drivers/net/ethernet/intel/ixgbevf/ixgbevf_main.c

   1 /*******************************************************************************
   2
   3   Intel 82599 Virtual Function driver
   4   Copyright(c) 1999 - 2012 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
  29 /******************************************************************************
  30  Copyright (c)2006 - 2007 Myricom, Inc. for some LRO specific code
  31 ******************************************************************************/
  32
  33 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  34
  35 #include <linux/types.h>
  36 #include <linux/bitops.h>
  37 #include <linux/module.h>
  38 #include <linux/pci.h>
  39 #include <linux/netdevice.h>
  40 #include <linux/vmalloc.h>
  41 #include <linux/string.h>
  42 #include <linux/in.h>
  43 #include <linux/ip.h>
  44 #include <linux/tcp.h>
  45 #include <linux/sctp.h>
  46 #include <linux/ipv6.h>
  47 #include <linux/slab.h>
  48 #include <net/checksum.h>
  49 #include <net/ip6_checksum.h>
  50 #include <linux/ethtool.h>
  51 #include <linux/if.h>
  52 #include <linux/if_vlan.h>
  53 #include <linux/prefetch.h>
  54
  55 #include "ixgbevf.h"
  56
  57 const char ixgbevf_driver_name[] = "ixgbevf";
  58 static const char ixgbevf_driver_string[] =
  59         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  60
  61 #define DRV_VERSION "2.11.3-k"
  62 const char ixgbevf_driver_version[] = DRV_VERSION;
  63 static char ixgbevf_copyright[] =
  64         "Copyright (c) 2009 - 2012 Intel Corporation.";
  65
  66 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  67         [board_82599_vf] = &ixgbevf_82599_vf_info,
  68         [board_X540_vf]  = &ixgbevf_X540_vf_info,
  69 };
  70
  71 /* ixgbevf_pci_tbl - PCI Device ID Table
  72  *
  73  * Wildcard entries (PCI_ANY_ID) should come last
  74  * Last entry must be all 0s
  75  *
  76  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  77  *   Class, Class Mask, private data (not used) }
  78  */
  79 static DEFINE_PCI_DEVICE_TABLE(ixgbevf_pci_tbl) = {
  80         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF), board_82599_vf },
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF), board_X540_vf },
  82         /* required last entry */
  83         {0, }
  84 };
  85 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  86
  87 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  88 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
  89 MODULE_LICENSE("GPL");
  90 MODULE_VERSION(DRV_VERSION);
  91
  92 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  93 static int debug = -1;
  94 module_param(debug, int, 0);
  95 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  96
  97 /* forward decls */
  98 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector);
  99 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter);
 100
 101 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
 102                                            struct ixgbevf_ring *rx_ring,
 103                                            u32 val)
 104 {
 105         /*
 106          * Force memory writes to complete before letting h/w
 107          * know there are new descriptors to fetch.  (Only
 108          * applicable for weak-ordered memory model archs,
 109          * such as IA-64).
 110          */
 111         wmb();
 112         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
 113 }
 114
 115 /**
 116  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 117  * @adapter: pointer to adapter struct
 118  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 119  * @queue: queue to map the corresponding interrupt to
 120  * @msix_vector: the vector to map to the corresponding queue
 121  */
 122 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 123                              u8 queue, u8 msix_vector)
 124 {
 125         u32 ivar, index;
 126         struct ixgbe_hw *hw = &adapter->hw;
 127         if (direction == -1) {
 128                 /* other causes */
 129                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 130                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 131                 ivar &= ~0xFF;
 132                 ivar |= msix_vector;
 133                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 134         } else {
 135                 /* tx or rx causes */
 136                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 137                 index = ((16 * (queue & 1)) + (8 * direction));
 138                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 139                 ivar &= ~(0xFF << index);
 140                 ivar |= (msix_vector << index);
 141                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 142         }
 143 }
 144
 145 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_ring *tx_ring,
 146                                                struct ixgbevf_tx_buffer
 147                                                *tx_buffer_info)
 148 {
 149         if (tx_buffer_info->dma) {
 150                 if (tx_buffer_info->mapped_as_page)
 151                         dma_unmap_page(tx_ring->dev,
 152                                        tx_buffer_info->dma,
 153                                        tx_buffer_info->length,
 154                                        DMA_TO_DEVICE);
 155                 else
 156                         dma_unmap_single(tx_ring->dev,
 157                                          tx_buffer_info->dma,
 158                                          tx_buffer_info->length,
 159                                          DMA_TO_DEVICE);
 160                 tx_buffer_info->dma = 0;
 161         }
 162         if (tx_buffer_info->skb) {
 163                 dev_kfree_skb_any(tx_buffer_info->skb);
 164                 tx_buffer_info->skb = NULL;
 165         }
 166         tx_buffer_info->time_stamp = 0;
 167         /* tx_buffer_info must be completely set up in the transmit path */
 168 }
 169
 170 #define IXGBE_MAX_TXD_PWR       14
 171 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 172
 173 /* Tx Descriptors needed, worst case */
 174 #define TXD_USE_COUNT(S) DIV_ROUND_UP((S), IXGBE_MAX_DATA_PER_TXD)
 175 #define DESC_NEEDED (MAX_SKB_FRAGS + 4)
 176
 177 static void ixgbevf_tx_timeout(struct net_device *netdev);
 178
 179 /**
 180  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 181  * @q_vector: board private structure
 182  * @tx_ring: tx ring to clean
 183  **/
 184 static bool ixgbevf_clean_tx_irq(struct ixgbevf_q_vector *q_vector,
 185                                  struct ixgbevf_ring *tx_ring)
 186 {
 187         struct ixgbevf_adapter *adapter = q_vector->adapter;
 188         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 189         struct ixgbevf_tx_buffer *tx_buffer_info;
 190         unsigned int i, count = 0;
 191         unsigned int total_bytes = 0, total_packets = 0;
 192
 193         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 194                 return true;
 195
 196         i = tx_ring->next_to_clean;
 197         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 198         eop_desc = tx_buffer_info->next_to_watch;
 199
 200         do {
 201                 bool cleaned = false;
 202
 203                 /* if next_to_watch is not set then there is no work pending */
 204                 if (!eop_desc)
 205                         break;
 206
 207                 /* prevent any other reads prior to eop_desc */
 208                 read_barrier_depends();
 209
 210                 /* if DD is not set pending work has not been completed */
 211                 if (!(eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)))
 212                         break;
 213
 214                 /* clear next_to_watch to prevent false hangs */
 215                 tx_buffer_info->next_to_watch = NULL;
 216
 217                 for ( ; !cleaned; count++) {
 218                         struct sk_buff *skb;
 219                         tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
 220                         cleaned = (tx_desc == eop_desc);
 221                         skb = tx_buffer_info->skb;
 222
 223                         if (cleaned && skb) {
 224                                 unsigned int segs, bytecount;
 225
 226                                 /* gso_segs is currently only valid for tcp */
 227                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 228                                 /* multiply data chunks by size of headers */
 229                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 230                                             skb->len;
 231                                 total_packets += segs;
 232                                 total_bytes += bytecount;
 233                         }
 234
 235                         ixgbevf_unmap_and_free_tx_resource(tx_ring,
 236                                                            tx_buffer_info);
 237
 238                         tx_desc->wb.status = 0;
 239
 240                         i++;
 241                         if (i == tx_ring->count)
 242                                 i = 0;
 243
 244                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 245                 }
 246
 247                 eop_desc = tx_buffer_info->next_to_watch;
 248         } while (count < tx_ring->count);
 249
 250         tx_ring->next_to_clean = i;
 251
 252 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 253         if (unlikely(count && netif_carrier_ok(tx_ring->netdev) &&
 254                      (ixgbevf_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
 255                 /* Make sure that anybody stopping the queue after this
 256                  * sees the new next_to_clean.
 257                  */
 258                 smp_mb();
 259                 if (__netif_subqueue_stopped(tx_ring->netdev,
 260                                              tx_ring->queue_index) &&
 261                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 262                         netif_wake_subqueue(tx_ring->netdev,
 263                                             tx_ring->queue_index);
 264                         ++adapter->restart_queue;
 265                 }
 266         }
 267
 268         u64_stats_update_begin(&tx_ring->syncp);
 269         tx_ring->total_bytes += total_bytes;
 270         tx_ring->total_packets += total_packets;
 271         u64_stats_update_end(&tx_ring->syncp);
 272         q_vector->tx.total_bytes += total_bytes;
 273         q_vector->tx.total_packets += total_packets;
 274
 275         return count < tx_ring->count;
 276 }
 277
 278 /**
 279  * ixgbevf_receive_skb - Send a completed packet up the stack
 280  * @q_vector: structure containing interrupt and ring information
 281  * @skb: packet to send up
 282  * @status: hardware indication of status of receive
 283  * @rx_desc: rx descriptor
 284  **/
 285 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
 286                                 struct sk_buff *skb, u8 status,
 287                                 union ixgbe_adv_rx_desc *rx_desc)
 288 {
 289         struct ixgbevf_adapter *adapter = q_vector->adapter;
 290         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 291         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 292
 293         if (is_vlan && test_bit(tag & VLAN_VID_MASK, adapter->active_vlans))
 294                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), tag);
 295
 296         if (!(adapter->flags & IXGBE_FLAG_IN_NETPOLL))
 297                 napi_gro_receive(&q_vector->napi, skb);
 298         else
 299                 netif_rx(skb);
 300 }
 301
 302 /**
 303  * ixgbevf_rx_skb - Helper function to determine proper Rx method
 304  * @q_vector: structure containing interrupt and ring information
 305  * @skb: packet to send up
 306  * @status: hardware indication of status of receive
 307  * @rx_desc: rx descriptor
 308  **/
 309 static void ixgbevf_rx_skb(struct ixgbevf_q_vector *q_vector,
 310                            struct sk_buff *skb, u8 status,
 311                            union ixgbe_adv_rx_desc *rx_desc)
 312 {
 313 #ifdef CONFIG_NET_RX_BUSY_POLL
 314         skb_mark_napi_id(skb, &q_vector->napi);
 315
 316         if (ixgbevf_qv_busy_polling(q_vector)) {
 317                 netif_receive_skb(skb);
 318                 /* exit early if we busy polled */
 319                 return;
 320         }
 321 #endif /* CONFIG_NET_RX_BUSY_POLL */
 322
 323         ixgbevf_receive_skb(q_vector, skb, status, rx_desc);
 324 }
 325
 326 /**
 327  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 328  * @ring: pointer to Rx descriptor ring structure
 329  * @status_err: hardware indication of status of receive
 330  * @skb: skb currently being received and modified
 331  **/
 332 static inline void ixgbevf_rx_checksum(struct ixgbevf_ring *ring,
 333                                        u32 status_err, struct sk_buff *skb)
 334 {
 335         skb_checksum_none_assert(skb);
 336
 337         /* Rx csum disabled */
 338         if (!(ring->netdev->features & NETIF_F_RXCSUM))
 339                 return;
 340
 341         /* if IP and error */
 342         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 343             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 344                 ring->hw_csum_rx_error++;
 345                 return;
 346         }
 347
 348         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 349                 return;
 350
 351         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 352                 ring->hw_csum_rx_error++;
 353                 return;
 354         }
 355
 356         /* It must be a TCP or UDP packet with a valid checksum */
 357         skb->ip_summed = CHECKSUM_UNNECESSARY;
 358         ring->hw_csum_rx_good++;
 359 }
 360
 361 /**
 362  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 363  * @adapter: address of board private structure
 364  **/
 365 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
 366                                      struct ixgbevf_ring *rx_ring,
 367                                      int cleaned_count)
 368 {
 369         struct pci_dev *pdev = adapter->pdev;
 370         union ixgbe_adv_rx_desc *rx_desc;
 371         struct ixgbevf_rx_buffer *bi;
 372         unsigned int i = rx_ring->next_to_use;
 373
 374         bi = &rx_ring->rx_buffer_info[i];
 375
 376         while (cleaned_count--) {
 377                 rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 378
 379                 if (!bi->skb) {
 380                         struct sk_buff *skb;
 381
 382                         skb = netdev_alloc_skb_ip_align(rx_ring->netdev,
 383                                                         rx_ring->rx_buf_len);
 384                         if (!skb) {
 385                                 adapter->alloc_rx_buff_failed++;
 386                                 goto no_buffers;
 387                         }
 388                         bi->skb = skb;
 389
 390                         bi->dma = dma_map_single(&pdev->dev, skb->data,
 391                                                  rx_ring->rx_buf_len,
 392                                                  DMA_FROM_DEVICE);
 393                         if (dma_mapping_error(&pdev->dev, bi->dma)) {
 394                                 dev_kfree_skb(skb);
 395                                 bi->skb = NULL;
 396                                 dev_err(&pdev->dev, "RX DMA map failed\n");
 397                                 break;
 398                         }
 399                 }
 400                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
 401
 402                 i++;
 403                 if (i == rx_ring->count)
 404                         i = 0;
 405                 bi = &rx_ring->rx_buffer_info[i];
 406         }
 407
 408 no_buffers:
 409         if (rx_ring->next_to_use != i) {
 410                 rx_ring->next_to_use = i;
 411                 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
 412         }
 413 }
 414
 415 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 416                                              u32 qmask)
 417 {
 418         struct ixgbe_hw *hw = &adapter->hw;
 419
 420         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, qmask);
 421 }
 422
 423 static int ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 424                                 struct ixgbevf_ring *rx_ring,
 425                                 int budget)
 426 {
 427         struct ixgbevf_adapter *adapter = q_vector->adapter;
 428         struct pci_dev *pdev = adapter->pdev;
 429         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 430         struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
 431         struct sk_buff *skb;
 432         unsigned int i;
 433         u32 len, staterr;
 434         int cleaned_count = 0;
 435         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 436
 437         i = rx_ring->next_to_clean;
 438         rx_desc = IXGBEVF_RX_DESC(rx_ring, i);
 439         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 440         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 441
 442         while (staterr & IXGBE_RXD_STAT_DD) {
 443                 if (!budget)
 444                         break;
 445                 budget--;
 446
 447                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 448                 len = le16_to_cpu(rx_desc->wb.upper.length);
 449                 skb = rx_buffer_info->skb;
 450                 prefetch(skb->data - NET_IP_ALIGN);
 451                 rx_buffer_info->skb = NULL;
 452
 453                 if (rx_buffer_info->dma) {
 454                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
 455                                          rx_ring->rx_buf_len,
 456                                          DMA_FROM_DEVICE);
 457                         rx_buffer_info->dma = 0;
 458                         skb_put(skb, len);
 459                 }
 460
 461                 i++;
 462                 if (i == rx_ring->count)
 463                         i = 0;
 464
 465                 next_rxd = IXGBEVF_RX_DESC(rx_ring, i);
 466                 prefetch(next_rxd);
 467                 cleaned_count++;
 468
 469                 next_buffer = &rx_ring->rx_buffer_info[i];
 470
 471                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
 472                         skb->next = next_buffer->skb;
 473                         IXGBE_CB(skb->next)->prev = skb;
 474                         adapter->non_eop_descs++;
 475                         goto next_desc;
 476                 }
 477
 478                 /* we should not be chaining buffers, if we did drop the skb */
 479                 if (IXGBE_CB(skb)->prev) {
 480                         do {
 481                                 struct sk_buff *this = skb;
 482                                 skb = IXGBE_CB(skb)->prev;
 483                                 dev_kfree_skb(this);
 484                         } while (skb);
 485                         goto next_desc;
 486                 }
 487
 488                 /* ERR_MASK will only have valid bits if EOP set */
 489                 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
 490                         dev_kfree_skb_irq(skb);
 491                         goto next_desc;
 492                 }
 493
 494                 ixgbevf_rx_checksum(rx_ring, staterr, skb);
 495
 496                 /* probably a little skewed due to removing CRC */
 497                 total_rx_bytes += skb->len;
 498                 total_rx_packets++;
 499
 500                 skb->protocol = eth_type_trans(skb, rx_ring->netdev);
 501
 502                 /* Workaround hardware that can't do proper VEPA multicast
 503                  * source pruning.
