drivers/net/igb/igb_main.c

   1 /*******************************************************************************
   2
   3   Intel(R) Gigabit Ethernet Linux driver
   4   Copyright(c) 2007-2009 Intel Corporation.
   5
   6   This program is free software; you can redistribute it and/or modify it
   7   under the terms and conditions of the GNU General Public License,
   8   version 2, as published by the Free Software Foundation.
   9
  10   This program is distributed in the hope it will be useful, but WITHOUT
  11   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  12   FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  13   more details.
  14
  15   You should have received a copy of the GNU General Public License along with
  16   this program; if not, write to the Free Software Foundation, Inc.,
  17   51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
  18
  19   The full GNU General Public License is included in this distribution in
  20   the file called "COPYING".
  21
  22   Contact Information:
  23   e1000-devel Mailing List <e1000-devel@lists.sourceforge.net>
  24   Intel Corporation, 5200 N.E. Elam Young Parkway, Hillsboro, OR 97124-6497
  25
  26 *******************************************************************************/
  27
  28 #include <linux/module.h>
  29 #include <linux/types.h>
  30 #include <linux/init.h>
  31 #include <linux/vmalloc.h>
  32 #include <linux/pagemap.h>
  33 #include <linux/netdevice.h>
  34 #include <linux/ipv6.h>
  35 #include <net/checksum.h>
  36 #include <net/ip6_checksum.h>
  37 #include <linux/net_tstamp.h>
  38 #include <linux/mii.h>
  39 #include <linux/ethtool.h>
  40 #include <linux/if_vlan.h>
  41 #include <linux/pci.h>
  42 #include <linux/pci-aspm.h>
  43 #include <linux/delay.h>
  44 #include <linux/interrupt.h>
  45 #include <linux/if_ether.h>
  46 #include <linux/aer.h>
  47 #ifdef CONFIG_IGB_DCA
  48 #include <linux/dca.h>
  49 #endif
  50 #include "igb.h"
  51
  52 #define DRV_VERSION "1.3.16-k2"
  53 char igb_driver_name[] = "igb";
  54 char igb_driver_version[] = DRV_VERSION;
  55 static const char igb_driver_string[] =
  56                                 "Intel(R) Gigabit Ethernet Network Driver";
  57 static const char igb_copyright[] = "Copyright (c) 2007-2009 Intel Corporation.";
  58
  59 static const struct e1000_info *igb_info_tbl[] = {
  60         [board_82575] = &e1000_82575_info,
  61 };
  62
  63 static struct pci_device_id igb_pci_tbl[] = {
  64         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
  65         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_NS), board_82575 },
  66         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
  67         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
  68         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
  69         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
  70         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
  71         { PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
  72         /* required last entry */
  73         {0, }
  74 };
  75
  76 MODULE_DEVICE_TABLE(pci, igb_pci_tbl);
  77
  78 void igb_reset(struct igb_adapter *);
  79 static int igb_setup_all_tx_resources(struct igb_adapter *);
  80 static int igb_setup_all_rx_resources(struct igb_adapter *);
  81 static void igb_free_all_tx_resources(struct igb_adapter *);
  82 static void igb_free_all_rx_resources(struct igb_adapter *);
  83 void igb_update_stats(struct igb_adapter *);
  84 static int igb_probe(struct pci_dev *, const struct pci_device_id *);
  85 static void __devexit igb_remove(struct pci_dev *pdev);
  86 static int igb_sw_init(struct igb_adapter *);
  87 static int igb_open(struct net_device *);
  88 static int igb_close(struct net_device *);
  89 static void igb_configure_tx(struct igb_adapter *);
  90 static void igb_configure_rx(struct igb_adapter *);
  91 static void igb_setup_rctl(struct igb_adapter *);
  92 static void igb_clean_all_tx_rings(struct igb_adapter *);
  93 static void igb_clean_all_rx_rings(struct igb_adapter *);
  94 static void igb_clean_tx_ring(struct igb_ring *);
  95 static void igb_clean_rx_ring(struct igb_ring *);
  96 static void igb_set_multi(struct net_device *);
  97 static void igb_update_phy_info(unsigned long);
  98 static void igb_watchdog(unsigned long);
  99 static void igb_watchdog_task(struct work_struct *);
 100 static int igb_xmit_frame_ring_adv(struct sk_buff *, struct net_device *,
 101                                   struct igb_ring *);
 102 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *);
 103 static struct net_device_stats *igb_get_stats(struct net_device *);
 104 static int igb_change_mtu(struct net_device *, int);
 105 static int igb_set_mac(struct net_device *, void *);
 106 static irqreturn_t igb_intr(int irq, void *);
 107 static irqreturn_t igb_intr_msi(int irq, void *);
 108 static irqreturn_t igb_msix_other(int irq, void *);
 109 static irqreturn_t igb_msix_rx(int irq, void *);
 110 static irqreturn_t igb_msix_tx(int irq, void *);
 111 #ifdef CONFIG_IGB_DCA
 112 static void igb_update_rx_dca(struct igb_ring *);
 113 static void igb_update_tx_dca(struct igb_ring *);
 114 static void igb_setup_dca(struct igb_adapter *);
 115 #endif /* CONFIG_IGB_DCA */
 116 static bool igb_clean_tx_irq(struct igb_ring *);
 117 static int igb_poll(struct napi_struct *, int);
 118 static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
 119 static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
 120 static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
 121 static void igb_tx_timeout(struct net_device *);
 122 static void igb_reset_task(struct work_struct *);
 123 static void igb_vlan_rx_register(struct net_device *, struct vlan_group *);
 124 static void igb_vlan_rx_add_vid(struct net_device *, u16);
 125 static void igb_vlan_rx_kill_vid(struct net_device *, u16);
 126 static void igb_restore_vlan(struct igb_adapter *);
 127 static void igb_ping_all_vfs(struct igb_adapter *);
 128 static void igb_msg_task(struct igb_adapter *);
 129 static int igb_rcv_msg_from_vf(struct igb_adapter *, u32);
 130 static inline void igb_set_rah_pool(struct e1000_hw *, int , int);
 131 static void igb_set_mc_list_pools(struct igb_adapter *, int, u16);
 132 static void igb_vmm_control(struct igb_adapter *);
 133 static inline void igb_set_vmolr(struct e1000_hw *, int);
 134 static inline int igb_set_vf_rlpml(struct igb_adapter *, int, int);
 135 static int igb_set_vf_mac(struct igb_adapter *adapter, int, unsigned char *);
 136 static void igb_restore_vf_multicasts(struct igb_adapter *adapter);
 137
 138 static int igb_suspend(struct pci_dev *, pm_message_t);
 139 #ifdef CONFIG_PM
 140 static int igb_resume(struct pci_dev *);
 141 #endif
 142 static void igb_shutdown(struct pci_dev *);
 143 #ifdef CONFIG_IGB_DCA
 144 static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
 145 static struct notifier_block dca_notifier = {
 146         .notifier_call  = igb_notify_dca,
 147         .next           = NULL,
 148         .priority       = 0
 149 };
 150 #endif
 151 #ifdef CONFIG_NET_POLL_CONTROLLER
 152 /* for netdump / net console */
 153 static void igb_netpoll(struct net_device *);
 154 #endif
 155
 156 #ifdef CONFIG_PCI_IOV
 157 static ssize_t igb_set_num_vfs(struct device *, struct device_attribute *,
 158                                const char *, size_t);
 159 static ssize_t igb_show_num_vfs(struct device *, struct device_attribute *,
 160                                char *);
 161 DEVICE_ATTR(num_vfs, S_IRUGO | S_IWUSR, igb_show_num_vfs, igb_set_num_vfs);
 162 #endif
 163 static pci_ers_result_t igb_io_error_detected(struct pci_dev *,
 164                      pci_channel_state_t);
 165 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *);
 166 static void igb_io_resume(struct pci_dev *);
 167
 168 static struct pci_error_handlers igb_err_handler = {
 169         .error_detected = igb_io_error_detected,
 170         .slot_reset = igb_io_slot_reset,
 171         .resume = igb_io_resume,
 172 };
 173
 174
 175 static struct pci_driver igb_driver = {
 176         .name     = igb_driver_name,
 177         .id_table = igb_pci_tbl,
 178         .probe    = igb_probe,
 179         .remove   = __devexit_p(igb_remove),
 180 #ifdef CONFIG_PM
 181         /* Power Managment Hooks */
 182         .suspend  = igb_suspend,
 183         .resume   = igb_resume,
 184 #endif
 185         .shutdown = igb_shutdown,
 186         .err_handler = &igb_err_handler
 187 };
 188
 189 static int global_quad_port_a; /* global quad port a indication */
 190
 191 MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
 192 MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
 193 MODULE_LICENSE("GPL");
 194 MODULE_VERSION(DRV_VERSION);
 195
 196 /**
 197  * Scale the NIC clock cycle by a large factor so that
 198  * relatively small clock corrections can be added or
 199  * substracted at each clock tick. The drawbacks of a
 200  * large factor are a) that the clock register overflows
 201  * more quickly (not such a big deal) and b) that the
 202  * increment per tick has to fit into 24 bits.
 203  *
 204  * Note that
 205  *   TIMINCA = IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS *
 206  *             IGB_TSYNC_SCALE
 207  *   TIMINCA += TIMINCA * adjustment [ppm] / 1e9
 208  *
 209  * The base scale factor is intentionally a power of two
 210  * so that the division in %struct timecounter can be done with
 211  * a shift.
 212  */
 213 #define IGB_TSYNC_SHIFT (19)
 214 #define IGB_TSYNC_SCALE (1<<IGB_TSYNC_SHIFT)
 215
 216 /**
 217  * The duration of one clock cycle of the NIC.
 218  *
 219  * @todo This hard-coded value is part of the specification and might change
 220  * in future hardware revisions. Add revision check.
 221  */
 222 #define IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS 16
 223
 224 #if (IGB_TSYNC_SCALE * IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS) >= (1<<24)
 225 # error IGB_TSYNC_SCALE and/or IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS are too large to fit into TIMINCA
 226 #endif
 227
 228 /**
 229  * igb_read_clock - read raw cycle counter (to be used by time counter)
 230  */
 231 static cycle_t igb_read_clock(const struct cyclecounter *tc)
 232 {
 233         struct igb_adapter *adapter =
 234                 container_of(tc, struct igb_adapter, cycles);
 235         struct e1000_hw *hw = &adapter->hw;
 236         u64 stamp;
 237
 238         stamp =  rd32(E1000_SYSTIML);
 239         stamp |= (u64)rd32(E1000_SYSTIMH) << 32ULL;
 240
 241         return stamp;
 242 }
 243
 244 #ifdef DEBUG
 245 /**
 246  * igb_get_hw_dev_name - return device name string
 247  * used by hardware layer to print debugging information
 248  **/
 249 char *igb_get_hw_dev_name(struct e1000_hw *hw)
 250 {
 251         struct igb_adapter *adapter = hw->back;
 252         return adapter->netdev->name;
 253 }
 254
 255 /**
 256  * igb_get_time_str - format current NIC and system time as string
 257  */
 258 static char *igb_get_time_str(struct igb_adapter *adapter,
 259                               char buffer[160])
 260 {
 261         cycle_t hw = adapter->cycles.read(&adapter->cycles);
 262         struct timespec nic = ns_to_timespec(timecounter_read(&adapter->clock));
 263         struct timespec sys;
 264         struct timespec delta;
 265         getnstimeofday(&sys);
 266
 267         delta = timespec_sub(nic, sys);
 268
 269         sprintf(buffer,
 270                 "HW %llu, NIC %ld.%09lus, SYS %ld.%09lus, NIC-SYS %lds + %09luns",
 271                 hw,
 272                 (long)nic.tv_sec, nic.tv_nsec,
 273                 (long)sys.tv_sec, sys.tv_nsec,
 274                 (long)delta.tv_sec, delta.tv_nsec);
 275
 276         return buffer;
 277 }
 278 #endif
 279
 280 /**
 281  * igb_init_module - Driver Registration Routine
 282  *
 283  * igb_init_module is the first routine called when the driver is
 284  * loaded. All it does is register with the PCI subsystem.
 285  **/
 286 static int __init igb_init_module(void)
 287 {
 288         int ret;
 289         printk(KERN_INFO "%s - version %s\n",
 290                igb_driver_string, igb_driver_version);
 291
 292         printk(KERN_INFO "%s\n", igb_copyright);
 293
 294         global_quad_port_a = 0;
 295
 296 #ifdef CONFIG_IGB_DCA
 297         dca_register_notify(&dca_notifier);
 298 #endif
 299
 300         ret = pci_register_driver(&igb_driver);
 301         return ret;
 302 }
 303
 304 module_init(igb_init_module);
 305
 306 /**
 307  * igb_exit_module - Driver Exit Cleanup Routine
 308  *
 309  * igb_exit_module is called just before the driver is removed
 310  * from memory.
 311  **/
 312 static void __exit igb_exit_module(void)
 313 {
 314 #ifdef CONFIG_IGB_DCA
 315         dca_unregister_notify(&dca_notifier);
 316 #endif
 317         pci_unregister_driver(&igb_driver);
 318 }
 319
 320 module_exit(igb_exit_module);
 321
 322 #define Q_IDX_82576(i) (((i & 0x1) << 3) + (i >> 1))
 323 /**
 324  * igb_cache_ring_register - Descriptor ring to register mapping
 325  * @adapter: board private structure to initialize
 326  *
 327  * Once we know the feature-set enabled for the device, we'll cache
 328  * the register offset the descriptor ring is assigned to.
 329  **/
 330 static void igb_cache_ring_register(struct igb_adapter *adapter)
 331 {
 332         int i;
 333         unsigned int rbase_offset = adapter->vfs_allocated_count;
 334
 335         switch (adapter->hw.mac.type) {
 336         case e1000_82576:
 337                 /* The queues are allocated for virtualization such that VF 0
 338                  * is allocated queues 0 and 8, VF 1 queues 1 and 9, etc.
 339                  * In order to avoid collision we start at the first free queue
 340                  * and continue consuming queues in the same sequence
 341                  */
 342                 for (i = 0; i < adapter->num_rx_queues; i++)
 343                         adapter->rx_ring[i].reg_idx = rbase_offset +
 344                                                       Q_IDX_82576(i);
 345                 for (i = 0; i < adapter->num_tx_queues; i++)
 346                         adapter->tx_ring[i].reg_idx = rbase_offset +
 347                                                       Q_IDX_82576(i);
 348                 break;
 349         case e1000_82575:
 350         default:
 351                 for (i = 0; i < adapter->num_rx_queues; i++)
 352                         adapter->rx_ring[i].reg_idx = i;
 353                 for (i = 0; i < adapter->num_tx_queues; i++)
 354                         adapter->tx_ring[i].reg_idx = i;
 355                 break;
 356         }
 357 }
 358
 359 /**
 360  * igb_alloc_queues - Allocate memory for all rings
 361  * @adapter: board private structure to initialize
 362  *
 363  * We allocate one ring per queue at run-time since we don't know the
 364  * number of queues at compile-time.
 365  **/
 366 static int igb_alloc_queues(struct igb_adapter *adapter)
 367 {
 368         int i;
 369
 370         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
 371                                    sizeof(struct igb_ring), GFP_KERNEL);
 372         if (!adapter->tx_ring)
 373                 return -ENOMEM;
 374
 375         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
 376                                    sizeof(struct igb_ring), GFP_KERNEL);
 377         if (!adapter->rx_ring) {
 378                 kfree(adapter->tx_ring);
 379                 return -ENOMEM;
 380         }
 381
 382         adapter->rx_ring->buddy = adapter->tx_ring;
 383
 384         for (i = 0; i < adapter->num_tx_queues; i++) {
 385                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 386                 ring->count = adapter->tx_ring_count;
 387                 ring->adapter = adapter;
 388                 ring->queue_index = i;
 389         }
 390         for (i = 0; i < adapter->num_rx_queues; i++) {
 391                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 392                 ring->count = adapter->rx_ring_count;
 393                 ring->adapter = adapter;
 394                 ring->queue_index = i;
 395                 ring->itr_register = E1000_ITR;
 396
 397                 /* set a default napi handler for each rx_ring */
 398                 netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
 399         }
 400
 401         igb_cache_ring_register(adapter);
 402         return 0;
 403 }
 404
 405 static void igb_free_queues(struct igb_adapter *adapter)
 406 {
 407         int i;
 408
 409         for (i = 0; i < adapter->num_rx_queues; i++)
 410                 netif_napi_del(&adapter->rx_ring[i].napi);
 411
 412         kfree(adapter->tx_ring);
 413         kfree(adapter->rx_ring);
 414 }
 415
 416 #define IGB_N0_QUEUE -1
 417 static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
 418                               int tx_queue, int msix_vector)
 419 {
 420         u32 msixbm = 0;
 421         struct e1000_hw *hw = &adapter->hw;
 422         u32 ivar, index;
 423
 424         switch (hw->mac.type) {
 425         case e1000_82575:
 426                 /* The 82575 assigns vectors using a bitmask, which matches the
 427                    bitmask for the EICR/EIMS/EIMC registers.  To assign one
 428                    or more queues to a vector, we write the appropriate bits
 429                    into the MSIXBM register for that vector. */
 430                 if (rx_queue > IGB_N0_QUEUE) {
 431                         msixbm = E1000_EICR_RX_QUEUE0 << rx_queue;
 432                         adapter->rx_ring[rx_queue].eims_value = msixbm;
 433                 }
 434                 if (tx_queue > IGB_N0_QUEUE) {
 435                         msixbm |= E1000_EICR_TX_QUEUE0 << tx_queue;
 436                         adapter->tx_ring[tx_queue].eims_value =
 437                                   E1000_EICR_TX_QUEUE0 << tx_queue;
 438                 }
 439                 array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
 440                 break;
 441         case e1000_82576:
 442                 /* 82576 uses a table-based method for assigning vectors.
 443                    Each queue has a single entry in the table to which we write
 444                    a vector number along with a "valid" bit.  Sadly, the layout
 445                    of the table is somewhat counterintuitive. */
 446                 if (rx_queue > IGB_N0_QUEUE) {
 447                         index = (rx_queue >> 1) + adapter->vfs_allocated_count;
 448                         ivar = array_rd32(E1000_IVAR0, index);
 449                         if (rx_queue & 0x1) {
 450                                 /* vector goes into third byte of register */
 451                                 ivar = ivar & 0xFF00FFFF;
 452                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
 453                         } else {
 454                                 /* vector goes into low byte of register */
 455                                 ivar = ivar & 0xFFFFFF00;
 456                                 ivar |= msix_vector | E1000_IVAR_VALID;
 457                         }
 458                         adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
 459                         array_wr32(E1000_IVAR0, index, ivar);
 460                 }
 461                 if (tx_queue > IGB_N0_QUEUE) {
 462                         index = (tx_queue >> 1) + adapter->vfs_allocated_count;
 463                         ivar = array_rd32(E1000_IVAR0, index);
 464                         if (tx_queue & 0x1) {
 465                                 /* vector goes into high byte of register */
 466                                 ivar = ivar & 0x00FFFFFF;
 467                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
 468                         } else {
 469                                 /* vector goes into second byte of register */
 470                                 ivar = ivar & 0xFFFF00FF;
 471                                 ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
 472                         }
 473                         adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
 474                         array_wr32(E1000_IVAR0, index, ivar);
 475                 }
 476                 break;
 477         default:
 478                 BUG();
 479                 break;
 480         }
 481 }
 482
 483 /**
 484  * igb_configure_msix - Configure MSI-X hardware
 485  *
 486  * igb_configure_msix sets up the hardware to properly
 487  * generate MSI-X interrupts.
 488  **/
 489 static void igb_configure_msix(struct igb_adapter *adapter)
 490 {
 491         u32 tmp;
 492         int i, vector = 0;
 493         struct e1000_hw *hw = &adapter->hw;
 494
 495         adapter->eims_enable_mask = 0;
 496         if (hw->mac.type == e1000_82576)
 497                 /* Turn on MSI-X capability first, or our settings
 498                  * won't stick.  And it will take days to debug. */
 499                 wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
 500                                    E1000_GPIE_PBA | E1000_GPIE_EIAME |
 501                                    E1000_GPIE_NSICR);
 502
 503         for (i = 0; i < adapter->num_tx_queues; i++) {
 504                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
 505                 igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
 506                 adapter->eims_enable_mask |= tx_ring->eims_value;
 507                 if (tx_ring->itr_val)
 508                         writel(tx_ring->itr_val,
 509                                hw->hw_addr + tx_ring->itr_register);
 510                 else
 511                         writel(1, hw->hw_addr + tx_ring->itr_register);
 512         }
 513
 514         for (i = 0; i < adapter->num_rx_queues; i++) {
 515                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
 516                 rx_ring->buddy = NULL;
 517                 igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
 518                 adapter->eims_enable_mask |= rx_ring->eims_value;
 519                 if (rx_ring->itr_val)
 520                         writel(rx_ring->itr_val,
 521                                hw->hw_addr + rx_ring->itr_register);
 522                 else
 523                         writel(1, hw->hw_addr + rx_ring->itr_register);
 524         }
 525
 526
 527         /* set vector for other causes, i.e. link changes */
 528         switch (hw->mac.type) {
 529         case e1000_82575:
 530                 array_wr32(E1000_MSIXBM(0), vector++,
 531                                       E1000_EIMS_OTHER);
 532
 533                 tmp = rd32(E1000_CTRL_EXT);
 534                 /* enable MSI-X PBA support*/
 535                 tmp |= E1000_CTRL_EXT_PBA_CLR;
 536
 537                 /* Auto-Mask interrupts upon ICR read. */
 538                 tmp |= E1000_CTRL_EXT_EIAME;
 539                 tmp |= E1000_CTRL_EXT_IRCA;
 540
 541                 wr32(E1000_CTRL_EXT, tmp);
 542                 adapter->eims_enable_mask |= E1000_EIMS_OTHER;
 543                 adapter->eims_other = E1000_EIMS_OTHER;
 544
 545                 break;
 546
 547         case e1000_82576:
 548                 tmp = (vector++ | E1000_IVAR_VALID) << 8;
 549                 wr32(E1000_IVAR_MISC, tmp);
 550
 551                 adapter->eims_enable_mask = (1 << (vector)) - 1;
 552                 adapter->eims_other = 1 << (vector - 1);
 553                 break;
 554         default:
 555                 /* do nothing, since nothing else supports MSI-X */
 556                 break;
 557         } /* switch (hw->mac.type) */
 558         wrfl();
 559 }
 560
 561 /**
 562  * igb_request_msix - Initialize MSI-X interrupts
 563  *
 564  * igb_request_msix allocates MSI-X vectors and requests interrupts from the
 565  * kernel.
 566  **/
 567 static int igb_request_msix(struct igb_adapter *adapter)
 568 {
 569         struct net_device *netdev = adapter->netdev;
 570         int i, err = 0, vector = 0;
 571
 572         vector = 0;
 573
 574         for (i = 0; i < adapter->num_tx_queues; i++) {
 575                 struct igb_ring *ring = &(adapter->tx_ring[i]);
 576                 sprintf(ring->name, "%s-tx-%d", netdev->name, i);
 577                 err = request_irq(adapter->msix_entries[vector].vector,
 578                                   &igb_msix_tx, 0, ring->name,
 579                                   &(adapter->tx_ring[i]));
 580                 if (err)
 581                         goto out;
 582                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 583                 ring->itr_val = 976; /* ~4000 ints/sec */
 584                 vector++;
 585         }
 586         for (i = 0; i < adapter->num_rx_queues; i++) {
 587                 struct igb_ring *ring = &(adapter->rx_ring[i]);
 588                 if (strlen(netdev->name) < (IFNAMSIZ - 5))
 589                         sprintf(ring->name, "%s-rx-%d", netdev->name, i);
 590                 else
 591                         memcpy(ring->name, netdev->name, IFNAMSIZ);
 592                 err = request_irq(adapter->msix_entries[vector].vector,
 593                                   &igb_msix_rx, 0, ring->name,
 594                                   &(adapter->rx_ring[i]));
 595                 if (err)
 596                         goto out;
 597                 ring->itr_register = E1000_EITR(0) + (vector << 2);
 598                 ring->itr_val = adapter->itr;
 599                 vector++;
 600         }
 601
 602         err = request_irq(adapter->msix_entries[vector].vector,
 603                           &igb_msix_other, 0, netdev->name, netdev);
 604         if (err)
 605                 goto out;
 606
 607         igb_configure_msix(adapter);
 608         return 0;
 609 out:
 610         return err;
 611 }
 612
 613 static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
 614 {
 615         if (adapter->msix_entries) {
 616                 pci_disable_msix(adapter->pdev);
 617                 kfree(adapter->msix_entries);
 618                 adapter->msix_entries = NULL;
 619         } else if (adapter->flags & IGB_FLAG_HAS_MSI)
 620                 pci_disable_msi(adapter->pdev);
 621         return;
 622 }
 623
 624
 625 /**
 626  * igb_set_interrupt_capability - set MSI or MSI-X if supported
 627  *
 628  * Attempt to configure interrupts using the best available
 629  * capabilities of the hardware and kernel.
 630  **/
 631 static void igb_set_interrupt_capability(struct igb_adapter *adapter)
 632 {
 633         int err;
 634         int numvecs, i;
 635
 636         /* Number of supported queues. */
 637         /* Having more queues than CPUs doesn't make sense. */
 638         adapter->num_rx_queues = min_t(u32, IGB_MAX_RX_QUEUES, num_online_cpus());
 639         adapter->num_tx_queues = min_t(u32, IGB_MAX_TX_QUEUES, num_online_cpus());
 640
 641         numvecs = adapter->num_tx_queues + adapter->num_rx_queues + 1;
 642         adapter->msix_entries = kcalloc(numvecs, sizeof(struct msix_entry),
 643                                         GFP_KERNEL);
 644         if (!adapter->msix_entries)
 645                 goto msi_only;
 646
 647         for (i = 0; i < numvecs; i++)
 648                 adapter->msix_entries[i].entry = i;
 649
 650         err = pci_enable_msix(adapter->pdev,
 651                               adapter->msix_entries,
 652                               numvecs);
 653         if (err == 0)
 654                 goto out;
 655
 656         igb_reset_interrupt_capability(adapter);
 657
 658         /* If we can't do MSI-X, try MSI */
 659 msi_only:
 660         adapter->num_rx_queues = 1;
 661         adapter->num_tx_queues = 1;
 662         if (!pci_enable_msi(adapter->pdev))
 663                 adapter->flags |= IGB_FLAG_HAS_MSI;
 664 out:
 665         /* Notify the stack of the (possibly) reduced Tx Queue count. */
 666         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
 667         return;
 668 }
 669
 670 /**
 671  * igb_request_irq - initialize interrupts
 672  *
 673  * Attempts to configure interrupts using the best available
 674  * capabilities of the hardware and kernel.
