drivers/net/ixgbe/ixgbe_main.c

   1 /*******************************************************************************
   2
   3   Intel 10 Gigabit PCI Express Linux driver
   4   Copyright(c) 1999 - 2008 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/types.h>
  29 #include <linux/module.h>
  30 #include <linux/pci.h>
  31 #include <linux/netdevice.h>
  32 #include <linux/vmalloc.h>
  33 #include <linux/string.h>
  34 #include <linux/in.h>
  35 #include <linux/ip.h>
  36 #include <linux/tcp.h>
  37 #include <linux/ipv6.h>
  38 #include <net/checksum.h>
  39 #include <net/ip6_checksum.h>
  40 #include <linux/ethtool.h>
  41 #include <linux/if_vlan.h>
  42
  43 #include "ixgbe.h"
  44 #include "ixgbe_common.h"
  45
  46 char ixgbe_driver_name[] = "ixgbe";
  47 static const char ixgbe_driver_string[] =
  48                               "Intel(R) 10 Gigabit PCI Express Network Driver";
  49
  50 #define DRV_VERSION "1.3.30-k2"
  51 const char ixgbe_driver_version[] = DRV_VERSION;
  52 static char ixgbe_copyright[] = "Copyright (c) 1999-2007 Intel Corporation.";
  53
  54 static const struct ixgbe_info *ixgbe_info_tbl[] = {
  55         [board_82598] = &ixgbe_82598_info,
  56 };
  57
  58 /* ixgbe_pci_tbl - PCI Device ID Table
  59  *
  60  * Wildcard entries (PCI_ANY_ID) should come last
  61  * Last entry must be all 0s
  62  *
  63  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  64  *   Class, Class Mask, private data (not used) }
  65  */
  66 static struct pci_device_id ixgbe_pci_tbl[] = {
  67         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_DUAL_PORT),
  68          board_82598 },
  69         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AF_SINGLE_PORT),
  70          board_82598 },
  71         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598AT),
  72          board_82598 },
  73         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_CX4),
  74          board_82598 },
  75         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598_CX4_DUAL_PORT),
  76          board_82598 },
  77         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82598EB_XF_LR),
  78          board_82598 },
  79
  80         /* required last entry */
  81         {0, }
  82 };
  83 MODULE_DEVICE_TABLE(pci, ixgbe_pci_tbl);
  84
  85 #ifdef CONFIG_IXGBE_DCA
  86 static int ixgbe_notify_dca(struct notifier_block *, unsigned long event,
  87                             void *p);
  88 static struct notifier_block dca_notifier = {
  89         .notifier_call = ixgbe_notify_dca,
  90         .next          = NULL,
  91         .priority      = 0
  92 };
  93 #endif
  94
  95 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  96 MODULE_DESCRIPTION("Intel(R) 10 Gigabit PCI Express Network Driver");
  97 MODULE_LICENSE("GPL");
  98 MODULE_VERSION(DRV_VERSION);
  99
 100 #define DEFAULT_DEBUG_LEVEL_SHIFT 3
 101
 102 static void ixgbe_release_hw_control(struct ixgbe_adapter *adapter)
 103 {
 104         u32 ctrl_ext;
 105
 106         /* Let firmware take over control of h/w */
 107         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 108         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 109                         ctrl_ext & ~IXGBE_CTRL_EXT_DRV_LOAD);
 110 }
 111
 112 static void ixgbe_get_hw_control(struct ixgbe_adapter *adapter)
 113 {
 114         u32 ctrl_ext;
 115
 116         /* Let firmware know the driver has taken over */
 117         ctrl_ext = IXGBE_READ_REG(&adapter->hw, IXGBE_CTRL_EXT);
 118         IXGBE_WRITE_REG(&adapter->hw, IXGBE_CTRL_EXT,
 119                         ctrl_ext | IXGBE_CTRL_EXT_DRV_LOAD);
 120 }
 121
 122 static void ixgbe_set_ivar(struct ixgbe_adapter *adapter, u16 int_alloc_entry,
 123                            u8 msix_vector)
 124 {
 125         u32 ivar, index;
 126
 127         msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 128         index = (int_alloc_entry >> 2) & 0x1F;
 129         ivar = IXGBE_READ_REG(&adapter->hw, IXGBE_IVAR(index));
 130         ivar &= ~(0xFF << (8 * (int_alloc_entry & 0x3)));
 131         ivar |= (msix_vector << (8 * (int_alloc_entry & 0x3)));
 132         IXGBE_WRITE_REG(&adapter->hw, IXGBE_IVAR(index), ivar);
 133 }
 134
 135 static void ixgbe_unmap_and_free_tx_resource(struct ixgbe_adapter *adapter,
 136                                              struct ixgbe_tx_buffer
 137                                              *tx_buffer_info)
 138 {
 139         if (tx_buffer_info->dma) {
 140                 pci_unmap_page(adapter->pdev, tx_buffer_info->dma,
 141                                tx_buffer_info->length, PCI_DMA_TODEVICE);
 142                 tx_buffer_info->dma = 0;
 143         }
 144         if (tx_buffer_info->skb) {
 145                 dev_kfree_skb_any(tx_buffer_info->skb);
 146                 tx_buffer_info->skb = NULL;
 147         }
 148         /* tx_buffer_info must be completely set up in the transmit path */
 149 }
 150
 151 static inline bool ixgbe_check_tx_hang(struct ixgbe_adapter *adapter,
 152                                        struct ixgbe_ring *tx_ring,
 153                                        unsigned int eop)
 154 {
 155         struct ixgbe_hw *hw = &adapter->hw;
 156         u32 head, tail;
 157
 158         /* Detect a transmit hang in hardware, this serializes the
 159          * check with the clearing of time_stamp and movement of eop */
 160         head = IXGBE_READ_REG(hw, tx_ring->head);
 161         tail = IXGBE_READ_REG(hw, tx_ring->tail);
 162         adapter->detect_tx_hung = false;
 163         if ((head != tail) &&
 164             tx_ring->tx_buffer_info[eop].time_stamp &&
 165             time_after(jiffies, tx_ring->tx_buffer_info[eop].time_stamp + HZ) &&
 166             !(IXGBE_READ_REG(&adapter->hw, IXGBE_TFCS) & IXGBE_TFCS_TXOFF)) {
 167                 /* detected Tx unit hang */
 168                 union ixgbe_adv_tx_desc *tx_desc;
 169                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
 170                 DPRINTK(DRV, ERR, "Detected Tx Unit Hang\n"
 171                         "  Tx Queue             <%d>\n"
 172                         "  TDH, TDT             <%x>, <%x>\n"
 173                         "  next_to_use          <%x>\n"
 174                         "  next_to_clean        <%x>\n"
 175                         "tx_buffer_info[next_to_clean]\n"
 176                         "  time_stamp           <%lx>\n"
 177                         "  jiffies              <%lx>\n",
 178                         tx_ring->queue_index,
 179                         head, tail,
 180                         tx_ring->next_to_use, eop,
 181                         tx_ring->tx_buffer_info[eop].time_stamp, jiffies);
 182                 return true;
 183         }
 184
 185         return false;
 186 }
 187
 188 #define IXGBE_MAX_TXD_PWR       14
 189 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 190
 191 /* Tx Descriptors needed, worst case */
 192 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 193                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 194 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 195         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1) /* for context */
 196
 197 #define GET_TX_HEAD_FROM_RING(ring) (\
 198         *(volatile u32 *) \
 199         ((union ixgbe_adv_tx_desc *)(ring)->desc + (ring)->count))
 200 static void ixgbe_tx_timeout(struct net_device *netdev);
 201
 202 /**
 203  * ixgbe_clean_tx_irq - Reclaim resources after transmit completes
 204  * @adapter: board private structure
 205  * @tx_ring: tx ring to clean
 206  **/
 207 static bool ixgbe_clean_tx_irq(struct ixgbe_adapter *adapter,
 208                                struct ixgbe_ring *tx_ring)
 209 {
 210         union ixgbe_adv_tx_desc *tx_desc;
 211         struct ixgbe_tx_buffer *tx_buffer_info;
 212         struct net_device *netdev = adapter->netdev;
 213         struct sk_buff *skb;
 214         unsigned int i;
 215         u32 head, oldhead;
 216         unsigned int count = 0;
 217         unsigned int total_bytes = 0, total_packets = 0;
 218
 219         rmb();
 220         head = GET_TX_HEAD_FROM_RING(tx_ring);
 221         head = le32_to_cpu(head);
 222         i = tx_ring->next_to_clean;
 223         while (1) {
 224                 while (i != head) {
 225                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
 226                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 227                         skb = tx_buffer_info->skb;
 228
 229                         if (skb) {
 230                                 unsigned int segs, bytecount;
 231
 232                                 /* gso_segs is currently only valid for tcp */
 233                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 234                                 /* multiply data chunks by size of headers */
 235                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 236                                             skb->len;
 237                                 total_packets += segs;
 238                                 total_bytes += bytecount;
 239                         }
 240
 241                         ixgbe_unmap_and_free_tx_resource(adapter,
 242                                                          tx_buffer_info);
 243
 244                         i++;
 245                         if (i == tx_ring->count)
 246                                 i = 0;
 247
 248                         count++;
 249                         if (count == tx_ring->count)
 250                                 goto done_cleaning;
 251                 }
 252                 oldhead = head;
 253                 rmb();
 254                 head = GET_TX_HEAD_FROM_RING(tx_ring);
 255                 head = le32_to_cpu(head);
 256                 if (head == oldhead)
 257                         goto done_cleaning;
 258         } /* while (1) */
 259
 260 done_cleaning:
 261         tx_ring->next_to_clean = i;
 262
 263 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 264         if (unlikely(count && netif_carrier_ok(netdev) &&
 265                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 266                 /* Make sure that anybody stopping the queue after this
 267                  * sees the new next_to_clean.
 268                  */
 269                 smp_mb();
 270                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
 271                     !test_bit(__IXGBE_DOWN, &adapter->state)) {
 272                         netif_wake_subqueue(netdev, tx_ring->queue_index);
 273                         ++adapter->restart_queue;
 274                 }
 275         }
 276
 277         if (adapter->detect_tx_hung) {
 278                 if (ixgbe_check_tx_hang(adapter, tx_ring, i)) {
 279                         /* schedule immediate reset if we believe we hung */
 280                         DPRINTK(PROBE, INFO,
 281                                 "tx hang %d detected, resetting adapter\n",
 282                                 adapter->tx_timeout_count + 1);
 283                         ixgbe_tx_timeout(adapter->netdev);
 284                 }
 285         }
 286
 287         /* re-arm the interrupt */
 288         if ((total_packets >= tx_ring->work_limit) ||
 289             (count == tx_ring->count))
 290                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EICS, tx_ring->v_idx);
 291
 292         tx_ring->total_bytes += total_bytes;
 293         tx_ring->total_packets += total_packets;
 294         tx_ring->stats.bytes += total_bytes;
 295         tx_ring->stats.packets += total_packets;
 296         adapter->net_stats.tx_bytes += total_bytes;
 297         adapter->net_stats.tx_packets += total_packets;
 298         return (total_packets ? true : false);
 299 }
 300
 301 #ifdef CONFIG_IXGBE_DCA
 302 static void ixgbe_update_rx_dca(struct ixgbe_adapter *adapter,
 303                                 struct ixgbe_ring *rx_ring)
 304 {
 305         u32 rxctrl;
 306         int cpu = get_cpu();
 307         int q = rx_ring - adapter->rx_ring;
 308
 309         if (rx_ring->cpu != cpu) {
 310                 rxctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q));
 311                 rxctrl &= ~IXGBE_DCA_RXCTRL_CPUID_MASK;
 312                 rxctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 313                 rxctrl |= IXGBE_DCA_RXCTRL_DESC_DCA_EN;
 314                 rxctrl |= IXGBE_DCA_RXCTRL_HEAD_DCA_EN;
 315                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_RXCTRL(q), rxctrl);
 316                 rx_ring->cpu = cpu;
 317         }
 318         put_cpu();
 319 }
 320
 321 static void ixgbe_update_tx_dca(struct ixgbe_adapter *adapter,
 322                                 struct ixgbe_ring *tx_ring)
 323 {
 324         u32 txctrl;
 325         int cpu = get_cpu();
 326         int q = tx_ring - adapter->tx_ring;
 327
 328         if (tx_ring->cpu != cpu) {
 329                 txctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q));
 330                 txctrl &= ~IXGBE_DCA_TXCTRL_CPUID_MASK;
 331                 txctrl |= dca3_get_tag(&adapter->pdev->dev, cpu);
 332                 txctrl |= IXGBE_DCA_TXCTRL_DESC_DCA_EN;
 333                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_TXCTRL(q), txctrl);
 334                 tx_ring->cpu = cpu;
 335         }
 336         put_cpu();
 337 }
 338
 339 static void ixgbe_setup_dca(struct ixgbe_adapter *adapter)
 340 {
 341         int i;
 342
 343         if (!(adapter->flags & IXGBE_FLAG_DCA_ENABLED))
 344                 return;
 345
 346         for (i = 0; i < adapter->num_tx_queues; i++) {
 347                 adapter->tx_ring[i].cpu = -1;
 348                 ixgbe_update_tx_dca(adapter, &adapter->tx_ring[i]);
 349         }
 350         for (i = 0; i < adapter->num_rx_queues; i++) {
 351                 adapter->rx_ring[i].cpu = -1;
 352                 ixgbe_update_rx_dca(adapter, &adapter->rx_ring[i]);
 353         }
 354 }
 355
 356 static int __ixgbe_notify_dca(struct device *dev, void *data)
 357 {
 358         struct net_device *netdev = dev_get_drvdata(dev);
 359         struct ixgbe_adapter *adapter = netdev_priv(netdev);
 360         unsigned long event = *(unsigned long *)data;
 361
 362         switch (event) {
 363         case DCA_PROVIDER_ADD:
 364                 /* if we're already enabled, don't do it again */
 365                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
 366                         break;
 367                 /* Always use CB2 mode, difference is masked
 368                  * in the CB driver. */
 369                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 2);
 370                 if (dca_add_requester(dev) == 0) {
 371                         adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
 372                         ixgbe_setup_dca(adapter);
 373                         break;
 374                 }
 375                 /* Fall Through since DCA is disabled. */
 376         case DCA_PROVIDER_REMOVE:
 377                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
 378                         dca_remove_requester(dev);
 379                         adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
 380                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
 381                 }
 382                 break;
 383         }
 384
 385         return 0;
 386 }
 387
 388 #endif /* CONFIG_IXGBE_DCA */
 389 /**
 390  * ixgbe_receive_skb - Send a completed packet up the stack
 391  * @adapter: board private structure
 392  * @skb: packet to send up
 393  * @status: hardware indication of status of receive
 394  * @rx_ring: rx descriptor ring (for a specific queue) to setup
 395  * @rx_desc: rx descriptor
 396  **/
 397 static void ixgbe_receive_skb(struct ixgbe_adapter *adapter,
 398                               struct sk_buff *skb, u8 status,
 399                               struct ixgbe_ring *ring,
 400                               union ixgbe_adv_rx_desc *rx_desc)
 401 {
 402         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 403         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 404
 405         if (adapter->netdev->features & NETIF_F_LRO &&
 406             skb->ip_summed == CHECKSUM_UNNECESSARY) {
 407                 if (adapter->vlgrp && is_vlan)
 408                         lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
 409                                                      adapter->vlgrp, tag,
 410                                                      rx_desc);
 411                 else
 412                         lro_receive_skb(&ring->lro_mgr, skb, rx_desc);
 413                 ring->lro_used = true;
 414         } else {
 415                 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL)) {
 416                         if (adapter->vlgrp && is_vlan)
 417                                 vlan_hwaccel_receive_skb(skb, adapter->vlgrp, tag);
 418                         else
 419                                 netif_receive_skb(skb);
 420                 } else {
 421                         if (adapter->vlgrp && is_vlan)
 422                                 vlan_hwaccel_rx(skb, adapter->vlgrp, tag);
 423                         else
 424                                 netif_rx(skb);
 425                 }
 426         }
 427 }
 428
 429 /**
 430  * ixgbe_rx_checksum - indicate in skb if hw indicated a good cksum
 431  * @adapter: address of board private structure
 432  * @status_err: hardware indication of status of receive
 433  * @skb: skb currently being received and modified
 434  **/
 435 static inline void ixgbe_rx_checksum(struct ixgbe_adapter *adapter,
 436                                      u32 status_err, struct sk_buff *skb)
 437 {
 438         skb->ip_summed = CHECKSUM_NONE;
 439
 440         /* Rx csum disabled */
 441         if (!(adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED))
 442                 return;
 443
 444         /* if IP and error */
 445         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 446             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 447                 adapter->hw_csum_rx_error++;
 448                 return;
 449         }
 450
 451         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 452                 return;
 453
 454         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 455                 adapter->hw_csum_rx_error++;
 456                 return;
 457         }
 458
 459         /* It must be a TCP or UDP packet with a valid checksum */
 460         skb->ip_summed = CHECKSUM_UNNECESSARY;
 461         adapter->hw_csum_rx_good++;
 462 }
 463
 464 /**
 465  * ixgbe_alloc_rx_buffers - Replace used receive buffers; packet split
 466  * @adapter: address of board private structure
 467  **/
 468 static void ixgbe_alloc_rx_buffers(struct ixgbe_adapter *adapter,
 469                                    struct ixgbe_ring *rx_ring,
 470                                    int cleaned_count)
 471 {
 472         struct pci_dev *pdev = adapter->pdev;
 473         union ixgbe_adv_rx_desc *rx_desc;
 474         struct ixgbe_rx_buffer *bi;
 475         unsigned int i;
 476         unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
 477
 478         i = rx_ring->next_to_use;
 479         bi = &rx_ring->rx_buffer_info[i];
 480
 481         while (cleaned_count--) {
 482                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 483
 484                 if (!bi->page_dma &&
 485                     (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED)) {
 486                         if (!bi->page) {
 487                                 bi->page = alloc_page(GFP_ATOMIC);
 488                                 if (!bi->page) {
 489                                         adapter->alloc_rx_page_failed++;
 490                                         goto no_buffers;
 491                                 }
 492                                 bi->page_offset = 0;
 493                         } else {
 494                                 /* use a half page if we're re-using */
 495                                 bi->page_offset ^= (PAGE_SIZE / 2);
 496                         }
 497
 498                         bi->page_dma = pci_map_page(pdev, bi->page,
 499                                                     bi->page_offset,
 500                                                     (PAGE_SIZE / 2),
 501                                                     PCI_DMA_FROMDEVICE);
 502                 }
 503
 504                 if (!bi->skb) {
 505                         struct sk_buff *skb = netdev_alloc_skb(adapter->netdev,
 506                                                                bufsz);
 507
 508                         if (!skb) {
 509                                 adapter->alloc_rx_buff_failed++;
 510                                 goto no_buffers;
 511                         }
 512
 513                         /*
 514                          * Make buffer alignment 2 beyond a 16 byte boundary
 515                          * this will result in a 16 byte aligned IP header after
 516                          * the 14 byte MAC header is removed
 517                          */
 518                         skb_reserve(skb, NET_IP_ALIGN);
 519
 520                         bi->skb = skb;
 521                         bi->dma = pci_map_single(pdev, skb->data, bufsz,
 522                                                  PCI_DMA_FROMDEVICE);
 523                 }
 524                 /* Refresh the desc even if buffer_addrs didn't change because
 525                  * each write-back erases this info. */
 526                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
 527                         rx_desc->read.pkt_addr = cpu_to_le64(bi->page_dma);
 528                         rx_desc->read.hdr_addr = cpu_to_le64(bi->dma);
 529                 } else {
 530                         rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
 531                 }
 532
 533                 i++;
 534                 if (i == rx_ring->count)
 535                         i = 0;
 536                 bi = &rx_ring->rx_buffer_info[i];
 537         }
 538
 539 no_buffers:
 540         if (rx_ring->next_to_use != i) {
 541                 rx_ring->next_to_use = i;
 542                 if (i-- == 0)
 543                         i = (rx_ring->count - 1);
 544
 545                 /*
 546                  * Force memory writes to complete before letting h/w
 547                  * know there are new descriptors to fetch.  (Only
 548                  * applicable for weak-ordered memory model archs,
 549                  * such as IA-64).
