drivers/net/vmxnet3/vmxnet3_drv.c

   1 /*
   2  * Linux driver for VMware's vmxnet3 ethernet NIC.
   3  *
   4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; version 2 of the License and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  * NON INFRINGEMENT. See the GNU General Public License for more
  14  * details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * The full GNU General Public License is included in this distribution in
  21  * the file called "COPYING".
  22  *
  23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
  24  *
  25  */
  26
  27 #include <net/ip6_checksum.h>
  28
  29 #include "vmxnet3_int.h"
  30
  31 char vmxnet3_driver_name[] = "vmxnet3";
  32 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
  33
  34 /*
  35  * PCI Device ID Table
  36  * Last entry must be all 0s
  37  */
  38 static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = {
  39         {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
  40         {0}
  41 };
  42
  43 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
  44
  45 static atomic_t devices_found;
  46
  47
  48 /*
  49  *    Enable/Disable the given intr
  50  */
  51 static void
  52 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  53 {
  54         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
  55 }
  56
  57
  58 static void
  59 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  60 {
  61         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
  62 }
  63
  64
  65 /*
  66  *    Enable/Disable all intrs used by the device
  67  */
  68 static void
  69 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
  70 {
  71         int i;
  72
  73         for (i = 0; i < adapter->intr.num_intrs; i++)
  74                 vmxnet3_enable_intr(adapter, i);
  75         adapter->shared->devRead.intrConf.intrCtrl &=
  76                                         cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
  77 }
  78
  79
  80 static void
  81 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
  82 {
  83         int i;
  84
  85         adapter->shared->devRead.intrConf.intrCtrl |=
  86                                         cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
  87         for (i = 0; i < adapter->intr.num_intrs; i++)
  88                 vmxnet3_disable_intr(adapter, i);
  89 }
  90
  91
  92 static void
  93 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
  94 {
  95         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
  96 }
  97
  98
  99 static bool
 100 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 101 {
 102         return netif_queue_stopped(adapter->netdev);
 103 }
 104
 105
 106 static void
 107 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 108 {
 109         tq->stopped = false;
 110         netif_start_queue(adapter->netdev);
 111 }
 112
 113
 114 static void
 115 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 116 {
 117         tq->stopped = false;
 118         netif_wake_queue(adapter->netdev);
 119 }
 120
 121
 122 static void
 123 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 124 {
 125         tq->stopped = true;
 126         tq->num_stop++;
 127         netif_stop_queue(adapter->netdev);
 128 }
 129
 130
 131 /*
 132  * Check the link state. This may start or stop the tx queue.
 133  */
 134 static void
 135 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
 136 {
 137         u32 ret;
 138
 139         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
 140         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
 141         adapter->link_speed = ret >> 16;
 142         if (ret & 1) { /* Link is up. */
 143                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
 144                        adapter->netdev->name, adapter->link_speed);
 145                 if (!netif_carrier_ok(adapter->netdev))
 146                         netif_carrier_on(adapter->netdev);
 147
 148                 if (affectTxQueue)
 149                         vmxnet3_tq_start(&adapter->tx_queue, adapter);
 150         } else {
 151                 printk(KERN_INFO "%s: NIC Link is Down\n",
 152                        adapter->netdev->name);
 153                 if (netif_carrier_ok(adapter->netdev))
 154                         netif_carrier_off(adapter->netdev);
 155
 156                 if (affectTxQueue)
 157                         vmxnet3_tq_stop(&adapter->tx_queue, adapter);
 158         }
 159 }
 160
 161 static void
 162 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 163 {
 164         u32 events = le32_to_cpu(adapter->shared->ecr);
 165         if (!events)
 166                 return;
 167
 168         vmxnet3_ack_events(adapter, events);
 169
 170         /* Check if link state has changed */
 171         if (events & VMXNET3_ECR_LINK)
 172                 vmxnet3_check_link(adapter, true);
 173
 174         /* Check if there is an error on xmit/recv queues */
 175         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 176                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 177                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 178
 179                 if (adapter->tqd_start->status.stopped) {
 180                         printk(KERN_ERR "%s: tq error 0x%x\n",
 181                                adapter->netdev->name,
 182                                le32_to_cpu(adapter->tqd_start->status.error));
 183                 }
 184                 if (adapter->rqd_start->status.stopped) {
 185                         printk(KERN_ERR "%s: rq error 0x%x\n",
 186                                adapter->netdev->name,
 187                                adapter->rqd_start->status.error);
 188                 }
 189
 190                 schedule_work(&adapter->work);
 191         }
 192 }
 193
 194 #ifdef __BIG_ENDIAN_BITFIELD
 195 /*
 196  * The device expects the bitfields in shared structures to be written in
 197  * little endian. When CPU is big endian, the following routines are used to
 198  * correctly read and write into ABI.
 199  * The general technique used here is : double word bitfields are defined in
 200  * opposite order for big endian architecture. Then before reading them in
 201  * driver the complete double word is translated using le32_to_cpu. Similarly
 202  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 203  * double words into required format.
 204  * In order to avoid touching bits in shared structure more than once, temporary
 205  * descriptors are used. These are passed as srcDesc to following functions.
 206  */
 207 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 208                                 struct Vmxnet3_RxDesc *dstDesc)
 209 {
 210         u32 *src = (u32 *)srcDesc + 2;
 211         u32 *dst = (u32 *)dstDesc + 2;
 212         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 213         *dst = le32_to_cpu(*src);
 214         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 215 }
 216
 217 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 218                                struct Vmxnet3_TxDesc *dstDesc)
 219 {
 220         int i;
 221         u32 *src = (u32 *)(srcDesc + 1);
 222         u32 *dst = (u32 *)(dstDesc + 1);
 223
 224         /* Working backwards so that the gen bit is set at the end. */
 225         for (i = 2; i > 0; i--) {
 226                 src--;
 227                 dst--;
 228                 *dst = cpu_to_le32(*src);
 229         }
 230 }
 231
 232
 233 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 234                                 struct Vmxnet3_RxCompDesc *dstDesc)
 235 {
 236         int i = 0;
 237         u32 *src = (u32 *)srcDesc;
 238         u32 *dst = (u32 *)dstDesc;
 239         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 240                 *dst = le32_to_cpu(*src);
 241                 src++;
 242                 dst++;
 243         }
 244 }
 245
 246
 247 /* Used to read bitfield values from double words. */
 248 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 249 {
 250         u32 temp = le32_to_cpu(*bitfield);
 251         u32 mask = ((1 << size) - 1) << pos;
 252         temp &= mask;
 253         temp >>= pos;
 254         return temp;
 255 }
 256
 257
 258
 259 #endif  /* __BIG_ENDIAN_BITFIELD */
 260
 261 #ifdef __BIG_ENDIAN_BITFIELD
 262
 263 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 264                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 265                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 266 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 267                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 268                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 269 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 270                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 271                         VMXNET3_TCD_GEN_SIZE)
 272 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 273                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 274 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 275                         (dstrcd) = (tmp); \
 276                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 277                 } while (0)
 278 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 279                         (dstrxd) = (tmp); \
 280                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 281                 } while (0)
 282
 283 #else
 284
 285 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 286 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 287 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 288 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 289 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 290 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 291
 292 #endif /* __BIG_ENDIAN_BITFIELD  */
 293
 294
 295 static void
 296 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 297                      struct pci_dev *pdev)
 298 {
 299         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 300                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
 301                                  PCI_DMA_TODEVICE);
 302         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 303                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
 304                                PCI_DMA_TODEVICE);
 305         else
 306                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 307
 308         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 309 }
 310
 311
 312 static int
 313 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 314                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 315 {
 316         struct sk_buff *skb;
 317         int entries = 0;
 318
 319         /* no out of order completion */
 320         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 321         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 322
 323         skb = tq->buf_info[eop_idx].skb;
 324         BUG_ON(skb == NULL);
 325         tq->buf_info[eop_idx].skb = NULL;
 326
 327         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 328
 329         while (tq->tx_ring.next2comp != eop_idx) {
 330                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 331                                      pdev);
 332
 333                 /* update next2comp w/o tx_lock. Since we are marking more,
 334                  * instead of less, tx ring entries avail, the worst case is
 335                  * that the tx routine incorrectly re-queues a pkt due to
 336                  * insufficient tx ring entries.
