SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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 "vmxnet3_int.h"
  28
  29 char vmxnet3_driver_name[] = "vmxnet3";
  30 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
  31
  32 /*
  33  * PCI Device ID Table
  34  * Last entry must be all 0s
  35  */
  36 static const struct pci_device_id vmxnet3_pciid_table[] = {
  37         {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
  38         {0}
  39 };
  40
  41 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
  42
  43 static atomic_t devices_found;
  44
  45
  46 /*
  47  *    Enable/Disable the given intr
  48  */
  49 static void
  50 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  51 {
  52         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
  53 }
  54
  55
  56 static void
  57 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  58 {
  59         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
  60 }
  61
  62
  63 /*
  64  *    Enable/Disable all intrs used by the device
  65  */
  66 static void
  67 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
  68 {
  69         int i;
  70
  71         for (i = 0; i < adapter->intr.num_intrs; i++)
  72                 vmxnet3_enable_intr(adapter, i);
  73 }
  74
  75
  76 static void
  77 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
  78 {
  79         int i;
  80
  81         for (i = 0; i < adapter->intr.num_intrs; i++)
  82                 vmxnet3_disable_intr(adapter, i);
  83 }
  84
  85
  86 static void
  87 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
  88 {
  89         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
  90 }
  91
  92
  93 static bool
  94 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
  95 {
  96         return netif_queue_stopped(adapter->netdev);
  97 }
  98
  99
 100 static void
 101 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 102 {
 103         tq->stopped = false;
 104         netif_start_queue(adapter->netdev);
 105 }
 106
 107
 108 static void
 109 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 110 {
 111         tq->stopped = false;
 112         netif_wake_queue(adapter->netdev);
 113 }
 114
 115
 116 static void
 117 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 118 {
 119         tq->stopped = true;
 120         tq->num_stop++;
 121         netif_stop_queue(adapter->netdev);
 122 }
 123
 124
 125 /*
 126  * Check the link state. This may start or stop the tx queue.
 127  */
 128 static void
 129 vmxnet3_check_link(struct vmxnet3_adapter *adapter)
 130 {
 131         u32 ret;
 132
 133         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
 134         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
 135         adapter->link_speed = ret >> 16;
 136         if (ret & 1) { /* Link is up. */
 137                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
 138                        adapter->netdev->name, adapter->link_speed);
 139                 if (!netif_carrier_ok(adapter->netdev))
 140                         netif_carrier_on(adapter->netdev);
 141
 142                 vmxnet3_tq_start(&adapter->tx_queue, adapter);
 143         } else {
 144                 printk(KERN_INFO "%s: NIC Link is Down\n",
 145                        adapter->netdev->name);
 146                 if (netif_carrier_ok(adapter->netdev))
 147                         netif_carrier_off(adapter->netdev);
 148
 149                 vmxnet3_tq_stop(&adapter->tx_queue, adapter);
 150         }
 151 }
 152
 153 static void
 154 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 155 {
 156         u32 events = le32_to_cpu(adapter->shared->ecr);
 157         if (!events)
 158                 return;
 159
 160         vmxnet3_ack_events(adapter, events);
 161
 162         /* Check if link state has changed */
 163         if (events & VMXNET3_ECR_LINK)
 164                 vmxnet3_check_link(adapter);
 165
 166         /* Check if there is an error on xmit/recv queues */
 167         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 168                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 169                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 170
 171                 if (adapter->tqd_start->status.stopped) {
 172                         printk(KERN_ERR "%s: tq error 0x%x\n",
 173                                adapter->netdev->name,
 174                                le32_to_cpu(adapter->tqd_start->status.error));
 175                 }
 176                 if (adapter->rqd_start->status.stopped) {
 177                         printk(KERN_ERR "%s: rq error 0x%x\n",
 178                                adapter->netdev->name,
 179                                adapter->rqd_start->status.error);
 180                 }
 181
 182                 schedule_work(&adapter->work);
 183         }
 184 }
 185
 186 #ifdef __BIG_ENDIAN_BITFIELD
 187 /*
 188  * The device expects the bitfields in shared structures to be written in
 189  * little endian. When CPU is big endian, the following routines are used to
 190  * correctly read and write into ABI.
 191  * The general technique used here is : double word bitfields are defined in
 192  * opposite order for big endian architecture. Then before reading them in
 193  * driver the complete double word is translated using le32_to_cpu. Similarly
 194  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 195  * double words into required format.
 196  * In order to avoid touching bits in shared structure more than once, temporary
 197  * descriptors are used. These are passed as srcDesc to following functions.
 198  */
 199 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 200                                 struct Vmxnet3_RxDesc *dstDesc)
 201 {
 202         u32 *src = (u32 *)srcDesc + 2;
 203         u32 *dst = (u32 *)dstDesc + 2;
 204         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 205         *dst = le32_to_cpu(*src);
 206         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 207 }
 208
 209 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 210                                struct Vmxnet3_TxDesc *dstDesc)
 211 {
 212         int i;
 213         u32 *src = (u32 *)(srcDesc + 1);
 214         u32 *dst = (u32 *)(dstDesc + 1);
 215
 216         /* Working backwards so that the gen bit is set at the end. */
 217         for (i = 2; i > 0; i--) {
 218                 src--;
 219                 dst--;
 220                 *dst = cpu_to_le32(*src);
 221         }
 222 }
 223
 224
 225 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 226                                 struct Vmxnet3_RxCompDesc *dstDesc)
 227 {
 228         int i = 0;
 229         u32 *src = (u32 *)srcDesc;
 230         u32 *dst = (u32 *)dstDesc;
 231         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 232                 *dst = le32_to_cpu(*src);
 233                 src++;
 234                 dst++;
 235         }
 236 }
 237
 238
 239 /* Used to read bitfield values from double words. */
 240 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 241 {
 242         u32 temp = le32_to_cpu(*bitfield);
 243         u32 mask = ((1 << size) - 1) << pos;
 244         temp &= mask;
 245         temp >>= pos;
 246         return temp;
 247 }
 248
 249
 250
 251 #endif  /* __BIG_ENDIAN_BITFIELD */
 252
 253 #ifdef __BIG_ENDIAN_BITFIELD
 254
 255 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 256                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 257                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 258 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 259                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 260                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 261 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 262                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 263                         VMXNET3_TCD_GEN_SIZE)
 264 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 265                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 266 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 267                         (dstrcd) = (tmp); \
 268                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 269                 } while (0)
 270 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 271                         (dstrxd) = (tmp); \
 272                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 273                 } while (0)
 274
 275 #else
 276
 277 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 278 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 279 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 280 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 281 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 282 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 283
 284 #endif /* __BIG_ENDIAN_BITFIELD  */
 285
 286
 287 static void
 288 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 289                      struct pci_dev *pdev)
 290 {
 291         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 292                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
 293                                  PCI_DMA_TODEVICE);
 294         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 295                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
 296                                PCI_DMA_TODEVICE);
 297         else
 298                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 299
 300         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 301 }
 302
 303
 304 static int
 305 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 306                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 307 {
 308         struct sk_buff *skb;
 309         int entries = 0;
 310
 311         /* no out of order completion */
 312         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 313         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 314
 315         skb = tq->buf_info[eop_idx].skb;
 316         BUG_ON(skb == NULL);
 317         tq->buf_info[eop_idx].skb = NULL;
 318
 319         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 320
 321         while (tq->tx_ring.next2comp != eop_idx) {
 322                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 323                                      pdev);
 324
 325                 /* update next2comp w/o tx_lock. Since we are marking more,
 326                  * instead of less, tx ring entries avail, the worst case is
 327                  * that the tx routine incorrectly re-queues a pkt due to
 328                  * insufficient tx ring entries.
