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
  48 /*
  49  *    Enable/Disable the given intr
  50  */
  51 static void
  52 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  53 {
  54         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
  55 }
  56
  57
  58 static void
  59 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  60 {
  61         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
  62 }
  63
  64
  65 /*
  66  *    Enable/Disable all intrs used by the device
  67  */
  68 static void
  69 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
  70 {
  71         int i;
  72
  73         for (i = 0; i < adapter->intr.num_intrs; i++)
  74                 vmxnet3_enable_intr(adapter, i);
  75         adapter->shared->devRead.intrConf.intrCtrl &=
  76                                         cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
  77 }
  78
  79
  80 static void
  81 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
  82 {
  83         int i;
  84
  85         adapter->shared->devRead.intrConf.intrCtrl |=
  86                                         cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
  87         for (i = 0; i < adapter->intr.num_intrs; i++)
  88                 vmxnet3_disable_intr(adapter, i);
  89 }
  90
  91
  92 static void
  93 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
  94 {
  95         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
  96 }
  97
  98
  99 static bool
 100 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 101 {
 102         return netif_queue_stopped(adapter->netdev);
 103 }
 104
 105
 106 static void
 107 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 108 {
 109         tq->stopped = false;
 110         netif_start_queue(adapter->netdev);
 111 }
 112
 113
 114 static void
 115 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 116 {
 117         tq->stopped = false;
 118         netif_wake_queue(adapter->netdev);
 119 }
 120
 121
 122 static void
 123 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 124 {
 125         tq->stopped = true;
 126         tq->num_stop++;
 127         netif_stop_queue(adapter->netdev);
 128 }
 129
 130
 131 /*
 132  * Check the link state. This may start or stop the tx queue.
 133  */
 134 static void
 135 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
 136 {
 137         u32 ret;
 138
 139         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
 140         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
 141         adapter->link_speed = ret >> 16;
 142         if (ret & 1) { /* Link is up. */
 143                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
 144                        adapter->netdev->name, adapter->link_speed);
 145                 if (!netif_carrier_ok(adapter->netdev))
 146                         netif_carrier_on(adapter->netdev);
 147
 148                 if (affectTxQueue)
 149                         vmxnet3_tq_start(&adapter->tx_queue, adapter);
 150         } else {
 151                 printk(KERN_INFO "%s: NIC Link is Down\n",
 152                        adapter->netdev->name);
 153                 if (netif_carrier_ok(adapter->netdev))
 154                         netif_carrier_off(adapter->netdev);
 155
 156                 if (affectTxQueue)
 157                         vmxnet3_tq_stop(&adapter->tx_queue, adapter);
 158         }
 159 }
 160
 161 static void
 162 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 163 {
 164         u32 events = le32_to_cpu(adapter->shared->ecr);
 165         if (!events)
 166                 return;
 167
 168         vmxnet3_ack_events(adapter, events);
 169
 170         /* Check if link state has changed */
 171         if (events & VMXNET3_ECR_LINK)
 172                 vmxnet3_check_link(adapter, true);
 173
 174         /* Check if there is an error on xmit/recv queues */
 175         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 176                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 177                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 178
 179                 if (adapter->tqd_start->status.stopped) {
 180                         printk(KERN_ERR "%s: tq error 0x%x\n",
 181                                adapter->netdev->name,
 182                                le32_to_cpu(adapter->tqd_start->status.error));
 183                 }
 184                 if (adapter->rqd_start->status.stopped) {
 185                         printk(KERN_ERR "%s: rq error 0x%x\n",
 186                                adapter->netdev->name,
 187                                adapter->rqd_start->status.error);
 188                 }
 189
 190                 schedule_work(&adapter->work);
 191         }
 192 }
 193
 194 #ifdef __BIG_ENDIAN_BITFIELD
 195 /*
 196  * The device expects the bitfields in shared structures to be written in
 197  * little endian. When CPU is big endian, the following routines are used to
 198  * correctly read and write into ABI.
 199  * The general technique used here is : double word bitfields are defined in
 200  * opposite order for big endian architecture. Then before reading them in
 201  * driver the complete double word is translated using le32_to_cpu. Similarly
 202  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 203  * double words into required format.
 204  * In order to avoid touching bits in shared structure more than once, temporary
 205  * descriptors are used. These are passed as srcDesc to following functions.
 206  */
 207 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 208                                 struct Vmxnet3_RxDesc *dstDesc)
 209 {
 210         u32 *src = (u32 *)srcDesc + 2;
 211         u32 *dst = (u32 *)dstDesc + 2;
 212         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 213         *dst = le32_to_cpu(*src);
 214         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 215 }
 216
 217 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 218                                struct Vmxnet3_TxDesc *dstDesc)
 219 {
 220         int i;
 221         u32 *src = (u32 *)(srcDesc + 1);
 222         u32 *dst = (u32 *)(dstDesc + 1);
 223
 224         /* Working backwards so that the gen bit is set at the end. */
 225         for (i = 2; i > 0; i--) {
 226                 src--;
 227                 dst--;
 228                 *dst = cpu_to_le32(*src);
 229         }
 230 }
 231
 232
 233 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 234                                 struct Vmxnet3_RxCompDesc *dstDesc)
 235 {
 236         int i = 0;
 237         u32 *src = (u32 *)srcDesc;
 238         u32 *dst = (u32 *)dstDesc;
 239         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 240                 *dst = le32_to_cpu(*src);
 241                 src++;
 242                 dst++;
 243         }
 244 }
 245
 246
 247 /* Used to read bitfield values from double words. */
 248 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 249 {
 250         u32 temp = le32_to_cpu(*bitfield);
 251         u32 mask = ((1 << size) - 1) << pos;
 252         temp &= mask;
 253         temp >>= pos;
 254         return temp;
 255 }
 256
 257
 258
 259 #endif  /* __BIG_ENDIAN_BITFIELD */
 260
 261 #ifdef __BIG_ENDIAN_BITFIELD
 262
 263 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 264                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 265                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 266 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 267                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 268                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 269 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 270                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 271                         VMXNET3_TCD_GEN_SIZE)
 272 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 273                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 274 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 275                         (dstrcd) = (tmp); \
 276                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 277                 } while (0)
 278 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 279                         (dstrxd) = (tmp); \
 280                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 281                 } while (0)
 282
 283 #else
 284
 285 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 286 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 287 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 288 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 289 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 290 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 291
 292 #endif /* __BIG_ENDIAN_BITFIELD  */
 293
 294
 295 static void
 296 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 297                      struct pci_dev *pdev)
 298 {
 299         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 300                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
 301                                  PCI_DMA_TODEVICE);
 302         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 303                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
 304                                PCI_DMA_TODEVICE);
 305         else
 306                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 307
 308         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 309 }
 310
 311
 312 static int
 313 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 314                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 315 {
 316         struct sk_buff *skb;
 317         int entries = 0;
 318
 319         /* no out of order completion */
 320         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 321         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 322
 323         skb = tq->buf_info[eop_idx].skb;
 324         BUG_ON(skb == NULL);
 325         tq->buf_info[eop_idx].skb = NULL;
 326
 327         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 328
 329         while (tq->tx_ring.next2comp != eop_idx) {
 330                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 331                                      pdev);
 332
 333                 /* update next2comp w/o tx_lock. Since we are marking more,
 334                  * instead of less, tx ring entries avail, the worst case is
 335                  * that the tx routine incorrectly re-queues a pkt due to
 336                  * insufficient tx ring entries.