 337                  */
 338                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 339                 entries++;
 340         }
 341
 342         dev_kfree_skb_any(skb);
 343         return entries;
 344 }
 345
 346
 347 static int
 348 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 349                         struct vmxnet3_adapter *adapter)
 350 {
 351         int completed = 0;
 352         union Vmxnet3_GenericDesc *gdesc;
 353
 354         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 355         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 356                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 357                                                &gdesc->tcd), tq, adapter->pdev,
 358                                                adapter);
 359
 360                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 361                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 362         }
 363
 364         if (completed) {
 365                 spin_lock(&tq->tx_lock);
 366                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 367                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 368                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 369                              netif_carrier_ok(adapter->netdev))) {
 370                         vmxnet3_tq_wake(tq, adapter);
 371                 }
 372                 spin_unlock(&tq->tx_lock);
 373         }
 374         return completed;
 375 }
 376
 377
 378 static void
 379 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 380                    struct vmxnet3_adapter *adapter)
 381 {
 382         int i;
 383
 384         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 385                 struct vmxnet3_tx_buf_info *tbi;
 386                 union Vmxnet3_GenericDesc *gdesc;
 387
 388                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 389                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
 390
 391                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 392                 if (tbi->skb) {
 393                         dev_kfree_skb_any(tbi->skb);
 394                         tbi->skb = NULL;
 395                 }
 396                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 397         }
 398
 399         /* sanity check, verify all buffers are indeed unmapped and freed */
 400         for (i = 0; i < tq->tx_ring.size; i++) {
 401                 BUG_ON(tq->buf_info[i].skb != NULL ||
 402                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 403         }
 404
 405         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 406         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 407
 408         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 409         tq->comp_ring.next2proc = 0;
 410 }
 411
 412
 413 void
 414 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 415                    struct vmxnet3_adapter *adapter)
 416 {
 417         if (tq->tx_ring.base) {
 418                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 419                                     sizeof(struct Vmxnet3_TxDesc),
 420                                     tq->tx_ring.base, tq->tx_ring.basePA);
 421                 tq->tx_ring.base = NULL;
 422         }
 423         if (tq->data_ring.base) {
 424                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 425                                     sizeof(struct Vmxnet3_TxDataDesc),
 426                                     tq->data_ring.base, tq->data_ring.basePA);
 427                 tq->data_ring.base = NULL;
 428         }
 429         if (tq->comp_ring.base) {
 430                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 431                                     sizeof(struct Vmxnet3_TxCompDesc),
 432                                     tq->comp_ring.base, tq->comp_ring.basePA);
 433                 tq->comp_ring.base = NULL;
 434         }
 435         kfree(tq->buf_info);
 436         tq->buf_info = NULL;
 437 }
 438
 439
 440 static void
 441 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 442                 struct vmxnet3_adapter *adapter)
 443 {
 444         int i;
 445
 446         /* reset the tx ring contents to 0 and reset the tx ring states */
 447         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 448                sizeof(struct Vmxnet3_TxDesc));
 449         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 450         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 451
 452         memset(tq->data_ring.base, 0, tq->data_ring.size *
 453                sizeof(struct Vmxnet3_TxDataDesc));
 454
 455         /* reset the tx comp ring contents to 0 and reset comp ring states */
 456         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 457                sizeof(struct Vmxnet3_TxCompDesc));
 458         tq->comp_ring.next2proc = 0;
 459         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 460
 461         /* reset the bookkeeping data */
 462         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 463         for (i = 0; i < tq->tx_ring.size; i++)
 464                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 465
 466         /* stats are not reset */
 467 }
 468
 469
 470 static int
 471 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 472                   struct vmxnet3_adapter *adapter)
 473 {
 474         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 475                tq->comp_ring.base || tq->buf_info);
 476
 477         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 478                            * sizeof(struct Vmxnet3_TxDesc),
 479                            &tq->tx_ring.basePA);
 480         if (!tq->tx_ring.base) {
 481                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 482                        adapter->netdev->name);
 483                 goto err;
 484         }
 485
 486         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 487                              tq->data_ring.size *
 488                              sizeof(struct Vmxnet3_TxDataDesc),
 489                              &tq->data_ring.basePA);
 490         if (!tq->data_ring.base) {
 491                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 492                        adapter->netdev->name);
 493                 goto err;
 494         }
 495
 496         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 497                              tq->comp_ring.size *
 498                              sizeof(struct Vmxnet3_TxCompDesc),
 499                              &tq->comp_ring.basePA);
 500         if (!tq->comp_ring.base) {
 501                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 502                        adapter->netdev->name);
 503                 goto err;
 504         }
 505
 506         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 507                                GFP_KERNEL);
 508         if (!tq->buf_info) {
 509                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
 510                        adapter->netdev->name);
 511                 goto err;
 512         }
 513
 514         return 0;
 515
 516 err:
 517         vmxnet3_tq_destroy(tq, adapter);
 518         return -ENOMEM;
 519 }
 520
 521
 522 /*
 523  *    starting from ring->next2fill, allocate rx buffers for the given ring
 524  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 525  *    are allocated or allocation fails
 526  */
 527
 528 static int
 529 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 530                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 531 {
 532         int num_allocated = 0;
 533         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 534         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 535         u32 val;
 536
 537         while (num_allocated < num_to_alloc) {
 538                 struct vmxnet3_rx_buf_info *rbi;
 539                 union Vmxnet3_GenericDesc *gd;
 540
 541                 rbi = rbi_base + ring->next2fill;
 542                 gd = ring->base + ring->next2fill;
 543
 544                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 545                         if (rbi->skb == NULL) {
 546                                 rbi->skb = dev_alloc_skb(rbi->len +
 547                                                          NET_IP_ALIGN);
 548                                 if (unlikely(rbi->skb == NULL)) {
 549                                         rq->stats.rx_buf_alloc_failure++;
 550                                         break;
 551                                 }
 552                                 rbi->skb->dev = adapter->netdev;
 553
 554                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 555                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 556                                                 rbi->skb->data, rbi->len,
 557                                                 PCI_DMA_FROMDEVICE);
 558                         } else {
 559                                 /* rx buffer skipped by the device */
 560                         }
 561                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 562                 } else {
 563                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 564                                rbi->len  != PAGE_SIZE);
 565
 566                         if (rbi->page == NULL) {
 567                                 rbi->page = alloc_page(GFP_ATOMIC);
 568                                 if (unlikely(rbi->page == NULL)) {
 569                                         rq->stats.rx_buf_alloc_failure++;
 570                                         break;
 571                                 }
 572                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 573                                                 rbi->page, 0, PAGE_SIZE,
 574                                                 PCI_DMA_FROMDEVICE);
 575                         } else {
 576                                 /* rx buffers skipped by the device */
 577                         }
 578                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 579                 }
 580
 581                 BUG_ON(rbi->dma_addr == 0);
 582                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 583                 gd->dword[2] = cpu_to_le32((ring->gen << VMXNET3_RXD_GEN_SHIFT)
 584                                            | val | rbi->len);
 585
 586                 num_allocated++;
 587                 vmxnet3_cmd_ring_adv_next2fill(ring);
 588         }
 589         rq->uncommitted[ring_idx] += num_allocated;
 590
 591         dev_dbg(&adapter->netdev->dev,
 592                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 593                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
 594                 ring->next2comp, rq->uncommitted[ring_idx]);
 595
 596         /* so that the device can distinguish a full ring and an empty ring */
 597         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 598
 599         return num_allocated;
 600 }
 601
 602
 603 static void
 604 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 605                     struct vmxnet3_rx_buf_info *rbi)
 606 {
 607         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 608                 skb_shinfo(skb)->nr_frags;
 609
 610         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 611
 612         frag->page = rbi->page;
 613         frag->page_offset = 0;
 614         frag->size = rcd->len;
 615         skb->data_len += frag->size;
 616         skb_shinfo(skb)->nr_frags++;
 617 }
 618
 619
 620 static void
 621 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 622                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 623                 struct vmxnet3_adapter *adapter)
 624 {
 625         u32 dw2, len;
 626         unsigned long buf_offset;
 627         int i;
 628         union Vmxnet3_GenericDesc *gdesc;
 629         struct vmxnet3_tx_buf_info *tbi = NULL;
 630
 631         BUG_ON(ctx->copy_size > skb_headlen(skb));
 632
 633         /* use the previous gen bit for the SOP desc */
 634         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 635
 636         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 637         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 638
 639         /* no need to map the buffer if headers are copied */
 640         if (ctx->copy_size) {
 641                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 642                                         tq->tx_ring.next2fill *
 643                                         sizeof(struct Vmxnet3_TxDataDesc));
 644                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 645                 ctx->sop_txd->dword[3] = 0;
 646
 647                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 648                 tbi->map_type = VMXNET3_MAP_NONE;
 649
 650                 dev_dbg(&adapter->netdev->dev,
 651                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 652                         tq->tx_ring.next2fill,
 653                         le64_to_cpu(ctx->sop_txd->txd.addr),
 654                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 655                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 656
 657                 /* use the right gen for non-SOP desc */
 658                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 659         }
 660
 661         /* linear part can use multiple tx desc if it's big */
 662         len = skb_headlen(skb) - ctx->copy_size;
 663         buf_offset = ctx->copy_size;
 664         while (len) {
 665                 u32 buf_size;
 666
 667                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 668                         buf_size = len;
 669                         dw2 |= len;
 670                 } else {
 671                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 672                         /* spec says that for TxDesc.len, 0 == 2^14 */
 673                 }
 674
 675                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 676                 tbi->map_type = VMXNET3_MAP_SINGLE;
 677                 tbi->dma_addr = pci_map_single(adapter->pdev,
 678                                 skb->data + buf_offset, buf_size,
 679                                 PCI_DMA_TODEVICE);
 680
 681                 tbi->len = buf_size;
 682
 683                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 684                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 685
 686                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 687                 gdesc->dword[2] = cpu_to_le32(dw2);
 688                 gdesc->dword[3] = 0;
 689
 690                 dev_dbg(&adapter->netdev->dev,
 691                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 692                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 693                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 694                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 695                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 696
 697                 len -= buf_size;
 698                 buf_offset += buf_size;
 699         }
 700
 701         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 702                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 703
 704                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 705                 tbi->map_type = VMXNET3_MAP_PAGE;
 706                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
 707                                              frag->page_offset, frag->size,
 708                                              PCI_DMA_TODEVICE);
 709
 710                 tbi->len = frag->size;
 711
 712                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 713                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 714
 715                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 716                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
 717                 gdesc->dword[3] = 0;
 718
 719                 dev_dbg(&adapter->netdev->dev,
 720                         "txd[%u]: 0x%llu %u %u\n",
 721                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 722                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 723                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 724                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 725         }
 726
 727         ctx->eop_txd = gdesc;
 728
 729         /* set the last buf_info for the pkt */
 730         tbi->skb = skb;
 731         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 732 }
 733
 734
 735 /*
 736  *    parse and copy relevant protocol headers:
 737  *      For a tso pkt, relevant headers are L2/3/4 including options
 738  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 739  *      if it's a TCP/UDP pkt
 740  *
 741  * Returns:
 742  *    -1:  error happens during parsing
 743  *     0:  protocol headers parsed, but too big to be copied
 744  *     1:  protocol headers parsed and copied
 745  *
 746  * Other effects:
 747  *    1. related *ctx fields are updated.
