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 = dev_alloc_skb(rbi->len +
 584                                                          NET_IP_ALIGN);
 585                                 if (unlikely(rbi->skb == NULL)) {
 586                                         rq->stats.rx_buf_alloc_failure++;
 587                                         break;
 588                                 }
 589                                 rbi->skb->dev = adapter->netdev;
 590
 591                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 592                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 593                                                 rbi->skb->data, rbi->len,
 594                                                 PCI_DMA_FROMDEVICE);
 595                         } else {
 596                                 /* rx buffer skipped by the device */
 597                         }
 598                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 599                 } else {
 600                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 601                                rbi->len  != PAGE_SIZE);
 602
 603                         if (rbi->page == NULL) {
 604                                 rbi->page = alloc_page(GFP_ATOMIC);
 605                                 if (unlikely(rbi->page == NULL)) {
 606                                         rq->stats.rx_buf_alloc_failure++;
 607                                         break;
 608                                 }
 609                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 610                                                 rbi->page, 0, PAGE_SIZE,
 611                                                 PCI_DMA_FROMDEVICE);
 612                         } else {
 613                                 /* rx buffers skipped by the device */
 614                         }
 615                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 616                 }
 617
 618                 BUG_ON(rbi->dma_addr == 0);
 619                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 620                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 621                                            | val | rbi->len);
 622
 623                 /* Fill the last buffer but dont mark it ready, or else the
 624                  * device will think that the queue is full */
 625                 if (num_allocated == num_to_alloc)
 626                         break;
 627
 628                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 629                 num_allocated++;
 630                 vmxnet3_cmd_ring_adv_next2fill(ring);
 631         }
 632         rq->uncommitted[ring_idx] += num_allocated;
 633
 634         dev_dbg(&adapter->netdev->dev,
 635                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 636                 "%u, uncommitted %u\n", num_allocated, ring->next2fill,
 637                 ring->next2comp, rq->uncommitted[ring_idx]);
 638
 639         /* so that the device can distinguish a full ring and an empty ring */
 640         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 641
 642         return num_allocated;
 643 }
 644
 645
 646 static void
 647 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 648                     struct vmxnet3_rx_buf_info *rbi)
 649 {
 650         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 651                 skb_shinfo(skb)->nr_frags;
 652
 653         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 654
 655         __skb_frag_set_page(frag, rbi->page);
 656         frag->page_offset = 0;
 657         skb_frag_size_set(frag, rcd->len);
 658         skb->data_len += rcd->len;
 659         skb->truesize += PAGE_SIZE;
 660         skb_shinfo(skb)->nr_frags++;
 661 }
 662
 663
 664 static void
 665 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 666                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 667                 struct vmxnet3_adapter *adapter)
 668 {
 669         u32 dw2, len;
 670         unsigned long buf_offset;
 671         int i;
 672         union Vmxnet3_GenericDesc *gdesc;
 673         struct vmxnet3_tx_buf_info *tbi = NULL;
 674
 675         BUG_ON(ctx->copy_size > skb_headlen(skb));
 676
 677         /* use the previous gen bit for the SOP desc */
 678         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 679
 680         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 681         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 682
 683         /* no need to map the buffer if headers are copied */
 684         if (ctx->copy_size) {
 685                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 686                                         tq->tx_ring.next2fill *
 687                                         sizeof(struct Vmxnet3_TxDataDesc));
 688                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 689                 ctx->sop_txd->dword[3] = 0;
 690
 691                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 692                 tbi->map_type = VMXNET3_MAP_NONE;
 693
 694                 dev_dbg(&adapter->netdev->dev,
 695                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 696                         tq->tx_ring.next2fill,
 697                         le64_to_cpu(ctx->sop_txd->txd.addr),
 698                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 699                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 700
 701                 /* use the right gen for non-SOP desc */
 702                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 703         }
 704
 705         /* linear part can use multiple tx desc if it's big */
 706         len = skb_headlen(skb) - ctx->copy_size;
 707         buf_offset = ctx->copy_size;
 708         while (len) {
 709                 u32 buf_size;
 710
 711                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 712                         buf_size = len;
 713                         dw2 |= len;
 714                 } else {
 715                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 716                         /* spec says that for TxDesc.len, 0 == 2^14 */
 717                 }
 718
 719                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 720                 tbi->map_type = VMXNET3_MAP_SINGLE;
 721                 tbi->dma_addr = pci_map_single(adapter->pdev,
 722                                 skb->data + buf_offset, buf_size,
 723                                 PCI_DMA_TODEVICE);
 724
 725                 tbi->len = buf_size;
 726
 727                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 728                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 729
 730                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 731                 gdesc->dword[2] = cpu_to_le32(dw2);
 732                 gdesc->dword[3] = 0;
 733
 734                 dev_dbg(&adapter->netdev->dev,
 735                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 736                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 737                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 738                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 739                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 740
 741                 len -= buf_size;
 742                 buf_offset += buf_size;
 743         }
 744
 745         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 746                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 747                 u32 buf_size;
 748
 749                 buf_offset = 0;
 750                 len = skb_frag_size(frag);
 751                 while (len) {
 752                         tbi = tq->buf_info + tq->tx_ring.next2fill;
 753                         if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 754                                 buf_size = len;
 755                                 dw2 |= len;
 756                         } else {
 757                                 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 758                                 /* spec says that for TxDesc.len, 0 == 2^14 */
 759                         }
 760                         tbi->map_type = VMXNET3_MAP_PAGE;
 761                         tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
 762                                                          buf_offset, buf_size,
 763                                                          DMA_TO_DEVICE);
 764
 765                         tbi->len = buf_size;
 766
 767                         gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 768                         BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 769
 770                         gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 771                         gdesc->dword[2] = cpu_to_le32(dw2);
 772                         gdesc->dword[3] = 0;
 773
 774                         dev_dbg(&adapter->netdev->dev,
 775                                 "txd[%u]: 0x%llu %u %u\n",
 776                                 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 777                                 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 778                         vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 779                         dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 780
 781                         len -= buf_size;
 782                         buf_offset += buf_size;
 783                 }
 784         }
 785
 786         ctx->eop_txd = gdesc;
 787
 788         /* set the last buf_info for the pkt */
 789         tbi->skb = skb;
 790         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 791 }
 792
 793
 794 /* Init all tx queues */
 795 static void
 796 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 797 {
 798         int i;
 799
 800         for (i = 0; i < adapter->num_tx_queues; i++)
 801                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 802 }
 803
 804
 805 /*
 806  *    parse and copy relevant protocol headers:
 807  *      For a tso pkt, relevant headers are L2/3/4 including options
 808  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 809  *      if it's a TCP/UDP pkt
 810  *
 811  * Returns:
 812  *    -1:  error happens during parsing
 813  *     0:  protocol headers parsed, but too big to be copied
 814  *     1:  protocol headers parsed and copied
 815  *
 816  * Other effects:
 817  *    1. related *ctx fields are updated.
 818  *    2. ctx->copy_size is # of bytes copied
 819  *    3. the portion copied is guaranteed to be in the linear part
 820  *
 821  */
 822 static int
 823 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 824                            struct vmxnet3_tx_ctx *ctx,
 825                            struct vmxnet3_adapter *adapter)
 826 {
 827         struct Vmxnet3_TxDataDesc *tdd;
 828
 829         if (ctx->mss) { /* TSO */
 830                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 831                 ctx->l4_hdr_size = tcp_hdrlen(skb);
 832                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 833         } else {
 834                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 835                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 836
 837                         if (ctx->ipv4) {
 838                                 const struct iphdr *iph = ip_hdr(skb);
 839
 840                                 if (iph->protocol == IPPROTO_TCP)
 841                                         ctx->l4_hdr_size = tcp_hdrlen(skb);
 842                                 else if (iph->protocol == IPPROTO_UDP)
 843                                         ctx->l4_hdr_size = sizeof(struct udphdr);
 844                                 else
 845                                         ctx->l4_hdr_size = 0;
 846                         } else {
 847                                 /* for simplicity, don't copy L4 headers */
 848                                 ctx->l4_hdr_size = 0;
 849                         }
 850                         ctx->copy_size = min(ctx->eth_ip_hdr_size +
 851                                          ctx->l4_hdr_size, skb->len);
 852                 } else {
 853                         ctx->eth_ip_hdr_size = 0;
 854                         ctx->l4_hdr_size = 0;
 855                         /* copy as much as allowed */
 856                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 857                                              , skb_headlen(skb));
 858                 }
 859
 860                 /* make sure headers are accessible directly */
 861                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 862                         goto err;
 863         }
 864
 865         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 866                 tq->stats.oversized_hdr++;
 867                 ctx->copy_size = 0;
 868                 return 0;
 869         }
 870
 871         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 872
 873         memcpy(tdd->data, skb->data, ctx->copy_size);
 874         dev_dbg(&adapter->netdev->dev,
 875                 "copy %u bytes to dataRing[%u]\n",
 876                 ctx->copy_size, tq->tx_ring.next2fill);
 877         return 1;
 878
 879 err:
 880         return -1;
 881 }
 882
 883
 884 static void
 885 vmxnet3_prepare_tso(struct sk_buff *skb,
 886                     struct vmxnet3_tx_ctx *ctx)
 887 {
 888         struct tcphdr *tcph = tcp_hdr(skb);
 889
 890         if (ctx->ipv4) {
 891                 struct iphdr *iph = ip_hdr(skb);
 892
 893                 iph->check = 0;
 894                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 895                                                  IPPROTO_TCP, 0);
 896         } else {
 897                 struct ipv6hdr *iph = ipv6_hdr(skb);
 898
 899                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 900                                                IPPROTO_TCP, 0);
 901         }
 902 }
 903
 904 static int txd_estimate(const struct sk_buff *skb)
 905 {
 906         int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 907         int i;
 908
 909         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 910                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 911
 912                 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
 913         }
 914         return count;
 915 }
 916
 917 /*
 918  * Transmits a pkt thru a given tq
 919  * Returns:
 920  *    NETDEV_TX_OK:      descriptors are setup successfully