 504                  */
 505                 if ((skb->pkt_type & (PACKET_BROADCAST | PACKET_MULTICAST)) &&
 506                     ether_addr_equal(adapter->netdev->dev_addr,
 507                                      eth_hdr(skb)->h_source)) {
 508                         dev_kfree_skb_irq(skb);
 509                         goto next_desc;
 510                 }
 511
 512                 ixgbevf_rx_skb(q_vector, skb, staterr, rx_desc);
 513
 514 next_desc:
 515                 rx_desc->wb.upper.status_error = 0;
 516
 517                 /* return some buffers to hardware, one at a time is too slow */
 518                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 519                         ixgbevf_alloc_rx_buffers(adapter, rx_ring,
 520                                                  cleaned_count);
 521                         cleaned_count = 0;
 522                 }
 523
 524                 /* use prefetched values */
 525                 rx_desc = next_rxd;
 526                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
 527
 528                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 529         }
 530
 531         rx_ring->next_to_clean = i;
 532         cleaned_count = ixgbevf_desc_unused(rx_ring);
 533
 534         if (cleaned_count)
 535                 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
 536
 537         u64_stats_update_begin(&rx_ring->syncp);
 538         rx_ring->total_packets += total_rx_packets;
 539         rx_ring->total_bytes += total_rx_bytes;
 540         u64_stats_update_end(&rx_ring->syncp);
 541         q_vector->rx.total_packets += total_rx_packets;
 542         q_vector->rx.total_bytes += total_rx_bytes;
 543
 544         return total_rx_packets;
 545 }
 546
 547 /**
 548  * ixgbevf_poll - NAPI polling calback
 549  * @napi: napi struct with our devices info in it
 550  * @budget: amount of work driver is allowed to do this pass, in packets
 551  *
 552  * This function will clean more than one or more rings associated with a
 553  * q_vector.
 554  **/
 555 static int ixgbevf_poll(struct napi_struct *napi, int budget)
 556 {
 557         struct ixgbevf_q_vector *q_vector =
 558                 container_of(napi, struct ixgbevf_q_vector, napi);
 559         struct ixgbevf_adapter *adapter = q_vector->adapter;
 560         struct ixgbevf_ring *ring;
 561         int per_ring_budget;
 562         bool clean_complete = true;
 563
 564         ixgbevf_for_each_ring(ring, q_vector->tx)
 565                 clean_complete &= ixgbevf_clean_tx_irq(q_vector, ring);
 566
 567 #ifdef CONFIG_NET_RX_BUSY_POLL
 568         if (!ixgbevf_qv_lock_napi(q_vector))
 569                 return budget;
 570 #endif
 571
 572         /* attempt to distribute budget to each queue fairly, but don't allow
 573          * the budget to go below 1 because we'll exit polling */
 574         if (q_vector->rx.count > 1)
 575                 per_ring_budget = max(budget/q_vector->rx.count, 1);
 576         else
 577                 per_ring_budget = budget;
 578
 579         adapter->flags |= IXGBE_FLAG_IN_NETPOLL;
 580         ixgbevf_for_each_ring(ring, q_vector->rx)
 581                 clean_complete &= (ixgbevf_clean_rx_irq(q_vector, ring,
 582                                                         per_ring_budget)
 583                                    < per_ring_budget);
 584         adapter->flags &= ~IXGBE_FLAG_IN_NETPOLL;
 585
 586 #ifdef CONFIG_NET_RX_BUSY_POLL
 587         ixgbevf_qv_unlock_napi(q_vector);
 588 #endif
 589
 590         /* If all work not completed, return budget and keep polling */
 591         if (!clean_complete)
 592                 return budget;
 593         /* all work done, exit the polling mode */
 594         napi_complete(napi);
 595         if (adapter->rx_itr_setting & 1)
 596                 ixgbevf_set_itr(q_vector);
 597         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
 598                 ixgbevf_irq_enable_queues(adapter,
 599                                           1 << q_vector->v_idx);
 600
 601         return 0;
 602 }
 603
 604 /**
 605  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
 606  * @q_vector: structure containing interrupt and ring information
 607  */
 608 void ixgbevf_write_eitr(struct ixgbevf_q_vector *q_vector)
 609 {
 610         struct ixgbevf_adapter *adapter = q_vector->adapter;
 611         struct ixgbe_hw *hw = &adapter->hw;
 612         int v_idx = q_vector->v_idx;
 613         u32 itr_reg = q_vector->itr & IXGBE_MAX_EITR;
 614
 615         /*
 616          * set the WDIS bit to not clear the timer bits and cause an
 617          * immediate assertion of the interrupt
 618          */
 619         itr_reg |= IXGBE_EITR_CNT_WDIS;
 620
 621         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
 622 }
 623
 624 #ifdef CONFIG_NET_RX_BUSY_POLL
 625 /* must be called with local_bh_disable()d */
 626 static int ixgbevf_busy_poll_recv(struct napi_struct *napi)
 627 {
 628         struct ixgbevf_q_vector *q_vector =
 629                         container_of(napi, struct ixgbevf_q_vector, napi);
 630         struct ixgbevf_adapter *adapter = q_vector->adapter;
 631         struct ixgbevf_ring  *ring;
 632         int found = 0;
 633
 634         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
 635                 return LL_FLUSH_FAILED;
 636
 637         if (!ixgbevf_qv_lock_poll(q_vector))
 638                 return LL_FLUSH_BUSY;
 639
 640         ixgbevf_for_each_ring(ring, q_vector->rx) {
 641                 found = ixgbevf_clean_rx_irq(q_vector, ring, 4);
 642 #ifdef BP_EXTENDED_STATS
 643                 if (found)
 644                         ring->bp_cleaned += found;
 645                 else
 646                         ring->bp_misses++;
 647 #endif
 648                 if (found)
 649                         break;
 650         }
 651
 652         ixgbevf_qv_unlock_poll(q_vector);
 653
 654         return found;
 655 }
 656 #endif /* CONFIG_NET_RX_BUSY_POLL */
 657
 658 /**
 659  * ixgbevf_configure_msix - Configure MSI-X hardware
 660  * @adapter: board private structure
 661  *
 662  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
 663  * interrupts.
 664  **/
 665 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
 666 {
 667         struct ixgbevf_q_vector *q_vector;
 668         int q_vectors, v_idx;
 669
 670         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 671         adapter->eims_enable_mask = 0;
 672
 673         /*
 674          * Populate the IVAR table and set the ITR values to the
 675          * corresponding register.
 676          */
 677         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
 678                 struct ixgbevf_ring *ring;
 679                 q_vector = adapter->q_vector[v_idx];
 680
 681                 ixgbevf_for_each_ring(ring, q_vector->rx)
 682                         ixgbevf_set_ivar(adapter, 0, ring->reg_idx, v_idx);
 683
 684                 ixgbevf_for_each_ring(ring, q_vector->tx)
 685                         ixgbevf_set_ivar(adapter, 1, ring->reg_idx, v_idx);
 686
 687                 if (q_vector->tx.ring && !q_vector->rx.ring) {
 688                         /* tx only vector */
 689                         if (adapter->tx_itr_setting == 1)
 690                                 q_vector->itr = IXGBE_10K_ITR;
 691                         else
 692                                 q_vector->itr = adapter->tx_itr_setting;
 693                 } else {
 694                         /* rx or rx/tx vector */
 695                         if (adapter->rx_itr_setting == 1)
 696                                 q_vector->itr = IXGBE_20K_ITR;
 697                         else
 698                                 q_vector->itr = adapter->rx_itr_setting;
 699                 }
 700
 701                 /* add q_vector eims value to global eims_enable_mask */
 702                 adapter->eims_enable_mask |= 1 << v_idx;
 703
 704                 ixgbevf_write_eitr(q_vector);
 705         }
 706
 707         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
 708         /* setup eims_other and add value to global eims_enable_mask */
 709         adapter->eims_other = 1 << v_idx;
 710         adapter->eims_enable_mask |= adapter->eims_other;
 711 }
 712
 713 enum latency_range {
 714         lowest_latency = 0,
 715         low_latency = 1,
 716         bulk_latency = 2,
 717         latency_invalid = 255
 718 };
 719
 720 /**
 721  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
 722  * @q_vector: structure containing interrupt and ring information
 723  * @ring_container: structure containing ring performance data
 724  *
 725  *      Stores a new ITR value based on packets and byte
 726  *      counts during the last interrupt.  The advantage of per interrupt
 727  *      computation is faster updates and more accurate ITR for the current
 728  *      traffic pattern.  Constants in this function were computed
 729  *      based on theoretical maximum wire speed and thresholds were set based
 730  *      on testing data as well as attempting to minimize response time
 731  *      while increasing bulk throughput.
 732  **/
 733 static void ixgbevf_update_itr(struct ixgbevf_q_vector *q_vector,
 734                                struct ixgbevf_ring_container *ring_container)
 735 {
 736         int bytes = ring_container->total_bytes;
 737         int packets = ring_container->total_packets;
 738         u32 timepassed_us;
 739         u64 bytes_perint;
 740         u8 itr_setting = ring_container->itr;
 741
 742         if (packets == 0)
 743                 return;
 744
 745         /* simple throttlerate management
 746          *    0-20MB/s lowest (100000 ints/s)
 747          *   20-100MB/s low   (20000 ints/s)
 748          *  100-1249MB/s bulk (8000 ints/s)
 749          */
 750         /* what was last interrupt timeslice? */
 751         timepassed_us = q_vector->itr >> 2;
 752         bytes_perint = bytes / timepassed_us; /* bytes/usec */
 753
 754         switch (itr_setting) {
 755         case lowest_latency:
 756                 if (bytes_perint > 10)
 757                         itr_setting = low_latency;
 758                 break;
 759         case low_latency:
 760                 if (bytes_perint > 20)
 761                         itr_setting = bulk_latency;
 762                 else if (bytes_perint <= 10)
 763                         itr_setting = lowest_latency;
 764                 break;
 765         case bulk_latency:
 766                 if (bytes_perint <= 20)
 767                         itr_setting = low_latency;
 768                 break;
 769         }
 770
 771         /* clear work counters since we have the values we need */
 772         ring_container->total_bytes = 0;
 773         ring_container->total_packets = 0;
 774
 775         /* write updated itr to ring container */
 776         ring_container->itr = itr_setting;
 777 }
 778
 779 static void ixgbevf_set_itr(struct ixgbevf_q_vector *q_vector)
 780 {
 781         u32 new_itr = q_vector->itr;
 782         u8 current_itr;
 783
 784         ixgbevf_update_itr(q_vector, &q_vector->tx);
 785         ixgbevf_update_itr(q_vector, &q_vector->rx);
 786
 787         current_itr = max(q_vector->rx.itr, q_vector->tx.itr);
 788
 789         switch (current_itr) {
 790         /* counts and packets in update_itr are dependent on these numbers */
 791         case lowest_latency:
 792                 new_itr = IXGBE_100K_ITR;
 793                 break;
 794         case low_latency:
 795                 new_itr = IXGBE_20K_ITR;
 796                 break;
 797         case bulk_latency:
 798         default:
 799                 new_itr = IXGBE_8K_ITR;
 800                 break;
 801         }
 802
 803         if (new_itr != q_vector->itr) {
 804                 /* do an exponential smoothing */
 805                 new_itr = (10 * new_itr * q_vector->itr) /
 806                           ((9 * new_itr) + q_vector->itr);
 807
 808                 /* save the algorithm value here */
 809                 q_vector->itr = new_itr;
 810
 811                 ixgbevf_write_eitr(q_vector);
 812         }
 813 }
 814
 815 static irqreturn_t ixgbevf_msix_other(int irq, void *data)
 816 {
 817         struct ixgbevf_adapter *adapter = data;
 818         struct ixgbe_hw *hw = &adapter->hw;
 819
 820         hw->mac.get_link_status = 1;
 821
 822         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
 823                 mod_timer(&adapter->watchdog_timer, jiffies);
 824
 825         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_other);
 826
 827         return IRQ_HANDLED;
 828 }
 829
 830 /**
 831  * ixgbevf_msix_clean_rings - single unshared vector rx clean (all queues)
 832  * @irq: unused
 833  * @data: pointer to our q_vector struct for this interrupt vector
 834  **/
 835 static irqreturn_t ixgbevf_msix_clean_rings(int irq, void *data)
 836 {
 837         struct ixgbevf_q_vector *q_vector = data;
 838
 839         /* EIAM disabled interrupts (on this vector) for us */
 840         if (q_vector->rx.ring || q_vector->tx.ring)
 841                 napi_schedule(&q_vector->napi);
 842
 843         return IRQ_HANDLED;
 844 }
 845
 846 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
 847                                      int r_idx)
 848 {
 849         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
 850
 851         a->rx_ring[r_idx].next = q_vector->rx.ring;
 852         q_vector->rx.ring = &a->rx_ring[r_idx];
 853         q_vector->rx.count++;
 854 }
 855
 856 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
 857                                      int t_idx)
 858 {
 859         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
 860
 861         a->tx_ring[t_idx].next = q_vector->tx.ring;
 862         q_vector->tx.ring = &a->tx_ring[t_idx];
 863         q_vector->tx.count++;
 864 }
 865
 866 /**
 867  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
 868  * @adapter: board private structure to initialize
 869  *
 870  * This function maps descriptor rings to the queue-specific vectors
 871  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
 872  * one vector per ring/queue, but on a constrained vector budget, we
 873  * group the rings as "efficiently" as possible.  You would add new
 874  * mapping configurations in here.
 875  **/
 876 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
 877 {
 878         int q_vectors;
 879         int v_start = 0;
 880         int rxr_idx = 0, txr_idx = 0;
 881         int rxr_remaining = adapter->num_rx_queues;
 882         int txr_remaining = adapter->num_tx_queues;
 883         int i, j;
 884         int rqpv, tqpv;
 885         int err = 0;
 886
 887         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 888
 889         /*
 890          * The ideal configuration...
 891          * We have enough vectors to map one per queue.
 892          */
 893         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
 894                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
 895                         map_vector_to_rxq(adapter, v_start, rxr_idx);
 896
 897                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
 898                         map_vector_to_txq(adapter, v_start, txr_idx);
 899                 goto out;
 900         }
 901
 902         /*
 903          * If we don't have enough vectors for a 1-to-1
 904          * mapping, we'll have to group them so there are
 905          * multiple queues per vector.