 329                  */
 330                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 331                 entries++;
 332         }
 333
 334         dev_kfree_skb_any(skb);
 335         return entries;
 336 }
 337
 338
 339 static int
 340 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 341                         struct vmxnet3_adapter *adapter)
 342 {
 343         int completed = 0;
 344         union Vmxnet3_GenericDesc *gdesc;
 345
 346         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 347         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 348                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 349                                                &gdesc->tcd), tq, adapter->pdev,
 350                                                adapter);
 351
 352                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 353                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 354         }
 355
 356         if (completed) {
 357                 spin_lock(&tq->tx_lock);
 358                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 359                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 360                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 361                              netif_carrier_ok(adapter->netdev))) {
 362                         vmxnet3_tq_wake(tq, adapter);
 363                 }
 364                 spin_unlock(&tq->tx_lock);
 365         }
 366         return completed;
 367 }
 368
 369
 370 static void
 371 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 372                    struct vmxnet3_adapter *adapter)
 373 {
 374         int i;
 375
 376         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 377                 struct vmxnet3_tx_buf_info *tbi;
 378                 union Vmxnet3_GenericDesc *gdesc;
 379
 380                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 381                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
 382
 383                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 384                 if (tbi->skb) {
 385                         dev_kfree_skb_any(tbi->skb);
 386                         tbi->skb = NULL;
 387                 }
 388                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 389         }
 390
 391         /* sanity check, verify all buffers are indeed unmapped and freed */
 392         for (i = 0; i < tq->tx_ring.size; i++) {
 393                 BUG_ON(tq->buf_info[i].skb != NULL ||
 394                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 395         }
 396
 397         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 398         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 399
 400         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 401         tq->comp_ring.next2proc = 0;
 402 }
 403
 404
 405 void
 406 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 407                    struct vmxnet3_adapter *adapter)
 408 {
 409         if (tq->tx_ring.base) {
 410                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 411                                     sizeof(struct Vmxnet3_TxDesc),
 412                                     tq->tx_ring.base, tq->tx_ring.basePA);
 413                 tq->tx_ring.base = NULL;
 414         }
 415         if (tq->data_ring.base) {
 416                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 417                                     sizeof(struct Vmxnet3_TxDataDesc),
 418                                     tq->data_ring.base, tq->data_ring.basePA);
 419                 tq->data_ring.base = NULL;
 420         }
 421         if (tq->comp_ring.base) {
 422                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 423                                     sizeof(struct Vmxnet3_TxCompDesc),
 424                                     tq->comp_ring.base, tq->comp_ring.basePA);
 425                 tq->comp_ring.base = NULL;
 426         }
 427         kfree(tq->buf_info);
 428         tq->buf_info = NULL;
 429 }
 430
 431
 432 static void
 433 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 434                 struct vmxnet3_adapter *adapter)
 435 {
 436         int i;
 437
 438         /* reset the tx ring contents to 0 and reset the tx ring states */
 439         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 440                sizeof(struct Vmxnet3_TxDesc));
 441         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 442         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 443
 444         memset(tq->data_ring.base, 0, tq->data_ring.size *
 445                sizeof(struct Vmxnet3_TxDataDesc));
 446
 447         /* reset the tx comp ring contents to 0 and reset comp ring states */
 448         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 449                sizeof(struct Vmxnet3_TxCompDesc));
 450         tq->comp_ring.next2proc = 0;
 451         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 452
 453         /* reset the bookkeeping data */
 454         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 455         for (i = 0; i < tq->tx_ring.size; i++)
 456                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 457
 458         /* stats are not reset */
 459 }
 460
 461
 462 static int
 463 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 464                   struct vmxnet3_adapter *adapter)
 465 {
 466         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 467                tq->comp_ring.base || tq->buf_info);
 468
 469         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 470                            * sizeof(struct Vmxnet3_TxDesc),
 471                            &tq->tx_ring.basePA);
 472         if (!tq->tx_ring.base) {
 473                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 474                        adapter->netdev->name);
 475                 goto err;
 476         }
 477
 478         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 479                              tq->data_ring.size *
 480                              sizeof(struct Vmxnet3_TxDataDesc),
 481                              &tq->data_ring.basePA);
 482         if (!tq->data_ring.base) {
 483                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 484                        adapter->netdev->name);
 485                 goto err;
 486         }
 487
 488         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 489                              tq->comp_ring.size *
 490                              sizeof(struct Vmxnet3_TxCompDesc),
 491                              &tq->comp_ring.basePA);
 492         if (!tq->comp_ring.base) {
 493                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 494                        adapter->netdev->name);
 495                 goto err;
 496         }
 497
 498         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 499                                GFP_KERNEL);
 500         if (!tq->buf_info) {
 501                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
 502                        adapter->netdev->name);
 503                 goto err;
 504         }
 505
 506         return 0;
 507
 508 err:
 509         vmxnet3_tq_destroy(tq, adapter);
 510         return -ENOMEM;
 511 }
 512
 513
 514 /*
 515  *    starting from ring->next2fill, allocate rx buffers for the given ring
 516  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 517  *    are allocated or allocation fails
 518  */
 519
 520 static int
 521 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 522                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 523 {
 524         int num_allocated = 0;
 525         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 526         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 527         u32 val;
 528
 529         while (num_allocated < num_to_alloc) {
 530                 struct vmxnet3_rx_buf_info *rbi;
 531                 union Vmxnet3_GenericDesc *gd;
 532
 533                 rbi = rbi_base + ring->next2fill;
 534                 gd = ring->base + ring->next2fill;
 535
 536                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 537                         if (rbi->skb == NULL) {
 538                                 rbi->skb = dev_alloc_skb(rbi->len +
 539                                                          NET_IP_ALIGN);
 540                                 if (unlikely(rbi->skb == NULL)) {
 541                                         rq->stats.rx_buf_alloc_failure++;
 542                                         break;
 543                                 }
 544                                 rbi->skb->dev = adapter->netdev;
 545
 546                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 547                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 548                                                 rbi->skb->data, rbi->len,
 549                                                 PCI_DMA_FROMDEVICE);
 550                         } else {
 551                                 /* rx buffer skipped by the device */
 552                         }
 553                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 554                 } else {
 555                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 556                                rbi->len  != PAGE_SIZE);
 557
 558                         if (rbi->page == NULL) {
 559                                 rbi->page = alloc_page(GFP_ATOMIC);
 560                                 if (unlikely(rbi->page == NULL)) {
 561                                         rq->stats.rx_buf_alloc_failure++;
 562                                         break;
 563                                 }
 564                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 565                                                 rbi->page, 0, PAGE_SIZE,
 566                                                 PCI_DMA_FROMDEVICE);
 567                         } else {
 568                                 /* rx buffers skipped by the device */
 569                         }
 570                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 571                 }
 572
 573                 BUG_ON(rbi->dma_addr == 0);
 574                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 575                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
 576                                            | val | rbi->len);
 577
 578                 num_allocated++;
 579                 vmxnet3_cmd_ring_adv_next2fill(ring);
 580         }
 581         rq->uncommitted[ring_idx] += num_allocated;
 582
 583         dev_dbg(&adapter->netdev->dev,
 584                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 585                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
 586                 ring->next2comp, rq->uncommitted[ring_idx]);
 587
 588         /* so that the device can distinguish a full ring and an empty ring */
 589         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 590
 591         return num_allocated;
 592 }
 593
 594
 595 static void
 596 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 597                     struct vmxnet3_rx_buf_info *rbi)
 598 {
 599         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 600                 skb_shinfo(skb)->nr_frags;
 601
 602         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 603
 604         frag->page = rbi->page;
 605         frag->page_offset = 0;
 606         frag->size = rcd->len;
 607         skb->data_len += frag->size;
 608         skb_shinfo(skb)->nr_frags++;
 609 }
 610
 611
 612 static void
 613 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 614                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 615                 struct vmxnet3_adapter *adapter)
 616 {
 617         u32 dw2, len;
 618         unsigned long buf_offset;
 619         int i;
 620         union Vmxnet3_GenericDesc *gdesc;
 621         struct vmxnet3_tx_buf_info *tbi = NULL;
 622
 623         BUG_ON(ctx->copy_size > skb_headlen(skb));
 624
 625         /* use the previous gen bit for the SOP desc */
 626         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 627
 628         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 629         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 630
 631         /* no need to map the buffer if headers are copied */
 632         if (ctx->copy_size) {
 633                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 634                                         tq->tx_ring.next2fill *
 635                                         sizeof(struct Vmxnet3_TxDataDesc));
 636                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 637                 ctx->sop_txd->dword[3] = 0;
 638
 639                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 640                 tbi->map_type = VMXNET3_MAP_NONE;
 641
 642                 dev_dbg(&adapter->netdev->dev,
 643                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 644                         tq->tx_ring.next2fill,
 645                         le64_to_cpu(ctx->sop_txd->txd.addr),
 646                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 647                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 648
 649                 /* use the right gen for non-SOP desc */
 650                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 651         }
 652
 653         /* linear part can use multiple tx desc if it's big */
 654         len = skb_headlen(skb) - ctx->copy_size;
 655         buf_offset = ctx->copy_size;
 656         while (len) {
 657                 u32 buf_size;
 658
 659                 buf_size = len > VMXNET3_MAX_TX_BUF_SIZE ?