 921  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 922  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 923  *
 924  * Side-effects:
 925  *    1. tx ring may be changed
 926  *    2. tq stats may be updated accordingly
 927  *    3. shared->txNumDeferred may be updated
 928  */
 929
 930 static int
 931 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 932                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 933 {
 934         int ret;
 935         u32 count;
 936         unsigned long flags;
 937         struct vmxnet3_tx_ctx ctx;
 938         union Vmxnet3_GenericDesc *gdesc;
 939 #ifdef __BIG_ENDIAN_BITFIELD
 940         /* Use temporary descriptor to avoid touching bits multiple times */
 941         union Vmxnet3_GenericDesc tempTxDesc;
 942 #endif
 943
 944         count = txd_estimate(skb);
 945
 946         ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
 947
 948         ctx.mss = skb_shinfo(skb)->gso_size;
 949         if (ctx.mss) {
 950                 if (skb_header_cloned(skb)) {
 951                         if (unlikely(pskb_expand_head(skb, 0, 0,
 952                                                       GFP_ATOMIC) != 0)) {
 953                                 tq->stats.drop_tso++;
 954                                 goto drop_pkt;
 955                         }
 956                         tq->stats.copy_skb_header++;
 957                 }
 958                 vmxnet3_prepare_tso(skb, &ctx);
 959         } else {
 960                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 961
 962                         /* non-tso pkts must not use more than
 963                          * VMXNET3_MAX_TXD_PER_PKT entries
 964                          */
 965                         if (skb_linearize(skb) != 0) {
 966                                 tq->stats.drop_too_many_frags++;
 967                                 goto drop_pkt;
 968                         }
 969                         tq->stats.linearized++;
 970
 971                         /* recalculate the # of descriptors to use */
 972                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 973                 }
 974         }
 975
 976         spin_lock_irqsave(&tq->tx_lock, flags);
 977
 978         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 979                 tq->stats.tx_ring_full++;
 980                 dev_dbg(&adapter->netdev->dev,
 981                         "tx queue stopped on %s, next2comp %u"
 982                         " next2fill %u\n", adapter->netdev->name,
 983                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 984
 985                 vmxnet3_tq_stop(tq, adapter);
 986                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 987                 return NETDEV_TX_BUSY;
 988         }
 989
 990
 991         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 992         if (ret >= 0) {
 993                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 994                 /* hdrs parsed, check against other limits */
 995                 if (ctx.mss) {
 996                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 997                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 998                                 goto hdr_too_big;
 999                         }
1000                 } else {
1001                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1002                                 if (unlikely(ctx.eth_ip_hdr_size +
1003                                              skb->csum_offset >
1004                                              VMXNET3_MAX_CSUM_OFFSET)) {
1005                                         goto hdr_too_big;
1006                                 }
1007                         }
1008                 }
1009         } else {
1010                 tq->stats.drop_hdr_inspect_err++;
1011                 goto unlock_drop_pkt;
1012         }
1013
1014         /* fill tx descs related to addr & len */
1015         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
1016
1017         /* setup the EOP desc */
1018         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1019
1020         /* setup the SOP desc */
1021 #ifdef __BIG_ENDIAN_BITFIELD
1022         gdesc = &tempTxDesc;
1023         gdesc->dword[2] = ctx.sop_txd->dword[2];
1024         gdesc->dword[3] = ctx.sop_txd->dword[3];
1025 #else
1026         gdesc = ctx.sop_txd;
1027 #endif
1028         if (ctx.mss) {
1029                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1030                 gdesc->txd.om = VMXNET3_OM_TSO;
1031                 gdesc->txd.msscof = ctx.mss;
1032                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1033                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1034         } else {
1035                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1036                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1037                         gdesc->txd.om = VMXNET3_OM_CSUM;
1038                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1039                                             skb->csum_offset;
1040                 } else {
1041                         gdesc->txd.om = 0;
1042                         gdesc->txd.msscof = 0;
1043                 }
1044                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1045         }
1046
1047         if (vlan_tx_tag_present(skb)) {
1048                 gdesc->txd.ti = 1;
1049                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1050         }
1051
1052         /* finally flips the GEN bit of the SOP desc. */
1053         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1054                                                   VMXNET3_TXD_GEN);
1055 #ifdef __BIG_ENDIAN_BITFIELD
1056         /* Finished updating in bitfields of Tx Desc, so write them in original
1057          * place.
1058          */
1059         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1060                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1061         gdesc = ctx.sop_txd;
1062 #endif
1063         dev_dbg(&adapter->netdev->dev,
1064                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1065                 (u32)(ctx.sop_txd -
1066                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1067                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1068
1069         spin_unlock_irqrestore(&tq->tx_lock, flags);
1070
1071         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1072                                         le32_to_cpu(tq->shared->txThreshold)) {
1073                 tq->shared->txNumDeferred = 0;
1074                 VMXNET3_WRITE_BAR0_REG(adapter,
1075                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1076                                        tq->tx_ring.next2fill);
1077         }
1078
1079         return NETDEV_TX_OK;
1080
1081 hdr_too_big:
1082         tq->stats.drop_oversized_hdr++;
1083 unlock_drop_pkt:
1084         spin_unlock_irqrestore(&tq->tx_lock, flags);
1085 drop_pkt:
1086         tq->stats.drop_total++;
1087         dev_kfree_skb(skb);
1088         return NETDEV_TX_OK;
1089 }
1090
1091
1092 static netdev_tx_t
1093 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1094 {
1095         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1096
1097                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1098                 return vmxnet3_tq_xmit(skb,
1099                                        &adapter->tx_queue[skb->queue_mapping],
1100                                        adapter, netdev);
1101 }
1102
1103
1104 static void
1105 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1106                 struct sk_buff *skb,
1107                 union Vmxnet3_GenericDesc *gdesc)
1108 {
1109         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1110                 /* typical case: TCP/UDP over IP and both csums are correct */
1111                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1112                                                         VMXNET3_RCD_CSUM_OK) {
1113                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1114                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1115                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1116                         BUG_ON(gdesc->rcd.frg);
1117                 } else {
1118                         if (gdesc->rcd.csum) {
1119                                 skb->csum = htons(gdesc->rcd.csum);
1120                                 skb->ip_summed = CHECKSUM_PARTIAL;
1121                         } else {
1122                                 skb_checksum_none_assert(skb);
1123                         }
1124                 }
1125         } else {
1126                 skb_checksum_none_assert(skb);
1127         }
1128 }
1129
1130
1131 static void
1132 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1133                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1134 {
1135         rq->stats.drop_err++;
1136         if (!rcd->fcs)
1137                 rq->stats.drop_fcs++;
1138
1139         rq->stats.drop_total++;
1140
1141         /*
1142          * We do not unmap and chain the rx buffer to the skb.
1143          * We basically pretend this buffer is not used and will be recycled
1144          * by vmxnet3_rq_alloc_rx_buf()
1145          */
1146
1147         /*
1148          * ctx->skb may be NULL if this is the first and the only one
1149          * desc for the pkt
1150          */
1151         if (ctx->skb)
1152                 dev_kfree_skb_irq(ctx->skb);
1153
1154         ctx->skb = NULL;
1155 }
1156
1157
1158 static int
1159 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1160                        struct vmxnet3_adapter *adapter, int quota)
1161 {
1162         static const u32 rxprod_reg[2] = {
1163                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1164         };
1165         u32 num_rxd = 0;
1166         bool skip_page_frags = false;
1167         struct Vmxnet3_RxCompDesc *rcd;
1168         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1169 #ifdef __BIG_ENDIAN_BITFIELD
1170         struct Vmxnet3_RxDesc rxCmdDesc;
1171         struct Vmxnet3_RxCompDesc rxComp;
1172 #endif
1173         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1174                           &rxComp);
1175         while (rcd->gen == rq->comp_ring.gen) {
1176                 struct vmxnet3_rx_buf_info *rbi;
1177                 struct sk_buff *skb, *new_skb = NULL;
1178                 struct page *new_page = NULL;
1179                 int num_to_alloc;
1180                 struct Vmxnet3_RxDesc *rxd;
1181                 u32 idx, ring_idx;
1182                 struct vmxnet3_cmd_ring *ring = NULL;
1183                 if (num_rxd >= quota) {
1184                         /* we may stop even before we see the EOP desc of
1185                          * the current pkt
1186                          */
1187                         break;
1188                 }
1189                 num_rxd++;
1190                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1191                 idx = rcd->rxdIdx;
1192                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1193                 ring = rq->rx_ring + ring_idx;
1194                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1195                                   &rxCmdDesc);
1196                 rbi = rq->buf_info[ring_idx] + idx;
1197
1198                 BUG_ON(rxd->addr != rbi->dma_addr ||
1199                        rxd->len != rbi->len);
1200
1201                 if (unlikely(rcd->eop && rcd->err)) {
1202                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1203                         goto rcd_done;
1204                 }
1205
1206                 if (rcd->sop) { /* first buf of the pkt */
1207                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1208                                rcd->rqID != rq->qid);
1209
1210                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1211                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1212
1213                         if (unlikely(rcd->len == 0)) {
1214                                 /* Pretend the rx buffer is skipped. */
1215                                 BUG_ON(!(rcd->sop && rcd->eop));
1216                                 dev_dbg(&adapter->netdev->dev,
1217                                         "rxRing[%u][%u] 0 length\n",
1218                                         ring_idx, idx);
1219                                 goto rcd_done;
1220                         }
1221
1222                         skip_page_frags = false;
1223                         ctx->skb = rbi->skb;
1224                         new_skb = dev_alloc_skb(rbi->len + NET_IP_ALIGN);
1225                         if (new_skb == NULL) {
1226                                 /* Skb allocation failed, do not handover this
1227                                  * skb to stack. Reuse it. Drop the existing pkt
1228                                  */
1229                                 rq->stats.rx_buf_alloc_failure++;
1230                                 ctx->skb = NULL;
1231                                 rq->stats.drop_total++;
1232                                 skip_page_frags = true;
1233                                 goto rcd_done;
1234                         }
1235
1236                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1237                                          PCI_DMA_FROMDEVICE);
1238
1239                         skb_put(ctx->skb, rcd->len);
1240
1241                         /* Immediate refill */
1242                         new_skb->dev = adapter->netdev;
1243                         skb_reserve(new_skb, NET_IP_ALIGN);
1244                         rbi->skb = new_skb;
1245                         rbi->dma_addr = pci_map_single(adapter->pdev,
1246                                         rbi->skb->data, rbi->len,
1247                                         PCI_DMA_FROMDEVICE);
1248                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1249                         rxd->len = rbi->len;
1250
1251                 } else {
1252                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1253
1254                         /* non SOP buffer must be type 1 in most cases */
1255                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1256                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1257
1258                         /* If an sop buffer was dropped, skip all
1259                          * following non-sop fragments. They will be reused.
