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                 goto err;
 542
 543         return 0;
 544
 545 err:
 546         vmxnet3_tq_destroy(tq, adapter);
 547         return -ENOMEM;
 548 }
 549
 550 static void
 551 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
 552 {
 553         int i;
 554
 555         for (i = 0; i < adapter->num_tx_queues; i++)
 556                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
 557 }
 558
 559 /*
 560  *    starting from ring->next2fill, allocate rx buffers for the given ring
 561  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 562  *    are allocated or allocation fails
 563  */
 564
 565 static int
 566 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 567                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 568 {
 569         int num_allocated = 0;
 570         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 571         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 572         u32 val;
 573
 574         while (num_allocated <= num_to_alloc) {
 575                 struct vmxnet3_rx_buf_info *rbi;
 576                 union Vmxnet3_GenericDesc *gd;
 577
 578                 rbi = rbi_base + ring->next2fill;
 579                 gd = ring->base + ring->next2fill;
 580
 581                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 582                         if (rbi->skb == NULL) {
 583                                 rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
 584                                                                        rbi->len,
 585                                                                        GFP_KERNEL);
 586                                 if (unlikely(rbi->skb == NULL)) {
 587                                         rq->stats.rx_buf_alloc_failure++;
 588                                         break;
 589                                 }
 590
 591                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 592                                                 rbi->skb->data, rbi->len,
 593                                                 PCI_DMA_FROMDEVICE);
 594                         } else {
 595                                 /* rx buffer skipped by the device */
 596                         }
 597                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 598                 } else {
 599                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 600                                rbi->len  != PAGE_SIZE);
 601
 602                         if (rbi->page == NULL) {
 603                                 rbi->page = alloc_page(GFP_ATOMIC);
 604                                 if (unlikely(rbi->page == NULL)) {
 605                                         rq->stats.rx_buf_alloc_failure++;
 606                                         break;
 607                                 }
 608                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 609                                                 rbi->page, 0, PAGE_SIZE,
 610                                                 PCI_DMA_FROMDEVICE);
 611                         } else {
 612                                 /* rx buffers skipped by the device */
 613                         }
 614                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 615                 }
 616
 617                 BUG_ON(rbi->dma_addr == 0);
 618                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 619                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 620                                            | val | rbi->len);
 621
 622                 /* Fill the last buffer but dont mark it ready, or else the
 623                  * device will think that the queue is full */
 624                 if (num_allocated == num_to_alloc)
 625                         break;
 626
 627                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 628                 num_allocated++;
 629                 vmxnet3_cmd_ring_adv_next2fill(ring);
 630         }
 631
 632         netdev_dbg(adapter->netdev,
 633                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
 634                 num_allocated, ring->next2fill, ring->next2comp);
 635
 636         /* so that the device can distinguish a full ring and an empty ring */
 637         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 638
 639         return num_allocated;
 640 }
 641
 642
 643 static void
 644 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 645                     struct vmxnet3_rx_buf_info *rbi)
 646 {
 647         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 648                 skb_shinfo(skb)->nr_frags;
 649
 650         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 651
 652         __skb_frag_set_page(frag, rbi->page);
 653         frag->page_offset = 0;
 654         skb_frag_size_set(frag, rcd->len);
 655         skb->data_len += rcd->len;
 656         skb->truesize += PAGE_SIZE;
 657         skb_shinfo(skb)->nr_frags++;
 658 }
 659
 660
 661 static void
 662 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 663                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 664                 struct vmxnet3_adapter *adapter)
 665 {
 666         u32 dw2, len;
 667         unsigned long buf_offset;
 668         int i;
 669         union Vmxnet3_GenericDesc *gdesc;
 670         struct vmxnet3_tx_buf_info *tbi = NULL;
 671
 672         BUG_ON(ctx->copy_size > skb_headlen(skb));
 673
 674         /* use the previous gen bit for the SOP desc */
 675         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 676
 677         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 678         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 679
 680         /* no need to map the buffer if headers are copied */
 681         if (ctx->copy_size) {
 682                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 683                                         tq->tx_ring.next2fill *
 684                                         sizeof(struct Vmxnet3_TxDataDesc));
 685                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 686                 ctx->sop_txd->dword[3] = 0;
 687
 688                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 689                 tbi->map_type = VMXNET3_MAP_NONE;
 690
 691                 netdev_dbg(adapter->netdev,
 692                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 693                         tq->tx_ring.next2fill,
 694                         le64_to_cpu(ctx->sop_txd->txd.addr),
 695                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 696                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 697
 698                 /* use the right gen for non-SOP desc */
 699                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 700         }
 701
 702         /* linear part can use multiple tx desc if it's big */
 703         len = skb_headlen(skb) - ctx->copy_size;
 704         buf_offset = ctx->copy_size;
 705         while (len) {
 706                 u32 buf_size;
 707
 708                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 709                         buf_size = len;
 710                         dw2 |= len;
 711                 } else {
 712                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 713                         /* spec says that for TxDesc.len, 0 == 2^14 */
 714                 }
 715
 716                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 717                 tbi->map_type = VMXNET3_MAP_SINGLE;
 718                 tbi->dma_addr = pci_map_single(adapter->pdev,
 719                                 skb->data + buf_offset, buf_size,
 720                                 PCI_DMA_TODEVICE);
 721
 722                 tbi->len = buf_size;
 723
 724                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 725                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 726
 727                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 728                 gdesc->dword[2] = cpu_to_le32(dw2);
 729                 gdesc->dword[3] = 0;
 730
 731                 netdev_dbg(adapter->netdev,
 732                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 733                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 734                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 735                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 736                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 737
 738                 len -= buf_size;
 739                 buf_offset += buf_size;
 740         }
 741
 742         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 743                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 744                 u32 buf_size;
 745
 746                 buf_offset = 0;
 747                 len = skb_frag_size(frag);
 748                 while (len) {
 749                         tbi = tq->buf_info + tq->tx_ring.next2fill;
 750                         if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 751                                 buf_size = len;
 752                                 dw2 |= len;
 753                         } else {
 754                                 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 755                                 /* spec says that for TxDesc.len, 0 == 2^14 */
 756                         }
 757                         tbi->map_type = VMXNET3_MAP_PAGE;
 758                         tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
 759                                                          buf_offset, buf_size,
 760                                                          DMA_TO_DEVICE);
 761
 762                         tbi->len = buf_size;
 763
 764                         gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 765                         BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 766
 767                         gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 768                         gdesc->dword[2] = cpu_to_le32(dw2);
 769                         gdesc->dword[3] = 0;
 770
 771                         netdev_dbg(adapter->netdev,
 772                                 "txd[%u]: 0x%llu %u %u\n",
 773                                 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 774                                 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 775                         vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 776                         dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 777
 778                         len -= buf_size;
 779                         buf_offset += buf_size;
 780                 }
 781         }
 782
 783         ctx->eop_txd = gdesc;
 784
 785         /* set the last buf_info for the pkt */
 786         tbi->skb = skb;
 787         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 788 }
 789
 790
 791 /* Init all tx queues */
 792 static void
 793 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 794 {
 795         int i;
 796
 797         for (i = 0; i < adapter->num_tx_queues; i++)
 798                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 799 }
 800
 801
 802 /*
 803  *    parse and copy relevant protocol headers:
 804  *      For a tso pkt, relevant headers are L2/3/4 including options
 805  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 806  *      if it's a TCP/UDP pkt
 807  *
 808  * Returns:
 809  *    -1:  error happens during parsing
 810  *     0:  protocol headers parsed, but too big to be copied
 811  *     1:  protocol headers parsed and copied
 812  *
 813  * Other effects:
 814  *    1. related *ctx fields are updated.
