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                 netif_carrier_on(adapter->netdev);
 158
 159                 if (affectTxQueue) {
 160                         for (i = 0; i < adapter->num_tx_queues; i++)
 161                                 vmxnet3_tq_start(&adapter->tx_queue[i],
 162                                                  adapter);
 163                 }
 164         } else {
 165                 printk(KERN_INFO "%s: NIC Link is Down\n",
 166                        adapter->netdev->name);
 167                 netif_carrier_off(adapter->netdev);
 168
 169                 if (affectTxQueue) {
 170                         for (i = 0; i < adapter->num_tx_queues; i++)
 171                                 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
 172                 }
 173         }
 174 }
 175
 176 static void
 177 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 178 {
 179         int i;
 180         unsigned long flags;
 181         u32 events = le32_to_cpu(adapter->shared->ecr);
 182         if (!events)
 183                 return;
 184
 185         vmxnet3_ack_events(adapter, events);
 186
 187         /* Check if link state has changed */
 188         if (events & VMXNET3_ECR_LINK)
 189                 vmxnet3_check_link(adapter, true);
 190
 191         /* Check if there is an error on xmit/recv queues */
 192         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 193                 spin_lock_irqsave(&adapter->cmd_lock, flags);
 194                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 195                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 196                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
 197
 198                 for (i = 0; i < adapter->num_tx_queues; i++)
 199                         if (adapter->tqd_start[i].status.stopped)
 200                                 dev_err(&adapter->netdev->dev,
 201                                         "%s: tq[%d] error 0x%x\n",
 202                                         adapter->netdev->name, i, le32_to_cpu(
 203                                         adapter->tqd_start[i].status.error));
 204                 for (i = 0; i < adapter->num_rx_queues; i++)
 205                         if (adapter->rqd_start[i].status.stopped)
 206                                 dev_err(&adapter->netdev->dev,
 207                                         "%s: rq[%d] error 0x%x\n",
 208                                         adapter->netdev->name, i,
 209                                         adapter->rqd_start[i].status.error);
 210
 211                 schedule_work(&adapter->work);
 212         }
 213 }
 214
 215 #ifdef __BIG_ENDIAN_BITFIELD
 216 /*
 217  * The device expects the bitfields in shared structures to be written in
 218  * little endian. When CPU is big endian, the following routines are used to
 219  * correctly read and write into ABI.
 220  * The general technique used here is : double word bitfields are defined in
 221  * opposite order for big endian architecture. Then before reading them in
 222  * driver the complete double word is translated using le32_to_cpu. Similarly
 223  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 224  * double words into required format.
 225  * In order to avoid touching bits in shared structure more than once, temporary
 226  * descriptors are used. These are passed as srcDesc to following functions.
 227  */
 228 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 229                                 struct Vmxnet3_RxDesc *dstDesc)
 230 {
 231         u32 *src = (u32 *)srcDesc + 2;
 232         u32 *dst = (u32 *)dstDesc + 2;
 233         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 234         *dst = le32_to_cpu(*src);
 235         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 236 }
 237
 238 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 239                                struct Vmxnet3_TxDesc *dstDesc)
 240 {
 241         int i;
 242         u32 *src = (u32 *)(srcDesc + 1);
 243         u32 *dst = (u32 *)(dstDesc + 1);
 244
 245         /* Working backwards so that the gen bit is set at the end. */
 246         for (i = 2; i > 0; i--) {
 247                 src--;
 248                 dst--;
 249                 *dst = cpu_to_le32(*src);
 250         }
 251 }
 252
 253
 254 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 255                                 struct Vmxnet3_RxCompDesc *dstDesc)
 256 {
 257         int i = 0;
 258         u32 *src = (u32 *)srcDesc;
 259         u32 *dst = (u32 *)dstDesc;
 260         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 261                 *dst = le32_to_cpu(*src);
 262                 src++;
 263                 dst++;
 264         }
 265 }
 266
 267
 268 /* Used to read bitfield values from double words. */
 269 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 270 {
 271         u32 temp = le32_to_cpu(*bitfield);
 272         u32 mask = ((1 << size) - 1) << pos;
 273         temp &= mask;
 274         temp >>= pos;
 275         return temp;
 276 }
 277
 278
 279
 280 #endif  /* __BIG_ENDIAN_BITFIELD */
 281
 282 #ifdef __BIG_ENDIAN_BITFIELD
 283
 284 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 285                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 286                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 287 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 288                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 289                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 290 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 291                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 292                         VMXNET3_TCD_GEN_SIZE)
 293 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 294                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 295 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 296                         (dstrcd) = (tmp); \
 297                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 298                 } while (0)
 299 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 300                         (dstrxd) = (tmp); \
 301                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 302                 } while (0)
 303
 304 #else
 305
 306 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 307 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 308 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 309 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 310 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 311 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 312
 313 #endif /* __BIG_ENDIAN_BITFIELD  */
 314
 315
 316 static void
 317 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 318                      struct pci_dev *pdev)
 319 {
 320         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 321                 pci_unmap_single(pdev, tbi->dma_addr, tbi->len,
 322                                  PCI_DMA_TODEVICE);
 323         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 324                 pci_unmap_page(pdev, tbi->dma_addr, tbi->len,
 325                                PCI_DMA_TODEVICE);
 326         else
 327                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 328
 329         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 330 }
 331
 332
 333 static int
 334 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 335                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 336 {
 337         struct sk_buff *skb;
 338         int entries = 0;
 339
 340         /* no out of order completion */
 341         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 342         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 343
 344         skb = tq->buf_info[eop_idx].skb;
 345         BUG_ON(skb == NULL);
 346         tq->buf_info[eop_idx].skb = NULL;
 347
 348         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 349
 350         while (tq->tx_ring.next2comp != eop_idx) {
 351                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 352                                      pdev);
 353
 354                 /* update next2comp w/o tx_lock. Since we are marking more,
 355                  * instead of less, tx ring entries avail, the worst case is
 356                  * that the tx routine incorrectly re-queues a pkt due to
 357                  * insufficient tx ring entries.
 358                  */
 359                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 360                 entries++;
 361         }
 362
 363         dev_kfree_skb_any(skb);
 364         return entries;
 365 }
 366
 367
 368 static int
 369 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 370                         struct vmxnet3_adapter *adapter)
 371 {
 372         int completed = 0;
 373         union Vmxnet3_GenericDesc *gdesc;
 374
 375         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 376         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 377                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 378                                                &gdesc->tcd), tq, adapter->pdev,
 379                                                adapter);
 380
 381                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 382                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 383         }
 384
 385         if (completed) {
 386                 spin_lock(&tq->tx_lock);
 387                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 388                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 389                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 390                              netif_carrier_ok(adapter->netdev))) {
 391                         vmxnet3_tq_wake(tq, adapter);
 392                 }
 393                 spin_unlock(&tq->tx_lock);
 394         }
 395         return completed;
 396 }
 397
 398
 399 static void
 400 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 401                    struct vmxnet3_adapter *adapter)
 402 {
 403         int i;
 404
 405         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 406                 struct vmxnet3_tx_buf_info *tbi;
 407
 408                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 409
 410                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 411                 if (tbi->skb) {
 412                         dev_kfree_skb_any(tbi->skb);
 413                         tbi->skb = NULL;
 414                 }
 415                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 416         }
 417
 418         /* sanity check, verify all buffers are indeed unmapped and freed */
 419         for (i = 0; i < tq->tx_ring.size; i++) {
 420                 BUG_ON(tq->buf_info[i].skb != NULL ||
 421                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 422         }
 423
 424         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 425         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 426
 427         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 428         tq->comp_ring.next2proc = 0;
 429 }
 430
 431
 432 static void
 433 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 434                    struct vmxnet3_adapter *adapter)
 435 {
 436         if (tq->tx_ring.base) {
 437                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 438                                     sizeof(struct Vmxnet3_TxDesc),
 439                                     tq->tx_ring.base, tq->tx_ring.basePA);
 440                 tq->tx_ring.base = NULL;
 441         }
 442         if (tq->data_ring.base) {
 443                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 444                                     sizeof(struct Vmxnet3_TxDataDesc),
 445                                     tq->data_ring.base, tq->data_ring.basePA);
 446                 tq->data_ring.base = NULL;
 447         }
 448         if (tq->comp_ring.base) {
 449                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 450                                     sizeof(struct Vmxnet3_TxCompDesc),
 451                                     tq->comp_ring.base, tq->comp_ring.basePA);
 452                 tq->comp_ring.base = NULL;
 453         }
 454         kfree(tq->buf_info);
 455         tq->buf_info = NULL;
 456 }
 457
 458
 459 /* Destroy all tx queues */
 460 void
 461 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
 462 {
 463         int i;
 464
 465         for (i = 0; i < adapter->num_tx_queues; i++)
 466                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
 467 }
 468
 469
 470 static void
 471 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 472                 struct vmxnet3_adapter *adapter)
 473 {
 474         int i;
 475
 476         /* reset the tx ring contents to 0 and reset the tx ring states */
 477         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 478                sizeof(struct Vmxnet3_TxDesc));
 479         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 480         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 481
 482         memset(tq->data_ring.base, 0, tq->data_ring.size *
 483                sizeof(struct Vmxnet3_TxDataDesc));
 484
 485         /* reset the tx comp ring contents to 0 and reset comp ring states */
 486         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 487                sizeof(struct Vmxnet3_TxCompDesc));
 488         tq->comp_ring.next2proc = 0;
 489         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 490
 491         /* reset the bookkeeping data */
 492         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 493         for (i = 0; i < tq->tx_ring.size; i++)
 494                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 495
 496         /* stats are not reset */
 497 }
 498
 499
 500 static int
 501 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 502                   struct vmxnet3_adapter *adapter)
 503 {
 504         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 505                tq->comp_ring.base || tq->buf_info);
 506
 507         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 508                            * sizeof(struct Vmxnet3_TxDesc),
 509                            &tq->tx_ring.basePA);
 510         if (!tq->tx_ring.base) {
 511                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 512                        adapter->netdev->name);
 513                 goto err;
 514         }
 515
 516         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 517                              tq->data_ring.size *
 518                              sizeof(struct Vmxnet3_TxDataDesc),
 519                              &tq->data_ring.basePA);
 520         if (!tq->data_ring.base) {
 521                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 522                        adapter->netdev->name);
 523                 goto err;
 524         }
 525
 526         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 527                              tq->comp_ring.size *
 528                              sizeof(struct Vmxnet3_TxCompDesc),
 529                              &tq->comp_ring.basePA);
 530         if (!tq->comp_ring.base) {
 531                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 532                        adapter->netdev->name);
 533                 goto err;
 534         }
 535
 536         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 537                                GFP_KERNEL);
 538         if (!tq->buf_info)
 539                 goto err;
 540
 541         return 0;
 542
 543 err:
 544         vmxnet3_tq_destroy(tq, adapter);
 545         return -ENOMEM;
 546 }
 547
 548 static void
 549 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
 550 {
 551         int i;
 552
 553         for (i = 0; i < adapter->num_tx_queues; i++)
 554                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
 555 }
 556
 557 /*
 558  *    starting from ring->next2fill, allocate rx buffers for the given ring
 559  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 560  *    are allocated or allocation fails
 561  */
 562
 563 static int
 564 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 565                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 566 {
 567         int num_allocated = 0;
 568         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 569         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 570         u32 val;
 571
 572         while (num_allocated <= num_to_alloc) {
 573                 struct vmxnet3_rx_buf_info *rbi;
 574                 union Vmxnet3_GenericDesc *gd;
 575
 576                 rbi = rbi_base + ring->next2fill;
 577                 gd = ring->base + ring->next2fill;
 578
 579                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 580                         if (rbi->skb == NULL) {
 581                                 rbi->skb = dev_alloc_skb(rbi->len +
 582                                                          NET_IP_ALIGN);
 583                                 if (unlikely(rbi->skb == NULL)) {
 584                                         rq->stats.rx_buf_alloc_failure++;
 585                                         break;
 586                                 }
 587                                 rbi->skb->dev = adapter->netdev;
 588
 589                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 590                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 591                                                 rbi->skb->data, rbi->len,
 592                                                 PCI_DMA_FROMDEVICE);
 593                         } else {
 594                                 /* rx buffer skipped by the device */
 595                         }
 596                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 597                 } else {
 598                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 599                                rbi->len  != PAGE_SIZE);
 600
 601                         if (rbi->page == NULL) {
 602                                 rbi->page = alloc_page(GFP_ATOMIC);
 603                                 if (unlikely(rbi->page == NULL)) {
 604                                         rq->stats.rx_buf_alloc_failure++;
 605                                         break;
 606                                 }
 607                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 608                                                 rbi->page, 0, PAGE_SIZE,
 609                                                 PCI_DMA_FROMDEVICE);