 550                  */
 551                 wmb();
 552                 writel(i, adapter->hw.hw_addr + rx_ring->tail);
 553         }
 554 }
 555
 556 static inline u16 ixgbe_get_hdr_info(union ixgbe_adv_rx_desc *rx_desc)
 557 {
 558         return rx_desc->wb.lower.lo_dword.hs_rss.hdr_info;
 559 }
 560
 561 static inline u16 ixgbe_get_pkt_info(union ixgbe_adv_rx_desc *rx_desc)
 562 {
 563         return rx_desc->wb.lower.lo_dword.hs_rss.pkt_info;
 564 }
 565
 566 static bool ixgbe_clean_rx_irq(struct ixgbe_adapter *adapter,
 567                                struct ixgbe_ring *rx_ring,
 568                                int *work_done, int work_to_do)
 569 {
 570         struct pci_dev *pdev = adapter->pdev;
 571         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 572         struct ixgbe_rx_buffer *rx_buffer_info, *next_buffer;
 573         struct sk_buff *skb;
 574         unsigned int i;
 575         u32 len, staterr;
 576         u16 hdr_info;
 577         bool cleaned = false;
 578         int cleaned_count = 0;
 579         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 580
 581         i = rx_ring->next_to_clean;
 582         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 583         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 584         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 585
 586         while (staterr & IXGBE_RXD_STAT_DD) {
 587                 u32 upper_len = 0;
 588                 if (*work_done >= work_to_do)
 589                         break;
 590                 (*work_done)++;
 591
 592                 if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
 593                         hdr_info = le16_to_cpu(ixgbe_get_hdr_info(rx_desc));
 594                         len = (hdr_info & IXGBE_RXDADV_HDRBUFLEN_MASK) >>
 595                                IXGBE_RXDADV_HDRBUFLEN_SHIFT;
 596                         if (hdr_info & IXGBE_RXDADV_SPH)
 597                                 adapter->rx_hdr_split++;
 598                         if (len > IXGBE_RX_HDR_SIZE)
 599                                 len = IXGBE_RX_HDR_SIZE;
 600                         upper_len = le16_to_cpu(rx_desc->wb.upper.length);
 601                 } else {
 602                         len = le16_to_cpu(rx_desc->wb.upper.length);
 603                 }
 604
 605                 cleaned = true;
 606                 skb = rx_buffer_info->skb;
 607                 prefetch(skb->data - NET_IP_ALIGN);
 608                 rx_buffer_info->skb = NULL;
 609
 610                 if (len && !skb_shinfo(skb)->nr_frags) {
 611                         pci_unmap_single(pdev, rx_buffer_info->dma,
 612                                          rx_ring->rx_buf_len + NET_IP_ALIGN,
 613                                          PCI_DMA_FROMDEVICE);
 614                         skb_put(skb, len);
 615                 }
 616
 617                 if (upper_len) {
 618                         pci_unmap_page(pdev, rx_buffer_info->page_dma,
 619                                        PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
 620                         rx_buffer_info->page_dma = 0;
 621                         skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags,
 622                                            rx_buffer_info->page,
 623                                            rx_buffer_info->page_offset,
 624                                            upper_len);
 625
 626                         if ((rx_ring->rx_buf_len > (PAGE_SIZE / 2)) ||
 627                             (page_count(rx_buffer_info->page) != 1))
 628                                 rx_buffer_info->page = NULL;
 629                         else
 630                                 get_page(rx_buffer_info->page);
 631
 632                         skb->len += upper_len;
 633                         skb->data_len += upper_len;
 634                         skb->truesize += upper_len;
 635                 }
 636
 637                 i++;
 638                 if (i == rx_ring->count)
 639                         i = 0;
 640                 next_buffer = &rx_ring->rx_buffer_info[i];
 641
 642                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
 643                 prefetch(next_rxd);
 644
 645                 cleaned_count++;
 646                 if (staterr & IXGBE_RXD_STAT_EOP) {
 647                         rx_ring->stats.packets++;
 648                         rx_ring->stats.bytes += skb->len;
 649                 } else {
 650                         rx_buffer_info->skb = next_buffer->skb;
 651                         rx_buffer_info->dma = next_buffer->dma;
 652                         next_buffer->skb = skb;
 653                         next_buffer->dma = 0;
 654                         adapter->non_eop_descs++;
 655                         goto next_desc;
 656                 }
 657
 658                 if (staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK) {
 659                         dev_kfree_skb_irq(skb);
 660                         goto next_desc;
 661                 }
 662
 663                 ixgbe_rx_checksum(adapter, staterr, skb);
 664
 665                 /* probably a little skewed due to removing CRC */
 666                 total_rx_bytes += skb->len;
 667                 total_rx_packets++;
 668
 669                 skb->protocol = eth_type_trans(skb, adapter->netdev);
 670                 ixgbe_receive_skb(adapter, skb, staterr, rx_ring, rx_desc);
 671                 adapter->netdev->last_rx = jiffies;
 672
 673 next_desc:
 674                 rx_desc->wb.upper.status_error = 0;
 675
 676                 /* return some buffers to hardware, one at a time is too slow */
 677                 if (cleaned_count >= IXGBE_RX_BUFFER_WRITE) {
 678                         ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
 679                         cleaned_count = 0;
 680                 }
 681
 682                 /* use prefetched values */
 683                 rx_desc = next_rxd;
 684                 rx_buffer_info = next_buffer;
 685
 686                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 687         }
 688
 689         if (rx_ring->lro_used) {
 690                 lro_flush_all(&rx_ring->lro_mgr);
 691                 rx_ring->lro_used = false;
 692         }
 693
 694         rx_ring->next_to_clean = i;
 695         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
 696
 697         if (cleaned_count)
 698                 ixgbe_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
 699
 700         rx_ring->total_packets += total_rx_packets;
 701         rx_ring->total_bytes += total_rx_bytes;
 702         adapter->net_stats.rx_bytes += total_rx_bytes;
 703         adapter->net_stats.rx_packets += total_rx_packets;
 704
 705         return cleaned;
 706 }
 707
 708 static int ixgbe_clean_rxonly(struct napi_struct *, int);
 709 /**
 710  * ixgbe_configure_msix - Configure MSI-X hardware
 711  * @adapter: board private structure
 712  *
 713  * ixgbe_configure_msix sets up the hardware to properly generate MSI-X
 714  * interrupts.
 715  **/
 716 static void ixgbe_configure_msix(struct ixgbe_adapter *adapter)
 717 {
 718         struct ixgbe_q_vector *q_vector;
 719         int i, j, q_vectors, v_idx, r_idx;
 720         u32 mask;
 721
 722         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 723
 724         /* Populate the IVAR table and set the ITR values to the
 725          * corresponding register.
 726          */
 727         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
 728                 q_vector = &adapter->q_vector[v_idx];
 729                 /* XXX for_each_bit(...) */
 730                 r_idx = find_first_bit(q_vector->rxr_idx,
 731                                        adapter->num_rx_queues);
 732
 733                 for (i = 0; i < q_vector->rxr_count; i++) {
 734                         j = adapter->rx_ring[r_idx].reg_idx;
 735                         ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(j), v_idx);
 736                         r_idx = find_next_bit(q_vector->rxr_idx,
 737                                               adapter->num_rx_queues,
 738                                               r_idx + 1);
 739                 }
 740                 r_idx = find_first_bit(q_vector->txr_idx,
 741                                        adapter->num_tx_queues);
 742
 743                 for (i = 0; i < q_vector->txr_count; i++) {
 744                         j = adapter->tx_ring[r_idx].reg_idx;
 745                         ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(j), v_idx);
 746                         r_idx = find_next_bit(q_vector->txr_idx,
 747                                               adapter->num_tx_queues,
 748                                               r_idx + 1);
 749                 }
 750
 751                 /* if this is a tx only vector halve the interrupt rate */
 752                 if (q_vector->txr_count && !q_vector->rxr_count)
 753                         q_vector->eitr = (adapter->eitr_param >> 1);
 754                 else
 755                         /* rx only */
 756                         q_vector->eitr = adapter->eitr_param;
 757
 758                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx),
 759                                 EITR_INTS_PER_SEC_TO_REG(q_vector->eitr));
 760         }
 761
 762         ixgbe_set_ivar(adapter, IXGBE_IVAR_OTHER_CAUSES_INDEX, v_idx);
 763         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EITR(v_idx), 1950);
 764
 765         /* set up to autoclear timer, and the vectors */
 766         mask = IXGBE_EIMS_ENABLE_MASK;
 767         mask &= ~(IXGBE_EIMS_OTHER | IXGBE_EIMS_LSC);
 768         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIAC, mask);
 769 }
 770
 771 enum latency_range {
 772         lowest_latency = 0,
 773         low_latency = 1,
 774         bulk_latency = 2,
 775         latency_invalid = 255
 776 };
 777
 778 /**
 779  * ixgbe_update_itr - update the dynamic ITR value based on statistics
 780  * @adapter: pointer to adapter
 781  * @eitr: eitr setting (ints per sec) to give last timeslice
 782  * @itr_setting: current throttle rate in ints/second
 783  * @packets: the number of packets during this measurement interval
 784  * @bytes: the number of bytes during this measurement interval
 785  *
 786  *      Stores a new ITR value based on packets and byte
 787  *      counts during the last interrupt.  The advantage of per interrupt
 788  *      computation is faster updates and more accurate ITR for the current
 789  *      traffic pattern.  Constants in this function were computed
 790  *      based on theoretical maximum wire speed and thresholds were set based
 791  *      on testing data as well as attempting to minimize response time
 792  *      while increasing bulk throughput.
 793  *      this functionality is controlled by the InterruptThrottleRate module
 794  *      parameter (see ixgbe_param.c)
 795  **/
 796 static u8 ixgbe_update_itr(struct ixgbe_adapter *adapter,
 797                            u32 eitr, u8 itr_setting,
 798                            int packets, int bytes)
 799 {
 800         unsigned int retval = itr_setting;
 801         u32 timepassed_us;
 802         u64 bytes_perint;
 803
 804         if (packets == 0)
 805                 goto update_itr_done;
 806
 807
 808         /* simple throttlerate management
 809          *    0-20MB/s lowest (100000 ints/s)
 810          *   20-100MB/s low   (20000 ints/s)
 811          *  100-1249MB/s bulk (8000 ints/s)
 812          */
 813         /* what was last interrupt timeslice? */
 814         timepassed_us = 1000000/eitr;
 815         bytes_perint = bytes / timepassed_us; /* bytes/usec */
 816
 817         switch (itr_setting) {
 818         case lowest_latency:
 819                 if (bytes_perint > adapter->eitr_low)
 820                         retval = low_latency;
 821                 break;
 822         case low_latency:
 823                 if (bytes_perint > adapter->eitr_high)
 824                         retval = bulk_latency;
 825                 else if (bytes_perint <= adapter->eitr_low)
 826                         retval = lowest_latency;
 827                 break;
 828         case bulk_latency:
 829                 if (bytes_perint <= adapter->eitr_high)
 830                         retval = low_latency;
 831                 break;
 832         }
 833
 834 update_itr_done:
 835         return retval;
 836 }
 837
 838 static void ixgbe_set_itr_msix(struct ixgbe_q_vector *q_vector)
 839 {
 840         struct ixgbe_adapter *adapter = q_vector->adapter;
 841         struct ixgbe_hw *hw = &adapter->hw;
 842         u32 new_itr;
 843         u8 current_itr, ret_itr;
 844         int i, r_idx, v_idx = ((void *)q_vector - (void *)(adapter->q_vector)) /
 845                                sizeof(struct ixgbe_q_vector);
 846         struct ixgbe_ring *rx_ring, *tx_ring;
 847
 848         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
 849         for (i = 0; i < q_vector->txr_count; i++) {
 850                 tx_ring = &(adapter->tx_ring[r_idx]);
 851                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
 852                                            q_vector->tx_itr,
 853                                            tx_ring->total_packets,
 854                                            tx_ring->total_bytes);
 855                 /* if the result for this queue would decrease interrupt
 856                  * rate for this vector then use that result */
 857                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
 858                                     q_vector->tx_itr - 1 : ret_itr);
 859                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
 860                                       r_idx + 1);
 861         }
 862
 863         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 864         for (i = 0; i < q_vector->rxr_count; i++) {
 865                 rx_ring = &(adapter->rx_ring[r_idx]);
 866                 ret_itr = ixgbe_update_itr(adapter, q_vector->eitr,
 867                                            q_vector->rx_itr,
 868                                            rx_ring->total_packets,
 869                                            rx_ring->total_bytes);
 870                 /* if the result for this queue would decrease interrupt
 871                  * rate for this vector then use that result */
 872                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
 873                                     q_vector->rx_itr - 1 : ret_itr);
 874                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
 875                                       r_idx + 1);
 876         }
 877
 878         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
 879
 880         switch (current_itr) {
 881         /* counts and packets in update_itr are dependent on these numbers */
 882         case lowest_latency:
 883                 new_itr = 100000;
 884                 break;
 885         case low_latency:
 886                 new_itr = 20000; /* aka hwitr = ~200 */
 887                 break;
 888         case bulk_latency:
 889         default:
 890                 new_itr = 8000;
 891                 break;
 892         }
 893
 894         if (new_itr != q_vector->eitr) {
 895                 u32 itr_reg;
 896                 /* do an exponential smoothing */
 897                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
 898                 q_vector->eitr = new_itr;
 899                 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
 900                 /* must write high and low 16 bits to reset counter */
 901                 DPRINTK(TX_ERR, DEBUG, "writing eitr(%d): %08X\n", v_idx,
 902                         itr_reg);
 903                 IXGBE_WRITE_REG(hw, IXGBE_EITR(v_idx), itr_reg | (itr_reg)<<16);
 904         }
 905
 906         return;
 907 }
 908
 909 static void ixgbe_check_fan_failure(struct ixgbe_adapter *adapter, u32 eicr)
 910 {
 911         struct ixgbe_hw *hw = &adapter->hw;
 912
 913         if ((adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) &&
 914             (eicr & IXGBE_EICR_GPI_SDP1)) {
 915                 DPRINTK(PROBE, CRIT, "Fan has stopped, replace the adapter\n");
 916                 /* write to clear the interrupt */
 917                 IXGBE_WRITE_REG(hw, IXGBE_EICR, IXGBE_EICR_GPI_SDP1);
 918         }
 919 }
 920
 921 static void ixgbe_check_lsc(struct ixgbe_adapter *adapter)
 922 {
 923         struct ixgbe_hw *hw = &adapter->hw;
 924
 925         adapter->lsc_int++;
 926         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
 927         adapter->link_check_timeout = jiffies;
 928         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
 929                 IXGBE_WRITE_REG(hw, IXGBE_EIMC, IXGBE_EIMC_LSC);
 930                 schedule_work(&adapter->watchdog_task);
 931         }
 932 }
 933
 934 static irqreturn_t ixgbe_msix_lsc(int irq, void *data)
 935 {
 936         struct net_device *netdev = data;
 937         struct ixgbe_adapter *adapter = netdev_priv(netdev);
 938         struct ixgbe_hw *hw = &adapter->hw;
 939         u32 eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
 940
 941         if (eicr & IXGBE_EICR_LSC)
 942                 ixgbe_check_lsc(adapter);
 943
 944         ixgbe_check_fan_failure(adapter, eicr);
 945
 946         if (!test_bit(__IXGBE_DOWN, &adapter->state))
 947                 IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMS_OTHER);
 948
 949         return IRQ_HANDLED;
 950 }
 951
 952 static irqreturn_t ixgbe_msix_clean_tx(int irq, void *data)
 953 {
 954         struct ixgbe_q_vector *q_vector = data;
 955         struct ixgbe_adapter  *adapter = q_vector->adapter;
 956         struct ixgbe_ring     *tx_ring;
 957         int i, r_idx;
 958
 959         if (!q_vector->txr_count)
 960                 return IRQ_HANDLED;
 961
 962         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
 963         for (i = 0; i < q_vector->txr_count; i++) {
 964                 tx_ring = &(adapter->tx_ring[r_idx]);
 965 #ifdef CONFIG_IXGBE_DCA
 966                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
 967                         ixgbe_update_tx_dca(adapter, tx_ring);
 968 #endif
 969                 tx_ring->total_bytes = 0;
 970                 tx_ring->total_packets = 0;
 971                 ixgbe_clean_tx_irq(adapter, tx_ring);
 972                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
 973                                       r_idx + 1);
 974         }
 975
 976         return IRQ_HANDLED;
 977 }
 978
 979 /**
 980  * ixgbe_msix_clean_rx - single unshared vector rx clean (all queues)
 981  * @irq: unused
 982  * @data: pointer to our q_vector struct for this interrupt vector
 983  **/
 984 static irqreturn_t ixgbe_msix_clean_rx(int irq, void *data)
 985 {
 986         struct ixgbe_q_vector *q_vector = data;
 987         struct ixgbe_adapter  *adapter = q_vector->adapter;
 988         struct ixgbe_ring  *rx_ring;
 989         int r_idx;
 990         int i;
 991
 992         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 993         for (i = 0;  i < q_vector->rxr_count; i++) {
 994                 rx_ring = &(adapter->rx_ring[r_idx]);
 995                 rx_ring->total_bytes = 0;
 996                 rx_ring->total_packets = 0;
 997                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
 998                                       r_idx + 1);
 999         }
1000
1001         if (!q_vector->rxr_count)
1002                 return IRQ_HANDLED;
1003
1004         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1005         rx_ring = &(adapter->rx_ring[r_idx]);
1006         /* disable interrupts on this vector only */
1007         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, rx_ring->v_idx);
1008         netif_rx_schedule(adapter->netdev, &q_vector->napi);
1009
1010         return IRQ_HANDLED;
1011 }
1012
1013 static irqreturn_t ixgbe_msix_clean_many(int irq, void *data)
1014 {
1015         ixgbe_msix_clean_rx(irq, data);
1016         ixgbe_msix_clean_tx(irq, data);
1017
1018         return IRQ_HANDLED;
1019 }
1020
1021 /**
1022  * ixgbe_clean_rxonly - msix (aka one shot) rx clean routine
1023  * @napi: napi struct with our devices info in it
1024  * @budget: amount of work driver is allowed to do this pass, in packets
1025  *
1026  * This function is optimized for cleaning one queue only on a single
1027  * q_vector!!!
1028  **/
1029 static int ixgbe_clean_rxonly(struct napi_struct *napi, int budget)
1030 {
1031         struct ixgbe_q_vector *q_vector =
1032                                container_of(napi, struct ixgbe_q_vector, napi);
1033         struct ixgbe_adapter *adapter = q_vector->adapter;
1034         struct ixgbe_ring *rx_ring = NULL;
1035         int work_done = 0;
1036         long r_idx;
1037
1038         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1039         rx_ring = &(adapter->rx_ring[r_idx]);
1040 #ifdef CONFIG_IXGBE_DCA
1041         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1042                 ixgbe_update_rx_dca(adapter, rx_ring);
1043 #endif
1044
1045         ixgbe_clean_rx_irq(adapter, rx_ring, &work_done, budget);
1046
1047         /* If all Rx work done, exit the polling mode */
1048         if (work_done < budget) {
1049                 netif_rx_complete(adapter->netdev, napi);
1050                 if (adapter->itr_setting & 3)
1051                         ixgbe_set_itr_msix(q_vector);
1052                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1053                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, rx_ring->v_idx);
1054         }
1055
1056         return work_done;
1057 }
1058
1059 /**
1060  * ixgbe_clean_rxonly_many - msix (aka one shot) rx clean routine
1061  * @napi: napi struct with our devices info in it
1062  * @budget: amount of work driver is allowed to do this pass, in packets
1063  *
1064  * This function will clean more than one rx queue associated with a
1065  * q_vector.