1260                          */
1261                         if (skip_page_frags)
1262                                 goto rcd_done;
1263
1264                         new_page = alloc_page(GFP_ATOMIC);
1265                         if (unlikely(new_page == NULL)) {
1266                                 /* Replacement page frag could not be allocated.
1267                                  * Reuse this page. Drop the pkt and free the
1268                                  * skb which contained this page as a frag. Skip
1269                                  * processing all the following non-sop frags.
1270                                  */
1271                                 rq->stats.rx_buf_alloc_failure++;
1272                                 dev_kfree_skb(ctx->skb);
1273                                 ctx->skb = NULL;
1274                                 skip_page_frags = true;
1275                                 goto rcd_done;
1276                         }
1277
1278                         if (rcd->len) {
1279                                 pci_unmap_page(adapter->pdev,
1280                                                rbi->dma_addr, rbi->len,
1281                                                PCI_DMA_FROMDEVICE);
1282
1283                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1284                         }
1285
1286                         /* Immediate refill */
1287                         rbi->page = new_page;
1288                         rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1289                                                      0, PAGE_SIZE,
1290                                                      PCI_DMA_FROMDEVICE);
1291                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1292                         rxd->len = rbi->len;
1293                 }
1294
1295
1296                 skb = ctx->skb;
1297                 if (rcd->eop) {
1298                         skb->len += skb->data_len;
1299
1300                         vmxnet3_rx_csum(adapter, skb,
1301                                         (union Vmxnet3_GenericDesc *)rcd);
1302                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1303
1304                         if (unlikely(rcd->ts))
1305                                 __vlan_hwaccel_put_tag(skb, rcd->tci);
1306
1307                         if (adapter->netdev->features & NETIF_F_LRO)
1308                                 netif_receive_skb(skb);
1309                         else
1310                                 napi_gro_receive(&rq->napi, skb);
1311
1312                         ctx->skb = NULL;
1313                 }
1314
1315 rcd_done:
1316                 /* device may have skipped some rx descs */
1317                 ring->next2comp = idx;
1318                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1319                 ring = rq->rx_ring + ring_idx;
1320                 while (num_to_alloc) {
1321                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1322                                           &rxCmdDesc);
1323                         BUG_ON(!rxd->addr);
1324
1325                         /* Recv desc is ready to be used by the device */
1326                         rxd->gen = ring->gen;
1327                         vmxnet3_cmd_ring_adv_next2fill(ring);
1328                         num_to_alloc--;
1329                 }
1330
1331                 /* if needed, update the register */
1332                 if (unlikely(rq->shared->updateRxProd)) {
1333                         VMXNET3_WRITE_BAR0_REG(adapter,
1334                                 rxprod_reg[ring_idx] + rq->qid * 8,
1335                                 ring->next2fill);
1336                         rq->uncommitted[ring_idx] = 0;
1337                 }
1338
1339                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1340                 vmxnet3_getRxComp(rcd,
1341                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1342         }
1343
1344         return num_rxd;
1345 }
1346
1347
1348 static void
1349 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1350                    struct vmxnet3_adapter *adapter)
1351 {
1352         u32 i, ring_idx;
1353         struct Vmxnet3_RxDesc *rxd;
1354
1355         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1356                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1357 #ifdef __BIG_ENDIAN_BITFIELD
1358                         struct Vmxnet3_RxDesc rxDesc;
1359 #endif
1360                         vmxnet3_getRxDesc(rxd,
1361                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1362
1363                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1364                                         rq->buf_info[ring_idx][i].skb) {
1365                                 pci_unmap_single(adapter->pdev, rxd->addr,
1366                                                  rxd->len, PCI_DMA_FROMDEVICE);
1367                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1368                                 rq->buf_info[ring_idx][i].skb = NULL;
1369                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1370                                         rq->buf_info[ring_idx][i].page) {
1371                                 pci_unmap_page(adapter->pdev, rxd->addr,
1372                                                rxd->len, PCI_DMA_FROMDEVICE);
1373                                 put_page(rq->buf_info[ring_idx][i].page);
1374                                 rq->buf_info[ring_idx][i].page = NULL;
1375                         }
1376                 }
1377
1378                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1379                 rq->rx_ring[ring_idx].next2fill =
1380                                         rq->rx_ring[ring_idx].next2comp = 0;
1381                 rq->uncommitted[ring_idx] = 0;
1382         }
1383
1384         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1385         rq->comp_ring.next2proc = 0;
1386 }
1387
1388
1389 static void
1390 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1391 {
1392         int i;
1393
1394         for (i = 0; i < adapter->num_rx_queues; i++)
1395                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1396 }
1397
1398
1399 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1400                         struct vmxnet3_adapter *adapter)
1401 {
1402         int i;
1403         int j;
1404
1405         /* all rx buffers must have already been freed */
1406         for (i = 0; i < 2; i++) {
1407                 if (rq->buf_info[i]) {
1408                         for (j = 0; j < rq->rx_ring[i].size; j++)
1409                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1410                 }
1411         }
1412
1413
1414         kfree(rq->buf_info[0]);
1415
1416         for (i = 0; i < 2; i++) {
1417                 if (rq->rx_ring[i].base) {
1418                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1419                                             * sizeof(struct Vmxnet3_RxDesc),
1420                                             rq->rx_ring[i].base,
1421                                             rq->rx_ring[i].basePA);
1422                         rq->rx_ring[i].base = NULL;
1423                 }
1424                 rq->buf_info[i] = NULL;
1425         }
1426
1427         if (rq->comp_ring.base) {
1428                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1429                                     sizeof(struct Vmxnet3_RxCompDesc),
1430                                     rq->comp_ring.base, rq->comp_ring.basePA);
1431                 rq->comp_ring.base = NULL;
1432         }
1433 }
1434
1435
1436 static int
1437 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1438                 struct vmxnet3_adapter  *adapter)
1439 {
1440         int i;
1441
1442         /* initialize buf_info */
1443         for (i = 0; i < rq->rx_ring[0].size; i++) {
1444
1445                 /* 1st buf for a pkt is skbuff */
1446                 if (i % adapter->rx_buf_per_pkt == 0) {
1447                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1448                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1449                 } else { /* subsequent bufs for a pkt is frag */
1450                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1451                         rq->buf_info[0][i].len = PAGE_SIZE;
1452                 }
1453         }
1454         for (i = 0; i < rq->rx_ring[1].size; i++) {
1455                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1456                 rq->buf_info[1][i].len = PAGE_SIZE;
1457         }
1458
1459         /* reset internal state and allocate buffers for both rings */
1460         for (i = 0; i < 2; i++) {
1461                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1462                 rq->uncommitted[i] = 0;
1463
1464                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1465                        sizeof(struct Vmxnet3_RxDesc));
1466                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1467         }
1468         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1469                                     adapter) == 0) {
1470                 /* at least has 1 rx buffer for the 1st ring */
1471                 return -ENOMEM;
1472         }
1473         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1474
1475         /* reset the comp ring */
1476         rq->comp_ring.next2proc = 0;
1477         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1478                sizeof(struct Vmxnet3_RxCompDesc));
1479         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1480
1481         /* reset rxctx */
1482         rq->rx_ctx.skb = NULL;
1483
1484         /* stats are not reset */
1485         return 0;
1486 }
1487
1488
1489 static int
1490 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1491 {
1492         int i, err = 0;
1493
1494         for (i = 0; i < adapter->num_rx_queues; i++) {
1495                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1496                 if (unlikely(err)) {
1497                         dev_err(&adapter->netdev->dev, "%s: failed to "
1498                                 "initialize rx queue%i\n",
1499                                 adapter->netdev->name, i);
1500                         break;
1501                 }
1502         }
1503         return err;
1504
1505 }
1506
1507
1508 static int
1509 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1510 {
1511         int i;
1512         size_t sz;
1513         struct vmxnet3_rx_buf_info *bi;
1514
1515         for (i = 0; i < 2; i++) {
1516
1517                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1518                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1519                                                         &rq->rx_ring[i].basePA);
1520                 if (!rq->rx_ring[i].base) {
1521                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1522                                adapter->netdev->name, i);
1523                         goto err;
1524                 }
1525         }
1526
1527         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1528         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1529                                                   &rq->comp_ring.basePA);
1530         if (!rq->comp_ring.base) {
1531                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1532                        adapter->netdev->name);
1533                 goto err;
1534         }
1535
1536         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1537                                                    rq->rx_ring[1].size);
1538         bi = kzalloc(sz, GFP_KERNEL);
1539         if (!bi)
1540                 goto err;
1541
1542         rq->buf_info[0] = bi;
1543         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1544
1545         return 0;
1546
1547 err:
1548         vmxnet3_rq_destroy(rq, adapter);
1549         return -ENOMEM;
1550 }
1551
1552
1553 static int
1554 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1555 {
1556         int i, err = 0;
1557
1558         for (i = 0; i < adapter->num_rx_queues; i++) {
1559                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1560                 if (unlikely(err)) {
1561                         dev_err(&adapter->netdev->dev,
1562                                 "%s: failed to create rx queue%i\n",
1563                                 adapter->netdev->name, i);
1564                         goto err_out;
1565                 }
1566         }
1567         return err;
1568 err_out:
1569         vmxnet3_rq_destroy_all(adapter);
1570         return err;
1571
1572 }
1573
1574 /* Multiple queue aware polling function for tx and rx */
1575
1576 static int
1577 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1578 {
1579         int rcd_done = 0, i;
1580         if (unlikely(adapter->shared->ecr))
1581                 vmxnet3_process_events(adapter);
1582         for (i = 0; i < adapter->num_tx_queues; i++)
1583                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1584
1585         for (i = 0; i < adapter->num_rx_queues; i++)
1586                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1587                                                    adapter, budget);
1588         return rcd_done;
1589 }
1590
1591
1592 static int
1593 vmxnet3_poll(struct napi_struct *napi, int budget)
1594 {
1595         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1596                                           struct vmxnet3_rx_queue, napi);
1597         int rxd_done;
1598
1599         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1600
1601         if (rxd_done < budget) {
1602                 napi_complete(napi);
1603                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1604         }
1605         return rxd_done;
1606 }
1607
1608 /*
1609  * NAPI polling function for MSI-X mode with multiple Rx queues