 610                         } else {
 611                                 /* rx buffers skipped by the device */
 612                         }
 613                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 614                 }
 615
 616                 BUG_ON(rbi->dma_addr == 0);
 617                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 618                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 619                                            | val | rbi->len);
 620
 621                 /* Fill the last buffer but dont mark it ready, or else the
 622                  * device will think that the queue is full */
 623                 if (num_allocated == num_to_alloc)
 624                         break;
 625
 626                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 627                 num_allocated++;
 628                 vmxnet3_cmd_ring_adv_next2fill(ring);
 629         }
 630         rq->uncommitted[ring_idx] += num_allocated;
 631
 632         dev_dbg(&adapter->netdev->dev,
 633                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 634                 "%u, uncommitted %u\n", num_allocated, ring->next2fill,
 635                 ring->next2comp, rq->uncommitted[ring_idx]);
 636
 637         /* so that the device can distinguish a full ring and an empty ring */
 638         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 639
 640         return num_allocated;
 641 }
 642
 643
 644 static void
 645 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 646                     struct vmxnet3_rx_buf_info *rbi)
 647 {
 648         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 649                 skb_shinfo(skb)->nr_frags;
 650
 651         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 652
 653         __skb_frag_set_page(frag, rbi->page);
 654         frag->page_offset = 0;
 655         skb_frag_size_set(frag, rcd->len);
 656         skb->data_len += rcd->len;
 657         skb->truesize += PAGE_SIZE;
 658         skb_shinfo(skb)->nr_frags++;
 659 }
 660
 661
 662 static void
 663 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 664                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 665                 struct vmxnet3_adapter *adapter)
 666 {
 667         u32 dw2, len;
 668         unsigned long buf_offset;
 669         int i;
 670         union Vmxnet3_GenericDesc *gdesc;
 671         struct vmxnet3_tx_buf_info *tbi = NULL;
 672
 673         BUG_ON(ctx->copy_size > skb_headlen(skb));
 674
 675         /* use the previous gen bit for the SOP desc */
 676         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 677
 678         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 679         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 680
 681         /* no need to map the buffer if headers are copied */
 682         if (ctx->copy_size) {
 683                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 684                                         tq->tx_ring.next2fill *
 685                                         sizeof(struct Vmxnet3_TxDataDesc));
 686                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 687                 ctx->sop_txd->dword[3] = 0;
 688
 689                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 690                 tbi->map_type = VMXNET3_MAP_NONE;
 691
 692                 dev_dbg(&adapter->netdev->dev,
 693                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 694                         tq->tx_ring.next2fill,
 695                         le64_to_cpu(ctx->sop_txd->txd.addr),
 696                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 697                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 698
 699                 /* use the right gen for non-SOP desc */
 700                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 701         }
 702
 703         /* linear part can use multiple tx desc if it's big */
 704         len = skb_headlen(skb) - ctx->copy_size;
 705         buf_offset = ctx->copy_size;
 706         while (len) {
 707                 u32 buf_size;
 708
 709                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 710                         buf_size = len;
 711                         dw2 |= len;
 712                 } else {
 713                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 714                         /* spec says that for TxDesc.len, 0 == 2^14 */
 715                 }
 716
 717                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 718                 tbi->map_type = VMXNET3_MAP_SINGLE;
 719                 tbi->dma_addr = pci_map_single(adapter->pdev,
 720                                 skb->data + buf_offset, buf_size,
 721                                 PCI_DMA_TODEVICE);
 722
 723                 tbi->len = buf_size;
 724
 725                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 726                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 727
 728                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 729                 gdesc->dword[2] = cpu_to_le32(dw2);
 730                 gdesc->dword[3] = 0;
 731
 732                 dev_dbg(&adapter->netdev->dev,
 733                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 734                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 735                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 736                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 737                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 738
 739                 len -= buf_size;
 740                 buf_offset += buf_size;
 741         }
 742
 743         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 744                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 745                 u32 buf_size;
 746
 747                 buf_offset = 0;
 748                 len = skb_frag_size(frag);
 749                 while (len) {
 750                         tbi = tq->buf_info + tq->tx_ring.next2fill;
 751                         if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 752                                 buf_size = len;
 753                                 dw2 |= len;
 754                         } else {
 755                                 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 756                                 /* spec says that for TxDesc.len, 0 == 2^14 */
 757                         }
 758                         tbi->map_type = VMXNET3_MAP_PAGE;
 759                         tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
 760                                                          buf_offset, buf_size,
 761                                                          DMA_TO_DEVICE);
 762
 763                         tbi->len = buf_size;
 764
 765                         gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 766                         BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 767
 768                         gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 769                         gdesc->dword[2] = cpu_to_le32(dw2);
 770                         gdesc->dword[3] = 0;
 771
 772                         dev_dbg(&adapter->netdev->dev,
 773                                 "txd[%u]: 0x%llu %u %u\n",
 774                                 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 775                                 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 776                         vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 777                         dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 778
 779                         len -= buf_size;
 780                         buf_offset += buf_size;
 781                 }
 782         }
 783
 784         ctx->eop_txd = gdesc;
 785
 786         /* set the last buf_info for the pkt */
 787         tbi->skb = skb;
 788         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 789 }
 790
 791
 792 /* Init all tx queues */
 793 static void
 794 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 795 {
 796         int i;
 797
 798         for (i = 0; i < adapter->num_tx_queues; i++)
 799                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 800 }
 801
 802
 803 /*
 804  *    parse and copy relevant protocol headers:
 805  *      For a tso pkt, relevant headers are L2/3/4 including options
 806  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 807  *      if it's a TCP/UDP pkt
 808  *
 809  * Returns:
 810  *    -1:  error happens during parsing
 811  *     0:  protocol headers parsed, but too big to be copied
 812  *     1:  protocol headers parsed and copied
 813  *
 814  * Other effects:
 815  *    1. related *ctx fields are updated.
 816  *    2. ctx->copy_size is # of bytes copied
 817  *    3. the portion copied is guaranteed to be in the linear part
 818  *
 819  */
 820 static int
 821 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 822                            struct vmxnet3_tx_ctx *ctx,
 823                            struct vmxnet3_adapter *adapter)
 824 {
 825         struct Vmxnet3_TxDataDesc *tdd;
 826
 827         if (ctx->mss) { /* TSO */
 828                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 829                 ctx->l4_hdr_size = tcp_hdrlen(skb);
 830                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 831         } else {
 832                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 833                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 834
 835                         if (ctx->ipv4) {
 836                                 const struct iphdr *iph = ip_hdr(skb);
 837
 838                                 if (iph->protocol == IPPROTO_TCP)
 839                                         ctx->l4_hdr_size = tcp_hdrlen(skb);
 840                                 else if (iph->protocol == IPPROTO_UDP)
 841                                         ctx->l4_hdr_size = sizeof(struct udphdr);
 842                                 else
 843                                         ctx->l4_hdr_size = 0;
 844                         } else {
 845                                 /* for simplicity, don't copy L4 headers */
 846                                 ctx->l4_hdr_size = 0;
 847                         }
 848                         ctx->copy_size = min(ctx->eth_ip_hdr_size +
 849                                          ctx->l4_hdr_size, skb->len);
 850                 } else {
 851                         ctx->eth_ip_hdr_size = 0;
 852                         ctx->l4_hdr_size = 0;
 853                         /* copy as much as allowed */
 854                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 855                                              , skb_headlen(skb));
 856                 }
 857
 858                 /* make sure headers are accessible directly */
 859                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 860                         goto err;
 861         }
 862
 863         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 864                 tq->stats.oversized_hdr++;
 865                 ctx->copy_size = 0;
 866                 return 0;
 867         }
 868
 869         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 870
 871         memcpy(tdd->data, skb->data, ctx->copy_size);
 872         dev_dbg(&adapter->netdev->dev,
 873                 "copy %u bytes to dataRing[%u]\n",
 874                 ctx->copy_size, tq->tx_ring.next2fill);
 875         return 1;
 876
 877 err:
 878         return -1;
 879 }
 880
 881
 882 static void
 883 vmxnet3_prepare_tso(struct sk_buff *skb,
 884                     struct vmxnet3_tx_ctx *ctx)
 885 {
 886         struct tcphdr *tcph = tcp_hdr(skb);
 887
 888         if (ctx->ipv4) {
 889                 struct iphdr *iph = ip_hdr(skb);
 890
 891                 iph->check = 0;
 892                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 893                                                  IPPROTO_TCP, 0);
 894         } else {
 895                 struct ipv6hdr *iph = ipv6_hdr(skb);
 896
 897                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 898                                                IPPROTO_TCP, 0);
 899         }
 900 }
 901
 902 static int txd_estimate(const struct sk_buff *skb)
 903 {
 904         int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 905         int i;
 906
 907         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 908                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 909
 910                 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
 911         }
 912         return count;
 913 }
 914
 915 /*
 916  * Transmits a pkt thru a given tq
 917  * Returns:
 918  *    NETDEV_TX_OK:      descriptors are setup successfully
 919  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 920  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 921  *
 922  * Side-effects:
 923  *    1. tx ring may be changed
 924  *    2. tq stats may be updated accordingly
 925  *    3. shared->txNumDeferred may be updated
 926  */
 927
 928 static int
 929 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 930                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 931 {
 932         int ret;
 933         u32 count;
 934         unsigned long flags;
 935         struct vmxnet3_tx_ctx ctx;
 936         union Vmxnet3_GenericDesc *gdesc;
 937 #ifdef __BIG_ENDIAN_BITFIELD
 938         /* Use temporary descriptor to avoid touching bits multiple times */
 939         union Vmxnet3_GenericDesc tempTxDesc;
 940 #endif
 941
 942         count = txd_estimate(skb);
 943
 944         ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
 945
 946         ctx.mss = skb_shinfo(skb)->gso_size;
 947         if (ctx.mss) {
 948                 if (skb_header_cloned(skb)) {
 949                         if (unlikely(pskb_expand_head(skb, 0, 0,
 950                                                       GFP_ATOMIC) != 0)) {
 951                                 tq->stats.drop_tso++;
 952                                 goto drop_pkt;
 953                         }
 954                         tq->stats.copy_skb_header++;
 955                 }
 956                 vmxnet3_prepare_tso(skb, &ctx);
 957         } else {
 958                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 959
 960                         /* non-tso pkts must not use more than
 961                          * VMXNET3_MAX_TXD_PER_PKT entries
 962                          */
 963                         if (skb_linearize(skb) != 0) {
 964                                 tq->stats.drop_too_many_frags++;
 965                                 goto drop_pkt;
 966                         }
 967                         tq->stats.linearized++;
 968
 969                         /* recalculate the # of descriptors to use */
 970                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 971                 }
 972         }
 973
 974         spin_lock_irqsave(&tq->tx_lock, flags);
 975
 976         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 977                 tq->stats.tx_ring_full++;
 978                 dev_dbg(&adapter->netdev->dev,
 979                         "tx queue stopped on %s, next2comp %u"
 980                         " next2fill %u\n", adapter->netdev->name,
 981                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 982
 983                 vmxnet3_tq_stop(tq, adapter);
 984                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 985                 return NETDEV_TX_BUSY;
 986         }
 987
 988
 989         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 990         if (ret >= 0) {
 991                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 992                 /* hdrs parsed, check against other limits */
 993                 if (ctx.mss) {
 994                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 995                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 996                                 goto hdr_too_big;
 997                         }
 998                 } else {
 999                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1000                                 if (unlikely(ctx.eth_ip_hdr_size +
1001                                              skb->csum_offset >
1002                                              VMXNET3_MAX_CSUM_OFFSET)) {
1003                                         goto hdr_too_big;
1004                                 }
1005                         }
1006                 }
1007         } else {
1008                 tq->stats.drop_hdr_inspect_err++;
1009                 goto unlock_drop_pkt;
1010         }
1011
1012         /* fill tx descs related to addr & len */
1013         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
1014
1015         /* setup the EOP desc */
1016         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1017
1018         /* setup the SOP desc */
1019 #ifdef __BIG_ENDIAN_BITFIELD
1020         gdesc = &tempTxDesc;
1021         gdesc->dword[2] = ctx.sop_txd->dword[2];
1022         gdesc->dword[3] = ctx.sop_txd->dword[3];
1023 #else
1024         gdesc = ctx.sop_txd;
1025 #endif
1026         if (ctx.mss) {
1027                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1028                 gdesc->txd.om = VMXNET3_OM_TSO;
1029                 gdesc->txd.msscof = ctx.mss;
1030                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1031                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1032         } else {
1033                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1034                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1035                         gdesc->txd.om = VMXNET3_OM_CSUM;
1036                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1037                                             skb->csum_offset;
1038                 } else {
1039                         gdesc->txd.om = 0;
1040                         gdesc->txd.msscof = 0;
1041                 }
1042                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1043         }
1044
1045         if (vlan_tx_tag_present(skb)) {
1046                 gdesc->txd.ti = 1;
1047                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1048         }
1049
1050         /* finally flips the GEN bit of the SOP desc. */
1051         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1052                                                   VMXNET3_TXD_GEN);
1053 #ifdef __BIG_ENDIAN_BITFIELD
1054         /* Finished updating in bitfields of Tx Desc, so write them in original
1055          * place.
