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