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/ipv6.h>
  46 #include <linux/slab.h>
  47 #include <net/checksum.h>
  48 #include <net/ip6_checksum.h>
  49 #include <linux/ethtool.h>
  50 #include <linux/if.h>
  51 #include <linux/if_vlan.h>
  52 #include <linux/prefetch.h>
  53
  54 #include "ixgbevf.h"
  55
  56 const char ixgbevf_driver_name[] = "ixgbevf";
  57 static const char ixgbevf_driver_string[] =
  58         "Intel(R) 10 Gigabit PCI Express Virtual Function Network Driver";
  59
  60 #define DRV_VERSION "2.6.0-k"
  61 const char ixgbevf_driver_version[] = DRV_VERSION;
  62 static char ixgbevf_copyright[] =
  63         "Copyright (c) 2009 - 2012 Intel Corporation.";
  64
  65 static const struct ixgbevf_info *ixgbevf_info_tbl[] = {
  66         [board_82599_vf] = &ixgbevf_82599_vf_info,
  67         [board_X540_vf]  = &ixgbevf_X540_vf_info,
  68 };
  69
  70 /* ixgbevf_pci_tbl - PCI Device ID Table
  71  *
  72  * Wildcard entries (PCI_ANY_ID) should come last
  73  * Last entry must be all 0s
  74  *
  75  * { Vendor ID, Device ID, SubVendor ID, SubDevice ID,
  76  *   Class, Class Mask, private data (not used) }
  77  */
  78 static struct pci_device_id ixgbevf_pci_tbl[] = {
  79         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_82599_VF),
  80         board_82599_vf},
  81         {PCI_VDEVICE(INTEL, IXGBE_DEV_ID_X540_VF),
  82         board_X540_vf},
  83
  84         /* required last entry */
  85         {0, }
  86 };
  87 MODULE_DEVICE_TABLE(pci, ixgbevf_pci_tbl);
  88
  89 MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
  90 MODULE_DESCRIPTION("Intel(R) 82599 Virtual Function Driver");
  91 MODULE_LICENSE("GPL");
  92 MODULE_VERSION(DRV_VERSION);
  93
  94 #define DEFAULT_MSG_ENABLE (NETIF_MSG_DRV|NETIF_MSG_PROBE|NETIF_MSG_LINK)
  95 static int debug = -1;
  96 module_param(debug, int, 0);
  97 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
  98
  99 /* forward decls */
 100 static void ixgbevf_set_itr_msix(struct ixgbevf_q_vector *q_vector);
 101 static void ixgbevf_write_eitr(struct ixgbevf_adapter *adapter, int v_idx,
 102                                u32 itr_reg);
 103
 104 static inline void ixgbevf_release_rx_desc(struct ixgbe_hw *hw,
 105                                            struct ixgbevf_ring *rx_ring,
 106                                            u32 val)
 107 {
 108         /*
 109          * Force memory writes to complete before letting h/w
 110          * know there are new descriptors to fetch.  (Only
 111          * applicable for weak-ordered memory model archs,
 112          * such as IA-64).
 113          */
 114         wmb();
 115         IXGBE_WRITE_REG(hw, IXGBE_VFRDT(rx_ring->reg_idx), val);
 116 }
 117
 118 /**
 119  * ixgbevf_set_ivar - set IVAR registers - maps interrupt causes to vectors
 120  * @adapter: pointer to adapter struct
 121  * @direction: 0 for Rx, 1 for Tx, -1 for other causes
 122  * @queue: queue to map the corresponding interrupt to
 123  * @msix_vector: the vector to map to the corresponding queue
 124  *
 125  */
 126 static void ixgbevf_set_ivar(struct ixgbevf_adapter *adapter, s8 direction,
 127                              u8 queue, u8 msix_vector)
 128 {
 129         u32 ivar, index;
 130         struct ixgbe_hw *hw = &adapter->hw;
 131         if (direction == -1) {
 132                 /* other causes */
 133                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 134                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR_MISC);
 135                 ivar &= ~0xFF;
 136                 ivar |= msix_vector;
 137                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR_MISC, ivar);
 138         } else {
 139                 /* tx or rx causes */
 140                 msix_vector |= IXGBE_IVAR_ALLOC_VAL;
 141                 index = ((16 * (queue & 1)) + (8 * direction));
 142                 ivar = IXGBE_READ_REG(hw, IXGBE_VTIVAR(queue >> 1));
 143                 ivar &= ~(0xFF << index);
 144                 ivar |= (msix_vector << index);
 145                 IXGBE_WRITE_REG(hw, IXGBE_VTIVAR(queue >> 1), ivar);
 146         }
 147 }
 148
 149 static void ixgbevf_unmap_and_free_tx_resource(struct ixgbevf_adapter *adapter,
 150                                                struct ixgbevf_tx_buffer
 151                                                *tx_buffer_info)
 152 {
 153         if (tx_buffer_info->dma) {
 154                 if (tx_buffer_info->mapped_as_page)
 155                         dma_unmap_page(&adapter->pdev->dev,
 156                                        tx_buffer_info->dma,
 157                                        tx_buffer_info->length,
 158                                        DMA_TO_DEVICE);
 159                 else
 160                         dma_unmap_single(&adapter->pdev->dev,
 161                                          tx_buffer_info->dma,
 162                                          tx_buffer_info->length,
 163                                          DMA_TO_DEVICE);
 164                 tx_buffer_info->dma = 0;
 165         }
 166         if (tx_buffer_info->skb) {
 167                 dev_kfree_skb_any(tx_buffer_info->skb);
 168                 tx_buffer_info->skb = NULL;
 169         }
 170         tx_buffer_info->time_stamp = 0;
 171         /* tx_buffer_info must be completely set up in the transmit path */
 172 }
 173
 174 #define IXGBE_MAX_TXD_PWR       14
 175 #define IXGBE_MAX_DATA_PER_TXD  (1 << IXGBE_MAX_TXD_PWR)
 176
 177 /* Tx Descriptors needed, worst case */
 178 #define TXD_USE_COUNT(S) (((S) >> IXGBE_MAX_TXD_PWR) + \
 179                          (((S) & (IXGBE_MAX_DATA_PER_TXD - 1)) ? 1 : 0))
 180 #define DESC_NEEDED (TXD_USE_COUNT(IXGBE_MAX_DATA_PER_TXD) /* skb->data */ + \
 181         MAX_SKB_FRAGS * TXD_USE_COUNT(PAGE_SIZE) + 1)      /* for context */
 182
 183 static void ixgbevf_tx_timeout(struct net_device *netdev);
 184
 185 /**
 186  * ixgbevf_clean_tx_irq - Reclaim resources after transmit completes
 187  * @adapter: board private structure
 188  * @tx_ring: tx ring to clean
 189  **/
 190 static bool ixgbevf_clean_tx_irq(struct ixgbevf_adapter *adapter,
 191                                  struct ixgbevf_ring *tx_ring)
 192 {
 193         struct net_device *netdev = adapter->netdev;
 194         struct ixgbe_hw *hw = &adapter->hw;
 195         union ixgbe_adv_tx_desc *tx_desc, *eop_desc;
 196         struct ixgbevf_tx_buffer *tx_buffer_info;
 197         unsigned int i, eop, count = 0;
 198         unsigned int total_bytes = 0, total_packets = 0;
 199
 200         i = tx_ring->next_to_clean;
 201         eop = tx_ring->tx_buffer_info[i].next_to_watch;
 202         eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
 203
 204         while ((eop_desc->wb.status & cpu_to_le32(IXGBE_TXD_STAT_DD)) &&
 205                (count < tx_ring->work_limit)) {
 206                 bool cleaned = false;
 207                 rmb(); /* read buffer_info after eop_desc */
 208                 /* eop could change between read and DD-check */
 209                 if (unlikely(eop != tx_ring->tx_buffer_info[i].next_to_watch))
 210                         goto cont_loop;
 211                 for ( ; !cleaned; count++) {
 212                         struct sk_buff *skb;
 213                         tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
 214                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
 215                         cleaned = (i == eop);
 216                         skb = tx_buffer_info->skb;
 217
 218                         if (cleaned && skb) {
 219                                 unsigned int segs, bytecount;
 220
 221                                 /* gso_segs is currently only valid for tcp */
 222                                 segs = skb_shinfo(skb)->gso_segs ?: 1;
 223                                 /* multiply data chunks by size of headers */
 224                                 bytecount = ((segs - 1) * skb_headlen(skb)) +
 225                                             skb->len;
 226                                 total_packets += segs;
 227                                 total_bytes += bytecount;
 228                         }
 229
 230                         ixgbevf_unmap_and_free_tx_resource(adapter,
 231                                                            tx_buffer_info);
 232
 233                         tx_desc->wb.status = 0;
 234
 235                         i++;
 236                         if (i == tx_ring->count)
 237                                 i = 0;
 238                 }
 239
 240 cont_loop:
 241                 eop = tx_ring->tx_buffer_info[i].next_to_watch;
 242                 eop_desc = IXGBE_TX_DESC_ADV(*tx_ring, eop);
 243         }
 244
 245         tx_ring->next_to_clean = i;
 246
 247 #define TX_WAKE_THRESHOLD (DESC_NEEDED * 2)
 248         if (unlikely(count && netif_carrier_ok(netdev) &&
 249                      (IXGBE_DESC_UNUSED(tx_ring) >= TX_WAKE_THRESHOLD))) {
 250                 /* Make sure that anybody stopping the queue after this
 251                  * sees the new next_to_clean.
 252                  */
 253                 smp_mb();
 254                 if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
 255                     !test_bit(__IXGBEVF_DOWN, &adapter->state)) {
 256                         netif_wake_subqueue(netdev, tx_ring->queue_index);
 257                         ++adapter->restart_queue;
 258                 }
 259         }
 260
 261         /* re-arm the interrupt */
 262         if ((count >= tx_ring->work_limit) &&
 263             (!test_bit(__IXGBEVF_DOWN, &adapter->state))) {
 264                 IXGBE_WRITE_REG(hw, IXGBE_VTEICS, tx_ring->v_idx);
 265         }
 266
 267         u64_stats_update_begin(&tx_ring->syncp);
 268         tx_ring->total_bytes += total_bytes;
 269         tx_ring->total_packets += total_packets;
 270         u64_stats_update_end(&tx_ring->syncp);
 271
 272         return count < tx_ring->work_limit;
 273 }
 274
 275 /**
 276  * ixgbevf_receive_skb - Send a completed packet up the stack
 277  * @q_vector: structure containing interrupt and ring information
 278  * @skb: packet to send up
 279  * @status: hardware indication of status of receive
 280  * @rx_ring: rx descriptor ring (for a specific queue) to setup
 281  * @rx_desc: rx descriptor
 282  **/
 283 static void ixgbevf_receive_skb(struct ixgbevf_q_vector *q_vector,
 284                                 struct sk_buff *skb, u8 status,
 285                                 struct ixgbevf_ring *ring,
 286                                 union ixgbe_adv_rx_desc *rx_desc)
 287 {
 288         struct ixgbevf_adapter *adapter = q_vector->adapter;
 289         bool is_vlan = (status & IXGBE_RXD_STAT_VP);
 290         u16 tag = le16_to_cpu(rx_desc->wb.upper.vlan);
 291
 292         if (is_vlan && test_bit(tag, adapter->active_vlans))
 293                 __vlan_hwaccel_put_tag(skb, tag);
 294
 295         napi_gro_receive(&q_vector->napi, skb);
 296 }
 297
 298 /**
 299  * ixgbevf_rx_checksum - indicate in skb if hw indicated a good cksum
 300  * @adapter: address of board private structure
 301  * @status_err: hardware indication of status of receive
 302  * @skb: skb currently being received and modified
 303  **/
 304 static inline void ixgbevf_rx_checksum(struct ixgbevf_adapter *adapter,
 305                                        u32 status_err, struct sk_buff *skb)
 306 {
 307         skb_checksum_none_assert(skb);
 308
 309         /* Rx csum disabled */
 310         if (!(adapter->netdev->features & NETIF_F_RXCSUM))
 311                 return;
 312
 313         /* if IP and error */
 314         if ((status_err & IXGBE_RXD_STAT_IPCS) &&
 315             (status_err & IXGBE_RXDADV_ERR_IPE)) {
 316                 adapter->hw_csum_rx_error++;
 317                 return;
 318         }
 319
 320         if (!(status_err & IXGBE_RXD_STAT_L4CS))
 321                 return;
 322
 323         if (status_err & IXGBE_RXDADV_ERR_TCPE) {
 324                 adapter->hw_csum_rx_error++;
 325                 return;
 326         }
 327
 328         /* It must be a TCP or UDP packet with a valid checksum */
 329         skb->ip_summed = CHECKSUM_UNNECESSARY;
 330         adapter->hw_csum_rx_good++;
 331 }
 332
 333 /**
 334  * ixgbevf_alloc_rx_buffers - Replace used receive buffers; packet split
 335  * @adapter: address of board private structure
 336  **/
 337 static void ixgbevf_alloc_rx_buffers(struct ixgbevf_adapter *adapter,
 338                                      struct ixgbevf_ring *rx_ring,
 339                                      int cleaned_count)
 340 {
 341         struct pci_dev *pdev = adapter->pdev;
 342         union ixgbe_adv_rx_desc *rx_desc;
 343         struct ixgbevf_rx_buffer *bi;
 344         struct sk_buff *skb;
 345         unsigned int i;
 346         unsigned int bufsz = rx_ring->rx_buf_len + NET_IP_ALIGN;
 347
 348         i = rx_ring->next_to_use;
 349         bi = &rx_ring->rx_buffer_info[i];
 350
 351         while (cleaned_count--) {
 352                 rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 353                 skb = bi->skb;
 354                 if (!skb) {
 355                         skb = netdev_alloc_skb(adapter->netdev,
 356                                                                bufsz);
 357
 358                         if (!skb) {
 359                                 adapter->alloc_rx_buff_failed++;
 360                                 goto no_buffers;
 361                         }
 362
 363                         /*
 364                          * Make buffer alignment 2 beyond a 16 byte boundary
 365                          * this will result in a 16 byte aligned IP header after
 366                          * the 14 byte MAC header is removed
 367                          */
 368                         skb_reserve(skb, NET_IP_ALIGN);
 369
 370                         bi->skb = skb;
 371                 }
 372                 if (!bi->dma) {
 373                         bi->dma = dma_map_single(&pdev->dev, skb->data,
 374                                                  rx_ring->rx_buf_len,
 375                                                  DMA_FROM_DEVICE);
 376                 }
 377                 rx_desc->read.pkt_addr = cpu_to_le64(bi->dma);
 378
 379                 i++;
 380                 if (i == rx_ring->count)
 381                         i = 0;
 382                 bi = &rx_ring->rx_buffer_info[i];
 383         }
 384
 385 no_buffers:
 386         if (rx_ring->next_to_use != i) {
 387                 rx_ring->next_to_use = i;
 388                 if (i-- == 0)
 389                         i = (rx_ring->count - 1);
 390
 391                 ixgbevf_release_rx_desc(&adapter->hw, rx_ring, i);
 392         }
 393 }
 394
 395 static inline void ixgbevf_irq_enable_queues(struct ixgbevf_adapter *adapter,
 396                                              u64 qmask)
 397 {
 398         u32 mask;
 399         struct ixgbe_hw *hw = &adapter->hw;
 400
 401         mask = (qmask & 0xFFFFFFFF);
 402         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
 403 }
 404
 405 static bool ixgbevf_clean_rx_irq(struct ixgbevf_q_vector *q_vector,
 406                                  struct ixgbevf_ring *rx_ring,
 407                                  int *work_done, int work_to_do)
 408 {
 409         struct ixgbevf_adapter *adapter = q_vector->adapter;
 410         struct pci_dev *pdev = adapter->pdev;
 411         union ixgbe_adv_rx_desc *rx_desc, *next_rxd;
 412         struct ixgbevf_rx_buffer *rx_buffer_info, *next_buffer;
 413         struct sk_buff *skb;
 414         unsigned int i;
 415         u32 len, staterr;
 416         bool cleaned = false;
 417         int cleaned_count = 0;
 418         unsigned int total_rx_bytes = 0, total_rx_packets = 0;
 419
 420         i = rx_ring->next_to_clean;
 421         rx_desc = IXGBE_RX_DESC_ADV(*rx_ring, i);
 422         staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 423         rx_buffer_info = &rx_ring->rx_buffer_info[i];
 424
 425         while (staterr & IXGBE_RXD_STAT_DD) {
 426                 if (*work_done >= work_to_do)
 427                         break;
 428                 (*work_done)++;
 429
 430                 rmb(); /* read descriptor and rx_buffer_info after status DD */
 431                 len = le16_to_cpu(rx_desc->wb.upper.length);
 432                 cleaned = true;
 433                 skb = rx_buffer_info->skb;
 434                 prefetch(skb->data - NET_IP_ALIGN);
 435                 rx_buffer_info->skb = NULL;
 436
 437                 if (rx_buffer_info->dma) {
 438                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
 439                                          rx_ring->rx_buf_len,
 440                                          DMA_FROM_DEVICE);
 441                         rx_buffer_info->dma = 0;
 442                         skb_put(skb, len);
 443                 }
 444
 445                 i++;
 446                 if (i == rx_ring->count)
 447                         i = 0;
 448
 449                 next_rxd = IXGBE_RX_DESC_ADV(*rx_ring, i);
 450                 prefetch(next_rxd);
 451                 cleaned_count++;
 452
 453                 next_buffer = &rx_ring->rx_buffer_info[i];
 454
 455                 if (!(staterr & IXGBE_RXD_STAT_EOP)) {
 456                         skb->next = next_buffer->skb;
 457                         skb->next->prev = skb;
 458                         adapter->non_eop_descs++;
 459                         goto next_desc;
 460                 }
 461
 462                 /* ERR_MASK will only have valid bits if EOP set */
 463                 if (unlikely(staterr & IXGBE_RXDADV_ERR_FRAME_ERR_MASK)) {
 464                         dev_kfree_skb_irq(skb);
 465                         goto next_desc;
 466                 }
 467
 468                 ixgbevf_rx_checksum(adapter, staterr, skb);
 469
 470                 /* probably a little skewed due to removing CRC */
 471                 total_rx_bytes += skb->len;
 472                 total_rx_packets++;
 473
 474                 /*
 475                  * Work around issue of some types of VM to VM loop back
 476                  * packets not getting split correctly
 477                  */
 478                 if (staterr & IXGBE_RXD_STAT_LB) {
 479                         u32 header_fixup_len = skb_headlen(skb);
 480                         if (header_fixup_len < 14)
 481                                 skb_push(skb, header_fixup_len);
 482                 }
 483                 skb->protocol = eth_type_trans(skb, adapter->netdev);
 484
 485                 ixgbevf_receive_skb(q_vector, skb, staterr, rx_ring, rx_desc);
 486
 487 next_desc:
 488                 rx_desc->wb.upper.status_error = 0;
 489
 490                 /* return some buffers to hardware, one at a time is too slow */
 491                 if (cleaned_count >= IXGBEVF_RX_BUFFER_WRITE) {
 492                         ixgbevf_alloc_rx_buffers(adapter, rx_ring,
 493                                                  cleaned_count);
 494                         cleaned_count = 0;
 495                 }
 496
 497                 /* use prefetched values */
 498                 rx_desc = next_rxd;
 499                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
 500
 501                 staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
 502         }
 503
 504         rx_ring->next_to_clean = i;
 505         cleaned_count = IXGBE_DESC_UNUSED(rx_ring);
 506
 507         if (cleaned_count)
 508                 ixgbevf_alloc_rx_buffers(adapter, rx_ring, cleaned_count);
 509
 510         u64_stats_update_begin(&rx_ring->syncp);
 511         rx_ring->total_packets += total_rx_packets;
 512         rx_ring->total_bytes += total_rx_bytes;
 513         u64_stats_update_end(&rx_ring->syncp);
 514
 515         return cleaned;
 516 }
 517
 518 /**
 519  * ixgbevf_clean_rxonly - msix (aka one shot) rx clean routine
 520  * @napi: napi struct with our devices info in it
 521  * @budget: amount of work driver is allowed to do this pass, in packets
 522  *
 523  * This function is optimized for cleaning one queue only on a single
 524  * q_vector!!!