 337                  */
 338                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 339                 entries++;
 340         }
 341
 342         dev_kfree_skb_any(skb);
 343         return entries;
 344 }
 345
 346
 347 static int
 348 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 349                         struct vmxnet3_adapter *adapter)
 350 {
 351         int completed = 0;
 352         union Vmxnet3_GenericDesc *gdesc;
 353
 354         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 355         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 356                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 357                                                &gdesc->tcd), tq, adapter->pdev,
 358                                                adapter);
 359
 360                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 361                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 362         }
 363
 364         if (completed) {
 365                 spin_lock(&tq->tx_lock);
 366                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 367                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 368                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 369                              netif_carrier_ok(adapter->netdev))) {
 370                         vmxnet3_tq_wake(tq, adapter);
 371                 }
 372                 spin_unlock(&tq->tx_lock);
 373         }
 374         return completed;
 375 }
 376
 377
 378 static void
 379 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 380                    struct vmxnet3_adapter *adapter)
 381 {
 382         int i;
 383
 384         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 385                 struct vmxnet3_tx_buf_info *tbi;
 386                 union Vmxnet3_GenericDesc *gdesc;
 387
 388                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 389                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
 390
 391                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 392                 if (tbi->skb) {
 393                         dev_kfree_skb_any(tbi->skb);
 394                         tbi->skb = NULL;
 395                 }
 396                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 397         }
 398
 399         /* sanity check, verify all buffers are indeed unmapped and freed */
 400         for (i = 0; i < tq->tx_ring.size; i++) {
 401                 BUG_ON(tq->buf_info[i].skb != NULL ||
 402                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 403         }
 404
 405         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 406         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 407
 408         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 409         tq->comp_ring.next2proc = 0;
 410 }
 411
 412
 413 void
 414 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 415                    struct vmxnet3_adapter *adapter)
 416 {
 417         if (tq->tx_ring.base) {
 418                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 419                                     sizeof(struct Vmxnet3_TxDesc),
 420                                     tq->tx_ring.base, tq->tx_ring.basePA);
 421                 tq->tx_ring.base = NULL;
 422         }
 423         if (tq->data_ring.base) {
 424                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 425                                     sizeof(struct Vmxnet3_TxDataDesc),
 426                                     tq->data_ring.base, tq->data_ring.basePA);
 427                 tq->data_ring.base = NULL;
 428         }
 429         if (tq->comp_ring.base) {
 430                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 431                                     sizeof(struct Vmxnet3_TxCompDesc),
 432                                     tq->comp_ring.base, tq->comp_ring.basePA);
 433                 tq->comp_ring.base = NULL;
 434         }
 435         kfree(tq->buf_info);
 436         tq->buf_info = NULL;
 437 }
 438
 439
 440 static void
 441 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 442                 struct vmxnet3_adapter *adapter)
 443 {
 444         int i;
 445
 446         /* reset the tx ring contents to 0 and reset the tx ring states */
 447         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 448                sizeof(struct Vmxnet3_TxDesc));
 449         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 450         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 451
 452         memset(tq->data_ring.base, 0, tq->data_ring.size *
 453                sizeof(struct Vmxnet3_TxDataDesc));
 454
 455         /* reset the tx comp ring contents to 0 and reset comp ring states */
 456         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 457                sizeof(struct Vmxnet3_TxCompDesc));
 458         tq->comp_ring.next2proc = 0;
 459         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 460
 461         /* reset the bookkeeping data */
 462         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 463         for (i = 0; i < tq->tx_ring.size; i++)
 464                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 465
 466         /* stats are not reset */
 467 }
 468
 469
 470 static int
 471 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 472                   struct vmxnet3_adapter *adapter)
 473 {
 474         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 475                tq->comp_ring.base || tq->buf_info);
 476
 477         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 478                            * sizeof(struct Vmxnet3_TxDesc),
 479                            &tq->tx_ring.basePA);
 480         if (!tq->tx_ring.base) {
 481                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 482                        adapter->netdev->name);
 483                 goto err;
 484         }
 485
 486         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 487                              tq->data_ring.size *
 488                              sizeof(struct Vmxnet3_TxDataDesc),
 489                              &tq->data_ring.basePA);
 490         if (!tq->data_ring.base) {
 491                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 492                        adapter->netdev->name);
 493                 goto err;
 494         }
 495
 496         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 497                              tq->comp_ring.size *
 498                              sizeof(struct Vmxnet3_TxCompDesc),
 499                              &tq->comp_ring.basePA);
 500         if (!tq->comp_ring.base) {
 501                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 502                        adapter->netdev->name);
 503                 goto err;
 504         }
 505
 506         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 507                                GFP_KERNEL);
 508         if (!tq->buf_info) {
 509                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
 510                        adapter->netdev->name);
 511                 goto err;
 512         }
 513
 514         return 0;
 515
 516 err:
 517         vmxnet3_tq_destroy(tq, adapter);
 518         return -ENOMEM;
 519 }
 520
 521
 522 /*
 523  *    starting from ring->next2fill, allocate rx buffers for the given ring
 524  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 525  *    are allocated or allocation fails
 526  */
 527
 528 static int
 529 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 530                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 531 {
 532         int num_allocated = 0;
 533         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 534         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 535         u32 val;
 536
 537         while (num_allocated < num_to_alloc) {
 538                 struct vmxnet3_rx_buf_info *rbi;
 539                 union Vmxnet3_GenericDesc *gd;
 540
 541                 rbi = rbi_base + ring->next2fill;
 542                 gd = ring->base + ring->next2fill;
 543
 544                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 545                         if (rbi->skb == NULL) {
 546                                 rbi->skb = dev_alloc_skb(rbi->len +
 547                                                          NET_IP_ALIGN);
 548                                 if (unlikely(rbi->skb == NULL)) {
 549                                         rq->stats.rx_buf_alloc_failure++;
 550                                         break;
 551                                 }
 552                                 rbi->skb->dev = adapter->netdev;
 553
 554                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 555                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 556                                                 rbi->skb->data, rbi->len,
 557                                                 PCI_DMA_FROMDEVICE);
 558                         } else {
 559                                 /* rx buffer skipped by the device */
 560                         }
 561                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 562                 } else {
 563                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 564                                rbi->len  != PAGE_SIZE);
 565
 566                         if (rbi->page == NULL) {
 567                                 rbi->page = alloc_page(GFP_ATOMIC);
 568                                 if (unlikely(rbi->page == NULL)) {
 569                                         rq->stats.rx_buf_alloc_failure++;
 570                                         break;
 571                                 }
 572                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 573                                                 rbi->page, 0, PAGE_SIZE,
 574                                                 PCI_DMA_FROMDEVICE);
 575                         } else {
 576                                 /* rx buffers skipped by the device */
 577                         }
 578                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 579                 }
 580
 581                 BUG_ON(rbi->dma_addr == 0);
 582                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 583                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
 584                                            | val | rbi->len);
 585
 586                 num_allocated++;
 587                 vmxnet3_cmd_ring_adv_next2fill(ring);
 588         }
 589         rq->uncommitted[ring_idx] += num_allocated;
 590
 591         dev_dbg(&adapter->netdev->dev,
 592                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 593                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
 594                 ring->next2comp, rq->uncommitted[ring_idx]);
 595
 596         /* so that the device can distinguish a full ring and an empty ring */
 597         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 598
 599         return num_allocated;
 600 }
 601
 602
 603 static void
 604 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 605                     struct vmxnet3_rx_buf_info *rbi)
 606 {
 607         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 608                 skb_shinfo(skb)->nr_frags;
 609
 610         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 611
 612         frag->page = rbi->page;
 613         frag->page_offset = 0;
 614         frag->size = rcd->len;
 615         skb->data_len += frag->size;
 616         skb_shinfo(skb)->nr_frags++;
 617 }
 618
 619
 620 static void
 621 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 622                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 623                 struct vmxnet3_adapter *adapter)
 624 {
 625         u32 dw2, len;
 626         unsigned long buf_offset;
 627         int i;
 628         union Vmxnet3_GenericDesc *gdesc;
 629         struct vmxnet3_tx_buf_info *tbi = NULL;
 630
 631         BUG_ON(ctx->copy_size > skb_headlen(skb));
 632
 633         /* use the previous gen bit for the SOP desc */
 634         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 635
 636         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 637         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 638
 639         /* no need to map the buffer if headers are copied */
 640         if (ctx->copy_size) {
 641                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 642                                         tq->tx_ring.next2fill *
 643                                         sizeof(struct Vmxnet3_TxDataDesc));
 644                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 645                 ctx->sop_txd->dword[3] = 0;
 646
 647                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 648                 tbi->map_type = VMXNET3_MAP_NONE;
 649
 650                 dev_dbg(&adapter->netdev->dev,
 651                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 652                         tq->tx_ring.next2fill,
 653                         le64_to_cpu(ctx->sop_txd->txd.addr),
 654                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 655                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 656
 657                 /* use the right gen for non-SOP desc */
 658                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 659         }
 660
 661         /* linear part can use multiple tx desc if it's big */
 662         len = skb_headlen(skb) - ctx->copy_size;
 663         buf_offset = ctx->copy_size;
 664         while (len) {
 665                 u32 buf_size;
 666
 667                 buf_size = len > VMXNET3_MAX_TX_BUF_SIZE ?
 668                            VMXNET3_MAX_TX_BUF_SIZE : len;
 669
 670                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 671                 tbi->map_type = VMXNET3_MAP_SINGLE;
 672                 tbi->dma_addr = pci_map_single(adapter->pdev,
 673                                 skb->data + buf_offset, buf_size,
 674                                 PCI_DMA_TODEVICE);
 675
 676                 tbi->len = buf_size; /* this automatically convert 2^14 to 0 */
 677
 678                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 679                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 680
 681                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 682                 gdesc->dword[2] = cpu_to_le32(dw2 | buf_size);
 683                 gdesc->dword[3] = 0;
 684
 685                 dev_dbg(&adapter->netdev->dev,
 686                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 687                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 688                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 689                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 690                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 691
 692                 len -= buf_size;
 693                 buf_offset += buf_size;
 694         }
 695
 696         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 697                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 698
 699                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 700                 tbi->map_type = VMXNET3_MAP_PAGE;
 701                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
 702                                              frag->page_offset, frag->size,
 703                                              PCI_DMA_TODEVICE);
 704
 705                 tbi->len = frag->size;
 706
 707                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 708                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 709
 710                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 711                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
 712                 gdesc->dword[3] = 0;
 713
 714                 dev_dbg(&adapter->netdev->dev,
 715                         "txd[%u]: 0x%llu %u %u\n",
 716                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 717                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 718                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 719                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 720         }
 721
 722         ctx->eop_txd = gdesc;
 723
 724         /* set the last buf_info for the pkt */
 725         tbi->skb = skb;
 726         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 727 }
 728
 729
 730 /*
 731  *    parse and copy relevant protocol headers:
 732  *      For a tso pkt, relevant headers are L2/3/4 including options
 733  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 734  *      if it's a TCP/UDP pkt
 735  *
 736  * Returns:
 737  *    -1:  error happens during parsing
 738  *     0:  protocol headers parsed, but too big to be copied
 739  *     1:  protocol headers parsed and copied
 740  *
 741  * Other effects:
 742  *    1. related *ctx fields are updated.