 906          */
 907         /* Re-adjusting *qpv takes care of the remainder. */
 908         for (i = v_start; i < q_vectors; i++) {
 909                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
 910                 for (j = 0; j < rqpv; j++) {
 911                         map_vector_to_rxq(adapter, i, rxr_idx);
 912                         rxr_idx++;
 913                         rxr_remaining--;
 914                 }
 915         }
 916         for (i = v_start; i < q_vectors; i++) {
 917                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
 918                 for (j = 0; j < tqpv; j++) {
 919                         map_vector_to_txq(adapter, i, txr_idx);
 920                         txr_idx++;
 921                         txr_remaining--;
 922                 }
 923         }
 924
 925 out:
 926         return err;
 927 }
 928
 929 /**
 930  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
 931  * @adapter: board private structure
 932  *
 933  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
 934  * interrupts from the kernel.
 935  **/
 936 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
 937 {
 938         struct net_device *netdev = adapter->netdev;
 939         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 940         int vector, err;
 941         int ri = 0, ti = 0;
 942
 943         for (vector = 0; vector < q_vectors; vector++) {
 944                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[vector];
 945                 struct msix_entry *entry = &adapter->msix_entries[vector];
 946
 947                 if (q_vector->tx.ring && q_vector->rx.ring) {
 948                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
 949                                  "%s-%s-%d", netdev->name, "TxRx", ri++);
 950                         ti++;
 951                 } else if (q_vector->rx.ring) {
 952                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
 953                                  "%s-%s-%d", netdev->name, "rx", ri++);
 954                 } else if (q_vector->tx.ring) {
 955                         snprintf(q_vector->name, sizeof(q_vector->name) - 1,
 956                                  "%s-%s-%d", netdev->name, "tx", ti++);
 957                 } else {
 958                         /* skip this unused q_vector */
 959                         continue;
 960                 }
 961                 err = request_irq(entry->vector, &ixgbevf_msix_clean_rings, 0,
 962                                   q_vector->name, q_vector);
 963                 if (err) {
 964                         hw_dbg(&adapter->hw,
 965                                "request_irq failed for MSIX interrupt "
 966                                "Error: %d\n", err);
 967                         goto free_queue_irqs;
 968                 }
 969         }
 970
 971         err = request_irq(adapter->msix_entries[vector].vector,
 972                           &ixgbevf_msix_other, 0, netdev->name, adapter);
 973         if (err) {
 974                 hw_dbg(&adapter->hw,
 975                        "request_irq for msix_other failed: %d\n", err);
 976                 goto free_queue_irqs;
 977         }
 978
 979         return 0;
 980
 981 free_queue_irqs:
 982         while (vector) {
 983                 vector--;
 984                 free_irq(adapter->msix_entries[vector].vector,
 985                          adapter->q_vector[vector]);
 986         }
 987         /* This failure is non-recoverable - it indicates the system is
 988          * out of MSIX vector resources and the VF driver cannot run
 989          * without them.  Set the number of msix vectors to zero
 990          * indicating that not enough can be allocated.  The error
 991          * will be returned to the user indicating device open failed.
 992          * Any further attempts to force the driver to open will also
 993          * fail.  The only way to recover is to unload the driver and
 994          * reload it again.  If the system has recovered some MSIX
 995          * vectors then it may succeed.
 996          */
 997         adapter->num_msix_vectors = 0;
 998         return err;
 999 }
1000
1001 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1002 {
1003         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1004
1005         for (i = 0; i < q_vectors; i++) {
1006                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1007                 q_vector->rx.ring = NULL;
1008                 q_vector->tx.ring = NULL;
1009                 q_vector->rx.count = 0;
1010                 q_vector->tx.count = 0;
1011         }
1012 }
1013
1014 /**
1015  * ixgbevf_request_irq - initialize interrupts
1016  * @adapter: board private structure
1017  *
1018  * Attempts to configure interrupts using the best available
1019  * capabilities of the hardware and kernel.
1020  **/
1021 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1022 {
1023         int err = 0;
1024
1025         err = ixgbevf_request_msix_irqs(adapter);
1026
1027         if (err)
1028                 hw_dbg(&adapter->hw,
1029                        "request_irq failed, Error %d\n", err);
1030
1031         return err;
1032 }
1033
1034 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1035 {
1036         int i, q_vectors;
1037
1038         q_vectors = adapter->num_msix_vectors;
1039         i = q_vectors - 1;
1040
1041         free_irq(adapter->msix_entries[i].vector, adapter);
1042         i--;
1043
1044         for (; i >= 0; i--) {
1045                 /* free only the irqs that were actually requested */
1046                 if (!adapter->q_vector[i]->rx.ring &&
1047                     !adapter->q_vector[i]->tx.ring)
1048                         continue;
1049
1050                 free_irq(adapter->msix_entries[i].vector,
1051                          adapter->q_vector[i]);
1052         }
1053
1054         ixgbevf_reset_q_vectors(adapter);
1055 }
1056
1057 /**
1058  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1059  * @adapter: board private structure
1060  **/
1061 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1062 {
1063         struct ixgbe_hw *hw = &adapter->hw;
1064         int i;
1065
1066         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, 0);
1067         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1068         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, 0);
1069
1070         IXGBE_WRITE_FLUSH(hw);
1071
1072         for (i = 0; i < adapter->num_msix_vectors; i++)
1073                 synchronize_irq(adapter->msix_entries[i].vector);
1074 }
1075
1076 /**
1077  * ixgbevf_irq_enable - Enable default interrupt generation settings
1078  * @adapter: board private structure
1079  **/
1080 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter)
1081 {
1082         struct ixgbe_hw *hw = &adapter->hw;
1083
1084         IXGBE_WRITE_REG(hw, IXGBE_VTEIAM, adapter->eims_enable_mask);
1085         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, adapter->eims_enable_mask);
1086         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, adapter->eims_enable_mask);
1087 }
1088
1089 /**
1090  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1091  * @adapter: board private structure
1092  *
1093  * Configure the Tx unit of the MAC after a reset.
1094  **/
1095 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1096 {
1097         u64 tdba;
1098         struct ixgbe_hw *hw = &adapter->hw;
1099         u32 i, j, tdlen, txctrl;
1100
1101         /* Setup the HW Tx Head and Tail descriptor pointers */
1102         for (i = 0; i < adapter->num_tx_queues; i++) {
1103                 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1104                 j = ring->reg_idx;
1105                 tdba = ring->dma;
1106                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1107                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1108                                 (tdba & DMA_BIT_MASK(32)));
1109                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1110                 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1111                 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1112                 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1113                 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1114                 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1115                 /* Disable Tx Head Writeback RO bit, since this hoses
1116                  * bookkeeping if things aren't delivered in order.
1117                  */
1118                 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1119                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1120                 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1121         }
1122 }
1123
1124 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1125
1126 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1127 {
1128         struct ixgbevf_ring *rx_ring;
1129         struct ixgbe_hw *hw = &adapter->hw;
1130         u32 srrctl;
1131
1132         rx_ring = &adapter->rx_ring[index];
1133
1134         srrctl = IXGBE_SRRCTL_DROP_EN;
1135
1136         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1137
1138         srrctl |= ALIGN(rx_ring->rx_buf_len, 1024) >>
1139                   IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1140
1141         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1142 }
1143
1144 static void ixgbevf_setup_psrtype(struct ixgbevf_adapter *adapter)
1145 {
1146         struct ixgbe_hw *hw = &adapter->hw;
1147
1148         /* PSRTYPE must be initialized in 82599 */
1149         u32 psrtype = IXGBE_PSRTYPE_TCPHDR | IXGBE_PSRTYPE_UDPHDR |
1150                       IXGBE_PSRTYPE_IPV4HDR | IXGBE_PSRTYPE_IPV6HDR |
1151                       IXGBE_PSRTYPE_L2HDR;
1152
1153         if (adapter->num_rx_queues > 1)
1154                 psrtype |= 1 << 29;
1155
1156         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, psrtype);
1157 }
1158
1159 static void ixgbevf_set_rx_buffer_len(struct ixgbevf_adapter *adapter)
1160 {
1161         struct ixgbe_hw *hw = &adapter->hw;
1162         struct net_device *netdev = adapter->netdev;
1163         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1164         int i;
1165         u16 rx_buf_len;
1166
1167         /* notify the PF of our intent to use this size of frame */
1168         ixgbevf_rlpml_set_vf(hw, max_frame);
1169
1170         /* PF will allow an extra 4 bytes past for vlan tagged frames */
1171         max_frame += VLAN_HLEN;
1172
1173         /*
1174          * Allocate buffer sizes that fit well into 32K and
1175          * take into account max frame size of 9.5K
1176          */
1177         if ((hw->mac.type == ixgbe_mac_X540_vf) &&
1178             (max_frame <= MAXIMUM_ETHERNET_VLAN_SIZE))
1179                 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1180         else if (max_frame <= IXGBEVF_RXBUFFER_2K)
1181                 rx_buf_len = IXGBEVF_RXBUFFER_2K;
1182         else if (max_frame <= IXGBEVF_RXBUFFER_4K)
1183                 rx_buf_len = IXGBEVF_RXBUFFER_4K;
1184         else if (max_frame <= IXGBEVF_RXBUFFER_8K)
1185                 rx_buf_len = IXGBEVF_RXBUFFER_8K;
1186         else
1187                 rx_buf_len = IXGBEVF_RXBUFFER_10K;
1188
1189         for (i = 0; i < adapter->num_rx_queues; i++)
1190                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1191 }
1192
1193 /**
1194  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1195  * @adapter: board private structure
1196  *
1197  * Configure the Rx unit of the MAC after a reset.
1198  **/
1199 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1200 {
1201         u64 rdba;
1202         struct ixgbe_hw *hw = &adapter->hw;
1203         int i, j;
1204         u32 rdlen;
1205
1206         ixgbevf_setup_psrtype(adapter);
1207
1208         /* set_rx_buffer_len must be called before ring initialization */
1209         ixgbevf_set_rx_buffer_len(adapter);
1210
1211         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1212         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1213          * the Base and Length of the Rx Descriptor Ring */
1214         for (i = 0; i < adapter->num_rx_queues; i++) {
1215                 rdba = adapter->rx_ring[i].dma;
1216                 j = adapter->rx_ring[i].reg_idx;
1217                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1218                                 (rdba & DMA_BIT_MASK(32)));
1219                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1220                 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1221                 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1222                 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1223                 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1224                 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1225
1226                 ixgbevf_configure_srrctl(adapter, j);
1227         }
1228 }
1229
1230 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev,
1231                                    __be16 proto, u16 vid)
1232 {
1233         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1234         struct ixgbe_hw *hw = &adapter->hw;
1235         int err;
1236
1237         spin_lock_bh(&adapter->mbx_lock);
1238
1239         /* add VID to filter table */
1240         err = hw->mac.ops.set_vfta(hw, vid, 0, true);
1241
1242         spin_unlock_bh(&adapter->mbx_lock);
1243
1244         /* translate error return types so error makes sense */
1245         if (err == IXGBE_ERR_MBX)
1246                 return -EIO;
1247
1248         if (err == IXGBE_ERR_INVALID_ARGUMENT)
1249                 return -EACCES;
1250
1251         set_bit(vid, adapter->active_vlans);
1252
1253         return err;
1254 }
1255
1256 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev,
1257                                     __be16 proto, u16 vid)
1258 {
1259         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1260         struct ixgbe_hw *hw = &adapter->hw;
1261         int err = -EOPNOTSUPP;
1262
1263         spin_lock_bh(&adapter->mbx_lock);
1264
1265         /* remove VID from filter table */
1266         err = hw->mac.ops.set_vfta(hw, vid, 0, false);
1267
1268         spin_unlock_bh(&adapter->mbx_lock);
1269
1270         clear_bit(vid, adapter->active_vlans);
1271
1272         return err;
1273 }
1274
1275 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1276 {
1277         u16 vid;
1278
1279         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1280                 ixgbevf_vlan_rx_add_vid(adapter->netdev,
1281                                         htons(ETH_P_8021Q), vid);
1282 }
1283
1284 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1285 {
1286         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1287         struct ixgbe_hw *hw = &adapter->hw;
1288         int count = 0;
1289
1290         if ((netdev_uc_count(netdev)) > 10) {
1291                 pr_err("Too many unicast filters - No Space\n");
1292                 return -ENOSPC;
1293         }
1294
1295         if (!netdev_uc_empty(netdev)) {
1296                 struct netdev_hw_addr *ha;
1297                 netdev_for_each_uc_addr(ha, netdev) {
1298                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1299                         udelay(200);
1300                 }
1301         } else {
1302                 /*
1303                  * If the list is empty then send message to PF driver to
1304                  * clear all macvlans on this VF.
1305                  */
1306                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1307         }
1308
1309         return count;
1310 }
1311
1312 /**
1313  * ixgbevf_set_rx_mode - Multicast and unicast set
1314  * @netdev: network interface device structure
1315  *
1316  * The set_rx_method entry point is called whenever the multicast address
1317  * list, unicast address list or the network interface flags are updated.
1318  * This routine is responsible for configuring the hardware for proper
1319  * multicast mode and configuring requested unicast filters.
