drivers/net/vmxnet3/vmxnet3_drv.c

   1 /*
   2  * Linux driver for VMware's vmxnet3 ethernet NIC.
   3  *
   4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; version 2 of the License and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  * NON INFRINGEMENT. See the GNU General Public License for more
  14  * details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * The full GNU General Public License is included in this distribution in
  21  * the file called "COPYING".
  22  *
  23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
  24  *
  25  */
  26
  27 #include <net/ip6_checksum.h>
  28
  29 #include "vmxnet3_int.h"
  30
  31 char vmxnet3_driver_name[] = "vmxnet3";
  32 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
  33
  34 /*
  35  * PCI Device ID Table
  36  * Last entry must be all 0s
  37  */
  38 static DEFINE_PCI_DEVICE_TABLE(vmxnet3_pciid_table) = {
  39         {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
  40         {0}
  41 };
  42
  43 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
  44
  45 static atomic_t devices_found;
  46
  47 #define VMXNET3_MAX_DEVICES 10
  48 static int enable_mq = 1;
  49 static int irq_share_mode;
  50
  51 static void
  52 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
  53
  54 /*
  55  *    Enable/Disable the given intr
  56  */
  57 static void
  58 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  59 {
  60         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
  61 }
  62
  63
  64 static void
  65 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  66 {
  67         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
  68 }
  69
  70
  71 /*
  72  *    Enable/Disable all intrs used by the device
  73  */
  74 static void
  75 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
  76 {
  77         int i;
  78
  79         for (i = 0; i < adapter->intr.num_intrs; i++)
  80                 vmxnet3_enable_intr(adapter, i);
  81         adapter->shared->devRead.intrConf.intrCtrl &=
  82                                         cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
  83 }
  84
  85
  86 static void
  87 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
  88 {
  89         int i;
  90
  91         adapter->shared->devRead.intrConf.intrCtrl |=
  92                                         cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
  93         for (i = 0; i < adapter->intr.num_intrs; i++)
  94                 vmxnet3_disable_intr(adapter, i);
  95 }
  96
  97
  98 static void
  99 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
 100 {
 101         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
 102 }
 103
 104
 105 static bool
 106 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 107 {
 108         return tq->stopped;
 109 }
 110
 111
 112 static void
 113 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 114 {
 115         tq->stopped = false;
 116         netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
 117 }
 118
 119
 120 static void
 121 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 122 {
 123         tq->stopped = false;
 124         netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
 125 }
 126
 127
 128 static void
 129 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 130 {
 131         tq->stopped = true;
 132         tq->num_stop++;
 133         netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
 134 }
 135
 136
 137 /*
 138  * Check the link state. This may start or stop the tx queue.
 139  */
 140 static void
 141 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
 142 {
 143         u32 ret;
 144         int i;
 145         unsigned long flags;
 146
 147         spin_lock_irqsave(&adapter->cmd_lock, flags);
 148         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
 149         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
 150         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
 151
 152         adapter->link_speed = ret >> 16;
 153         if (ret & 1) { /* Link is up. */
 154                 printk(KERN_INFO "%s: NIC Link is Up %d Mbps\n",
 155                        adapter->netdev->name, adapter->link_speed);
 156                 if (!netif_carrier_ok(adapter->netdev))
 157                         netif_carrier_on(adapter->netdev);
 158
 159                 if (affectTxQueue) {
 160                         for (i = 0; i < adapter->num_tx_queues; i++)
 161                                 vmxnet3_tq_start(&adapter->tx_queue[i],
 162                                                  adapter);
 163                 }
 164         } else {
 165                 printk(KERN_INFO "%s: NIC Link is Down\n",
 166                        adapter->netdev->name);
 167                 if (netif_carrier_ok(adapter->netdev))
 168                         netif_carrier_off(adapter->netdev);
 169
 170                 if (affectTxQueue) {
 171                         for (i = 0; i < adapter->num_tx_queues; i++)
 172                                 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
 173                 }
 174         }
 175 }
 176
 177 static void
 178 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 179 {
 180         int i;
 181         unsigned long flags;
 182         u32 events = le32_to_cpu(adapter->shared->ecr);
 183         if (!events)
 184                 return;
 185
 186         vmxnet3_ack_events(adapter, events);
 187
 188         /* Check if link state has changed */
 189         if (events & VMXNET3_ECR_LINK)
 190                 vmxnet3_check_link(adapter, true);
 191
 192         /* Check if there is an error on xmit/recv queues */
 193         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 194                 spin_lock_irqsave(&adapter->cmd_lock, flags);
 195                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 196                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 197                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
 198
 199                 for (i = 0; i < adapter->num_tx_queues; i++)
 200                         if (adapter->tqd_start[i].status.stopped)
 201                                 dev_err(&adapter->netdev->dev,
 202                                         "%s: tq[%d] error 0x%x\n",
 203                                         adapter->netdev->name, i, le32_to_cpu(
 204                                         adapter->tqd_start[i].status.error));
 205                 for (i = 0; i < adapter->num_rx_queues; i++)
 206                         if (adapter->rqd_start[i].status.stopped)
 207                                 dev_err(&adapter->netdev->dev,
 208                                         "%s: rq[%d] error 0x%x\n",
 209                                         adapter->netdev->name, i,
 210                                         adapter->rqd_start[i].status.error);
 211
 212                 schedule_work(&adapter->work);
 213         }
 214 }
 215
 216 #ifdef __BIG_ENDIAN_BITFIELD
 217 /*
 218  * The device expects the bitfields in shared structures to be written in
 219  * little endian. When CPU is big endian, the following routines are used to
 220  * correctly read and write into ABI.
 221  * The general technique used here is : double word bitfields are defined in
 222  * opposite order for big endian architecture. Then before reading them in
 223  * driver the complete double word is translated using le32_to_cpu. Similarly
 224  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 225  * double words into required format.
 226  * In order to avoid touching bits in shared structure more than once, temporary
 227  * descriptors are used. These are passed as srcDesc to following functions.
 228  */
 229 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 230                                 struct Vmxnet3_RxDesc *dstDesc)
 231 {
 232         u32 *src = (u32 *)srcDesc + 2;
 233         u32 *dst = (u32 *)dstDesc + 2;
 234         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 235         *dst = le32_to_cpu(*src);
 236         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 237 }
 238
 239 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 240                                struct Vmxnet3_TxDesc *dstDesc)
 241 {
 242         int i;
 243         u32 *src = (u32 *)(srcDesc + 1);
 244         u32 *dst = (u32 *)(dstDesc + 1);
 245
 246         /* Working backwards so that the gen bit is set at the end. */
 247         for (i = 2; i > 0; i--) {
 248                 src--;
 249                 dst--;
 250                 *dst = cpu_to_le32(*src);
 251         }
 252 }
 253
 254
 255 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 256                                 struct Vmxnet3_RxCompDesc *dstDesc)
 257 {
 258         int i = 0;
 259         u32 *src = (u32 *)srcDesc;
 260         u32 *dst = (u32 *)dstDesc;
 261         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 262                 *dst = le32_to_cpu(*src);
 263                 src++;
 264                 dst++;
 265         }
 266 }
 267
 268
 269 /* Used to read bitfield values from double words. */
 270 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 271 {
 272         u32 temp = le32_to_cpu(*bitfield);
 273         u32 mask = ((1 << size) - 1) << pos;
 274         temp &= mask;
 275         temp >>= pos;
 276         return temp;
 277 }
 278
 279
 280
 281 #endif  /* __BIG_ENDIAN_BITFIELD */
 282
 283 #ifdef __BIG_ENDIAN_BITFIELD
 284
 285 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 286                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 287                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 288 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 289                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 290                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 291 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 292                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 293                         VMXNET3_TCD_GEN_SIZE)
 294 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 295                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 296 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 297                         (dstrcd) = (tmp); \
 298                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 299                 } while (0)
 300 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 301                         (dstrxd) = (tmp); \
 302                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 303                 } while (0)
 304
 305 #else
 306
 307 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 308 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 309 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 310 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 311 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 312 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 313
 314 #endif /* __BIG_ENDIAN_BITFIELD  */
 315
 316
 317 static void
 318 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 319                      struct pci_dev *pdev)
 320 {
 321         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 322                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
 323                                  PCI_DMA_TODEVICE);
 324         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 325                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
 326                                PCI_DMA_TODEVICE);
 327         else
 328                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 329
 330         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 331 }
 332
 333
 334 static int
 335 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 336                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 337 {
 338         struct sk_buff *skb;
 339         int entries = 0;
 340
 341         /* no out of order completion */
 342         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 343         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 344
 345         skb = tq->buf_info[eop_idx].skb;
 346         BUG_ON(skb == NULL);
 347         tq->buf_info[eop_idx].skb = NULL;
 348
 349         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 350
 351         while (tq->tx_ring.next2comp != eop_idx) {
 352                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 353                                      pdev);
 354
 355                 /* update next2comp w/o tx_lock. Since we are marking more,
 356                  * instead of less, tx ring entries avail, the worst case is
 357                  * that the tx routine incorrectly re-queues a pkt due to
 358                  * insufficient tx ring entries.