 748  *    2. ctx->copy_size is # of bytes copied
 749  *    3. the portion copied is guaranteed to be in the linear part
 750  *
 751  */
 752 static int
 753 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 754                            struct vmxnet3_tx_ctx *ctx,
 755                            struct vmxnet3_adapter *adapter)
 756 {
 757         struct Vmxnet3_TxDataDesc *tdd;
 758
 759         if (ctx->mss) {
 760                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 761                 ctx->l4_hdr_size = ((struct tcphdr *)
 762                                    skb_transport_header(skb))->doff * 4;
 763                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 764         } else {
 765                 unsigned int pull_size;
 766
 767                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 768                         ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 769
 770                         if (ctx->ipv4) {
 771                                 struct iphdr *iph = (struct iphdr *)
 772                                                     skb_network_header(skb);
 773                                 if (iph->protocol == IPPROTO_TCP) {
 774                                         pull_size = ctx->eth_ip_hdr_size +
 775                                                     sizeof(struct tcphdr);
 776
 777                                         if (unlikely(!pskb_may_pull(skb,
 778                                                                 pull_size))) {
 779                                                 goto err;
 780                                         }
 781                                         ctx->l4_hdr_size = ((struct tcphdr *)
 782                                            skb_transport_header(skb))->doff * 4;
 783                                 } else if (iph->protocol == IPPROTO_UDP) {
 784                                         ctx->l4_hdr_size =
 785                                                         sizeof(struct udphdr);
 786                                 } else {
 787                                         ctx->l4_hdr_size = 0;
 788                                 }
 789                         } else {
 790                                 /* for simplicity, don't copy L4 headers */
 791                                 ctx->l4_hdr_size = 0;
 792                         }
 793                         ctx->copy_size = ctx->eth_ip_hdr_size +
 794                                          ctx->l4_hdr_size;
 795                 } else {
 796                         ctx->eth_ip_hdr_size = 0;
 797                         ctx->l4_hdr_size = 0;
 798                         /* copy as much as allowed */
 799                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 800                                              , skb_headlen(skb));
 801                 }
 802
 803                 /* make sure headers are accessible directly */
 804                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 805                         goto err;
 806         }
 807
 808         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 809                 tq->stats.oversized_hdr++;
 810                 ctx->copy_size = 0;
 811                 return 0;
 812         }
 813
 814         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 815
 816         memcpy(tdd->data, skb->data, ctx->copy_size);
 817         dev_dbg(&adapter->netdev->dev,
 818                 "copy %u bytes to dataRing[%u]\n",
 819                 ctx->copy_size, tq->tx_ring.next2fill);
 820         return 1;
 821
 822 err:
 823         return -1;
 824 }
 825
 826
 827 static void
 828 vmxnet3_prepare_tso(struct sk_buff *skb,
 829                     struct vmxnet3_tx_ctx *ctx)
 830 {
 831         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
 832         if (ctx->ipv4) {
 833                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
 834                 iph->check = 0;
 835                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 836                                                  IPPROTO_TCP, 0);
 837         } else {
 838                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
 839                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 840                                                IPPROTO_TCP, 0);
 841         }
 842 }
 843
 844
 845 /*
 846  * Transmits a pkt thru a given tq
 847  * Returns:
 848  *    NETDEV_TX_OK:      descriptors are setup successfully
 849  *    NETDEV_TX_OK:      error occured, the pkt is dropped
 850  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 851  *
 852  * Side-effects:
 853  *    1. tx ring may be changed
 854  *    2. tq stats may be updated accordingly
 855  *    3. shared->txNumDeferred may be updated
 856  */
 857
 858 static int
 859 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 860                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 861 {
 862         int ret;
 863         u32 count;
 864         unsigned long flags;
 865         struct vmxnet3_tx_ctx ctx;
 866         union Vmxnet3_GenericDesc *gdesc;
 867 #ifdef __BIG_ENDIAN_BITFIELD
 868         /* Use temporary descriptor to avoid touching bits multiple times */
 869         union Vmxnet3_GenericDesc tempTxDesc;
 870 #endif
 871
 872         /* conservatively estimate # of descriptors to use */
 873         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
 874                 skb_shinfo(skb)->nr_frags + 1;
 875
 876         ctx.ipv4 = (skb->protocol == __constant_ntohs(ETH_P_IP));
 877
 878         ctx.mss = skb_shinfo(skb)->gso_size;
 879         if (ctx.mss) {
 880                 if (skb_header_cloned(skb)) {
 881                         if (unlikely(pskb_expand_head(skb, 0, 0,
 882                                                       GFP_ATOMIC) != 0)) {
 883                                 tq->stats.drop_tso++;
 884                                 goto drop_pkt;
 885                         }
 886                         tq->stats.copy_skb_header++;
 887                 }
 888                 vmxnet3_prepare_tso(skb, &ctx);
 889         } else {
 890                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 891
 892                         /* non-tso pkts must not use more than
 893                          * VMXNET3_MAX_TXD_PER_PKT entries
 894                          */
 895                         if (skb_linearize(skb) != 0) {
 896                                 tq->stats.drop_too_many_frags++;
 897                                 goto drop_pkt;
 898                         }
 899                         tq->stats.linearized++;
 900
 901                         /* recalculate the # of descriptors to use */
 902                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 903                 }
 904         }
 905
 906         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 907         if (ret >= 0) {
 908                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 909                 /* hdrs parsed, check against other limits */
 910                 if (ctx.mss) {
 911                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 912                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 913                                 goto hdr_too_big;
 914                         }
 915                 } else {
 916                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 917                                 if (unlikely(ctx.eth_ip_hdr_size +
 918                                              skb->csum_offset >
 919                                              VMXNET3_MAX_CSUM_OFFSET)) {
 920                                         goto hdr_too_big;
 921                                 }
 922                         }
 923                 }
 924         } else {
 925                 tq->stats.drop_hdr_inspect_err++;
 926                 goto drop_pkt;
 927         }
 928
 929         spin_lock_irqsave(&tq->tx_lock, flags);
 930
 931         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 932                 tq->stats.tx_ring_full++;
 933                 dev_dbg(&adapter->netdev->dev,
 934                         "tx queue stopped on %s, next2comp %u"
 935                         " next2fill %u\n", adapter->netdev->name,
 936                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 937
 938                 vmxnet3_tq_stop(tq, adapter);
 939                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 940                 return NETDEV_TX_BUSY;
 941         }
 942
 943         /* fill tx descs related to addr & len */
 944         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
 945
 946         /* setup the EOP desc */
 947         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
 948
 949         /* setup the SOP desc */
 950 #ifdef __BIG_ENDIAN_BITFIELD
 951         gdesc = &tempTxDesc;
 952         gdesc->dword[2] = ctx.sop_txd->dword[2];
 953         gdesc->dword[3] = ctx.sop_txd->dword[3];
 954 #else
 955         gdesc = ctx.sop_txd;
 956 #endif
 957         if (ctx.mss) {
 958                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
 959                 gdesc->txd.om = VMXNET3_OM_TSO;
 960                 gdesc->txd.msscof = ctx.mss;
 961                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
 962                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
 963         } else {
 964                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 965                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
 966                         gdesc->txd.om = VMXNET3_OM_CSUM;
 967                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
 968                                             skb->csum_offset;
 969                 } else {
 970                         gdesc->txd.om = 0;
 971                         gdesc->txd.msscof = 0;
 972                 }
 973                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
 974         }
 975
 976         if (vlan_tx_tag_present(skb)) {
 977                 gdesc->txd.ti = 1;
 978                 gdesc->txd.tci = vlan_tx_tag_get(skb);
 979         }
 980
 981         /* finally flips the GEN bit of the SOP desc. */
 982         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
 983                                                   VMXNET3_TXD_GEN);
 984 #ifdef __BIG_ENDIAN_BITFIELD
 985         /* Finished updating in bitfields of Tx Desc, so write them in original
 986          * place.
 987          */
 988         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
 989                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
 990         gdesc = ctx.sop_txd;
 991 #endif
 992         dev_dbg(&adapter->netdev->dev,
 993                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
 994                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
 995                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
 996                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
 997
 998         spin_unlock_irqrestore(&tq->tx_lock, flags);
 999
1000         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1001                                         le32_to_cpu(tq->shared->txThreshold)) {
1002                 tq->shared->txNumDeferred = 0;
1003                 VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_TXPROD,
1004                                        tq->tx_ring.next2fill);
1005         }
1006
1007         return NETDEV_TX_OK;
1008
1009 hdr_too_big:
1010         tq->stats.drop_oversized_hdr++;
1011 drop_pkt:
1012         tq->stats.drop_total++;
1013         dev_kfree_skb(skb);
1014         return NETDEV_TX_OK;
1015 }
1016
1017
1018 static netdev_tx_t
1019 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1020 {
1021         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1022
1023         return vmxnet3_tq_xmit(skb, &adapter->tx_queue, adapter, netdev);
1024 }
1025
1026
1027 static void
1028 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1029                 struct sk_buff *skb,
1030                 union Vmxnet3_GenericDesc *gdesc)
1031 {
1032         if (!gdesc->rcd.cnc && adapter->rxcsum) {
1033                 /* typical case: TCP/UDP over IP and both csums are correct */
1034                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1035                                                         VMXNET3_RCD_CSUM_OK) {
1036                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1037                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1038                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1039                         BUG_ON(gdesc->rcd.frg);
1040                 } else {
1041                         if (gdesc->rcd.csum) {
1042                                 skb->csum = htons(gdesc->rcd.csum);
1043                                 skb->ip_summed = CHECKSUM_PARTIAL;
1044                         } else {
1045                                 skb_checksum_none_assert(skb);
1046                         }
1047                 }
1048         } else {
1049                 skb_checksum_none_assert(skb);
1050         }
1051 }
1052
1053
1054 static void
1055 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1056                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1057 {
1058         rq->stats.drop_err++;
1059         if (!rcd->fcs)
1060                 rq->stats.drop_fcs++;
1061
1062         rq->stats.drop_total++;
1063
1064         /*
1065          * We do not unmap and chain the rx buffer to the skb.