1056          */
1057         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1058                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1059         gdesc = ctx.sop_txd;
1060 #endif
1061         dev_dbg(&adapter->netdev->dev,
1062                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1063                 (u32)(ctx.sop_txd -
1064                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1065                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1066
1067         spin_unlock_irqrestore(&tq->tx_lock, flags);
1068
1069         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1070                                         le32_to_cpu(tq->shared->txThreshold)) {
1071                 tq->shared->txNumDeferred = 0;
1072                 VMXNET3_WRITE_BAR0_REG(adapter,
1073                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1074                                        tq->tx_ring.next2fill);
1075         }
1076
1077         return NETDEV_TX_OK;
1078
1079 hdr_too_big:
1080         tq->stats.drop_oversized_hdr++;
1081 unlock_drop_pkt:
1082         spin_unlock_irqrestore(&tq->tx_lock, flags);
1083 drop_pkt:
1084         tq->stats.drop_total++;
1085         dev_kfree_skb(skb);
1086         return NETDEV_TX_OK;
1087 }
1088
1089
1090 static netdev_tx_t
1091 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1092 {
1093         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1094
1095         BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1096         return vmxnet3_tq_xmit(skb,
1097                                &adapter->tx_queue[skb->queue_mapping],
1098                                adapter, netdev);
1099 }
1100
1101
1102 static void
1103 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1104                 struct sk_buff *skb,
1105                 union Vmxnet3_GenericDesc *gdesc)
1106 {
1107         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1108                 /* typical case: TCP/UDP over IP and both csums are correct */
1109                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1110                                                         VMXNET3_RCD_CSUM_OK) {
1111                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1112                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1113                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1114                         BUG_ON(gdesc->rcd.frg);
1115                 } else {
1116                         if (gdesc->rcd.csum) {
1117                                 skb->csum = htons(gdesc->rcd.csum);
1118                                 skb->ip_summed = CHECKSUM_PARTIAL;
1119                         } else {
1120                                 skb_checksum_none_assert(skb);
1121                         }
1122                 }
1123         } else {
1124                 skb_checksum_none_assert(skb);
1125         }
1126 }
1127
1128
1129 static void
1130 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1131                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1132 {
1133         rq->stats.drop_err++;
1134         if (!rcd->fcs)
1135                 rq->stats.drop_fcs++;
1136
1137         rq->stats.drop_total++;
1138
1139         /*
1140          * We do not unmap and chain the rx buffer to the skb.
1141          * We basically pretend this buffer is not used and will be recycled
1142          * by vmxnet3_rq_alloc_rx_buf()
1143          */
1144
1145         /*
1146          * ctx->skb may be NULL if this is the first and the only one
1147          * desc for the pkt
1148          */
1149         if (ctx->skb)
1150                 dev_kfree_skb_irq(ctx->skb);
1151
1152         ctx->skb = NULL;
1153 }
1154
1155
1156 static int
1157 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1158                        struct vmxnet3_adapter *adapter, int quota)
1159 {
1160         static const u32 rxprod_reg[2] = {
1161                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1162         };
1163         u32 num_rxd = 0;
1164         bool skip_page_frags = false;
1165         struct Vmxnet3_RxCompDesc *rcd;
1166         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1167 #ifdef __BIG_ENDIAN_BITFIELD
1168         struct Vmxnet3_RxDesc rxCmdDesc;
1169         struct Vmxnet3_RxCompDesc rxComp;
1170 #endif
1171         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1172                           &rxComp);
1173         while (rcd->gen == rq->comp_ring.gen) {
1174                 struct vmxnet3_rx_buf_info *rbi;
1175                 struct sk_buff *skb, *new_skb = NULL;
1176                 struct page *new_page = NULL;
1177                 int num_to_alloc;
1178                 struct Vmxnet3_RxDesc *rxd;
1179                 u32 idx, ring_idx;
1180                 struct vmxnet3_cmd_ring *ring = NULL;
1181                 if (num_rxd >= quota) {
1182                         /* we may stop even before we see the EOP desc of
1183                          * the current pkt
1184                          */
1185                         break;
1186                 }
1187                 num_rxd++;
1188                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1189                 idx = rcd->rxdIdx;
1190                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1191                 ring = rq->rx_ring + ring_idx;
1192                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1193                                   &rxCmdDesc);
1194                 rbi = rq->buf_info[ring_idx] + idx;
1195
1196                 BUG_ON(rxd->addr != rbi->dma_addr ||
1197                        rxd->len != rbi->len);
1198
1199                 if (unlikely(rcd->eop && rcd->err)) {
1200                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1201                         goto rcd_done;
1202                 }
1203
1204                 if (rcd->sop) { /* first buf of the pkt */
1205                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1206                                rcd->rqID != rq->qid);
1207
1208                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1209                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1210
1211                         if (unlikely(rcd->len == 0)) {
1212                                 /* Pretend the rx buffer is skipped. */
1213                                 BUG_ON(!(rcd->sop && rcd->eop));
1214                                 dev_dbg(&adapter->netdev->dev,
1215                                         "rxRing[%u][%u] 0 length\n",
1216                                         ring_idx, idx);
1217                                 goto rcd_done;
1218                         }
1219
1220                         skip_page_frags = false;
1221                         ctx->skb = rbi->skb;
1222                         new_skb = dev_alloc_skb(rbi->len + NET_IP_ALIGN);
1223                         if (new_skb == NULL) {
1224                                 /* Skb allocation failed, do not handover this
1225                                  * skb to stack. Reuse it. Drop the existing pkt
1226                                  */
1227                                 rq->stats.rx_buf_alloc_failure++;
1228                                 ctx->skb = NULL;
1229                                 rq->stats.drop_total++;
1230                                 skip_page_frags = true;
1231                                 goto rcd_done;
1232                         }
1233
1234                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1235                                          PCI_DMA_FROMDEVICE);
1236
1237                         skb_put(ctx->skb, rcd->len);
1238
1239                         /* Immediate refill */
1240                         new_skb->dev = adapter->netdev;
1241                         skb_reserve(new_skb, NET_IP_ALIGN);
1242                         rbi->skb = new_skb;
1243                         rbi->dma_addr = pci_map_single(adapter->pdev,
1244                                                        rbi->skb->data, rbi->len,
1245                                                        PCI_DMA_FROMDEVICE);
1246                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1247                         rxd->len = rbi->len;
1248
1249                 } else {
1250                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1251
1252                         /* non SOP buffer must be type 1 in most cases */
1253                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1254                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1255
1256                         /* If an sop buffer was dropped, skip all
1257                          * following non-sop fragments. They will be reused.
1258                          */
1259                         if (skip_page_frags)
1260                                 goto rcd_done;
1261
1262                         new_page = alloc_page(GFP_ATOMIC);
1263                         if (unlikely(new_page == NULL)) {
1264                                 /* Replacement page frag could not be allocated.
1265                                  * Reuse this page. Drop the pkt and free the
1266                                  * skb which contained this page as a frag. Skip
1267                                  * processing all the following non-sop frags.
