drivers/net/vmxnet3/vmxnet3_drv.c

   1 /*
   2  * Linux driver for VMware's vmxnet3 ethernet NIC.
   3  *
   4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; version 2 of the License and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  * NON INFRINGEMENT. See the GNU General Public License for more
  14  * details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * The full GNU General Public License is included in this distribution in
  21  * the file called "COPYING".
  22  *
  23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
  24  *
  25  */
  26
  27 #include <net/ip6_checksum.h>
  28
  29 #include "vmxnet3_int.h"
  30
  31 char vmxnet3_driver_name[] = "vmxnet3";
  32 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
  33
  34 /*
  35  * PCI Device ID Table
  36  * Last entry must be all 0s
  37  */
  38 static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = {
  39         {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
  40         {0}
  41 };
  42
  43 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
  44
  45 static atomic_t devices_found;
  46
  47 #define VMXNET3_MAX_DEVICES 10
  48 static int enable_mq = 1;
  49 static int irq_share_mode;
  50
  51 /*
  52  *    Enable/Disable the given intr
  53  */
  54 static void
  55 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  56 {
  57         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
  58 }
  59
  60
  61 static void
  62 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  63 {
  64         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
  65 }
  66
  67
  68 /*
  69  *    Enable/Disable all intrs used by the device
  70  */
  71 static void
  72 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
  73 {
  74         int i;
  75
  76         for (i = 0; i < adapter->intr.num_intrs; i++)
  77                 vmxnet3_enable_intr(adapter, i);
  78         adapter->shared->devRead.intrConf.intrCtrl &=
  79                                         cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
  80 }
  81
  82
  83 static void
  84 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
  85 {
  86         int i;
  87
  88         adapter->shared->devRead.intrConf.intrCtrl |=
  89                                         cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
  90         for (i = 0; i < adapter->intr.num_intrs; i++)
  91                 vmxnet3_disable_intr(adapter, i);
  92 }
  93
  94
  95 static void
  96 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
  97 {
  98         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
  99 }
 100
 101
 102 static bool
 103 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 104 {
 105         return tq->stopped;
 106 }
 107
 108
 109 static void
 110 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 111 {
 112         tq->stopped = false;
 113         netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
 114 }
 115
 116
 117 static void
 118 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 119 {
 120         tq->stopped = false;
 121         netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
 122 }
 123
 124
 125 static void
 126 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 127 {
 128         tq->stopped = true;
 129         tq->num_stop++;
 130         netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
 131 }
 132
 133
 134 /*
 135  * Check the link state. This may start or stop the tx queue.
 136  */
 137 static void
 138 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
 139 {
 140         u32 ret;
 141         int i;
 142
 143         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
 144         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
 145         adapter->link_speed = ret >> 16;
 146         if (ret & 1) { /* Link is up. */
 147                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
 148                        adapter->netdev->name, adapter->link_speed);
 149                 if (!netif_carrier_ok(adapter->netdev))
 150                         netif_carrier_on(adapter->netdev);
 151
 152                 if (affectTxQueue) {
 153                         for (i = 0; i < adapter->num_tx_queues; i++)
 154                                 vmxnet3_tq_start(&adapter->tx_queue[i],
 155                                                  adapter);
 156                 }
 157         } else {
 158                 printk(KERN_INFO "%s: NIC Link is Down\n",
 159                        adapter->netdev->name);
 160                 if (netif_carrier_ok(adapter->netdev))
 161                         netif_carrier_off(adapter->netdev);
 162
 163                 if (affectTxQueue) {
 164                         for (i = 0; i < adapter->num_tx_queues; i++)
 165                                 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
 166                 }
 167         }
 168 }
 169
 170 static void
 171 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 172 {
 173         int i;
 174         u32 events = le32_to_cpu(adapter->shared->ecr);
 175         if (!events)
 176                 return;
 177
 178         vmxnet3_ack_events(adapter, events);
 179
 180         /* Check if link state has changed */
 181         if (events & VMXNET3_ECR_LINK)
 182                 vmxnet3_check_link(adapter, true);
 183
 184         /* Check if there is an error on xmit/recv queues */
 185         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 186                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 187                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 188
 189                 for (i = 0; i < adapter->num_tx_queues; i++)
 190                         if (adapter->tqd_start[i].status.stopped)
 191                                 dev_err(&adapter->netdev->dev,
 192                                         "%s: tq[%d] error 0x%x\n",
 193                                         adapter->netdev->name, i, le32_to_cpu(
 194                                         adapter->tqd_start[i].status.error));
 195                 for (i = 0; i < adapter->num_rx_queues; i++)
 196                         if (adapter->rqd_start[i].status.stopped)
 197                                 dev_err(&adapter->netdev->dev,
 198                                         "%s: rq[%d] error 0x%x\n",
 199                                         adapter->netdev->name, i,
 200                                         adapter->rqd_start[i].status.error);
 201
 202                 schedule_work(&adapter->work);
 203         }
 204 }
 205
 206 #ifdef __BIG_ENDIAN_BITFIELD
 207 /*
 208  * The device expects the bitfields in shared structures to be written in
 209  * little endian. When CPU is big endian, the following routines are used to
 210  * correctly read and write into ABI.
 211  * The general technique used here is : double word bitfields are defined in
 212  * opposite order for big endian architecture. Then before reading them in
 213  * driver the complete double word is translated using le32_to_cpu. Similarly
 214  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 215  * double words into required format.
 216  * In order to avoid touching bits in shared structure more than once, temporary
 217  * descriptors are used. These are passed as srcDesc to following functions.
 218  */
 219 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 220                                 struct Vmxnet3_RxDesc *dstDesc)
 221 {
 222         u32 *src = (u32 *)srcDesc + 2;
 223         u32 *dst = (u32 *)dstDesc + 2;
 224         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 225         *dst = le32_to_cpu(*src);
 226         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 227 }
 228
 229 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 230                                struct Vmxnet3_TxDesc *dstDesc)
 231 {
 232         int i;
 233         u32 *src = (u32 *)(srcDesc + 1);
 234         u32 *dst = (u32 *)(dstDesc + 1);
 235
 236         /* Working backwards so that the gen bit is set at the end. */
 237         for (i = 2; i > 0; i--) {
 238                 src--;
 239                 dst--;
 240                 *dst = cpu_to_le32(*src);
 241         }
 242 }
 243
 244
 245 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 246                                 struct Vmxnet3_RxCompDesc *dstDesc)
 247 {
 248         int i = 0;
 249         u32 *src = (u32 *)srcDesc;
 250         u32 *dst = (u32 *)dstDesc;
 251         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 252                 *dst = le32_to_cpu(*src);
 253                 src++;
 254                 dst++;
 255         }
 256 }
 257
 258
 259 /* Used to read bitfield values from double words. */
 260 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 261 {
 262         u32 temp = le32_to_cpu(*bitfield);
 263         u32 mask = ((1 << size) - 1) << pos;
 264         temp &= mask;
 265         temp >>= pos;
 266         return temp;
 267 }
 268
 269
 270
 271 #endif  /* __BIG_ENDIAN_BITFIELD */
 272
 273 #ifdef __BIG_ENDIAN_BITFIELD
 274
 275 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 276                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 277                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 278 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 279                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 280                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 281 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 282                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 283                         VMXNET3_TCD_GEN_SIZE)
 284 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 285                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 286 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 287                         (dstrcd) = (tmp); \
 288                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 289                 } while (0)
 290 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 291                         (dstrxd) = (tmp); \
 292                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 293                 } while (0)
 294
 295 #else
 296
 297 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 298 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 299 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 300 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 301 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 302 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 303
 304 #endif /* __BIG_ENDIAN_BITFIELD  */
 305
 306
 307 static void
 308 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 309                      struct pci_dev *pdev)
 310 {
 311         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 312                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
 313                                  PCI_DMA_TODEVICE);
 314         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 315                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
 316                                PCI_DMA_TODEVICE);
 317         else
 318                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 319
 320         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 321 }
 322
 323
 324 static int
 325 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 326                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 327 {
 328         struct sk_buff *skb;
 329         int entries = 0;
 330
 331         /* no out of order completion */
 332         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 333         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 334
 335         skb = tq->buf_info[eop_idx].skb;
 336         BUG_ON(skb == NULL);
 337         tq->buf_info[eop_idx].skb = NULL;
 338
 339         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 340
 341         while (tq->tx_ring.next2comp != eop_idx) {
 342                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 343                                      pdev);
 344
 345                 /* update next2comp w/o tx_lock. Since we are marking more,
 346                  * instead of less, tx ring entries avail, the worst case is
 347                  * that the tx routine incorrectly re-queues a pkt due to
 348                  * insufficient tx ring entries.
