ALIGN);
3695                                 new_skb->dev = netdev;
3696                                 memcpy(new_skb->data - NET_IP_ALIGN,
3697                                        skb->data - NET_IP_ALIGN,
3698                                        length + NET_IP_ALIGN);
3699                                 /* save the skb in buffer_info as good */
3700                                 buffer_info->skb = skb;
3701                                 skb = new_skb;
3702                                 skb_put(skb, length);
3703                         }
3704                 } else
3705                         skb_put(skb, length);
3706
3707                 /* end copybreak code */
3708
3709                 /* Receive Checksum Offload */
3710                 e1000_rx_checksum(adapter,
3711                                   (uint32_t)(status) |
3712                                   ((uint32_t)(rx_desc->errors) << 24),
3713                                   le16_to_cpu(rx_desc->csum), skb);
3714
3715                 skb->protocol = eth_type_trans(skb, netdev);
3716 #ifdef CONFIG_E1000_NAPI
3717                 if (unlikely(adapter->vlgrp &&
3718                             (status & E1000_RXD_STAT_VP))) {
3719                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3720                                                  le16_to_cpu(rx_desc->special) &
3721                                                  E1000_RXD_SPC_VLAN_MASK);
3722                 } else {
3723                         netif_receive_skb(skb);
3724                 }
3725 #else /* CONFIG_E1000_NAPI */
3726                 if (unlikely(adapter->vlgrp &&
3727                             (status & E1000_RXD_STAT_VP))) {
3728                         vlan_hwaccel_rx(skb, adapter->vlgrp,
3729                                         le16_to_cpu(rx_desc->special) &
3730                                         E1000_RXD_SPC_VLAN_MASK);
3731                 } else {
3732                         netif_rx(skb);
3733                 }
3734 #endif /* CONFIG_E1000_NAPI */
3735                 netdev->last_rx = jiffies;
3736 #ifdef CONFIG_E1000_MQ
3737                 rx_ring->rx_stats.packets++;
3738                 rx_ring->rx_stats.bytes += length;
3739 #endif
3740
3741 next_desc:
3742                 rx_desc->status = 0;
3743
3744                 /* return some buffers to hardware, one at a time is too slow */
3745                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3746                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3747                         cleaned_count = 0;
3748                 }
3749
3750                 rx_desc = next_rxd;
3751                 buffer_info = next_buffer;
3752         }
3753         rx_ring->next_to_clean = i;
3754
3755         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3756         if (cleaned_count)
3757                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3758
3759         return cleaned;
3760 }
3761
3762 /**
3763  * e1000_clean_rx_irq_ps - Send received data up the network stack; packet split
3764  * @adapter: board private structure
3765  **/
3766
3767 static boolean_t
3768 #ifdef CONFIG_E1000_NAPI
3769 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
3770                       struct e1000_rx_ring *rx_ring,
3771                       int *work_done, int work_to_do)
3772 #else
3773 e1000_clean_rx_irq_ps(struct e1000_adapter *adapter,
3774                       struct e1000_rx_ring *rx_ring)
3775 #endif
3776 {
3777         union e1000_rx_desc_packet_split *rx_desc, *next_rxd;
3778         struct net_device *netdev = adapter->netdev;
3779         struct pci_dev *pdev = adapter->pdev;
3780         struct e1000_buffer *buffer_info, *next_buffer;
3781         struct e1000_ps_page *ps_page;
3782         struct e1000_ps_page_dma *ps_page_dma;
3783         struct sk_buff *skb, *next_skb;
3784         unsigned int i, j;
3785         uint32_t length, staterr;
3786         int cleaned_count = 0;
3787         boolean_t cleaned = FALSE;
3788
3789         i = rx_ring->next_to_clean;
3790         rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
3791         staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
3792         buffer_info = &rx_ring->buffer_info[i];
3793
3794         while (staterr & E1000_RXD_STAT_DD) {
3795                 ps_page = &rx_ring->ps_page[i];
3796                 ps_page_dma = &rx_ring->ps_page_dma[i];
3797 #ifdef CONFIG_E1000_NAPI
3798                 if (unlikely(*work_done >= work_to_do))
3799                         break;
3800                 (*work_done)++;
3801 #endif
3802                 skb = buffer_info->skb;
3803
3804                 if (++i == rx_ring->count) i = 0;
3805                 next_rxd = E1000_RX_DESC_PS(*rx_ring, i);
3806                 next_buffer = &rx_ring->buffer_info[i];
3807                 next_skb = next_buffer->skb;
3808
3809                 cleaned = TRUE;
3810                 cleaned_count++;
3811                 pci_unmap_single(pdev, buffer_info->dma,
3812                                  buffer_info->length,
3813                                  PCI_DMA_FROMDEVICE);
3814
3815                 if (unlikely(!(staterr & E1000_RXD_STAT_EOP))) {
3816                         E1000_DBG("%s: Packet Split buffers didn't pick up"
3817                                   " the full packet\n", netdev->name);
3818                         dev_kfree_skb_irq(skb);
3819                         goto next_desc;
3820                 }
3821
3822                 if (unlikely(staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK)) {
3823                         dev_kfree_skb_irq(skb);
3824                         goto next_desc;