1268                                  */
1269                                 rq->stats.rx_buf_alloc_failure++;
1270                                 dev_kfree_skb(ctx->skb);
1271                                 ctx->skb = NULL;
1272                                 skip_page_frags = true;
1273                                 goto rcd_done;
1274                         }
1275
1276                         if (rcd->len) {
1277                                 pci_unmap_page(adapter->pdev,
1278                                                rbi->dma_addr, rbi->len,
1279                                                PCI_DMA_FROMDEVICE);
1280
1281                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1282                         }
1283
1284                         /* Immediate refill */
1285                         rbi->page = new_page;
1286                         rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1287                                                      0, PAGE_SIZE,
1288                                                      PCI_DMA_FROMDEVICE);
1289                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1290                         rxd->len = rbi->len;
1291                 }
1292
1293
1294                 skb = ctx->skb;
1295                 if (rcd->eop) {
1296                         skb->len += skb->data_len;
1297
1298                         vmxnet3_rx_csum(adapter, skb,
1299                                         (union Vmxnet3_GenericDesc *)rcd);
1300                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1301
1302                         if (unlikely(rcd->ts))
1303                                 __vlan_hwaccel_put_tag(skb, rcd->tci);
1304
1305                         if (adapter->netdev->features & NETIF_F_LRO)
1306                                 netif_receive_skb(skb);
1307                         else
1308                                 napi_gro_receive(&rq->napi, skb);
1309
1310                         ctx->skb = NULL;
1311                 }
1312
1313 rcd_done:
1314                 /* device may have skipped some rx descs */
1315                 ring->next2comp = idx;
1316                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1317                 ring = rq->rx_ring + ring_idx;
1318                 while (num_to_alloc) {
1319                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1320                                           &rxCmdDesc);
1321                         BUG_ON(!rxd->addr);
1322
1323                         /* Recv desc is ready to be used by the device */
1324                         rxd->gen = ring->gen;
1325                         vmxnet3_cmd_ring_adv_next2fill(ring);
1326                         num_to_alloc--;
1327                 }
1328
1329                 /* if needed, update the register */
1330                 if (unlikely(rq->shared->updateRxProd)) {
1331                         VMXNET3_WRITE_BAR0_REG(adapter,
1332                                                rxprod_reg[ring_idx] + rq->qid * 8,
1333                                                ring->next2fill);
1334                         rq->uncommitted[ring_idx] = 0;
1335                 }
1336
1337                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1338                 vmxnet3_getRxComp(rcd,
1339                                   &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1340         }
1341
1342         return num_rxd;
1343 }
1344
1345
1346 static void
1347 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1348                    struct vmxnet3_adapter *adapter)
1349 {
1350         u32 i, ring_idx;
1351         struct Vmxnet3_RxDesc *rxd;
1352
1353         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1354                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1355 #ifdef __BIG_ENDIAN_BITFIELD
1356                         struct Vmxnet3_RxDesc rxDesc;
1357 #endif
1358                         vmxnet3_getRxDesc(rxd,
1359                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1360
1361                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1362                                         rq->buf_info[ring_idx][i].skb) {
1363                                 pci_unmap_single(adapter->pdev, rxd->addr,
1364                                                  rxd->len, PCI_DMA_FROMDEVICE);
1365                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1366                                 rq->buf_info[ring_idx][i].skb = NULL;
1367                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1368                                         rq->buf_info[ring_idx][i].page) {
1369                                 pci_unmap_page(adapter->pdev, rxd->addr,
1370                                                rxd->len, PCI_DMA_FROMDEVICE);
1371                                 put_page(rq->buf_info[ring_idx][i].page);
1372                                 rq->buf_info[ring_idx][i].page = NULL;
1373                         }
1374                 }
1375
1376                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1377                 rq->rx_ring[ring_idx].next2fill =
1378                                         rq->rx_ring[ring_idx].next2comp = 0;
1379                 rq->uncommitted[ring_idx] = 0;
1380         }
1381
1382         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1383         rq->comp_ring.next2proc = 0;
1384 }
1385
1386
1387 static void
1388 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1389 {
1390         int i;
1391
1392         for (i = 0; i < adapter->num_rx_queues; i++)
1393                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1394 }
1395
1396
1397 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1398                         struct vmxnet3_adapter *adapter)
1399 {
1400         int i;
1401         int j;
1402
1403         /* all rx buffers must have already been freed */
1404         for (i = 0; i < 2; i++) {
1405                 if (rq->buf_info[i]) {
1406                         for (j = 0; j < rq->rx_ring[i].size; j++)
1407                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1408                 }
1409         }
1410
1411
1412         kfree(rq->buf_info[0]);
1413
1414         for (i = 0; i < 2; i++) {
1415                 if (rq->rx_ring[i].base) {
1416                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1417                                             * sizeof(struct Vmxnet3_RxDesc),
1418                                             rq->rx_ring[i].base,
1419                                             rq->rx_ring[i].basePA);
1420                         rq->rx_ring[i].base = NULL;
1421                 }
1422                 rq->buf_info[i] = NULL;
1423         }
1424
1425         if (rq->comp_ring.base) {
1426                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1427                                     sizeof(struct Vmxnet3_RxCompDesc),
1428                                     rq->comp_ring.base, rq->comp_ring.basePA);
1429                 rq->comp_ring.base = NULL;
1430         }
1431 }
1432
1433
1434 static int
1435 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1436                 struct vmxnet3_adapter  *adapter)
1437 {
1438         int i;
1439
1440         /* initialize buf_info */
1441         for (i = 0; i < rq->rx_ring[0].size; i++) {
1442
1443                 /* 1st buf for a pkt is skbuff */
1444                 if (i % adapter->rx_buf_per_pkt == 0) {
1445                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1446                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1447                 } else { /* subsequent bufs for a pkt is frag */
1448                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1449                         rq->buf_info[0][i].len = PAGE_SIZE;
1450                 }
1451         }
1452         for (i = 0; i < rq->rx_ring[1].size; i++) {
1453                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1454                 rq->buf_info[1][i].len = PAGE_SIZE;
1455         }
1456
1457         /* reset internal state and allocate buffers for both rings */
1458         for (i = 0; i < 2; i++) {
1459                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1460                 rq->uncommitted[i] = 0;
1461
1462                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1463                        sizeof(struct Vmxnet3_RxDesc));
1464                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1465         }
1466         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1467                                     adapter) == 0) {
1468                 /* at least has 1 rx buffer for the 1st ring */
1469                 return -ENOMEM;
1470         }
1471         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1472
1473         /* reset the comp ring */
1474         rq->comp_ring.next2proc = 0;
1475         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1476                sizeof(struct Vmxnet3_RxCompDesc));
1477         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1478
1479         /* reset rxctx */
1480         rq->rx_ctx.skb = NULL;
1481
1482         /* stats are not reset */
1483         return 0;
1484 }
1485
1486
1487 static int
1488 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1489 {
1490         int i, err = 0;
1491
1492         for (i = 0; i < adapter->num_rx_queues; i++) {
1493                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1494                 if (unlikely(err)) {
1495                         dev_err(&adapter->netdev->dev, "%s: failed to "
1496                                 "initialize rx queue%i\n",
1497                                 adapter->netdev->name, i);
1498                         break;
1499                 }
1500         }
1501         return err;
1502
1503 }
1504
1505
1506 static int
1507 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1508 {
1509         int i;
1510         size_t sz;
1511         struct vmxnet3_rx_buf_info *bi;
1512
1513         for (i = 0; i < 2; i++) {
1514
1515                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1516                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1517                                                         &rq->rx_ring[i].basePA);
1518                 if (!rq->rx_ring[i].base) {
1519                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1520                                adapter->netdev->name, i);
1521                         goto err;
1522                 }
1523         }
1524
1525         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1526         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1527                                                   &rq->comp_ring.basePA);
1528         if (!rq->comp_ring.base) {
1529                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1530                        adapter->netdev->name);
1531                 goto err;
1532         }
1533
1534         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1535                                                    rq->rx_ring[1].size);
1536         bi = kzalloc(sz, GFP_KERNEL);
1537         if (!bi)
1538                 goto err;
1539
1540         rq->buf_info[0] = bi;
1541         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1542
1543         return 0;
1544
1545 err:
1546         vmxnet3_rq_destroy(rq, adapter);
1547         return -ENOMEM;
1548 }
1549
1550
1551 static int
1552 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1553 {
1554         int i, err = 0;
1555
1556         for (i = 0; i < adapter->num_rx_queues; i++) {
1557                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1558                 if (unlikely(err)) {
1559                         dev_err(&adapter->netdev->dev,
1560                                 "%s: failed to create rx queue%i\n",
1561                                 adapter->netdev->name, i);
1562                         goto err_out;
1563                 }
1564         }
1565         return err;
1566 err_out:
1567         vmxnet3_rq_destroy_all(adapter);
1568         return err;
1569
1570 }
1571
1572 /* Multiple queue aware polling function for tx and rx */
1573
1574 static int
1575 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1576 {
1577         int rcd_done = 0, i;
1578         if (unlikely(adapter->shared->ecr))
1579                 vmxnet3_process_events(adapter);
1580         for (i = 0; i < adapter->num_tx_queues; i++)
1581                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1582
1583         for (i = 0; i < adapter->num_rx_queues; i++)
1584                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1585                                                    adapter, budget);
1586         return rcd_done;
1587 }
1588
1589
1590 static int
1591 vmxnet3_poll(struct napi_struct *napi, int budget)
1592 {
1593         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1594                                           struct vmxnet3_rx_queue, napi);
1595         int rxd_done;
1596
1597         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1598
1599         if (rxd_done < budget) {
1600                 napi_complete(napi);
1601                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1602         }
1603         return rxd_done;
1604 }
1605
1606 /*
1607  * NAPI polling function for MSI-X mode with multiple Rx queues
1608  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1609  */
1610
1611 static int
1612 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1613 {
1614         struct vmxnet3_rx_queue *rq = container_of(napi,
1615                                                 struct vmxnet3_rx_queue, napi);
1616         struct vmxnet3_adapter *adapter = rq->adapter;
1617         int rxd_done;
1618
1619         /* When sharing interrupt with corresponding tx queue, process
1620          * tx completions in that queue as well
1621          */
1622         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1623                 struct vmxnet3_tx_queue *tq =
1624                                 &adapter->tx_queue[rq - adapter->rx_queue];
1625                 vmxnet3_tq_tx_complete(tq, adapter);
1626         }
1627
1628         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1629
1630         if (rxd_done < budget) {
1631                 napi_complete(napi);
1632                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1633         }
1634         return rxd_done;
1635 }
1636
1637
1638 #ifdef CONFIG_PCI_MSI
1639
1640 /*
1641  * Handle completion interrupts on tx queues
1642  * Returns whether or not the intr is handled
1643  */
1644
1645 static irqreturn_t
1646 vmxnet3_msix_tx(int irq, void *data)
1647 {
1648         struct vmxnet3_tx_queue *tq = data;
1649         struct vmxnet3_adapter *adapter = tq->adapter;
1650
1651         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1652                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1653
1654         /* Handle the case where only one irq is allocate for all tx queues */
1655         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1656                 int i;
1657                 for (i = 0; i < adapter->num_tx_queues; i++) {
1658                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1659                         vmxnet3_tq_tx_complete(txq, adapter);
1660                 }
1661         } else {
1662                 vmxnet3_tq_tx_complete(tq, adapter);
1663         }
1664         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1665
1666         return IRQ_HANDLED;
1667 }
1668
1669
1670 /*
1671  * Handle completion interrupts on rx queues. Returns whether or not the
1672  * intr is handled
1673  */
1674
1675 static irqreturn_t
1676 vmxnet3_msix_rx(int irq, void *data)
1677 {
1678         struct vmxnet3_rx_queue *rq = data;
1679         struct vmxnet3_adapter *adapter = rq->adapter;
1680
1681         /* disable intr if needed */