 675  **/
 676 static int igb_request_irq(struct igb_adapter *adapter)
 677 {
 678         struct net_device *netdev = adapter->netdev;
 679         struct e1000_hw *hw = &adapter->hw;
 680         int err = 0;
 681
 682         if (adapter->msix_entries) {
 683                 err = igb_request_msix(adapter);
 684                 if (!err)
 685                         goto request_done;
 686                 /* fall back to MSI */
 687                 igb_reset_interrupt_capability(adapter);
 688                 if (!pci_enable_msi(adapter->pdev))
 689                         adapter->flags |= IGB_FLAG_HAS_MSI;
 690                 igb_free_all_tx_resources(adapter);
 691                 igb_free_all_rx_resources(adapter);
 692                 adapter->num_rx_queues = 1;
 693                 igb_alloc_queues(adapter);
 694         } else {
 695                 switch (hw->mac.type) {
 696                 case e1000_82575:
 697                         wr32(E1000_MSIXBM(0),
 698                              (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
 699                         break;
 700                 case e1000_82576:
 701                         wr32(E1000_IVAR0, E1000_IVAR_VALID);
 702                         break;
 703                 default:
 704                         break;
 705                 }
 706         }
 707
 708         if (adapter->flags & IGB_FLAG_HAS_MSI) {
 709                 err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
 710                                   netdev->name, netdev);
 711                 if (!err)
 712                         goto request_done;
 713                 /* fall back to legacy interrupts */
 714                 igb_reset_interrupt_capability(adapter);
 715                 adapter->flags &= ~IGB_FLAG_HAS_MSI;
 716         }
 717
 718         err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
 719                           netdev->name, netdev);
 720
 721         if (err)
 722                 dev_err(&adapter->pdev->dev, "Error %d getting interrupt\n",
 723                         err);
 724
 725 request_done:
 726         return err;
 727 }
 728
 729 static void igb_free_irq(struct igb_adapter *adapter)
 730 {
 731         struct net_device *netdev = adapter->netdev;
 732
 733         if (adapter->msix_entries) {
 734                 int vector = 0, i;
 735
 736                 for (i = 0; i < adapter->num_tx_queues; i++)
 737                         free_irq(adapter->msix_entries[vector++].vector,
 738                                 &(adapter->tx_ring[i]));
 739                 for (i = 0; i < adapter->num_rx_queues; i++)
 740                         free_irq(adapter->msix_entries[vector++].vector,
 741                                 &(adapter->rx_ring[i]));
 742
 743                 free_irq(adapter->msix_entries[vector++].vector, netdev);
 744                 return;
 745         }
 746
 747         free_irq(adapter->pdev->irq, netdev);
 748 }
 749
 750 /**
 751  * igb_irq_disable - Mask off interrupt generation on the NIC
 752  * @adapter: board private structure
 753  **/
 754 static void igb_irq_disable(struct igb_adapter *adapter)
 755 {
 756         struct e1000_hw *hw = &adapter->hw;
 757
 758         if (adapter->msix_entries) {
 759                 wr32(E1000_EIAM, 0);
 760                 wr32(E1000_EIMC, ~0);
 761                 wr32(E1000_EIAC, 0);
 762         }
 763
 764         wr32(E1000_IAM, 0);
 765         wr32(E1000_IMC, ~0);
 766         wrfl();
 767         synchronize_irq(adapter->pdev->irq);
 768 }
 769
 770 /**
 771  * igb_irq_enable - Enable default interrupt generation settings
 772  * @adapter: board private structure
 773  **/
 774 static void igb_irq_enable(struct igb_adapter *adapter)
 775 {
 776         struct e1000_hw *hw = &adapter->hw;
 777
 778         if (adapter->msix_entries) {
 779                 wr32(E1000_EIAC, adapter->eims_enable_mask);
 780                 wr32(E1000_EIAM, adapter->eims_enable_mask);
 781                 wr32(E1000_EIMS, adapter->eims_enable_mask);
 782                 if (adapter->vfs_allocated_count)
 783                         wr32(E1000_MBVFIMR, 0xFF);
 784                 wr32(E1000_IMS, (E1000_IMS_LSC | E1000_IMS_VMMB |
 785                                  E1000_IMS_DOUTSYNC));
 786         } else {
 787                 wr32(E1000_IMS, IMS_ENABLE_MASK);
 788                 wr32(E1000_IAM, IMS_ENABLE_MASK);
 789         }
 790 }
 791
 792 static void igb_update_mng_vlan(struct igb_adapter *adapter)
 793 {
 794         struct net_device *netdev = adapter->netdev;
 795         u16 vid = adapter->hw.mng_cookie.vlan_id;
 796         u16 old_vid = adapter->mng_vlan_id;
 797         if (adapter->vlgrp) {
 798                 if (!vlan_group_get_device(adapter->vlgrp, vid)) {
 799                         if (adapter->hw.mng_cookie.status &
 800                                 E1000_MNG_DHCP_COOKIE_STATUS_VLAN) {
 801                                 igb_vlan_rx_add_vid(netdev, vid);
 802                                 adapter->mng_vlan_id = vid;
 803                         } else
 804                                 adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
 805
 806                         if ((old_vid != (u16)IGB_MNG_VLAN_NONE) &&
 807                                         (vid != old_vid) &&
 808                             !vlan_group_get_device(adapter->vlgrp, old_vid))
 809                                 igb_vlan_rx_kill_vid(netdev, old_vid);
 810                 } else
 811                         adapter->mng_vlan_id = vid;
 812         }
 813 }
 814
 815 /**
 816  * igb_release_hw_control - release control of the h/w to f/w
 817  * @adapter: address of board private structure
 818  *
 819  * igb_release_hw_control resets CTRL_EXT:DRV_LOAD bit.
 820  * For ASF and Pass Through versions of f/w this means that the
 821  * driver is no longer loaded.
 822  *
 823  **/
 824 static void igb_release_hw_control(struct igb_adapter *adapter)
 825 {
 826         struct e1000_hw *hw = &adapter->hw;
 827         u32 ctrl_ext;
 828
 829         /* Let firmware take over control of h/w */
 830         ctrl_ext = rd32(E1000_CTRL_EXT);
 831         wr32(E1000_CTRL_EXT,
 832                         ctrl_ext & ~E1000_CTRL_EXT_DRV_LOAD);
 833 }
 834
 835
 836 /**
 837  * igb_get_hw_control - get control of the h/w from f/w
 838  * @adapter: address of board private structure
 839  *
 840  * igb_get_hw_control sets CTRL_EXT:DRV_LOAD bit.
 841  * For ASF and Pass Through versions of f/w this means that
 842  * the driver is loaded.
 843  *
 844  **/
 845 static void igb_get_hw_control(struct igb_adapter *adapter)
 846 {
 847         struct e1000_hw *hw = &adapter->hw;
 848         u32 ctrl_ext;
 849
 850         /* Let firmware know the driver has taken over */
 851         ctrl_ext = rd32(E1000_CTRL_EXT);
 852         wr32(E1000_CTRL_EXT,
 853                         ctrl_ext | E1000_CTRL_EXT_DRV_LOAD);
 854 }
 855
 856 /**
 857  * igb_configure - configure the hardware for RX and TX
 858  * @adapter: private board structure
 859  **/
 860 static void igb_configure(struct igb_adapter *adapter)
 861 {
 862         struct net_device *netdev = adapter->netdev;
 863         int i;
 864
 865         igb_get_hw_control(adapter);
 866         igb_set_multi(netdev);
 867
 868         igb_restore_vlan(adapter);
 869
 870         igb_configure_tx(adapter);
 871         igb_setup_rctl(adapter);
 872         igb_configure_rx(adapter);
 873
 874         igb_rx_fifo_flush_82575(&adapter->hw);
 875
 876         /* call IGB_DESC_UNUSED which always leaves
 877          * at least 1 descriptor unused to make sure
 878          * next_to_use != next_to_clean */
 879         for (i = 0; i < adapter->num_rx_queues; i++) {
 880                 struct igb_ring *ring = &adapter->rx_ring[i];
 881                 igb_alloc_rx_buffers_adv(ring, IGB_DESC_UNUSED(ring));
 882         }
 883
 884
 885         adapter->tx_queue_len = netdev->tx_queue_len;
 886 }
 887
 888
 889 /**
 890  * igb_up - Open the interface and prepare it to handle traffic
 891  * @adapter: board private structure
 892  **/
 893
 894 int igb_up(struct igb_adapter *adapter)
 895 {
 896         struct e1000_hw *hw = &adapter->hw;
 897         int i;
 898
 899         /* hardware has been reset, we need to reload some things */
 900         igb_configure(adapter);
 901
 902         clear_bit(__IGB_DOWN, &adapter->state);
 903
 904         for (i = 0; i < adapter->num_rx_queues; i++)
 905                 napi_enable(&adapter->rx_ring[i].napi);
 906         if (adapter->msix_entries)
 907                 igb_configure_msix(adapter);
 908
 909         igb_vmm_control(adapter);
 910         igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
 911         igb_set_vmolr(hw, adapter->vfs_allocated_count);
 912
 913         /* Clear any pending interrupts. */
 914         rd32(E1000_ICR);
 915         igb_irq_enable(adapter);
 916
 917         /* Fire a link change interrupt to start the watchdog. */
 918         wr32(E1000_ICS, E1000_ICS_LSC);
 919         return 0;
 920 }
 921
 922 void igb_down(struct igb_adapter *adapter)
 923 {
 924         struct e1000_hw *hw = &adapter->hw;
 925         struct net_device *netdev = adapter->netdev;
 926         u32 tctl, rctl;
 927         int i;
 928
 929         /* signal that we're down so the interrupt handler does not
 930          * reschedule our watchdog timer */
 931         set_bit(__IGB_DOWN, &adapter->state);
 932
 933         /* disable receives in the hardware */
 934         rctl = rd32(E1000_RCTL);
 935         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
 936         /* flush and sleep below */
 937
 938         netif_tx_stop_all_queues(netdev);
 939
 940         /* disable transmits in the hardware */
 941         tctl = rd32(E1000_TCTL);
 942         tctl &= ~E1000_TCTL_EN;
 943         wr32(E1000_TCTL, tctl);
 944         /* flush both disables and wait for them to finish */
 945         wrfl();
 946         msleep(10);
 947
 948         for (i = 0; i < adapter->num_rx_queues; i++)
 949                 napi_disable(&adapter->rx_ring[i].napi);
 950
 951         igb_irq_disable(adapter);
 952
 953         del_timer_sync(&adapter->watchdog_timer);
 954         del_timer_sync(&adapter->phy_info_timer);
 955
 956         netdev->tx_queue_len = adapter->tx_queue_len;
 957         netif_carrier_off(netdev);
 958
 959         /* record the stats before reset*/
 960         igb_update_stats(adapter);
 961
 962         adapter->link_speed = 0;
 963         adapter->link_duplex = 0;
 964
 965         if (!pci_channel_offline(adapter->pdev))
 966                 igb_reset(adapter);
 967         igb_clean_all_tx_rings(adapter);
 968         igb_clean_all_rx_rings(adapter);
 969 }
 970
 971 void igb_reinit_locked(struct igb_adapter *adapter)
 972 {
 973         WARN_ON(in_interrupt());
 974         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
 975                 msleep(1);
 976         igb_down(adapter);
 977         igb_up(adapter);
 978         clear_bit(__IGB_RESETTING, &adapter->state);
 979 }
 980
 981 void igb_reset(struct igb_adapter *adapter)
 982 {
 983         struct e1000_hw *hw = &adapter->hw;
 984         struct e1000_mac_info *mac = &hw->mac;
 985         struct e1000_fc_info *fc = &hw->fc;
 986         u32 pba = 0, tx_space, min_tx_space, min_rx_space;
 987         u16 hwm;
 988
 989         /* Repartition Pba for greater than 9k mtu
 990          * To take effect CTRL.RST is required.
 991          */
 992         switch (mac->type) {
 993         case e1000_82576:
 994                 pba = E1000_PBA_64K;
 995                 break;
 996         case e1000_82575:
 997         default:
 998                 pba = E1000_PBA_34K;
 999                 break;
1000         }
1001
1002         if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
1003             (mac->type < e1000_82576)) {
1004                 /* adjust PBA for jumbo frames */
1005                 wr32(E1000_PBA, pba);
1006
1007                 /* To maintain wire speed transmits, the Tx FIFO should be
1008                  * large enough to accommodate two full transmit packets,
1009                  * rounded up to the next 1KB and expressed in KB.  Likewise,
1010                  * the Rx FIFO should be large enough to accommodate at least
1011                  * one full receive packet and is similarly rounded up and
1012                  * expressed in KB. */
1013                 pba = rd32(E1000_PBA);
1014                 /* upper 16 bits has Tx packet buffer allocation size in KB */
1015                 tx_space = pba >> 16;
1016                 /* lower 16 bits has Rx packet buffer allocation size in KB */
1017                 pba &= 0xffff;
1018                 /* the tx fifo also stores 16 bytes of information about the tx
1019                  * but don't include ethernet FCS because hardware appends it */
1020                 min_tx_space = (adapter->max_frame_size +
1021                                 sizeof(union e1000_adv_tx_desc) -
1022                                 ETH_FCS_LEN) * 2;
1023                 min_tx_space = ALIGN(min_tx_space, 1024);
1024                 min_tx_space >>= 10;
1025                 /* software strips receive CRC, so leave room for it */
1026                 min_rx_space = adapter->max_frame_size;
1027                 min_rx_space = ALIGN(min_rx_space, 1024);
1028                 min_rx_space >>= 10;
1029
1030                 /* If current Tx allocation is less than the min Tx FIFO size,
1031                  * and the min Tx FIFO size is less than the current Rx FIFO
1032                  * allocation, take space away from current Rx allocation */
1033                 if (tx_space < min_tx_space &&
1034                     ((min_tx_space - tx_space) < pba)) {
1035                         pba = pba - (min_tx_space - tx_space);
1036
1037                         /* if short on rx space, rx wins and must trump tx
1038                          * adjustment */
1039                         if (pba < min_rx_space)
1040                                 pba = min_rx_space;
1041                 }
1042                 wr32(E1000_PBA, pba);
1043         }
1044
1045         /* flow control settings */
1046         /* The high water mark must be low enough to fit one full frame
1047          * (or the size used for early receive) above it in the Rx FIFO.
1048          * Set it to the lower of:
1049          * - 90% of the Rx FIFO size, or
1050          * - the full Rx FIFO size minus one full frame */
1051         hwm = min(((pba << 10) * 9 / 10),
1052                         ((pba << 10) - 2 * adapter->max_frame_size));
1053
1054         if (mac->type < e1000_82576) {
1055                 fc->high_water = hwm & 0xFFF8;  /* 8-byte granularity */
1056                 fc->low_water = fc->high_water - 8;
1057         } else {
1058                 fc->high_water = hwm & 0xFFF0;  /* 16-byte granularity */
1059                 fc->low_water = fc->high_water - 16;
1060         }
1061         fc->pause_time = 0xFFFF;
1062         fc->send_xon = 1;
1063         fc->type = fc->original_type;
1064
1065         /* disable receive for all VFs and wait one second */
1066         if (adapter->vfs_allocated_count) {
1067                 int i;
1068                 for (i = 0 ; i < adapter->vfs_allocated_count; i++)
1069                         adapter->vf_data[i].clear_to_send = false;
1070
1071                 /* ping all the active vfs to let them know we are going down */
1072                         igb_ping_all_vfs(adapter);
1073
1074                 /* disable transmits and receives */
1075                 wr32(E1000_VFRE, 0);
1076                 wr32(E1000_VFTE, 0);
1077         }
1078
1079         /* Allow time for pending master requests to run */
1080         adapter->hw.mac.ops.reset_hw(&adapter->hw);
1081         wr32(E1000_WUC, 0);
1082
1083         if (adapter->hw.mac.ops.init_hw(&adapter->hw))
1084                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
1085
1086         igb_update_mng_vlan(adapter);
1087
1088         /* Enable h/w to recognize an 802.1Q VLAN Ethernet packet */
1089         wr32(E1000_VET, ETHERNET_IEEE_VLAN_TYPE);
1090
1091         igb_reset_adaptive(&adapter->hw);
1092         igb_get_phy_info(&adapter->hw);
1093 }
1094
1095 static const struct net_device_ops igb_netdev_ops = {
1096         .ndo_open               = igb_open,
1097         .ndo_stop               = igb_close,
1098         .ndo_start_xmit         = igb_xmit_frame_adv,
1099         .ndo_get_stats          = igb_get_stats,
1100         .ndo_set_multicast_list = igb_set_multi,
1101         .ndo_set_mac_address    = igb_set_mac,
1102         .ndo_change_mtu         = igb_change_mtu,
1103         .ndo_do_ioctl           = igb_ioctl,
1104         .ndo_tx_timeout         = igb_tx_timeout,
1105         .ndo_validate_addr      = eth_validate_addr,
1106         .ndo_vlan_rx_register   = igb_vlan_rx_register,
1107         .ndo_vlan_rx_add_vid    = igb_vlan_rx_add_vid,
1108         .ndo_vlan_rx_kill_vid   = igb_vlan_rx_kill_vid,
1109 #ifdef CONFIG_NET_POLL_CONTROLLER
1110         .ndo_poll_controller    = igb_netpoll,
1111 #endif
1112 };
1113
1114 /**
1115  * igb_probe - Device Initialization Routine
1116  * @pdev: PCI device information struct
1117  * @ent: entry in igb_pci_tbl
1118  *
1119  * Returns 0 on success, negative on failure
1120  *
1121  * igb_probe initializes an adapter identified by a pci_dev structure.
1122  * The OS initialization, configuring of the adapter private structure,
1123  * and a hardware reset occur.
1124  **/
1125 static int __devinit igb_probe(struct pci_dev *pdev,
1126                                const struct pci_device_id *ent)
1127 {
1128         struct net_device *netdev;
1129         struct igb_adapter *adapter;
1130         struct e1000_hw *hw;
1131         struct pci_dev *us_dev;
1132         const struct e1000_info *ei = igb_info_tbl[ent->driver_data];
1133         unsigned long mmio_start, mmio_len;
1134         int err, pci_using_dac, pos;
1135         u16 eeprom_data = 0, state = 0;
1136         u16 eeprom_apme_mask = IGB_EEPROM_APME;
1137         u32 part_num;
1138
1139         err = pci_enable_device_mem(pdev);
1140         if (err)
1141                 return err;
1142
1143         pci_using_dac = 0;
1144         err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
1145         if (!err) {
1146                 err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
1147                 if (!err)
1148                         pci_using_dac = 1;
1149         } else {
1150                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
1151                 if (err) {
1152                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
1153                         if (err) {
1154                                 dev_err(&pdev->dev, "No usable DMA "
1155                                         "configuration, aborting\n");
1156                                 goto err_dma;
1157                         }
1158                 }
1159         }
1160
1161         /* 82575 requires that the pci-e link partner disable the L0s state */
1162         switch (pdev->device) {
1163         case E1000_DEV_ID_82575EB_COPPER:
1164         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1165         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1166                 us_dev = pdev->bus->self;
1167                 pos = pci_find_capability(us_dev, PCI_CAP_ID_EXP);
1168                 if (pos) {
1169                         pci_read_config_word(us_dev, pos + PCI_EXP_LNKCTL,
1170                                              &state);
1171                         state &= ~PCIE_LINK_STATE_L0S;
1172                         pci_write_config_word(us_dev, pos + PCI_EXP_LNKCTL,
1173                                               state);
1174                         dev_info(&pdev->dev,
1175                                  "Disabling ASPM L0s upstream switch port %s\n",
1176                                  pci_name(us_dev));
1177                 }
1178         default:
1179                 break;
1180         }
1181
1182         err = pci_request_selected_regions(pdev, pci_select_bars(pdev,
1183                                            IORESOURCE_MEM),
1184                                            igb_driver_name);
1185         if (err)
1186                 goto err_pci_reg;
1187
1188         err = pci_enable_pcie_error_reporting(pdev);
1189         if (err) {
1190                 dev_err(&pdev->dev, "pci_enable_pcie_error_reporting failed "
1191                         "0x%x\n", err);
1192                 /* non-fatal, continue */
1193         }
1194
1195         pci_set_master(pdev);
1196         pci_save_state(pdev);
1197
1198         err = -ENOMEM;
1199         netdev = alloc_etherdev_mq(sizeof(struct igb_adapter),
1200                                    IGB_ABS_MAX_TX_QUEUES);
1201         if (!netdev)
1202                 goto err_alloc_etherdev;
1203
1204         SET_NETDEV_DEV(netdev, &pdev->dev);
1205
1206         pci_set_drvdata(pdev, netdev);
1207         adapter = netdev_priv(netdev);
1208         adapter->netdev = netdev;
1209         adapter->pdev = pdev;
1210         hw = &adapter->hw;
1211         hw->back = adapter;
1212         adapter->msg_enable = NETIF_MSG_DRV | NETIF_MSG_PROBE;
1213
1214         mmio_start = pci_resource_start(pdev, 0);
1215         mmio_len = pci_resource_len(pdev, 0);
1216
1217         err = -EIO;
1218         hw->hw_addr = ioremap(mmio_start, mmio_len);
1219         if (!hw->hw_addr)
1220                 goto err_ioremap;
1221
1222         netdev->netdev_ops = &igb_netdev_ops;
1223         igb_set_ethtool_ops(netdev);
1224         netdev->watchdog_timeo = 5 * HZ;
1225
1226         strncpy(netdev->name, pci_name(pdev), sizeof(netdev->name) - 1);
1227
1228         netdev->mem_start = mmio_start;
1229         netdev->mem_end = mmio_start + mmio_len;
1230
1231         /* PCI config space info */
1232         hw->vendor_id = pdev->vendor;
1233         hw->device_id = pdev->device;
1234         hw->revision_id = pdev->revision;
1235         hw->subsystem_vendor_id = pdev->subsystem_vendor;
1236         hw->subsystem_device_id = pdev->subsystem_device;
1237
1238         /* setup the private structure */
1239         hw->back = adapter;
1240         /* Copy the default MAC, PHY and NVM function pointers */
1241         memcpy(&hw->mac.ops, ei->mac_ops, sizeof(hw->mac.ops));
1242         memcpy(&hw->phy.ops, ei->phy_ops, sizeof(hw->phy.ops));
1243         memcpy(&hw->nvm.ops, ei->nvm_ops, sizeof(hw->nvm.ops));
1244         /* Initialize skew-specific constants */
1245         err = ei->get_invariants(hw);
1246         if (err)
1247                 goto err_sw_init;
1248
1249         /* setup the private structure */
1250         err = igb_sw_init(adapter);
1251         if (err)
1252                 goto err_sw_init;
1253
1254         igb_get_bus_info_pcie(hw);
1255
1256         /* set flags */
1257         switch (hw->mac.type) {
1258         case e1000_82575:
1259                 adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
1260                 break;
1261         case e1000_82576:
1262         default:
1263                 break;
1264         }
1265
1266         hw->phy.autoneg_wait_to_complete = false;
1267         hw->mac.adaptive_ifs = true;
1268
1269         /* Copper options */
1270         if (hw->phy.media_type == e1000_media_type_copper) {
1271                 hw->phy.mdix = AUTO_ALL_MODES;
1272                 hw->phy.disable_polarity_correction = false;
1273                 hw->phy.ms_type = e1000_ms_hw_default;
1274         }
1275
1276         if (igb_check_reset_block(hw))
1277                 dev_info(&pdev->dev,
1278                         "PHY reset is blocked due to SOL/IDER session.\n");
1279
1280         netdev->features = NETIF_F_SG |
1281                            NETIF_F_IP_CSUM |
1282                            NETIF_F_HW_VLAN_TX |
1283                            NETIF_F_HW_VLAN_RX |
1284                            NETIF_F_HW_VLAN_FILTER;
1285
1286         netdev->features |= NETIF_F_IPV6_CSUM;
1287         netdev->features |= NETIF_F_TSO;
1288         netdev->features |= NETIF_F_TSO6;
1289
1290         netdev->features |= NETIF_F_GRO;
1291
1292         netdev->vlan_features |= NETIF_F_TSO;
1293         netdev->vlan_features |= NETIF_F_TSO6;
1294         netdev->vlan_features |= NETIF_F_IP_CSUM;
1295         netdev->vlan_features |= NETIF_F_SG;
1296
1297         if (pci_using_dac)
1298                 netdev->features |= NETIF_F_HIGHDMA;
1299
1300         adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
1301
1302         /* before reading the NVM, reset the controller to put the device in a
1303          * known good starting state */
1304         hw->mac.ops.reset_hw(hw);
1305
1306         /* make sure the NVM is good */
1307         if (igb_validate_nvm_checksum(hw) < 0) {
1308                 dev_err(&pdev->dev, "The NVM Checksum Is Not Valid\n");
1309                 err = -EIO;
1310                 goto err_eeprom;
1311         }
1312
1313         /* copy the MAC address out of the NVM */
1314         if (hw->mac.ops.read_mac_addr(hw))
1315                 dev_err(&pdev->dev, "NVM Read Error\n");
1316
1317         memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
1318         memcpy(netdev->perm_addr, hw->mac.addr, netdev->addr_len);
1319
1320         if (!is_valid_ether_addr(netdev->perm_addr)) {
1321                 dev_err(&pdev->dev, "Invalid MAC Address\n");
1322                 err = -EIO;
1323                 goto err_eeprom;
1324         }
1325
1326         init_timer(&adapter->watchdog_timer);
1327         adapter->watchdog_timer.function = &igb_watchdog;
1328         adapter->watchdog_timer.data = (unsigned long) adapter;
1329
1330         init_timer(&adapter->phy_info_timer);
1331         adapter->phy_info_timer.function = &igb_update_phy_info;
1332         adapter->phy_info_timer.data = (unsigned long) adapter;
1333
1334         INIT_WORK(&adapter->reset_task, igb_reset_task);
1335         INIT_WORK(&adapter->watchdog_task, igb_watchdog_task);
1336
1337         /* Initialize link properties that are user-changeable */
1338         adapter->fc_autoneg = true;
1339         hw->mac.autoneg = true;
1340         hw->phy.autoneg_advertised = 0x2f;
1341
1342         hw->fc.original_type = e1000_fc_default;
1343         hw->fc.type = e1000_fc_default;
1344
1345         adapter->itr_setting = IGB_DEFAULT_ITR;
1346         adapter->itr = IGB_START_ITR;
1347
1348         igb_validate_mdi_setting(hw);
1349
1350         adapter->rx_csum = 1;
1351
1352         /* Initial Wake on LAN setting If APM wake is enabled in the EEPROM,
1353          * enable the ACPI Magic Packet filter
1354          */
1355
1356         if (hw->bus.func == 0)
1357                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_A, 1, &eeprom_data);
1358         else if (hw->bus.func == 1)
1359                 hw->nvm.ops.read(hw, NVM_INIT_CONTROL3_PORT_B, 1, &eeprom_data);
1360
1361         if (eeprom_data & eeprom_apme_mask)
1362                 adapter->eeprom_wol |= E1000_WUFC_MAG;
1363
1364         /* now that we have the eeprom settings, apply the special cases where
1365          * the eeprom may be wrong or the board simply won't support wake on
1366          * lan on a particular port */
1367         switch (pdev->device) {
1368         case E1000_DEV_ID_82575GB_QUAD_COPPER:
1369                 adapter->eeprom_wol = 0;
1370                 break;
1371         case E1000_DEV_ID_82575EB_FIBER_SERDES:
1372         case E1000_DEV_ID_82576_FIBER:
1373         case E1000_DEV_ID_82576_SERDES:
1374                 /* Wake events only supported on port A for dual fiber
1375                  * regardless of eeprom setting */
1376                 if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
1377                         adapter->eeprom_wol = 0;
1378                 break;
1379         case E1000_DEV_ID_82576_QUAD_COPPER:
1380                 /* if quad port adapter, disable WoL on all but port A */
1381                 if (global_quad_port_a != 0)
1382                         adapter->eeprom_wol = 0;
1383                 else
1384                         adapter->flags |= IGB_FLAG_QUAD_PORT_A;
1385                 /* Reset for multiple quad port adapters */
1386                 if (++global_quad_port_a == 4)
1387                         global_quad_port_a = 0;
1388                 break;
1389         }
1390
1391         /* initialize the wol settings based on the eeprom settings */
1392         adapter->wol = adapter->eeprom_wol;
1393         device_set_wakeup_enable(&adapter->pdev->dev, adapter->wol);
1394
1395         /* reset the hardware with the new settings */
1396         igb_reset(adapter);
1397
1398         /* let the f/w know that the h/w is now under the control of the
1399          * driver. */
1400         igb_get_hw_control(adapter);
1401
1402         /* tell the stack to leave us alone until igb_open() is called */
1403         netif_carrier_off(netdev);
1404         netif_tx_stop_all_queues(netdev);
1405
1406         strcpy(netdev->name, "eth%d");
1407         err = register_netdev(netdev);
1408         if (err)
1409                 goto err_register;
1410
1411 #ifdef CONFIG_PCI_IOV
1412         /* since iov functionality isn't critical to base device function we
1413          * can accept failure.  If it fails we don't allow iov to be enabled */
1414         if (hw->mac.type == e1000_82576) {
1415                 err = pci_enable_sriov(pdev, 0);
1416                 if (!err)
1417                         err = device_create_file(&netdev->dev,
1418                                                  &dev_attr_num_vfs);
1419                 if (err)
1420                         dev_err(&pdev->dev, "Failed to initialize IOV\n");
1421         }
1422
1423 #endif
1424 #ifdef CONFIG_IGB_DCA
1425         if (dca_add_requester(&pdev->dev) == 0) {
1426                 adapter->flags |= IGB_FLAG_DCA_ENABLED;
1427                 dev_info(&pdev->dev, "DCA enabled\n");
1428                 /* Always use CB2 mode, difference is masked
1429                  * in the CB driver. */
1430                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
1431                 igb_setup_dca(adapter);
1432         }
1433 #endif
1434
1435         /*
1436          * Initialize hardware timer: we keep it running just in case
1437          * that some program needs it later on.