1320  **/
1321 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1322 {
1323         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1324         struct ixgbe_hw *hw = &adapter->hw;
1325
1326         spin_lock_bh(&adapter->mbx_lock);
1327
1328         /* reprogram multicast list */
1329         hw->mac.ops.update_mc_addr_list(hw, netdev);
1330
1331         ixgbevf_write_uc_addr_list(netdev);
1332
1333         spin_unlock_bh(&adapter->mbx_lock);
1334 }
1335
1336 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1337 {
1338         int q_idx;
1339         struct ixgbevf_q_vector *q_vector;
1340         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1341
1342         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1343                 q_vector = adapter->q_vector[q_idx];
1344 #ifdef CONFIG_NET_RX_BUSY_POLL
1345                 ixgbevf_qv_init_lock(adapter->q_vector[q_idx]);
1346 #endif
1347                 napi_enable(&q_vector->napi);
1348         }
1349 }
1350
1351 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1352 {
1353         int q_idx;
1354         struct ixgbevf_q_vector *q_vector;
1355         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1356
1357         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1358                 q_vector = adapter->q_vector[q_idx];
1359                 napi_disable(&q_vector->napi);
1360 #ifdef CONFIG_NET_RX_BUSY_POLL
1361                 while (!ixgbevf_qv_disable(adapter->q_vector[q_idx])) {
1362                         pr_info("QV %d locked\n", q_idx);
1363                         usleep_range(1000, 20000);
1364                 }
1365 #endif /* CONFIG_NET_RX_BUSY_POLL */
1366         }
1367 }
1368
1369 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1370 {
1371         struct net_device *netdev = adapter->netdev;
1372         int i;
1373
1374         ixgbevf_set_rx_mode(netdev);
1375
1376         ixgbevf_restore_vlan(adapter);
1377
1378         ixgbevf_configure_tx(adapter);
1379         ixgbevf_configure_rx(adapter);
1380         for (i = 0; i < adapter->num_rx_queues; i++) {
1381                 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1382                 ixgbevf_alloc_rx_buffers(adapter, ring,
1383                                          ixgbevf_desc_unused(ring));
1384         }
1385 }
1386
1387 #define IXGBEVF_MAX_RX_DESC_POLL 10
1388 static void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1389                                          int rxr)
1390 {
1391         struct ixgbe_hw *hw = &adapter->hw;
1392         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1393         u32 rxdctl;
1394         int j = adapter->rx_ring[rxr].reg_idx;
1395
1396         do {
1397                 usleep_range(1000, 2000);
1398                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1399         } while (--wait_loop && !(rxdctl & IXGBE_RXDCTL_ENABLE));
1400
1401         if (!wait_loop)
1402                 hw_dbg(hw, "RXDCTL.ENABLE queue %d not set while polling\n",
1403                        rxr);
1404
1405         ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1406                                 (adapter->rx_ring[rxr].count - 1));
1407 }
1408
1409 static void ixgbevf_disable_rx_queue(struct ixgbevf_adapter *adapter,
1410                                      struct ixgbevf_ring *ring)
1411 {
1412         struct ixgbe_hw *hw = &adapter->hw;
1413         int wait_loop = IXGBEVF_MAX_RX_DESC_POLL;
1414         u32 rxdctl;
1415         u8 reg_idx = ring->reg_idx;
1416
1417         rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1418         rxdctl &= ~IXGBE_RXDCTL_ENABLE;
1419
1420         /* write value back with RXDCTL.ENABLE bit cleared */
1421         IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(reg_idx), rxdctl);
1422
1423         /* the hardware may take up to 100us to really disable the rx queue */
1424         do {
1425                 udelay(10);
1426                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(reg_idx));
1427         } while (--wait_loop && (rxdctl & IXGBE_RXDCTL_ENABLE));
1428
1429         if (!wait_loop)
1430                 hw_dbg(hw, "RXDCTL.ENABLE queue %d not cleared while polling\n",
1431                        reg_idx);
1432 }
1433
1434 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1435 {
1436         /* Only save pre-reset stats if there are some */
1437         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1438                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1439                         adapter->stats.base_vfgprc;
1440                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1441                         adapter->stats.base_vfgptc;
1442                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1443                         adapter->stats.base_vfgorc;
1444                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1445                         adapter->stats.base_vfgotc;
1446                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1447                         adapter->stats.base_vfmprc;
1448         }
1449 }
1450
1451 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1452 {
1453         struct ixgbe_hw *hw = &adapter->hw;
1454
1455         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1456         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1457         adapter->stats.last_vfgorc |=
1458                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1459         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1460         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1461         adapter->stats.last_vfgotc |=
1462                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1463         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1464
1465         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1466         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1467         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1468         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1469         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1470 }
1471
1472 static void ixgbevf_negotiate_api(struct ixgbevf_adapter *adapter)
1473 {
1474         struct ixgbe_hw *hw = &adapter->hw;
1475         int api[] = { ixgbe_mbox_api_11,
1476                       ixgbe_mbox_api_10,
1477                       ixgbe_mbox_api_unknown };
1478         int err = 0, idx = 0;
1479
1480         spin_lock_bh(&adapter->mbx_lock);
1481
1482         while (api[idx] != ixgbe_mbox_api_unknown) {
1483                 err = ixgbevf_negotiate_api_version(hw, api[idx]);
1484                 if (!err)
1485                         break;
1486                 idx++;
1487         }
1488
1489         spin_unlock_bh(&adapter->mbx_lock);
1490 }
1491
1492 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1493 {
1494         struct net_device *netdev = adapter->netdev;
1495         struct ixgbe_hw *hw = &adapter->hw;
1496         int i, j = 0;
1497         int num_rx_rings = adapter->num_rx_queues;
1498         u32 txdctl, rxdctl;
1499
1500         for (i = 0; i < adapter->num_tx_queues; i++) {
1501                 j = adapter->tx_ring[i].reg_idx;
1502                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1503                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1504                 txdctl |= (8 << 16);
1505                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1506         }
1507
1508         for (i = 0; i < adapter->num_tx_queues; i++) {
1509                 j = adapter->tx_ring[i].reg_idx;
1510                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1511                 txdctl |= IXGBE_TXDCTL_ENABLE;
1512                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1513         }
1514
1515         for (i = 0; i < num_rx_rings; i++) {
1516                 j = adapter->rx_ring[i].reg_idx;
1517                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1518                 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1519                 if (hw->mac.type == ixgbe_mac_X540_vf) {
1520                         rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1521                         rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1522                                    IXGBE_RXDCTL_RLPML_EN);
1523                 }
1524                 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1525                 ixgbevf_rx_desc_queue_enable(adapter, i);
1526         }
1527
1528         ixgbevf_configure_msix(adapter);
1529
1530         spin_lock_bh(&adapter->mbx_lock);
1531
1532         if (is_valid_ether_addr(hw->mac.addr))
1533                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1534         else
1535                 hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1536
1537         spin_unlock_bh(&adapter->mbx_lock);
1538
1539         clear_bit(__IXGBEVF_DOWN, &adapter->state);
1540         ixgbevf_napi_enable_all(adapter);
1541
1542         /* enable transmits */
1543         netif_tx_start_all_queues(netdev);
1544
1545         ixgbevf_save_reset_stats(adapter);
1546         ixgbevf_init_last_counter_stats(adapter);
1547
1548         hw->mac.get_link_status = 1;
1549         mod_timer(&adapter->watchdog_timer, jiffies);
1550 }
1551
1552 static int ixgbevf_reset_queues(struct ixgbevf_adapter *adapter)
1553 {
1554         struct ixgbe_hw *hw = &adapter->hw;
1555         struct ixgbevf_ring *rx_ring;
1556         unsigned int def_q = 0;
1557         unsigned int num_tcs = 0;
1558         unsigned int num_rx_queues = 1;
1559         int err, i;
1560
1561         spin_lock_bh(&adapter->mbx_lock);
1562
1563         /* fetch queue configuration from the PF */
1564         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
1565
1566         spin_unlock_bh(&adapter->mbx_lock);
1567
1568         if (err)
1569                 return err;
1570
1571         if (num_tcs > 1) {
1572                 /* update default Tx ring register index */
1573                 adapter->tx_ring[0].reg_idx = def_q;
1574
1575                 /* we need as many queues as traffic classes */
1576                 num_rx_queues = num_tcs;
1577         }
1578
1579         /* nothing to do if we have the correct number of queues */
1580         if (adapter->num_rx_queues == num_rx_queues)
1581                 return 0;
1582
1583         /* allocate new rings */
1584         rx_ring = kcalloc(num_rx_queues,
1585                           sizeof(struct ixgbevf_ring), GFP_KERNEL);
1586         if (!rx_ring)
1587                 return -ENOMEM;
1588
1589         /* setup ring fields */
1590         for (i = 0; i < num_rx_queues; i++) {
1591                 rx_ring[i].count = adapter->rx_ring_count;
1592                 rx_ring[i].queue_index = i;
1593                 rx_ring[i].reg_idx = i;
1594                 rx_ring[i].dev = &adapter->pdev->dev;
1595                 rx_ring[i].netdev = adapter->netdev;
1596
1597                 /* allocate resources on the ring */
1598                 err = ixgbevf_setup_rx_resources(adapter, &rx_ring[i]);
1599                 if (err) {
1600                         while (i) {
1601                                 i--;
1602                                 ixgbevf_free_rx_resources(adapter, &rx_ring[i]);
1603                         }
1604                         kfree(rx_ring);
1605                         return err;
1606                 }
1607         }
1608
1609         /* free the existing rings and queues */
1610         ixgbevf_free_all_rx_resources(adapter);
1611         adapter->num_rx_queues = 0;
1612         kfree(adapter->rx_ring);
1613
1614         /* move new rings into position on the adapter struct */
1615         adapter->rx_ring = rx_ring;
1616         adapter->num_rx_queues = num_rx_queues;
1617
1618         /* reset ring to vector mapping */
1619         ixgbevf_reset_q_vectors(adapter);
1620         ixgbevf_map_rings_to_vectors(adapter);
1621
1622         return 0;
1623 }
1624
1625 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1626 {
1627         struct ixgbe_hw *hw = &adapter->hw;
1628
1629         ixgbevf_reset_queues(adapter);
1630
1631         ixgbevf_configure(adapter);
1632
1633         ixgbevf_up_complete(adapter);
1634
1635         /* clear any pending interrupts, may auto mask */
1636         IXGBE_READ_REG(hw, IXGBE_VTEICR);
1637
1638         ixgbevf_irq_enable(adapter);
1639 }
1640
1641 /**
1642  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1643  * @adapter: board private structure
1644  * @rx_ring: ring to free buffers from
1645  **/
1646 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1647                                   struct ixgbevf_ring *rx_ring)
1648 {
1649         struct pci_dev *pdev = adapter->pdev;
1650         unsigned long size;
1651         unsigned int i;
1652
1653         if (!rx_ring->rx_buffer_info)
1654                 return;
1655
1656         /* Free all the Rx ring sk_buffs */
1657         for (i = 0; i < rx_ring->count; i++) {
1658                 struct ixgbevf_rx_buffer *rx_buffer_info;
1659
1660                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1661                 if (rx_buffer_info->dma) {
1662                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1663                                          rx_ring->rx_buf_len,
1664                                          DMA_FROM_DEVICE);
1665                         rx_buffer_info->dma = 0;
1666                 }
1667                 if (rx_buffer_info->skb) {
1668                         struct sk_buff *skb = rx_buffer_info->skb;
1669                         rx_buffer_info->skb = NULL;
1670                         do {
1671                                 struct sk_buff *this = skb;
1672                                 skb = IXGBE_CB(skb)->prev;
1673                                 dev_kfree_skb(this);
1674                         } while (skb);
1675                 }
1676         }
1677
1678         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1679         memset(rx_ring->rx_buffer_info, 0, size);
1680
1681         /* Zero out the descriptor ring */
1682         memset(rx_ring->desc, 0, rx_ring->size);
1683
1684         rx_ring->next_to_clean = 0;
1685         rx_ring->next_to_use = 0;
1686
1687         if (rx_ring->head)
1688                 writel(0, adapter->hw.hw_addr + rx_ring->head);
1689         if (rx_ring->tail)
1690                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1691 }
1692
1693 /**
1694  * ixgbevf_clean_tx_ring - Free Tx Buffers
1695  * @adapter: board private structure
1696  * @tx_ring: ring to be cleaned
1697  **/
1698 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1699                                   struct ixgbevf_ring *tx_ring)
1700 {
1701         struct ixgbevf_tx_buffer *tx_buffer_info;
1702         unsigned long size;
1703         unsigned int i;
1704
1705         if (!tx_ring->tx_buffer_info)
1706                 return;
1707
1708         /* Free all the Tx ring sk_buffs */
1709         for (i = 0; i < tx_ring->count; i++) {
1710                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1711                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
1712         }
1713
1714         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1715         memset(tx_ring->tx_buffer_info, 0, size);
1716
1717         memset(tx_ring->desc, 0, tx_ring->size);
1718
1719         tx_ring->next_to_use = 0;
1720         tx_ring->next_to_clean = 0;
1721
1722         if (tx_ring->head)
1723                 writel(0, adapter->hw.hw_addr + tx_ring->head);
1724         if (tx_ring->tail)
1725                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1726 }
1727
1728 /**
1729  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1730  * @adapter: board private structure
1731  **/
1732 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1733 {
1734         int i;
1735
1736         for (i = 0; i < adapter->num_rx_queues; i++)
1737                 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1738 }
1739
1740 /**
1741  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1742  * @adapter: board private structure
1743  **/
1744 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1745 {
1746         int i;
1747
1748         for (i = 0; i < adapter->num_tx_queues; i++)
1749                 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1750 }
1751
1752 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1753 {
1754         struct net_device *netdev = adapter->netdev;
1755         struct ixgbe_hw *hw = &adapter->hw;
1756         u32 txdctl;
1757         int i, j;
1758
1759         /* signal that we are down to the interrupt handler */
1760         set_bit(__IXGBEVF_DOWN, &adapter->state);
1761
1762         /* disable all enabled rx queues */
1763         for (i = 0; i < adapter->num_rx_queues; i++)
1764                 ixgbevf_disable_rx_queue(adapter, &adapter->rx_ring[i]);
1765
1766         netif_tx_disable(netdev);
1767
1768         msleep(10);
1769
1770         netif_tx_stop_all_queues(netdev);
1771
1772         ixgbevf_irq_disable(adapter);
1773
1774         ixgbevf_napi_disable_all(adapter);
1775
1776         del_timer_sync(&adapter->watchdog_timer);
1777         /* can't call flush scheduled work here because it can deadlock
1778          * if linkwatch_event tries to acquire the rtnl_lock which we are
1779          * holding */
1780         while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1781                 msleep(1);
1782
1783         /* disable transmits in the hardware now that interrupts are off */
1784         for (i = 0; i < adapter->num_tx_queues; i++) {
1785                 j = adapter->tx_ring[i].reg_idx;
1786                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1787                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1788                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1789         }
1790
1791         netif_carrier_off(netdev);
1792
1793         if (!pci_channel_offline(adapter->pdev))
1794                 ixgbevf_reset(adapter);
1795
1796         ixgbevf_clean_all_tx_rings(adapter);
1797         ixgbevf_clean_all_rx_rings(adapter);
1798 }
1799
1800 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1801 {
1802         WARN_ON(in_interrupt());
1803
1804         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1805                 msleep(1);
1806
1807         ixgbevf_down(adapter);
1808         ixgbevf_up(adapter);
1809
1810         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1811 }
1812
1813 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1814 {
1815         struct ixgbe_hw *hw = &adapter->hw;
1816         struct net_device *netdev = adapter->netdev;
1817
1818         if (hw->mac.ops.reset_hw(hw)) {
1819                 hw_dbg(hw, "PF still resetting\n");
1820         } else {
1821                 hw->mac.ops.init_hw(hw);
1822                 ixgbevf_negotiate_api(adapter);
1823         }
1824
1825         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1826                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1827                        netdev->addr_len);
1828                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1829                        netdev->addr_len);
1830         }
1831 }
1832
1833 static int ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1834                                         int vectors)
1835 {
1836         int err = 0;
1837         int vector_threshold;
1838
1839         /* We'll want at least 2 (vector_threshold):
1840          * 1) TxQ[0] + RxQ[0] handler
1841          * 2) Other (Link Status Change, etc.)
1842          */
1843         vector_threshold = MIN_MSIX_COUNT;
1844
1845         /* The more we get, the more we will assign to Tx/Rx Cleanup
1846          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1847          * Right now, we simply care about how many we'll get; we'll
1848          * set them up later while requesting irq's.
1849          */
1850         while (vectors >= vector_threshold) {
1851                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1852                                       vectors);
1853                 if (!err || err < 0) /* Success or a nasty failure. */
1854                         break;
1855                 else /* err == number of vectors we should try again with */
1856                         vectors = err;
1857         }
1858
1859         if (vectors < vector_threshold)
1860                 err = -ENOMEM;
1861
1862         if (err) {
1863                 dev_err(&adapter->pdev->dev,
1864                         "Unable to allocate MSI-X interrupts\n");
1865                 kfree(adapter->msix_entries);
1866                 adapter->msix_entries = NULL;
1867         } else {
1868                 /*
1869                  * Adjust for only the vectors we'll use, which is minimum
1870                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1871                  * vectors we were allocated.
1872                  */
1873                 adapter->num_msix_vectors = vectors;
1874         }
1875
1876         return err;
1877 }
1878
1879 /**
1880  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1881  * @adapter: board private structure to initialize
1882  *
1883  * This is the top level queue allocation routine.  The order here is very
1884  * important, starting with the "most" number of features turned on at once,
1885  * and ending with the smallest set of features.  This way large combinations
1886  * can be allocated if they're turned on, and smaller combinations are the
1887  * fallthrough conditions.
