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                 union Vmxnet3_GenericDesc *gdesc;
 409
 410                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 411                 gdesc = tq->tx_ring.base + tq->tx_ring.next2comp;
 412
 413                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 414                 if (tbi->skb) {
 415                         dev_kfree_skb_any(tbi->skb);
 416                         tbi->skb = NULL;
 417                 }
 418                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 419         }
 420
 421         /* sanity check, verify all buffers are indeed unmapped and freed */
 422         for (i = 0; i < tq->tx_ring.size; i++) {
 423                 BUG_ON(tq->buf_info[i].skb != NULL ||
 424                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 425         }
 426
 427         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 428         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 429
 430         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 431         tq->comp_ring.next2proc = 0;
 432 }
 433
 434
 435 static void
 436 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 437                    struct vmxnet3_adapter *adapter)
 438 {
 439         if (tq->tx_ring.base) {
 440                 pci_free_consistent(adapter->pdev, tq->tx_ring.size *
 441                                     sizeof(struct Vmxnet3_TxDesc),
 442                                     tq->tx_ring.base, tq->tx_ring.basePA);
 443                 tq->tx_ring.base = NULL;
 444         }
 445         if (tq->data_ring.base) {
 446                 pci_free_consistent(adapter->pdev, tq->data_ring.size *
 447                                     sizeof(struct Vmxnet3_TxDataDesc),
 448                                     tq->data_ring.base, tq->data_ring.basePA);
 449                 tq->data_ring.base = NULL;
 450         }
 451         if (tq->comp_ring.base) {
 452                 pci_free_consistent(adapter->pdev, tq->comp_ring.size *
 453                                     sizeof(struct Vmxnet3_TxCompDesc),
 454                                     tq->comp_ring.base, tq->comp_ring.basePA);
 455                 tq->comp_ring.base = NULL;
 456         }
 457         kfree(tq->buf_info);
 458         tq->buf_info = NULL;
 459 }
 460
 461
 462 /* Destroy all tx queues */
 463 void
 464 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
 465 {
 466         int i;
 467
 468         for (i = 0; i < adapter->num_tx_queues; i++)
 469                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
 470 }
 471
 472
 473 static void
 474 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 475                 struct vmxnet3_adapter *adapter)
 476 {
 477         int i;
 478
 479         /* reset the tx ring contents to 0 and reset the tx ring states */
 480         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 481                sizeof(struct Vmxnet3_TxDesc));
 482         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 483         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 484
 485         memset(tq->data_ring.base, 0, tq->data_ring.size *
 486                sizeof(struct Vmxnet3_TxDataDesc));
 487
 488         /* reset the tx comp ring contents to 0 and reset comp ring states */
 489         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 490                sizeof(struct Vmxnet3_TxCompDesc));
 491         tq->comp_ring.next2proc = 0;
 492         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 493
 494         /* reset the bookkeeping data */
 495         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 496         for (i = 0; i < tq->tx_ring.size; i++)
 497                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 498
 499         /* stats are not reset */
 500 }
 501
 502
 503 static int
 504 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 505                   struct vmxnet3_adapter *adapter)
 506 {
 507         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 508                tq->comp_ring.base || tq->buf_info);
 509
 510         tq->tx_ring.base = pci_alloc_consistent(adapter->pdev, tq->tx_ring.size
 511                            * sizeof(struct Vmxnet3_TxDesc),
 512                            &tq->tx_ring.basePA);
 513         if (!tq->tx_ring.base) {
 514                 printk(KERN_ERR "%s: failed to allocate tx ring\n",
 515                        adapter->netdev->name);
 516                 goto err;
 517         }
 518
 519         tq->data_ring.base = pci_alloc_consistent(adapter->pdev,
 520                              tq->data_ring.size *
 521                              sizeof(struct Vmxnet3_TxDataDesc),
 522                              &tq->data_ring.basePA);
 523         if (!tq->data_ring.base) {
 524                 printk(KERN_ERR "%s: failed to allocate data ring\n",
 525                        adapter->netdev->name);
 526                 goto err;
 527         }
 528
 529         tq->comp_ring.base = pci_alloc_consistent(adapter->pdev,
 530                              tq->comp_ring.size *
 531                              sizeof(struct Vmxnet3_TxCompDesc),
 532                              &tq->comp_ring.basePA);
 533         if (!tq->comp_ring.base) {
 534                 printk(KERN_ERR "%s: failed to allocate tx comp ring\n",
 535                        adapter->netdev->name);
 536                 goto err;
 537         }
 538
 539         tq->buf_info = kcalloc(tq->tx_ring.size, sizeof(tq->buf_info[0]),
 540                                GFP_KERNEL);
 541         if (!tq->buf_info) {
 542                 printk(KERN_ERR "%s: failed to allocate tx bufinfo\n",
 543                        adapter->netdev->name);
 544                 goto err;
 545         }
 546
 547         return 0;
 548
 549 err:
 550         vmxnet3_tq_destroy(tq, adapter);
 551         return -ENOMEM;
 552 }
 553
 554 static void
 555 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
 556 {
 557         int i;
 558
 559         for (i = 0; i < adapter->num_tx_queues; i++)
 560                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
 561 }
 562
 563 /*
 564  *    starting from ring->next2fill, allocate rx buffers for the given ring
 565  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 566  *    are allocated or allocation fails
 567  */
 568
 569 static int
 570 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 571                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 572 {
 573         int num_allocated = 0;
 574         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 575         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 576         u32 val;
 577
 578         while (num_allocated <= num_to_alloc) {
 579                 struct vmxnet3_rx_buf_info *rbi;
 580                 union Vmxnet3_GenericDesc *gd;
 581
 582                 rbi = rbi_base + ring->next2fill;
 583                 gd = ring->base + ring->next2fill;
 584
 585                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 586                         if (rbi->skb == NULL) {
 587                                 rbi->skb = dev_alloc_skb(rbi->len +
 588                                                          NET_IP_ALIGN);
 589                                 if (unlikely(rbi->skb == NULL)) {
 590                                         rq->stats.rx_buf_alloc_failure++;
 591                                         break;
 592                                 }
 593                                 rbi->skb->dev = adapter->netdev;
 594
 595                                 skb_reserve(rbi->skb, NET_IP_ALIGN);
 596                                 rbi->dma_addr = pci_map_single(adapter->pdev,
 597                                                 rbi->skb->data, rbi->len,
 598                                                 PCI_DMA_FROMDEVICE);
 599                         } else {
 600                                 /* rx buffer skipped by the device */
 601                         }
 602                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 603                 } else {
 604                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 605                                rbi->len  != PAGE_SIZE);
 606
 607                         if (rbi->page == NULL) {
 608                                 rbi->page = alloc_page(GFP_ATOMIC);
 609                                 if (unlikely(rbi->page == NULL)) {
 610                                         rq->stats.rx_buf_alloc_failure++;
 611                                         break;
 612                                 }
 613                                 rbi->dma_addr = pci_map_page(adapter->pdev,
 614                                                 rbi->page, 0, PAGE_SIZE,
 615                                                 PCI_DMA_FROMDEVICE);
 616                         } else {
 617                                 /* rx buffers skipped by the device */
 618                         }
 619                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 620                 }
 621
 622                 BUG_ON(rbi->dma_addr == 0);
 623                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 624                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 625                                            | val | rbi->len);
 626
 627                 /* Fill the last buffer but dont mark it ready, or else the
 628                  * device will think that the queue is full */
 629                 if (num_allocated == num_to_alloc)
 630                         break;
 631
 632                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 633                 num_allocated++;
 634                 vmxnet3_cmd_ring_adv_next2fill(ring);
 635         }
 636         rq->uncommitted[ring_idx] += num_allocated;
 637
 638         dev_dbg(&adapter->netdev->dev,
 639                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp "
 640                 "%u, uncommited %u\n", num_allocated, ring->next2fill,
 641                 ring->next2comp, rq->uncommitted[ring_idx]);
 642
 643         /* so that the device can distinguish a full ring and an empty ring */
 644         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 645
 646         return num_allocated;
 647 }
 648
 649
 650 static void
 651 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 652                     struct vmxnet3_rx_buf_info *rbi)
 653 {
 654         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 655                 skb_shinfo(skb)->nr_frags;
 656
 657         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 658
 659         frag->page = rbi->page;
 660         frag->page_offset = 0;
 661         frag->size = rcd->len;
 662         skb->data_len += frag->size;
 663         skb_shinfo(skb)->nr_frags++;
 664 }
 665
 666
 667 static void
 668 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 669                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 670                 struct vmxnet3_adapter *adapter)
 671 {
 672         u32 dw2, len;
 673         unsigned long buf_offset;
 674         int i;
 675         union Vmxnet3_GenericDesc *gdesc;
 676         struct vmxnet3_tx_buf_info *tbi = NULL;
 677
 678         BUG_ON(ctx->copy_size > skb_headlen(skb));
 679
 680         /* use the previous gen bit for the SOP desc */
 681         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 682
 683         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 684         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 685
 686         /* no need to map the buffer if headers are copied */
 687         if (ctx->copy_size) {
 688                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 689                                         tq->tx_ring.next2fill *
 690                                         sizeof(struct Vmxnet3_TxDataDesc));
 691                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 692                 ctx->sop_txd->dword[3] = 0;
 693
 694                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 695                 tbi->map_type = VMXNET3_MAP_NONE;
 696
 697                 dev_dbg(&adapter->netdev->dev,
 698                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 699                         tq->tx_ring.next2fill,
 700                         le64_to_cpu(ctx->sop_txd->txd.addr),
 701                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 702                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 703
 704                 /* use the right gen for non-SOP desc */
 705                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 706         }
 707
 708         /* linear part can use multiple tx desc if it's big */
 709         len = skb_headlen(skb) - ctx->copy_size;
 710         buf_offset = ctx->copy_size;
 711         while (len) {
 712                 u32 buf_size;
 713
 714                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 715                         buf_size = len;
 716                         dw2 |= len;
 717                 } else {
 718                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 719                         /* spec says that for TxDesc.len, 0 == 2^14 */
 720                 }
 721
 722                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 723                 tbi->map_type = VMXNET3_MAP_SINGLE;
 724                 tbi->dma_addr = pci_map_single(adapter->pdev,
 725                                 skb->data + buf_offset, buf_size,