 349                  */
 350                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 351                 entries++;
 352         }
 353
 354         dev_kfree_skb_any(skb);
 355         return entries;
 356 }
 357
 358
 359 static int
 360 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 361                         struct vmxnet3_adapter *adapter)
 362 {
 363         int completed = 0;
 364         union Vmxnet3_GenericDesc *gdesc;
 365
 366         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 367         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 368                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 369                                                &gdesc->tcd), tq, adapter->pdev,
 370                                                adapter);
 371
 372                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 373                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 374         }
 375
 376         if (completed) {
 377                 spin_lock(&tq->tx_lock);
 378                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 379                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 380                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 381                              netif_carrier_ok(adapter->netdev))) {
 382                         vmxnet3_tq_wake(tq, adapter);
 383                 }
 384                 spin_unlock(&tq->tx_lock);
 385         }
 386         return completed;
 387 }
 388
 389
 390 static void
 391 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 392                    struct vmxnet3_adapter *adapter)
 393 {
 394         int i;
 395
 396         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 397                 struct vmxnet3_tx_buf_info *tbi;
 398                 union Vmxnet3_GenericDesc *gdesc;
 399
 400                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 401                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
 402
 403                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 404                 if (tbi->skb) {
 405                         dev_kfree_skb_any(tbi->skb);
 406                         tbi->skb = NULL;
 407                 }
 408                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 409         }
 410
 411         /* sanity check, verify all buffers are indeed unmapped and freed */
 412         for (i = 0; i < tq->tx_ring.size; i++) {
 413                 BUG_ON(tq->buf_info[i].skb != NULL ||
 414                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 415         }
 416
 417         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 418         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 419
 420         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 421         tq->comp_ring.next2proc = 0;
 422 }
 423
 424
 425 static void
 426 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 427                    struct vmxnet3_adapter *adapter)
 428 {
 429         if (tq->tx_ring.base) {
 430                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 431                                     sizeof(struct Vmxnet3_TxDesc),
 432                                     tq->tx_ring.base, tq->tx_ring.basePA);
 433                 tq->tx_ring.base = NULL;
 434         }
 435         if (tq->data_ring.base) {
 436                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 437                                     sizeof(struct Vmxnet3_TxDataDesc),
 438                                     tq->data_ring.base, tq->data_ring.basePA);
 439                 tq->data_ring.base = NULL;
 440         }
 441         if (tq->comp_ring.base) {
 442                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 443                                     sizeof(struct Vmxnet3_TxCompDesc),
 444                                     tq->comp_ring.base, tq->comp_ring.basePA);
 445                 tq->comp_ring.base = NULL;
 446         }
 447         kfree(tq->buf_info);
 448         tq->buf_info = NULL;
 449 }
 450
 451
 452 /* Destroy all tx queues */
 453 void
 454 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
 455 {
 456         int i;
 457
 458         for (i = 0; i < adapter->num_tx_queues; i++)
 459                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
 460 }
 461
 462
 463 static void
 464 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 465                 struct vmxnet3_adapter *adapter)
 466 {
 467         int i;
 468
 469         /* reset the tx ring contents to 0 and reset the tx ring states */
 470         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 471                sizeof(struct Vmxnet3_TxDesc));
 472         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 473         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 474
 475         memset(tq->data_ring.base, 0, tq->data_ring.size *
 476                sizeof(struct Vmxnet3_TxDataDesc));
 477
 478         /* reset the tx comp ring contents to 0 and reset comp ring states */
 479         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 480                sizeof(struct Vmxnet3_TxCompDesc));
 481         tq->comp_ring.next2proc = 0;
 482         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 483
 484         /* reset the bookkeeping data */
 485         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 486         for (i = 0; i < tq->tx_ring.size; i++)
 487                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 488
 489         /* stats are not reset */
 490 }
 491
 492
 493 static int
 494 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 495                   struct vmxnet3_adapter *adapter)
 496 {
 497         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 498                tq->comp_ring.base || tq->buf_info);
 499
 500         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 501                            * sizeof(struct Vmxnet3_TxDesc),
 502                            &tq->tx_ring.basePA);
 503         if (!tq->tx_ring.base) {
 504                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 505                        adapter->netdev->name);
 506                 goto err;
 507         }
 508
 509         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 510                              tq->data_ring.size *
 511                              sizeof(struct Vmxnet3_TxDataDesc),
 512                              &tq->data_ring.basePA);
 513         if (!tq->data_ring.base) {
 514                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 515                        adapter->netdev->name);
 516                 goto err;
 517         }
 518
 519         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 520                              tq->comp_ring.size *
 521                              sizeof(struct Vmxnet3_TxCompDesc),
 522                              &tq->comp_ring.basePA);
 523         if (!tq->comp_ring.base) {
 524                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 525                        adapter->netdev->name);
 526                 goto err;
 527         }
 528
 529         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 530                                GFP_KERNEL);
 531         if (!tq->buf_info) {
 532                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
 533                        adapter->netdev->name);
 534                 goto err;
 535         }
 536
 537         return 0;
 538
 539 err:
 540         vmxnet3_tq_destroy(tq, adapter);
 541         return -ENOMEM;
 542 }
 543
 544 static void
 545 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
 546 {
 547         int i;
 548
 549         for (i = 0; i < adapter->num_tx_queues; i++)
 550                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
 551 }
 552
 553 /*
 554  *    starting from ring->next2fill, allocate rx buffers for the given ring
 555  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 556  *    are allocated or allocation fails
 557  */
 558
 559 static int
 560 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 561                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 562 {
 563         int num_allocated = 0;
 564         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 565         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 566         u32 val;
 567
 568         while (num_allocated < num_to_alloc) {
 569                 struct vmxnet3_rx_buf_info *rbi;
 570                 union Vmxnet3_GenericDesc *gd;
 571
 572                 rbi = rbi_base + ring->next2fill;
 573                 gd = ring->base + ring->next2fill;
 574
 575                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 576                         if (rbi->skb == NULL) {
 577                                 rbi->skb = dev_alloc_skb(rbi->len +
 578                                                          NET_IP_ALIGN);
 579                                 if (unlikely(rbi->skb == NULL)) {
 580                                         rq->stats.rx_buf_alloc_failure++;
 581                                         break;
 582                                 }
 583                                 rbi->skb->dev = adapter->netdev;
 584
 585                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 586                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 587                                                 rbi->skb->data, rbi->len,
 588                                                 PCI_DMA_FROMDEVICE);
 589                         } else {
 590                                 /* rx buffer skipped by the device */
 591                         }
 592                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 593                 } else {
 594                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 595                                rbi->len  != PAGE_SIZE);
 596
 597                         if (rbi->page == NULL) {
 598                                 rbi->page = alloc_page(GFP_ATOMIC);
 599                                 if (unlikely(rbi->page == NULL)) {
 600                                         rq->stats.rx_buf_alloc_failure++;
 601                                         break;
 602                                 }
 603                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 604                                                 rbi->page, 0, PAGE_SIZE,
 605                                                 PCI_DMA_FROMDEVICE);
 606                         } else {
 607                                 /* rx buffers skipped by the device */
 608                         }
 609                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 610                 }
 611
 612                 BUG_ON(rbi->dma_addr == 0);
 613                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 614                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
 615                                            | val | rbi->len);
 616
 617                 num_allocated++;
 618                 vmxnet3_cmd_ring_adv_next2fill(ring);
 619         }
 620         rq->uncommitted[ring_idx] += num_allocated;
 621
 622         dev_dbg(&adapter->netdev->dev,
 623                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 624                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
 625                 ring->next2comp, rq->uncommitted[ring_idx]);
 626
 627         /* so that the device can distinguish a full ring and an empty ring */
 628         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 629
 630         return num_allocated;
 631 }
 632
 633
 634 static void
 635 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 636                     struct vmxnet3_rx_buf_info *rbi)
 637 {
 638         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 639                 skb_shinfo(skb)->nr_frags;
 640
 641         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 642
 643         frag->page = rbi->page;
 644         frag->page_offset = 0;
 645         frag->size = rcd->len;
 646         skb->data_len += frag->size;
 647         skb_shinfo(skb)->nr_frags++;
 648 }
 649
 650
 651 static void
 652 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 653                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 654                 struct vmxnet3_adapter *adapter)
 655 {
 656         u32 dw2, len;
 657         unsigned long buf_offset;
 658         int i;
 659         union Vmxnet3_GenericDesc *gdesc;
 660         struct vmxnet3_tx_buf_info *tbi = NULL;
 661
 662         BUG_ON(ctx->copy_size > skb_headlen(skb));
 663
 664         /* use the previous gen bit for the SOP desc */
 665         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 666
 667         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 668         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 669
 670         /* no need to map the buffer if headers are copied */
 671         if (ctx->copy_size) {
 672                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 673                                         tq->tx_ring.next2fill *
 674                                         sizeof(struct Vmxnet3_TxDataDesc));
 675                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 676                 ctx->sop_txd->dword[3] = 0;
 677
 678                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 679                 tbi->map_type = VMXNET3_MAP_NONE;
 680
 681                 dev_dbg(&adapter->netdev->dev,
 682                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 683                         tq->tx_ring.next2fill,
 684                         le64_to_cpu(ctx->sop_txd->txd.addr),
 685                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 686                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 687
 688                 /* use the right gen for non-SOP desc */
 689                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 690         }
 691
 692         /* linear part can use multiple tx desc if it's big */
 693         len = skb_headlen(skb) - ctx->copy_size;
 694         buf_offset = ctx->copy_size;
 695         while (len) {
 696                 u32 buf_size;
 697
 698                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 699                         buf_size = len;
 700                         dw2 |= len;
 701                 } else {
 702                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 703                         /* spec says that for TxDesc.len, 0 == 2^14 */
 704                 }
 705
 706                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 707                 tbi->map_type = VMXNET3_MAP_SINGLE;
 708                 tbi->dma_addr = pci_map_single(adapter->pdev,
 709                                 skb->data + buf_offset, buf_size,
 710                                 PCI_DMA_TODEVICE);
 711
 712                 tbi->len = buf_size;
 713
 714                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 715                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 716
 717                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 718                 gdesc->dword[2] = cpu_to_le32(dw2);
 719                 gdesc->dword[3] = 0;
 720
 721                 dev_dbg(&adapter->netdev->dev,
 722                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 723                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 724                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 725                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 726                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 727
 728                 len -= buf_size;
 729                 buf_offset += buf_size;
 730         }
 731
 732         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 733                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 734
 735                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 736                 tbi->map_type = VMXNET3_MAP_PAGE;
 737                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
 738                                              frag->page_offset, frag->size,
 739                                              PCI_DMA_TODEVICE);
 740
 741                 tbi->len = frag->size;
 742
 743                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 744                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 745
 746                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 747                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
 748                 gdesc->dword[3] = 0;
 749
 750                 dev_dbg(&adapter->netdev->dev,
 751                         "txd[%u]: 0x%llu %u %u\n",
 752                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 753                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 754                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 755                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 756         }
 757
 758         ctx->eop_txd = gdesc;
 759
 760         /* set the last buf_info for the pkt */
 761         tbi->skb = skb;
 762         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 763 }
 764
 765
 766 /* Init all tx queues */
 767 static void
 768 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 769 {
 770         int i;
 771
 772         for (i = 0; i < adapter->num_tx_queues; i++)
 773                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 774 }
 775
 776
 777 /*
 778  *    parse and copy relevant protocol headers:
 779  *      For a tso pkt, relevant headers are L2/3/4 including options
 780  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 781  *      if it's a TCP/UDP pkt
 782  *
 783  * Returns:
 784  *    -1:  error happens during parsing
 785  *     0:  protocol headers parsed, but too big to be copied
 786  *     1:  protocol headers parsed and copied
 787  *
 788  * Other effects:
 789  *    1. related *ctx fields are updated.