 359                  */
 360                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 361                 entries++;
 362         }
 363
 364         dev_kfree_skb_any(skb);
 365         return entries;
 366 }
 367
 368
 369 static int
 370 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 371                         struct vmxnet3_adapter *adapter)
 372 {
 373         int completed = 0;
 374         union Vmxnet3_GenericDesc *gdesc;
 375
 376         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 377         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 378                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 379                                                &gdesc->tcd), tq, adapter->pdev,
 380                                                adapter);
 381
 382                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 383                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 384         }
 385
 386         if (completed) {
 387                 spin_lock(&tq->tx_lock);
 388                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 389                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 390                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 391                              netif_carrier_ok(adapter->netdev))) {
 392                         vmxnet3_tq_wake(tq, adapter);
 393                 }
 394                 spin_unlock(&tq->tx_lock);
 395         }
 396         return completed;
 397 }
 398
 399
 400 static void
 401 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 402                    struct vmxnet3_adapter *adapter)
 403 {
 404         int i;
 405
 406         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 407                 struct vmxnet3_tx_buf_info *tbi;
 408
 409                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 410
 411                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 412                 if (tbi->skb) {
 413                         dev_kfree_skb_any(tbi->skb);
 414                         tbi->skb = NULL;
 415                 }
 416                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 417         }
 418
 419         /* sanity check, verify all buffers are indeed unmapped and freed */
 420         for (i = 0; i < tq->tx_ring.size; i++) {
 421                 BUG_ON(tq->buf_info[i].skb != NULL ||
 422                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 423         }
 424
 425         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 426         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 427
 428         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 429         tq->comp_ring.next2proc = 0;
 430 }
 431
 432
 433 static void
 434 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 435                    struct vmxnet3_adapter *adapter)
 436 {
 437         if (tq->tx_ring.base) {
 438                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 439                                     sizeof(struct Vmxnet3_TxDesc),
 440                                     tq->tx_ring.base, tq->tx_ring.basePA);
 441                 tq->tx_ring.base = NULL;
 442         }
 443         if (tq->data_ring.base) {
 444                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 445                                     sizeof(struct Vmxnet3_TxDataDesc),
 446                                     tq->data_ring.base, tq->data_ring.basePA);
 447                 tq->data_ring.base = NULL;
 448         }
 449         if (tq->comp_ring.base) {
 450                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 451                                     sizeof(struct Vmxnet3_TxCompDesc),
 452                                     tq->comp_ring.base, tq->comp_ring.basePA);
 453                 tq->comp_ring.base = NULL;
 454         }
 455         kfree(tq->buf_info);
 456         tq->buf_info = NULL;
 457 }
 458
 459
 460 /* Destroy all tx queues */
 461 void
 462 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
 463 {
 464         int i;
 465
 466         for (i = 0; i < adapter->num_tx_queues; i++)
 467                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
 468 }
 469
 470
 471 static void
 472 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 473                 struct vmxnet3_adapter *adapter)
 474 {
 475         int i;
 476
 477         /* reset the tx ring contents to 0 and reset the tx ring states */
 478         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 479                sizeof(struct Vmxnet3_TxDesc));
 480         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 481         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 482
 483         memset(tq->data_ring.base, 0, tq->data_ring.size *
 484                sizeof(struct Vmxnet3_TxDataDesc));
 485
 486         /* reset the tx comp ring contents to 0 and reset comp ring states */
 487         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 488                sizeof(struct Vmxnet3_TxCompDesc));
 489         tq->comp_ring.next2proc = 0;
 490         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 491
 492         /* reset the bookkeeping data */
 493         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 494         for (i = 0; i < tq->tx_ring.size; i++)
 495                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 496
 497         /* stats are not reset */
 498 }
 499
 500
 501 static int
 502 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 503                   struct vmxnet3_adapter *adapter)
 504 {
 505         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 506                tq->comp_ring.base || tq->buf_info);
 507
 508         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 509                            * sizeof(struct Vmxnet3_TxDesc),
 510                            &tq->tx_ring.basePA);
 511         if (!tq->tx_ring.base) {
 512                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 513                        adapter->netdev->name);
 514                 goto err;
 515         }
 516
 517         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 518                              tq->data_ring.size *
 519                              sizeof(struct Vmxnet3_TxDataDesc),
 520                              &tq->data_ring.basePA);
 521         if (!tq->data_ring.base) {
 522                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 523                        adapter->netdev->name);
 524                 goto err;
 525         }
 526
 527         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 528                              tq->comp_ring.size *
 529                              sizeof(struct Vmxnet3_TxCompDesc),
 530                              &tq->comp_ring.basePA);
 531         if (!tq->comp_ring.base) {
 532                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 533                        adapter->netdev->name);
 534                 goto err;
 535         }
 536
 537         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 538                                GFP_KERNEL);
 539         if (!tq->buf_info) {
 540                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
 541                        adapter->netdev->name);
 542                 goto err;
 543         }
 544
 545         return 0;
 546
 547 err:
 548         vmxnet3_tq_destroy(tq, adapter);
 549         return -ENOMEM;
 550 }
 551
 552 static void
 553 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
 554 {
 555         int i;
 556
 557         for (i = 0; i < adapter->num_tx_queues; i++)
 558                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
 559 }
 560
 561 /*
 562  *    starting from ring->next2fill, allocate rx buffers for the given ring
 563  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 564  *    are allocated or allocation fails
 565  */
 566
 567 static int
 568 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 569                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 570 {
 571         int num_allocated = 0;
 572         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 573         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 574         u32 val;
 575
 576         while (num_allocated <= num_to_alloc) {
 577                 struct vmxnet3_rx_buf_info *rbi;
 578                 union Vmxnet3_GenericDesc *gd;
 579
 580                 rbi = rbi_base + ring->next2fill;
 581                 gd = ring->base + ring->next2fill;
 582
 583                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 584                         if (rbi->skb == NULL) {
 585                                 rbi->skb = dev_alloc_skb(rbi->len +
 586                                                          NET_IP_ALIGN);
 587                                 if (unlikely(rbi->skb == NULL)) {
 588                                         rq->stats.rx_buf_alloc_failure++;
 589                                         break;
 590                                 }
 591                                 rbi->skb->dev = adapter->netdev;
 592
 593                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 594                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 595                                                 rbi->skb->data, rbi->len,
 596                                                 PCI_DMA_FROMDEVICE);
 597                         } else {
 598                                 /* rx buffer skipped by the device */
 599                         }
 600                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 601                 } else {
 602                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 603                                rbi->len  != PAGE_SIZE);
 604
 605                         if (rbi->page == NULL) {
 606                                 rbi->page = alloc_page(GFP_ATOMIC);
 607                                 if (unlikely(rbi->page == NULL)) {
 608                                         rq->stats.rx_buf_alloc_failure++;
 609                                         break;
 610                                 }
 611                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 612                                                 rbi->page, 0, PAGE_SIZE,
 613                                                 PCI_DMA_FROMDEVICE);
 614                         } else {
 615                                 /* rx buffers skipped by the device */
 616                         }
 617                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 618                 }
 619
 620                 BUG_ON(rbi->dma_addr == 0);
 621                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 622                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 623                                            | val | rbi->len);
 624
 625                 /* Fill the last buffer but dont mark it ready, or else the
 626                  * device will think that the queue is full */
 627                 if (num_allocated == num_to_alloc)
 628                         break;
 629
 630                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 631                 num_allocated++;
 632                 vmxnet3_cmd_ring_adv_next2fill(ring);
 633         }
 634         rq->uncommitted[ring_idx] += num_allocated;
 635
 636         dev_dbg(&adapter->netdev->dev,
 637                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 638                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
 639                 ring->next2comp, rq->uncommitted[ring_idx]);
 640
 641         /* so that the device can distinguish a full ring and an empty ring */
 642         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 643
 644         return num_allocated;
 645 }
 646
 647
 648 static void
 649 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 650                     struct vmxnet3_rx_buf_info *rbi)
 651 {
 652         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 653                 skb_shinfo(skb)->nr_frags;
 654
 655         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 656
 657         __skb_frag_set_page(frag, rbi->page);
 658         frag->page_offset = 0;
 659         skb_frag_size_set(frag, rcd->len);
 660         skb->data_len += rcd->len;
 661         skb->truesize += PAGE_SIZE;
 662         skb_shinfo(skb)->nr_frags++;
 663 }
 664
 665
 666 static void
 667 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 668                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 669                 struct vmxnet3_adapter *adapter)
 670 {
 671         u32 dw2, len;
 672         unsigned long buf_offset;
 673         int i;
 674         union Vmxnet3_GenericDesc *gdesc;
 675         struct vmxnet3_tx_buf_info *tbi = NULL;
 676
 677         BUG_ON(ctx->copy_size > skb_headlen(skb));
 678
 679         /* use the previous gen bit for the SOP desc */
 680         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 681
 682         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 683         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 684
 685         /* no need to map the buffer if headers are copied */
 686         if (ctx->copy_size) {
 687                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 688                                         tq->tx_ring.next2fill *
 689                                         sizeof(struct Vmxnet3_TxDataDesc));
 690                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 691                 ctx->sop_txd->dword[3] = 0;
 692
 693                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 694                 tbi->map_type = VMXNET3_MAP_NONE;
 695
 696                 dev_dbg(&adapter->netdev->dev,
 697                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 698                         tq->tx_ring.next2fill,
 699                         le64_to_cpu(ctx->sop_txd->txd.addr),
 700                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 701                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 702
 703                 /* use the right gen for non-SOP desc */
 704                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 705         }
 706
 707         /* linear part can use multiple tx desc if it's big */
 708         len = skb_headlen(skb) - ctx->copy_size;
 709         buf_offset = ctx->copy_size;
 710         while (len) {
 711                 u32 buf_size;
 712
 713                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 714                         buf_size = len;
 715                         dw2 |= len;
 716                 } else {
 717                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 718                         /* spec says that for TxDesc.len, 0 == 2^14 */
 719                 }
 720
 721                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 722                 tbi->map_type = VMXNET3_MAP_SINGLE;
 723                 tbi->dma_addr = pci_map_single(adapter->pdev,
 724                                 skb->data + buf_offset, buf_size,
 725                                 PCI_DMA_TODEVICE);
 726
 727                 tbi->len = buf_size;
 728
 729                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 730                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 731
 732                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 733                 gdesc->dword[2] = cpu_to_le32(dw2);
 734                 gdesc->dword[3] = 0;
 735
 736                 dev_dbg(&adapter->netdev->dev,
 737                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 738                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 739                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 740                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 741                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 742
 743                 len -= buf_size;
 744                 buf_offset += buf_size;
 745         }
 746
 747         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 748                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 749
 750                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 751                 tbi->map_type = VMXNET3_MAP_PAGE;
 752                 tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
 753                                                  0, skb_frag_size(frag),
 754                                                  DMA_TO_DEVICE);
 755
 756                 tbi->len = skb_frag_size(frag);
 757
 758                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 759                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 760
 761                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 762                 gdesc->dword[2] = cpu_to_le32(dw2 | skb_frag_size(frag));
 763                 gdesc->dword[3] = 0;
 764
 765                 dev_dbg(&adapter->netdev->dev,
 766                         "txd[%u]: 0x%llu %u %u\n",
 767                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 768                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 769                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 770                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 771         }
 772
 773         ctx->eop_txd = gdesc;
 774
 775         /* set the last buf_info for the pkt */
 776         tbi->skb = skb;
 777         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 778 }
 779
 780
 781 /* Init all tx queues */
 782 static void
 783 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 784 {
 785         int i;
 786
 787         for (i = 0; i < adapter->num_tx_queues; i++)
 788                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 789 }
 790
 791
 792 /*
 793  *    parse and copy relevant protocol headers:
 794  *      For a tso pkt, relevant headers are L2/3/4 including options
 795  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 796  *      if it's a TCP/UDP pkt
 797  *
 798  * Returns:
 799  *    -1:  error happens during parsing
 800  *     0:  protocol headers parsed, but too big to be copied
 801  *     1:  protocol headers parsed and copied
 802  *
 803  * Other effects:
 804  *    1. related *ctx fields are updated.
 805  *    2. ctx->copy_size is # of bytes copied
 806  *    3. the portion copied is guaranteed to be in the linear part
 807  *
 808  */
 809 static int
 810 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 811                            struct vmxnet3_tx_ctx *ctx,
 812                            struct vmxnet3_adapter *adapter)
 813 {
 814         struct Vmxnet3_TxDataDesc *tdd;
 815
 816         if (ctx->mss) { /* TSO */
 817                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 818                 ctx->l4_hdr_size = ((struct tcphdr *)
 819                                    skb_transport_header(skb))->doff * 4;
 820                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 821         } else {
 822                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 823                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 824
 825                         if (ctx->ipv4) {
 826                                 struct iphdr *iph = (struct iphdr *)
 827                                                     skb_network_header(skb);
 828                                 if (iph->protocol == IPPROTO_TCP)
 829                                         ctx->l4_hdr_size = ((struct tcphdr *)
 830                                            skb_transport_header(skb))->doff * 4;
 831                                 else if (iph->protocol == IPPROTO_UDP)
 832                                         /*
 833                                          * Use tcp header size so that bytes to
 834                                          * be copied are more than required by
 835                                          * the device.