 525  **/
 526 static int ixgbevf_clean_rxonly(struct napi_struct *napi, int budget)
 527 {
 528         struct ixgbevf_q_vector *q_vector =
 529                 container_of(napi, struct ixgbevf_q_vector, napi);
 530         struct ixgbevf_adapter *adapter = q_vector->adapter;
 531         struct ixgbevf_ring *rx_ring = NULL;
 532         int work_done = 0;
 533         long r_idx;
 534
 535         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 536         rx_ring = &(adapter->rx_ring[r_idx]);
 537
 538         ixgbevf_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
 539
 540         /* If all Rx work done, exit the polling mode */
 541         if (work_done < budget) {
 542                 napi_complete(napi);
 543                 if (adapter->itr_setting & 1)
 544                         ixgbevf_set_itr_msix(q_vector);
 545                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
 546                         ixgbevf_irq_enable_queues(adapter, rx_ring->v_idx);
 547         }
 548
 549         return work_done;
 550 }
 551
 552 /**
 553  * ixgbevf_clean_rxonly_many - msix (aka one shot) rx clean routine
 554  * @napi: napi struct with our devices info in it
 555  * @budget: amount of work driver is allowed to do this pass, in packets
 556  *
 557  * This function will clean more than one rx queue associated with a
 558  * q_vector.
 559  **/
 560 static int ixgbevf_clean_rxonly_many(struct napi_struct *napi, int budget)
 561 {
 562         struct ixgbevf_q_vector *q_vector =
 563                 container_of(napi, struct ixgbevf_q_vector, napi);
 564         struct ixgbevf_adapter *adapter = q_vector->adapter;
 565         struct ixgbevf_ring *rx_ring = NULL;
 566         int work_done = 0, i;
 567         long r_idx;
 568         u64 enable_mask = 0;
 569
 570         /* attempt to distribute budget to each queue fairly, but don't allow
 571          * the budget to go below 1 because we'll exit polling */
 572         budget /= (q_vector->rxr_count ?: 1);
 573         budget = max(budget, 1);
 574         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 575         for (i = 0; i < q_vector->rxr_count; i++) {
 576                 rx_ring = &(adapter->rx_ring[r_idx]);
 577                 ixgbevf_clean_rx_irq(q_vector, rx_ring, &work_done, budget);
 578                 enable_mask |= rx_ring->v_idx;
 579                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
 580                                       r_idx + 1);
 581         }
 582
 583         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 584         rx_ring = &(adapter->rx_ring[r_idx]);
 585
 586         /* If all Rx work done, exit the polling mode */
 587         if (work_done < budget) {
 588                 napi_complete(napi);
 589                 if (adapter->itr_setting & 1)
 590                         ixgbevf_set_itr_msix(q_vector);
 591                 if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
 592                         ixgbevf_irq_enable_queues(adapter, enable_mask);
 593         }
 594
 595         return work_done;
 596 }
 597
 598
 599 /**
 600  * ixgbevf_configure_msix - Configure MSI-X hardware
 601  * @adapter: board private structure
 602  *
 603  * ixgbevf_configure_msix sets up the hardware to properly generate MSI-X
 604  * interrupts.
 605  **/
 606 static void ixgbevf_configure_msix(struct ixgbevf_adapter *adapter)
 607 {
 608         struct ixgbevf_q_vector *q_vector;
 609         struct ixgbe_hw *hw = &adapter->hw;
 610         int i, j, q_vectors, v_idx, r_idx;
 611         u32 mask;
 612
 613         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 614
 615         /*
 616          * Populate the IVAR table and set the ITR values to the
 617          * corresponding register.
 618          */
 619         for (v_idx = 0; v_idx < q_vectors; v_idx++) {
 620                 q_vector = adapter->q_vector[v_idx];
 621                 /* XXX for_each_set_bit(...) */
 622                 r_idx = find_first_bit(q_vector->rxr_idx,
 623                                        adapter->num_rx_queues);
 624
 625                 for (i = 0; i < q_vector->rxr_count; i++) {
 626                         j = adapter->rx_ring[r_idx].reg_idx;
 627                         ixgbevf_set_ivar(adapter, 0, j, v_idx);
 628                         r_idx = find_next_bit(q_vector->rxr_idx,
 629                                               adapter->num_rx_queues,
 630                                               r_idx + 1);
 631                 }
 632                 r_idx = find_first_bit(q_vector->txr_idx,
 633                                        adapter->num_tx_queues);
 634
 635                 for (i = 0; i < q_vector->txr_count; i++) {
 636                         j = adapter->tx_ring[r_idx].reg_idx;
 637                         ixgbevf_set_ivar(adapter, 1, j, v_idx);
 638                         r_idx = find_next_bit(q_vector->txr_idx,
 639                                               adapter->num_tx_queues,
 640                                               r_idx + 1);
 641                 }
 642
 643                 /* if this is a tx only vector halve the interrupt rate */
 644                 if (q_vector->txr_count && !q_vector->rxr_count)
 645                         q_vector->eitr = (adapter->eitr_param >> 1);
 646                 else if (q_vector->rxr_count)
 647                         /* rx only */
 648                         q_vector->eitr = adapter->eitr_param;
 649
 650                 ixgbevf_write_eitr(adapter, v_idx, q_vector->eitr);
 651         }
 652
 653         ixgbevf_set_ivar(adapter, -1, 1, v_idx);
 654
 655         /* set up to autoclear timer, and the vectors */
 656         mask = IXGBE_EIMS_ENABLE_MASK;
 657         mask &= ~IXGBE_EIMS_OTHER;
 658         IXGBE_WRITE_REG(hw, IXGBE_VTEIAC, mask);
 659 }
 660
 661 enum latency_range {
 662         lowest_latency = 0,
 663         low_latency = 1,
 664         bulk_latency = 2,
 665         latency_invalid = 255
 666 };
 667
 668 /**
 669  * ixgbevf_update_itr - update the dynamic ITR value based on statistics
 670  * @adapter: pointer to adapter
 671  * @eitr: eitr setting (ints per sec) to give last timeslice
 672  * @itr_setting: current throttle rate in ints/second
 673  * @packets: the number of packets during this measurement interval
 674  * @bytes: the number of bytes during this measurement interval
 675  *
 676  *      Stores a new ITR value based on packets and byte
 677  *      counts during the last interrupt.  The advantage of per interrupt
 678  *      computation is faster updates and more accurate ITR for the current
 679  *      traffic pattern.  Constants in this function were computed
 680  *      based on theoretical maximum wire speed and thresholds were set based
 681  *      on testing data as well as attempting to minimize response time
 682  *      while increasing bulk throughput.
 683  **/
 684 static u8 ixgbevf_update_itr(struct ixgbevf_adapter *adapter,
 685                              u32 eitr, u8 itr_setting,
 686                              int packets, int bytes)
 687 {
 688         unsigned int retval = itr_setting;
 689         u32 timepassed_us;
 690         u64 bytes_perint;
 691
 692         if (packets == 0)
 693                 goto update_itr_done;
 694
 695
 696         /* simple throttlerate management
 697          *    0-20MB/s lowest (100000 ints/s)
 698          *   20-100MB/s low   (20000 ints/s)
 699          *  100-1249MB/s bulk (8000 ints/s)
 700          */
 701         /* what was last interrupt timeslice? */
 702         timepassed_us = 1000000/eitr;
 703         bytes_perint = bytes / timepassed_us; /* bytes/usec */
 704
 705         switch (itr_setting) {
 706         case lowest_latency:
 707                 if (bytes_perint > 10)
 708                         retval = low_latency;
 709                 break;
 710         case low_latency:
 711                 if (bytes_perint > 20)
 712                         retval = bulk_latency;
 713                 else if (bytes_perint <= 10)
 714                         retval = lowest_latency;
 715                 break;
 716         case bulk_latency:
 717                 if (bytes_perint <= 20)
 718                         retval = low_latency;
 719                 break;
 720         }
 721
 722 update_itr_done:
 723         return retval;
 724 }
 725
 726 /**
 727  * ixgbevf_write_eitr - write VTEITR register in hardware specific way
 728  * @adapter: pointer to adapter struct
 729  * @v_idx: vector index into q_vector array
 730  * @itr_reg: new value to be written in *register* format, not ints/s
 731  *
 732  * This function is made to be called by ethtool and by the driver
 733  * when it needs to update VTEITR registers at runtime.  Hardware
 734  * specific quirks/differences are taken care of here.
 735  */
 736 static void ixgbevf_write_eitr(struct ixgbevf_adapter *adapter, int v_idx,
 737                                u32 itr_reg)
 738 {
 739         struct ixgbe_hw *hw = &adapter->hw;
 740
 741         itr_reg = EITR_INTS_PER_SEC_TO_REG(itr_reg);
 742
 743         /*
 744          * set the WDIS bit to not clear the timer bits and cause an
 745          * immediate assertion of the interrupt
 746          */
 747         itr_reg |= IXGBE_EITR_CNT_WDIS;
 748
 749         IXGBE_WRITE_REG(hw, IXGBE_VTEITR(v_idx), itr_reg);
 750 }
 751
 752 static void ixgbevf_set_itr_msix(struct ixgbevf_q_vector *q_vector)
 753 {
 754         struct ixgbevf_adapter *adapter = q_vector->adapter;
 755         u32 new_itr;
 756         u8 current_itr, ret_itr;
 757         int i, r_idx, v_idx = q_vector->v_idx;
 758         struct ixgbevf_ring *rx_ring, *tx_ring;
 759
 760         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
 761         for (i = 0; i < q_vector->txr_count; i++) {
 762                 tx_ring = &(adapter->tx_ring[r_idx]);
 763                 ret_itr = ixgbevf_update_itr(adapter, q_vector->eitr,
 764                                              q_vector->tx_itr,
 765                                              tx_ring->total_packets,
 766                                              tx_ring->total_bytes);
 767                 /* if the result for this queue would decrease interrupt
 768                  * rate for this vector then use that result */
 769                 q_vector->tx_itr = ((q_vector->tx_itr > ret_itr) ?
 770                                     q_vector->tx_itr - 1 : ret_itr);
 771                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
 772                                       r_idx + 1);
 773         }
 774
 775         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 776         for (i = 0; i < q_vector->rxr_count; i++) {
 777                 rx_ring = &(adapter->rx_ring[r_idx]);
 778                 ret_itr = ixgbevf_update_itr(adapter, q_vector->eitr,
 779                                              q_vector->rx_itr,
 780                                              rx_ring->total_packets,
 781                                              rx_ring->total_bytes);
 782                 /* if the result for this queue would decrease interrupt
 783                  * rate for this vector then use that result */
 784                 q_vector->rx_itr = ((q_vector->rx_itr > ret_itr) ?