 815  *    2. ctx->copy_size is # of bytes copied
 816  *    3. the portion copied is guaranteed to be in the linear part
 817  *
 818  */
 819 static int
 820 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 821                            struct vmxnet3_tx_ctx *ctx,
 822                            struct vmxnet3_adapter *adapter)
 823 {
 824         struct Vmxnet3_TxDataDesc *tdd;
 825
 826         if (ctx->mss) { /* TSO */
 827                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 828                 ctx->l4_hdr_size = tcp_hdrlen(skb);
 829                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 830         } else {
 831                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 832                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 833
 834                         if (ctx->ipv4) {
 835                                 const struct iphdr *iph = ip_hdr(skb);
 836
 837                                 if (iph->protocol == IPPROTO_TCP)
 838                                         ctx->l4_hdr_size = tcp_hdrlen(skb);
 839                                 else if (iph->protocol == IPPROTO_UDP)
 840                                         ctx->l4_hdr_size = sizeof(struct udphdr);
 841                                 else
 842                                         ctx->l4_hdr_size = 0;
 843                         } else {
 844                                 /* for simplicity, don't copy L4 headers */
 845                                 ctx->l4_hdr_size = 0;
 846                         }
 847                         ctx->copy_size = min(ctx->eth_ip_hdr_size +
 848                                          ctx->l4_hdr_size, skb->len);
 849                 } else {
 850                         ctx->eth_ip_hdr_size = 0;
 851                         ctx->l4_hdr_size = 0;
 852                         /* copy as much as allowed */
 853                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 854                                              , skb_headlen(skb));
 855                 }
 856
 857                 /* make sure headers are accessible directly */
 858                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 859                         goto err;
 860         }
 861
 862         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 863                 tq->stats.oversized_hdr++;
 864                 ctx->copy_size = 0;
 865                 return 0;
 866         }
 867
 868         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 869
 870         memcpy(tdd->data, skb->data, ctx->copy_size);
 871         netdev_dbg(adapter->netdev,
 872                 "copy %u bytes to dataRing[%u]\n",
 873                 ctx->copy_size, tq->tx_ring.next2fill);
 874         return 1;
 875
 876 err:
 877         return -1;
 878 }
 879
 880
 881 static void
 882 vmxnet3_prepare_tso(struct sk_buff *skb,
 883                     struct vmxnet3_tx_ctx *ctx)
 884 {
 885         struct tcphdr *tcph = tcp_hdr(skb);
 886
 887         if (ctx->ipv4) {
 888                 struct iphdr *iph = ip_hdr(skb);
 889
 890                 iph->check = 0;
 891                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 892                                                  IPPROTO_TCP, 0);
 893         } else {
 894                 struct ipv6hdr *iph = ipv6_hdr(skb);
 895
 896                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 897                                                IPPROTO_TCP, 0);
 898         }
 899 }
 900
 901 static int txd_estimate(const struct sk_buff *skb)
 902 {
 903         int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 904         int i;
 905
 906         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 907                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 908
 909                 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
 910         }
 911         return count;
 912 }
 913
 914 /*
 915  * Transmits a pkt thru a given tq
 916  * Returns:
 917  *    NETDEV_TX_OK:      descriptors are setup successfully
 918  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 919  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 920  *
 921  * Side-effects:
 922  *    1. tx ring may be changed
 923  *    2. tq stats may be updated accordingly
 924  *    3. shared->txNumDeferred may be updated
 925  */
 926
 927 static int
 928 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 929                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 930 {
 931         int ret;
 932         u32 count;
 933         unsigned long flags;
 934         struct vmxnet3_tx_ctx ctx;
 935         union Vmxnet3_GenericDesc *gdesc;
 936 #ifdef __BIG_ENDIAN_BITFIELD
 937         /* Use temporary descriptor to avoid touching bits multiple times */
 938         union Vmxnet3_GenericDesc tempTxDesc;
 939 #endif
 940
 941         count = txd_estimate(skb);
 942
 943         ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
 944
 945         ctx.mss = skb_shinfo(skb)->gso_size;
 946         if (ctx.mss) {
 947                 if (skb_header_cloned(skb)) {
 948                         if (unlikely(pskb_expand_head(skb, 0, 0,
 949                                                       GFP_ATOMIC) != 0)) {
 950                                 tq->stats.drop_tso++;
 951                                 goto drop_pkt;
 952                         }
 953                         tq->stats.copy_skb_header++;
 954                 }
 955                 vmxnet3_prepare_tso(skb, &ctx);
 956         } else {
 957                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 958
 959                         /* non-tso pkts must not use more than
 960                          * VMXNET3_MAX_TXD_PER_PKT entries
 961                          */
 962                         if (skb_linearize(skb) != 0) {
 963                                 tq->stats.drop_too_many_frags++;
 964                                 goto drop_pkt;
 965                         }
 966                         tq->stats.linearized++;
 967
 968                         /* recalculate the # of descriptors to use */
 969                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 970                 }
 971         }
 972
 973         spin_lock_irqsave(&tq->tx_lock, flags);
 974
 975         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 976                 tq->stats.tx_ring_full++;
 977                 netdev_dbg(adapter->netdev,
 978                         "tx queue stopped on %s, next2comp %u"
 979                         " next2fill %u\n", adapter->netdev->name,
 980                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 981
 982                 vmxnet3_tq_stop(tq, adapter);
 983                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 984                 return NETDEV_TX_BUSY;
 985         }
 986
 987
 988         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 989         if (ret >= 0) {
 990                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 991                 /* hdrs parsed, check against other limits */
 992                 if (ctx.mss) {
 993                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 994                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 995                                 goto hdr_too_big;
 996                         }
 997                 } else {
 998                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 999                                 if (unlikely(ctx.eth_ip_hdr_size +
1000                                              skb->csum_offset >
1001                                              VMXNET3_MAX_CSUM_OFFSET)) {
1002                                         goto hdr_too_big;
1003                                 }
1004                         }
1005                 }
1006         } else {
1007                 tq->stats.drop_hdr_inspect_err++;
1008                 goto unlock_drop_pkt;
1009         }
1010
1011         /* fill tx descs related to addr & len */
1012         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
1013
1014         /* setup the EOP desc */
1015         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1016
1017         /* setup the SOP desc */
1018 #ifdef __BIG_ENDIAN_BITFIELD
1019         gdesc = &tempTxDesc;
1020         gdesc->dword[2] = ctx.sop_txd->dword[2];
1021         gdesc->dword[3] = ctx.sop_txd->dword[3];
1022 #else
1023         gdesc = ctx.sop_txd;
1024 #endif
1025         if (ctx.mss) {
1026                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1027                 gdesc->txd.om = VMXNET3_OM_TSO;
1028                 gdesc->txd.msscof = ctx.mss;
1029                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1030                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1031         } else {
1032                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1033                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1034                         gdesc->txd.om = VMXNET3_OM_CSUM;
1035                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1036                                             skb->csum_offset;
1037                 } else {
1038                         gdesc->txd.om = 0;
1039                         gdesc->txd.msscof = 0;
1040                 }
1041                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1042         }
1043
1044         if (vlan_tx_tag_present(skb)) {
1045                 gdesc->txd.ti = 1;
1046                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1047         }
1048
1049         /* finally flips the GEN bit of the SOP desc. */
1050         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1051                                                   VMXNET3_TXD_GEN);
1052 #ifdef __BIG_ENDIAN_BITFIELD
1053         /* Finished updating in bitfields of Tx Desc, so write them in original
1054          * place.
1055          */
1056         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1057                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1058         gdesc = ctx.sop_txd;
1059 #endif
1060         netdev_dbg(adapter->netdev,
1061                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1062                 (u32)(ctx.sop_txd -
1063                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1064                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1065
1066         spin_unlock_irqrestore(&tq->tx_lock, flags);
1067
1068         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1069                                         le32_to_cpu(tq->shared->txThreshold)) {
1070                 tq->shared->txNumDeferred = 0;
1071                 VMXNET3_WRITE_BAR0_REG(adapter,
1072                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1073                                        tq->tx_ring.next2fill);
1074         }
1075
1076         return NETDEV_TX_OK;
1077
1078 hdr_too_big:
1079         tq->stats.drop_oversized_hdr++;
1080 unlock_drop_pkt:
1081         spin_unlock_irqrestore(&tq->tx_lock, flags);
1082 drop_pkt:
1083         tq->stats.drop_total++;
1084         dev_kfree_skb(skb);
1085         return NETDEV_TX_OK;
1086 }
1087
1088
1089 static netdev_tx_t
1090 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1091 {
1092         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1093
1094         BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1095         return vmxnet3_tq_xmit(skb,
1096                                &adapter->tx_queue[skb->queue_mapping],
1097                                adapter, netdev);
1098 }
1099
1100
1101 static void
1102 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1103                 struct sk_buff *skb,
1104                 union Vmxnet3_GenericDesc *gdesc)
1105 {
1106         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1107                 /* typical case: TCP/UDP over IP and both csums are correct */
1108                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1109                                                         VMXNET3_RCD_CSUM_OK) {
1110                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1111                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1112                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1113                         BUG_ON(gdesc->rcd.frg);
1114                 } else {
1115                         if (gdesc->rcd.csum) {
1116                                 skb->csum = htons(gdesc->rcd.csum);
1117                                 skb->ip_summed = CHECKSUM_PARTIAL;
1118                         } else {
1119                                 skb_checksum_none_assert(skb);
1120                         }
1121                 }
1122         } else {
1123                 skb_checksum_none_assert(skb);
1124         }
1125 }
1126
1127
1128 static void
1129 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1130                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1131 {
1132         rq->stats.drop_err++;
1133         if (!rcd->fcs)
1134                 rq->stats.drop_fcs++;
1135
1136         rq->stats.drop_total++;
1137
1138         /*
1139          * We do not unmap and chain the rx buffer to the skb.
1140          * We basically pretend this buffer is not used and will be recycled
1141          * by vmxnet3_rq_alloc_rx_buf()
1142          */
1143
1144         /*
1145          * ctx->skb may be NULL if this is the first and the only one
1146          * desc for the pkt
1147          */
1148         if (ctx->skb)
1149                 dev_kfree_skb_irq(ctx->skb);
1150
1151         ctx->skb = NULL;
1152 }
1153
1154
1155 static int
1156 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1157                        struct vmxnet3_adapter *adapter, int quota)
1158 {
1159         static const u32 rxprod_reg[2] = {
1160                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1161         };
1162         u32 num_rxd = 0;
1163         bool skip_page_frags = false;
1164         struct Vmxnet3_RxCompDesc *rcd;
1165         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1166 #ifdef __BIG_ENDIAN_BITFIELD
1167         struct Vmxnet3_RxDesc rxCmdDesc;
1168         struct Vmxnet3_RxCompDesc rxComp;
1169 #endif
1170         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1171                           &rxComp);
1172         while (rcd->gen == rq->comp_ring.gen) {
1173                 struct vmxnet3_rx_buf_info *rbi;
1174                 struct sk_buff *skb, *new_skb = NULL;
1175                 struct page *new_page = NULL;
1176                 int num_to_alloc;
1177                 struct Vmxnet3_RxDesc *rxd;
1178                 u32 idx, ring_idx;
1179                 struct vmxnet3_cmd_ring *ring = NULL;
1180                 if (num_rxd >= quota) {
1181                         /* we may stop even before we see the EOP desc of
1182                          * the current pkt
1183                          */
1184                         break;
1185                 }
1186                 num_rxd++;
1187                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1188                 idx = rcd->rxdIdx;
1189                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1190                 ring = rq->rx_ring + ring_idx;
1191                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1192                                   &rxCmdDesc);
1193                 rbi = rq->buf_info[ring_idx] + idx;
1194
1195                 BUG_ON(rxd->addr != rbi->dma_addr ||
1196                        rxd->len != rbi->len);
1197
1198                 if (unlikely(rcd->eop && rcd->err)) {
1199                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1200                         goto rcd_done;
1201                 }
1202
1203                 if (rcd->sop) { /* first buf of the pkt */
1204                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1205                                rcd->rqID != rq->qid);
1206
1207                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1208                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1209
1210                         if (unlikely(rcd->len == 0)) {
1211                                 /* Pretend the rx buffer is skipped. */
1212                                 BUG_ON(!(rcd->sop && rcd->eop));
1213                                 netdev_dbg(adapter->netdev,
1214                                         "rxRing[%u][%u] 0 length\n",
1215                                         ring_idx, idx);
1216                                 goto rcd_done;
1217                         }
1218
1219                         skip_page_frags = false;
1220                         ctx->skb = rbi->skb;
1221                         new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1222                                                             rbi->len);
1223                         if (new_skb == NULL) {
1224                                 /* Skb allocation failed, do not handover this
1225                                  * skb to stack. Reuse it. Drop the existing pkt
1226                                  */
1227                                 rq->stats.rx_buf_alloc_failure++;
1228                                 ctx->skb = NULL;
1229                                 rq->stats.drop_total++;
1230                                 skip_page_frags = true;
1231                                 goto rcd_done;
1232                         }
1233
1234                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1235                                          PCI_DMA_FROMDEVICE);
1236
1237                         skb_put(ctx->skb, rcd->len);
1238
1239                         /* Immediate refill */
1240                         rbi->skb = new_skb;
1241                         rbi->dma_addr = pci_map_single(adapter->pdev,
1242                                                        rbi->skb->data, rbi->len,
1243                                                        PCI_DMA_FROMDEVICE);
1244                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1245                         rxd->len = rbi->len;
1246
1247                 } else {
1248                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1249
1250                         /* non SOP buffer must be type 1 in most cases */
1251                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1252                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1253
1254                         /* If an sop buffer was dropped, skip all
1255                          * following non-sop fragments. They will be reused.