 660                            VMXNET3_MAX_TX_BUF_SIZE : len;
 661
 662                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 663                 tbi->map_type = VMXNET3_MAP_SINGLE;
 664                 tbi->dma_addr = pci_map_single(adapter->pdev,
 665                                 skb->data + buf_offset, buf_size,
 666                                 PCI_DMA_TODEVICE);
 667
 668                 tbi->len = buf_size; /* this automatically convert 2^14 to 0 */
 669
 670                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 671                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 672
 673                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 674                 gdesc->dword[2] = cpu_to_le32(dw2 | buf_size);
 675                 gdesc->dword[3] = 0;
 676
 677                 dev_dbg(&adapter->netdev->dev,
 678                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 679                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 680                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 681                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 682                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 683
 684                 len -= buf_size;
 685                 buf_offset += buf_size;
 686         }
 687
 688         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 689                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 690
 691                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 692                 tbi->map_type = VMXNET3_MAP_PAGE;
 693                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
 694                                              frag->page_offset, frag->size,
 695                                              PCI_DMA_TODEVICE);
 696
 697                 tbi->len = frag->size;
 698
 699                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 700                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 701
 702                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 703                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
 704                 gdesc->dword[3] = 0;
 705
 706                 dev_dbg(&adapter->netdev->dev,
 707                         "txd[%u]: 0x%llu %u %u\n",
 708                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 709                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 710                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 711                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 712         }
 713
 714         ctx->eop_txd = gdesc;
 715
 716         /* set the last buf_info for the pkt */
 717         tbi->skb = skb;
 718         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 719 }
 720
 721
 722 /*
 723  *    parse and copy relevant protocol headers:
 724  *      For a tso pkt, relevant headers are L2/3/4 including options
 725  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 726  *      if it's a TCP/UDP pkt
 727  *
 728  * Returns:
 729  *    -1:  error happens during parsing
 730  *     0:  protocol headers parsed, but too big to be copied
 731  *     1:  protocol headers parsed and copied
 732  *
 733  * Other effects:
 734  *    1. related *ctx fields are updated.
 735  *    2. ctx->copy_size is # of bytes copied
 736  *    3. the portion copied is guaranteed to be in the linear part
 737  *
 738  */
 739 static int
 740 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 741                            struct vmxnet3_tx_ctx *ctx,
 742                            struct vmxnet3_adapter *adapter)
 743 {
 744         struct Vmxnet3_TxDataDesc *tdd;
 745
 746         if (ctx->mss) {
 747                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 748                 ctx->l4_hdr_size = ((struct tcphdr *)
 749                                    skb_transport_header(skb))->doff * 4;
 750                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 751         } else {
 752                 unsigned int pull_size;
 753
 754                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 755                         ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 756
 757                         if (ctx->ipv4) {
 758                                 struct iphdr *iph = (struct iphdr *)
 759                                                     skb_network_header(skb);
 760                                 if (iph->protocol == IPPROTO_TCP) {
 761                                         pull_size = ctx->eth_ip_hdr_size +
 762                                                     sizeof(struct tcphdr);
 763
 764                                         if (unlikely(!pskb_may_pull(skb,
 765                                                                 pull_size))) {
 766                                                 goto err;
 767                                         }
 768                                         ctx->l4_hdr_size = ((struct tcphdr *)
 769                                            skb_transport_header(skb))->doff * 4;
 770                                 } else if (iph->protocol == IPPROTO_UDP) {
 771                                         ctx->l4_hdr_size =
 772                                                         sizeof(struct udphdr);
 773                                 } else {
 774                                         ctx->l4_hdr_size = 0;
 775                                 }
 776                         } else {
 777                                 /* for simplicity, don't copy L4 headers */
 778                                 ctx->l4_hdr_size = 0;
 779                         }
 780                         ctx->copy_size = ctx->eth_ip_hdr_size +
 781                                          ctx->l4_hdr_size;
 782                 } else {
 783                         ctx->eth_ip_hdr_size = 0;
 784                         ctx->l4_hdr_size = 0;
 785                         /* copy as much as allowed */
 786                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 787                                              , skb_headlen(skb));
 788                 }
 789
 790                 /* make sure headers are accessible directly */
 791                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 792                         goto err;
 793         }
 794
 795         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 796                 tq->stats.oversized_hdr++;
 797                 ctx->copy_size = 0;
 798                 return 0;
 799         }
 800
 801         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 802
 803         memcpy(tdd->data, skb->data, ctx->copy_size);
 804         dev_dbg(&adapter->netdev->dev,
 805                 "copy %u bytes to dataRing[%u]\n",
 806                 ctx->copy_size, tq->tx_ring.next2fill);
 807         return 1;
 808
 809 err:
 810         return -1;
 811 }
 812
 813
 814 static void
 815 vmxnet3_prepare_tso(struct sk_buff *skb,
 816                     struct vmxnet3_tx_ctx *ctx)
 817 {
 818         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
 819         if (ctx->ipv4) {
 820                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
 821                 iph->check = 0;
 822                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 823                                                  IPPROTO_TCP, 0);
 824         } else {
 825                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
 826                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 827                                                IPPROTO_TCP, 0);
 828         }
 829 }
 830
 831
 832 /*
 833  * Transmits a pkt thru a given tq
 834  * Returns:
 835  *    NETDEV_TX_OK:      descriptors are setup successfully
 836  *    NETDEV_TX_OK:      error occured, the pkt is dropped
 837  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 838  *
 839  * Side-effects:
 840  *    1. tx ring may be changed
 841  *    2. tq stats may be updated accordingly
 842  *    3. shared->txNumDeferred may be updated
 843  */
 844
 845 static int
 846 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 847                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 848 {
 849         int ret;
 850         u32 count;
 851         unsigned long flags;
 852         struct vmxnet3_tx_ctx ctx;
 853         union Vmxnet3_GenericDesc *gdesc;
 854 #ifdef __BIG_ENDIAN_BITFIELD
 855         /* Use temporary descriptor to avoid touching bits multiple times */
 856         union Vmxnet3_GenericDesc tempTxDesc;
 857 #endif
 858
 859         /* conservatively estimate # of descriptors to use */
 860         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
 861                 skb_shinfo(skb)->nr_frags + 1;
 862
 863         ctx.ipv4 = (skb->protocol == __constant_ntohs(ETH_P_IP));
 864
 865         ctx.mss = skb_shinfo(skb)->gso_size;
 866         if (ctx.mss) {
 867                 if (skb_header_cloned(skb)) {
 868                         if (unlikely(pskb_expand_head(skb, 0, 0,
 869                                                       GFP_ATOMIC) != 0)) {
 870                                 tq->stats.drop_tso++;
 871                                 goto drop_pkt;
 872                         }
 873                         tq->stats.copy_skb_header++;
 874                 }
 875                 vmxnet3_prepare_tso(skb, &ctx);
 876         } else {
 877                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 878
 879                         /* non-tso pkts must not use more than
 880                          * VMXNET3_MAX_TXD_PER_PKT entries
 881                          */
 882                         if (skb_linearize(skb) != 0) {
 883                                 tq->stats.drop_too_many_frags++;
 884                                 goto drop_pkt;
 885                         }
 886                         tq->stats.linearized++;
 887
 888                         /* recalculate the # of descriptors to use */
 889                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 890                 }
 891         }
 892
 893         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 894         if (ret >= 0) {
 895                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 896                 /* hdrs parsed, check against other limits */
 897                 if (ctx.mss) {
 898                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 899                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 900                                 goto hdr_too_big;
 901                         }
 902                 } else {
 903                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 904                                 if (unlikely(ctx.eth_ip_hdr_size +
 905                                              skb->csum_offset >
 906                                              VMXNET3_MAX_CSUM_OFFSET)) {
 907                                         goto hdr_too_big;
 908                                 }
 909                         }
 910                 }
 911         } else {
 912                 tq->stats.drop_hdr_inspect_err++;
 913                 goto drop_pkt;
 914         }
 915
 916         spin_lock_irqsave(&tq->tx_lock, flags);
 917
 918         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 919                 tq->stats.tx_ring_full++;
 920                 dev_dbg(&adapter->netdev->dev,
 921                         "tx queue stopped on %s, next2comp %u"
 922                         " next2fill %u\n", adapter->netdev->name,
 923                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 924
 925                 vmxnet3_tq_stop(tq, adapter);
 926                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 927                 return NETDEV_TX_BUSY;
 928         }
 929
 930         /* fill tx descs related to addr & len */
 931         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
 932
 933         /* setup the EOP desc */
 934         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
 935
 936         /* setup the SOP desc */
 937 #ifdef __BIG_ENDIAN_BITFIELD
 938         gdesc = &tempTxDesc;
 939         gdesc->dword[2] = ctx.sop_txd->dword[2];
 940         gdesc->dword[3] = ctx.sop_txd->dword[3];
 941 #else
 942         gdesc = ctx.sop_txd;
 943 #endif
 944         if (ctx.mss) {
 945                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
 946                 gdesc->txd.om = VMXNET3_OM_TSO;
 947                 gdesc->txd.msscof = ctx.mss;
 948                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
 949                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
 950         } else {
 951                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 952                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
 953                         gdesc->txd.om = VMXNET3_OM_CSUM;
 954                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
 955                                             skb->csum_offset;
 956                 } else {
 957                         gdesc->txd.om = 0;
 958                         gdesc->txd.msscof = 0;
 959                 }
 960                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
 961         }
 962
 963         if (vlan_tx_tag_present(skb)) {
 964                 gdesc->txd.ti = 1;
 965                 gdesc->txd.tci = vlan_tx_tag_get(skb);
 966         }
 967
 968         /* finally flips the GEN bit of the SOP desc. */
 969         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
 970                                                   VMXNET3_TXD_GEN);
 971 #ifdef __BIG_ENDIAN_BITFIELD
 972         /* Finished updating in bitfields of Tx Desc, so write them in original
 973          * place.