1682         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1683                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1684         napi_schedule(&rq->napi);
1685
1686         return IRQ_HANDLED;
1687 }
1688
1689 /*
1690  *----------------------------------------------------------------------------
1691  *
1692  * vmxnet3_msix_event --
1693  *
1694  *    vmxnet3 msix event intr handler
1695  *
1696  * Result:
1697  *    whether or not the intr is handled
1698  *
1699  *----------------------------------------------------------------------------
1700  */
1701
1702 static irqreturn_t
1703 vmxnet3_msix_event(int irq, void *data)
1704 {
1705         struct net_device *dev = data;
1706         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1707
1708         /* disable intr if needed */
1709         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1710                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1711
1712         if (adapter->shared->ecr)
1713                 vmxnet3_process_events(adapter);
1714
1715         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1716
1717         return IRQ_HANDLED;
1718 }
1719
1720 #endif /* CONFIG_PCI_MSI  */
1721
1722
1723 /* Interrupt handler for vmxnet3  */
1724 static irqreturn_t
1725 vmxnet3_intr(int irq, void *dev_id)
1726 {
1727         struct net_device *dev = dev_id;
1728         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1729
1730         if (adapter->intr.type == VMXNET3_IT_INTX) {
1731                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1732                 if (unlikely(icr == 0))
1733                         /* not ours */
1734                         return IRQ_NONE;
1735         }
1736
1737
1738         /* disable intr if needed */
1739         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1740                 vmxnet3_disable_all_intrs(adapter);
1741
1742         napi_schedule(&adapter->rx_queue[0].napi);
1743
1744         return IRQ_HANDLED;
1745 }
1746
1747 #ifdef CONFIG_NET_POLL_CONTROLLER
1748
1749 /* netpoll callback. */
1750 static void
1751 vmxnet3_netpoll(struct net_device *netdev)
1752 {
1753         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1754
1755         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1756                 vmxnet3_disable_all_intrs(adapter);
1757
1758         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1759         vmxnet3_enable_all_intrs(adapter);
1760
1761 }
1762 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1763
1764 static int
1765 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1766 {
1767         struct vmxnet3_intr *intr = &adapter->intr;
1768         int err = 0, i;
1769         int vector = 0;
1770
1771 #ifdef CONFIG_PCI_MSI
1772         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1773                 for (i = 0; i < adapter->num_tx_queues; i++) {
1774                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1775                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1776                                         adapter->netdev->name, vector);
1777                                 err = request_irq(
1778                                               intr->msix_entries[vector].vector,
1779                                               vmxnet3_msix_tx, 0,
1780                                               adapter->tx_queue[i].name,
1781                                               &adapter->tx_queue[i]);
1782                         } else {
1783                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1784                                         adapter->netdev->name, vector);
1785                         }
1786                         if (err) {
1787                                 dev_err(&adapter->netdev->dev,
1788                                         "Failed to request irq for MSIX, %s, "
1789                                         "error %d\n",
1790                                         adapter->tx_queue[i].name, err);
1791                                 return err;
1792                         }
1793
1794                         /* Handle the case where only 1 MSIx was allocated for
1795                          * all tx queues */
1796                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1797                                 for (; i < adapter->num_tx_queues; i++)
1798                                         adapter->tx_queue[i].comp_ring.intr_idx
1799                                                                 = vector;
1800                                 vector++;
1801                                 break;
1802                         } else {
1803                                 adapter->tx_queue[i].comp_ring.intr_idx
1804                                                                 = vector++;
1805                         }
1806                 }
1807                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1808                         vector = 0;
1809
1810                 for (i = 0; i < adapter->num_rx_queues; i++) {
1811                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1812                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1813                                         adapter->netdev->name, vector);
1814                         else
1815                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1816                                         adapter->netdev->name, vector);
1817                         err = request_irq(intr->msix_entries[vector].vector,
1818                                           vmxnet3_msix_rx, 0,
1819                                           adapter->rx_queue[i].name,
1820                                           &(adapter->rx_queue[i]));
1821                         if (err) {
1822                                 printk(KERN_ERR "Failed to request irq for MSIX"
1823                                        ", %s, error %d\n",
1824                                        adapter->rx_queue[i].name, err);
1825                                 return err;
1826                         }
1827
1828                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1829                 }
1830
1831                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1832                         adapter->netdev->name, vector);
1833                 err = request_irq(intr->msix_entries[vector].vector,
1834                                   vmxnet3_msix_event, 0,
1835                                   intr->event_msi_vector_name, adapter->netdev);
1836                 intr->event_intr_idx = vector;
1837
1838         } else if (intr->type == VMXNET3_IT_MSI) {
1839                 adapter->num_rx_queues = 1;
1840                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1841                                   adapter->netdev->name, adapter->netdev);
1842         } else {
1843 #endif
1844                 adapter->num_rx_queues = 1;
1845                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1846                                   IRQF_SHARED, adapter->netdev->name,
1847                                   adapter->netdev);
1848 #ifdef CONFIG_PCI_MSI
1849         }
1850 #endif
1851         intr->num_intrs = vector + 1;
1852         if (err) {
1853                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1854                        ":%d\n", adapter->netdev->name, intr->type, err);
1855         } else {
1856                 /* Number of rx queues will not change after this */
1857                 for (i = 0; i < adapter->num_rx_queues; i++) {
1858                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1859                         rq->qid = i;
1860                         rq->qid2 = i + adapter->num_rx_queues;
1861                 }
1862
1863
1864
1865                 /* init our intr settings */
1866                 for (i = 0; i < intr->num_intrs; i++)
1867                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1868                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1869                         adapter->intr.event_intr_idx = 0;
1870                         for (i = 0; i < adapter->num_tx_queues; i++)
1871                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1872                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1873                 }
1874
1875                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1876                        "allocated\n", adapter->netdev->name, intr->type,
1877                        intr->mask_mode, intr->num_intrs);
1878         }
1879
1880         return err;
1881 }
1882
1883
1884 static void
1885 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1886 {
1887         struct vmxnet3_intr *intr = &adapter->intr;
1888         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1889
1890         switch (intr->type) {
1891 #ifdef CONFIG_PCI_MSI
1892         case VMXNET3_IT_MSIX:
1893         {
1894                 int i, vector = 0;
1895
1896                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1897                         for (i = 0; i < adapter->num_tx_queues; i++) {
1898                                 free_irq(intr->msix_entries[vector++].vector,
1899                                          &(adapter->tx_queue[i]));
1900                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1901                                         break;
1902                         }
1903                 }
1904
1905                 for (i = 0; i < adapter->num_rx_queues; i++) {
1906                         free_irq(intr->msix_entries[vector++].vector,
1907                                  &(adapter->rx_queue[i]));
1908                 }
1909
1910                 free_irq(intr->msix_entries[vector].vector,
1911                          adapter->netdev);
1912                 BUG_ON(vector >= intr->num_intrs);
1913                 break;
1914         }
1915 #endif
1916         case VMXNET3_IT_MSI:
1917                 free_irq(adapter->pdev->irq, adapter->netdev);
1918                 break;
1919         case VMXNET3_IT_INTX:
1920                 free_irq(adapter->pdev->irq, adapter->netdev);
1921                 break;
1922         default:
1923                 BUG();
1924         }
1925 }
1926
1927
1928 static void
1929 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1930 {
1931         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1932         u16 vid;
1933
1934         /* allow untagged pkts */
1935         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1936
1937         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1938                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1939 }
1940
1941
1942 static int
1943 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1944 {
1945         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1946
1947         if (!(netdev->flags & IFF_PROMISC)) {
1948                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1949                 unsigned long flags;
1950
1951                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1952                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1953                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1954                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1955                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1956         }
1957
1958         set_bit(vid, adapter->active_vlans);
1959
1960         return 0;
1961 }
1962
1963
1964 static int
1965 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1966 {
1967         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1968
1969         if (!(netdev->flags & IFF_PROMISC)) {
1970                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1971                 unsigned long flags;
1972
1973                 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1974                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1975                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1976                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1977                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1978         }
1979
1980         clear_bit(vid, adapter->active_vlans);
1981
1982         return 0;
1983 }
1984
1985
1986 static u8 *
1987 vmxnet3_copy_mc(struct net_device *netdev)
1988 {
1989         u8 *buf = NULL;
1990         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
1991
1992         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
1993         if (sz <= 0xffff) {
1994                 /* We may be called with BH disabled */
1995                 buf = kmalloc(sz, GFP_ATOMIC);
1996                 if (buf) {
1997                         struct netdev_hw_addr *ha;
1998                         int i = 0;
1999
2000                         netdev_for_each_mc_addr(ha, netdev)
2001                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2002                                        ETH_ALEN);
2003                 }
2004         }
2005         return buf;
2006 }
2007
2008
2009 static void
2010 vmxnet3_set_mc(struct net_device *netdev)
2011 {
2012         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2013         unsigned long flags;
2014         struct Vmxnet3_RxFilterConf *rxConf =
2015                                         &adapter->shared->devRead.rxFilterConf;
2016         u8 *new_table = NULL;
2017         u32 new_mode = VMXNET3_RXM_UCAST;
2018
2019         if (netdev->flags & IFF_PROMISC) {
2020                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2021                 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2022
2023                 new_mode |= VMXNET3_RXM_PROMISC;
2024         } else {
2025                 vmxnet3_restore_vlan(adapter);
2026         }
2027
2028         if (netdev->flags & IFF_BROADCAST)
2029                 new_mode |= VMXNET3_RXM_BCAST;
2030
2031         if (netdev->flags & IFF_ALLMULTI)
2032                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2033         else
2034                 if (!netdev_mc_empty(netdev)) {
2035                         new_table = vmxnet3_copy_mc(netdev);
2036                         if (new_table) {
2037                                 new_mode |= VMXNET3_RXM_MCAST;
2038                                 rxConf->mfTableLen = cpu_to_le16(
2039                                         netdev_mc_count(netdev) * ETH_ALEN);
2040                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2041                                                     new_table));
2042                         } else {
2043                                 printk(KERN_INFO "%s: failed to copy mcast list"
2044                                        ", setting ALL_MULTI\n", netdev->name);
2045                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2046                         }
2047                 }
2048
2049
2050         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2051                 rxConf->mfTableLen = 0;
2052                 rxConf->mfTablePA = 0;
2053         }
2054
2055         spin_lock_irqsave(&adapter->cmd_lock, flags);
2056         if (new_mode != rxConf->rxMode) {
2057                 rxConf->rxMode = cpu_to_le32(new_mode);
2058                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2059                                        VMXNET3_CMD_UPDATE_RX_MODE);
2060                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2061                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2062         }
2063
2064         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2065                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2066         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2067
2068         kfree(new_table);
2069 }
2070
2071 void
2072 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2073 {
2074         int i;
2075
2076         for (i = 0; i < adapter->num_rx_queues; i++)
2077                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2078 }
2079
2080
2081 /*
2082  *   Set up driver_shared based on settings in adapter.