 790  *    2. ctx->copy_size is # of bytes copied
 791  *    3. the portion copied is guaranteed to be in the linear part
 792  *
 793  */
 794 static int
 795 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 796                            struct vmxnet3_tx_ctx *ctx,
 797                            struct vmxnet3_adapter *adapter)
 798 {
 799         struct Vmxnet3_TxDataDesc *tdd;
 800
 801         if (ctx->mss) { /* TSO */
 802                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 803                 ctx->l4_hdr_size = ((struct tcphdr *)
 804                                    skb_transport_header(skb))->doff * 4;
 805                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 806         } else {
 807                 unsigned int pull_size;
 808
 809                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 810                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 811
 812                         if (ctx->ipv4) {
 813                                 struct iphdr *iph = (struct iphdr *)
 814                                                     skb_network_header(skb);
 815                                 if (iph->protocol == IPPROTO_TCP) {
 816                                         pull_size = ctx->eth_ip_hdr_size +
 817                                                     sizeof(struct tcphdr);
 818
 819                                         if (unlikely(!pskb_may_pull(skb,
 820                                                                 pull_size))) {
 821                                                 goto err;
 822                                         }
 823                                         ctx->l4_hdr_size = ((struct tcphdr *)
 824                                            skb_transport_header(skb))->doff * 4;
 825                                 } else if (iph->protocol == IPPROTO_UDP) {
 826                                         ctx->l4_hdr_size =
 827                                                         sizeof(struct udphdr);
 828                                 } else {
 829                                         ctx->l4_hdr_size = 0;
 830                                 }
 831                         } else {
 832                                 /* for simplicity, don't copy L4 headers */
 833                                 ctx->l4_hdr_size = 0;
 834                         }
 835                         ctx->copy_size = ctx->eth_ip_hdr_size +
 836                                          ctx->l4_hdr_size;
 837                 } else {
 838                         ctx->eth_ip_hdr_size = 0;
 839                         ctx->l4_hdr_size = 0;
 840                         /* copy as much as allowed */
 841                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 842                                              , skb_headlen(skb));
 843                 }
 844
 845                 /* make sure headers are accessible directly */
 846                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 847                         goto err;
 848         }
 849
 850         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 851                 tq->stats.oversized_hdr++;
 852                 ctx->copy_size = 0;
 853                 return 0;
 854         }
 855
 856         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 857
 858         memcpy(tdd->data, skb->data, ctx->copy_size);
 859         dev_dbg(&adapter->netdev->dev,
 860                 "copy %u bytes to dataRing[%u]\n",
 861                 ctx->copy_size, tq->tx_ring.next2fill);
 862         return 1;
 863
 864 err:
 865         return -1;
 866 }
 867
 868
 869 static void
 870 vmxnet3_prepare_tso(struct sk_buff *skb,
 871                     struct vmxnet3_tx_ctx *ctx)
 872 {
 873         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
 874         if (ctx->ipv4) {
 875                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
 876                 iph->check = 0;
 877                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 878                                                  IPPROTO_TCP, 0);
 879         } else {
 880                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
 881                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 882                                                IPPROTO_TCP, 0);
 883         }
 884 }
 885
 886
 887 /*
 888  * Transmits a pkt thru a given tq
 889  * Returns:
 890  *    NETDEV_TX_OK:      descriptors are setup successfully
 891  *    NETDEV_TX_OK:      error occured, the pkt is dropped
 892  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 893  *
 894  * Side-effects:
 895  *    1. tx ring may be changed
 896  *    2. tq stats may be updated accordingly
 897  *    3. shared->txNumDeferred may be updated
 898  */
 899
 900 static int
 901 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 902                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 903 {
 904         int ret;
 905         u32 count;
 906         unsigned long flags;
 907         struct vmxnet3_tx_ctx ctx;
 908         union Vmxnet3_GenericDesc *gdesc;
 909 #ifdef __BIG_ENDIAN_BITFIELD
 910         /* Use temporary descriptor to avoid touching bits multiple times */
 911         union Vmxnet3_GenericDesc tempTxDesc;
 912 #endif
 913
 914         /* conservatively estimate # of descriptors to use */
 915         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
 916                 skb_shinfo(skb)->nr_frags + 1;
 917
 918         ctx.ipv4 = (skb->protocol == cpu_to_be16(ETH_P_IP));
 919
 920         ctx.mss = skb_shinfo(skb)->gso_size;
 921         if (ctx.mss) {
 922                 if (skb_header_cloned(skb)) {
 923                         if (unlikely(pskb_expand_head(skb, 0, 0,
 924                                                       GFP_ATOMIC) != 0)) {
 925                                 tq->stats.drop_tso++;
 926                                 goto drop_pkt;
 927                         }
 928                         tq->stats.copy_skb_header++;
 929                 }
 930                 vmxnet3_prepare_tso(skb, &ctx);
 931         } else {
 932                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 933
 934                         /* non-tso pkts must not use more than
 935                          * VMXNET3_MAX_TXD_PER_PKT entries
 936                          */
 937                         if (skb_linearize(skb) != 0) {
 938                                 tq->stats.drop_too_many_frags++;
 939                                 goto drop_pkt;
 940                         }
 941                         tq->stats.linearized++;
 942
 943                         /* recalculate the # of descriptors to use */
 944                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 945                 }
 946         }
 947
 948         spin_lock_irqsave(&tq->tx_lock, flags);
 949
 950         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 951                 tq->stats.tx_ring_full++;
 952                 dev_dbg(&adapter->netdev->dev,
 953                         "tx queue stopped on %s, next2comp %u"
 954                         " next2fill %u\n", adapter->netdev->name,
 955                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 956
 957                 vmxnet3_tq_stop(tq, adapter);
 958                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 959                 return NETDEV_TX_BUSY;
 960         }
 961
 962
 963         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 964         if (ret >= 0) {
 965                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 966                 /* hdrs parsed, check against other limits */
 967                 if (ctx.mss) {
 968                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 969                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 970                                 goto hdr_too_big;
 971                         }
 972                 } else {
 973                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 974                                 if (unlikely(ctx.eth_ip_hdr_size +
 975                                              skb->csum_offset >
 976                                              VMXNET3_MAX_CSUM_OFFSET)) {
 977                                         goto hdr_too_big;
 978                                 }
 979                         }
 980                 }
 981         } else {
 982                 tq->stats.drop_hdr_inspect_err++;
 983                 goto drop_pkt;
 984         }
 985
 986         /* fill tx descs related to addr & len */
 987         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
 988
 989         /* setup the EOP desc */
 990         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
 991
 992         /* setup the SOP desc */
 993 #ifdef __BIG_ENDIAN_BITFIELD
 994         gdesc = &tempTxDesc;
 995         gdesc->dword[2] = ctx.sop_txd->dword[2];
 996         gdesc->dword[3] = ctx.sop_txd->dword[3];
 997 #else
 998         gdesc = ctx.sop_txd;
 999 #endif
1000         if (ctx.mss) {
1001                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1002                 gdesc->txd.om = VMXNET3_OM_TSO;
1003                 gdesc->txd.msscof = ctx.mss;
1004                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1005                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1006         } else {
1007                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1008                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1009                         gdesc->txd.om = VMXNET3_OM_CSUM;
1010                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1011                                             skb->csum_offset;
1012                 } else {
1013                         gdesc->txd.om = 0;
1014                         gdesc->txd.msscof = 0;
1015                 }
1016                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1017         }
1018
1019         if (vlan_tx_tag_present(skb)) {
1020                 gdesc->txd.ti = 1;
1021                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1022         }
1023
1024         /* finally flips the GEN bit of the SOP desc. */
1025         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1026                                                   VMXNET3_TXD_GEN);
1027 #ifdef __BIG_ENDIAN_BITFIELD
1028         /* Finished updating in bitfields of Tx Desc, so write them in original
1029          * place.
1030          */
1031         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1032                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1033         gdesc = ctx.sop_txd;
1034 #endif
1035         dev_dbg(&adapter->netdev->dev,
1036                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1037                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1038                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1039                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1040
1041         spin_unlock_irqrestore(&tq->tx_lock, flags);
1042
1043         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1044                                         le32_to_cpu(tq->shared->txThreshold)) {
1045                 tq->shared->txNumDeferred = 0;
1046                 VMXNET3_WRITE_BAR0_REG(adapter,
1047                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1048                                        tq->tx_ring.next2fill);
1049         }
1050
1051         return NETDEV_TX_OK;
1052
1053 hdr_too_big:
1054         tq->stats.drop_oversized_hdr++;
1055 drop_pkt:
1056         tq->stats.drop_total++;
1057         dev_kfree_skb(skb);
1058         return NETDEV_TX_OK;
1059 }
1060
1061
1062 static netdev_tx_t
1063 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1064 {
1065         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1066
1067                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1068                 return vmxnet3_tq_xmit(skb,
1069                                        &adapter->tx_queue[skb->queue_mapping],
1070                                        adapter, netdev);
1071 }
1072
1073
1074 static void
1075 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1076                 struct sk_buff *skb,
1077                 union Vmxnet3_GenericDesc *gdesc)
1078 {
1079         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1080                 /* typical case: TCP/UDP over IP and both csums are correct */
1081                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1082                                                         VMXNET3_RCD_CSUM_OK) {
1083                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1084                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1085                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1086                         BUG_ON(gdesc->rcd.frg);
1087                 } else {
1088                         if (gdesc->rcd.csum) {
1089                                 skb->csum = htons(gdesc->rcd.csum);
1090                                 skb->ip_summed = CHECKSUM_PARTIAL;
1091                         } else {
1092                                 skb_checksum_none_assert(skb);
1093                         }
1094                 }
1095         } else {
1096                 skb_checksum_none_assert(skb);
1097         }
1098 }
1099
1100
1101 static void
1102 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1103                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1104 {
1105         rq->stats.drop_err++;
1106         if (!rcd->fcs)
1107                 rq->stats.drop_fcs++;
1108
1109         rq->stats.drop_total++;
1110
1111         /*
1112          * We do not unmap and chain the rx buffer to the skb.
1113          * We basically pretend this buffer is not used and will be recycled
1114          * by vmxnet3_rq_alloc_rx_buf()
1115          */
1116
1117         /*
1118          * ctx->skb may be NULL if this is the first and the only one
1119          * desc for the pkt
1120          */
1121         if (ctx->skb)
1122                 dev_kfree_skb_irq(ctx->skb);
1123
1124         ctx->skb = NULL;
1125 }
1126
1127
1128 static int
1129 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1130                        struct vmxnet3_adapter *adapter, int quota)
1131 {
1132         static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
1133         u32 num_rxd = 0;
1134         struct Vmxnet3_RxCompDesc *rcd;
1135         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1136 #ifdef __BIG_ENDIAN_BITFIELD
1137         struct Vmxnet3_RxDesc rxCmdDesc;
1138         struct Vmxnet3_RxCompDesc rxComp;
1139 #endif
1140         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1141                           &rxComp);
1142         while (rcd->gen == rq->comp_ring.gen) {
1143                 struct vmxnet3_rx_buf_info *rbi;
1144                 struct sk_buff *skb;
1145                 int num_to_alloc;
1146                 struct Vmxnet3_RxDesc *rxd;
1147                 u32 idx, ring_idx;
1148
1149                 if (num_rxd >= quota) {
1150                         /* we may stop even before we see the EOP desc of
1151                          * the current pkt
1152                          */
1153                         break;
1154                 }
1155                 num_rxd++;
1156                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1157                 idx = rcd->rxdIdx;
1158                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1159                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1160                                   &rxCmdDesc);
1161                 rbi = rq->buf_info[ring_idx] + idx;
1162
1163                 BUG_ON(rxd->addr != rbi->dma_addr ||
1164                        rxd->len != rbi->len);
1165
1166                 if (unlikely(rcd->eop && rcd->err)) {
1167                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1168                         goto rcd_done;
1169                 }
1170
1171                 if (rcd->sop) { /* first buf of the pkt */
1172                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1173                                rcd->rqID != rq->qid);
1174
1175                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1176                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1177
1178                         if (unlikely(rcd->len == 0)) {
1179                                 /* Pretend the rx buffer is skipped. */
1180                                 BUG_ON(!(rcd->sop && rcd->eop));
1181                                 dev_dbg(&adapter->netdev->dev,
1182                                         "rxRing[%u][%u] 0 length\n",
1183                                         ring_idx, idx);
1184                                 goto rcd_done;
1185                         }
1186
1187                         ctx->skb = rbi->skb;
1188                         rbi->skb = NULL;
1189
1190                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1191                                          PCI_DMA_FROMDEVICE);
1192
1193                         skb_put(ctx->skb, rcd->len);
1194                 } else {
1195                         BUG_ON(ctx->skb == NULL);
1196                         /* non SOP buffer must be type 1 in most cases */
1197                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1198                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1199
1200                                 if (rcd->len) {
1201                                         pci_unmap_page(adapter->pdev,
1202                                                        rbi->dma_addr, rbi->len,
1203                                                        PCI_DMA_FROMDEVICE);
1204
1205                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1206                                         rbi->page = NULL;
1207                                 }
1208                         } else {
1209                                 /*
1210                                  * The only time a non-SOP buffer is type 0 is
1211                                  * when it's EOP and error flag is raised, which
1212                                  * has already been handled.