 974          */
 975         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
 976                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
 977         gdesc = ctx.sop_txd;
 978 #endif
 979         dev_dbg(&adapter->netdev->dev,
 980                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
 981                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
 982                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
 983                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
 984
 985         spin_unlock_irqrestore(&tq->tx_lock, flags);
 986
 987         if (le32_to_cpu(tq->shared->txNumDeferred) >=
 988                                         le32_to_cpu(tq->shared->txThreshold)) {
 989                 tq->shared->txNumDeferred = 0;
 990                 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_TXPROD,
 991                                        tq->tx_ring.next2fill);
 992         }
 993         netdev->trans_start = jiffies;
 994
 995         return NETDEV_TX_OK;
 996
 997 hdr_too_big:
 998         tq->stats.drop_oversized_hdr++;
 999 drop_pkt:
1000         tq->stats.drop_total++;
1001         dev_kfree_skb(skb);
1002         return NETDEV_TX_OK;
1003 }
1004
1005
1006 static netdev_tx_t
1007 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1008 {
1009         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1010
1011         return vmxnet3_tq_xmit(skb, &adapter->tx_queue, adapter, netdev);
1012 }
1013
1014
1015 static void
1016 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1017                 struct sk_buff *skb,
1018                 union Vmxnet3_GenericDesc *gdesc)
1019 {
1020         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1021                 /* typical case: TCP/UDP over IP and both csums are correct */
1022                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1023                                                         VMXNET3_RCD_CSUM_OK) {
1024                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1025                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1026                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1027                         BUG_ON(gdesc->rcd.frg);
1028                 } else {
1029                         if (gdesc->rcd.csum) {
1030                                 skb->csum = htons(gdesc->rcd.csum);
1031                                 skb->ip_summed = CHECKSUM_PARTIAL;
1032                         } else {
1033                                 skb->ip_summed = CHECKSUM_NONE;
1034                         }
1035                 }
1036         } else {
1037                 skb->ip_summed = CHECKSUM_NONE;
1038         }
1039 }
1040
1041
1042 static void
1043 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1044                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1045 {
1046         rq->stats.drop_err++;
1047         if (!rcd->fcs)
1048                 rq->stats.drop_fcs++;
1049
1050         rq->stats.drop_total++;
1051
1052         /*
1053          * We do not unmap and chain the rx buffer to the skb.
1054          * We basically pretend this buffer is not used and will be recycled
1055          * by vmxnet3_rq_alloc_rx_buf()
1056          */
1057
1058         /*
1059          * ctx->skb may be NULL if this is the first and the only one
1060          * desc for the pkt
1061          */
1062         if (ctx->skb)
1063                 dev_kfree_skb_irq(ctx->skb);
1064
1065         ctx->skb = NULL;
1066 }
1067
1068
1069 static int
1070 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1071                        struct vmxnet3_adapter *adapter, int quota)
1072 {
1073         static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
1074         u32 num_rxd = 0;
1075         struct Vmxnet3_RxCompDesc *rcd;
1076         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1077 #ifdef __BIG_ENDIAN_BITFIELD
1078         struct Vmxnet3_RxDesc rxCmdDesc;
1079         struct Vmxnet3_RxCompDesc rxComp;
1080 #endif
1081         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1082                           &rxComp);
1083         while (rcd->gen == rq->comp_ring.gen) {
1084                 struct vmxnet3_rx_buf_info *rbi;
1085                 struct sk_buff *skb;
1086                 int num_to_alloc;
1087                 struct Vmxnet3_RxDesc *rxd;
1088                 u32 idx, ring_idx;
1089
1090                 if (num_rxd >= quota) {
1091                         /* we may stop even before we see the EOP desc of
1092                          * the current pkt
1093                          */
1094                         break;
1095                 }
1096                 num_rxd++;
1097
1098                 idx = rcd->rxdIdx;
1099                 ring_idx = rcd->rqID == rq->qid ? 0 : 1;
1100                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1101                                   &rxCmdDesc);
1102                 rbi = rq->buf_info[ring_idx] + idx;
1103
1104                 BUG_ON(rxd->addr != rbi->dma_addr ||
1105                        rxd->len != rbi->len);
1106
1107                 if (unlikely(rcd->eop && rcd->err)) {
1108                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1109                         goto rcd_done;
1110                 }
1111
1112                 if (rcd->sop) { /* first buf of the pkt */
1113                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1114                                rcd->rqID != rq->qid);
1115
1116                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1117                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1118
1119                         if (unlikely(rcd->len == 0)) {
1120                                 /* Pretend the rx buffer is skipped. */
1121                                 BUG_ON(!(rcd->sop && rcd->eop));
1122                                 dev_dbg(&adapter->netdev->dev,
1123                                         "rxRing[%u][%u] 0 length\n",
1124                                         ring_idx, idx);
1125                                 goto rcd_done;
1126                         }
1127
1128                         ctx->skb = rbi->skb;
1129                         rbi->skb = NULL;
1130
1131                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1132                                          PCI_DMA_FROMDEVICE);
1133
1134                         skb_put(ctx->skb, rcd->len);
1135                 } else {
1136                         BUG_ON(ctx->skb == NULL);
1137                         /* non SOP buffer must be type 1 in most cases */
1138                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1139                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1140
1141                                 if (rcd->len) {
1142                                         pci_unmap_page(adapter->pdev,
1143                                                        rbi->dma_addr, rbi->len,
1144                                                        PCI_DMA_FROMDEVICE);
1145
1146                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1147                                         rbi->page = NULL;
1148                                 }
1149                         } else {
1150                                 /*
1151                                  * The only time a non-SOP buffer is type 0 is
1152                                  * when it's EOP and error flag is raised, which
1153                                  * has already been handled.