 726                                 PCI_DMA_TODEVICE);
 727
 728                 tbi->len = buf_size;
 729
 730                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 731                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 732
 733                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 734                 gdesc->dword[2] = cpu_to_le32(dw2);
 735                 gdesc->dword[3] = 0;
 736
 737                 dev_dbg(&adapter->netdev->dev,
 738                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 739                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 740                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 741                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 742                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 743
 744                 len -= buf_size;
 745                 buf_offset += buf_size;
 746         }
 747
 748         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 749                 struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 750
 751                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 752                 tbi->map_type = VMXNET3_MAP_PAGE;
 753                 tbi->dma_addr = pci_map_page(adapter->pdev, frag->page,
 754                                              frag->page_offset, frag->size,
 755                                              PCI_DMA_TODEVICE);
 756
 757                 tbi->len = frag->size;
 758
 759                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 760                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 761
 762                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 763                 gdesc->dword[2] = cpu_to_le32(dw2 | frag->size);
 764                 gdesc->dword[3] = 0;
 765
 766                 dev_dbg(&adapter->netdev->dev,
 767                         "txd[%u]: 0x%llu %u %u\n",
 768                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 769                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 770                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 771                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 772         }
 773
 774         ctx->eop_txd = gdesc;
 775
 776         /* set the last buf_info for the pkt */
 777         tbi->skb = skb;
 778         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 779 }
 780
 781
 782 /* Init all tx queues */
 783 static void
 784 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 785 {
 786         int i;
 787
 788         for (i = 0; i < adapter->num_tx_queues; i++)
 789                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 790 }
 791
 792
 793 /*
 794  *    parse and copy relevant protocol headers:
 795  *      For a tso pkt, relevant headers are L2/3/4 including options
 796  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 797  *      if it's a TCP/UDP pkt
 798  *
 799  * Returns:
 800  *    -1:  error happens during parsing
 801  *     0:  protocol headers parsed, but too big to be copied
 802  *     1:  protocol headers parsed and copied
 803  *
 804  * Other effects:
 805  *    1. related *ctx fields are updated.
 806  *    2. ctx->copy_size is # of bytes copied
 807  *    3. the portion copied is guaranteed to be in the linear part
 808  *
 809  */
 810 static int
 811 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 812                            struct vmxnet3_tx_ctx *ctx,
 813                            struct vmxnet3_adapter *adapter)
 814 {
 815         struct Vmxnet3_TxDataDesc *tdd;
 816
 817         if (ctx->mss) { /* TSO */
 818                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 819                 ctx->l4_hdr_size = ((struct tcphdr *)
 820                                    skb_transport_header(skb))->doff * 4;
 821                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 822         } else {
 823                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 824                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 825
 826                         if (ctx->ipv4) {
 827                                 struct iphdr *iph = (struct iphdr *)
 828                                                     skb_network_header(skb);
 829                                 if (iph->protocol == IPPROTO_TCP)
 830                                         ctx->l4_hdr_size = ((struct tcphdr *)
 831                                            skb_transport_header(skb))->doff * 4;
 832                                 else if (iph->protocol == IPPROTO_UDP)
 833                                         /*
 834                                          * Use tcp header size so that bytes to
 835                                          * be copied are more than required by
 836                                          * the device.
 837                                          */
 838                                         ctx->l4_hdr_size =
 839                                                         sizeof(struct tcphdr);
 840                                 else
 841                                         ctx->l4_hdr_size = 0;
 842                         } else {
 843                                 /* for simplicity, don't copy L4 headers */
 844                                 ctx->l4_hdr_size = 0;
 845                         }
 846                         ctx->copy_size = ctx->eth_ip_hdr_size +
 847                                          ctx->l4_hdr_size;
 848                 } else {
 849                         ctx->eth_ip_hdr_size = 0;
 850                         ctx->l4_hdr_size = 0;
 851                         /* copy as much as allowed */
 852                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 853                                              , skb_headlen(skb));
 854                 }
 855
 856                 /* make sure headers are accessible directly */
 857                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 858                         goto err;
 859         }
 860
 861         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 862                 tq->stats.oversized_hdr++;
 863                 ctx->copy_size = 0;
 864                 return 0;
 865         }
 866
 867         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 868
 869         memcpy(tdd->data, skb->data, ctx->copy_size);
 870         dev_dbg(&adapter->netdev->dev,
 871                 "copy %u bytes to dataRing[%u]\n",
 872                 ctx->copy_size, tq->tx_ring.next2fill);
 873         return 1;
 874
 875 err:
 876         return -1;
 877 }
 878
 879
 880 static void
 881 vmxnet3_prepare_tso(struct sk_buff *skb,
 882                     struct vmxnet3_tx_ctx *ctx)
 883 {
 884         struct tcphdr *tcph = (struct tcphdr *)skb_transport_header(skb);
 885         if (ctx->ipv4) {
 886                 struct iphdr *iph = (struct iphdr *)skb_network_header(skb);
 887                 iph->check = 0;
 888                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 889                                                  IPPROTO_TCP, 0);
 890         } else {
 891                 struct ipv6hdr *iph = (struct ipv6hdr *)skb_network_header(skb);
 892                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 893                                                IPPROTO_TCP, 0);
 894         }
 895 }
 896
 897
 898 /*
 899  * Transmits a pkt thru a given tq
 900  * Returns:
 901  *    NETDEV_TX_OK:      descriptors are setup successfully
 902  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 903  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 904  *
 905  * Side-effects:
 906  *    1. tx ring may be changed
 907  *    2. tq stats may be updated accordingly
 908  *    3. shared->txNumDeferred may be updated
 909  */
 910
 911 static int
 912 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 913                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 914 {
 915         int ret;
 916         u32 count;
 917         unsigned long flags;
 918         struct vmxnet3_tx_ctx ctx;
 919         union Vmxnet3_GenericDesc *gdesc;
 920 #ifdef __BIG_ENDIAN_BITFIELD
 921         /* Use temporary descriptor to avoid touching bits multiple times */
 922         union Vmxnet3_GenericDesc tempTxDesc;
 923 #endif
 924
 925         /* conservatively estimate # of descriptors to use */
 926         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) +
 927                 skb_shinfo(skb)->nr_frags + 1;
 928
 929         ctx.ipv4 = (skb->protocol == cpu_to_be16(ETH_P_IP));
 930
 931         ctx.mss = skb_shinfo(skb)->gso_size;
 932         if (ctx.mss) {
 933                 if (skb_header_cloned(skb)) {
 934                         if (unlikely(pskb_expand_head(skb, 0, 0,
 935                                                       GFP_ATOMIC) != 0)) {
 936                                 tq->stats.drop_tso++;
 937                                 goto drop_pkt;
 938                         }
 939                         tq->stats.copy_skb_header++;
 940                 }
 941                 vmxnet3_prepare_tso(skb, &ctx);
 942         } else {
 943                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 944
 945                         /* non-tso pkts must not use more than
 946                          * VMXNET3_MAX_TXD_PER_PKT entries
 947                          */
 948                         if (skb_linearize(skb) != 0) {
 949                                 tq->stats.drop_too_many_frags++;
 950                                 goto drop_pkt;
 951                         }
 952                         tq->stats.linearized++;
 953
 954                         /* recalculate the # of descriptors to use */
 955                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 956                 }
 957         }
 958
 959         spin_lock_irqsave(&tq->tx_lock, flags);
 960
 961         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 962                 tq->stats.tx_ring_full++;
 963                 dev_dbg(&adapter->netdev->dev,
 964                         "tx queue stopped on %s, next2comp %u"
 965                         " next2fill %u\n", adapter->netdev->name,
 966                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 967
 968                 vmxnet3_tq_stop(tq, adapter);
 969                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 970                 return NETDEV_TX_BUSY;
 971         }
 972
 973
 974         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 975         if (ret >= 0) {
 976                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 977                 /* hdrs parsed, check against other limits */
 978                 if (ctx.mss) {
 979                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 980                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 981                                 goto hdr_too_big;
 982                         }
 983                 } else {
 984                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 985                                 if (unlikely(ctx.eth_ip_hdr_size +
 986                                              skb->csum_offset >
 987                                              VMXNET3_MAX_CSUM_OFFSET)) {
 988                                         goto hdr_too_big;
 989                                 }
 990                         }
 991                 }
 992         } else {
 993                 tq->stats.drop_hdr_inspect_err++;
 994                 goto unlock_drop_pkt;
 995         }
 996
 997         /* fill tx descs related to addr & len */
 998         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
 999
1000         /* setup the EOP desc */
1001         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1002
1003         /* setup the SOP desc */
1004 #ifdef __BIG_ENDIAN_BITFIELD
1005         gdesc = &tempTxDesc;
1006         gdesc->dword[2] = ctx.sop_txd->dword[2];
1007         gdesc->dword[3] = ctx.sop_txd->dword[3];
1008 #else
1009         gdesc = ctx.sop_txd;
1010 #endif
1011         if (ctx.mss) {
1012                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1013                 gdesc->txd.om = VMXNET3_OM_TSO;
1014                 gdesc->txd.msscof = ctx.mss;
1015                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1016                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1017         } else {
1018                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1019                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1020                         gdesc->txd.om = VMXNET3_OM_CSUM;
1021                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1022                                             skb->csum_offset;
1023                 } else {
1024                         gdesc->txd.om = 0;
1025                         gdesc->txd.msscof = 0;
1026                 }
1027                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1028         }
1029
1030         if (vlan_tx_tag_present(skb)) {
1031                 gdesc->txd.ti = 1;
1032                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1033         }
1034
1035         /* finally flips the GEN bit of the SOP desc. */
1036         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1037                                                   VMXNET3_TXD_GEN);
1038 #ifdef __BIG_ENDIAN_BITFIELD
1039         /* Finished updating in bitfields of Tx Desc, so write them in original
1040          * place.