 836                                          */
 837                                         ctx->l4_hdr_size =
 838                                                         sizeof(struct tcphdr);
 839                                 else
 840                                         ctx->l4_hdr_size = 0;
 841                         } else {
 842                                 /* for simplicity, don't copy L4 headers */
 843                                 ctx->l4_hdr_size = 0;
 844                         }
 845                         ctx->copy_size = ctx->eth_ip_hdr_size +
 846                                          ctx->l4_hdr_size;
 847                 } else {
 848                         ctx->eth_ip_hdr_size = 0;
 849                         ctx->l4_hdr_size = 0;
 850                         /* copy as much as allowed */
 851                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 852                                              , skb_headlen(skb));
 853                 }
 854
 855                 /* make sure headers are accessible directly */
 856                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 857                         goto err;
 858         }
 859
 860         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 861                 tq->stats.oversized_hdr++;
 862                 ctx->copy_size = 0;
 863                 return 0;
 864         }
 865
 866         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 867
 868         memcpy(tdd->data, skb->data, ctx->copy_size);
 869         dev_dbg(&adapter->netdev->dev,
 870                 "copy %u bytes to dataRing[%u]\n",
 871                 ctx->copy_size, tq->tx_ring.next2fill);
 872         return 1;
 873
 874 err:
 875         return -1;
 876 }
 877
 878
 879 static void
 880 vmxnet3_prepare_tso(struct sk_buff *skb,
 881                     struct vmxnet3_tx_ctx *ctx)
 882 {
 883         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
 884         if (ctx->ipv4) {
 885                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
 886                 iph->check = 0;
 887                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 888                                                  IPPROTO_TCP, 0);
 889         } else {
 890                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
 891                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 892                                                IPPROTO_TCP, 0);
 893         }
 894 }
 895
 896
 897 /*
 898  * Transmits a pkt thru a given tq
 899  * Returns:
 900  *    NETDEV_TX_OK:      descriptors are setup successfully
 901  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 902  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 903  *
 904  * Side-effects:
 905  *    1. tx ring may be changed
 906  *    2. tq stats may be updated accordingly
 907  *    3. shared->txNumDeferred may be updated
 908  */
 909
 910 static int
 911 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 912                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 913 {
 914         int ret;
 915         u32 count;
 916         unsigned long flags;
 917         struct vmxnet3_tx_ctx ctx;
 918         union Vmxnet3_GenericDesc *gdesc;
 919 #ifdef __BIG_ENDIAN_BITFIELD
 920         /* Use temporary descriptor to avoid touching bits multiple times */
 921         union Vmxnet3_GenericDesc tempTxDesc;
 922 #endif
 923
 924         /* conservatively estimate # of descriptors to use */
 925         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
 926                 skb_shinfo(skb)->nr_frags + 1;
 927
 928         ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
 929
 930         ctx.mss = skb_shinfo(skb)->gso_size;
 931         if (ctx.mss) {
 932                 if (skb_header_cloned(skb)) {
 933                         if (unlikely(pskb_expand_head(skb, 0, 0,
 934                                                       GFP_ATOMIC) != 0)) {
 935                                 tq->stats.drop_tso++;
 936                                 goto drop_pkt;
 937                         }
 938                         tq->stats.copy_skb_header++;
 939                 }
 940                 vmxnet3_prepare_tso(skb, &ctx);
 941         } else {
 942                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 943
 944                         /* non-tso pkts must not use more than
 945                          * VMXNET3_MAX_TXD_PER_PKT entries
 946                          */
 947                         if (skb_linearize(skb) != 0) {
 948                                 tq->stats.drop_too_many_frags++;
 949                                 goto drop_pkt;
 950                         }
 951                         tq->stats.linearized++;
 952
 953                         /* recalculate the # of descriptors to use */
 954                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 955                 }
 956         }
 957
 958         spin_lock_irqsave(&tq->tx_lock, flags);
 959
 960         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 961                 tq->stats.tx_ring_full++;
 962                 dev_dbg(&adapter->netdev->dev,
 963                         "tx queue stopped on %s, next2comp %u"
 964                         " next2fill %u\n", adapter->netdev->name,
 965                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 966
 967                 vmxnet3_tq_stop(tq, adapter);
 968                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 969                 return NETDEV_TX_BUSY;
 970         }
 971
 972
 973         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 974         if (ret >= 0) {
 975                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 976                 /* hdrs parsed, check against other limits */
 977                 if (ctx.mss) {
 978                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 979                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 980                                 goto hdr_too_big;
 981                         }
 982                 } else {
 983                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 984                                 if (unlikely(ctx.eth_ip_hdr_size +
 985                                              skb->csum_offset >
 986                                              VMXNET3_MAX_CSUM_OFFSET)) {
 987                                         goto hdr_too_big;
 988                                 }
 989                         }
 990                 }
 991         } else {
 992                 tq->stats.drop_hdr_inspect_err++;
 993                 goto unlock_drop_pkt;
 994         }
 995
 996         /* fill tx descs related to addr & len */
 997         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
 998
 999         /* setup the EOP desc */
1000         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1001
1002         /* setup the SOP desc */
1003 #ifdef __BIG_ENDIAN_BITFIELD
1004         gdesc = &tempTxDesc;
1005         gdesc->dword[2] = ctx.sop_txd->dword[2];
1006         gdesc->dword[3] = ctx.sop_txd->dword[3];
1007 #else
1008         gdesc = ctx.sop_txd;
1009 #endif
1010         if (ctx.mss) {
1011                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1012                 gdesc->txd.om = VMXNET3_OM_TSO;
1013                 gdesc->txd.msscof = ctx.mss;
1014                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1015                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1016         } else {
1017                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1018                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1019                         gdesc->txd.om = VMXNET3_OM_CSUM;
1020                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1021                                             skb->csum_offset;
1022                 } else {
1023                         gdesc->txd.om = 0;
1024                         gdesc->txd.msscof = 0;
1025                 }
1026                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1027         }
1028
1029         if (vlan_tx_tag_present(skb)) {
1030                 gdesc->txd.ti = 1;
1031                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1032         }
1033
1034         /* finally flips the GEN bit of the SOP desc. */
1035         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1036                                                   VMXNET3_TXD_GEN);
1037 #ifdef __BIG_ENDIAN_BITFIELD
1038         /* Finished updating in bitfields of Tx Desc, so write them in original
1039          * place.
1040          */
1041         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1042                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1043         gdesc = ctx.sop_txd;
1044 #endif
1045         dev_dbg(&adapter->netdev->dev,
1046                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1047                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1048                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1049                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1050
1051         spin_unlock_irqrestore(&tq->tx_lock, flags);
1052
1053         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1054                                         le32_to_cpu(tq->shared->txThreshold)) {
1055                 tq->shared->txNumDeferred = 0;
1056                 VMXNET3_WRITE_BAR0_REG(adapter,
1057                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1058                                        tq->tx_ring.next2fill);
1059         }
1060
1061         return NETDEV_TX_OK;
1062
1063 hdr_too_big:
1064         tq->stats.drop_oversized_hdr++;
1065 unlock_drop_pkt:
1066         spin_unlock_irqrestore(&tq->tx_lock, flags);
1067 drop_pkt:
1068         tq->stats.drop_total++;
1069         dev_kfree_skb(skb);
1070         return NETDEV_TX_OK;
1071 }
1072
1073
1074 static netdev_tx_t
1075 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1076 {
1077         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1078
1079                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1080                 return vmxnet3_tq_xmit(skb,
1081                                        &adapter->tx_queue[skb->queue_mapping],
1082                                        adapter, netdev);
1083 }
1084
1085
1086 static void
1087 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1088                 struct sk_buff *skb,
1089                 union Vmxnet3_GenericDesc *gdesc)
1090 {
1091         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1092                 /* typical case: TCP/UDP over IP and both csums are correct */
1093                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1094                                                         VMXNET3_RCD_CSUM_OK) {
1095                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1096                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1097                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1098                         BUG_ON(gdesc->rcd.frg);
1099                 } else {
1100                         if (gdesc->rcd.csum) {
1101                                 skb->csum = htons(gdesc->rcd.csum);
1102                                 skb->ip_summed = CHECKSUM_PARTIAL;
1103                         } else {
1104                                 skb_checksum_none_assert(skb);
1105                         }
1106                 }
1107         } else {
1108                 skb_checksum_none_assert(skb);
1109         }
1110 }
1111
1112
1113 static void
1114 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1115                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1116 {
1117         rq->stats.drop_err++;
1118         if (!rcd->fcs)
1119                 rq->stats.drop_fcs++;
1120
1121         rq->stats.drop_total++;
1122
1123         /*
1124          * We do not unmap and chain the rx buffer to the skb.
1125          * We basically pretend this buffer is not used and will be recycled
1126          * by vmxnet3_rq_alloc_rx_buf()
1127          */
1128
1129         /*
1130          * ctx->skb may be NULL if this is the first and the only one
1131          * desc for the pkt
1132          */
1133         if (ctx->skb)
1134                 dev_kfree_skb_irq(ctx->skb);
1135
1136         ctx->skb = NULL;
1137 }
1138
1139
1140 static int
1141 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1142                        struct vmxnet3_adapter *adapter, int quota)
1143 {
1144         static const u32 rxprod_reg[2] = {
1145                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1146         };
1147         u32 num_rxd = 0;
1148         bool skip_page_frags = false;
1149         struct Vmxnet3_RxCompDesc *rcd;
1150         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1151 #ifdef __BIG_ENDIAN_BITFIELD
1152         struct Vmxnet3_RxDesc rxCmdDesc;
1153         struct Vmxnet3_RxCompDesc rxComp;
1154 #endif
1155         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1156                           &rxComp);
1157         while (rcd->gen == rq->comp_ring.gen) {
1158                 struct vmxnet3_rx_buf_info *rbi;
1159                 struct sk_buff *skb, *new_skb = NULL;
1160                 struct page *new_page = NULL;
1161                 int num_to_alloc;
1162                 struct Vmxnet3_RxDesc *rxd;
1163                 u32 idx, ring_idx;
1164                 struct vmxnet3_cmd_ring *ring = NULL;
1165                 if (num_rxd >= quota) {
1166                         /* we may stop even before we see the EOP desc of
1167                          * the current pkt
1168                          */
1169                         break;
1170                 }
1171                 num_rxd++;
1172                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1173                 idx = rcd->rxdIdx;
1174                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1175                 ring = rq->rx_ring + ring_idx;
1176                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1177                                   &rxCmdDesc);
1178                 rbi = rq->buf_info[ring_idx] + idx;
1179
1180                 BUG_ON(rxd->addr != rbi->dma_addr ||
1181                        rxd->len != rbi->len);
1182
1183                 if (unlikely(rcd->eop && rcd->err)) {
1184                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1185                         goto rcd_done;
1186                 }
1187
1188                 if (rcd->sop) { /* first buf of the pkt */
1189                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1190                                rcd->rqID != rq->qid);
1191
1192                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1193                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1194
1195                         if (unlikely(rcd->len == 0)) {
1196                                 /* Pretend the rx buffer is skipped. */
1197                                 BUG_ON(!(rcd->sop && rcd->eop));
1198                                 dev_dbg(&adapter->netdev->dev,
1199                                         "rxRing[%u][%u] 0 length\n",
1200                                         ring_idx, idx);
1201                                 goto rcd_done;
1202                         }
1203
1204                         skip_page_frags = false;
1205                         ctx->skb = rbi->skb;
1206                         new_skb = dev_alloc_skb(rbi->len + NET_IP_ALIGN);
1207                         if (new_skb == NULL) {
1208                                 /* Skb allocation failed, do not handover this
1209                                  * skb to stack. Reuse it. Drop the existing pkt
1210                                  */
1211                                 rq->stats.rx_buf_alloc_failure++;
1212                                 ctx->skb = NULL;
1213                                 rq->stats.drop_total++;
1214                                 skip_page_frags = true;
1215                                 goto rcd_done;
1216                         }
1217
1218                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1219                                          PCI_DMA_FROMDEVICE);
1220
1221                         skb_put(ctx->skb, rcd->len);
1222
1223                         /* Immediate refill */
1224                         new_skb->dev = adapter->netdev;
1225                         skb_reserve(new_skb, NET_IP_ALIGN);
1226                         rbi->skb = new_skb;
1227                         rbi->dma_addr = pci_map_single(adapter->pdev,
1228                                         rbi->skb->data, rbi->len,
1229                                         PCI_DMA_FROMDEVICE);
1230                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1231                         rxd->len = rbi->len;
1232
1233                 } else {
1234                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1235
1236                         /* non SOP buffer must be type 1 in most cases */
1237                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1238                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1239
1240                         /* If an sop buffer was dropped, skip all
1241                          * following non-sop fragments. They will be reused.