3825                 }
3826
3827                 length = le16_to_cpu(rx_desc->wb.middle.length0);
3828
3829                 if (unlikely(!length)) {
3830                         E1000_DBG("%s: Last part of the packet spanning"
3831                                   " multiple descriptors\n", netdev->name);
3832                         dev_kfree_skb_irq(skb);
3833                         goto next_desc;
3834                 }
3835
3836                 /* Good Receive */
3837                 skb_put(skb, length);
3838
3839                 for (j = 0; j < adapter->rx_ps_pages; j++) {
3840                         if (!(length = le16_to_cpu(rx_desc->wb.upper.length[j])))
3841                                 break;
3842
3843                         pci_unmap_page(pdev, ps_page_dma->ps_page_dma[j],
3844                                         PAGE_SIZE, PCI_DMA_FROMDEVICE);
3845                         ps_page_dma->ps_page_dma[j] = 0;
3846                         skb_shinfo(skb)->frags[j].page =
3847                                 ps_page->ps_page[j];
3848                         ps_page->ps_page[j] = NULL;
3849                         skb_shinfo(skb)->frags[j].page_offset = 0;
3850                         skb_shinfo(skb)->frags[j].size = length;
3851                         skb_shinfo(skb)->nr_frags++;
3852                         skb->len += length;
3853                         skb->data_len += length;
3854                 }
3855
3856                 e1000_rx_checksum(adapter, staterr,
3857                                   le16_to_cpu(rx_desc->wb.lower.hi_dword.csum_ip.csum), skb);
3858                 skb->protocol = eth_type_trans(skb, netdev);
3859
3860                 if (likely(rx_desc->wb.upper.header_status &
3861                            cpu_to_le16(E1000_RXDPS_HDRSTAT_HDRSP)))
3862                         adapter->rx_hdr_split++;
3863 #ifdef CONFIG_E1000_NAPI
3864                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
3865                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
3866                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
3867                                 E1000_RXD_SPC_VLAN_MASK);
3868                 } else {
3869                         netif_receive_skb(skb);
3870                 }
3871 #else /* CONFIG_E1000_NAPI */
3872                 if (unlikely(adapter->vlgrp && (staterr & E1000_RXD_STAT_VP))) {
3873                         vlan_hwaccel_rx(skb, adapter->vlgrp,
3874                                 le16_to_cpu(rx_desc->wb.middle.vlan) &
3875                                 E1000_RXD_SPC_VLAN_MASK);
3876                 } else {
3877                         netif_rx(skb);
3878                 }
3879 #endif /* CONFIG_E1000_NAPI */
3880                 netdev->last_rx = jiffies;
3881 #ifdef CONFIG_E1000_MQ
3882                 rx_ring->rx_stats.packets++;
3883                 rx_ring->rx_stats.bytes += length;
3884 #endif
3885
3886 next_desc:
3887                 rx_desc->wb.middle.status_error &= cpu_to_le32(~0xFF);
3888                 buffer_info->skb = NULL;
3889
3890                 /* return some buffers to hardware, one at a time is too slow */
3891                 if (unlikely(cleaned_count >= E1000_RX_BUFFER_WRITE)) {
3892                         adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3893                         cleaned_count = 0;
3894                 }
3895
3896                 rx_desc = next_rxd;
3897                 buffer_info = next_buffer;
3898
3899                 staterr = le32_to_cpu(rx_desc->wb.middle.status_error);
3900         }
3901         rx_ring->next_to_clean = i;
3902
3903         cleaned_count = E1000_DESC_UNUSED(rx_ring);
3904         if (cleaned_count)
3905                 adapter->alloc_rx_buf(adapter, rx_ring, cleaned_count);
3906
3907         return cleaned;
3908 }
3909
3910 /**
3911  * e1000_alloc_rx_buffers - Replace used receive buffers; legacy & extended
3912  * @adapter: address of board private structure
3913  **/
3914
3915 static void
3916 e1000_alloc_rx_buffers(struct e1000_adapter *adapter,
3917                        struct e1000_rx_ring *rx_ring,
3918                        int cleaned_count)
3919 {
3920         struct net_device *netdev = adapter->netdev;
3921         struct pci_dev *pdev = adapter->pdev;
3922         struct e1000_rx_desc *rx_desc;
3923         struct e1000_buffer *buffer_info;
3924         struct sk_buff *skb;
3925         unsigned int i;
3926         unsigned int bufsz = adapter->rx_buffer_len + NET_IP_ALIGN;
3927
3928         i = rx_ring->next_to_use;
3929         buffer_info = &rx_ring->buffer_info[i];
3930
3931         while (cleaned_count--) {
3932                 if (!(skb = buffer_info->skb))
3933                         skb = dev_alloc_skb(bufsz);
3934                 else {
3935                         skb_trim(skb, 0);
3936                         goto map_skb;
3937                 }
3938
3939
3940                 if (unlikely(!skb)) {
3941                         /* Better luck next round */
3942                         adapter->alloc_rx_buff_failed++;
3943                         break;
3944                 }
3945
3946                 /* Fix for errata 23, can't cross 64kB boundary */
3947                 if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3948                         struct sk_buff *oldskb = skb;
3949                         DPRINTK(RX_ERR, ERR, "skb align check failed: %u bytes "
3950                                              "at %p\n", bufsz, skb->data);
3951                         /* Try again, without freeing the previous */
3952                         skb = dev_alloc_skb(bufsz);