1438          */
1439         memset(&adapter->cycles, 0, sizeof(adapter->cycles));
1440         adapter->cycles.read = igb_read_clock;
1441         adapter->cycles.mask = CLOCKSOURCE_MASK(64);
1442         adapter->cycles.mult = 1;
1443         adapter->cycles.shift = IGB_TSYNC_SHIFT;
1444         wr32(E1000_TIMINCA,
1445              (1<<24) |
1446              IGB_TSYNC_CYCLE_TIME_IN_NANOSECONDS * IGB_TSYNC_SCALE);
1447 #if 0
1448         /*
1449          * Avoid rollover while we initialize by resetting the time counter.
1450          */
1451         wr32(E1000_SYSTIML, 0x00000000);
1452         wr32(E1000_SYSTIMH, 0x00000000);
1453 #else
1454         /*
1455          * Set registers so that rollover occurs soon to test this.
1456          */
1457         wr32(E1000_SYSTIML, 0x00000000);
1458         wr32(E1000_SYSTIMH, 0xFF800000);
1459 #endif
1460         wrfl();
1461         timecounter_init(&adapter->clock,
1462                          &adapter->cycles,
1463                          ktime_to_ns(ktime_get_real()));
1464
1465         /*
1466          * Synchronize our NIC clock against system wall clock. NIC
1467          * time stamp reading requires ~3us per sample, each sample
1468          * was pretty stable even under load => only require 10
1469          * samples for each offset comparison.
1470          */
1471         memset(&adapter->compare, 0, sizeof(adapter->compare));
1472         adapter->compare.source = &adapter->clock;
1473         adapter->compare.target = ktime_get_real;
1474         adapter->compare.num_samples = 10;
1475         timecompare_update(&adapter->compare, 0);
1476
1477 #ifdef DEBUG
1478         {
1479                 char buffer[160];
1480                 printk(KERN_DEBUG
1481                         "igb: %s: hw %p initialized timer\n",
1482                         igb_get_time_str(adapter, buffer),
1483                         &adapter->hw);
1484         }
1485 #endif
1486
1487         dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
1488         /* print bus type/speed/width info */
1489         dev_info(&pdev->dev, "%s: (PCIe:%s:%s) %pM\n",
1490                  netdev->name,
1491                  ((hw->bus.speed == e1000_bus_speed_2500)
1492                   ? "2.5Gb/s" : "unknown"),
1493                  ((hw->bus.width == e1000_bus_width_pcie_x4)
1494                   ? "Width x4" : (hw->bus.width == e1000_bus_width_pcie_x1)
1495                   ? "Width x1" : "unknown"),
1496                  netdev->dev_addr);
1497
1498         igb_read_part_num(hw, &part_num);
1499         dev_info(&pdev->dev, "%s: PBA No: %06x-%03x\n", netdev->name,
1500                 (part_num >> 8), (part_num & 0xff));
1501
1502         dev_info(&pdev->dev,
1503                 "Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
1504                 adapter->msix_entries ? "MSI-X" :
1505                 (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
1506                 adapter->num_rx_queues, adapter->num_tx_queues);
1507
1508         return 0;
1509
1510 err_register:
1511         igb_release_hw_control(adapter);
1512 err_eeprom:
1513         if (!igb_check_reset_block(hw))
1514                 igb_reset_phy(hw);
1515
1516         if (hw->flash_address)
1517                 iounmap(hw->flash_address);
1518
1519         igb_free_queues(adapter);
1520 err_sw_init:
1521         iounmap(hw->hw_addr);
1522 err_ioremap:
1523         free_netdev(netdev);
1524 err_alloc_etherdev:
1525         pci_release_selected_regions(pdev, pci_select_bars(pdev,
1526                                      IORESOURCE_MEM));
1527 err_pci_reg:
1528 err_dma:
1529         pci_disable_device(pdev);
1530         return err;
1531 }
1532
1533 /**
1534  * igb_remove - Device Removal Routine
1535  * @pdev: PCI device information struct
1536  *
1537  * igb_remove is called by the PCI subsystem to alert the driver
1538  * that it should release a PCI device.  The could be caused by a
1539  * Hot-Plug event, or because the driver is going to be removed from
1540  * memory.
1541  **/
1542 static void __devexit igb_remove(struct pci_dev *pdev)
1543 {
1544         struct net_device *netdev = pci_get_drvdata(pdev);
1545         struct igb_adapter *adapter = netdev_priv(netdev);
1546         struct e1000_hw *hw = &adapter->hw;
1547         int err;
1548
1549         /* flush_scheduled work may reschedule our watchdog task, so
1550          * explicitly disable watchdog tasks from being rescheduled  */
1551         set_bit(__IGB_DOWN, &adapter->state);
1552         del_timer_sync(&adapter->watchdog_timer);
1553         del_timer_sync(&adapter->phy_info_timer);
1554
1555         flush_scheduled_work();
1556
1557 #ifdef CONFIG_IGB_DCA
1558         if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
1559                 dev_info(&pdev->dev, "DCA disabled\n");
1560                 dca_remove_requester(&pdev->dev);
1561                 adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
1562                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
1563         }
1564 #endif
1565
1566         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
1567          * would have already happened in close and is redundant. */
1568         igb_release_hw_control(adapter);
1569
1570         unregister_netdev(netdev);
1571
1572         if (!igb_check_reset_block(&adapter->hw))
1573                 igb_reset_phy(&adapter->hw);
1574
1575         igb_reset_interrupt_capability(adapter);
1576
1577         igb_free_queues(adapter);
1578
1579 #ifdef CONFIG_PCI_IOV
1580         /* reclaim resources allocated to VFs */
1581         if (adapter->vf_data) {
1582                 /* disable iov and allow time for transactions to clear */
1583                 pci_disable_sriov(pdev);
1584                 msleep(500);
1585
1586                 kfree(adapter->vf_data);
1587                 adapter->vf_data = NULL;
1588                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
1589                 msleep(100);
1590                 dev_info(&pdev->dev, "IOV Disabled\n");
1591         }
1592 #endif
1593         iounmap(hw->hw_addr);
1594         if (hw->flash_address)
1595                 iounmap(hw->flash_address);
1596         pci_release_selected_regions(pdev, pci_select_bars(pdev,
1597                                      IORESOURCE_MEM));
1598
1599         free_netdev(netdev);
1600
1601         err = pci_disable_pcie_error_reporting(pdev);
1602         if (err)
1603                 dev_err(&pdev->dev,
1604                         "pci_disable_pcie_error_reporting failed 0x%x\n", err);
1605
1606         pci_disable_device(pdev);
1607 }
1608
1609 /**
1610  * igb_sw_init - Initialize general software structures (struct igb_adapter)
1611  * @adapter: board private structure to initialize
1612  *
1613  * igb_sw_init initializes the Adapter private data structure.
1614  * Fields are initialized based on PCI device information and
1615  * OS network device settings (MTU size).
1616  **/
1617 static int __devinit igb_sw_init(struct igb_adapter *adapter)
1618 {
1619         struct e1000_hw *hw = &adapter->hw;
1620         struct net_device *netdev = adapter->netdev;
1621         struct pci_dev *pdev = adapter->pdev;
1622
1623         pci_read_config_word(pdev, PCI_COMMAND, &hw->bus.pci_cmd_word);
1624
1625         adapter->tx_ring_count = IGB_DEFAULT_TXD;
1626         adapter->rx_ring_count = IGB_DEFAULT_RXD;
1627         adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1628         adapter->rx_ps_hdr_size = 0; /* disable packet split */
1629         adapter->max_frame_size = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1630         adapter->min_frame_size = ETH_ZLEN + ETH_FCS_LEN;
1631
1632         /* This call may decrease the number of queues depending on
1633          * interrupt mode. */
1634         igb_set_interrupt_capability(adapter);
1635
1636         if (igb_alloc_queues(adapter)) {
1637                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
1638                 return -ENOMEM;
1639         }
1640
1641         /* Explicitly disable IRQ since the NIC can be in any state. */
1642         igb_irq_disable(adapter);
1643
1644         set_bit(__IGB_DOWN, &adapter->state);
1645         return 0;
1646 }
1647
1648 /**
1649  * igb_open - Called when a network interface is made active
1650  * @netdev: network interface device structure
1651  *
1652  * Returns 0 on success, negative value on failure
1653  *
1654  * The open entry point is called when a network interface is made
1655  * active by the system (IFF_UP).  At this point all resources needed
1656  * for transmit and receive operations are allocated, the interrupt
1657  * handler is registered with the OS, the watchdog timer is started,
1658  * and the stack is notified that the interface is ready.
1659  **/
1660 static int igb_open(struct net_device *netdev)
1661 {
1662         struct igb_adapter *adapter = netdev_priv(netdev);
1663         struct e1000_hw *hw = &adapter->hw;
1664         int err;
1665         int i;
1666
1667         /* disallow open during test */
1668         if (test_bit(__IGB_TESTING, &adapter->state))
1669                 return -EBUSY;
1670
1671         /* allocate transmit descriptors */
1672         err = igb_setup_all_tx_resources(adapter);
1673         if (err)
1674                 goto err_setup_tx;
1675
1676         /* allocate receive descriptors */
1677         err = igb_setup_all_rx_resources(adapter);
1678         if (err)
1679                 goto err_setup_rx;
1680
1681         /* e1000_power_up_phy(adapter); */
1682
1683         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
1684         if ((adapter->hw.mng_cookie.status &
1685              E1000_MNG_DHCP_COOKIE_STATUS_VLAN))
1686                 igb_update_mng_vlan(adapter);
1687
1688         /* before we allocate an interrupt, we must be ready to handle it.
1689          * Setting DEBUG_SHIRQ in the kernel makes it fire an interrupt
1690          * as soon as we call pci_request_irq, so we have to setup our
1691          * clean_rx handler before we do so.  */
1692         igb_configure(adapter);
1693
1694         igb_vmm_control(adapter);
1695         igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
1696         igb_set_vmolr(hw, adapter->vfs_allocated_count);
1697
1698         err = igb_request_irq(adapter);
1699         if (err)
1700                 goto err_req_irq;
1701
1702         /* From here on the code is the same as igb_up() */
1703         clear_bit(__IGB_DOWN, &adapter->state);
1704
1705         for (i = 0; i < adapter->num_rx_queues; i++)
1706                 napi_enable(&adapter->rx_ring[i].napi);
1707
1708         /* Clear any pending interrupts. */
1709         rd32(E1000_ICR);
1710
1711         igb_irq_enable(adapter);
1712
1713         netif_tx_start_all_queues(netdev);
1714
1715         /* Fire a link status change interrupt to start the watchdog. */
1716         wr32(E1000_ICS, E1000_ICS_LSC);
1717
1718         return 0;
1719
1720 err_req_irq:
1721         igb_release_hw_control(adapter);
1722         /* e1000_power_down_phy(adapter); */
1723         igb_free_all_rx_resources(adapter);
1724 err_setup_rx:
1725         igb_free_all_tx_resources(adapter);
1726 err_setup_tx:
1727         igb_reset(adapter);
1728
1729         return err;
1730 }
1731
1732 /**
1733  * igb_close - Disables a network interface
1734  * @netdev: network interface device structure
1735  *
1736  * Returns 0, this is not allowed to fail
1737  *
1738  * The close entry point is called when an interface is de-activated
1739  * by the OS.  The hardware is still under the driver's control, but
1740  * needs to be disabled.  A global MAC reset is issued to stop the
1741  * hardware, and all transmit and receive resources are freed.
1742  **/
1743 static int igb_close(struct net_device *netdev)
1744 {
1745         struct igb_adapter *adapter = netdev_priv(netdev);
1746
1747         WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
1748         igb_down(adapter);
1749
1750         igb_free_irq(adapter);
1751
1752         igb_free_all_tx_resources(adapter);
1753         igb_free_all_rx_resources(adapter);
1754
1755         /* kill manageability vlan ID if supported, but not if a vlan with
1756          * the same ID is registered on the host OS (let 8021q kill it) */
1757         if ((adapter->hw.mng_cookie.status &
1758                           E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
1759              !(adapter->vlgrp &&
1760                vlan_group_get_device(adapter->vlgrp, adapter->mng_vlan_id)))
1761                 igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
1762
1763         return 0;
1764 }
1765
1766 /**
1767  * igb_setup_tx_resources - allocate Tx resources (Descriptors)
1768  * @adapter: board private structure
1769  * @tx_ring: tx descriptor ring (for a specific queue) to setup
1770  *
1771  * Return 0 on success, negative on failure
1772  **/
1773 int igb_setup_tx_resources(struct igb_adapter *adapter,
1774                            struct igb_ring *tx_ring)
1775 {
1776         struct pci_dev *pdev = adapter->pdev;
1777         int size;
1778
1779         size = sizeof(struct igb_buffer) * tx_ring->count;
1780         tx_ring->buffer_info = vmalloc(size);
1781         if (!tx_ring->buffer_info)
1782                 goto err;
1783         memset(tx_ring->buffer_info, 0, size);
1784
1785         /* round up to nearest 4K */
1786         tx_ring->size = tx_ring->count * sizeof(union e1000_adv_tx_desc);
1787         tx_ring->size = ALIGN(tx_ring->size, 4096);
1788
1789         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
1790                                              &tx_ring->dma);
1791
1792         if (!tx_ring->desc)
1793                 goto err;
1794
1795         tx_ring->adapter = adapter;
1796         tx_ring->next_to_use = 0;
1797         tx_ring->next_to_clean = 0;
1798         return 0;
1799
1800 err:
1801         vfree(tx_ring->buffer_info);
1802         dev_err(&adapter->pdev->dev,
1803                 "Unable to allocate memory for the transmit descriptor ring\n");
1804         return -ENOMEM;
1805 }
1806
1807 /**
1808  * igb_setup_all_tx_resources - wrapper to allocate Tx resources
1809  *                                (Descriptors) for all queues
1810  * @adapter: board private structure
1811  *
1812  * Return 0 on success, negative on failure
1813  **/
1814 static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
1815 {
1816         int i, err = 0;
1817         int r_idx;
1818
1819         for (i = 0; i < adapter->num_tx_queues; i++) {
1820                 err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
1821                 if (err) {
1822                         dev_err(&adapter->pdev->dev,
1823                                 "Allocation for Tx Queue %u failed\n", i);
1824                         for (i--; i >= 0; i--)
1825                                 igb_free_tx_resources(&adapter->tx_ring[i]);
1826                         break;
1827                 }
1828         }
1829
1830         for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
1831                 r_idx = i % adapter->num_tx_queues;
1832                 adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
1833         }
1834         return err;
1835 }
1836
1837 /**
1838  * igb_configure_tx - Configure transmit Unit after Reset
1839  * @adapter: board private structure
1840  *
1841  * Configure the Tx unit of the MAC after a reset.
1842  **/
1843 static void igb_configure_tx(struct igb_adapter *adapter)
1844 {
1845         u64 tdba;
1846         struct e1000_hw *hw = &adapter->hw;
1847         u32 tctl;
1848         u32 txdctl, txctrl;
1849         int i, j;
1850
1851         for (i = 0; i < adapter->num_tx_queues; i++) {
1852                 struct igb_ring *ring = &adapter->tx_ring[i];
1853                 j = ring->reg_idx;
1854                 wr32(E1000_TDLEN(j),
1855                      ring->count * sizeof(union e1000_adv_tx_desc));
1856                 tdba = ring->dma;
1857                 wr32(E1000_TDBAL(j),
1858                      tdba & 0x00000000ffffffffULL);
1859                 wr32(E1000_TDBAH(j), tdba >> 32);
1860
1861                 ring->head = E1000_TDH(j);
1862                 ring->tail = E1000_TDT(j);
1863                 writel(0, hw->hw_addr + ring->tail);
1864                 writel(0, hw->hw_addr + ring->head);
1865                 txdctl = rd32(E1000_TXDCTL(j));
1866                 txdctl |= E1000_TXDCTL_QUEUE_ENABLE;
1867                 wr32(E1000_TXDCTL(j), txdctl);
1868
1869                 /* Turn off Relaxed Ordering on head write-backs.  The
1870                  * writebacks MUST be delivered in order or it will
1871                  * completely screw up our bookeeping.
1872                  */
1873                 txctrl = rd32(E1000_DCA_TXCTRL(j));
1874                 txctrl &= ~E1000_DCA_TXCTRL_TX_WB_RO_EN;
1875                 wr32(E1000_DCA_TXCTRL(j), txctrl);
1876         }
1877
1878         /* disable queue 0 to prevent tail bump w/o re-configuration */
1879         if (adapter->vfs_allocated_count)
1880                 wr32(E1000_TXDCTL(0), 0);
1881
1882         /* Program the Transmit Control Register */
1883         tctl = rd32(E1000_TCTL);
1884         tctl &= ~E1000_TCTL_CT;
1885         tctl |= E1000_TCTL_PSP | E1000_TCTL_RTLC |
1886                 (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
1887
1888         igb_config_collision_dist(hw);
1889
1890         /* Setup Transmit Descriptor Settings for eop descriptor */
1891         adapter->txd_cmd = E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS;
1892
1893         /* Enable transmits */
1894         tctl |= E1000_TCTL_EN;
1895
1896         wr32(E1000_TCTL, tctl);
1897 }
1898
1899 /**
1900  * igb_setup_rx_resources - allocate Rx resources (Descriptors)
1901  * @adapter: board private structure
1902  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
1903  *
1904  * Returns 0 on success, negative on failure
1905  **/
1906 int igb_setup_rx_resources(struct igb_adapter *adapter,
1907                            struct igb_ring *rx_ring)
1908 {
1909         struct pci_dev *pdev = adapter->pdev;
1910         int size, desc_len;
1911
1912         size = sizeof(struct igb_buffer) * rx_ring->count;
1913         rx_ring->buffer_info = vmalloc(size);
1914         if (!rx_ring->buffer_info)
1915                 goto err;
1916         memset(rx_ring->buffer_info, 0, size);
1917
1918         desc_len = sizeof(union e1000_adv_rx_desc);
1919
1920         /* Round up to nearest 4K */
1921         rx_ring->size = rx_ring->count * desc_len;
1922         rx_ring->size = ALIGN(rx_ring->size, 4096);
1923
1924         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size,
1925                                              &rx_ring->dma);
1926
1927         if (!rx_ring->desc)
1928                 goto err;
1929
1930         rx_ring->next_to_clean = 0;
1931         rx_ring->next_to_use = 0;
1932
1933         rx_ring->adapter = adapter;
1934
1935         return 0;
1936
1937 err:
1938         vfree(rx_ring->buffer_info);
1939         dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
1940                 "the receive descriptor ring\n");
1941         return -ENOMEM;
1942 }
1943
1944 /**
1945  * igb_setup_all_rx_resources - wrapper to allocate Rx resources
1946  *                                (Descriptors) for all queues
1947  * @adapter: board private structure
1948  *
1949  * Return 0 on success, negative on failure
1950  **/
1951 static int igb_setup_all_rx_resources(struct igb_adapter *adapter)
1952 {
1953         int i, err = 0;
1954
1955         for (i = 0; i < adapter->num_rx_queues; i++) {
1956                 err = igb_setup_rx_resources(adapter, &adapter->rx_ring[i]);
1957                 if (err) {
1958                         dev_err(&adapter->pdev->dev,
1959                                 "Allocation for Rx Queue %u failed\n", i);
1960                         for (i--; i >= 0; i--)
1961                                 igb_free_rx_resources(&adapter->rx_ring[i]);
1962                         break;
1963                 }
1964         }
1965
1966         return err;
1967 }
1968
1969 /**
1970  * igb_setup_rctl - configure the receive control registers
1971  * @adapter: Board private structure
1972  **/
1973 static void igb_setup_rctl(struct igb_adapter *adapter)
1974 {
1975         struct e1000_hw *hw = &adapter->hw;
1976         u32 rctl;
1977         u32 srrctl = 0;
1978         int i, j;
1979
1980         rctl = rd32(E1000_RCTL);
1981
1982         rctl &= ~(3 << E1000_RCTL_MO_SHIFT);
1983         rctl &= ~(E1000_RCTL_LBM_TCVR | E1000_RCTL_LBM_MAC);
1984
1985         rctl |= E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_RDMTS_HALF |
1986                 (hw->mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
1987
1988         /*
1989          * enable stripping of CRC. It's unlikely this will break BMC
1990          * redirection as it did with e1000. Newer features require
1991          * that the HW strips the CRC.
1992          */
1993         rctl |= E1000_RCTL_SECRC;
1994
1995         /*
1996          * disable store bad packets and clear size bits.
1997          */
1998         rctl &= ~(E1000_RCTL_SBP | E1000_RCTL_SZ_256);
1999
2000         /* enable LPE when to prevent packets larger than max_frame_size */
2001                 rctl |= E1000_RCTL_LPE;
2002
2003         /* Setup buffer sizes */
2004         switch (adapter->rx_buffer_len) {
2005         case IGB_RXBUFFER_256:
2006                 rctl |= E1000_RCTL_SZ_256;
2007                 break;
2008         case IGB_RXBUFFER_512:
2009                 rctl |= E1000_RCTL_SZ_512;
2010                 break;
2011         default:
2012                 srrctl = ALIGN(adapter->rx_buffer_len, 1024)
2013                          >> E1000_SRRCTL_BSIZEPKT_SHIFT;
2014                 break;
2015         }
2016
2017         /* 82575 and greater support packet-split where the protocol
2018          * header is placed in skb->data and the packet data is
2019          * placed in pages hanging off of skb_shinfo(skb)->nr_frags.
2020          * In the case of a non-split, skb->data is linearly filled,
2021          * followed by the page buffers.  Therefore, skb->data is
2022          * sized to hold the largest protocol header.
2023          */
2024         /* allocations using alloc_page take too long for regular MTU
2025          * so only enable packet split for jumbo frames */
2026         if (adapter->netdev->mtu > ETH_DATA_LEN) {
2027                 adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
2028                 srrctl |= adapter->rx_ps_hdr_size <<
2029                          E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
2030                 srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
2031         } else {
2032                 adapter->rx_ps_hdr_size = 0;
2033                 srrctl |= E1000_SRRCTL_DESCTYPE_ADV_ONEBUF;
2034         }
2035
2036         /* Attention!!!  For SR-IOV PF driver operations you must enable
2037          * queue drop for all VF and PF queues to prevent head of line blocking
2038          * if an un-trusted VF does not provide descriptors to hardware.
2039          */
2040         if (adapter->vfs_allocated_count) {
2041                 u32 vmolr;
2042
2043                 j = adapter->rx_ring[0].reg_idx;
2044
2045                 /* set all queue drop enable bits */
2046                 wr32(E1000_QDE, ALL_QUEUES);
2047                 srrctl |= E1000_SRRCTL_DROP_EN;
2048
2049                 /* disable queue 0 to prevent tail write w/o re-config */
2050                 wr32(E1000_RXDCTL(0), 0);
2051
2052                 vmolr = rd32(E1000_VMOLR(j));
2053                 if (rctl & E1000_RCTL_LPE)
2054                         vmolr |= E1000_VMOLR_LPE;
2055                 if (adapter->num_rx_queues > 0)
2056                         vmolr |= E1000_VMOLR_RSSE;
2057                 wr32(E1000_VMOLR(j), vmolr);
2058         }
2059
2060         for (i = 0; i < adapter->num_rx_queues; i++) {
2061                 j = adapter->rx_ring[i].reg_idx;
2062                 wr32(E1000_SRRCTL(j), srrctl);
2063         }
2064
2065         wr32(E1000_RCTL, rctl);
2066 }
2067
2068 /**
2069  * igb_rlpml_set - set maximum receive packet size
2070  * @adapter: board private structure
2071  *
2072  * Configure maximum receivable packet size.