1066          * We basically pretend this buffer is not used and will be recycled
1067          * by vmxnet3_rq_alloc_rx_buf()
1068          */
1069
1070         /*
1071          * ctx->skb may be NULL if this is the first and the only one
1072          * desc for the pkt
1073          */
1074         if (ctx->skb)
1075                 dev_kfree_skb_irq(ctx->skb);
1076
1077         ctx->skb = NULL;
1078 }
1079
1080
1081 static int
1082 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1083                        struct vmxnet3_adapter *adapter, int quota)
1084 {
1085         static u32 rxprod_reg[2] = {VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2};
1086         u32 num_rxd = 0;
1087         struct Vmxnet3_RxCompDesc *rcd;
1088         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1089 #ifdef __BIG_ENDIAN_BITFIELD
1090         struct Vmxnet3_RxDesc rxCmdDesc;
1091         struct Vmxnet3_RxCompDesc rxComp;
1092 #endif
1093         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1094                           &rxComp);
1095         while (rcd->gen == rq->comp_ring.gen) {
1096                 struct vmxnet3_rx_buf_info *rbi;
1097                 struct sk_buff *skb;
1098                 int num_to_alloc;
1099                 struct Vmxnet3_RxDesc *rxd;
1100                 u32 idx, ring_idx;
1101
1102                 if (num_rxd >= quota) {
1103                         /* we may stop even before we see the EOP desc of
1104                          * the current pkt
1105                          */
1106                         break;
1107                 }
1108                 num_rxd++;
1109
1110                 idx = rcd->rxdIdx;
1111                 ring_idx = rcd->rqID == rq->qid ? 0 : 1;
1112                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1113                                   &rxCmdDesc);
1114                 rbi = rq->buf_info[ring_idx] + idx;
1115
1116                 BUG_ON(rxd->addr != rbi->dma_addr ||
1117                        rxd->len != rbi->len);
1118
1119                 if (unlikely(rcd->eop && rcd->err)) {
1120                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1121                         goto rcd_done;
1122                 }
1123
1124                 if (rcd->sop) { /* first buf of the pkt */
1125                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1126                                rcd->rqID != rq->qid);
1127
1128                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1129                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1130
1131                         if (unlikely(rcd->len == 0)) {
1132                                 /* Pretend the rx buffer is skipped. */
1133                                 BUG_ON(!(rcd->sop && rcd->eop));
1134                                 dev_dbg(&adapter->netdev->dev,
1135                                         "rxRing[%u][%u] 0 length\n",
1136                                         ring_idx, idx);
1137                                 goto rcd_done;
1138                         }
1139
1140                         ctx->skb = rbi->skb;
1141                         rbi->skb = NULL;
1142
1143                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1144                                          PCI_DMA_FROMDEVICE);
1145
1146                         skb_put(ctx->skb, rcd->len);
1147                 } else {
1148                         BUG_ON(ctx->skb == NULL);
1149                         /* non SOP buffer must be type 1 in most cases */
1150                         if (rbi->buf_type == VMXNET3_RX_BUF_PAGE) {
1151                                 BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1152
1153                                 if (rcd->len) {
1154                                         pci_unmap_page(adapter->pdev,
1155                                                        rbi->dma_addr, rbi->len,
1156                                                        PCI_DMA_FROMDEVICE);
1157
1158                                         vmxnet3_append_frag(ctx->skb, rcd, rbi);
1159                                         rbi->page = NULL;
1160                                 }
1161                         } else {
1162                                 /*
1163                                  * The only time a non-SOP buffer is type 0 is
1164                                  * when it's EOP and error flag is raised, which
1165                                  * has already been handled.
1166                                  */
1167                                 BUG_ON(true);
1168                         }
1169                 }
1170
1171                 skb = ctx->skb;
1172                 if (rcd->eop) {
1173                         skb->len += skb->data_len;
1174                         skb->truesize += skb->data_len;
1175
1176                         vmxnet3_rx_csum(adapter, skb,
1177                                         (union Vmxnet3_GenericDesc *)rcd);
1178                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1179
1180                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1181                                 vlan_hwaccel_receive_skb(skb,
1182                                                 adapter->vlan_grp, rcd->tci);
1183                         } else {
1184                                 netif_receive_skb(skb);
1185                         }
1186
1187                         ctx->skb = NULL;
1188                 }
1189
1190 rcd_done:
1191                 /* device may skip some rx descs */
1192                 rq->rx_ring[ring_idx].next2comp = idx;
1193                 VMXNET3_INC_RING_IDX_ONLY(rq->rx_ring[ring_idx].next2comp,
1194                                           rq->rx_ring[ring_idx].size);
1195
1196                 /* refill rx buffers frequently to avoid starving the h/w */
1197                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(rq->rx_ring +
1198                                                            ring_idx);
1199                 if (unlikely(num_to_alloc > VMXNET3_RX_ALLOC_THRESHOLD(rq,
1200                                                         ring_idx, adapter))) {
1201                         vmxnet3_rq_alloc_rx_buf(rq, ring_idx, num_to_alloc,
1202                                                 adapter);
1203
1204                         /* if needed, update the register */
1205                         if (unlikely(rq->shared->updateRxProd)) {
1206                                 VMXNET3_WRITE_BAR0_REG(adapter,
1207                                         rxprod_reg[ring_idx] + rq->qid * 8,
1208                                         rq->rx_ring[ring_idx].next2fill);
1209                                 rq->uncommitted[ring_idx] = 0;
1210                         }
1211                 }
1212
1213                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1214                 vmxnet3_getRxComp(rcd,
1215                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1216         }
1217
1218         return num_rxd;
1219 }
1220
1221
1222 static void
1223 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1224                    struct vmxnet3_adapter *adapter)
1225 {
1226         u32 i, ring_idx;
1227         struct Vmxnet3_RxDesc *rxd;
1228
1229         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1230                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1231 #ifdef __BIG_ENDIAN_BITFIELD
1232                         struct Vmxnet3_RxDesc rxDesc;
1233 #endif
1234                         vmxnet3_getRxDesc(rxd,
1235                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1236
1237                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1238                                         rq->buf_info[ring_idx][i].skb) {
1239                                 pci_unmap_single(adapter->pdev, rxd->addr,
1240                                                  rxd->len, PCI_DMA_FROMDEVICE);
1241                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1242                                 rq->buf_info[ring_idx][i].skb = NULL;
1243                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1244                                         rq->buf_info[ring_idx][i].page) {
1245                                 pci_unmap_page(adapter->pdev, rxd->addr,
1246                                                rxd->len, PCI_DMA_FROMDEVICE);
1247                                 put_page(rq->buf_info[ring_idx][i].page);
1248                                 rq->buf_info[ring_idx][i].page = NULL;
1249                         }
1250                 }
1251
1252                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1253                 rq->rx_ring[ring_idx].next2fill =
1254                                         rq->rx_ring[ring_idx].next2comp = 0;
1255                 rq->uncommitted[ring_idx] = 0;
1256         }
1257
1258         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1259         rq->comp_ring.next2proc = 0;
1260 }
1261
1262
1263 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1264                         struct vmxnet3_adapter *adapter)
1265 {
1266         int i;
1267         int j;
1268
1269         /* all rx buffers must have already been freed */
1270         for (i = 0; i < 2; i++) {
1271                 if (rq->buf_info[i]) {
1272                         for (j = 0; j < rq->rx_ring[i].size; j++)
1273                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1274                 }
1275         }
1276
1277
1278         kfree(rq->buf_info[0]);
1279
1280         for (i = 0; i < 2; i++) {
1281                 if (rq->rx_ring[i].base) {
1282                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1283                                             * sizeof(struct Vmxnet3_RxDesc),
1284                                             rq->rx_ring[i].base,
1285                                             rq->rx_ring[i].basePA);
1286                         rq->rx_ring[i].base = NULL;
1287                 }
1288                 rq->buf_info[i] = NULL;
1289         }
1290
1291         if (rq->comp_ring.base) {
1292                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1293                                     sizeof(struct Vmxnet3_RxCompDesc),
1294                                     rq->comp_ring.base, rq->comp_ring.basePA);
1295                 rq->comp_ring.base = NULL;
1296         }
1297 }
1298
1299
1300 static int
1301 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1302                 struct vmxnet3_adapter  *adapter)
1303 {
1304         int i;
1305
1306         /* initialize buf_info */
1307         for (i = 0; i < rq->rx_ring[0].size; i++) {
1308
1309                 /* 1st buf for a pkt is skbuff */
1310                 if (i % adapter->rx_buf_per_pkt == 0) {
1311                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1312                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1313                 } else { /* subsequent bufs for a pkt is frag */
1314                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1315                         rq->buf_info[0][i].len = PAGE_SIZE;
1316                 }
1317         }
1318         for (i = 0; i < rq->rx_ring[1].size; i++) {
1319                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1320                 rq->buf_info[1][i].len = PAGE_SIZE;
1321         }
1322
1323         /* reset internal state and allocate buffers for both rings */
1324         for (i = 0; i < 2; i++) {
1325                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1326                 rq->uncommitted[i] = 0;
1327
1328                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1329                        sizeof(struct Vmxnet3_RxDesc));
1330                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1331         }
1332         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1333                                     adapter) == 0) {
1334                 /* at least has 1 rx buffer for the 1st ring */
1335                 return -ENOMEM;
1336         }
1337         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1338
1339         /* reset the comp ring */
1340         rq->comp_ring.next2proc = 0;
1341         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1342                sizeof(struct Vmxnet3_RxCompDesc));
1343         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1344
1345         /* reset rxctx */
1346         rq->rx_ctx.skb = NULL;
1347
1348         /* stats are not reset */
1349         return 0;
1350 }
1351
1352
1353 static int
1354 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1355 {
1356         int i;
1357         size_t sz;
1358         struct vmxnet3_rx_buf_info *bi;
1359
1360         for (i = 0; i < 2; i++) {
1361
1362                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1363                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1364                                                         &rq->rx_ring[i].basePA);
1365                 if (!rq->rx_ring[i].base) {
1366                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1367                                adapter->netdev->name, i);
1368                         goto err;
1369                 }
1370         }
1371
1372         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1373         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1374                                                   &rq->comp_ring.basePA);
1375         if (!rq->comp_ring.base) {