1066  **/
1067 static int ixgbe_clean_rxonly_many(struct napi_struct *napi, int budget)
1068 {
1069         struct ixgbe_q_vector *q_vector =
1070                                container_of(napi, struct ixgbe_q_vector, napi);
1071         struct ixgbe_adapter *adapter = q_vector->adapter;
1072         struct ixgbe_ring *rx_ring = NULL;
1073         int work_done = 0, i;
1074         long r_idx;
1075         u16 enable_mask = 0;
1076
1077         /* attempt to distribute budget to each queue fairly, but don't allow
1078          * the budget to go below 1 because we'll exit polling */
1079         budget /= (q_vector->rxr_count ?: 1);
1080         budget = max(budget, 1);
1081         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1082         for (i = 0; i < q_vector->rxr_count; i++) {
1083                 rx_ring = &(adapter->rx_ring[r_idx]);
1084 #ifdef CONFIG_IXGBE_DCA
1085                 if (adapter->flags & IXGBE_FLAG_DCA_ENABLED)
1086                         ixgbe_update_rx_dca(adapter, rx_ring);
1087 #endif
1088                 ixgbe_clean_rx_irq(adapter, rx_ring, &work_done, budget);
1089                 enable_mask |= rx_ring->v_idx;
1090                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
1091                                       r_idx + 1);
1092         }
1093
1094         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
1095         rx_ring = &(adapter->rx_ring[r_idx]);
1096         /* If all Rx work done, exit the polling mode */
1097         if (work_done < budget) {
1098                 netif_rx_complete(adapter->netdev, napi);
1099                 if (adapter->itr_setting & 3)
1100                         ixgbe_set_itr_msix(q_vector);
1101                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
1102                         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, enable_mask);
1103                 return 0;
1104         }
1105
1106         return work_done;
1107 }
1108 static inline void map_vector_to_rxq(struct ixgbe_adapter *a, int v_idx,
1109                                      int r_idx)
1110 {
1111         a->q_vector[v_idx].adapter = a;
1112         set_bit(r_idx, a->q_vector[v_idx].rxr_idx);
1113         a->q_vector[v_idx].rxr_count++;
1114         a->rx_ring[r_idx].v_idx = 1 << v_idx;
1115 }
1116
1117 static inline void map_vector_to_txq(struct ixgbe_adapter *a, int v_idx,
1118                                      int r_idx)
1119 {
1120         a->q_vector[v_idx].adapter = a;
1121         set_bit(r_idx, a->q_vector[v_idx].txr_idx);
1122         a->q_vector[v_idx].txr_count++;
1123         a->tx_ring[r_idx].v_idx = 1 << v_idx;
1124 }
1125
1126 /**
1127  * ixgbe_map_rings_to_vectors - Maps descriptor rings to vectors
1128  * @adapter: board private structure to initialize
1129  * @vectors: allotted vector count for descriptor rings
1130  *
1131  * This function maps descriptor rings to the queue-specific vectors
1132  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
1133  * one vector per ring/queue, but on a constrained vector budget, we
1134  * group the rings as "efficiently" as possible.  You would add new
1135  * mapping configurations in here.
1136  **/
1137 static int ixgbe_map_rings_to_vectors(struct ixgbe_adapter *adapter,
1138                                       int vectors)
1139 {
1140         int v_start = 0;
1141         int rxr_idx = 0, txr_idx = 0;
1142         int rxr_remaining = adapter->num_rx_queues;
1143         int txr_remaining = adapter->num_tx_queues;
1144         int i, j;
1145         int rqpv, tqpv;
1146         int err = 0;
1147
1148         /* No mapping required if MSI-X is disabled. */
1149         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1150                 goto out;
1151
1152         /*
1153          * The ideal configuration...
1154          * We have enough vectors to map one per queue.
1155          */
1156         if (vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
1157                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
1158                         map_vector_to_rxq(adapter, v_start, rxr_idx);
1159
1160                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
1161                         map_vector_to_txq(adapter, v_start, txr_idx);
1162
1163                 goto out;
1164         }
1165
1166         /*
1167          * If we don't have enough vectors for a 1-to-1
1168          * mapping, we'll have to group them so there are
1169          * multiple queues per vector.
1170          */
1171         /* Re-adjusting *qpv takes care of the remainder. */
1172         for (i = v_start; i < vectors; i++) {
1173                 rqpv = DIV_ROUND_UP(rxr_remaining, vectors - i);
1174                 for (j = 0; j < rqpv; j++) {
1175                         map_vector_to_rxq(adapter, i, rxr_idx);
1176                         rxr_idx++;
1177                         rxr_remaining--;
1178                 }
1179         }
1180         for (i = v_start; i < vectors; i++) {
1181                 tqpv = DIV_ROUND_UP(txr_remaining, vectors - i);
1182                 for (j = 0; j < tqpv; j++) {
1183                         map_vector_to_txq(adapter, i, txr_idx);
1184                         txr_idx++;
1185                         txr_remaining--;
1186                 }
1187         }
1188
1189 out:
1190         return err;
1191 }
1192
1193 /**
1194  * ixgbe_request_msix_irqs - Initialize MSI-X interrupts
1195  * @adapter: board private structure
1196  *
1197  * ixgbe_request_msix_irqs allocates MSI-X vectors and requests
1198  * interrupts from the kernel.
1199  **/
1200 static int ixgbe_request_msix_irqs(struct ixgbe_adapter *adapter)
1201 {
1202         struct net_device *netdev = adapter->netdev;
1203         irqreturn_t (*handler)(int, void *);
1204         int i, vector, q_vectors, err;
1205
1206         /* Decrement for Other and TCP Timer vectors */
1207         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1208
1209         /* Map the Tx/Rx rings to the vectors we were allotted. */
1210         err = ixgbe_map_rings_to_vectors(adapter, q_vectors);
1211         if (err)
1212                 goto out;
1213
1214 #define SET_HANDLER(_v) ((!(_v)->rxr_count) ? &ixgbe_msix_clean_tx : \
1215                          (!(_v)->txr_count) ? &ixgbe_msix_clean_rx : \
1216                          &ixgbe_msix_clean_many)
1217         for (vector = 0; vector < q_vectors; vector++) {
1218                 handler = SET_HANDLER(&adapter->q_vector[vector]);
1219                 sprintf(adapter->name[vector], "%s:v%d-%s",
1220                         netdev->name, vector,
1221                         (handler == &ixgbe_msix_clean_rx) ? "Rx" :
1222                          ((handler == &ixgbe_msix_clean_tx) ? "Tx" : "TxRx"));
1223                 err = request_irq(adapter->msix_entries[vector].vector,
1224                                   handler, 0, adapter->name[vector],
1225                                   &(adapter->q_vector[vector]));
1226                 if (err) {
1227                         DPRINTK(PROBE, ERR,
1228                                 "request_irq failed for MSIX interrupt "
1229                                 "Error: %d\n", err);
1230                         goto free_queue_irqs;
1231                 }
1232         }
1233
1234         sprintf(adapter->name[vector], "%s:lsc", netdev->name);
1235         err = request_irq(adapter->msix_entries[vector].vector,
1236                           &ixgbe_msix_lsc, 0, adapter->name[vector], netdev);
1237         if (err) {
1238                 DPRINTK(PROBE, ERR,
1239                         "request_irq for msix_lsc failed: %d\n", err);
1240                 goto free_queue_irqs;
1241         }
1242
1243         return 0;
1244
1245 free_queue_irqs:
1246         for (i = vector - 1; i >= 0; i--)
1247                 free_irq(adapter->msix_entries[--vector].vector,
1248                          &(adapter->q_vector[i]));
1249         adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
1250         pci_disable_msix(adapter->pdev);
1251         kfree(adapter->msix_entries);
1252         adapter->msix_entries = NULL;
1253 out:
1254         return err;
1255 }
1256
1257 static void ixgbe_set_itr(struct ixgbe_adapter *adapter)
1258 {
1259         struct ixgbe_hw *hw = &adapter->hw;
1260         struct ixgbe_q_vector *q_vector = adapter->q_vector;
1261         u8 current_itr;
1262         u32 new_itr = q_vector->eitr;
1263         struct ixgbe_ring *rx_ring = &adapter->rx_ring[0];
1264         struct ixgbe_ring *tx_ring = &adapter->tx_ring[0];
1265
1266         q_vector->tx_itr = ixgbe_update_itr(adapter, new_itr,
1267                                             q_vector->tx_itr,
1268                                             tx_ring->total_packets,
1269                                             tx_ring->total_bytes);
1270         q_vector->rx_itr = ixgbe_update_itr(adapter, new_itr,
1271                                             q_vector->rx_itr,
1272                                             rx_ring->total_packets,
1273                                             rx_ring->total_bytes);
1274
1275         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
1276
1277         switch (current_itr) {
1278         /* counts and packets in update_itr are dependent on these numbers */
1279         case lowest_latency:
1280                 new_itr = 100000;
1281                 break;
1282         case low_latency:
1283                 new_itr = 20000; /* aka hwitr = ~200 */
1284                 break;
1285         case bulk_latency:
1286                 new_itr = 8000;
1287                 break;
1288         default:
1289                 break;
1290         }
1291
1292         if (new_itr != q_vector->eitr) {
1293                 u32 itr_reg;
1294                 /* do an exponential smoothing */
1295                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
1296                 q_vector->eitr = new_itr;
1297                 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
1298                 /* must write high and low 16 bits to reset counter */
1299                 IXGBE_WRITE_REG(hw, IXGBE_EITR(0), itr_reg | (itr_reg)<<16);
1300         }
1301
1302         return;
1303 }
1304
1305 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter);
1306
1307 /**
1308  * ixgbe_intr - legacy mode Interrupt Handler
1309  * @irq: interrupt number
1310  * @data: pointer to a network interface device structure
1311  * @pt_regs: CPU registers structure
1312  **/
1313 static irqreturn_t ixgbe_intr(int irq, void *data)
1314 {
1315         struct net_device *netdev = data;
1316         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1317         struct ixgbe_hw *hw = &adapter->hw;
1318         u32 eicr;
1319
1320         /* for NAPI, using EIAM to auto-mask tx/rx interrupt bits on read
1321          * therefore no explict interrupt disable is necessary */
1322         eicr = IXGBE_READ_REG(hw, IXGBE_EICR);
1323         if (!eicr) {
1324                 /* shared interrupt alert!
1325                  * make sure interrupts are enabled because the read will
1326                  * have disabled interrupts due to EIAM */
1327                 ixgbe_irq_enable(adapter);
1328                 return IRQ_NONE;        /* Not our interrupt */
1329         }
1330
1331         if (eicr & IXGBE_EICR_LSC)
1332                 ixgbe_check_lsc(adapter);
1333
1334         ixgbe_check_fan_failure(adapter, eicr);
1335
1336         if (netif_rx_schedule_prep(netdev, &adapter->q_vector[0].napi)) {
1337                 adapter->tx_ring[0].total_packets = 0;
1338                 adapter->tx_ring[0].total_bytes = 0;
1339                 adapter->rx_ring[0].total_packets = 0;
1340                 adapter->rx_ring[0].total_bytes = 0;
1341                 /* would disable interrupts here but EIAM disabled it */
1342                 __netif_rx_schedule(netdev, &adapter->q_vector[0].napi);
1343         }
1344
1345         return IRQ_HANDLED;
1346 }
1347
1348 static inline void ixgbe_reset_q_vectors(struct ixgbe_adapter *adapter)
1349 {
1350         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1351
1352         for (i = 0; i < q_vectors; i++) {
1353                 struct ixgbe_q_vector *q_vector = &adapter->q_vector[i];
1354                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1355                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1356                 q_vector->rxr_count = 0;
1357                 q_vector->txr_count = 0;
1358         }
1359 }
1360
1361 /**
1362  * ixgbe_request_irq - initialize interrupts
1363  * @adapter: board private structure
1364  *
1365  * Attempts to configure interrupts using the best available
1366  * capabilities of the hardware and kernel.
1367  **/
1368 static int ixgbe_request_irq(struct ixgbe_adapter *adapter)
1369 {
1370         struct net_device *netdev = adapter->netdev;
1371         int err;
1372
1373         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1374                 err = ixgbe_request_msix_irqs(adapter);
1375         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
1376                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, 0,
1377                                   netdev->name, netdev);
1378         } else {
1379                 err = request_irq(adapter->pdev->irq, &ixgbe_intr, IRQF_SHARED,
1380                                   netdev->name, netdev);
1381         }
1382
1383         if (err)
1384                 DPRINTK(PROBE, ERR, "request_irq failed, Error %d\n", err);
1385
1386         return err;
1387 }
1388
1389 static void ixgbe_free_irq(struct ixgbe_adapter *adapter)
1390 {
1391         struct net_device *netdev = adapter->netdev;
1392
1393         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1394                 int i, q_vectors;
1395
1396                 q_vectors = adapter->num_msix_vectors;
1397
1398                 i = q_vectors - 1;
1399                 free_irq(adapter->msix_entries[i].vector, netdev);
1400
1401                 i--;
1402                 for (; i >= 0; i--) {
1403                         free_irq(adapter->msix_entries[i].vector,
1404                                  &(adapter->q_vector[i]));
1405                 }
1406
1407                 ixgbe_reset_q_vectors(adapter);
1408         } else {
1409                 free_irq(adapter->pdev->irq, netdev);
1410         }
1411 }
1412
1413 /**
1414  * ixgbe_irq_disable - Mask off interrupt generation on the NIC
1415  * @adapter: board private structure
1416  **/
1417 static inline void ixgbe_irq_disable(struct ixgbe_adapter *adapter)
1418 {
1419         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMC, ~0);
1420         IXGBE_WRITE_FLUSH(&adapter->hw);
1421         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1422                 int i;
1423                 for (i = 0; i < adapter->num_msix_vectors; i++)
1424                         synchronize_irq(adapter->msix_entries[i].vector);
1425         } else {
1426                 synchronize_irq(adapter->pdev->irq);
1427         }
1428 }
1429
1430 /**
1431  * ixgbe_irq_enable - Enable default interrupt generation settings
1432  * @adapter: board private structure
1433  **/
1434 static inline void ixgbe_irq_enable(struct ixgbe_adapter *adapter)
1435 {
1436         u32 mask;
1437         mask = IXGBE_EIMS_ENABLE_MASK;
1438         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE)
1439                 mask |= IXGBE_EIMS_GPI_SDP1;
1440         IXGBE_WRITE_REG(&adapter->hw, IXGBE_EIMS, mask);
1441         IXGBE_WRITE_FLUSH(&adapter->hw);
1442 }
1443
1444 /**
1445  * ixgbe_configure_msi_and_legacy - Initialize PIN (INTA...) and MSI interrupts
1446  *
1447  **/
1448 static void ixgbe_configure_msi_and_legacy(struct ixgbe_adapter *adapter)
1449 {
1450         struct ixgbe_hw *hw = &adapter->hw;
1451
1452         IXGBE_WRITE_REG(hw, IXGBE_EITR(0),
1453                         EITR_INTS_PER_SEC_TO_REG(adapter->eitr_param));
1454
1455         ixgbe_set_ivar(adapter, IXGBE_IVAR_RX_QUEUE(0), 0);
1456         ixgbe_set_ivar(adapter, IXGBE_IVAR_TX_QUEUE(0), 0);
1457
1458         map_vector_to_rxq(adapter, 0, 0);
1459         map_vector_to_txq(adapter, 0, 0);
1460
1461         DPRINTK(HW, INFO, "Legacy interrupt IVAR setup done\n");
1462 }
1463
1464 /**
1465  * ixgbe_configure_tx - Configure 8259x Transmit Unit after Reset
1466  * @adapter: board private structure
1467  *
1468  * Configure the Tx unit of the MAC after a reset.
1469  **/
1470 static void ixgbe_configure_tx(struct ixgbe_adapter *adapter)
1471 {
1472         u64 tdba, tdwba;
1473         struct ixgbe_hw *hw = &adapter->hw;
1474         u32 i, j, tdlen, txctrl;
1475
1476         /* Setup the HW Tx Head and Tail descriptor pointers */
1477         for (i = 0; i < adapter->num_tx_queues; i++) {
1478                 struct ixgbe_ring *ring = &adapter->tx_ring[i];
1479                 j = ring->reg_idx;
1480                 tdba = ring->dma;
1481                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1482                 IXGBE_WRITE_REG(hw, IXGBE_TDBAL(j),
1483                                 (tdba & DMA_32BIT_MASK));
1484                 IXGBE_WRITE_REG(hw, IXGBE_TDBAH(j), (tdba >> 32));
1485                 tdwba = ring->dma +
1486                         (ring->count * sizeof(union ixgbe_adv_tx_desc));
1487                 tdwba |= IXGBE_TDWBAL_HEAD_WB_ENABLE;
1488                 IXGBE_WRITE_REG(hw, IXGBE_TDWBAL(j), tdwba & DMA_32BIT_MASK);
1489                 IXGBE_WRITE_REG(hw, IXGBE_TDWBAH(j), (tdwba >> 32));
1490                 IXGBE_WRITE_REG(hw, IXGBE_TDLEN(j), tdlen);
1491                 IXGBE_WRITE_REG(hw, IXGBE_TDH(j), 0);
1492                 IXGBE_WRITE_REG(hw, IXGBE_TDT(j), 0);
1493                 adapter->tx_ring[i].head = IXGBE_TDH(j);
1494                 adapter->tx_ring[i].tail = IXGBE_TDT(j);
1495                 /* Disable Tx Head Writeback RO bit, since this hoses
1496                  * bookkeeping if things aren't delivered in order.
1497                  */
1498                 txctrl = IXGBE_READ_REG(hw, IXGBE_DCA_TXCTRL(j));
1499                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1500                 IXGBE_WRITE_REG(hw, IXGBE_DCA_TXCTRL(j), txctrl);
1501         }
1502 }
1503
1504 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1505
1506 static void ixgbe_configure_srrctl(struct ixgbe_adapter *adapter, int index)
1507 {
1508         struct ixgbe_ring *rx_ring;
1509         u32 srrctl;
1510         int queue0;
1511         unsigned long mask;
1512
1513         /* program one srrctl register per VMDq index */
1514         if (adapter->flags & IXGBE_FLAG_VMDQ_ENABLED) {
1515                 long shift, len;
1516                 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1517                 len = sizeof(adapter->ring_feature[RING_F_VMDQ].mask) * 8;
1518                 shift = find_first_bit(&mask, len);
1519                 queue0 = index & mask;
1520                 index = (index & mask) >> shift;
1521         /* program one srrctl per RSS queue since RDRXCTL.MVMEN is enabled */
1522         } else {
1523                 mask = (unsigned long) adapter->ring_feature[RING_F_RSS].mask;
1524                 queue0 = index & mask;
1525                 index = index & mask;
1526         }
1527
1528         rx_ring = &adapter->rx_ring[queue0];
1529
1530         srrctl = IXGBE_READ_REG(&adapter->hw, IXGBE_SRRCTL(index));
1531
1532         srrctl &= ~IXGBE_SRRCTL_BSIZEHDR_MASK;
1533         srrctl &= ~IXGBE_SRRCTL_BSIZEPKT_MASK;
1534
1535         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1536                 srrctl |= IXGBE_RXBUFFER_2048 >> IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1537                 srrctl |= IXGBE_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
1538                 srrctl |= ((IXGBE_RX_HDR_SIZE <<
1539                             IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT) &
1540                            IXGBE_SRRCTL_BSIZEHDR_MASK);
1541         } else {
1542                 srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1543
1544                 if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1545                         srrctl |= IXGBE_RXBUFFER_2048 >>
1546                                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1547                 else
1548                         srrctl |= rx_ring->rx_buf_len >>
1549                                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1550         }
1551         IXGBE_WRITE_REG(&adapter->hw, IXGBE_SRRCTL(index), srrctl);
1552 }
1553
1554 /**
1555  * ixgbe_get_skb_hdr - helper function for LRO header processing
1556  * @skb: pointer to sk_buff to be added to LRO packet
1557  * @iphdr: pointer to ip header structure
1558  * @tcph: pointer to tcp header structure
1559  * @hdr_flags: pointer to header flags
1560  * @priv: private data
1561  **/
1562 static int ixgbe_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
1563                              u64 *hdr_flags, void *priv)
1564 {
1565         union ixgbe_adv_rx_desc *rx_desc = priv;
1566
1567         /* Verify that this is a valid IPv4 TCP packet */
1568         if (!((ixgbe_get_pkt_info(rx_desc) & IXGBE_RXDADV_PKTTYPE_IPV4) &&
1569              (ixgbe_get_pkt_info(rx_desc) & IXGBE_RXDADV_PKTTYPE_TCP)))
1570                 return -1;
1571
1572         /* Set network headers */
1573         skb_reset_network_header(skb);
1574         skb_set_transport_header(skb, ip_hdrlen(skb));
1575         *iphdr = ip_hdr(skb);
1576         *tcph = tcp_hdr(skb);
1577         *hdr_flags = LRO_IPV4 | LRO_TCP;
1578         return 0;
1579 }
1580
1581 #define PAGE_USE_COUNT(S) (((S) >> PAGE_SHIFT) + \
1582                            (((S) & (PAGE_SIZE - 1)) ? 1 : 0))
1583
1584 /**
1585  * ixgbe_configure_rx - Configure 8259x Receive Unit after Reset
1586  * @adapter: board private structure
1587  *
1588  * Configure the Rx unit of the MAC after a reset.