2073  **/
2074 static void igb_rlpml_set(struct igb_adapter *adapter)
2075 {
2076         u32 max_frame_size = adapter->max_frame_size;
2077         struct e1000_hw *hw = &adapter->hw;
2078         u16 pf_id = adapter->vfs_allocated_count;
2079
2080         if (adapter->vlgrp)
2081                 max_frame_size += VLAN_TAG_SIZE;
2082
2083         /* if vfs are enabled we set RLPML to the largest possible request
2084          * size and set the VMOLR RLPML to the size we need */
2085         if (pf_id) {
2086                 igb_set_vf_rlpml(adapter, max_frame_size, pf_id);
2087                 max_frame_size = MAX_STD_JUMBO_FRAME_SIZE + VLAN_TAG_SIZE;
2088         }
2089
2090         wr32(E1000_RLPML, max_frame_size);
2091 }
2092
2093 /**
2094  * igb_configure_vt_default_pool - Configure VT default pool
2095  * @adapter: board private structure
2096  *
2097  * Configure the default pool
2098  **/
2099 static void igb_configure_vt_default_pool(struct igb_adapter *adapter)
2100 {
2101         struct e1000_hw *hw = &adapter->hw;
2102         u16 pf_id = adapter->vfs_allocated_count;
2103         u32 vtctl;
2104
2105         /* not in sr-iov mode - do nothing */
2106         if (!pf_id)
2107                 return;
2108
2109         vtctl = rd32(E1000_VT_CTL);
2110         vtctl &= ~(E1000_VT_CTL_DEFAULT_POOL_MASK |
2111                    E1000_VT_CTL_DISABLE_DEF_POOL);
2112         vtctl |= pf_id << E1000_VT_CTL_DEFAULT_POOL_SHIFT;
2113         wr32(E1000_VT_CTL, vtctl);
2114 }
2115
2116 /**
2117  * igb_configure_rx - Configure receive Unit after Reset
2118  * @adapter: board private structure
2119  *
2120  * Configure the Rx unit of the MAC after a reset.
2121  **/
2122 static void igb_configure_rx(struct igb_adapter *adapter)
2123 {
2124         u64 rdba;
2125         struct e1000_hw *hw = &adapter->hw;
2126         u32 rctl, rxcsum;
2127         u32 rxdctl;
2128         int i;
2129
2130         /* disable receives while setting up the descriptors */
2131         rctl = rd32(E1000_RCTL);
2132         wr32(E1000_RCTL, rctl & ~E1000_RCTL_EN);
2133         wrfl();
2134         mdelay(10);
2135
2136         if (adapter->itr_setting > 3)
2137                 wr32(E1000_ITR, adapter->itr);
2138
2139         /* Setup the HW Rx Head and Tail Descriptor Pointers and
2140          * the Base and Length of the Rx Descriptor Ring */
2141         for (i = 0; i < adapter->num_rx_queues; i++) {
2142                 struct igb_ring *ring = &adapter->rx_ring[i];
2143                 int j = ring->reg_idx;
2144                 rdba = ring->dma;
2145                 wr32(E1000_RDBAL(j),
2146                      rdba & 0x00000000ffffffffULL);
2147                 wr32(E1000_RDBAH(j), rdba >> 32);
2148                 wr32(E1000_RDLEN(j),
2149                      ring->count * sizeof(union e1000_adv_rx_desc));
2150
2151                 ring->head = E1000_RDH(j);
2152                 ring->tail = E1000_RDT(j);
2153                 writel(0, hw->hw_addr + ring->tail);
2154                 writel(0, hw->hw_addr + ring->head);
2155
2156                 rxdctl = rd32(E1000_RXDCTL(j));
2157                 rxdctl |= E1000_RXDCTL_QUEUE_ENABLE;
2158                 rxdctl &= 0xFFF00000;
2159                 rxdctl |= IGB_RX_PTHRESH;
2160                 rxdctl |= IGB_RX_HTHRESH << 8;
2161                 rxdctl |= IGB_RX_WTHRESH << 16;
2162                 wr32(E1000_RXDCTL(j), rxdctl);
2163         }
2164
2165         if (adapter->num_rx_queues > 1) {
2166                 u32 random[10];
2167                 u32 mrqc;
2168                 u32 j, shift;
2169                 union e1000_reta {
2170                         u32 dword;
2171                         u8  bytes[4];
2172                 } reta;
2173
2174                 get_random_bytes(&random[0], 40);
2175
2176                 if (hw->mac.type >= e1000_82576)
2177                         shift = 0;
2178                 else
2179                         shift = 6;
2180                 for (j = 0; j < (32 * 4); j++) {
2181                         reta.bytes[j & 3] =
2182                                 adapter->rx_ring[(j % adapter->num_rx_queues)].reg_idx << shift;
2183                         if ((j & 3) == 3)
2184                                 writel(reta.dword,
2185                                        hw->hw_addr + E1000_RETA(0) + (j & ~3));
2186                 }
2187                 if (adapter->vfs_allocated_count)
2188                         mrqc = E1000_MRQC_ENABLE_VMDQ_RSS_2Q;
2189                 else
2190                         mrqc = E1000_MRQC_ENABLE_RSS_4Q;
2191
2192                 /* Fill out hash function seeds */
2193                 for (j = 0; j < 10; j++)
2194                         array_wr32(E1000_RSSRK(0), j, random[j]);
2195
2196                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4 |
2197                          E1000_MRQC_RSS_FIELD_IPV4_TCP);
2198                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6 |
2199                          E1000_MRQC_RSS_FIELD_IPV6_TCP);
2200                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV4_UDP |
2201                          E1000_MRQC_RSS_FIELD_IPV6_UDP);
2202                 mrqc |= (E1000_MRQC_RSS_FIELD_IPV6_UDP_EX |
2203                          E1000_MRQC_RSS_FIELD_IPV6_TCP_EX);
2204
2205
2206                 wr32(E1000_MRQC, mrqc);
2207
2208                 /* Multiqueue and raw packet checksumming are mutually
2209                  * exclusive.  Note that this not the same as TCP/IP
2210                  * checksumming, which works fine. */
2211                 rxcsum = rd32(E1000_RXCSUM);
2212                 rxcsum |= E1000_RXCSUM_PCSD;
2213                 wr32(E1000_RXCSUM, rxcsum);
2214         } else {
2215                 /* Enable multi-queue for sr-iov */
2216                 if (adapter->vfs_allocated_count)
2217                         wr32(E1000_MRQC, E1000_MRQC_ENABLE_VMDQ);
2218                 /* Enable Receive Checksum Offload for TCP and UDP */
2219                 rxcsum = rd32(E1000_RXCSUM);
2220                 if (adapter->rx_csum)
2221                         rxcsum |= E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE;
2222                 else
2223                         rxcsum &= ~(E1000_RXCSUM_TUOFL | E1000_RXCSUM_IPPCSE);
2224
2225                 wr32(E1000_RXCSUM, rxcsum);
2226         }
2227
2228         /* Set the default pool for the PF's first queue */
2229         igb_configure_vt_default_pool(adapter);
2230
2231         igb_rlpml_set(adapter);
2232
2233         /* Enable Receives */
2234         wr32(E1000_RCTL, rctl);
2235 }
2236
2237 /**
2238  * igb_free_tx_resources - Free Tx Resources per Queue
2239  * @tx_ring: Tx descriptor ring for a specific queue
2240  *
2241  * Free all transmit software resources
2242  **/
2243 void igb_free_tx_resources(struct igb_ring *tx_ring)
2244 {
2245         struct pci_dev *pdev = tx_ring->adapter->pdev;
2246
2247         igb_clean_tx_ring(tx_ring);
2248
2249         vfree(tx_ring->buffer_info);
2250         tx_ring->buffer_info = NULL;
2251
2252         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2253
2254         tx_ring->desc = NULL;
2255 }
2256
2257 /**
2258  * igb_free_all_tx_resources - Free Tx Resources for All Queues
2259  * @adapter: board private structure
2260  *
2261  * Free all transmit software resources
2262  **/
2263 static void igb_free_all_tx_resources(struct igb_adapter *adapter)
2264 {
2265         int i;
2266
2267         for (i = 0; i < adapter->num_tx_queues; i++)
2268                 igb_free_tx_resources(&adapter->tx_ring[i]);
2269 }
2270
2271 static void igb_unmap_and_free_tx_resource(struct igb_adapter *adapter,
2272                                            struct igb_buffer *buffer_info)
2273 {
2274         if (buffer_info->dma) {
2275                 pci_unmap_page(adapter->pdev,
2276                                 buffer_info->dma,
2277                                 buffer_info->length,
2278                                 PCI_DMA_TODEVICE);
2279                 buffer_info->dma = 0;
2280         }
2281         if (buffer_info->skb) {
2282                 dev_kfree_skb_any(buffer_info->skb);
2283                 buffer_info->skb = NULL;
2284         }
2285         buffer_info->time_stamp = 0;
2286         buffer_info->next_to_watch = 0;
2287         /* buffer_info must be completely set up in the transmit path */
2288 }
2289
2290 /**
2291  * igb_clean_tx_ring - Free Tx Buffers
2292  * @tx_ring: ring to be cleaned
2293  **/
2294 static void igb_clean_tx_ring(struct igb_ring *tx_ring)
2295 {
2296         struct igb_adapter *adapter = tx_ring->adapter;
2297         struct igb_buffer *buffer_info;
2298         unsigned long size;
2299         unsigned int i;
2300
2301         if (!tx_ring->buffer_info)
2302                 return;
2303         /* Free all the Tx ring sk_buffs */
2304
2305         for (i = 0; i < tx_ring->count; i++) {
2306                 buffer_info = &tx_ring->buffer_info[i];
2307                 igb_unmap_and_free_tx_resource(adapter, buffer_info);
2308         }
2309
2310         size = sizeof(struct igb_buffer) * tx_ring->count;
2311         memset(tx_ring->buffer_info, 0, size);
2312
2313         /* Zero out the descriptor ring */
2314
2315         memset(tx_ring->desc, 0, tx_ring->size);
2316
2317         tx_ring->next_to_use = 0;
2318         tx_ring->next_to_clean = 0;
2319
2320         writel(0, adapter->hw.hw_addr + tx_ring->head);
2321         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2322 }
2323
2324 /**
2325  * igb_clean_all_tx_rings - Free Tx Buffers for all queues
2326  * @adapter: board private structure
2327  **/
2328 static void igb_clean_all_tx_rings(struct igb_adapter *adapter)
2329 {
2330         int i;
2331
2332         for (i = 0; i < adapter->num_tx_queues; i++)
2333                 igb_clean_tx_ring(&adapter->tx_ring[i]);
2334 }
2335
2336 /**
2337  * igb_free_rx_resources - Free Rx Resources
2338  * @rx_ring: ring to clean the resources from
2339  *
2340  * Free all receive software resources
2341  **/
2342 void igb_free_rx_resources(struct igb_ring *rx_ring)
2343 {
2344         struct pci_dev *pdev = rx_ring->adapter->pdev;
2345
2346         igb_clean_rx_ring(rx_ring);
2347
2348         vfree(rx_ring->buffer_info);
2349         rx_ring->buffer_info = NULL;
2350
2351         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2352
2353         rx_ring->desc = NULL;
2354 }
2355
2356 /**
2357  * igb_free_all_rx_resources - Free Rx Resources for All Queues
2358  * @adapter: board private structure
2359  *
2360  * Free all receive software resources
2361  **/
2362 static void igb_free_all_rx_resources(struct igb_adapter *adapter)
2363 {
2364         int i;
2365
2366         for (i = 0; i < adapter->num_rx_queues; i++)
2367                 igb_free_rx_resources(&adapter->rx_ring[i]);
2368 }
2369
2370 /**
2371  * igb_clean_rx_ring - Free Rx Buffers per Queue
2372  * @rx_ring: ring to free buffers from
2373  **/
2374 static void igb_clean_rx_ring(struct igb_ring *rx_ring)
2375 {
2376         struct igb_adapter *adapter = rx_ring->adapter;
2377         struct igb_buffer *buffer_info;
2378         struct pci_dev *pdev = adapter->pdev;
2379         unsigned long size;
2380         unsigned int i;
2381
2382         if (!rx_ring->buffer_info)
2383                 return;
2384         /* Free all the Rx ring sk_buffs */
2385         for (i = 0; i < rx_ring->count; i++) {
2386                 buffer_info = &rx_ring->buffer_info[i];
2387                 if (buffer_info->dma) {
2388                         if (adapter->rx_ps_hdr_size)
2389                                 pci_unmap_single(pdev, buffer_info->dma,
2390                                                  adapter->rx_ps_hdr_size,
2391                                                  PCI_DMA_FROMDEVICE);
2392                         else
2393                                 pci_unmap_single(pdev, buffer_info->dma,
2394                                                  adapter->rx_buffer_len,
2395                                                  PCI_DMA_FROMDEVICE);
2396                         buffer_info->dma = 0;
2397                 }
2398
2399                 if (buffer_info->skb) {
2400                         dev_kfree_skb(buffer_info->skb);
2401                         buffer_info->skb = NULL;
2402                 }
2403                 if (buffer_info->page) {
2404                         if (buffer_info->page_dma)
2405                                 pci_unmap_page(pdev, buffer_info->page_dma,
2406                                                PAGE_SIZE / 2,
2407                                                PCI_DMA_FROMDEVICE);
2408                         put_page(buffer_info->page);
2409                         buffer_info->page = NULL;
2410                         buffer_info->page_dma = 0;
2411                         buffer_info->page_offset = 0;
2412                 }
2413         }
2414
2415         size = sizeof(struct igb_buffer) * rx_ring->count;
2416         memset(rx_ring->buffer_info, 0, size);
2417
2418         /* Zero out the descriptor ring */
2419         memset(rx_ring->desc, 0, rx_ring->size);
2420
2421         rx_ring->next_to_clean = 0;
2422         rx_ring->next_to_use = 0;
2423
2424         writel(0, adapter->hw.hw_addr + rx_ring->head);
2425         writel(0, adapter->hw.hw_addr + rx_ring->tail);
2426 }
2427
2428 /**
2429  * igb_clean_all_rx_rings - Free Rx Buffers for all queues
2430  * @adapter: board private structure
2431  **/
2432 static void igb_clean_all_rx_rings(struct igb_adapter *adapter)
2433 {
2434         int i;
2435
2436         for (i = 0; i < adapter->num_rx_queues; i++)
2437                 igb_clean_rx_ring(&adapter->rx_ring[i]);
2438 }
2439
2440 /**
2441  * igb_set_mac - Change the Ethernet Address of the NIC
2442  * @netdev: network interface device structure
2443  * @p: pointer to an address structure
2444  *
2445  * Returns 0 on success, negative on failure
2446  **/
2447 static int igb_set_mac(struct net_device *netdev, void *p)
2448 {
2449         struct igb_adapter *adapter = netdev_priv(netdev);
2450         struct e1000_hw *hw = &adapter->hw;
2451         struct sockaddr *addr = p;
2452
2453         if (!is_valid_ether_addr(addr->sa_data))
2454                 return -EADDRNOTAVAIL;
2455
2456         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2457         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
2458
2459         hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
2460
2461         igb_set_rah_pool(hw, adapter->vfs_allocated_count, 0);
2462
2463         return 0;
2464 }
2465
2466 /**
2467  * igb_set_multi - Multicast and Promiscuous mode set
2468  * @netdev: network interface device structure
2469  *
2470  * The set_multi entry point is called whenever the multicast address
2471  * list or the network interface flags are updated.  This routine is
2472  * responsible for configuring the hardware for proper multicast,
2473  * promiscuous mode, and all-multi behavior.
2474  **/
2475 static void igb_set_multi(struct net_device *netdev)
2476 {
2477         struct igb_adapter *adapter = netdev_priv(netdev);
2478         struct e1000_hw *hw = &adapter->hw;
2479         struct e1000_mac_info *mac = &hw->mac;
2480         struct dev_mc_list *mc_ptr;
2481         u8  *mta_list;
2482         u32 rctl;
2483         int i;
2484
2485         /* Check for Promiscuous and All Multicast modes */
2486
2487         rctl = rd32(E1000_RCTL);
2488
2489         if (netdev->flags & IFF_PROMISC) {
2490                 rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
2491                 rctl &= ~E1000_RCTL_VFE;
2492         } else {
2493                 if (netdev->flags & IFF_ALLMULTI) {
2494                         rctl |= E1000_RCTL_MPE;
2495                         rctl &= ~E1000_RCTL_UPE;
2496                 } else
2497                         rctl &= ~(E1000_RCTL_UPE | E1000_RCTL_MPE);
2498                 rctl |= E1000_RCTL_VFE;
2499         }
2500         wr32(E1000_RCTL, rctl);
2501
2502         if (!netdev->mc_count) {
2503                 /* nothing to program, so clear mc list */
2504                 igb_update_mc_addr_list(hw, NULL, 0, 1,
2505                                         mac->rar_entry_count);
2506                 return;
2507         }
2508
2509         mta_list = kzalloc(netdev->mc_count * 6, GFP_ATOMIC);
2510         if (!mta_list)
2511                 return;
2512
2513         /* The shared function expects a packed array of only addresses. */
2514         mc_ptr = netdev->mc_list;
2515
2516         for (i = 0; i < netdev->mc_count; i++) {
2517                 if (!mc_ptr)
2518                         break;
2519                 memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
2520                 mc_ptr = mc_ptr->next;
2521         }
2522         igb_update_mc_addr_list(hw, mta_list, i,
2523                                 adapter->vfs_allocated_count + 1,
2524                                 mac->rar_entry_count);
2525
2526         igb_set_mc_list_pools(adapter, i, mac->rar_entry_count);
2527         igb_restore_vf_multicasts(adapter);
2528
2529         kfree(mta_list);
2530 }
2531
2532 /* Need to wait a few seconds after link up to get diagnostic information from
2533  * the phy */
2534 static void igb_update_phy_info(unsigned long data)
2535 {
2536         struct igb_adapter *adapter = (struct igb_adapter *) data;
2537         igb_get_phy_info(&adapter->hw);
2538 }
2539
2540 /**
2541  * igb_has_link - check shared code for link and determine up/down
2542  * @adapter: pointer to driver private info
2543  **/
2544 static bool igb_has_link(struct igb_adapter *adapter)
2545 {
2546         struct e1000_hw *hw = &adapter->hw;
2547         bool link_active = false;
2548         s32 ret_val = 0;
2549
2550         /* get_link_status is set on LSC (link status) interrupt or
2551          * rx sequence error interrupt.  get_link_status will stay
2552          * false until the e1000_check_for_link establishes link
2553          * for copper adapters ONLY
2554          */
2555         switch (hw->phy.media_type) {
2556         case e1000_media_type_copper:
2557                 if (hw->mac.get_link_status) {
2558                         ret_val = hw->mac.ops.check_for_link(hw);
2559                         link_active = !hw->mac.get_link_status;
2560                 } else {
2561                         link_active = true;
2562                 }
2563                 break;
2564         case e1000_media_type_fiber:
2565                 ret_val = hw->mac.ops.check_for_link(hw);
2566                 link_active = !!(rd32(E1000_STATUS) & E1000_STATUS_LU);
2567                 break;
2568         case e1000_media_type_internal_serdes:
2569                 ret_val = hw->mac.ops.check_for_link(hw);
2570                 link_active = hw->mac.serdes_has_link;
2571                 break;
2572         default:
2573         case e1000_media_type_unknown:
2574                 break;
2575         }
2576
2577         return link_active;
2578 }
2579
2580 /**
2581  * igb_watchdog - Timer Call-back
2582  * @data: pointer to adapter cast into an unsigned long
2583  **/
2584 static void igb_watchdog(unsigned long data)
2585 {
2586         struct igb_adapter *adapter = (struct igb_adapter *)data;
2587         /* Do the rest outside of interrupt context */
2588         schedule_work(&adapter->watchdog_task);
2589 }
2590
2591 static void igb_watchdog_task(struct work_struct *work)
2592 {
2593         struct igb_adapter *adapter = container_of(work,
2594                                         struct igb_adapter, watchdog_task);
2595         struct e1000_hw *hw = &adapter->hw;
2596         struct net_device *netdev = adapter->netdev;
2597         struct igb_ring *tx_ring = adapter->tx_ring;
2598         u32 link;
2599         u32 eics = 0;
2600         int i;
2601
2602         link = igb_has_link(adapter);
2603         if ((netif_carrier_ok(netdev)) && link)
2604                 goto link_up;
2605
2606         if (link) {
2607                 if (!netif_carrier_ok(netdev)) {
2608                         u32 ctrl;
2609                         hw->mac.ops.get_speed_and_duplex(&adapter->hw,
2610                                                    &adapter->link_speed,
2611                                                    &adapter->link_duplex);
2612
2613                         ctrl = rd32(E1000_CTRL);
2614                         /* Links status message must follow this format */
2615                         printk(KERN_INFO "igb: %s NIC Link is Up %d Mbps %s, "
2616                                  "Flow Control: %s\n",
2617                                  netdev->name,
2618                                  adapter->link_speed,
2619                                  adapter->link_duplex == FULL_DUPLEX ?
2620                                  "Full Duplex" : "Half Duplex",
2621                                  ((ctrl & E1000_CTRL_TFCE) && (ctrl &
2622                                  E1000_CTRL_RFCE)) ? "RX/TX" : ((ctrl &
2623                                  E1000_CTRL_RFCE) ? "RX" : ((ctrl &
2624                                  E1000_CTRL_TFCE) ? "TX" : "None")));
2625
2626                         /* tweak tx_queue_len according to speed/duplex and
2627                          * adjust the timeout factor */
2628                         netdev->tx_queue_len = adapter->tx_queue_len;
2629                         adapter->tx_timeout_factor = 1;
2630                         switch (adapter->link_speed) {
2631                         case SPEED_10:
2632                                 netdev->tx_queue_len = 10;
2633                                 adapter->tx_timeout_factor = 14;
2634                                 break;
2635                         case SPEED_100:
2636                                 netdev->tx_queue_len = 100;
2637                                 /* maybe add some timeout factor ? */
2638                                 break;
2639                         }
2640
2641                         netif_carrier_on(netdev);
2642                         netif_tx_wake_all_queues(netdev);
2643
2644                         igb_ping_all_vfs(adapter);
2645
2646                         /* link state has changed, schedule phy info update */
2647                         if (!test_bit(__IGB_DOWN, &adapter->state))
2648                                 mod_timer(&adapter->phy_info_timer,
2649                                           round_jiffies(jiffies + 2 * HZ));
2650                 }
2651         } else {
2652                 if (netif_carrier_ok(netdev)) {
2653                         adapter->link_speed = 0;
2654                         adapter->link_duplex = 0;
2655                         /* Links status message must follow this format */
2656                         printk(KERN_INFO "igb: %s NIC Link is Down\n",
2657                                netdev->name);
2658                         netif_carrier_off(netdev);
2659                         netif_tx_stop_all_queues(netdev);
2660
2661                         igb_ping_all_vfs(adapter);
2662
2663                         /* link state has changed, schedule phy info update */
2664                         if (!test_bit(__IGB_DOWN, &adapter->state))
2665                                 mod_timer(&adapter->phy_info_timer,
2666                                           round_jiffies(jiffies + 2 * HZ));
2667                 }
2668         }
2669
2670 link_up:
2671         igb_update_stats(adapter);
2672
2673         hw->mac.tx_packet_delta = adapter->stats.tpt - adapter->tpt_old;
2674         adapter->tpt_old = adapter->stats.tpt;
2675         hw->mac.collision_delta = adapter->stats.colc - adapter->colc_old;
2676         adapter->colc_old = adapter->stats.colc;
2677
2678         adapter->gorc = adapter->stats.gorc - adapter->gorc_old;
2679         adapter->gorc_old = adapter->stats.gorc;
2680         adapter->gotc = adapter->stats.gotc - adapter->gotc_old;
2681         adapter->gotc_old = adapter->stats.gotc;
2682
2683         igb_update_adaptive(&adapter->hw);
2684
2685         if (!netif_carrier_ok(netdev)) {
2686                 if (IGB_DESC_UNUSED(tx_ring) + 1 < tx_ring->count) {
2687                         /* We've lost link, so the controller stops DMA,
2688                          * but we've got queued Tx work that's never going
2689                          * to get done, so reset controller to flush Tx.
2690                          * (Do the reset outside of interrupt context). */
2691                         adapter->tx_timeout_count++;
2692                         schedule_work(&adapter->reset_task);
2693                 }
2694         }
2695
2696         /* Cause software interrupt to ensure rx ring is cleaned */
2697         if (adapter->msix_entries) {
2698                 for (i = 0; i < adapter->num_rx_queues; i++)
2699                         eics |= adapter->rx_ring[i].eims_value;
2700                 wr32(E1000_EICS, eics);
2701         } else {
2702                 wr32(E1000_ICS, E1000_ICS_RXDMT0);
2703         }
2704
2705         /* Force detection of hung controller every watchdog period */
2706         tx_ring->detect_tx_hung = true;
2707
2708         /* Reset the timer */
2709         if (!test_bit(__IGB_DOWN, &adapter->state))
2710                 mod_timer(&adapter->watchdog_timer,
2711                           round_jiffies(jiffies + 2 * HZ));
2712 }
2713
2714 enum latency_range {
2715         lowest_latency = 0,
2716         low_latency = 1,
2717         bulk_latency = 2,
2718         latency_invalid = 255
2719 };
2720
2721
2722 /**
2723  * igb_update_ring_itr - update the dynamic ITR value based on packet size
2724  *
2725  *      Stores a new ITR value based on strictly on packet size.  This
2726  *      algorithm is less sophisticated than that used in igb_update_itr,
2727  *      due to the difficulty of synchronizing statistics across multiple
2728  *      receive rings.  The divisors and thresholds used by this fuction
2729  *      were determined based on theoretical maximum wire speed and testing
2730  *      data, in order to minimize response time while increasing bulk
2731  *      throughput.
2732  *      This functionality is controlled by the InterruptThrottleRate module
2733  *      parameter (see igb_param.c)
2734  *      NOTE:  This function is called only when operating in a multiqueue
2735  *             receive environment.
2736  * @rx_ring: pointer to ring
2737  **/
2738 static void igb_update_ring_itr(struct igb_ring *rx_ring)
2739 {
2740         int new_val = rx_ring->itr_val;
2741         int avg_wire_size = 0;
2742         struct igb_adapter *adapter = rx_ring->adapter;
2743
2744         if (!rx_ring->total_packets)
2745                 goto clear_counts; /* no packets, so don't do anything */
2746
2747         /* For non-gigabit speeds, just fix the interrupt rate at 4000
2748          * ints/sec - ITR timer value of 120 ticks.