1242                          */
1243                         if (skip_page_frags)
1244                                 goto rcd_done;
1245
1246                         new_page = alloc_page(GFP_ATOMIC);
1247                         if (unlikely(new_page == NULL)) {
1248                                 /* Replacement page frag could not be allocated.
1249                                  * Reuse this page. Drop the pkt and free the
1250                                  * skb which contained this page as a frag. Skip
1251                                  * processing all the following non-sop frags.
1252                                  */
1253                                 rq->stats.rx_buf_alloc_failure++;
1254                                 dev_kfree_skb(ctx->skb);
1255                                 ctx->skb = NULL;
1256                                 skip_page_frags = true;
1257                                 goto rcd_done;
1258                         }
1259
1260                         if (rcd->len) {
1261                                 pci_unmap_page(adapter->pdev,
1262                                                rbi->dma_addr, rbi->len,
1263                                                PCI_DMA_FROMDEVICE);
1264
1265                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1266                         }
1267
1268                         /* Immediate refill */
1269                         rbi->page = new_page;
1270                         rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1271                                                      0, PAGE_SIZE,
1272                                                      PCI_DMA_FROMDEVICE);
1273                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1274                         rxd->len = rbi->len;
1275                 }
1276
1277
1278                 skb = ctx->skb;
1279                 if (rcd->eop) {
1280                         skb->len += skb->data_len;
1281
1282                         vmxnet3_rx_csum(adapter, skb,
1283                                         (union Vmxnet3_GenericDesc *)rcd);
1284                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1285
1286                         if (unlikely(rcd->ts))
1287                                 __vlan_hwaccel_put_tag(skb, rcd->tci);
1288
1289                         if (adapter->netdev->features & NETIF_F_LRO)
1290                                 netif_receive_skb(skb);
1291                         else
1292                                 napi_gro_receive(&rq->napi, skb);
1293
1294                         ctx->skb = NULL;
1295                 }
1296
1297 rcd_done:
1298                 /* device may have skipped some rx descs */
1299                 ring->next2comp = idx;
1300                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1301                 ring = rq->rx_ring + ring_idx;
1302                 while (num_to_alloc) {
1303                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1304                                           &rxCmdDesc);
1305                         BUG_ON(!rxd->addr);
1306
1307                         /* Recv desc is ready to be used by the device */
1308                         rxd->gen = ring->gen;
1309                         vmxnet3_cmd_ring_adv_next2fill(ring);
1310                         num_to_alloc--;
1311                 }
1312
1313                 /* if needed, update the register */
1314                 if (unlikely(rq->shared->updateRxProd)) {
1315                         VMXNET3_WRITE_BAR0_REG(adapter,
1316                                 rxprod_reg[ring_idx] + rq->qid * 8,
1317                                 ring->next2fill);
1318                         rq->uncommitted[ring_idx] = 0;
1319                 }
1320
1321                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1322                 vmxnet3_getRxComp(rcd,
1323                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1324         }
1325
1326         return num_rxd;
1327 }
1328
1329
1330 static void
1331 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1332                    struct vmxnet3_adapter *adapter)
1333 {
1334         u32 i, ring_idx;
1335         struct Vmxnet3_RxDesc *rxd;
1336
1337         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1338                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1339 #ifdef __BIG_ENDIAN_BITFIELD
1340                         struct Vmxnet3_RxDesc rxDesc;
1341 #endif
1342                         vmxnet3_getRxDesc(rxd,
1343                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1344
1345                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1346                                         rq->buf_info[ring_idx][i].skb) {
1347                                 pci_unmap_single(adapter->pdev, rxd->addr,
1348                                                  rxd->len, PCI_DMA_FROMDEVICE);
1349                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1350                                 rq->buf_info[ring_idx][i].skb = NULL;
1351                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1352                                         rq->buf_info[ring_idx][i].page) {
1353                                 pci_unmap_page(adapter->pdev, rxd->addr,
1354                                                rxd->len, PCI_DMA_FROMDEVICE);
1355                                 put_page(rq->buf_info[ring_idx][i].page);
1356                                 rq->buf_info[ring_idx][i].page = NULL;
1357                         }
1358                 }
1359
1360                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1361                 rq->rx_ring[ring_idx].next2fill =
1362                                         rq->rx_ring[ring_idx].next2comp = 0;
1363                 rq->uncommitted[ring_idx] = 0;
1364         }
1365
1366         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1367         rq->comp_ring.next2proc = 0;
1368 }
1369
1370
1371 static void
1372 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1373 {
1374         int i;
1375
1376         for (i = 0; i < adapter->num_rx_queues; i++)
1377                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1378 }
1379
1380
1381 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1382                         struct vmxnet3_adapter *adapter)
1383 {
1384         int i;
1385         int j;
1386
1387         /* all rx buffers must have already been freed */
1388         for (i = 0; i < 2; i++) {
1389                 if (rq->buf_info[i]) {
1390                         for (j = 0; j < rq->rx_ring[i].size; j++)
1391                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1392                 }
1393         }
1394
1395
1396         kfree(rq->buf_info[0]);
1397
1398         for (i = 0; i < 2; i++) {
1399                 if (rq->rx_ring[i].base) {
1400                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1401                                             * sizeof(struct Vmxnet3_RxDesc),
1402                                             rq->rx_ring[i].base,
1403                                             rq->rx_ring[i].basePA);
1404                         rq->rx_ring[i].base = NULL;
1405                 }
1406                 rq->buf_info[i] = NULL;
1407         }
1408
1409         if (rq->comp_ring.base) {
1410                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1411                                     sizeof(struct Vmxnet3_RxCompDesc),
1412                                     rq->comp_ring.base, rq->comp_ring.basePA);
1413                 rq->comp_ring.base = NULL;
1414         }
1415 }
1416
1417
1418 static int
1419 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1420                 struct vmxnet3_adapter  *adapter)
1421 {
1422         int i;
1423
1424         /* initialize buf_info */
1425         for (i = 0; i < rq->rx_ring[0].size; i++) {
1426
1427                 /* 1st buf for a pkt is skbuff */
1428                 if (i % adapter->rx_buf_per_pkt == 0) {
1429                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1430                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1431                 } else { /* subsequent bufs for a pkt is frag */
1432                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1433                         rq->buf_info[0][i].len = PAGE_SIZE;
1434                 }
1435         }
1436         for (i = 0; i < rq->rx_ring[1].size; i++) {
1437                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1438                 rq->buf_info[1][i].len = PAGE_SIZE;
1439         }
1440
1441         /* reset internal state and allocate buffers for both rings */
1442         for (i = 0; i < 2; i++) {
1443                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1444                 rq->uncommitted[i] = 0;
1445
1446                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1447                        sizeof(struct Vmxnet3_RxDesc));
1448                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1449         }
1450         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1451                                     adapter) == 0) {
1452                 /* at least has 1 rx buffer for the 1st ring */
1453                 return -ENOMEM;
1454         }
1455         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1456
1457         /* reset the comp ring */
1458         rq->comp_ring.next2proc = 0;
1459         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1460                sizeof(struct Vmxnet3_RxCompDesc));
1461         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1462
1463         /* reset rxctx */
1464         rq->rx_ctx.skb = NULL;
1465
1466         /* stats are not reset */
1467         return 0;
1468 }
1469
1470
1471 static int
1472 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1473 {
1474         int i, err = 0;
1475
1476         for (i = 0; i < adapter->num_rx_queues; i++) {
1477                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1478                 if (unlikely(err)) {
1479                         dev_err(&adapter->netdev->dev, "%s: failed to "
1480                                 "initialize rx queue%i\n",
1481                                 adapter->netdev->name, i);
1482                         break;
1483                 }
1484         }
1485         return err;
1486
1487 }
1488
1489
1490 static int
1491 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1492 {
1493         int i;
1494         size_t sz;
1495         struct vmxnet3_rx_buf_info *bi;
1496
1497         for (i = 0; i < 2; i++) {
1498
1499                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1500                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1501                                                         &rq->rx_ring[i].basePA);
1502                 if (!rq->rx_ring[i].base) {
1503                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1504                                adapter->netdev->name, i);
1505                         goto err;
1506                 }
1507         }
1508
1509         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1510         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1511                                                   &rq->comp_ring.basePA);
1512         if (!rq->comp_ring.base) {
1513                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1514                        adapter->netdev->name);
1515                 goto err;
1516         }
1517
1518         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1519                                                    rq->rx_ring[1].size);
1520         bi = kzalloc(sz, GFP_KERNEL);
1521         if (!bi) {
1522                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1523                        adapter->netdev->name);
1524                 goto err;
1525         }
1526         rq->buf_info[0] = bi;
1527         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1528
1529         return 0;
1530
1531 err:
1532         vmxnet3_rq_destroy(rq, adapter);
1533         return -ENOMEM;
1534 }
1535
1536
1537 static int
1538 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1539 {
1540         int i, err = 0;
1541
1542         for (i = 0; i < adapter->num_rx_queues; i++) {
1543                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1544                 if (unlikely(err)) {
1545                         dev_err(&adapter->netdev->dev,
1546                                 "%s: failed to create rx queue%i\n",
1547                                 adapter->netdev->name, i);
1548                         goto err_out;
1549                 }
1550         }
1551         return err;
1552 err_out:
1553         vmxnet3_rq_destroy_all(adapter);
1554         return err;
1555
1556 }
1557
1558 /* Multiple queue aware polling function for tx and rx */
1559
1560 static int
1561 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1562 {
1563         int rcd_done = 0, i;
1564         if (unlikely(adapter->shared->ecr))
1565                 vmxnet3_process_events(adapter);
1566         for (i = 0; i < adapter->num_tx_queues; i++)
1567                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1568
1569         for (i = 0; i < adapter->num_rx_queues; i++)
1570                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1571                                                    adapter, budget);
1572         return rcd_done;
1573 }
1574
1575
1576 static int
1577 vmxnet3_poll(struct napi_struct *napi, int budget)
1578 {
1579         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1580                                           struct vmxnet3_rx_queue, napi);
1581         int rxd_done;
1582
1583         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1584
1585         if (rxd_done < budget) {
1586                 napi_complete(napi);
1587                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1588         }
1589         return rxd_done;
1590 }
1591
1592 /*
1593  * NAPI polling function for MSI-X mode with multiple Rx queues
1594  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1595  */
1596
1597 static int
1598 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1599 {
1600         struct vmxnet3_rx_queue *rq = container_of(napi,
1601                                                 struct vmxnet3_rx_queue, napi);
1602         struct vmxnet3_adapter *adapter = rq->adapter;
1603         int rxd_done;
1604
1605         /* When sharing interrupt with corresponding tx queue, process
1606          * tx completions in that queue as well
1607          */
1608         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1609                 struct vmxnet3_tx_queue *tq =
1610                                 &adapter->tx_queue[rq - adapter->rx_queue];
1611                 vmxnet3_tq_tx_complete(tq, adapter);
1612         }
1613
1614         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1615
1616         if (rxd_done < budget) {
1617                 napi_complete(napi);
1618                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1619         }
1620         return rxd_done;
1621 }
1622
1623
1624 #ifdef CONFIG_PCI_MSI
1625
1626 /*
1627  * Handle completion interrupts on tx queues
1628  * Returns whether or not the intr is handled
1629  */
1630
1631 static irqreturn_t
1632 vmxnet3_msix_tx(int irq, void *data)
1633 {
1634         struct vmxnet3_tx_queue *tq = data;
1635         struct vmxnet3_adapter *adapter = tq->adapter;
1636
1637         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1638                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1639
1640         /* Handle the case where only one irq is allocate for all tx queues */
1641         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1642                 int i;
1643                 for (i = 0; i < adapter->num_tx_queues; i++) {
1644                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1645                         vmxnet3_tq_tx_complete(txq, adapter);
1646                 }
1647         } else {
1648                 vmxnet3_tq_tx_complete(tq, adapter);
1649         }
1650         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1651
1652         return IRQ_HANDLED;
1653 }
1654
1655
1656 /*
1657  * Handle completion interrupts on rx queues. Returns whether or not the