1589  **/
1590 static void ixgbe_configure_rx(struct ixgbe_adapter *adapter)
1591 {
1592         u64 rdba;
1593         struct ixgbe_hw *hw = &adapter->hw;
1594         struct net_device *netdev = adapter->netdev;
1595         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1596         int i, j;
1597         u32 rdlen, rxctrl, rxcsum;
1598         static const u32 seed[10] = { 0xE291D73D, 0x1805EC6C, 0x2A94B30D,
1599                           0xA54F2BEC, 0xEA49AF7C, 0xE214AD3D, 0xB855AABE,
1600                           0x6A3E67EA, 0x14364D17, 0x3BED200D};
1601         u32 fctrl, hlreg0;
1602         u32 pages;
1603         u32 reta = 0, mrqc;
1604         u32 rdrxctl;
1605         int rx_buf_len;
1606
1607         /* Decide whether to use packet split mode or not */
1608         adapter->flags |= IXGBE_FLAG_RX_PS_ENABLED;
1609
1610         /* Set the RX buffer length according to the mode */
1611         if (adapter->flags & IXGBE_FLAG_RX_PS_ENABLED) {
1612                 rx_buf_len = IXGBE_RX_HDR_SIZE;
1613         } else {
1614                 if (netdev->mtu <= ETH_DATA_LEN)
1615                         rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1616                 else
1617                         rx_buf_len = ALIGN(max_frame, 1024);
1618         }
1619
1620         fctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_FCTRL);
1621         fctrl |= IXGBE_FCTRL_BAM;
1622         fctrl |= IXGBE_FCTRL_DPF; /* discard pause frames when FC enabled */
1623         IXGBE_WRITE_REG(&adapter->hw, IXGBE_FCTRL, fctrl);
1624
1625         hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
1626         if (adapter->netdev->mtu <= ETH_DATA_LEN)
1627                 hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
1628         else
1629                 hlreg0 |= IXGBE_HLREG0_JUMBOEN;
1630         IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
1631
1632         pages = PAGE_USE_COUNT(adapter->netdev->mtu);
1633
1634         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1635         /* disable receives while setting up the descriptors */
1636         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1637         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
1638
1639         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1640          * the Base and Length of the Rx Descriptor Ring */
1641         for (i = 0; i < adapter->num_rx_queues; i++) {
1642                 rdba = adapter->rx_ring[i].dma;
1643                 j = adapter->rx_ring[i].reg_idx;
1644                 IXGBE_WRITE_REG(hw, IXGBE_RDBAL(j), (rdba & DMA_32BIT_MASK));
1645                 IXGBE_WRITE_REG(hw, IXGBE_RDBAH(j), (rdba >> 32));
1646                 IXGBE_WRITE_REG(hw, IXGBE_RDLEN(j), rdlen);
1647                 IXGBE_WRITE_REG(hw, IXGBE_RDH(j), 0);
1648                 IXGBE_WRITE_REG(hw, IXGBE_RDT(j), 0);
1649                 adapter->rx_ring[i].head = IXGBE_RDH(j);
1650                 adapter->rx_ring[i].tail = IXGBE_RDT(j);
1651                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1652                 /* Intitial LRO Settings */
1653                 adapter->rx_ring[i].lro_mgr.max_aggr = IXGBE_MAX_LRO_AGGREGATE;
1654                 adapter->rx_ring[i].lro_mgr.max_desc = IXGBE_MAX_LRO_DESCRIPTORS;
1655                 adapter->rx_ring[i].lro_mgr.get_skb_header = ixgbe_get_skb_hdr;
1656                 adapter->rx_ring[i].lro_mgr.features = LRO_F_EXTRACT_VLAN_ID;
1657                 if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
1658                         adapter->rx_ring[i].lro_mgr.features |= LRO_F_NAPI;
1659                 adapter->rx_ring[i].lro_mgr.dev = adapter->netdev;
1660                 adapter->rx_ring[i].lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
1661                 adapter->rx_ring[i].lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1662
1663                 ixgbe_configure_srrctl(adapter, j);
1664         }
1665
1666         /*
1667          * For VMDq support of different descriptor types or
1668          * buffer sizes through the use of multiple SRRCTL
1669          * registers, RDRXCTL.MVMEN must be set to 1
1670          *
1671          * also, the manual doesn't mention it clearly but DCA hints
1672          * will only use queue 0's tags unless this bit is set.  Side
1673          * effects of setting this bit are only that SRRCTL must be
1674          * fully programmed [0..15]
1675          */
1676         rdrxctl = IXGBE_READ_REG(hw, IXGBE_RDRXCTL);
1677         rdrxctl |= IXGBE_RDRXCTL_MVMEN;
1678         IXGBE_WRITE_REG(hw, IXGBE_RDRXCTL, rdrxctl);
1679
1680
1681         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED) {
1682                 /* Fill out redirection table */
1683                 for (i = 0, j = 0; i < 128; i++, j++) {
1684                         if (j == adapter->ring_feature[RING_F_RSS].indices)
1685                                 j = 0;
1686                         /* reta = 4-byte sliding window of
1687                          * 0x00..(indices-1)(indices-1)00..etc. */
1688                         reta = (reta << 8) | (j * 0x11);
1689                         if ((i & 3) == 3)
1690                                 IXGBE_WRITE_REG(hw, IXGBE_RETA(i >> 2), reta);
1691                 }
1692
1693                 /* Fill out hash function seeds */
1694                 for (i = 0; i < 10; i++)
1695                         IXGBE_WRITE_REG(hw, IXGBE_RSSRK(i), seed[i]);
1696
1697                 mrqc = IXGBE_MRQC_RSSEN
1698                     /* Perform hash on these packet types */
1699                        | IXGBE_MRQC_RSS_FIELD_IPV4
1700                        | IXGBE_MRQC_RSS_FIELD_IPV4_TCP
1701                        | IXGBE_MRQC_RSS_FIELD_IPV4_UDP
1702                        | IXGBE_MRQC_RSS_FIELD_IPV6_EX_TCP
1703                        | IXGBE_MRQC_RSS_FIELD_IPV6_EX
1704                        | IXGBE_MRQC_RSS_FIELD_IPV6
1705                        | IXGBE_MRQC_RSS_FIELD_IPV6_TCP
1706                        | IXGBE_MRQC_RSS_FIELD_IPV6_UDP
1707                        | IXGBE_MRQC_RSS_FIELD_IPV6_EX_UDP;
1708                 IXGBE_WRITE_REG(hw, IXGBE_MRQC, mrqc);
1709         }
1710
1711         rxcsum = IXGBE_READ_REG(hw, IXGBE_RXCSUM);
1712
1713         if (adapter->flags & IXGBE_FLAG_RSS_ENABLED ||
1714             adapter->flags & IXGBE_FLAG_RX_CSUM_ENABLED) {
1715                 /* Disable indicating checksum in descriptor, enables
1716                  * RSS hash */
1717                 rxcsum |= IXGBE_RXCSUM_PCSD;
1718         }
1719         if (!(rxcsum & IXGBE_RXCSUM_PCSD)) {
1720                 /* Enable IPv4 payload checksum for UDP fragments
1721                  * if PCSD is not set */
1722                 rxcsum |= IXGBE_RXCSUM_IPPCSE;
1723         }
1724
1725         IXGBE_WRITE_REG(hw, IXGBE_RXCSUM, rxcsum);
1726 }
1727
1728 static void ixgbe_vlan_rx_register(struct net_device *netdev,
1729                                    struct vlan_group *grp)
1730 {
1731         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1732         u32 ctrl;
1733
1734         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1735                 ixgbe_irq_disable(adapter);
1736         adapter->vlgrp = grp;
1737
1738         if (grp) {
1739                 /* enable VLAN tag insert/strip */
1740                 ctrl = IXGBE_READ_REG(&adapter->hw, IXGBE_VLNCTRL);
1741                 ctrl |= IXGBE_VLNCTRL_VME;
1742                 ctrl &= ~IXGBE_VLNCTRL_CFIEN;
1743                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_VLNCTRL, ctrl);
1744         }
1745
1746         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1747                 ixgbe_irq_enable(adapter);
1748 }
1749
1750 static void ixgbe_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1751 {
1752         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1753         struct ixgbe_hw *hw = &adapter->hw;
1754
1755         /* add VID to filter table */
1756         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, true);
1757 }
1758
1759 static void ixgbe_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1760 {
1761         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1762         struct ixgbe_hw *hw = &adapter->hw;
1763
1764         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1765                 ixgbe_irq_disable(adapter);
1766
1767         vlan_group_set_device(adapter->vlgrp, vid, NULL);
1768
1769         if (!test_bit(__IXGBE_DOWN, &adapter->state))
1770                 ixgbe_irq_enable(adapter);
1771
1772         /* remove VID from filter table */
1773         hw->mac.ops.set_vfta(&adapter->hw, vid, 0, false);
1774 }
1775
1776 static void ixgbe_restore_vlan(struct ixgbe_adapter *adapter)
1777 {
1778         ixgbe_vlan_rx_register(adapter->netdev, adapter->vlgrp);
1779
1780         if (adapter->vlgrp) {
1781                 u16 vid;
1782                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1783                         if (!vlan_group_get_device(adapter->vlgrp, vid))
1784                                 continue;
1785                         ixgbe_vlan_rx_add_vid(adapter->netdev, vid);
1786                 }
1787         }
1788 }
1789
1790 static u8 *ixgbe_addr_list_itr(struct ixgbe_hw *hw, u8 **mc_addr_ptr, u32 *vmdq)
1791 {
1792         struct dev_mc_list *mc_ptr;
1793         u8 *addr = *mc_addr_ptr;
1794         *vmdq = 0;
1795
1796         mc_ptr = container_of(addr, struct dev_mc_list, dmi_addr[0]);
1797         if (mc_ptr->next)
1798                 *mc_addr_ptr = mc_ptr->next->dmi_addr;
1799         else
1800                 *mc_addr_ptr = NULL;
1801
1802         return addr;
1803 }
1804
1805 /**
1806  * ixgbe_set_rx_mode - Unicast, Multicast and Promiscuous mode set
1807  * @netdev: network interface device structure
1808  *
1809  * The set_rx_method entry point is called whenever the unicast/multicast
1810  * address list or the network interface flags are updated.  This routine is
1811  * responsible for configuring the hardware for proper unicast, multicast and
1812  * promiscuous mode.
1813  **/
1814 static void ixgbe_set_rx_mode(struct net_device *netdev)
1815 {
1816         struct ixgbe_adapter *adapter = netdev_priv(netdev);
1817         struct ixgbe_hw *hw = &adapter->hw;
1818         u32 fctrl, vlnctrl;
1819         u8 *addr_list = NULL;
1820         int addr_count = 0;
1821
1822         /* Check for Promiscuous and All Multicast modes */
1823
1824         fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
1825         vlnctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
1826
1827         if (netdev->flags & IFF_PROMISC) {
1828                 hw->addr_ctrl.user_set_promisc = 1;
1829                 fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1830                 vlnctrl &= ~IXGBE_VLNCTRL_VFE;
1831         } else {
1832                 if (netdev->flags & IFF_ALLMULTI) {
1833                         fctrl |= IXGBE_FCTRL_MPE;
1834                         fctrl &= ~IXGBE_FCTRL_UPE;
1835                 } else {
1836                         fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
1837                 }
1838                 vlnctrl |= IXGBE_VLNCTRL_VFE;
1839                 hw->addr_ctrl.user_set_promisc = 0;
1840         }
1841
1842         IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
1843         IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlnctrl);
1844
1845         /* reprogram secondary unicast list */
1846         addr_count = netdev->uc_count;
1847         if (addr_count)
1848                 addr_list = netdev->uc_list->dmi_addr;
1849         hw->mac.ops.update_uc_addr_list(hw, addr_list, addr_count,
1850                                           ixgbe_addr_list_itr);
1851
1852         /* reprogram multicast list */
1853         addr_count = netdev->mc_count;
1854         if (addr_count)
1855                 addr_list = netdev->mc_list->dmi_addr;
1856         hw->mac.ops.update_mc_addr_list(hw, addr_list, addr_count,
1857                                         ixgbe_addr_list_itr);
1858 }
1859
1860 static void ixgbe_napi_enable_all(struct ixgbe_adapter *adapter)
1861 {
1862         int q_idx;
1863         struct ixgbe_q_vector *q_vector;
1864         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1865
1866         /* legacy and MSI only use one vector */
1867         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1868                 q_vectors = 1;
1869
1870         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1871                 struct napi_struct *napi;
1872                 q_vector = &adapter->q_vector[q_idx];
1873                 if (!q_vector->rxr_count)
1874                         continue;
1875                 napi = &q_vector->napi;
1876                 if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) &&
1877                     (q_vector->rxr_count > 1))
1878                         napi->poll = &ixgbe_clean_rxonly_many;
1879
1880                 napi_enable(napi);
1881         }
1882 }
1883
1884 static void ixgbe_napi_disable_all(struct ixgbe_adapter *adapter)
1885 {
1886         int q_idx;
1887         struct ixgbe_q_vector *q_vector;
1888         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1889
1890         /* legacy and MSI only use one vector */
1891         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
1892                 q_vectors = 1;
1893
1894         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1895                 q_vector = &adapter->q_vector[q_idx];
1896                 if (!q_vector->rxr_count)
1897                         continue;
1898                 napi_disable(&q_vector->napi);
1899         }
1900 }
1901
1902 static void ixgbe_configure(struct ixgbe_adapter *adapter)
1903 {
1904         struct net_device *netdev = adapter->netdev;
1905         int i;
1906
1907         ixgbe_set_rx_mode(netdev);
1908
1909         ixgbe_restore_vlan(adapter);
1910
1911         ixgbe_configure_tx(adapter);
1912         ixgbe_configure_rx(adapter);
1913         for (i = 0; i < adapter->num_rx_queues; i++)
1914                 ixgbe_alloc_rx_buffers(adapter, &adapter->rx_ring[i],
1915                                        (adapter->rx_ring[i].count - 1));
1916 }
1917
1918 static int ixgbe_up_complete(struct ixgbe_adapter *adapter)
1919 {
1920         struct net_device *netdev = adapter->netdev;
1921         struct ixgbe_hw *hw = &adapter->hw;
1922         int i, j = 0;
1923         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1924         u32 txdctl, rxdctl, mhadd;
1925         u32 gpie;
1926
1927         ixgbe_get_hw_control(adapter);
1928
1929         if ((adapter->flags & IXGBE_FLAG_MSIX_ENABLED) ||
1930             (adapter->flags & IXGBE_FLAG_MSI_ENABLED)) {
1931                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
1932                         gpie = (IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_EIAME |
1933                                 IXGBE_GPIE_PBA_SUPPORT | IXGBE_GPIE_OCD);
1934                 } else {
1935                         /* MSI only */
1936                         gpie = 0;
1937                 }
1938                 /* XXX: to interrupt immediately for EICS writes, enable this */
1939                 /* gpie |= IXGBE_GPIE_EIMEN; */
1940                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1941         }
1942
1943         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED)) {
1944                 /* legacy interrupts, use EIAM to auto-mask when reading EICR,
1945                  * specifically only auto mask tx and rx interrupts */
1946                 IXGBE_WRITE_REG(hw, IXGBE_EIAM, IXGBE_EICS_RTX_QUEUE);
1947         }
1948
1949         /* Enable fan failure interrupt if media type is copper */
1950         if (adapter->flags & IXGBE_FLAG_FAN_FAIL_CAPABLE) {
1951                 gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
1952                 gpie |= IXGBE_SDP1_GPIEN;
1953                 IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
1954         }
1955
1956         mhadd = IXGBE_READ_REG(hw, IXGBE_MHADD);
1957         if (max_frame != (mhadd >> IXGBE_MHADD_MFS_SHIFT)) {
1958                 mhadd &= ~IXGBE_MHADD_MFS_MASK;
1959                 mhadd |= max_frame << IXGBE_MHADD_MFS_SHIFT;
1960
1961                 IXGBE_WRITE_REG(hw, IXGBE_MHADD, mhadd);
1962         }
1963
1964         for (i = 0; i < adapter->num_tx_queues; i++) {
1965                 j = adapter->tx_ring[i].reg_idx;
1966                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
1967                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1968                 txdctl |= (8 << 16);
1969                 txdctl |= IXGBE_TXDCTL_ENABLE;
1970                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j), txdctl);
1971         }
1972
1973         for (i = 0; i < adapter->num_rx_queues; i++) {
1974                 j = adapter->rx_ring[i].reg_idx;
1975                 rxdctl = IXGBE_READ_REG(hw, IXGBE_RXDCTL(j));
1976                 /* enable PTHRESH=32 descriptors (half the internal cache)
1977                  * and HTHRESH=0 descriptors (to minimize latency on fetch),
1978                  * this also removes a pesky rx_no_buffer_count increment */
1979                 rxdctl |= 0x0020;
1980                 rxdctl |= IXGBE_RXDCTL_ENABLE;
1981                 IXGBE_WRITE_REG(hw, IXGBE_RXDCTL(j), rxdctl);
1982         }
1983         /* enable all receives */
1984         rxdctl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
1985         rxdctl |= (IXGBE_RXCTRL_DMBYPS | IXGBE_RXCTRL_RXEN);
1986         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxdctl);
1987
1988         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
1989                 ixgbe_configure_msix(adapter);
1990         else
1991                 ixgbe_configure_msi_and_legacy(adapter);
1992
1993         clear_bit(__IXGBE_DOWN, &adapter->state);
1994         ixgbe_napi_enable_all(adapter);
1995
1996         /* clear any pending interrupts, may auto mask */
1997         IXGBE_READ_REG(hw, IXGBE_EICR);
1998
1999         ixgbe_irq_enable(adapter);
2000
2001         /* bring the link up in the watchdog, this could race with our first
2002          * link up interrupt but shouldn't be a problem */
2003         adapter->flags |= IXGBE_FLAG_NEED_LINK_UPDATE;
2004         adapter->link_check_timeout = jiffies;
2005         mod_timer(&adapter->watchdog_timer, jiffies);
2006         return 0;
2007 }
2008
2009 void ixgbe_reinit_locked(struct ixgbe_adapter *adapter)
2010 {
2011         WARN_ON(in_interrupt());
2012         while (test_and_set_bit(__IXGBE_RESETTING, &adapter->state))
2013                 msleep(1);
2014         ixgbe_down(adapter);
2015         ixgbe_up(adapter);
2016         clear_bit(__IXGBE_RESETTING, &adapter->state);
2017 }
2018
2019 int ixgbe_up(struct ixgbe_adapter *adapter)
2020 {
2021         /* hardware has been reset, we need to reload some things */
2022         ixgbe_configure(adapter);
2023
2024         return ixgbe_up_complete(adapter);
2025 }
2026
2027 void ixgbe_reset(struct ixgbe_adapter *adapter)
2028 {
2029         struct ixgbe_hw *hw = &adapter->hw;
2030         if (hw->mac.ops.init_hw(hw))
2031                 dev_err(&adapter->pdev->dev, "Hardware Error\n");
2032
2033         /* reprogram the RAR[0] in case user changed it. */
2034         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
2035
2036 }
2037
2038 /**
2039  * ixgbe_clean_rx_ring - Free Rx Buffers per Queue
2040  * @adapter: board private structure
2041  * @rx_ring: ring to free buffers from
2042  **/
2043 static void ixgbe_clean_rx_ring(struct ixgbe_adapter *adapter,
2044                                 struct ixgbe_ring *rx_ring)
2045 {
2046         struct pci_dev *pdev = adapter->pdev;
2047         unsigned long size;
2048         unsigned int i;
2049
2050         /* Free all the Rx ring sk_buffs */
2051
2052         for (i = 0; i < rx_ring->count; i++) {
2053                 struct ixgbe_rx_buffer *rx_buffer_info;
2054
2055                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
2056                 if (rx_buffer_info->dma) {