1154                                  */
1155                                 BUG_ON(true);
1156                         }
1157                 }
1158
1159                 skb = ctx->skb;
1160                 if (rcd->eop) {
1161                         skb->len += skb->data_len;
1162                         skb->truesize += skb->data_len;
1163
1164                         vmxnet3_rx_csum(adapter, skb,
1165                                         (union Vmxnet3_GenericDesc *)rcd);
1166                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1167
1168                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1169                                 vlan_hwaccel_receive_skb(skb,
1170                                                 adapter->vlan_grp, rcd->tci);
1171                         } else {
1172                                 netif_receive_skb(skb);
1173                         }
1174
1175                         adapter->netdev->last_rx = jiffies;
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 * 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                         int i;
1677                         struct dev_mc_list *mc = netdev->mc_list;
1678
1679                         for (i = 0; i < netdev->mc_count; i++) {
1680                                 BUG_ON(!mc);
1681                                 memcpy(buf + i * ETH_ALEN, mc->dmi_addr,
1682                                        ETH_ALEN);
1683                                 mc = mc->next;
1684                         }
1685                 }
1686         }
1687         return buf;
1688 }
1689
1690
1691 static void
1692 vmxnet3_set_mc(struct net_device *netdev)
1693 {
1694         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1695         struct Vmxnet3_RxFilterConf *rxConf =
1696                                         &adapter->shared->devRead.rxFilterConf;
1697         u8 *new_table = NULL;
1698         u32 new_mode = VMXNET3_RXM_UCAST;
1699
1700         if (netdev->flags & IFF_PROMISC)
1701                 new_mode |= VMXNET3_RXM_PROMISC;
1702
1703         if (netdev->flags & IFF_BROADCAST)
1704                 new_mode |= VMXNET3_RXM_BCAST;
1705
1706         if (netdev->flags & IFF_ALLMULTI)
1707                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1708         else
1709                 if (netdev->mc_count > 0) {
1710                         new_table = vmxnet3_copy_mc(netdev);
1711                         if (new_table) {
1712                                 new_mode |= VMXNET3_RXM_MCAST;
1713                                 rxConf->mfTableLen = cpu_to_le16(
1714                                                 netdev->mc_count * ETH_ALEN);
1715                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
1716                                                     new_table));
1717                         } else {
1718                                 printk(KERN_INFO "%s: failed to copy mcast list"
1719                                        ", setting ALL_MULTI\n", netdev->name);
1720                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1721                         }
1722                 }
1723
1724
1725         if (!(new_mode & VMXNET3_RXM_MCAST)) {
1726                 rxConf->mfTableLen = 0;
1727                 rxConf->mfTablePA = 0;
1728         }
1729
1730         if (new_mode != rxConf->rxMode) {
1731                 rxConf->rxMode = cpu_to_le32(new_mode);
1732                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1733                                        VMXNET3_CMD_UPDATE_RX_MODE);
1734         }
1735
1736         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1737                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
1738
1739         kfree(new_table);
1740 }
1741
1742
1743 /*
1744  *   Set up driver_shared based on settings in adapter.
1745  */
1746
1747 static void
1748 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
1749 {
1750         struct Vmxnet3_DriverShared *shared = adapter->shared;
1751         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1752         struct Vmxnet3_TxQueueConf *tqc;
1753         struct Vmxnet3_RxQueueConf *rqc;
1754         int i;
1755
1756         memset(shared, 0, sizeof(*shared));
1757
1758         /* driver settings */
1759         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
1760         devRead->misc.driverInfo.version = cpu_to_le32(
1761                                                 VMXNET3_DRIVER_VERSION_NUM);
1762         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
1763                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
1764         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
1765         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
1766                                 *((u32 *)&devRead->misc.driverInfo.gos));
1767         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
1768         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
1769
1770         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
1771         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
1772
1773         /* set up feature flags */
1774         if (adapter->rxcsum)
1775                 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXCSUM);
1776
1777         if (adapter->lro) {
1778                 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_LRO);
1779                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
1780         }
1781         if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX)
1782                         && adapter->vlan_grp) {
1783                 set_flag_le64(&devRead->misc.uptFeatures, UPT1_F_RXVLAN);
1784         }
1785
1786         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
1787         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
1788         devRead->misc.queueDescLen = cpu_to_le32(
1789                                      sizeof(struct Vmxnet3_TxQueueDesc) +
1790                                      sizeof(struct Vmxnet3_RxQueueDesc));
1791
1792         /* tx queue settings */
1793         BUG_ON(adapter->tx_queue.tx_ring.base == NULL);
1794
1795         devRead->misc.numTxQueues = 1;
1796         tqc = &adapter->tqd_start->conf;
1797         tqc->txRingBasePA   = cpu_to_le64(adapter->tx_queue.tx_ring.basePA);
1798         tqc->dataRingBasePA = cpu_to_le64(adapter->tx_queue.data_ring.basePA);
1799         tqc->compRingBasePA = cpu_to_le64(adapter->tx_queue.comp_ring.basePA);
1800         tqc->ddPA           = cpu_to_le64(virt_to_phys(
1801                                                 adapter->tx_queue.buf_info));
1802         tqc->txRingSize     = cpu_to_le32(adapter->tx_queue.tx_ring.size);
1803         tqc->dataRingSize   = cpu_to_le32(adapter->tx_queue.data_ring.size);
1804         tqc->compRingSize   = cpu_to_le32(adapter->tx_queue.comp_ring.size);
1805         tqc->ddLen          = cpu_to_le32(sizeof(struct vmxnet3_tx_buf_info) *
1806                               tqc->txRingSize);
1807         tqc->intrIdx        = adapter->tx_queue.comp_ring.intr_idx;
1808
1809         /* rx queue settings */
1810         devRead->misc.numRxQueues = 1;
1811         rqc = &adapter->rqd_start->conf;
1812         rqc->rxRingBasePA[0] = cpu_to_le64(adapter->rx_queue.rx_ring[0].basePA);
1813         rqc->rxRingBasePA[1] = cpu_to_le64(adapter->rx_queue.rx_ring[1].basePA);
1814         rqc->compRingBasePA  = cpu_to_le64(adapter->rx_queue.comp_ring.basePA);
1815         rqc->ddPA            = cpu_to_le64(virt_to_phys(
1816                                                 adapter->rx_queue.buf_info));
1817         rqc->rxRingSize[0]   = cpu_to_le32(adapter->rx_queue.rx_ring[0].size);
1818         rqc->rxRingSize[1]   = cpu_to_le32(adapter->rx_queue.rx_ring[1].size);
1819         rqc->compRingSize    = cpu_to_le32(adapter->rx_queue.comp_ring.size);
1820         rqc->ddLen           = cpu_to_le32(sizeof(struct vmxnet3_rx_buf_info) *
1821                                (rqc->rxRingSize[0] + rqc->rxRingSize[1]));
1822         rqc->intrIdx         = adapter->rx_queue.comp_ring.intr_idx;
1823
1824         /* intr settings */
1825         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
1826                                      VMXNET3_IMM_AUTO;
1827         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
1828         for (i = 0; i < adapter->intr.num_intrs; i++)
1829                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
1830
1831         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
1832
1833         /* rx filter settings */
1834         devRead->rxFilterConf.rxMode = 0;
1835         vmxnet3_restore_vlan(adapter);
1836         /* the rest are already zeroed */
1837 }
1838
1839
1840 int
1841 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
1842 {
1843         int err;
1844         u32 ret;
1845
1846         dev_dbg(&adapter->netdev->dev,
1847                 "%s: skb_buf_size %d, rx_buf_per_pkt %d, ring sizes"
1848                 " %u %u %u\n", adapter->netdev->name, adapter->skb_buf_size,
1849                 adapter->rx_buf_per_pkt, adapter->tx_queue.tx_ring.size,
1850                 adapter->rx_queue.rx_ring[0].size,
1851                 adapter->rx_queue.rx_ring[1].size);
1852
1853         vmxnet3_tq_init(&adapter->tx_queue, adapter);
1854         err = vmxnet3_rq_init(&adapter->rx_queue, adapter);
1855         if (err) {
1856                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
1857                        adapter->netdev->name, err);
1858                 goto rq_err;
1859         }
1860
1861         err = vmxnet3_request_irqs(adapter);
1862         if (err) {
1863                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
1864                        adapter->netdev->name, err);
1865                 goto irq_err;
1866         }
1867
1868         vmxnet3_setup_driver_shared(adapter);
1869
1870         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
1871                                adapter->shared_pa));
1872         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
1873                                adapter->shared_pa));
1874         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1875                                VMXNET3_CMD_ACTIVATE_DEV);
1876         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
1877
1878         if (ret != 0) {
1879                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
1880                        adapter->netdev->name, ret);
1881                 err = -EINVAL;
1882                 goto activate_err;
1883         }
1884         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD,
1885                                adapter->rx_queue.rx_ring[0].next2fill);
1886         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD2,
1887                                adapter->rx_queue.rx_ring[1].next2fill);
1888
1889         /* Apply the rx filter settins last. */
1890         vmxnet3_set_mc(adapter->netdev);
1891
1892         /*
1893          * Check link state when first activating device. It will start the
1894          * tx queue if the link is up.