2083  */
2084
2085 static void
2086 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2087 {
2088         struct Vmxnet3_DriverShared *shared = adapter->shared;
2089         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2090         struct Vmxnet3_TxQueueConf *tqc;
2091         struct Vmxnet3_RxQueueConf *rqc;
2092         int i;
2093
2094         memset(shared, 0, sizeof(*shared));
2095
2096         /* driver settings */
2097         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2098         devRead->misc.driverInfo.version = cpu_to_le32(
2099                                                 VMXNET3_DRIVER_VERSION_NUM);
2100         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2101                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2102         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2103         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2104                                 *((u32 *)&devRead->misc.driverInfo.gos));
2105         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2106         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2107
2108         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2109         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2110
2111         /* set up feature flags */
2112         if (adapter->netdev->features & NETIF_F_RXCSUM)
2113                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2114
2115         if (adapter->netdev->features & NETIF_F_LRO) {
2116                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2117                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2118         }
2119         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2120                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2121
2122         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2123         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2124         devRead->misc.queueDescLen = cpu_to_le32(
2125                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2126                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2127
2128         /* tx queue settings */
2129         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2130         for (i = 0; i < adapter->num_tx_queues; i++) {
2131                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2132                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2133                 tqc = &adapter->tqd_start[i].conf;
2134                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2135                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2136                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2137                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2138                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2139                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2140                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2141                 tqc->ddLen          = cpu_to_le32(
2142                                         sizeof(struct vmxnet3_tx_buf_info) *
2143                                         tqc->txRingSize);
2144                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2145         }
2146
2147         /* rx queue settings */
2148         devRead->misc.numRxQueues = adapter->num_rx_queues;
2149         for (i = 0; i < adapter->num_rx_queues; i++) {
2150                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2151                 rqc = &adapter->rqd_start[i].conf;
2152                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2153                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2154                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2155                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2156                                                         rq->buf_info));
2157                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2158                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2159                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2160                 rqc->ddLen           = cpu_to_le32(
2161                                         sizeof(struct vmxnet3_rx_buf_info) *
2162                                         (rqc->rxRingSize[0] +
2163                                          rqc->rxRingSize[1]));
2164                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2165         }
2166
2167 #ifdef VMXNET3_RSS
2168         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2169
2170         if (adapter->rss) {
2171                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2172                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2173                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2174                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2175                                     UPT1_RSS_HASH_TYPE_IPV4 |
2176                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2177                                     UPT1_RSS_HASH_TYPE_IPV6;
2178                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2179                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2180                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2181                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2182                 for (i = 0; i < rssConf->indTableSize; i++)
2183                         rssConf->indTable[i] = ethtool_rxfh_indir_default(
2184                                 i, adapter->num_rx_queues);
2185
2186                 devRead->rssConfDesc.confVer = 1;
2187                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2188                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2189         }
2190
2191 #endif /* VMXNET3_RSS */
2192
2193         /* intr settings */
2194         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2195                                      VMXNET3_IMM_AUTO;
2196         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2197         for (i = 0; i < adapter->intr.num_intrs; i++)
2198                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2199
2200         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2201         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2202
2203         /* rx filter settings */
2204         devRead->rxFilterConf.rxMode = 0;
2205         vmxnet3_restore_vlan(adapter);
2206         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2207
2208         /* the rest are already zeroed */
2209 }
2210
2211
2212 int
2213 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2214 {
2215         int err, i;
2216         u32 ret;
2217         unsigned long flags;
2218
2219         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2220                 " ring sizes %u %u %u\n", adapter->netdev->name,
2221                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2222                 adapter->tx_queue[0].tx_ring.size,
2223                 adapter->rx_queue[0].rx_ring[0].size,
2224                 adapter->rx_queue[0].rx_ring[1].size);
2225
2226         vmxnet3_tq_init_all(adapter);
2227         err = vmxnet3_rq_init_all(adapter);
2228         if (err) {
2229                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2230                        adapter->netdev->name, err);
2231                 goto rq_err;
2232         }
2233
2234         err = vmxnet3_request_irqs(adapter);
2235         if (err) {
2236                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2237                        adapter->netdev->name, err);
2238                 goto irq_err;
2239         }
2240
2241         vmxnet3_setup_driver_shared(adapter);
2242
2243         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2244                                adapter->shared_pa));
2245         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2246                                adapter->shared_pa));
2247         spin_lock_irqsave(&adapter->cmd_lock, flags);
2248         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2249                                VMXNET3_CMD_ACTIVATE_DEV);
2250         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2251         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2252
2253         if (ret != 0) {
2254                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2255                        adapter->netdev->name, ret);
2256                 err = -EINVAL;
2257                 goto activate_err;
2258         }
2259
2260         for (i = 0; i < adapter->num_rx_queues; i++) {
2261                 VMXNET3_WRITE_BAR0_REG(adapter,
2262                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2263                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2264                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2265                                 (i * VMXNET3_REG_ALIGN)),
2266                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2267         }
2268
2269         /* Apply the rx filter settins last. */
2270         vmxnet3_set_mc(adapter->netdev);
2271
2272         /*
2273          * Check link state when first activating device. It will start the
2274          * tx queue if the link is up.
2275          */
2276         vmxnet3_check_link(adapter, true);
2277         for (i = 0; i < adapter->num_rx_queues; i++)
2278                 napi_enable(&adapter->rx_queue[i].napi);
2279         vmxnet3_enable_all_intrs(adapter);
2280         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2281         return 0;
2282
2283 activate_err:
2284         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2285         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2286         vmxnet3_free_irqs(adapter);
2287 irq_err:
2288 rq_err:
2289         /* free up buffers we allocated */
2290         vmxnet3_rq_cleanup_all(adapter);
2291         return err;
2292 }
2293
2294
2295 void
2296 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2297 {
2298         unsigned long flags;
2299         spin_lock_irqsave(&adapter->cmd_lock, flags);
2300         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2301         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2302 }
2303
2304
2305 int
2306 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2307 {
2308         int i;
2309         unsigned long flags;
2310         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2311                 return 0;
2312
2313
2314         spin_lock_irqsave(&adapter->cmd_lock, flags);
2315         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2316                                VMXNET3_CMD_QUIESCE_DEV);
2317         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2318         vmxnet3_disable_all_intrs(adapter);
2319
2320         for (i = 0; i < adapter->num_rx_queues; i++)
2321                 napi_disable(&adapter->rx_queue[i].napi);
2322         netif_tx_disable(adapter->netdev);
2323         adapter->link_speed = 0;
2324         netif_carrier_off(adapter->netdev);
2325
2326         vmxnet3_tq_cleanup_all(adapter);
2327         vmxnet3_rq_cleanup_all(adapter);
2328         vmxnet3_free_irqs(adapter);
2329         return 0;
2330 }
2331
2332
2333 static void
2334 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2335 {
2336         u32 tmp;
2337
2338         tmp = *(u32 *)mac;
2339         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2340
2341         tmp = (mac[5] << 8) | mac[4];
2342         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2343 }
2344
2345
2346 static int
2347 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2348 {
2349         struct sockaddr *addr = p;
2350         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2351
2352         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2353         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2354
2355         return 0;
2356 }
2357
2358
2359 /* ==================== initialization and cleanup routines ============ */
2360
2361 static int
2362 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2363 {
2364         int err;
2365         unsigned long mmio_start, mmio_len;
2366         struct pci_dev *pdev = adapter->pdev;
2367
2368         err = pci_enable_device(pdev);
2369         if (err) {
2370                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2371                        pci_name(pdev), err);
2372                 return err;
2373         }
2374
2375         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2376                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2377                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2378                                "for adapter %s\n", pci_name(pdev));
2379                         err = -EIO;
2380                         goto err_set_mask;
2381                 }
2382                 *dma64 = true;
2383         } else {
2384                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2385                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2386                                "%s\n",  pci_name(pdev));
2387                         err = -EIO;
2388                         goto err_set_mask;
2389                 }
2390                 *dma64 = false;
2391         }
2392
2393         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2394                                            vmxnet3_driver_name);
2395         if (err) {
2396                 printk(KERN_ERR "Failed to request region for adapter %s: "
2397                        "error %d\n", pci_name(pdev), err);
2398                 goto err_set_mask;
2399         }
2400
2401         pci_set_master(pdev);
2402
2403         mmio_start = pci_resource_start(pdev, 0);
2404         mmio_len = pci_resource_len(pdev, 0);
2405         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2406         if (!adapter->hw_addr0) {
2407                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2408                        pci_name(pdev));
2409                 err = -EIO;
2410                 goto err_ioremap;
2411         }
2412
2413         mmio_start = pci_resource_start(pdev, 1);
2414         mmio_len = pci_resource_len(pdev, 1);
2415         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2416         if (!adapter->hw_addr1) {
2417                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2418                        pci_name(pdev));
2419                 err = -EIO;
2420                 goto err_bar1;
2421         }
2422         return 0;
2423
2424 err_bar1:
2425         iounmap(adapter->hw_addr0);
2426 err_ioremap:
2427         pci_release_selected_regions(pdev, (1 << 2) - 1);
2428 err_set_mask:
2429         pci_disable_device(pdev);
2430         return err;
2431 }
2432
2433
2434 static void
2435 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2436 {
2437         BUG_ON(!adapter->pdev);
2438
2439         iounmap(adapter->hw_addr0);
2440         iounmap(adapter->hw_addr1);
2441         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2442         pci_disable_device(adapter->pdev);
2443 }
2444
2445
2446 static void
2447 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2448 {
2449         size_t sz, i, ring0_size, ring1_size, comp_size;
2450         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2451
2452
2453         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2454                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2455                 adapter->skb_buf_size = adapter->netdev->mtu +
2456                                         VMXNET3_MAX_ETH_HDR_SIZE;
2457                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2458                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2459
2460                 adapter->rx_buf_per_pkt = 1;
2461         } else {
2462                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2463                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2464                                             VMXNET3_MAX_ETH_HDR_SIZE;
2465                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2466         }
2467
2468         /*
2469          * for simplicity, force the ring0 size to be a multiple of
2470          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2471          */
2472         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2473         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2474         ring0_size = (ring0_size + sz - 1) / sz * sz;
2475         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2476                            sz * sz);
2477         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2478         comp_size = ring0_size + ring1_size;
2479
2480         for (i = 0; i < adapter->num_rx_queues; i++) {
2481                 rq = &adapter->rx_queue[i];
2482                 rq->rx_ring[0].size = ring0_size;
2483                 rq->rx_ring[1].size = ring1_size;
2484                 rq->comp_ring.size = comp_size;
2485         }
2486 }
2487
2488
2489 int
2490 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2491                       u32 rx_ring_size, u32 rx_ring2_size)
2492 {
2493         int err = 0, i;
2494
2495         for (i = 0; i < adapter->num_tx_queues; i++) {
2496                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2497                 tq->tx_ring.size   = tx_ring_size;
2498                 tq->data_ring.size = tx_ring_size;
2499                 tq->comp_ring.size = tx_ring_size;
2500                 tq->shared = &adapter->tqd_start[i].ctrl;
2501                 tq->stopped = true;
2502                 tq->adapter = adapter;
2503                 tq->qid = i;
2504                 err = vmxnet3_tq_create(tq, adapter);
2505                 /*
2506                  * Too late to change num_tx_queues. We cannot do away with
2507                  * lesser number of queues than what we asked for
2508                  */
2509                 if (err)
2510                         goto queue_err;
2511         }
2512
2513         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2514         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2515         vmxnet3_adjust_rx_ring_size(adapter);
2516         for (i = 0; i < adapter->num_rx_queues; i++) {
2517                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2518                 /* qid and qid2 for rx queues will be assigned later when num
2519                  * of rx queues is finalized after allocating intrs */
2520                 rq->shared = &adapter->rqd_start[i].ctrl;
2521                 rq->adapter = adapter;
2522                 err = vmxnet3_rq_create(rq, adapter);
2523                 if (err) {
2524                         if (i == 0) {
2525                                 printk(KERN_ERR "Could not allocate any rx"
2526                                        "queues. Aborting.\n");
2527                                 goto queue_err;
2528                         } else {
2529                                 printk(KERN_INFO "Number of rx queues changed "
2530                                        "to : %d.\n", i);
2531                                 adapter->num_rx_queues = i;
2532                                 err = 0;
2533                                 break;
2534                         }
2535                 }
2536         }
2537         return err;
2538 queue_err:
2539         vmxnet3_tq_destroy_all(adapter);
2540         return err;
2541 }
2542
2543 static int
2544 vmxnet3_open(struct net_device *netdev)
2545 {
2546         struct vmxnet3_adapter *adapter;
2547         int err, i;
2548
2549         adapter = netdev_priv(netdev);
2550
2551         for (i = 0; i < adapter->num_tx_queues; i++)
2552                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2553
2554         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2555                                     VMXNET3_DEF_RX_RING_SIZE,
2556                                     VMXNET3_DEF_RX_RING_SIZE);
2557         if (err)
2558                 goto queue_err;
2559
2560         err = vmxnet3_activate_dev(adapter);
2561         if (err)
2562                 goto activate_err;
2563
2564         return 0;
2565
2566 activate_err:
2567         vmxnet3_rq_destroy_all(adapter);
2568         vmxnet3_tq_destroy_all(adapter);
2569 queue_err:
2570         return err;
2571 }
2572
2573
2574 static int
2575 vmxnet3_close(struct net_device *netdev)
2576 {
2577         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2578
2579         /*
2580          * Reset_work may be in the middle of resetting the device, wait for its
2581          * completion.