 785                                     q_vector->rx_itr - 1 : ret_itr);
 786                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
 787                                       r_idx + 1);
 788         }
 789
 790         current_itr = max(q_vector->rx_itr, q_vector->tx_itr);
 791
 792         switch (current_itr) {
 793         /* counts and packets in update_itr are dependent on these numbers */
 794         case lowest_latency:
 795                 new_itr = 100000;
 796                 break;
 797         case low_latency:
 798                 new_itr = 20000; /* aka hwitr = ~200 */
 799                 break;
 800         case bulk_latency:
 801         default:
 802                 new_itr = 8000;
 803                 break;
 804         }
 805
 806         if (new_itr != q_vector->eitr) {
 807                 u32 itr_reg;
 808
 809                 /* save the algorithm value here, not the smoothed one */
 810                 q_vector->eitr = new_itr;
 811                 /* do an exponential smoothing */
 812                 new_itr = ((q_vector->eitr * 90)/100) + ((new_itr * 10)/100);
 813                 itr_reg = EITR_INTS_PER_SEC_TO_REG(new_itr);
 814                 ixgbevf_write_eitr(adapter, v_idx, itr_reg);
 815         }
 816 }
 817
 818 static irqreturn_t ixgbevf_msix_mbx(int irq, void *data)
 819 {
 820         struct net_device *netdev = data;
 821         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
 822         struct ixgbe_hw *hw = &adapter->hw;
 823         u32 eicr;
 824         u32 msg;
 825         bool got_ack = false;
 826
 827         eicr = IXGBE_READ_REG(hw, IXGBE_VTEICS);
 828         IXGBE_WRITE_REG(hw, IXGBE_VTEICR, eicr);
 829
 830         if (!hw->mbx.ops.check_for_ack(hw))
 831                 got_ack = true;
 832
 833         if (!hw->mbx.ops.check_for_msg(hw)) {
 834                 hw->mbx.ops.read(hw, &msg, 1);
 835
 836                 if ((msg & IXGBE_MBVFICR_VFREQ_MASK) == IXGBE_PF_CONTROL_MSG)
 837                         mod_timer(&adapter->watchdog_timer,
 838                                   round_jiffies(jiffies + 1));
 839
 840                 if (msg & IXGBE_VT_MSGTYPE_NACK)
 841                         pr_warn("Last Request of type %2.2x to PF Nacked\n",
 842                                 msg & 0xFF);
 843                 /*
 844                  * Restore the PFSTS bit in case someone is polling for a
 845                  * return message from the PF
 846                  */
 847                 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFSTS;
 848         }
 849
 850         /*
 851          * checking for the ack clears the PFACK bit.  Place
 852          * it back in the v2p_mailbox cache so that anyone
 853          * polling for an ack will not miss it
 854          */
 855         if (got_ack)
 856                 hw->mbx.v2p_mailbox |= IXGBE_VFMAILBOX_PFACK;
 857
 858         return IRQ_HANDLED;
 859 }
 860
 861 static irqreturn_t ixgbevf_msix_clean_tx(int irq, void *data)
 862 {
 863         struct ixgbevf_q_vector *q_vector = data;
 864         struct ixgbevf_adapter  *adapter = q_vector->adapter;
 865         struct ixgbevf_ring     *tx_ring;
 866         int i, r_idx;
 867
 868         if (!q_vector->txr_count)
 869                 return IRQ_HANDLED;
 870
 871         r_idx = find_first_bit(q_vector->txr_idx, adapter->num_tx_queues);
 872         for (i = 0; i < q_vector->txr_count; i++) {
 873                 tx_ring = &(adapter->tx_ring[r_idx]);
 874                 tx_ring->total_bytes = 0;
 875                 tx_ring->total_packets = 0;
 876                 ixgbevf_clean_tx_irq(adapter, tx_ring);
 877                 r_idx = find_next_bit(q_vector->txr_idx, adapter->num_tx_queues,
 878                                       r_idx + 1);
 879         }
 880
 881         if (adapter->itr_setting & 1)
 882                 ixgbevf_set_itr_msix(q_vector);
 883
 884         return IRQ_HANDLED;
 885 }
 886
 887 /**
 888  * ixgbevf_msix_clean_rx - single unshared vector rx clean (all queues)
 889  * @irq: unused
 890  * @data: pointer to our q_vector struct for this interrupt vector
 891  **/
 892 static irqreturn_t ixgbevf_msix_clean_rx(int irq, void *data)
 893 {
 894         struct ixgbevf_q_vector *q_vector = data;
 895         struct ixgbevf_adapter  *adapter = q_vector->adapter;
 896         struct ixgbe_hw *hw = &adapter->hw;
 897         struct ixgbevf_ring  *rx_ring;
 898         int r_idx;
 899         int i;
 900
 901         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 902         for (i = 0; i < q_vector->rxr_count; i++) {
 903                 rx_ring = &(adapter->rx_ring[r_idx]);
 904                 rx_ring->total_bytes = 0;
 905                 rx_ring->total_packets = 0;
 906                 r_idx = find_next_bit(q_vector->rxr_idx, adapter->num_rx_queues,
 907                                       r_idx + 1);
 908         }
 909
 910         if (!q_vector->rxr_count)
 911                 return IRQ_HANDLED;
 912
 913         r_idx = find_first_bit(q_vector->rxr_idx, adapter->num_rx_queues);
 914         rx_ring = &(adapter->rx_ring[r_idx]);
 915         /* disable interrupts on this vector only */
 916         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, rx_ring->v_idx);
 917         napi_schedule(&q_vector->napi);
 918
 919
 920         return IRQ_HANDLED;
 921 }
 922
 923 static irqreturn_t ixgbevf_msix_clean_many(int irq, void *data)
 924 {
 925         ixgbevf_msix_clean_rx(irq, data);
 926         ixgbevf_msix_clean_tx(irq, data);
 927
 928         return IRQ_HANDLED;
 929 }
 930
 931 static inline void map_vector_to_rxq(struct ixgbevf_adapter *a, int v_idx,
 932                                      int r_idx)
 933 {
 934         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
 935
 936         set_bit(r_idx, q_vector->rxr_idx);
 937         q_vector->rxr_count++;
 938         a->rx_ring[r_idx].v_idx = 1 << v_idx;
 939 }
 940
 941 static inline void map_vector_to_txq(struct ixgbevf_adapter *a, int v_idx,
 942                                      int t_idx)
 943 {
 944         struct ixgbevf_q_vector *q_vector = a->q_vector[v_idx];
 945
 946         set_bit(t_idx, q_vector->txr_idx);
 947         q_vector->txr_count++;
 948         a->tx_ring[t_idx].v_idx = 1 << v_idx;
 949 }
 950
 951 /**
 952  * ixgbevf_map_rings_to_vectors - Maps descriptor rings to vectors
 953  * @adapter: board private structure to initialize
 954  *
 955  * This function maps descriptor rings to the queue-specific vectors
 956  * we were allotted through the MSI-X enabling code.  Ideally, we'd have
 957  * one vector per ring/queue, but on a constrained vector budget, we
 958  * group the rings as "efficiently" as possible.  You would add new
 959  * mapping configurations in here.
 960  **/
 961 static int ixgbevf_map_rings_to_vectors(struct ixgbevf_adapter *adapter)
 962 {
 963         int q_vectors;
 964         int v_start = 0;
 965         int rxr_idx = 0, txr_idx = 0;
 966         int rxr_remaining = adapter->num_rx_queues;
 967         int txr_remaining = adapter->num_tx_queues;
 968         int i, j;
 969         int rqpv, tqpv;
 970         int err = 0;
 971
 972         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
 973
 974         /*
 975          * The ideal configuration...
 976          * We have enough vectors to map one per queue.
 977          */
 978         if (q_vectors == adapter->num_rx_queues + adapter->num_tx_queues) {
 979                 for (; rxr_idx < rxr_remaining; v_start++, rxr_idx++)
 980                         map_vector_to_rxq(adapter, v_start, rxr_idx);
 981
 982                 for (; txr_idx < txr_remaining; v_start++, txr_idx++)
 983                         map_vector_to_txq(adapter, v_start, txr_idx);
 984                 goto out;
 985         }
 986
 987         /*
 988          * If we don't have enough vectors for a 1-to-1
 989          * mapping, we'll have to group them so there are
 990          * multiple queues per vector.
 991          */
 992         /* Re-adjusting *qpv takes care of the remainder. */
 993         for (i = v_start; i < q_vectors; i++) {
 994                 rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - i);
 995                 for (j = 0; j < rqpv; j++) {
 996                         map_vector_to_rxq(adapter, i, rxr_idx);
 997                         rxr_idx++;
 998                         rxr_remaining--;
 999                 }
1000         }
1001         for (i = v_start; i < q_vectors; i++) {
1002                 tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - i);
1003                 for (j = 0; j < tqpv; j++) {
1004                         map_vector_to_txq(adapter, i, txr_idx);
1005                         txr_idx++;
1006                         txr_remaining--;
1007                 }
1008         }
1009
1010 out:
1011         return err;
1012 }
1013
1014 /**
1015  * ixgbevf_request_msix_irqs - Initialize MSI-X interrupts
1016  * @adapter: board private structure
1017  *
1018  * ixgbevf_request_msix_irqs allocates MSI-X vectors and requests
1019  * interrupts from the kernel.
1020  **/
1021 static int ixgbevf_request_msix_irqs(struct ixgbevf_adapter *adapter)
1022 {
1023         struct net_device *netdev = adapter->netdev;
1024         irqreturn_t (*handler)(int, void *);
1025         int i, vector, q_vectors, err;
1026         int ri = 0, ti = 0;
1027
1028         /* Decrement for Other and TCP Timer vectors */
1029         q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1030
1031 #define SET_HANDLER(_v) (((_v)->rxr_count && (_v)->txr_count)          \
1032                                           ? &ixgbevf_msix_clean_many : \
1033                           (_v)->rxr_count ? &ixgbevf_msix_clean_rx   : \
1034                           (_v)->txr_count ? &ixgbevf_msix_clean_tx   : \
1035                           NULL)
1036         for (vector = 0; vector < q_vectors; vector++) {
1037                 handler = SET_HANDLER(adapter->q_vector[vector]);
1038
1039                 if (handler == &ixgbevf_msix_clean_rx) {
1040                         sprintf(adapter->name[vector], "%s-%s-%d",
1041                                 netdev->name, "rx", ri++);
1042                 } else if (handler == &ixgbevf_msix_clean_tx) {
1043                         sprintf(adapter->name[vector], "%s-%s-%d",
1044                                 netdev->name, "tx", ti++);
1045                 } else if (handler == &ixgbevf_msix_clean_many) {
1046                         sprintf(adapter->name[vector], "%s-%s-%d",
1047                                 netdev->name, "TxRx", vector);
1048                 } else {
1049                         /* skip this unused q_vector */
1050                         continue;
1051                 }
1052                 err = request_irq(adapter->msix_entries[vector].vector,
1053                                   handler, 0, adapter->name[vector],
1054                                   adapter->q_vector[vector]);
1055                 if (err) {
1056                         hw_dbg(&adapter->hw,
1057                                "request_irq failed for MSIX interrupt "
1058                                "Error: %d\n", err);
1059                         goto free_queue_irqs;
1060                 }
1061         }
1062
1063         sprintf(adapter->name[vector], "%s:mbx", netdev->name);
1064         err = request_irq(adapter->msix_entries[vector].vector,
1065                           &ixgbevf_msix_mbx, 0, adapter->name[vector], netdev);
1066         if (err) {
1067                 hw_dbg(&adapter->hw,
1068                        "request_irq for msix_mbx failed: %d\n", err);
1069                 goto free_queue_irqs;
1070         }
1071
1072         return 0;
1073
1074 free_queue_irqs:
1075         for (i = vector - 1; i >= 0; i--)
1076                 free_irq(adapter->msix_entries[--vector].vector,
1077                          &(adapter->q_vector[i]));
1078         pci_disable_msix(adapter->pdev);
1079         kfree(adapter->msix_entries);
1080         adapter->msix_entries = NULL;
1081         return err;
1082 }
1083
1084 static inline void ixgbevf_reset_q_vectors(struct ixgbevf_adapter *adapter)
1085 {
1086         int i, q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1087
1088         for (i = 0; i < q_vectors; i++) {
1089                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[i];
1090                 bitmap_zero(q_vector->rxr_idx, MAX_RX_QUEUES);
1091                 bitmap_zero(q_vector->txr_idx, MAX_TX_QUEUES);
1092                 q_vector->rxr_count = 0;
1093                 q_vector->txr_count = 0;
1094                 q_vector->eitr = adapter->eitr_param;
1095         }
1096 }
1097
1098 /**
1099  * ixgbevf_request_irq - initialize interrupts
1100  * @adapter: board private structure
1101  *
1102  * Attempts to configure interrupts using the best available
1103  * capabilities of the hardware and kernel.
1104  **/
1105 static int ixgbevf_request_irq(struct ixgbevf_adapter *adapter)
1106 {
1107         int err = 0;
1108
1109         err = ixgbevf_request_msix_irqs(adapter);
1110
1111         if (err)
1112                 hw_dbg(&adapter->hw,
1113                        "request_irq failed, Error %d\n", err);
1114
1115         return err;
1116 }
1117
1118 static void ixgbevf_free_irq(struct ixgbevf_adapter *adapter)
1119 {
1120         struct net_device *netdev = adapter->netdev;
1121         int i, q_vectors;
1122
1123         q_vectors = adapter->num_msix_vectors;
1124
1125         i = q_vectors - 1;
1126
1127         free_irq(adapter->msix_entries[i].vector, netdev);
1128         i--;
1129
1130         for (; i >= 0; i--) {
1131                 free_irq(adapter->msix_entries[i].vector,
1132                          adapter->q_vector[i]);
1133         }
1134
1135         ixgbevf_reset_q_vectors(adapter);
1136 }
1137
1138 /**
1139  * ixgbevf_irq_disable - Mask off interrupt generation on the NIC
1140  * @adapter: board private structure
1141  **/
1142 static inline void ixgbevf_irq_disable(struct ixgbevf_adapter *adapter)
1143 {
1144         int i;
1145         struct ixgbe_hw *hw = &adapter->hw;
1146
1147         IXGBE_WRITE_REG(hw, IXGBE_VTEIMC, ~0);
1148
1149         IXGBE_WRITE_FLUSH(hw);
1150
1151         for (i = 0; i < adapter->num_msix_vectors; i++)
1152                 synchronize_irq(adapter->msix_entries[i].vector);
1153 }
1154
1155 /**
1156  * ixgbevf_irq_enable - Enable default interrupt generation settings
1157  * @adapter: board private structure
1158  **/
1159 static inline void ixgbevf_irq_enable(struct ixgbevf_adapter *adapter,
1160                                       bool queues, bool flush)
1161 {
1162         struct ixgbe_hw *hw = &adapter->hw;
1163         u32 mask;
1164         u64 qmask;
1165
1166         mask = (IXGBE_EIMS_ENABLE_MASK & ~IXGBE_EIMS_RTX_QUEUE);
1167         qmask = ~0;
1168
1169         IXGBE_WRITE_REG(hw, IXGBE_VTEIMS, mask);
1170
1171         if (queues)
1172                 ixgbevf_irq_enable_queues(adapter, qmask);
1173
1174         if (flush)
1175                 IXGBE_WRITE_FLUSH(hw);
1176 }
1177
1178 /**
1179  * ixgbevf_configure_tx - Configure 82599 VF Transmit Unit after Reset
1180  * @adapter: board private structure
1181  *
1182  * Configure the Tx unit of the MAC after a reset.
1183  **/
1184 static void ixgbevf_configure_tx(struct ixgbevf_adapter *adapter)
1185 {
1186         u64 tdba;
1187         struct ixgbe_hw *hw = &adapter->hw;
1188         u32 i, j, tdlen, txctrl;
1189
1190         /* Setup the HW Tx Head and Tail descriptor pointers */
1191         for (i = 0; i < adapter->num_tx_queues; i++) {
1192                 struct ixgbevf_ring *ring = &adapter->tx_ring[i];
1193                 j = ring->reg_idx;
1194                 tdba = ring->dma;
1195                 tdlen = ring->count * sizeof(union ixgbe_adv_tx_desc);
1196                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAL(j),
1197                                 (tdba & DMA_BIT_MASK(32)));
1198                 IXGBE_WRITE_REG(hw, IXGBE_VFTDBAH(j), (tdba >> 32));
1199                 IXGBE_WRITE_REG(hw, IXGBE_VFTDLEN(j), tdlen);
1200                 IXGBE_WRITE_REG(hw, IXGBE_VFTDH(j), 0);
1201                 IXGBE_WRITE_REG(hw, IXGBE_VFTDT(j), 0);
1202                 adapter->tx_ring[i].head = IXGBE_VFTDH(j);
1203                 adapter->tx_ring[i].tail = IXGBE_VFTDT(j);
1204                 /* Disable Tx Head Writeback RO bit, since this hoses
1205                  * bookkeeping if things aren't delivered in order.
1206                  */
1207                 txctrl = IXGBE_READ_REG(hw, IXGBE_VFDCA_TXCTRL(j));
1208                 txctrl &= ~IXGBE_DCA_TXCTRL_TX_WB_RO_EN;
1209                 IXGBE_WRITE_REG(hw, IXGBE_VFDCA_TXCTRL(j), txctrl);
1210         }
1211 }
1212
1213 #define IXGBE_SRRCTL_BSIZEHDRSIZE_SHIFT 2
1214
1215 static void ixgbevf_configure_srrctl(struct ixgbevf_adapter *adapter, int index)
1216 {
1217         struct ixgbevf_ring *rx_ring;
1218         struct ixgbe_hw *hw = &adapter->hw;
1219         u32 srrctl;
1220
1221         rx_ring = &adapter->rx_ring[index];
1222
1223         srrctl = IXGBE_SRRCTL_DROP_EN;
1224
1225         srrctl |= IXGBE_SRRCTL_DESCTYPE_ADV_ONEBUF;
1226
1227         if (rx_ring->rx_buf_len == MAXIMUM_ETHERNET_VLAN_SIZE)
1228                 srrctl |= IXGBEVF_RXBUFFER_2048 >>
1229                         IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1230         else
1231                 srrctl |= rx_ring->rx_buf_len >>
1232                         IXGBE_SRRCTL_BSIZEPKT_SHIFT;
1233         IXGBE_WRITE_REG(hw, IXGBE_VFSRRCTL(index), srrctl);
1234 }
1235
1236 /**
1237  * ixgbevf_configure_rx - Configure 82599 VF Receive Unit after Reset
1238  * @adapter: board private structure
1239  *
1240  * Configure the Rx unit of the MAC after a reset.