1895          */
1896         vmxnet3_check_link(adapter);
1897
1898         napi_enable(&adapter->napi);
1899         vmxnet3_enable_all_intrs(adapter);
1900         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
1901         return 0;
1902
1903 activate_err:
1904         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
1905         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
1906         vmxnet3_free_irqs(adapter);
1907 irq_err:
1908 rq_err:
1909         /* free up buffers we allocated */
1910         vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1911         return err;
1912 }
1913
1914
1915 void
1916 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
1917 {
1918         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
1919 }
1920
1921
1922 int
1923 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
1924 {
1925         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
1926                 return 0;
1927
1928
1929         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1930                                VMXNET3_CMD_QUIESCE_DEV);
1931         vmxnet3_disable_all_intrs(adapter);
1932
1933         napi_disable(&adapter->napi);
1934         netif_tx_disable(adapter->netdev);
1935         adapter->link_speed = 0;
1936         netif_carrier_off(adapter->netdev);
1937
1938         vmxnet3_tq_cleanup(&adapter->tx_queue, adapter);
1939         vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1940         vmxnet3_free_irqs(adapter);
1941         return 0;
1942 }
1943
1944
1945 static void
1946 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
1947 {
1948         u32 tmp;
1949
1950         tmp = *(u32 *)mac;
1951         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
1952
1953         tmp = (mac[5] << 8) | mac[4];
1954         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
1955 }
1956
1957
1958 static int
1959 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
1960 {
1961         struct sockaddr *addr = p;
1962         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1963
1964         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1965         vmxnet3_write_mac_addr(adapter, addr->sa_data);
1966
1967         return 0;
1968 }
1969
1970
1971 /* ==================== initialization and cleanup routines ============ */
1972
1973 static int
1974 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
1975 {
1976         int err;
1977         unsigned long mmio_start, mmio_len;
1978         struct pci_dev *pdev = adapter->pdev;
1979
1980         err = pci_enable_device(pdev);
1981         if (err) {
1982                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
1983                        pci_name(pdev), err);
1984                 return err;
1985         }
1986
1987         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1988                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
1989                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
1990                                "for adapter %s\n", pci_name(pdev));
1991                         err = -EIO;
1992                         goto err_set_mask;
1993                 }
1994                 *dma64 = true;
1995         } else {
1996                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
1997                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
1998                                "%s\n",  pci_name(pdev));
1999                         err = -EIO;
2000                         goto err_set_mask;
2001                 }
2002                 *dma64 = false;
2003         }
2004
2005         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2006                                            vmxnet3_driver_name);
2007         if (err) {
2008                 printk(KERN_ERR "Failed to request region for adapter %s: "
2009                        "error %d\n", pci_name(pdev), err);
2010                 goto err_set_mask;
2011         }
2012
2013         pci_set_master(pdev);
2014
2015         mmio_start = pci_resource_start(pdev, 0);
2016         mmio_len = pci_resource_len(pdev, 0);
2017         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2018         if (!adapter->hw_addr0) {
2019                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2020                        pci_name(pdev));
2021                 err = -EIO;
2022                 goto err_ioremap;
2023         }
2024
2025         mmio_start = pci_resource_start(pdev, 1);
2026         mmio_len = pci_resource_len(pdev, 1);
2027         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2028         if (!adapter->hw_addr1) {
2029                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2030                        pci_name(pdev));
2031                 err = -EIO;
2032                 goto err_bar1;
2033         }
2034         return 0;
2035
2036 err_bar1:
2037         iounmap(adapter->hw_addr0);
2038 err_ioremap:
2039         pci_release_selected_regions(pdev, (1 << 2) - 1);
2040 err_set_mask:
2041         pci_disable_device(pdev);
2042         return err;
2043 }
2044
2045
2046 static void
2047 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2048 {
2049         BUG_ON(!adapter->pdev);
2050
2051         iounmap(adapter->hw_addr0);
2052         iounmap(adapter->hw_addr1);
2053         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2054         pci_disable_device(adapter->pdev);
2055 }
2056
2057
2058 static void
2059 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2060 {
2061         size_t sz;
2062
2063         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2064                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2065                 adapter->skb_buf_size = adapter->netdev->mtu +
2066                                         VMXNET3_MAX_ETH_HDR_SIZE;
2067                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2068                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2069
2070                 adapter->rx_buf_per_pkt = 1;
2071         } else {
2072                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2073                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2074                                             VMXNET3_MAX_ETH_HDR_SIZE;
2075                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2076         }
2077
2078         /*
2079          * for simplicity, force the ring0 size to be a multiple of
2080          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2081          */
2082         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2083         adapter->rx_queue.rx_ring[0].size = (adapter->rx_queue.rx_ring[0].size +
2084                                              sz - 1) / sz * sz;
2085         adapter->rx_queue.rx_ring[0].size = min_t(u32,
2086                                             adapter->rx_queue.rx_ring[0].size,
2087                                             VMXNET3_RX_RING_MAX_SIZE / sz * sz);
2088 }
2089
2090
2091 int
2092 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2093                       u32 rx_ring_size, u32 rx_ring2_size)
2094 {
2095         int err;
2096
2097         adapter->tx_queue.tx_ring.size   = tx_ring_size;
2098         adapter->tx_queue.data_ring.size = tx_ring_size;
2099         adapter->tx_queue.comp_ring.size = tx_ring_size;
2100         adapter->tx_queue.shared = &adapter->tqd_start->ctrl;
2101         adapter->tx_queue.stopped = true;
2102         err = vmxnet3_tq_create(&adapter->tx_queue, adapter);
2103         if (err)
2104                 return err;
2105
2106         adapter->rx_queue.rx_ring[0].size = rx_ring_size;
2107         adapter->rx_queue.rx_ring[1].size = rx_ring2_size;
2108         vmxnet3_adjust_rx_ring_size(adapter);
2109         adapter->rx_queue.comp_ring.size  = adapter->rx_queue.rx_ring[0].size +
2110                                             adapter->rx_queue.rx_ring[1].size;
2111         adapter->rx_queue.qid  = 0;
2112         adapter->rx_queue.qid2 = 1;
2113         adapter->rx_queue.shared = &adapter->rqd_start->ctrl;
2114         err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2115         if (err)
2116                 vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2117
2118         return err;
2119 }
2120
2121 static int
2122 vmxnet3_open(struct net_device *netdev)
2123 {
2124         struct vmxnet3_adapter *adapter;
2125         int err;
2126
2127         adapter = netdev_priv(netdev);
2128
2129         spin_lock_init(&adapter->tx_queue.tx_lock);
2130
2131         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2132                                     VMXNET3_DEF_RX_RING_SIZE,
2133                                     VMXNET3_DEF_RX_RING_SIZE);
2134         if (err)
2135                 goto queue_err;
2136
2137         err = vmxnet3_activate_dev(adapter);
2138         if (err)
2139                 goto activate_err;
2140
2141         return 0;
2142
2143 activate_err:
2144         vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2145         vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2146 queue_err:
2147         return err;
2148 }
2149
2150
2151 static int
2152 vmxnet3_close(struct net_device *netdev)
2153 {
2154         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2155
2156         /*
2157          * Reset_work may be in the middle of resetting the device, wait for its
2158          * completion.
2159          */
2160         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2161                 msleep(1);
2162
2163         vmxnet3_quiesce_dev(adapter);
2164
2165         vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2166         vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2167
2168         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2169
2170
2171         return 0;
2172 }
2173
2174
2175 void
2176 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2177 {
2178         /*
2179          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2180          * vmxnet3_close() will deadlock.
2181          */
2182         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2183
2184         /* we need to enable NAPI, otherwise dev_close will deadlock */
2185         napi_enable(&adapter->napi);
2186         dev_close(adapter->netdev);
2187 }
2188
2189
2190 static int
2191 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2192 {
2193         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2194         int err = 0;
2195
2196         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2197                 return -EINVAL;
2198
2199         if (new_mtu > 1500 && !adapter->jumbo_frame)
2200                 return -EINVAL;
2201
2202         netdev->mtu = new_mtu;
2203
2204         /*
2205          * Reset_work may be in the middle of resetting the device, wait for its
2206          * completion.