1658  * intr is handled
1659  */
1660
1661 static irqreturn_t
1662 vmxnet3_msix_rx(int irq, void *data)
1663 {
1664         struct vmxnet3_rx_queue *rq = data;
1665         struct vmxnet3_adapter *adapter = rq->adapter;
1666
1667         /* disable intr if needed */
1668         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1669                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1670         napi_schedule(&rq->napi);
1671
1672         return IRQ_HANDLED;
1673 }
1674
1675 /*
1676  *----------------------------------------------------------------------------
1677  *
1678  * vmxnet3_msix_event --
1679  *
1680  *    vmxnet3 msix event intr handler
1681  *
1682  * Result:
1683  *    whether or not the intr is handled
1684  *
1685  *----------------------------------------------------------------------------
1686  */
1687
1688 static irqreturn_t
1689 vmxnet3_msix_event(int irq, void *data)
1690 {
1691         struct net_device *dev = data;
1692         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1693
1694         /* disable intr if needed */
1695         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1696                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1697
1698         if (adapter->shared->ecr)
1699                 vmxnet3_process_events(adapter);
1700
1701         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1702
1703         return IRQ_HANDLED;
1704 }
1705
1706 #endif /* CONFIG_PCI_MSI  */
1707
1708
1709 /* Interrupt handler for vmxnet3  */
1710 static irqreturn_t
1711 vmxnet3_intr(int irq, void *dev_id)
1712 {
1713         struct net_device *dev = dev_id;
1714         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1715
1716         if (adapter->intr.type == VMXNET3_IT_INTX) {
1717                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1718                 if (unlikely(icr == 0))
1719                         /* not ours */
1720                         return IRQ_NONE;
1721         }
1722
1723
1724         /* disable intr if needed */
1725         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1726                 vmxnet3_disable_all_intrs(adapter);
1727
1728         napi_schedule(&adapter->rx_queue[0].napi);
1729
1730         return IRQ_HANDLED;
1731 }
1732
1733 #ifdef CONFIG_NET_POLL_CONTROLLER
1734
1735 /* netpoll callback. */
1736 static void
1737 vmxnet3_netpoll(struct net_device *netdev)
1738 {
1739         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1740
1741         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1742                 vmxnet3_disable_all_intrs(adapter);
1743
1744         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1745         vmxnet3_enable_all_intrs(adapter);
1746
1747 }
1748 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1749
1750 static int
1751 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1752 {
1753         struct vmxnet3_intr *intr = &adapter->intr;
1754         int err = 0, i;
1755         int vector = 0;
1756
1757 #ifdef CONFIG_PCI_MSI
1758         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1759                 for (i = 0; i < adapter->num_tx_queues; i++) {
1760                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1761                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1762                                         adapter->netdev->name, vector);
1763                                 err = request_irq(
1764                                               intr->msix_entries[vector].vector,
1765                                               vmxnet3_msix_tx, 0,
1766                                               adapter->tx_queue[i].name,
1767                                               &adapter->tx_queue[i]);
1768                         } else {
1769                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1770                                         adapter->netdev->name, vector);
1771                         }
1772                         if (err) {
1773                                 dev_err(&adapter->netdev->dev,
1774                                         "Failed to request irq for MSIX, %s, "
1775                                         "error %d\n",
1776                                         adapter->tx_queue[i].name, err);
1777                                 return err;
1778                         }
1779
1780                         /* Handle the case where only 1 MSIx was allocated for
1781                          * all tx queues */
1782                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1783                                 for (; i < adapter->num_tx_queues; i++)
1784                                         adapter->tx_queue[i].comp_ring.intr_idx
1785                                                                 = vector;
1786                                 vector++;
1787                                 break;
1788                         } else {
1789                                 adapter->tx_queue[i].comp_ring.intr_idx
1790                                                                 = vector++;
1791                         }
1792                 }
1793                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1794                         vector = 0;
1795
1796                 for (i = 0; i < adapter->num_rx_queues; i++) {
1797                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1798                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1799                                         adapter->netdev->name, vector);
1800                         else
1801                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1802                                         adapter->netdev->name, vector);
1803                         err = request_irq(intr->msix_entries[vector].vector,
1804                                           vmxnet3_msix_rx, 0,
1805                                           adapter->rx_queue[i].name,
1806                                           &(adapter->rx_queue[i]));
1807                         if (err) {
1808                                 printk(KERN_ERR "Failed to request irq for MSIX"
1809                                        ", %s, error %d\n",
1810                                        adapter->rx_queue[i].name, err);
1811                                 return err;
1812                         }
1813
1814                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1815                 }
1816
1817                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1818                         adapter->netdev->name, vector);
1819                 err = request_irq(intr->msix_entries[vector].vector,
1820                                   vmxnet3_msix_event, 0,
1821                                   intr->event_msi_vector_name, adapter->netdev);
1822                 intr->event_intr_idx = vector;
1823
1824         } else if (intr->type == VMXNET3_IT_MSI) {
1825                 adapter->num_rx_queues = 1;
1826                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1827                                   adapter->netdev->name, adapter->netdev);
1828         } else {
1829 #endif
1830                 adapter->num_rx_queues = 1;
1831                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1832                                   IRQF_SHARED, adapter->netdev->name,
1833                                   adapter->netdev);
1834 #ifdef CONFIG_PCI_MSI
1835         }
1836 #endif
1837         intr->num_intrs = vector + 1;
1838         if (err) {
1839                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1840                        ":%d\n", adapter->netdev->name, intr->type, err);
1841         } else {
1842                 /* Number of rx queues will not change after this */
1843                 for (i = 0; i < adapter->num_rx_queues; i++) {
1844                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1845                         rq->qid = i;
1846                         rq->qid2 = i + adapter->num_rx_queues;
1847                 }
1848
1849
1850
1851                 /* init our intr settings */
1852                 for (i = 0; i < intr->num_intrs; i++)
1853                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1854                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1855                         adapter->intr.event_intr_idx = 0;
1856                         for (i = 0; i < adapter->num_tx_queues; i++)
1857                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1858                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1859                 }
1860
1861                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1862                        "allocated\n", adapter->netdev->name, intr->type,
1863                        intr->mask_mode, intr->num_intrs);
1864         }
1865
1866         return err;
1867 }
1868
1869
1870 static void
1871 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1872 {
1873         struct vmxnet3_intr *intr = &adapter->intr;
1874         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1875
1876         switch (intr->type) {
1877 #ifdef CONFIG_PCI_MSI
1878         case VMXNET3_IT_MSIX:
1879         {
1880                 int i, vector = 0;
1881
1882                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1883                         for (i = 0; i < adapter->num_tx_queues; i++) {
1884                                 free_irq(intr->msix_entries[vector++].vector,
1885                                          &(adapter->tx_queue[i]));
1886                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1887                                         break;
1888                         }
1889                 }
1890
1891                 for (i = 0; i < adapter->num_rx_queues; i++) {
1892                         free_irq(intr->msix_entries[vector++].vector,
1893                                  &(adapter->rx_queue[i]));
1894                 }
1895
1896                 free_irq(intr->msix_entries[vector].vector,
1897                          adapter->netdev);
1898                 BUG_ON(vector >= intr->num_intrs);
1899                 break;
1900         }
1901 #endif
1902         case VMXNET3_IT_MSI:
1903                 free_irq(adapter->pdev->irq, adapter->netdev);
1904                 break;
1905         case VMXNET3_IT_INTX:
1906                 free_irq(adapter->pdev->irq, adapter->netdev);
1907                 break;
1908         default:
1909                 BUG_ON(true);
1910         }
1911 }
1912
1913
1914 static void
1915 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1916 {
1917         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1918         u16 vid;
1919
1920         /* allow untagged pkts */
1921         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1922
1923         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1924                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1925 }
1926
1927
1928 static void
1929 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1930 {
1931         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1932
1933         if (!(netdev->flags & IFF_PROMISC)) {
1934                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1935                 unsigned long flags;
1936
1937                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1938                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1939                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1940                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1941                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1942         }
1943
1944         set_bit(vid, adapter->active_vlans);
1945 }
1946
1947
1948 static void
1949 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1950 {
1951         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1952
1953         if (!(netdev->flags & IFF_PROMISC)) {
1954                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1955                 unsigned long flags;
1956
1957                 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1958                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1959                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1960                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1961                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1962         }
1963
1964         clear_bit(vid, adapter->active_vlans);
1965 }
1966
1967
1968 static u8 *
1969 vmxnet3_copy_mc(struct net_device *netdev)
1970 {
1971         u8 *buf = NULL;
1972         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1973
1974         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1975         if (sz <= 0xffff) {
1976                 /* We may be called with BH disabled */
1977                 buf = kmalloc(sz, GFP_ATOMIC);
1978                 if (buf) {
1979                         struct netdev_hw_addr *ha;
1980                         int i = 0;
1981
1982                         netdev_for_each_mc_addr(ha, netdev)
1983                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1984                                        ETH_ALEN);
1985                 }
1986         }
1987         return buf;
1988 }
1989
1990
1991 static void
1992 vmxnet3_set_mc(struct net_device *netdev)
1993 {
1994         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1995         unsigned long flags;
1996         struct Vmxnet3_RxFilterConf *rxConf =
1997                                         &adapter->shared->devRead.rxFilterConf;
1998         u8 *new_table = NULL;
1999         u32 new_mode = VMXNET3_RXM_UCAST;
2000
2001         if (netdev->flags & IFF_PROMISC) {
2002                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2003                 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2004
2005                 new_mode |= VMXNET3_RXM_PROMISC;
2006         } else {
2007                 vmxnet3_restore_vlan(adapter);
2008         }
2009
2010         if (netdev->flags & IFF_BROADCAST)
2011                 new_mode |= VMXNET3_RXM_BCAST;
2012
2013         if (netdev->flags & IFF_ALLMULTI)
2014                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2015         else
2016                 if (!netdev_mc_empty(netdev)) {
2017                         new_table = vmxnet3_copy_mc(netdev);
2018                         if (new_table) {
2019                                 new_mode |= VMXNET3_RXM_MCAST;
2020                                 rxConf->mfTableLen = cpu_to_le16(
2021                                         netdev_mc_count(netdev) * ETH_ALEN);
2022                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2023                                                     new_table));
2024                         } else {
2025                                 printk(KERN_INFO "%s: failed to copy mcast list"
2026                                        ", setting ALL_MULTI\n", netdev->name);
2027                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2028                         }
2029                 }
2030
2031
2032         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2033                 rxConf->mfTableLen = 0;
2034                 rxConf->mfTablePA = 0;
2035         }
2036
2037         spin_lock_irqsave(&adapter->cmd_lock, flags);
2038         if (new_mode != rxConf->rxMode) {
2039                 rxConf->rxMode = cpu_to_le32(new_mode);
2040                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2041                                        VMXNET3_CMD_UPDATE_RX_MODE);
2042                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2043                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2044         }
2045
2046         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2047                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2048         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2049
2050         kfree(new_table);
2051 }
2052
2053 void
2054 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2055 {
2056         int i;
2057
2058         for (i = 0; i < adapter->num_rx_queues; i++)
2059                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2060 }
2061
2062
2063 /*
2064  *   Set up driver_shared based on settings in adapter.