1241  **/
1242 static void ixgbevf_configure_rx(struct ixgbevf_adapter *adapter)
1243 {
1244         u64 rdba;
1245         struct ixgbe_hw *hw = &adapter->hw;
1246         struct net_device *netdev = adapter->netdev;
1247         int max_frame = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1248         int i, j;
1249         u32 rdlen;
1250         int rx_buf_len;
1251
1252         /* PSRTYPE must be initialized in 82599 */
1253         IXGBE_WRITE_REG(hw, IXGBE_VFPSRTYPE, 0);
1254         if (netdev->mtu <= ETH_DATA_LEN)
1255                 rx_buf_len = MAXIMUM_ETHERNET_VLAN_SIZE;
1256         else
1257                 rx_buf_len = ALIGN(max_frame, 1024);
1258
1259         rdlen = adapter->rx_ring[0].count * sizeof(union ixgbe_adv_rx_desc);
1260         /* Setup the HW Rx Head and Tail Descriptor Pointers and
1261          * the Base and Length of the Rx Descriptor Ring */
1262         for (i = 0; i < adapter->num_rx_queues; i++) {
1263                 rdba = adapter->rx_ring[i].dma;
1264                 j = adapter->rx_ring[i].reg_idx;
1265                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAL(j),
1266                                 (rdba & DMA_BIT_MASK(32)));
1267                 IXGBE_WRITE_REG(hw, IXGBE_VFRDBAH(j), (rdba >> 32));
1268                 IXGBE_WRITE_REG(hw, IXGBE_VFRDLEN(j), rdlen);
1269                 IXGBE_WRITE_REG(hw, IXGBE_VFRDH(j), 0);
1270                 IXGBE_WRITE_REG(hw, IXGBE_VFRDT(j), 0);
1271                 adapter->rx_ring[i].head = IXGBE_VFRDH(j);
1272                 adapter->rx_ring[i].tail = IXGBE_VFRDT(j);
1273                 adapter->rx_ring[i].rx_buf_len = rx_buf_len;
1274
1275                 ixgbevf_configure_srrctl(adapter, j);
1276         }
1277 }
1278
1279 static int ixgbevf_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1280 {
1281         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1282         struct ixgbe_hw *hw = &adapter->hw;
1283
1284         /* add VID to filter table */
1285         if (hw->mac.ops.set_vfta)
1286                 hw->mac.ops.set_vfta(hw, vid, 0, true);
1287         set_bit(vid, adapter->active_vlans);
1288
1289         return 0;
1290 }
1291
1292 static int ixgbevf_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1293 {
1294         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1295         struct ixgbe_hw *hw = &adapter->hw;
1296
1297         /* remove VID from filter table */
1298         if (hw->mac.ops.set_vfta)
1299                 hw->mac.ops.set_vfta(hw, vid, 0, false);
1300         clear_bit(vid, adapter->active_vlans);
1301
1302         return 0;
1303 }
1304
1305 static void ixgbevf_restore_vlan(struct ixgbevf_adapter *adapter)
1306 {
1307         u16 vid;
1308
1309         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1310                 ixgbevf_vlan_rx_add_vid(adapter->netdev, vid);
1311 }
1312
1313 static int ixgbevf_write_uc_addr_list(struct net_device *netdev)
1314 {
1315         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1316         struct ixgbe_hw *hw = &adapter->hw;
1317         int count = 0;
1318
1319         if ((netdev_uc_count(netdev)) > 10) {
1320                 pr_err("Too many unicast filters - No Space\n");
1321                 return -ENOSPC;
1322         }
1323
1324         if (!netdev_uc_empty(netdev)) {
1325                 struct netdev_hw_addr *ha;
1326                 netdev_for_each_uc_addr(ha, netdev) {
1327                         hw->mac.ops.set_uc_addr(hw, ++count, ha->addr);
1328                         udelay(200);
1329                 }
1330         } else {
1331                 /*
1332                  * If the list is empty then send message to PF driver to
1333                  * clear all macvlans on this VF.
1334                  */
1335                 hw->mac.ops.set_uc_addr(hw, 0, NULL);
1336         }
1337
1338         return count;
1339 }
1340
1341 /**
1342  * ixgbevf_set_rx_mode - Multicast set
1343  * @netdev: network interface device structure
1344  *
1345  * The set_rx_method entry point is called whenever the multicast address
1346  * list or the network interface flags are updated.  This routine is
1347  * responsible for configuring the hardware for proper multicast mode.
1348  **/
1349 static void ixgbevf_set_rx_mode(struct net_device *netdev)
1350 {
1351         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
1352         struct ixgbe_hw *hw = &adapter->hw;
1353
1354         /* reprogram multicast list */
1355         if (hw->mac.ops.update_mc_addr_list)
1356                 hw->mac.ops.update_mc_addr_list(hw, netdev);
1357
1358         ixgbevf_write_uc_addr_list(netdev);
1359 }
1360
1361 static void ixgbevf_napi_enable_all(struct ixgbevf_adapter *adapter)
1362 {
1363         int q_idx;
1364         struct ixgbevf_q_vector *q_vector;
1365         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1366
1367         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1368                 struct napi_struct *napi;
1369                 q_vector = adapter->q_vector[q_idx];
1370                 if (!q_vector->rxr_count)
1371                         continue;
1372                 napi = &q_vector->napi;
1373                 if (q_vector->rxr_count > 1)
1374                         napi->poll = &ixgbevf_clean_rxonly_many;
1375
1376                 napi_enable(napi);
1377         }
1378 }
1379
1380 static void ixgbevf_napi_disable_all(struct ixgbevf_adapter *adapter)
1381 {
1382         int q_idx;
1383         struct ixgbevf_q_vector *q_vector;
1384         int q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1385
1386         for (q_idx = 0; q_idx < q_vectors; q_idx++) {
1387                 q_vector = adapter->q_vector[q_idx];
1388                 if (!q_vector->rxr_count)
1389                         continue;
1390                 napi_disable(&q_vector->napi);
1391         }
1392 }
1393
1394 static void ixgbevf_configure(struct ixgbevf_adapter *adapter)
1395 {
1396         struct net_device *netdev = adapter->netdev;
1397         int i;
1398
1399         ixgbevf_set_rx_mode(netdev);
1400
1401         ixgbevf_restore_vlan(adapter);
1402
1403         ixgbevf_configure_tx(adapter);
1404         ixgbevf_configure_rx(adapter);
1405         for (i = 0; i < adapter->num_rx_queues; i++) {
1406                 struct ixgbevf_ring *ring = &adapter->rx_ring[i];
1407                 ixgbevf_alloc_rx_buffers(adapter, ring, ring->count);
1408                 ring->next_to_use = ring->count - 1;
1409                 writel(ring->next_to_use, adapter->hw.hw_addr + ring->tail);
1410         }
1411 }
1412
1413 #define IXGBE_MAX_RX_DESC_POLL 10
1414 static inline void ixgbevf_rx_desc_queue_enable(struct ixgbevf_adapter *adapter,
1415                                                 int rxr)
1416 {
1417         struct ixgbe_hw *hw = &adapter->hw;
1418         int j = adapter->rx_ring[rxr].reg_idx;
1419         int k;
1420
1421         for (k = 0; k < IXGBE_MAX_RX_DESC_POLL; k++) {
1422                 if (IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j)) & IXGBE_RXDCTL_ENABLE)
1423                         break;
1424                 else
1425                         msleep(1);
1426         }
1427         if (k >= IXGBE_MAX_RX_DESC_POLL) {
1428                 hw_dbg(hw, "RXDCTL.ENABLE on Rx queue %d "
1429                        "not set within the polling period\n", rxr);
1430         }
1431
1432         ixgbevf_release_rx_desc(&adapter->hw, &adapter->rx_ring[rxr],
1433                                 (adapter->rx_ring[rxr].count - 1));
1434 }
1435
1436 static void ixgbevf_save_reset_stats(struct ixgbevf_adapter *adapter)
1437 {
1438         /* Only save pre-reset stats if there are some */
1439         if (adapter->stats.vfgprc || adapter->stats.vfgptc) {
1440                 adapter->stats.saved_reset_vfgprc += adapter->stats.vfgprc -
1441                         adapter->stats.base_vfgprc;
1442                 adapter->stats.saved_reset_vfgptc += adapter->stats.vfgptc -
1443                         adapter->stats.base_vfgptc;
1444                 adapter->stats.saved_reset_vfgorc += adapter->stats.vfgorc -
1445                         adapter->stats.base_vfgorc;
1446                 adapter->stats.saved_reset_vfgotc += adapter->stats.vfgotc -
1447                         adapter->stats.base_vfgotc;
1448                 adapter->stats.saved_reset_vfmprc += adapter->stats.vfmprc -
1449                         adapter->stats.base_vfmprc;
1450         }
1451 }
1452
1453 static void ixgbevf_init_last_counter_stats(struct ixgbevf_adapter *adapter)
1454 {
1455         struct ixgbe_hw *hw = &adapter->hw;
1456
1457         adapter->stats.last_vfgprc = IXGBE_READ_REG(hw, IXGBE_VFGPRC);
1458         adapter->stats.last_vfgorc = IXGBE_READ_REG(hw, IXGBE_VFGORC_LSB);
1459         adapter->stats.last_vfgorc |=
1460                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGORC_MSB))) << 32);
1461         adapter->stats.last_vfgptc = IXGBE_READ_REG(hw, IXGBE_VFGPTC);
1462         adapter->stats.last_vfgotc = IXGBE_READ_REG(hw, IXGBE_VFGOTC_LSB);
1463         adapter->stats.last_vfgotc |=
1464                 (((u64)(IXGBE_READ_REG(hw, IXGBE_VFGOTC_MSB))) << 32);
1465         adapter->stats.last_vfmprc = IXGBE_READ_REG(hw, IXGBE_VFMPRC);
1466
1467         adapter->stats.base_vfgprc = adapter->stats.last_vfgprc;
1468         adapter->stats.base_vfgorc = adapter->stats.last_vfgorc;
1469         adapter->stats.base_vfgptc = adapter->stats.last_vfgptc;
1470         adapter->stats.base_vfgotc = adapter->stats.last_vfgotc;
1471         adapter->stats.base_vfmprc = adapter->stats.last_vfmprc;
1472 }
1473
1474 static void ixgbevf_up_complete(struct ixgbevf_adapter *adapter)
1475 {
1476         struct net_device *netdev = adapter->netdev;
1477         struct ixgbe_hw *hw = &adapter->hw;
1478         int i, j = 0;
1479         int num_rx_rings = adapter->num_rx_queues;
1480         u32 txdctl, rxdctl;
1481         u32 msg[2];
1482
1483         for (i = 0; i < adapter->num_tx_queues; i++) {
1484                 j = adapter->tx_ring[i].reg_idx;
1485                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1486                 /* enable WTHRESH=8 descriptors, to encourage burst writeback */
1487                 txdctl |= (8 << 16);
1488                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1489         }
1490
1491         for (i = 0; i < adapter->num_tx_queues; i++) {
1492                 j = adapter->tx_ring[i].reg_idx;
1493                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1494                 txdctl |= IXGBE_TXDCTL_ENABLE;
1495                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j), txdctl);
1496         }
1497
1498         for (i = 0; i < num_rx_rings; i++) {
1499                 j = adapter->rx_ring[i].reg_idx;
1500                 rxdctl = IXGBE_READ_REG(hw, IXGBE_VFRXDCTL(j));
1501                 rxdctl |= IXGBE_RXDCTL_ENABLE | IXGBE_RXDCTL_VME;
1502                 if (hw->mac.type == ixgbe_mac_X540_vf) {
1503                         rxdctl &= ~IXGBE_RXDCTL_RLPMLMASK;
1504                         rxdctl |= ((netdev->mtu + ETH_HLEN + ETH_FCS_LEN) |
1505                                    IXGBE_RXDCTL_RLPML_EN);
1506                 }
1507                 IXGBE_WRITE_REG(hw, IXGBE_VFRXDCTL(j), rxdctl);
1508                 ixgbevf_rx_desc_queue_enable(adapter, i);
1509         }
1510
1511         ixgbevf_configure_msix(adapter);
1512
1513         if (hw->mac.ops.set_rar) {
1514                 if (is_valid_ether_addr(hw->mac.addr))
1515                         hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
1516                 else
1517                         hw->mac.ops.set_rar(hw, 0, hw->mac.perm_addr, 0);
1518         }
1519
1520         msg[0] = IXGBE_VF_SET_LPE;
1521         msg[1] = netdev->mtu + ETH_HLEN + ETH_FCS_LEN;
1522         hw->mbx.ops.write_posted(hw, msg, 2);
1523
1524         clear_bit(__IXGBEVF_DOWN, &adapter->state);
1525         ixgbevf_napi_enable_all(adapter);
1526
1527         /* enable transmits */
1528         netif_tx_start_all_queues(netdev);
1529
1530         ixgbevf_save_reset_stats(adapter);
1531         ixgbevf_init_last_counter_stats(adapter);
1532
1533         mod_timer(&adapter->watchdog_timer, jiffies);
1534 }
1535
1536 void ixgbevf_up(struct ixgbevf_adapter *adapter)
1537 {
1538         struct ixgbe_hw *hw = &adapter->hw;
1539
1540         ixgbevf_configure(adapter);
1541
1542         ixgbevf_up_complete(adapter);
1543
1544         /* clear any pending interrupts, may auto mask */
1545         IXGBE_READ_REG(hw, IXGBE_VTEICR);
1546
1547         ixgbevf_irq_enable(adapter, true, true);
1548 }
1549
1550 /**
1551  * ixgbevf_clean_rx_ring - Free Rx Buffers per Queue
1552  * @adapter: board private structure
1553  * @rx_ring: ring to free buffers from
1554  **/
1555 static void ixgbevf_clean_rx_ring(struct ixgbevf_adapter *adapter,
1556                                   struct ixgbevf_ring *rx_ring)
1557 {
1558         struct pci_dev *pdev = adapter->pdev;
1559         unsigned long size;
1560         unsigned int i;
1561
1562         if (!rx_ring->rx_buffer_info)
1563                 return;
1564
1565         /* Free all the Rx ring sk_buffs */
1566         for (i = 0; i < rx_ring->count; i++) {
1567                 struct ixgbevf_rx_buffer *rx_buffer_info;
1568
1569                 rx_buffer_info = &rx_ring->rx_buffer_info[i];
1570                 if (rx_buffer_info->dma) {
1571                         dma_unmap_single(&pdev->dev, rx_buffer_info->dma,
1572                                          rx_ring->rx_buf_len,
1573                                          DMA_FROM_DEVICE);
1574                         rx_buffer_info->dma = 0;
1575                 }
1576                 if (rx_buffer_info->skb) {
1577                         struct sk_buff *skb = rx_buffer_info->skb;
1578                         rx_buffer_info->skb = NULL;
1579                         do {
1580                                 struct sk_buff *this = skb;
1581                                 skb = skb->prev;
1582                                 dev_kfree_skb(this);
1583                         } while (skb);
1584                 }
1585         }
1586
1587         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
1588         memset(rx_ring->rx_buffer_info, 0, size);
1589
1590         /* Zero out the descriptor ring */
1591         memset(rx_ring->desc, 0, rx_ring->size);
1592
1593         rx_ring->next_to_clean = 0;
1594         rx_ring->next_to_use = 0;
1595
1596         if (rx_ring->head)
1597                 writel(0, adapter->hw.hw_addr + rx_ring->head);
1598         if (rx_ring->tail)
1599                 writel(0, adapter->hw.hw_addr + rx_ring->tail);
1600 }
1601
1602 /**
1603  * ixgbevf_clean_tx_ring - Free Tx Buffers
1604  * @adapter: board private structure
1605  * @tx_ring: ring to be cleaned
1606  **/
1607 static void ixgbevf_clean_tx_ring(struct ixgbevf_adapter *adapter,
1608                                   struct ixgbevf_ring *tx_ring)
1609 {
1610         struct ixgbevf_tx_buffer *tx_buffer_info;
1611         unsigned long size;
1612         unsigned int i;
1613
1614         if (!tx_ring->tx_buffer_info)
1615                 return;
1616
1617         /* Free all the Tx ring sk_buffs */
1618
1619         for (i = 0; i < tx_ring->count; i++) {
1620                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
1621                 ixgbevf_unmap_and_free_tx_resource(adapter, tx_buffer_info);
1622         }
1623
1624         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
1625         memset(tx_ring->tx_buffer_info, 0, size);
1626
1627         memset(tx_ring->desc, 0, tx_ring->size);
1628
1629         tx_ring->next_to_use = 0;
1630         tx_ring->next_to_clean = 0;
1631
1632         if (tx_ring->head)
1633                 writel(0, adapter->hw.hw_addr + tx_ring->head);
1634         if (tx_ring->tail)
1635                 writel(0, adapter->hw.hw_addr + tx_ring->tail);
1636 }
1637
1638 /**
1639  * ixgbevf_clean_all_rx_rings - Free Rx Buffers for all queues
1640  * @adapter: board private structure
1641  **/
1642 static void ixgbevf_clean_all_rx_rings(struct ixgbevf_adapter *adapter)
1643 {
1644         int i;
1645
1646         for (i = 0; i < adapter->num_rx_queues; i++)
1647                 ixgbevf_clean_rx_ring(adapter, &adapter->rx_ring[i]);
1648 }
1649
1650 /**
1651  * ixgbevf_clean_all_tx_rings - Free Tx Buffers for all queues
1652  * @adapter: board private structure
1653  **/
1654 static void ixgbevf_clean_all_tx_rings(struct ixgbevf_adapter *adapter)
1655 {
1656         int i;
1657
1658         for (i = 0; i < adapter->num_tx_queues; i++)
1659                 ixgbevf_clean_tx_ring(adapter, &adapter->tx_ring[i]);
1660 }
1661
1662 void ixgbevf_down(struct ixgbevf_adapter *adapter)
1663 {
1664         struct net_device *netdev = adapter->netdev;
1665         struct ixgbe_hw *hw = &adapter->hw;
1666         u32 txdctl;
1667         int i, j;
1668
1669         /* signal that we are down to the interrupt handler */
1670         set_bit(__IXGBEVF_DOWN, &adapter->state);
1671         /* disable receives */
1672
1673         netif_tx_disable(netdev);
1674
1675         msleep(10);
1676
1677         netif_tx_stop_all_queues(netdev);
1678
1679         ixgbevf_irq_disable(adapter);
1680
1681         ixgbevf_napi_disable_all(adapter);
1682
1683         del_timer_sync(&adapter->watchdog_timer);
1684         /* can't call flush scheduled work here because it can deadlock
1685          * if linkwatch_event tries to acquire the rtnl_lock which we are
1686          * holding */
1687         while (adapter->flags & IXGBE_FLAG_IN_WATCHDOG_TASK)
1688                 msleep(1);
1689
1690         /* disable transmits in the hardware now that interrupts are off */
1691         for (i = 0; i < adapter->num_tx_queues; i++) {
1692                 j = adapter->tx_ring[i].reg_idx;
1693                 txdctl = IXGBE_READ_REG(hw, IXGBE_VFTXDCTL(j));
1694                 IXGBE_WRITE_REG(hw, IXGBE_VFTXDCTL(j),
1695                                 (txdctl & ~IXGBE_TXDCTL_ENABLE));
1696         }
1697
1698         netif_carrier_off(netdev);
1699
1700         if (!pci_channel_offline(adapter->pdev))
1701                 ixgbevf_reset(adapter);
1702
1703         ixgbevf_clean_all_tx_rings(adapter);
1704         ixgbevf_clean_all_rx_rings(adapter);
1705 }
1706
1707 void ixgbevf_reinit_locked(struct ixgbevf_adapter *adapter)
1708 {
1709         struct ixgbe_hw *hw = &adapter->hw;
1710
1711         WARN_ON(in_interrupt());
1712
1713         while (test_and_set_bit(__IXGBEVF_RESETTING, &adapter->state))
1714                 msleep(1);
1715
1716         /*
1717          * Check if PF is up before re-init.  If not then skip until
1718          * later when the PF is up and ready to service requests from
1719          * the VF via mailbox.  If the VF is up and running then the
1720          * watchdog task will continue to schedule reset tasks until
1721          * the PF is up and running.