3953                         /* Failed allocation, critical failure */
3954                         if (!skb) {
3955                                 dev_kfree_skb(oldskb);
3956                                 break;
3957                         }
3958
3959                         if (!e1000_check_64k_bound(adapter, skb->data, bufsz)) {
3960                                 /* give up */
3961                                 dev_kfree_skb(skb);
3962                                 dev_kfree_skb(oldskb);
3963                                 break; /* while !buffer_info->skb */
3964                         } else {
3965                                 /* Use new allocation */
3966                                 dev_kfree_skb(oldskb);
3967                         }
3968                 }
3969                 /* Make buffer alignment 2 beyond a 16 byte boundary
3970                  * this will result in a 16 byte aligned IP header after
3971                  * the 14 byte MAC header is removed
3972                  */
3973                 skb_reserve(skb, NET_IP_ALIGN);
3974
3975                 skb->dev = netdev;
3976
3977                 buffer_info->skb = skb;
3978                 buffer_info->length = adapter->rx_buffer_len;
3979 map_skb:
3980                 buffer_info->dma = pci_map_single(pdev,
3981                                                   skb->data,
3982                                                   adapter->rx_buffer_len,
3983                                                   PCI_DMA_FROMDEVICE);
3984
3985                 /* Fix for errata 23, can't cross 64kB boundary */
3986                 if (!e1000_check_64k_bound(adapter,
3987                                         (void *)(unsigned long)buffer_info->dma,
3988                                         adapter->rx_buffer_len)) {
3989                         DPRINTK(RX_ERR, ERR,
3990                                 "dma align check failed: %u bytes at %p\n",
3991                                 adapter->rx_buffer_len,
3992                                 (void *)(unsigned long)buffer_info->dma);
3993                         dev_kfree_skb(skb);
3994                         buffer_info->skb = NULL;
3995
3996                         pci_unmap_single(pdev, buffer_info->dma,
3997                                          adapter->rx_buffer_len,
3998                                          PCI_DMA_FROMDEVICE);
3999
4000                         break; /* while !buffer_info->skb */
4001                 }
4002                 rx_desc = E1000_RX_DESC(*rx_ring, i);
4003                 rx_desc->buffer_addr = cpu_to_le64(buffer_info->dma);
4004
4005                 if (unlikely(++i == rx_ring->count))
4006                         i = 0;
4007                 buffer_info = &rx_ring->buffer_info[i];
4008         }
4009
4010         if (likely(rx_ring->next_to_use != i)) {
4011                 rx_ring->next_to_use = i;
4012                 if (unlikely(i-- == 0))
4013                         i = (rx_ring->count - 1);
4014
4015                 /* Force memory writes to complete before letting h/w
4016                  * know there are new descriptors to fetch.  (Only
4017                  * applicable for weak-ordered memory model archs,
4018                  * such as IA-64). */
4019                 wmb();
4020                 writel(i, adapter->hw.hw_addr + rx_ring->rdt);
4021         }
4022 }
4023
4024 /**
4025  * e1000_alloc_rx_buffers_ps - Replace used receive buffers; packet split
4026  * @adapter: address of board private structure
4027  **/
4028
4029 static void
4030 e1000_alloc_rx_buffers_ps(struct e1000_adapter *adapter,
4031                           struct e1000_rx_ring *rx_ring,
4032                           int cleaned_count)
4033 {
4034         struct net_device *netdev = adapter->netdev;
4035         struct pci_dev *pdev = adapter->pdev;
4036         union e1000_rx_desc_packet_split *rx_desc;
4037         struct e1000_buffer *buffer_info;
4038         struct e1000_ps_page *ps_page;
4039         struct e1000_ps_page_dma *ps_page_dma;
4040         struct sk_buff *skb;
4041         unsigned int i, j;
4042
4043         i = rx_ring->next_to_use;
4044         buffer_info = &rx_ring->buffer_info[i];
4045         ps_page = &rx_ring->ps_page[i];
4046         ps_page_dma = &rx_ring->ps_page_dma[i];
4047
4048         while (cleaned_count--) {
4049                 rx_desc = E1000_RX_DESC_PS(*rx_ring, i);
4050
4051                 for (j = 0; j < PS_PAGE_BUFFERS; j++) {
4052                         if (j < adapter->rx_ps_pages) {
4053                                 if (likely(!ps_page->ps_page[j])) {
4054                                         ps_page->ps_page[j] =
4055                                                 alloc_page(GFP_ATOMIC);
4056                                         if (unlikely(!ps_page->ps_page[j])) {
4057                                                 adapter->alloc_rx_buff_failed++;
4058                                                 goto no_buffers;
4059                                         }
4060                                         ps_page_dma->ps_page_dma[j] =
4061                                                 pci_map_page(pdev,
4062                                                             ps_page->ps_page[j],
4063                                                             0, PAGE_SIZE,
4064                                                             PCI_DMA_FROMDEVICE);
4065                                 }
4066                                 /* Refresh the desc even if buffer_addrs didn't
4067                                  * change because each write-back erases
4068                                  * this info.