1213                                  */
1214                                 BUG_ON(true);
1215                         }
1216                 }
1217
1218                 skb = ctx->skb;
1219                 if (rcd->eop) {
1220                         skb->len += skb->data_len;
1221                         skb->truesize += skb->data_len;
1222
1223                         vmxnet3_rx_csum(adapter, skb,
1224                                         (union Vmxnet3_GenericDesc *)rcd);
1225                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1226
1227                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1228                                 vlan_hwaccel_receive_skb(skb,
1229                                                 adapter->vlan_grp, rcd->tci);
1230                         } else {
1231                                 netif_receive_skb(skb);
1232                         }
1233
1234                         ctx->skb = NULL;
1235                 }
1236
1237 rcd_done:
1238                 /* device may skip some rx descs */
1239                 rq->rx_ring[ring_idx].next2comp = idx;
1240                 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1241                                           rq->rx_ring[ring_idx].size);
1242
1243                 /* refill rx buffers frequently to avoid starving the h/w */
1244                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1245                                                            ring_idx);
1246                 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1247                                                         ring_idx, adapter))) {
1248                         vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1249                                                 adapter);
1250
1251                         /* if needed, update the register */
1252                         if (unlikely(rq->shared->updateRxProd)) {
1253                                 VMXNET3_WRITE_BAR0_REG(adapter,
1254                                         rxprod_reg[ring_idx] + rq->qid * 8,
1255                                         rq->rx_ring[ring_idx].next2fill);
1256                                 rq->uncommitted[ring_idx] = 0;
1257                         }
1258                 }
1259
1260                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1261                 vmxnet3_getRxComp(rcd,
1262                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1263         }
1264
1265         return num_rxd;
1266 }
1267
1268
1269 static void
1270 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1271                    struct vmxnet3_adapter *adapter)
1272 {
1273         u32 i, ring_idx;
1274         struct Vmxnet3_RxDesc *rxd;
1275
1276         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1277                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1278 #ifdef __BIG_ENDIAN_BITFIELD
1279                         struct Vmxnet3_RxDesc rxDesc;
1280 #endif
1281                         vmxnet3_getRxDesc(rxd,
1282                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1283
1284                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1285                                         rq->buf_info[ring_idx][i].skb) {
1286                                 pci_unmap_single(adapter->pdev, rxd->addr,
1287                                                  rxd->len, PCI_DMA_FROMDEVICE);
1288                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1289                                 rq->buf_info[ring_idx][i].skb = NULL;
1290                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1291                                         rq->buf_info[ring_idx][i].page) {
1292                                 pci_unmap_page(adapter->pdev, rxd->addr,
1293                                                rxd->len, PCI_DMA_FROMDEVICE);
1294                                 put_page(rq->buf_info[ring_idx][i].page);
1295                                 rq->buf_info[ring_idx][i].page = NULL;
1296                         }
1297                 }
1298
1299                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1300                 rq->rx_ring[ring_idx].next2fill =
1301                                         rq->rx_ring[ring_idx].next2comp = 0;
1302                 rq->uncommitted[ring_idx] = 0;
1303         }
1304
1305         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1306         rq->comp_ring.next2proc = 0;
1307 }
1308
1309
1310 static void
1311 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1312 {
1313         int i;
1314
1315         for (i = 0; i < adapter->num_rx_queues; i++)
1316                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1317 }
1318
1319
1320 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1321                         struct vmxnet3_adapter *adapter)
1322 {
1323         int i;
1324         int j;
1325
1326         /* all rx buffers must have already been freed */
1327         for (i = 0; i < 2; i++) {
1328                 if (rq->buf_info[i]) {
1329                         for (j = 0; j < rq->rx_ring[i].size; j++)
1330                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1331                 }
1332         }
1333
1334
1335         kfree(rq->buf_info[0]);
1336
1337         for (i = 0; i < 2; i++) {
1338                 if (rq->rx_ring[i].base) {
1339                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1340                                             * sizeof(struct Vmxnet3_RxDesc),
1341                                             rq->rx_ring[i].base,
1342                                             rq->rx_ring[i].basePA);
1343                         rq->rx_ring[i].base = NULL;
1344                 }
1345                 rq->buf_info[i] = NULL;
1346         }
1347
1348         if (rq->comp_ring.base) {
1349                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1350                                     sizeof(struct Vmxnet3_RxCompDesc),
1351                                     rq->comp_ring.base, rq->comp_ring.basePA);
1352                 rq->comp_ring.base = NULL;
1353         }
1354 }
1355
1356
1357 static int
1358 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1359                 struct vmxnet3_adapter  *adapter)
1360 {
1361         int i;
1362
1363         /* initialize buf_info */
1364         for (i = 0; i < rq->rx_ring[0].size; i++) {
1365
1366                 /* 1st buf for a pkt is skbuff */
1367                 if (i % adapter->rx_buf_per_pkt == 0) {
1368                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1369                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1370                 } else { /* subsequent bufs for a pkt is frag */
1371                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1372                         rq->buf_info[0][i].len = PAGE_SIZE;
1373                 }
1374         }
1375         for (i = 0; i < rq->rx_ring[1].size; i++) {
1376                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1377                 rq->buf_info[1][i].len = PAGE_SIZE;
1378         }
1379
1380         /* reset internal state and allocate buffers for both rings */
1381         for (i = 0; i < 2; i++) {
1382                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1383                 rq->uncommitted[i] = 0;
1384
1385                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1386                        sizeof(struct Vmxnet3_RxDesc));
1387                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1388         }
1389         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1390                                     adapter) == 0) {
1391                 /* at least has 1 rx buffer for the 1st ring */
1392                 return -ENOMEM;
1393         }
1394         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1395
1396         /* reset the comp ring */
1397         rq->comp_ring.next2proc = 0;
1398         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1399                sizeof(struct Vmxnet3_RxCompDesc));
1400         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1401
1402         /* reset rxctx */
1403         rq->rx_ctx.skb = NULL;
1404
1405         /* stats are not reset */
1406         return 0;
1407 }
1408
1409
1410 static int
1411 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1412 {
1413         int i, err = 0;
1414
1415         for (i = 0; i < adapter->num_rx_queues; i++) {
1416                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1417                 if (unlikely(err)) {
1418                         dev_err(&adapter->netdev->dev, "%s: failed to "
1419                                 "initialize rx queue%i\n",
1420                                 adapter->netdev->name, i);
1421                         break;
1422                 }
1423         }
1424         return err;
1425
1426 }
1427
1428
1429 static int
1430 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1431 {
1432         int i;
1433         size_t sz;
1434         struct vmxnet3_rx_buf_info *bi;
1435
1436         for (i = 0; i < 2; i++) {
1437
1438                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1439                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1440                                                         &rq->rx_ring[i].basePA);
1441                 if (!rq->rx_ring[i].base) {
1442                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1443                                adapter->netdev->name, i);
1444                         goto err;
1445                 }
1446         }
1447
1448         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1449         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1450                                                   &rq->comp_ring.basePA);
1451         if (!rq->comp_ring.base) {
1452                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1453                        adapter->netdev->name);
1454                 goto err;
1455         }
1456
1457         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1458                                                    rq->rx_ring[1].size);
1459         bi = kzalloc(sz, GFP_KERNEL);
1460         if (!bi) {
1461                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1462                        adapter->netdev->name);
1463                 goto err;
1464         }
1465         rq->buf_info[0] = bi;
1466         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1467
1468         return 0;
1469
1470 err:
1471         vmxnet3_rq_destroy(rq, adapter);
1472         return -ENOMEM;
1473 }
1474
1475
1476 static int
1477 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1478 {
1479         int i, err = 0;
1480
1481         for (i = 0; i < adapter->num_rx_queues; i++) {
1482                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1483                 if (unlikely(err)) {
1484                         dev_err(&adapter->netdev->dev,
1485                                 "%s: failed to create rx queue%i\n",
1486                                 adapter->netdev->name, i);
1487                         goto err_out;
1488                 }
1489         }
1490         return err;
1491 err_out:
1492         vmxnet3_rq_destroy_all(adapter);
1493         return err;
1494
1495 }
1496
1497 /* Multiple queue aware polling function for tx and rx */
1498
1499 static int
1500 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1501 {
1502         int rcd_done = 0, i;
1503         if (unlikely(adapter->shared->ecr))
1504                 vmxnet3_process_events(adapter);
1505         for (i = 0; i < adapter->num_tx_queues; i++)
1506                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1507
1508         for (i = 0; i < adapter->num_rx_queues; i++)
1509                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1510                                                    adapter, budget);
1511         return rcd_done;
1512 }
1513
1514
1515 static int
1516 vmxnet3_poll(struct napi_struct *napi, int budget)
1517 {
1518         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1519                                           struct vmxnet3_rx_queue, napi);
1520         int rxd_done;
1521
1522         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1523
1524         if (rxd_done < budget) {
1525                 napi_complete(napi);
1526                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1527         }
1528         return rxd_done;
1529 }
1530
1531 /*
1532  * NAPI polling function for MSI-X mode with multiple Rx queues
1533  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1534  */
1535
1536 static int
1537 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1538 {
1539         struct vmxnet3_rx_queue *rq = container_of(napi,
1540                                                 struct vmxnet3_rx_queue, napi);
1541         struct vmxnet3_adapter *adapter = rq->adapter;
1542         int rxd_done;
1543
1544         /* When sharing interrupt with corresponding tx queue, process
1545          * tx completions in that queue as well
1546          */
1547         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1548                 struct vmxnet3_tx_queue *tq =
1549                                 &adapter->tx_queue[rq - adapter->rx_queue];
1550                 vmxnet3_tq_tx_complete(tq, adapter);
1551         }
1552
1553         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1554
1555         if (rxd_done < budget) {
1556                 napi_complete(napi);
1557                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1558         }
1559         return rxd_done;
1560 }
1561
1562
1563 #ifdef CONFIG_PCI_MSI
1564
1565 /*
1566  * Handle completion interrupts on tx queues
1567  * Returns whether or not the intr is handled
1568  */
1569
1570 static irqreturn_t
1571 vmxnet3_msix_tx(int irq, void *data)
1572 {
1573         struct vmxnet3_tx_queue *tq = data;
1574         struct vmxnet3_adapter *adapter = tq->adapter;
1575
1576         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1577                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1578
1579         /* Handle the case where only one irq is allocate for all tx queues */
1580         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1581                 int i;
1582                 for (i = 0; i < adapter->num_tx_queues; i++) {
1583                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1584                         vmxnet3_tq_tx_complete(txq, adapter);
1585                 }
1586         } else {
1587                 vmxnet3_tq_tx_complete(tq, adapter);
1588         }
1589         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1590
1591         return IRQ_HANDLED;
1592 }
1593
1594
1595 /*
1596  * Handle completion interrupts on rx queues. Returns whether or not the
1597  * intr is handled
1598  */
1599
1600 static irqreturn_t
1601 vmxnet3_msix_rx(int irq, void *data)
1602 {
1603         struct vmxnet3_rx_queue *rq = data;
1604         struct vmxnet3_adapter *adapter = rq->adapter;
1605
1606         /* disable intr if needed */
1607         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1608                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1609         napi_schedule(&rq->napi);
1610
1611         return IRQ_HANDLED;
1612 }
1613
1614 /*
1615  *----------------------------------------------------------------------------
1616  *
1617  * vmxnet3_msix_event --
1618  *
1619  *    vmxnet3 msix event intr handler
1620  *
1621  * Result:
1622  *    whether or not the intr is handled
1623  *
1624  *----------------------------------------------------------------------------
1625  */
1626
1627 static irqreturn_t
1628 vmxnet3_msix_event(int irq, void *data)
1629 {
1630         struct net_device *dev = data;
1631         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1632
1633         /* disable intr if needed */
1634         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1635                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1636
1637         if (adapter->shared->ecr)
1638                 vmxnet3_process_events(adapter);
1639
1640         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1641
1642         return IRQ_HANDLED;
1643 }
1644
1645 #endif /* CONFIG_PCI_MSI  */
1646
1647
1648 /* Interrupt handler for vmxnet3  */
1649 static irqreturn_t
1650 vmxnet3_intr(int irq, void *dev_id)
1651 {
1652         struct net_device *dev = dev_id;
1653         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1654
1655         if (adapter->intr.type == VMXNET3_IT_INTX) {