2057                         pci_unmap_single(pdev, rx_buffer_info->dma,
2058                                          rx_ring->rx_buf_len,
2059                                          PCI_DMA_FROMDEVICE);
2060                         rx_buffer_info->dma = 0;
2061                 }
2062                 if (rx_buffer_info->skb) {
2063                         dev_kfree_skb(rx_buffer_info->skb);
2064                         rx_buffer_info->skb = NULL;
2065                 }
2066                 if (!rx_buffer_info->page)
2067                         continue;
2068                 pci_unmap_page(pdev, rx_buffer_info->page_dma, PAGE_SIZE / 2,
2069                                PCI_DMA_FROMDEVICE);
2070                 rx_buffer_info->page_dma = 0;
2071                 put_page(rx_buffer_info->page);
2072                 rx_buffer_info->page = NULL;
2073                 rx_buffer_info->page_offset = 0;
2074         }
2075
2076         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2077         memset(rx_ring->rx_buffer_info, 0, size);
2078
2079         /* Zero out the descriptor ring */
2080         memset(rx_ring->desc, 0, rx_ring->size);
2081
2082         rx_ring->next_to_clean = 0;
2083         rx_ring->next_to_use = 0;
2084
2085         writel(0, adapter->hw.hw_addr + rx_ring->head);
2086         writel(0, adapter->hw.hw_addr + rx_ring->tail);
2087 }
2088
2089 /**
2090  * ixgbe_clean_tx_ring - Free Tx Buffers
2091  * @adapter: board private structure
2092  * @tx_ring: ring to be cleaned
2093  **/
2094 static void ixgbe_clean_tx_ring(struct ixgbe_adapter *adapter,
2095                                 struct ixgbe_ring *tx_ring)
2096 {
2097         struct ixgbe_tx_buffer *tx_buffer_info;
2098         unsigned long size;
2099         unsigned int i;
2100
2101         /* Free all the Tx ring sk_buffs */
2102
2103         for (i = 0; i < tx_ring->count; i++) {
2104                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2105                 ixgbe_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2106         }
2107
2108         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2109         memset(tx_ring->tx_buffer_info, 0, size);
2110
2111         /* Zero out the descriptor ring */
2112         memset(tx_ring->desc, 0, tx_ring->size);
2113
2114         tx_ring->next_to_use = 0;
2115         tx_ring->next_to_clean = 0;
2116
2117         writel(0, adapter->hw.hw_addr + tx_ring->head);
2118         writel(0, adapter->hw.hw_addr + tx_ring->tail);
2119 }
2120
2121 /**
2122  * ixgbe_clean_all_rx_rings - Free Rx Buffers for all queues
2123  * @adapter: board private structure
2124  **/
2125 static void ixgbe_clean_all_rx_rings(struct ixgbe_adapter *adapter)
2126 {
2127         int i;
2128
2129         for (i = 0; i < adapter->num_rx_queues; i++)
2130                 ixgbe_clean_rx_ring(adapter, &adapter->rx_ring[i]);
2131 }
2132
2133 /**
2134  * ixgbe_clean_all_tx_rings - Free Tx Buffers for all queues
2135  * @adapter: board private structure
2136  **/
2137 static void ixgbe_clean_all_tx_rings(struct ixgbe_adapter *adapter)
2138 {
2139         int i;
2140
2141         for (i = 0; i < adapter->num_tx_queues; i++)
2142                 ixgbe_clean_tx_ring(adapter, &adapter->tx_ring[i]);
2143 }
2144
2145 void ixgbe_down(struct ixgbe_adapter *adapter)
2146 {
2147         struct net_device *netdev = adapter->netdev;
2148         struct ixgbe_hw *hw = &adapter->hw;
2149         u32 rxctrl;
2150         u32 txdctl;
2151         int i, j;
2152
2153         /* signal that we are down to the interrupt handler */
2154         set_bit(__IXGBE_DOWN, &adapter->state);
2155
2156         /* disable receives */
2157         rxctrl = IXGBE_READ_REG(hw, IXGBE_RXCTRL);
2158         IXGBE_WRITE_REG(hw, IXGBE_RXCTRL, rxctrl & ~IXGBE_RXCTRL_RXEN);
2159
2160         netif_tx_disable(netdev);
2161
2162         IXGBE_WRITE_FLUSH(hw);
2163         msleep(10);
2164
2165         netif_tx_stop_all_queues(netdev);
2166
2167         ixgbe_irq_disable(adapter);
2168
2169         ixgbe_napi_disable_all(adapter);
2170
2171         del_timer_sync(&adapter->watchdog_timer);
2172         cancel_work_sync(&adapter->watchdog_task);
2173
2174         /* disable transmits in the hardware now that interrupts are off */
2175         for (i = 0; i < adapter->num_tx_queues; i++) {
2176                 j = adapter->tx_ring[i].reg_idx;
2177                 txdctl = IXGBE_READ_REG(hw, IXGBE_TXDCTL(j));
2178                 IXGBE_WRITE_REG(hw, IXGBE_TXDCTL(j),
2179                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
2180         }
2181
2182         netif_carrier_off(netdev);
2183
2184 #ifdef CONFIG_IXGBE_DCA
2185         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2186                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
2187                 dca_remove_requester(&adapter->pdev->dev);
2188         }
2189
2190 #endif
2191         if (!pci_channel_offline(adapter->pdev))
2192                 ixgbe_reset(adapter);
2193         ixgbe_clean_all_tx_rings(adapter);
2194         ixgbe_clean_all_rx_rings(adapter);
2195
2196 #ifdef CONFIG_IXGBE_DCA
2197         /* since we reset the hardware DCA settings were cleared */
2198         if (dca_add_requester(&adapter->pdev->dev) == 0) {
2199                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
2200                 /* always use CB2 mode, difference is masked
2201                  * in the CB driver */
2202                 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
2203                 ixgbe_setup_dca(adapter);
2204         }
2205 #endif
2206 }
2207
2208 /**
2209  * ixgbe_poll - NAPI Rx polling callback
2210  * @napi: structure for representing this polling device
2211  * @budget: how many packets driver is allowed to clean
2212  *
2213  * This function is used for legacy and MSI, NAPI mode
2214  **/
2215 static int ixgbe_poll(struct napi_struct *napi, int budget)
2216 {
2217         struct ixgbe_q_vector *q_vector = container_of(napi,
2218                                                   struct ixgbe_q_vector, napi);
2219         struct ixgbe_adapter *adapter = q_vector->adapter;
2220         int tx_cleaned, work_done = 0;
2221
2222 #ifdef CONFIG_IXGBE_DCA
2223         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
2224                 ixgbe_update_tx_dca(adapter, adapter->tx_ring);
2225                 ixgbe_update_rx_dca(adapter, adapter->rx_ring);
2226         }
2227 #endif
2228
2229         tx_cleaned = ixgbe_clean_tx_irq(adapter, adapter->tx_ring);
2230         ixgbe_clean_rx_irq(adapter, adapter->rx_ring, &work_done, budget);
2231
2232         if (tx_cleaned)
2233                 work_done = budget;
2234
2235         /* If budget not fully consumed, exit the polling mode */
2236         if (work_done < budget) {
2237                 netif_rx_complete(adapter->netdev, napi);
2238                 if (adapter->itr_setting & 3)
2239                         ixgbe_set_itr(adapter);
2240                 if (!test_bit(__IXGBE_DOWN, &adapter->state))
2241                         ixgbe_irq_enable(adapter);
2242         }
2243         return work_done;
2244 }
2245
2246 /**
2247  * ixgbe_tx_timeout - Respond to a Tx Hang
2248  * @netdev: network interface device structure
2249  **/
2250 static void ixgbe_tx_timeout(struct net_device *netdev)
2251 {
2252         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2253
2254         /* Do the reset outside of interrupt context */
2255         schedule_work(&adapter->reset_task);
2256 }
2257
2258 static void ixgbe_reset_task(struct work_struct *work)
2259 {
2260         struct ixgbe_adapter *adapter;
2261         adapter = container_of(work, struct ixgbe_adapter, reset_task);
2262
2263         adapter->tx_timeout_count++;
2264
2265         ixgbe_reinit_locked(adapter);
2266 }
2267
2268 static void ixgbe_set_num_queues(struct ixgbe_adapter *adapter)
2269 {
2270         int nrq = 1, ntq = 1;
2271         int feature_mask = 0, rss_i, rss_m;
2272
2273         /* Number of supported queues */
2274         switch (adapter->hw.mac.type) {
2275         case ixgbe_mac_82598EB:
2276                 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2277                 rss_m = 0;
2278                 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2279
2280                 switch (adapter->flags & feature_mask) {
2281                 case (IXGBE_FLAG_RSS_ENABLED):
2282                         rss_m = 0xF;
2283                         nrq = rss_i;
2284                         ntq = rss_i;
2285                         break;
2286                 case 0:
2287                 default:
2288                         rss_i = 0;
2289                         rss_m = 0;
2290                         nrq = 1;
2291                         ntq = 1;
2292                         break;
2293                 }
2294
2295                 adapter->ring_feature[RING_F_RSS].indices = rss_i;
2296                 adapter->ring_feature[RING_F_RSS].mask = rss_m;
2297                 break;
2298         default:
2299                 nrq = 1;
2300                 ntq = 1;
2301                 break;
2302         }
2303
2304         adapter->num_rx_queues = nrq;
2305         adapter->num_tx_queues = ntq;
2306 }
2307
2308 static void ixgbe_acquire_msix_vectors(struct ixgbe_adapter *adapter,
2309                                        int vectors)
2310 {
2311         int err, vector_threshold;
2312
2313         /* We'll want at least 3 (vector_threshold):
2314          * 1) TxQ[0] Cleanup
2315          * 2) RxQ[0] Cleanup
2316          * 3) Other (Link Status Change, etc.)
2317          * 4) TCP Timer (optional)
2318          */
2319         vector_threshold = MIN_MSIX_COUNT;
2320
2321         /* The more we get, the more we will assign to Tx/Rx Cleanup
2322          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
2323          * Right now, we simply care about how many we'll get; we'll
2324          * set them up later while requesting irq's.
2325          */
2326         while (vectors >= vector_threshold) {
2327                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
2328                                       vectors);
2329                 if (!err) /* Success in acquiring all requested vectors. */
2330                         break;
2331                 else if (err < 0)
2332                         vectors = 0; /* Nasty failure, quit now */
2333                 else /* err == number of vectors we should try again with */
2334                         vectors = err;
2335         }
2336
2337         if (vectors < vector_threshold) {
2338                 /* Can't allocate enough MSI-X interrupts?  Oh well.
2339                  * This just means we'll go with either a single MSI
2340                  * vector or fall back to legacy interrupts.
2341                  */
2342                 DPRINTK(HW, DEBUG, "Unable to allocate MSI-X interrupts\n");
2343                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2344                 kfree(adapter->msix_entries);
2345                 adapter->msix_entries = NULL;
2346                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2347                 ixgbe_set_num_queues(adapter);
2348         } else {
2349                 adapter->flags |= IXGBE_FLAG_MSIX_ENABLED; /* Woot! */
2350                 adapter->num_msix_vectors = vectors;
2351         }
2352 }
2353
2354 /**
2355  * ixgbe_cache_ring_register - Descriptor ring to register mapping
2356  * @adapter: board private structure to initialize
2357  *
2358  * Once we know the feature-set enabled for the device, we'll cache
2359  * the register offset the descriptor ring is assigned to.
2360  **/
2361 static void __devinit ixgbe_cache_ring_register(struct ixgbe_adapter *adapter)
2362 {
2363         int feature_mask = 0, rss_i;
2364         int i, txr_idx, rxr_idx;
2365
2366         /* Number of supported queues */
2367         switch (adapter->hw.mac.type) {
2368         case ixgbe_mac_82598EB:
2369                 rss_i = adapter->ring_feature[RING_F_RSS].indices;
2370                 txr_idx = 0;
2371                 rxr_idx = 0;
2372                 feature_mask |= IXGBE_FLAG_RSS_ENABLED;
2373                 switch (adapter->flags & feature_mask) {
2374                 case (IXGBE_FLAG_RSS_ENABLED):
2375                         for (i = 0; i < adapter->num_rx_queues; i++)
2376                                 adapter->rx_ring[i].reg_idx = i;
2377                         for (i = 0; i < adapter->num_tx_queues; i++)
2378                                 adapter->tx_ring[i].reg_idx = i;
2379                         break;
2380                 case 0:
2381                 default:
2382                         break;
2383                 }
2384                 break;
2385         default:
2386                 break;
2387         }
2388 }
2389
2390 /**
2391  * ixgbe_alloc_queues - Allocate memory for all rings
2392  * @adapter: board private structure to initialize
2393  *
2394  * We allocate one ring per queue at run-time since we don't know the
2395  * number of queues at compile-time.  The polling_netdev array is
2396  * intended for Multiqueue, but should work fine with a single queue.
2397  **/
2398 static int __devinit ixgbe_alloc_queues(struct ixgbe_adapter *adapter)
2399 {
2400         int i;
2401
2402         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
2403                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
2404         if (!adapter->tx_ring)
2405                 goto err_tx_ring_allocation;
2406
2407         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
2408                                    sizeof(struct ixgbe_ring), GFP_KERNEL);
2409         if (!adapter->rx_ring)
2410                 goto err_rx_ring_allocation;
2411
2412         for (i = 0; i < adapter->num_tx_queues; i++) {
2413                 adapter->tx_ring[i].count = adapter->tx_ring_count;
2414                 adapter->tx_ring[i].queue_index = i;
2415         }
2416
2417         for (i = 0; i < adapter->num_rx_queues; i++) {
2418                 adapter->rx_ring[i].count = adapter->rx_ring_count;
2419                 adapter->rx_ring[i].queue_index = i;
2420         }
2421
2422         ixgbe_cache_ring_register(adapter);
2423
2424         return 0;
2425
2426 err_rx_ring_allocation:
2427         kfree(adapter->tx_ring);
2428 err_tx_ring_allocation:
2429         return -ENOMEM;
2430 }
2431
2432 /**
2433  * ixgbe_set_interrupt_capability - set MSI-X or MSI if supported
2434  * @adapter: board private structure to initialize
2435  *
2436  * Attempt to configure the interrupts using the best available
2437  * capabilities of the hardware and the kernel.
2438  **/
2439 static int __devinit ixgbe_set_interrupt_capability(struct ixgbe_adapter
2440                                                     *adapter)
2441 {
2442         int err = 0;
2443         int vector, v_budget;
2444
2445         /*
2446          * It's easy to be greedy for MSI-X vectors, but it really
2447          * doesn't do us much good if we have a lot more vectors
2448          * than CPU's.  So let's be conservative and only ask for
2449          * (roughly) twice the number of vectors as there are CPU's.
2450          */
2451         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
2452                        (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
2453
2454         /*
2455          * At the same time, hardware can only support a maximum of
2456          * MAX_MSIX_COUNT vectors.  With features such as RSS and VMDq,
2457          * we can easily reach upwards of 64 Rx descriptor queues and
2458          * 32 Tx queues.  Thus, we cap it off in those rare cases where
2459          * the cpu count also exceeds our vector limit.
2460          */
2461         v_budget = min(v_budget, MAX_MSIX_COUNT);
2462
2463         /* A failure in MSI-X entry allocation isn't fatal, but it does
2464          * mean we disable MSI-X capabilities of the adapter. */
2465         adapter->msix_entries = kcalloc(v_budget,
2466                                         sizeof(struct msix_entry), GFP_KERNEL);
2467         if (!adapter->msix_entries) {
2468                 adapter->flags &= ~IXGBE_FLAG_RSS_ENABLED;
2469                 ixgbe_set_num_queues(adapter);
2470                 kfree(adapter->tx_ring);
2471                 kfree(adapter->rx_ring);
2472                 err = ixgbe_alloc_queues(adapter);
2473                 if (err) {
2474                         DPRINTK(PROBE, ERR, "Unable to allocate memory "
2475                                 "for queues\n");
2476                         goto out;
2477                 }
2478
2479                 goto try_msi;
2480         }
2481
2482         for (vector = 0; vector < v_budget; vector++)
2483                 adapter->msix_entries[vector].entry = vector;
2484
2485         ixgbe_acquire_msix_vectors(adapter, v_budget);
2486
2487         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED)
2488                 goto out;
2489
2490 try_msi:
2491         err = pci_enable_msi(adapter->pdev);
2492         if (!err) {
2493                 adapter->flags |= IXGBE_FLAG_MSI_ENABLED;
2494         } else {
2495                 DPRINTK(HW, DEBUG, "Unable to allocate MSI interrupt, "
2496                         "falling back to legacy.  Error: %d\n", err);
2497                 /* reset err */
2498                 err = 0;
2499         }
2500
2501 out:
2502         /* Notify the stack of the (possibly) reduced Tx Queue count. */
2503         adapter->netdev->real_num_tx_queues = adapter->num_tx_queues;
2504
2505         return err;
2506 }
2507
2508 static void ixgbe_reset_interrupt_capability(struct ixgbe_adapter *adapter)
2509 {
2510         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2511                 adapter->flags &= ~IXGBE_FLAG_MSIX_ENABLED;
2512                 pci_disable_msix(adapter->pdev);
2513                 kfree(adapter->msix_entries);
2514                 adapter->msix_entries = NULL;
2515         } else if (adapter->flags & IXGBE_FLAG_MSI_ENABLED) {
2516                 adapter->flags &= ~IXGBE_FLAG_MSI_ENABLED;
2517                 pci_disable_msi(adapter->pdev);
2518         }
2519         return;
2520 }
2521
2522 /**
2523  * ixgbe_init_interrupt_scheme - Determine proper interrupt scheme
2524  * @adapter: board private structure to initialize
2525  *
2526  * We determine which interrupt scheme to use based on...
2527  * - Kernel support (MSI, MSI-X)
2528  *   - which can be user-defined (via MODULE_PARAM)
2529  * - Hardware queue count (num_*_queues)
2530  *   - defined by miscellaneous hardware support/features (RSS, etc.)
2531  **/
2532 static int __devinit ixgbe_init_interrupt_scheme(struct ixgbe_adapter *adapter)
2533 {
2534         int err;
2535
2536         /* Number of supported queues */
2537         ixgbe_set_num_queues(adapter);
2538
2539         err = ixgbe_alloc_queues(adapter);
2540         if (err) {
2541                 DPRINTK(PROBE, ERR, "Unable to allocate memory for queues\n");
2542                 goto err_alloc_queues;
2543         }
2544
2545         err = ixgbe_set_interrupt_capability(adapter);
2546         if (err) {
2547                 DPRINTK(PROBE, ERR, "Unable to setup interrupt capabilities\n");
2548                 goto err_set_interrupt;
2549         }
2550
2551         DPRINTK(DRV, INFO, "Multiqueue %s: Rx Queue count = %u, "
2552                 "Tx Queue count = %u\n",
2553                 (adapter->num_rx_queues > 1) ? "Enabled" :
2554                 "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2555
2556         set_bit(__IXGBE_DOWN, &adapter->state);
2557
2558         return 0;
2559
2560 err_set_interrupt:
2561         kfree(adapter->tx_ring);
2562         kfree(adapter->rx_ring);
2563 err_alloc_queues:
2564         return err;
2565 }
2566
2567 /**
2568  * ixgbe_sw_init - Initialize general software structures (struct ixgbe_adapter)
2569  * @adapter: board private structure to initialize
2570  *
2571  * ixgbe_sw_init initializes the Adapter private data structure.
2572  * Fields are initialized based on PCI device information and
2573  * OS network device settings (MTU size).