4069                                  */
4070                                 rx_desc->read.buffer_addr[j+1] =
4071                                      cpu_to_le64(ps_page_dma->ps_page_dma[j]);
4072                         } else
4073                                 rx_desc->read.buffer_addr[j+1] = ~0;
4074                 }
4075
4076                 skb = dev_alloc_skb(adapter->rx_ps_bsize0 + NET_IP_ALIGN);
4077
4078                 if (unlikely(!skb)) {
4079                         adapter->alloc_rx_buff_failed++;
4080                         break;
4081                 }
4082
4083                 /* Make buffer alignment 2 beyond a 16 byte boundary
4084                  * this will result in a 16 byte aligned IP header after
4085                  * the 14 byte MAC header is removed
4086                  */
4087                 skb_reserve(skb, NET_IP_ALIGN);
4088
4089                 skb->dev = netdev;
4090
4091                 buffer_info->skb = skb;
4092                 buffer_info->length = adapter->rx_ps_bsize0;
4093                 buffer_info->dma = pci_map_single(pdev, skb->data,
4094                                                   adapter->rx_ps_bsize0,
4095                                                   PCI_DMA_FROMDEVICE);
4096
4097                 rx_desc->read.buffer_addr[0] = cpu_to_le64(buffer_info->dma);
4098
4099                 if (unlikely(++i == rx_ring->count)) i = 0;
4100                 buffer_info = &rx_ring->buffer_info[i];
4101                 ps_page = &rx_ring->ps_page[i];
4102                 ps_page_dma = &rx_ring->ps_page_dma[i];
4103         }
4104
4105 no_buffers:
4106         if (likely(rx_ring->next_to_use != i)) {
4107                 rx_ring->next_to_use = i;
4108                 if (unlikely(i-- == 0)) i = (rx_ring->count - 1);
4109
4110                 /* Force memory writes to complete before letting h/w
4111                  * know there are new descriptors to fetch.  (Only
4112                  * applicable for weak-ordered memory model archs,
4113                  * such as IA-64). */
4114                 wmb();
4115                 /* Hardware increments by 16 bytes, but packet split
4116                  * descriptors are 32 bytes...so we increment tail
4117                  * twice as much.
4118                  */
4119                 writel(i<<1, adapter->hw.hw_addr + rx_ring->rdt);
4120         }
4121 }
4122
4123 /**
4124  * e1000_smartspeed - Workaround for SmartSpeed on 82541 and 82547 controllers.
4125  * @adapter:
4126  **/
4127
4128 static void
4129 e1000_smartspeed(struct e1000_adapter *adapter)
4130 {
4131         uint16_t phy_status;
4132         uint16_t phy_ctrl;
4133
4134         if ((adapter->hw.phy_type != e1000_phy_igp) || !adapter->hw.autoneg ||
4135            !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
4136                 return;
4137
4138         if (adapter->smartspeed == 0) {
4139                 /* If Master/Slave config fault is asserted twice,
4140                  * we assume back-to-back */
4141                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4142                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4143                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_status);
4144                 if (!(phy_status & SR_1000T_MS_CONFIG_FAULT)) return;
4145                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4146                 if (phy_ctrl & CR_1000T_MS_ENABLE) {
4147                         phy_ctrl &= ~CR_1000T_MS_ENABLE;
4148                         e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL,
4149                                             phy_ctrl);
4150                         adapter->smartspeed++;
4151                         if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4152                            !e1000_read_phy_reg(&adapter->hw, PHY_CTRL,
4153                                                &phy_ctrl)) {
4154                                 phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4155                                              MII_CR_RESTART_AUTO_NEG);
4156                                 e1000_write_phy_reg(&adapter->hw, PHY_CTRL,
4157                                                     phy_ctrl);
4158                         }
4159                 }
4160                 return;
4161         } else if (adapter->smartspeed == E1000_SMARTSPEED_DOWNSHIFT) {
4162                 /* If still no link, perhaps using 2/3 pair cable */
4163                 e1000_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_ctrl);
4164                 phy_ctrl |= CR_1000T_MS_ENABLE;
4165                 e1000_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_ctrl);
4166                 if (!e1000_phy_setup_autoneg(&adapter->hw) &&
4167                    !e1000_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_ctrl)) {
4168                         phy_ctrl |= (MII_CR_AUTO_NEG_EN |
4169                                      MII_CR_RESTART_AUTO_NEG);
4170                         e1000_write_phy_reg(&adapter->hw, PHY_CTRL, phy_ctrl);
4171                 }
4172         }
4173         /* Restart process after E1000_SMARTSPEED_MAX iterations */
4174         if (adapter->smartspeed++ == E1000_SMARTSPEED_MAX)
4175                 adapter->smartspeed = 0;
4176 }
4177
4178 /**
4179  * e1000_ioctl -
4180  * @netdev:
4181  * @ifreq:
4182  * @cmd:
4183  **/
4184
4185 static int