1888  *
1889  **/
1890 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1891 {
1892         /* Start with base case */
1893         adapter->num_rx_queues = 1;
1894         adapter->num_tx_queues = 1;
1895 }
1896
1897 /**
1898  * ixgbevf_alloc_queues - Allocate memory for all rings
1899  * @adapter: board private structure to initialize
1900  *
1901  * We allocate one ring per queue at run-time since we don't know the
1902  * number of queues at compile-time.  The polling_netdev array is
1903  * intended for Multiqueue, but should work fine with a single queue.
1904  **/
1905 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1906 {
1907         int i;
1908
1909         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1910                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1911         if (!adapter->tx_ring)
1912                 goto err_tx_ring_allocation;
1913
1914         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1915                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1916         if (!adapter->rx_ring)
1917                 goto err_rx_ring_allocation;
1918
1919         for (i = 0; i < adapter->num_tx_queues; i++) {
1920                 adapter->tx_ring[i].count = adapter->tx_ring_count;
1921                 adapter->tx_ring[i].queue_index = i;
1922                 /* reg_idx may be remapped later by DCB config */
1923                 adapter->tx_ring[i].reg_idx = i;
1924                 adapter->tx_ring[i].dev = &adapter->pdev->dev;
1925                 adapter->tx_ring[i].netdev = adapter->netdev;
1926         }
1927
1928         for (i = 0; i < adapter->num_rx_queues; i++) {
1929                 adapter->rx_ring[i].count = adapter->rx_ring_count;
1930                 adapter->rx_ring[i].queue_index = i;
1931                 adapter->rx_ring[i].reg_idx = i;
1932                 adapter->rx_ring[i].dev = &adapter->pdev->dev;
1933                 adapter->rx_ring[i].netdev = adapter->netdev;
1934         }
1935
1936         return 0;
1937
1938 err_rx_ring_allocation:
1939         kfree(adapter->tx_ring);
1940 err_tx_ring_allocation:
1941         return -ENOMEM;
1942 }
1943
1944 /**
1945  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1946  * @adapter: board private structure to initialize
1947  *
1948  * Attempt to configure the interrupts using the best available
1949  * capabilities of the hardware and the kernel.
1950  **/
1951 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1952 {
1953         struct net_device *netdev = adapter->netdev;
1954         int err = 0;
1955         int vector, v_budget;
1956
1957         /*
1958          * It's easy to be greedy for MSI-X vectors, but it really
1959          * doesn't do us much good if we have a lot more vectors
1960          * than CPU's.  So let's be conservative and only ask for
1961          * (roughly) the same number of vectors as there are CPU's.
1962          * The default is to use pairs of vectors.
1963          */
1964         v_budget = max(adapter->num_rx_queues, adapter->num_tx_queues);
1965         v_budget = min_t(int, v_budget, num_online_cpus());
1966         v_budget += NON_Q_VECTORS;
1967
1968         /* A failure in MSI-X entry allocation isn't fatal, but it does
1969          * mean we disable MSI-X capabilities of the adapter. */
1970         adapter->msix_entries = kcalloc(v_budget,
1971                                         sizeof(struct msix_entry), GFP_KERNEL);
1972         if (!adapter->msix_entries) {
1973                 err = -ENOMEM;
1974                 goto out;
1975         }
1976
1977         for (vector = 0; vector < v_budget; vector++)
1978                 adapter->msix_entries[vector].entry = vector;
1979
1980         err = ixgbevf_acquire_msix_vectors(adapter, v_budget);
1981         if (err)
1982                 goto out;
1983
1984         err = netif_set_real_num_tx_queues(netdev, adapter->num_tx_queues);
1985         if (err)
1986                 goto out;
1987
1988         err = netif_set_real_num_rx_queues(netdev, adapter->num_rx_queues);
1989
1990 out:
1991         return err;
1992 }
1993
1994 /**
1995  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1996  * @adapter: board private structure to initialize
1997  *
1998  * We allocate one q_vector per queue interrupt.  If allocation fails we
1999  * return -ENOMEM.
2000  **/
2001 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
2002 {
2003         int q_idx, num_q_vectors;
2004         struct ixgbevf_q_vector *q_vector;
2005
2006         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2007
2008         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2009                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
2010                 if (!q_vector)
2011                         goto err_out;
2012                 q_vector->adapter = adapter;
2013                 q_vector->v_idx = q_idx;
2014                 netif_napi_add(adapter->netdev, &q_vector->napi,
2015                                ixgbevf_poll, 64);
2016 #ifdef CONFIG_NET_RX_BUSY_POLL
2017                 napi_hash_add(&q_vector->napi);
2018 #endif
2019                 adapter->q_vector[q_idx] = q_vector;
2020         }
2021
2022         return 0;
2023
2024 err_out:
2025         while (q_idx) {
2026                 q_idx--;
2027                 q_vector = adapter->q_vector[q_idx];
2028 #ifdef CONFIG_NET_RX_BUSY_POLL
2029                 napi_hash_del(&q_vector->napi);
2030 #endif
2031                 netif_napi_del(&q_vector->napi);
2032                 kfree(q_vector);
2033                 adapter->q_vector[q_idx] = NULL;
2034         }
2035         return -ENOMEM;
2036 }
2037
2038 /**
2039  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
2040  * @adapter: board private structure to initialize
2041  *
2042  * This function frees the memory allocated to the q_vectors.  In addition if
2043  * NAPI is enabled it will delete any references to the NAPI struct prior
2044  * to freeing the q_vector.
2045  **/
2046 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
2047 {
2048         int q_idx, num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
2049
2050         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
2051                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
2052
2053                 adapter->q_vector[q_idx] = NULL;
2054 #ifdef CONFIG_NET_RX_BUSY_POLL
2055                 napi_hash_del(&q_vector->napi);
2056 #endif
2057                 netif_napi_del(&q_vector->napi);
2058                 kfree(q_vector);
2059         }
2060 }
2061
2062 /**
2063  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
2064  * @adapter: board private structure
2065  *
2066  **/
2067 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
2068 {
2069         pci_disable_msix(adapter->pdev);
2070         kfree(adapter->msix_entries);
2071         adapter->msix_entries = NULL;
2072 }
2073
2074 /**
2075  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
2076  * @adapter: board private structure to initialize
2077  *
2078  **/
2079 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
2080 {
2081         int err;
2082
2083         /* Number of supported queues */
2084         ixgbevf_set_num_queues(adapter);
2085
2086         err = ixgbevf_set_interrupt_capability(adapter);
2087         if (err) {
2088                 hw_dbg(&adapter->hw,
2089                        "Unable to setup interrupt capabilities\n");
2090                 goto err_set_interrupt;
2091         }
2092
2093         err = ixgbevf_alloc_q_vectors(adapter);
2094         if (err) {
2095                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
2096                        "vectors\n");
2097                 goto err_alloc_q_vectors;
2098         }
2099
2100         err = ixgbevf_alloc_queues(adapter);
2101         if (err) {
2102                 pr_err("Unable to allocate memory for queues\n");
2103                 goto err_alloc_queues;
2104         }
2105
2106         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2107                "Tx Queue count = %u\n",
2108                (adapter->num_rx_queues > 1) ? "Enabled" :
2109                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2110
2111         set_bit(__IXGBEVF_DOWN, &adapter->state);
2112
2113         return 0;
2114 err_alloc_queues:
2115         ixgbevf_free_q_vectors(adapter);
2116 err_alloc_q_vectors:
2117         ixgbevf_reset_interrupt_capability(adapter);
2118 err_set_interrupt:
2119         return err;
2120 }
2121
2122 /**
2123  * ixgbevf_clear_interrupt_scheme - Clear the current interrupt scheme settings
2124  * @adapter: board private structure to clear interrupt scheme on
2125  *
2126  * We go through and clear interrupt specific resources and reset the structure
2127  * to pre-load conditions
2128  **/
2129 static void ixgbevf_clear_interrupt_scheme(struct ixgbevf_adapter *adapter)
2130 {
2131         adapter->num_tx_queues = 0;
2132         adapter->num_rx_queues = 0;
2133
2134         ixgbevf_free_q_vectors(adapter);
2135         ixgbevf_reset_interrupt_capability(adapter);
2136 }
2137
2138 /**
2139  * ixgbevf_sw_init - Initialize general software structures
2140  * (struct ixgbevf_adapter)
2141  * @adapter: board private structure to initialize
2142  *
2143  * ixgbevf_sw_init initializes the Adapter private data structure.
2144  * Fields are initialized based on PCI device information and
2145  * OS network device settings (MTU size).
2146  **/
2147 static int ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2148 {
2149         struct ixgbe_hw *hw = &adapter->hw;
2150         struct pci_dev *pdev = adapter->pdev;
2151         struct net_device *netdev = adapter->netdev;
2152         int err;
2153
2154         /* PCI config space info */
2155
2156         hw->vendor_id = pdev->vendor;
2157         hw->device_id = pdev->device;
2158         hw->revision_id = pdev->revision;
2159         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2160         hw->subsystem_device_id = pdev->subsystem_device;
2161
2162         hw->mbx.ops.init_params(hw);
2163
2164         /* assume legacy case in which PF would only give VF 2 queues */
2165         hw->mac.max_tx_queues = 2;
2166         hw->mac.max_rx_queues = 2;
2167
2168         /* lock to protect mailbox accesses */
2169         spin_lock_init(&adapter->mbx_lock);
2170
2171         err = hw->mac.ops.reset_hw(hw);
2172         if (err) {
2173                 dev_info(&pdev->dev,
2174                          "PF still in reset state.  Is the PF interface up?\n");
2175         } else {
2176                 err = hw->mac.ops.init_hw(hw);
2177                 if (err) {
2178                         pr_err("init_shared_code failed: %d\n", err);
2179                         goto out;
2180                 }
2181                 ixgbevf_negotiate_api(adapter);
2182                 err = hw->mac.ops.get_mac_addr(hw, hw->mac.addr);
2183                 if (err)
2184                         dev_info(&pdev->dev, "Error reading MAC address\n");
2185                 else if (is_zero_ether_addr(adapter->hw.mac.addr))
2186                         dev_info(&pdev->dev,
2187                                  "MAC address not assigned by administrator.\n");
2188                 memcpy(netdev->dev_addr, hw->mac.addr, netdev->addr_len);
2189         }
2190
2191         if (!is_valid_ether_addr(netdev->dev_addr)) {
2192                 dev_info(&pdev->dev, "Assigning random MAC address\n");
2193                 eth_hw_addr_random(netdev);
2194                 memcpy(hw->mac.addr, netdev->dev_addr, netdev->addr_len);
2195         }
2196
2197         /* Enable dynamic interrupt throttling rates */
2198         adapter->rx_itr_setting = 1;
2199         adapter->tx_itr_setting = 1;
2200
2201         /* set default ring sizes */
2202         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2203         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2204
2205         set_bit(__IXGBEVF_DOWN, &adapter->state);
2206         return 0;
2207
2208 out:
2209         return err;
2210 }
2211
2212 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2213         {                                                       \
2214                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2215                 if (current_counter < last_counter)             \
2216                         counter += 0x100000000LL;               \
2217                 last_counter = current_counter;                 \
2218                 counter &= 0xFFFFFFFF00000000LL;                \
2219                 counter |= current_counter;                     \
2220         }
2221
2222 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2223         {                                                                \
2224                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2225                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2226                 u64 current_counter = (current_counter_msb << 32) |      \
2227                         current_counter_lsb;                             \
2228                 if (current_counter < last_counter)                      \
2229                         counter += 0x1000000000LL;                       \
2230                 last_counter = current_counter;                          \
2231                 counter &= 0xFFFFFFF000000000LL;                         \
2232                 counter |= current_counter;                              \
2233         }
2234 /**
2235  * ixgbevf_update_stats - Update the board statistics counters.