1656                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1657                 if (unlikely(icr == 0))
1658                         /* not ours */
1659                         return IRQ_NONE;
1660         }
1661
1662
1663         /* disable intr if needed */
1664         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1665                 vmxnet3_disable_all_intrs(adapter);
1666
1667         napi_schedule(&adapter->rx_queue[0].napi);
1668
1669         return IRQ_HANDLED;
1670 }
1671
1672 #ifdef CONFIG_NET_POLL_CONTROLLER
1673
1674 /* netpoll callback. */
1675 static void
1676 vmxnet3_netpoll(struct net_device *netdev)
1677 {
1678         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1679
1680         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1681                 vmxnet3_disable_all_intrs(adapter);
1682
1683         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1684         vmxnet3_enable_all_intrs(adapter);
1685
1686 }
1687 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1688
1689 static int
1690 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1691 {
1692         struct vmxnet3_intr *intr = &adapter->intr;
1693         int err = 0, i;
1694         int vector = 0;
1695
1696 #ifdef CONFIG_PCI_MSI
1697         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1698                 for (i = 0; i < adapter->num_tx_queues; i++) {
1699                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1700                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1701                                         adapter->netdev->name, vector);
1702                                 err = request_irq(
1703                                               intr->msix_entries[vector].vector,
1704                                               vmxnet3_msix_tx, 0,
1705                                               adapter->tx_queue[i].name,
1706                                               &adapter->tx_queue[i]);
1707                         } else {
1708                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1709                                         adapter->netdev->name, vector);
1710                         }
1711                         if (err) {
1712                                 dev_err(&adapter->netdev->dev,
1713                                         "Failed to request irq for MSIX, %s, "
1714                                         "error %d\n",
1715                                         adapter->tx_queue[i].name, err);
1716                                 return err;
1717                         }
1718
1719                         /* Handle the case where only 1 MSIx was allocated for
1720                          * all tx queues */
1721                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1722                                 for (; i < adapter->num_tx_queues; i++)
1723                                         adapter->tx_queue[i].comp_ring.intr_idx
1724                                                                 = vector;
1725                                 vector++;
1726                                 break;
1727                         } else {
1728                                 adapter->tx_queue[i].comp_ring.intr_idx
1729                                                                 = vector++;
1730                         }
1731                 }
1732                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1733                         vector = 0;
1734
1735                 for (i = 0; i < adapter->num_rx_queues; i++) {
1736                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1737                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1738                                         adapter->netdev->name, vector);
1739                         else
1740                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1741                                         adapter->netdev->name, vector);
1742                         err = request_irq(intr->msix_entries[vector].vector,
1743                                           vmxnet3_msix_rx, 0,
1744                                           adapter->rx_queue[i].name,
1745                                           &(adapter->rx_queue[i]));
1746                         if (err) {
1747                                 printk(KERN_ERR "Failed to request irq for MSIX"
1748                                        ", %s, error %d\n",
1749                                        adapter->rx_queue[i].name, err);
1750                                 return err;
1751                         }
1752
1753                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1754                 }
1755
1756                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1757                         adapter->netdev->name, vector);
1758                 err = request_irq(intr->msix_entries[vector].vector,
1759                                   vmxnet3_msix_event, 0,
1760                                   intr->event_msi_vector_name, adapter->netdev);
1761                 intr->event_intr_idx = vector;
1762
1763         } else if (intr->type == VMXNET3_IT_MSI) {
1764                 adapter->num_rx_queues = 1;
1765                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1766                                   adapter->netdev->name, adapter->netdev);
1767         } else {
1768 #endif
1769                 adapter->num_rx_queues = 1;
1770                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1771                                   IRQF_SHARED, adapter->netdev->name,
1772                                   adapter->netdev);
1773 #ifdef CONFIG_PCI_MSI
1774         }
1775 #endif
1776         intr->num_intrs = vector + 1;
1777         if (err) {
1778                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1779                        ":%d\n", adapter->netdev->name, intr->type, err);
1780         } else {
1781                 /* Number of rx queues will not change after this */
1782                 for (i = 0; i < adapter->num_rx_queues; i++) {
1783                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1784                         rq->qid = i;
1785                         rq->qid2 = i + adapter->num_rx_queues;
1786                 }
1787
1788
1789
1790                 /* init our intr settings */
1791                 for (i = 0; i < intr->num_intrs; i++)
1792                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1793                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1794                         adapter->intr.event_intr_idx = 0;
1795                         for (i = 0; i < adapter->num_tx_queues; i++)
1796                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1797                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1798                 }
1799
1800                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1801                        "allocated\n", adapter->netdev->name, intr->type,
1802                        intr->mask_mode, intr->num_intrs);
1803         }
1804
1805         return err;
1806 }
1807
1808
1809 static void
1810 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1811 {
1812         struct vmxnet3_intr *intr = &adapter->intr;
1813         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1814
1815         switch (intr->type) {
1816 #ifdef CONFIG_PCI_MSI
1817         case VMXNET3_IT_MSIX:
1818         {
1819                 int i, vector = 0;
1820
1821                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1822                         for (i = 0; i < adapter->num_tx_queues; i++) {
1823                                 free_irq(intr->msix_entries[vector++].vector,
1824                                          &(adapter->tx_queue[i]));
1825                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1826                                         break;
1827                         }
1828                 }
1829
1830                 for (i = 0; i < adapter->num_rx_queues; i++) {
1831                         free_irq(intr->msix_entries[vector++].vector,
1832                                  &(adapter->rx_queue[i]));
1833                 }
1834
1835                 free_irq(intr->msix_entries[vector].vector,
1836                          adapter->netdev);
1837                 BUG_ON(vector >= intr->num_intrs);
1838                 break;
1839         }
1840 #endif
1841         case VMXNET3_IT_MSI:
1842                 free_irq(adapter->pdev->irq, adapter->netdev);
1843                 break;
1844         case VMXNET3_IT_INTX:
1845                 free_irq(adapter->pdev->irq, adapter->netdev);
1846                 break;
1847         default:
1848                 BUG_ON(true);
1849         }
1850 }
1851
1852 static void
1853 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1854 {
1855         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1856         struct Vmxnet3_DriverShared *shared = adapter->shared;
1857         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1858
1859         if (grp) {
1860                 /* add vlan rx stripping. */
1861                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1862                         int i;
1863                         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1864                         adapter->vlan_grp = grp;
1865
1866                         /* update FEATURES to device */
1867                         devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1868                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1869                                                VMXNET3_CMD_UPDATE_FEATURE);
1870                         /*
1871                          *  Clear entire vfTable; then enable untagged pkts.
1872                          *  Note: setting one entry in vfTable to non-zero turns
1873                          *  on VLAN rx filtering.
1874                          */
1875                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1876                                 vfTable[i] = 0;
1877
1878                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1879                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1880                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1881                 } else {
1882                         printk(KERN_ERR "%s: vlan_rx_register when device has "
1883                                "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1884                 }
1885         } else {
1886                 /* remove vlan rx stripping. */
1887                 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1888                 adapter->vlan_grp = NULL;
1889
1890                 if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1891                         int i;
1892
1893                         for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1894                                 /* clear entire vfTable; this also disables
1895                                  * VLAN rx filtering
1896                                  */
1897                                 vfTable[i] = 0;
1898                         }
1899                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1900                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1901
1902                         /* update FEATURES to device */
1903                         devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1904                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1905                                                VMXNET3_CMD_UPDATE_FEATURE);
1906                 }
1907         }
1908 }
1909
1910
1911 static void
1912 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1913 {
1914         if (adapter->vlan_grp) {
1915                 u16 vid;
1916                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1917                 bool activeVlan = false;
1918
1919                 for (vid = 0; vid < VLAN_N_VID; vid++) {
1920                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1921                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1922                                 activeVlan = true;
1923                         }
1924                 }
1925                 if (activeVlan) {
1926                         /* continue to allow untagged pkts */
1927                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1928                 }
1929         }
1930 }
1931
1932
1933 static void
1934 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1935 {
1936         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1937         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1938
1939         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1940         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1941                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1942 }
1943
1944
1945 static void
1946 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1947 {
1948         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1949         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1950
1951         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1952         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1953                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1954 }
1955
1956
1957 static u8 *
1958 vmxnet3_copy_mc(struct net_device *netdev)
1959 {
1960         u8 *buf = NULL;
1961         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1962
1963         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1964         if (sz <= 0xffff) {
1965                 /* We may be called with BH disabled */
1966                 buf = kmalloc(sz, GFP_ATOMIC);
1967                 if (buf) {
1968                         struct netdev_hw_addr *ha;
1969                         int i = 0;
1970
1971                         netdev_for_each_mc_addr(ha, netdev)
1972                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1973                                        ETH_ALEN);
1974                 }
1975         }
1976         return buf;
1977 }
1978
1979
1980 static void
1981 vmxnet3_set_mc(struct net_device *netdev)
1982 {
1983         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1984         struct Vmxnet3_RxFilterConf *rxConf =
1985                                         &adapter->shared->devRead.rxFilterConf;
1986         u8 *new_table = NULL;
1987         u32 new_mode = VMXNET3_RXM_UCAST;
1988
1989         if (netdev->flags & IFF_PROMISC)
1990                 new_mode |= VMXNET3_RXM_PROMISC;
1991
1992         if (netdev->flags & IFF_BROADCAST)
1993                 new_mode |= VMXNET3_RXM_BCAST;
1994
1995         if (netdev->flags & IFF_ALLMULTI)
1996                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1997         else
1998                 if (!netdev_mc_empty(netdev)) {
1999                         new_table = vmxnet3_copy_mc(netdev);
2000                         if (new_table) {
2001                                 new_mode |= VMXNET3_RXM_MCAST;
2002                                 rxConf->mfTableLen = cpu_to_le16(
2003                                         netdev_mc_count(netdev) * ETH_ALEN);
2004                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2005                                                     new_table));
2006                         } else {
2007                                 printk(KERN_INFO "%s: failed to copy mcast list"
2008                                        ", setting ALL_MULTI\n", netdev->name);
2009                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2010                         }
2011                 }
2012
2013
2014         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2015                 rxConf->mfTableLen = 0;
2016                 rxConf->mfTablePA = 0;
2017         }
2018
2019         if (new_mode != rxConf->rxMode) {
2020                 rxConf->rxMode = cpu_to_le32(new_mode);
2021                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2022                                        VMXNET3_CMD_UPDATE_RX_MODE);
2023         }
2024
2025         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2026                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2027
2028         kfree(new_table);
2029 }
2030
2031 void
2032 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2033 {
2034         int i;
2035
2036         for (i = 0; i < adapter->num_rx_queues; i++)
2037                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2038 }
2039
2040
2041 /*
2042  *   Set up driver_shared based on settings in adapter.