2582          */
2583         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2584                 msleep(1);
2585
2586         vmxnet3_quiesce_dev(adapter);
2587
2588         vmxnet3_rq_destroy_all(adapter);
2589         vmxnet3_tq_destroy_all(adapter);
2590
2591         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2592
2593
2594         return 0;
2595 }
2596
2597
2598 void
2599 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2600 {
2601         int i;
2602
2603         /*
2604          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2605          * vmxnet3_close() will deadlock.
2606          */
2607         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2608
2609         /* we need to enable NAPI, otherwise dev_close will deadlock */
2610         for (i = 0; i < adapter->num_rx_queues; i++)
2611                 napi_enable(&adapter->rx_queue[i].napi);
2612         dev_close(adapter->netdev);
2613 }
2614
2615
2616 static int
2617 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2618 {
2619         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2620         int err = 0;
2621
2622         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2623                 return -EINVAL;
2624
2625         netdev->mtu = new_mtu;
2626
2627         /*
2628          * Reset_work may be in the middle of resetting the device, wait for its
2629          * completion.
2630          */
2631         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2632                 msleep(1);
2633
2634         if (netif_running(netdev)) {
2635                 vmxnet3_quiesce_dev(adapter);
2636                 vmxnet3_reset_dev(adapter);
2637
2638                 /* we need to re-create the rx queue based on the new mtu */
2639                 vmxnet3_rq_destroy_all(adapter);
2640                 vmxnet3_adjust_rx_ring_size(adapter);
2641                 err = vmxnet3_rq_create_all(adapter);
2642                 if (err) {
2643                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2644                                 " error %d. Closing it.\n", netdev->name, err);
2645                         goto out;
2646                 }
2647
2648                 err = vmxnet3_activate_dev(adapter);
2649                 if (err) {
2650                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2651                                 "Closing it\n", netdev->name, err);
2652                         goto out;
2653                 }
2654         }
2655
2656 out:
2657         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2658         if (err)
2659                 vmxnet3_force_close(adapter);
2660
2661         return err;
2662 }
2663
2664
2665 static void
2666 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2667 {
2668         struct net_device *netdev = adapter->netdev;
2669
2670         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2671                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2672                 NETIF_F_HW_VLAN_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2673                 NETIF_F_LRO;
2674         if (dma64)
2675                 netdev->hw_features |= NETIF_F_HIGHDMA;
2676         netdev->vlan_features = netdev->hw_features &
2677                                 ~(NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
2678         netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_FILTER;
2679
2680         netdev_info(adapter->netdev,
2681                 "features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2682                 dma64 ? " highDMA" : "");
2683 }
2684
2685
2686 static void
2687 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2688 {
2689         u32 tmp;
2690
2691         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2692         *(u32 *)mac = tmp;
2693
2694         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2695         mac[4] = tmp & 0xff;
2696         mac[5] = (tmp >> 8) & 0xff;
2697 }
2698
2699 #ifdef CONFIG_PCI_MSI
2700
2701 /*
2702  * Enable MSIx vectors.
2703  * Returns :
2704  *      0 on successful enabling of required vectors,
2705  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2706  *       could be enabled.
2707  *      number of vectors which can be enabled otherwise (this number is smaller
2708  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2709  */
2710
2711 static int
2712 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2713                              int vectors)
2714 {
2715         int err = 0, vector_threshold;
2716         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2717
2718         while (vectors >= vector_threshold) {
2719                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2720                                       vectors);
2721                 if (!err) {
2722                         adapter->intr.num_intrs = vectors;
2723                         return 0;
2724                 } else if (err < 0) {
2725                         netdev_err(adapter->netdev,
2726                                    "Failed to enable MSI-X, error: %d\n", err);
2727                         vectors = 0;
2728                 } else if (err < vector_threshold) {
2729                         break;
2730                 } else {
2731                         /* If fails to enable required number of MSI-x vectors
2732                          * try enabling minimum number of vectors required.
2733                          */
2734                         netdev_err(adapter->netdev,
2735                                    "Failed to enable %d MSI-X, trying %d instead\n",
2736                                     vectors, vector_threshold);
2737                         vectors = vector_threshold;
2738                 }
2739         }
2740
2741         netdev_info(adapter->netdev,
2742                     "Number of MSI-X interrupts which can be allocated are lower than min threshold required.\n");
2743         return err;
2744 }
2745
2746
2747 #endif /* CONFIG_PCI_MSI */
2748
2749 static void
2750 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2751 {
2752         u32 cfg;
2753         unsigned long flags;
2754
2755         /* intr settings */
2756         spin_lock_irqsave(&adapter->cmd_lock, flags);
2757         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2758                                VMXNET3_CMD_GET_CONF_INTR);
2759         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2760         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2761         adapter->intr.type = cfg & 0x3;
2762         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2763
2764         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2765                 adapter->intr.type = VMXNET3_IT_MSIX;
2766         }
2767
2768 #ifdef CONFIG_PCI_MSI
2769         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2770                 int vector, err = 0;
2771
2772                 adapter->intr.num_intrs = (adapter->share_intr ==
2773                                            VMXNET3_INTR_TXSHARE) ? 1 :
2774                                            adapter->num_tx_queues;
2775                 adapter->intr.num_intrs += (adapter->share_intr ==
2776                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2777                                            adapter->num_rx_queues;
2778                 adapter->intr.num_intrs += 1;           /* for link event */
2779
2780                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2781                                            VMXNET3_LINUX_MIN_MSIX_VECT
2782                                            ? adapter->intr.num_intrs :
2783                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2784
2785                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2786                         adapter->intr.msix_entries[vector].entry = vector;
2787
2788                 err = vmxnet3_acquire_msix_vectors(adapter,
2789                                                    adapter->intr.num_intrs);
2790                 /* If we cannot allocate one MSIx vector per queue
2791                  * then limit the number of rx queues to 1
2792                  */
2793                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2794                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2795                             || adapter->num_rx_queues != 1) {
2796                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2797                                 printk(KERN_ERR "Number of rx queues : 1\n");
2798                                 adapter->num_rx_queues = 1;
2799                                 adapter->intr.num_intrs =
2800                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2801                         }
2802                         return;
2803                 }
2804                 if (!err)
2805                         return;
2806
2807                 /* If we cannot allocate MSIx vectors use only one rx queue */
2808                 netdev_info(adapter->netdev,
2809                             "Failed to enable MSI-X, error %d . Limiting #rx queues to 1, try MSI.\n",
2810                             err);
2811
2812                 adapter->intr.type = VMXNET3_IT_MSI;
2813         }
2814
2815         if (adapter->intr.type == VMXNET3_IT_MSI) {
2816                 int err;
2817                 err = pci_enable_msi(adapter->pdev);
2818                 if (!err) {
2819                         adapter->num_rx_queues = 1;
2820                         adapter->intr.num_intrs = 1;
2821                         return;
2822                 }
2823         }
2824 #endif /* CONFIG_PCI_MSI */
2825
2826         adapter->num_rx_queues = 1;
2827         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2828         adapter->intr.type = VMXNET3_IT_INTX;
2829
2830         /* INT-X related setting */
2831         adapter->intr.num_intrs = 1;
2832 }
2833
2834
2835 static void
2836 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2837 {
2838         if (adapter->intr.type == VMXNET3_IT_MSIX)
2839                 pci_disable_msix(adapter->pdev);
2840         else if (adapter->intr.type == VMXNET3_IT_MSI)
2841                 pci_disable_msi(adapter->pdev);
2842         else
2843                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2844 }
2845
2846
2847 static void
2848 vmxnet3_tx_timeout(struct net_device *netdev)
2849 {
2850         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2851         adapter->tx_timeout_count++;
2852
2853         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2854         schedule_work(&adapter->work);
2855         netif_wake_queue(adapter->netdev);
2856 }
2857
2858
2859 static void
2860 vmxnet3_reset_work(struct work_struct *data)
2861 {
2862         struct vmxnet3_adapter *adapter;
2863
2864         adapter = container_of(data, struct vmxnet3_adapter, work);
2865
2866         /* if another thread is resetting the device, no need to proceed */
2867         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2868                 return;
2869
2870         /* if the device is closed, we must leave it alone */
2871         rtnl_lock();
2872         if (netif_running(adapter->netdev)) {
2873                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2874                 vmxnet3_quiesce_dev(adapter);
2875                 vmxnet3_reset_dev(adapter);
2876                 vmxnet3_activate_dev(adapter);
2877         } else {
2878                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2879         }
2880         rtnl_unlock();
2881
2882         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2883 }
2884
2885
2886 static int
2887 vmxnet3_probe_device(struct pci_dev *pdev,
2888                      const struct pci_device_id *id)
2889 {
2890         static const struct net_device_ops vmxnet3_netdev_ops = {
2891                 .ndo_open = vmxnet3_open,
2892                 .ndo_stop = vmxnet3_close,
2893                 .ndo_start_xmit = vmxnet3_xmit_frame,
2894                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2895                 .ndo_change_mtu = vmxnet3_change_mtu,
2896                 .ndo_set_features = vmxnet3_set_features,
2897                 .ndo_get_stats64 = vmxnet3_get_stats64,
2898                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2899                 .ndo_set_rx_mode = vmxnet3_set_mc,
2900                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2901                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2902 #ifdef CONFIG_NET_POLL_CONTROLLER
2903                 .ndo_poll_controller = vmxnet3_netpoll,
2904 #endif
2905         };
2906         int err;
2907         bool dma64 = false; /* stupid gcc */
2908         u32 ver;
2909         struct net_device *netdev;
2910         struct vmxnet3_adapter *adapter;
2911         u8 mac[ETH_ALEN];
2912         int size;
2913         int num_tx_queues;
2914         int num_rx_queues;
2915
2916         if (!pci_msi_enabled())
2917                 enable_mq = 0;
2918
2919 #ifdef VMXNET3_RSS
2920         if (enable_mq)
2921                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2922                                     (int)num_online_cpus());
2923         else
2924 #endif
2925                 num_rx_queues = 1;
2926         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2927
2928         if (enable_mq)
2929                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2930                                     (int)num_online_cpus());
2931         else
2932                 num_tx_queues = 1;
2933
2934         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2935         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2936                                    max(num_tx_queues, num_rx_queues));
2937         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2938                num_tx_queues, num_rx_queues);
2939
2940         if (!netdev)
2941                 return -ENOMEM;
2942
2943         pci_set_drvdata(pdev, netdev);
2944         adapter = netdev_priv(netdev);
2945         adapter->netdev = netdev;
2946         adapter->pdev = pdev;
2947
2948         spin_lock_init(&adapter->cmd_lock);
2949         adapter->shared = pci_alloc_consistent(adapter->pdev,
2950                                                sizeof(struct Vmxnet3_DriverShared),
2951                                                &adapter->shared_pa);
2952         if (!adapter->shared) {