2236  * @adapter: board private structure
2237  **/
2238 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2239 {
2240         struct ixgbe_hw *hw = &adapter->hw;
2241         int i;
2242
2243         if (!adapter->link_up)
2244                 return;
2245
2246         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2247                                 adapter->stats.vfgprc);
2248         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2249                                 adapter->stats.vfgptc);
2250         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2251                                 adapter->stats.last_vfgorc,
2252                                 adapter->stats.vfgorc);
2253         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2254                                 adapter->stats.last_vfgotc,
2255                                 adapter->stats.vfgotc);
2256         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2257                                 adapter->stats.vfmprc);
2258
2259         for (i = 0;  i  < adapter->num_rx_queues;  i++) {
2260                 adapter->hw_csum_rx_error +=
2261                         adapter->rx_ring[i].hw_csum_rx_error;
2262                 adapter->hw_csum_rx_good +=
2263                         adapter->rx_ring[i].hw_csum_rx_good;
2264                 adapter->rx_ring[i].hw_csum_rx_error = 0;
2265                 adapter->rx_ring[i].hw_csum_rx_good = 0;
2266         }
2267 }
2268
2269 /**
2270  * ixgbevf_watchdog - Timer Call-back
2271  * @data: pointer to adapter cast into an unsigned long
2272  **/
2273 static void ixgbevf_watchdog(unsigned long data)
2274 {
2275         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2276         struct ixgbe_hw *hw = &adapter->hw;
2277         u32 eics = 0;
2278         int i;
2279
2280         /*
2281          * Do the watchdog outside of interrupt context due to the lovely
2282          * delays that some of the newer hardware requires
2283          */
2284
2285         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2286                 goto watchdog_short_circuit;
2287
2288         /* get one bit for every active tx/rx interrupt vector */
2289         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2290                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2291                 if (qv->rx.ring || qv->tx.ring)
2292                         eics |= 1 << i;
2293         }
2294
2295         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, eics);
2296
2297 watchdog_short_circuit:
2298         schedule_work(&adapter->watchdog_task);
2299 }
2300
2301 /**
2302  * ixgbevf_tx_timeout - Respond to a Tx Hang
2303  * @netdev: network interface device structure
2304  **/
2305 static void ixgbevf_tx_timeout(struct net_device *netdev)
2306 {
2307         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2308
2309         /* Do the reset outside of interrupt context */
2310         schedule_work(&adapter->reset_task);
2311 }
2312
2313 static void ixgbevf_reset_task(struct work_struct *work)
2314 {
2315         struct ixgbevf_adapter *adapter;
2316         adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2317
2318         /* If we're already down or resetting, just bail */
2319         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2320             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2321                 return;
2322
2323         adapter->tx_timeout_count++;
2324
2325         ixgbevf_reinit_locked(adapter);
2326 }
2327
2328 /**
2329  * ixgbevf_watchdog_task - worker thread to bring link up
2330  * @work: pointer to work_struct containing our data
2331  **/
2332 static void ixgbevf_watchdog_task(struct work_struct *work)
2333 {
2334         struct ixgbevf_adapter *adapter = container_of(work,
2335                                                        struct ixgbevf_adapter,
2336                                                        watchdog_task);
2337         struct net_device *netdev = adapter->netdev;
2338         struct ixgbe_hw *hw = &adapter->hw;
2339         u32 link_speed = adapter->link_speed;
2340         bool link_up = adapter->link_up;
2341         s32 need_reset;
2342
2343         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2344
2345         /*
2346          * Always check the link on the watchdog because we have
2347          * no LSC interrupt
2348          */
2349         spin_lock_bh(&adapter->mbx_lock);
2350
2351         need_reset = hw->mac.ops.check_link(hw, &link_speed, &link_up, false);
2352
2353         spin_unlock_bh(&adapter->mbx_lock);
2354
2355         if (need_reset) {
2356                 adapter->link_up = link_up;
2357                 adapter->link_speed = link_speed;
2358                 netif_carrier_off(netdev);
2359                 netif_tx_stop_all_queues(netdev);
2360                 schedule_work(&adapter->reset_task);
2361                 goto pf_has_reset;
2362         }
2363         adapter->link_up = link_up;
2364         adapter->link_speed = link_speed;
2365
2366         if (link_up) {
2367                 if (!netif_carrier_ok(netdev)) {
2368                         char *link_speed_string;
2369                         switch (link_speed) {
2370                         case IXGBE_LINK_SPEED_10GB_FULL:
2371                                 link_speed_string = "10 Gbps";
2372                                 break;
2373                         case IXGBE_LINK_SPEED_1GB_FULL:
2374                                 link_speed_string = "1 Gbps";
2375                                 break;
2376                         case IXGBE_LINK_SPEED_100_FULL:
2377                                 link_speed_string = "100 Mbps";
2378                                 break;
2379                         default:
2380                                 link_speed_string = "unknown speed";
2381                                 break;
2382                         }
2383                         dev_info(&adapter->pdev->dev,
2384                                 "NIC Link is Up, %s\n", link_speed_string);
2385                         netif_carrier_on(netdev);
2386                         netif_tx_wake_all_queues(netdev);
2387                 }
2388         } else {
2389                 adapter->link_up = false;
2390                 adapter->link_speed = 0;
2391                 if (netif_carrier_ok(netdev)) {
2392                         dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
2393                         netif_carrier_off(netdev);
2394                         netif_tx_stop_all_queues(netdev);
2395                 }
2396         }
2397
2398         ixgbevf_update_stats(adapter);
2399
2400 pf_has_reset:
2401         /* Reset the timer */
2402         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2403                 mod_timer(&adapter->watchdog_timer,
2404                           round_jiffies(jiffies + (2 * HZ)));
2405
2406         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2407 }
2408
2409 /**
2410  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2411  * @adapter: board private structure
2412  * @tx_ring: Tx descriptor ring for a specific queue
2413  *
2414  * Free all transmit software resources
2415  **/
2416 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2417                                struct ixgbevf_ring *tx_ring)
2418 {
2419         struct pci_dev *pdev = adapter->pdev;
2420
2421         ixgbevf_clean_tx_ring(adapter, tx_ring);
2422
2423         vfree(tx_ring->tx_buffer_info);
2424         tx_ring->tx_buffer_info = NULL;
2425
2426         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2427                           tx_ring->dma);
2428
2429         tx_ring->desc = NULL;
2430 }
2431
2432 /**
2433  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2434  * @adapter: board private structure
2435  *
2436  * Free all transmit software resources
2437  **/
2438 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2439 {
2440         int i;
2441
2442         for (i = 0; i < adapter->num_tx_queues; i++)
2443                 if (adapter->tx_ring[i].desc)
2444                         ixgbevf_free_tx_resources(adapter,
2445                                                   &adapter->tx_ring[i]);
2446
2447 }
2448
2449 /**
2450  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2451  * @adapter: board private structure
2452  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2453  *
2454  * Return 0 on success, negative on failure
2455  **/
2456 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2457                                struct ixgbevf_ring *tx_ring)
2458 {
2459         struct pci_dev *pdev = adapter->pdev;
2460         int size;
2461
2462         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2463         tx_ring->tx_buffer_info = vzalloc(size);
2464         if (!tx_ring->tx_buffer_info)
2465                 goto err;
2466
2467         /* round up to nearest 4K */
2468         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2469         tx_ring->size = ALIGN(tx_ring->size, 4096);
2470
2471         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2472                                            &tx_ring->dma, GFP_KERNEL);
2473         if (!tx_ring->desc)
2474                 goto err;
2475
2476         tx_ring->next_to_use = 0;
2477         tx_ring->next_to_clean = 0;
2478         return 0;
2479
2480 err:
2481         vfree(tx_ring->tx_buffer_info);
2482         tx_ring->tx_buffer_info = NULL;
2483         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2484                "descriptor ring\n");
2485         return -ENOMEM;
2486 }
2487
2488 /**
2489  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2490  * @adapter: board private structure
2491  *
2492  * If this function returns with an error, then it's possible one or
2493  * more of the rings is populated (while the rest are not).  It is the
2494  * callers duty to clean those orphaned rings.
2495  *
2496  * Return 0 on success, negative on failure
2497  **/
2498 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2499 {
2500         int i, err = 0;
2501
2502         for (i = 0; i < adapter->num_tx_queues; i++) {
2503                 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2504                 if (!err)
2505                         continue;
2506                 hw_dbg(&adapter->hw,
2507                        "Allocation for Tx Queue %u failed\n", i);
2508                 break;
2509         }
2510
2511         return err;
2512 }
2513
2514 /**
2515  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2516  * @adapter: board private structure
2517  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2518  *
2519  * Returns 0 on success, negative on failure
2520  **/
2521 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2522                                struct ixgbevf_ring *rx_ring)
2523 {
2524         struct pci_dev *pdev = adapter->pdev;
2525         int size;
2526
2527         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2528         rx_ring->rx_buffer_info = vzalloc(size);
2529         if (!rx_ring->rx_buffer_info)
2530                 goto alloc_failed;
2531
2532         /* Round up to nearest 4K */
2533         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2534         rx_ring->size = ALIGN(rx_ring->size, 4096);
2535
2536         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2537                                            &rx_ring->dma, GFP_KERNEL);
2538
2539         if (!rx_ring->desc) {
2540                 vfree(rx_ring->rx_buffer_info);
2541                 rx_ring->rx_buffer_info = NULL;
2542                 goto alloc_failed;
2543         }
2544
2545         rx_ring->next_to_clean = 0;
2546         rx_ring->next_to_use = 0;
2547
2548         return 0;
2549 alloc_failed:
2550         return -ENOMEM;
2551 }
2552
2553 /**
2554  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2555  * @adapter: board private structure
2556  *
2557  * If this function returns with an error, then it's possible one or
2558  * more of the rings is populated (while the rest are not).  It is the
2559  * callers duty to clean those orphaned rings.
2560  *
2561  * Return 0 on success, negative on failure
2562  **/
2563 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2564 {
2565         int i, err = 0;
2566
2567         for (i = 0; i < adapter->num_rx_queues; i++) {
2568                 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2569                 if (!err)
2570                         continue;
2571                 hw_dbg(&adapter->hw,
2572                        "Allocation for Rx Queue %u failed\n", i);
2573                 break;
2574         }
2575         return err;
2576 }
2577
2578 /**
2579  * ixgbevf_free_rx_resources - Free Rx Resources
2580  * @adapter: board private structure
2581  * @rx_ring: ring to clean the resources from
2582  *
2583  * Free all receive software resources
2584  **/
2585 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2586                                struct ixgbevf_ring *rx_ring)
2587 {
2588         struct pci_dev *pdev = adapter->pdev;
2589
2590         ixgbevf_clean_rx_ring(adapter, rx_ring);
2591
2592         vfree(rx_ring->rx_buffer_info);
2593         rx_ring->rx_buffer_info = NULL;
2594
2595         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2596                           rx_ring->dma);
2597
2598         rx_ring->desc = NULL;
2599 }
2600
2601 /**
2602  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2603  * @adapter: board private structure
2604  *
2605  * Free all receive software resources
2606  **/
2607 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2608 {
2609         int i;
2610
2611         for (i = 0; i < adapter->num_rx_queues; i++)
2612                 if (adapter->rx_ring[i].desc)
2613                         ixgbevf_free_rx_resources(adapter,
2614                                                   &adapter->rx_ring[i]);
2615 }
2616
2617 static int ixgbevf_setup_queues(struct ixgbevf_adapter *adapter)
2618 {
2619         struct ixgbe_hw *hw = &adapter->hw;
2620         struct ixgbevf_ring *rx_ring;
2621         unsigned int def_q = 0;
2622         unsigned int num_tcs = 0;
2623         unsigned int num_rx_queues = 1;
2624         int err, i;
2625
2626         spin_lock_bh(&adapter->mbx_lock);
2627
2628         /* fetch queue configuration from the PF */
2629         err = ixgbevf_get_queues(hw, &num_tcs, &def_q);
2630
2631         spin_unlock_bh(&adapter->mbx_lock);
2632
2633         if (err)
2634                 return err;
2635
2636         if (num_tcs > 1) {
2637                 /* update default Tx ring register index */
2638                 adapter->tx_ring[0].reg_idx = def_q;
2639
2640                 /* we need as many queues as traffic classes */
2641                 num_rx_queues = num_tcs;
2642         }
2643
2644         /* nothing to do if we have the correct number of queues */
2645         if (adapter->num_rx_queues == num_rx_queues)
2646                 return 0;
2647
2648         /* allocate new rings */
2649         rx_ring = kcalloc(num_rx_queues,
2650                           sizeof(struct ixgbevf_ring), GFP_KERNEL);
2651         if (!rx_ring)
2652                 return -ENOMEM;
2653
2654         /* setup ring fields */
2655         for (i = 0; i < num_rx_queues; i++) {
2656                 rx_ring[i].count = adapter->rx_ring_count;
2657                 rx_ring[i].queue_index = i;
2658                 rx_ring[i].reg_idx = i;
2659                 rx_ring[i].dev = &adapter->pdev->dev;
2660                 rx_ring[i].netdev = adapter->netdev;
2661         }
2662
2663         /* free the existing ring and queues */
2664         adapter->num_rx_queues = 0;
2665         kfree(adapter->rx_ring);
2666
2667         /* move new rings into position on the adapter struct */
2668         adapter->rx_ring = rx_ring;
2669         adapter->num_rx_queues = num_rx_queues;
2670
2671         return 0;
2672 }
2673
2674 /**
2675  * ixgbevf_open - Called when a network interface is made active
2676  * @netdev: network interface device structure
2677  *
2678  * Returns 0 on success, negative value on failure
2679  *
2680  * The open entry point is called when a network interface is made
2681  * active by the system (IFF_UP).  At this point all resources needed
2682  * for transmit and receive operations are allocated, the interrupt
2683  * handler is registered with the OS, the watchdog timer is started,
2684  * and the stack is notified that the interface is ready.
2685  **/
2686 static int ixgbevf_open(struct net_device *netdev)
2687 {
2688         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2689         struct ixgbe_hw *hw = &adapter->hw;
2690         int err;
2691
2692         /* A previous failure to open the device because of a lack of
2693          * available MSIX vector resources may have reset the number
2694          * of msix vectors variable to zero.  The only way to recover
2695          * is to unload/reload the driver and hope that the system has
2696          * been able to recover some MSIX vector resources.
2697          */
2698         if (!adapter->num_msix_vectors)
2699                 return -ENOMEM;
2700
2701         /* disallow open during test */
2702         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2703                 return -EBUSY;
2704
2705         if (hw->adapter_stopped) {
2706                 ixgbevf_reset(adapter);
2707                 /* if adapter is still stopped then PF isn't up and
2708                  * the vf can't start. */
2709                 if (hw->adapter_stopped) {
2710                         err = IXGBE_ERR_MBX;
2711                         pr_err("Unable to start - perhaps the PF Driver isn't "
2712                                "up yet\n");
2713                         goto err_setup_reset;
2714                 }
2715         }
2716
2717         /* setup queue reg_idx and Rx queue count */
2718         err = ixgbevf_setup_queues(adapter);
2719         if (err)
2720                 goto err_setup_queues;
2721
2722         /* allocate transmit descriptors */
2723         err = ixgbevf_setup_all_tx_resources(adapter);
2724         if (err)
2725                 goto err_setup_tx;
2726
2727         /* allocate receive descriptors */
2728         err = ixgbevf_setup_all_rx_resources(adapter);
2729         if (err)
2730                 goto err_setup_rx;
2731
2732         ixgbevf_configure(adapter);
2733
2734         /*
2735          * Map the Tx/Rx rings to the vectors we were allotted.
2736          * if request_irq will be called in this function map_rings
2737          * must be called *before* up_complete
2738          */
2739         ixgbevf_map_rings_to_vectors(adapter);
2740
2741         ixgbevf_up_complete(adapter);
2742
2743         /* clear any pending interrupts, may auto mask */
2744         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2745         err = ixgbevf_request_irq(adapter);
2746         if (err)
2747                 goto err_req_irq;
2748
2749         ixgbevf_irq_enable(adapter);
2750
2751         return 0;
2752
2753 err_req_irq:
2754         ixgbevf_down(adapter);
2755 err_setup_rx:
2756         ixgbevf_free_all_rx_resources(adapter);
2757 err_setup_tx:
2758         ixgbevf_free_all_tx_resources(adapter);
2759 err_setup_queues:
2760         ixgbevf_reset(adapter);
2761
2762 err_setup_reset:
2763
2764         return err;
2765 }
2766
2767 /**
2768  * ixgbevf_close - Disables a network interface
2769  * @netdev: network interface device structure
2770  *
2771  * Returns 0, this is not allowed to fail
2772  *
2773  * The close entry point is called when an interface is de-activated
2774  * by the OS.  The hardware is still under the drivers control, but
2775  * needs to be disabled.  A global MAC reset is issued to stop the
2776  * hardware, and all transmit and receive resources are freed.
