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