1256                          */
1257                         if (skip_page_frags)
1258                                 goto rcd_done;
1259
1260                         new_page = alloc_page(GFP_ATOMIC);
1261                         if (unlikely(new_page == NULL)) {
1262                                 /* Replacement page frag could not be allocated.
1263                                  * Reuse this page. Drop the pkt and free the
1264                                  * skb which contained this page as a frag. Skip
1265                                  * processing all the following non-sop frags.
1266                                  */
1267                                 rq->stats.rx_buf_alloc_failure++;
1268                                 dev_kfree_skb(ctx->skb);
1269                                 ctx->skb = NULL;
1270                                 skip_page_frags = true;
1271                                 goto rcd_done;
1272                         }
1273
1274                         if (rcd->len) {
1275                                 pci_unmap_page(adapter->pdev,
1276                                                rbi->dma_addr, rbi->len,
1277                                                PCI_DMA_FROMDEVICE);
1278
1279                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1280                         }
1281
1282                         /* Immediate refill */
1283                         rbi->page = new_page;
1284                         rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1285                                                      0, PAGE_SIZE,
1286                                                      PCI_DMA_FROMDEVICE);
1287                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1288                         rxd->len = rbi->len;
1289                 }
1290
1291
1292                 skb = ctx->skb;
1293                 if (rcd->eop) {
1294                         skb->len += skb->data_len;
1295
1296                         vmxnet3_rx_csum(adapter, skb,
1297                                         (union Vmxnet3_GenericDesc *)rcd);
1298                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1299
1300                         if (unlikely(rcd->ts))
1301                                 __vlan_hwaccel_put_tag(skb, rcd->tci);
1302
1303                         if (adapter->netdev->features & NETIF_F_LRO)
1304                                 netif_receive_skb(skb);
1305                         else
1306                                 napi_gro_receive(&rq->napi, skb);
1307
1308                         ctx->skb = NULL;
1309                 }
1310
1311 rcd_done:
1312                 /* device may have skipped some rx descs */
1313                 ring->next2comp = idx;
1314                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1315                 ring = rq->rx_ring + ring_idx;
1316                 while (num_to_alloc) {
1317                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1318                                           &rxCmdDesc);
1319                         BUG_ON(!rxd->addr);
1320
1321                         /* Recv desc is ready to be used by the device */
1322                         rxd->gen = ring->gen;
1323                         vmxnet3_cmd_ring_adv_next2fill(ring);
1324                         num_to_alloc--;
1325                 }
1326
1327                 /* if needed, update the register */
1328                 if (unlikely(rq->shared->updateRxProd)) {
1329                         VMXNET3_WRITE_BAR0_REG(adapter,
1330                                                rxprod_reg[ring_idx] + rq->qid * 8,
1331                                                ring->next2fill);
1332                 }
1333
1334                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1335                 vmxnet3_getRxComp(rcd,
1336                                   &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1337         }
1338
1339         return num_rxd;
1340 }
1341
1342
1343 static void
1344 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1345                    struct vmxnet3_adapter *adapter)
1346 {
1347         u32 i, ring_idx;
1348         struct Vmxnet3_RxDesc *rxd;
1349
1350         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1351                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1352 #ifdef __BIG_ENDIAN_BITFIELD
1353                         struct Vmxnet3_RxDesc rxDesc;
1354 #endif
1355                         vmxnet3_getRxDesc(rxd,
1356                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1357
1358                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1359                                         rq->buf_info[ring_idx][i].skb) {
1360                                 pci_unmap_single(adapter->pdev, rxd->addr,
1361                                                  rxd->len, PCI_DMA_FROMDEVICE);
1362                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1363                                 rq->buf_info[ring_idx][i].skb = NULL;
1364                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1365                                         rq->buf_info[ring_idx][i].page) {
1366                                 pci_unmap_page(adapter->pdev, rxd->addr,
1367                                                rxd->len, PCI_DMA_FROMDEVICE);
1368                                 put_page(rq->buf_info[ring_idx][i].page);
1369                                 rq->buf_info[ring_idx][i].page = NULL;
1370                         }
1371                 }
1372
1373                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1374                 rq->rx_ring[ring_idx].next2fill =
1375                                         rq->rx_ring[ring_idx].next2comp = 0;
1376         }
1377
1378         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1379         rq->comp_ring.next2proc = 0;
1380 }
1381
1382
1383 static void
1384 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1385 {
1386         int i;
1387
1388         for (i = 0; i < adapter->num_rx_queues; i++)
1389                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1390 }
1391
1392
1393 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1394                         struct vmxnet3_adapter *adapter)
1395 {
1396         int i;
1397         int j;
1398
1399         /* all rx buffers must have already been freed */
1400         for (i = 0; i < 2; i++) {
1401                 if (rq->buf_info[i]) {
1402                         for (j = 0; j < rq->rx_ring[i].size; j++)
1403                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1404                 }
1405         }
1406
1407
1408         kfree(rq->buf_info[0]);
1409
1410         for (i = 0; i < 2; i++) {
1411                 if (rq->rx_ring[i].base) {
1412                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1413                                             * sizeof(struct Vmxnet3_RxDesc),
1414                                             rq->rx_ring[i].base,
1415                                             rq->rx_ring[i].basePA);
1416                         rq->rx_ring[i].base = NULL;
1417                 }
1418                 rq->buf_info[i] = NULL;
1419         }
1420
1421         if (rq->comp_ring.base) {
1422                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1423                                     sizeof(struct Vmxnet3_RxCompDesc),
1424                                     rq->comp_ring.base, rq->comp_ring.basePA);
1425                 rq->comp_ring.base = NULL;
1426         }
1427 }
1428
1429
1430 static int
1431 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1432                 struct vmxnet3_adapter  *adapter)
1433 {
1434         int i;
1435
1436         /* initialize buf_info */
1437         for (i = 0; i < rq->rx_ring[0].size; i++) {
1438
1439                 /* 1st buf for a pkt is skbuff */
1440                 if (i % adapter->rx_buf_per_pkt == 0) {
1441                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1442                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1443                 } else { /* subsequent bufs for a pkt is frag */
1444                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1445                         rq->buf_info[0][i].len = PAGE_SIZE;
1446                 }
1447         }
1448         for (i = 0; i < rq->rx_ring[1].size; i++) {
1449                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1450                 rq->buf_info[1][i].len = PAGE_SIZE;
1451         }
1452
1453         /* reset internal state and allocate buffers for both rings */
1454         for (i = 0; i < 2; i++) {
1455                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1456
1457                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1458                        sizeof(struct Vmxnet3_RxDesc));
1459                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1460         }
1461         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1462                                     adapter) == 0) {
1463                 /* at least has 1 rx buffer for the 1st ring */
1464                 return -ENOMEM;
1465         }
1466         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1467
1468         /* reset the comp ring */
1469         rq->comp_ring.next2proc = 0;
1470         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1471                sizeof(struct Vmxnet3_RxCompDesc));
1472         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1473
1474         /* reset rxctx */
1475         rq->rx_ctx.skb = NULL;
1476
1477         /* stats are not reset */
1478         return 0;
1479 }
1480
1481
1482 static int
1483 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1484 {
1485         int i, err = 0;
1486
1487         for (i = 0; i < adapter->num_rx_queues; i++) {
1488                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1489                 if (unlikely(err)) {
1490                         dev_err(&adapter->netdev->dev, "%s: failed to "
1491                                 "initialize rx queue%i\n",
1492                                 adapter->netdev->name, i);
1493                         break;
1494                 }
1495         }
1496         return err;
1497
1498 }
1499
1500
1501 static int
1502 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1503 {
1504         int i;
1505         size_t sz;
1506         struct vmxnet3_rx_buf_info *bi;
1507
1508         for (i = 0; i < 2; i++) {
1509
1510                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1511                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1512                                                         &rq->rx_ring[i].basePA);
1513                 if (!rq->rx_ring[i].base) {
1514                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1515                                adapter->netdev->name, i);
1516                         goto err;
1517                 }
1518         }
1519
1520         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1521         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1522                                                   &rq->comp_ring.basePA);
1523         if (!rq->comp_ring.base) {
1524                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1525                        adapter->netdev->name);
1526                 goto err;
1527         }
1528
1529         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1530                                                    rq->rx_ring[1].size);
1531         bi = kzalloc(sz, GFP_KERNEL);
1532         if (!bi)
1533                 goto err;
1534
1535         rq->buf_info[0] = bi;
1536         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1537
1538         return 0;
1539
1540 err:
1541         vmxnet3_rq_destroy(rq, adapter);
1542         return -ENOMEM;
1543 }
1544
1545
1546 static int
1547 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1548 {
1549         int i, err = 0;
1550
1551         for (i = 0; i < adapter->num_rx_queues; i++) {
1552                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1553                 if (unlikely(err)) {
1554                         dev_err(&adapter->netdev->dev,
1555                                 "%s: failed to create rx queue%i\n",
1556                                 adapter->netdev->name, i);
1557                         goto err_out;
1558                 }
1559         }
1560         return err;
1561 err_out:
1562         vmxnet3_rq_destroy_all(adapter);
1563         return err;
1564
1565 }
1566
1567 /* Multiple queue aware polling function for tx and rx */
1568
1569 static int
1570 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1571 {
1572         int rcd_done = 0, i;
1573         if (unlikely(adapter->shared->ecr))
1574                 vmxnet3_process_events(adapter);
1575         for (i = 0; i < adapter->num_tx_queues; i++)
1576                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1577
1578         for (i = 0; i < adapter->num_rx_queues; i++)
1579                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1580                                                    adapter, budget);
1581         return rcd_done;
1582 }
1583
1584
1585 static int
1586 vmxnet3_poll(struct napi_struct *napi, int budget)
1587 {
1588         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1589                                           struct vmxnet3_rx_queue, napi);
1590         int rxd_done;
1591
1592         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1593
1594         if (rxd_done < budget) {
1595                 napi_complete(napi);
1596                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1597         }
1598         return rxd_done;
1599 }
1600
1601 /*
1602  * NAPI polling function for MSI-X mode with multiple Rx queues
1603  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1604  */
1605
1606 static int