1610  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1611  */
1612
1613 static int
1614 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1615 {
1616         struct vmxnet3_rx_queue *rq = container_of(napi,
1617                                                 struct vmxnet3_rx_queue, napi);
1618         struct vmxnet3_adapter *adapter = rq->adapter;
1619         int rxd_done;
1620
1621         /* When sharing interrupt with corresponding tx queue, process
1622          * tx completions in that queue as well
1623          */
1624         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1625                 struct vmxnet3_tx_queue *tq =
1626                                 &adapter->tx_queue[rq - adapter->rx_queue];
1627                 vmxnet3_tq_tx_complete(tq, adapter);
1628         }
1629
1630         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1631
1632         if (rxd_done < budget) {
1633                 napi_complete(napi);
1634                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1635         }
1636         return rxd_done;
1637 }
1638
1639
1640 #ifdef CONFIG_PCI_MSI
1641
1642 /*
1643  * Handle completion interrupts on tx queues
1644  * Returns whether or not the intr is handled
1645  */
1646
1647 static irqreturn_t
1648 vmxnet3_msix_tx(int irq, void *data)
1649 {
1650         struct vmxnet3_tx_queue *tq = data;
1651         struct vmxnet3_adapter *adapter = tq->adapter;
1652
1653         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1654                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1655
1656         /* Handle the case where only one irq is allocate for all tx queues */
1657         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1658                 int i;
1659                 for (i = 0; i < adapter->num_tx_queues; i++) {
1660                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1661                         vmxnet3_tq_tx_complete(txq, adapter);
1662                 }
1663         } else {
1664                 vmxnet3_tq_tx_complete(tq, adapter);
1665         }
1666         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1667
1668         return IRQ_HANDLED;
1669 }
1670
1671
1672 /*
1673  * Handle completion interrupts on rx queues. Returns whether or not the
1674  * intr is handled
1675  */
1676
1677 static irqreturn_t
1678 vmxnet3_msix_rx(int irq, void *data)
1679 {
1680         struct vmxnet3_rx_queue *rq = data;
1681         struct vmxnet3_adapter *adapter = rq->adapter;
1682
1683         /* disable intr if needed */
1684         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1685                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1686         napi_schedule(&rq->napi);
1687
1688         return IRQ_HANDLED;
1689 }
1690
1691 /*
1692  *----------------------------------------------------------------------------
1693  *
1694  * vmxnet3_msix_event --
1695  *
1696  *    vmxnet3 msix event intr handler
1697  *
1698  * Result:
1699  *    whether or not the intr is handled
1700  *
1701  *----------------------------------------------------------------------------
1702  */
1703
1704 static irqreturn_t
1705 vmxnet3_msix_event(int irq, void *data)
1706 {
1707         struct net_device *dev = data;
1708         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1709
1710         /* disable intr if needed */
1711         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1712                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1713
1714         if (adapter->shared->ecr)
1715                 vmxnet3_process_events(adapter);
1716
1717         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1718
1719         return IRQ_HANDLED;
1720 }
1721
1722 #endif /* CONFIG_PCI_MSI  */
1723
1724
1725 /* Interrupt handler for vmxnet3  */
1726 static irqreturn_t
1727 vmxnet3_intr(int irq, void *dev_id)
1728 {
1729         struct net_device *dev = dev_id;
1730         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1731
1732         if (adapter->intr.type == VMXNET3_IT_INTX) {
1733                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1734                 if (unlikely(icr == 0))
1735                         /* not ours */
1736                         return IRQ_NONE;
1737         }
1738
1739
1740         /* disable intr if needed */
1741         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1742                 vmxnet3_disable_all_intrs(adapter);
1743
1744         napi_schedule(&adapter->rx_queue[0].napi);
1745
1746         return IRQ_HANDLED;
1747 }
1748
1749 #ifdef CONFIG_NET_POLL_CONTROLLER
1750
1751 /* netpoll callback. */
1752 static void
1753 vmxnet3_netpoll(struct net_device *netdev)
1754 {
1755         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1756
1757         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1758                 vmxnet3_disable_all_intrs(adapter);
1759
1760         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1761         vmxnet3_enable_all_intrs(adapter);
1762
1763 }
1764 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1765
1766 static int
1767 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1768 {
1769         struct vmxnet3_intr *intr = &adapter->intr;
1770         int err = 0, i;
1771         int vector = 0;
1772
1773 #ifdef CONFIG_PCI_MSI
1774         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1775                 for (i = 0; i < adapter->num_tx_queues; i++) {
1776                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1777                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1778                                         adapter->netdev->name, vector);
1779                                 err = request_irq(
1780                                               intr->msix_entries[vector].vector,
1781                                               vmxnet3_msix_tx, 0,
1782                                               adapter->tx_queue[i].name,
1783                                               &adapter->tx_queue[i]);
1784                         } else {
1785                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1786                                         adapter->netdev->name, vector);
1787                         }
1788                         if (err) {
1789                                 dev_err(&adapter->netdev->dev,
1790                                         "Failed to request irq for MSIX, %s, "
1791                                         "error %d\n",
1792                                         adapter->tx_queue[i].name, err);
1793                                 return err;
1794                         }
1795
1796                         /* Handle the case where only 1 MSIx was allocated for
1797                          * all tx queues */
1798                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1799                                 for (; i < adapter->num_tx_queues; i++)
1800                                         adapter->tx_queue[i].comp_ring.intr_idx
1801                                                                 = vector;
1802                                 vector++;
1803                                 break;
1804                         } else {
1805                                 adapter->tx_queue[i].comp_ring.intr_idx
1806                                                                 = vector++;
1807                         }
1808                 }
1809                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1810                         vector = 0;
1811
1812                 for (i = 0; i < adapter->num_rx_queues; i++) {
1813                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1814                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1815                                         adapter->netdev->name, vector);
1816                         else
1817                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1818                                         adapter->netdev->name, vector);
1819                         err = request_irq(intr->msix_entries[vector].vector,
1820                                           vmxnet3_msix_rx, 0,
1821                                           adapter->rx_queue[i].name,
1822                                           &(adapter->rx_queue[i]));
1823                         if (err) {
1824                                 printk(KERN_ERR "Failed to request irq for MSIX"
1825                                        ", %s, error %d\n",
1826                                        adapter->rx_queue[i].name, err);
1827                                 return err;
1828                         }
1829
1830                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1831                 }
1832
1833                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1834                         adapter->netdev->name, vector);
1835                 err = request_irq(intr->msix_entries[vector].vector,
1836                                   vmxnet3_msix_event, 0,
1837                                   intr->event_msi_vector_name, adapter->netdev);
1838                 intr->event_intr_idx = vector;
1839
1840         } else if (intr->type == VMXNET3_IT_MSI) {
1841                 adapter->num_rx_queues = 1;
1842                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1843                                   adapter->netdev->name, adapter->netdev);
1844         } else {
1845 #endif
1846                 adapter->num_rx_queues = 1;
1847                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1848                                   IRQF_SHARED, adapter->netdev->name,
1849                                   adapter->netdev);
1850 #ifdef CONFIG_PCI_MSI
1851         }
1852 #endif
1853         intr->num_intrs = vector + 1;
1854         if (err) {
1855                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1856                        ":%d\n", adapter->netdev->name, intr->type, err);
1857         } else {
1858                 /* Number of rx queues will not change after this */
1859                 for (i = 0; i < adapter->num_rx_queues; i++) {
1860                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1861                         rq->qid = i;
1862                         rq->qid2 = i + adapter->num_rx_queues;
1863                 }
1864
1865
1866
1867                 /* init our intr settings */
1868                 for (i = 0; i < intr->num_intrs; i++)
1869                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1870                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1871                         adapter->intr.event_intr_idx = 0;
1872                         for (i = 0; i < adapter->num_tx_queues; i++)
1873                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1874                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1875                 }
1876
1877                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1878                        "allocated\n", adapter->netdev->name, intr->type,
1879                        intr->mask_mode, intr->num_intrs);
1880         }
1881
1882         return err;
1883 }
1884
1885
1886 static void
1887 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1888 {
1889         struct vmxnet3_intr *intr = &adapter->intr;
1890         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1891
1892         switch (intr->type) {
1893 #ifdef CONFIG_PCI_MSI
1894         case VMXNET3_IT_MSIX:
1895         {
1896                 int i, vector = 0;
1897
1898                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1899                         for (i = 0; i < adapter->num_tx_queues; i++) {
1900                                 free_irq(intr->msix_entries[vector++].vector,
1901                                          &(adapter->tx_queue[i]));
1902                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1903                                         break;
1904                         }
1905                 }
1906
1907                 for (i = 0; i < adapter->num_rx_queues; i++) {
1908                         free_irq(intr->msix_entries[vector++].vector,
1909                                  &(adapter->rx_queue[i]));
1910                 }
1911
1912                 free_irq(intr->msix_entries[vector].vector,
1913                          adapter->netdev);
1914                 BUG_ON(vector >= intr->num_intrs);
1915                 break;
1916         }
1917 #endif
1918         case VMXNET3_IT_MSI:
1919                 free_irq(adapter->pdev->irq, adapter->netdev);
1920                 break;
1921         case VMXNET3_IT_INTX:
1922                 free_irq(adapter->pdev->irq, adapter->netdev);
1923                 break;
1924         default:
1925                 BUG_ON(true);
1926         }
1927 }
1928
1929
1930 static void
1931 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1932 {
1933         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1934         u16 vid;
1935
1936         /* allow untagged pkts */
1937         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1938
1939         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1940                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1941 }
1942
1943
1944 static int
1945 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1946 {
1947         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1948
1949         if (!(netdev->flags & IFF_PROMISC)) {
1950                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1951                 unsigned long flags;
1952
1953                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1954                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1955                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1956                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1957                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1958         }
1959
1960         set_bit(vid, adapter->active_vlans);
1961