1376                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1377                        adapter->netdev->name);
1378                 goto err;
1379         }
1380
1381         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1382                                                    rq->rx_ring[1].size);
1383         bi = kzalloc(sz, GFP_KERNEL);
1384         if (!bi) {
1385                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1386                        adapter->netdev->name);
1387                 goto err;
1388         }
1389         rq->buf_info[0] = bi;
1390         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1391
1392         return 0;
1393
1394 err:
1395         vmxnet3_rq_destroy(rq, adapter);
1396         return -ENOMEM;
1397 }
1398
1399
1400 static int
1401 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1402 {
1403         if (unlikely(adapter->shared->ecr))
1404                 vmxnet3_process_events(adapter);
1405
1406         vmxnet3_tq_tx_complete(&adapter->tx_queue, adapter);
1407         return vmxnet3_rq_rx_complete(&adapter->rx_queue, adapter, budget);
1408 }
1409
1410
1411 static int
1412 vmxnet3_poll(struct napi_struct *napi, int budget)
1413 {
1414         struct vmxnet3_adapter *adapter = container_of(napi,
1415                                           struct vmxnet3_adapter, napi);
1416         int rxd_done;
1417
1418         rxd_done = vmxnet3_do_poll(adapter, budget);
1419
1420         if (rxd_done < budget) {
1421                 napi_complete(napi);
1422                 vmxnet3_enable_intr(adapter, 0);
1423         }
1424         return rxd_done;
1425 }
1426
1427
1428 /* Interrupt handler for vmxnet3  */
1429 static irqreturn_t
1430 vmxnet3_intr(int irq, void *dev_id)
1431 {
1432         struct net_device *dev = dev_id;
1433         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1434
1435         if (unlikely(adapter->intr.type == VMXNET3_IT_INTX)) {
1436                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1437                 if (unlikely(icr == 0))
1438                         /* not ours */
1439                         return IRQ_NONE;
1440         }
1441
1442
1443         /* disable intr if needed */
1444         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1445                 vmxnet3_disable_intr(adapter, 0);
1446
1447         napi_schedule(&adapter->napi);
1448
1449         return IRQ_HANDLED;
1450 }
1451
1452 #ifdef CONFIG_NET_POLL_CONTROLLER
1453
1454
1455 /* netpoll callback. */
1456 static void
1457 vmxnet3_netpoll(struct net_device *netdev)
1458 {
1459         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1460         int irq;
1461
1462 #ifdef CONFIG_PCI_MSI
1463         if (adapter->intr.type == VMXNET3_IT_MSIX)
1464                 irq = adapter->intr.msix_entries[0].vector;
1465         else
1466 #endif
1467                 irq = adapter->pdev->irq;
1468
1469         disable_irq(irq);
1470         vmxnet3_intr(irq, netdev);
1471         enable_irq(irq);
1472 }
1473 #endif
1474
1475 static int
1476 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1477 {
1478         int err;
1479
1480 #ifdef CONFIG_PCI_MSI
1481         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1482                 /* we only use 1 MSI-X vector */
1483                 err = request_irq(adapter->intr.msix_entries[0].vector,
1484                                   vmxnet3_intr, 0, adapter->netdev->name,
1485                                   adapter->netdev);
1486         } else if (adapter->intr.type == VMXNET3_IT_MSI) {
1487                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1488                                   adapter->netdev->name, adapter->netdev);
1489         } else
1490 #endif
1491         {
1492                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1493                                   IRQF_SHARED, adapter->netdev->name,
1494                                   adapter->netdev);
1495         }
1496
1497         if (err)
1498                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1499                        ":%d\n", adapter->netdev->name, adapter->intr.type, err);
1500
1501
1502         if (!err) {
1503                 int i;
1504                 /* init our intr settings */
1505                 for (i = 0; i < adapter->intr.num_intrs; i++)
1506                         adapter->intr.mod_levels[i] = UPT1_IML_ADAPTIVE;
1507
1508                 /* next setup intr index for all intr sources */
1509                 adapter->tx_queue.comp_ring.intr_idx = 0;
1510                 adapter->rx_queue.comp_ring.intr_idx = 0;
1511                 adapter->intr.event_intr_idx = 0;
1512
1513                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1514                        "allocated\n", adapter->netdev->name, adapter->intr.type,
1515                        adapter->intr.mask_mode, adapter->intr.num_intrs);
1516         }
1517
1518         return err;
1519 }
1520
1521
1522 static void
1523 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1524 {
1525         BUG_ON(adapter->intr.type == VMXNET3_IT_AUTO ||
1526                adapter->intr.num_intrs <= 0);
1527
1528         switch (adapter->intr.type) {
1529 #ifdef CONFIG_PCI_MSI
1530         case VMXNET3_IT_MSIX:
1531         {
1532                 int i;
1533
1534                 for (i = 0; i < adapter->intr.num_intrs; i++)
1535                         free_irq(adapter->intr.msix_entries[i].vector,
1536                                  adapter->netdev);
1537                 break;
1538         }
1539 #endif
1540         case VMXNET3_IT_MSI:
1541                 free_irq(adapter->pdev->irq, adapter->netdev);
1542                 break;
1543         case VMXNET3_IT_INTX:
1544                 free_irq(adapter->pdev->irq, adapter->netdev);
1545                 break;
1546         default:
1547                 BUG_ON(true);
1548         }
1549 }
1550
1551 static void
1552 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1553 {
1554         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1555         struct Vmxnet3_DriverShared *shared = adapter->shared;
1556         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1557
1558         if (grp) {
1559                 /* add vlan rx stripping. */
1560                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1561                         int i;
1562                         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1563                         adapter->vlan_grp = grp;
1564
1565                         /* update FEATURES to device */
1566                         devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1567                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1568                                                VMXNET3_CMD_UPDATE_FEATURE);
1569                         /*
1570                          *  Clear entire vfTable; then enable untagged pkts.
1571                          *  Note: setting one entry in vfTable to non-zero turns
1572                          *  on VLAN rx filtering.
1573                          */
1574                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1575                                 vfTable[i] = 0;
1576
1577                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1578                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1579                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1580                 } else {
1581                         printk(KERN_ERR "%s: vlan_rx_register when device has "
1582                                "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1583                 }
1584         } else {
1585                 /* remove vlan rx stripping. */
1586                 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1587                 adapter->vlan_grp = NULL;
1588
1589                 if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1590                         int i;
1591
1592                         for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1593                                 /* clear entire vfTable; this also disables
1594                                  * VLAN rx filtering
1595                                  */
1596                                 vfTable[i] = 0;
1597                         }
1598                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1599                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1600
1601                         /* update FEATURES to device */
1602                         devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1603                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1604                                                VMXNET3_CMD_UPDATE_FEATURE);
1605                 }
1606         }
1607 }
1608
1609
1610 static void
1611 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1612 {
1613         if (adapter->vlan_grp) {
1614                 u16 vid;
1615                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1616                 bool activeVlan = false;
1617
1618                 for (vid = 0; vid < VLAN_N_VID; vid++) {
1619                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1620                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1621                                 activeVlan = true;
1622                         }
1623                 }
1624                 if (activeVlan) {
1625                         /* continue to allow untagged pkts */
1626                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1627                 }
1628         }
1629 }
1630
1631
1632 static void
1633 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1634 {
1635         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1636         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1637
1638         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1639         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1640                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1641 }
1642
1643
1644 static void
1645 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1646 {
1647         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1648         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1649
1650         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1651         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1652                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1653 }
1654
1655
1656 static u8 *
1657 vmxnet3_copy_mc(struct net_device *netdev)
1658 {
1659         u8 *buf = NULL;
1660         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1661
1662         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1663         if (sz <= 0xffff) {
1664                 /* We may be called with BH disabled */
1665                 buf = kmalloc(sz, GFP_ATOMIC);
1666                 if (buf) {
1667                         struct netdev_hw_addr *ha;
1668                         int i = 0;
1669
1670                         netdev_for_each_mc_addr(ha, netdev)
1671                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1672                                        ETH_ALEN);
1673                 }
1674         }
1675         return buf;
1676 }
1677
1678
1679 static void
1680 vmxnet3_set_mc(struct net_device *netdev)
1681 {
1682         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1683         struct Vmxnet3_RxFilterConf *rxConf =
1684                                         &adapter->shared->devRead.rxFilterConf;
1685         u8 *new_table = NULL;
1686         u32 new_mode = VMXNET3_RXM_UCAST;
1687
1688         if (netdev->flags & IFF_PROMISC)
1689                 new_mode |= VMXNET3_RXM_PROMISC;
1690
1691         if (netdev->flags & IFF_BROADCAST)
1692                 new_mode |= VMXNET3_RXM_BCAST;
1693
1694         if (netdev->flags & IFF_ALLMULTI)
1695                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1696         else
1697                 if (!netdev_mc_empty(netdev)) {
1698                         new_table = vmxnet3_copy_mc(netdev);
1699                         if (new_table) {
1700                                 new_mode |= VMXNET3_RXM_MCAST;
1701                                 rxConf->mfTableLen = cpu_to_le16(
1702                                         netdev_mc_count(netdev) * ETH_ALEN);
1703                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
1704                                                     new_table));
1705                         } else {
1706                                 printk(KERN_INFO "%s: failed to copy mcast list"
1707                                        ", setting ALL_MULTI\n", netdev->name);
1708                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
1709                         }
1710                 }
1711
1712
1713         if (!(new_mode & VMXNET3_RXM_MCAST)) {
1714                 rxConf->mfTableLen = 0;
1715                 rxConf->mfTablePA = 0;
1716         }
1717
1718         if (new_mode != rxConf->rxMode) {
1719                 rxConf->rxMode = cpu_to_le32(new_mode);
1720                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1721                                        VMXNET3_CMD_UPDATE_RX_MODE);
1722         }
1723
1724         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1725                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
1726
1727         kfree(new_table);
1728 }
1729
1730
1731 /*
1732  *   Set up driver_shared based on settings in adapter.