1722          */
1723         if (!hw->mac.ops.reset_hw(hw)) {
1724                 ixgbevf_down(adapter);
1725                 ixgbevf_up(adapter);
1726         }
1727
1728         clear_bit(__IXGBEVF_RESETTING, &adapter->state);
1729 }
1730
1731 void ixgbevf_reset(struct ixgbevf_adapter *adapter)
1732 {
1733         struct ixgbe_hw *hw = &adapter->hw;
1734         struct net_device *netdev = adapter->netdev;
1735
1736         if (hw->mac.ops.reset_hw(hw))
1737                 hw_dbg(hw, "PF still resetting\n");
1738         else
1739                 hw->mac.ops.init_hw(hw);
1740
1741         if (is_valid_ether_addr(adapter->hw.mac.addr)) {
1742                 memcpy(netdev->dev_addr, adapter->hw.mac.addr,
1743                        netdev->addr_len);
1744                 memcpy(netdev->perm_addr, adapter->hw.mac.addr,
1745                        netdev->addr_len);
1746         }
1747 }
1748
1749 static void ixgbevf_acquire_msix_vectors(struct ixgbevf_adapter *adapter,
1750                                          int vectors)
1751 {
1752         int err, vector_threshold;
1753
1754         /* We'll want at least 3 (vector_threshold):
1755          * 1) TxQ[0] Cleanup
1756          * 2) RxQ[0] Cleanup
1757          * 3) Other (Link Status Change, etc.)
1758          */
1759         vector_threshold = MIN_MSIX_COUNT;
1760
1761         /* The more we get, the more we will assign to Tx/Rx Cleanup
1762          * for the separate queues...where Rx Cleanup >= Tx Cleanup.
1763          * Right now, we simply care about how many we'll get; we'll
1764          * set them up later while requesting irq's.
1765          */
1766         while (vectors >= vector_threshold) {
1767                 err = pci_enable_msix(adapter->pdev, adapter->msix_entries,
1768                                       vectors);
1769                 if (!err) /* Success in acquiring all requested vectors. */
1770                         break;
1771                 else if (err < 0)
1772                         vectors = 0; /* Nasty failure, quit now */
1773                 else /* err == number of vectors we should try again with */
1774                         vectors = err;
1775         }
1776
1777         if (vectors < vector_threshold) {
1778                 /* Can't allocate enough MSI-X interrupts?  Oh well.
1779                  * This just means we'll go with either a single MSI
1780                  * vector or fall back to legacy interrupts.
1781                  */
1782                 hw_dbg(&adapter->hw,
1783                        "Unable to allocate MSI-X interrupts\n");
1784                 kfree(adapter->msix_entries);
1785                 adapter->msix_entries = NULL;
1786         } else {
1787                 /*
1788                  * Adjust for only the vectors we'll use, which is minimum
1789                  * of max_msix_q_vectors + NON_Q_VECTORS, or the number of
1790                  * vectors we were allocated.
1791                  */
1792                 adapter->num_msix_vectors = vectors;
1793         }
1794 }
1795
1796 /**
1797  * ixgbevf_set_num_queues - Allocate queues for device, feature dependent
1798  * @adapter: board private structure to initialize
1799  *
1800  * This is the top level queue allocation routine.  The order here is very
1801  * important, starting with the "most" number of features turned on at once,
1802  * and ending with the smallest set of features.  This way large combinations
1803  * can be allocated if they're turned on, and smaller combinations are the
1804  * fallthrough conditions.
1805  *
1806  **/
1807 static void ixgbevf_set_num_queues(struct ixgbevf_adapter *adapter)
1808 {
1809         /* Start with base case */
1810         adapter->num_rx_queues = 1;
1811         adapter->num_tx_queues = 1;
1812 }
1813
1814 /**
1815  * ixgbevf_alloc_queues - Allocate memory for all rings
1816  * @adapter: board private structure to initialize
1817  *
1818  * We allocate one ring per queue at run-time since we don't know the
1819  * number of queues at compile-time.  The polling_netdev array is
1820  * intended for Multiqueue, but should work fine with a single queue.
1821  **/
1822 static int ixgbevf_alloc_queues(struct ixgbevf_adapter *adapter)
1823 {
1824         int i;
1825
1826         adapter->tx_ring = kcalloc(adapter->num_tx_queues,
1827                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1828         if (!adapter->tx_ring)
1829                 goto err_tx_ring_allocation;
1830
1831         adapter->rx_ring = kcalloc(adapter->num_rx_queues,
1832                                    sizeof(struct ixgbevf_ring), GFP_KERNEL);
1833         if (!adapter->rx_ring)
1834                 goto err_rx_ring_allocation;
1835
1836         for (i = 0; i < adapter->num_tx_queues; i++) {
1837                 adapter->tx_ring[i].count = adapter->tx_ring_count;
1838                 adapter->tx_ring[i].queue_index = i;
1839                 adapter->tx_ring[i].reg_idx = i;
1840         }
1841
1842         for (i = 0; i < adapter->num_rx_queues; i++) {
1843                 adapter->rx_ring[i].count = adapter->rx_ring_count;
1844                 adapter->rx_ring[i].queue_index = i;
1845                 adapter->rx_ring[i].reg_idx = i;
1846         }
1847
1848         return 0;
1849
1850 err_rx_ring_allocation:
1851         kfree(adapter->tx_ring);
1852 err_tx_ring_allocation:
1853         return -ENOMEM;
1854 }
1855
1856 /**
1857  * ixgbevf_set_interrupt_capability - set MSI-X or FAIL if not supported
1858  * @adapter: board private structure to initialize
1859  *
1860  * Attempt to configure the interrupts using the best available
1861  * capabilities of the hardware and the kernel.
1862  **/
1863 static int ixgbevf_set_interrupt_capability(struct ixgbevf_adapter *adapter)
1864 {
1865         int err = 0;
1866         int vector, v_budget;
1867
1868         /*
1869          * It's easy to be greedy for MSI-X vectors, but it really
1870          * doesn't do us much good if we have a lot more vectors
1871          * than CPU's.  So let's be conservative and only ask for
1872          * (roughly) twice the number of vectors as there are CPU's.
1873          */
1874         v_budget = min(adapter->num_rx_queues + adapter->num_tx_queues,
1875                        (int)(num_online_cpus() * 2)) + NON_Q_VECTORS;
1876
1877         /* A failure in MSI-X entry allocation isn't fatal, but it does
1878          * mean we disable MSI-X capabilities of the adapter. */
1879         adapter->msix_entries = kcalloc(v_budget,
1880                                         sizeof(struct msix_entry), GFP_KERNEL);
1881         if (!adapter->msix_entries) {
1882                 err = -ENOMEM;
1883                 goto out;
1884         }
1885
1886         for (vector = 0; vector < v_budget; vector++)
1887                 adapter->msix_entries[vector].entry = vector;
1888
1889         ixgbevf_acquire_msix_vectors(adapter, v_budget);
1890
1891 out:
1892         return err;
1893 }
1894
1895 /**
1896  * ixgbevf_alloc_q_vectors - Allocate memory for interrupt vectors
1897  * @adapter: board private structure to initialize
1898  *
1899  * We allocate one q_vector per queue interrupt.  If allocation fails we
1900  * return -ENOMEM.
1901  **/
1902 static int ixgbevf_alloc_q_vectors(struct ixgbevf_adapter *adapter)
1903 {
1904         int q_idx, num_q_vectors;
1905         struct ixgbevf_q_vector *q_vector;
1906         int napi_vectors;
1907         int (*poll)(struct napi_struct *, int);
1908
1909         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1910         napi_vectors = adapter->num_rx_queues;
1911         poll = &ixgbevf_clean_rxonly;
1912
1913         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1914                 q_vector = kzalloc(sizeof(struct ixgbevf_q_vector), GFP_KERNEL);
1915                 if (!q_vector)
1916                         goto err_out;
1917                 q_vector->adapter = adapter;
1918                 q_vector->v_idx = q_idx;
1919                 q_vector->eitr = adapter->eitr_param;
1920                 if (q_idx < napi_vectors)
1921                         netif_napi_add(adapter->netdev, &q_vector->napi,
1922                                        (*poll), 64);
1923                 adapter->q_vector[q_idx] = q_vector;
1924         }
1925
1926         return 0;
1927
1928 err_out:
1929         while (q_idx) {
1930                 q_idx--;
1931                 q_vector = adapter->q_vector[q_idx];
1932                 netif_napi_del(&q_vector->napi);
1933                 kfree(q_vector);
1934                 adapter->q_vector[q_idx] = NULL;
1935         }
1936         return -ENOMEM;
1937 }
1938
1939 /**
1940  * ixgbevf_free_q_vectors - Free memory allocated for interrupt vectors
1941  * @adapter: board private structure to initialize
1942  *
1943  * This function frees the memory allocated to the q_vectors.  In addition if
1944  * NAPI is enabled it will delete any references to the NAPI struct prior
1945  * to freeing the q_vector.
1946  **/
1947 static void ixgbevf_free_q_vectors(struct ixgbevf_adapter *adapter)
1948 {
1949         int q_idx, num_q_vectors;
1950         int napi_vectors;
1951
1952         num_q_vectors = adapter->num_msix_vectors - NON_Q_VECTORS;
1953         napi_vectors = adapter->num_rx_queues;
1954
1955         for (q_idx = 0; q_idx < num_q_vectors; q_idx++) {
1956                 struct ixgbevf_q_vector *q_vector = adapter->q_vector[q_idx];
1957
1958                 adapter->q_vector[q_idx] = NULL;
1959                 if (q_idx < napi_vectors)
1960                         netif_napi_del(&q_vector->napi);
1961                 kfree(q_vector);
1962         }
1963 }
1964
1965 /**
1966  * ixgbevf_reset_interrupt_capability - Reset MSIX setup
1967  * @adapter: board private structure
1968  *
1969  **/
1970 static void ixgbevf_reset_interrupt_capability(struct ixgbevf_adapter *adapter)
1971 {
1972         pci_disable_msix(adapter->pdev);
1973         kfree(adapter->msix_entries);
1974         adapter->msix_entries = NULL;
1975 }
1976
1977 /**
1978  * ixgbevf_init_interrupt_scheme - Determine if MSIX is supported and init
1979  * @adapter: board private structure to initialize
1980  *
1981  **/
1982 static int ixgbevf_init_interrupt_scheme(struct ixgbevf_adapter *adapter)
1983 {
1984         int err;
1985
1986         /* Number of supported queues */
1987         ixgbevf_set_num_queues(adapter);
1988
1989         err = ixgbevf_set_interrupt_capability(adapter);
1990         if (err) {
1991                 hw_dbg(&adapter->hw,
1992                        "Unable to setup interrupt capabilities\n");
1993                 goto err_set_interrupt;
1994         }
1995
1996         err = ixgbevf_alloc_q_vectors(adapter);
1997         if (err) {
1998                 hw_dbg(&adapter->hw, "Unable to allocate memory for queue "
1999                        "vectors\n");
2000                 goto err_alloc_q_vectors;
2001         }
2002
2003         err = ixgbevf_alloc_queues(adapter);
2004         if (err) {
2005                 pr_err("Unable to allocate memory for queues\n");
2006                 goto err_alloc_queues;
2007         }
2008
2009         hw_dbg(&adapter->hw, "Multiqueue %s: Rx Queue count = %u, "
2010                "Tx Queue count = %u\n",
2011                (adapter->num_rx_queues > 1) ? "Enabled" :
2012                "Disabled", adapter->num_rx_queues, adapter->num_tx_queues);
2013
2014         set_bit(__IXGBEVF_DOWN, &adapter->state);
2015
2016         return 0;
2017 err_alloc_queues:
2018         ixgbevf_free_q_vectors(adapter);
2019 err_alloc_q_vectors:
2020         ixgbevf_reset_interrupt_capability(adapter);
2021 err_set_interrupt:
2022         return err;
2023 }
2024
2025 /**
2026  * ixgbevf_sw_init - Initialize general software structures
2027  * (struct ixgbevf_adapter)
2028  * @adapter: board private structure to initialize
2029  *
2030  * ixgbevf_sw_init initializes the Adapter private data structure.
2031  * Fields are initialized based on PCI device information and
2032  * OS network device settings (MTU size).
2033  **/
2034 static int __devinit ixgbevf_sw_init(struct ixgbevf_adapter *adapter)
2035 {
2036         struct ixgbe_hw *hw = &adapter->hw;
2037         struct pci_dev *pdev = adapter->pdev;
2038         int err;
2039
2040         /* PCI config space info */
2041
2042         hw->vendor_id = pdev->vendor;
2043         hw->device_id = pdev->device;
2044         hw->revision_id = pdev->revision;
2045         hw->subsystem_vendor_id = pdev->subsystem_vendor;
2046         hw->subsystem_device_id = pdev->subsystem_device;
2047
2048         hw->mbx.ops.init_params(hw);
2049         hw->mac.max_tx_queues = MAX_TX_QUEUES;
2050         hw->mac.max_rx_queues = MAX_RX_QUEUES;
2051         err = hw->mac.ops.reset_hw(hw);
2052         if (err) {
2053                 dev_info(&pdev->dev,
2054                          "PF still in reset state, assigning new address\n");
2055                 eth_hw_addr_random(adapter->netdev);
2056                 memcpy(adapter->hw.mac.addr, adapter->netdev->dev_addr,
2057                         adapter->netdev->addr_len);
2058         } else {
2059                 err = hw->mac.ops.init_hw(hw);
2060                 if (err) {
2061                         pr_err("init_shared_code failed: %d\n", err);
2062                         goto out;
2063                 }
2064                 memcpy(adapter->netdev->dev_addr, adapter->hw.mac.addr,
2065                         adapter->netdev->addr_len);
2066         }
2067
2068         /* Enable dynamic interrupt throttling rates */
2069         adapter->eitr_param = 20000;
2070         adapter->itr_setting = 1;
2071
2072         /* set default ring sizes */
2073         adapter->tx_ring_count = IXGBEVF_DEFAULT_TXD;
2074         adapter->rx_ring_count = IXGBEVF_DEFAULT_RXD;
2075
2076         set_bit(__IXGBEVF_DOWN, &adapter->state);
2077         return 0;
2078
2079 out:
2080         return err;
2081 }
2082
2083 #define UPDATE_VF_COUNTER_32bit(reg, last_counter, counter)     \
2084         {                                                       \
2085                 u32 current_counter = IXGBE_READ_REG(hw, reg);  \
2086                 if (current_counter < last_counter)             \
2087                         counter += 0x100000000LL;               \
2088                 last_counter = current_counter;                 \
2089                 counter &= 0xFFFFFFFF00000000LL;                \
2090                 counter |= current_counter;                     \
2091         }
2092
2093 #define UPDATE_VF_COUNTER_36bit(reg_lsb, reg_msb, last_counter, counter) \
2094         {                                                                \
2095                 u64 current_counter_lsb = IXGBE_READ_REG(hw, reg_lsb);   \
2096                 u64 current_counter_msb = IXGBE_READ_REG(hw, reg_msb);   \
2097                 u64 current_counter = (current_counter_msb << 32) |      \
2098                         current_counter_lsb;                             \
2099                 if (current_counter < last_counter)                      \
2100                         counter += 0x1000000000LL;                       \
2101                 last_counter = current_counter;                          \
2102                 counter &= 0xFFFFFFF000000000LL;                         \
2103                 counter |= current_counter;                              \
2104         }
2105 /**
2106  * ixgbevf_update_stats - Update the board statistics counters.