1962         return 0;
1963 }
1964
1965
1966 static int
1967 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1968 {
1969         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1970
1971         if (!(netdev->flags & IFF_PROMISC)) {
1972                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1973                 unsigned long flags;
1974
1975                 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1976                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1977                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1978                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1979                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1980         }
1981
1982         clear_bit(vid, adapter->active_vlans);
1983
1984         return 0;
1985 }
1986
1987
1988 static u8 *
1989 vmxnet3_copy_mc(struct net_device *netdev)
1990 {
1991         u8 *buf = NULL;
1992         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1993
1994         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1995         if (sz <= 0xffff) {
1996                 /* We may be called with BH disabled */
1997                 buf = kmalloc(sz, GFP_ATOMIC);
1998                 if (buf) {
1999                         struct netdev_hw_addr *ha;
2000                         int i = 0;
2001
2002                         netdev_for_each_mc_addr(ha, netdev)
2003                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2004                                        ETH_ALEN);
2005                 }
2006         }
2007         return buf;
2008 }
2009
2010
2011 static void
2012 vmxnet3_set_mc(struct net_device *netdev)
2013 {
2014         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2015         unsigned long flags;
2016         struct Vmxnet3_RxFilterConf *rxConf =
2017                                         &adapter->shared->devRead.rxFilterConf;
2018         u8 *new_table = NULL;
2019         u32 new_mode = VMXNET3_RXM_UCAST;
2020
2021         if (netdev->flags & IFF_PROMISC) {
2022                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2023                 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2024
2025                 new_mode |= VMXNET3_RXM_PROMISC;
2026         } else {
2027                 vmxnet3_restore_vlan(adapter);
2028         }
2029
2030         if (netdev->flags & IFF_BROADCAST)
2031                 new_mode |= VMXNET3_RXM_BCAST;
2032
2033         if (netdev->flags & IFF_ALLMULTI)
2034                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2035         else
2036                 if (!netdev_mc_empty(netdev)) {
2037                         new_table = vmxnet3_copy_mc(netdev);
2038                         if (new_table) {
2039                                 new_mode |= VMXNET3_RXM_MCAST;
2040                                 rxConf->mfTableLen = cpu_to_le16(
2041                                         netdev_mc_count(netdev) * ETH_ALEN);
2042                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2043                                                     new_table));
2044                         } else {
2045                                 printk(KERN_INFO "%s: failed to copy mcast list"
2046                                        ", setting ALL_MULTI\n", netdev->name);
2047                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2048                         }
2049                 }
2050
2051
2052         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2053                 rxConf->mfTableLen = 0;
2054                 rxConf->mfTablePA = 0;
2055         }
2056
2057         spin_lock_irqsave(&adapter->cmd_lock, flags);
2058         if (new_mode != rxConf->rxMode) {
2059                 rxConf->rxMode = cpu_to_le32(new_mode);
2060                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2061                                        VMXNET3_CMD_UPDATE_RX_MODE);
2062                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2063                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2064         }
2065
2066         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2067                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2068         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2069
2070         kfree(new_table);
2071 }
2072
2073 void
2074 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2075 {
2076         int i;
2077
2078         for (i = 0; i < adapter->num_rx_queues; i++)
2079                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2080 }
2081
2082
2083 /*
2084  *   Set up driver_shared based on settings in adapter.
2085  */
2086
2087 static void
2088 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2089 {
2090         struct Vmxnet3_DriverShared *shared = adapter->shared;
2091         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2092         struct Vmxnet3_TxQueueConf *tqc;
2093         struct Vmxnet3_RxQueueConf *rqc;
2094         int i;
2095
2096         memset(shared, 0, sizeof(*shared));
2097
2098         /* driver settings */
2099         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2100         devRead->misc.driverInfo.version = cpu_to_le32(
2101                                                 VMXNET3_DRIVER_VERSION_NUM);
2102         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2103                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2104         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2105         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2106                                 *((u32 *)&devRead->misc.driverInfo.gos));
2107         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2108         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2109
2110         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2111         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2112
2113         /* set up feature flags */
2114         if (adapter->netdev->features & NETIF_F_RXCSUM)
2115                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2116
2117         if (adapter->netdev->features & NETIF_F_LRO) {
2118                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2119                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2120         }
2121         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2122                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2123
2124         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2125         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2126         devRead->misc.queueDescLen = cpu_to_le32(
2127                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2128                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2129
2130         /* tx queue settings */
2131         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2132         for (i = 0; i < adapter->num_tx_queues; i++) {
2133                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2134                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2135                 tqc = &adapter->tqd_start[i].conf;
2136                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2137                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2138                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2139                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2140                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2141                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2142                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2143                 tqc->ddLen          = cpu_to_le32(
2144                                         sizeof(struct vmxnet3_tx_buf_info) *
2145                                         tqc->txRingSize);
2146                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2147         }
2148
2149         /* rx queue settings */
2150         devRead->misc.numRxQueues = adapter->num_rx_queues;
2151         for (i = 0; i < adapter->num_rx_queues; i++) {
2152                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2153                 rqc = &adapter->rqd_start[i].conf;
2154                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2155                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2156                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2157                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2158                                                         rq->buf_info));
2159                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2160                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2161                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2162                 rqc->ddLen           = cpu_to_le32(
2163                                         sizeof(struct vmxnet3_rx_buf_info) *
2164                                         (rqc->rxRingSize[0] +
2165                                          rqc->rxRingSize[1]));
2166                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2167         }
2168
2169 #ifdef VMXNET3_RSS
2170         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2171
2172         if (adapter->rss) {
2173                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2174                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2175                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2176                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2177                                     UPT1_RSS_HASH_TYPE_IPV4 |
2178                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2179                                     UPT1_RSS_HASH_TYPE_IPV6;
2180                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2181                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2182                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2183                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2184                 for (i = 0; i < rssConf->indTableSize; i++)
2185                         rssConf->indTable[i] = ethtool_rxfh_indir_default(
2186                                 i, adapter->num_rx_queues);
2187
2188                 devRead->rssConfDesc.confVer = 1;
2189                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2190                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2191         }
2192
2193 #endif /* VMXNET3_RSS */
2194
2195         /* intr settings */
2196         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2197                                      VMXNET3_IMM_AUTO;
2198         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2199         for (i = 0; i < adapter->intr.num_intrs; i++)
2200                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2201
2202         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2203         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2204
2205         /* rx filter settings */
2206         devRead->rxFilterConf.rxMode = 0;
2207         vmxnet3_restore_vlan(adapter);
2208         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2209
2210         /* the rest are already zeroed */
2211 }
2212
2213
2214 int
2215 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2216 {
2217         int err, i;
2218         u32 ret;
2219         unsigned long flags;
2220
2221         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2222                 " ring sizes %u %u %u\n", adapter->netdev->name,
2223                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2224                 adapter->tx_queue[0].tx_ring.size,
2225                 adapter->rx_queue[0].rx_ring[0].size,
2226                 adapter->rx_queue[0].rx_ring[1].size);
2227
2228         vmxnet3_tq_init_all(adapter);
2229         err = vmxnet3_rq_init_all(adapter);
2230         if (err) {
2231                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2232                        adapter->netdev->name, err);
2233                 goto rq_err;
2234         }
2235
2236         err = vmxnet3_request_irqs(adapter);
2237         if (err) {
2238                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2239                        adapter->netdev->name, err);
2240                 goto irq_err;
2241         }
2242
2243         vmxnet3_setup_driver_shared(adapter);
2244
2245         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2246                                adapter->shared_pa));
2247         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2248                                adapter->shared_pa));
2249         spin_lock_irqsave(&adapter->cmd_lock, flags);
2250         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2251                                VMXNET3_CMD_ACTIVATE_DEV);
2252         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2253         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2254
2255         if (ret != 0) {
2256                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2257                        adapter->netdev->name, ret);
2258                 err = -EINVAL;
2259                 goto activate_err;
2260         }
2261
2262         for (i = 0; i < adapter->num_rx_queues; i++) {
2263                 VMXNET3_WRITE_BAR0_REG(adapter,
2264                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2265                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2266                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2267                                 (i * VMXNET3_REG_ALIGN)),
2268                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2269         }
2270
2271         /* Apply the rx filter settins last. */
2272         vmxnet3_set_mc(adapter->netdev);
2273
2274         /*
2275          * Check link state when first activating device. It will start the
2276          * tx queue if the link is up.