2207          */
2208         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2209                 msleep(1);
2210
2211         if (netif_running(netdev)) {
2212                 vmxnet3_quiesce_dev(adapter);
2213                 vmxnet3_reset_dev(adapter);
2214
2215                 /* we need to re-create the rx queue based on the new mtu */
2216                 vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2217                 vmxnet3_adjust_rx_ring_size(adapter);
2218                 adapter->rx_queue.comp_ring.size  =
2219                                         adapter->rx_queue.rx_ring[0].size +
2220                                         adapter->rx_queue.rx_ring[1].size;
2221                 err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2222                 if (err) {
2223                         printk(KERN_ERR "%s: failed to re-create rx queue,"
2224                                 " error %d. Closing it.\n", netdev->name, err);
2225                         goto out;
2226                 }
2227
2228                 err = vmxnet3_activate_dev(adapter);
2229                 if (err) {
2230                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2231                                 "Closing it\n", netdev->name, err);
2232                         goto out;
2233                 }
2234         }
2235
2236 out:
2237         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2238         if (err)
2239                 vmxnet3_force_close(adapter);
2240
2241         return err;
2242 }
2243
2244
2245 static void
2246 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2247 {
2248         struct net_device *netdev = adapter->netdev;
2249
2250         netdev->features = NETIF_F_SG |
2251                 NETIF_F_HW_CSUM |
2252                 NETIF_F_HW_VLAN_TX |
2253                 NETIF_F_HW_VLAN_RX |
2254                 NETIF_F_HW_VLAN_FILTER |
2255                 NETIF_F_TSO |
2256                 NETIF_F_TSO6 |
2257                 NETIF_F_LRO;
2258
2259         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2260
2261         adapter->rxcsum = true;
2262         adapter->jumbo_frame = true;
2263         adapter->lro = true;
2264
2265         if (dma64) {
2266                 netdev->features |= NETIF_F_HIGHDMA;
2267                 printk(" highDMA");
2268         }
2269
2270         netdev->vlan_features = netdev->features;
2271         printk("\n");
2272 }
2273
2274
2275 static void
2276 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2277 {
2278         u32 tmp;
2279
2280         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2281         *(u32 *)mac = tmp;
2282
2283         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2284         mac[4] = tmp & 0xff;
2285         mac[5] = (tmp >> 8) & 0xff;
2286 }
2287
2288
2289 static void
2290 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2291 {
2292         u32 cfg;
2293
2294         /* intr settings */
2295         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2296                                VMXNET3_CMD_GET_CONF_INTR);
2297         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2298         adapter->intr.type = cfg & 0x3;
2299         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2300
2301         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2302                 int err;
2303
2304 #ifdef CONFIG_PCI_MSI
2305                 adapter->intr.msix_entries[0].entry = 0;
2306                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2307                                       VMXNET3_LINUX_MAX_MSIX_VECT);
2308                 if (!err) {
2309                         adapter->intr.num_intrs = 1;
2310                         adapter->intr.type = VMXNET3_IT_MSIX;
2311                         return;
2312                 }
2313 #endif
2314
2315                 err = pci_enable_msi(adapter->pdev);
2316                 if (!err) {
2317                         adapter->intr.num_intrs = 1;
2318                         adapter->intr.type = VMXNET3_IT_MSI;
2319                         return;
2320                 }
2321         }
2322
2323         adapter->intr.type = VMXNET3_IT_INTX;
2324
2325         /* INT-X related setting */
2326         adapter->intr.num_intrs = 1;
2327 }
2328
2329
2330 static void
2331 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2332 {
2333         if (adapter->intr.type == VMXNET3_IT_MSIX)
2334                 pci_disable_msix(adapter->pdev);
2335         else if (adapter->intr.type == VMXNET3_IT_MSI)
2336                 pci_disable_msi(adapter->pdev);
2337         else
2338                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2339 }
2340
2341
2342 static void
2343 vmxnet3_tx_timeout(struct net_device *netdev)
2344 {
2345         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2346         adapter->tx_timeout_count++;
2347
2348         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2349         schedule_work(&adapter->work);
2350 }
2351
2352
2353 static void
2354 vmxnet3_reset_work(struct work_struct *data)
2355 {
2356         struct vmxnet3_adapter *adapter;
2357
2358         adapter = container_of(data, struct vmxnet3_adapter, work);
2359
2360         /* if another thread is resetting the device, no need to proceed */
2361         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2362                 return;
2363
2364         /* if the device is closed, we must leave it alone */
2365         if (netif_running(adapter->netdev)) {
2366                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2367                 vmxnet3_quiesce_dev(adapter);
2368                 vmxnet3_reset_dev(adapter);
2369                 vmxnet3_activate_dev(adapter);
2370         } else {
2371                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2372         }
2373
2374         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2375 }
2376
2377
2378 static int __devinit
2379 vmxnet3_probe_device(struct pci_dev *pdev,
2380                      const struct pci_device_id *id)
2381 {
2382         static const struct net_device_ops vmxnet3_netdev_ops = {
2383                 .ndo_open = vmxnet3_open,
2384                 .ndo_stop = vmxnet3_close,
2385                 .ndo_start_xmit = vmxnet3_xmit_frame,
2386                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2387                 .ndo_change_mtu = vmxnet3_change_mtu,
2388                 .ndo_get_stats = vmxnet3_get_stats,
2389                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2390                 .ndo_set_multicast_list = vmxnet3_set_mc,
2391                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2392                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2393                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2394 #ifdef CONFIG_NET_POLL_CONTROLLER
2395                 .ndo_poll_controller = vmxnet3_netpoll,
2396 #endif
2397         };
2398         int err;
2399         bool dma64 = false; /* stupid gcc */
2400         u32 ver;
2401         struct net_device *netdev;
2402         struct vmxnet3_adapter *adapter;
2403         u8 mac[ETH_ALEN];
2404
2405         netdev = alloc_etherdev(sizeof(struct vmxnet3_adapter));
2406         if (!netdev) {
2407                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2408                         "%s\n", pci_name(pdev));
2409                 return -ENOMEM;
2410         }
2411
2412         pci_set_drvdata(pdev, netdev);
2413         adapter = netdev_priv(netdev);
2414         adapter->netdev = netdev;
2415         adapter->pdev = pdev;
2416
2417         adapter->shared = pci_alloc_consistent(adapter->pdev,
2418                           sizeof(struct Vmxnet3_DriverShared),
2419                           &adapter->shared_pa);
2420         if (!adapter->shared) {
2421                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2422                         pci_name(pdev));
2423                 err = -ENOMEM;
2424                 goto err_alloc_shared;
2425         }
2426
2427         adapter->tqd_start = pci_alloc_consistent(adapter->pdev,
2428                              sizeof(struct Vmxnet3_TxQueueDesc) +
2429                              sizeof(struct Vmxnet3_RxQueueDesc),
2430                              &adapter->queue_desc_pa);
2431
2432         if (!adapter->tqd_start) {
2433                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2434                         pci_name(pdev));
2435                 err = -ENOMEM;
2436                 goto err_alloc_queue_desc;
2437         }
2438         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start
2439                                                             + 1);
2440
2441         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2442         if (adapter->pm_conf == NULL) {
2443                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2444                         pci_name(pdev));
2445                 err = -ENOMEM;
2446                 goto err_alloc_pm;
2447         }
2448
2449         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2450         if (err < 0)
2451                 goto err_alloc_pci;
2452
2453         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2454         if (ver & 1) {
2455                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2456         } else {
2457                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2458                        " %s\n", ver, pci_name(pdev));
2459                 err = -EBUSY;
2460                 goto err_ver;
2461         }
2462
2463         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2464         if (ver & 1) {
2465                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2466         } else {