2107  * @adapter: board private structure
2108  **/
2109 void ixgbevf_update_stats(struct ixgbevf_adapter *adapter)
2110 {
2111         struct ixgbe_hw *hw = &adapter->hw;
2112
2113         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPRC, adapter->stats.last_vfgprc,
2114                                 adapter->stats.vfgprc);
2115         UPDATE_VF_COUNTER_32bit(IXGBE_VFGPTC, adapter->stats.last_vfgptc,
2116                                 adapter->stats.vfgptc);
2117         UPDATE_VF_COUNTER_36bit(IXGBE_VFGORC_LSB, IXGBE_VFGORC_MSB,
2118                                 adapter->stats.last_vfgorc,
2119                                 adapter->stats.vfgorc);
2120         UPDATE_VF_COUNTER_36bit(IXGBE_VFGOTC_LSB, IXGBE_VFGOTC_MSB,
2121                                 adapter->stats.last_vfgotc,
2122                                 adapter->stats.vfgotc);
2123         UPDATE_VF_COUNTER_32bit(IXGBE_VFMPRC, adapter->stats.last_vfmprc,
2124                                 adapter->stats.vfmprc);
2125 }
2126
2127 /**
2128  * ixgbevf_watchdog - Timer Call-back
2129  * @data: pointer to adapter cast into an unsigned long
2130  **/
2131 static void ixgbevf_watchdog(unsigned long data)
2132 {
2133         struct ixgbevf_adapter *adapter = (struct ixgbevf_adapter *)data;
2134         struct ixgbe_hw *hw = &adapter->hw;
2135         u64 eics = 0;
2136         int i;
2137
2138         /*
2139          * Do the watchdog outside of interrupt context due to the lovely
2140          * delays that some of the newer hardware requires
2141          */
2142
2143         if (test_bit(__IXGBEVF_DOWN, &adapter->state))
2144                 goto watchdog_short_circuit;
2145
2146         /* get one bit for every active tx/rx interrupt vector */
2147         for (i = 0; i < adapter->num_msix_vectors - NON_Q_VECTORS; i++) {
2148                 struct ixgbevf_q_vector *qv = adapter->q_vector[i];
2149                 if (qv->rxr_count || qv->txr_count)
2150                         eics |= (1 << i);
2151         }
2152
2153         IXGBE_WRITE_REG(hw, IXGBE_VTEICS, (u32)eics);
2154
2155 watchdog_short_circuit:
2156         schedule_work(&adapter->watchdog_task);
2157 }
2158
2159 /**
2160  * ixgbevf_tx_timeout - Respond to a Tx Hang
2161  * @netdev: network interface device structure
2162  **/
2163 static void ixgbevf_tx_timeout(struct net_device *netdev)
2164 {
2165         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2166
2167         /* Do the reset outside of interrupt context */
2168         schedule_work(&adapter->reset_task);
2169 }
2170
2171 static void ixgbevf_reset_task(struct work_struct *work)
2172 {
2173         struct ixgbevf_adapter *adapter;
2174         adapter = container_of(work, struct ixgbevf_adapter, reset_task);
2175
2176         /* If we're already down or resetting, just bail */
2177         if (test_bit(__IXGBEVF_DOWN, &adapter->state) ||
2178             test_bit(__IXGBEVF_RESETTING, &adapter->state))
2179                 return;
2180
2181         adapter->tx_timeout_count++;
2182
2183         ixgbevf_reinit_locked(adapter);
2184 }
2185
2186 /**
2187  * ixgbevf_watchdog_task - worker thread to bring link up
2188  * @work: pointer to work_struct containing our data
2189  **/
2190 static void ixgbevf_watchdog_task(struct work_struct *work)
2191 {
2192         struct ixgbevf_adapter *adapter = container_of(work,
2193                                                        struct ixgbevf_adapter,
2194                                                        watchdog_task);
2195         struct net_device *netdev = adapter->netdev;
2196         struct ixgbe_hw *hw = &adapter->hw;
2197         u32 link_speed = adapter->link_speed;
2198         bool link_up = adapter->link_up;
2199
2200         adapter->flags |= IXGBE_FLAG_IN_WATCHDOG_TASK;
2201
2202         /*
2203          * Always check the link on the watchdog because we have
2204          * no LSC interrupt
2205          */
2206         if (hw->mac.ops.check_link) {
2207                 if ((hw->mac.ops.check_link(hw, &link_speed,
2208                                             &link_up, false)) != 0) {
2209                         adapter->link_up = link_up;
2210                         adapter->link_speed = link_speed;
2211                         netif_carrier_off(netdev);
2212                         netif_tx_stop_all_queues(netdev);
2213                         schedule_work(&adapter->reset_task);
2214                         goto pf_has_reset;
2215                 }
2216         } else {
2217                 /* always assume link is up, if no check link
2218                  * function */
2219                 link_speed = IXGBE_LINK_SPEED_10GB_FULL;
2220                 link_up = true;
2221         }
2222         adapter->link_up = link_up;
2223         adapter->link_speed = link_speed;
2224
2225         if (link_up) {
2226                 if (!netif_carrier_ok(netdev)) {
2227                         hw_dbg(&adapter->hw, "NIC Link is Up, %u Gbps\n",
2228                                (link_speed == IXGBE_LINK_SPEED_10GB_FULL) ?
2229                                10 : 1);
2230                         netif_carrier_on(netdev);
2231                         netif_tx_wake_all_queues(netdev);
2232                 }
2233         } else {
2234                 adapter->link_up = false;
2235                 adapter->link_speed = 0;
2236                 if (netif_carrier_ok(netdev)) {
2237                         hw_dbg(&adapter->hw, "NIC Link is Down\n");
2238                         netif_carrier_off(netdev);
2239                         netif_tx_stop_all_queues(netdev);
2240                 }
2241         }
2242
2243         ixgbevf_update_stats(adapter);
2244
2245 pf_has_reset:
2246         /* Reset the timer */
2247         if (!test_bit(__IXGBEVF_DOWN, &adapter->state))
2248                 mod_timer(&adapter->watchdog_timer,
2249                           round_jiffies(jiffies + (2 * HZ)));
2250
2251         adapter->flags &= ~IXGBE_FLAG_IN_WATCHDOG_TASK;
2252 }
2253
2254 /**
2255  * ixgbevf_free_tx_resources - Free Tx Resources per Queue
2256  * @adapter: board private structure
2257  * @tx_ring: Tx descriptor ring for a specific queue
2258  *
2259  * Free all transmit software resources
2260  **/
2261 void ixgbevf_free_tx_resources(struct ixgbevf_adapter *adapter,
2262                                struct ixgbevf_ring *tx_ring)
2263 {
2264         struct pci_dev *pdev = adapter->pdev;
2265
2266         ixgbevf_clean_tx_ring(adapter, tx_ring);
2267
2268         vfree(tx_ring->tx_buffer_info);
2269         tx_ring->tx_buffer_info = NULL;
2270
2271         dma_free_coherent(&pdev->dev, tx_ring->size, tx_ring->desc,
2272                           tx_ring->dma);
2273
2274         tx_ring->desc = NULL;
2275 }
2276
2277 /**
2278  * ixgbevf_free_all_tx_resources - Free Tx Resources for All Queues
2279  * @adapter: board private structure
2280  *
2281  * Free all transmit software resources
2282  **/
2283 static void ixgbevf_free_all_tx_resources(struct ixgbevf_adapter *adapter)
2284 {
2285         int i;
2286
2287         for (i = 0; i < adapter->num_tx_queues; i++)
2288                 if (adapter->tx_ring[i].desc)
2289                         ixgbevf_free_tx_resources(adapter,
2290                                                   &adapter->tx_ring[i]);
2291
2292 }
2293
2294 /**
2295  * ixgbevf_setup_tx_resources - allocate Tx resources (Descriptors)
2296  * @adapter: board private structure
2297  * @tx_ring:    tx descriptor ring (for a specific queue) to setup
2298  *
2299  * Return 0 on success, negative on failure
2300  **/
2301 int ixgbevf_setup_tx_resources(struct ixgbevf_adapter *adapter,
2302                                struct ixgbevf_ring *tx_ring)
2303 {
2304         struct pci_dev *pdev = adapter->pdev;
2305         int size;
2306
2307         size = sizeof(struct ixgbevf_tx_buffer) * tx_ring->count;
2308         tx_ring->tx_buffer_info = vzalloc(size);
2309         if (!tx_ring->tx_buffer_info)
2310                 goto err;
2311
2312         /* round up to nearest 4K */
2313         tx_ring->size = tx_ring->count * sizeof(union ixgbe_adv_tx_desc);
2314         tx_ring->size = ALIGN(tx_ring->size, 4096);
2315
2316         tx_ring->desc = dma_alloc_coherent(&pdev->dev, tx_ring->size,
2317                                            &tx_ring->dma, GFP_KERNEL);
2318         if (!tx_ring->desc)
2319                 goto err;
2320
2321         tx_ring->next_to_use = 0;
2322         tx_ring->next_to_clean = 0;
2323         tx_ring->work_limit = tx_ring->count;
2324         return 0;
2325
2326 err:
2327         vfree(tx_ring->tx_buffer_info);
2328         tx_ring->tx_buffer_info = NULL;
2329         hw_dbg(&adapter->hw, "Unable to allocate memory for the transmit "
2330                "descriptor ring\n");
2331         return -ENOMEM;
2332 }
2333
2334 /**
2335  * ixgbevf_setup_all_tx_resources - allocate all queues Tx resources
2336  * @adapter: board private structure
2337  *
2338  * If this function returns with an error, then it's possible one or
2339  * more of the rings is populated (while the rest are not).  It is the
2340  * callers duty to clean those orphaned rings.
2341  *
2342  * Return 0 on success, negative on failure
2343  **/
2344 static int ixgbevf_setup_all_tx_resources(struct ixgbevf_adapter *adapter)
2345 {
2346         int i, err = 0;
2347
2348         for (i = 0; i < adapter->num_tx_queues; i++) {
2349                 err = ixgbevf_setup_tx_resources(adapter, &adapter->tx_ring[i]);
2350                 if (!err)
2351                         continue;
2352                 hw_dbg(&adapter->hw,
2353                        "Allocation for Tx Queue %u failed\n", i);
2354                 break;
2355         }
2356
2357         return err;
2358 }
2359
2360 /**
2361  * ixgbevf_setup_rx_resources - allocate Rx resources (Descriptors)
2362  * @adapter: board private structure
2363  * @rx_ring:    rx descriptor ring (for a specific queue) to setup
2364  *
2365  * Returns 0 on success, negative on failure
2366  **/
2367 int ixgbevf_setup_rx_resources(struct ixgbevf_adapter *adapter,
2368                                struct ixgbevf_ring *rx_ring)
2369 {
2370         struct pci_dev *pdev = adapter->pdev;
2371         int size;
2372
2373         size = sizeof(struct ixgbevf_rx_buffer) * rx_ring->count;
2374         rx_ring->rx_buffer_info = vzalloc(size);
2375         if (!rx_ring->rx_buffer_info)
2376                 goto alloc_failed;
2377
2378         /* Round up to nearest 4K */
2379         rx_ring->size = rx_ring->count * sizeof(union ixgbe_adv_rx_desc);
2380         rx_ring->size = ALIGN(rx_ring->size, 4096);
2381
2382         rx_ring->desc = dma_alloc_coherent(&pdev->dev, rx_ring->size,
2383                                            &rx_ring->dma, GFP_KERNEL);
2384
2385         if (!rx_ring->desc) {
2386                 hw_dbg(&adapter->hw,
2387                        "Unable to allocate memory for "
2388                        "the receive descriptor ring\n");
2389                 vfree(rx_ring->rx_buffer_info);
2390                 rx_ring->rx_buffer_info = NULL;
2391                 goto alloc_failed;
2392         }
2393
2394         rx_ring->next_to_clean = 0;
2395         rx_ring->next_to_use = 0;
2396
2397         return 0;
2398 alloc_failed:
2399         return -ENOMEM;
2400 }
2401
2402 /**
2403  * ixgbevf_setup_all_rx_resources - allocate all queues Rx resources
2404  * @adapter: board private structure
2405  *
2406  * If this function returns with an error, then it's possible one or
2407  * more of the rings is populated (while the rest are not).  It is the
2408  * callers duty to clean those orphaned rings.
2409  *
2410  * Return 0 on success, negative on failure
2411  **/
2412 static int ixgbevf_setup_all_rx_resources(struct ixgbevf_adapter *adapter)
2413 {
2414         int i, err = 0;
2415
2416         for (i = 0; i < adapter->num_rx_queues; i++) {
2417                 err = ixgbevf_setup_rx_resources(adapter, &adapter->rx_ring[i]);
2418                 if (!err)
2419                         continue;
2420                 hw_dbg(&adapter->hw,
2421                        "Allocation for Rx Queue %u failed\n", i);
2422                 break;
2423         }
2424         return err;
2425 }
2426
2427 /**
2428  * ixgbevf_free_rx_resources - Free Rx Resources
2429  * @adapter: board private structure
2430  * @rx_ring: ring to clean the resources from
2431  *
2432  * Free all receive software resources
2433  **/
2434 void ixgbevf_free_rx_resources(struct ixgbevf_adapter *adapter,
2435                                struct ixgbevf_ring *rx_ring)
2436 {
2437         struct pci_dev *pdev = adapter->pdev;
2438
2439         ixgbevf_clean_rx_ring(adapter, rx_ring);
2440
2441         vfree(rx_ring->rx_buffer_info);
2442         rx_ring->rx_buffer_info = NULL;
2443
2444         dma_free_coherent(&pdev->dev, rx_ring->size, rx_ring->desc,
2445                           rx_ring->dma);
2446
2447         rx_ring->desc = NULL;
2448 }
2449
2450 /**
2451  * ixgbevf_free_all_rx_resources - Free Rx Resources for All Queues
2452  * @adapter: board private structure
2453  *
2454  * Free all receive software resources
2455  **/
2456 static void ixgbevf_free_all_rx_resources(struct ixgbevf_adapter *adapter)
2457 {
2458         int i;
2459
2460         for (i = 0; i < adapter->num_rx_queues; i++)
2461                 if (adapter->rx_ring[i].desc)
2462                         ixgbevf_free_rx_resources(adapter,
2463                                                   &adapter->rx_ring[i]);
2464 }
2465
2466 /**
2467  * ixgbevf_open - Called when a network interface is made active
2468  * @netdev: network interface device structure
2469  *
2470  * Returns 0 on success, negative value on failure
2471  *
2472  * The open entry point is called when a network interface is made
2473  * active by the system (IFF_UP).  At this point all resources needed
2474  * for transmit and receive operations are allocated, the interrupt
2475  * handler is registered with the OS, the watchdog timer is started,
2476  * and the stack is notified that the interface is ready.
2477  **/
2478 static int ixgbevf_open(struct net_device *netdev)
2479 {
2480         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2481         struct ixgbe_hw *hw = &adapter->hw;
2482         int err;
2483
2484         /* disallow open during test */
2485         if (test_bit(__IXGBEVF_TESTING, &adapter->state))
2486                 return -EBUSY;
2487
2488         if (hw->adapter_stopped) {
2489                 ixgbevf_reset(adapter);
2490                 /* if adapter is still stopped then PF isn't up and
2491                  * the vf can't start. */
2492                 if (hw->adapter_stopped) {
2493                         err = IXGBE_ERR_MBX;
2494                         pr_err("Unable to start - perhaps the PF Driver isn't "
2495                                "up yet\n");
2496                         goto err_setup_reset;
2497                 }
2498         }
2499
2500         /* allocate transmit descriptors */
2501         err = ixgbevf_setup_all_tx_resources(adapter);
2502         if (err)
2503                 goto err_setup_tx;
2504
2505         /* allocate receive descriptors */
2506         err = ixgbevf_setup_all_rx_resources(adapter);
2507         if (err)
2508                 goto err_setup_rx;
2509
2510         ixgbevf_configure(adapter);
2511
2512         /*
2513          * Map the Tx/Rx rings to the vectors we were allotted.
2514          * if request_irq will be called in this function map_rings
2515          * must be called *before* up_complete
2516          */
2517         ixgbevf_map_rings_to_vectors(adapter);
2518
2519         ixgbevf_up_complete(adapter);
2520
2521         /* clear any pending interrupts, may auto mask */
2522         IXGBE_READ_REG(hw, IXGBE_VTEICR);
2523         err = ixgbevf_request_irq(adapter);
2524         if (err)
2525                 goto err_req_irq;
2526
2527         ixgbevf_irq_enable(adapter, true, true);
2528
2529         return 0;
2530
2531 err_req_irq:
2532         ixgbevf_down(adapter);
2533         ixgbevf_free_irq(adapter);
2534 err_setup_rx:
2535         ixgbevf_free_all_rx_resources(adapter);
2536 err_setup_tx:
2537         ixgbevf_free_all_tx_resources(adapter);
2538         ixgbevf_reset(adapter);
2539
2540 err_setup_reset:
2541
2542         return err;
2543 }
2544
2545 /**
2546  * ixgbevf_close - Disables a network interface
2547  * @netdev: network interface device structure
2548  *
2549  * Returns 0, this is not allowed to fail
2550  *
2551  * The close entry point is called when an interface is de-activated
2552  * by the OS.  The hardware is still under the drivers control, but
2553  * needs to be disabled.  A global MAC reset is issued to stop the
2554  * hardware, and all transmit and receive resources are freed.