1607 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1608 {
1609         struct vmxnet3_rx_queue *rq = container_of(napi,
1610                                                 struct vmxnet3_rx_queue, napi);
1611         struct vmxnet3_adapter *adapter = rq->adapter;
1612         int rxd_done;
1613
1614         /* When sharing interrupt with corresponding tx queue, process
1615          * tx completions in that queue as well
1616          */
1617         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1618                 struct vmxnet3_tx_queue *tq =
1619                                 &adapter->tx_queue[rq - adapter->rx_queue];
1620                 vmxnet3_tq_tx_complete(tq, adapter);
1621         }
1622
1623         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1624
1625         if (rxd_done < budget) {
1626                 napi_complete(napi);
1627                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1628         }
1629         return rxd_done;
1630 }
1631
1632
1633 #ifdef CONFIG_PCI_MSI
1634
1635 /*
1636  * Handle completion interrupts on tx queues
1637  * Returns whether or not the intr is handled
1638  */
1639
1640 static irqreturn_t
1641 vmxnet3_msix_tx(int irq, void *data)
1642 {
1643         struct vmxnet3_tx_queue *tq = data;
1644         struct vmxnet3_adapter *adapter = tq->adapter;
1645
1646         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1647                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1648
1649         /* Handle the case where only one irq is allocate for all tx queues */
1650         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1651                 int i;
1652                 for (i = 0; i < adapter->num_tx_queues; i++) {
1653                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1654                         vmxnet3_tq_tx_complete(txq, adapter);
1655                 }
1656         } else {
1657                 vmxnet3_tq_tx_complete(tq, adapter);
1658         }
1659         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1660
1661         return IRQ_HANDLED;
1662 }
1663
1664
1665 /*
1666  * Handle completion interrupts on rx queues. Returns whether or not the
1667  * intr is handled
1668  */
1669
1670 static irqreturn_t
1671 vmxnet3_msix_rx(int irq, void *data)
1672 {
1673         struct vmxnet3_rx_queue *rq = data;
1674         struct vmxnet3_adapter *adapter = rq->adapter;
1675
1676         /* disable intr if needed */
1677         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1678                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1679         napi_schedule(&rq->napi);
1680
1681         return IRQ_HANDLED;
1682 }
1683
1684 /*
1685  *----------------------------------------------------------------------------
1686  *
1687  * vmxnet3_msix_event --
1688  *
1689  *    vmxnet3 msix event intr handler
1690  *
1691  * Result:
1692  *    whether or not the intr is handled
1693  *
1694  *----------------------------------------------------------------------------
1695  */
1696
1697 static irqreturn_t
1698 vmxnet3_msix_event(int irq, void *data)
1699 {
1700         struct net_device *dev = data;
1701         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1702
1703         /* disable intr if needed */
1704         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1705                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1706
1707         if (adapter->shared->ecr)
1708                 vmxnet3_process_events(adapter);
1709
1710         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1711
1712         return IRQ_HANDLED;
1713 }
1714
1715 #endif /* CONFIG_PCI_MSI  */
1716
1717
1718 /* Interrupt handler for vmxnet3  */
1719 static irqreturn_t
1720 vmxnet3_intr(int irq, void *dev_id)
1721 {
1722         struct net_device *dev = dev_id;
1723         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1724
1725         if (adapter->intr.type == VMXNET3_IT_INTX) {
1726                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1727                 if (unlikely(icr == 0))
1728                         /* not ours */
1729                         return IRQ_NONE;
1730         }
1731
1732
1733         /* disable intr if needed */
1734         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1735                 vmxnet3_disable_all_intrs(adapter);
1736
1737         napi_schedule(&adapter->rx_queue[0].napi);
1738
1739         return IRQ_HANDLED;
1740 }
1741
1742 #ifdef CONFIG_NET_POLL_CONTROLLER
1743
1744 /* netpoll callback. */
1745 static void
1746 vmxnet3_netpoll(struct net_device *netdev)
1747 {
1748         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1749
1750         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1751                 vmxnet3_disable_all_intrs(adapter);
1752
1753         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1754         vmxnet3_enable_all_intrs(adapter);
1755
1756 }
1757 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1758
1759 static int
1760 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1761 {
1762         struct vmxnet3_intr *intr = &adapter->intr;
1763         int err = 0, i;
1764         int vector = 0;
1765
1766 #ifdef CONFIG_PCI_MSI
1767         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1768                 for (i = 0; i < adapter->num_tx_queues; i++) {
1769                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1770                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1771                                         adapter->netdev->name, vector);
1772                                 err = request_irq(
1773                                               intr->msix_entries[vector].vector,
1774                                               vmxnet3_msix_tx, 0,
1775                                               adapter->tx_queue[i].name,
1776                                               &adapter->tx_queue[i]);
1777                         } else {
1778                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1779                                         adapter->netdev->name, vector);
1780                         }
1781                         if (err) {
1782                                 dev_err(&adapter->netdev->dev,
1783                                         "Failed to request irq for MSIX, %s, "
1784                                         "error %d\n",
1785                                         adapter->tx_queue[i].name, err);
1786                                 return err;
1787                         }
1788
1789                         /* Handle the case where only 1 MSIx was allocated for
1790                          * all tx queues */
1791                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1792                                 for (; i < adapter->num_tx_queues; i++)
1793                                         adapter->tx_queue[i].comp_ring.intr_idx
1794                                                                 = vector;
1795                                 vector++;
1796                                 break;
1797                         } else {
1798                                 adapter->tx_queue[i].comp_ring.intr_idx
1799                                                                 = vector++;
1800                         }
1801                 }
1802                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1803                         vector = 0;
1804
1805                 for (i = 0; i < adapter->num_rx_queues; i++) {
1806                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1807                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1808                                         adapter->netdev->name, vector);
1809                         else
1810                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1811                                         adapter->netdev->name, vector);
1812                         err = request_irq(intr->msix_entries[vector].vector,
1813                                           vmxnet3_msix_rx, 0,
1814                                           adapter->rx_queue[i].name,
1815                                           &(adapter->rx_queue[i]));
1816                         if (err) {
1817                                 printk(KERN_ERR "Failed to request irq for MSIX"
1818                                        ", %s, error %d\n",
1819                                        adapter->rx_queue[i].name, err);
1820                                 return err;
1821                         }
1822
1823                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1824                 }
1825
1826                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1827                         adapter->netdev->name, vector);
1828                 err = request_irq(intr->msix_entries[vector].vector,
1829                                   vmxnet3_msix_event, 0,
1830                                   intr->event_msi_vector_name, adapter->netdev);
1831                 intr->event_intr_idx = vector;
1832
1833         } else if (intr->type == VMXNET3_IT_MSI) {
1834                 adapter->num_rx_queues = 1;
1835                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1836                                   adapter->netdev->name, adapter->netdev);
1837         } else {
1838 #endif
1839                 adapter->num_rx_queues = 1;
1840                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1841                                   IRQF_SHARED, adapter->netdev->name,
1842                                   adapter->netdev);
1843 #ifdef CONFIG_PCI_MSI
1844         }
1845 #endif
1846         intr->num_intrs = vector + 1;
1847         if (err) {
1848                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1849                        ":%d\n", adapter->netdev->name, intr->type, err);
1850         } else {
1851                 /* Number of rx queues will not change after this */
1852                 for (i = 0; i < adapter->num_rx_queues; i++) {
1853                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1854                         rq->qid = i;
1855                         rq->qid2 = i + adapter->num_rx_queues;
1856                 }
1857
1858
1859
1860                 /* init our intr settings */
1861                 for (i = 0; i < intr->num_intrs; i++)
1862                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1863                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1864                         adapter->intr.event_intr_idx = 0;
1865                         for (i = 0; i < adapter->num_tx_queues; i++)
1866                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1867                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1868                 }
1869
1870                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1871                        "allocated\n", adapter->netdev->name, intr->type,
1872                        intr->mask_mode, intr->num_intrs);
1873         }
1874
1875         return err;
1876 }
1877
1878
1879 static void
1880 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1881 {
1882         struct vmxnet3_intr *intr = &adapter->intr;
1883         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1884
1885         switch (intr->type) {
1886 #ifdef CONFIG_PCI_MSI
1887         case VMXNET3_IT_MSIX:
1888         {
1889                 int i, vector = 0;
1890
1891                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1892                         for (i = 0; i < adapter->num_tx_queues; i++) {
1893                                 free_irq(intr->msix_entries[vector++].vector,
1894                                          &(adapter->tx_queue[i]));
1895                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1896                                         break;
1897                         }
1898                 }
1899
1900                 for (i = 0; i < adapter->num_rx_queues; i++) {
1901                         free_irq(intr->msix_entries[vector++].vector,
1902                                  &(adapter->rx_queue[i]));
1903                 }
1904
1905                 free_irq(intr->msix_entries[vector].vector,
1906                          adapter->netdev);
1907                 BUG_ON(vector >= intr->num_intrs);
1908                 break;
1909         }
1910 #endif
1911         case VMXNET3_IT_MSI:
1912                 free_irq(adapter->pdev->irq, adapter->netdev);
1913                 break;
1914         case VMXNET3_IT_INTX:
1915                 free_irq(adapter->pdev->irq, adapter->netdev);
1916                 break;
1917         default:
1918                 BUG();
1919         }
1920 }
1921
1922
1923 static void
1924 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1925 {
1926         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1927         u16 vid;
1928
1929         /* allow untagged pkts */
1930         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1931
1932         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1933                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1934 }
1935
1936
1937 static int
1938 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1939 {
1940         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1941
1942         if (!(netdev->flags & IFF_PROMISC)) {
1943                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1944                 unsigned long flags;
1945
1946                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1947                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1948                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1949                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1950                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1951         }
1952
1953         set_bit(vid, adapter->active_vlans);
1954
1955         return 0;
1956 }
1957
1958
1959 static int