2043  */
2044
2045 static void
2046 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2047 {
2048         struct Vmxnet3_DriverShared *shared = adapter->shared;
2049         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2050         struct Vmxnet3_TxQueueConf *tqc;
2051         struct Vmxnet3_RxQueueConf *rqc;
2052         int i;
2053
2054         memset(shared, 0, sizeof(*shared));
2055
2056         /* driver settings */
2057         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2058         devRead->misc.driverInfo.version = cpu_to_le32(
2059                                                 VMXNET3_DRIVER_VERSION_NUM);
2060         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2061                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2062         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2063         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2064                                 *((u32 *)&devRead->misc.driverInfo.gos));
2065         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2066         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2067
2068         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2069         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2070
2071         /* set up feature flags */
2072         if (adapter->rxcsum)
2073                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2074
2075         if (adapter->lro) {
2076                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2077                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2078         }
2079         if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX) &&
2080             adapter->vlan_grp) {
2081                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2082         }
2083
2084         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2085         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2086         devRead->misc.queueDescLen = cpu_to_le32(
2087                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2088                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2089
2090         /* tx queue settings */
2091         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2092         for (i = 0; i < adapter->num_tx_queues; i++) {
2093                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2094                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2095                 tqc = &adapter->tqd_start[i].conf;
2096                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2097                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2098                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2099                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2100                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2101                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2102                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2103                 tqc->ddLen          = cpu_to_le32(
2104                                         sizeof(struct vmxnet3_tx_buf_info) *
2105                                         tqc->txRingSize);
2106                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2107         }
2108
2109         /* rx queue settings */
2110         devRead->misc.numRxQueues = adapter->num_rx_queues;
2111         for (i = 0; i < adapter->num_rx_queues; i++) {
2112                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2113                 rqc = &adapter->rqd_start[i].conf;
2114                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2115                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2116                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2117                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2118                                                         rq->buf_info));
2119                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2120                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2121                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2122                 rqc->ddLen           = cpu_to_le32(
2123                                         sizeof(struct vmxnet3_rx_buf_info) *
2124                                         (rqc->rxRingSize[0] +
2125                                          rqc->rxRingSize[1]));
2126                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2127         }
2128
2129 #ifdef VMXNET3_RSS
2130         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2131
2132         if (adapter->rss) {
2133                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2134                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2135                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2136                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2137                                     UPT1_RSS_HASH_TYPE_IPV4 |
2138                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2139                                     UPT1_RSS_HASH_TYPE_IPV6;
2140                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2141                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2142                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2143                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2144                 for (i = 0; i < rssConf->indTableSize; i++)
2145                         rssConf->indTable[i] = i % adapter->num_rx_queues;
2146
2147                 devRead->rssConfDesc.confVer = 1;
2148                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2149                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2150         }
2151
2152 #endif /* VMXNET3_RSS */
2153
2154         /* intr settings */
2155         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2156                                      VMXNET3_IMM_AUTO;
2157         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2158         for (i = 0; i < adapter->intr.num_intrs; i++)
2159                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2160
2161         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2162         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2163
2164         /* rx filter settings */
2165         devRead->rxFilterConf.rxMode = 0;
2166         vmxnet3_restore_vlan(adapter);
2167         /* the rest are already zeroed */
2168 }
2169
2170
2171 int
2172 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2173 {
2174         int err, i;
2175         u32 ret;
2176
2177         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2178                 " ring sizes %u %u %u\n", adapter->netdev->name,
2179                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2180                 adapter->tx_queue[0].tx_ring.size,
2181                 adapter->rx_queue[0].rx_ring[0].size,
2182                 adapter->rx_queue[0].rx_ring[1].size);
2183
2184         vmxnet3_tq_init_all(adapter);
2185         err = vmxnet3_rq_init_all(adapter);
2186         if (err) {
2187                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2188                        adapter->netdev->name, err);
2189                 goto rq_err;
2190         }
2191
2192         err = vmxnet3_request_irqs(adapter);
2193         if (err) {
2194                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2195                        adapter->netdev->name, err);
2196                 goto irq_err;
2197         }
2198
2199         vmxnet3_setup_driver_shared(adapter);
2200
2201         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2202                                adapter->shared_pa));
2203         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2204                                adapter->shared_pa));
2205         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2206                                VMXNET3_CMD_ACTIVATE_DEV);
2207         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2208
2209         if (ret != 0) {
2210                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2211                        adapter->netdev->name, ret);
2212                 err = -EINVAL;
2213                 goto activate_err;
2214         }
2215
2216         for (i = 0; i < adapter->num_rx_queues; i++) {
2217                 VMXNET3_WRITE_BAR0_REG(adapter,
2218                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2219                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2220                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2221                                 (i * VMXNET3_REG_ALIGN)),
2222                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2223         }
2224
2225         /* Apply the rx filter settins last. */
2226         vmxnet3_set_mc(adapter->netdev);
2227
2228         /*
2229          * Check link state when first activating device. It will start the
2230          * tx queue if the link is up.
2231          */
2232         vmxnet3_check_link(adapter, true);
2233         for (i = 0; i < adapter->num_rx_queues; i++)
2234                 napi_enable(&adapter->rx_queue[i].napi);
2235         vmxnet3_enable_all_intrs(adapter);
2236         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2237         return 0;
2238
2239 activate_err:
2240         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2241         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2242         vmxnet3_free_irqs(adapter);
2243 irq_err:
2244 rq_err:
2245         /* free up buffers we allocated */
2246         vmxnet3_rq_cleanup_all(adapter);
2247         return err;
2248 }
2249
2250
2251 void
2252 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2253 {
2254         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2255 }
2256
2257
2258 int
2259 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2260 {
2261         int i;
2262         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2263                 return 0;
2264
2265
2266         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2267                                VMXNET3_CMD_QUIESCE_DEV);
2268         vmxnet3_disable_all_intrs(adapter);
2269
2270         for (i = 0; i < adapter->num_rx_queues; i++)
2271                 napi_disable(&adapter->rx_queue[i].napi);
2272         netif_tx_disable(adapter->netdev);
2273         adapter->link_speed = 0;
2274         netif_carrier_off(adapter->netdev);
2275
2276         vmxnet3_tq_cleanup_all(adapter);
2277         vmxnet3_rq_cleanup_all(adapter);
2278         vmxnet3_free_irqs(adapter);
2279         return 0;
2280 }
2281
2282
2283 static void
2284 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2285 {
2286         u32 tmp;
2287
2288         tmp = *(u32 *)mac;
2289         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2290
2291         tmp = (mac[5] << 8) | mac[4];
2292         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2293 }
2294
2295
2296 static int
2297 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2298 {
2299         struct sockaddr *addr = p;
2300         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2301
2302         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2303         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2304
2305         return 0;
2306 }
2307
2308
2309 /* ==================== initialization and cleanup routines ============ */
2310
2311 static int
2312 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2313 {
2314         int err;
2315         unsigned long mmio_start, mmio_len;
2316         struct pci_dev *pdev = adapter->pdev;
2317
2318         err = pci_enable_device(pdev);
2319         if (err) {
2320                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2321                        pci_name(pdev), err);
2322                 return err;
2323         }
2324
2325         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2326                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2327                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2328                                "for adapter %s\n", pci_name(pdev));
2329                         err = -EIO;
2330                         goto err_set_mask;
2331                 }
2332                 *dma64 = true;
2333         } else {
2334                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2335                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2336                                "%s\n",  pci_name(pdev));
2337                         err = -EIO;
2338                         goto err_set_mask;
2339                 }
2340                 *dma64 = false;
2341         }
2342
2343         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2344                                            vmxnet3_driver_name);
2345         if (err) {
2346                 printk(KERN_ERR "Failed to request region for adapter %s: "
2347                        "error %d\n", pci_name(pdev), err);
2348                 goto err_set_mask;
2349         }
2350
2351         pci_set_master(pdev);
2352
2353         mmio_start = pci_resource_start(pdev, 0);
2354         mmio_len = pci_resource_len(pdev, 0);
2355         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2356         if (!adapter->hw_addr0) {
2357                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2358                        pci_name(pdev));
2359                 err = -EIO;
2360                 goto err_ioremap;
2361         }
2362
2363         mmio_start = pci_resource_start(pdev, 1);
2364         mmio_len = pci_resource_len(pdev, 1);
2365         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2366         if (!adapter->hw_addr1) {
2367                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2368                        pci_name(pdev));
2369                 err = -EIO;
2370                 goto err_bar1;
2371         }
2372         return 0;
2373
2374 err_bar1:
2375         iounmap(adapter->hw_addr0);
2376 err_ioremap:
2377         pci_release_selected_regions(pdev, (1 << 2) - 1);
2378 err_set_mask:
2379         pci_disable_device(pdev);
2380         return err;
2381 }
2382
2383
2384 static void
2385 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2386 {
2387         BUG_ON(!adapter->pdev);
2388
2389         iounmap(adapter->hw_addr0);
2390         iounmap(adapter->hw_addr1);
2391         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2392         pci_disable_device(adapter->pdev);
2393 }
2394
2395
2396 static void
2397 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2398 {
2399         size_t sz, i, ring0_size, ring1_size, comp_size;
2400         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2401
2402
2403         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2404                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2405                 adapter->skb_buf_size = adapter->netdev->mtu +
2406                                         VMXNET3_MAX_ETH_HDR_SIZE;
2407                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2408                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2409
2410                 adapter->rx_buf_per_pkt = 1;
2411         } else {
2412                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2413                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2414                                             VMXNET3_MAX_ETH_HDR_SIZE;
2415                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2416         }
2417
2418         /*
2419          * for simplicity, force the ring0 size to be a multiple of
2420          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2421          */
2422         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2423         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2424         ring0_size = (ring0_size + sz - 1) / sz * sz;
2425         ring0_size = min_t(u32, rq->rx_ring[0].size, VMXNET3_RX_RING_MAX_SIZE /
2426                            sz * sz);
2427         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2428         comp_size = ring0_size + ring1_size;
2429
2430         for (i = 0; i < adapter->num_rx_queues; i++) {
2431                 rq = &adapter->rx_queue[i];
2432                 rq->rx_ring[0].size = ring0_size;
2433                 rq->rx_ring[1].size = ring1_size;
2434                 rq->comp_ring.size = comp_size;
2435         }
2436 }
2437
2438
2439 int
2440 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2441                       u32 rx_ring_size, u32 rx_ring2_size)
2442 {
2443         int err = 0, i;
2444
2445         for (i = 0; i < adapter->num_tx_queues; i++) {
2446                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2447                 tq->tx_ring.size   = tx_ring_size;
2448                 tq->data_ring.size = tx_ring_size;
2449                 tq->comp_ring.size = tx_ring_size;
2450                 tq->shared = &adapter->tqd_start[i].ctrl;
2451                 tq->stopped = true;
2452                 tq->adapter = adapter;
2453                 tq->qid = i;
2454                 err = vmxnet3_tq_create(tq, adapter);
2455                 /*
2456                  * Too late to change num_tx_queues. We cannot do away with
2457                  * lesser number of queues than what we asked for
2458                  */
2459                 if (err)
2460                         goto queue_err;
2461         }
2462
2463         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2464         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2465         vmxnet3_adjust_rx_ring_size(adapter);
2466         for (i = 0; i < adapter->num_rx_queues; i++) {
2467                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2468                 /* qid and qid2 for rx queues will be assigned later when num
2469                  * of rx queues is finalized after allocating intrs */
2470                 rq->shared = &adapter->rqd_start[i].ctrl;
2471                 rq->adapter = adapter;
2472                 err = vmxnet3_rq_create(rq, adapter);
2473                 if (err) {
2474                         if (i == 0) {
2475                                 printk(KERN_ERR "Could not allocate any rx"
2476                                        "queues. Aborting.\n");
2477                                 goto queue_err;
2478                         } else {
2479                                 printk(KERN_INFO "Number of rx queues changed "
2480                                        "to : %d.\n", i);
2481                                 adapter->num_rx_queues = i;
2482                                 err = 0;
2483                                 break;
2484                         }
2485                 }
2486         }
2487         return err;
2488 queue_err:
2489         vmxnet3_tq_destroy_all(adapter);
2490         return err;
2491 }
2492
2493 static int
2494 vmxnet3_open(struct net_device *netdev)
2495 {
2496         struct vmxnet3_adapter *adapter;
2497         int err, i;
2498
2499         adapter = netdev_priv(netdev);
2500
2501         for (i = 0; i < adapter->num_tx_queues; i++)
2502                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2503
2504         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2505                                     VMXNET3_DEF_RX_RING_SIZE,
2506                                     VMXNET3_DEF_RX_RING_SIZE);
2507         if (err)
2508                 goto queue_err;
2509
2510         err = vmxnet3_activate_dev(adapter);
2511         if (err)
2512                 goto activate_err;
2513
2514         return 0;
2515
2516 activate_err:
2517         vmxnet3_rq_destroy_all(adapter);
2518         vmxnet3_tq_destroy_all(adapter);
2519 queue_err:
2520         return err;
2521 }
2522
2523
2524 static int
2525 vmxnet3_close(struct net_device *netdev)
2526 {
2527         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2528
2529         /*
2530          * Reset_work may be in the middle of resetting the device, wait for its
2531          * completion.