2777  **/
2778 static int ixgbevf_close(struct net_device *netdev)
2779 {
2780         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2781
2782         ixgbevf_down(adapter);
2783         ixgbevf_free_irq(adapter);
2784
2785         ixgbevf_free_all_tx_resources(adapter);
2786         ixgbevf_free_all_rx_resources(adapter);
2787
2788         return 0;
2789 }
2790
2791 static void ixgbevf_tx_ctxtdesc(struct ixgbevf_ring *tx_ring,
2792                                 u32 vlan_macip_lens, u32 type_tucmd,
2793                                 u32 mss_l4len_idx)
2794 {
2795         struct ixgbe_adv_tx_context_desc *context_desc;
2796         u16 i = tx_ring->next_to_use;
2797
2798         context_desc = IXGBEVF_TX_CTXTDESC(tx_ring, i);
2799
2800         i++;
2801         tx_ring->next_to_use = (i < tx_ring->count) ? i : 0;
2802
2803         /* set bits to identify this as an advanced context descriptor */
2804         type_tucmd |= IXGBE_TXD_CMD_DEXT | IXGBE_ADVTXD_DTYP_CTXT;
2805
2806         context_desc->vlan_macip_lens   = cpu_to_le32(vlan_macip_lens);
2807         context_desc->seqnum_seed       = 0;
2808         context_desc->type_tucmd_mlhl   = cpu_to_le32(type_tucmd);
2809         context_desc->mss_l4len_idx     = cpu_to_le32(mss_l4len_idx);
2810 }
2811
2812 static int ixgbevf_tso(struct ixgbevf_ring *tx_ring,
2813                        struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2814 {
2815         u32 vlan_macip_lens, type_tucmd;
2816         u32 mss_l4len_idx, l4len;
2817
2818         if (!skb_is_gso(skb))
2819                 return 0;
2820
2821         if (skb_header_cloned(skb)) {
2822                 int err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2823                 if (err)
2824                         return err;
2825         }
2826
2827         /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2828         type_tucmd = IXGBE_ADVTXD_TUCMD_L4T_TCP;
2829
2830         if (skb->protocol == htons(ETH_P_IP)) {
2831                 struct iphdr *iph = ip_hdr(skb);
2832                 iph->tot_len = 0;
2833                 iph->check = 0;
2834                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2835                                                          iph->daddr, 0,
2836                                                          IPPROTO_TCP,
2837                                                          0);
2838                 type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2839         } else if (skb_is_gso_v6(skb)) {
2840                 ipv6_hdr(skb)->payload_len = 0;
2841                 tcp_hdr(skb)->check =
2842                     ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2843                                      &ipv6_hdr(skb)->daddr,
2844                                      0, IPPROTO_TCP, 0);
2845         }
2846
2847         /* compute header lengths */
2848         l4len = tcp_hdrlen(skb);
2849         *hdr_len += l4len;
2850         *hdr_len = skb_transport_offset(skb) + l4len;
2851
2852         /* mss_l4len_id: use 1 as index for TSO */
2853         mss_l4len_idx = l4len << IXGBE_ADVTXD_L4LEN_SHIFT;
2854         mss_l4len_idx |= skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT;
2855         mss_l4len_idx |= 1 << IXGBE_ADVTXD_IDX_SHIFT;
2856
2857         /* vlan_macip_lens: HEADLEN, MACLEN, VLAN tag */
2858         vlan_macip_lens = skb_network_header_len(skb);
2859         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2860         vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2861
2862         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2863                             type_tucmd, mss_l4len_idx);
2864
2865         return 1;
2866 }
2867
2868 static bool ixgbevf_tx_csum(struct ixgbevf_ring *tx_ring,
2869                             struct sk_buff *skb, u32 tx_flags)
2870 {
2871         u32 vlan_macip_lens = 0;
2872         u32 mss_l4len_idx = 0;
2873         u32 type_tucmd = 0;
2874
2875         if (skb->ip_summed == CHECKSUM_PARTIAL) {
2876                 u8 l4_hdr = 0;
2877                 switch (skb->protocol) {
2878                 case __constant_htons(ETH_P_IP):
2879                         vlan_macip_lens |= skb_network_header_len(skb);
2880                         type_tucmd |= IXGBE_ADVTXD_TUCMD_IPV4;
2881                         l4_hdr = ip_hdr(skb)->protocol;
2882                         break;
2883                 case __constant_htons(ETH_P_IPV6):
2884                         vlan_macip_lens |= skb_network_header_len(skb);
2885                         l4_hdr = ipv6_hdr(skb)->nexthdr;
2886                         break;
2887                 default:
2888                         if (unlikely(net_ratelimit())) {
2889                                 dev_warn(tx_ring->dev,
2890                                  "partial checksum but proto=%x!\n",
2891                                  skb->protocol);
2892                         }
2893                         break;
2894                 }
2895
2896                 switch (l4_hdr) {
2897                 case IPPROTO_TCP:
2898                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2899                         mss_l4len_idx = tcp_hdrlen(skb) <<
2900                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2901                         break;
2902                 case IPPROTO_SCTP:
2903                         type_tucmd |= IXGBE_ADVTXD_TUCMD_L4T_SCTP;
2904                         mss_l4len_idx = sizeof(struct sctphdr) <<
2905                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2906                         break;
2907                 case IPPROTO_UDP:
2908                         mss_l4len_idx = sizeof(struct udphdr) <<
2909                                         IXGBE_ADVTXD_L4LEN_SHIFT;
2910                         break;
2911                 default:
2912                         if (unlikely(net_ratelimit())) {
2913                                 dev_warn(tx_ring->dev,
2914                                  "partial checksum but l4 proto=%x!\n",
2915                                  l4_hdr);
2916                         }
2917                         break;
2918                 }
2919         }
2920
2921         /* vlan_macip_lens: MACLEN, VLAN tag */
2922         vlan_macip_lens |= skb_network_offset(skb) << IXGBE_ADVTXD_MACLEN_SHIFT;
2923         vlan_macip_lens |= tx_flags & IXGBE_TX_FLAGS_VLAN_MASK;
2924
2925         ixgbevf_tx_ctxtdesc(tx_ring, vlan_macip_lens,
2926                             type_tucmd, mss_l4len_idx);
2927
2928         return (skb->ip_summed == CHECKSUM_PARTIAL);
2929 }
2930
2931 static int ixgbevf_tx_map(struct ixgbevf_ring *tx_ring,
2932                           struct sk_buff *skb, u32 tx_flags)
2933 {
2934         struct ixgbevf_tx_buffer *tx_buffer_info;
2935         unsigned int len;
2936         unsigned int total = skb->len;
2937         unsigned int offset = 0, size;
2938         int count = 0;
2939         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2940         unsigned int f;
2941         int i;
2942
2943         i = tx_ring->next_to_use;
2944
2945         len = min(skb_headlen(skb), total);
2946         while (len) {
2947                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2948                 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2949
2950                 tx_buffer_info->length = size;
2951                 tx_buffer_info->mapped_as_page = false;
2952                 tx_buffer_info->dma = dma_map_single(tx_ring->dev,
2953                                                      skb->data + offset,
2954                                                      size, DMA_TO_DEVICE);
2955                 if (dma_mapping_error(tx_ring->dev, tx_buffer_info->dma))
2956                         goto dma_error;
2957
2958                 len -= size;
2959                 total -= size;
2960                 offset += size;
2961                 count++;
2962                 i++;
2963                 if (i == tx_ring->count)
2964                         i = 0;
2965         }
2966
2967         for (f = 0; f < nr_frags; f++) {
2968                 const struct skb_frag_struct *frag;
2969
2970                 frag = &skb_shinfo(skb)->frags[f];
2971                 len = min((unsigned int)skb_frag_size(frag), total);
2972                 offset = 0;
2973
2974                 while (len) {
2975                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
2976                         size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2977
2978                         tx_buffer_info->length = size;
2979                         tx_buffer_info->dma =
2980                                 skb_frag_dma_map(tx_ring->dev, frag,
2981                                                  offset, size, DMA_TO_DEVICE);
2982                         if (dma_mapping_error(tx_ring->dev,
2983                                               tx_buffer_info->dma))
2984                                 goto dma_error;
2985                         tx_buffer_info->mapped_as_page = true;
2986
2987                         len -= size;
2988                         total -= size;
2989                         offset += size;
2990                         count++;
2991                         i++;
2992                         if (i == tx_ring->count)
2993                                 i = 0;
2994                 }
2995                 if (total == 0)
2996                         break;
2997         }
2998
2999         if (i == 0)
3000                 i = tx_ring->count - 1;
3001         else
3002                 i = i - 1;
3003         tx_ring->tx_buffer_info[i].skb = skb;
3004
3005         return count;
3006
3007 dma_error:
3008         dev_err(tx_ring->dev, "TX DMA map failed\n");
3009
3010         /* clear timestamp and dma mappings for failed tx_buffer_info map */
3011         tx_buffer_info->dma = 0;
3012         count--;
3013
3014         /* clear timestamp and dma mappings for remaining portion of packet */
3015         while (count >= 0) {
3016                 count--;
3017                 i--;
3018                 if (i < 0)
3019                         i += tx_ring->count;
3020                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3021                 ixgbevf_unmap_and_free_tx_resource(tx_ring, tx_buffer_info);
3022         }
3023
3024         return count;
3025 }
3026
3027 static void ixgbevf_tx_queue(struct ixgbevf_ring *tx_ring, int tx_flags,
3028                              int count, unsigned int first, u32 paylen,
3029                              u8 hdr_len)
3030 {
3031         union ixgbe_adv_tx_desc *tx_desc = NULL;
3032         struct ixgbevf_tx_buffer *tx_buffer_info;
3033         u32 olinfo_status = 0, cmd_type_len = 0;
3034         unsigned int i;
3035
3036         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
3037
3038         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
3039
3040         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
3041
3042         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
3043                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
3044
3045         if (tx_flags & IXGBE_TX_FLAGS_CSUM)
3046                 olinfo_status |= IXGBE_ADVTXD_POPTS_TXSM;
3047
3048         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
3049                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
3050
3051                 /* use index 1 context for tso */
3052                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
3053                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
3054                         olinfo_status |= IXGBE_ADVTXD_POPTS_IXSM;
3055         }
3056
3057         /*
3058          * Check Context must be set if Tx switch is enabled, which it
3059          * always is for case where virtual functions are running
3060          */
3061         olinfo_status |= IXGBE_ADVTXD_CC;
3062
3063         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
3064
3065         i = tx_ring->next_to_use;
3066         while (count--) {
3067                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
3068                 tx_desc = IXGBEVF_TX_DESC(tx_ring, i);
3069                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
3070                 tx_desc->read.cmd_type_len =
3071                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
3072                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
3073                 i++;
3074                 if (i == tx_ring->count)
3075                         i = 0;
3076         }
3077
3078         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
3079
3080         tx_ring->tx_buffer_info[first].time_stamp = jiffies;
3081
3082         /* Force memory writes to complete before letting h/w
3083          * know there are new descriptors to fetch.  (Only
3084          * applicable for weak-ordered memory model archs,
3085          * such as IA-64).
3086          */
3087         wmb();
3088
3089         tx_ring->tx_buffer_info[first].next_to_watch = tx_desc;
3090         tx_ring->next_to_use = i;
3091 }
3092
3093 static int __ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3094 {
3095         struct ixgbevf_adapter *adapter = netdev_priv(tx_ring->netdev);
3096
3097         netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
3098         /* Herbert's original patch had:
3099          *  smp_mb__after_netif_stop_queue();
3100          * but since that doesn't exist yet, just open code it. */
3101         smp_mb();
3102
3103         /* We need to check again in a case another CPU has just
3104          * made room available. */
3105         if (likely(ixgbevf_desc_unused(tx_ring) < size))
3106                 return -EBUSY;
3107
3108         /* A reprieve! - use start_queue because it doesn't call schedule */
3109         netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
3110         ++adapter->restart_queue;
3111         return 0;
3112 }
3113
3114 static int ixgbevf_maybe_stop_tx(struct ixgbevf_ring *tx_ring, int size)
3115 {
3116         if (likely(ixgbevf_desc_unused(tx_ring) >= size))
3117                 return 0;
3118         return __ixgbevf_maybe_stop_tx(tx_ring, size);
3119 }
3120
3121 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
3122 {
3123         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3124         struct ixgbevf_ring *tx_ring;
3125         unsigned int first;
3126         unsigned int tx_flags = 0;
3127         u8 hdr_len = 0;
3128         int r_idx = 0, tso;
3129         u16 count = TXD_USE_COUNT(skb_headlen(skb));
3130 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3131         unsigned short f;
3132 #endif
3133         u8 *dst_mac = skb_header_pointer(skb, 0, 0, NULL);
3134         if (!dst_mac || is_link_local_ether_addr(dst_mac)) {
3135                 dev_kfree_skb(skb);
3136                 return NETDEV_TX_OK;
3137         }
3138
3139         tx_ring = &adapter->tx_ring[r_idx];
3140
3141         /*
3142          * need: 1 descriptor per page * PAGE_SIZE/IXGBE_MAX_DATA_PER_TXD,
3143          *       + 1 desc for skb_headlen/IXGBE_MAX_DATA_PER_TXD,
3144          *       + 2 desc gap to keep tail from touching head,
3145          *       + 1 desc for context descriptor,
3146          * otherwise try next time
3147          */
3148 #if PAGE_SIZE > IXGBE_MAX_DATA_PER_TXD
3149         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
3150                 count += TXD_USE_COUNT(skb_shinfo(skb)->frags[f].size);
3151 #else
3152         count += skb_shinfo(skb)->nr_frags;
3153 #endif
3154         if (ixgbevf_maybe_stop_tx(tx_ring, count + 3)) {
3155                 adapter->tx_busy++;
3156                 return NETDEV_TX_BUSY;
3157         }
3158
3159         if (vlan_tx_tag_present(skb)) {
3160                 tx_flags |= vlan_tx_tag_get(skb);
3161                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
3162                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
3163         }
3164
3165         first = tx_ring->next_to_use;
3166
3167         if (skb->protocol == htons(ETH_P_IP))
3168                 tx_flags |= IXGBE_TX_FLAGS_IPV4;
3169         tso = ixgbevf_tso(tx_ring, skb, tx_flags, &hdr_len);
3170         if (tso < 0) {
3171                 dev_kfree_skb_any(skb);
3172                 return NETDEV_TX_OK;
3173         }
3174
3175         if (tso)
3176                 tx_flags |= IXGBE_TX_FLAGS_TSO | IXGBE_TX_FLAGS_CSUM;
3177         else if (ixgbevf_tx_csum(tx_ring, skb, tx_flags))
3178                 tx_flags |= IXGBE_TX_FLAGS_CSUM;
3179
3180         ixgbevf_tx_queue(tx_ring, tx_flags,
3181                          ixgbevf_tx_map(tx_ring, skb, tx_flags),
3182                          first, skb->len, hdr_len);
3183
3184         writel(tx_ring->next_to_use, adapter->hw.hw_addr + tx_ring->tail);
3185
3186         ixgbevf_maybe_stop_tx(tx_ring, DESC_NEEDED);
3187
3188         return NETDEV_TX_OK;
3189 }
3190
3191 /**
3192  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
3193  * @netdev: network interface device structure
3194  * @p: pointer to an address structure
3195  *
3196  * Returns 0 on success, negative on failure
3197  **/
3198 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
3199 {
3200         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3201         struct ixgbe_hw *hw = &adapter->hw;
3202         struct sockaddr *addr = p;
3203
3204         if (!is_valid_ether_addr(addr->sa_data))
3205                 return -EADDRNOTAVAIL;
3206
3207         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3208         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3209
3210         spin_lock_bh(&adapter->mbx_lock);
3211
3212         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3213
3214         spin_unlock_bh(&adapter->mbx_lock);
3215
3216         return 0;
3217 }
3218
3219 /**
3220  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3221  * @netdev: network interface device structure
3222  * @new_mtu: new value for maximum frame size
3223  *
3224  * Returns 0 on success, negative on failure
3225  **/
3226 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3227 {
3228         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3229         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3230         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3231
3232         switch (adapter->hw.api_version) {
3233         case ixgbe_mbox_api_11:
3234                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3235                 break;
3236         default:
3237                 if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3238                         max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3239                 break;
3240         }
3241
3242         /* MTU < 68 is an error and causes problems on some kernels */
3243         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3244                 return -EINVAL;
3245
3246         hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3247                netdev->mtu, new_mtu);
3248         /* must set new MTU before calling down or up */
3249         netdev->mtu = new_mtu;
3250
3251         if (netif_running(netdev))
3252                 ixgbevf_reinit_locked(adapter);
3253
3254         return 0;
3255 }
3256
3257 static int ixgbevf_suspend(struct pci_dev *pdev, pm_message_t state)
3258 {
3259         struct net_device *netdev = pci_get_drvdata(pdev);
3260         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3261 #ifdef CONFIG_PM
3262         int retval = 0;
3263 #endif
3264
3265         netif_device_detach(netdev);
3266
3267         if (netif_running(netdev)) {
3268                 rtnl_lock();
3269                 ixgbevf_down(adapter);
3270                 ixgbevf_free_irq(adapter);
3271                 ixgbevf_free_all_tx_resources(adapter);
3272                 ixgbevf_free_all_rx_resources(adapter);
3273                 rtnl_unlock();
3274         }
3275
3276         ixgbevf_clear_interrupt_scheme(adapter);
3277
3278 #ifdef CONFIG_PM
3279         retval = pci_save_state(pdev);
3280         if (retval)
3281                 return retval;
3282
3283 #endif
3284         pci_disable_device(pdev);
3285
3286         return 0;
3287 }
3288
3289 #ifdef CONFIG_PM
3290 static int ixgbevf_resume(struct pci_dev *pdev)
3291 {
3292         struct ixgbevf_adapter *adapter = pci_get_drvdata(pdev);
3293         struct net_device *netdev = adapter->netdev;
3294         u32 err;
3295
3296         pci_set_power_state(pdev, PCI_D0);
3297         pci_restore_state(pdev);
3298         /*
3299          * pci_restore_state clears dev->state_saved so call
3300          * pci_save_state to restore it.