1960 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1961 {
1962         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1963
1964         if (!(netdev->flags & IFF_PROMISC)) {
1965                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1966                 unsigned long flags;
1967
1968                 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1969                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1970                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1971                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1972                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1973         }
1974
1975         clear_bit(vid, adapter->active_vlans);
1976
1977         return 0;
1978 }
1979
1980
1981 static u8 *
1982 vmxnet3_copy_mc(struct net_device *netdev)
1983 {
1984         u8 *buf = NULL;
1985         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1986
1987         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1988         if (sz <= 0xffff) {
1989                 /* We may be called with BH disabled */
1990                 buf = kmalloc(sz, GFP_ATOMIC);
1991                 if (buf) {
1992                         struct netdev_hw_addr *ha;
1993                         int i = 0;
1994
1995                         netdev_for_each_mc_addr(ha, netdev)
1996                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
1997                                        ETH_ALEN);
1998                 }
1999         }
2000         return buf;
2001 }
2002
2003
2004 static void
2005 vmxnet3_set_mc(struct net_device *netdev)
2006 {
2007         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2008         unsigned long flags;
2009         struct Vmxnet3_RxFilterConf *rxConf =
2010                                         &adapter->shared->devRead.rxFilterConf;
2011         u8 *new_table = NULL;
2012         u32 new_mode = VMXNET3_RXM_UCAST;
2013
2014         if (netdev->flags & IFF_PROMISC) {
2015                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2016                 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2017
2018                 new_mode |= VMXNET3_RXM_PROMISC;
2019         } else {
2020                 vmxnet3_restore_vlan(adapter);
2021         }
2022
2023         if (netdev->flags & IFF_BROADCAST)
2024                 new_mode |= VMXNET3_RXM_BCAST;
2025
2026         if (netdev->flags & IFF_ALLMULTI)
2027                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2028         else
2029                 if (!netdev_mc_empty(netdev)) {
2030                         new_table = vmxnet3_copy_mc(netdev);
2031                         if (new_table) {
2032                                 new_mode |= VMXNET3_RXM_MCAST;
2033                                 rxConf->mfTableLen = cpu_to_le16(
2034                                         netdev_mc_count(netdev) * ETH_ALEN);
2035                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2036                                                     new_table));
2037                         } else {
2038                                 printk(KERN_INFO "%s: failed to copy mcast list"
2039                                        ", setting ALL_MULTI\n", netdev->name);
2040                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2041                         }
2042                 }
2043
2044
2045         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2046                 rxConf->mfTableLen = 0;
2047                 rxConf->mfTablePA = 0;
2048         }
2049
2050         spin_lock_irqsave(&adapter->cmd_lock, flags);
2051         if (new_mode != rxConf->rxMode) {
2052                 rxConf->rxMode = cpu_to_le32(new_mode);
2053                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2054                                        VMXNET3_CMD_UPDATE_RX_MODE);
2055                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2056                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2057         }
2058
2059         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2060                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2061         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2062
2063         kfree(new_table);
2064 }
2065
2066 void
2067 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2068 {
2069         int i;
2070
2071         for (i = 0; i < adapter->num_rx_queues; i++)
2072                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2073 }
2074
2075
2076 /*
2077  *   Set up driver_shared based on settings in adapter.
2078  */
2079
2080 static void
2081 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2082 {
2083         struct Vmxnet3_DriverShared *shared = adapter->shared;
2084         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2085         struct Vmxnet3_TxQueueConf *tqc;
2086         struct Vmxnet3_RxQueueConf *rqc;
2087         int i;
2088
2089         memset(shared, 0, sizeof(*shared));
2090
2091         /* driver settings */
2092         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2093         devRead->misc.driverInfo.version = cpu_to_le32(
2094                                                 VMXNET3_DRIVER_VERSION_NUM);
2095         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2096                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2097         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2098         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2099                                 *((u32 *)&devRead->misc.driverInfo.gos));
2100         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2101         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2102
2103         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2104         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2105
2106         /* set up feature flags */
2107         if (adapter->netdev->features & NETIF_F_RXCSUM)
2108                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2109
2110         if (adapter->netdev->features & NETIF_F_LRO) {
2111                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2112                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2113         }
2114         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2115                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2116
2117         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2118         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2119         devRead->misc.queueDescLen = cpu_to_le32(
2120                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2121                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2122
2123         /* tx queue settings */
2124         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2125         for (i = 0; i < adapter->num_tx_queues; i++) {
2126                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2127                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2128                 tqc = &adapter->tqd_start[i].conf;
2129                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2130                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2131                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2132                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2133                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2134                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2135                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2136                 tqc->ddLen          = cpu_to_le32(
2137                                         sizeof(struct vmxnet3_tx_buf_info) *
2138                                         tqc->txRingSize);
2139                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2140         }
2141
2142         /* rx queue settings */
2143         devRead->misc.numRxQueues = adapter->num_rx_queues;
2144         for (i = 0; i < adapter->num_rx_queues; i++) {
2145                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2146                 rqc = &adapter->rqd_start[i].conf;
2147                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2148                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2149                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2150                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2151                                                         rq->buf_info));
2152                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2153                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2154                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2155                 rqc->ddLen           = cpu_to_le32(
2156                                         sizeof(struct vmxnet3_rx_buf_info) *
2157                                         (rqc->rxRingSize[0] +
2158                                          rqc->rxRingSize[1]));
2159                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2160         }
2161
2162 #ifdef VMXNET3_RSS
2163         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2164
2165         if (adapter->rss) {
2166                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2167                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2168                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2169                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2170                                     UPT1_RSS_HASH_TYPE_IPV4 |
2171                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2172                                     UPT1_RSS_HASH_TYPE_IPV6;
2173                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2174                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2175                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2176                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2177                 for (i = 0; i < rssConf->indTableSize; i++)
2178                         rssConf->indTable[i] = ethtool_rxfh_indir_default(
2179                                 i, adapter->num_rx_queues);
2180
2181                 devRead->rssConfDesc.confVer = 1;
2182                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2183                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2184         }
2185
2186 #endif /* VMXNET3_RSS */
2187
2188         /* intr settings */
2189         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2190                                      VMXNET3_IMM_AUTO;
2191         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2192         for (i = 0; i < adapter->intr.num_intrs; i++)
2193                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2194
2195         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2196         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2197
2198         /* rx filter settings */
2199         devRead->rxFilterConf.rxMode = 0;
2200         vmxnet3_restore_vlan(adapter);
2201         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2202
2203         /* the rest are already zeroed */
2204 }
2205
2206
2207 int
2208 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2209 {
2210         int err, i;
2211         u32 ret;
2212         unsigned long flags;
2213
2214         netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2215                 " ring sizes %u %u %u\n", adapter->netdev->name,
2216                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2217                 adapter->tx_queue[0].tx_ring.size,
2218                 adapter->rx_queue[0].rx_ring[0].size,
2219                 adapter->rx_queue[0].rx_ring[1].size);
2220
2221         vmxnet3_tq_init_all(adapter);
2222         err = vmxnet3_rq_init_all(adapter);
2223         if (err) {
2224                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2225                        adapter->netdev->name, err);
2226                 goto rq_err;
2227         }
2228
2229         err = vmxnet3_request_irqs(adapter);
2230         if (err) {
2231                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2232                        adapter->netdev->name, err);
2233                 goto irq_err;
2234         }
2235
2236         vmxnet3_setup_driver_shared(adapter);
2237
2238         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2239                                adapter->shared_pa));
2240         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2241                                adapter->shared_pa));
2242         spin_lock_irqsave(&adapter->cmd_lock, flags);
2243         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2244                                VMXNET3_CMD_ACTIVATE_DEV);
2245         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2246         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2247
2248         if (ret != 0) {
2249                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2250                        adapter->netdev->name, ret);
2251                 err = -EINVAL;
2252                 goto activate_err;
2253         }
2254
2255         for (i = 0; i < adapter->num_rx_queues; i++) {
2256                 VMXNET3_WRITE_BAR0_REG(adapter,
2257                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2258                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2259                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2260                                 (i * VMXNET3_REG_ALIGN)),
2261                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2262         }
2263
2264         /* Apply the rx filter settins last. */
2265         vmxnet3_set_mc(adapter->netdev);
2266
2267         /*
2268          * Check link state when first activating device. It will start the
2269          * tx queue if the link is up.