2749          */
2750         if (adapter->link_speed != SPEED_1000) {
2751                 new_val = 120;
2752                 goto set_itr_val;
2753         }
2754         avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
2755
2756         /* Add 24 bytes to size to account for CRC, preamble, and gap */
2757         avg_wire_size += 24;
2758
2759         /* Don't starve jumbo frames */
2760         avg_wire_size = min(avg_wire_size, 3000);
2761
2762         /* Give a little boost to mid-size frames */
2763         if ((avg_wire_size > 300) && (avg_wire_size < 1200))
2764                 new_val = avg_wire_size / 3;
2765         else
2766                 new_val = avg_wire_size / 2;
2767
2768 set_itr_val:
2769         if (new_val != rx_ring->itr_val) {
2770                 rx_ring->itr_val = new_val;
2771                 rx_ring->set_itr = 1;
2772         }
2773 clear_counts:
2774         rx_ring->total_bytes = 0;
2775         rx_ring->total_packets = 0;
2776 }
2777
2778 /**
2779  * igb_update_itr - update the dynamic ITR value based on statistics
2780  *      Stores a new ITR value based on packets and byte
2781  *      counts during the last interrupt.  The advantage of per interrupt
2782  *      computation is faster updates and more accurate ITR for the current
2783  *      traffic pattern.  Constants in this function were computed
2784  *      based on theoretical maximum wire speed and thresholds were set based
2785  *      on testing data as well as attempting to minimize response time
2786  *      while increasing bulk throughput.
2787  *      this functionality is controlled by the InterruptThrottleRate module
2788  *      parameter (see igb_param.c)
2789  *      NOTE:  These calculations are only valid when operating in a single-
2790  *             queue environment.
2791  * @adapter: pointer to adapter
2792  * @itr_setting: current adapter->itr
2793  * @packets: the number of packets during this measurement interval
2794  * @bytes: the number of bytes during this measurement interval
2795  **/
2796 static unsigned int igb_update_itr(struct igb_adapter *adapter, u16 itr_setting,
2797                                    int packets, int bytes)
2798 {
2799         unsigned int retval = itr_setting;
2800
2801         if (packets == 0)
2802                 goto update_itr_done;
2803
2804         switch (itr_setting) {
2805         case lowest_latency:
2806                 /* handle TSO and jumbo frames */
2807                 if (bytes/packets > 8000)
2808                         retval = bulk_latency;
2809                 else if ((packets < 5) && (bytes > 512))
2810                         retval = low_latency;
2811                 break;
2812         case low_latency:  /* 50 usec aka 20000 ints/s */
2813                 if (bytes > 10000) {
2814                         /* this if handles the TSO accounting */
2815                         if (bytes/packets > 8000) {
2816                                 retval = bulk_latency;
2817                         } else if ((packets < 10) || ((bytes/packets) > 1200)) {
2818                                 retval = bulk_latency;
2819                         } else if ((packets > 35)) {
2820                                 retval = lowest_latency;
2821                         }
2822                 } else if (bytes/packets > 2000) {
2823                         retval = bulk_latency;
2824                 } else if (packets <= 2 && bytes < 512) {
2825                         retval = lowest_latency;
2826                 }
2827                 break;
2828         case bulk_latency: /* 250 usec aka 4000 ints/s */
2829                 if (bytes > 25000) {
2830                         if (packets > 35)
2831                                 retval = low_latency;
2832                 } else if (bytes < 1500) {
2833                         retval = low_latency;
2834                 }
2835                 break;
2836         }
2837
2838 update_itr_done:
2839         return retval;
2840 }
2841
2842 static void igb_set_itr(struct igb_adapter *adapter)
2843 {
2844         u16 current_itr;
2845         u32 new_itr = adapter->itr;
2846
2847         /* for non-gigabit speeds, just fix the interrupt rate at 4000 */
2848         if (adapter->link_speed != SPEED_1000) {
2849                 current_itr = 0;
2850                 new_itr = 4000;
2851                 goto set_itr_now;
2852         }
2853
2854         adapter->rx_itr = igb_update_itr(adapter,
2855                                     adapter->rx_itr,
2856                                     adapter->rx_ring->total_packets,
2857                                     adapter->rx_ring->total_bytes);
2858
2859         if (adapter->rx_ring->buddy) {
2860                 adapter->tx_itr = igb_update_itr(adapter,
2861                                             adapter->tx_itr,
2862                                             adapter->tx_ring->total_packets,
2863                                             adapter->tx_ring->total_bytes);
2864                 current_itr = max(adapter->rx_itr, adapter->tx_itr);
2865         } else {
2866                 current_itr = adapter->rx_itr;
2867         }
2868
2869         /* conservative mode (itr 3) eliminates the lowest_latency setting */
2870         if (adapter->itr_setting == 3 && current_itr == lowest_latency)
2871                 current_itr = low_latency;
2872
2873         switch (current_itr) {
2874         /* counts and packets in update_itr are dependent on these numbers */
2875         case lowest_latency:
2876                 new_itr = 70000;
2877                 break;
2878         case low_latency:
2879                 new_itr = 20000; /* aka hwitr = ~200 */
2880                 break;
2881         case bulk_latency:
2882                 new_itr = 4000;
2883                 break;
2884         default:
2885                 break;
2886         }
2887
2888 set_itr_now:
2889         adapter->rx_ring->total_bytes = 0;
2890         adapter->rx_ring->total_packets = 0;
2891         if (adapter->rx_ring->buddy) {
2892                 adapter->rx_ring->buddy->total_bytes = 0;
2893                 adapter->rx_ring->buddy->total_packets = 0;
2894         }
2895
2896         if (new_itr != adapter->itr) {
2897                 /* this attempts to bias the interrupt rate towards Bulk
2898                  * by adding intermediate steps when interrupt rate is
2899                  * increasing */
2900                 new_itr = new_itr > adapter->itr ?
2901                              min(adapter->itr + (new_itr >> 2), new_itr) :
2902                              new_itr;
2903                 /* Don't write the value here; it resets the adapter's
2904                  * internal timer, and causes us to delay far longer than
2905                  * we should between interrupts.  Instead, we write the ITR
2906                  * value at the beginning of the next interrupt so the timing
2907                  * ends up being correct.
2908                  */
2909                 adapter->itr = new_itr;
2910                 adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
2911                 adapter->rx_ring->set_itr = 1;
2912         }
2913
2914         return;
2915 }
2916
2917
2918 #define IGB_TX_FLAGS_CSUM               0x00000001
2919 #define IGB_TX_FLAGS_VLAN               0x00000002
2920 #define IGB_TX_FLAGS_TSO                0x00000004
2921 #define IGB_TX_FLAGS_IPV4               0x00000008
2922 #define IGB_TX_FLAGS_TSTAMP             0x00000010
2923 #define IGB_TX_FLAGS_VLAN_MASK  0xffff0000
2924 #define IGB_TX_FLAGS_VLAN_SHIFT 16
2925
2926 static inline int igb_tso_adv(struct igb_adapter *adapter,
2927                               struct igb_ring *tx_ring,
2928                               struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2929 {
2930         struct e1000_adv_tx_context_desc *context_desc;
2931         unsigned int i;
2932         int err;
2933         struct igb_buffer *buffer_info;
2934         u32 info = 0, tu_cmd = 0;
2935         u32 mss_l4len_idx, l4len;
2936         *hdr_len = 0;
2937
2938         if (skb_header_cloned(skb)) {
2939                 err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2940                 if (err)
2941                         return err;
2942         }
2943
2944         l4len = tcp_hdrlen(skb);
2945         *hdr_len += l4len;
2946
2947         if (skb->protocol == htons(ETH_P_IP)) {
2948                 struct iphdr *iph = ip_hdr(skb);
2949                 iph->tot_len = 0;
2950                 iph->check = 0;
2951                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2952                                                          iph->daddr, 0,
2953                                                          IPPROTO_TCP,
2954                                                          0);
2955         } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
2956                 ipv6_hdr(skb)->payload_len = 0;
2957                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2958                                                        &ipv6_hdr(skb)->daddr,
2959                                                        0, IPPROTO_TCP, 0);
2960         }
2961
2962         i = tx_ring->next_to_use;
2963
2964         buffer_info = &tx_ring->buffer_info[i];
2965         context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
2966         /* VLAN MACLEN IPLEN */
2967         if (tx_flags & IGB_TX_FLAGS_VLAN)
2968                 info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
2969         info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
2970         *hdr_len += skb_network_offset(skb);
2971         info |= skb_network_header_len(skb);
2972         *hdr_len += skb_network_header_len(skb);
2973         context_desc->vlan_macip_lens = cpu_to_le32(info);
2974
2975         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2976         tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
2977
2978         if (skb->protocol == htons(ETH_P_IP))
2979                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
2980         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
2981
2982         context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
2983
2984         /* MSS L4LEN IDX */
2985         mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
2986         mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
2987
2988         /* For 82575, context index must be unique per ring. */
2989         if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
2990                 mss_l4len_idx |= tx_ring->queue_index << 4;
2991
2992         context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2993         context_desc->seqnum_seed = 0;
2994
2995         buffer_info->time_stamp = jiffies;
2996         buffer_info->next_to_watch = i;
2997         buffer_info->dma = 0;
2998         i++;
2999         if (i == tx_ring->count)
3000                 i = 0;
3001
3002         tx_ring->next_to_use = i;
3003
3004         return true;
3005 }
3006
3007 static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
3008                                         struct igb_ring *tx_ring,
3009                                         struct sk_buff *skb, u32 tx_flags)
3010 {
3011         struct e1000_adv_tx_context_desc *context_desc;
3012         unsigned int i;
3013         struct igb_buffer *buffer_info;
3014         u32 info = 0, tu_cmd = 0;
3015
3016         if ((skb->ip_summed == CHECKSUM_PARTIAL) ||
3017             (tx_flags & IGB_TX_FLAGS_VLAN)) {
3018                 i = tx_ring->next_to_use;
3019                 buffer_info = &tx_ring->buffer_info[i];
3020                 context_desc = E1000_TX_CTXTDESC_ADV(*tx_ring, i);
3021
3022                 if (tx_flags & IGB_TX_FLAGS_VLAN)
3023                         info |= (tx_flags & IGB_TX_FLAGS_VLAN_MASK);
3024                 info |= (skb_network_offset(skb) << E1000_ADVTXD_MACLEN_SHIFT);
3025                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3026                         info |= skb_network_header_len(skb);
3027
3028                 context_desc->vlan_macip_lens = cpu_to_le32(info);
3029
3030                 tu_cmd |= (E1000_TXD_CMD_DEXT | E1000_ADVTXD_DTYP_CTXT);
3031
3032                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3033                         switch (skb->protocol) {
3034                         case cpu_to_be16(ETH_P_IP):
3035                                 tu_cmd |= E1000_ADVTXD_TUCMD_IPV4;
3036                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3037                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3038                                 break;
3039                         case cpu_to_be16(ETH_P_IPV6):
3040                                 /* XXX what about other V6 headers?? */
3041                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3042                                         tu_cmd |= E1000_ADVTXD_TUCMD_L4T_TCP;
3043                                 break;
3044                         default:
3045                                 if (unlikely(net_ratelimit()))
3046                                         dev_warn(&adapter->pdev->dev,
3047                                             "partial checksum but proto=%x!\n",
3048                                             skb->protocol);
3049                                 break;
3050                         }
3051                 }
3052
3053                 context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
3054                 context_desc->seqnum_seed = 0;
3055                 if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
3056                         context_desc->mss_l4len_idx =
3057                                 cpu_to_le32(tx_ring->queue_index << 4);
3058                 else
3059                         context_desc->mss_l4len_idx = 0;
3060
3061                 buffer_info->time_stamp = jiffies;
3062                 buffer_info->next_to_watch = i;
3063                 buffer_info->dma = 0;
3064
3065                 i++;
3066                 if (i == tx_ring->count)
3067                         i = 0;
3068                 tx_ring->next_to_use = i;
3069
3070                 return true;
3071         }
3072         return false;
3073 }
3074
3075 #define IGB_MAX_TXD_PWR 16
3076 #define IGB_MAX_DATA_PER_TXD    (1<<IGB_MAX_TXD_PWR)
3077
3078 static inline int igb_tx_map_adv(struct igb_adapter *adapter,
3079                                  struct igb_ring *tx_ring, struct sk_buff *skb,
3080                                  unsigned int first)
3081 {
3082         struct igb_buffer *buffer_info;
3083         unsigned int len = skb_headlen(skb);
3084         unsigned int count = 0, i;
3085         unsigned int f;
3086
3087         i = tx_ring->next_to_use;
3088
3089         buffer_info = &tx_ring->buffer_info[i];
3090         BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3091         buffer_info->length = len;
3092         /* set time_stamp *before* dma to help avoid a possible race */
3093         buffer_info->time_stamp = jiffies;
3094         buffer_info->next_to_watch = i;
3095         buffer_info->dma = pci_map_single(adapter->pdev, skb->data, len,
3096                                           PCI_DMA_TODEVICE);
3097         count++;
3098         i++;
3099         if (i == tx_ring->count)
3100                 i = 0;
3101
3102         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
3103                 struct skb_frag_struct *frag;
3104
3105                 frag = &skb_shinfo(skb)->frags[f];
3106                 len = frag->size;
3107
3108                 buffer_info = &tx_ring->buffer_info[i];
3109                 BUG_ON(len >= IGB_MAX_DATA_PER_TXD);
3110                 buffer_info->length = len;
3111                 buffer_info->time_stamp = jiffies;
3112                 buffer_info->next_to_watch = i;
3113                 buffer_info->dma = pci_map_page(adapter->pdev,
3114                                                 frag->page,
3115                                                 frag->page_offset,
3116                                                 len,
3117                                                 PCI_DMA_TODEVICE);
3118
3119                 count++;
3120                 i++;
3121                 if (i == tx_ring->count)
3122                         i = 0;
3123         }
3124
3125         i = ((i == 0) ? tx_ring->count - 1 : i - 1);
3126         tx_ring->buffer_info[i].skb = skb;
3127         tx_ring->buffer_info[first].next_to_watch = i;
3128
3129         return count;
3130 }
3131
3132 static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
3133                                     struct igb_ring *tx_ring,
3134                                     int tx_flags, int count, u32 paylen,
3135                                     u8 hdr_len)
3136 {
3137         union e1000_adv_tx_desc *tx_desc = NULL;
3138         struct igb_buffer *buffer_info;
3139         u32 olinfo_status = 0, cmd_type_len;
3140         unsigned int i;
3141
3142         cmd_type_len = (E1000_ADVTXD_DTYP_DATA | E1000_ADVTXD_DCMD_IFCS |
3143                         E1000_ADVTXD_DCMD_DEXT);
3144
3145         if (tx_flags & IGB_TX_FLAGS_VLAN)
3146                 cmd_type_len |= E1000_ADVTXD_DCMD_VLE;
3147
3148         if (tx_flags & IGB_TX_FLAGS_TSTAMP)
3149                 cmd_type_len |= E1000_ADVTXD_MAC_TSTAMP;
3150
3151         if (tx_flags & IGB_TX_FLAGS_TSO) {
3152                 cmd_type_len |= E1000_ADVTXD_DCMD_TSE;
3153
3154                 /* insert tcp checksum */
3155                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3156
3157                 /* insert ip checksum */
3158                 if (tx_flags & IGB_TX_FLAGS_IPV4)
3159                         olinfo_status |= E1000_TXD_POPTS_IXSM << 8;
3160
3161         } else if (tx_flags & IGB_TX_FLAGS_CSUM) {
3162                 olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
3163         }
3164
3165         if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
3166             (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
3167                          IGB_TX_FLAGS_VLAN)))
3168                 olinfo_status |= tx_ring->queue_index << 4;
3169
3170         olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
3171
3172         i = tx_ring->next_to_use;
3173         while (count--) {
3174                 buffer_info = &tx_ring->buffer_info[i];
3175                 tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
3176                 tx_desc->read.buffer_addr = cpu_to_le64(buffer_info->dma);
3177                 tx_desc->read.cmd_type_len =
3178                         cpu_to_le32(cmd_type_len | buffer_info->length);
3179                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3180                 i++;
3181                 if (i == tx_ring->count)
3182                         i = 0;
3183         }
3184
3185         tx_desc->read.cmd_type_len |= cpu_to_le32(adapter->txd_cmd);
3186         /* Force memory writes to complete before letting h/w
3187          * know there are new descriptors to fetch.  (Only
3188          * applicable for weak-ordered memory model archs,
3189          * such as IA-64). */
3190         wmb();
3191
3192         tx_ring->next_to_use = i;
3193         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3194         /* we need this if more than one processor can write to our tail
3195          * at a time, it syncronizes IO on IA64/Altix systems */
3196         mmiowb();
3197 }
3198
3199 static int __igb_maybe_stop_tx(struct net_device *netdev,
3200                                struct igb_ring *tx_ring, int size)
3201 {
3202         struct igb_adapter *adapter = netdev_priv(netdev);
3203
3204         netif_stop_subqueue(netdev, tx_ring->queue_index);
3205
3206         /* Herbert's original patch had:
3207          *  smp_mb__after_netif_stop_queue();
3208          * but since that doesn't exist yet, just open code it. */
3209         smp_mb();
3210
3211         /* We need to check again in a case another CPU has just
3212          * made room available. */
3213         if (IGB_DESC_UNUSED(tx_ring) < size)
3214                 return -EBUSY;
3215
3216         /* A reprieve! */
3217         netif_wake_subqueue(netdev, tx_ring->queue_index);
3218         ++adapter->restart_queue;
3219         return 0;
3220 }
3221
3222 static int igb_maybe_stop_tx(struct net_device *netdev,
3223                              struct igb_ring *tx_ring, int size)
3224 {
3225         if (IGB_DESC_UNUSED(tx_ring) >= size)
3226                 return 0;
3227         return __igb_maybe_stop_tx(netdev, tx_ring, size);
3228 }
3229
3230 static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
3231                                    struct net_device *netdev,
3232                                    struct igb_ring *tx_ring)
3233 {
3234         struct igb_adapter *adapter = netdev_priv(netdev);
3235         unsigned int first;
3236         unsigned int tx_flags = 0;
3237         u8 hdr_len = 0;
3238         int tso = 0;
3239         union skb_shared_tx *shtx;
3240
3241         if (test_bit(__IGB_DOWN, &adapter->state)) {
3242                 dev_kfree_skb_any(skb);
3243                 return NETDEV_TX_OK;
3244         }
3245
3246         if (skb->len <= 0) {
3247                 dev_kfree_skb_any(skb);
3248                 return NETDEV_TX_OK;
3249         }
3250
3251         /* need: 1 descriptor per page,
3252          *       + 2 desc gap to keep tail from touching head,
3253          *       + 1 desc for skb->data,
3254          *       + 1 desc for context descriptor,
3255          * otherwise try next time */
3256         if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
3257                 /* this is a hard error */
3258                 return NETDEV_TX_BUSY;
3259         }
3260
3261         /*
3262          * TODO: check that there currently is no other packet with
3263          * time stamping in the queue
3264          *
3265          * When doing time stamping, keep the connection to the socket
3266          * a while longer: it is still needed by skb_hwtstamp_tx(),
3267          * called either in igb_tx_hwtstamp() or by our caller when
3268          * doing software time stamping.
3269          */
3270         shtx = skb_tx(skb);
3271         if (unlikely(shtx->hardware)) {
3272                 shtx->in_progress = 1;
3273                 tx_flags |= IGB_TX_FLAGS_TSTAMP;
3274         }
3275
3276         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3277                 tx_flags |= IGB_TX_FLAGS_VLAN;
3278                 tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
3279         }
3280
3281         if (skb->protocol == htons(ETH_P_IP))
3282                 tx_flags |= IGB_TX_FLAGS_IPV4;
3283
3284         first = tx_ring->next_to_use;
3285         tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
3286                                               &hdr_len) : 0;
3287
3288         if (tso < 0) {
3289                 dev_kfree_skb_any(skb);
3290                 return NETDEV_TX_OK;
3291         }
3292
3293         if (tso)
3294                 tx_flags |= IGB_TX_FLAGS_TSO;
3295         else if (igb_tx_csum_adv(adapter, tx_ring, skb, tx_flags) &&
3296                  (skb->ip_summed == CHECKSUM_PARTIAL))
3297                 tx_flags |= IGB_TX_FLAGS_CSUM;
3298
3299         igb_tx_queue_adv(adapter, tx_ring, tx_flags,
3300                          igb_tx_map_adv(adapter, tx_ring, skb, first),
3301                          skb->len, hdr_len);
3302
3303         netdev->trans_start = jiffies;
3304
3305         /* Make sure there is space in the ring for the next send. */
3306         igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
3307
3308         return NETDEV_TX_OK;
3309 }
3310
3311 static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
3312 {
3313         struct igb_adapter *adapter = netdev_priv(netdev);
3314         struct igb_ring *tx_ring;
3315
3316         int r_idx = 0;
3317         r_idx = skb->queue_mapping & (IGB_ABS_MAX_TX_QUEUES - 1);
3318         tx_ring = adapter->multi_tx_table[r_idx];
3319
3320         /* This goes back to the question of how to logically map a tx queue
3321          * to a flow.  Right now, performance is impacted slightly negatively
3322          * if using multiple tx queues.  If the stack breaks away from a
3323          * single qdisc implementation, we can look at this again. */
3324         return (igb_xmit_frame_ring_adv(skb, netdev, tx_ring));
3325 }
3326
3327 /**
3328  * igb_tx_timeout - Respond to a Tx Hang
3329  * @netdev: network interface device structure
3330  **/
3331 static void igb_tx_timeout(struct net_device *netdev)
3332 {
3333         struct igb_adapter *adapter = netdev_priv(netdev);
3334         struct e1000_hw *hw = &adapter->hw;
3335
3336         /* Do the reset outside of interrupt context */
3337         adapter->tx_timeout_count++;
3338         schedule_work(&adapter->reset_task);
3339         wr32(E1000_EICS,
3340              (adapter->eims_enable_mask & ~adapter->eims_other));
3341 }
3342
3343 static void igb_reset_task(struct work_struct *work)
3344 {
3345         struct igb_adapter *adapter;
3346         adapter = container_of(work, struct igb_adapter, reset_task);
3347
3348         igb_reinit_locked(adapter);
3349 }
3350
3351 /**
3352  * igb_get_stats - Get System Network Statistics
3353  * @netdev: network interface device structure
3354  *
3355  * Returns the address of the device statistics structure.
3356  * The statistics are actually updated from the timer callback.
3357  **/
3358 static struct net_device_stats *igb_get_stats(struct net_device *netdev)
3359 {
3360         struct igb_adapter *adapter = netdev_priv(netdev);
3361
3362         /* only return the current stats */
3363         return &adapter->net_stats;
3364 }
3365
3366 /**
3367  * igb_change_mtu - Change the Maximum Transfer Unit
3368  * @netdev: network interface device structure
3369  * @new_mtu: new value for maximum frame size
3370  *
3371  * Returns 0 on success, negative on failure
3372  **/
3373 static int igb_change_mtu(struct net_device *netdev, int new_mtu)
3374 {
3375         struct igb_adapter *adapter = netdev_priv(netdev);
3376         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3377
3378         if ((max_frame < ETH_ZLEN + ETH_FCS_LEN) ||
3379             (max_frame > MAX_JUMBO_FRAME_SIZE)) {
3380                 dev_err(&adapter->pdev->dev, "Invalid MTU setting\n");
3381                 return -EINVAL;
3382         }
3383
3384         if (max_frame > MAX_STD_JUMBO_FRAME_SIZE) {
3385                 dev_err(&adapter->pdev->dev, "MTU > 9216 not supported.\n");
3386                 return -EINVAL;
3387         }
3388
3389         while (test_and_set_bit(__IGB_RESETTING, &adapter->state))
3390                 msleep(1);
3391
3392         /* igb_down has a dependency on max_frame_size */
3393         adapter->max_frame_size = max_frame;
3394         if (netif_running(netdev))
3395                 igb_down(adapter);
3396
3397         /* NOTE: netdev_alloc_skb reserves 16 bytes, and typically NET_IP_ALIGN
3398          * means we reserve 2 more, this pushes us to allocate from the next
3399          * larger slab size.
3400          * i.e. RXBUFFER_2048 --> size-4096 slab
3401          */
3402
3403         if (max_frame <= IGB_RXBUFFER_256)
3404                 adapter->rx_buffer_len = IGB_RXBUFFER_256;
3405         else if (max_frame <= IGB_RXBUFFER_512)
3406                 adapter->rx_buffer_len = IGB_RXBUFFER_512;
3407         else if (max_frame <= IGB_RXBUFFER_1024)
3408                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3409         else if (max_frame <= IGB_RXBUFFER_2048)
3410                 adapter->rx_buffer_len = IGB_RXBUFFER_2048;
3411         else
3412 #if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
3413                 adapter->rx_buffer_len = IGB_RXBUFFER_16384;
3414 #else
3415                 adapter->rx_buffer_len = PAGE_SIZE / 2;
3416 #endif
3417
3418         /* if sr-iov is enabled we need to force buffer size to 1K or larger */
3419         if (adapter->vfs_allocated_count &&
3420             (adapter->rx_buffer_len < IGB_RXBUFFER_1024))
3421                 adapter->rx_buffer_len = IGB_RXBUFFER_1024;
3422
3423         /* adjust allocation if LPE protects us, and we aren't using SBP */
3424         if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
3425              (max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
3426                 adapter->rx_buffer_len = MAXIMUM_ETHERNET_VLAN_SIZE;
3427
3428         dev_info(&adapter->pdev->dev, "changing MTU from %d to %d\n",
3429                  netdev->mtu, new_mtu);
3430         netdev->mtu = new_mtu;
3431
3432         if (netif_running(netdev))
3433                 igb_up(adapter);
3434         else
3435                 igb_reset(adapter);
3436
3437         clear_bit(__IGB_RESETTING, &adapter->state);
3438
3439         return 0;
3440 }
3441
3442 /**
3443  * igb_update_stats - Update the board statistics counters
3444  * @adapter: board private structure
3445  **/
3446
3447 void igb_update_stats(struct igb_adapter *adapter)
3448 {
3449         struct e1000_hw *hw = &adapter->hw;
3450         struct pci_dev *pdev = adapter->pdev;
3451         u16 phy_tmp;
3452
3453 #define PHY_IDLE_ERROR_COUNT_MASK 0x00FF
3454
3455         /*
3456          * Prevent stats update while adapter is being reset, or if the pci
3457          * connection is down.