2277          */
2278         vmxnet3_check_link(adapter, true);
2279         for (i = 0; i < adapter->num_rx_queues; i++)
2280                 napi_enable(&adapter->rx_queue[i].napi);
2281         vmxnet3_enable_all_intrs(adapter);
2282         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2283         return 0;
2284
2285 activate_err:
2286         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2287         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2288         vmxnet3_free_irqs(adapter);
2289 irq_err:
2290 rq_err:
2291         /* free up buffers we allocated */
2292         vmxnet3_rq_cleanup_all(adapter);
2293         return err;
2294 }
2295
2296
2297 void
2298 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2299 {
2300         unsigned long flags;
2301         spin_lock_irqsave(&adapter->cmd_lock, flags);
2302         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2303         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2304 }
2305
2306
2307 int
2308 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2309 {
2310         int i;
2311         unsigned long flags;
2312         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2313                 return 0;
2314
2315
2316         spin_lock_irqsave(&adapter->cmd_lock, flags);
2317         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2318                                VMXNET3_CMD_QUIESCE_DEV);
2319         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2320         vmxnet3_disable_all_intrs(adapter);
2321
2322         for (i = 0; i < adapter->num_rx_queues; i++)
2323                 napi_disable(&adapter->rx_queue[i].napi);
2324         netif_tx_disable(adapter->netdev);
2325         adapter->link_speed = 0;
2326         netif_carrier_off(adapter->netdev);
2327
2328         vmxnet3_tq_cleanup_all(adapter);
2329         vmxnet3_rq_cleanup_all(adapter);
2330         vmxnet3_free_irqs(adapter);
2331         return 0;
2332 }
2333
2334
2335 static void
2336 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2337 {
2338         u32 tmp;
2339
2340         tmp = *(u32 *)mac;
2341         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2342
2343         tmp = (mac[5] << 8) | mac[4];
2344         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2345 }
2346
2347
2348 static int
2349 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2350 {
2351         struct sockaddr *addr = p;
2352         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2353
2354         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2355         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2356
2357         return 0;
2358 }
2359
2360
2361 /* ==================== initialization and cleanup routines ============ */
2362
2363 static int
2364 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2365 {
2366         int err;
2367         unsigned long mmio_start, mmio_len;
2368         struct pci_dev *pdev = adapter->pdev;
2369
2370         err = pci_enable_device(pdev);
2371         if (err) {
2372                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2373                        pci_name(pdev), err);
2374                 return err;
2375         }
2376
2377         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2378                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2379                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2380                                "for adapter %s\n", pci_name(pdev));
2381                         err = -EIO;
2382                         goto err_set_mask;
2383                 }
2384                 *dma64 = true;
2385         } else {
2386                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2387                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2388                                "%s\n",  pci_name(pdev));
2389                         err = -EIO;
2390                         goto err_set_mask;
2391                 }
2392                 *dma64 = false;
2393         }
2394
2395         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2396                                            vmxnet3_driver_name);
2397         if (err) {
2398                 printk(KERN_ERR "Failed to request region for adapter %s: "
2399                        "error %d\n", pci_name(pdev), err);
2400                 goto err_set_mask;
2401         }
2402
2403         pci_set_master(pdev);
2404
2405         mmio_start = pci_resource_start(pdev, 0);
2406         mmio_len = pci_resource_len(pdev, 0);
2407         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2408         if (!adapter->hw_addr0) {
2409                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2410                        pci_name(pdev));
2411                 err = -EIO;
2412                 goto err_ioremap;
2413         }
2414
2415         mmio_start = pci_resource_start(pdev, 1);
2416         mmio_len = pci_resource_len(pdev, 1);
2417         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2418         if (!adapter->hw_addr1) {
2419                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2420                        pci_name(pdev));
2421                 err = -EIO;
2422                 goto err_bar1;
2423         }
2424         return 0;
2425
2426 err_bar1:
2427         iounmap(adapter->hw_addr0);
2428 err_ioremap:
2429         pci_release_selected_regions(pdev, (1 << 2) - 1);
2430 err_set_mask:
2431         pci_disable_device(pdev);
2432         return err;
2433 }
2434
2435
2436 static void
2437 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2438 {
2439         BUG_ON(!adapter->pdev);
2440
2441         iounmap(adapter->hw_addr0);
2442         iounmap(adapter->hw_addr1);
2443         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2444         pci_disable_device(adapter->pdev);
2445 }
2446
2447
2448 static void
2449 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2450 {
2451         size_t sz, i, ring0_size, ring1_size, comp_size;
2452         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2453
2454
2455         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2456                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2457                 adapter->skb_buf_size = adapter->netdev->mtu +
2458                                         VMXNET3_MAX_ETH_HDR_SIZE;
2459                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2460                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2461
2462                 adapter->rx_buf_per_pkt = 1;
2463         } else {
2464                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2465                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2466                                             VMXNET3_MAX_ETH_HDR_SIZE;
2467                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2468         }
2469
2470         /*
2471          * for simplicity, force the ring0 size to be a multiple of
2472          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2473          */
2474         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2475         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2476         ring0_size = (ring0_size + sz - 1) / sz * sz;
2477         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2478                            sz * sz);
2479         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2480         comp_size = ring0_size + ring1_size;
2481
2482         for (i = 0; i < adapter->num_rx_queues; i++) {
2483                 rq = &adapter->rx_queue[i];
2484                 rq->rx_ring[0].size = ring0_size;
2485                 rq->rx_ring[1].size = ring1_size;
2486                 rq->comp_ring.size = comp_size;
2487         }
2488 }
2489
2490
2491 int
2492 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2493                       u32 rx_ring_size, u32 rx_ring2_size)
2494 {
2495         int err = 0, i;
2496
2497         for (i = 0; i < adapter->num_tx_queues; i++) {
2498                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2499                 tq->tx_ring.size   = tx_ring_size;
2500                 tq->data_ring.size = tx_ring_size;
2501                 tq->comp_ring.size = tx_ring_size;
2502                 tq->shared = &adapter->tqd_start[i].ctrl;
2503                 tq->stopped = true;
2504                 tq->adapter = adapter;
2505                 tq->qid = i;
2506                 err = vmxnet3_tq_create(tq, adapter);
2507                 /*
2508                  * Too late to change num_tx_queues. We cannot do away with
2509                  * lesser number of queues than what we asked for
2510                  */
2511                 if (err)
2512                         goto queue_err;
2513         }
2514
2515         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2516         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2517         vmxnet3_adjust_rx_ring_size(adapter);
2518         for (i = 0; i < adapter->num_rx_queues; i++) {
2519                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2520                 /* qid and qid2 for rx queues will be assigned later when num
2521                  * of rx queues is finalized after allocating intrs */
2522                 rq->shared = &adapter->rqd_start[i].ctrl;
2523                 rq->adapter = adapter;
2524                 err = vmxnet3_rq_create(rq, adapter);
2525                 if (err) {
2526                         if (i == 0) {
2527                                 printk(KERN_ERR "Could not allocate any rx"
2528                                        "queues. Aborting.\n");
2529                                 goto queue_err;
2530                         } else {
2531                                 printk(KERN_INFO "Number of rx queues changed "
2532                                        "to : %d.\n", i);
2533                                 adapter->num_rx_queues = i;
2534                                 err = 0;
2535                                 break;
2536                         }
2537                 }
2538         }
2539         return err;
2540 queue_err:
2541         vmxnet3_tq_destroy_all(adapter);
2542         return err;
2543 }
2544
2545 static int
2546 vmxnet3_open(struct net_device *netdev)
2547 {
2548         struct vmxnet3_adapter *adapter;
2549         int err, i;
2550
2551         adapter = netdev_priv(netdev);
2552
2553         for (i = 0; i < adapter->num_tx_queues; i++)
2554                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2555
2556         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2557                                     VMXNET3_DEF_RX_RING_SIZE,
2558                                     VMXNET3_DEF_RX_RING_SIZE);
2559         if (err)
2560                 goto queue_err;
2561
2562         err = vmxnet3_activate_dev(adapter);
2563         if (err)
2564                 goto activate_err;
2565
2566         return 0;
2567
2568 activate_err:
2569         vmxnet3_rq_destroy_all(adapter);
2570         vmxnet3_tq_destroy_all(adapter);
2571 queue_err:
2572         return err;
2573 }
2574
2575
2576 static int
2577 vmxnet3_close(struct net_device *netdev)
2578 {
2579         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2580
2581         /*
2582          * Reset_work may be in the middle of resetting the device, wait for its
2583          * completion.
2584          */
2585         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2586                 msleep(1);
2587
2588         vmxnet3_quiesce_dev(adapter);
2589
2590         vmxnet3_rq_destroy_all(adapter);
2591         vmxnet3_tq_destroy_all(adapter);
2592
2593         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2594
2595
2596         return 0;
2597 }
2598
2599
2600 void
2601 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2602 {
2603         int i;
2604
2605         /*
2606          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2607          * vmxnet3_close() will deadlock.
2608          */
2609         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2610
2611         /* we need to enable NAPI, otherwise dev_close will deadlock */
2612         for (i = 0; i < adapter->num_rx_queues; i++)
2613                 napi_enable(&adapter->rx_queue[i].napi);
2614         dev_close(adapter->netdev);
2615 }
2616
2617
2618 static int
2619 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2620 {
2621         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2622         int err = 0;
2623
2624         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2625                 return -EINVAL;
2626
2627         netdev->mtu = new_mtu;
2628
2629         /*
2630          * Reset_work may be in the middle of resetting the device, wait for its
2631          * completion.
2632          */
2633         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2634                 msleep(1);
2635
2636         if (netif_running(netdev)) {
2637                 vmxnet3_quiesce_dev(adapter);
2638                 vmxnet3_reset_dev(adapter);
2639
2640                 /* we need to re-create the rx queue based on the new mtu */
2641                 vmxnet3_rq_destroy_all(adapter);
2642                 vmxnet3_adjust_rx_ring_size(adapter);
2643                 err = vmxnet3_rq_create_all(adapter);
2644                 if (err) {
2645                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2646                                 " error %d. Closing it.\n", netdev->name, err);
2647                         goto out;
2648                 }
2649
2650                 err = vmxnet3_activate_dev(adapter);
2651                 if (err) {
2652                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2653                                 "Closing it\n", netdev->name, err);
2654                         goto out;
2655                 }
2656         }
2657
2658 out:
2659         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2660         if (err)
2661                 vmxnet3_force_close(adapter);
2662
2663         return err;
2664 }
2665
2666
2667 static void
2668 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2669 {
2670         struct net_device *netdev = adapter->netdev;
2671
2672         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2673                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2674                 NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2675                 NETIF_F_LRO;
2676         if (dma64)
2677                 netdev->hw_features |= NETIF_F_HIGHDMA;
2678         netdev->vlan_features = netdev->hw_features &
2679                                 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
2680         netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER;
2681
2682         netdev_info(adapter->netdev,
2683                 "features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2684                 dma64 ? " highDMA" : "");
2685 }
2686
2687
2688 static void
2689 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2690 {
2691         u32 tmp;
2692
2693         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2694         *(u32 *)mac = tmp;
2695
2696         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2697         mac[4] = tmp & 0xff;
2698         mac[5] = (tmp >> 8) & 0xff;
2699 }
2700
2701 #ifdef CONFIG_PCI_MSI
2702
2703 /*
2704  * Enable MSIx vectors.