2532          */
2533         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2534                 msleep(1);
2535
2536         vmxnet3_quiesce_dev(adapter);
2537
2538         vmxnet3_rq_destroy_all(adapter);
2539         vmxnet3_tq_destroy_all(adapter);
2540
2541         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2542
2543
2544         return 0;
2545 }
2546
2547
2548 void
2549 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2550 {
2551         int i;
2552
2553         /*
2554          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2555          * vmxnet3_close() will deadlock.
2556          */
2557         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2558
2559         /* we need to enable NAPI, otherwise dev_close will deadlock */
2560         for (i = 0; i < adapter->num_rx_queues; i++)
2561                 napi_enable(&adapter->rx_queue[i].napi);
2562         dev_close(adapter->netdev);
2563 }
2564
2565
2566 static int
2567 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2568 {
2569         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2570         int err = 0;
2571
2572         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2573                 return -EINVAL;
2574
2575         if (new_mtu > 1500 && !adapter->jumbo_frame)
2576                 return -EINVAL;
2577
2578         netdev->mtu = new_mtu;
2579
2580         /*
2581          * Reset_work may be in the middle of resetting the device, wait for its
2582          * completion.
2583          */
2584         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2585                 msleep(1);
2586
2587         if (netif_running(netdev)) {
2588                 vmxnet3_quiesce_dev(adapter);
2589                 vmxnet3_reset_dev(adapter);
2590
2591                 /* we need to re-create the rx queue based on the new mtu */
2592                 vmxnet3_rq_destroy_all(adapter);
2593                 vmxnet3_adjust_rx_ring_size(adapter);
2594                 err = vmxnet3_rq_create_all(adapter);
2595                 if (err) {
2596                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2597                                 " error %d. Closing it.\n", netdev->name, err);
2598                         goto out;
2599                 }
2600
2601                 err = vmxnet3_activate_dev(adapter);
2602                 if (err) {
2603                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2604                                 "Closing it\n", netdev->name, err);
2605                         goto out;
2606                 }
2607         }
2608
2609 out:
2610         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2611         if (err)
2612                 vmxnet3_force_close(adapter);
2613
2614         return err;
2615 }
2616
2617
2618 static void
2619 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2620 {
2621         struct net_device *netdev = adapter->netdev;
2622
2623         netdev->features = NETIF_F_SG |
2624                 NETIF_F_HW_CSUM |
2625                 NETIF_F_HW_VLAN_TX |
2626                 NETIF_F_HW_VLAN_RX |
2627                 NETIF_F_HW_VLAN_FILTER |
2628                 NETIF_F_TSO |
2629                 NETIF_F_TSO6 |
2630                 NETIF_F_LRO;
2631
2632         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2633
2634         adapter->rxcsum = true;
2635         adapter->jumbo_frame = true;
2636         adapter->lro = true;
2637
2638         if (dma64) {
2639                 netdev->features |= NETIF_F_HIGHDMA;
2640                 printk(" highDMA");
2641         }
2642
2643         netdev->vlan_features = netdev->features;
2644         printk("\n");
2645 }
2646
2647
2648 static void
2649 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2650 {
2651         u32 tmp;
2652
2653         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2654         *(u32 *)mac = tmp;
2655
2656         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2657         mac[4] = tmp & 0xff;
2658         mac[5] = (tmp >> 8) & 0xff;
2659 }
2660
2661 #ifdef CONFIG_PCI_MSI
2662
2663 /*
2664  * Enable MSIx vectors.
2665  * Returns :
2666  *      0 on successful enabling of required vectors,
2667  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minumum number of vectors required
2668  *       could be enabled.
2669  *      number of vectors which can be enabled otherwise (this number is smaller
2670  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2671  */
2672
2673 static int
2674 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2675                              int vectors)
2676 {
2677         int err = 0, vector_threshold;
2678         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2679
2680         while (vectors >= vector_threshold) {
2681                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2682                                       vectors);
2683                 if (!err) {
2684                         adapter->intr.num_intrs = vectors;
2685                         return 0;
2686                 } else if (err < 0) {
2687                         printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2688                                " %d\n", adapter->netdev->name, err);
2689                         vectors = 0;
2690                 } else if (err < vector_threshold) {
2691                         break;
2692                 } else {
2693                         /* If fails to enable required number of MSI-x vectors
2694                          * try enabling 3 of them. One each for rx, tx and event
2695                          */
2696                         vectors = vector_threshold;
2697                         printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2698                                " %d instead\n", vectors, adapter->netdev->name,
2699                                vector_threshold);
2700                 }
2701         }
2702
2703         printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2704                " are lower than min threshold required.\n");
2705         return err;
2706 }
2707
2708
2709 #endif /* CONFIG_PCI_MSI */
2710
2711 static void
2712 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2713 {
2714         u32 cfg;
2715
2716         /* intr settings */
2717         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2718                                VMXNET3_CMD_GET_CONF_INTR);
2719         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2720         adapter->intr.type = cfg & 0x3;
2721         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2722
2723         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2724                 adapter->intr.type = VMXNET3_IT_MSIX;
2725         }
2726
2727 #ifdef CONFIG_PCI_MSI
2728         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2729                 int vector, err = 0;
2730
2731                 adapter->intr.num_intrs = (adapter->share_intr ==
2732                                            VMXNET3_INTR_TXSHARE) ? 1 :
2733                                            adapter->num_tx_queues;
2734                 adapter->intr.num_intrs += (adapter->share_intr ==
2735                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2736                                            adapter->num_rx_queues;
2737                 adapter->intr.num_intrs += 1;           /* for link event */
2738
2739                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2740                                            VMXNET3_LINUX_MIN_MSIX_VECT
2741                                            ? adapter->intr.num_intrs :
2742                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2743
2744                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2745                         adapter->intr.msix_entries[vector].entry = vector;
2746
2747                 err = vmxnet3_acquire_msix_vectors(adapter,
2748                                                    adapter->intr.num_intrs);
2749                 /* If we cannot allocate one MSIx vector per queue
2750                  * then limit the number of rx queues to 1
2751                  */
2752                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2753                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2754                             || adapter->num_rx_queues != 2) {
2755                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2756                                 printk(KERN_ERR "Number of rx queues : 1\n");
2757                                 adapter->num_rx_queues = 1;
2758                                 adapter->intr.num_intrs =
2759                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2760                         }
2761                         return;
2762                 }
2763                 if (!err)
2764                         return;
2765
2766                 /* If we cannot allocate MSIx vectors use only one rx queue */
2767                 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2768                        "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2769
2770                 adapter->intr.type = VMXNET3_IT_MSI;
2771         }
2772
2773         if (adapter->intr.type == VMXNET3_IT_MSI) {
2774                 int err;
2775                 err = pci_enable_msi(adapter->pdev);
2776                 if (!err) {
2777                         adapter->num_rx_queues = 1;
2778                         adapter->intr.num_intrs = 1;
2779                         return;
2780                 }
2781         }
2782 #endif /* CONFIG_PCI_MSI */
2783
2784         adapter->num_rx_queues = 1;
2785         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2786         adapter->intr.type = VMXNET3_IT_INTX;
2787
2788         /* INT-X related setting */
2789         adapter->intr.num_intrs = 1;
2790 }
2791
2792
2793 static void
2794 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2795 {
2796         if (adapter->intr.type == VMXNET3_IT_MSIX)
2797                 pci_disable_msix(adapter->pdev);
2798         else if (adapter->intr.type == VMXNET3_IT_MSI)
2799                 pci_disable_msi(adapter->pdev);
2800         else
2801                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2802 }
2803
2804
2805 static void
2806 vmxnet3_tx_timeout(struct net_device *netdev)
2807 {
2808         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2809         adapter->tx_timeout_count++;
2810
2811         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2812         schedule_work(&adapter->work);
2813         netif_wake_queue(adapter->netdev);
2814 }
2815
2816
2817 static void
2818 vmxnet3_reset_work(struct work_struct *data)
2819 {
2820         struct vmxnet3_adapter *adapter;
2821
2822         adapter = container_of(data, struct vmxnet3_adapter, work);
2823
2824         /* if another thread is resetting the device, no need to proceed */
2825         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2826                 return;
2827
2828         /* if the device is closed, we must leave it alone */
2829         rtnl_lock();
2830         if (netif_running(adapter->netdev)) {
2831                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2832                 vmxnet3_quiesce_dev(adapter);
2833                 vmxnet3_reset_dev(adapter);
2834                 vmxnet3_activate_dev(adapter);
2835         } else {
2836                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2837         }
2838         rtnl_unlock();
2839
2840         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2841 }
2842
2843
2844 static int __devinit
2845 vmxnet3_probe_device(struct pci_dev *pdev,
2846                      const struct pci_device_id *id)
2847 {
2848         static const struct net_device_ops vmxnet3_netdev_ops = {
2849                 .ndo_open = vmxnet3_open,
2850                 .ndo_stop = vmxnet3_close,
2851                 .ndo_start_xmit = vmxnet3_xmit_frame,
2852                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2853                 .ndo_change_mtu = vmxnet3_change_mtu,
2854                 .ndo_get_stats = vmxnet3_get_stats,
2855                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2856                 .ndo_set_multicast_list = vmxnet3_set_mc,
2857                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2858                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2859                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2860 #ifdef CONFIG_NET_POLL_CONTROLLER
2861                 .ndo_poll_controller = vmxnet3_netpoll,
2862 #endif
2863         };
2864         int err;
2865         bool dma64 = false; /* stupid gcc */
2866         u32 ver;
2867         struct net_device *netdev;
2868         struct vmxnet3_adapter *adapter;
2869         u8 mac[ETH_ALEN];
2870         int size;
2871         int num_tx_queues;
2872         int num_rx_queues;
2873
2874 #ifdef VMXNET3_RSS
2875         if (enable_mq)
2876                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2877                                     (int)num_online_cpus());
2878         else
2879 #endif
2880                 num_rx_queues = 1;
2881
2882         if (enable_mq)
2883                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2884                                     (int)num_online_cpus());
2885         else
2886                 num_tx_queues = 1;
2887
2888         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2889                                    max(num_tx_queues, num_rx_queues));
2890         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2891                num_tx_queues, num_rx_queues);
2892
2893         if (!netdev) {
2894                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2895                         "%s\n", pci_name(pdev));
2896                 return -ENOMEM;
2897         }
2898
2899         pci_set_drvdata(pdev, netdev);
2900         adapter = netdev_priv(netdev);
2901         adapter->netdev = netdev;
2902         adapter->pdev = pdev;
2903
2904         adapter->shared = pci_alloc_consistent(adapter->pdev,