1041          */
1042         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1043                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1044         gdesc = ctx.sop_txd;
1045 #endif
1046         dev_dbg(&adapter->netdev->dev,
1047                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1048                 (u32)((union Vmxnet3_GenericDesc *)ctx.sop_txd -
1049                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1050                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1051
1052         spin_unlock_irqrestore(&tq->tx_lock, flags);
1053
1054         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1055                                         le32_to_cpu(tq->shared->txThreshold)) {
1056                 tq->shared->txNumDeferred = 0;
1057                 VMXNET3_WRITE_BAR0_REG(adapter,
1058                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1059                                        tq->tx_ring.next2fill);
1060         }
1061
1062         return NETDEV_TX_OK;
1063
1064 hdr_too_big:
1065         tq->stats.drop_oversized_hdr++;
1066 unlock_drop_pkt:
1067         spin_unlock_irqrestore(&tq->tx_lock, flags);
1068 drop_pkt:
1069         tq->stats.drop_total++;
1070         dev_kfree_skb(skb);
1071         return NETDEV_TX_OK;
1072 }
1073
1074
1075 static netdev_tx_t
1076 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1077 {
1078         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1079
1080                 BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1081                 return vmxnet3_tq_xmit(skb,
1082                                        &adapter->tx_queue[skb->queue_mapping],
1083                                        adapter, netdev);
1084 }
1085
1086
1087 static void
1088 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1089                 struct sk_buff *skb,
1090                 union Vmxnet3_GenericDesc *gdesc)
1091 {
1092         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1093                 /* typical case: TCP/UDP over IP and both csums are correct */
1094                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1095                                                         VMXNET3_RCD_CSUM_OK) {
1096                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1097                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1098                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1099                         BUG_ON(gdesc->rcd.frg);
1100                 } else {
1101                         if (gdesc->rcd.csum) {
1102                                 skb->csum = htons(gdesc->rcd.csum);
1103                                 skb->ip_summed = CHECKSUM_PARTIAL;
1104                         } else {
1105                                 skb_checksum_none_assert(skb);
1106                         }
1107                 }
1108         } else {
1109                 skb_checksum_none_assert(skb);
1110         }
1111 }
1112
1113
1114 static void
1115 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1116                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1117 {
1118         rq->stats.drop_err++;
1119         if (!rcd->fcs)
1120                 rq->stats.drop_fcs++;
1121
1122         rq->stats.drop_total++;
1123
1124         /*
1125          * We do not unmap and chain the rx buffer to the skb.
1126          * We basically pretend this buffer is not used and will be recycled
1127          * by vmxnet3_rq_alloc_rx_buf()
1128          */
1129
1130         /*
1131          * ctx->skb may be NULL if this is the first and the only one
1132          * desc for the pkt
1133          */
1134         if (ctx->skb)
1135                 dev_kfree_skb_irq(ctx->skb);
1136
1137         ctx->skb = NULL;
1138 }
1139
1140
1141 static int
1142 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1143                        struct vmxnet3_adapter *adapter, int quota)
1144 {
1145         static const u32 rxprod_reg[2] = {
1146                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1147         };
1148         u32 num_rxd = 0;
1149         bool skip_page_frags = false;
1150         struct Vmxnet3_RxCompDesc *rcd;
1151         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1152 #ifdef __BIG_ENDIAN_BITFIELD
1153         struct Vmxnet3_RxDesc rxCmdDesc;
1154         struct Vmxnet3_RxCompDesc rxComp;
1155 #endif
1156         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1157                           &rxComp);
1158         while (rcd->gen == rq->comp_ring.gen) {
1159                 struct vmxnet3_rx_buf_info *rbi;
1160                 struct sk_buff *skb, *new_skb = NULL;
1161                 struct page *new_page = NULL;
1162                 int num_to_alloc;
1163                 struct Vmxnet3_RxDesc *rxd;
1164                 u32 idx, ring_idx;
1165                 struct vmxnet3_cmd_ring *ring = NULL;
1166                 if (num_rxd >= quota) {
1167                         /* we may stop even before we see the EOP desc of
1168                          * the current pkt
1169                          */
1170                         break;
1171                 }
1172                 num_rxd++;
1173                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1174                 idx = rcd->rxdIdx;
1175                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1176                 ring = rq->rx_ring + ring_idx;
1177                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1178                                   &rxCmdDesc);
1179                 rbi = rq->buf_info[ring_idx] + idx;
1180
1181                 BUG_ON(rxd->addr != rbi->dma_addr ||
1182                        rxd->len != rbi->len);
1183
1184                 if (unlikely(rcd->eop && rcd->err)) {
1185                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1186                         goto rcd_done;
1187                 }
1188
1189                 if (rcd->sop) { /* first buf of the pkt */
1190                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1191                                rcd->rqID != rq->qid);
1192
1193                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1194                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1195
1196                         if (unlikely(rcd->len == 0)) {
1197                                 /* Pretend the rx buffer is skipped. */
1198                                 BUG_ON(!(rcd->sop && rcd->eop));
1199                                 dev_dbg(&adapter->netdev->dev,
1200                                         "rxRing[%u][%u] 0 length\n",
1201                                         ring_idx, idx);
1202                                 goto rcd_done;
1203                         }
1204
1205                         skip_page_frags = false;
1206                         ctx->skb = rbi->skb;
1207                         new_skb = dev_alloc_skb(rbi->len + NET_IP_ALIGN);
1208                         if (new_skb == NULL) {
1209                                 /* Skb allocation failed, do not handover this
1210                                  * skb to stack. Reuse it. Drop the existing pkt
1211                                  */
1212                                 rq->stats.rx_buf_alloc_failure++;
1213                                 ctx->skb = NULL;
1214                                 rq->stats.drop_total++;
1215                                 skip_page_frags = true;
1216                                 goto rcd_done;
1217                         }
1218
1219                         pci_unmap_single(adapter->pdev, rbi->dma_addr, rbi->len,
1220                                          PCI_DMA_FROMDEVICE);
1221
1222                         skb_put(ctx->skb, rcd->len);
1223
1224                         /* Immediate refill */
1225                         new_skb->dev = adapter->netdev;
1226                         skb_reserve(new_skb, NET_IP_ALIGN);
1227                         rbi->skb = new_skb;
1228                         rbi->dma_addr = pci_map_single(adapter->pdev,
1229                                         rbi->skb->data, rbi->len,
1230                                         PCI_DMA_FROMDEVICE);
1231                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1232                         rxd->len = rbi->len;
1233
1234                 } else {
1235                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1236
1237                         /* non SOP buffer must be type 1 in most cases */
1238                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1239                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1240
1241                         /* If an sop buffer was dropped, skip all
1242                          * following non-sop fragments. They will be reused.