1733  */
1734
1735 static void
1736 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
1737 {
1738         struct Vmxnet3_DriverShared *shared = adapter->shared;
1739         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1740         struct Vmxnet3_TxQueueConf *tqc;
1741         struct Vmxnet3_RxQueueConf *rqc;
1742         int i;
1743
1744         memset(shared, 0, sizeof(*shared));
1745
1746         /* driver settings */
1747         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
1748         devRead->misc.driverInfo.version = cpu_to_le32(
1749                                                 VMXNET3_DRIVER_VERSION_NUM);
1750         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
1751                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
1752         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
1753         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
1754                                 *((u32 *)&devRead->misc.driverInfo.gos));
1755         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
1756         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
1757
1758         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
1759         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
1760
1761         /* set up feature flags */
1762         if (adapter->rxcsum)
1763                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
1764
1765         if (adapter->lro) {
1766                 devRead->misc.uptFeatures |= UPT1_F_LRO;
1767                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
1768         }
1769         if ((adapter->netdev->features & NETIF_F_HW_VLAN_RX) &&
1770             adapter->vlan_grp) {
1771                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1772         }
1773
1774         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
1775         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
1776         devRead->misc.queueDescLen = cpu_to_le32(
1777                                      sizeof(struct Vmxnet3_TxQueueDesc) +
1778                                      sizeof(struct Vmxnet3_RxQueueDesc));
1779
1780         /* tx queue settings */
1781         BUG_ON(adapter->tx_queue.tx_ring.base == NULL);
1782
1783         devRead->misc.numTxQueues = 1;
1784         tqc = &adapter->tqd_start->conf;
1785         tqc->txRingBasePA   = cpu_to_le64(adapter->tx_queue.tx_ring.basePA);
1786         tqc->dataRingBasePA = cpu_to_le64(adapter->tx_queue.data_ring.basePA);
1787         tqc->compRingBasePA = cpu_to_le64(adapter->tx_queue.comp_ring.basePA);
1788         tqc->ddPA           = cpu_to_le64(virt_to_phys(
1789                                                 adapter->tx_queue.buf_info));
1790         tqc->txRingSize     = cpu_to_le32(adapter->tx_queue.tx_ring.size);
1791         tqc->dataRingSize   = cpu_to_le32(adapter->tx_queue.data_ring.size);
1792         tqc->compRingSize   = cpu_to_le32(adapter->tx_queue.comp_ring.size);
1793         tqc->ddLen          = cpu_to_le32(sizeof(struct vmxnet3_tx_buf_info) *
1794                               tqc->txRingSize);
1795         tqc->intrIdx        = adapter->tx_queue.comp_ring.intr_idx;
1796
1797         /* rx queue settings */
1798         devRead->misc.numRxQueues = 1;
1799         rqc = &adapter->rqd_start->conf;
1800         rqc->rxRingBasePA[0] = cpu_to_le64(adapter->rx_queue.rx_ring[0].basePA);
1801         rqc->rxRingBasePA[1] = cpu_to_le64(adapter->rx_queue.rx_ring[1].basePA);
1802         rqc->compRingBasePA  = cpu_to_le64(adapter->rx_queue.comp_ring.basePA);
1803         rqc->ddPA            = cpu_to_le64(virt_to_phys(
1804                                                 adapter->rx_queue.buf_info));
1805         rqc->rxRingSize[0]   = cpu_to_le32(adapter->rx_queue.rx_ring[0].size);
1806         rqc->rxRingSize[1]   = cpu_to_le32(adapter->rx_queue.rx_ring[1].size);
1807         rqc->compRingSize    = cpu_to_le32(adapter->rx_queue.comp_ring.size);
1808         rqc->ddLen           = cpu_to_le32(sizeof(struct vmxnet3_rx_buf_info) *
1809                                (rqc->rxRingSize[0] + rqc->rxRingSize[1]));
1810         rqc->intrIdx         = adapter->rx_queue.comp_ring.intr_idx;
1811
1812         /* intr settings */
1813         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
1814                                      VMXNET3_IMM_AUTO;
1815         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
1816         for (i = 0; i < adapter->intr.num_intrs; i++)
1817                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
1818
1819         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
1820         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
1821
1822         /* rx filter settings */
1823         devRead->rxFilterConf.rxMode = 0;
1824         vmxnet3_restore_vlan(adapter);
1825         /* the rest are already zeroed */
1826 }
1827
1828
1829 int
1830 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
1831 {
1832         int err;
1833         u32 ret;
1834
1835         dev_dbg(&adapter->netdev->dev,
1836                 "%s: skb_buf_size %d, rx_buf_per_pkt %d, ring sizes"
1837                 " %u %u %u\n", adapter->netdev->name, adapter->skb_buf_size,
1838                 adapter->rx_buf_per_pkt, adapter->tx_queue.tx_ring.size,
1839                 adapter->rx_queue.rx_ring[0].size,
1840                 adapter->rx_queue.rx_ring[1].size);
1841
1842         vmxnet3_tq_init(&adapter->tx_queue, adapter);
1843         err = vmxnet3_rq_init(&adapter->rx_queue, adapter);
1844         if (err) {
1845                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
1846                        adapter->netdev->name, err);
1847                 goto rq_err;
1848         }
1849
1850         err = vmxnet3_request_irqs(adapter);
1851         if (err) {
1852                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
1853                        adapter->netdev->name, err);
1854                 goto irq_err;
1855         }
1856
1857         vmxnet3_setup_driver_shared(adapter);
1858
1859         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
1860                                adapter->shared_pa));
1861         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
1862                                adapter->shared_pa));
1863         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1864                                VMXNET3_CMD_ACTIVATE_DEV);
1865         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
1866
1867         if (ret != 0) {
1868                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
1869                        adapter->netdev->name, ret);
1870                 err = -EINVAL;
1871                 goto activate_err;
1872         }
1873         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD,
1874                                adapter->rx_queue.rx_ring[0].next2fill);
1875         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_RXPROD2,
1876                                adapter->rx_queue.rx_ring[1].next2fill);
1877
1878         /* Apply the rx filter settins last. */
1879         vmxnet3_set_mc(adapter->netdev);
1880
1881         /*
1882          * Check link state when first activating device. It will start the
1883          * tx queue if the link is up.
1884          */
1885         vmxnet3_check_link(adapter, true);
1886
1887         napi_enable(&adapter->napi);
1888         vmxnet3_enable_all_intrs(adapter);
1889         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
1890         return 0;
1891
1892 activate_err:
1893         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
1894         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
1895         vmxnet3_free_irqs(adapter);
1896 irq_err:
1897 rq_err:
1898         /* free up buffers we allocated */
1899         vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1900         return err;
1901 }
1902
1903
1904 void
1905 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
1906 {
1907         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
1908 }
1909
1910
1911 int
1912 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
1913 {
1914         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
1915                 return 0;
1916
1917
1918         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1919                                VMXNET3_CMD_QUIESCE_DEV);
1920         vmxnet3_disable_all_intrs(adapter);
1921
1922         napi_disable(&adapter->napi);
1923         netif_tx_disable(adapter->netdev);
1924         adapter->link_speed = 0;
1925         netif_carrier_off(adapter->netdev);
1926
1927         vmxnet3_tq_cleanup(&adapter->tx_queue, adapter);
1928         vmxnet3_rq_cleanup(&adapter->rx_queue, adapter);
1929         vmxnet3_free_irqs(adapter);
1930         return 0;
1931 }
1932
1933
1934 static void
1935 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
1936 {
1937         u32 tmp;
1938
1939         tmp = *(u32 *)mac;
1940         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
1941
1942         tmp = (mac[5] << 8) | mac[4];
1943         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
1944 }
1945
1946
1947 static int
1948 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
1949 {
1950         struct sockaddr *addr = p;
1951         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1952
1953         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
1954         vmxnet3_write_mac_addr(adapter, addr->sa_data);
1955
1956         return 0;
1957 }
1958
1959
1960 /* ==================== initialization and cleanup routines ============ */
1961
1962 static int
1963 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
1964 {
1965         int err;
1966         unsigned long mmio_start, mmio_len;
1967         struct pci_dev *pdev = adapter->pdev;
1968
1969         err = pci_enable_device(pdev);
1970         if (err) {
1971                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
1972                        pci_name(pdev), err);
1973                 return err;
1974         }
1975
1976         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
1977                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
1978                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
1979                                "for adapter %s\n", pci_name(pdev));
1980                         err = -EIO;
1981                         goto err_set_mask;
1982                 }
1983                 *dma64 = true;
1984         } else {
1985                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
1986                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
1987                                "%s\n",  pci_name(pdev));
1988                         err = -EIO;
1989                         goto err_set_mask;
1990                 }
1991                 *dma64 = false;
1992         }
1993
1994         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
1995                                            vmxnet3_driver_name);
1996         if (err) {
1997                 printk(KERN_ERR "Failed to request region for adapter %s: "
1998                        "error %d\n", pci_name(pdev), err);
1999                 goto err_set_mask;
2000         }
2001
2002         pci_set_master(pdev);
2003
2004         mmio_start = pci_resource_start(pdev, 0);
2005         mmio_len = pci_resource_len(pdev, 0);
2006         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2007         if (!adapter->hw_addr0) {
2008                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2009                        pci_name(pdev));
2010                 err = -EIO;
2011                 goto err_ioremap;
2012         }
2013
2014         mmio_start = pci_resource_start(pdev, 1);
2015         mmio_len = pci_resource_len(pdev, 1);
2016         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2017         if (!adapter->hw_addr1) {
2018                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2019                        pci_name(pdev));
2020                 err = -EIO;
2021                 goto err_bar1;
2022         }
2023         return 0;
2024
2025 err_bar1:
2026         iounmap(adapter->hw_addr0);
2027 err_ioremap:
2028         pci_release_selected_regions(pdev, (1 << 2) - 1);
2029 err_set_mask:
2030         pci_disable_device(pdev);
2031         return err;
2032 }
2033
2034
2035 static void
2036 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2037 {
2038         BUG_ON(!adapter->pdev);
2039
2040         iounmap(adapter->hw_addr0);
2041         iounmap(adapter->hw_addr1);
2042         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2043         pci_disable_device(adapter->pdev);
2044 }
2045
2046
2047 static void
2048 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2049 {
2050         size_t sz;
2051
2052         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2053                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2054                 adapter->skb_buf_size = adapter->netdev->mtu +
2055                                         VMXNET3_MAX_ETH_HDR_SIZE;
2056                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2057                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2058
2059                 adapter->rx_buf_per_pkt = 1;
2060         } else {
2061                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2062                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2063                                             VMXNET3_MAX_ETH_HDR_SIZE;
2064                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2065         }
2066
2067         /*
2068          * for simplicity, force the ring0 size to be a multiple of
2069          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2070          */
2071         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2072         adapter->rx_queue.rx_ring[0].size = (adapter->rx_queue.rx_ring[0].size +
2073                                              sz - 1) / sz * sz;
2074         adapter->rx_queue.rx_ring[0].size = min_t(u32,
2075                                             adapter->rx_queue.rx_ring[0].size,
2076                                             VMXNET3_RX_RING_MAX_SIZE / sz * sz);
2077 }
2078
2079
2080 int
2081 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2082                       u32 rx_ring_size, u32 rx_ring2_size)
2083 {
2084         int err;
2085
2086         adapter->tx_queue.tx_ring.size   = tx_ring_size;
2087         adapter->tx_queue.data_ring.size = tx_ring_size;
2088         adapter->tx_queue.comp_ring.size = tx_ring_size;
2089         adapter->tx_queue.shared = &adapter->tqd_start->ctrl;
2090         adapter->tx_queue.stopped = true;
2091         err = vmxnet3_tq_create(&adapter->tx_queue, adapter);
2092         if (err)
2093                 return err;
2094
2095         adapter->rx_queue.rx_ring[0].size = rx_ring_size;
2096         adapter->rx_queue.rx_ring[1].size = rx_ring2_size;
2097         vmxnet3_adjust_rx_ring_size(adapter);
2098         adapter->rx_queue.comp_ring.size  = adapter->rx_queue.rx_ring[0].size +
2099                                             adapter->rx_queue.rx_ring[1].size;
2100         adapter->rx_queue.qid  = 0;
2101         adapter->rx_queue.qid2 = 1;
2102         adapter->rx_queue.shared = &adapter->rqd_start->ctrl;
2103         err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2104         if (err)
2105                 vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2106
2107         return err;
2108 }
2109
2110 static int
2111 vmxnet3_open(struct net_device *netdev)
2112 {
2113         struct vmxnet3_adapter *adapter;
2114         int err;
2115
2116         adapter = netdev_priv(netdev);
2117
2118         spin_lock_init(&adapter->tx_queue.tx_lock);
2119
2120         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2121                                     VMXNET3_DEF_RX_RING_SIZE,
2122                                     VMXNET3_DEF_RX_RING_SIZE);
2123         if (err)
2124                 goto queue_err;
2125
2126         err = vmxnet3_activate_dev(adapter);
2127         if (err)
2128                 goto activate_err;
2129
2130         return 0;
2131
2132 activate_err:
2133         vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2134         vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2135 queue_err:
2136         return err;
2137 }
2138
2139
2140 static int
2141 vmxnet3_close(struct net_device *netdev)
2142 {
2143         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2144
2145         /*
2146          * Reset_work may be in the middle of resetting the device, wait for its
2147          * completion.