4186 e1000_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4187 {
4188         switch (cmd) {
4189         case SIOCGMIIPHY:
4190         case SIOCGMIIREG:
4191         case SIOCSMIIREG:
4192                 return e1000_mii_ioctl(netdev, ifr, cmd);
4193         default:
4194                 return -EOPNOTSUPP;
4195         }
4196 }
4197
4198 /**
4199  * e1000_mii_ioctl -
4200  * @netdev:
4201  * @ifreq:
4202  * @cmd:
4203  **/
4204
4205 static int
4206 e1000_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4207 {
4208         struct e1000_adapter *adapter = netdev_priv(netdev);
4209         struct mii_ioctl_data *data = if_mii(ifr);
4210         int retval;
4211         uint16_t mii_reg;
4212         uint16_t spddplx;
4213         unsigned long flags;
4214
4215         if (adapter->hw.media_type != e1000_media_type_copper)
4216                 return -EOPNOTSUPP;
4217
4218         switch (cmd) {
4219         case SIOCGMIIPHY:
4220                 data->phy_id = adapter->hw.phy_addr;
4221                 break;
4222         case SIOCGMIIREG:
4223                 if (!capable(CAP_NET_ADMIN))
4224                         return -EPERM;
4225                 spin_lock_irqsave(&adapter->stats_lock, flags);
4226                 if (e1000_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4227                                    &data->val_out)) {
4228                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4229                         return -EIO;
4230                 }
4231                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4232                 break;
4233         case SIOCSMIIREG:
4234                 if (!capable(CAP_NET_ADMIN))
4235                         return -EPERM;
4236                 if (data->reg_num & ~(0x1F))
4237                         return -EFAULT;
4238                 mii_reg = data->val_in;
4239                 spin_lock_irqsave(&adapter->stats_lock, flags);
4240                 if (e1000_write_phy_reg(&adapter->hw, data->reg_num,
4241                                         mii_reg)) {
4242                         spin_unlock_irqrestore(&adapter->stats_lock, flags);
4243                         return -EIO;
4244                 }
4245                 if (adapter->hw.phy_type == e1000_phy_m88) {
4246                         switch (data->reg_num) {
4247                         case PHY_CTRL:
4248                                 if (mii_reg & MII_CR_POWER_DOWN)
4249                                         break;
4250                                 if (mii_reg & MII_CR_AUTO_NEG_EN) {
4251                                         adapter->hw.autoneg = 1;
4252                                         adapter->hw.autoneg_advertised = 0x2F;
4253                                 } else {
4254                                         if (mii_reg & 0x40)
4255                                                 spddplx = SPEED_1000;
4256                                         else if (mii_reg & 0x2000)
4257                                                 spddplx = SPEED_100;
4258                                         else
4259                                                 spddplx = SPEED_10;
4260                                         spddplx += (mii_reg & 0x100)
4261                                                    ? FULL_DUPLEX :
4262                                                    HALF_DUPLEX;
4263                                         retval = e1000_set_spd_dplx(adapter,
4264                                                                     spddplx);
4265                                         if (retval) {
4266                                                 spin_unlock_irqrestore(
4267                                                         &adapter->stats_lock,
4268                                                         flags);
4269                                                 return retval;
4270                                         }
4271                                 }
4272                                 if (netif_running(adapter->netdev)) {
4273                                         e1000_down(adapter);
4274                                         e1000_up(adapter);
4275                                 } else
4276                                         e1000_reset(adapter);
4277                                 break;
4278                         case M88E1000_PHY_SPEC_CTRL:
4279                         case M88E1000_EXT_PHY_SPEC_CTRL:
4280                                 if (e1000_phy_reset(&adapter->hw)) {
4281                                         spin_unlock_irqrestore(
4282                                                 &adapter->stats_lock, flags);
4283                                         return -EIO;
4284                                 }
4285                                 break;
4286                         }
4287                 } else {
4288                         switch (data->reg_num) {
4289                         case PHY_CTRL:
4290                                 if (mii_reg & MII_CR_POWER_DOWN)
4291                                         break;
4292                                 if (netif_running(adapter->netdev)) {
4293                                         e1000_down(adapter);
4294                                         e1000_up(adapter);
4295                                 } else
4296                                         e1000_reset(adapter);