1243                          */
1244                         if (skip_page_frags)
1245                                 goto rcd_done;
1246
1247                         new_page = alloc_page(GFP_ATOMIC);
1248                         if (unlikely(new_page == NULL)) {
1249                                 /* Replacement page frag could not be allocated.
1250                                  * Reuse this page. Drop the pkt and free the
1251                                  * skb which contained this page as a frag. Skip
1252                                  * processing all the following non-sop frags.
1253                                  */
1254                                 rq->stats.rx_buf_alloc_failure++;
1255                                 dev_kfree_skb(ctx->skb);
1256                                 ctx->skb = NULL;
1257                                 skip_page_frags = true;
1258                                 goto rcd_done;
1259                         }
1260
1261                         if (rcd->len) {
1262                                 pci_unmap_page(adapter->pdev,
1263                                                rbi->dma_addr, rbi->len,
1264                                                PCI_DMA_FROMDEVICE);
1265
1266                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1267                         }
1268
1269                         /* Immediate refill */
1270                         rbi->page = new_page;
1271                         rbi->dma_addr = pci_map_page(adapter->pdev, rbi->page,
1272                                                      0, PAGE_SIZE,
1273                                                      PCI_DMA_FROMDEVICE);
1274                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1275                         rxd->len = rbi->len;
1276                 }
1277
1278
1279                 skb = ctx->skb;
1280                 if (rcd->eop) {
1281                         skb->len += skb->data_len;
1282                         skb->truesize += skb->data_len;
1283
1284                         vmxnet3_rx_csum(adapter, skb,
1285                                         (union Vmxnet3_GenericDesc *)rcd);
1286                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1287
1288                         if (unlikely(adapter->vlan_grp && rcd->ts)) {
1289                                 vlan_hwaccel_receive_skb(skb,
1290                                                 adapter->vlan_grp, rcd->tci);
1291                         } else {
1292                                 netif_receive_skb(skb);
1293                         }
1294
1295                         ctx->skb = NULL;
1296                 }
1297
1298 rcd_done:
1299                 /* device may have skipped some rx descs */
1300                 ring->next2comp = idx;
1301                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1302                 ring = rq->rx_ring + ring_idx;
1303                 while (num_to_alloc) {
1304                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1305                                           &rxCmdDesc);
1306                         BUG_ON(!rxd->addr);
1307
1308                         /* Recv desc is ready to be used by the device */
1309                         rxd->gen = ring->gen;
1310                         vmxnet3_cmd_ring_adv_next2fill(ring);
1311                         num_to_alloc--;
1312                 }
1313
1314                 /* if needed, update the register */
1315                 if (unlikely(rq->shared->updateRxProd)) {
1316                         VMXNET3_WRITE_BAR0_REG(adapter,
1317                                 rxprod_reg[ring_idx] + rq->qid * 8,
1318                                 ring->next2fill);
1319                         rq->uncommitted[ring_idx] = 0;
1320                 }
1321
1322                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1323                 vmxnet3_getRxComp(rcd,
1324                      &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1325         }
1326
1327         return num_rxd;
1328 }
1329
1330
1331 static void
1332 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1333                    struct vmxnet3_adapter *adapter)
1334 {
1335         u32 i, ring_idx;
1336         struct Vmxnet3_RxDesc *rxd;
1337
1338         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1339                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1340 #ifdef __BIG_ENDIAN_BITFIELD
1341                         struct Vmxnet3_RxDesc rxDesc;
1342 #endif
1343                         vmxnet3_getRxDesc(rxd,
1344                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1345
1346                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1347                                         rq->buf_info[ring_idx][i].skb) {
1348                                 pci_unmap_single(adapter->pdev, rxd->addr,
1349                                                  rxd->len, PCI_DMA_FROMDEVICE);
1350                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1351                                 rq->buf_info[ring_idx][i].skb = NULL;
1352                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1353                                         rq->buf_info[ring_idx][i].page) {
1354                                 pci_unmap_page(adapter->pdev, rxd->addr,
1355                                                rxd->len, PCI_DMA_FROMDEVICE);
1356                                 put_page(rq->buf_info[ring_idx][i].page);
1357                                 rq->buf_info[ring_idx][i].page = NULL;
1358                         }
1359                 }
1360
1361                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1362                 rq->rx_ring[ring_idx].next2fill =
1363                                         rq->rx_ring[ring_idx].next2comp = 0;
1364                 rq->uncommitted[ring_idx] = 0;
1365         }
1366
1367         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1368         rq->comp_ring.next2proc = 0;
1369 }
1370
1371
1372 static void
1373 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1374 {
1375         int i;
1376
1377         for (i = 0; i < adapter->num_rx_queues; i++)
1378                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1379 }
1380
1381
1382 void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1383                         struct vmxnet3_adapter *adapter)
1384 {
1385         int i;
1386         int j;
1387
1388         /* all rx buffers must have already been freed */
1389         for (i = 0; i < 2; i++) {
1390                 if (rq->buf_info[i]) {
1391                         for (j = 0; j < rq->rx_ring[i].size; j++)
1392                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1393                 }
1394         }
1395
1396
1397         kfree(rq->buf_info[0]);
1398
1399         for (i = 0; i < 2; i++) {
1400                 if (rq->rx_ring[i].base) {
1401                         pci_free_consistent(adapter->pdev, rq->rx_ring[i].size
1402                                             * sizeof(struct Vmxnet3_RxDesc),
1403                                             rq->rx_ring[i].base,
1404                                             rq->rx_ring[i].basePA);
1405                         rq->rx_ring[i].base = NULL;
1406                 }
1407                 rq->buf_info[i] = NULL;
1408         }
1409
1410         if (rq->comp_ring.base) {
1411                 pci_free_consistent(adapter->pdev, rq->comp_ring.size *
1412                                     sizeof(struct Vmxnet3_RxCompDesc),
1413                                     rq->comp_ring.base, rq->comp_ring.basePA);
1414                 rq->comp_ring.base = NULL;
1415         }
1416 }
1417
1418
1419 static int
1420 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1421                 struct vmxnet3_adapter  *adapter)
1422 {
1423         int i;
1424
1425         /* initialize buf_info */
1426         for (i = 0; i < rq->rx_ring[0].size; i++) {
1427
1428                 /* 1st buf for a pkt is skbuff */
1429                 if (i % adapter->rx_buf_per_pkt == 0) {
1430                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1431                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1432                 } else { /* subsequent bufs for a pkt is frag */
1433                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1434                         rq->buf_info[0][i].len = PAGE_SIZE;
1435                 }
1436         }
1437         for (i = 0; i < rq->rx_ring[1].size; i++) {
1438                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1439                 rq->buf_info[1][i].len = PAGE_SIZE;
1440         }
1441
1442         /* reset internal state and allocate buffers for both rings */
1443         for (i = 0; i < 2; i++) {
1444                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1445                 rq->uncommitted[i] = 0;
1446
1447                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1448                        sizeof(struct Vmxnet3_RxDesc));
1449                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1450         }
1451         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1452                                     adapter) == 0) {
1453                 /* at least has 1 rx buffer for the 1st ring */
1454                 return -ENOMEM;
1455         }
1456         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1457
1458         /* reset the comp ring */
1459         rq->comp_ring.next2proc = 0;
1460         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1461                sizeof(struct Vmxnet3_RxCompDesc));
1462         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1463
1464         /* reset rxctx */
1465         rq->rx_ctx.skb = NULL;
1466
1467         /* stats are not reset */
1468         return 0;
1469 }
1470
1471
1472 static int
1473 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1474 {
1475         int i, err = 0;
1476
1477         for (i = 0; i < adapter->num_rx_queues; i++) {
1478                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1479                 if (unlikely(err)) {
1480                         dev_err(&adapter->netdev->dev, "%s: failed to "
1481                                 "initialize rx queue%i\n",
1482                                 adapter->netdev->name, i);
1483                         break;
1484                 }
1485         }
1486         return err;
1487
1488 }
1489
1490
1491 static int
1492 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1493 {
1494         int i;
1495         size_t sz;
1496         struct vmxnet3_rx_buf_info *bi;
1497
1498         for (i = 0; i < 2; i++) {
1499
1500                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1501                 rq->rx_ring[i].base = pci_alloc_consistent(adapter->pdev, sz,
1502                                                         &rq->rx_ring[i].basePA);
1503                 if (!rq->rx_ring[i].base) {
1504                         printk(KERN_ERR "%s: failed to allocate rx ring %d\n",
1505                                adapter->netdev->name, i);
1506                         goto err;
1507                 }
1508         }
1509
1510         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1511         rq->comp_ring.base = pci_alloc_consistent(adapter->pdev, sz,
1512                                                   &rq->comp_ring.basePA);
1513         if (!rq->comp_ring.base) {
1514                 printk(KERN_ERR "%s: failed to allocate rx comp ring\n",
1515                        adapter->netdev->name);
1516                 goto err;
1517         }
1518