3458          */
3459         if (adapter->link_speed == 0)
3460                 return;
3461         if (pci_channel_offline(pdev))
3462                 return;
3463
3464         adapter->stats.crcerrs += rd32(E1000_CRCERRS);
3465         adapter->stats.gprc += rd32(E1000_GPRC);
3466         adapter->stats.gorc += rd32(E1000_GORCL);
3467         rd32(E1000_GORCH); /* clear GORCL */
3468         adapter->stats.bprc += rd32(E1000_BPRC);
3469         adapter->stats.mprc += rd32(E1000_MPRC);
3470         adapter->stats.roc += rd32(E1000_ROC);
3471
3472         adapter->stats.prc64 += rd32(E1000_PRC64);
3473         adapter->stats.prc127 += rd32(E1000_PRC127);
3474         adapter->stats.prc255 += rd32(E1000_PRC255);
3475         adapter->stats.prc511 += rd32(E1000_PRC511);
3476         adapter->stats.prc1023 += rd32(E1000_PRC1023);
3477         adapter->stats.prc1522 += rd32(E1000_PRC1522);
3478         adapter->stats.symerrs += rd32(E1000_SYMERRS);
3479         adapter->stats.sec += rd32(E1000_SEC);
3480
3481         adapter->stats.mpc += rd32(E1000_MPC);
3482         adapter->stats.scc += rd32(E1000_SCC);
3483         adapter->stats.ecol += rd32(E1000_ECOL);
3484         adapter->stats.mcc += rd32(E1000_MCC);
3485         adapter->stats.latecol += rd32(E1000_LATECOL);
3486         adapter->stats.dc += rd32(E1000_DC);
3487         adapter->stats.rlec += rd32(E1000_RLEC);
3488         adapter->stats.xonrxc += rd32(E1000_XONRXC);
3489         adapter->stats.xontxc += rd32(E1000_XONTXC);
3490         adapter->stats.xoffrxc += rd32(E1000_XOFFRXC);
3491         adapter->stats.xofftxc += rd32(E1000_XOFFTXC);
3492         adapter->stats.fcruc += rd32(E1000_FCRUC);
3493         adapter->stats.gptc += rd32(E1000_GPTC);
3494         adapter->stats.gotc += rd32(E1000_GOTCL);
3495         rd32(E1000_GOTCH); /* clear GOTCL */
3496         adapter->stats.rnbc += rd32(E1000_RNBC);
3497         adapter->stats.ruc += rd32(E1000_RUC);
3498         adapter->stats.rfc += rd32(E1000_RFC);
3499         adapter->stats.rjc += rd32(E1000_RJC);
3500         adapter->stats.tor += rd32(E1000_TORH);
3501         adapter->stats.tot += rd32(E1000_TOTH);
3502         adapter->stats.tpr += rd32(E1000_TPR);
3503
3504         adapter->stats.ptc64 += rd32(E1000_PTC64);
3505         adapter->stats.ptc127 += rd32(E1000_PTC127);
3506         adapter->stats.ptc255 += rd32(E1000_PTC255);
3507         adapter->stats.ptc511 += rd32(E1000_PTC511);
3508         adapter->stats.ptc1023 += rd32(E1000_PTC1023);
3509         adapter->stats.ptc1522 += rd32(E1000_PTC1522);
3510
3511         adapter->stats.mptc += rd32(E1000_MPTC);
3512         adapter->stats.bptc += rd32(E1000_BPTC);
3513
3514         /* used for adaptive IFS */
3515
3516         hw->mac.tx_packet_delta = rd32(E1000_TPT);
3517         adapter->stats.tpt += hw->mac.tx_packet_delta;
3518         hw->mac.collision_delta = rd32(E1000_COLC);
3519         adapter->stats.colc += hw->mac.collision_delta;
3520
3521         adapter->stats.algnerrc += rd32(E1000_ALGNERRC);
3522         adapter->stats.rxerrc += rd32(E1000_RXERRC);
3523         adapter->stats.tncrs += rd32(E1000_TNCRS);
3524         adapter->stats.tsctc += rd32(E1000_TSCTC);
3525         adapter->stats.tsctfc += rd32(E1000_TSCTFC);
3526
3527         adapter->stats.iac += rd32(E1000_IAC);
3528         adapter->stats.icrxoc += rd32(E1000_ICRXOC);
3529         adapter->stats.icrxptc += rd32(E1000_ICRXPTC);
3530         adapter->stats.icrxatc += rd32(E1000_ICRXATC);
3531         adapter->stats.ictxptc += rd32(E1000_ICTXPTC);
3532         adapter->stats.ictxatc += rd32(E1000_ICTXATC);
3533         adapter->stats.ictxqec += rd32(E1000_ICTXQEC);
3534         adapter->stats.ictxqmtc += rd32(E1000_ICTXQMTC);
3535         adapter->stats.icrxdmtc += rd32(E1000_ICRXDMTC);
3536
3537         /* Fill out the OS statistics structure */
3538         adapter->net_stats.multicast = adapter->stats.mprc;
3539         adapter->net_stats.collisions = adapter->stats.colc;
3540
3541         /* Rx Errors */
3542
3543         /* RLEC on some newer hardware can be incorrect so build
3544         * our own version based on RUC and ROC */
3545         adapter->net_stats.rx_errors = adapter->stats.rxerrc +
3546                 adapter->stats.crcerrs + adapter->stats.algnerrc +
3547                 adapter->stats.ruc + adapter->stats.roc +
3548                 adapter->stats.cexterr;
3549         adapter->net_stats.rx_length_errors = adapter->stats.ruc +
3550                                               adapter->stats.roc;
3551         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3552         adapter->net_stats.rx_frame_errors = adapter->stats.algnerrc;
3553         adapter->net_stats.rx_missed_errors = adapter->stats.mpc;
3554
3555         /* Tx Errors */
3556         adapter->net_stats.tx_errors = adapter->stats.ecol +
3557                                        adapter->stats.latecol;
3558         adapter->net_stats.tx_aborted_errors = adapter->stats.ecol;
3559         adapter->net_stats.tx_window_errors = adapter->stats.latecol;
3560         adapter->net_stats.tx_carrier_errors = adapter->stats.tncrs;
3561
3562         /* Tx Dropped needs to be maintained elsewhere */
3563
3564         /* Phy Stats */
3565         if (hw->phy.media_type == e1000_media_type_copper) {
3566                 if ((adapter->link_speed == SPEED_1000) &&
3567                    (!igb_read_phy_reg(hw, PHY_1000T_STATUS, &phy_tmp))) {
3568                         phy_tmp &= PHY_IDLE_ERROR_COUNT_MASK;
3569                         adapter->phy_stats.idle_errors += phy_tmp;
3570                 }
3571         }
3572
3573         /* Management Stats */
3574         adapter->stats.mgptc += rd32(E1000_MGTPTC);
3575         adapter->stats.mgprc += rd32(E1000_MGTPRC);
3576         adapter->stats.mgpdc += rd32(E1000_MGTPDC);
3577 }
3578
3579 static irqreturn_t igb_msix_other(int irq, void *data)
3580 {
3581         struct net_device *netdev = data;
3582         struct igb_adapter *adapter = netdev_priv(netdev);
3583         struct e1000_hw *hw = &adapter->hw;
3584         u32 icr = rd32(E1000_ICR);
3585
3586         /* reading ICR causes bit 31 of EICR to be cleared */
3587
3588         if(icr & E1000_ICR_DOUTSYNC) {
3589                 /* HW is reporting DMA is out of sync */
3590                 adapter->stats.doosync++;
3591         }
3592
3593         /* Check for a mailbox event */
3594         if (icr & E1000_ICR_VMMB)
3595                 igb_msg_task(adapter);
3596
3597         if (icr & E1000_ICR_LSC) {
3598                 hw->mac.get_link_status = 1;
3599                 /* guard against interrupt when we're going down */
3600                 if (!test_bit(__IGB_DOWN, &adapter->state))
3601                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
3602         }
3603
3604         wr32(E1000_IMS, E1000_IMS_LSC | E1000_IMS_DOUTSYNC | E1000_IMS_VMMB);
3605         wr32(E1000_EIMS, adapter->eims_other);
3606
3607         return IRQ_HANDLED;
3608 }
3609
3610 static irqreturn_t igb_msix_tx(int irq, void *data)
3611 {
3612         struct igb_ring *tx_ring = data;
3613         struct igb_adapter *adapter = tx_ring->adapter;
3614         struct e1000_hw *hw = &adapter->hw;
3615
3616 #ifdef CONFIG_IGB_DCA
3617         if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3618                 igb_update_tx_dca(tx_ring);
3619 #endif
3620
3621         tx_ring->total_bytes = 0;
3622         tx_ring->total_packets = 0;
3623
3624         /* auto mask will automatically reenable the interrupt when we write
3625          * EICS */
3626         if (!igb_clean_tx_irq(tx_ring))
3627                 /* Ring was not completely cleaned, so fire another interrupt */
3628                 wr32(E1000_EICS, tx_ring->eims_value);
3629         else
3630                 wr32(E1000_EIMS, tx_ring->eims_value);
3631
3632         return IRQ_HANDLED;
3633 }
3634
3635 static void igb_write_itr(struct igb_ring *ring)
3636 {
3637         struct e1000_hw *hw = &ring->adapter->hw;
3638         if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
3639                 switch (hw->mac.type) {
3640                 case e1000_82576:
3641                         wr32(ring->itr_register, ring->itr_val |
3642                              0x80000000);
3643                         break;
3644                 default:
3645                         wr32(ring->itr_register, ring->itr_val |
3646                              (ring->itr_val << 16));
3647                         break;
3648                 }
3649                 ring->set_itr = 0;
3650         }
3651 }
3652
3653 static irqreturn_t igb_msix_rx(int irq, void *data)
3654 {
3655         struct igb_ring *rx_ring = data;
3656
3657         /* Write the ITR value calculated at the end of the
3658          * previous interrupt.
3659          */
3660
3661         igb_write_itr(rx_ring);
3662
3663         if (napi_schedule_prep(&rx_ring->napi))
3664                 __napi_schedule(&rx_ring->napi);
3665
3666 #ifdef CONFIG_IGB_DCA
3667         if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
3668                 igb_update_rx_dca(rx_ring);
3669 #endif
3670                 return IRQ_HANDLED;
3671 }
3672
3673 #ifdef CONFIG_IGB_DCA
3674 static void igb_update_rx_dca(struct igb_ring *rx_ring)
3675 {
3676         u32 dca_rxctrl;
3677         struct igb_adapter *adapter = rx_ring->adapter;
3678         struct e1000_hw *hw = &adapter->hw;
3679         int cpu = get_cpu();
3680         int q = rx_ring->reg_idx;
3681
3682         if (rx_ring->cpu != cpu) {
3683                 dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
3684                 if (hw->mac.type == e1000_82576) {
3685                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
3686                         dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3687                                       E1000_DCA_RXCTRL_CPUID_SHIFT;
3688                 } else {
3689                         dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
3690                         dca_rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3691                 }
3692                 dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
3693                 dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
3694                 dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
3695                 wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
3696                 rx_ring->cpu = cpu;
3697         }
3698         put_cpu();
3699 }
3700
3701 static void igb_update_tx_dca(struct igb_ring *tx_ring)
3702 {
3703         u32 dca_txctrl;
3704         struct igb_adapter *adapter = tx_ring->adapter;
3705         struct e1000_hw *hw = &adapter->hw;
3706         int cpu = get_cpu();
3707         int q = tx_ring->reg_idx;
3708
3709         if (tx_ring->cpu != cpu) {
3710                 dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
3711                 if (hw->mac.type == e1000_82576) {
3712                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
3713                         dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu) <<
3714                                       E1000_DCA_TXCTRL_CPUID_SHIFT;
3715                 } else {
3716                         dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
3717                         dca_txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
3718                 }
3719                 dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
3720                 wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
3721                 tx_ring->cpu = cpu;
3722         }
3723         put_cpu();
3724 }
3725
3726 static void igb_setup_dca(struct igb_adapter *adapter)
3727 {
3728         int i;
3729
3730         if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
3731                 return;
3732
3733         for (i = 0; i < adapter->num_tx_queues; i++) {
3734                 adapter->tx_ring[i].cpu = -1;
3735                 igb_update_tx_dca(&adapter->tx_ring[i]);
3736         }
3737         for (i = 0; i < adapter->num_rx_queues; i++) {
3738                 adapter->rx_ring[i].cpu = -1;
3739                 igb_update_rx_dca(&adapter->rx_ring[i]);
3740         }
3741 }
3742
3743 static int __igb_notify_dca(struct device *dev, void *data)
3744 {
3745         struct net_device *netdev = dev_get_drvdata(dev);
3746         struct igb_adapter *adapter = netdev_priv(netdev);
3747         struct e1000_hw *hw = &adapter->hw;
3748         unsigned long event = *(unsigned long *)data;
3749
3750         switch (event) {
3751         case DCA_PROVIDER_ADD:
3752                 /* if already enabled, don't do it again */
3753                 if (adapter->flags & IGB_FLAG_DCA_ENABLED)
3754                         break;
3755                 /* Always use CB2 mode, difference is masked
3756                  * in the CB driver. */
3757                 wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_CB2);
3758                 if (dca_add_requester(dev) == 0) {
3759                         adapter->flags |= IGB_FLAG_DCA_ENABLED;
3760                         dev_info(&adapter->pdev->dev, "DCA enabled\n");
3761                         igb_setup_dca(adapter);
3762                         break;
3763                 }
3764                 /* Fall Through since DCA is disabled. */
3765         case DCA_PROVIDER_REMOVE:
3766                 if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
3767                         /* without this a class_device is left
3768                          * hanging around in the sysfs model */
3769                         dca_remove_requester(dev);
3770                         dev_info(&adapter->pdev->dev, "DCA disabled\n");
3771                         adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
3772                         wr32(E1000_DCA_CTRL, E1000_DCA_CTRL_DCA_MODE_DISABLE);
3773                 }
3774                 break;
3775         }
3776
3777         return 0;
3778 }
3779
3780 static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
3781                           void *p)
3782 {
3783         int ret_val;
3784
3785         ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
3786                                          __igb_notify_dca);
3787
3788         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
3789 }
3790 #endif /* CONFIG_IGB_DCA */
3791
3792 static void igb_ping_all_vfs(struct igb_adapter *adapter)
3793 {
3794         struct e1000_hw *hw = &adapter->hw;
3795         u32 ping;
3796         int i;
3797
3798         for (i = 0 ; i < adapter->vfs_allocated_count; i++) {
3799                 ping = E1000_PF_CONTROL_MSG;
3800                 if (adapter->vf_data[i].clear_to_send)
3801                         ping |= E1000_VT_MSGTYPE_CTS;
3802                 igb_write_mbx(hw, &ping, 1, i);
3803         }
3804 }
3805
3806 static int igb_set_vf_multicasts(struct igb_adapter *adapter,
3807                                   u32 *msgbuf, u32 vf)
3808 {
3809         int n = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3810         u16 *hash_list = (u16 *)&msgbuf[1];
3811         struct vf_data_storage *vf_data = &adapter->vf_data[vf];
3812         int i;
3813
3814         /* only up to 30 hash values supported */
3815         if (n > 30)
3816                 n = 30;
3817
3818         /* salt away the number of multi cast addresses assigned
3819          * to this VF for later use to restore when the PF multi cast
3820          * list changes
3821          */
3822         vf_data->num_vf_mc_hashes = n;
3823
3824         /* VFs are limited to using the MTA hash table for their multicast
3825          * addresses */
3826         for (i = 0; i < n; i++)
3827                 vf_data->vf_mc_hashes[i] = hash_list[i];;
3828
3829         /* Flush and reset the mta with the new values */
3830         igb_set_multi(adapter->netdev);
3831
3832         return 0;
3833 }
3834
3835 static void igb_restore_vf_multicasts(struct igb_adapter *adapter)
3836 {
3837         struct e1000_hw *hw = &adapter->hw;
3838         struct vf_data_storage *vf_data;
3839         int i, j;
3840
3841         for (i = 0; i < adapter->vfs_allocated_count; i++) {
3842                 vf_data = &adapter->vf_data[i];
3843                 for (j = 0; j < vf_data->num_vf_mc_hashes; j++)
3844                         igb_mta_set(hw, vf_data->vf_mc_hashes[j]);
3845         }
3846 }
3847
3848 static void igb_clear_vf_vfta(struct igb_adapter *adapter, u32 vf)
3849 {
3850         struct e1000_hw *hw = &adapter->hw;
3851         u32 pool_mask, reg, vid;
3852         int i;
3853
3854         pool_mask = 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3855
3856         /* Find the vlan filter for this id */
3857         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3858                 reg = rd32(E1000_VLVF(i));
3859
3860                 /* remove the vf from the pool */
3861                 reg &= ~pool_mask;
3862
3863                 /* if pool is empty then remove entry from vfta */
3864                 if (!(reg & E1000_VLVF_POOLSEL_MASK) &&
3865                     (reg & E1000_VLVF_VLANID_ENABLE)) {
3866                         reg = 0;
3867                         vid = reg & E1000_VLVF_VLANID_MASK;
3868                         igb_vfta_set(hw, vid, false);
3869                 }
3870
3871                 wr32(E1000_VLVF(i), reg);
3872         }
3873 }
3874
3875 static s32 igb_vlvf_set(struct igb_adapter *adapter, u32 vid, bool add, u32 vf)
3876 {
3877         struct e1000_hw *hw = &adapter->hw;
3878         u32 reg, i;
3879
3880         /* It is an error to call this function when VFs are not enabled */
3881         if (!adapter->vfs_allocated_count)
3882                 return -1;
3883
3884         /* Find the vlan filter for this id */
3885         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3886                 reg = rd32(E1000_VLVF(i));
3887                 if ((reg & E1000_VLVF_VLANID_ENABLE) &&
3888                     vid == (reg & E1000_VLVF_VLANID_MASK))
3889                         break;
3890         }
3891
3892         if (add) {
3893                 if (i == E1000_VLVF_ARRAY_SIZE) {
3894                         /* Did not find a matching VLAN ID entry that was
3895                          * enabled.  Search for a free filter entry, i.e.
3896                          * one without the enable bit set
3897                          */
3898                         for (i = 0; i < E1000_VLVF_ARRAY_SIZE; i++) {
3899                                 reg = rd32(E1000_VLVF(i));
3900                                 if (!(reg & E1000_VLVF_VLANID_ENABLE))
3901                                         break;
3902                         }
3903                 }
3904                 if (i < E1000_VLVF_ARRAY_SIZE) {
3905                         /* Found an enabled/available entry */
3906                         reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT + vf);
3907
3908                         /* if !enabled we need to set this up in vfta */
3909                         if (!(reg & E1000_VLVF_VLANID_ENABLE)) {
3910                                 /* add VID to filter table, if bit already set
3911                                  * PF must have added it outside of table */
3912                                 if (igb_vfta_set(hw, vid, true))
3913                                         reg |= 1 << (E1000_VLVF_POOLSEL_SHIFT +
3914                                                 adapter->vfs_allocated_count);
3915                                 reg |= E1000_VLVF_VLANID_ENABLE;
3916                         }
3917                         reg &= ~E1000_VLVF_VLANID_MASK;
3918                         reg |= vid;
3919
3920                         wr32(E1000_VLVF(i), reg);
3921                         return 0;
3922                 }
3923         } else {
3924                 if (i < E1000_VLVF_ARRAY_SIZE) {
3925                         /* remove vf from the pool */
3926                         reg &= ~(1 << (E1000_VLVF_POOLSEL_SHIFT + vf));
3927                         /* if pool is empty then remove entry from vfta */
3928                         if (!(reg & E1000_VLVF_POOLSEL_MASK)) {
3929                                 reg = 0;
3930                                 igb_vfta_set(hw, vid, false);
3931                         }
3932                         wr32(E1000_VLVF(i), reg);
3933                         return 0;
3934                 }
3935         }
3936         return -1;
3937 }
3938
3939 static int igb_set_vf_vlan(struct igb_adapter *adapter, u32 *msgbuf, u32 vf)
3940 {
3941         int add = (msgbuf[0] & E1000_VT_MSGINFO_MASK) >> E1000_VT_MSGINFO_SHIFT;
3942         int vid = (msgbuf[1] & E1000_VLVF_VLANID_MASK);
3943
3944         return igb_vlvf_set(adapter, vid, add, vf);
3945 }
3946
3947 static inline void igb_vf_reset_event(struct igb_adapter *adapter, u32 vf)
3948 {
3949         struct e1000_hw *hw = &adapter->hw;
3950
3951         /* disable mailbox functionality for vf */
3952         adapter->vf_data[vf].clear_to_send = false;
3953
3954         /* reset offloads to defaults */
3955         igb_set_vmolr(hw, vf);
3956
3957         /* reset vlans for device */
3958         igb_clear_vf_vfta(adapter, vf);
3959
3960         /* reset multicast table array for vf */
3961         adapter->vf_data[vf].num_vf_mc_hashes = 0;
3962
3963         /* Flush and reset the mta with the new values */
3964         igb_set_multi(adapter->netdev);
3965 }
3966
3967 static inline void igb_vf_reset_msg(struct igb_adapter *adapter, u32 vf)
3968 {
3969         struct e1000_hw *hw = &adapter->hw;
3970         unsigned char *vf_mac = adapter->vf_data[vf].vf_mac_addresses;
3971         u32 reg, msgbuf[3];
3972         u8 *addr = (u8 *)(&msgbuf[1]);
3973
3974         /* process all the same items cleared in a function level reset */
3975         igb_vf_reset_event(adapter, vf);
3976
3977         /* set vf mac address */
3978         igb_rar_set(hw, vf_mac, vf + 1);
3979         igb_set_rah_pool(hw, vf, vf + 1);
3980
3981         /* enable transmit and receive for vf */
3982         reg = rd32(E1000_VFTE);
3983         wr32(E1000_VFTE, reg | (1 << vf));
3984         reg = rd32(E1000_VFRE);
3985         wr32(E1000_VFRE, reg | (1 << vf));
3986
3987         /* enable mailbox functionality for vf */
3988         adapter->vf_data[vf].clear_to_send = true;
3989
3990         /* reply to reset with ack and vf mac address */
3991         msgbuf[0] = E1000_VF_RESET | E1000_VT_MSGTYPE_ACK;
3992         memcpy(addr, vf_mac, 6);
3993         igb_write_mbx(hw, msgbuf, 3, vf);
3994 }
3995
3996 static int igb_set_vf_mac_addr(struct igb_adapter *adapter, u32 *msg, int vf)
3997 {
3998                 unsigned char *addr = (char *)&msg[1];
3999                 int err = -1;
4000
4001                 if (is_valid_ether_addr(addr))
4002                         err = igb_set_vf_mac(adapter, vf, addr);
4003
4004                 return err;
4005
4006 }
4007
4008 static void igb_rcv_ack_from_vf(struct igb_adapter *adapter, u32 vf)
4009 {
4010         struct e1000_hw *hw = &adapter->hw;
4011         u32 msg = E1000_VT_MSGTYPE_NACK;
4012
4013         /* if device isn't clear to send it shouldn't be reading either */
4014         if (!adapter->vf_data[vf].clear_to_send)
4015                 igb_write_mbx(hw, &msg, 1, vf);
4016 }
4017
4018
4019 static void igb_msg_task(struct igb_adapter *adapter)
4020 {
4021         struct e1000_hw *hw = &adapter->hw;
4022         u32 vf;
4023
4024         for (vf = 0; vf < adapter->vfs_allocated_count; vf++) {
4025                 /* process any reset requests */
4026                 if (!igb_check_for_rst(hw, vf)) {
4027                         adapter->vf_data[vf].clear_to_send = false;
4028                         igb_vf_reset_event(adapter, vf);
4029                 }
4030
4031                 /* process any messages pending */
4032                 if (!igb_check_for_msg(hw, vf))
4033                         igb_rcv_msg_from_vf(adapter, vf);
4034
4035                 /* process any acks */
4036                 if (!igb_check_for_ack(hw, vf))
4037                         igb_rcv_ack_from_vf(adapter, vf);
4038
4039         }
4040 }
4041
4042 static int igb_rcv_msg_from_vf(struct igb_adapter *adapter, u32 vf)
4043 {
4044         u32 mbx_size = E1000_VFMAILBOX_SIZE;
4045         u32 msgbuf[mbx_size];
4046         struct e1000_hw *hw = &adapter->hw;
4047         s32 retval;
4048
4049         retval = igb_read_mbx(hw, msgbuf, mbx_size, vf);
4050
4051         if (retval)
4052                 dev_err(&adapter->pdev->dev,
4053                         "Error receiving message from VF\n");
4054
4055         /* this is a message we already processed, do nothing */
4056         if (msgbuf[0] & (E1000_VT_MSGTYPE_ACK | E1000_VT_MSGTYPE_NACK))
4057                 return retval;
4058
4059         /*
4060          * until the vf completes a reset it should not be
4061          * allowed to start any configuration.