3301          */
3302         pci_save_state(pdev);
3303
3304         err = pci_enable_device_mem(pdev);
3305         if (err) {
3306                 dev_err(&pdev->dev, "Cannot enable PCI device from suspend\n");
3307                 return err;
3308         }
3309         pci_set_master(pdev);
3310
3311         ixgbevf_reset(adapter);
3312
3313         rtnl_lock();
3314         err = ixgbevf_init_interrupt_scheme(adapter);
3315         rtnl_unlock();
3316         if (err) {
3317                 dev_err(&pdev->dev, "Cannot initialize interrupts\n");
3318                 return err;
3319         }
3320
3321         if (netif_running(netdev)) {
3322                 err = ixgbevf_open(netdev);
3323                 if (err)
3324                         return err;
3325         }
3326
3327         netif_device_attach(netdev);
3328
3329         return err;
3330 }
3331
3332 #endif /* CONFIG_PM */
3333 static void ixgbevf_shutdown(struct pci_dev *pdev)
3334 {
3335         ixgbevf_suspend(pdev, PMSG_SUSPEND);
3336 }
3337
3338 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3339                                                 struct rtnl_link_stats64 *stats)
3340 {
3341         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3342         unsigned int start;
3343         u64 bytes, packets;
3344         const struct ixgbevf_ring *ring;
3345         int i;
3346
3347         ixgbevf_update_stats(adapter);
3348
3349         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3350
3351         for (i = 0; i < adapter->num_rx_queues; i++) {
3352                 ring = &adapter->rx_ring[i];
3353                 do {
3354                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3355                         bytes = ring->total_bytes;
3356                         packets = ring->total_packets;
3357                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3358                 stats->rx_bytes += bytes;
3359                 stats->rx_packets += packets;
3360         }
3361
3362         for (i = 0; i < adapter->num_tx_queues; i++) {
3363                 ring = &adapter->tx_ring[i];
3364                 do {
3365                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3366                         bytes = ring->total_bytes;
3367                         packets = ring->total_packets;
3368                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3369                 stats->tx_bytes += bytes;
3370                 stats->tx_packets += packets;
3371         }
3372
3373         return stats;
3374 }
3375
3376 static const struct net_device_ops ixgbevf_netdev_ops = {
3377         .ndo_open               = ixgbevf_open,
3378         .ndo_stop               = ixgbevf_close,
3379         .ndo_start_xmit         = ixgbevf_xmit_frame,
3380         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3381         .ndo_get_stats64        = ixgbevf_get_stats,
3382         .ndo_validate_addr      = eth_validate_addr,
3383         .ndo_set_mac_address    = ixgbevf_set_mac,
3384         .ndo_change_mtu         = ixgbevf_change_mtu,
3385         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3386         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3387         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3388 #ifdef CONFIG_NET_RX_BUSY_POLL
3389         .ndo_busy_poll          = ixgbevf_busy_poll_recv,
3390 #endif
3391 };
3392
3393 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3394 {
3395         dev->netdev_ops = &ixgbevf_netdev_ops;
3396         ixgbevf_set_ethtool_ops(dev);
3397         dev->watchdog_timeo = 5 * HZ;
3398 }
3399
3400 /**
3401  * ixgbevf_probe - Device Initialization Routine
3402  * @pdev: PCI device information struct
3403  * @ent: entry in ixgbevf_pci_tbl
3404  *
3405  * Returns 0 on success, negative on failure
3406  *
3407  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3408  * The OS initialization, configuring of the adapter private structure,
3409  * and a hardware reset occur.
3410  **/
3411 static int ixgbevf_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
3412 {
3413         struct net_device *netdev;
3414         struct ixgbevf_adapter *adapter = NULL;
3415         struct ixgbe_hw *hw = NULL;
3416         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3417         static int cards_found;
3418         int err, pci_using_dac;
3419
3420         err = pci_enable_device(pdev);
3421         if (err)
3422                 return err;
3423
3424         if (!dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64))) {
3425                 pci_using_dac = 1;
3426         } else {
3427                 err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
3428                 if (err) {
3429                         dev_err(&pdev->dev, "No usable DMA "
3430                                 "configuration, aborting\n");
3431                         goto err_dma;
3432                 }
3433                 pci_using_dac = 0;
3434         }
3435
3436         err = pci_request_regions(pdev, ixgbevf_driver_name);
3437         if (err) {
3438                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3439                 goto err_pci_reg;
3440         }
3441
3442         pci_set_master(pdev);
3443
3444         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3445                                    MAX_TX_QUEUES);
3446         if (!netdev) {
3447                 err = -ENOMEM;
3448                 goto err_alloc_etherdev;
3449         }
3450
3451         SET_NETDEV_DEV(netdev, &pdev->dev);
3452
3453         pci_set_drvdata(pdev, netdev);
3454         adapter = netdev_priv(netdev);
3455
3456         adapter->netdev = netdev;
3457         adapter->pdev = pdev;
3458         hw = &adapter->hw;
3459         hw->back = adapter;
3460         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3461
3462         /*
3463          * call save state here in standalone driver because it relies on
3464          * adapter struct to exist, and needs to call netdev_priv
3465          */
3466         pci_save_state(pdev);
3467
3468         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3469                               pci_resource_len(pdev, 0));
3470         if (!hw->hw_addr) {
3471                 err = -EIO;
3472                 goto err_ioremap;
3473         }
3474
3475         ixgbevf_assign_netdev_ops(netdev);
3476
3477         adapter->bd_number = cards_found;
3478
3479         /* Setup hw api */
3480         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3481         hw->mac.type  = ii->mac;
3482
3483         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3484                sizeof(struct ixgbe_mbx_operations));
3485
3486         /* setup the private structure */
3487         err = ixgbevf_sw_init(adapter);
3488         if (err)
3489                 goto err_sw_init;
3490
3491         /* The HW MAC address was set and/or determined in sw_init */
3492         if (!is_valid_ether_addr(netdev->dev_addr)) {
3493                 pr_err("invalid MAC address\n");
3494                 err = -EIO;
3495                 goto err_sw_init;
3496         }
3497
3498         netdev->hw_features = NETIF_F_SG |
3499                            NETIF_F_IP_CSUM |
3500                            NETIF_F_IPV6_CSUM |
3501                            NETIF_F_TSO |
3502                            NETIF_F_TSO6 |
3503                            NETIF_F_RXCSUM;
3504
3505         netdev->features = netdev->hw_features |
3506                            NETIF_F_HW_VLAN_CTAG_TX |
3507                            NETIF_F_HW_VLAN_CTAG_RX |
3508                            NETIF_F_HW_VLAN_CTAG_FILTER;
3509
3510         netdev->vlan_features |= NETIF_F_TSO;
3511         netdev->vlan_features |= NETIF_F_TSO6;
3512         netdev->vlan_features |= NETIF_F_IP_CSUM;
3513         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3514         netdev->vlan_features |= NETIF_F_SG;
3515
3516         if (pci_using_dac)
3517                 netdev->features |= NETIF_F_HIGHDMA;
3518
3519         netdev->priv_flags |= IFF_UNICAST_FLT;
3520
3521         init_timer(&adapter->watchdog_timer);
3522         adapter->watchdog_timer.function = ixgbevf_watchdog;
3523         adapter->watchdog_timer.data = (unsigned long)adapter;
3524
3525         INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3526         INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3527
3528         err = ixgbevf_init_interrupt_scheme(adapter);
3529         if (err)
3530                 goto err_sw_init;
3531
3532         strcpy(netdev->name, "eth%d");
3533
3534         err = register_netdev(netdev);
3535         if (err)
3536                 goto err_register;
3537
3538         netif_carrier_off(netdev);
3539
3540         ixgbevf_init_last_counter_stats(adapter);
3541
3542         /* print the MAC address */
3543         hw_dbg(hw, "%pM\n", netdev->dev_addr);
3544
3545         hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3546
3547         hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3548         cards_found++;
3549         return 0;
3550
3551 err_register:
3552         ixgbevf_clear_interrupt_scheme(adapter);
3553 err_sw_init:
3554         ixgbevf_reset_interrupt_capability(adapter);
3555         iounmap(hw->hw_addr);
3556 err_ioremap:
3557         free_netdev(netdev);
3558 err_alloc_etherdev:
3559         pci_release_regions(pdev);
3560 err_pci_reg:
3561 err_dma:
3562         pci_disable_device(pdev);
3563         return err;
3564 }
3565
3566 /**
3567  * ixgbevf_remove - Device Removal Routine
3568  * @pdev: PCI device information struct
3569  *
3570  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3571  * that it should release a PCI device.  The could be caused by a
3572  * Hot-Plug event, or because the driver is going to be removed from
3573  * memory.
3574  **/
3575 static void ixgbevf_remove(struct pci_dev *pdev)
3576 {
3577         struct net_device *netdev = pci_get_drvdata(pdev);
3578         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3579
3580         set_bit(__IXGBEVF_DOWN, &adapter->state);
3581
3582         del_timer_sync(&adapter->watchdog_timer);
3583
3584         cancel_work_sync(&adapter->reset_task);
3585         cancel_work_sync(&adapter->watchdog_task);
3586
3587         if (netdev->reg_state == NETREG_REGISTERED)
3588                 unregister_netdev(netdev);
3589
3590         ixgbevf_clear_interrupt_scheme(adapter);
3591         ixgbevf_reset_interrupt_capability(adapter);
3592
3593         iounmap(adapter->hw.hw_addr);
3594         pci_release_regions(pdev);
3595
3596         hw_dbg(&adapter->hw, "Remove complete\n");
3597
3598         kfree(adapter->tx_ring);
3599         kfree(adapter->rx_ring);
3600
3601         free_netdev(netdev);
3602
3603         pci_disable_device(pdev);
3604 }
3605
3606 /**
3607  * ixgbevf_io_error_detected - called when PCI error is detected
3608  * @pdev: Pointer to PCI device
3609  * @state: The current pci connection state
3610  *
3611  * This function is called after a PCI bus error affecting
3612  * this device has been detected.
3613  */
3614 static pci_ers_result_t ixgbevf_io_error_detected(struct pci_dev *pdev,
3615                                                   pci_channel_state_t state)
3616 {
3617         struct net_device *netdev = pci_get_drvdata(pdev);
3618         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3619
3620         netif_device_detach(netdev);
3621
3622         if (state == pci_channel_io_perm_failure)
3623                 return PCI_ERS_RESULT_DISCONNECT;
3624
3625         if (netif_running(netdev))
3626                 ixgbevf_down(adapter);
3627
3628         pci_disable_device(pdev);
3629
3630         /* Request a slot slot reset. */
3631         return PCI_ERS_RESULT_NEED_RESET;
3632 }
3633
3634 /**
3635  * ixgbevf_io_slot_reset - called after the pci bus has been reset.
3636  * @pdev: Pointer to PCI device
3637  *
3638  * Restart the card from scratch, as if from a cold-boot. Implementation
3639  * resembles the first-half of the ixgbevf_resume routine.
3640  */
3641 static pci_ers_result_t ixgbevf_io_slot_reset(struct pci_dev *pdev)
3642 {
3643         struct net_device *netdev = pci_get_drvdata(pdev);
3644         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3645
3646         if (pci_enable_device_mem(pdev)) {
3647                 dev_err(&pdev->dev,
3648                         "Cannot re-enable PCI device after reset.\n");
3649                 return PCI_ERS_RESULT_DISCONNECT;
3650         }
3651
3652         pci_set_master(pdev);
3653
3654         ixgbevf_reset(adapter);
3655
3656         return PCI_ERS_RESULT_RECOVERED;
3657 }
3658
3659 /**
3660  * ixgbevf_io_resume - called when traffic can start flowing again.
3661  * @pdev: Pointer to PCI device
3662  *
3663  * This callback is called when the error recovery driver tells us that
3664  * its OK to resume normal operation. Implementation resembles the
3665  * second-half of the ixgbevf_resume routine.
3666  */
3667 static void ixgbevf_io_resume(struct pci_dev *pdev)
3668 {
3669         struct net_device *netdev = pci_get_drvdata(pdev);
3670         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3671
3672         if (netif_running(netdev))
3673                 ixgbevf_up(adapter);
3674
3675         netif_device_attach(netdev);
3676 }
3677
3678 /* PCI Error Recovery (ERS) */
3679 static const struct pci_error_handlers ixgbevf_err_handler = {
3680         .error_detected = ixgbevf_io_error_detected,
3681         .slot_reset = ixgbevf_io_slot_reset,
3682         .resume = ixgbevf_io_resume,
3683 };
3684
3685 static struct pci_driver ixgbevf_driver = {
3686         .name     = ixgbevf_driver_name,
3687         .id_table = ixgbevf_pci_tbl,
3688         .probe    = ixgbevf_probe,
3689         .remove   = ixgbevf_remove,
3690 #ifdef CONFIG_PM
3691         /* Power Management Hooks */
3692         .suspend  = ixgbevf_suspend,
3693         .resume   = ixgbevf_resume,
3694 #endif
3695         .shutdown = ixgbevf_shutdown,
3696         .err_handler = &ixgbevf_err_handler
3697 };
3698
3699 /**
3700  * ixgbevf_init_module - Driver Registration Routine
3701  *
3702  * ixgbevf_init_module is the first routine called when the driver is
3703  * loaded. All it does is register with the PCI subsystem.
3704  **/
3705 static int __init ixgbevf_init_module(void)
3706 {
3707         int ret;
3708         pr_info("%s - version %s\n", ixgbevf_driver_string,
3709                 ixgbevf_driver_version);
3710
3711         pr_info("%s\n", ixgbevf_copyright);
3712
3713         ret = pci_register_driver(&ixgbevf_driver);
3714         return ret;
3715 }
3716
3717 module_init(ixgbevf_init_module);
3718
3719 /**
3720  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3721  *
3722  * ixgbevf_exit_module is called just before the driver is removed
3723  * from memory.
3724  **/
3725 static void __exit ixgbevf_exit_module(void)
3726 {
3727         pci_unregister_driver(&ixgbevf_driver);
3728 }
3729
3730 #ifdef DEBUG
3731 /**
3732  * ixgbevf_get_hw_dev_name - return device name string
3733  * used by hardware layer to print debugging information
3734  **/
3735 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3736 {
3737         struct ixgbevf_adapter *adapter = hw->back;
3738         return adapter->netdev->name;
3739 }
3740
3741 #endif
3742 module_exit(ixgbevf_exit_module);
3743
3744 /* ixgbevf_main.c */