2270          */
2271         vmxnet3_check_link(adapter, true);
2272         for (i = 0; i < adapter->num_rx_queues; i++)
2273                 napi_enable(&adapter->rx_queue[i].napi);
2274         vmxnet3_enable_all_intrs(adapter);
2275         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2276         return 0;
2277
2278 activate_err:
2279         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2280         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2281         vmxnet3_free_irqs(adapter);
2282 irq_err:
2283 rq_err:
2284         /* free up buffers we allocated */
2285         vmxnet3_rq_cleanup_all(adapter);
2286         return err;
2287 }
2288
2289
2290 void
2291 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2292 {
2293         unsigned long flags;
2294         spin_lock_irqsave(&adapter->cmd_lock, flags);
2295         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2296         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2297 }
2298
2299
2300 int
2301 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2302 {
2303         int i;
2304         unsigned long flags;
2305         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2306                 return 0;
2307
2308
2309         spin_lock_irqsave(&adapter->cmd_lock, flags);
2310         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2311                                VMXNET3_CMD_QUIESCE_DEV);
2312         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2313         vmxnet3_disable_all_intrs(adapter);
2314
2315         for (i = 0; i < adapter->num_rx_queues; i++)
2316                 napi_disable(&adapter->rx_queue[i].napi);
2317         netif_tx_disable(adapter->netdev);
2318         adapter->link_speed = 0;
2319         netif_carrier_off(adapter->netdev);
2320
2321         vmxnet3_tq_cleanup_all(adapter);
2322         vmxnet3_rq_cleanup_all(adapter);
2323         vmxnet3_free_irqs(adapter);
2324         return 0;
2325 }
2326
2327
2328 static void
2329 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2330 {
2331         u32 tmp;
2332
2333         tmp = *(u32 *)mac;
2334         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2335
2336         tmp = (mac[5] << 8) | mac[4];
2337         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2338 }
2339
2340
2341 static int
2342 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2343 {
2344         struct sockaddr *addr = p;
2345         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2346
2347         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2348         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2349
2350         return 0;
2351 }
2352
2353
2354 /* ==================== initialization and cleanup routines ============ */
2355
2356 static int
2357 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2358 {
2359         int err;
2360         unsigned long mmio_start, mmio_len;
2361         struct pci_dev *pdev = adapter->pdev;
2362
2363         err = pci_enable_device(pdev);
2364         if (err) {
2365                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2366                        pci_name(pdev), err);
2367                 return err;
2368         }
2369
2370         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2371                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2372                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2373                                "for adapter %s\n", pci_name(pdev));
2374                         err = -EIO;
2375                         goto err_set_mask;
2376                 }
2377                 *dma64 = true;
2378         } else {
2379                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2380                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2381                                "%s\n",  pci_name(pdev));
2382                         err = -EIO;
2383                         goto err_set_mask;
2384                 }
2385                 *dma64 = false;
2386         }
2387
2388         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2389                                            vmxnet3_driver_name);
2390         if (err) {
2391                 printk(KERN_ERR "Failed to request region for adapter %s: "
2392                        "error %d\n", pci_name(pdev), err);
2393                 goto err_set_mask;
2394         }
2395
2396         pci_set_master(pdev);
2397
2398         mmio_start = pci_resource_start(pdev, 0);
2399         mmio_len = pci_resource_len(pdev, 0);
2400         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2401         if (!adapter->hw_addr0) {
2402                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2403                        pci_name(pdev));
2404                 err = -EIO;
2405                 goto err_ioremap;
2406         }
2407
2408         mmio_start = pci_resource_start(pdev, 1);
2409         mmio_len = pci_resource_len(pdev, 1);
2410         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2411         if (!adapter->hw_addr1) {
2412                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2413                        pci_name(pdev));
2414                 err = -EIO;
2415                 goto err_bar1;
2416         }
2417         return 0;
2418
2419 err_bar1:
2420         iounmap(adapter->hw_addr0);
2421 err_ioremap:
2422         pci_release_selected_regions(pdev, (1 << 2) - 1);
2423 err_set_mask:
2424         pci_disable_device(pdev);
2425         return err;
2426 }
2427
2428
2429 static void
2430 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2431 {
2432         BUG_ON(!adapter->pdev);
2433
2434         iounmap(adapter->hw_addr0);
2435         iounmap(adapter->hw_addr1);
2436         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2437         pci_disable_device(adapter->pdev);
2438 }
2439
2440
2441 static void
2442 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2443 {
2444         size_t sz, i, ring0_size, ring1_size, comp_size;
2445         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2446
2447
2448         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2449                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2450                 adapter->skb_buf_size = adapter->netdev->mtu +
2451                                         VMXNET3_MAX_ETH_HDR_SIZE;
2452                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2453                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2454
2455                 adapter->rx_buf_per_pkt = 1;
2456         } else {
2457                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2458                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2459                                             VMXNET3_MAX_ETH_HDR_SIZE;
2460                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2461         }
2462
2463         /*
2464          * for simplicity, force the ring0 size to be a multiple of
2465          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2466          */
2467         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2468         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2469         ring0_size = (ring0_size + sz - 1) / sz * sz;
2470         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2471                            sz * sz);
2472         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2473         comp_size = ring0_size + ring1_size;
2474
2475         for (i = 0; i < adapter->num_rx_queues; i++) {
2476                 rq = &adapter->rx_queue[i];
2477                 rq->rx_ring[0].size = ring0_size;
2478                 rq->rx_ring[1].size = ring1_size;
2479                 rq->comp_ring.size = comp_size;
2480         }
2481 }
2482
2483
2484 int
2485 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2486                       u32 rx_ring_size, u32 rx_ring2_size)
2487 {
2488         int err = 0, i;
2489
2490         for (i = 0; i < adapter->num_tx_queues; i++) {
2491                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2492                 tq->tx_ring.size   = tx_ring_size;
2493                 tq->data_ring.size = tx_ring_size;
2494                 tq->comp_ring.size = tx_ring_size;
2495                 tq->shared = &adapter->tqd_start[i].ctrl;
2496                 tq->stopped = true;
2497                 tq->adapter = adapter;
2498                 tq->qid = i;
2499                 err = vmxnet3_tq_create(tq, adapter);
2500                 /*
2501                  * Too late to change num_tx_queues. We cannot do away with
2502                  * lesser number of queues than what we asked for
2503                  */
2504                 if (err)
2505                         goto queue_err;
2506         }
2507
2508         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2509         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2510         vmxnet3_adjust_rx_ring_size(adapter);
2511         for (i = 0; i < adapter->num_rx_queues; i++) {
2512                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2513                 /* qid and qid2 for rx queues will be assigned later when num
2514                  * of rx queues is finalized after allocating intrs */
2515                 rq->shared = &adapter->rqd_start[i].ctrl;
2516                 rq->adapter = adapter;
2517                 err = vmxnet3_rq_create(rq, adapter);
2518                 if (err) {
2519                         if (i == 0) {
2520                                 printk(KERN_ERR "Could not allocate any rx"
2521                                        "queues. Aborting.\n");
2522                                 goto queue_err;
2523                         } else {
2524                                 printk(KERN_INFO "Number of rx queues changed "
2525                                        "to : %d.\n", i);
2526                                 adapter->num_rx_queues = i;
2527                                 err = 0;
2528                                 break;
2529                         }
2530                 }
2531         }
2532         return err;
2533 queue_err:
2534         vmxnet3_tq_destroy_all(adapter);
2535         return err;
2536 }
2537
2538 static int
2539 vmxnet3_open(struct net_device *netdev)
2540 {
2541         struct vmxnet3_adapter *adapter;
2542         int err, i;
2543
2544         adapter = netdev_priv(netdev);
2545
2546         for (i = 0; i < adapter->num_tx_queues; i++)
2547                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2548
2549         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2550                                     VMXNET3_DEF_RX_RING_SIZE,
2551                                     VMXNET3_DEF_RX_RING_SIZE);
2552         if (err)
2553                 goto queue_err;
2554
2555         err = vmxnet3_activate_dev(adapter);
2556         if (err)
2557                 goto activate_err;
2558
2559         return 0;
2560
2561 activate_err:
2562         vmxnet3_rq_destroy_all(adapter);
2563         vmxnet3_tq_destroy_all(adapter);
2564 queue_err:
2565         return err;
2566 }
2567
2568
2569 static int
2570 vmxnet3_close(struct net_device *netdev)
2571 {
2572         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2573
2574         /*
2575          * Reset_work may be in the middle of resetting the device, wait for its
2576          * completion.
2577          */
2578         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2579                 msleep(1);
2580
2581         vmxnet3_quiesce_dev(adapter);
2582
2583         vmxnet3_rq_destroy_all(adapter);
2584         vmxnet3_tq_destroy_all(adapter);
2585
2586         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2587
2588
2589         return 0;
2590 }
2591
2592
2593 void
2594 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2595 {
2596         int i;
2597
2598         /*
2599          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2600          * vmxnet3_close() will deadlock.
2601          */
2602         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2603
2604         /* we need to enable NAPI, otherwise dev_close will deadlock */
2605         for (i = 0; i < adapter->num_rx_queues; i++)
2606                 napi_enable(&adapter->rx_queue[i].napi);
2607         dev_close(adapter->netdev);
2608 }
2609
2610
2611 static int
2612 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2613 {
2614         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2615         int err = 0;
2616
2617         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2618                 return -EINVAL;
2619
2620         netdev->mtu = new_mtu;
2621
2622         /*
2623          * Reset_work may be in the middle of resetting the device, wait for its
2624          * completion.
2625          */
2626         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2627                 msleep(1);
2628
2629         if (netif_running(netdev)) {
2630                 vmxnet3_quiesce_dev(adapter);
2631                 vmxnet3_reset_dev(adapter);
2632
2633                 /* we need to re-create the rx queue based on the new mtu */
2634                 vmxnet3_rq_destroy_all(adapter);
2635                 vmxnet3_adjust_rx_ring_size(adapter);
2636                 err = vmxnet3_rq_create_all(adapter);
2637                 if (err) {
2638                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2639                                 " error %d. Closing it.\n", netdev->name, err);
2640                         goto out;
2641                 }
2642
2643                 err = vmxnet3_activate_dev(adapter);
2644                 if (err) {
2645                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2646                                 "Closing it\n", netdev->name, err);
2647                         goto out;
2648                 }
2649         }
2650
2651 out:
2652         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2653         if (err)
2654                 vmxnet3_force_close(adapter);
2655
2656         return err;
2657 }
2658
2659
2660 static void
2661 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2662 {
2663         struct net_device *netdev = adapter->netdev;
2664
2665         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2666                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2667                 NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2668                 NETIF_F_LRO;
2669         if (dma64)
2670                 netdev->hw_features |= NETIF_F_HIGHDMA;
2671         netdev->vlan_features = netdev->hw_features &
2672                                 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
2673         netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER;
2674 }
2675
2676
2677 static void
2678 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2679 {
2680         u32 tmp;
2681
2682         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2683         *(u32 *)mac = tmp;
2684
2685         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2686         mac[4] = tmp & 0xff;
2687         mac[5] = (tmp >> 8) & 0xff;
2688 }
2689
2690 #ifdef CONFIG_PCI_MSI
2691
2692 /*
2693  * Enable MSIx vectors.