 743  *    2. ctx->copy_size is # of bytes copied
 744  *    3. the portion copied is guaranteed to be in the linear part
 745  *
 746  */
 747 static int
 748 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 749                            struct vmxnet3_tx_ctx *ctx,
 750                            struct vmxnet3_adapter *adapter)
 751 {
 752         struct Vmxnet3_TxDataDesc *tdd;
 753
 754         if (ctx->mss) {
 755                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 756                 ctx->l4_hdr_size = ((struct tcphdr *)
 757                                    skb_transport_header(skb))->doff * 4;
 758                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 759         } else {
 760                 unsigned int pull_size;
 761
 762                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 763                         ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 764
 765                         if (ctx->ipv4) {
 766                                 struct iphdr *iph = (struct iphdr *)
 767                                                     skb_network_header(skb);
 768                                 if (iph->protocol == IPPROTO_TCP) {
 769                                         pull_size = ctx->eth_ip_hdr_size +
 770                                                     sizeof(struct tcphdr);
 771
 772                                         if (unlikely(!pskb_may_pull(skb,
 773                                                                 pull_size))) {
 774                                                 goto err;
 775                                         }
 776                                         ctx->l4_hdr_size = ((struct tcphdr *)
 777                                            skb_transport_header(skb))->doff * 4;
 778                                 } else if (iph->protocol == IPPROTO_UDP) {
 779                                         ctx->l4_hdr_size =
 780                                                         sizeof(struct udphdr);
 781                                 } else {
 782                                         ctx->l4_hdr_size = 0;
 783                                 }
 784                         } else {
 785                                 /* for simplicity, don't copy L4 headers */
 786                                 ctx->l4_hdr_size = 0;
 787                         }
 788                         ctx->copy_size = ctx->eth_ip_hdr_size +
 789                                          ctx->l4_hdr_size;
 790                 } else {
 791                         ctx->eth_ip_hdr_size = 0;
 792                         ctx->l4_hdr_size = 0;
 793                         /* copy as much as allowed */
 794                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 795                                              , skb_headlen(skb));
 796                 }
 797
 798                 /* make sure headers are accessible directly */
 799                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 800                         goto err;
 801         }
 802
 803         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 804                 tq->stats.oversized_hdr++;
 805                 ctx->copy_size = 0;
 806                 return 0;
 807         }
 808
 809         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 810
 811         memcpy(tdd->data, skb->data, ctx->copy_size);
 812         dev_dbg(&adapter->netdev->dev,
 813                 "copy %u bytes to dataRing[%u]\n",
 814                 ctx->copy_size, tq->tx_ring.next2fill);
 815         return 1;
 816
 817 err:
 818         return -1;
 819 }
 820
 821
 822 static void
 823 vmxnet3_prepare_tso(struct sk_buff *skb,
 824                     struct vmxnet3_tx_ctx *ctx)
 825 {
 826         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
 827         if (ctx->ipv4) {
 828                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
 829                 iph->check = 0;
 830                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 831                                                  IPPROTO_TCP, 0);
 832         } else {
 833                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
 834                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 835                                                IPPROTO_TCP, 0);
 836         }
 837 }
 838
 839
 840 /*
 841  * Transmits a pkt thru a given tq
 842  * Returns:
 843  *    NETDEV_TX_OK:      descriptors are setup successfully
 844  *    NETDEV_TX_OK:      error occured, the pkt is dropped
 845  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 846  *
 847  * Side-effects:
 848  *    1. tx ring may be changed
 849  *    2. tq stats may be updated accordingly
 850  *    3. shared->txNumDeferred may be updated
 851  */
 852
 853 static int
 854 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 855                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 856 {
 857         int ret;
 858         u32 count;
 859         unsigned long flags;
 860         struct vmxnet3_tx_ctx ctx;
 861         union Vmxnet3_GenericDesc *gdesc;
 862 #ifdef __BIG_ENDIAN_BITFIELD
 863         /* Use temporary descriptor to avoid touching bits multiple times */
 864         union Vmxnet3_GenericDesc tempTxDesc;
 865 #endif
 866
 867         /* conservatively estimate # of descriptors to use */
 868         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
 869                 skb_shinfo(skb)->nr_frags + 1;
 870
 871         ctx.ipv4 = (skb->protocol == __constant_ntohs(ETH_P_IP));
 872
 873         ctx.mss = skb_shinfo(skb)->gso_size;
 874         if (ctx.mss) {
 875                 if (skb_header_cloned(skb)) {
 876                         if (unlikely(pskb_expand_head(skb, 0, 0,
 877                                                       GFP_ATOMIC) != 0)) {
 878                                 tq->stats.drop_tso++;
 879                                 goto drop_pkt;
 880                         }
 881                         tq->stats.copy_skb_header++;
 882                 }
 883                 vmxnet3_prepare_tso(skb, &ctx);
 884         } else {
 885                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 886
 887                         /* non-tso pkts must not use more than
 888                          * VMXNET3_MAX_TXD_PER_PKT entries
 889                          */
 890                         if (skb_linearize(skb) != 0) {
 891                                 tq->stats.drop_too_many_frags++;
 892                                 goto drop_pkt;
 893                         }
 894                         tq->stats.linearized++;
 895
 896                         /* recalculate the # of descriptors to use */
 897                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 898                 }
 899         }
 900
 901         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 902         if (ret >= 0) {
 903                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 904                 /* hdrs parsed, check against other limits */
 905                 if (ctx.mss) {
 906                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 907                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 908                                 goto hdr_too_big;
 909                         }
 910                 } else {
 911                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 912                                 if (unlikely(ctx.eth_ip_hdr_size +
 913                                              skb->csum_offset >
 914                                              VMXNET3_MAX_CSUM_OFFSET)) {
 915                                         goto hdr_too_big;
 916                                 }
 917                         }
 918                 }
 919         } else {
 920                 tq->stats.drop_hdr_inspect_err++;
 921                 goto drop_pkt;
 922         }
 923
 924         spin_lock_irqsave(&tq->tx_lock, flags);
 925
 926         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 927                 tq->stats.tx_ring_full++;
 928                 dev_dbg(&adapter->netdev->dev,
 929                         "tx queue stopped on %s, next2comp %u"
 930                         " next2fill %u\n", adapter->netdev->name,
 931                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 932
 933                 vmxnet3_tq_stop(tq, adapter);
 934                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 935                 return NETDEV_TX_BUSY;
 936         }
 937
 938         /* fill tx descs related to addr & len */
 939         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
 940
 941         /* setup the EOP desc */
 942         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
 943
 944         /* setup the SOP desc */
 945 #ifdef __BIG_ENDIAN_BITFIELD
 946         gdesc = &tempTxDesc;
 947         gdesc->dword[2] = ctx.sop_txd->dword[2];
 948         gdesc->dword[3] = ctx.sop_txd->dword[3];
 949 #else
 950         gdesc = ctx.sop_txd;
 951 #endif
 952         if (ctx.mss) {
 953                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
 954                 gdesc->txd.om = VMXNET3_OM_TSO;
 955                 gdesc->txd.msscof = ctx.mss;
 956                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
 957                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
 958         } else {
 959                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 960                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
 961                         gdesc->txd.om = VMXNET3_OM_CSUM;
 962                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
 963                                             skb->csum_offset;
 964                 } else {
 965                         gdesc->txd.om = 0;
 966                         gdesc->txd.msscof = 0;
 967                 }
 968                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
 969         }
 970
 971         if (vlan_tx_tag_present(skb)) {
 972                 gdesc->txd.ti = 1;
 973                 gdesc->txd.tci = vlan_tx_tag_get(skb);
 974         }
 975
 976         /* finally flips the GEN bit of the SOP desc. */
 977         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
 978                                                   VMXNET3_TXD_GEN);
 979 #ifdef __BIG_ENDIAN_BITFIELD
 980         /* Finished updating in bitfields of Tx Desc, so write them in original
 981          * place.
 982          */
 983         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
 984                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
 985         gdesc = ctx.sop_txd;
 986 #endif
 987         dev_dbg(&adapter->netdev->dev,
 988                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
 989                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
 990                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
 991                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
 992
 993         spin_unlock_irqrestore(&tq->tx_lock, flags);
 994
 995         if (le32_to_cpu(tq->shared->txNumDeferred) >=
 996                                         le32_to_cpu(tq->shared->txThreshold)) {
 997                 tq->shared->txNumDeferred = 0;
 998                 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_TXPROD,
 999                                        tq->tx_ring.next2fill);
1000         }
1001
1002         return NETDEV_TX_OK;
1003
1004 hdr_too_big:
1005         tq->stats.drop_oversized_hdr++;
1006 drop_pkt:
1007         tq->stats.drop_total++;
1008         dev_kfree_skb(skb);
1009         return NETDEV_TX_OK;
1010 }
1011
1012
1013 static netdev_tx_t
1014 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1015 {
1016         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1017
1018         return vmxnet3_tq_xmit(skb, &adapter->tx_queue, adapter, netdev);
1019 }
1020
1021
1022 static void
1023 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1024                 struct sk_buff *skb,
1025                 union Vmxnet3_GenericDesc *gdesc)
1026 {
1027         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1028                 /* typical case: TCP/UDP over IP and both csums are correct */
1029                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1030                                                         VMXNET3_RCD_CSUM_OK) {
1031                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1032                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1033                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1034                         BUG_ON(gdesc->rcd.frg);
1035                 } else {
1036                         if (gdesc->rcd.csum) {
1037                                 skb->csum = htons(gdesc->rcd.csum);
1038                                 skb->ip_summed = CHECKSUM_PARTIAL;
1039                         } else {
1040                                 skb->ip_summed = CHECKSUM_NONE;
1041                         }
1042                 }
1043         } else {
1044                 skb->ip_summed = CHECKSUM_NONE;
1045         }
1046 }
1047
1048
1049 static void
1050 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1051                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1052 {
1053         rq->stats.drop_err++;
1054         if (!rcd->fcs)
1055                 rq->stats.drop_fcs++;
1056
1057         rq->stats.drop_total++;
1058
1059         /*
1060          * We do not unmap and chain the rx buffer to the skb.