4297                                 break;
4298                         }
4299                 }
4300                 spin_unlock_irqrestore(&adapter->stats_lock, flags);
4301                 break;
4302         default:
4303                 return -EOPNOTSUPP;
4304         }
4305         return E1000_SUCCESS;
4306 }
4307
4308 void
4309 e1000_pci_set_mwi(struct e1000_hw *hw)
4310 {
4311         struct e1000_adapter *adapter = hw->back;
4312         int ret_val = pci_set_mwi(adapter->pdev);
4313
4314         if (ret_val)
4315                 DPRINTK(PROBE, ERR, "Error in setting MWI\n");
4316 }
4317
4318 void
4319 e1000_pci_clear_mwi(struct e1000_hw *hw)
4320 {
4321         struct e1000_adapter *adapter = hw->back;
4322
4323         pci_clear_mwi(adapter->pdev);
4324 }
4325
4326 void
4327 e1000_read_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4328 {
4329         struct e1000_adapter *adapter = hw->back;
4330
4331         pci_read_config_word(adapter->pdev, reg, value);
4332 }
4333
4334 void
4335 e1000_write_pci_cfg(struct e1000_hw *hw, uint32_t reg, uint16_t *value)
4336 {
4337         struct e1000_adapter *adapter = hw->back;
4338
4339         pci_write_config_word(adapter->pdev, reg, *value);
4340 }
4341
4342 uint32_t
4343 e1000_io_read(struct e1000_hw *hw, unsigned long port)
4344 {
4345         return inl(port);
4346 }
4347
4348 void
4349 e1000_io_write(struct e1000_hw *hw, unsigned long port, uint32_t value)
4350 {
4351         outl(value, port);
4352 }
4353
4354 static void
4355 e1000_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
4356 {
4357         struct e1000_adapter *adapter = netdev_priv(netdev);
4358         uint32_t ctrl, rctl;
4359
4360         e1000_irq_disable(adapter);
4361         adapter->vlgrp = grp;
4362
4363         if (grp) {
4364                 /* enable VLAN tag insert/strip */
4365                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4366                 ctrl |= E1000_CTRL_VME;
4367                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4368
4369                 /* enable VLAN receive filtering */
4370                 rctl = E1000_READ_REG(&adapter->hw, RCTL);
4371                 rctl |= E1000_RCTL_VFE;
4372                 rctl &= ~E1000_RCTL_CFIEN;
4373                 E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4374                 e1000_update_mng_vlan(adapter);
4375         } else {
4376                 /* disable VLAN tag insert/strip */
4377                 ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4378                 ctrl &= ~E1000_CTRL_VME;
4379                 E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4380
4381                 /* disable VLAN filtering */
4382                 rctl = E1000_READ_REG(&adapter->hw, RCTL);
4383                 rctl &= ~E1000_RCTL_VFE;
4384                 E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4385                 if (adapter->mng_vlan_id != (uint16_t)E1000_MNG_VLAN_NONE) {
4386                         e1000_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4387                         adapter->mng_vlan_id = E1000_MNG_VLAN_NONE;
4388                 }
4389         }
4390
4391         e1000_irq_enable(adapter);
4392 }
4393
4394 static void
4395 e1000_vlan_rx_add_vid(struct net_device *netdev, uint16_t vid)
4396 {
4397         struct e1000_adapter *adapter = netdev_priv(netdev);
4398         uint32_t vfta, index;
4399
4400         if ((adapter->hw.mng_cookie.status &
4401              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4402             (vid == adapter->mng_vlan_id))
4403                 return;
4404         /* add VID to filter table */
4405         index = (vid >> 5) & 0x7F;
4406         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4407         vfta |= (1 << (vid & 0x1F));
4408         e1000_write_vfta(&adapter->hw, index, vfta);
4409 }
4410
4411 static void
4412 e1000_vlan_rx_kill_vid(struct net_device *netdev, uint16_t vid)
4413 {
4414         struct e1000_adapter *adapter = netdev_priv(netdev);
4415         uint32_t vfta, index;
4416
4417         e1000_irq_disable(adapter);
4418
4419         if (adapter->vlgrp)
4420                 adapter->vlgrp->vlan_devices[vid] = NULL;
4421
4422         e1000_irq_enable(adapter);
4423
4424         if ((adapter->hw.mng_cookie.status &
4425              E1000_MNG_DHCP_COOKIE_STATUS_VLAN_SUPPORT) &&
4426             (vid == adapter->mng_vlan_id)) {
4427                 /* release control to f/w */
4428                 e1000_release_hw_control(adapter);
4429                 return;
4430         }
4431
4432         /* remove VID from filter table */
4433         index = (vid >> 5) & 0x7F;
4434         vfta = E1000_READ_REG_ARRAY(&adapter->hw, VFTA, index);
4435         vfta &= ~(1 << (vid & 0x1F));
4436         e1000_write_vfta(&adapter->hw, index, vfta);
4437 }
4438
4439 static void
4440 e1000_restore_vlan(struct e1000_adapter *adapter)
4441 {
4442         e1000_vlan_rx_register(adapter->netdev, adapter->vlgrp);
4443
4444         if (adapter->vlgrp) {
4445                 uint16_t vid;
4446                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
4447                         if (!adapter->vlgrp->vlan_devices[vid])
4448                                 continue;
4449                         e1000_vlan_rx_add_vid(adapter->netdev, vid);
4450                 }
4451         }
4452 }
4453
4454 int
4455 e1000_set_spd_dplx(struct e1000_adapter *adapter, uint16_t spddplx)