2694  * Returns :
2695  *      0 on successful enabling of required vectors,
2696  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2697  *       could be enabled.
2698  *      number of vectors which can be enabled otherwise (this number is smaller
2699  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2700  */
2701
2702 static int
2703 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2704                              int vectors)
2705 {
2706         int err = 0, vector_threshold;
2707         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2708
2709         while (vectors >= vector_threshold) {
2710                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2711                                       vectors);
2712                 if (!err) {
2713                         adapter->intr.num_intrs = vectors;
2714                         return 0;
2715                 } else if (err < 0) {
2716                         dev_err(&adapter->netdev->dev,
2717                                    "Failed to enable MSI-X, error: %d\n", err);
2718                         vectors = 0;
2719                 } else if (err < vector_threshold) {
2720                         break;
2721                 } else {
2722                         /* If fails to enable required number of MSI-x vectors
2723                          * try enabling minimum number of vectors required.
2724                          */
2725                         dev_err(&adapter->netdev->dev,
2726                                 "Failed to enable %d MSI-X, trying %d instead\n",
2727                                     vectors, vector_threshold);
2728                         vectors = vector_threshold;
2729                 }
2730         }
2731
2732         dev_info(&adapter->pdev->dev,
2733                  "Number of MSI-X interrupts which can be allocated "
2734                  "is lower than min threshold required.\n");
2735         return err;
2736 }
2737
2738
2739 #endif /* CONFIG_PCI_MSI */
2740
2741 static void
2742 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2743 {
2744         u32 cfg;
2745         unsigned long flags;
2746
2747         /* intr settings */
2748         spin_lock_irqsave(&adapter->cmd_lock, flags);
2749         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2750                                VMXNET3_CMD_GET_CONF_INTR);
2751         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2752         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2753         adapter->intr.type = cfg & 0x3;
2754         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2755
2756         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2757                 adapter->intr.type = VMXNET3_IT_MSIX;
2758         }
2759
2760 #ifdef CONFIG_PCI_MSI
2761         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2762                 int vector, err = 0;
2763
2764                 adapter->intr.num_intrs = (adapter->share_intr ==
2765                                            VMXNET3_INTR_TXSHARE) ? 1 :
2766                                            adapter->num_tx_queues;
2767                 adapter->intr.num_intrs += (adapter->share_intr ==
2768                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2769                                            adapter->num_rx_queues;
2770                 adapter->intr.num_intrs += 1;           /* for link event */
2771
2772                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2773                                            VMXNET3_LINUX_MIN_MSIX_VECT
2774                                            ? adapter->intr.num_intrs :
2775                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2776
2777                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2778                         adapter->intr.msix_entries[vector].entry = vector;
2779
2780                 err = vmxnet3_acquire_msix_vectors(adapter,
2781                                                    adapter->intr.num_intrs);
2782                 /* If we cannot allocate one MSIx vector per queue
2783                  * then limit the number of rx queues to 1
2784                  */
2785                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2786                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2787                             || adapter->num_rx_queues != 1) {
2788                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2789                                 printk(KERN_ERR "Number of rx queues : 1\n");
2790                                 adapter->num_rx_queues = 1;
2791                                 adapter->intr.num_intrs =
2792                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2793                         }
2794                         return;
2795                 }
2796                 if (!err)
2797                         return;
2798
2799                 /* If we cannot allocate MSIx vectors use only one rx queue */
2800                 dev_info(&adapter->pdev->dev,
2801                          "Failed to enable MSI-X, error %d. "
2802                          "Limiting #rx queues to 1, try MSI.\n", err);
2803
2804                 adapter->intr.type = VMXNET3_IT_MSI;
2805         }
2806
2807         if (adapter->intr.type == VMXNET3_IT_MSI) {
2808                 int err;
2809                 err = pci_enable_msi(adapter->pdev);
2810                 if (!err) {
2811                         adapter->num_rx_queues = 1;
2812                         adapter->intr.num_intrs = 1;
2813                         return;
2814                 }
2815         }
2816 #endif /* CONFIG_PCI_MSI */
2817
2818         adapter->num_rx_queues = 1;
2819         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2820         adapter->intr.type = VMXNET3_IT_INTX;
2821
2822         /* INT-X related setting */
2823         adapter->intr.num_intrs = 1;
2824 }
2825
2826
2827 static void
2828 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2829 {
2830         if (adapter->intr.type == VMXNET3_IT_MSIX)
2831                 pci_disable_msix(adapter->pdev);
2832         else if (adapter->intr.type == VMXNET3_IT_MSI)
2833                 pci_disable_msi(adapter->pdev);
2834         else
2835                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2836 }
2837
2838
2839 static void
2840 vmxnet3_tx_timeout(struct net_device *netdev)
2841 {
2842         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2843         adapter->tx_timeout_count++;
2844
2845         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2846         schedule_work(&adapter->work);
2847         netif_wake_queue(adapter->netdev);
2848 }
2849
2850
2851 static void
2852 vmxnet3_reset_work(struct work_struct *data)
2853 {
2854         struct vmxnet3_adapter *adapter;
2855
2856         adapter = container_of(data, struct vmxnet3_adapter, work);
2857
2858         /* if another thread is resetting the device, no need to proceed */
2859         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2860                 return;
2861
2862         /* if the device is closed, we must leave it alone */
2863         rtnl_lock();
2864         if (netif_running(adapter->netdev)) {
2865                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2866                 vmxnet3_quiesce_dev(adapter);
2867                 vmxnet3_reset_dev(adapter);
2868                 vmxnet3_activate_dev(adapter);
2869         } else {
2870                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2871         }
2872         rtnl_unlock();
2873
2874         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2875 }
2876
2877
2878 static int
2879 vmxnet3_probe_device(struct pci_dev *pdev,
2880                      const struct pci_device_id *id)
2881 {
2882         static const struct net_device_ops vmxnet3_netdev_ops = {
2883                 .ndo_open = vmxnet3_open,
2884                 .ndo_stop = vmxnet3_close,
2885                 .ndo_start_xmit = vmxnet3_xmit_frame,
2886                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2887                 .ndo_change_mtu = vmxnet3_change_mtu,
2888                 .ndo_set_features = vmxnet3_set_features,
2889                 .ndo_get_stats64 = vmxnet3_get_stats64,
2890                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2891                 .ndo_set_rx_mode = vmxnet3_set_mc,
2892                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2893                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2894 #ifdef CONFIG_NET_POLL_CONTROLLER
2895                 .ndo_poll_controller = vmxnet3_netpoll,
2896 #endif
2897         };
2898         int err;
2899         bool dma64 = false; /* stupid gcc */
2900         u32 ver;
2901         struct net_device *netdev;
2902         struct vmxnet3_adapter *adapter;
2903         u8 mac[ETH_ALEN];
2904         int size;
2905         int num_tx_queues;
2906         int num_rx_queues;
2907
2908         if (!pci_msi_enabled())
2909                 enable_mq = 0;
2910
2911 #ifdef VMXNET3_RSS
2912         if (enable_mq)
2913                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2914                                     (int)num_online_cpus());
2915         else
2916 #endif
2917                 num_rx_queues = 1;
2918         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2919
2920         if (enable_mq)
2921                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2922                                     (int)num_online_cpus());
2923         else
2924                 num_tx_queues = 1;
2925
2926         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2927         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2928                                    max(num_tx_queues, num_rx_queues));
2929         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2930                num_tx_queues, num_rx_queues);
2931
2932         if (!netdev)
2933                 return -ENOMEM;
2934
2935         pci_set_drvdata(pdev, netdev);
2936         adapter = netdev_priv(netdev);
2937         adapter->netdev = netdev;
2938         adapter->pdev = pdev;
2939
2940         spin_lock_init(&adapter->cmd_lock);
2941         adapter->shared = pci_alloc_consistent(adapter->pdev,
2942                                                sizeof(struct Vmxnet3_DriverShared),
2943                                                &adapter->shared_pa);
2944         if (!adapter->shared) {
2945                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2946                        pci_name(pdev));
2947                 err = -ENOMEM;
2948                 goto err_alloc_shared;
2949         }
2950
2951         adapter->num_rx_queues = num_rx_queues;
2952         adapter->num_tx_queues = num_tx_queues;
2953
2954         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2955         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2956         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2957                                                   &adapter->queue_desc_pa);
2958
2959         if (!adapter->tqd_start) {
2960                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2961                        pci_name(pdev));
2962                 err = -ENOMEM;
2963                 goto err_alloc_queue_desc;
2964         }
2965         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2966                                                             adapter->num_tx_queues);
2967
2968         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2969         if (adapter->pm_conf == NULL) {
2970                 err = -ENOMEM;
2971                 goto err_alloc_pm;
2972         }
2973
2974 #ifdef VMXNET3_RSS
2975
2976         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2977         if (adapter->rss_conf == NULL) {
2978                 err = -ENOMEM;
2979                 goto err_alloc_rss;
2980         }
2981 #endif /* VMXNET3_RSS */
2982
2983         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2984         if (err < 0)
2985                 goto err_alloc_pci;
2986
2987         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2988         if (ver & 1) {
2989                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2990         } else {
2991                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
2992                        " %s\n", ver, pci_name(pdev));