2065  */
2066
2067 static void
2068 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2069 {
2070         struct Vmxnet3_DriverShared *shared = adapter->shared;
2071         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2072         struct Vmxnet3_TxQueueConf *tqc;
2073         struct Vmxnet3_RxQueueConf *rqc;
2074         int i;
2075
2076         memset(shared, 0, sizeof(*shared));
2077
2078         /* driver settings */
2079         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2080         devRead->misc.driverInfo.version = cpu_to_le32(
2081                                                 VMXNET3_DRIVER_VERSION_NUM);
2082         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2083                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2084         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2085         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2086                                 *((u32 *)&devRead->misc.driverInfo.gos));
2087         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2088         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2089
2090         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2091         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2092
2093         /* set up feature flags */
2094         if (adapter->netdev->features & NETIF_F_RXCSUM)
2095                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2096
2097         if (adapter->netdev->features & NETIF_F_LRO) {
2098                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2099                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2100         }
2101         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2102                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2103
2104         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2105         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2106         devRead->misc.queueDescLen = cpu_to_le32(
2107                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2108                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2109
2110         /* tx queue settings */
2111         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2112         for (i = 0; i < adapter->num_tx_queues; i++) {
2113                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2114                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2115                 tqc = &adapter->tqd_start[i].conf;
2116                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2117                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2118                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2119                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2120                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2121                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2122                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2123                 tqc->ddLen          = cpu_to_le32(
2124                                         sizeof(struct vmxnet3_tx_buf_info) *
2125                                         tqc->txRingSize);
2126                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2127         }
2128
2129         /* rx queue settings */
2130         devRead->misc.numRxQueues = adapter->num_rx_queues;
2131         for (i = 0; i < adapter->num_rx_queues; i++) {
2132                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2133                 rqc = &adapter->rqd_start[i].conf;
2134                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2135                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2136                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2137                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2138                                                         rq->buf_info));
2139                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2140                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2141                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2142                 rqc->ddLen           = cpu_to_le32(
2143                                         sizeof(struct vmxnet3_rx_buf_info) *
2144                                         (rqc->rxRingSize[0] +
2145                                          rqc->rxRingSize[1]));
2146                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2147         }
2148
2149 #ifdef VMXNET3_RSS
2150         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2151
2152         if (adapter->rss) {
2153                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2154                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2155                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2156                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2157                                     UPT1_RSS_HASH_TYPE_IPV4 |
2158                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2159                                     UPT1_RSS_HASH_TYPE_IPV6;
2160                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2161                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2162                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2163                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2164                 for (i = 0; i < rssConf->indTableSize; i++)
2165                         rssConf->indTable[i] = i % adapter->num_rx_queues;
2166
2167                 devRead->rssConfDesc.confVer = 1;
2168                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2169                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2170         }
2171
2172 #endif /* VMXNET3_RSS */
2173
2174         /* intr settings */
2175         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2176                                      VMXNET3_IMM_AUTO;
2177         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2178         for (i = 0; i < adapter->intr.num_intrs; i++)
2179                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2180
2181         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2182         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2183
2184         /* rx filter settings */
2185         devRead->rxFilterConf.rxMode = 0;
2186         vmxnet3_restore_vlan(adapter);
2187         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2188
2189         /* the rest are already zeroed */
2190 }
2191
2192
2193 int
2194 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2195 {
2196         int err, i;
2197         u32 ret;
2198         unsigned long flags;
2199
2200         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2201                 " ring sizes %u %u %u\n", adapter->netdev->name,
2202                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2203                 adapter->tx_queue[0].tx_ring.size,
2204                 adapter->rx_queue[0].rx_ring[0].size,
2205                 adapter->rx_queue[0].rx_ring[1].size);
2206
2207         vmxnet3_tq_init_all(adapter);
2208         err = vmxnet3_rq_init_all(adapter);
2209         if (err) {
2210                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2211                        adapter->netdev->name, err);
2212                 goto rq_err;
2213         }
2214
2215         err = vmxnet3_request_irqs(adapter);
2216         if (err) {
2217                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2218                        adapter->netdev->name, err);
2219                 goto irq_err;
2220         }
2221
2222         vmxnet3_setup_driver_shared(adapter);
2223
2224         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2225                                adapter->shared_pa));
2226         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2227                                adapter->shared_pa));
2228         spin_lock_irqsave(&adapter->cmd_lock, flags);
2229         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2230                                VMXNET3_CMD_ACTIVATE_DEV);
2231         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2232         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2233
2234         if (ret != 0) {
2235                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2236                        adapter->netdev->name, ret);
2237                 err = -EINVAL;
2238                 goto activate_err;
2239         }
2240
2241         for (i = 0; i < adapter->num_rx_queues; i++) {
2242                 VMXNET3_WRITE_BAR0_REG(adapter,
2243                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2244                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2245                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2246                                 (i * VMXNET3_REG_ALIGN)),
2247                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2248         }
2249
2250         /* Apply the rx filter settins last. */
2251         vmxnet3_set_mc(adapter->netdev);
2252
2253         /*
2254          * Check link state when first activating device. It will start the
2255          * tx queue if the link is up.
2256          */
2257         vmxnet3_check_link(adapter, true);
2258         for (i = 0; i < adapter->num_rx_queues; i++)
2259                 napi_enable(&adapter->rx_queue[i].napi);
2260         vmxnet3_enable_all_intrs(adapter);
2261         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2262         return 0;
2263
2264 activate_err:
2265         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2266         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2267         vmxnet3_free_irqs(adapter);
2268 irq_err:
2269 rq_err:
2270         /* free up buffers we allocated */
2271         vmxnet3_rq_cleanup_all(adapter);
2272         return err;
2273 }
2274
2275
2276 void
2277 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2278 {
2279         unsigned long flags;
2280         spin_lock_irqsave(&adapter->cmd_lock, flags);
2281         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2282         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2283 }
2284
2285
2286 int
2287 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2288 {
2289         int i;
2290         unsigned long flags;
2291         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2292                 return 0;
2293
2294
2295         spin_lock_irqsave(&adapter->cmd_lock, flags);
2296         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2297                                VMXNET3_CMD_QUIESCE_DEV);
2298         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2299         vmxnet3_disable_all_intrs(adapter);
2300
2301         for (i = 0; i < adapter->num_rx_queues; i++)
2302                 napi_disable(&adapter->rx_queue[i].napi);
2303         netif_tx_disable(adapter->netdev);
2304         adapter->link_speed = 0;
2305         netif_carrier_off(adapter->netdev);
2306
2307         vmxnet3_tq_cleanup_all(adapter);
2308         vmxnet3_rq_cleanup_all(adapter);
2309         vmxnet3_free_irqs(adapter);
2310         return 0;
2311 }
2312
2313
2314 static void
2315 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2316 {
2317         u32 tmp;
2318
2319         tmp = *(u32 *)mac;
2320         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2321
2322         tmp = (mac[5] << 8) | mac[4];
2323         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2324 }
2325
2326
2327 static int
2328 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2329 {
2330         struct sockaddr *addr = p;
2331         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2332
2333         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2334         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2335
2336         return 0;
2337 }
2338
2339
2340 /* ==================== initialization and cleanup routines ============ */
2341
2342 static int
2343 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2344 {
2345         int err;
2346         unsigned long mmio_start, mmio_len;
2347         struct pci_dev *pdev = adapter->pdev;
2348
2349         err = pci_enable_device(pdev);
2350         if (err) {
2351                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2352                        pci_name(pdev), err);
2353                 return err;
2354         }
2355
2356         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2357                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2358                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2359                                "for adapter %s\n", pci_name(pdev));
2360                         err = -EIO;
2361                         goto err_set_mask;
2362                 }
2363                 *dma64 = true;
2364         } else {
2365                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2366                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2367                                "%s\n",  pci_name(pdev));
2368                         err = -EIO;
2369                         goto err_set_mask;
2370                 }
2371                 *dma64 = false;
2372         }
2373
2374         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2375                                            vmxnet3_driver_name);
2376         if (err) {
2377                 printk(KERN_ERR "Failed to request region for adapter %s: "
2378                        "error %d\n", pci_name(pdev), err);
2379                 goto err_set_mask;
2380         }
2381
2382         pci_set_master(pdev);
2383
2384         mmio_start = pci_resource_start(pdev, 0);
2385         mmio_len = pci_resource_len(pdev, 0);
2386         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2387         if (!adapter->hw_addr0) {
2388                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2389                        pci_name(pdev));
2390                 err = -EIO;
2391                 goto err_ioremap;
2392         }
2393
2394         mmio_start = pci_resource_start(pdev, 1);
2395         mmio_len = pci_resource_len(pdev, 1);
2396         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2397         if (!adapter->hw_addr1) {
2398                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2399                        pci_name(pdev));
2400                 err = -EIO;
2401                 goto err_bar1;
2402         }
2403         return 0;
2404
2405 err_bar1:
2406         iounmap(adapter->hw_addr0);
2407 err_ioremap:
2408         pci_release_selected_regions(pdev, (1 << 2) - 1);
2409 err_set_mask:
2410         pci_disable_device(pdev);
2411         return err;
2412 }
2413
2414
2415 static void
2416 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2417 {
2418         BUG_ON(!adapter->pdev);
2419
2420         iounmap(adapter->hw_addr0);
2421         iounmap(adapter->hw_addr1);
2422         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2423         pci_disable_device(adapter->pdev);
2424 }
2425
2426
2427 static void
2428 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2429 {
2430         size_t sz, i, ring0_size, ring1_size, comp_size;
2431         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2432
2433
2434         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2435                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2436                 adapter->skb_buf_size = adapter->netdev->mtu +
2437                                         VMXNET3_MAX_ETH_HDR_SIZE;
2438                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2439                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2440
2441                 adapter->rx_buf_per_pkt = 1;
2442         } else {
2443                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2444                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2445                                             VMXNET3_MAX_ETH_HDR_SIZE;
2446                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2447         }
2448
2449         /*
2450          * for simplicity, force the ring0 size to be a multiple of
2451          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2452          */
2453         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2454         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2455         ring0_size = (ring0_size + sz - 1) / sz * sz;
2456         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2457                            sz * sz);
2458         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2459         comp_size = ring0_size + ring1_size;
2460
2461         for (i = 0; i < adapter->num_rx_queues; i++) {
2462                 rq = &adapter->rx_queue[i];
2463                 rq->rx_ring[0].size = ring0_size;
2464                 rq->rx_ring[1].size = ring1_size;
2465                 rq->comp_ring.size = comp_size;
2466         }
2467 }
2468
2469
2470 int
2471 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2472                       u32 rx_ring_size, u32 rx_ring2_size)
2473 {
2474         int err = 0, i;
2475
2476         for (i = 0; i < adapter->num_tx_queues; i++) {
2477                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2478                 tq->tx_ring.size   = tx_ring_size;
2479                 tq->data_ring.size = tx_ring_size;
2480                 tq->comp_ring.size = tx_ring_size;
2481                 tq->shared = &adapter->tqd_start[i].ctrl;
2482                 tq->stopped = true;
2483                 tq->adapter = adapter;
2484                 tq->qid = i;
2485                 err = vmxnet3_tq_create(tq, adapter);
2486                 /*
2487                  * Too late to change num_tx_queues. We cannot do away with
2488                  * lesser number of queues than what we asked for
2489                  */
2490                 if (err)
2491                         goto queue_err;
2492         }
2493
2494         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2495         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2496         vmxnet3_adjust_rx_ring_size(adapter);
2497         for (i = 0; i < adapter->num_rx_queues; i++) {
2498                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2499                 /* qid and qid2 for rx queues will be assigned later when num
2500                  * of rx queues is finalized after allocating intrs */
2501                 rq->shared = &adapter->rqd_start[i].ctrl;
2502                 rq->adapter = adapter;
2503                 err = vmxnet3_rq_create(rq, adapter);
2504                 if (err) {
2505                         if (i == 0) {
2506                                 printk(KERN_ERR "Could not allocate any rx"
2507                                        "queues. Aborting.\n");
2508                                 goto queue_err;
2509                         } else {
2510                                 printk(KERN_INFO "Number of rx queues changed "
2511                                        "to : %d.\n", i);
2512                                 adapter->num_rx_queues = i;
2513                                 err = 0;
2514                                 break;
2515                         }
2516                 }
2517         }
2518         return err;
2519 queue_err:
2520         vmxnet3_tq_destroy_all(adapter);
2521         return err;
2522 }
2523
2524 static int
2525 vmxnet3_open(struct net_device *netdev)
2526 {
2527         struct vmxnet3_adapter *adapter;
2528         int err, i;
2529
2530         adapter = netdev_priv(netdev);
2531
2532         for (i = 0; i < adapter->num_tx_queues; i++)
2533                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2534
2535         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2536                                     VMXNET3_DEF_RX_RING_SIZE,
2537                                     VMXNET3_DEF_RX_RING_SIZE);
2538         if (err)
2539                 goto queue_err;
2540
2541         err = vmxnet3_activate_dev(adapter);
2542         if (err)
2543                 goto activate_err;
2544
2545         return 0;
2546
2547 activate_err:
2548         vmxnet3_rq_destroy_all(adapter);
2549         vmxnet3_tq_destroy_all(adapter);
2550 queue_err:
2551         return err;
2552 }
2553
2554
2555 static int
2556 vmxnet3_close(struct net_device *netdev)
2557 {
2558         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2559
2560         /*
2561          * Reset_work may be in the middle of resetting the device, wait for its
2562          * completion.