4062          */
4063
4064         if (msgbuf[0] == E1000_VF_RESET) {
4065                 igb_vf_reset_msg(adapter, vf);
4066
4067                 return retval;
4068         }
4069
4070         if (!adapter->vf_data[vf].clear_to_send) {
4071                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4072                 igb_write_mbx(hw, msgbuf, 1, vf);
4073                 return retval;
4074         }
4075
4076         switch ((msgbuf[0] & 0xFFFF)) {
4077         case E1000_VF_SET_MAC_ADDR:
4078                 retval = igb_set_vf_mac_addr(adapter, msgbuf, vf);
4079                 break;
4080         case E1000_VF_SET_MULTICAST:
4081                 retval = igb_set_vf_multicasts(adapter, msgbuf, vf);
4082                 break;
4083         case E1000_VF_SET_LPE:
4084                 retval = igb_set_vf_rlpml(adapter, msgbuf[1], vf);
4085                 break;
4086         case E1000_VF_SET_VLAN:
4087                 retval = igb_set_vf_vlan(adapter, msgbuf, vf);
4088                 break;
4089         default:
4090                 dev_err(&adapter->pdev->dev, "Unhandled Msg %08x\n", msgbuf[0]);
4091                 retval = -1;
4092                 break;
4093         }
4094
4095         /* notify the VF of the results of what it sent us */
4096         if (retval)
4097                 msgbuf[0] |= E1000_VT_MSGTYPE_NACK;
4098         else
4099                 msgbuf[0] |= E1000_VT_MSGTYPE_ACK;
4100
4101         msgbuf[0] |= E1000_VT_MSGTYPE_CTS;
4102
4103         igb_write_mbx(hw, msgbuf, 1, vf);
4104
4105         return retval;
4106 }
4107
4108 /**
4109  * igb_intr_msi - Interrupt Handler
4110  * @irq: interrupt number
4111  * @data: pointer to a network interface device structure
4112  **/
4113 static irqreturn_t igb_intr_msi(int irq, void *data)
4114 {
4115         struct net_device *netdev = data;
4116         struct igb_adapter *adapter = netdev_priv(netdev);
4117         struct e1000_hw *hw = &adapter->hw;
4118         /* read ICR disables interrupts using IAM */
4119         u32 icr = rd32(E1000_ICR);
4120
4121         igb_write_itr(adapter->rx_ring);
4122
4123         if(icr & E1000_ICR_DOUTSYNC) {
4124                 /* HW is reporting DMA is out of sync */
4125                 adapter->stats.doosync++;
4126         }
4127
4128         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4129                 hw->mac.get_link_status = 1;
4130                 if (!test_bit(__IGB_DOWN, &adapter->state))
4131                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
4132         }
4133
4134         napi_schedule(&adapter->rx_ring[0].napi);
4135
4136         return IRQ_HANDLED;
4137 }
4138
4139 /**
4140  * igb_intr - Legacy Interrupt Handler
4141  * @irq: interrupt number
4142  * @data: pointer to a network interface device structure
4143  **/
4144 static irqreturn_t igb_intr(int irq, void *data)
4145 {
4146         struct net_device *netdev = data;
4147         struct igb_adapter *adapter = netdev_priv(netdev);
4148         struct e1000_hw *hw = &adapter->hw;
4149         /* Interrupt Auto-Mask...upon reading ICR, interrupts are masked.  No
4150          * need for the IMC write */
4151         u32 icr = rd32(E1000_ICR);
4152         if (!icr)
4153                 return IRQ_NONE;  /* Not our interrupt */
4154
4155         igb_write_itr(adapter->rx_ring);
4156
4157         /* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
4158          * not set, then the adapter didn't send an interrupt */
4159         if (!(icr & E1000_ICR_INT_ASSERTED))
4160                 return IRQ_NONE;
4161
4162         if(icr & E1000_ICR_DOUTSYNC) {
4163                 /* HW is reporting DMA is out of sync */
4164                 adapter->stats.doosync++;
4165         }
4166
4167         if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
4168                 hw->mac.get_link_status = 1;
4169                 /* guard against interrupt when we're going down */
4170                 if (!test_bit(__IGB_DOWN, &adapter->state))
4171                         mod_timer(&adapter->watchdog_timer, jiffies + 1);
4172         }
4173
4174         napi_schedule(&adapter->rx_ring[0].napi);
4175
4176         return IRQ_HANDLED;
4177 }
4178
4179 static inline void igb_rx_irq_enable(struct igb_ring *rx_ring)
4180 {
4181         struct igb_adapter *adapter = rx_ring->adapter;
4182         struct e1000_hw *hw = &adapter->hw;
4183
4184         if (adapter->itr_setting & 3) {
4185                 if (adapter->num_rx_queues == 1)
4186                         igb_set_itr(adapter);
4187                 else
4188                         igb_update_ring_itr(rx_ring);
4189         }
4190
4191         if (!test_bit(__IGB_DOWN, &adapter->state)) {
4192                 if (adapter->msix_entries)
4193                         wr32(E1000_EIMS, rx_ring->eims_value);
4194                 else
4195                         igb_irq_enable(adapter);
4196         }
4197 }
4198
4199 /**
4200  * igb_poll - NAPI Rx polling callback
4201  * @napi: napi polling structure
4202  * @budget: count of how many packets we should handle
4203  **/
4204 static int igb_poll(struct napi_struct *napi, int budget)
4205 {
4206         struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
4207         int work_done = 0;
4208
4209 #ifdef CONFIG_IGB_DCA
4210         if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4211                 igb_update_rx_dca(rx_ring);
4212 #endif
4213         igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
4214
4215         if (rx_ring->buddy) {
4216 #ifdef CONFIG_IGB_DCA
4217                 if (rx_ring->adapter->flags & IGB_FLAG_DCA_ENABLED)
4218                         igb_update_tx_dca(rx_ring->buddy);
4219 #endif
4220                 if (!igb_clean_tx_irq(rx_ring->buddy))
4221                         work_done = budget;
4222         }
4223
4224         /* If not enough Rx work done, exit the polling mode */
4225         if (work_done < budget) {
4226                 napi_complete(napi);
4227                 igb_rx_irq_enable(rx_ring);
4228         }
4229
4230         return work_done;
4231 }
4232
4233 /**
4234  * igb_hwtstamp - utility function which checks for TX time stamp
4235  * @adapter: board private structure
4236  * @skb: packet that was just sent
4237  *
4238  * If we were asked to do hardware stamping and such a time stamp is
4239  * available, then it must have been for this skb here because we only
4240  * allow only one such packet into the queue.
4241  */
4242 static void igb_tx_hwtstamp(struct igb_adapter *adapter, struct sk_buff *skb)
4243 {
4244         union skb_shared_tx *shtx = skb_tx(skb);
4245         struct e1000_hw *hw = &adapter->hw;
4246
4247         if (unlikely(shtx->hardware)) {
4248                 u32 valid = rd32(E1000_TSYNCTXCTL) & E1000_TSYNCTXCTL_VALID;
4249                 if (valid) {
4250                         u64 regval = rd32(E1000_TXSTMPL);
4251                         u64 ns;
4252                         struct skb_shared_hwtstamps shhwtstamps;
4253
4254                         memset(&shhwtstamps, 0, sizeof(shhwtstamps));
4255                         regval |= (u64)rd32(E1000_TXSTMPH) << 32;
4256                         ns = timecounter_cyc2time(&adapter->clock,
4257                                                   regval);
4258                         timecompare_update(&adapter->compare, ns);
4259                         shhwtstamps.hwtstamp = ns_to_ktime(ns);
4260                         shhwtstamps.syststamp =
4261                                 timecompare_transform(&adapter->compare, ns);
4262                         skb_tstamp_tx(skb, &shhwtstamps);
4263                 }
4264         }
4265 }
4266
4267 /**
4268  * igb_clean_tx_irq - Reclaim resources after transmit completes
4269  * @adapter: board private structure
4270  * returns true if ring is completely cleaned
4271  **/
4272 static bool igb_clean_tx_irq(struct igb_ring *tx_ring)
4273 {
4274         struct igb_adapter *adapter = tx_ring->adapter;
4275         struct net_device *netdev = adapter->netdev;
4276         struct e1000_hw *hw = &adapter->hw;
4277         struct igb_buffer *buffer_info;
4278         struct sk_buff *skb;
4279         union e1000_adv_tx_desc *tx_desc, *eop_desc;
4280         unsigned int total_bytes = 0, total_packets = 0;
4281         unsigned int i, eop, count = 0;
4282         bool cleaned = false;
4283
4284         i = tx_ring->next_to_clean;
4285         eop = tx_ring->buffer_info[i].next_to_watch;
4286         eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4287
4288         while ((eop_desc->wb.status & cpu_to_le32(E1000_TXD_STAT_DD)) &&
4289                (count < tx_ring->count)) {
4290                 for (cleaned = false; !cleaned; count++) {
4291                         tx_desc = E1000_TX_DESC_ADV(*tx_ring, i);
4292                         buffer_info = &tx_ring->buffer_info[i];
4293                         cleaned = (i == eop);
4294                         skb = buffer_info->skb;
4295
4296                         if (skb) {
4297                                 unsigned int segs, bytecount;
4298                                 /* gso_segs is currently only valid for tcp */
4299                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
4300                                 /* multiply data chunks by size of headers */
4301                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
4302                                             skb->len;
4303                                 total_packets += segs;
4304                                 total_bytes += bytecount;
4305
4306                                 igb_tx_hwtstamp(adapter, skb);
4307                         }
4308
4309                         igb_unmap_and_free_tx_resource(adapter, buffer_info);
4310                         tx_desc->wb.status = 0;
4311
4312                         i++;
4313                         if (i == tx_ring->count)
4314                                 i = 0;
4315                 }
4316                 eop = tx_ring->buffer_info[i].next_to_watch;
4317                 eop_desc = E1000_TX_DESC_ADV(*tx_ring, eop);
4318         }
4319
4320         tx_ring->next_to_clean = i;
4321
4322         if (unlikely(count &&
4323                      netif_carrier_ok(netdev) &&
4324                      IGB_DESC_UNUSED(tx_ring) >= IGB_TX_QUEUE_WAKE)) {
4325                 /* Make sure that anybody stopping the queue after this
4326                  * sees the new next_to_clean.
4327                  */
4328                 smp_mb();
4329                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
4330                     !(test_bit(__IGB_DOWN, &adapter->state))) {
4331                         netif_wake_subqueue(netdev, tx_ring->queue_index);
4332                         ++adapter->restart_queue;
4333                 }
4334         }
4335
4336         if (tx_ring->detect_tx_hung) {
4337                 /* Detect a transmit hang in hardware, this serializes the
4338                  * check with the clearing of time_stamp and movement of i */
4339                 tx_ring->detect_tx_hung = false;
4340                 if (tx_ring->buffer_info[i].time_stamp &&
4341                     time_after(jiffies, tx_ring->buffer_info[i].time_stamp +
4342                                (adapter->tx_timeout_factor * HZ))
4343                     && !(rd32(E1000_STATUS) &
4344                          E1000_STATUS_TXOFF)) {
4345
4346                         /* detected Tx unit hang */
4347                         dev_err(&adapter->pdev->dev,
4348                                 "Detected Tx Unit Hang\n"
4349                                 "  Tx Queue             <%d>\n"
4350                                 "  TDH                  <%x>\n"
4351                                 "  TDT                  <%x>\n"
4352                                 "  next_to_use          <%x>\n"
4353                                 "  next_to_clean        <%x>\n"
4354                                 "buffer_info[next_to_clean]\n"
4355                                 "  time_stamp           <%lx>\n"
4356                                 "  next_to_watch        <%x>\n"
4357                                 "  jiffies              <%lx>\n"
4358                                 "  desc.status          <%x>\n",
4359                                 tx_ring->queue_index,
4360                                 readl(adapter->hw.hw_addr + tx_ring->head),
4361                                 readl(adapter->hw.hw_addr + tx_ring->tail),
4362                                 tx_ring->next_to_use,
4363                                 tx_ring->next_to_clean,
4364                                 tx_ring->buffer_info[i].time_stamp,
4365                                 eop,
4366                                 jiffies,
4367                                 eop_desc->wb.status);
4368                         netif_stop_subqueue(netdev, tx_ring->queue_index);
4369                 }
4370         }
4371         tx_ring->total_bytes += total_bytes;
4372         tx_ring->total_packets += total_packets;
4373         tx_ring->tx_stats.bytes += total_bytes;
4374         tx_ring->tx_stats.packets += total_packets;
4375         adapter->net_stats.tx_bytes += total_bytes;
4376         adapter->net_stats.tx_packets += total_packets;
4377         return (count < tx_ring->count);
4378 }
4379
4380 /**
4381  * igb_receive_skb - helper function to handle rx indications
4382  * @ring: pointer to receive ring receving this packet
4383  * @status: descriptor status field as written by hardware
4384  * @rx_desc: receive descriptor containing vlan and type information.
4385  * @skb: pointer to sk_buff to be indicated to stack
4386  **/
4387 static void igb_receive_skb(struct igb_ring *ring, u8 status,
4388                             union e1000_adv_rx_desc * rx_desc,
4389                             struct sk_buff *skb)
4390 {
4391         struct igb_adapter * adapter = ring->adapter;
4392         bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
4393
4394         skb_record_rx_queue(skb, ring->queue_index);
4395         if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
4396                 if (vlan_extracted)
4397                         vlan_gro_receive(&ring->napi, adapter->vlgrp,
4398                                          le16_to_cpu(rx_desc->wb.upper.vlan),
4399                                          skb);
4400                 else
4401                         napi_gro_receive(&ring->napi, skb);
4402         } else {
4403                 if (vlan_extracted)
4404                         vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
4405                                           le16_to_cpu(rx_desc->wb.upper.vlan));
4406                 else
4407                         netif_receive_skb(skb);
4408         }
4409 }
4410
4411 static inline void igb_rx_checksum_adv(struct igb_adapter *adapter,
4412                                        u32 status_err, struct sk_buff *skb)
4413 {
4414         skb->ip_summed = CHECKSUM_NONE;
4415
4416         /* Ignore Checksum bit is set or checksum is disabled through ethtool */
4417         if ((status_err & E1000_RXD_STAT_IXSM) || !adapter->rx_csum)
4418                 return;
4419         /* TCP/UDP checksum error bit is set */
4420         if (status_err &
4421             (E1000_RXDEXT_STATERR_TCPE | E1000_RXDEXT_STATERR_IPE)) {
4422                 /* let the stack verify checksum errors */
4423                 adapter->hw_csum_err++;
4424                 return;
4425         }
4426         /* It must be a TCP or UDP packet with a valid checksum */
4427         if (status_err & (E1000_RXD_STAT_TCPCS | E1000_RXD_STAT_UDPCS))
4428                 skb->ip_summed = CHECKSUM_UNNECESSARY;
4429
4430         adapter->hw_csum_good++;
4431 }
4432
4433 static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
4434                                  int *work_done, int budget)
4435 {
4436         struct igb_adapter *adapter = rx_ring->adapter;
4437         struct net_device *netdev = adapter->netdev;
4438         struct e1000_hw *hw = &adapter->hw;
4439         struct pci_dev *pdev = adapter->pdev;
4440         union e1000_adv_rx_desc *rx_desc , *next_rxd;
4441         struct igb_buffer *buffer_info , *next_buffer;
4442         struct sk_buff *skb;
4443         bool cleaned = false;
4444         int cleaned_count = 0;
4445         unsigned int total_bytes = 0, total_packets = 0;
4446         unsigned int i;
4447         u32 length, hlen, staterr;
4448
4449         i = rx_ring->next_to_clean;
4450         buffer_info = &rx_ring->buffer_info[i];
4451         rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4452         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4453
4454         while (staterr & E1000_RXD_STAT_DD) {
4455                 if (*work_done >= budget)
4456                         break;
4457                 (*work_done)++;
4458
4459                 skb = buffer_info->skb;
4460                 prefetch(skb->data - NET_IP_ALIGN);
4461                 buffer_info->skb = NULL;
4462
4463                 i++;
4464                 if (i == rx_ring->count)
4465                         i = 0;
4466                 next_rxd = E1000_RX_DESC_ADV(*rx_ring, i);
4467                 prefetch(next_rxd);
4468                 next_buffer = &rx_ring->buffer_info[i];
4469
4470                 length = le16_to_cpu(rx_desc->wb.upper.length);
4471                 cleaned = true;
4472                 cleaned_count++;
4473
4474                 if (!adapter->rx_ps_hdr_size) {
4475                         pci_unmap_single(pdev, buffer_info->dma,
4476                                          adapter->rx_buffer_len +
4477                                            NET_IP_ALIGN,
4478                                          PCI_DMA_FROMDEVICE);
4479                         skb_put(skb, length);
4480                         goto send_up;
4481                 }
4482
4483                 /* HW will not DMA in data larger than the given buffer, even
4484                  * if it parses the (NFS, of course) header to be larger.  In
4485                  * that case, it fills the header buffer and spills the rest
4486                  * into the page.
4487                  */
4488                 hlen = (le16_to_cpu(rx_desc->wb.lower.lo_dword.hdr_info) &
4489                   E1000_RXDADV_HDRBUFLEN_MASK) >> E1000_RXDADV_HDRBUFLEN_SHIFT;
4490                 if (hlen > adapter->rx_ps_hdr_size)
4491                         hlen = adapter->rx_ps_hdr_size;
4492
4493                 if (!skb_shinfo(skb)->nr_frags) {
4494                         pci_unmap_single(pdev, buffer_info->dma,
4495                                          adapter->rx_ps_hdr_size + NET_IP_ALIGN,
4496                                          PCI_DMA_FROMDEVICE);
4497                         skb_put(skb, hlen);
4498                 }
4499
4500                 if (length) {
4501                         pci_unmap_page(pdev, buffer_info->page_dma,
4502                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
4503                         buffer_info->page_dma = 0;
4504
4505                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
4506                                                 buffer_info->page,
4507                                                 buffer_info->page_offset,
4508                                                 length);
4509
4510                         if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
4511                             (page_count(buffer_info->page) != 1))
4512                                 buffer_info->page = NULL;
4513                         else
4514                                 get_page(buffer_info->page);
4515
4516                         skb->len += length;
4517                         skb->data_len += length;
4518
4519                         skb->truesize += length;
4520                 }
4521
4522                 if (!(staterr & E1000_RXD_STAT_EOP)) {
4523                         buffer_info->skb = next_buffer->skb;
4524                         buffer_info->dma = next_buffer->dma;
4525                         next_buffer->skb = skb;
4526                         next_buffer->dma = 0;
4527                         goto next_desc;
4528                 }
4529 send_up:
4530                 /*
4531                  * If this bit is set, then the RX registers contain
4532                  * the time stamp. No other packet will be time
4533                  * stamped until we read these registers, so read the
4534                  * registers to make them available again. Because
4535                  * only one packet can be time stamped at a time, we
4536                  * know that the register values must belong to this
4537                  * one here and therefore we don't need to compare
4538                  * any of the additional attributes stored for it.
4539                  *
4540                  * If nothing went wrong, then it should have a
4541                  * skb_shared_tx that we can turn into a
4542                  * skb_shared_hwtstamps.
4543                  *
4544                  * TODO: can time stamping be triggered (thus locking
4545                  * the registers) without the packet reaching this point
4546                  * here? In that case RX time stamping would get stuck.
4547                  *
4548                  * TODO: in "time stamp all packets" mode this bit is
4549                  * not set. Need a global flag for this mode and then
4550                  * always read the registers. Cannot be done without
4551                  * a race condition.
4552                  */
4553                 if (unlikely(staterr & E1000_RXD_STAT_TS)) {
4554                         u64 regval;
4555                         u64 ns;
4556                         struct skb_shared_hwtstamps *shhwtstamps =
4557                                 skb_hwtstamps(skb);
4558
4559                         WARN(!(rd32(E1000_TSYNCRXCTL) & E1000_TSYNCRXCTL_VALID),
4560                              "igb: no RX time stamp available for time stamped packet");
4561                         regval = rd32(E1000_RXSTMPL);
4562                         regval |= (u64)rd32(E1000_RXSTMPH) << 32;
4563                         ns = timecounter_cyc2time(&adapter->clock, regval);
4564                         timecompare_update(&adapter->compare, ns);
4565                         memset(shhwtstamps, 0, sizeof(*shhwtstamps));
4566                         shhwtstamps->hwtstamp = ns_to_ktime(ns);
4567                         shhwtstamps->syststamp =
4568                                 timecompare_transform(&adapter->compare, ns);
4569                 }
4570
4571                 if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
4572                         dev_kfree_skb_irq(skb);
4573                         goto next_desc;
4574                 }
4575
4576                 total_bytes += skb->len;
4577                 total_packets++;
4578
4579                 igb_rx_checksum_adv(adapter, staterr, skb);
4580
4581                 skb->protocol = eth_type_trans(skb, netdev);
4582
4583                 igb_receive_skb(rx_ring, staterr, rx_desc, skb);
4584
4585 next_desc:
4586                 rx_desc->wb.upper.status_error = 0;
4587
4588                 /* return some buffers to hardware, one at a time is too slow */
4589                 if (cleaned_count >= IGB_RX_BUFFER_WRITE) {
4590                         igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4591                         cleaned_count = 0;
4592                 }
4593
4594                 /* use prefetched values */
4595                 rx_desc = next_rxd;
4596                 buffer_info = next_buffer;
4597                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
4598         }
4599
4600         rx_ring->next_to_clean = i;
4601         cleaned_count = IGB_DESC_UNUSED(rx_ring);
4602
4603         if (cleaned_count)
4604                 igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
4605
4606         rx_ring->total_packets += total_packets;
4607         rx_ring->total_bytes += total_bytes;
4608         rx_ring->rx_stats.packets += total_packets;
4609         rx_ring->rx_stats.bytes += total_bytes;
4610         adapter->net_stats.rx_bytes += total_bytes;
4611         adapter->net_stats.rx_packets += total_packets;
4612         return cleaned;
4613 }
4614
4615 /**
4616  * igb_alloc_rx_buffers_adv - Replace used receive buffers; packet split
4617  * @adapter: address of board private structure
4618  **/
4619 static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
4620                                      int cleaned_count)
4621 {
4622         struct igb_adapter *adapter = rx_ring->adapter;
4623         struct net_device *netdev = adapter->netdev;
4624         struct pci_dev *pdev = adapter->pdev;
4625         union e1000_adv_rx_desc *rx_desc;
4626         struct igb_buffer *buffer_info;
4627         struct sk_buff *skb;
4628         unsigned int i;
4629         int bufsz;
4630
4631         i = rx_ring->next_to_use;
4632         buffer_info = &rx_ring->buffer_info[i];
4633
4634         if (adapter->rx_ps_hdr_size)
4635                 bufsz = adapter->rx_ps_hdr_size;
4636         else
4637                 bufsz = adapter->rx_buffer_len;
4638         bufsz += NET_IP_ALIGN;
4639
4640         while (cleaned_count--) {
4641                 rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
4642
4643                 if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
4644                         if (!buffer_info->page) {
4645                                 buffer_info->page = alloc_page(GFP_ATOMIC);
4646                                 if (!buffer_info->page) {
4647                                         adapter->alloc_rx_buff_failed++;
4648                                         goto no_buffers;
4649                                 }
4650                                 buffer_info->page_offset = 0;
4651                         } else {
4652                                 buffer_info->page_offset ^= PAGE_SIZE / 2;
4653                         }
4654                         buffer_info->page_dma =
4655                                 pci_map_page(pdev, buffer_info->page,
4656                                              buffer_info->page_offset,
4657                                              PAGE_SIZE / 2,
4658                                              PCI_DMA_FROMDEVICE);
4659                 }
4660
4661                 if (!buffer_info->skb) {
4662                         skb = netdev_alloc_skb(netdev, bufsz);
4663                         if (!skb) {
4664                                 adapter->alloc_rx_buff_failed++;
4665                                 goto no_buffers;
4666                         }
4667
4668                         /* Make buffer alignment 2 beyond a 16 byte boundary
4669                          * this will result in a 16 byte aligned IP header after
4670                          * the 14 byte MAC header is removed
4671                          */
4672                         skb_reserve(skb, NET_IP_ALIGN);
4673
4674                         buffer_info->skb = skb;
4675                         buffer_info->dma = pci_map_single(pdev, skb->data,
4676                                                           bufsz,
4677                                                           PCI_DMA_FROMDEVICE);
4678                 }
4679                 /* Refresh the desc even if buffer_addrs didn't change because
4680                  * each write-back erases this info. */
4681                 if (adapter->rx_ps_hdr_size) {
4682                         rx_desc->read.pkt_addr =
4683                              cpu_to_le64(buffer_info->page_dma);
4684                         rx_desc->read.hdr_addr = cpu_to_le64(buffer_info->dma);
4685                 } else {
4686                         rx_desc->read.pkt_addr =
4687                              cpu_to_le64(buffer_info->dma);
4688                         rx_desc->read.hdr_addr = 0;
4689                 }
4690
4691                 i++;
4692                 if (i == rx_ring->count)
4693                         i = 0;
4694                 buffer_info = &rx_ring->buffer_info[i];
4695         }
4696
4697 no_buffers:
4698         if (rx_ring->next_to_use != i) {
4699                 rx_ring->next_to_use = i;
4700                 if (i == 0)
4701                         i = (rx_ring->count - 1);
4702                 else
4703                         i--;
4704
4705                 /* Force memory writes to complete before letting h/w
4706                  * know there are new descriptors to fetch.  (Only
4707                  * applicable for weak-ordered memory model archs,
4708                  * such as IA-64). */
4709                 wmb();
4710                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
4711         }
4712 }
4713
4714 /**
4715  * igb_mii_ioctl -
4716  * @netdev:
4717  * @ifreq:
4718  * @cmd:
4719  **/
4720 static int igb_mii_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4721 {
4722         struct igb_adapter *adapter = netdev_priv(netdev);
4723         struct mii_ioctl_data *data = if_mii(ifr);
4724
4725         if (adapter->hw.phy.media_type != e1000_media_type_copper)
4726                 return -EOPNOTSUPP;
4727
4728         switch (cmd) {
4729         case SIOCGMIIPHY:
4730                 data->phy_id = adapter->hw.phy.addr;
4731                 break;
4732         case SIOCGMIIREG:
4733                 if (!capable(CAP_NET_ADMIN))
4734                         return -EPERM;
4735                 if (igb_read_phy_reg(&adapter->hw, data->reg_num & 0x1F,
4736                                      &data->val_out))
4737                         return -EIO;
4738                 break;
4739         case SIOCSMIIREG:
4740         default:
4741                 return -EOPNOTSUPP;
4742         }
4743         return 0;
4744 }
4745
4746 /**
4747  * igb_hwtstamp_ioctl - control hardware time stamping
4748  * @netdev:
4749  * @ifreq:
4750  * @cmd:
4751  *
4752  * Outgoing time stamping can be enabled and disabled. Play nice and
4753  * disable it when requested, although it shouldn't case any overhead
4754  * when no packet needs it. At most one packet in the queue may be
4755  * marked for time stamping, otherwise it would be impossible to tell
4756  * for sure to which packet the hardware time stamp belongs.
4757  *
4758  * Incoming time stamping has to be configured via the hardware
4759  * filters. Not all combinations are supported, in particular event
4760  * type has to be specified. Matching the kind of event packet is
4761  * not supported, with the exception of "all V2 events regardless of
4762  * level 2 or 4".
