drivers/net/vmxnet3/vmxnet3_drv.c

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