2574  **/
2575 static int __devinit ixgbe_sw_init(struct ixgbe_adapter *adapter)
2576 {
2577         struct ixgbe_hw *hw = &adapter->hw;
2578         struct pci_dev *pdev = adapter->pdev;
2579         unsigned int rss;
2580
2581         /* PCI config space info */
2582
2583         hw->vendor_id = pdev->vendor;
2584         hw->device_id = pdev->device;
2585         hw->revision_id = pdev->revision;
2586         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2587         hw->subsystem_device_id = pdev->subsystem_device;
2588
2589         /* Set capability flags */
2590         rss = min(IXGBE_MAX_RSS_INDICES, (int)num_online_cpus());
2591         adapter->ring_feature[RING_F_RSS].indices = rss;
2592         adapter->flags |= IXGBE_FLAG_RSS_ENABLED;
2593         if (hw->mac.ops.get_media_type &&
2594             (hw->mac.ops.get_media_type(hw) == ixgbe_media_type_copper))
2595                 adapter->flags |= IXGBE_FLAG_FAN_FAIL_CAPABLE;
2596
2597         /* default flow control settings */
2598         hw->fc.original_type = ixgbe_fc_none;
2599         hw->fc.type = ixgbe_fc_none;
2600         hw->fc.high_water = IXGBE_DEFAULT_FCRTH;
2601         hw->fc.low_water = IXGBE_DEFAULT_FCRTL;
2602         hw->fc.pause_time = IXGBE_DEFAULT_FCPAUSE;
2603         hw->fc.send_xon = true;
2604
2605         /* select 10G link by default */
2606         hw->mac.link_mode_select = IXGBE_AUTOC_LMS_10G_LINK_NO_AN;
2607
2608         /* enable itr by default in dynamic mode */
2609         adapter->itr_setting = 1;
2610         adapter->eitr_param = 20000;
2611
2612         /* set defaults for eitr in MegaBytes */
2613         adapter->eitr_low = 10;
2614         adapter->eitr_high = 20;
2615
2616         /* set default ring sizes */
2617         adapter->tx_ring_count = IXGBE_DEFAULT_TXD;
2618         adapter->rx_ring_count = IXGBE_DEFAULT_RXD;
2619
2620         /* initialize eeprom parameters */
2621         if (ixgbe_init_eeprom_params_generic(hw)) {
2622                 dev_err(&pdev->dev, "EEPROM initialization failed\n");
2623                 return -EIO;
2624         }
2625
2626         /* enable rx csum by default */
2627         adapter->flags |= IXGBE_FLAG_RX_CSUM_ENABLED;
2628
2629         set_bit(__IXGBE_DOWN, &adapter->state);
2630
2631         return 0;
2632 }
2633
2634 /**
2635  * ixgbe_setup_tx_resources - allocate Tx resources (Descriptors)
2636  * @adapter: board private structure
2637  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2638  *
2639  * Return 0 on success, negative on failure
2640  **/
2641 int ixgbe_setup_tx_resources(struct ixgbe_adapter *adapter,
2642                              struct ixgbe_ring *tx_ring)
2643 {
2644         struct pci_dev *pdev = adapter->pdev;
2645         int size;
2646
2647         size = sizeof(struct ixgbe_tx_buffer) * tx_ring->count;
2648         tx_ring->tx_buffer_info = vmalloc(size);
2649         if (!tx_ring->tx_buffer_info)
2650                 goto err;
2651         memset(tx_ring->tx_buffer_info, 0, size);
2652
2653         /* round up to nearest 4K */
2654         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc) +
2655                         sizeof(u32);
2656         tx_ring->size = ALIGN(tx_ring->size, 4096);
2657
2658         tx_ring->desc = pci_alloc_consistent(pdev, tx_ring->size,
2659                                              &tx_ring->dma);
2660         if (!tx_ring->desc)
2661                 goto err;
2662
2663         tx_ring->next_to_use = 0;
2664         tx_ring->next_to_clean = 0;
2665         tx_ring->work_limit = tx_ring->count;
2666         return 0;
2667
2668 err:
2669         vfree(tx_ring->tx_buffer_info);
2670         tx_ring->tx_buffer_info = NULL;
2671         DPRINTK(PROBE, ERR, "Unable to allocate memory for the transmit "
2672                             "descriptor ring\n");
2673         return -ENOMEM;
2674 }
2675
2676 /**
2677  * ixgbe_setup_all_tx_resources - allocate all queues Tx resources
2678  * @adapter: board private structure
2679  *
2680  * If this function returns with an error, then it's possible one or
2681  * more of the rings is populated (while the rest are not).  It is the
2682  * callers duty to clean those orphaned rings.
2683  *
2684  * Return 0 on success, negative on failure
2685  **/
2686 static int ixgbe_setup_all_tx_resources(struct ixgbe_adapter *adapter)
2687 {
2688         int i, err = 0;
2689
2690         for (i = 0; i < adapter->num_tx_queues; i++) {
2691                 err = ixgbe_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2692                 if (!err)
2693                         continue;
2694                 DPRINTK(PROBE, ERR, "Allocation for Tx Queue %u failed\n", i);
2695                 break;
2696         }
2697
2698         return err;
2699 }
2700
2701 /**
2702  * ixgbe_setup_rx_resources - allocate Rx resources (Descriptors)
2703  * @adapter: board private structure
2704  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2705  *
2706  * Returns 0 on success, negative on failure
2707  **/
2708 int ixgbe_setup_rx_resources(struct ixgbe_adapter *adapter,
2709                              struct ixgbe_ring *rx_ring)
2710 {
2711         struct pci_dev *pdev = adapter->pdev;
2712         int size;
2713
2714         size = sizeof(struct net_lro_desc) * IXGBE_MAX_LRO_DESCRIPTORS;
2715         rx_ring->lro_mgr.lro_arr = vmalloc(size);
2716         if (!rx_ring->lro_mgr.lro_arr)
2717                 return -ENOMEM;
2718         memset(rx_ring->lro_mgr.lro_arr, 0, size);
2719
2720         size = sizeof(struct ixgbe_rx_buffer) * rx_ring->count;
2721         rx_ring->rx_buffer_info = vmalloc(size);
2722         if (!rx_ring->rx_buffer_info) {
2723                 DPRINTK(PROBE, ERR,
2724                         "vmalloc allocation failed for the rx desc ring\n");
2725                 goto alloc_failed;
2726         }
2727         memset(rx_ring->rx_buffer_info, 0, size);
2728
2729         /* Round up to nearest 4K */
2730         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2731         rx_ring->size = ALIGN(rx_ring->size, 4096);
2732
2733         rx_ring->desc = pci_alloc_consistent(pdev, rx_ring->size, &rx_ring->dma);
2734
2735         if (!rx_ring->desc) {
2736                 DPRINTK(PROBE, ERR,
2737                         "Memory allocation failed for the rx desc ring\n");
2738                 vfree(rx_ring->rx_buffer_info);
2739                 goto alloc_failed;
2740         }
2741
2742         rx_ring->next_to_clean = 0;
2743         rx_ring->next_to_use = 0;
2744
2745         return 0;
2746
2747 alloc_failed:
2748         vfree(rx_ring->lro_mgr.lro_arr);
2749         rx_ring->lro_mgr.lro_arr = NULL;
2750         return -ENOMEM;
2751 }
2752
2753 /**
2754  * ixgbe_setup_all_rx_resources - allocate all queues Rx resources
2755  * @adapter: board private structure
2756  *
2757  * If this function returns with an error, then it's possible one or
2758  * more of the rings is populated (while the rest are not).  It is the
2759  * callers duty to clean those orphaned rings.
2760  *
2761  * Return 0 on success, negative on failure
2762  **/
2763
2764 static int ixgbe_setup_all_rx_resources(struct ixgbe_adapter *adapter)
2765 {
2766         int i, err = 0;
2767
2768         for (i = 0; i < adapter->num_rx_queues; i++) {
2769                 err = ixgbe_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2770                 if (!err)
2771                         continue;
2772                 DPRINTK(PROBE, ERR, "Allocation for Rx Queue %u failed\n", i);
2773                 break;
2774         }
2775
2776         return err;
2777 }
2778
2779 /**
2780  * ixgbe_free_tx_resources - Free Tx Resources per Queue
2781  * @adapter: board private structure
2782  * @tx_ring: Tx descriptor ring for a specific queue
2783  *
2784  * Free all transmit software resources
2785  **/
2786 void ixgbe_free_tx_resources(struct ixgbe_adapter *adapter,
2787                              struct ixgbe_ring *tx_ring)
2788 {
2789         struct pci_dev *pdev = adapter->pdev;
2790
2791         ixgbe_clean_tx_ring(adapter, tx_ring);
2792
2793         vfree(tx_ring->tx_buffer_info);
2794         tx_ring->tx_buffer_info = NULL;
2795
2796         pci_free_consistent(pdev, tx_ring->size, tx_ring->desc, tx_ring->dma);
2797
2798         tx_ring->desc = NULL;
2799 }
2800
2801 /**
2802  * ixgbe_free_all_tx_resources - Free Tx Resources for All Queues
2803  * @adapter: board private structure
2804  *
2805  * Free all transmit software resources
2806  **/
2807 static void ixgbe_free_all_tx_resources(struct ixgbe_adapter *adapter)
2808 {
2809         int i;
2810
2811         for (i = 0; i < adapter->num_tx_queues; i++)
2812                 ixgbe_free_tx_resources(adapter, &adapter->tx_ring[i]);
2813 }
2814
2815 /**
2816  * ixgbe_free_rx_resources - Free Rx Resources
2817  * @adapter: board private structure
2818  * @rx_ring: ring to clean the resources from
2819  *
2820  * Free all receive software resources
2821  **/
2822 void ixgbe_free_rx_resources(struct ixgbe_adapter *adapter,
2823                              struct ixgbe_ring *rx_ring)
2824 {
2825         struct pci_dev *pdev = adapter->pdev;
2826
2827         vfree(rx_ring->lro_mgr.lro_arr);
2828         rx_ring->lro_mgr.lro_arr = NULL;
2829
2830         ixgbe_clean_rx_ring(adapter, rx_ring);
2831
2832         vfree(rx_ring->rx_buffer_info);
2833         rx_ring->rx_buffer_info = NULL;
2834
2835         pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
2836
2837         rx_ring->desc = NULL;
2838 }
2839
2840 /**
2841  * ixgbe_free_all_rx_resources - Free Rx Resources for All Queues
2842  * @adapter: board private structure
2843  *
2844  * Free all receive software resources
2845  **/
2846 static void ixgbe_free_all_rx_resources(struct ixgbe_adapter *adapter)
2847 {
2848         int i;
2849
2850         for (i = 0; i < adapter->num_rx_queues; i++)
2851                 ixgbe_free_rx_resources(adapter, &adapter->rx_ring[i]);
2852 }
2853
2854 /**
2855  * ixgbe_change_mtu - Change the Maximum Transfer Unit
2856  * @netdev: network interface device structure
2857  * @new_mtu: new value for maximum frame size
2858  *
2859  * Returns 0 on success, negative on failure
2860  **/
2861 static int ixgbe_change_mtu(struct net_device *netdev, int new_mtu)
2862 {
2863         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2864         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
2865
2866         /* MTU < 68 is an error and causes problems on some kernels */
2867         if ((new_mtu < 68) || (max_frame > IXGBE_MAX_JUMBO_FRAME_SIZE))
2868                 return -EINVAL;
2869
2870         DPRINTK(PROBE, INFO, "changing MTU from %d to %d\n",
2871                 netdev->mtu, new_mtu);
2872         /* must set new MTU before calling down or up */
2873         netdev->mtu = new_mtu;
2874
2875         if (netif_running(netdev))
2876                 ixgbe_reinit_locked(adapter);
2877
2878         return 0;
2879 }
2880
2881 /**
2882  * ixgbe_open - Called when a network interface is made active
2883  * @netdev: network interface device structure
2884  *
2885  * Returns 0 on success, negative value on failure
2886  *
2887  * The open entry point is called when a network interface is made
2888  * active by the system (IFF_UP).  At this point all resources needed
2889  * for transmit and receive operations are allocated, the interrupt
2890  * handler is registered with the OS, the watchdog timer is started,
2891  * and the stack is notified that the interface is ready.
2892  **/
2893 static int ixgbe_open(struct net_device *netdev)
2894 {
2895         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2896         int err;
2897
2898         /* disallow open during test */
2899         if (test_bit(__IXGBE_TESTING, &adapter->state))
2900                 return -EBUSY;
2901
2902         /* allocate transmit descriptors */
2903         err = ixgbe_setup_all_tx_resources(adapter);
2904         if (err)
2905                 goto err_setup_tx;
2906
2907         /* allocate receive descriptors */
2908         err = ixgbe_setup_all_rx_resources(adapter);
2909         if (err)
2910                 goto err_setup_rx;
2911
2912         ixgbe_configure(adapter);
2913
2914         err = ixgbe_request_irq(adapter);
2915         if (err)
2916                 goto err_req_irq;
2917
2918         err = ixgbe_up_complete(adapter);
2919         if (err)
2920                 goto err_up;
2921
2922         netif_tx_start_all_queues(netdev);
2923
2924         return 0;
2925
2926 err_up:
2927         ixgbe_release_hw_control(adapter);
2928         ixgbe_free_irq(adapter);
2929 err_req_irq:
2930         ixgbe_free_all_rx_resources(adapter);
2931 err_setup_rx:
2932         ixgbe_free_all_tx_resources(adapter);
2933 err_setup_tx:
2934         ixgbe_reset(adapter);
2935
2936         return err;
2937 }
2938
2939 /**
2940  * ixgbe_close - Disables a network interface
2941  * @netdev: network interface device structure
2942  *
2943  * Returns 0, this is not allowed to fail
2944  *
2945  * The close entry point is called when an interface is de-activated
2946  * by the OS.  The hardware is still under the drivers control, but
2947  * needs to be disabled.  A global MAC reset is issued to stop the
2948  * hardware, and all transmit and receive resources are freed.
2949  **/
2950 static int ixgbe_close(struct net_device *netdev)
2951 {
2952         struct ixgbe_adapter *adapter = netdev_priv(netdev);
2953
2954         ixgbe_down(adapter);
2955         ixgbe_free_irq(adapter);
2956
2957         ixgbe_free_all_tx_resources(adapter);
2958         ixgbe_free_all_rx_resources(adapter);
2959
2960         ixgbe_release_hw_control(adapter);
2961
2962         return 0;
2963 }
2964
2965 /**
2966  * ixgbe_napi_add_all - prep napi structs for use
2967  * @adapter: private struct
2968  * helper function to napi_add each possible q_vector->napi
2969  */
2970 static void ixgbe_napi_add_all(struct ixgbe_adapter *adapter)
2971 {
2972         int q_idx, q_vectors;
2973         int (*poll)(struct napi_struct *, int);
2974
2975         if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
2976                 poll = &ixgbe_clean_rxonly;
2977                 /* Only enable as many vectors as we have rx queues. */
2978                 q_vectors = adapter->num_rx_queues;
2979         } else {
2980                 poll = &ixgbe_poll;
2981                 /* only one q_vector for legacy modes */
2982                 q_vectors = 1;
2983         }
2984
2985         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
2986                 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
2987                 netif_napi_add(adapter->netdev, &q_vector->napi, (*poll), 64);
2988         }
2989 }
2990
2991 static void ixgbe_napi_del_all(struct ixgbe_adapter *adapter)
2992 {
2993         int q_idx;
2994         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2995
2996         /* legacy and MSI only use one vector */
2997         if (!(adapter->flags & IXGBE_FLAG_MSIX_ENABLED))
2998                 q_vectors = 1;
2999
3000         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
3001                 struct ixgbe_q_vector *q_vector = &adapter->q_vector[q_idx];
3002                 if (!q_vector->rxr_count)
3003                         continue;
3004                 netif_napi_del(&q_vector->napi);
3005         }
3006 }
3007
3008 #ifdef CONFIG_PM
3009 static int ixgbe_resume(struct pci_dev *pdev)
3010 {
3011         struct net_device *netdev = pci_get_drvdata(pdev);
3012         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3013         u32 err;
3014
3015         pci_set_power_state(pdev, PCI_D0);
3016         pci_restore_state(pdev);
3017         err = pci_enable_device(pdev);
3018         if (err) {
3019                 printk(KERN_ERR "ixgbe: Cannot enable PCI device from "
3020                                 "suspend\n");
3021                 return err;
3022         }
3023         pci_set_master(pdev);
3024
3025         pci_enable_wake(pdev, PCI_D3hot, 0);
3026         pci_enable_wake(pdev, PCI_D3cold, 0);
3027
3028         err = ixgbe_init_interrupt_scheme(adapter);
3029         if (err) {
3030                 printk(KERN_ERR "ixgbe: Cannot initialize interrupts for "
3031                                 "device\n");
3032                 return err;
3033         }
3034
3035         ixgbe_napi_add_all(adapter);
3036         ixgbe_reset(adapter);
3037
3038         if (netif_running(netdev)) {
3039                 err = ixgbe_open(adapter->netdev);
3040                 if (err)
3041                         return err;
3042         }
3043
3044         netif_device_attach(netdev);
3045
3046         return 0;
3047 }
3048
3049 #endif /* CONFIG_PM */
3050 static int ixgbe_suspend(struct pci_dev *pdev, pm_message_t state)
3051 {
3052         struct net_device *netdev = pci_get_drvdata(pdev);
3053         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3054 #ifdef CONFIG_PM
3055         int retval = 0;
3056 #endif
3057
3058         netif_device_detach(netdev);
3059
3060         if (netif_running(netdev)) {
3061                 ixgbe_down(adapter);
3062                 ixgbe_free_irq(adapter);
3063                 ixgbe_free_all_tx_resources(adapter);
3064                 ixgbe_free_all_rx_resources(adapter);
3065         }
3066         ixgbe_reset_interrupt_capability(adapter);
3067         ixgbe_napi_del_all(adapter);
3068         kfree(adapter->tx_ring);
3069         kfree(adapter->rx_ring);
3070
3071 #ifdef CONFIG_PM
3072         retval = pci_save_state(pdev);
3073         if (retval)
3074                 return retval;
3075 #endif
3076
3077         pci_enable_wake(pdev, PCI_D3hot, 0);
3078         pci_enable_wake(pdev, PCI_D3cold, 0);
3079
3080         ixgbe_release_hw_control(adapter);
3081
3082         pci_disable_device(pdev);
3083
3084         pci_set_power_state(pdev, pci_choose_state(pdev, state));
3085
3086         return 0;
3087 }
3088
3089 static void ixgbe_shutdown(struct pci_dev *pdev)
3090 {
3091         ixgbe_suspend(pdev, PMSG_SUSPEND);
3092 }
3093
3094 /**
3095  * ixgbe_update_stats - Update the board statistics counters.