4763  *
4764  **/
4765 static int igb_hwtstamp_ioctl(struct net_device *netdev,
4766                               struct ifreq *ifr, int cmd)
4767 {
4768         struct igb_adapter *adapter = netdev_priv(netdev);
4769         struct e1000_hw *hw = &adapter->hw;
4770         struct hwtstamp_config config;
4771         u32 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4772         u32 tsync_rx_ctl_bit = E1000_TSYNCRXCTL_ENABLED;
4773         u32 tsync_rx_ctl_type = 0;
4774         u32 tsync_rx_cfg = 0;
4775         int is_l4 = 0;
4776         int is_l2 = 0;
4777         short port = 319; /* PTP */
4778         u32 regval;
4779
4780         if (copy_from_user(&config, ifr->ifr_data, sizeof(config)))
4781                 return -EFAULT;
4782
4783         /* reserved for future extensions */
4784         if (config.flags)
4785                 return -EINVAL;
4786
4787         switch (config.tx_type) {
4788         case HWTSTAMP_TX_OFF:
4789                 tsync_tx_ctl_bit = 0;
4790                 break;
4791         case HWTSTAMP_TX_ON:
4792                 tsync_tx_ctl_bit = E1000_TSYNCTXCTL_ENABLED;
4793                 break;
4794         default:
4795                 return -ERANGE;
4796         }
4797
4798         switch (config.rx_filter) {
4799         case HWTSTAMP_FILTER_NONE:
4800                 tsync_rx_ctl_bit = 0;
4801                 break;
4802         case HWTSTAMP_FILTER_PTP_V1_L4_EVENT:
4803         case HWTSTAMP_FILTER_PTP_V2_L4_EVENT:
4804         case HWTSTAMP_FILTER_PTP_V2_L2_EVENT:
4805         case HWTSTAMP_FILTER_ALL:
4806                 /*
4807                  * register TSYNCRXCFG must be set, therefore it is not
4808                  * possible to time stamp both Sync and Delay_Req messages
4809                  * => fall back to time stamping all packets
4810                  */
4811                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_ALL;
4812                 config.rx_filter = HWTSTAMP_FILTER_ALL;
4813                 break;
4814         case HWTSTAMP_FILTER_PTP_V1_L4_SYNC:
4815                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4816                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_SYNC_MESSAGE;
4817                 is_l4 = 1;
4818                 break;
4819         case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ:
4820                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L4_V1;
4821                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V1_DELAY_REQ_MESSAGE;
4822                 is_l4 = 1;
4823                 break;
4824         case HWTSTAMP_FILTER_PTP_V2_L2_SYNC:
4825         case HWTSTAMP_FILTER_PTP_V2_L4_SYNC:
4826                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4827                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_SYNC_MESSAGE;
4828                 is_l2 = 1;
4829                 is_l4 = 1;
4830                 config.rx_filter = HWTSTAMP_FILTER_SOME;
4831                 break;
4832         case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ:
4833         case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ:
4834                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_L2_L4_V2;
4835                 tsync_rx_cfg = E1000_TSYNCRXCFG_PTP_V2_DELAY_REQ_MESSAGE;
4836                 is_l2 = 1;
4837                 is_l4 = 1;
4838                 config.rx_filter = HWTSTAMP_FILTER_SOME;
4839                 break;
4840         case HWTSTAMP_FILTER_PTP_V2_EVENT:
4841         case HWTSTAMP_FILTER_PTP_V2_SYNC:
4842         case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ:
4843                 tsync_rx_ctl_type = E1000_TSYNCRXCTL_TYPE_EVENT_V2;
4844                 config.rx_filter = HWTSTAMP_FILTER_PTP_V2_EVENT;
4845                 is_l2 = 1;
4846                 break;
4847         default:
4848                 return -ERANGE;
4849         }
4850
4851         /* enable/disable TX */
4852         regval = rd32(E1000_TSYNCTXCTL);
4853         regval = (regval & ~E1000_TSYNCTXCTL_ENABLED) | tsync_tx_ctl_bit;
4854         wr32(E1000_TSYNCTXCTL, regval);
4855
4856         /* enable/disable RX, define which PTP packets are time stamped */
4857         regval = rd32(E1000_TSYNCRXCTL);
4858         regval = (regval & ~E1000_TSYNCRXCTL_ENABLED) | tsync_rx_ctl_bit;
4859         regval = (regval & ~0xE) | tsync_rx_ctl_type;
4860         wr32(E1000_TSYNCRXCTL, regval);
4861         wr32(E1000_TSYNCRXCFG, tsync_rx_cfg);
4862
4863         /*
4864          * Ethertype Filter Queue Filter[0][15:0] = 0x88F7
4865          *                                          (Ethertype to filter on)
4866          * Ethertype Filter Queue Filter[0][26] = 0x1 (Enable filter)
4867          * Ethertype Filter Queue Filter[0][30] = 0x1 (Enable Timestamping)
4868          */
4869         wr32(E1000_ETQF0, is_l2 ? 0x440088f7 : 0);
4870
4871         /* L4 Queue Filter[0]: only filter by source and destination port */
4872         wr32(E1000_SPQF0, htons(port));
4873         wr32(E1000_IMIREXT(0), is_l4 ?
4874              ((1<<12) | (1<<19) /* bypass size and control flags */) : 0);
4875         wr32(E1000_IMIR(0), is_l4 ?
4876              (htons(port)
4877               | (0<<16) /* immediate interrupt disabled */
4878               | 0 /* (1<<17) bit cleared: do not bypass
4879                      destination port check */)
4880                 : 0);
4881         wr32(E1000_FTQF0, is_l4 ?
4882              (0x11 /* UDP */
4883               | (1<<15) /* VF not compared */
4884               | (1<<27) /* Enable Timestamping */
4885               | (7<<28) /* only source port filter enabled,
4886                            source/target address and protocol
4887                            masked */)
4888              : ((1<<15) | (15<<28) /* all mask bits set = filter not
4889                                       enabled */));
4890
4891         wrfl();
4892
4893         adapter->hwtstamp_config = config;
4894
4895         /* clear TX/RX time stamp registers, just to be sure */
4896         regval = rd32(E1000_TXSTMPH);
4897         regval = rd32(E1000_RXSTMPH);
4898
4899         return copy_to_user(ifr->ifr_data, &config, sizeof(config)) ?
4900                 -EFAULT : 0;
4901 }
4902
4903 /**
4904  * igb_ioctl -
4905  * @netdev:
4906  * @ifreq:
4907  * @cmd:
4908  **/
4909 static int igb_ioctl(struct net_device *netdev, struct ifreq *ifr, int cmd)
4910 {
4911         switch (cmd) {
4912         case SIOCGMIIPHY:
4913         case SIOCGMIIREG:
4914         case SIOCSMIIREG:
4915                 return igb_mii_ioctl(netdev, ifr, cmd);
4916         case SIOCSHWTSTAMP:
4917                 return igb_hwtstamp_ioctl(netdev, ifr, cmd);
4918         default:
4919                 return -EOPNOTSUPP;
4920         }
4921 }
4922
4923 static void igb_vlan_rx_register(struct net_device *netdev,
4924                                  struct vlan_group *grp)
4925 {
4926         struct igb_adapter *adapter = netdev_priv(netdev);
4927         struct e1000_hw *hw = &adapter->hw;
4928         u32 ctrl, rctl;
4929
4930         igb_irq_disable(adapter);
4931         adapter->vlgrp = grp;
4932
4933         if (grp) {
4934                 /* enable VLAN tag insert/strip */
4935                 ctrl = rd32(E1000_CTRL);
4936                 ctrl |= E1000_CTRL_VME;
4937                 wr32(E1000_CTRL, ctrl);
4938
4939                 /* enable VLAN receive filtering */
4940                 rctl = rd32(E1000_RCTL);
4941                 rctl &= ~E1000_RCTL_CFIEN;
4942                 wr32(E1000_RCTL, rctl);
4943                 igb_update_mng_vlan(adapter);
4944         } else {
4945                 /* disable VLAN tag insert/strip */
4946                 ctrl = rd32(E1000_CTRL);
4947                 ctrl &= ~E1000_CTRL_VME;
4948                 wr32(E1000_CTRL, ctrl);
4949
4950                 if (adapter->mng_vlan_id != (u16)IGB_MNG_VLAN_NONE) {
4951                         igb_vlan_rx_kill_vid(netdev, adapter->mng_vlan_id);
4952                         adapter->mng_vlan_id = IGB_MNG_VLAN_NONE;
4953                 }
4954         }
4955
4956         igb_rlpml_set(adapter);
4957
4958         if (!test_bit(__IGB_DOWN, &adapter->state))
4959                 igb_irq_enable(adapter);
4960 }
4961
4962 static void igb_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
4963 {
4964         struct igb_adapter *adapter = netdev_priv(netdev);
4965         struct e1000_hw *hw = &adapter->hw;
4966         int pf_id = adapter->vfs_allocated_count;
4967
4968         if ((hw->mng_cookie.status &
4969              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4970             (vid == adapter->mng_vlan_id))
4971                 return;
4972
4973         /* add vid to vlvf if sr-iov is enabled,
4974          * if that fails add directly to filter table */
4975         if (igb_vlvf_set(adapter, vid, true, pf_id))
4976                 igb_vfta_set(hw, vid, true);
4977
4978 }
4979
4980 static void igb_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
4981 {
4982         struct igb_adapter *adapter = netdev_priv(netdev);
4983         struct e1000_hw *hw = &adapter->hw;
4984         int pf_id = adapter->vfs_allocated_count;
4985
4986         igb_irq_disable(adapter);
4987         vlan_group_set_device(adapter->vlgrp, vid, NULL);
4988
4989         if (!test_bit(__IGB_DOWN, &adapter->state))
4990                 igb_irq_enable(adapter);
4991
4992         if ((adapter->hw.mng_cookie.status &
4993              E1000_MNG_DHCP_COOKIE_STATUS_VLAN) &&
4994             (vid == adapter->mng_vlan_id)) {
4995                 /* release control to f/w */
4996                 igb_release_hw_control(adapter);
4997                 return;
4998         }
4999
5000         /* remove vid from vlvf if sr-iov is enabled,
5001          * if not in vlvf remove from vfta */
5002         if (igb_vlvf_set(adapter, vid, false, pf_id))
5003                 igb_vfta_set(hw, vid, false);
5004 }
5005
5006 static void igb_restore_vlan(struct igb_adapter *adapter)
5007 {
5008         igb_vlan_rx_register(adapter->netdev, adapter->vlgrp);
5009
5010         if (adapter->vlgrp) {
5011                 u16 vid;
5012                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
5013                         if (!vlan_group_get_device(adapter->vlgrp, vid))
5014                                 continue;
5015                         igb_vlan_rx_add_vid(adapter->netdev, vid);
5016                 }
5017         }
5018 }
5019
5020 int igb_set_spd_dplx(struct igb_adapter *adapter, u16 spddplx)
5021 {
5022         struct e1000_mac_info *mac = &adapter->hw.mac;
5023
5024         mac->autoneg = 0;
5025
5026         /* Fiber NICs only allow 1000 gbps Full duplex */
5027         if ((adapter->hw.phy.media_type == e1000_media_type_fiber) &&
5028                 spddplx != (SPEED_1000 + DUPLEX_FULL)) {
5029                 dev_err(&adapter->pdev->dev,
5030                         "Unsupported Speed/Duplex configuration\n");
5031                 return -EINVAL;
5032         }
5033
5034         switch (spddplx) {
5035         case SPEED_10 + DUPLEX_HALF:
5036                 mac->forced_speed_duplex = ADVERTISE_10_HALF;
5037                 break;
5038         case SPEED_10 + DUPLEX_FULL:
5039                 mac->forced_speed_duplex = ADVERTISE_10_FULL;
5040                 break;
5041         case SPEED_100 + DUPLEX_HALF:
5042                 mac->forced_speed_duplex = ADVERTISE_100_HALF;
5043                 break;
5044         case SPEED_100 + DUPLEX_FULL:
5045                 mac->forced_speed_duplex = ADVERTISE_100_FULL;
5046                 break;
5047         case SPEED_1000 + DUPLEX_FULL:
5048                 mac->autoneg = 1;
5049                 adapter->hw.phy.autoneg_advertised = ADVERTISE_1000_FULL;
5050                 break;
5051         case SPEED_1000 + DUPLEX_HALF: /* not supported */
5052         default:
5053                 dev_err(&adapter->pdev->dev,
5054                         "Unsupported Speed/Duplex configuration\n");
5055                 return -EINVAL;
5056         }
5057         return 0;
5058 }
5059
5060 static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
5061 {
5062         struct net_device *netdev = pci_get_drvdata(pdev);
5063         struct igb_adapter *adapter = netdev_priv(netdev);
5064         struct e1000_hw *hw = &adapter->hw;
5065         u32 ctrl, rctl, status;
5066         u32 wufc = adapter->wol;
5067 #ifdef CONFIG_PM
5068         int retval = 0;
5069 #endif
5070
5071         netif_device_detach(netdev);
5072
5073         if (netif_running(netdev))
5074                 igb_close(netdev);
5075
5076         igb_reset_interrupt_capability(adapter);
5077
5078         igb_free_queues(adapter);
5079
5080 #ifdef CONFIG_PM
5081         retval = pci_save_state(pdev);
5082         if (retval)
5083                 return retval;
5084 #endif
5085
5086         status = rd32(E1000_STATUS);
5087         if (status & E1000_STATUS_LU)
5088                 wufc &= ~E1000_WUFC_LNKC;
5089
5090         if (wufc) {
5091                 igb_setup_rctl(adapter);
5092                 igb_set_multi(netdev);
5093
5094                 /* turn on all-multi mode if wake on multicast is enabled */
5095                 if (wufc & E1000_WUFC_MC) {
5096                         rctl = rd32(E1000_RCTL);
5097                         rctl |= E1000_RCTL_MPE;
5098                         wr32(E1000_RCTL, rctl);
5099                 }
5100
5101                 ctrl = rd32(E1000_CTRL);
5102                 /* advertise wake from D3Cold */
5103                 #define E1000_CTRL_ADVD3WUC 0x00100000
5104                 /* phy power management enable */
5105                 #define E1000_CTRL_EN_PHY_PWR_MGMT 0x00200000
5106                 ctrl |= E1000_CTRL_ADVD3WUC;
5107                 wr32(E1000_CTRL, ctrl);
5108
5109                 /* Allow time for pending master requests to run */
5110                 igb_disable_pcie_master(&adapter->hw);
5111
5112                 wr32(E1000_WUC, E1000_WUC_PME_EN);
5113                 wr32(E1000_WUFC, wufc);
5114         } else {
5115                 wr32(E1000_WUC, 0);
5116                 wr32(E1000_WUFC, 0);
5117         }
5118
5119         /* make sure adapter isn't asleep if manageability/wol is enabled */
5120         if (wufc || adapter->en_mng_pt) {
5121                 pci_enable_wake(pdev, PCI_D3hot, 1);
5122                 pci_enable_wake(pdev, PCI_D3cold, 1);
5123         } else {
5124                 igb_shutdown_fiber_serdes_link_82575(hw);
5125                 pci_enable_wake(pdev, PCI_D3hot, 0);
5126                 pci_enable_wake(pdev, PCI_D3cold, 0);
5127         }
5128
5129         /* Release control of h/w to f/w.  If f/w is AMT enabled, this
5130          * would have already happened in close and is redundant. */
5131         igb_release_hw_control(adapter);
5132
5133         pci_disable_device(pdev);
5134
5135         pci_set_power_state(pdev, pci_choose_state(pdev, state));
5136
5137         return 0;
5138 }
5139
5140 #ifdef CONFIG_PM
5141 static int igb_resume(struct pci_dev *pdev)
5142 {
5143         struct net_device *netdev = pci_get_drvdata(pdev);
5144         struct igb_adapter *adapter = netdev_priv(netdev);
5145         struct e1000_hw *hw = &adapter->hw;
5146         u32 err;
5147
5148         pci_set_power_state(pdev, PCI_D0);
5149         pci_restore_state(pdev);
5150
5151         err = pci_enable_device_mem(pdev);
5152         if (err) {
5153                 dev_err(&pdev->dev,
5154                         "igb: Cannot enable PCI device from suspend\n");
5155                 return err;
5156         }
5157         pci_set_master(pdev);
5158
5159         pci_enable_wake(pdev, PCI_D3hot, 0);
5160         pci_enable_wake(pdev, PCI_D3cold, 0);
5161
5162         igb_set_interrupt_capability(adapter);
5163
5164         if (igb_alloc_queues(adapter)) {
5165                 dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
5166                 return -ENOMEM;
5167         }
5168
5169         /* e1000_power_up_phy(adapter); */
5170
5171         igb_reset(adapter);
5172
5173         /* let the f/w know that the h/w is now under the control of the
5174          * driver. */
5175         igb_get_hw_control(adapter);
5176
5177         wr32(E1000_WUS, ~0);
5178
5179         if (netif_running(netdev)) {
5180                 err = igb_open(netdev);
5181                 if (err)
5182                         return err;
5183         }
5184
5185         netif_device_attach(netdev);
5186
5187         return 0;
5188 }
5189 #endif
5190
5191 static void igb_shutdown(struct pci_dev *pdev)
5192 {
5193         igb_suspend(pdev, PMSG_SUSPEND);
5194 }
5195
5196 #ifdef CONFIG_NET_POLL_CONTROLLER
5197 /*
5198  * Polling 'interrupt' - used by things like netconsole to send skbs
5199  * without having to re-enable interrupts. It's not called while
5200  * the interrupt routine is executing.
5201  */
5202 static void igb_netpoll(struct net_device *netdev)
5203 {
5204         struct igb_adapter *adapter = netdev_priv(netdev);
5205         struct e1000_hw *hw = &adapter->hw;
5206         int i;
5207
5208         if (!adapter->msix_entries) {
5209                 igb_irq_disable(adapter);
5210                 napi_schedule(&adapter->rx_ring[0].napi);
5211                 return;
5212         }
5213
5214         for (i = 0; i < adapter->num_tx_queues; i++) {
5215                 struct igb_ring *tx_ring = &adapter->tx_ring[i];
5216                 wr32(E1000_EIMC, tx_ring->eims_value);
5217                 igb_clean_tx_irq(tx_ring);
5218                 wr32(E1000_EIMS, tx_ring->eims_value);
5219         }
5220
5221         for (i = 0; i < adapter->num_rx_queues; i++) {
5222                 struct igb_ring *rx_ring = &adapter->rx_ring[i];
5223                 wr32(E1000_EIMC, rx_ring->eims_value);
5224                 napi_schedule(&rx_ring->napi);
5225         }
5226 }
5227 #endif /* CONFIG_NET_POLL_CONTROLLER */
5228
5229 /**
5230  * igb_io_error_detected - called when PCI error is detected
5231  * @pdev: Pointer to PCI device
5232  * @state: The current pci connection state
5233  *
5234  * This function is called after a PCI bus error affecting
5235  * this device has been detected.
5236  */
5237 static pci_ers_result_t igb_io_error_detected(struct pci_dev *pdev,
5238                                               pci_channel_state_t state)
5239 {
5240         struct net_device *netdev = pci_get_drvdata(pdev);
5241         struct igb_adapter *adapter = netdev_priv(netdev);
5242
5243         netif_device_detach(netdev);
5244
5245         if (netif_running(netdev))
5246                 igb_down(adapter);
5247         pci_disable_device(pdev);
5248
5249         /* Request a slot slot reset. */
5250         return PCI_ERS_RESULT_NEED_RESET;
5251 }
5252
5253 /**
5254  * igb_io_slot_reset - called after the pci bus has been reset.
5255  * @pdev: Pointer to PCI device
5256  *
5257  * Restart the card from scratch, as if from a cold-boot. Implementation
5258  * resembles the first-half of the igb_resume routine.
5259  */
5260 static pci_ers_result_t igb_io_slot_reset(struct pci_dev *pdev)
5261 {
5262         struct net_device *netdev = pci_get_drvdata(pdev);
5263         struct igb_adapter *adapter = netdev_priv(netdev);
5264         struct e1000_hw *hw = &adapter->hw;
5265         pci_ers_result_t result;
5266         int err;
5267
5268         if (pci_enable_device_mem(pdev)) {
5269                 dev_err(&pdev->dev,
5270                         "Cannot re-enable PCI device after reset.\n");
5271                 result = PCI_ERS_RESULT_DISCONNECT;
5272         } else {
5273                 pci_set_master(pdev);
5274                 pci_restore_state(pdev);
5275
5276                 pci_enable_wake(pdev, PCI_D3hot, 0);
5277                 pci_enable_wake(pdev, PCI_D3cold, 0);
5278
5279                 igb_reset(adapter);
5280                 wr32(E1000_WUS, ~0);
5281                 result = PCI_ERS_RESULT_RECOVERED;
5282         }
5283
5284         err = pci_cleanup_aer_uncorrect_error_status(pdev);
5285         if (err) {
5286                 dev_err(&pdev->dev, "pci_cleanup_aer_uncorrect_error_status "
5287                         "failed 0x%0x\n", err);
5288                 /* non-fatal, continue */
5289         }
5290
5291         return result;
5292 }
5293
5294 /**
5295  * igb_io_resume - called when traffic can start flowing again.
5296  * @pdev: Pointer to PCI device
5297  *
5298  * This callback is called when the error recovery driver tells us that
5299  * its OK to resume normal operation. Implementation resembles the
5300  * second-half of the igb_resume routine.
5301  */
5302 static void igb_io_resume(struct pci_dev *pdev)
5303 {
5304         struct net_device *netdev = pci_get_drvdata(pdev);
5305         struct igb_adapter *adapter = netdev_priv(netdev);
5306
5307         if (netif_running(netdev)) {
5308                 if (igb_up(adapter)) {
5309                         dev_err(&pdev->dev, "igb_up failed after reset\n");
5310                         return;
5311                 }
5312         }
5313
5314         netif_device_attach(netdev);
5315
5316         /* let the f/w know that the h/w is now under the control of the
5317          * driver. */
5318         igb_get_hw_control(adapter);
5319 }
5320
5321 static inline void igb_set_vmolr(struct e1000_hw *hw, int vfn)
5322 {
5323         u32 reg_data;
5324
5325         reg_data = rd32(E1000_VMOLR(vfn));
5326         reg_data |= E1000_VMOLR_BAM |    /* Accept broadcast */
5327                     E1000_VMOLR_ROPE |   /* Accept packets matched in UTA */
5328                     E1000_VMOLR_ROMPE |  /* Accept packets matched in MTA */
5329                     E1000_VMOLR_AUPE |   /* Accept untagged packets */
5330                     E1000_VMOLR_STRVLAN; /* Strip vlan tags */
5331         wr32(E1000_VMOLR(vfn), reg_data);
5332 }
5333
5334 static inline int igb_set_vf_rlpml(struct igb_adapter *adapter, int size,
5335                                  int vfn)
5336 {
5337         struct e1000_hw *hw = &adapter->hw;
5338         u32 vmolr;
5339
5340         vmolr = rd32(E1000_VMOLR(vfn));
5341         vmolr &= ~E1000_VMOLR_RLPML_MASK;
5342         vmolr |= size | E1000_VMOLR_LPE;
5343         wr32(E1000_VMOLR(vfn), vmolr);
5344
5345         return 0;
5346 }
5347
5348 static inline void igb_set_rah_pool(struct e1000_hw *hw, int pool, int entry)
5349 {
5350         u32 reg_data;
5351
5352         reg_data = rd32(E1000_RAH(entry));
5353         reg_data &= ~E1000_RAH_POOL_MASK;
5354         reg_data |= E1000_RAH_POOL_1 << pool;;
5355         wr32(E1000_RAH(entry), reg_data);
5356 }
5357
5358 static void igb_set_mc_list_pools(struct igb_adapter *adapter,
5359                                   int entry_count, u16 total_rar_filters)
5360 {
5361         struct e1000_hw *hw = &adapter->hw;
5362         int i = adapter->vfs_allocated_count + 1;
5363
5364         if ((i + entry_count) < total_rar_filters)
5365                 total_rar_filters = i + entry_count;
5366
5367         for (; i < total_rar_filters; i++)
5368                 igb_set_rah_pool(hw, adapter->vfs_allocated_count, i);
5369 }
5370
5371 static int igb_set_vf_mac(struct igb_adapter *adapter,
5372                           int vf, unsigned char *mac_addr)
5373 {
5374         struct e1000_hw *hw = &adapter->hw;
5375         int rar_entry = vf + 1; /* VF MAC addresses start at entry 1 */
5376
5377         igb_rar_set(hw, mac_addr, rar_entry);
5378
5379         memcpy(adapter->vf_data[vf].vf_mac_addresses, mac_addr, ETH_ALEN);
5380
5381         igb_set_rah_pool(hw, vf, rar_entry);
5382
5383         return 0;
5384 }
5385
5386 static void igb_vmm_control(struct igb_adapter *adapter)
5387 {
5388         struct e1000_hw *hw = &adapter->hw;
5389         u32 reg_data;
5390
5391         if (!adapter->vfs_allocated_count)
5392                 return;
5393
5394         /* VF's need PF reset indication before they
5395          * can send/receive mail */
5396         reg_data = rd32(E1000_CTRL_EXT);
5397         reg_data |= E1000_CTRL_EXT_PFRSTD;
5398         wr32(E1000_CTRL_EXT, reg_data);
5399
5400         igb_vmdq_set_loopback_pf(hw, true);
5401         igb_vmdq_set_replication_pf(hw, true);
5402 }
5403
5404 #ifdef CONFIG_PCI_IOV
5405 static ssize_t igb_show_num_vfs(struct device *dev,
5406                                 struct device_attribute *attr, char *buf)
5407 {
5408         struct igb_adapter *adapter = netdev_priv(to_net_dev(dev));
5409
5410         return sprintf(buf, "%d\n", adapter->vfs_allocated_count);
5411 }
5412
5413 static ssize_t igb_set_num_vfs(struct device *dev,
5414                                struct device_attribute *attr,
5415                                const char *buf, size_t count)
5416 {
5417         struct net_device *netdev = to_net_dev(dev);
5418         struct igb_adapter *adapter = netdev_priv(netdev);
5419         struct e1000_hw *hw = &adapter->hw;
5420         struct pci_dev *pdev = adapter->pdev;
5421         unsigned int num_vfs, i;
5422         unsigned char mac_addr[ETH_ALEN];
5423         int err;
5424
5425         sscanf(buf, "%u", &num_vfs);
5426
5427         if (num_vfs > 7)
5428                 num_vfs = 7;
5429
5430         /* value unchanged do nothing */
5431         if (num_vfs == adapter->vfs_allocated_count)
5432                 return count;
5433
5434         if (netdev->flags & IFF_UP)
5435                 igb_close(netdev);
5436
5437         igb_reset_interrupt_capability(adapter);
5438         igb_free_queues(adapter);
5439         adapter->tx_ring = NULL;
5440         adapter->rx_ring = NULL;
5441         adapter->vfs_allocated_count = 0;
5442
5443         /* reclaim resources allocated to VFs since we are changing count */
5444         if (adapter->vf_data) {
5445                 /* disable iov and allow time for transactions to clear */
5446                 pci_disable_sriov(pdev);
5447                 msleep(500);
5448
5449                 kfree(adapter->vf_data);
5450                 adapter->vf_data = NULL;
5451                 wr32(E1000_IOVCTL, E1000_IOVCTL_REUSE_VFQ);
5452                 msleep(100);
5453                 dev_info(&pdev->dev, "IOV Disabled\n");
5454         }
5455
5456         if (num_vfs) {
5457                 adapter->vf_data = kcalloc(num_vfs,
5458                                            sizeof(struct vf_data_storage),
5459                                            GFP_KERNEL);
5460                 if (!adapter->vf_data) {
5461                         dev_err(&pdev->dev, "Could not allocate VF private "
5462                                 "data - IOV enable failed\n");
5463                 } else {
5464                         err = pci_enable_sriov(pdev, num_vfs);
5465                         if (!err) {
5466                                 adapter->vfs_allocated_count = num_vfs;
5467                                 dev_info(&pdev->dev, "%d vfs allocated\n", num_vfs);
5468                                 for (i = 0; i < adapter->vfs_allocated_count; i++) {
5469                                         random_ether_addr(mac_addr);
5470                                         igb_set_vf_mac(adapter, i, mac_addr);
5471                                 }
5472                         } else {
5473                                 kfree(adapter->vf_data);
5474                                 adapter->vf_data = NULL;
5475                         }
5476                 }
5477         }
5478
5479         igb_set_interrupt_capability(adapter);
5480         igb_alloc_queues(adapter);
5481         igb_reset(adapter);
5482
5483         if (netdev->flags & IFF_UP)
5484                 igb_open(netdev);
5485
5486         return count;
5487 }
5488 #endif /* CONFIG_PCI_IOV */
5489 /* igb_main.c */