2993                 err = -EBUSY;
2994                 goto err_ver;
2995         }
2996
2997         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
2998         if (ver & 1) {
2999                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3000         } else {
3001                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
3002                        "adapter %s\n", ver, pci_name(pdev));
3003                 err = -EBUSY;
3004                 goto err_ver;
3005         }
3006
3007         SET_NETDEV_DEV(netdev, &pdev->dev);
3008         vmxnet3_declare_features(adapter, dma64);
3009
3010         adapter->dev_number = atomic_read(&devices_found);
3011
3012         adapter->share_intr = irq_share_mode;
3013         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3014             adapter->num_tx_queues != adapter->num_rx_queues)
3015                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3016
3017         vmxnet3_alloc_intr_resources(adapter);
3018
3019 #ifdef VMXNET3_RSS
3020         if (adapter->num_rx_queues > 1 &&
3021             adapter->intr.type == VMXNET3_IT_MSIX) {
3022                 adapter->rss = true;
3023                 printk(KERN_INFO "RSS is enabled.\n");
3024         } else {
3025                 adapter->rss = false;
3026         }
3027 #endif
3028
3029         vmxnet3_read_mac_addr(adapter, mac);
3030         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3031
3032         netdev->netdev_ops = &vmxnet3_netdev_ops;
3033         vmxnet3_set_ethtool_ops(netdev);
3034         netdev->watchdog_timeo = 5 * HZ;
3035
3036         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3037         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3038
3039         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3040                 int i;
3041                 for (i = 0; i < adapter->num_rx_queues; i++) {
3042                         netif_napi_add(adapter->netdev,
3043                                        &adapter->rx_queue[i].napi,
3044                                        vmxnet3_poll_rx_only, 64);
3045                 }
3046         } else {
3047                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3048                                vmxnet3_poll, 64);
3049         }
3050
3051         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3052         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3053
3054         err = register_netdev(netdev);
3055
3056         if (err) {
3057                 printk(KERN_ERR "Failed to register adapter %s\n",
3058                        pci_name(pdev));
3059                 goto err_register;
3060         }
3061
3062         vmxnet3_check_link(adapter, false);
3063         atomic_inc(&devices_found);
3064         return 0;
3065
3066 err_register:
3067         vmxnet3_free_intr_resources(adapter);
3068 err_ver:
3069         vmxnet3_free_pci_resources(adapter);
3070 err_alloc_pci:
3071 #ifdef VMXNET3_RSS
3072         kfree(adapter->rss_conf);
3073 err_alloc_rss:
3074 #endif
3075         kfree(adapter->pm_conf);
3076 err_alloc_pm:
3077         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3078                             adapter->queue_desc_pa);
3079 err_alloc_queue_desc:
3080         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3081                             adapter->shared, adapter->shared_pa);
3082 err_alloc_shared:
3083         pci_set_drvdata(pdev, NULL);
3084         free_netdev(netdev);
3085         return err;
3086 }
3087
3088
3089 static void
3090 vmxnet3_remove_device(struct pci_dev *pdev)
3091 {
3092         struct net_device *netdev = pci_get_drvdata(pdev);
3093         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3094         int size = 0;
3095         int num_rx_queues;
3096
3097 #ifdef VMXNET3_RSS
3098         if (enable_mq)
3099                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3100                                     (int)num_online_cpus());
3101         else
3102 #endif
3103                 num_rx_queues = 1;
3104         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3105
3106         cancel_work_sync(&adapter->work);
3107
3108         unregister_netdev(netdev);
3109
3110         vmxnet3_free_intr_resources(adapter);
3111         vmxnet3_free_pci_resources(adapter);
3112 #ifdef VMXNET3_RSS
3113         kfree(adapter->rss_conf);
3114 #endif
3115         kfree(adapter->pm_conf);
3116
3117         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3118         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3119         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3120                             adapter->queue_desc_pa);
3121         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3122                             adapter->shared, adapter->shared_pa);
3123         free_netdev(netdev);
3124 }
3125
3126
3127 #ifdef CONFIG_PM
3128
3129 static int
3130 vmxnet3_suspend(struct device *device)
3131 {
3132         struct pci_dev *pdev = to_pci_dev(device);
3133         struct net_device *netdev = pci_get_drvdata(pdev);
3134         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3135         struct Vmxnet3_PMConf *pmConf;
3136         struct ethhdr *ehdr;
3137         struct arphdr *ahdr;
3138         u8 *arpreq;
3139         struct in_device *in_dev;
3140         struct in_ifaddr *ifa;
3141         unsigned long flags;
3142         int i = 0;
3143
3144         if (!netif_running(netdev))
3145                 return 0;
3146
3147         for (i = 0; i < adapter->num_rx_queues; i++)
3148                 napi_disable(&adapter->rx_queue[i].napi);
3149
3150         vmxnet3_disable_all_intrs(adapter);
3151         vmxnet3_free_irqs(adapter);
3152         vmxnet3_free_intr_resources(adapter);
3153
3154         netif_device_detach(netdev);
3155         netif_tx_stop_all_queues(netdev);
3156
3157         /* Create wake-up filters. */
3158         pmConf = adapter->pm_conf;
3159         memset(pmConf, 0, sizeof(*pmConf));
3160
3161         if (adapter->wol & WAKE_UCAST) {
3162                 pmConf->filters[i].patternSize = ETH_ALEN;
3163                 pmConf->filters[i].maskSize = 1;
3164                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3165                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3166
3167                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3168                 i++;
3169         }
3170
3171         if (adapter->wol & WAKE_ARP) {
3172                 in_dev = in_dev_get(netdev);
3173                 if (!in_dev)
3174                         goto skip_arp;
3175
3176                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3177                 if (!ifa)
3178                         goto skip_arp;
3179
3180                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3181                         sizeof(struct arphdr) +         /* ARP header */
3182                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3183                         2 * sizeof(u32);        /*2 IPv4 addresses */
3184                 pmConf->filters[i].maskSize =
3185                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3186
3187                 /* ETH_P_ARP in Ethernet header. */
3188                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3189                 ehdr->h_proto = htons(ETH_P_ARP);
3190
3191                 /* ARPOP_REQUEST in ARP header. */
3192                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3193                 ahdr->ar_op = htons(ARPOP_REQUEST);
3194                 arpreq = (u8 *)(ahdr + 1);
3195
3196                 /* The Unicast IPv4 address in 'tip' field. */
3197                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3198                 *(u32 *)arpreq = ifa->ifa_address;
3199
3200                 /* The mask for the relevant bits. */
3201                 pmConf->filters[i].mask[0] = 0x00;
3202                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3203                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3204                 pmConf->filters[i].mask[3] = 0x00;
3205                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3206                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3207                 in_dev_put(in_dev);
3208
3209                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3210                 i++;
3211         }
3212
3213 skip_arp:
3214         if (adapter->wol & WAKE_MAGIC)
3215                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3216
3217         pmConf->numFilters = i;
3218
3219         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3220         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3221                                                                   *pmConf));
3222         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3223                                                                  pmConf));
3224
3225         spin_lock_irqsave(&adapter->cmd_lock, flags);
3226         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3227                                VMXNET3_CMD_UPDATE_PMCFG);
3228         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3229
3230         pci_save_state(pdev);
3231         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3232                         adapter->wol);
3233         pci_disable_device(pdev);
3234         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3235
3236         return 0;
3237 }
3238
3239
3240 static int
3241 vmxnet3_resume(struct device *device)
3242 {
3243         int err, i = 0;
3244         unsigned long flags;
3245         struct pci_dev *pdev = to_pci_dev(device);
3246         struct net_device *netdev = pci_get_drvdata(pdev);
3247         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3248         struct Vmxnet3_PMConf *pmConf;
3249
3250         if (!netif_running(netdev))
3251                 return 0;
3252
3253         /* Destroy wake-up filters. */
3254         pmConf = adapter->pm_conf;
3255         memset(pmConf, 0, sizeof(*pmConf));
3256
3257         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3258         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3259                                                                   *pmConf));
3260         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3261                                                                  pmConf));
3262
3263         netif_device_attach(netdev);
3264         pci_set_power_state(pdev, PCI_D0);
3265         pci_restore_state(pdev);
3266         err = pci_enable_device_mem(pdev);
3267         if (err != 0)
3268                 return err;
3269
3270         pci_enable_wake(pdev, PCI_D0, 0);
3271
3272         spin_lock_irqsave(&adapter->cmd_lock, flags);
3273         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3274                                VMXNET3_CMD_UPDATE_PMCFG);
3275         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3276         vmxnet3_alloc_intr_resources(adapter);
3277         vmxnet3_request_irqs(adapter);
3278         for (i = 0; i < adapter->num_rx_queues; i++)
3279                 napi_enable(&adapter->rx_queue[i].napi);
3280         vmxnet3_enable_all_intrs(adapter);
3281
3282         return 0;
3283 }
3284
3285 static const struct dev_pm_ops vmxnet3_pm_ops = {
3286         .suspend = vmxnet3_suspend,
3287         .resume = vmxnet3_resume,
3288 };
3289 #endif
3290
3291 static struct pci_driver vmxnet3_driver = {
3292         .name           = vmxnet3_driver_name,
3293         .id_table       = vmxnet3_pciid_table,
3294         .probe          = vmxnet3_probe_device,
3295         .remove         = vmxnet3_remove_device,
3296 #ifdef CONFIG_PM
3297         .driver.pm      = &vmxnet3_pm_ops,
3298 #endif
3299 };
3300
3301
3302 static int __init
3303 vmxnet3_init_module(void)
3304 {
3305         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3306                 VMXNET3_DRIVER_VERSION_REPORT);
3307         return pci_register_driver(&vmxnet3_driver);
3308 }
3309
3310 module_init(vmxnet3_init_module);
3311
3312
3313 static void
3314 vmxnet3_exit_module(void)
3315 {
3316         pci_unregister_driver(&vmxnet3_driver);
3317 }
3318
3319 module_exit(vmxnet3_exit_module);
3320
3321 MODULE_AUTHOR("VMware, Inc.");
3322 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3323 MODULE_LICENSE("GPL v2");
3324 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);