1519         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1520                                                    rq->rx_ring[1].size);
1521         bi = kzalloc(sz, GFP_KERNEL);
1522         if (!bi) {
1523                 printk(KERN_ERR "%s: failed to allocate rx bufinfo\n",
1524                        adapter->netdev->name);
1525                 goto err;
1526         }
1527         rq->buf_info[0] = bi;
1528         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1529
1530         return 0;
1531
1532 err:
1533         vmxnet3_rq_destroy(rq, adapter);
1534         return -ENOMEM;
1535 }
1536
1537
1538 static int
1539 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1540 {
1541         int i, err = 0;
1542
1543         for (i = 0; i < adapter->num_rx_queues; i++) {
1544                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1545                 if (unlikely(err)) {
1546                         dev_err(&adapter->netdev->dev,
1547                                 "%s: failed to create rx queue%i\n",
1548                                 adapter->netdev->name, i);
1549                         goto err_out;
1550                 }
1551         }
1552         return err;
1553 err_out:
1554         vmxnet3_rq_destroy_all(adapter);
1555         return err;
1556
1557 }
1558
1559 /* Multiple queue aware polling function for tx and rx */
1560
1561 static int
1562 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1563 {
1564         int rcd_done = 0, i;
1565         if (unlikely(adapter->shared->ecr))
1566                 vmxnet3_process_events(adapter);
1567         for (i = 0; i < adapter->num_tx_queues; i++)
1568                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1569
1570         for (i = 0; i < adapter->num_rx_queues; i++)
1571                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1572                                                    adapter, budget);
1573         return rcd_done;
1574 }
1575
1576
1577 static int
1578 vmxnet3_poll(struct napi_struct *napi, int budget)
1579 {
1580         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1581                                           struct vmxnet3_rx_queue, napi);
1582         int rxd_done;
1583
1584         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1585
1586         if (rxd_done < budget) {
1587                 napi_complete(napi);
1588                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1589         }
1590         return rxd_done;
1591 }
1592
1593 /*
1594  * NAPI polling function for MSI-X mode with multiple Rx queues
1595  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1596  */
1597
1598 static int
1599 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1600 {
1601         struct vmxnet3_rx_queue *rq = container_of(napi,
1602                                                 struct vmxnet3_rx_queue, napi);
1603         struct vmxnet3_adapter *adapter = rq->adapter;
1604         int rxd_done;
1605
1606         /* When sharing interrupt with corresponding tx queue, process
1607          * tx completions in that queue as well
1608          */
1609         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1610                 struct vmxnet3_tx_queue *tq =
1611                                 &adapter->tx_queue[rq - adapter->rx_queue];
1612                 vmxnet3_tq_tx_complete(tq, adapter);
1613         }
1614
1615         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1616
1617         if (rxd_done < budget) {
1618                 napi_complete(napi);
1619                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1620         }
1621         return rxd_done;
1622 }
1623
1624
1625 #ifdef CONFIG_PCI_MSI
1626
1627 /*
1628  * Handle completion interrupts on tx queues
1629  * Returns whether or not the intr is handled
1630  */
1631
1632 static irqreturn_t
1633 vmxnet3_msix_tx(int irq, void *data)
1634 {
1635         struct vmxnet3_tx_queue *tq = data;
1636         struct vmxnet3_adapter *adapter = tq->adapter;
1637
1638         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1639                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1640
1641         /* Handle the case where only one irq is allocate for all tx queues */
1642         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1643                 int i;
1644                 for (i = 0; i < adapter->num_tx_queues; i++) {
1645                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1646                         vmxnet3_tq_tx_complete(txq, adapter);
1647                 }
1648         } else {
1649                 vmxnet3_tq_tx_complete(tq, adapter);
1650         }
1651         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1652
1653         return IRQ_HANDLED;
1654 }
1655
1656
1657 /*
1658  * Handle completion interrupts on rx queues. Returns whether or not the
1659  * intr is handled
1660  */
1661
1662 static irqreturn_t
1663 vmxnet3_msix_rx(int irq, void *data)
1664 {
1665         struct vmxnet3_rx_queue *rq = data;
1666         struct vmxnet3_adapter *adapter = rq->adapter;
1667
1668         /* disable intr if needed */
1669         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1670                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1671         napi_schedule(&rq->napi);
1672
1673         return IRQ_HANDLED;
1674 }
1675
1676 /*
1677  *----------------------------------------------------------------------------
1678  *
1679  * vmxnet3_msix_event --
1680  *
1681  *    vmxnet3 msix event intr handler
1682  *
1683  * Result:
1684  *    whether or not the intr is handled
1685  *
1686  *----------------------------------------------------------------------------
1687  */
1688
1689 static irqreturn_t
1690 vmxnet3_msix_event(int irq, void *data)
1691 {
1692         struct net_device *dev = data;
1693         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1694
1695         /* disable intr if needed */
1696         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1697                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1698
1699         if (adapter->shared->ecr)
1700                 vmxnet3_process_events(adapter);
1701
1702         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1703
1704         return IRQ_HANDLED;
1705 }
1706
1707 #endif /* CONFIG_PCI_MSI  */
1708
1709
1710 /* Interrupt handler for vmxnet3  */
1711 static irqreturn_t
1712 vmxnet3_intr(int irq, void *dev_id)
1713 {
1714         struct net_device *dev = dev_id;
1715         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1716
1717         if (adapter->intr.type == VMXNET3_IT_INTX) {
1718                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1719                 if (unlikely(icr == 0))
1720                         /* not ours */
1721                         return IRQ_NONE;
1722         }
1723
1724
1725         /* disable intr if needed */
1726         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1727                 vmxnet3_disable_all_intrs(adapter);
1728
1729         napi_schedule(&adapter->rx_queue[0].napi);
1730
1731         return IRQ_HANDLED;
1732 }
1733
1734 #ifdef CONFIG_NET_POLL_CONTROLLER
1735
1736 /* netpoll callback. */
1737 static void
1738 vmxnet3_netpoll(struct net_device *netdev)
1739 {
1740         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1741
1742         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1743                 vmxnet3_disable_all_intrs(adapter);
1744
1745         vmxnet3_do_poll(adapter, adapter->rx_queue[0].rx_ring[0].size);
1746         vmxnet3_enable_all_intrs(adapter);
1747
1748 }
1749 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1750
1751 static int
1752 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1753 {
1754         struct vmxnet3_intr *intr = &adapter->intr;
1755         int err = 0, i;
1756         int vector = 0;
1757
1758 #ifdef CONFIG_PCI_MSI
1759         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1760                 for (i = 0; i < adapter->num_tx_queues; i++) {
1761                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1762                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1763                                         adapter->netdev->name, vector);
1764                                 err = request_irq(
1765                                               intr->msix_entries[vector].vector,
1766                                               vmxnet3_msix_tx, 0,
1767                                               adapter->tx_queue[i].name,
1768                                               &adapter->tx_queue[i]);
1769                         } else {
1770                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1771                                         adapter->netdev->name, vector);
1772                         }
1773                         if (err) {
1774                                 dev_err(&adapter->netdev->dev,
1775                                         "Failed to request irq for MSIX, %s, "
1776                                         "error %d\n",
1777                                         adapter->tx_queue[i].name, err);
1778                                 return err;
1779                         }
1780
1781                         /* Handle the case where only 1 MSIx was allocated for
1782                          * all tx queues */
1783                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1784                                 for (; i < adapter->num_tx_queues; i++)
1785                                         adapter->tx_queue[i].comp_ring.intr_idx
1786                                                                 = vector;
1787                                 vector++;
1788                                 break;
1789                         } else {
1790                                 adapter->tx_queue[i].comp_ring.intr_idx
1791                                                                 = vector++;
1792                         }
1793                 }
1794                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1795                         vector = 0;
1796
1797                 for (i = 0; i < adapter->num_rx_queues; i++) {
1798                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1799                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1800                                         adapter->netdev->name, vector);
1801                         else
1802                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1803                                         adapter->netdev->name, vector);
1804                         err = request_irq(intr->msix_entries[vector].vector,
1805                                           vmxnet3_msix_rx, 0,
1806                                           adapter->rx_queue[i].name,
1807                                           &(adapter->rx_queue[i]));
1808                         if (err) {
1809                                 printk(KERN_ERR "Failed to request irq for MSIX"
1810                                        ", %s, error %d\n",
1811                                        adapter->rx_queue[i].name, err);
1812                                 return err;
1813                         }
1814
1815                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1816                 }
1817
1818                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1819                         adapter->netdev->name, vector);
1820                 err = request_irq(intr->msix_entries[vector].vector,
1821                                   vmxnet3_msix_event, 0,
1822                                   intr->event_msi_vector_name, adapter->netdev);
1823                 intr->event_intr_idx = vector;
1824
1825         } else if (intr->type == VMXNET3_IT_MSI) {