1061          * We basically pretend this buffer is not used and will be recycled
1062          * by vmxnet3_rq_alloc_rx_buf()
1063          */
1064
1065         /*
1066          * ctx->skb may be NULL if this is the first and the only one
1067          * desc for the pkt
1068          */
1069         if (ctx->skb)
1070                 dev_kfree_skb_irq(ctx->skb);
1071
1072         ctx->skb = NULL;
1073 }
1074
1075
1076 static int
1077 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1078                        struct vmxnet3_adapter *adapter, int quota)
1079 {
1080         static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
1081         u32 num_rxd = 0;
1082         struct Vmxnet3_RxCompDesc *rcd;
1083         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1084 #ifdef __BIG_ENDIAN_BITFIELD
1085         struct Vmxnet3_RxDesc rxCmdDesc;
1086         struct Vmxnet3_RxCompDesc rxComp;
1087 #endif
1088         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1089                           &rxComp);
1090         while (rcd->gen == rq->comp_ring.gen) {
1091                 struct vmxnet3_rx_buf_info *rbi;
1092                 struct sk_buff *skb;
1093                 int num_to_alloc;
1094                 struct Vmxnet3_RxDesc *rxd;
1095                 u32 idx, ring_idx;
1096
1097                 if (num_rxd >= quota) {
1098                         /* we may stop even before we see the EOP desc of
1099                          * the current pkt
1100                          */
1101                         break;
1102                 }
1103                 num_rxd++;
1104
1105                 idx = rcd->rxdIdx;
1106                 ring_idx = rcd->rqID == rq->qid ? 0 : 1;
1107                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1108                                   &rxCmdDesc);
1109                 rbi = rq->buf_info[ring_idx] + idx;
1110
1111                 BUG_ON(rxd->addr != rbi->dma_addr ||
1112                        rxd->len != rbi->len);
1113
1114                 if (unlikely(rcd->eop && rcd->err)) {
1115                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1116                         goto rcd_done;
1117                 }
1118
1119                 if (rcd->sop) { /* first buf of the pkt */
1120                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1121                                rcd->rqID != rq->qid);
1122
1123                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1124                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1125
1126                         if (unlikely(rcd->len == 0)) {
1127                                 /* Pretend the rx buffer is skipped. */
1128                                 BUG_ON(!(rcd->sop && rcd->eop));
1129                                 dev_dbg(&adapter->netdev->dev,
1130                                         "rxRing[%u][%u] 0 length\n",
1131                                         ring_idx, idx);
1132                                 goto rcd_done;
1133                         }
1134
1135                         ctx->skb = rbi->skb;
1136                         rbi->skb = NULL;
1137
1138                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1139                                          PCI_DMA_FROMDEVICE);
1140
1141                         skb_put(ctx->skb, rcd->len);
1142                 } else {
1143                         BUG_ON(ctx->skb == NULL);
1144                         /* non SOP buffer must be type 1 in most cases */
1145                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1146                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1147
1148                                 if (rcd->len) {
1149                                         pci_unmap_page(adapter->pdev,
1150                                                        rbi->dma_addr, rbi->len,
1151                                                        PCI_DMA_FROMDEVICE);
1152
1153                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1154                                         rbi->page = NULL;
1155                                 }
1156                         } else {
1157                                 /*
1158                                  * The only time a non-SOP buffer is type 0 is
1159                                  * when it's EOP and error flag is raised, which
1160                                  * has already been handled.
1161                                  */
1162                                 BUG_ON(true);
1163                         }
1164                 }
1165
1166                 skb = ctx->skb;
1167                 if (rcd->eop) {
1168                         skb->len += skb->data_len;
1169                         skb->truesize += skb->data_len;
1170
1171                         vmxnet3_rx_csum(adapter, skb,
1172                                         (union Vmxnet3_GenericDesc *)rcd);
1173                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1174
1175                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1176                                 vlan_hwaccel_receive_skb(skb,
1177                                                 adapter->vlan_grp, rcd->tci);
1178                         } else {
1179                                 netif_receive_skb(skb);
1180                         }
1181
1182                         ctx->skb = NULL;
1183                 }
1184
1185 rcd_done:
1186                 /* device may skip some rx descs */
1187                 rq->rx_ring[ring_idx].next2comp = idx;
1188                 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1189                                           rq->rx_ring[ring_idx].size);
1190
1191                 /* refill rx buffers frequently to avoid starving the h/w */
1192                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1193                                                            ring_idx);
1194                 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1195                                                         ring_idx, adapter))) {
1196                         vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1197                                                 adapter);
1198
1199                         /* if needed, update the register */
1200                         if (unlikely(rq->shared->updateRxProd)) {
1201                                 VMXNET3_WRITE_BAR0_REG(adapter,
1202                                         rxprod_reg[ring_idx] + rq->qid * 8,
1203                                         rq->rx_ring[ring_idx].next2fill);
1204                                 rq->uncommitted[ring_idx] = 0;
1205                         }
1206                 }
1207
1208                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1209                 vmxnet3_getRxComp(rcd,
1210                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1211         }
1212
1213         return num_rxd;
1214 }
1215
1216
1217 static void
1218 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1219                    struct vmxnet3_adapter *adapter)
1220 {
1221         u32 i, ring_idx;
1222         struct Vmxnet3_RxDesc *rxd;
1223
1224         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1225                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1226 #ifdef __BIG_ENDIAN_BITFIELD
1227                         struct Vmxnet3_RxDesc rxDesc;
1228 #endif
1229                         vmxnet3_getRxDesc(rxd,
1230                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1231
1232                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1233                                         rq->buf_info[ring_idx][i].skb) {
1234                                 pci_unmap_single(adapter->pdev, rxd->addr,
1235                                                  rxd->len, PCI_DMA_FROMDEVICE);
1236                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1237                                 rq->buf_info[ring_idx][i].skb = NULL;
1238                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1239                                         rq->buf_info[ring_idx][i].page) {
1240                                 pci_unmap_page(adapter->pdev, rxd->addr,
1241                                                rxd->len, PCI_DMA_FROMDEVICE);
1242                                 put_page(rq->buf_info[ring_idx][i].page);
1243                                 rq->buf_info[ring_idx][i].page = NULL;
1244                         }
1245                 }
1246
1247                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1248                 rq->rx_ring[ring_idx].next2fill =
1249                                         rq->rx_ring[ring_idx].next2comp = 0;
1250                 rq->uncommitted[ring_idx] = 0;
1251         }
1252
1253         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1254         rq->comp_ring.next2proc = 0;
1255 }
1256
1257
1258 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1259                         struct vmxnet3_adapter *adapter)
1260 {
1261         int i;
1262         int j;
1263
1264         /* all rx buffers must have already been freed */
1265         for (i = 0; i < 2; i++) {
1266                 if (rq->buf_info[i]) {
1267                         for (j = 0; j < rq->rx_ring[i].size; j++)
1268                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1269                 }
1270         }
1271
1272
1273         kfree(rq->buf_info[0]);
1274
1275         for (i = 0; i < 2; i++) {
1276                 if (rq->rx_ring[i].base) {
1277                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1278                                             * sizeof(struct Vmxnet3_RxDesc),
1279                                             rq->rx_ring[i].base,
1280                                             rq->rx_ring[i].basePA);
1281                         rq->rx_ring[i].base = NULL;
1282                 }
1283                 rq->buf_info[i] = NULL;
1284         }
1285
1286         if (rq->comp_ring.base) {
1287                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1288                                     sizeof(struct Vmxnet3_RxCompDesc),
1289                                     rq->comp_ring.base, rq->comp_ring.basePA);
1290                 rq->comp_ring.base = NULL;
1291         }
1292 }
1293
1294
1295 static int
1296 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1297                 struct vmxnet3_adapter  *adapter)
1298 {
1299         int i;
1300
1301         /* initialize buf_info */
1302         for (i = 0; i < rq->rx_ring[0].size; i++) {
1303
1304                 /* 1st buf for a pkt is skbuff */
1305                 if (i % adapter->rx_buf_per_pkt == 0) {
1306                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1307                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1308                 } else { /* subsequent bufs for a pkt is frag */
1309                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1310                         rq->buf_info[0][i].len = PAGE_SIZE;
1311                 }
1312         }
1313         for (i = 0; i < rq->rx_ring[1].size; i++) {
1314                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1315                 rq->buf_info[1][i].len = PAGE_SIZE;
1316         }
1317
1318         /* reset internal state and allocate buffers for both rings */
1319         for (i = 0; i < 2; i++) {
1320                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1321                 rq->uncommitted[i] = 0;
1322
1323                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1324                        sizeof(struct Vmxnet3_RxDesc));
1325                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1326         }
1327         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1328                                     adapter) == 0) {
1329                 /* at least has 1 rx buffer for the 1st ring */
1330                 return -ENOMEM;
1331         }
1332         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1333
1334         /* reset the comp ring */
1335         rq->comp_ring.next2proc = 0;
1336         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1337                sizeof(struct Vmxnet3_RxCompDesc));
1338         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1339
1340         /* reset rxctx */
1341         rq->rx_ctx.skb = NULL;
1342
1343         /* stats are not reset */
1344         return 0;
1345 }
1346
1347
1348 static int
1349 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1350 {
1351         int i;
1352         size_t sz;
1353         struct vmxnet3_rx_buf_info *bi;
1354
1355         for (i = 0; i < 2; i++) {
1356
1357                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1358                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1359                                                         &rq->rx_ring[i].basePA);
1360                 if (!rq->rx_ring[i].base) {
1361                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1362                                adapter->netdev->name, i);
1363                         goto err;
1364                 }
1365         }
1366
1367         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1368         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1369                                                   &rq->comp_ring.basePA);
1370         if (!rq->comp_ring.base) {
1371                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1372                        adapter->netdev->name);
1373                 goto err;
1374         }
1375
1376         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1377                                                    rq->rx_ring[1].size);