3096  * @adapter: board private structure
3097  **/
3098 void ixgbe_update_stats(struct ixgbe_adapter *adapter)
3099 {
3100         struct ixgbe_hw *hw = &adapter->hw;
3101         u64 total_mpc = 0;
3102         u32 i, missed_rx = 0, mpc, bprc, lxon, lxoff, xon_off_tot;
3103
3104         adapter->stats.crcerrs += IXGBE_READ_REG(hw, IXGBE_CRCERRS);
3105         for (i = 0; i < 8; i++) {
3106                 /* for packet buffers not used, the register should read 0 */
3107                 mpc = IXGBE_READ_REG(hw, IXGBE_MPC(i));
3108                 missed_rx += mpc;
3109                 adapter->stats.mpc[i] += mpc;
3110                 total_mpc += adapter->stats.mpc[i];
3111                 adapter->stats.rnbc[i] += IXGBE_READ_REG(hw, IXGBE_RNBC(i));
3112         }
3113         adapter->stats.gprc += IXGBE_READ_REG(hw, IXGBE_GPRC);
3114         /* work around hardware counting issue */
3115         adapter->stats.gprc -= missed_rx;
3116
3117         /* 82598 hardware only has a 32 bit counter in the high register */
3118         adapter->stats.gorc += IXGBE_READ_REG(hw, IXGBE_GORCH);
3119         adapter->stats.gotc += IXGBE_READ_REG(hw, IXGBE_GOTCH);
3120         adapter->stats.tor += IXGBE_READ_REG(hw, IXGBE_TORH);
3121         bprc = IXGBE_READ_REG(hw, IXGBE_BPRC);
3122         adapter->stats.bprc += bprc;
3123         adapter->stats.mprc += IXGBE_READ_REG(hw, IXGBE_MPRC);
3124         adapter->stats.mprc -= bprc;
3125         adapter->stats.roc += IXGBE_READ_REG(hw, IXGBE_ROC);
3126         adapter->stats.prc64 += IXGBE_READ_REG(hw, IXGBE_PRC64);
3127         adapter->stats.prc127 += IXGBE_READ_REG(hw, IXGBE_PRC127);
3128         adapter->stats.prc255 += IXGBE_READ_REG(hw, IXGBE_PRC255);
3129         adapter->stats.prc511 += IXGBE_READ_REG(hw, IXGBE_PRC511);
3130         adapter->stats.prc1023 += IXGBE_READ_REG(hw, IXGBE_PRC1023);
3131         adapter->stats.prc1522 += IXGBE_READ_REG(hw, IXGBE_PRC1522);
3132         adapter->stats.rlec += IXGBE_READ_REG(hw, IXGBE_RLEC);
3133         adapter->stats.lxonrxc += IXGBE_READ_REG(hw, IXGBE_LXONRXC);
3134         adapter->stats.lxoffrxc += IXGBE_READ_REG(hw, IXGBE_LXOFFRXC);
3135         lxon = IXGBE_READ_REG(hw, IXGBE_LXONTXC);
3136         adapter->stats.lxontxc += lxon;
3137         lxoff = IXGBE_READ_REG(hw, IXGBE_LXOFFTXC);
3138         adapter->stats.lxofftxc += lxoff;
3139         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3140         adapter->stats.gptc += IXGBE_READ_REG(hw, IXGBE_GPTC);
3141         adapter->stats.mptc += IXGBE_READ_REG(hw, IXGBE_MPTC);
3142         /*
3143          * 82598 errata - tx of flow control packets is included in tx counters
3144          */
3145         xon_off_tot = lxon + lxoff;
3146         adapter->stats.gptc -= xon_off_tot;
3147         adapter->stats.mptc -= xon_off_tot;
3148         adapter->stats.gotc -= (xon_off_tot * (ETH_ZLEN + ETH_FCS_LEN));
3149         adapter->stats.ruc += IXGBE_READ_REG(hw, IXGBE_RUC);
3150         adapter->stats.rfc += IXGBE_READ_REG(hw, IXGBE_RFC);
3151         adapter->stats.rjc += IXGBE_READ_REG(hw, IXGBE_RJC);
3152         adapter->stats.tpr += IXGBE_READ_REG(hw, IXGBE_TPR);
3153         adapter->stats.ptc64 += IXGBE_READ_REG(hw, IXGBE_PTC64);
3154         adapter->stats.ptc64 -= xon_off_tot;
3155         adapter->stats.ptc127 += IXGBE_READ_REG(hw, IXGBE_PTC127);
3156         adapter->stats.ptc255 += IXGBE_READ_REG(hw, IXGBE_PTC255);
3157         adapter->stats.ptc511 += IXGBE_READ_REG(hw, IXGBE_PTC511);
3158         adapter->stats.ptc1023 += IXGBE_READ_REG(hw, IXGBE_PTC1023);
3159         adapter->stats.ptc1522 += IXGBE_READ_REG(hw, IXGBE_PTC1522);
3160         adapter->stats.bptc += IXGBE_READ_REG(hw, IXGBE_BPTC);
3161
3162         /* Fill out the OS statistics structure */
3163         adapter->net_stats.multicast = adapter->stats.mprc;
3164
3165         /* Rx Errors */
3166         adapter->net_stats.rx_errors = adapter->stats.crcerrs +
3167                                        adapter->stats.rlec;
3168         adapter->net_stats.rx_dropped = 0;
3169         adapter->net_stats.rx_length_errors = adapter->stats.rlec;
3170         adapter->net_stats.rx_crc_errors = adapter->stats.crcerrs;
3171         adapter->net_stats.rx_missed_errors = total_mpc;
3172 }
3173
3174 /**
3175  * ixgbe_watchdog - Timer Call-back
3176  * @data: pointer to adapter cast into an unsigned long
3177  **/
3178 static void ixgbe_watchdog(unsigned long data)
3179 {
3180         struct ixgbe_adapter *adapter = (struct ixgbe_adapter *)data;
3181         struct ixgbe_hw *hw = &adapter->hw;
3182
3183         /* Do the watchdog outside of interrupt context due to the lovely
3184          * delays that some of the newer hardware requires */
3185         if (!test_bit(__IXGBE_DOWN, &adapter->state)) {
3186                 /* Cause software interrupt to ensure rx rings are cleaned */
3187                 if (adapter->flags & IXGBE_FLAG_MSIX_ENABLED) {
3188                         u32 eics =
3189                          (1 << (adapter->num_msix_vectors - NON_Q_VECTORS)) - 1;
3190                         IXGBE_WRITE_REG(hw, IXGBE_EICS, eics);
3191                 } else {
3192                         /* For legacy and MSI interrupts don't set any bits that
3193                          * are enabled for EIAM, because this operation would
3194                          * set *both* EIMS and EICS for any bit in EIAM */
3195                         IXGBE_WRITE_REG(hw, IXGBE_EICS,
3196                                     (IXGBE_EICS_TCP_TIMER | IXGBE_EICS_OTHER));
3197                 }
3198                 /* Reset the timer */
3199                 mod_timer(&adapter->watchdog_timer,
3200                           round_jiffies(jiffies + 2 * HZ));
3201         }
3202
3203         schedule_work(&adapter->watchdog_task);
3204 }
3205
3206 /**
3207  * ixgbe_watchdog_task - worker thread to bring link up
3208  * @work: pointer to work_struct containing our data
3209  **/
3210 static void ixgbe_watchdog_task(struct work_struct *work)
3211 {
3212         struct ixgbe_adapter *adapter = container_of(work,
3213                                                      struct ixgbe_adapter,
3214                                                      watchdog_task);
3215         struct net_device *netdev = adapter->netdev;
3216         struct ixgbe_hw *hw = &adapter->hw;
3217         u32 link_speed = adapter->link_speed;
3218         bool link_up = adapter->link_up;
3219
3220         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
3221
3222         if (adapter->flags & IXGBE_FLAG_NEED_LINK_UPDATE) {
3223                 hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
3224                 if (link_up ||
3225                     time_after(jiffies, (adapter->link_check_timeout +
3226                                          IXGBE_TRY_LINK_TIMEOUT))) {
3227                         IXGBE_WRITE_REG(hw, IXGBE_EIMS, IXGBE_EIMC_LSC);
3228                         adapter->flags &= ~IXGBE_FLAG_NEED_LINK_UPDATE;
3229                 }
3230                 adapter->link_up = link_up;
3231                 adapter->link_speed = link_speed;
3232         }
3233
3234         if (link_up) {
3235                 if (!netif_carrier_ok(netdev)) {
3236                         u32 frctl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
3237                         u32 rmcs = IXGBE_READ_REG(hw, IXGBE_RMCS);
3238 #define FLOW_RX (frctl & IXGBE_FCTRL_RFCE)
3239 #define FLOW_TX (rmcs & IXGBE_RMCS_TFCE_802_3X)
3240                         DPRINTK(LINK, INFO, "NIC Link is Up %s, "
3241                                 "Flow Control: %s\n",
3242                                 (link_speed == IXGBE_LINK_SPEED_10GB_FULL ?
3243                                  "10 Gbps" :
3244                                  (link_speed == IXGBE_LINK_SPEED_1GB_FULL ?
3245                                   "1 Gbps" : "unknown speed")),
3246                                 ((FLOW_RX && FLOW_TX) ? "RX/TX" :
3247                                  (FLOW_RX ? "RX" :
3248                                  (FLOW_TX ? "TX" : "None"))));
3249
3250                         netif_carrier_on(netdev);
3251                         netif_tx_wake_all_queues(netdev);
3252                 } else {
3253                         /* Force detection of hung controller */
3254                         adapter->detect_tx_hung = true;
3255                 }
3256         } else {
3257                 adapter->link_up = false;
3258                 adapter->link_speed = 0;
3259                 if (netif_carrier_ok(netdev)) {
3260                         DPRINTK(LINK, INFO, "NIC Link is Down\n");
3261                         netif_carrier_off(netdev);
3262                         netif_tx_stop_all_queues(netdev);
3263                 }
3264         }
3265
3266         ixgbe_update_stats(adapter);
3267         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
3268 }
3269
3270 static int ixgbe_tso(struct ixgbe_adapter *adapter,
3271                      struct ixgbe_ring *tx_ring, struct sk_buff *skb,
3272                      u32 tx_flags, u8 *hdr_len)
3273 {
3274         struct ixgbe_adv_tx_context_desc *context_desc;
3275         unsigned int i;
3276         int err;
3277         struct ixgbe_tx_buffer *tx_buffer_info;
3278         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
3279         u32 mss_l4len_idx, l4len;
3280
3281         if (skb_is_gso(skb)) {
3282                 if (skb_header_cloned(skb)) {
3283                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3284                         if (err)
3285                                 return err;
3286                 }
3287                 l4len = tcp_hdrlen(skb);
3288                 *hdr_len += l4len;
3289
3290                 if (skb->protocol == htons(ETH_P_IP)) {
3291                         struct iphdr *iph = ip_hdr(skb);
3292                         iph->tot_len = 0;
3293                         iph->check = 0;
3294                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
3295                                                                  iph->daddr, 0,
3296                                                                  IPPROTO_TCP,
3297                                                                  0);
3298                         adapter->hw_tso_ctxt++;
3299                 } else if (skb_shinfo(skb)->gso_type == SKB_GSO_TCPV6) {
3300                         ipv6_hdr(skb)->payload_len = 0;
3301                         tcp_hdr(skb)->check =
3302                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3303                                              &ipv6_hdr(skb)->daddr,
3304                                              0, IPPROTO_TCP, 0);
3305                         adapter->hw_tso6_ctxt++;
3306                 }
3307
3308                 i = tx_ring->next_to_use;
3309
3310                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3311                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3312
3313                 /* VLAN MACLEN IPLEN */
3314                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3315                         vlan_macip_lens |=
3316                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3317                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
3318                                     IXGBE_ADVTXD_MACLEN_SHIFT);
3319                 *hdr_len += skb_network_offset(skb);
3320                 vlan_macip_lens |=
3321                     (skb_transport_header(skb) - skb_network_header(skb));
3322                 *hdr_len +=
3323                     (skb_transport_header(skb) - skb_network_header(skb));
3324                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3325                 context_desc->seqnum_seed = 0;
3326
3327                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
3328                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
3329                                    IXGBE_ADVTXD_DTYP_CTXT);
3330
3331                 if (skb->protocol == htons(ETH_P_IP))
3332                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3333                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
3334                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3335
3336                 /* MSS L4LEN IDX */
3337                 mss_l4len_idx =
3338                     (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
3339                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
3340                 /* use index 1 for TSO */
3341                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3342                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
3343
3344                 tx_buffer_info->time_stamp = jiffies;
3345                 tx_buffer_info->next_to_watch = i;
3346
3347                 i++;
3348                 if (i == tx_ring->count)
3349                         i = 0;
3350                 tx_ring->next_to_use = i;
3351
3352                 return true;
3353         }
3354         return false;
3355 }
3356
3357 static bool ixgbe_tx_csum(struct ixgbe_adapter *adapter,
3358                           struct ixgbe_ring *tx_ring,
3359                           struct sk_buff *skb, u32 tx_flags)
3360 {
3361         struct ixgbe_adv_tx_context_desc *context_desc;
3362         unsigned int i;
3363         struct ixgbe_tx_buffer *tx_buffer_info;
3364         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
3365
3366         if (skb->ip_summed == CHECKSUM_PARTIAL ||
3367             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
3368                 i = tx_ring->next_to_use;
3369                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3370                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
3371
3372                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3373                         vlan_macip_lens |=
3374                             (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
3375                 vlan_macip_lens |= (skb_network_offset(skb) <<
3376                                     IXGBE_ADVTXD_MACLEN_SHIFT);
3377                 if (skb->ip_summed == CHECKSUM_PARTIAL)
3378                         vlan_macip_lens |= (skb_transport_header(skb) -
3379                                             skb_network_header(skb));
3380
3381                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
3382                 context_desc->seqnum_seed = 0;
3383
3384                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
3385                                     IXGBE_ADVTXD_DTYP_CTXT);
3386
3387                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3388                         switch (skb->protocol) {
3389                         case __constant_htons(ETH_P_IP):
3390                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
3391                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
3392                                         type_tucmd_mlhl |=
3393                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3394                                 break;
3395                         case __constant_htons(ETH_P_IPV6):
3396                                 /* XXX what about other V6 headers?? */
3397                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
3398                                         type_tucmd_mlhl |=
3399                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
3400                                 break;
3401                         default:
3402                                 if (unlikely(net_ratelimit())) {
3403                                         DPRINTK(PROBE, WARNING,
3404                                          "partial checksum but proto=%x!\n",
3405                                          skb->protocol);
3406                                 }
3407                                 break;
3408                         }
3409                 }
3410
3411                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
3412                 /* use index zero for tx checksum offload */
3413                 context_desc->mss_l4len_idx = 0;
3414
3415                 tx_buffer_info->time_stamp = jiffies;
3416                 tx_buffer_info->next_to_watch = i;
3417
3418                 adapter->hw_csum_tx_good++;
3419                 i++;
3420                 if (i == tx_ring->count)
3421                         i = 0;
3422                 tx_ring->next_to_use = i;
3423
3424                 return true;
3425         }
3426
3427         return false;
3428 }
3429
3430 static int ixgbe_tx_map(struct ixgbe_adapter *adapter,
3431                         struct ixgbe_ring *tx_ring,
3432                         struct sk_buff *skb, unsigned int first)
3433 {
3434         struct ixgbe_tx_buffer *tx_buffer_info;
3435         unsigned int len = skb->len;
3436         unsigned int offset = 0, size, count = 0, i;
3437         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
3438         unsigned int f;
3439
3440         len -= skb->data_len;
3441
3442         i = tx_ring->next_to_use;
3443
3444         while (len) {
3445                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3446                 size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3447
3448                 tx_buffer_info->length = size;
3449                 tx_buffer_info->dma = pci_map_single(adapter->pdev,
3450                                                      skb->data + offset,
3451                                                      size, PCI_DMA_TODEVICE);
3452                 tx_buffer_info->time_stamp = jiffies;
3453                 tx_buffer_info->next_to_watch = i;
3454
3455                 len -= size;
3456                 offset += size;
3457                 count++;
3458                 i++;
3459                 if (i == tx_ring->count)
3460                         i = 0;
3461         }
3462
3463         for (f = 0; f < nr_frags; f++) {
3464                 struct skb_frag_struct *frag;
3465
3466                 frag = &skb_shinfo(skb)->frags[f];
3467                 len = frag->size;
3468                 offset = frag->page_offset;
3469
3470                 while (len) {
3471                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
3472                         size = min(len, (uint)IXGBE_MAX_DATA_PER_TXD);
3473
3474                         tx_buffer_info->length = size;
3475                         tx_buffer_info->dma = pci_map_page(adapter->pdev,
3476                                                            frag->page,
3477                                                            offset,
3478                                                            size,
3479                                                            PCI_DMA_TODEVICE);
3480                         tx_buffer_info->time_stamp = jiffies;
3481                         tx_buffer_info->next_to_watch = i;
3482
3483                         len -= size;
3484                         offset += size;
3485                         count++;
3486                         i++;
3487                         if (i == tx_ring->count)
3488                                 i = 0;
3489                 }
3490         }
3491         if (i == 0)
3492                 i = tx_ring->count - 1;
3493         else
3494                 i = i - 1;
3495         tx_ring->tx_buffer_info[i].skb = skb;
3496         tx_ring->tx_buffer_info[first].next_to_watch = i;
3497
3498         return count;
3499 }
3500
3501 static void ixgbe_tx_queue(struct ixgbe_adapter *adapter,
3502                            struct ixgbe_ring *tx_ring,
3503                            int tx_flags, int count, u32 paylen, u8 hdr_len)
3504 {
3505         union ixgbe_adv_tx_desc *tx_desc = NULL;
3506         struct ixgbe_tx_buffer *tx_buffer_info;
3507         u32 olinfo_status = 0, cmd_type_len = 0;
3508         unsigned int i;
3509         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3510
3511         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3512
3513         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3514
3515         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3516                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3517
3518         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3519                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3520
3521                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3522                                  IXGBE_ADVTXD_POPTS_SHIFT;
3523
3524                 /* use index 1 context for tso */
3525                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3526                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3527                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
3528                                          IXGBE_ADVTXD_POPTS_SHIFT;
3529
3530         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3531                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
3532                                  IXGBE_ADVTXD_POPTS_SHIFT;
3533
3534         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3535
3536         i = tx_ring->next_to_use;
3537         while (count--) {
3538                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3539                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
3540                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3541                 tx_desc->read.cmd_type_len =
3542                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3543                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3544                 i++;
3545                 if (i == tx_ring->count)
3546                         i = 0;
3547         }
3548
3549         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3550
3551         /*
3552          * Force memory writes to complete before letting h/w
3553          * know there are new descriptors to fetch.  (Only
3554          * applicable for weak-ordered memory model archs,
3555          * such as IA-64).
3556          */
3557         wmb();
3558
3559         tx_ring->next_to_use = i;
3560         writel(i, adapter->hw.hw_addr + tx_ring->tail);
3561 }
3562
3563 static int __ixgbe_maybe_stop_tx(struct net_device *netdev,
3564                                  struct ixgbe_ring *tx_ring, int size)
3565 {
3566         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3567
3568         netif_stop_subqueue(netdev, tx_ring->queue_index);
3569         /* Herbert's original patch had:
3570          *  smp_mb__after_netif_stop_queue();
3571          * but since that doesn't exist yet, just open code it. */
3572         smp_mb();
3573
3574         /* We need to check again in a case another CPU has just
3575          * made room available. */
3576         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
3577                 return -EBUSY;
3578
3579         /* A reprieve! - use start_queue because it doesn't call schedule */
3580         netif_start_subqueue(netdev, tx_ring->queue_index);
3581         ++adapter->restart_queue;
3582         return 0;
3583 }
3584
3585 static int ixgbe_maybe_stop_tx(struct net_device *netdev,
3586                               struct ixgbe_ring *tx_ring, int size)
3587 {
3588         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
3589                 return 0;
3590         return __ixgbe_maybe_stop_tx(netdev, tx_ring, size);
3591 }
3592
3593 static int ixgbe_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3594 {
3595         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3596         struct ixgbe_ring *tx_ring;
3597         unsigned int first;
3598         unsigned int tx_flags = 0;
3599         u8 hdr_len = 0;
3600         int r_idx = 0, tso;
3601         int count = 0;
3602         unsigned int f;
3603
3604         r_idx = (adapter->num_tx_queues - 1) & skb->queue_mapping;
3605         tx_ring = &adapter->tx_ring[r_idx];
3606
3607         if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
3608                 tx_flags |= vlan_tx_tag_get(skb);
3609                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3610                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3611         }
3612         /* three things can cause us to need a context descriptor */
3613         if (skb_is_gso(skb) ||
3614             (skb->ip_summed == CHECKSUM_PARTIAL) ||
3615             (tx_flags & IXGBE_TX_FLAGS_VLAN))
3616                 count++;
3617
3618         count += TXD_USE_COUNT(skb_headlen(skb));
3619         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3620                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3621
3622         if (ixgbe_maybe_stop_tx(netdev, tx_ring, count)) {
3623                 adapter->tx_busy++;
3624                 return NETDEV_TX_BUSY;
3625         }
3626
3627         if (skb->protocol == htons(ETH_P_IP))
3628                 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3629         first = tx_ring->next_to_use;
3630         tso = ixgbe_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
3631         if (tso < 0) {
3632                 dev_kfree_skb_any(skb);
3633                 return NETDEV_TX_OK;
3634         }
3635
3636         if (tso)
3637                 tx_flags |= IXGBE_TX_FLAGS_TSO;
3638         else if (ixgbe_tx_csum(adapter, tx_ring, skb, tx_flags) &&
3639                  (skb->ip_summed == CHECKSUM_PARTIAL))
3640                 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3641
3642         ixgbe_tx_queue(adapter, tx_ring, tx_flags,
3643                        ixgbe_tx_map(adapter, tx_ring, skb, first),
3644                        skb->len, hdr_len);
3645
3646         netdev->trans_start = jiffies;
3647
3648         ixgbe_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
3649
3650         return NETDEV_TX_OK;
3651 }
3652
3653 /**
3654  * ixgbe_get_stats - Get System Network Statistics
3655  * @netdev: network interface device structure
3656  *
3657  * Returns the address of the device statistics structure.
3658  * The statistics are actually updated from the timer callback.