2953                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2954                        pci_name(pdev));
2955                 err = -ENOMEM;
2956                 goto err_alloc_shared;
2957         }
2958
2959         adapter->num_rx_queues = num_rx_queues;
2960         adapter->num_tx_queues = num_tx_queues;
2961
2962         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2963         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2964         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2965                                                   &adapter->queue_desc_pa);
2966
2967         if (!adapter->tqd_start) {
2968                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2969                        pci_name(pdev));
2970                 err = -ENOMEM;
2971                 goto err_alloc_queue_desc;
2972         }
2973         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2974                                                             adapter->num_tx_queues);
2975
2976         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
2977         if (adapter->pm_conf == NULL) {
2978                 err = -ENOMEM;
2979                 goto err_alloc_pm;
2980         }
2981
2982 #ifdef VMXNET3_RSS
2983
2984         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
2985         if (adapter->rss_conf == NULL) {
2986                 err = -ENOMEM;
2987                 goto err_alloc_rss;
2988         }
2989 #endif /* VMXNET3_RSS */
2990
2991         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
2992         if (err < 0)
2993                 goto err_alloc_pci;
2994
2995         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
2996         if (ver & 1) {
2997                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
2998         } else {
2999                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
3000                        " %s\n", ver, pci_name(pdev));
3001                 err = -EBUSY;
3002                 goto err_ver;
3003         }
3004
3005         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3006         if (ver & 1) {
3007                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3008         } else {
3009                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
3010                        "adapter %s\n", ver, pci_name(pdev));
3011                 err = -EBUSY;
3012                 goto err_ver;
3013         }
3014
3015         SET_NETDEV_DEV(netdev, &pdev->dev);
3016         vmxnet3_declare_features(adapter, dma64);
3017
3018         adapter->dev_number = atomic_read(&devices_found);
3019
3020         adapter->share_intr = irq_share_mode;
3021         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3022             adapter->num_tx_queues != adapter->num_rx_queues)
3023                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3024
3025         vmxnet3_alloc_intr_resources(adapter);
3026
3027 #ifdef VMXNET3_RSS
3028         if (adapter->num_rx_queues > 1 &&
3029             adapter->intr.type == VMXNET3_IT_MSIX) {
3030                 adapter->rss = true;
3031                 printk(KERN_INFO "RSS is enabled.\n");
3032         } else {
3033                 adapter->rss = false;
3034         }
3035 #endif
3036
3037         vmxnet3_read_mac_addr(adapter, mac);
3038         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3039
3040         netdev->netdev_ops = &vmxnet3_netdev_ops;
3041         vmxnet3_set_ethtool_ops(netdev);
3042         netdev->watchdog_timeo = 5 * HZ;
3043
3044         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3045         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3046
3047         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3048                 int i;
3049                 for (i = 0; i < adapter->num_rx_queues; i++) {
3050                         netif_napi_add(adapter->netdev,
3051                                        &adapter->rx_queue[i].napi,
3052                                        vmxnet3_poll_rx_only, 64);
3053                 }
3054         } else {
3055                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3056                                vmxnet3_poll, 64);
3057         }
3058
3059         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3060         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3061
3062         netif_carrier_off(netdev);
3063         err = register_netdev(netdev);
3064
3065         if (err) {
3066                 printk(KERN_ERR "Failed to register adapter %s\n",
3067                        pci_name(pdev));
3068                 goto err_register;
3069         }
3070
3071         vmxnet3_check_link(adapter, false);
3072         atomic_inc(&devices_found);
3073         return 0;
3074
3075 err_register:
3076         vmxnet3_free_intr_resources(adapter);
3077 err_ver:
3078         vmxnet3_free_pci_resources(adapter);
3079 err_alloc_pci:
3080 #ifdef VMXNET3_RSS
3081         kfree(adapter->rss_conf);
3082 err_alloc_rss:
3083 #endif
3084         kfree(adapter->pm_conf);
3085 err_alloc_pm:
3086         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3087                             adapter->queue_desc_pa);
3088 err_alloc_queue_desc:
3089         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3090                             adapter->shared, adapter->shared_pa);
3091 err_alloc_shared:
3092         pci_set_drvdata(pdev, NULL);
3093         free_netdev(netdev);
3094         return err;
3095 }
3096
3097
3098 static void
3099 vmxnet3_remove_device(struct pci_dev *pdev)
3100 {
3101         struct net_device *netdev = pci_get_drvdata(pdev);
3102         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3103         int size = 0;
3104         int num_rx_queues;
3105
3106 #ifdef VMXNET3_RSS
3107         if (enable_mq)
3108                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3109                                     (int)num_online_cpus());
3110         else
3111 #endif
3112                 num_rx_queues = 1;
3113         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3114
3115         cancel_work_sync(&adapter->work);
3116
3117         unregister_netdev(netdev);
3118
3119         vmxnet3_free_intr_resources(adapter);
3120         vmxnet3_free_pci_resources(adapter);
3121 #ifdef VMXNET3_RSS
3122         kfree(adapter->rss_conf);
3123 #endif
3124         kfree(adapter->pm_conf);
3125
3126         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3127         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3128         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3129                             adapter->queue_desc_pa);
3130         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3131                             adapter->shared, adapter->shared_pa);
3132         free_netdev(netdev);
3133 }
3134
3135
3136 #ifdef CONFIG_PM
3137
3138 static int
3139 vmxnet3_suspend(struct device *device)
3140 {
3141         struct pci_dev *pdev = to_pci_dev(device);
3142         struct net_device *netdev = pci_get_drvdata(pdev);
3143         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3144         struct Vmxnet3_PMConf *pmConf;
3145         struct ethhdr *ehdr;
3146         struct arphdr *ahdr;
3147         u8 *arpreq;
3148         struct in_device *in_dev;
3149         struct in_ifaddr *ifa;
3150         unsigned long flags;
3151         int i = 0;
3152
3153         if (!netif_running(netdev))
3154                 return 0;
3155
3156         for (i = 0; i < adapter->num_rx_queues; i++)
3157                 napi_disable(&adapter->rx_queue[i].napi);
3158
3159         vmxnet3_disable_all_intrs(adapter);
3160         vmxnet3_free_irqs(adapter);
3161         vmxnet3_free_intr_resources(adapter);
3162
3163         netif_device_detach(netdev);
3164         netif_tx_stop_all_queues(netdev);
3165
3166         /* Create wake-up filters. */
3167         pmConf = adapter->pm_conf;
3168         memset(pmConf, 0, sizeof(*pmConf));
3169
3170         if (adapter->wol & WAKE_UCAST) {
3171                 pmConf->filters[i].patternSize = ETH_ALEN;
3172                 pmConf->filters[i].maskSize = 1;
3173                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3174                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3175
3176                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3177                 i++;
3178         }
3179
3180         if (adapter->wol & WAKE_ARP) {
3181                 in_dev = in_dev_get(netdev);
3182                 if (!in_dev)
3183                         goto skip_arp;
3184
3185                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3186                 if (!ifa)
3187                         goto skip_arp;
3188
3189                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3190                         sizeof(struct arphdr) +         /* ARP header */
3191                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3192                         2 * sizeof(u32);        /*2 IPv4 addresses */
3193                 pmConf->filters[i].maskSize =
3194                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3195
3196                 /* ETH_P_ARP in Ethernet header. */
3197                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3198                 ehdr->h_proto = htons(ETH_P_ARP);
3199
3200                 /* ARPOP_REQUEST in ARP header. */
3201                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3202                 ahdr->ar_op = htons(ARPOP_REQUEST);
3203                 arpreq = (u8 *)(ahdr + 1);
3204
3205                 /* The Unicast IPv4 address in 'tip' field. */
3206                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3207                 *(u32 *)arpreq = ifa->ifa_address;
3208
3209                 /* The mask for the relevant bits. */
3210                 pmConf->filters[i].mask[0] = 0x00;
3211                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3212                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3213                 pmConf->filters[i].mask[3] = 0x00;
3214                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3215                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3216                 in_dev_put(in_dev);
3217
3218                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3219                 i++;
3220         }
3221
3222 skip_arp:
3223         if (adapter->wol & WAKE_MAGIC)
3224                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3225
3226         pmConf->numFilters = i;
3227
3228         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3229         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3230                                                                   *pmConf));
3231         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3232                                                                  pmConf));
3233
3234         spin_lock_irqsave(&adapter->cmd_lock, flags);
3235         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3236                                VMXNET3_CMD_UPDATE_PMCFG);
3237         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3238
3239         pci_save_state(pdev);
3240         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3241                         adapter->wol);
3242         pci_disable_device(pdev);
3243         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3244
3245         return 0;
3246 }
3247
3248
3249 static int
3250 vmxnet3_resume(struct device *device)
3251 {
3252         int err, i = 0;
3253         unsigned long flags;
3254         struct pci_dev *pdev = to_pci_dev(device);
3255         struct net_device *netdev = pci_get_drvdata(pdev);
3256         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3257         struct Vmxnet3_PMConf *pmConf;
3258
3259         if (!netif_running(netdev))
3260                 return 0;
3261
3262         /* Destroy wake-up filters. */
3263         pmConf = adapter->pm_conf;
3264         memset(pmConf, 0, sizeof(*pmConf));
3265
3266         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3267         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3268                                                                   *pmConf));
3269         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3270                                                                  pmConf));
3271
3272         netif_device_attach(netdev);
3273         pci_set_power_state(pdev, PCI_D0);
3274         pci_restore_state(pdev);
3275         err = pci_enable_device_mem(pdev);
3276         if (err != 0)
3277                 return err;
3278
3279         pci_enable_wake(pdev, PCI_D0, 0);
3280
3281         spin_lock_irqsave(&adapter->cmd_lock, flags);
3282         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3283                                VMXNET3_CMD_UPDATE_PMCFG);
3284         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3285         vmxnet3_alloc_intr_resources(adapter);
3286         vmxnet3_request_irqs(adapter);
3287         for (i = 0; i < adapter->num_rx_queues; i++)
3288                 napi_enable(&adapter->rx_queue[i].napi);
3289         vmxnet3_enable_all_intrs(adapter);
3290
3291         return 0;
3292 }
3293
3294 static const struct dev_pm_ops vmxnet3_pm_ops = {
3295         .suspend = vmxnet3_suspend,
3296         .resume = vmxnet3_resume,
3297 };
3298 #endif
3299
3300 static struct pci_driver vmxnet3_driver = {
3301         .name           = vmxnet3_driver_name,
3302         .id_table       = vmxnet3_pciid_table,
3303         .probe          = vmxnet3_probe_device,
3304         .remove         = vmxnet3_remove_device,
3305 #ifdef CONFIG_PM
3306         .driver.pm      = &vmxnet3_pm_ops,
3307 #endif
3308 };
3309
3310
3311 static int __init
3312 vmxnet3_init_module(void)
3313 {
3314         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3315                 VMXNET3_DRIVER_VERSION_REPORT);
3316         return pci_register_driver(&vmxnet3_driver);
3317 }
3318
3319 module_init(vmxnet3_init_module);
3320
3321
3322 static void
3323 vmxnet3_exit_module(void)
3324 {
3325         pci_unregister_driver(&vmxnet3_driver);
3326 }
3327
3328 module_exit(vmxnet3_exit_module);
3329
3330 MODULE_AUTHOR("VMware, Inc.");
3331 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3332 MODULE_LICENSE("GPL v2");
3333 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);