1826                 adapter->num_rx_queues = 1;
1827                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1828                                   adapter->netdev->name, adapter->netdev);
1829         } else {
1830 #endif
1831                 adapter->num_rx_queues = 1;
1832                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1833                                   IRQF_SHARED, adapter->netdev->name,
1834                                   adapter->netdev);
1835 #ifdef CONFIG_PCI_MSI
1836         }
1837 #endif
1838         intr->num_intrs = vector + 1;
1839         if (err) {
1840                 printk(KERN_ERR "Failed to request irq %s (intr type:%d), error"
1841                        ":%d\n", adapter->netdev->name, intr->type, err);
1842         } else {
1843                 /* Number of rx queues will not change after this */
1844                 for (i = 0; i < adapter->num_rx_queues; i++) {
1845                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1846                         rq->qid = i;
1847                         rq->qid2 = i + adapter->num_rx_queues;
1848                 }
1849
1850
1851
1852                 /* init our intr settings */
1853                 for (i = 0; i < intr->num_intrs; i++)
1854                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1855                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1856                         adapter->intr.event_intr_idx = 0;
1857                         for (i = 0; i < adapter->num_tx_queues; i++)
1858                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1859                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1860                 }
1861
1862                 printk(KERN_INFO "%s: intr type %u, mode %u, %u vectors "
1863                        "allocated\n", adapter->netdev->name, intr->type,
1864                        intr->mask_mode, intr->num_intrs);
1865         }
1866
1867         return err;
1868 }
1869
1870
1871 static void
1872 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1873 {
1874         struct vmxnet3_intr *intr = &adapter->intr;
1875         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1876
1877         switch (intr->type) {
1878 #ifdef CONFIG_PCI_MSI
1879         case VMXNET3_IT_MSIX:
1880         {
1881                 int i, vector = 0;
1882
1883                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1884                         for (i = 0; i < adapter->num_tx_queues; i++) {
1885                                 free_irq(intr->msix_entries[vector++].vector,
1886                                          &(adapter->tx_queue[i]));
1887                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1888                                         break;
1889                         }
1890                 }
1891
1892                 for (i = 0; i < adapter->num_rx_queues; i++) {
1893                         free_irq(intr->msix_entries[vector++].vector,
1894                                  &(adapter->rx_queue[i]));
1895                 }
1896
1897                 free_irq(intr->msix_entries[vector].vector,
1898                          adapter->netdev);
1899                 BUG_ON(vector >= intr->num_intrs);
1900                 break;
1901         }
1902 #endif
1903         case VMXNET3_IT_MSI:
1904                 free_irq(adapter->pdev->irq, adapter->netdev);
1905                 break;
1906         case VMXNET3_IT_INTX:
1907                 free_irq(adapter->pdev->irq, adapter->netdev);
1908                 break;
1909         default:
1910                 BUG_ON(true);
1911         }
1912 }
1913
1914 static void
1915 vmxnet3_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
1916 {
1917         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1918         struct Vmxnet3_DriverShared *shared = adapter->shared;
1919         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1920         unsigned long flags;
1921
1922         if (grp) {
1923                 /* add vlan rx stripping. */
1924                 if (adapter->netdev->features & NETIF_F_HW_VLAN_RX) {
1925                         int i;
1926                         adapter->vlan_grp = grp;
1927
1928                         /*
1929                          *  Clear entire vfTable; then enable untagged pkts.
1930                          *  Note: setting one entry in vfTable to non-zero turns
1931                          *  on VLAN rx filtering.
1932                          */
1933                         for (i = 0; i < VMXNET3_VFT_SIZE; i++)
1934                                 vfTable[i] = 0;
1935
1936                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1937                         spin_lock_irqsave(&adapter->cmd_lock, flags);
1938                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1939                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1940                         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1941                 } else {
1942                         printk(KERN_ERR "%s: vlan_rx_register when device has "
1943                                "no NETIF_F_HW_VLAN_RX\n", netdev->name);
1944                 }
1945         } else {
1946                 /* remove vlan rx stripping. */
1947                 struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
1948                 adapter->vlan_grp = NULL;
1949
1950                 if (devRead->misc.uptFeatures & UPT1_F_RXVLAN) {
1951                         int i;
1952
1953                         for (i = 0; i < VMXNET3_VFT_SIZE; i++) {
1954                                 /* clear entire vfTable; this also disables
1955                                  * VLAN rx filtering
1956                                  */
1957                                 vfTable[i] = 0;
1958                         }
1959                         spin_lock_irqsave(&adapter->cmd_lock, flags);
1960                         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1961                                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1962                         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1963                 }
1964         }
1965 }
1966
1967
1968 static void
1969 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1970 {
1971         if (adapter->vlan_grp) {
1972                 u16 vid;
1973                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1974                 bool activeVlan = false;
1975
1976                 for (vid = 0; vid < VLAN_N_VID; vid++) {
1977                         if (vlan_group_get_device(adapter->vlan_grp, vid)) {
1978                                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1979                                 activeVlan = true;
1980                         }
1981                 }
1982                 if (activeVlan) {
1983                         /* continue to allow untagged pkts */
1984                         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1985                 }
1986         }
1987 }
1988
1989
1990 static void
1991 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1992 {
1993         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1994         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1995         unsigned long flags;
1996
1997         VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1998         spin_lock_irqsave(&adapter->cmd_lock, flags);
1999         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2000                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2001         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2002 }
2003
2004
2005 static void
2006 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
2007 {
2008         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2009         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2010         unsigned long flags;
2011
2012         VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2013         spin_lock_irqsave(&adapter->cmd_lock, flags);
2014         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2015                                VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2016         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2017 }
2018
2019
2020 static u8 *
2021 vmxnet3_copy_mc(struct net_device *netdev)
2022 {
2023         u8 *buf = NULL;
2024         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2025
2026         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2027         if (sz <= 0xffff) {
2028                 /* We may be called with BH disabled */
2029                 buf = kmalloc(sz, GFP_ATOMIC);
2030                 if (buf) {
2031                         struct netdev_hw_addr *ha;
2032                         int i = 0;
2033
2034                         netdev_for_each_mc_addr(ha, netdev)
2035                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2036                                        ETH_ALEN);
2037                 }
2038         }
2039         return buf;
2040 }
2041
2042
2043 static void
2044 vmxnet3_set_mc(struct net_device *netdev)
2045 {
2046         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2047         unsigned long flags;
2048         struct Vmxnet3_RxFilterConf *rxConf =
2049                                         &adapter->shared->devRead.rxFilterConf;
2050         u8 *new_table = NULL;
2051         u32 new_mode = VMXNET3_RXM_UCAST;
2052
2053         if (netdev->flags & IFF_PROMISC)
2054                 new_mode |= VMXNET3_RXM_PROMISC;
2055
2056         if (netdev->flags & IFF_BROADCAST)
2057                 new_mode |= VMXNET3_RXM_BCAST;
2058
2059         if (netdev->flags & IFF_ALLMULTI)
2060                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2061         else
2062                 if (!netdev_mc_empty(netdev)) {
2063                         new_table = vmxnet3_copy_mc(netdev);
2064                         if (new_table) {
2065                                 new_mode |= VMXNET3_RXM_MCAST;
2066                                 rxConf->mfTableLen = cpu_to_le16(
2067                                         netdev_mc_count(netdev) * ETH_ALEN);
2068                                 rxConf->mfTablePA = cpu_to_le64(virt_to_phys(
2069                                                     new_table));
2070                         } else {
2071                                 printk(KERN_INFO "%s: failed to copy mcast list"
2072                                        ", setting ALL_MULTI\n", netdev->name);
2073                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2074                         }
2075                 }
2076
2077
2078         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2079                 rxConf->mfTableLen = 0;
2080                 rxConf->mfTablePA = 0;
2081         }
2082
2083         spin_lock_irqsave(&adapter->cmd_lock, flags);
2084         if (new_mode != rxConf->rxMode) {
2085                 rxConf->rxMode = cpu_to_le32(new_mode);
2086                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2087                                        VMXNET3_CMD_UPDATE_RX_MODE);
2088         }
2089
2090         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2091                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2092         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2093
2094         kfree(new_table);
2095 }
2096
2097 void
2098 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2099 {
2100         int i;
2101
2102         for (i = 0; i < adapter->num_rx_queues; i++)
2103                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2104 }
2105
2106
2107 /*
2108  *   Set up driver_shared based on settings in adapter.