2705  * Returns :
2706  *      0 on successful enabling of required vectors,
2707  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2708  *       could be enabled.
2709  *      number of vectors which can be enabled otherwise (this number is smaller
2710  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2711  */
2712
2713 static int
2714 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2715                              int vectors)
2716 {
2717         int err = 0, vector_threshold;
2718         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2719
2720         while (vectors >= vector_threshold) {
2721                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2722                                       vectors);
2723                 if (!err) {
2724                         adapter->intr.num_intrs = vectors;
2725                         return 0;
2726                 } else if (err < 0) {
2727                         netdev_err(adapter->netdev,
2728                                    "Failed to enable MSI-X, error: %d\n", err);
2729                         vectors = 0;
2730                 } else if (err < vector_threshold) {
2731                         break;
2732                 } else {
2733                         /* If fails to enable required number of MSI-x vectors
2734                          * try enabling minimum number of vectors required.
2735                          */
2736                         netdev_err(adapter->netdev,
2737                                    "Failed to enable %d MSI-X, trying %d instead\n",
2738                                     vectors, vector_threshold);
2739                         vectors = vector_threshold;
2740                 }
2741         }
2742
2743         netdev_info(adapter->netdev,
2744                     "Number of MSI-X interrupts which can be allocated are lower than min threshold required.\n");
2745         return err;
2746 }
2747
2748
2749 #endif /* CONFIG_PCI_MSI */
2750
2751 static void
2752 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2753 {
2754         u32 cfg;
2755         unsigned long flags;
2756
2757         /* intr settings */
2758         spin_lock_irqsave(&adapter->cmd_lock, flags);
2759         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2760                                VMXNET3_CMD_GET_CONF_INTR);
2761         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2762         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2763         adapter->intr.type = cfg & 0x3;
2764         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2765
2766         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2767                 adapter->intr.type = VMXNET3_IT_MSIX;
2768         }
2769
2770 #ifdef CONFIG_PCI_MSI
2771         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2772                 int vector, err = 0;
2773
2774                 adapter->intr.num_intrs = (adapter->share_intr ==
2775                                            VMXNET3_INTR_TXSHARE) ? 1 :
2776                                            adapter->num_tx_queues;
2777                 adapter->intr.num_intrs += (adapter->share_intr ==
2778                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2779                                            adapter->num_rx_queues;
2780                 adapter->intr.num_intrs += 1;           /* for link event */
2781
2782                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2783                                            VMXNET3_LINUX_MIN_MSIX_VECT
2784                                            ? adapter->intr.num_intrs :
2785                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2786
2787                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2788                         adapter->intr.msix_entries[vector].entry = vector;
2789
2790                 err = vmxnet3_acquire_msix_vectors(adapter,
2791                                                    adapter->intr.num_intrs);
2792                 /* If we cannot allocate one MSIx vector per queue
2793                  * then limit the number of rx queues to 1
2794                  */
2795                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2796                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2797                             || adapter->num_rx_queues != 1) {
2798                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2799                                 printk(KERN_ERR "Number of rx queues : 1\n");
2800                                 adapter->num_rx_queues = 1;
2801                                 adapter->intr.num_intrs =
2802                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2803                         }
2804                         return;
2805                 }
2806                 if (!err)
2807                         return;
2808
2809                 /* If we cannot allocate MSIx vectors use only one rx queue */
2810                 netdev_info(adapter->netdev,
2811                             "Failed to enable MSI-X, error %d . Limiting #rx queues to 1, try MSI.\n",
2812                             err);
2813
2814                 adapter->intr.type = VMXNET3_IT_MSI;
2815         }
2816
2817         if (adapter->intr.type == VMXNET3_IT_MSI) {
2818                 int err;
2819                 err = pci_enable_msi(adapter->pdev);
2820                 if (!err) {
2821                         adapter->num_rx_queues = 1;
2822                         adapter->intr.num_intrs = 1;
2823                         return;
2824                 }
2825         }
2826 #endif /* CONFIG_PCI_MSI */
2827
2828         adapter->num_rx_queues = 1;
2829         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2830         adapter->intr.type = VMXNET3_IT_INTX;
2831
2832         /* INT-X related setting */
2833         adapter->intr.num_intrs = 1;
2834 }
2835
2836
2837 static void
2838 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2839 {
2840         if (adapter->intr.type == VMXNET3_IT_MSIX)
2841                 pci_disable_msix(adapter->pdev);
2842         else if (adapter->intr.type == VMXNET3_IT_MSI)
2843                 pci_disable_msi(adapter->pdev);
2844         else
2845                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2846 }
2847
2848
2849 static void
2850 vmxnet3_tx_timeout(struct net_device *netdev)
2851 {
2852         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2853         adapter->tx_timeout_count++;
2854
2855         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2856         schedule_work(&adapter->work);
2857         netif_wake_queue(adapter->netdev);
2858 }
2859
2860
2861 static void
2862 vmxnet3_reset_work(struct work_struct *data)
2863 {
2864         struct vmxnet3_adapter *adapter;
2865
2866         adapter = container_of(data, struct vmxnet3_adapter, work);
2867
2868         /* if another thread is resetting the device, no need to proceed */
2869         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2870                 return;
2871
2872         /* if the device is closed, we must leave it alone */
2873         rtnl_lock();
2874         if (netif_running(adapter->netdev)) {
2875                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2876                 vmxnet3_quiesce_dev(adapter);
2877                 vmxnet3_reset_dev(adapter);
2878                 vmxnet3_activate_dev(adapter);
2879         } else {
2880                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2881         }
2882         rtnl_unlock();
2883
2884         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2885 }
2886
2887
2888 static int __devinit
2889 vmxnet3_probe_device(struct pci_dev *pdev,
2890                      const struct pci_device_id *id)
2891 {
2892         static const struct net_device_ops vmxnet3_netdev_ops = {
2893                 .ndo_open = vmxnet3_open,
2894                 .ndo_stop = vmxnet3_close,
2895                 .ndo_start_xmit = vmxnet3_xmit_frame,
2896                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2897                 .ndo_change_mtu = vmxnet3_change_mtu,
2898                 .ndo_set_features = vmxnet3_set_features,
2899                 .ndo_get_stats64 = vmxnet3_get_stats64,
2900                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2901                 .ndo_set_rx_mode = vmxnet3_set_mc,
2902                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2903                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2904 #ifdef CONFIG_NET_POLL_CONTROLLER
2905                 .ndo_poll_controller = vmxnet3_netpoll,
2906 #endif
2907         };
2908         int err;
2909         bool dma64 = false; /* stupid gcc */
2910         u32 ver;
2911         struct net_device *netdev;
2912         struct vmxnet3_adapter *adapter;
2913         u8 mac[ETH_ALEN];
2914         int size;
2915         int num_tx_queues;
2916         int num_rx_queues;
2917
2918         if (!pci_msi_enabled())
2919                 enable_mq = 0;
2920
2921 #ifdef VMXNET3_RSS
2922         if (enable_mq)
2923                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2924                                     (int)num_online_cpus());
2925         else
2926 #endif
2927                 num_rx_queues = 1;
2928         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2929
2930         if (enable_mq)
2931                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2932                                     (int)num_online_cpus());
2933         else
2934                 num_tx_queues = 1;
2935
2936         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2937         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2938                                    max(num_tx_queues, num_rx_queues));
2939         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2940                num_tx_queues, num_rx_queues);
2941
2942         if (!netdev)
2943                 return -ENOMEM;
2944
2945         pci_set_drvdata(pdev, netdev);
2946         adapter = netdev_priv(netdev);
2947         adapter->netdev = netdev;
2948         adapter->pdev = pdev;
2949
2950         spin_lock_init(&adapter->cmd_lock);
2951         adapter->shared = pci_alloc_consistent(adapter->pdev,
2952                           sizeof(struct Vmxnet3_DriverShared),
2953                           &adapter->shared_pa);
2954         if (!adapter->shared) {
2955                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2956                         pci_name(pdev));
2957                 err = -ENOMEM;
2958                 goto err_alloc_shared;
2959         }
2960
2961         adapter->num_rx_queues = num_rx_queues;
2962         adapter->num_tx_queues = num_tx_queues;
2963
2964         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2965         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2966         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2967                              &adapter->queue_desc_pa);
2968
2969         if (!adapter->tqd_start) {
2970                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2971                         pci_name(pdev));
2972                 err = -ENOMEM;
2973                 goto err_alloc_queue_desc;
2974         }
2975         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2976                                                         adapter->num_tx_queues);
2977
2978         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2979         if (adapter->pm_conf == NULL) {
2980                 err = -ENOMEM;
2981                 goto err_alloc_pm;
2982         }
2983
2984 #ifdef VMXNET3_RSS
2985
2986         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2987         if (adapter->rss_conf == NULL) {
2988                 err = -ENOMEM;
2989                 goto err_alloc_rss;
2990         }
2991 #endif /* VMXNET3_RSS */
2992
2993         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2994         if (err < 0)
2995                 goto err_alloc_pci;
2996
2997         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2998         if (ver & 1) {
2999                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
3000         } else {
3001                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
3002                        " %s\n", ver, pci_name(pdev));
3003                 err = -EBUSY;
3004                 goto err_ver;