2148          */
2149         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2150                 msleep(1);
2151
2152         vmxnet3_quiesce_dev(adapter);
2153
2154         vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2155         vmxnet3_tq_destroy(&adapter->tx_queue, adapter);
2156
2157         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2158
2159
2160         return 0;
2161 }
2162
2163
2164 void
2165 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2166 {
2167         /*
2168          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2169          * vmxnet3_close() will deadlock.
2170          */
2171         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2172
2173         /* we need to enable NAPI, otherwise dev_close will deadlock */
2174         napi_enable(&adapter->napi);
2175         dev_close(adapter->netdev);
2176 }
2177
2178
2179 static int
2180 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2181 {
2182         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2183         int err = 0;
2184
2185         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2186                 return -EINVAL;
2187
2188         if (new_mtu > 1500 && !adapter->jumbo_frame)
2189                 return -EINVAL;
2190
2191         netdev->mtu = new_mtu;
2192
2193         /*
2194          * Reset_work may be in the middle of resetting the device, wait for its
2195          * completion.
2196          */
2197         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2198                 msleep(1);
2199
2200         if (netif_running(netdev)) {
2201                 vmxnet3_quiesce_dev(adapter);
2202                 vmxnet3_reset_dev(adapter);
2203
2204                 /* we need to re-create the rx queue based on the new mtu */
2205                 vmxnet3_rq_destroy(&adapter->rx_queue, adapter);
2206                 vmxnet3_adjust_rx_ring_size(adapter);
2207                 adapter->rx_queue.comp_ring.size  =
2208                                         adapter->rx_queue.rx_ring[0].size +
2209                                         adapter->rx_queue.rx_ring[1].size;
2210                 err = vmxnet3_rq_create(&adapter->rx_queue, adapter);
2211                 if (err) {
2212                         printk(KERN_ERR "%s: failed to re-create rx queue,"
2213                                 " error %d. Closing it.\n", netdev->name, err);
2214                         goto out;
2215                 }
2216
2217                 err = vmxnet3_activate_dev(adapter);
2218                 if (err) {
2219                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2220                                 "Closing it\n", netdev->name, err);
2221                         goto out;
2222                 }
2223         }
2224
2225 out:
2226         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2227         if (err)
2228                 vmxnet3_force_close(adapter);
2229
2230         return err;
2231 }
2232
2233
2234 static void
2235 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2236 {
2237         struct net_device *netdev = adapter->netdev;
2238
2239         netdev->features = NETIF_F_SG |
2240                 NETIF_F_HW_CSUM |
2241                 NETIF_F_HW_VLAN_TX |
2242                 NETIF_F_HW_VLAN_RX |
2243                 NETIF_F_HW_VLAN_FILTER |
2244                 NETIF_F_TSO |
2245                 NETIF_F_TSO6 |
2246                 NETIF_F_LRO;
2247
2248         printk(KERN_INFO "features: sg csum vlan jf tso tsoIPv6 lro");
2249
2250         adapter->rxcsum = true;
2251         adapter->jumbo_frame = true;
2252         adapter->lro = true;
2253
2254         if (dma64) {
2255                 netdev->features |= NETIF_F_HIGHDMA;
2256                 printk(" highDMA");
2257         }
2258
2259         netdev->vlan_features = netdev->features;
2260         printk("\n");
2261 }
2262
2263
2264 static void
2265 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2266 {
2267         u32 tmp;
2268
2269         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2270         *(u32 *)mac = tmp;
2271
2272         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2273         mac[4] = tmp & 0xff;
2274         mac[5] = (tmp >> 8) & 0xff;
2275 }
2276
2277
2278 static void
2279 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2280 {
2281         u32 cfg;
2282
2283         /* intr settings */
2284         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2285                                VMXNET3_CMD_GET_CONF_INTR);
2286         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2287         adapter->intr.type = cfg & 0x3;
2288         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2289
2290         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2291                 adapter->intr.type = VMXNET3_IT_MSIX;
2292         }
2293
2294 #ifdef CONFIG_PCI_MSI
2295         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2296                 int err;
2297
2298                 adapter->intr.msix_entries[0].entry = 0;
2299                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2300                                       VMXNET3_LINUX_MAX_MSIX_VECT);
2301                 if (!err) {
2302                         adapter->intr.num_intrs = 1;
2303                         adapter->intr.type = VMXNET3_IT_MSIX;
2304                         return;
2305                 }
2306                 adapter->intr.type = VMXNET3_IT_MSI;
2307         }
2308
2309         if (adapter->intr.type == VMXNET3_IT_MSI) {
2310                 int err;
2311                 err = pci_enable_msi(adapter->pdev);
2312                 if (!err) {
2313                         adapter->intr.num_intrs = 1;
2314                         return;
2315                 }
2316         }
2317 #endif /* CONFIG_PCI_MSI */
2318
2319         adapter->intr.type = VMXNET3_IT_INTX;
2320
2321         /* INT-X related setting */
2322         adapter->intr.num_intrs = 1;
2323 }
2324
2325
2326 static void
2327 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2328 {
2329         if (adapter->intr.type == VMXNET3_IT_MSIX)
2330                 pci_disable_msix(adapter->pdev);
2331         else if (adapter->intr.type == VMXNET3_IT_MSI)
2332                 pci_disable_msi(adapter->pdev);
2333         else
2334                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2335 }
2336
2337
2338 static void
2339 vmxnet3_tx_timeout(struct net_device *netdev)
2340 {
2341         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2342         adapter->tx_timeout_count++;
2343
2344         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2345         schedule_work(&adapter->work);
2346 }
2347
2348
2349 static void
2350 vmxnet3_reset_work(struct work_struct *data)
2351 {
2352         struct vmxnet3_adapter *adapter;
2353
2354         adapter = container_of(data, struct vmxnet3_adapter, work);
2355
2356         /* if another thread is resetting the device, no need to proceed */
2357         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2358                 return;
2359
2360         /* if the device is closed, we must leave it alone */
2361         rtnl_lock();
2362         if (netif_running(adapter->netdev)) {
2363                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2364                 vmxnet3_quiesce_dev(adapter);
2365                 vmxnet3_reset_dev(adapter);
2366                 vmxnet3_activate_dev(adapter);
2367         } else {
2368                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2369         }
2370         rtnl_unlock();
2371
2372         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2373 }
2374
2375
2376 static int __devinit
2377 vmxnet3_probe_device(struct pci_dev *pdev,
2378                      const struct pci_device_id *id)
2379 {
2380         static const struct net_device_ops vmxnet3_netdev_ops = {
2381                 .ndo_open = vmxnet3_open,
2382                 .ndo_stop = vmxnet3_close,
2383                 .ndo_start_xmit = vmxnet3_xmit_frame,
2384                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2385                 .ndo_change_mtu = vmxnet3_change_mtu,
2386                 .ndo_get_stats = vmxnet3_get_stats,
2387                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2388                 .ndo_set_multicast_list = vmxnet3_set_mc,
2389                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2390                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2391                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2392 #ifdef CONFIG_NET_POLL_CONTROLLER
2393                 .ndo_poll_controller = vmxnet3_netpoll,
2394 #endif
2395         };
2396         int err;
2397         bool dma64 = false; /* stupid gcc */
2398         u32 ver;
2399         struct net_device *netdev;
2400         struct vmxnet3_adapter *adapter;
2401         u8 mac[ETH_ALEN];
2402
2403         netdev = alloc_etherdev(sizeof(struct vmxnet3_adapter));
2404         if (!netdev) {
2405                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2406                         "%s\n", pci_name(pdev));
2407                 return -ENOMEM;
2408         }
2409
2410         pci_set_drvdata(pdev, netdev);
2411         adapter = netdev_priv(netdev);
2412         adapter->netdev = netdev;
2413         adapter->pdev = pdev;
2414
2415         adapter->shared = pci_alloc_consistent(adapter->pdev,
2416                           sizeof(struct Vmxnet3_DriverShared),
2417                           &adapter->shared_pa);
2418         if (!adapter->shared) {
2419                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2420                         pci_name(pdev));
2421                 err = -ENOMEM;
2422                 goto err_alloc_shared;
2423         }
2424
2425         adapter->tqd_start = pci_alloc_consistent(adapter->pdev,
2426                              sizeof(struct Vmxnet3_TxQueueDesc) +
2427                              sizeof(struct Vmxnet3_RxQueueDesc),
2428                              &adapter->queue_desc_pa);
2429
2430         if (!adapter->tqd_start) {
2431                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2432                         pci_name(pdev));
2433                 err = -ENOMEM;
2434                 goto err_alloc_queue_desc;
2435         }
2436         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start
2437                                                             + 1);
2438
2439         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2440         if (adapter->pm_conf == NULL) {
2441                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2442                         pci_name(pdev));
2443                 err = -ENOMEM;