4456 {
4457         adapter->hw.autoneg = 0;
4458
4459         /* Fiber NICs only allow 1000 gbps Full duplex */
4460         if ((adapter->hw.media_type == e1000_media_type_fiber) &&
4461                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
4462                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4463                 return -EINVAL;
4464         }
4465
4466         switch (spddplx) {
4467         case SPEED_10 + DUPLEX_HALF:
4468                 adapter->hw.forced_speed_duplex = e1000_10_half;
4469                 break;
4470         case SPEED_10 + DUPLEX_FULL:
4471                 adapter->hw.forced_speed_duplex = e1000_10_full;
4472                 break;
4473         case SPEED_100 + DUPLEX_HALF:
4474                 adapter->hw.forced_speed_duplex = e1000_100_half;
4475                 break;
4476         case SPEED_100 + DUPLEX_FULL:
4477                 adapter->hw.forced_speed_duplex = e1000_100_full;
4478                 break;
4479         case SPEED_1000 + DUPLEX_FULL:
4480                 adapter->hw.autoneg = 1;
4481                 adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
4482                 break;
4483         case SPEED_1000 + DUPLEX_HALF: /* not supported */
4484         default:
4485                 DPRINTK(PROBE, ERR, "Unsupported Speed/Duplex configuration\n");
4486                 return -EINVAL;
4487         }
4488         return 0;
4489 }
4490
4491 #ifdef CONFIG_PM
4492 /* these functions save and restore 16 or 64 dwords (64-256 bytes) of config
4493  * space versus the 64 bytes that pci_[save|restore]_state handle
4494  */
4495 #define PCIE_CONFIG_SPACE_LEN 256
4496 #define PCI_CONFIG_SPACE_LEN 64
4497 static int
4498 e1000_pci_save_state(struct e1000_adapter *adapter)
4499 {
4500         struct pci_dev *dev = adapter->pdev;
4501         int size;
4502         int i;
4503         if (adapter->hw.mac_type >= e1000_82571)
4504                 size = PCIE_CONFIG_SPACE_LEN;
4505         else
4506                 size = PCI_CONFIG_SPACE_LEN;
4507
4508         WARN_ON(adapter->config_space != NULL);
4509
4510         adapter->config_space = kmalloc(size, GFP_KERNEL);
4511         if (!adapter->config_space) {
4512                 DPRINTK(PROBE, ERR, "unable to allocate %d bytes\n", size);
4513                 return -ENOMEM;
4514         }
4515         for (i = 0; i < (size / 4); i++)
4516                 pci_read_config_dword(dev, i * 4, &adapter->config_space[i]);
4517         return 0;
4518 }
4519
4520 static void
4521 e1000_pci_restore_state(struct e1000_adapter *adapter)
4522 {
4523         struct pci_dev *dev = adapter->pdev;
4524         int size;
4525         int i;
4526         if (adapter->config_space == NULL)
4527                 return;
4528         if (adapter->hw.mac_type >= e1000_82571)
4529                 size = PCIE_CONFIG_SPACE_LEN;
4530         else
4531                 size = PCI_CONFIG_SPACE_LEN;
4532         for (i = 0; i < (size / 4); i++)
4533                 pci_write_config_dword(dev, i * 4, adapter->config_space[i]);
4534         kfree(adapter->config_space);
4535         adapter->config_space = NULL;
4536         return;
4537 }
4538 #endif /* CONFIG_PM */
4539
4540 static int
4541 e1000_suspend(struct pci_dev *pdev, pm_message_t state)
4542 {
4543         struct net_device *netdev = pci_get_drvdata(pdev);
4544         struct e1000_adapter *adapter = netdev_priv(netdev);
4545         uint32_t ctrl, ctrl_ext, rctl, manc, status;
4546         uint32_t wufc = adapter->wol;
4547         int retval = 0;
4548
4549         netif_device_detach(netdev);
4550
4551         if (netif_running(netdev))
4552                 e1000_down(adapter);
4553
4554 #ifdef CONFIG_PM
4555         /* implement our own version of pci_save_state(pdev) because pci
4556          * express adapters have larger 256 byte config spaces */
4557         retval = e1000_pci_save_state(adapter);
4558         if (retval)
4559                 return retval;
4560 #endif
4561
4562         status = E1000_READ_REG(&adapter->hw, STATUS);
4563         if (status & E1000_STATUS_LU)
4564                 wufc &= ~E1000_WUFC_LNKC;
4565
4566         if (wufc) {
4567                 e1000_setup_rctl(adapter);
4568                 e1000_set_multi(netdev);
4569
4570                 /* turn on all-multi mode if wake on multicast is enabled */
4571                 if (adapter->wol & E1000_WUFC_MC) {
4572                         rctl = E1000_READ_REG(&adapter->hw, RCTL);
4573                         rctl |= E1000_RCTL_MPE;
4574                         E1000_WRITE_REG(&adapter->hw, RCTL, rctl);
4575                 }
4576
4577                 if (adapter->hw.mac_type >= e1000_82540) {
4578                         ctrl = E1000_READ_REG(&adapter->hw, CTRL);
4579                         /* advertise wake from D3Cold */
4580                         #define E1000_CTRL_ADVD3WUC 0x00100000
4581                         /* phy power management enable */
4582                         #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
4583                         ctrl |= E1000_CTRL_ADVD3WUC |
4584                                 E1000_CTRL_EN_PHY_PWR_MGMT;
4585                         E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
4586                 }
4587
4588                 if (adapter->hw.media_type == e1000_media_type_fiber ||
4589                    adapter->hw.media_type == e1000_media_type_internal_serdes) {