2467                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2468                        "adapter %s\n", ver, pci_name(pdev));
2469                 err = -EBUSY;
2470                 goto err_ver;
2471         }
2472
2473         vmxnet3_declare_features(adapter, dma64);
2474
2475         adapter->dev_number = atomic_read(&devices_found);
2476         vmxnet3_alloc_intr_resources(adapter);
2477
2478         vmxnet3_read_mac_addr(adapter, mac);
2479         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
2480
2481         netdev->netdev_ops = &vmxnet3_netdev_ops;
2482         netdev->watchdog_timeo = 5 * HZ;
2483         vmxnet3_set_ethtool_ops(netdev);
2484
2485         INIT_WORK(&adapter->work, vmxnet3_reset_work);
2486
2487         netif_napi_add(netdev, &adapter->napi, vmxnet3_poll, 64);
2488         SET_NETDEV_DEV(netdev, &pdev->dev);
2489         err = register_netdev(netdev);
2490
2491         if (err) {
2492                 printk(KERN_ERR "Failed to register adapter %s\n",
2493                         pci_name(pdev));
2494                 goto err_register;
2495         }
2496
2497         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2498         atomic_inc(&devices_found);
2499         return 0;
2500
2501 err_register:
2502         vmxnet3_free_intr_resources(adapter);
2503 err_ver:
2504         vmxnet3_free_pci_resources(adapter);
2505 err_alloc_pci:
2506         kfree(adapter->pm_conf);
2507 err_alloc_pm:
2508         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2509                             sizeof(struct Vmxnet3_RxQueueDesc),
2510                             adapter->tqd_start, adapter->queue_desc_pa);
2511 err_alloc_queue_desc:
2512         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2513                             adapter->shared, adapter->shared_pa);
2514 err_alloc_shared:
2515         pci_set_drvdata(pdev, NULL);
2516         free_netdev(netdev);
2517         return err;
2518 }
2519
2520
2521 static void __devexit
2522 vmxnet3_remove_device(struct pci_dev *pdev)
2523 {
2524         struct net_device *netdev = pci_get_drvdata(pdev);
2525         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2526
2527         flush_scheduled_work();
2528
2529         unregister_netdev(netdev);
2530
2531         vmxnet3_free_intr_resources(adapter);
2532         vmxnet3_free_pci_resources(adapter);
2533         kfree(adapter->pm_conf);
2534         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2535                             sizeof(struct Vmxnet3_RxQueueDesc),
2536                             adapter->tqd_start, adapter->queue_desc_pa);
2537         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2538                             adapter->shared, adapter->shared_pa);
2539         free_netdev(netdev);
2540 }
2541
2542
2543 #ifdef CONFIG_PM
2544
2545 static int
2546 vmxnet3_suspend(struct device *device)
2547 {
2548         struct pci_dev *pdev = to_pci_dev(device);
2549         struct net_device *netdev = pci_get_drvdata(pdev);
2550         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2551         struct Vmxnet3_PMConf *pmConf;
2552         struct ethhdr *ehdr;
2553         struct arphdr *ahdr;
2554         u8 *arpreq;
2555         struct in_device *in_dev;
2556         struct in_ifaddr *ifa;
2557         int i = 0;
2558
2559         if (!netif_running(netdev))
2560                 return 0;
2561
2562         vmxnet3_disable_all_intrs(adapter);
2563         vmxnet3_free_irqs(adapter);
2564         vmxnet3_free_intr_resources(adapter);
2565
2566         netif_device_detach(netdev);
2567         netif_stop_queue(netdev);
2568
2569         /* Create wake-up filters. */
2570         pmConf = adapter->pm_conf;
2571         memset(pmConf, 0, sizeof(*pmConf));
2572
2573         if (adapter->wol & WAKE_UCAST) {
2574                 pmConf->filters[i].patternSize = ETH_ALEN;
2575                 pmConf->filters[i].maskSize = 1;
2576                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
2577                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
2578
2579                 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER);
2580                 i++;
2581         }
2582
2583         if (adapter->wol & WAKE_ARP) {
2584                 in_dev = in_dev_get(netdev);
2585                 if (!in_dev)
2586                         goto skip_arp;
2587
2588                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
2589                 if (!ifa)
2590                         goto skip_arp;
2591
2592                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
2593                         sizeof(struct arphdr) +         /* ARP header */
2594                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
2595                         2 * sizeof(u32);        /*2 IPv4 addresses */
2596                 pmConf->filters[i].maskSize =
2597                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
2598
2599                 /* ETH_P_ARP in Ethernet header. */
2600                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
2601                 ehdr->h_proto = htons(ETH_P_ARP);
2602
2603                 /* ARPOP_REQUEST in ARP header. */
2604                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
2605                 ahdr->ar_op = htons(ARPOP_REQUEST);
2606                 arpreq = (u8 *)(ahdr + 1);
2607
2608                 /* The Unicast IPv4 address in 'tip' field. */
2609                 arpreq += 2 * ETH_ALEN + sizeof(u32);
2610                 *(u32 *)arpreq = ifa->ifa_address;
2611
2612                 /* The mask for the relevant bits. */
2613                 pmConf->filters[i].mask[0] = 0x00;
2614                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
2615                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
2616                 pmConf->filters[i].mask[3] = 0x00;
2617                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
2618                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
2619                 in_dev_put(in_dev);
2620
2621                 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_FILTER);
2622                 i++;
2623         }
2624
2625 skip_arp:
2626         if (adapter->wol & WAKE_MAGIC)
2627                 set_flag_le16(&pmConf->wakeUpEvents, VMXNET3_PM_WAKEUP_MAGIC);
2628
2629         pmConf->numFilters = i;
2630
2631         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2632         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2633                                                                   *pmConf));
2634         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
2635                                                                  pmConf));
2636
2637         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2638                                VMXNET3_CMD_UPDATE_PMCFG);
2639
2640         pci_save_state(pdev);
2641         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
2642                         adapter->wol);
2643         pci_disable_device(pdev);
2644         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
2645
2646         return 0;
2647 }
2648
2649
2650 static int
2651 vmxnet3_resume(struct device *device)
2652 {
2653         int err;
2654         struct pci_dev *pdev = to_pci_dev(device);
2655         struct net_device *netdev = pci_get_drvdata(pdev);
2656         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2657         struct Vmxnet3_PMConf *pmConf;
2658
2659         if (!netif_running(netdev))
2660                 return 0;
2661
2662         /* Destroy wake-up filters. */
2663         pmConf = adapter->pm_conf;
2664         memset(pmConf, 0, sizeof(*pmConf));
2665
2666         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2667         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2668                                                                   *pmConf));
2669         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le32(virt_to_phys(
2670                                                                  pmConf));
2671
2672         netif_device_attach(netdev);
2673         pci_set_power_state(pdev, PCI_D0);
2674         pci_restore_state(pdev);
2675         err = pci_enable_device_mem(pdev);
2676         if (err != 0)
2677                 return err;
2678
2679         pci_enable_wake(pdev, PCI_D0, 0);
2680
2681         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2682                                VMXNET3_CMD_UPDATE_PMCFG);
2683         vmxnet3_alloc_intr_resources(adapter);
2684         vmxnet3_request_irqs(adapter);
2685         vmxnet3_enable_all_intrs(adapter);
2686
2687         return 0;
2688 }
2689
2690 static struct dev_pm_ops vmxnet3_pm_ops = {
2691         .suspend = vmxnet3_suspend,
2692         .resume = vmxnet3_resume,
2693 };
2694 #endif
2695
2696 static struct pci_driver vmxnet3_driver = {
2697         .name           = vmxnet3_driver_name,
2698         .id_table       = vmxnet3_pciid_table,
2699         .probe          = vmxnet3_probe_device,
2700         .remove         = __devexit_p(vmxnet3_remove_device),
2701 #ifdef CONFIG_PM
2702         .driver.pm      = &vmxnet3_pm_ops,
2703 #endif
2704 };
2705
2706
2707 static int __init
2708 vmxnet3_init_module(void)
2709 {
2710         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
2711                 VMXNET3_DRIVER_VERSION_REPORT);
2712         return pci_register_driver(&vmxnet3_driver);
2713 }
2714
2715 module_init(vmxnet3_init_module);
2716
2717
2718 static void
2719 vmxnet3_exit_module(void)
2720 {
2721         pci_unregister_driver(&vmxnet3_driver);
2722 }
2723
2724 module_exit(vmxnet3_exit_module);
2725
2726 MODULE_AUTHOR("VMware, Inc.");
2727 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
2728 MODULE_LICENSE("GPL v2");
2729 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);