2109  */
2110
2111 static void
2112 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2113 {
2114         struct Vmxnet3_DriverShared *shared = adapter->shared;
2115         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2116         struct Vmxnet3_TxQueueConf *tqc;
2117         struct Vmxnet3_RxQueueConf *rqc;
2118         int i;
2119
2120         memset(shared, 0, sizeof(*shared));
2121
2122         /* driver settings */
2123         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2124         devRead->misc.driverInfo.version = cpu_to_le32(
2125                                                 VMXNET3_DRIVER_VERSION_NUM);
2126         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2127                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2128         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2129         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2130                                 *((u32 *)&devRead->misc.driverInfo.gos));
2131         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2132         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2133
2134         devRead->misc.ddPA = cpu_to_le64(virt_to_phys(adapter));
2135         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2136
2137         /* set up feature flags */
2138         if (adapter->netdev->features & NETIF_F_RXCSUM)
2139                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2140
2141         if (adapter->netdev->features & NETIF_F_LRO) {
2142                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2143                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2144         }
2145         if (adapter->netdev->features & NETIF_F_HW_VLAN_RX)
2146                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2147
2148         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2149         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2150         devRead->misc.queueDescLen = cpu_to_le32(
2151                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2152                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2153
2154         /* tx queue settings */
2155         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2156         for (i = 0; i < adapter->num_tx_queues; i++) {
2157                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2158                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2159                 tqc = &adapter->tqd_start[i].conf;
2160                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2161                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2162                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2163                 tqc->ddPA           = cpu_to_le64(virt_to_phys(tq->buf_info));
2164                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2165                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2166                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2167                 tqc->ddLen          = cpu_to_le32(
2168                                         sizeof(struct vmxnet3_tx_buf_info) *
2169                                         tqc->txRingSize);
2170                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2171         }
2172
2173         /* rx queue settings */
2174         devRead->misc.numRxQueues = adapter->num_rx_queues;
2175         for (i = 0; i < adapter->num_rx_queues; i++) {
2176                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2177                 rqc = &adapter->rqd_start[i].conf;
2178                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2179                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2180                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2181                 rqc->ddPA            = cpu_to_le64(virt_to_phys(
2182                                                         rq->buf_info));
2183                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2184                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2185                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2186                 rqc->ddLen           = cpu_to_le32(
2187                                         sizeof(struct vmxnet3_rx_buf_info) *
2188                                         (rqc->rxRingSize[0] +
2189                                          rqc->rxRingSize[1]));
2190                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2191         }
2192
2193 #ifdef VMXNET3_RSS
2194         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2195
2196         if (adapter->rss) {
2197                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2198                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2199                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2200                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2201                                     UPT1_RSS_HASH_TYPE_IPV4 |
2202                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2203                                     UPT1_RSS_HASH_TYPE_IPV6;
2204                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2205                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2206                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2207                 get_random_bytes(&rssConf->hashKey[0], rssConf->hashKeySize);
2208                 for (i = 0; i < rssConf->indTableSize; i++)
2209                         rssConf->indTable[i] = i % adapter->num_rx_queues;
2210
2211                 devRead->rssConfDesc.confVer = 1;
2212                 devRead->rssConfDesc.confLen = sizeof(*rssConf);
2213                 devRead->rssConfDesc.confPA  = virt_to_phys(rssConf);
2214         }
2215
2216 #endif /* VMXNET3_RSS */
2217
2218         /* intr settings */
2219         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2220                                      VMXNET3_IMM_AUTO;
2221         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2222         for (i = 0; i < adapter->intr.num_intrs; i++)
2223                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2224
2225         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2226         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2227
2228         /* rx filter settings */
2229         devRead->rxFilterConf.rxMode = 0;
2230         vmxnet3_restore_vlan(adapter);
2231         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2232
2233         /* the rest are already zeroed */
2234 }
2235
2236
2237 int
2238 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2239 {
2240         int err, i;
2241         u32 ret;
2242         unsigned long flags;
2243
2244         dev_dbg(&adapter->netdev->dev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2245                 " ring sizes %u %u %u\n", adapter->netdev->name,
2246                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2247                 adapter->tx_queue[0].tx_ring.size,
2248                 adapter->rx_queue[0].rx_ring[0].size,
2249                 adapter->rx_queue[0].rx_ring[1].size);
2250
2251         vmxnet3_tq_init_all(adapter);
2252         err = vmxnet3_rq_init_all(adapter);
2253         if (err) {
2254                 printk(KERN_ERR "Failed to init rx queue for %s: error %d\n",
2255                        adapter->netdev->name, err);
2256                 goto rq_err;
2257         }
2258
2259         err = vmxnet3_request_irqs(adapter);
2260         if (err) {
2261                 printk(KERN_ERR "Failed to setup irq for %s: error %d\n",
2262                        adapter->netdev->name, err);
2263                 goto irq_err;
2264         }
2265
2266         vmxnet3_setup_driver_shared(adapter);
2267
2268         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2269                                adapter->shared_pa));
2270         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2271                                adapter->shared_pa));
2272         spin_lock_irqsave(&adapter->cmd_lock, flags);
2273         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2274                                VMXNET3_CMD_ACTIVATE_DEV);
2275         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2276         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2277
2278         if (ret != 0) {
2279                 printk(KERN_ERR "Failed to activate dev %s: error %u\n",
2280                        adapter->netdev->name, ret);
2281                 err = -EINVAL;
2282                 goto activate_err;
2283         }
2284
2285         for (i = 0; i < adapter->num_rx_queues; i++) {
2286                 VMXNET3_WRITE_BAR0_REG(adapter,
2287                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2288                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2289                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2290                                 (i * VMXNET3_REG_ALIGN)),
2291                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2292         }
2293
2294         /* Apply the rx filter settins last. */
2295         vmxnet3_set_mc(adapter->netdev);
2296
2297         /*
2298          * Check link state when first activating device. It will start the
2299          * tx queue if the link is up.
2300          */
2301         vmxnet3_check_link(adapter, true);
2302         for (i = 0; i < adapter->num_rx_queues; i++)
2303                 napi_enable(&adapter->rx_queue[i].napi);
2304         vmxnet3_enable_all_intrs(adapter);
2305         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2306         return 0;
2307
2308 activate_err:
2309         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2310         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2311         vmxnet3_free_irqs(adapter);
2312 irq_err:
2313 rq_err:
2314         /* free up buffers we allocated */
2315         vmxnet3_rq_cleanup_all(adapter);
2316         return err;
2317 }
2318
2319
2320 void
2321 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2322 {
2323         unsigned long flags;
2324         spin_lock_irqsave(&adapter->cmd_lock, flags);
2325         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2326         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2327 }
2328
2329
2330 int
2331 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2332 {
2333         int i;
2334         unsigned long flags;
2335         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2336                 return 0;
2337
2338
2339         spin_lock_irqsave(&adapter->cmd_lock, flags);
2340         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2341                                VMXNET3_CMD_QUIESCE_DEV);
2342         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2343         vmxnet3_disable_all_intrs(adapter);
2344
2345         for (i = 0; i < adapter->num_rx_queues; i++)
2346                 napi_disable(&adapter->rx_queue[i].napi);
2347         netif_tx_disable(adapter->netdev);
2348         adapter->link_speed = 0;
2349         netif_carrier_off(adapter->netdev);
2350
2351         vmxnet3_tq_cleanup_all(adapter);
2352         vmxnet3_rq_cleanup_all(adapter);
2353         vmxnet3_free_irqs(adapter);
2354         return 0;
2355 }
2356
2357
2358 static void
2359 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2360 {
2361         u32 tmp;
2362
2363         tmp = *(u32 *)mac;
2364         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2365
2366         tmp = (mac[5] << 8) | mac[4];
2367         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2368 }
2369
2370
2371 static int
2372 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2373 {
2374         struct sockaddr *addr = p;
2375         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2376
2377         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2378         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2379
2380         return 0;
2381 }
2382
2383
2384 /* ==================== initialization and cleanup routines ============ */
2385
2386 static int
2387 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2388 {
2389         int err;
2390         unsigned long mmio_start, mmio_len;
2391         struct pci_dev *pdev = adapter->pdev;
2392