1378         bi = kzalloc(sz, GFP_KERNEL);
1379         if (!bi) {
1380                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1381                        adapter->netdev->name);
1382                 goto err;
1383         }
1384         rq->buf_info[0] = bi;
1385         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1386
1387         return 0;
1388
1389 err:
1390         vmxnet3_rq_destroy(rq, adapter);
1391         return -ENOMEM;
1392 }
1393
1394
1395 static int
1396 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1397 {
1398         if (unlikely(adapter->shared->ecr))
1399                 vmxnet3_process_events(adapter);
1400
1401         vmxnet3_tq_tx_complete(&adapter->tx_queue, adapter);
1402         return vmxnet3_rq_rx_complete(&adapter->rx_queue, adapter, budget);
1403 }
1404
1405
1406 static int
1407 vmxnet3_poll(struct napi_struct *napi, int budget)
1408 {
1409         struct vmxnet3_adapter *adapter = container_of(napi,
1410                                           struct vmxnet3_adapter, napi);
1411         int rxd_done;
1412
1413         rxd_done = vmxnet3_do_poll(adapter, budget);
1414
1415         if (rxd_done < budget) {
1416                 napi_complete(napi);
1417                 vmxnet3_enable_intr(adapter, 0);
1418         }
1419         return rxd_done;
1420 }
1421
1422
1423 /* Interrupt handler for vmxnet3  */
1424 static irqreturn_t
1425 vmxnet3_intr(int irq, void *dev_id)
1426 {
1427         struct net_device *dev = dev_id;
1428         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1429
1430         if (unlikely(adapter->intr.type == VMXNET3_IT_INTX)) {
1431                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1432                 if (unlikely(icr == 0))
1433                         /* not ours */
1434                         return IRQ_NONE;
1435         }
1436
1437
1438         /* disable intr if needed */
1439         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1440                 vmxnet3_disable_intr(adapter, 0);
1441
1442         napi_schedule(&adapter->napi);
1443
1444         return IRQ_HANDLED;
1445 }
1446
1447 #ifdef CONFIG_NET_POLL_CONTROLLER
1448
1449
1450 /* netpoll callback. */
1451 static void
1452 vmxnet3_netpoll(struct net_device *netdev)
1453 {
1454         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1455         int irq;
1456
1457 #ifdef CONFIG_PCI_MSI
1458         if (adapter->intr.type == VMXNET3_IT_MSIX)
1459                 irq = adapter->intr.msix_entries[0].vector;
1460         else
1461 #endif
1462                 irq = adapter->pdev->irq;
1463
1464         disable_irq(irq);
1465         vmxnet3_intr(irq, netdev);
1466         enable_irq(irq);
1467 }
1468 #endif
1469
1470 static int
1471 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1472 {
1473         int err;
1474
1475 #ifdef CONFIG_PCI_MSI
1476         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1477                 /* we only use 1 MSI-X vector */
1478                 err = request_irq(adapter->intr.msix_entries[0].vector,
1479                                   vmxnet3_intr, 0, adapter->netdev->name,
1480                                   adapter->netdev);
1481         } else if (adapter->intr.type == VMXNET3_IT_MSI) {
1482                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1483                                   adapter->netdev->name, adapter->netdev);
1484         } else
1485 #endif
1486         {
1487                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1488                                   IRQF_SHARED, adapter->netdev->name,
1489                                   adapter->netdev);
1490         }
1491
1492         if (err)
1493                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1494                        ":%d\n", adapter->netdev->name, adapter->intr.type, err);
1495
1496
1497         if (!err) {
1498                 int i;
1499                 /* init our intr settings */
1500                 for (i = 0; i < adapter->intr.num_intrs; i++)
1501                         adapter->intr.mod_levels[i] = UPT1_IML_ADAPTIVE;
1502
1503                 /* next setup intr index for all intr sources */
1504                 adapter->tx_queue.comp_ring.intr_idx = 0;
1505                 adapter->rx_queue.comp_ring.intr_idx = 0;
1506                 adapter->intr.event_intr_idx = 0;
1507
1508                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1509                        "allocated\n", adapter->netdev->name, adapter->intr.type,
1510                        adapter->intr.mask_mode, adapter->intr.num_intrs);
1511         }
1512
1513         return err;
1514 }
1515
1516
1517 static void
1518 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1519 {
1520         BUG_ON(adapter->intr.type == VMXNET3_IT_AUTO ||
1521                adapter->intr.num_intrs <= 0);
1522
1523         switch (adapter->intr.type) {
1524 #ifdef CONFIG_PCI_MSI
1525         case VMXNET3_IT_MSIX:
1526         {
1527                 int i;
1528
1529                 for (i = 0; i < adapter->intr.num_intrs; i++)
1530                         free_irq(adapter->intr.msix_entries[i].vector,
1531                                  adapter->netdev);
1532                 break;
1533         }
1534 #endif
1535         case VMXNET3_IT_MSI:
1536                 free_irq(adapter->pdev->irq, adapter->netdev);
1537                 break;
1538         case VMXNET3_IT_INTX:
1539                 free_irq(adapter->pdev->irq, adapter->netdev);
1540                 break;
1541         default:
1542                 BUG_ON(true);
1543         }
1544 }
1545
1546
1547 inline void set_flag_le16(__le16 *data, u16 flag)
1548 {
1549         *data = cpu_to_le16(le16_to_cpu(*data) | flag);
1550 }
1551
1552 inline void set_flag_le64(__le64 *data, u64 flag)
1553 {
1554         *data = cpu_to_le64(le64_to_cpu(*data) | flag);
1555 }
1556
1557 inline void reset_flag_le64(__le64 *data, u64 flag)
1558 {
1559         *data = cpu_to_le64(le64_to_cpu(*data) & ~flag);
1560 }
1561
1562
1563 static void
1564 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1565 {
1566         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1567         struct Vmxnet3_DriverShared *shared = adapter->shared;
1568         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1569
1570         if (grp) {
1571                 /* add vlan rx stripping. */
1572                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1573                         int i;
1574                         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1575                         adapter->vlan_grp = grp;
1576
1577                         /* update FEATURES to device */
1578                         set_flag_le64(&devRead->misc.uptFeatures,
1579                                       UPT1_F_RXVLAN);
1580                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1581                                                VMXNET3_CMD_UPDATE_FEATURE);
1582                         /*
1583                          *  Clear entire vfTable; then enable untagged pkts.
1584                          *  Note: setting one entry in vfTable to non-zero turns
1585                          *  on VLAN rx filtering.
1586                          */
1587                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1588                                 vfTable[i] = 0;
1589
1590                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1591                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1592                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1593                 } else {
1594                         printk(KERN_ERR "%s: vlan_rx_register when device has "
1595                                "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1596                 }
1597         } else {
1598                 /* remove vlan rx stripping. */
1599                 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1600                 adapter->vlan_grp = NULL;
1601
1602                 if (le64_to_cpu(devRead->misc.uptFeatures) & UPT1_F_RXVLAN) {
1603                         int i;
1604
1605                         for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1606                                 /* clear entire vfTable; this also disables
1607                                  * VLAN rx filtering
1608                                  */
1609                                 vfTable[i] = 0;
1610                         }
1611                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1612                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1613
1614                         /* update FEATURES to device */
1615                         reset_flag_le64(&devRead->misc.uptFeatures,
1616                                         UPT1_F_RXVLAN);
1617                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1618                                                VMXNET3_CMD_UPDATE_FEATURE);
1619                 }
1620         }
1621 }
1622
1623
1624 static void
1625 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1626 {
1627         if (adapter->vlan_grp) {
1628                 u16 vid;
1629                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1630                 bool activeVlan = false;
1631
1632                 for (vid = 0; vid < VLAN_GROUP_ARRAY_LEN; vid++) {
1633                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1634                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1635                                 activeVlan = true;
1636                         }
1637                 }
1638                 if (activeVlan) {
1639                         /* continue to allow untagged pkts */
1640                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1641                 }
1642         }
1643 }
1644
1645
1646 static void
1647 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1648 {
1649         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1650         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1651
1652         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1653         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1654                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1655 }
1656
1657
1658 static void
1659 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1660 {
1661         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1662         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1663
1664         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1665         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1666                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1667 }
1668
1669
1670 static u8 *
1671 vmxnet3_copy_mc(struct net_device *netdev)
1672 {
1673         u8 *buf = NULL;
1674         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1675
1676         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1677         if (sz <= 0xffff) {
1678                 /* We may be called with BH disabled */
1679                 buf = kmalloc(sz, GFP_ATOMIC);
1680                 if (buf) {
1681                         struct netdev_hw_addr *ha;
1682                         int i = 0;
1683
1684                         netdev_for_each_mc_addr(ha, netdev)
1685                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1686                                        ETH_ALEN);
1687                 }
1688         }
1689         return buf;
1690 }
1691
1692
1693 static void
1694 vmxnet3_set_mc(struct net_device *netdev)
1695 {
1696         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1697         struct Vmxnet3_RxFilterConf *rxConf =
1698                                         &adapter->shared->devRead.rxFilterConf;
1699         u8 *new_table = NULL;
1700         u32 new_mode = VMXNET3_RXM_UCAST;
1701
1702         if (netdev->flags & IFF_PROMISC)
1703                 new_mode |= VMXNET3_RXM_PROMISC;
1704
1705         if (netdev->flags & IFF_BROADCAST)
1706                 new_mode |= VMXNET3_RXM_BCAST;
1707
1708         if (netdev->flags & IFF_ALLMULTI)
1709                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1710         else
1711                 if (!netdev_mc_empty(netdev)) {
1712                         new_table = vmxnet3_copy_mc(netdev);
1713                         if (new_table) {
1714                                 new_mode |= VMXNET3_RXM_MCAST;
1715                                 rxConf->mfTableLen = cpu_to_le16(
1716                                         netdev_mc_count(netdev) * ETH_ALEN);
1717                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
1718                                                     new_table));
1719                         } else {
1720                                 printk(KERN_INFO "%s: failed to copy mcast list"
1721                                        ", setting ALL_MULTI\n", netdev->name);
1722                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1723                         }
1724                 }
1725
1726
1727         if (!(new_mode & VMXNET3_RXM_MCAST)) {
1728                 rxConf->mfTableLen = 0;
1729                 rxConf->mfTablePA = 0;
1730         }
1731
1732         if (new_mode != rxConf->rxMode) {
1733                 rxConf->rxMode = cpu_to_le32(new_mode);
1734                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1735                                        VMXNET3_CMD_UPDATE_RX_MODE);
1736         }
1737
1738         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1739                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
1740
1741         kfree(new_table);
1742 }
1743
1744
1745 /*
1746  *   Set up driver_shared based on settings in adapter.