2555  **/
2556 static int ixgbevf_close(struct net_device *netdev)
2557 {
2558         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2559
2560         ixgbevf_down(adapter);
2561         ixgbevf_free_irq(adapter);
2562
2563         ixgbevf_free_all_tx_resources(adapter);
2564         ixgbevf_free_all_rx_resources(adapter);
2565
2566         return 0;
2567 }
2568
2569 static int ixgbevf_tso(struct ixgbevf_adapter *adapter,
2570                        struct ixgbevf_ring *tx_ring,
2571                        struct sk_buff *skb, u32 tx_flags, u8 *hdr_len)
2572 {
2573         struct ixgbe_adv_tx_context_desc *context_desc;
2574         unsigned int i;
2575         int err;
2576         struct ixgbevf_tx_buffer *tx_buffer_info;
2577         u32 vlan_macip_lens = 0, type_tucmd_mlhl;
2578         u32 mss_l4len_idx, l4len;
2579
2580         if (skb_is_gso(skb)) {
2581                 if (skb_header_cloned(skb)) {
2582                         err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2583                         if (err)
2584                                 return err;
2585                 }
2586                 l4len = tcp_hdrlen(skb);
2587                 *hdr_len += l4len;
2588
2589                 if (skb->protocol == htons(ETH_P_IP)) {
2590                         struct iphdr *iph = ip_hdr(skb);
2591                         iph->tot_len = 0;
2592                         iph->check = 0;
2593                         tcp_hdr(skb)->check = ~csum_tcpudp_magic(iph->saddr,
2594                                                                  iph->daddr, 0,
2595                                                                  IPPROTO_TCP,
2596                                                                  0);
2597                         adapter->hw_tso_ctxt++;
2598                 } else if (skb_is_gso_v6(skb)) {
2599                         ipv6_hdr(skb)->payload_len = 0;
2600                         tcp_hdr(skb)->check =
2601                             ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
2602                                              &ipv6_hdr(skb)->daddr,
2603                                              0, IPPROTO_TCP, 0);
2604                         adapter->hw_tso6_ctxt++;
2605                 }
2606
2607                 i = tx_ring->next_to_use;
2608
2609                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2610                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
2611
2612                 /* VLAN MACLEN IPLEN */
2613                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2614                         vlan_macip_lens |=
2615                                 (tx_flags & IXGBE_TX_FLAGS_VLAN_MASK);
2616                 vlan_macip_lens |= ((skb_network_offset(skb)) <<
2617                                     IXGBE_ADVTXD_MACLEN_SHIFT);
2618                 *hdr_len += skb_network_offset(skb);
2619                 vlan_macip_lens |=
2620                         (skb_transport_header(skb) - skb_network_header(skb));
2621                 *hdr_len +=
2622                         (skb_transport_header(skb) - skb_network_header(skb));
2623                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2624                 context_desc->seqnum_seed = 0;
2625
2626                 /* ADV DTYP TUCMD MKRLOC/ISCSIHEDLEN */
2627                 type_tucmd_mlhl = (IXGBE_TXD_CMD_DEXT |
2628                                     IXGBE_ADVTXD_DTYP_CTXT);
2629
2630                 if (skb->protocol == htons(ETH_P_IP))
2631                         type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
2632                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_L4T_TCP;
2633                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
2634
2635                 /* MSS L4LEN IDX */
2636                 mss_l4len_idx =
2637                         (skb_shinfo(skb)->gso_size << IXGBE_ADVTXD_MSS_SHIFT);
2638                 mss_l4len_idx |= (l4len << IXGBE_ADVTXD_L4LEN_SHIFT);
2639                 /* use index 1 for TSO */
2640                 mss_l4len_idx |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2641                 context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
2642
2643                 tx_buffer_info->time_stamp = jiffies;
2644                 tx_buffer_info->next_to_watch = i;
2645
2646                 i++;
2647                 if (i == tx_ring->count)
2648                         i = 0;
2649                 tx_ring->next_to_use = i;
2650
2651                 return true;
2652         }
2653
2654         return false;
2655 }
2656
2657 static bool ixgbevf_tx_csum(struct ixgbevf_adapter *adapter,
2658                             struct ixgbevf_ring *tx_ring,
2659                             struct sk_buff *skb, u32 tx_flags)
2660 {
2661         struct ixgbe_adv_tx_context_desc *context_desc;
2662         unsigned int i;
2663         struct ixgbevf_tx_buffer *tx_buffer_info;
2664         u32 vlan_macip_lens = 0, type_tucmd_mlhl = 0;
2665
2666         if (skb->ip_summed == CHECKSUM_PARTIAL ||
2667             (tx_flags & IXGBE_TX_FLAGS_VLAN)) {
2668                 i = tx_ring->next_to_use;
2669                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2670                 context_desc = IXGBE_TX_CTXTDESC_ADV(*tx_ring, i);
2671
2672                 if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2673                         vlan_macip_lens |= (tx_flags &
2674                                             IXGBE_TX_FLAGS_VLAN_MASK);
2675                 vlan_macip_lens |= (skb_network_offset(skb) <<
2676                                     IXGBE_ADVTXD_MACLEN_SHIFT);
2677                 if (skb->ip_summed == CHECKSUM_PARTIAL)
2678                         vlan_macip_lens |= (skb_transport_header(skb) -
2679                                             skb_network_header(skb));
2680
2681                 context_desc->vlan_macip_lens = cpu_to_le32(vlan_macip_lens);
2682                 context_desc->seqnum_seed = 0;
2683
2684                 type_tucmd_mlhl |= (IXGBE_TXD_CMD_DEXT |
2685                                     IXGBE_ADVTXD_DTYP_CTXT);
2686
2687                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2688                         switch (skb->protocol) {
2689                         case __constant_htons(ETH_P_IP):
2690                                 type_tucmd_mlhl |= IXGBE_ADVTXD_TUCMD_IPV4;
2691                                 if (ip_hdr(skb)->protocol == IPPROTO_TCP)
2692                                         type_tucmd_mlhl |=
2693                                             IXGBE_ADVTXD_TUCMD_L4T_TCP;
2694                                 break;
2695                         case __constant_htons(ETH_P_IPV6):
2696                                 /* XXX what about other V6 headers?? */
2697                                 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
2698                                         type_tucmd_mlhl |=
2699                                                 IXGBE_ADVTXD_TUCMD_L4T_TCP;
2700                                 break;
2701                         default:
2702                                 if (unlikely(net_ratelimit())) {
2703                                         pr_warn("partial checksum but "
2704                                                 "proto=%x!\n", skb->protocol);
2705                                 }
2706                                 break;
2707                         }
2708                 }
2709
2710                 context_desc->type_tucmd_mlhl = cpu_to_le32(type_tucmd_mlhl);
2711                 /* use index zero for tx checksum offload */
2712                 context_desc->mss_l4len_idx = 0;
2713
2714                 tx_buffer_info->time_stamp = jiffies;
2715                 tx_buffer_info->next_to_watch = i;
2716
2717                 adapter->hw_csum_tx_good++;
2718                 i++;
2719                 if (i == tx_ring->count)
2720                         i = 0;
2721                 tx_ring->next_to_use = i;
2722
2723                 return true;
2724         }
2725
2726         return false;
2727 }
2728
2729 static int ixgbevf_tx_map(struct ixgbevf_adapter *adapter,
2730                           struct ixgbevf_ring *tx_ring,
2731                           struct sk_buff *skb, u32 tx_flags,
2732                           unsigned int first)
2733 {
2734         struct pci_dev *pdev = adapter->pdev;
2735         struct ixgbevf_tx_buffer *tx_buffer_info;
2736         unsigned int len;
2737         unsigned int total = skb->len;
2738         unsigned int offset = 0, size;
2739         int count = 0;
2740         unsigned int nr_frags = skb_shinfo(skb)->nr_frags;
2741         unsigned int f;
2742         int i;
2743
2744         i = tx_ring->next_to_use;
2745
2746         len = min(skb_headlen(skb), total);
2747         while (len) {
2748                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2749                 size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2750
2751                 tx_buffer_info->length = size;
2752                 tx_buffer_info->mapped_as_page = false;
2753                 tx_buffer_info->dma = dma_map_single(&adapter->pdev->dev,
2754                                                      skb->data + offset,
2755                                                      size, DMA_TO_DEVICE);
2756                 if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
2757                         goto dma_error;
2758                 tx_buffer_info->time_stamp = jiffies;
2759                 tx_buffer_info->next_to_watch = i;
2760
2761                 len -= size;
2762                 total -= size;
2763                 offset += size;
2764                 count++;
2765                 i++;
2766                 if (i == tx_ring->count)
2767                         i = 0;
2768         }
2769
2770         for (f = 0; f < nr_frags; f++) {
2771                 const struct skb_frag_struct *frag;
2772
2773                 frag = &skb_shinfo(skb)->frags[f];
2774                 len = min((unsigned int)skb_frag_size(frag), total);
2775                 offset = 0;
2776
2777                 while (len) {
2778                         tx_buffer_info = &tx_ring->tx_buffer_info[i];
2779                         size = min(len, (unsigned int)IXGBE_MAX_DATA_PER_TXD);
2780
2781                         tx_buffer_info->length = size;
2782                         tx_buffer_info->dma =
2783                                 skb_frag_dma_map(&adapter->pdev->dev, frag,
2784                                                  offset, size, DMA_TO_DEVICE);
2785                         tx_buffer_info->mapped_as_page = true;
2786                         if (dma_mapping_error(&pdev->dev, tx_buffer_info->dma))
2787                                 goto dma_error;
2788                         tx_buffer_info->time_stamp = jiffies;
2789                         tx_buffer_info->next_to_watch = i;
2790
2791                         len -= size;
2792                         total -= size;
2793                         offset += size;
2794                         count++;
2795                         i++;
2796                         if (i == tx_ring->count)
2797                                 i = 0;
2798                 }
2799                 if (total == 0)
2800                         break;
2801         }
2802
2803         if (i == 0)
2804                 i = tx_ring->count - 1;
2805         else
2806                 i = i - 1;
2807         tx_ring->tx_buffer_info[i].skb = skb;
2808         tx_ring->tx_buffer_info[first].next_to_watch = i;
2809
2810         return count;
2811
2812 dma_error:
2813         dev_err(&pdev->dev, "TX DMA map failed\n");
2814
2815         /* clear timestamp and dma mappings for failed tx_buffer_info map */
2816         tx_buffer_info->dma = 0;
2817         tx_buffer_info->time_stamp = 0;
2818         tx_buffer_info->next_to_watch = 0;
2819         count--;
2820
2821         /* clear timestamp and dma mappings for remaining portion of packet */
2822         while (count >= 0) {
2823                 count--;
2824                 i--;
2825                 if (i < 0)
2826                         i += tx_ring->count;
2827                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2828                 ixgbevf_unmap_and_free_tx_resource(adapter, tx_buffer_info);
2829         }
2830
2831         return count;
2832 }
2833
2834 static void ixgbevf_tx_queue(struct ixgbevf_adapter *adapter,
2835                              struct ixgbevf_ring *tx_ring, int tx_flags,
2836                              int count, u32 paylen, u8 hdr_len)
2837 {
2838         union ixgbe_adv_tx_desc *tx_desc = NULL;
2839         struct ixgbevf_tx_buffer *tx_buffer_info;
2840         u32 olinfo_status = 0, cmd_type_len = 0;
2841         unsigned int i;
2842
2843         u32 txd_cmd = IXGBE_TXD_CMD_EOP | IXGBE_TXD_CMD_RS | IXGBE_TXD_CMD_IFCS;
2844
2845         cmd_type_len |= IXGBE_ADVTXD_DTYP_DATA;
2846
2847         cmd_type_len |= IXGBE_ADVTXD_DCMD_IFCS | IXGBE_ADVTXD_DCMD_DEXT;
2848
2849         if (tx_flags & IXGBE_TX_FLAGS_VLAN)
2850                 cmd_type_len |= IXGBE_ADVTXD_DCMD_VLE;
2851
2852         if (tx_flags & IXGBE_TX_FLAGS_TSO) {
2853                 cmd_type_len |= IXGBE_ADVTXD_DCMD_TSE;
2854
2855                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
2856                         IXGBE_ADVTXD_POPTS_SHIFT;
2857
2858                 /* use index 1 context for tso */
2859                 olinfo_status |= (1 << IXGBE_ADVTXD_IDX_SHIFT);
2860                 if (tx_flags & IXGBE_TX_FLAGS_IPV4)
2861                         olinfo_status |= IXGBE_TXD_POPTS_IXSM <<
2862                                 IXGBE_ADVTXD_POPTS_SHIFT;
2863
2864         } else if (tx_flags & IXGBE_TX_FLAGS_CSUM)
2865                 olinfo_status |= IXGBE_TXD_POPTS_TXSM <<
2866                         IXGBE_ADVTXD_POPTS_SHIFT;
2867
2868         olinfo_status |= ((paylen - hdr_len) << IXGBE_ADVTXD_PAYLEN_SHIFT);
2869
2870         i = tx_ring->next_to_use;
2871         while (count--) {
2872                 tx_buffer_info = &tx_ring->tx_buffer_info[i];
2873                 tx_desc = IXGBE_TX_DESC_ADV(*tx_ring, i);
2874                 tx_desc->read.buffer_addr = cpu_to_le64(tx_buffer_info->dma);
2875                 tx_desc->read.cmd_type_len =
2876                         cpu_to_le32(cmd_type_len | tx_buffer_info->length);
2877                 tx_desc->read.olinfo_status = cpu_to_le32(olinfo_status);
2878                 i++;
2879                 if (i == tx_ring->count)
2880                         i = 0;
2881         }
2882
2883         tx_desc->read.cmd_type_len |= cpu_to_le32(txd_cmd);
2884
2885         /*
2886          * Force memory writes to complete before letting h/w
2887          * know there are new descriptors to fetch.  (Only
2888          * applicable for weak-ordered memory model archs,
2889          * such as IA-64).