3659  **/
3660 static struct net_device_stats *ixgbe_get_stats(struct net_device *netdev)
3661 {
3662         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3663
3664         /* only return the current stats */
3665         return &adapter->net_stats;
3666 }
3667
3668 /**
3669  * ixgbe_set_mac - Change the Ethernet Address of the NIC
3670  * @netdev: network interface device structure
3671  * @p: pointer to an address structure
3672  *
3673  * Returns 0 on success, negative on failure
3674  **/
3675 static int ixgbe_set_mac(struct net_device *netdev, void *p)
3676 {
3677         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3678         struct ixgbe_hw *hw = &adapter->hw;
3679         struct sockaddr *addr = p;
3680
3681         if (!is_valid_ether_addr(addr->sa_data))
3682                 return -EADDRNOTAVAIL;
3683
3684         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3685         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3686
3687         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0, IXGBE_RAH_AV);
3688
3689         return 0;
3690 }
3691
3692 #ifdef CONFIG_NET_POLL_CONTROLLER
3693 /*
3694  * Polling 'interrupt' - used by things like netconsole to send skbs
3695  * without having to re-enable interrupts. It's not called while
3696  * the interrupt routine is executing.
3697  */
3698 static void ixgbe_netpoll(struct net_device *netdev)
3699 {
3700         struct ixgbe_adapter *adapter = netdev_priv(netdev);
3701
3702         disable_irq(adapter->pdev->irq);
3703         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
3704         ixgbe_intr(adapter->pdev->irq, netdev);
3705         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
3706         enable_irq(adapter->pdev->irq);
3707 }
3708 #endif
3709
3710 /**
3711  * ixgbe_link_config - set up initial link with default speed and duplex
3712  * @hw: pointer to private hardware struct
3713  *
3714  * Returns 0 on success, negative on failure
3715  **/
3716 static int ixgbe_link_config(struct ixgbe_hw *hw)
3717 {
3718         u32 autoneg = IXGBE_LINK_SPEED_10GB_FULL;
3719
3720         /* must always autoneg for both 1G and 10G link */
3721         hw->mac.autoneg = true;
3722
3723         if ((hw->mac.type == ixgbe_mac_82598EB) &&
3724             (hw->phy.media_type == ixgbe_media_type_copper))
3725                 autoneg = IXGBE_LINK_SPEED_82598_AUTONEG;
3726
3727         return hw->mac.ops.setup_link_speed(hw, autoneg, true, true);
3728 }
3729
3730 /**
3731  * ixgbe_probe - Device Initialization Routine
3732  * @pdev: PCI device information struct
3733  * @ent: entry in ixgbe_pci_tbl
3734  *
3735  * Returns 0 on success, negative on failure
3736  *
3737  * ixgbe_probe initializes an adapter identified by a pci_dev structure.
3738  * The OS initialization, configuring of the adapter private structure,
3739  * and a hardware reset occur.
3740  **/
3741 static int __devinit ixgbe_probe(struct pci_dev *pdev,
3742                                  const struct pci_device_id *ent)
3743 {
3744         struct net_device *netdev;
3745         struct ixgbe_adapter *adapter = NULL;
3746         struct ixgbe_hw *hw;
3747         const struct ixgbe_info *ii = ixgbe_info_tbl[ent->driver_data];
3748         static int cards_found;
3749         int i, err, pci_using_dac;
3750         u16 link_status, link_speed, link_width;
3751         u32 part_num, eec;
3752
3753         err = pci_enable_device(pdev);
3754         if (err)
3755                 return err;
3756
3757         if (!pci_set_dma_mask(pdev, DMA_64BIT_MASK) &&
3758             !pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK)) {
3759                 pci_using_dac = 1;
3760         } else {
3761                 err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
3762                 if (err) {
3763                         err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
3764                         if (err) {
3765                                 dev_err(&pdev->dev, "No usable DMA "
3766                                         "configuration, aborting\n");
3767                                 goto err_dma;
3768                         }
3769                 }
3770                 pci_using_dac = 0;
3771         }
3772
3773         err = pci_request_regions(pdev, ixgbe_driver_name);
3774         if (err) {
3775                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3776                 goto err_pci_reg;
3777         }
3778
3779         pci_set_master(pdev);
3780         pci_save_state(pdev);
3781
3782         netdev = alloc_etherdev_mq(sizeof(struct ixgbe_adapter), MAX_TX_QUEUES);
3783         if (!netdev) {
3784                 err = -ENOMEM;
3785                 goto err_alloc_etherdev;
3786         }
3787
3788         SET_NETDEV_DEV(netdev, &pdev->dev);
3789
3790         pci_set_drvdata(pdev, netdev);
3791         adapter = netdev_priv(netdev);
3792
3793         adapter->netdev = netdev;
3794         adapter->pdev = pdev;
3795         hw = &adapter->hw;
3796         hw->back = adapter;
3797         adapter->msg_enable = (1 << DEFAULT_DEBUG_LEVEL_SHIFT) - 1;
3798
3799         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3800                               pci_resource_len(pdev, 0));
3801         if (!hw->hw_addr) {
3802                 err = -EIO;
3803                 goto err_ioremap;
3804         }
3805
3806         for (i = 1; i <= 5; i++) {
3807                 if (pci_resource_len(pdev, i) == 0)
3808                         continue;
3809         }
3810
3811         netdev->open = &ixgbe_open;
3812         netdev->stop = &ixgbe_close;
3813         netdev->hard_start_xmit = &ixgbe_xmit_frame;
3814         netdev->get_stats = &ixgbe_get_stats;
3815         netdev->set_rx_mode = &ixgbe_set_rx_mode;
3816         netdev->set_multicast_list = &ixgbe_set_rx_mode;
3817         netdev->set_mac_address = &ixgbe_set_mac;
3818         netdev->change_mtu = &ixgbe_change_mtu;
3819         ixgbe_set_ethtool_ops(netdev);
3820         netdev->tx_timeout = &ixgbe_tx_timeout;
3821         netdev->watchdog_timeo = 5 * HZ;
3822         netdev->vlan_rx_register = ixgbe_vlan_rx_register;
3823         netdev->vlan_rx_add_vid = ixgbe_vlan_rx_add_vid;
3824         netdev->vlan_rx_kill_vid = ixgbe_vlan_rx_kill_vid;
3825 #ifdef CONFIG_NET_POLL_CONTROLLER
3826         netdev->poll_controller = ixgbe_netpoll;
3827 #endif
3828         strcpy(netdev->name, pci_name(pdev));
3829
3830         adapter->bd_number = cards_found;
3831
3832         /* Setup hw api */
3833         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3834         hw->mac.type  = ii->mac;
3835
3836         /* EEPROM */
3837         memcpy(&hw->eeprom.ops, ii->eeprom_ops, sizeof(hw->eeprom.ops));
3838         eec = IXGBE_READ_REG(hw, IXGBE_EEC);
3839         /* If EEPROM is valid (bit 8 = 1), use default otherwise use bit bang */
3840         if (!(eec & (1 << 8)))
3841                 hw->eeprom.ops.read = &ixgbe_read_eeprom_bit_bang_generic;
3842
3843         /* PHY */
3844         memcpy(&hw->phy.ops, ii->phy_ops, sizeof(hw->phy.ops));
3845         /* phy->sfp_type = ixgbe_sfp_type_unknown; */
3846
3847         err = ii->get_invariants(hw);
3848         if (err)
3849                 goto err_hw_init;
3850
3851         /* setup the private structure */
3852         err = ixgbe_sw_init(adapter);
3853         if (err)
3854                 goto err_sw_init;
3855
3856         /* reset_hw fills in the perm_addr as well */
3857         err = hw->mac.ops.reset_hw(hw);
3858         if (err) {
3859                 dev_err(&adapter->pdev->dev, "HW Init failed: %d\n", err);
3860                 goto err_sw_init;
3861         }
3862
3863         netdev->features = NETIF_F_SG |
3864                            NETIF_F_IP_CSUM |
3865                            NETIF_F_HW_VLAN_TX |
3866                            NETIF_F_HW_VLAN_RX |
3867                            NETIF_F_HW_VLAN_FILTER;
3868
3869         netdev->features |= NETIF_F_IPV6_CSUM;
3870         netdev->features |= NETIF_F_TSO;
3871         netdev->features |= NETIF_F_TSO6;
3872         netdev->features |= NETIF_F_LRO;
3873
3874         netdev->vlan_features |= NETIF_F_TSO;
3875         netdev->vlan_features |= NETIF_F_TSO6;
3876         netdev->vlan_features |= NETIF_F_IP_CSUM;
3877         netdev->vlan_features |= NETIF_F_SG;
3878
3879         if (pci_using_dac)
3880                 netdev->features |= NETIF_F_HIGHDMA;
3881
3882         /* make sure the EEPROM is good */
3883         if (hw->eeprom.ops.validate_checksum(hw, NULL) < 0) {
3884                 dev_err(&pdev->dev, "The EEPROM Checksum Is Not Valid\n");
3885                 err = -EIO;
3886                 goto err_eeprom;
3887         }
3888
3889         memcpy(netdev->dev_addr, hw->mac.perm_addr, netdev->addr_len);
3890         memcpy(netdev->perm_addr, hw->mac.perm_addr, netdev->addr_len);
3891
3892         if (ixgbe_validate_mac_addr(netdev->perm_addr)) {
3893                 dev_err(&pdev->dev, "invalid MAC address\n");
3894                 err = -EIO;
3895                 goto err_eeprom;
3896         }
3897
3898         init_timer(&adapter->watchdog_timer);
3899         adapter->watchdog_timer.function = &ixgbe_watchdog;
3900         adapter->watchdog_timer.data = (unsigned long)adapter;
3901
3902         INIT_WORK(&adapter->reset_task, ixgbe_reset_task);
3903         INIT_WORK(&adapter->watchdog_task, ixgbe_watchdog_task);
3904
3905         err = ixgbe_init_interrupt_scheme(adapter);
3906         if (err)
3907                 goto err_sw_init;
3908
3909         /* print bus type/speed/width info */
3910         pci_read_config_word(pdev, IXGBE_PCI_LINK_STATUS, &link_status);
3911         link_speed = link_status & IXGBE_PCI_LINK_SPEED;
3912         link_width = link_status & IXGBE_PCI_LINK_WIDTH;
3913         dev_info(&pdev->dev, "(PCI Express:%s:%s) %pM\n",
3914                 ((link_speed == IXGBE_PCI_LINK_SPEED_5000) ? "5.0Gb/s" :
3915                  (link_speed == IXGBE_PCI_LINK_SPEED_2500) ? "2.5Gb/s" :
3916                  "Unknown"),
3917                 ((link_width == IXGBE_PCI_LINK_WIDTH_8) ? "Width x8" :
3918                  (link_width == IXGBE_PCI_LINK_WIDTH_4) ? "Width x4" :
3919                  (link_width == IXGBE_PCI_LINK_WIDTH_2) ? "Width x2" :
3920                  (link_width == IXGBE_PCI_LINK_WIDTH_1) ? "Width x1" :
3921                  "Unknown"),
3922                 netdev->dev_addr);
3923         ixgbe_read_pba_num_generic(hw, &part_num);
3924         dev_info(&pdev->dev, "MAC: %d, PHY: %d, PBA No: %06x-%03x\n",
3925                  hw->mac.type, hw->phy.type,
3926                  (part_num >> 8), (part_num & 0xff));
3927
3928         if (link_width <= IXGBE_PCI_LINK_WIDTH_4) {
3929                 dev_warn(&pdev->dev, "PCI-Express bandwidth available for "
3930                          "this card is not sufficient for optimal "
3931                          "performance.\n");
3932                 dev_warn(&pdev->dev, "For optimal performance a x8 "
3933                          "PCI-Express slot is required.\n");
3934         }
3935
3936         /* reset the hardware with the new settings */
3937         hw->mac.ops.start_hw(hw);
3938
3939         /* link_config depends on start_hw being called at least once */
3940         err = ixgbe_link_config(hw);
3941         if (err) {
3942                 dev_err(&pdev->dev, "setup_link_speed FAILED %d\n", err);
3943                 goto err_register;
3944         }
3945
3946         netif_carrier_off(netdev);
3947         netif_tx_stop_all_queues(netdev);
3948
3949         ixgbe_napi_add_all(adapter);
3950
3951         strcpy(netdev->name, "eth%d");
3952         err = register_netdev(netdev);
3953         if (err)
3954                 goto err_register;
3955
3956 #ifdef CONFIG_IXGBE_DCA
3957         if (dca_add_requester(&pdev->dev) == 0) {
3958                 adapter->flags |= IXGBE_FLAG_DCA_ENABLED;
3959                 /* always use CB2 mode, difference is masked
3960                  * in the CB driver */
3961                 IXGBE_WRITE_REG(hw, IXGBE_DCA_CTRL, 2);
3962                 ixgbe_setup_dca(adapter);
3963         }
3964 #endif
3965
3966         dev_info(&pdev->dev, "Intel(R) 10 Gigabit Network Connection\n");
3967         cards_found++;
3968         return 0;
3969
3970 err_register:
3971         ixgbe_release_hw_control(adapter);
3972 err_hw_init:
3973 err_sw_init:
3974         ixgbe_reset_interrupt_capability(adapter);
3975 err_eeprom:
3976         iounmap(hw->hw_addr);
3977 err_ioremap:
3978         free_netdev(netdev);
3979 err_alloc_etherdev:
3980         pci_release_regions(pdev);
3981 err_pci_reg:
3982 err_dma:
3983         pci_disable_device(pdev);
3984         return err;
3985 }
3986
3987 /**
3988  * ixgbe_remove - Device Removal Routine
3989  * @pdev: PCI device information struct
3990  *
3991  * ixgbe_remove is called by the PCI subsystem to alert the driver
3992  * that it should release a PCI device.  The could be caused by a
3993  * Hot-Plug event, or because the driver is going to be removed from
3994  * memory.
3995  **/
3996 static void __devexit ixgbe_remove(struct pci_dev *pdev)
3997 {
3998         struct net_device *netdev = pci_get_drvdata(pdev);
3999         struct ixgbe_adapter *adapter = netdev_priv(netdev);
4000
4001         set_bit(__IXGBE_DOWN, &adapter->state);
4002         del_timer_sync(&adapter->watchdog_timer);
4003
4004         flush_scheduled_work();
4005
4006 #ifdef CONFIG_IXGBE_DCA
4007         if (adapter->flags & IXGBE_FLAG_DCA_ENABLED) {
4008                 adapter->flags &= ~IXGBE_FLAG_DCA_ENABLED;
4009                 dca_remove_requester(&pdev->dev);
4010                 IXGBE_WRITE_REG(&adapter->hw, IXGBE_DCA_CTRL, 1);
4011         }
4012
4013 #endif
4014         unregister_netdev(netdev);
4015
4016         ixgbe_reset_interrupt_capability(adapter);
4017
4018         ixgbe_release_hw_control(adapter);
4019
4020         iounmap(adapter->hw.hw_addr);
4021         pci_release_regions(pdev);
4022
4023         DPRINTK(PROBE, INFO, "complete\n");
4024         ixgbe_napi_del_all(adapter);
4025         kfree(adapter->tx_ring);
4026         kfree(adapter->rx_ring);
4027
4028         free_netdev(netdev);
4029
4030         pci_disable_device(pdev);
4031 }
4032
4033 /**
4034  * ixgbe_io_error_detected - called when PCI error is detected
4035  * @pdev: Pointer to PCI device
4036  * @state: The current pci connection state
4037  *
4038  * This function is called after a PCI bus error affecting
4039  * this device has been detected.
4040  */
4041 static pci_ers_result_t ixgbe_io_error_detected(struct pci_dev *pdev,
4042                                                 pci_channel_state_t state)
4043 {
4044         struct net_device *netdev = pci_get_drvdata(pdev);
4045         struct ixgbe_adapter *adapter = netdev->priv;
4046
4047         netif_device_detach(netdev);
4048
4049         if (netif_running(netdev))
4050                 ixgbe_down(adapter);
4051         pci_disable_device(pdev);
4052
4053         /* Request a slot reset. */
4054         return PCI_ERS_RESULT_NEED_RESET;
4055 }
4056
4057 /**
4058  * ixgbe_io_slot_reset - called after the pci bus has been reset.
4059  * @pdev: Pointer to PCI device
4060  *
4061  * Restart the card from scratch, as if from a cold-boot.
4062  */
4063 static pci_ers_result_t ixgbe_io_slot_reset(struct pci_dev *pdev)
4064 {
4065         struct net_device *netdev = pci_get_drvdata(pdev);
4066         struct ixgbe_adapter *adapter = netdev->priv;
4067
4068         if (pci_enable_device(pdev)) {
4069                 DPRINTK(PROBE, ERR,
4070                         "Cannot re-enable PCI device after reset.\n");
4071                 return PCI_ERS_RESULT_DISCONNECT;
4072         }
4073         pci_set_master(pdev);
4074         pci_restore_state(pdev);
4075
4076         pci_enable_wake(pdev, PCI_D3hot, 0);
4077         pci_enable_wake(pdev, PCI_D3cold, 0);
4078
4079         ixgbe_reset(adapter);
4080
4081         return PCI_ERS_RESULT_RECOVERED;
4082 }
4083
4084 /**
4085  * ixgbe_io_resume - called when traffic can start flowing again.
4086  * @pdev: Pointer to PCI device
4087  *
4088  * This callback is called when the error recovery driver tells us that
4089  * its OK to resume normal operation.
4090  */
4091 static void ixgbe_io_resume(struct pci_dev *pdev)
4092 {
4093         struct net_device *netdev = pci_get_drvdata(pdev);
4094         struct ixgbe_adapter *adapter = netdev->priv;
4095
4096         if (netif_running(netdev)) {
4097                 if (ixgbe_up(adapter)) {
4098                         DPRINTK(PROBE, INFO, "ixgbe_up failed after reset\n");
4099                         return;
4100                 }
4101         }
4102
4103         netif_device_attach(netdev);
4104 }
4105
4106 static struct pci_error_handlers ixgbe_err_handler = {
4107         .error_detected = ixgbe_io_error_detected,
4108         .slot_reset = ixgbe_io_slot_reset,
4109         .resume = ixgbe_io_resume,
4110 };
4111
4112 static struct pci_driver ixgbe_driver = {
4113         .name     = ixgbe_driver_name,
4114         .id_table = ixgbe_pci_tbl,
4115         .probe    = ixgbe_probe,
4116         .remove   = __devexit_p(ixgbe_remove),
4117 #ifdef CONFIG_PM
4118         .suspend  = ixgbe_suspend,
4119         .resume   = ixgbe_resume,
4120 #endif
4121         .shutdown = ixgbe_shutdown,
4122         .err_handler = &ixgbe_err_handler
4123 };
4124
4125 /**
4126  * ixgbe_init_module - Driver Registration Routine
4127  *
4128  * ixgbe_init_module is the first routine called when the driver is
4129  * loaded. All it does is register with the PCI subsystem.
4130  **/
4131 static int __init ixgbe_init_module(void)
4132 {
4133         int ret;
4134         printk(KERN_INFO "%s: %s - version %s\n", ixgbe_driver_name,
4135                ixgbe_driver_string, ixgbe_driver_version);
4136
4137         printk(KERN_INFO "%s: %s\n", ixgbe_driver_name, ixgbe_copyright);
4138
4139 #ifdef CONFIG_IXGBE_DCA
4140         dca_register_notify(&dca_notifier);
4141 #endif
4142
4143         ret = pci_register_driver(&ixgbe_driver);
4144         return ret;
4145 }
4146
4147 module_init(ixgbe_init_module);
4148
4149 /**
4150  * ixgbe_exit_module - Driver Exit Cleanup Routine
4151  *
4152  * ixgbe_exit_module is called just before the driver is removed
4153  * from memory.
4154  **/
4155 static void __exit ixgbe_exit_module(void)
4156 {
4157 #ifdef CONFIG_IXGBE_DCA
4158         dca_unregister_notify(&dca_notifier);
4159 #endif
4160         pci_unregister_driver(&ixgbe_driver);
4161 }
4162
4163 #ifdef CONFIG_IXGBE_DCA
4164 static int ixgbe_notify_dca(struct notifier_block *nb, unsigned long event,
4165                             void *p)
4166 {
4167         int ret_val;
4168
4169         ret_val = driver_for_each_device(&ixgbe_driver.driver, NULL, &event,
4170                                          __ixgbe_notify_dca);
4171
4172         return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
4173 }
4174 #endif /* CONFIG_IXGBE_DCA */
4175
4176 module_exit(ixgbe_exit_module);
4177
4178 /* ixgbe_main.c */