2444                 goto err_alloc_pm;
2445         }
2446
2447         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2448         if (err < 0)
2449                 goto err_alloc_pci;
2450
2451         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2452         if (ver & 1) {
2453                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2454         } else {
2455                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2456                        " %s\n", ver, pci_name(pdev));
2457                 err = -EBUSY;
2458                 goto err_ver;
2459         }
2460
2461         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2462         if (ver & 1) {
2463                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2464         } else {
2465                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2466                        "adapter %s\n", ver, pci_name(pdev));
2467                 err = -EBUSY;
2468                 goto err_ver;
2469         }
2470
2471         vmxnet3_declare_features(adapter, dma64);
2472
2473         adapter->dev_number = atomic_read(&devices_found);
2474         vmxnet3_alloc_intr_resources(adapter);
2475
2476         vmxnet3_read_mac_addr(adapter, mac);
2477         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
2478
2479         netdev->netdev_ops = &vmxnet3_netdev_ops;
2480         netdev->watchdog_timeo = 5 * HZ;
2481         vmxnet3_set_ethtool_ops(netdev);
2482
2483         INIT_WORK(&adapter->work, vmxnet3_reset_work);
2484
2485         netif_napi_add(netdev, &adapter->napi, vmxnet3_poll, 64);
2486         SET_NETDEV_DEV(netdev, &pdev->dev);
2487         err = register_netdev(netdev);
2488
2489         if (err) {
2490                 printk(KERN_ERR "Failed to register adapter %s\n",
2491                         pci_name(pdev));
2492                 goto err_register;
2493         }
2494
2495         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2496         vmxnet3_check_link(adapter, false);
2497         atomic_inc(&devices_found);
2498         return 0;
2499
2500 err_register:
2501         vmxnet3_free_intr_resources(adapter);
2502 err_ver:
2503         vmxnet3_free_pci_resources(adapter);
2504 err_alloc_pci:
2505         kfree(adapter->pm_conf);
2506 err_alloc_pm:
2507         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2508                             sizeof(struct Vmxnet3_RxQueueDesc),
2509                             adapter->tqd_start, adapter->queue_desc_pa);
2510 err_alloc_queue_desc:
2511         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2512                             adapter->shared, adapter->shared_pa);
2513 err_alloc_shared:
2514         pci_set_drvdata(pdev, NULL);
2515         free_netdev(netdev);
2516         return err;
2517 }
2518
2519
2520 static void __devexit
2521 vmxnet3_remove_device(struct pci_dev *pdev)
2522 {
2523         struct net_device *netdev = pci_get_drvdata(pdev);
2524         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2525
2526         flush_scheduled_work();
2527
2528         unregister_netdev(netdev);
2529
2530         vmxnet3_free_intr_resources(adapter);
2531         vmxnet3_free_pci_resources(adapter);
2532         kfree(adapter->pm_conf);
2533         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_TxQueueDesc) +
2534                             sizeof(struct Vmxnet3_RxQueueDesc),
2535                             adapter->tqd_start, adapter->queue_desc_pa);
2536         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
2537                             adapter->shared, adapter->shared_pa);
2538         free_netdev(netdev);
2539 }
2540
2541
2542 #ifdef CONFIG_PM
2543
2544 static int
2545 vmxnet3_suspend(struct device *device)
2546 {
2547         struct pci_dev *pdev = to_pci_dev(device);
2548         struct net_device *netdev = pci_get_drvdata(pdev);
2549         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2550         struct Vmxnet3_PMConf *pmConf;
2551         struct ethhdr *ehdr;
2552         struct arphdr *ahdr;
2553         u8 *arpreq;
2554         struct in_device *in_dev;
2555         struct in_ifaddr *ifa;
2556         int i = 0;
2557
2558         if (!netif_running(netdev))
2559                 return 0;
2560
2561         vmxnet3_disable_all_intrs(adapter);
2562         vmxnet3_free_irqs(adapter);
2563         vmxnet3_free_intr_resources(adapter);
2564
2565         netif_device_detach(netdev);
2566         netif_stop_queue(netdev);
2567
2568         /* Create wake-up filters. */
2569         pmConf = adapter->pm_conf;
2570         memset(pmConf, 0, sizeof(*pmConf));
2571
2572         if (adapter->wol & WAKE_UCAST) {
2573                 pmConf->filters[i].patternSize = ETH_ALEN;
2574                 pmConf->filters[i].maskSize = 1;
2575                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
2576                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
2577
2578                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
2579                 i++;
2580         }
2581
2582         if (adapter->wol & WAKE_ARP) {
2583                 in_dev = in_dev_get(netdev);
2584                 if (!in_dev)
2585                         goto skip_arp;
2586
2587                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
2588                 if (!ifa)
2589                         goto skip_arp;
2590
2591                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
2592                         sizeof(struct arphdr) +         /* ARP header */
2593                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
2594                         2 * sizeof(u32);        /*2 IPv4 addresses */
2595                 pmConf->filters[i].maskSize =
2596                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
2597
2598                 /* ETH_P_ARP in Ethernet header. */
2599                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
2600                 ehdr->h_proto = htons(ETH_P_ARP);
2601
2602                 /* ARPOP_REQUEST in ARP header. */
2603                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
2604                 ahdr->ar_op = htons(ARPOP_REQUEST);
2605                 arpreq = (u8 *)(ahdr + 1);
2606
2607                 /* The Unicast IPv4 address in 'tip' field. */
2608                 arpreq += 2 * ETH_ALEN + sizeof(u32);
2609                 *(u32 *)arpreq = ifa->ifa_address;
2610
2611                 /* The mask for the relevant bits. */
2612                 pmConf->filters[i].mask[0] = 0x00;
2613                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
2614                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
2615                 pmConf->filters[i].mask[3] = 0x00;
2616                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
2617                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
2618                 in_dev_put(in_dev);
2619
2620                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
2621                 i++;
2622         }
2623
2624 skip_arp:
2625         if (adapter->wol & WAKE_MAGIC)
2626                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
2627
2628         pmConf->numFilters = i;
2629
2630         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2631         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2632                                                                   *pmConf));
2633         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
2634                                                                  pmConf));
2635
2636         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2637                                VMXNET3_CMD_UPDATE_PMCFG);
2638
2639         pci_save_state(pdev);
2640         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
2641                         adapter->wol);
2642         pci_disable_device(pdev);
2643         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
2644
2645         return 0;
2646 }
2647
2648
2649 static int
2650 vmxnet3_resume(struct device *device)
2651 {
2652         int err;
2653         struct pci_dev *pdev = to_pci_dev(device);
2654         struct net_device *netdev = pci_get_drvdata(pdev);
2655         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2656         struct Vmxnet3_PMConf *pmConf;
2657
2658         if (!netif_running(netdev))
2659                 return 0;
2660
2661         /* Destroy wake-up filters. */
2662         pmConf = adapter->pm_conf;
2663         memset(pmConf, 0, sizeof(*pmConf));
2664
2665         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
2666         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
2667                                                                   *pmConf));
2668         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
2669                                                                  pmConf));
2670
2671         netif_device_attach(netdev);
2672         pci_set_power_state(pdev, PCI_D0);
2673         pci_restore_state(pdev);
2674         err = pci_enable_device_mem(pdev);
2675         if (err != 0)
2676                 return err;
2677
2678         pci_enable_wake(pdev, PCI_D0, 0);
2679
2680         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2681                                VMXNET3_CMD_UPDATE_PMCFG);
2682         vmxnet3_alloc_intr_resources(adapter);
2683         vmxnet3_request_irqs(adapter);
2684         vmxnet3_enable_all_intrs(adapter);
2685
2686         return 0;
2687 }
2688
2689 static const struct dev_pm_ops vmxnet3_pm_ops = {
2690         .suspend = vmxnet3_suspend,
2691         .resume = vmxnet3_resume,
2692 };
2693 #endif
2694
2695 static struct pci_driver vmxnet3_driver = {
2696         .name           = vmxnet3_driver_name,
2697         .id_table       = vmxnet3_pciid_table,
2698         .probe          = vmxnet3_probe_device,
2699         .remove         = __devexit_p(vmxnet3_remove_device),
2700 #ifdef CONFIG_PM
2701         .driver.pm      = &vmxnet3_pm_ops,
2702 #endif
2703 };
2704
2705
2706 static int __init
2707 vmxnet3_init_module(void)
2708 {
2709         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
2710                 VMXNET3_DRIVER_VERSION_REPORT);
2711         return pci_register_driver(&vmxnet3_driver);
2712 }
2713
2714 module_init(vmxnet3_init_module);
2715
2716
2717 static void
2718 vmxnet3_exit_module(void)
2719 {
2720         pci_unregister_driver(&vmxnet3_driver);
2721 }
2722
2723 module_exit(vmxnet3_exit_module);
2724
2725 MODULE_AUTHOR("VMware, Inc.");
2726 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
2727 MODULE_LICENSE("GPL v2");
2728 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);