2890          */
2891         wmb();
2892
2893         tx_ring->next_to_use = i;
2894         writel(i, adapter->hw.hw_addr + tx_ring->tail);
2895 }
2896
2897 static int __ixgbevf_maybe_stop_tx(struct net_device *netdev,
2898                                    struct ixgbevf_ring *tx_ring, int size)
2899 {
2900         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2901
2902         netif_stop_subqueue(netdev, tx_ring->queue_index);
2903         /* Herbert's original patch had:
2904          *  smp_mb__after_netif_stop_queue();
2905          * but since that doesn't exist yet, just open code it. */
2906         smp_mb();
2907
2908         /* We need to check again in a case another CPU has just
2909          * made room available. */
2910         if (likely(IXGBE_DESC_UNUSED(tx_ring) < size))
2911                 return -EBUSY;
2912
2913         /* A reprieve! - use start_queue because it doesn't call schedule */
2914         netif_start_subqueue(netdev, tx_ring->queue_index);
2915         ++adapter->restart_queue;
2916         return 0;
2917 }
2918
2919 static int ixgbevf_maybe_stop_tx(struct net_device *netdev,
2920                                  struct ixgbevf_ring *tx_ring, int size)
2921 {
2922         if (likely(IXGBE_DESC_UNUSED(tx_ring) >= size))
2923                 return 0;
2924         return __ixgbevf_maybe_stop_tx(netdev, tx_ring, size);
2925 }
2926
2927 static int ixgbevf_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
2928 {
2929         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2930         struct ixgbevf_ring *tx_ring;
2931         unsigned int first;
2932         unsigned int tx_flags = 0;
2933         u8 hdr_len = 0;
2934         int r_idx = 0, tso;
2935         int count = 0;
2936
2937         unsigned int f;
2938
2939         tx_ring = &adapter->tx_ring[r_idx];
2940
2941         if (vlan_tx_tag_present(skb)) {
2942                 tx_flags |= vlan_tx_tag_get(skb);
2943                 tx_flags <<= IXGBE_TX_FLAGS_VLAN_SHIFT;
2944                 tx_flags |= IXGBE_TX_FLAGS_VLAN;
2945         }
2946
2947         /* four things can cause us to need a context descriptor */
2948         if (skb_is_gso(skb) ||
2949             (skb->ip_summed == CHECKSUM_PARTIAL) ||
2950             (tx_flags & IXGBE_TX_FLAGS_VLAN))
2951                 count++;
2952
2953         count += TXD_USE_COUNT(skb_headlen(skb));
2954         for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
2955                 count += TXD_USE_COUNT(skb_frag_size(&skb_shinfo(skb)->frags[f]));
2956
2957         if (ixgbevf_maybe_stop_tx(netdev, tx_ring, count)) {
2958                 adapter->tx_busy++;
2959                 return NETDEV_TX_BUSY;
2960         }
2961
2962         first = tx_ring->next_to_use;
2963
2964         if (skb->protocol == htons(ETH_P_IP))
2965                 tx_flags |= IXGBE_TX_FLAGS_IPV4;
2966         tso = ixgbevf_tso(adapter, tx_ring, skb, tx_flags, &hdr_len);
2967         if (tso < 0) {
2968                 dev_kfree_skb_any(skb);
2969                 return NETDEV_TX_OK;
2970         }
2971
2972         if (tso)
2973                 tx_flags |= IXGBE_TX_FLAGS_TSO;
2974         else if (ixgbevf_tx_csum(adapter, tx_ring, skb, tx_flags) &&
2975                  (skb->ip_summed == CHECKSUM_PARTIAL))
2976                 tx_flags |= IXGBE_TX_FLAGS_CSUM;
2977
2978         ixgbevf_tx_queue(adapter, tx_ring, tx_flags,
2979                          ixgbevf_tx_map(adapter, tx_ring, skb, tx_flags, first),
2980                          skb->len, hdr_len);
2981
2982         ixgbevf_maybe_stop_tx(netdev, tx_ring, DESC_NEEDED);
2983
2984         return NETDEV_TX_OK;
2985 }
2986
2987 /**
2988  * ixgbevf_set_mac - Change the Ethernet Address of the NIC
2989  * @netdev: network interface device structure
2990  * @p: pointer to an address structure
2991  *
2992  * Returns 0 on success, negative on failure
2993  **/
2994 static int ixgbevf_set_mac(struct net_device *netdev, void *p)
2995 {
2996         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
2997         struct ixgbe_hw *hw = &adapter->hw;
2998         struct sockaddr *addr = p;
2999
3000         if (!is_valid_ether_addr(addr->sa_data))
3001                 return -EADDRNOTAVAIL;
3002
3003         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
3004         memcpy(hw->mac.addr, addr->sa_data, netdev->addr_len);
3005
3006         if (hw->mac.ops.set_rar)
3007                 hw->mac.ops.set_rar(hw, 0, hw->mac.addr, 0);
3008
3009         return 0;
3010 }
3011
3012 /**
3013  * ixgbevf_change_mtu - Change the Maximum Transfer Unit
3014  * @netdev: network interface device structure
3015  * @new_mtu: new value for maximum frame size
3016  *
3017  * Returns 0 on success, negative on failure
3018  **/
3019 static int ixgbevf_change_mtu(struct net_device *netdev, int new_mtu)
3020 {
3021         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3022         struct ixgbe_hw *hw = &adapter->hw;
3023         int max_frame = new_mtu + ETH_HLEN + ETH_FCS_LEN;
3024         int max_possible_frame = MAXIMUM_ETHERNET_VLAN_SIZE;
3025         u32 msg[2];
3026
3027         if (adapter->hw.mac.type == ixgbe_mac_X540_vf)
3028                 max_possible_frame = IXGBE_MAX_JUMBO_FRAME_SIZE;
3029
3030         /* MTU < 68 is an error and causes problems on some kernels */
3031         if ((new_mtu < 68) || (max_frame > max_possible_frame))
3032                 return -EINVAL;
3033
3034         hw_dbg(&adapter->hw, "changing MTU from %d to %d\n",
3035                netdev->mtu, new_mtu);
3036         /* must set new MTU before calling down or up */
3037         netdev->mtu = new_mtu;
3038
3039         if (!netif_running(netdev)) {
3040                 msg[0] = IXGBE_VF_SET_LPE;
3041                 msg[1] = max_frame;
3042                 hw->mbx.ops.write_posted(hw, msg, 2);
3043         }
3044
3045         if (netif_running(netdev))
3046                 ixgbevf_reinit_locked(adapter);
3047
3048         return 0;
3049 }
3050
3051 static void ixgbevf_shutdown(struct pci_dev *pdev)
3052 {
3053         struct net_device *netdev = pci_get_drvdata(pdev);
3054         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3055
3056         netif_device_detach(netdev);
3057
3058         if (netif_running(netdev)) {
3059                 ixgbevf_down(adapter);
3060                 ixgbevf_free_irq(adapter);
3061                 ixgbevf_free_all_tx_resources(adapter);
3062                 ixgbevf_free_all_rx_resources(adapter);
3063         }
3064
3065         pci_save_state(pdev);
3066
3067         pci_disable_device(pdev);
3068 }
3069
3070 static struct rtnl_link_stats64 *ixgbevf_get_stats(struct net_device *netdev,
3071                                                 struct rtnl_link_stats64 *stats)
3072 {
3073         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3074         unsigned int start;
3075         u64 bytes, packets;
3076         const struct ixgbevf_ring *ring;
3077         int i;
3078
3079         ixgbevf_update_stats(adapter);
3080
3081         stats->multicast = adapter->stats.vfmprc - adapter->stats.base_vfmprc;
3082
3083         for (i = 0; i < adapter->num_rx_queues; i++) {
3084                 ring = &adapter->rx_ring[i];
3085                 do {
3086                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3087                         bytes = ring->total_bytes;
3088                         packets = ring->total_packets;
3089                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3090                 stats->rx_bytes += bytes;
3091                 stats->rx_packets += packets;
3092         }
3093
3094         for (i = 0; i < adapter->num_tx_queues; i++) {
3095                 ring = &adapter->tx_ring[i];
3096                 do {
3097                         start = u64_stats_fetch_begin_bh(&ring->syncp);
3098                         bytes = ring->total_bytes;
3099                         packets = ring->total_packets;
3100                 } while (u64_stats_fetch_retry_bh(&ring->syncp, start));
3101                 stats->tx_bytes += bytes;
3102                 stats->tx_packets += packets;
3103         }
3104
3105         return stats;
3106 }
3107
3108 static const struct net_device_ops ixgbe_netdev_ops = {
3109         .ndo_open               = ixgbevf_open,
3110         .ndo_stop               = ixgbevf_close,
3111         .ndo_start_xmit         = ixgbevf_xmit_frame,
3112         .ndo_set_rx_mode        = ixgbevf_set_rx_mode,
3113         .ndo_get_stats64        = ixgbevf_get_stats,
3114         .ndo_validate_addr      = eth_validate_addr,
3115         .ndo_set_mac_address    = ixgbevf_set_mac,
3116         .ndo_change_mtu         = ixgbevf_change_mtu,
3117         .ndo_tx_timeout         = ixgbevf_tx_timeout,
3118         .ndo_vlan_rx_add_vid    = ixgbevf_vlan_rx_add_vid,
3119         .ndo_vlan_rx_kill_vid   = ixgbevf_vlan_rx_kill_vid,
3120 };
3121
3122 static void ixgbevf_assign_netdev_ops(struct net_device *dev)
3123 {
3124         dev->netdev_ops = &ixgbe_netdev_ops;
3125         ixgbevf_set_ethtool_ops(dev);
3126         dev->watchdog_timeo = 5 * HZ;
3127 }
3128
3129 /**
3130  * ixgbevf_probe - Device Initialization Routine
3131  * @pdev: PCI device information struct
3132  * @ent: entry in ixgbevf_pci_tbl
3133  *
3134  * Returns 0 on success, negative on failure
3135  *
3136  * ixgbevf_probe initializes an adapter identified by a pci_dev structure.
3137  * The OS initialization, configuring of the adapter private structure,
3138  * and a hardware reset occur.
3139  **/
3140 static int __devinit ixgbevf_probe(struct pci_dev *pdev,
3141                                    const struct pci_device_id *ent)
3142 {
3143         struct net_device *netdev;
3144         struct ixgbevf_adapter *adapter = NULL;
3145         struct ixgbe_hw *hw = NULL;
3146         const struct ixgbevf_info *ii = ixgbevf_info_tbl[ent->driver_data];
3147         static int cards_found;
3148         int err, pci_using_dac;
3149
3150         err = pci_enable_device(pdev);
3151         if (err)
3152                 return err;
3153
3154         if (!dma_set_mask(&pdev->dev, DMA_BIT_MASK(64)) &&
3155             !dma_set_coherent_mask(&pdev->dev, DMA_BIT_MASK(64))) {
3156                 pci_using_dac = 1;
3157         } else {
3158                 err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
3159                 if (err) {
3160                         err = dma_set_coherent_mask(&pdev->dev,
3161                                                     DMA_BIT_MASK(32));
3162                         if (err) {
3163                                 dev_err(&pdev->dev, "No usable DMA "
3164                                         "configuration, aborting\n");
3165                                 goto err_dma;
3166                         }
3167                 }
3168                 pci_using_dac = 0;
3169         }
3170
3171         err = pci_request_regions(pdev, ixgbevf_driver_name);
3172         if (err) {
3173                 dev_err(&pdev->dev, "pci_request_regions failed 0x%x\n", err);
3174                 goto err_pci_reg;
3175         }
3176
3177         pci_set_master(pdev);
3178
3179         netdev = alloc_etherdev_mq(sizeof(struct ixgbevf_adapter),
3180                                    MAX_TX_QUEUES);
3181         if (!netdev) {
3182                 err = -ENOMEM;
3183                 goto err_alloc_etherdev;
3184         }
3185
3186         SET_NETDEV_DEV(netdev, &pdev->dev);
3187
3188         pci_set_drvdata(pdev, netdev);
3189         adapter = netdev_priv(netdev);
3190
3191         adapter->netdev = netdev;
3192         adapter->pdev = pdev;
3193         hw = &adapter->hw;
3194         hw->back = adapter;
3195         adapter->msg_enable = netif_msg_init(debug, DEFAULT_MSG_ENABLE);
3196
3197         /*
3198          * call save state here in standalone driver because it relies on
3199          * adapter struct to exist, and needs to call netdev_priv
3200          */
3201         pci_save_state(pdev);
3202
3203         hw->hw_addr = ioremap(pci_resource_start(pdev, 0),
3204                               pci_resource_len(pdev, 0));
3205         if (!hw->hw_addr) {
3206                 err = -EIO;
3207                 goto err_ioremap;
3208         }
3209
3210         ixgbevf_assign_netdev_ops(netdev);
3211
3212         adapter->bd_number = cards_found;
3213
3214         /* Setup hw api */
3215         memcpy(&hw->mac.ops, ii->mac_ops, sizeof(hw->mac.ops));
3216         hw->mac.type  = ii->mac;
3217
3218         memcpy(&hw->mbx.ops, &ixgbevf_mbx_ops,
3219                sizeof(struct ixgbe_mbx_operations));
3220
3221         /* setup the private structure */
3222         err = ixgbevf_sw_init(adapter);
3223         if (err)
3224                 goto err_sw_init;
3225
3226         /* The HW MAC address was set and/or determined in sw_init */
3227         memcpy(netdev->perm_addr, adapter->hw.mac.addr, netdev->addr_len);
3228
3229         if (!is_valid_ether_addr(netdev->dev_addr)) {
3230                 pr_err("invalid MAC address\n");
3231                 err = -EIO;
3232                 goto err_sw_init;
3233         }
3234
3235         netdev->hw_features = NETIF_F_SG |
3236                            NETIF_F_IP_CSUM |
3237                            NETIF_F_IPV6_CSUM |
3238                            NETIF_F_TSO |
3239                            NETIF_F_TSO6 |
3240                            NETIF_F_RXCSUM;
3241
3242         netdev->features = netdev->hw_features |
3243                            NETIF_F_HW_VLAN_TX |
3244                            NETIF_F_HW_VLAN_RX |
3245                            NETIF_F_HW_VLAN_FILTER;
3246
3247         netdev->vlan_features |= NETIF_F_TSO;
3248         netdev->vlan_features |= NETIF_F_TSO6;
3249         netdev->vlan_features |= NETIF_F_IP_CSUM;
3250         netdev->vlan_features |= NETIF_F_IPV6_CSUM;
3251         netdev->vlan_features |= NETIF_F_SG;
3252
3253         if (pci_using_dac)
3254                 netdev->features |= NETIF_F_HIGHDMA;
3255
3256         netdev->priv_flags |= IFF_UNICAST_FLT;
3257
3258         init_timer(&adapter->watchdog_timer);
3259         adapter->watchdog_timer.function = ixgbevf_watchdog;
3260         adapter->watchdog_timer.data = (unsigned long)adapter;
3261
3262         INIT_WORK(&adapter->reset_task, ixgbevf_reset_task);
3263         INIT_WORK(&adapter->watchdog_task, ixgbevf_watchdog_task);
3264
3265         err = ixgbevf_init_interrupt_scheme(adapter);
3266         if (err)
3267                 goto err_sw_init;
3268
3269         /* pick up the PCI bus settings for reporting later */
3270         if (hw->mac.ops.get_bus_info)
3271                 hw->mac.ops.get_bus_info(hw);
3272
3273         strcpy(netdev->name, "eth%d");
3274
3275         err = register_netdev(netdev);
3276         if (err)
3277                 goto err_register;
3278
3279         netif_carrier_off(netdev);
3280
3281         ixgbevf_init_last_counter_stats(adapter);
3282
3283         /* print the MAC address */
3284         hw_dbg(hw, "%pM\n", netdev->dev_addr);
3285
3286         hw_dbg(hw, "MAC: %d\n", hw->mac.type);
3287
3288         hw_dbg(hw, "Intel(R) 82599 Virtual Function\n");
3289         cards_found++;
3290         return 0;
3291
3292 err_register:
3293 err_sw_init:
3294         ixgbevf_reset_interrupt_capability(adapter);
3295         iounmap(hw->hw_addr);
3296 err_ioremap:
3297         free_netdev(netdev);
3298 err_alloc_etherdev:
3299         pci_release_regions(pdev);
3300 err_pci_reg:
3301 err_dma:
3302         pci_disable_device(pdev);
3303         return err;
3304 }
3305
3306 /**
3307  * ixgbevf_remove - Device Removal Routine
3308  * @pdev: PCI device information struct
3309  *
3310  * ixgbevf_remove is called by the PCI subsystem to alert the driver
3311  * that it should release a PCI device.  The could be caused by a
3312  * Hot-Plug event, or because the driver is going to be removed from
3313  * memory.
3314  **/
3315 static void __devexit ixgbevf_remove(struct pci_dev *pdev)
3316 {
3317         struct net_device *netdev = pci_get_drvdata(pdev);
3318         struct ixgbevf_adapter *adapter = netdev_priv(netdev);
3319
3320         set_bit(__IXGBEVF_DOWN, &adapter->state);
3321
3322         del_timer_sync(&adapter->watchdog_timer);
3323
3324         cancel_work_sync(&adapter->reset_task);
3325         cancel_work_sync(&adapter->watchdog_task);
3326
3327         if (netdev->reg_state == NETREG_REGISTERED)
3328                 unregister_netdev(netdev);
3329
3330         ixgbevf_reset_interrupt_capability(adapter);
3331
3332         iounmap(adapter->hw.hw_addr);
3333         pci_release_regions(pdev);
3334
3335         hw_dbg(&adapter->hw, "Remove complete\n");
3336
3337         kfree(adapter->tx_ring);
3338         kfree(adapter->rx_ring);
3339
3340         free_netdev(netdev);
3341
3342         pci_disable_device(pdev);
3343 }
3344
3345 static struct pci_driver ixgbevf_driver = {
3346         .name     = ixgbevf_driver_name,
3347         .id_table = ixgbevf_pci_tbl,
3348         .probe    = ixgbevf_probe,
3349         .remove   = __devexit_p(ixgbevf_remove),
3350         .shutdown = ixgbevf_shutdown,
3351 };
3352
3353 /**
3354  * ixgbevf_init_module - Driver Registration Routine
3355  *
3356  * ixgbevf_init_module is the first routine called when the driver is
3357  * loaded. All it does is register with the PCI subsystem.
3358  **/
3359 static int __init ixgbevf_init_module(void)
3360 {
3361         int ret;
3362         pr_info("%s - version %s\n", ixgbevf_driver_string,
3363                 ixgbevf_driver_version);
3364
3365         pr_info("%s\n", ixgbevf_copyright);
3366
3367         ret = pci_register_driver(&ixgbevf_driver);
3368         return ret;
3369 }
3370
3371 module_init(ixgbevf_init_module);
3372
3373 /**
3374  * ixgbevf_exit_module - Driver Exit Cleanup Routine
3375  *
3376  * ixgbevf_exit_module is called just before the driver is removed
3377  * from memory.
3378  **/
3379 static void __exit ixgbevf_exit_module(void)
3380 {
3381         pci_unregister_driver(&ixgbevf_driver);
3382 }
3383
3384 #ifdef DEBUG
3385 /**
3386  * ixgbevf_get_hw_dev_name - return device name string
3387  * used by hardware layer to print debugging information
3388  **/
3389 char *ixgbevf_get_hw_dev_name(struct ixgbe_hw *hw)
3390 {
3391         struct ixgbevf_adapter *adapter = hw->back;
3392         return adapter->netdev->name;
3393 }
3394
3395 #endif
3396 module_exit(ixgbevf_exit_module);
3397
3398 /* ixgbevf_main.c */