2905                           sizeof(struct Vmxnet3_DriverShared),
2906                           &adapter->shared_pa);
2907         if (!adapter->shared) {
2908                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2909                         pci_name(pdev));
2910                 err = -ENOMEM;
2911                 goto err_alloc_shared;
2912         }
2913
2914         adapter->num_rx_queues = num_rx_queues;
2915         adapter->num_tx_queues = num_tx_queues;
2916
2917         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2918         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2919         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2920                              &adapter->queue_desc_pa);
2921
2922         if (!adapter->tqd_start) {
2923                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2924                         pci_name(pdev));
2925                 err = -ENOMEM;
2926                 goto err_alloc_queue_desc;
2927         }
2928         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2929                                                         adapter->num_tx_queues);
2930
2931         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2932         if (adapter->pm_conf == NULL) {
2933                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2934                         pci_name(pdev));
2935                 err = -ENOMEM;
2936                 goto err_alloc_pm;
2937         }
2938
2939 #ifdef VMXNET3_RSS
2940
2941         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2942         if (adapter->rss_conf == NULL) {
2943                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2944                        pci_name(pdev));
2945                 err = -ENOMEM;
2946                 goto err_alloc_rss;
2947         }
2948 #endif /* VMXNET3_RSS */
2949
2950         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2951         if (err < 0)
2952                 goto err_alloc_pci;
2953
2954         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2955         if (ver & 1) {
2956                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2957         } else {
2958                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2959                        " %s\n", ver, pci_name(pdev));
2960                 err = -EBUSY;
2961                 goto err_ver;
2962         }
2963
2964         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2965         if (ver & 1) {
2966                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2967         } else {
2968                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2969                        "adapter %s\n", ver, pci_name(pdev));
2970                 err = -EBUSY;
2971                 goto err_ver;
2972         }
2973
2974         vmxnet3_declare_features(adapter, dma64);
2975
2976         adapter->dev_number = atomic_read(&devices_found);
2977
2978          adapter->share_intr = irq_share_mode;
2979         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
2980             adapter->num_tx_queues != adapter->num_rx_queues)
2981                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
2982
2983         vmxnet3_alloc_intr_resources(adapter);
2984
2985 #ifdef VMXNET3_RSS
2986         if (adapter->num_rx_queues > 1 &&
2987             adapter->intr.type == VMXNET3_IT_MSIX) {
2988                 adapter->rss = true;
2989                 printk(KERN_INFO "RSS is enabled.\n");
2990         } else {
2991                 adapter->rss = false;
2992         }
2993 #endif
2994
2995         vmxnet3_read_mac_addr(adapter, mac);
2996         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
2997
2998         netdev->netdev_ops = &vmxnet3_netdev_ops;
2999         vmxnet3_set_ethtool_ops(netdev);
3000         netdev->watchdog_timeo = 5 * HZ;
3001
3002         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3003
3004         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3005                 int i;
3006                 for (i = 0; i < adapter->num_rx_queues; i++) {
3007                         netif_napi_add(adapter->netdev,
3008                                        &adapter->rx_queue[i].napi,
3009                                        vmxnet3_poll_rx_only, 64);
3010                 }
3011         } else {
3012                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3013                                vmxnet3_poll, 64);
3014         }
3015
3016         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3017         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3018
3019         SET_NETDEV_DEV(netdev, &pdev->dev);
3020         err = register_netdev(netdev);
3021
3022         if (err) {
3023                 printk(KERN_ERR "Failed to register adapter %s\n",
3024                         pci_name(pdev));
3025                 goto err_register;
3026         }
3027
3028         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3029         vmxnet3_check_link(adapter, false);
3030         atomic_inc(&devices_found);
3031         return 0;
3032
3033 err_register:
3034         vmxnet3_free_intr_resources(adapter);
3035 err_ver:
3036         vmxnet3_free_pci_resources(adapter);
3037 err_alloc_pci:
3038 #ifdef VMXNET3_RSS
3039         kfree(adapter->rss_conf);
3040 err_alloc_rss:
3041 #endif
3042         kfree(adapter->pm_conf);
3043 err_alloc_pm:
3044         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3045                             adapter->queue_desc_pa);
3046 err_alloc_queue_desc:
3047         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3048                             adapter->shared, adapter->shared_pa);
3049 err_alloc_shared:
3050         pci_set_drvdata(pdev, NULL);
3051         free_netdev(netdev);
3052         return err;
3053 }
3054
3055
3056 static void __devexit
3057 vmxnet3_remove_device(struct pci_dev *pdev)
3058 {
3059         struct net_device *netdev = pci_get_drvdata(pdev);
3060         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3061         int size = 0;
3062         int num_rx_queues;
3063
3064 #ifdef VMXNET3_RSS
3065         if (enable_mq)
3066                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3067                                     (int)num_online_cpus());
3068         else
3069 #endif
3070                 num_rx_queues = 1;
3071
3072         cancel_work_sync(&adapter->work);
3073
3074         unregister_netdev(netdev);
3075
3076         vmxnet3_free_intr_resources(adapter);
3077         vmxnet3_free_pci_resources(adapter);
3078 #ifdef VMXNET3_RSS
3079         kfree(adapter->rss_conf);
3080 #endif
3081         kfree(adapter->pm_conf);
3082
3083         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3084         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3085         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3086                             adapter->queue_desc_pa);
3087         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3088                             adapter->shared, adapter->shared_pa);
3089         free_netdev(netdev);
3090 }
3091
3092
3093 #ifdef CONFIG_PM
3094
3095 static int
3096 vmxnet3_suspend(struct device *device)
3097 {
3098         struct pci_dev *pdev = to_pci_dev(device);
3099         struct net_device *netdev = pci_get_drvdata(pdev);
3100         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3101         struct Vmxnet3_PMConf *pmConf;
3102         struct ethhdr *ehdr;
3103         struct arphdr *ahdr;
3104         u8 *arpreq;
3105         struct in_device *in_dev;
3106         struct in_ifaddr *ifa;
3107         int i = 0;
3108
3109         if (!netif_running(netdev))
3110                 return 0;
3111
3112         vmxnet3_disable_all_intrs(adapter);
3113         vmxnet3_free_irqs(adapter);
3114         vmxnet3_free_intr_resources(adapter);
3115
3116         netif_device_detach(netdev);
3117         netif_tx_stop_all_queues(netdev);
3118
3119         /* Create wake-up filters. */
3120         pmConf = adapter->pm_conf;
3121         memset(pmConf, 0, sizeof(*pmConf));
3122
3123         if (adapter->wol & WAKE_UCAST) {
3124                 pmConf->filters[i].patternSize = ETH_ALEN;
3125                 pmConf->filters[i].maskSize = 1;
3126                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3127                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3128
3129                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3130                 i++;
3131         }
3132
3133         if (adapter->wol & WAKE_ARP) {
3134                 in_dev = in_dev_get(netdev);
3135                 if (!in_dev)
3136                         goto skip_arp;
3137
3138                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3139                 if (!ifa)
3140                         goto skip_arp;
3141
3142                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3143                         sizeof(struct arphdr) +         /* ARP header */
3144                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3145                         2 * sizeof(u32);        /*2 IPv4 addresses */
3146                 pmConf->filters[i].maskSize =
3147                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3148
3149                 /* ETH_P_ARP in Ethernet header. */
3150                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3151                 ehdr->h_proto = htons(ETH_P_ARP);
3152
3153                 /* ARPOP_REQUEST in ARP header. */
3154                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3155                 ahdr->ar_op = htons(ARPOP_REQUEST);
3156                 arpreq = (u8 *)(ahdr + 1);
3157
3158                 /* The Unicast IPv4 address in 'tip' field. */
3159                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3160                 *(u32 *)arpreq = ifa->ifa_address;
3161
3162                 /* The mask for the relevant bits. */
3163                 pmConf->filters[i].mask[0] = 0x00;
3164                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3165                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3166                 pmConf->filters[i].mask[3] = 0x00;
3167                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3168                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3169                 in_dev_put(in_dev);
3170
3171                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3172                 i++;
3173         }
3174
3175 skip_arp:
3176         if (adapter->wol & WAKE_MAGIC)
3177                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3178
3179         pmConf->numFilters = i;
3180
3181         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3182         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3183                                                                   *pmConf));
3184         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3185                                                                  pmConf));
3186
3187         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3188                                VMXNET3_CMD_UPDATE_PMCFG);
3189
3190         pci_save_state(pdev);
3191         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3192                         adapter->wol);
3193         pci_disable_device(pdev);
3194         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3195
3196         return 0;
3197 }
3198
3199
3200 static int
3201 vmxnet3_resume(struct device *device)
3202 {
3203         int err;
3204         struct pci_dev *pdev = to_pci_dev(device);
3205         struct net_device *netdev = pci_get_drvdata(pdev);
3206         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3207         struct Vmxnet3_PMConf *pmConf;
3208
3209         if (!netif_running(netdev))
3210                 return 0;
3211
3212         /* Destroy wake-up filters. */
3213         pmConf = adapter->pm_conf;
3214         memset(pmConf, 0, sizeof(*pmConf));
3215
3216         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3217         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3218                                                                   *pmConf));
3219         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3220                                                                  pmConf));
3221
3222         netif_device_attach(netdev);
3223         pci_set_power_state(pdev, PCI_D0);
3224         pci_restore_state(pdev);
3225         err = pci_enable_device_mem(pdev);
3226         if (err != 0)
3227                 return err;
3228
3229         pci_enable_wake(pdev, PCI_D0, 0);
3230
3231         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3232                                VMXNET3_CMD_UPDATE_PMCFG);
3233         vmxnet3_alloc_intr_resources(adapter);
3234         vmxnet3_request_irqs(adapter);
3235         vmxnet3_enable_all_intrs(adapter);
3236
3237         return 0;
3238 }
3239
3240 static const struct dev_pm_ops vmxnet3_pm_ops = {
3241         .suspend = vmxnet3_suspend,
3242         .resume = vmxnet3_resume,
3243 };
3244 #endif
3245
3246 static struct pci_driver vmxnet3_driver = {
3247         .name           = vmxnet3_driver_name,
3248         .id_table       = vmxnet3_pciid_table,
3249         .probe          = vmxnet3_probe_device,
3250         .remove         = __devexit_p(vmxnet3_remove_device),
3251 #ifdef CONFIG_PM
3252         .driver.pm      = &vmxnet3_pm_ops,
3253 #endif
3254 };
3255
3256
3257 static int __init
3258 vmxnet3_init_module(void)
3259 {
3260         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3261                 VMXNET3_DRIVER_VERSION_REPORT);
3262         return pci_register_driver(&vmxnet3_driver);
3263 }
3264
3265 module_init(vmxnet3_init_module);
3266
3267
3268 static void
3269 vmxnet3_exit_module(void)
3270 {
3271         pci_unregister_driver(&vmxnet3_driver);
3272 }
3273
3274 module_exit(vmxnet3_exit_module);
3275
3276 MODULE_AUTHOR("VMware, Inc.");
3277 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3278 MODULE_LICENSE("GPL v2");
3279 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);