2563          */
2564         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2565                 msleep(1);
2566
2567         vmxnet3_quiesce_dev(adapter);
2568
2569         vmxnet3_rq_destroy_all(adapter);
2570         vmxnet3_tq_destroy_all(adapter);
2571
2572         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2573
2574
2575         return 0;
2576 }
2577
2578
2579 void
2580 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2581 {
2582         int i;
2583
2584         /*
2585          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2586          * vmxnet3_close() will deadlock.
2587          */
2588         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2589
2590         /* we need to enable NAPI, otherwise dev_close will deadlock */
2591         for (i = 0; i < adapter->num_rx_queues; i++)
2592                 napi_enable(&adapter->rx_queue[i].napi);
2593         dev_close(adapter->netdev);
2594 }
2595
2596
2597 static int
2598 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2599 {
2600         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2601         int err = 0;
2602
2603         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2604                 return -EINVAL;
2605
2606         netdev->mtu = new_mtu;
2607
2608         /*
2609          * Reset_work may be in the middle of resetting the device, wait for its
2610          * completion.
2611          */
2612         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2613                 msleep(1);
2614
2615         if (netif_running(netdev)) {
2616                 vmxnet3_quiesce_dev(adapter);
2617                 vmxnet3_reset_dev(adapter);
2618
2619                 /* we need to re-create the rx queue based on the new mtu */
2620                 vmxnet3_rq_destroy_all(adapter);
2621                 vmxnet3_adjust_rx_ring_size(adapter);
2622                 err = vmxnet3_rq_create_all(adapter);
2623                 if (err) {
2624                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2625                                 " error %d. Closing it.\n", netdev->name, err);
2626                         goto out;
2627                 }
2628
2629                 err = vmxnet3_activate_dev(adapter);
2630                 if (err) {
2631                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2632                                 "Closing it\n", netdev->name, err);
2633                         goto out;
2634                 }
2635         }
2636
2637 out:
2638         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2639         if (err)
2640                 vmxnet3_force_close(adapter);
2641
2642         return err;
2643 }
2644
2645
2646 static void
2647 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2648 {
2649         struct net_device *netdev = adapter->netdev;
2650
2651         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2652                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2653                 NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2654                 NETIF_F_LRO;
2655         if (dma64)
2656                 netdev->hw_features |= NETIF_F_HIGHDMA;
2657         netdev->vlan_features = netdev->hw_features &
2658                                 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
2659         netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER;
2660
2661         netdev_info(adapter->netdev,
2662                 "features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2663                 dma64 ? " highDMA" : "");
2664 }
2665
2666
2667 static void
2668 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2669 {
2670         u32 tmp;
2671
2672         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2673         *(u32 *)mac = tmp;
2674
2675         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2676         mac[4] = tmp & 0xff;
2677         mac[5] = (tmp >> 8) & 0xff;
2678 }
2679
2680 #ifdef CONFIG_PCI_MSI
2681
2682 /*
2683  * Enable MSIx vectors.
2684  * Returns :
2685  *      0 on successful enabling of required vectors,
2686  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2687  *       could be enabled.
2688  *      number of vectors which can be enabled otherwise (this number is smaller
2689  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2690  */
2691
2692 static int
2693 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2694                              int vectors)
2695 {
2696         int err = 0, vector_threshold;
2697         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2698
2699         while (vectors >= vector_threshold) {
2700                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2701                                       vectors);
2702                 if (!err) {
2703                         adapter->intr.num_intrs = vectors;
2704                         return 0;
2705                 } else if (err < 0) {
2706                         printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2707                                " %d\n", adapter->netdev->name, err);
2708                         vectors = 0;
2709                 } else if (err < vector_threshold) {
2710                         break;
2711                 } else {
2712                         /* If fails to enable required number of MSI-x vectors
2713                          * try enabling minimum number of vectors required.
2714                          */
2715                         vectors = vector_threshold;
2716                         printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2717                                " %d instead\n", vectors, adapter->netdev->name,
2718                                vector_threshold);
2719                 }
2720         }
2721
2722         printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2723                " are lower than min threshold required.\n");
2724         return err;
2725 }
2726
2727
2728 #endif /* CONFIG_PCI_MSI */
2729
2730 static void
2731 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2732 {
2733         u32 cfg;
2734         unsigned long flags;
2735
2736         /* intr settings */
2737         spin_lock_irqsave(&adapter->cmd_lock, flags);
2738         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2739                                VMXNET3_CMD_GET_CONF_INTR);
2740         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2741         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2742         adapter->intr.type = cfg & 0x3;
2743         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2744
2745         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2746                 adapter->intr.type = VMXNET3_IT_MSIX;
2747         }
2748
2749 #ifdef CONFIG_PCI_MSI
2750         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2751                 int vector, err = 0;
2752
2753                 adapter->intr.num_intrs = (adapter->share_intr ==
2754                                            VMXNET3_INTR_TXSHARE) ? 1 :
2755                                            adapter->num_tx_queues;
2756                 adapter->intr.num_intrs += (adapter->share_intr ==
2757                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2758                                            adapter->num_rx_queues;
2759                 adapter->intr.num_intrs += 1;           /* for link event */
2760
2761                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2762                                            VMXNET3_LINUX_MIN_MSIX_VECT
2763                                            ? adapter->intr.num_intrs :
2764                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2765
2766                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2767                         adapter->intr.msix_entries[vector].entry = vector;
2768
2769                 err = vmxnet3_acquire_msix_vectors(adapter,
2770                                                    adapter->intr.num_intrs);
2771                 /* If we cannot allocate one MSIx vector per queue
2772                  * then limit the number of rx queues to 1
2773                  */
2774                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2775                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2776                             || adapter->num_rx_queues != 1) {
2777                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2778                                 printk(KERN_ERR "Number of rx queues : 1\n");
2779                                 adapter->num_rx_queues = 1;
2780                                 adapter->intr.num_intrs =
2781                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2782                         }
2783                         return;
2784                 }
2785                 if (!err)
2786                         return;
2787
2788                 /* If we cannot allocate MSIx vectors use only one rx queue */
2789                 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2790                        "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2791
2792                 adapter->intr.type = VMXNET3_IT_MSI;
2793         }
2794
2795         if (adapter->intr.type == VMXNET3_IT_MSI) {
2796                 int err;
2797                 err = pci_enable_msi(adapter->pdev);
2798                 if (!err) {
2799                         adapter->num_rx_queues = 1;
2800                         adapter->intr.num_intrs = 1;
2801                         return;
2802                 }
2803         }
2804 #endif /* CONFIG_PCI_MSI */
2805
2806         adapter->num_rx_queues = 1;
2807         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2808         adapter->intr.type = VMXNET3_IT_INTX;
2809
2810         /* INT-X related setting */
2811         adapter->intr.num_intrs = 1;
2812 }
2813
2814
2815 static void
2816 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2817 {
2818         if (adapter->intr.type == VMXNET3_IT_MSIX)
2819                 pci_disable_msix(adapter->pdev);
2820         else if (adapter->intr.type == VMXNET3_IT_MSI)
2821                 pci_disable_msi(adapter->pdev);
2822         else
2823                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2824 }
2825
2826
2827 static void
2828 vmxnet3_tx_timeout(struct net_device *netdev)
2829 {
2830         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2831         adapter->tx_timeout_count++;
2832
2833         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2834         schedule_work(&adapter->work);
2835         netif_wake_queue(adapter->netdev);
2836 }
2837
2838
2839 static void
2840 vmxnet3_reset_work(struct work_struct *data)
2841 {
2842         struct vmxnet3_adapter *adapter;
2843
2844         adapter = container_of(data, struct vmxnet3_adapter, work);
2845
2846         /* if another thread is resetting the device, no need to proceed */
2847         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2848                 return;
2849
2850         /* if the device is closed, we must leave it alone */
2851         rtnl_lock();
2852         if (netif_running(adapter->netdev)) {
2853                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2854                 vmxnet3_quiesce_dev(adapter);
2855                 vmxnet3_reset_dev(adapter);
2856                 vmxnet3_activate_dev(adapter);
2857         } else {
2858                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2859         }
2860         rtnl_unlock();
2861
2862         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2863 }
2864
2865
2866 static int __devinit
2867 vmxnet3_probe_device(struct pci_dev *pdev,
2868                      const struct pci_device_id *id)
2869 {
2870         static const struct net_device_ops vmxnet3_netdev_ops = {
2871                 .ndo_open = vmxnet3_open,
2872                 .ndo_stop = vmxnet3_close,
2873                 .ndo_start_xmit = vmxnet3_xmit_frame,
2874                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2875                 .ndo_change_mtu = vmxnet3_change_mtu,
2876                 .ndo_set_features = vmxnet3_set_features,
2877                 .ndo_get_stats64 = vmxnet3_get_stats64,
2878                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2879                 .ndo_set_rx_mode = vmxnet3_set_mc,
2880                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2881                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2882 #ifdef CONFIG_NET_POLL_CONTROLLER
2883                 .ndo_poll_controller = vmxnet3_netpoll,
2884 #endif
2885         };
2886         int err;
2887         bool dma64 = false; /* stupid gcc */
2888         u32 ver;
2889         struct net_device *netdev;
2890         struct vmxnet3_adapter *adapter;
2891         u8 mac[ETH_ALEN];
2892         int size;
2893         int num_tx_queues;
2894         int num_rx_queues;
2895
2896         if (!pci_msi_enabled())
2897                 enable_mq = 0;
2898
2899 #ifdef VMXNET3_RSS
2900         if (enable_mq)
2901                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2902                                     (int)num_online_cpus());
2903         else
2904 #endif
2905                 num_rx_queues = 1;
2906         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2907
2908         if (enable_mq)
2909                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2910                                     (int)num_online_cpus());
2911         else
2912                 num_tx_queues = 1;
2913
2914         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2915         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2916                                    max(num_tx_queues, num_rx_queues));
2917         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2918                num_tx_queues, num_rx_queues);
2919
2920         if (!netdev) {
2921                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2922                         "%s\n", pci_name(pdev));
2923                 return -ENOMEM;
2924         }
2925
2926         pci_set_drvdata(pdev, netdev);
2927         adapter = netdev_priv(netdev);
2928         adapter->netdev = netdev;
2929         adapter->pdev = pdev;
2930
2931         spin_lock_init(&adapter->cmd_lock);
2932         adapter->shared = pci_alloc_consistent(adapter->pdev,
2933                           sizeof(struct Vmxnet3_DriverShared),
2934                           &adapter->shared_pa);
2935         if (!adapter->shared) {