1747  */
1748
1749 static void
1750 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
1751 {
1752         struct Vmxnet3_DriverShared *shared = adapter->shared;
1753         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1754         struct Vmxnet3_TxQueueConf *tqc;
1755         struct Vmxnet3_RxQueueConf *rqc;
1756         int i;
1757
1758         memset(shared, 0, sizeof(*shared));
1759
1760         /* driver settings */
1761         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
1762         devRead->misc.driverInfo.version = cpu_to_le32(
1763                                                 VMXNET3_DRIVER_VERSION_NUM);
1764         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
1765                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
1766         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
1767         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
1768                                 *((u32 *)&devRead->misc.driverInfo.gos));
1769         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
1770         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
1771
1772         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
1773         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
1774
1775         /* set up feature flags */
1776         if (adapter->rxcsum)
1777                 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXCSUM);
1778
1779         if (adapter->lro) {
1780                 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_LRO);
1781                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
1782         }
1783         if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX) &&
1784             adapter->vlan_grp) {
1785                 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXVLAN);
1786         }
1787
1788         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
1789         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
1790         devRead->misc.queueDescLen = cpu_to_le32(
1791                                      sizeof(struct Vmxnet3_TxQueueDesc) +
1792                                      sizeof(struct Vmxnet3_RxQueueDesc));
1793
1794         /* tx queue settings */
1795         BUG_ON(adapter->tx_queue.tx_ring.base == NULL);
1796
1797         devRead->misc.numTxQueues = 1;
1798         tqc = &adapter->tqd_start->conf;
1799         tqc->txRingBasePA   = cpu_to_le64(adapter->tx_queue.tx_ring.basePA);
1800         tqc->dataRingBasePA = cpu_to_le64(adapter->tx_queue.data_ring.basePA);
1801         tqc->compRingBasePA = cpu_to_le64(adapter->tx_queue.comp_ring.basePA);
1802         tqc->ddPA           = cpu_to_le64(virt_to_phys(
1803                                                 adapter->tx_queue.buf_info));
1804         tqc->txRingSize     = cpu_to_le32(adapter->tx_queue.tx_ring.size);
1805         tqc->dataRingSize   = cpu_to_le32(adapter->tx_queue.data_ring.size);
1806         tqc->compRingSize   = cpu_to_le32(adapter->tx_queue.comp_ring.size);
1807         tqc->ddLen          = cpu_to_le32(sizeof(struct vmxnet3_tx_buf_info) *
1808                               tqc->txRingSize);
1809         tqc->intrIdx        = adapter->tx_queue.comp_ring.intr_idx;
1810
1811         /* rx queue settings */
1812         devRead->misc.numRxQueues = 1;
1813         rqc = &adapter->rqd_start->conf;
1814         rqc->rxRingBasePA[0] = cpu_to_le64(adapter->rx_queue.rx_ring[0].basePA);
1815         rqc->rxRingBasePA[1] = cpu_to_le64(adapter->rx_queue.rx_ring[1].basePA);
1816         rqc->compRingBasePA  = cpu_to_le64(adapter->rx_queue.comp_ring.basePA);
1817         rqc->ddPA            = cpu_to_le64(virt_to_phys(
1818                                                 adapter->rx_queue.buf_info));
1819         rqc->rxRingSize[0]   = cpu_to_le32(adapter->rx_queue.rx_ring[0].size);
1820         rqc->rxRingSize[1]   = cpu_to_le32(adapter->rx_queue.rx_ring[1].size);
1821         rqc->compRingSize    = cpu_to_le32(adapter->rx_queue.comp_ring.size);
1822         rqc->ddLen           = cpu_to_le32(sizeof(struct vmxnet3_rx_buf_info) *
1823                                (rqc->rxRingSize[0] + rqc->rxRingSize[1]));
1824         rqc->intrIdx         = adapter->rx_queue.comp_ring.intr_idx;
1825
1826         /* intr settings */
1827         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
1828                                      VMXNET3_IMM_AUTO;
1829         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
1830         for (i = 0; i < adapter->intr.num_intrs; i++)
1831                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
1832
1833         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
1834         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
1835
1836         /* rx filter settings */
1837         devRead->rxFilterConf.rxMode = 0;
1838         vmxnet3_restore_vlan(adapter);
1839         /* the rest are already zeroed */
1840 }
1841
1842
1843 int
1844 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
1845 {
1846         int err;
1847         u32 ret;
1848
1849         dev_dbg(&adapter->netdev->dev,
1850                 "%s: skb_buf_size %d, rx_buf_per_pkt %d, ring sizes"
1851                 " %u %u %u\n", adapter->netdev->name, adapter->skb_buf_size,
1852                 adapter->rx_buf_per_pkt, adapter->tx_queue.tx_ring.size,
1853                 adapter->rx_queue.rx_ring[0].size,
1854                 adapter->rx_queue.rx_ring[1].size);
1855
1856         vmxnet3_tq_init(&adapter->tx_queue, adapter);
1857         err = vmxnet3_rq_init(&adapter->rx_queue, adapter);
1858         if (err) {
1859                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
1860                        adapter->netdev->name, err);
1861                 goto rq_err;
1862         }
1863
1864         err = vmxnet3_request_irqs(adapter);
1865         if (err) {
1866                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
1867                        adapter->netdev->name, err);
1868                 goto irq_err;
1869         }
1870
1871         vmxnet3_setup_driver_shared(adapter);
1872
1873         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
1874                                adapter->shared_pa));
1875         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
1876                                adapter->shared_pa));
1877         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1878                                VMXNET3_CMD_ACTIVATE_DEV);
1879         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
1880
1881         if (ret != 0) {
1882                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
1883                        adapter->netdev->name, ret);
1884                 err = -EINVAL;
1885                 goto activate_err;
1886         }
1887         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD,
1888                                adapter->rx_queue.rx_ring[0].next2fill);
1889         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD2,
1890                                adapter->rx_queue.rx_ring[1].next2fill);
1891
1892         /* Apply the rx filter settins last. */
1893         vmxnet3_set_mc(adapter->netdev);
1894
1895         /*
1896          * Check link state when first activating device. It will start the
1897          * tx queue if the link is up.
1898          */
1899         vmxnet3_check_link(adapter, true);
1900
1901         napi_enable(&adapter->napi);
1902         vmxnet3_enable_all_intrs(adapter);
1903         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
1904         return 0;
1905
1906 activate_err:
1907         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
1908         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
1909         vmxnet3_free_irqs(adapter);
1910 irq_err:
1911 rq_err:
1912         /* free up buffers we allocated */
1913         vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1914         return err;
1915 }
1916
1917
1918 void
1919 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
1920 {
1921         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
1922 }
1923
1924
1925 int
1926 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
1927 {
1928         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
1929                 return 0;
1930
1931
1932         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1933                                VMXNET3_CMD_QUIESCE_DEV);
1934         vmxnet3_disable_all_intrs(adapter);
1935
1936         napi_disable(&adapter->napi);
1937         netif_tx_disable(adapter->netdev);
1938         adapter->link_speed = 0;
1939         netif_carrier_off(adapter->netdev);
1940
1941         vmxnet3_tq_cleanup(&adapter->tx_queue, adapter);
1942         vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1943         vmxnet3_free_irqs(adapter);
1944         return 0;
1945 }
1946
1947
1948 static void
1949 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
1950 {
1951         u32 tmp;
1952
1953         tmp = *(u32 *)mac;
1954         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
1955
1956         tmp = (mac[5] << 8) | mac[4];
1957         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
1958 }
1959
1960
1961 static int
1962 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
1963 {
1964         struct sockaddr *addr = p;
1965         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1966
1967         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1968         vmxnet3_write_mac_addr(adapter, addr->sa_data);
1969
1970         return 0;
1971 }
1972
1973
1974 /* ==================== initialization and cleanup routines ============ */
1975
1976 static int
1977 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
1978 {
1979         int err;
1980         unsigned long mmio_start, mmio_len;
1981         struct pci_dev *pdev = adapter->pdev;
1982
1983         err = pci_enable_device(pdev);
1984         if (err) {
1985                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
1986                        pci_name(pdev), err);
1987                 return err;
1988         }
1989
1990         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1991                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
1992                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
1993                                "for adapter %s\n", pci_name(pdev));
1994                         err = -EIO;
1995                         goto err_set_mask;
1996                 }
1997                 *dma64 = true;
1998         } else {
1999                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2000                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2001                                "%s\n",  pci_name(pdev));
2002                         err = -EIO;
2003                         goto err_set_mask;
2004                 }
2005                 *dma64 = false;
2006         }
2007
2008         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2009                                            vmxnet3_driver_name);
2010         if (err) {
2011                 printk(KERN_ERR "Failed to request region for adapter %s: "
2012                        "error %d\n", pci_name(pdev), err);
2013                 goto err_set_mask;
2014         }
2015
2016         pci_set_master(pdev);
2017
2018         mmio_start = pci_resource_start(pdev, 0);
2019         mmio_len = pci_resource_len(pdev, 0);
2020         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2021         if (!adapter->hw_addr0) {
2022                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2023                        pci_name(pdev));
2024                 err = -EIO;
2025                 goto err_ioremap;
2026         }
2027
2028         mmio_start = pci_resource_start(pdev, 1);
2029         mmio_len = pci_resource_len(pdev, 1);
2030         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2031         if (!adapter->hw_addr1) {
2032                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2033                        pci_name(pdev));
2034                 err = -EIO;
2035                 goto err_bar1;
2036         }
2037         return 0;
2038
2039 err_bar1:
2040         iounmap(adapter->hw_addr0);
2041 err_ioremap:
2042         pci_release_selected_regions(pdev, (1 << 2) - 1);
2043 err_set_mask:
2044         pci_disable_device(pdev);
2045         return err;
2046 }
2047
2048
2049 static void
2050 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2051 {
2052         BUG_ON(!adapter->pdev);
2053
2054         iounmap(adapter->hw_addr0);
2055         iounmap(adapter->hw_addr1);
2056         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2057         pci_disable_device(adapter->pdev);
2058 }
2059
2060
2061 static void
2062 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2063 {
2064         size_t sz;
2065
2066         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2067                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2068                 adapter->skb_buf_size = adapter->netdev->mtu +
2069                                         VMXNET3_MAX_ETH_HDR_SIZE;
2070                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2071                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2072
2073                 adapter->rx_buf_per_pkt = 1;
2074         } else {
2075                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2076                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2077                                             VMXNET3_MAX_ETH_HDR_SIZE;
2078                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2079         }
2080
2081         /*
2082          * for simplicity, force the ring0 size to be a multiple of
2083          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2084          */
2085         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2086         adapter->rx_queue.rx_ring[0].size = (adapter->rx_queue.rx_ring[0].size +
2087                                              sz - 1) / sz * sz;
2088         adapter->rx_queue.rx_ring[0].size = min_t(u32,
2089                                             adapter->rx_queue.rx_ring[0].size,
2090                                             VMXNET3_RX_RING_MAX_SIZE / sz * sz);
2091 }
2092
2093
2094 int
2095 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2096                       u32 rx_ring_size, u32 rx_ring2_size)
2097 {
2098         int err;
2099
2100         adapter->tx_queue.tx_ring.size   = tx_ring_size;
2101         adapter->tx_queue.data_ring.size = tx_ring_size;
2102         adapter->tx_queue.comp_ring.size = tx_ring_size;
2103         adapter->tx_queue.shared = &adapter->tqd_start->ctrl;
2104         adapter->tx_queue.stopped = true;
2105         err = vmxnet3_tq_create(&adapter->tx_queue, adapter);
2106         if (err)
2107                 return err;
2108
2109         adapter->rx_queue.rx_ring[0].size = rx_ring_size;
2110         adapter->rx_queue.rx_ring[1].size = rx_ring2_size;
2111         vmxnet3_adjust_rx_ring_size(adapter);
2112         adapter->rx_queue.comp_ring.size  = adapter->rx_queue.rx_ring[0].size +
2113                                             adapter->rx_queue.rx_ring[1].size;
2114         adapter->rx_queue.qid  = 0;
2115         adapter->rx_queue.qid2 = 1;
2116         adapter->rx_queue.shared = &adapter->rqd_start->ctrl;
2117         err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2118         if (err)
2119                 vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2120
2121         return err;
2122 }
2123
2124 static int
2125 vmxnet3_open(struct net_device *netdev)
2126 {
2127         struct vmxnet3_adapter *adapter;
2128         int err;
2129
2130         adapter = netdev_priv(netdev);
2131
2132         spin_lock_init(&adapter->tx_queue.tx_lock);
2133
2134         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2135                                     VMXNET3_DEF_RX_RING_SIZE,
2136                                     VMXNET3_DEF_RX_RING_SIZE);
2137         if (err)
2138                 goto queue_err;
2139
2140         err = vmxnet3_activate_dev(adapter);
2141         if (err)
2142                 goto activate_err;
2143
2144         return 0;
2145
2146 activate_err:
2147         vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2148         vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2149 queue_err:
2150         return err;
2151 }
2152
2153
2154 static int
2155 vmxnet3_close(struct net_device *netdev)
2156 {
2157         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2158
2159         /*
2160          * Reset_work may be in the middle of resetting the device, wait for its
2161          * completion.