3005         }
3006
3007         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3008         if (ver & 1) {
3009                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3010         } else {
3011                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
3012                        "adapter %s\n", ver, pci_name(pdev));
3013                 err = -EBUSY;
3014                 goto err_ver;
3015         }
3016
3017         SET_NETDEV_DEV(netdev, &pdev->dev);
3018         vmxnet3_declare_features(adapter, dma64);
3019
3020         adapter->dev_number = atomic_read(&devices_found);
3021
3022          adapter->share_intr = irq_share_mode;
3023         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3024             adapter->num_tx_queues != adapter->num_rx_queues)
3025                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3026
3027         vmxnet3_alloc_intr_resources(adapter);
3028
3029 #ifdef VMXNET3_RSS
3030         if (adapter->num_rx_queues > 1 &&
3031             adapter->intr.type == VMXNET3_IT_MSIX) {
3032                 adapter->rss = true;
3033                 printk(KERN_INFO "RSS is enabled.\n");
3034         } else {
3035                 adapter->rss = false;
3036         }
3037 #endif
3038
3039         vmxnet3_read_mac_addr(adapter, mac);
3040         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3041
3042         netdev->netdev_ops = &vmxnet3_netdev_ops;
3043         vmxnet3_set_ethtool_ops(netdev);
3044         netdev->watchdog_timeo = 5 * HZ;
3045
3046         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3047         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3048
3049         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3050                 int i;
3051                 for (i = 0; i < adapter->num_rx_queues; i++) {
3052                         netif_napi_add(adapter->netdev,
3053                                        &adapter->rx_queue[i].napi,
3054                                        vmxnet3_poll_rx_only, 64);
3055                 }
3056         } else {
3057                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3058                                vmxnet3_poll, 64);
3059         }
3060
3061         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3062         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3063
3064         err = register_netdev(netdev);
3065
3066         if (err) {
3067                 printk(KERN_ERR "Failed to register adapter %s\n",
3068                         pci_name(pdev));
3069                 goto err_register;
3070         }
3071
3072         vmxnet3_check_link(adapter, false);
3073         atomic_inc(&devices_found);
3074         return 0;
3075
3076 err_register:
3077         vmxnet3_free_intr_resources(adapter);
3078 err_ver:
3079         vmxnet3_free_pci_resources(adapter);
3080 err_alloc_pci:
3081 #ifdef VMXNET3_RSS
3082         kfree(adapter->rss_conf);
3083 err_alloc_rss:
3084 #endif
3085         kfree(adapter->pm_conf);
3086 err_alloc_pm:
3087         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3088                             adapter->queue_desc_pa);
3089 err_alloc_queue_desc:
3090         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3091                             adapter->shared, adapter->shared_pa);
3092 err_alloc_shared:
3093         pci_set_drvdata(pdev, NULL);
3094         free_netdev(netdev);
3095         return err;
3096 }
3097
3098
3099 static void __devexit
3100 vmxnet3_remove_device(struct pci_dev *pdev)
3101 {
3102         struct net_device *netdev = pci_get_drvdata(pdev);
3103         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3104         int size = 0;
3105         int num_rx_queues;
3106
3107 #ifdef VMXNET3_RSS
3108         if (enable_mq)
3109                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3110                                     (int)num_online_cpus());
3111         else
3112 #endif
3113                 num_rx_queues = 1;
3114         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3115
3116         cancel_work_sync(&adapter->work);
3117
3118         unregister_netdev(netdev);
3119
3120         vmxnet3_free_intr_resources(adapter);
3121         vmxnet3_free_pci_resources(adapter);
3122 #ifdef VMXNET3_RSS
3123         kfree(adapter->rss_conf);
3124 #endif
3125         kfree(adapter->pm_conf);
3126
3127         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3128         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3129         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3130                             adapter->queue_desc_pa);
3131         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3132                             adapter->shared, adapter->shared_pa);
3133         free_netdev(netdev);
3134 }
3135
3136
3137 #ifdef CONFIG_PM
3138
3139 static int
3140 vmxnet3_suspend(struct device *device)
3141 {
3142         struct pci_dev *pdev = to_pci_dev(device);
3143         struct net_device *netdev = pci_get_drvdata(pdev);
3144         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3145         struct Vmxnet3_PMConf *pmConf;
3146         struct ethhdr *ehdr;
3147         struct arphdr *ahdr;
3148         u8 *arpreq;
3149         struct in_device *in_dev;
3150         struct in_ifaddr *ifa;
3151         unsigned long flags;
3152         int i = 0;
3153
3154         if (!netif_running(netdev))
3155                 return 0;
3156
3157         for (i = 0; i < adapter->num_rx_queues; i++)
3158                 napi_disable(&adapter->rx_queue[i].napi);
3159
3160         vmxnet3_disable_all_intrs(adapter);
3161         vmxnet3_free_irqs(adapter);
3162         vmxnet3_free_intr_resources(adapter);
3163
3164         netif_device_detach(netdev);
3165         netif_tx_stop_all_queues(netdev);
3166
3167         /* Create wake-up filters. */
3168         pmConf = adapter->pm_conf;
3169         memset(pmConf, 0, sizeof(*pmConf));
3170
3171         if (adapter->wol & WAKE_UCAST) {
3172                 pmConf->filters[i].patternSize = ETH_ALEN;
3173                 pmConf->filters[i].maskSize = 1;
3174                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3175                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3176
3177                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3178                 i++;
3179         }
3180
3181         if (adapter->wol & WAKE_ARP) {
3182                 in_dev = in_dev_get(netdev);
3183                 if (!in_dev)
3184                         goto skip_arp;
3185
3186                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3187                 if (!ifa)
3188                         goto skip_arp;
3189
3190                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3191                         sizeof(struct arphdr) +         /* ARP header */
3192                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3193                         2 * sizeof(u32);        /*2 IPv4 addresses */
3194                 pmConf->filters[i].maskSize =
3195                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3196
3197                 /* ETH_P_ARP in Ethernet header. */
3198                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3199                 ehdr->h_proto = htons(ETH_P_ARP);
3200
3201                 /* ARPOP_REQUEST in ARP header. */
3202                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3203                 ahdr->ar_op = htons(ARPOP_REQUEST);
3204                 arpreq = (u8 *)(ahdr + 1);
3205
3206                 /* The Unicast IPv4 address in 'tip' field. */
3207                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3208                 *(u32 *)arpreq = ifa->ifa_address;
3209
3210                 /* The mask for the relevant bits. */
3211                 pmConf->filters[i].mask[0] = 0x00;
3212                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3213                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3214                 pmConf->filters[i].mask[3] = 0x00;
3215                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3216                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3217                 in_dev_put(in_dev);
3218
3219                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3220                 i++;
3221         }
3222
3223 skip_arp:
3224         if (adapter->wol & WAKE_MAGIC)
3225                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3226
3227         pmConf->numFilters = i;
3228
3229         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3230         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3231                                                                   *pmConf));
3232         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3233                                                                  pmConf));
3234
3235         spin_lock_irqsave(&adapter->cmd_lock, flags);
3236         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3237                                VMXNET3_CMD_UPDATE_PMCFG);
3238         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3239
3240         pci_save_state(pdev);
3241         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3242                         adapter->wol);
3243         pci_disable_device(pdev);
3244         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3245
3246         return 0;
3247 }
3248
3249
3250 static int
3251 vmxnet3_resume(struct device *device)
3252 {
3253         int err, i = 0;
3254         unsigned long flags;
3255         struct pci_dev *pdev = to_pci_dev(device);
3256         struct net_device *netdev = pci_get_drvdata(pdev);
3257         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3258         struct Vmxnet3_PMConf *pmConf;
3259
3260         if (!netif_running(netdev))
3261                 return 0;
3262
3263         /* Destroy wake-up filters. */
3264         pmConf = adapter->pm_conf;
3265         memset(pmConf, 0, sizeof(*pmConf));
3266
3267         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3268         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3269                                                                   *pmConf));
3270         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3271                                                                  pmConf));
3272
3273         netif_device_attach(netdev);
3274         pci_set_power_state(pdev, PCI_D0);
3275         pci_restore_state(pdev);
3276         err = pci_enable_device_mem(pdev);
3277         if (err != 0)
3278                 return err;
3279
3280         pci_enable_wake(pdev, PCI_D0, 0);
3281
3282         spin_lock_irqsave(&adapter->cmd_lock, flags);
3283         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3284                                VMXNET3_CMD_UPDATE_PMCFG);
3285         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3286         vmxnet3_alloc_intr_resources(adapter);
3287         vmxnet3_request_irqs(adapter);
3288         for (i = 0; i < adapter->num_rx_queues; i++)
3289                 napi_enable(&adapter->rx_queue[i].napi);
3290         vmxnet3_enable_all_intrs(adapter);
3291
3292         return 0;
3293 }
3294
3295 static const struct dev_pm_ops vmxnet3_pm_ops = {
3296         .suspend = vmxnet3_suspend,
3297         .resume = vmxnet3_resume,
3298 };
3299 #endif
3300
3301 static struct pci_driver vmxnet3_driver = {
3302         .name           = vmxnet3_driver_name,
3303         .id_table       = vmxnet3_pciid_table,
3304         .probe          = vmxnet3_probe_device,
3305         .remove         = __devexit_p(vmxnet3_remove_device),
3306 #ifdef CONFIG_PM
3307         .driver.pm      = &vmxnet3_pm_ops,
3308 #endif
3309 };
3310
3311
3312 static int __init
3313 vmxnet3_init_module(void)
3314 {
3315         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3316                 VMXNET3_DRIVER_VERSION_REPORT);
3317         return pci_register_driver(&vmxnet3_driver);
3318 }
3319
3320 module_init(vmxnet3_init_module);
3321
3322
3323 static void
3324 vmxnet3_exit_module(void)
3325 {
3326         pci_unregister_driver(&vmxnet3_driver);
3327 }
3328
3329 module_exit(vmxnet3_exit_module);
3330
3331 MODULE_AUTHOR("VMware, Inc.");
3332 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3333 MODULE_LICENSE("GPL v2");
3334 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);