2936                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2937                         pci_name(pdev));
2938                 err = -ENOMEM;
2939                 goto err_alloc_shared;
2940         }
2941
2942         adapter->num_rx_queues = num_rx_queues;
2943         adapter->num_tx_queues = num_tx_queues;
2944
2945         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2946         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2947         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2948                              &adapter->queue_desc_pa);
2949
2950         if (!adapter->tqd_start) {
2951                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2952                         pci_name(pdev));
2953                 err = -ENOMEM;
2954                 goto err_alloc_queue_desc;
2955         }
2956         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2957                                                         adapter->num_tx_queues);
2958
2959         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2960         if (adapter->pm_conf == NULL) {
2961                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2962                         pci_name(pdev));
2963                 err = -ENOMEM;
2964                 goto err_alloc_pm;
2965         }
2966
2967 #ifdef VMXNET3_RSS
2968
2969         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2970         if (adapter->rss_conf == NULL) {
2971                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2972                        pci_name(pdev));
2973                 err = -ENOMEM;
2974                 goto err_alloc_rss;
2975         }
2976 #endif /* VMXNET3_RSS */
2977
2978         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2979         if (err < 0)
2980                 goto err_alloc_pci;
2981
2982         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2983         if (ver & 1) {
2984                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2985         } else {
2986                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2987                        " %s\n", ver, pci_name(pdev));
2988                 err = -EBUSY;
2989                 goto err_ver;
2990         }
2991
2992         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2993         if (ver & 1) {
2994                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
2995         } else {
2996                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
2997                        "adapter %s\n", ver, pci_name(pdev));
2998                 err = -EBUSY;
2999                 goto err_ver;
3000         }
3001
3002         SET_NETDEV_DEV(netdev, &pdev->dev);
3003         vmxnet3_declare_features(adapter, dma64);
3004
3005         adapter->dev_number = atomic_read(&devices_found);
3006
3007          adapter->share_intr = irq_share_mode;
3008         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3009             adapter->num_tx_queues != adapter->num_rx_queues)
3010                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3011
3012         vmxnet3_alloc_intr_resources(adapter);
3013
3014 #ifdef VMXNET3_RSS
3015         if (adapter->num_rx_queues > 1 &&
3016             adapter->intr.type == VMXNET3_IT_MSIX) {
3017                 adapter->rss = true;
3018                 printk(KERN_INFO "RSS is enabled.\n");
3019         } else {
3020                 adapter->rss = false;
3021         }
3022 #endif
3023
3024         vmxnet3_read_mac_addr(adapter, mac);
3025         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3026
3027         netdev->netdev_ops = &vmxnet3_netdev_ops;
3028         vmxnet3_set_ethtool_ops(netdev);
3029         netdev->watchdog_timeo = 5 * HZ;
3030
3031         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3032
3033         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3034                 int i;
3035                 for (i = 0; i < adapter->num_rx_queues; i++) {
3036                         netif_napi_add(adapter->netdev,
3037                                        &adapter->rx_queue[i].napi,
3038                                        vmxnet3_poll_rx_only, 64);
3039                 }
3040         } else {
3041                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3042                                vmxnet3_poll, 64);
3043         }
3044
3045         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3046         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3047
3048         err = register_netdev(netdev);
3049
3050         if (err) {
3051                 printk(KERN_ERR "Failed to register adapter %s\n",
3052                         pci_name(pdev));
3053                 goto err_register;
3054         }
3055
3056         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3057         vmxnet3_check_link(adapter, false);
3058         atomic_inc(&devices_found);
3059         return 0;
3060
3061 err_register:
3062         vmxnet3_free_intr_resources(adapter);
3063 err_ver:
3064         vmxnet3_free_pci_resources(adapter);
3065 err_alloc_pci:
3066 #ifdef VMXNET3_RSS
3067         kfree(adapter->rss_conf);
3068 err_alloc_rss:
3069 #endif
3070         kfree(adapter->pm_conf);
3071 err_alloc_pm:
3072         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3073                             adapter->queue_desc_pa);
3074 err_alloc_queue_desc:
3075         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3076                             adapter->shared, adapter->shared_pa);
3077 err_alloc_shared:
3078         pci_set_drvdata(pdev, NULL);
3079         free_netdev(netdev);
3080         return err;
3081 }
3082
3083
3084 static void __devexit
3085 vmxnet3_remove_device(struct pci_dev *pdev)
3086 {
3087         struct net_device *netdev = pci_get_drvdata(pdev);
3088         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3089         int size = 0;
3090         int num_rx_queues;
3091
3092 #ifdef VMXNET3_RSS
3093         if (enable_mq)
3094                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3095                                     (int)num_online_cpus());
3096         else
3097 #endif
3098                 num_rx_queues = 1;
3099         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3100
3101         cancel_work_sync(&adapter->work);
3102
3103         unregister_netdev(netdev);
3104
3105         vmxnet3_free_intr_resources(adapter);
3106         vmxnet3_free_pci_resources(adapter);
3107 #ifdef VMXNET3_RSS
3108         kfree(adapter->rss_conf);
3109 #endif
3110         kfree(adapter->pm_conf);
3111
3112         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3113         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3114         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3115                             adapter->queue_desc_pa);
3116         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3117                             adapter->shared, adapter->shared_pa);
3118         free_netdev(netdev);
3119 }
3120
3121
3122 #ifdef CONFIG_PM
3123
3124 static int
3125 vmxnet3_suspend(struct device *device)
3126 {
3127         struct pci_dev *pdev = to_pci_dev(device);
3128         struct net_device *netdev = pci_get_drvdata(pdev);
3129         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3130         struct Vmxnet3_PMConf *pmConf;
3131         struct ethhdr *ehdr;
3132         struct arphdr *ahdr;
3133         u8 *arpreq;
3134         struct in_device *in_dev;
3135         struct in_ifaddr *ifa;
3136         unsigned long flags;
3137         int i = 0;
3138
3139         if (!netif_running(netdev))
3140                 return 0;
3141
3142         for (i = 0; i < adapter->num_rx_queues; i++)
3143                 napi_disable(&adapter->rx_queue[i].napi);
3144
3145         vmxnet3_disable_all_intrs(adapter);
3146         vmxnet3_free_irqs(adapter);
3147         vmxnet3_free_intr_resources(adapter);
3148
3149         netif_device_detach(netdev);
3150         netif_tx_stop_all_queues(netdev);
3151
3152         /* Create wake-up filters. */
3153         pmConf = adapter->pm_conf;
3154         memset(pmConf, 0, sizeof(*pmConf));
3155
3156         if (adapter->wol & WAKE_UCAST) {
3157                 pmConf->filters[i].patternSize = ETH_ALEN;
3158                 pmConf->filters[i].maskSize = 1;
3159                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3160                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3161
3162                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3163                 i++;
3164         }
3165
3166         if (adapter->wol & WAKE_ARP) {
3167                 in_dev = in_dev_get(netdev);
3168                 if (!in_dev)
3169                         goto skip_arp;
3170
3171                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3172                 if (!ifa)
3173                         goto skip_arp;
3174
3175                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3176                         sizeof(struct arphdr) +         /* ARP header */
3177                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3178                         2 * sizeof(u32);        /*2 IPv4 addresses */
3179                 pmConf->filters[i].maskSize =
3180                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3181
3182                 /* ETH_P_ARP in Ethernet header. */
3183                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3184                 ehdr->h_proto = htons(ETH_P_ARP);
3185
3186                 /* ARPOP_REQUEST in ARP header. */
3187                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3188                 ahdr->ar_op = htons(ARPOP_REQUEST);
3189                 arpreq = (u8 *)(ahdr + 1);
3190
3191                 /* The Unicast IPv4 address in 'tip' field. */
3192                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3193                 *(u32 *)arpreq = ifa->ifa_address;
3194
3195                 /* The mask for the relevant bits. */
3196                 pmConf->filters[i].mask[0] = 0x00;
3197                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3198                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3199                 pmConf->filters[i].mask[3] = 0x00;
3200                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3201                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3202                 in_dev_put(in_dev);
3203
3204                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3205                 i++;
3206         }
3207
3208 skip_arp:
3209         if (adapter->wol & WAKE_MAGIC)
3210                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3211
3212         pmConf->numFilters = i;
3213
3214         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3215         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3216                                                                   *pmConf));
3217         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3218                                                                  pmConf));
3219
3220         spin_lock_irqsave(&adapter->cmd_lock, flags);
3221         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3222                                VMXNET3_CMD_UPDATE_PMCFG);
3223         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3224
3225         pci_save_state(pdev);
3226         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3227                         adapter->wol);
3228         pci_disable_device(pdev);
3229         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3230
3231         return 0;
3232 }
3233
3234
3235 static int
3236 vmxnet3_resume(struct device *device)
3237 {
3238         int err, i = 0;
3239         unsigned long flags;
3240         struct pci_dev *pdev = to_pci_dev(device);
3241         struct net_device *netdev = pci_get_drvdata(pdev);
3242         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3243         struct Vmxnet3_PMConf *pmConf;
3244
3245         if (!netif_running(netdev))
3246                 return 0;
3247
3248         /* Destroy wake-up filters. */
3249         pmConf = adapter->pm_conf;
3250         memset(pmConf, 0, sizeof(*pmConf));
3251
3252         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3253         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3254                                                                   *pmConf));
3255         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3256                                                                  pmConf));
3257
3258         netif_device_attach(netdev);
3259         pci_set_power_state(pdev, PCI_D0);
3260         pci_restore_state(pdev);
3261         err = pci_enable_device_mem(pdev);
3262         if (err != 0)
3263                 return err;
3264
3265         pci_enable_wake(pdev, PCI_D0, 0);
3266
3267         spin_lock_irqsave(&adapter->cmd_lock, flags);
3268         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3269                                VMXNET3_CMD_UPDATE_PMCFG);
3270         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3271         vmxnet3_alloc_intr_resources(adapter);
3272         vmxnet3_request_irqs(adapter);
3273         for (i = 0; i < adapter->num_rx_queues; i++)
3274                 napi_enable(&adapter->rx_queue[i].napi);
3275         vmxnet3_enable_all_intrs(adapter);
3276
3277         return 0;
3278 }
3279
3280 static const struct dev_pm_ops vmxnet3_pm_ops = {
3281         .suspend = vmxnet3_suspend,
3282         .resume = vmxnet3_resume,
3283 };
3284 #endif
3285
3286 static struct pci_driver vmxnet3_driver = {
3287         .name           = vmxnet3_driver_name,
3288         .id_table       = vmxnet3_pciid_table,
3289         .probe          = vmxnet3_probe_device,
3290         .remove         = __devexit_p(vmxnet3_remove_device),
3291 #ifdef CONFIG_PM
3292         .driver.pm      = &vmxnet3_pm_ops,
3293 #endif
3294 };
3295
3296
3297 static int __init
3298 vmxnet3_init_module(void)
3299 {
3300         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3301                 VMXNET3_DRIVER_VERSION_REPORT);
3302         return pci_register_driver(&vmxnet3_driver);
3303 }
3304
3305 module_init(vmxnet3_init_module);
3306
3307
3308 static void
3309 vmxnet3_exit_module(void)
3310 {
3311         pci_unregister_driver(&vmxnet3_driver);
3312 }
3313
3314 module_exit(vmxnet3_exit_module);
3315
3316 MODULE_AUTHOR("VMware, Inc.");
3317 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3318 MODULE_LICENSE("GPL v2");
3319 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);