2162          */
2163         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2164                 msleep(1);
2165
2166         vmxnet3_quiesce_dev(adapter);
2167
2168         vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2169         vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2170
2171         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2172
2173
2174         return 0;
2175 }
2176
2177
2178 void
2179 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2180 {
2181         /*
2182          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2183          * vmxnet3_close() will deadlock.
2184          */
2185         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2186
2187         /* we need to enable NAPI, otherwise dev_close will deadlock */
2188         napi_enable(&adapter->napi);
2189         dev_close(adapter->netdev);
2190 }
2191
2192
2193 static int
2194 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2195 {
2196         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2197         int err = 0;
2198
2199         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2200                 return -EINVAL;
2201
2202         if (new_mtu > 1500 && !adapter->jumbo_frame)
2203                 return -EINVAL;
2204
2205         netdev->mtu = new_mtu;
2206
2207         /*
2208          * Reset_work may be in the middle of resetting the device, wait for its
2209          * completion.
2210          */
2211         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2212                 msleep(1);
2213
2214         if (netif_running(netdev)) {
2215                 vmxnet3_quiesce_dev(adapter);
2216                 vmxnet3_reset_dev(adapter);
2217
2218                 /* we need to re-create the rx queue based on the new mtu */
2219                 vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2220                 vmxnet3_adjust_rx_ring_size(adapter);
2221                 adapter->rx_queue.comp_ring.size  =
2222                                         adapter->rx_queue.rx_ring[0].size +
2223                                         adapter->rx_queue.rx_ring[1].size;
2224                 err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2225                 if (err) {
2226                         printk(KERN_ERR "%s: failed to re-create rx queue,"
2227                                 " error %d. Closing it.\n", netdev->name, err);
2228                         goto out;
2229                 }
2230
2231                 err = vmxnet3_activate_dev(adapter);
2232                 if (err) {
2233                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2234                                 "Closing it\n", netdev->name, err);
2235                         goto out;
2236                 }
2237         }
2238
2239 out:
2240         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2241         if (err)
2242                 vmxnet3_force_close(adapter);
2243
2244         return err;
2245 }
2246
2247
2248 static void
2249 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2250 {
2251         struct net_device *netdev = adapter->netdev;
2252
2253         netdev->features = NETIF_F_SG |
2254                 NETIF_F_HW_CSUM |
2255                 NETIF_F_HW_VLAN_TX |
2256                 NETIF_F_HW_VLAN_RX |
2257                 NETIF_F_HW_VLAN_FILTER |
2258                 NETIF_F_TSO |
2259                 NETIF_F_TSO6 |
2260                 NETIF_F_LRO;
2261
2262         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2263
2264         adapter->rxcsum = true;
2265         adapter->jumbo_frame = true;
2266         adapter->lro = true;
2267
2268         if (dma64) {
2269                 netdev->features |= NETIF_F_HIGHDMA;
2270                 printk(" highDMA");
2271         }
2272
2273         netdev->vlan_features = netdev->features;
2274         printk("\n");
2275 }
2276
2277
2278 static void
2279 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2280 {
2281         u32 tmp;
2282
2283         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2284         *(u32 *)mac = tmp;
2285
2286         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2287         mac[4] = tmp & 0xff;
2288         mac[5] = (tmp >> 8) & 0xff;
2289 }
2290
2291
2292 static void
2293 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2294 {
2295         u32 cfg;
2296
2297         /* intr settings */
2298         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2299                                VMXNET3_CMD_GET_CONF_INTR);
2300         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2301         adapter->intr.type = cfg & 0x3;
2302         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2303
2304         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2305                 int err;
2306
2307 #ifdef CONFIG_PCI_MSI
2308                 adapter->intr.msix_entries[0].entry = 0;
2309                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2310                                       VMXNET3_LINUX_MAX_MSIX_VECT);
2311                 if (!err) {
2312                         adapter->intr.num_intrs = 1;
2313                         adapter->intr.type = VMXNET3_IT_MSIX;
2314                         return;
2315                 }
2316 #endif
2317
2318                 err = pci_enable_msi(adapter->pdev);
2319                 if (!err) {
2320                         adapter->intr.num_intrs = 1;
2321                         adapter->intr.type = VMXNET3_IT_MSI;
2322                         return;
2323                 }
2324         }
2325
2326         adapter->intr.type = VMXNET3_IT_INTX;
2327
2328         /* INT-X related setting */
2329         adapter->intr.num_intrs = 1;
2330 }
2331
2332
2333 static void
2334 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2335 {
2336         if (adapter->intr.type == VMXNET3_IT_MSIX)
2337                 pci_disable_msix(adapter->pdev);
2338         else if (adapter->intr.type == VMXNET3_IT_MSI)
2339                 pci_disable_msi(adapter->pdev);
2340         else
2341                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2342 }
2343
2344
2345 static void
2346 vmxnet3_tx_timeout(struct net_device *netdev)
2347 {
2348         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2349         adapter->tx_timeout_count++;
2350
2351         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2352         schedule_work(&adapter->work);
2353 }
2354
2355
2356 static void
2357 vmxnet3_reset_work(struct work_struct *data)
2358 {
2359         struct vmxnet3_adapter *adapter;
2360
2361         adapter = container_of(data, struct vmxnet3_adapter, work);
2362
2363         /* if another thread is resetting the device, no need to proceed */
2364         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2365                 return;
2366
2367         /* if the device is closed, we must leave it alone */
2368         if (netif_running(adapter->netdev)) {
2369                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2370                 vmxnet3_quiesce_dev(adapter);
2371                 vmxnet3_reset_dev(adapter);
2372                 vmxnet3_activate_dev(adapter);
2373         } else {
2374                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2375         }
2376
2377         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2378 }
2379
2380
2381 static int __devinit
2382 vmxnet3_probe_device(struct pci_dev *pdev,
2383                      const struct pci_device_id *id)
2384 {
2385         static const struct net_device_ops vmxnet3_netdev_ops = {
2386                 .ndo_open = vmxnet3_open,
2387                 .ndo_stop = vmxnet3_close,
2388                 .ndo_start_xmit = vmxnet3_xmit_frame,
2389                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2390                 .ndo_change_mtu = vmxnet3_change_mtu,
2391                 .ndo_get_stats = vmxnet3_get_stats,
2392                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2393                 .ndo_set_multicast_list = vmxnet3_set_mc,
2394                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2395                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2396                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2397 #ifdef CONFIG_NET_POLL_CONTROLLER
2398                 .ndo_poll_controller = vmxnet3_netpoll,
2399 #endif
2400         };
2401         int err;
2402         bool dma64 = false; /* stupid gcc */
2403         u32 ver;
2404         struct net_device *netdev;
2405         struct vmxnet3_adapter *adapter;
2406         u8 mac[ETH_ALEN];
2407
2408         netdev = alloc_etherdev(sizeof(struct vmxnet3_adapter));
2409         if (!netdev) {
2410                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2411                         "%s\n", pci_name(pdev));
2412                 return -ENOMEM;
2413         }
2414
2415         pci_set_drvdata(pdev, netdev);
2416         adapter = netdev_priv(netdev);
2417         adapter->netdev = netdev;
2418         adapter->pdev = pdev;
2419
2420         adapter->shared = pci_alloc_consistent(adapter->pdev,
2421                           sizeof(struct Vmxnet3_DriverShared),
2422                           &adapter->shared_pa);
2423         if (!adapter->shared) {
2424                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2425                         pci_name(pdev));
2426                 err = -ENOMEM;
2427                 goto err_alloc_shared;
2428         }
2429
2430         adapter->tqd_start = pci_alloc_consistent(adapter->pdev,
2431                              sizeof(struct Vmxnet3_TxQueueDesc) +
2432                              sizeof(struct Vmxnet3_RxQueueDesc),
2433                              &adapter->queue_desc_pa);
2434
2435         if (!adapter->tqd_start) {
2436                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2437                         pci_name(pdev));
2438                 err = -ENOMEM;
2439                 goto err_alloc_queue_desc;
2440         }
2441         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start
2442                                                             + 1);
2443
2444         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2445         if (adapter->pm_conf == NULL) {
2446                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2447                         pci_name(pdev));
2448                 err = -ENOMEM;
2449                 goto err_alloc_pm;
2450         }
2451
2452         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2453         if (err < 0)