4590                         /* keep the laser running in D3 */
4591                         ctrl_ext = E1000_READ_REG(&adapter->hw, CTRL_EXT);
4592                         ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
4593                         E1000_WRITE_REG(&adapter->hw, CTRL_EXT, ctrl_ext);
4594                 }
4595
4596                 /* Allow time for pending master requests to run */
4597                 e1000_disable_pciex_master(&adapter->hw);
4598
4599                 E1000_WRITE_REG(&adapter->hw, WUC, E1000_WUC_PME_EN);
4600                 E1000_WRITE_REG(&adapter->hw, WUFC, wufc);
4601                 retval = pci_enable_wake(pdev, PCI_D3hot, 1);
4602                 if (retval)
4603                         DPRINTK(PROBE, ERR, "Error enabling D3 wake\n");
4604                 retval = pci_enable_wake(pdev, PCI_D3cold, 1);
4605                 if (retval)
4606                         DPRINTK(PROBE, ERR, "Error enabling D3 cold wake\n");
4607         } else {
4608                 E1000_WRITE_REG(&adapter->hw, WUC, 0);
4609                 E1000_WRITE_REG(&adapter->hw, WUFC, 0);
4610                 retval = pci_enable_wake(pdev, PCI_D3hot, 0);
4611                 if (retval)
4612                         DPRINTK(PROBE, ERR, "Error enabling D3 wake\n");
4613                 retval = pci_enable_wake(pdev, PCI_D3cold, 0); /* 4 == D3 cold */
4614                 if (retval)
4615                         DPRINTK(PROBE, ERR, "Error enabling D3 cold wake\n");
4616         }
4617
4618         if (adapter->hw.mac_type >= e1000_82540 &&
4619            adapter->hw.media_type == e1000_media_type_copper) {
4620                 manc = E1000_READ_REG(&adapter->hw, MANC);
4621                 if (manc & E1000_MANC_SMBUS_EN) {
4622                         manc |= E1000_MANC_ARP_EN;
4623                         E1000_WRITE_REG(&adapter->hw, MANC, manc);
4624                         retval = pci_enable_wake(pdev, PCI_D3hot, 1);
4625                         if (retval)
4626                                 DPRINTK(PROBE, ERR, "Error enabling D3 wake\n");
4627                         retval = pci_enable_wake(pdev, PCI_D3cold, 1);
4628                         if (retval)
4629                                 DPRINTK(PROBE, ERR, "Error enabling D3 cold wake\n");
4630                 }
4631         }
4632
4633         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
4634          * would have already happened in close and is redundant. */
4635         e1000_release_hw_control(adapter);
4636
4637         pci_disable_device(pdev);
4638
4639         retval = pci_set_power_state(pdev, pci_choose_state(pdev, state));
4640         if (retval)
4641                 DPRINTK(PROBE, ERR, "Error in setting power state\n");
4642
4643         return 0;
4644 }
4645
4646 #ifdef CONFIG_PM
4647 static int
4648 e1000_resume(struct pci_dev *pdev)
4649 {
4650         struct net_device *netdev = pci_get_drvdata(pdev);
4651         struct e1000_adapter *adapter = netdev_priv(netdev);
4652         int retval;
4653         uint32_t manc, ret_val;
4654
4655         retval = pci_set_power_state(pdev, PCI_D0);
4656         if (retval)
4657                 DPRINTK(PROBE, ERR, "Error in setting power state\n");
4658         e1000_pci_restore_state(adapter);
4659         ret_val = pci_enable_device(pdev);
4660         pci_set_master(pdev);
4661
4662         retval = pci_enable_wake(pdev, PCI_D3hot, 0);
4663         if (retval)
4664                 DPRINTK(PROBE, ERR, "Error enabling D3 wake\n");
4665         retval = pci_enable_wake(pdev, PCI_D3cold, 0);
4666         if (retval)
4667                 DPRINTK(PROBE, ERR, "Error enabling D3 cold wake\n");
4668
4669         e1000_reset(adapter);
4670         E1000_WRITE_REG(&adapter->hw, WUS, ~0);
4671
4672         if (netif_running(netdev))
4673                 e1000_up(adapter);
4674
4675         netif_device_attach(netdev);
4676
4677         if (adapter->hw.mac_type >= e1000_82540 &&
4678            adapter->hw.media_type == e1000_media_type_copper) {
4679                 manc = E1000_READ_REG(&adapter->hw, MANC);
4680                 manc &= ~(E1000_MANC_ARP_EN);
4681                 E1000_WRITE_REG(&adapter->hw, MANC, manc);
4682         }
4683
4684         /* If the controller is 82573 and f/w is AMT, do not set
4685          * DRV_LOAD until the interface is up.  For all other cases,
4686          * let the f/w know that the h/w is now under the control
4687          * of the driver. */
4688         if (adapter->hw.mac_type != e1000_82573 ||
4689             !e1000_check_mng_mode(&adapter->hw))
4690                 e1000_get_hw_control(adapter);
4691
4692         return 0;
4693 }
4694 #endif
4695 #ifdef CONFIG_NET_POLL_CONTROLLER
4696 /*
4697  * Polling 'interrupt' - used by things like netconsole to send skbs
4698  * without having to re-enable interrupts. It's not called while
4699  * the interrupt routine is executing.
4700  */
4701 static void
4702 e1000_netpoll(struct net_device *netdev)
4703 {
4704         struct e1000_adapter *adapter = netdev_priv(netdev);
4705         disable_irq(adapter->pdev->irq);
4706         e1000_intr(adapter->pdev->irq, netdev, NULL);
4707         e1000_clean_tx_irq(adapter, adapter->tx_ring);
4708 #ifndef CONFIG_E1000_NAPI
4709         adapter->clean_rx(adapter, adapter->rx_ring);
4710 #endif
4711         enable_irq(adapter->pdev->irq);
4712 }
4713 #endif
4714
4715 /* e1000_main.c */