2454                 goto err_alloc_pci;
2455
2456         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2457         if (ver & 1) {
2458                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2459         } else {
2460                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2461                        " %s\n", ver, pci_name(pdev));
2462                 err = -EBUSY;
2463                 goto err_ver;
2464         }
2465
2466         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2467         if (ver & 1) {
2468                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2469         } else {
2470                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2471                        "adapter %s\n", ver, pci_name(pdev));
2472                 err = -EBUSY;
2473                 goto err_ver;
2474         }
2475
2476         vmxnet3_declare_features(adapter, dma64);
2477
2478         adapter->dev_number = atomic_read(&devices_found);
2479         vmxnet3_alloc_intr_resources(adapter);
2480
2481         vmxnet3_read_mac_addr(adapter, mac);
2482         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
2483
2484         netdev->netdev_ops = &vmxnet3_netdev_ops;
2485         netdev->watchdog_timeo = 5 * HZ;
2486         vmxnet3_set_ethtool_ops(netdev);
2487
2488         INIT_WORK(&adapter->work, vmxnet3_reset_work);
2489
2490         netif_napi_add(netdev, &adapter->napi, vmxnet3_poll, 64);
2491         SET_NETDEV_DEV(netdev, &pdev->dev);
2492         err = register_netdev(netdev);
2493
2494         if (err) {
2495                 printk(KERN_ERR "Failed to register adapter %s\n",
2496                         pci_name(pdev));
2497                 goto err_register;
2498         }
2499
2500         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2501         vmxnet3_check_link(adapter, false);
2502         atomic_inc(&devices_found);
2503         return 0;
2504
2505 err_register:
2506         vmxnet3_free_intr_resources(adapter);
2507 err_ver:
2508         vmxnet3_free_pci_resources(adapter);
2509 err_alloc_pci:
2510         kfree(adapter->pm_conf);
2511 err_alloc_pm:
2512         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2513                             sizeof(struct Vmxnet3_RxQueueDesc),
2514                             adapter->tqd_start, adapter->queue_desc_pa);
2515 err_alloc_queue_desc:
2516         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2517                             adapter->shared, adapter->shared_pa);
2518 err_alloc_shared:
2519         pci_set_drvdata(pdev, NULL);
2520         free_netdev(netdev);
2521         return err;
2522 }
2523
2524
2525 static void __devexit
2526 vmxnet3_remove_device(struct pci_dev *pdev)
2527 {
2528         struct net_device *netdev = pci_get_drvdata(pdev);
2529         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2530
2531         flush_scheduled_work();
2532
2533         unregister_netdev(netdev);
2534
2535         vmxnet3_free_intr_resources(adapter);
2536         vmxnet3_free_pci_resources(adapter);
2537         kfree(adapter->pm_conf);
2538         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2539                             sizeof(struct Vmxnet3_RxQueueDesc),
2540                             adapter->tqd_start, adapter->queue_desc_pa);
2541         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2542                             adapter->shared, adapter->shared_pa);
2543         free_netdev(netdev);
2544 }
2545
2546
2547 #ifdef CONFIG_PM
2548
2549 static int
2550 vmxnet3_suspend(struct device *device)
2551 {
2552         struct pci_dev *pdev = to_pci_dev(device);
2553         struct net_device *netdev = pci_get_drvdata(pdev);
2554         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2555         struct Vmxnet3_PMConf *pmConf;
2556         struct ethhdr *ehdr;
2557         struct arphdr *ahdr;
2558         u8 *arpreq;
2559         struct in_device *in_dev;
2560         struct in_ifaddr *ifa;
2561         int i = 0;
2562
2563         if (!netif_running(netdev))
2564                 return 0;
2565
2566         vmxnet3_disable_all_intrs(adapter);
2567         vmxnet3_free_irqs(adapter);
2568         vmxnet3_free_intr_resources(adapter);
2569
2570         netif_device_detach(netdev);
2571         netif_stop_queue(netdev);
2572
2573         /* Create wake-up filters. */
2574         pmConf = adapter->pm_conf;
2575         memset(pmConf, 0, sizeof(*pmConf));
2576
2577         if (adapter->wol & WAKE_UCAST) {
2578                 pmConf->filters[i].patternSize = ETH_ALEN;
2579                 pmConf->filters[i].maskSize = 1;
2580                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
2581                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
2582
2583                 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER);
2584                 i++;
2585         }
2586
2587         if (adapter->wol & WAKE_ARP) {
2588                 in_dev = in_dev_get(netdev);
2589                 if (!in_dev)
2590                         goto skip_arp;
2591
2592                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
2593                 if (!ifa)
2594                         goto skip_arp;
2595
2596                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
2597                         sizeof(struct arphdr) +         /* ARP header */
2598                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
2599                         2 * sizeof(u32);        /*2 IPv4 addresses */
2600                 pmConf->filters[i].maskSize =
2601                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
2602
2603                 /* ETH_P_ARP in Ethernet header. */
2604                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
2605                 ehdr->h_proto = htons(ETH_P_ARP);
2606
2607                 /* ARPOP_REQUEST in ARP header. */
2608                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
2609                 ahdr->ar_op = htons(ARPOP_REQUEST);
2610                 arpreq = (u8 *)(ahdr + 1);
2611
2612                 /* The Unicast IPv4 address in 'tip' field. */
2613                 arpreq += 2 * ETH_ALEN + sizeof(u32);
2614                 *(u32 *)arpreq = ifa->ifa_address;
2615
2616                 /* The mask for the relevant bits. */
2617                 pmConf->filters[i].mask[0] = 0x00;
2618                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
2619                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
2620                 pmConf->filters[i].mask[3] = 0x00;
2621                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
2622                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
2623                 in_dev_put(in_dev);
2624
2625                 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER);
2626                 i++;
2627         }
2628
2629 skip_arp:
2630         if (adapter->wol & WAKE_MAGIC)
2631                 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_MAGIC);
2632
2633         pmConf->numFilters = i;
2634
2635         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2636         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2637                                                                   *pmConf));
2638         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
2639                                                                  pmConf));
2640
2641         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2642                                VMXNET3_CMD_UPDATE_PMCFG);
2643
2644         pci_save_state(pdev);
2645         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
2646                         adapter->wol);
2647         pci_disable_device(pdev);
2648         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
2649
2650         return 0;
2651 }
2652
2653
2654 static int
2655 vmxnet3_resume(struct device *device)
2656 {
2657         int err;
2658         struct pci_dev *pdev = to_pci_dev(device);
2659         struct net_device *netdev = pci_get_drvdata(pdev);
2660         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2661         struct Vmxnet3_PMConf *pmConf;
2662
2663         if (!netif_running(netdev))
2664                 return 0;
2665
2666         /* Destroy wake-up filters. */
2667         pmConf = adapter->pm_conf;
2668         memset(pmConf, 0, sizeof(*pmConf));
2669
2670         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2671         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2672                                                                   *pmConf));
2673         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le32(virt_to_phys(
2674                                                                  pmConf));
2675
2676         netif_device_attach(netdev);
2677         pci_set_power_state(pdev, PCI_D0);
2678         pci_restore_state(pdev);
2679         err = pci_enable_device_mem(pdev);
2680         if (err != 0)
2681                 return err;
2682
2683         pci_enable_wake(pdev, PCI_D0, 0);
2684
2685         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2686                                VMXNET3_CMD_UPDATE_PMCFG);
2687         vmxnet3_alloc_intr_resources(adapter);
2688         vmxnet3_request_irqs(adapter);
2689         vmxnet3_enable_all_intrs(adapter);
2690
2691         return 0;
2692 }
2693
2694 static const struct dev_pm_ops vmxnet3_pm_ops = {
2695         .suspend = vmxnet3_suspend,
2696         .resume = vmxnet3_resume,
2697 };
2698 #endif
2699
2700 static struct pci_driver vmxnet3_driver = {
2701         .name           = vmxnet3_driver_name,
2702         .id_table       = vmxnet3_pciid_table,
2703         .probe          = vmxnet3_probe_device,
2704         .remove         = __devexit_p(vmxnet3_remove_device),
2705 #ifdef CONFIG_PM
2706         .driver.pm      = &vmxnet3_pm_ops,
2707 #endif
2708 };
2709
2710
2711 static int __init
2712 vmxnet3_init_module(void)
2713 {
2714         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
2715                 VMXNET3_DRIVER_VERSION_REPORT);
2716         return pci_register_driver(&vmxnet3_driver);
2717 }
2718
2719 module_init(vmxnet3_init_module);
2720
2721
2722 static void
2723 vmxnet3_exit_module(void)
2724 {
2725         pci_unregister_driver(&vmxnet3_driver);
2726 }
2727
2728 module_exit(vmxnet3_exit_module);
2729
2730 MODULE_AUTHOR("VMware, Inc.");
2731 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
2732 MODULE_LICENSE("GPL v2");
2733 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);