2393         err = pci_enable_device(pdev);
2394         if (err) {
2395                 printk(KERN_ERR "Failed to enable adapter %s: error %d\n",
2396                        pci_name(pdev), err);
2397                 return err;
2398         }
2399
2400         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2401                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2402                         printk(KERN_ERR "pci_set_consistent_dma_mask failed "
2403                                "for adapter %s\n", pci_name(pdev));
2404                         err = -EIO;
2405                         goto err_set_mask;
2406                 }
2407                 *dma64 = true;
2408         } else {
2409                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2410                         printk(KERN_ERR "pci_set_dma_mask failed for adapter "
2411                                "%s\n",  pci_name(pdev));
2412                         err = -EIO;
2413                         goto err_set_mask;
2414                 }
2415                 *dma64 = false;
2416         }
2417
2418         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2419                                            vmxnet3_driver_name);
2420         if (err) {
2421                 printk(KERN_ERR "Failed to request region for adapter %s: "
2422                        "error %d\n", pci_name(pdev), err);
2423                 goto err_set_mask;
2424         }
2425
2426         pci_set_master(pdev);
2427
2428         mmio_start = pci_resource_start(pdev, 0);
2429         mmio_len = pci_resource_len(pdev, 0);
2430         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2431         if (!adapter->hw_addr0) {
2432                 printk(KERN_ERR "Failed to map bar0 for adapter %s\n",
2433                        pci_name(pdev));
2434                 err = -EIO;
2435                 goto err_ioremap;
2436         }
2437
2438         mmio_start = pci_resource_start(pdev, 1);
2439         mmio_len = pci_resource_len(pdev, 1);
2440         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2441         if (!adapter->hw_addr1) {
2442                 printk(KERN_ERR "Failed to map bar1 for adapter %s\n",
2443                        pci_name(pdev));
2444                 err = -EIO;
2445                 goto err_bar1;
2446         }
2447         return 0;
2448
2449 err_bar1:
2450         iounmap(adapter->hw_addr0);
2451 err_ioremap:
2452         pci_release_selected_regions(pdev, (1 << 2) - 1);
2453 err_set_mask:
2454         pci_disable_device(pdev);
2455         return err;
2456 }
2457
2458
2459 static void
2460 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2461 {
2462         BUG_ON(!adapter->pdev);
2463
2464         iounmap(adapter->hw_addr0);
2465         iounmap(adapter->hw_addr1);
2466         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2467         pci_disable_device(adapter->pdev);
2468 }
2469
2470
2471 static void
2472 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2473 {
2474         size_t sz, i, ring0_size, ring1_size, comp_size;
2475         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2476
2477
2478         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2479                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2480                 adapter->skb_buf_size = adapter->netdev->mtu +
2481                                         VMXNET3_MAX_ETH_HDR_SIZE;
2482                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2483                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2484
2485                 adapter->rx_buf_per_pkt = 1;
2486         } else {
2487                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2488                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2489                                             VMXNET3_MAX_ETH_HDR_SIZE;
2490                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2491         }
2492
2493         /*
2494          * for simplicity, force the ring0 size to be a multiple of
2495          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2496          */
2497         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2498         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2499         ring0_size = (ring0_size + sz - 1) / sz * sz;
2500         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2501                            sz * sz);
2502         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2503         comp_size = ring0_size + ring1_size;
2504
2505         for (i = 0; i < adapter->num_rx_queues; i++) {
2506                 rq = &adapter->rx_queue[i];
2507                 rq->rx_ring[0].size = ring0_size;
2508                 rq->rx_ring[1].size = ring1_size;
2509                 rq->comp_ring.size = comp_size;
2510         }
2511 }
2512
2513
2514 int
2515 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2516                       u32 rx_ring_size, u32 rx_ring2_size)
2517 {
2518         int err = 0, i;
2519
2520         for (i = 0; i < adapter->num_tx_queues; i++) {
2521                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2522                 tq->tx_ring.size   = tx_ring_size;
2523                 tq->data_ring.size = tx_ring_size;
2524                 tq->comp_ring.size = tx_ring_size;
2525                 tq->shared = &adapter->tqd_start[i].ctrl;
2526                 tq->stopped = true;
2527                 tq->adapter = adapter;
2528                 tq->qid = i;
2529                 err = vmxnet3_tq_create(tq, adapter);
2530                 /*
2531                  * Too late to change num_tx_queues. We cannot do away with
2532                  * lesser number of queues than what we asked for
2533                  */
2534                 if (err)
2535                         goto queue_err;
2536         }
2537
2538         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2539         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2540         vmxnet3_adjust_rx_ring_size(adapter);
2541         for (i = 0; i < adapter->num_rx_queues; i++) {
2542                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2543                 /* qid and qid2 for rx queues will be assigned later when num
2544                  * of rx queues is finalized after allocating intrs */
2545                 rq->shared = &adapter->rqd_start[i].ctrl;
2546                 rq->adapter = adapter;
2547                 err = vmxnet3_rq_create(rq, adapter);
2548                 if (err) {
2549                         if (i == 0) {
2550                                 printk(KERN_ERR "Could not allocate any rx"
2551                                        "queues. Aborting.\n");
2552                                 goto queue_err;
2553                         } else {
2554                                 printk(KERN_INFO "Number of rx queues changed "
2555                                        "to : %d.\n", i);
2556                                 adapter->num_rx_queues = i;
2557                                 err = 0;
2558                                 break;
2559                         }
2560                 }
2561         }
2562         return err;
2563 queue_err:
2564         vmxnet3_tq_destroy_all(adapter);
2565         return err;
2566 }
2567
2568 static int
2569 vmxnet3_open(struct net_device *netdev)
2570 {
2571         struct vmxnet3_adapter *adapter;
2572         int err, i;
2573
2574         adapter = netdev_priv(netdev);
2575
2576         for (i = 0; i < adapter->num_tx_queues; i++)
2577                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2578
2579         err = vmxnet3_create_queues(adapter, VMXNET3_DEF_TX_RING_SIZE,
2580                                     VMXNET3_DEF_RX_RING_SIZE,
2581                                     VMXNET3_DEF_RX_RING_SIZE);
2582         if (err)
2583                 goto queue_err;
2584
2585         err = vmxnet3_activate_dev(adapter);
2586         if (err)
2587                 goto activate_err;
2588
2589         return 0;
2590
2591 activate_err:
2592         vmxnet3_rq_destroy_all(adapter);
2593         vmxnet3_tq_destroy_all(adapter);
2594 queue_err:
2595         return err;
2596 }
2597
2598
2599 static int
2600 vmxnet3_close(struct net_device *netdev)
2601 {
2602         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2603
2604         /*
2605          * Reset_work may be in the middle of resetting the device, wait for its
2606          * completion.
2607          */
2608         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2609                 msleep(1);
2610
2611         vmxnet3_quiesce_dev(adapter);
2612
2613         vmxnet3_rq_destroy_all(adapter);
2614         vmxnet3_tq_destroy_all(adapter);
2615
2616         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2617
2618
2619         return 0;
2620 }
2621
2622
2623 void
2624 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2625 {
2626         int i;
2627
2628         /*
2629          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2630          * vmxnet3_close() will deadlock.
2631          */
2632         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2633
2634         /* we need to enable NAPI, otherwise dev_close will deadlock */
2635         for (i = 0; i < adapter->num_rx_queues; i++)
2636                 napi_enable(&adapter->rx_queue[i].napi);
2637         dev_close(adapter->netdev);
2638 }
2639
2640
2641 static int
2642 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2643 {
2644         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2645         int err = 0;
2646
2647         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2648                 return -EINVAL;
2649
2650         netdev->mtu = new_mtu;
2651
2652         /*
2653          * Reset_work may be in the middle of resetting the device, wait for its
2654          * completion.
2655          */
2656         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2657                 msleep(1);
2658
2659         if (netif_running(netdev)) {
2660                 vmxnet3_quiesce_dev(adapter);
2661                 vmxnet3_reset_dev(adapter);
2662
2663                 /* we need to re-create the rx queue based on the new mtu */
2664                 vmxnet3_rq_destroy_all(adapter);
2665                 vmxnet3_adjust_rx_ring_size(adapter);
2666                 err = vmxnet3_rq_create_all(adapter);
2667                 if (err) {
2668                         printk(KERN_ERR "%s: failed to re-create rx queues,"
2669                                 " error %d. Closing it.\n", netdev->name, err);
2670                         goto out;
2671                 }
2672
2673                 err = vmxnet3_activate_dev(adapter);
2674                 if (err) {
2675                         printk(KERN_ERR "%s: failed to re-activate, error %d. "
2676                                 "Closing it\n", netdev->name, err);
2677                         goto out;
2678                 }
2679         }
2680
2681 out:
2682         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2683         if (err)
2684                 vmxnet3_force_close(adapter);
2685
2686         return err;
2687 }
2688
2689
2690 static void
2691 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2692 {
2693         struct net_device *netdev = adapter->netdev;
2694
2695         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2696                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_TX |
2697                 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_LRO;
2698         if (dma64)
2699                 netdev->features |= NETIF_F_HIGHDMA;
2700         netdev->vlan_features = netdev->hw_features & ~NETIF_F_HW_VLAN_TX;
2701         netdev->features = netdev->hw_features |
2702                 NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_FILTER;
2703
2704         netdev_info(adapter->netdev,
2705                 "features: sg csum vlan jf tso tsoIPv6 lro%s\n",
2706                 dma64 ? " highDMA" : "");
2707 }
2708
2709
2710 static void
2711 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2712 {
2713         u32 tmp;
2714
2715         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2716         *(u32 *)mac = tmp;
2717
2718         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2719         mac[4] = tmp & 0xff;
2720         mac[5] = (tmp >> 8) & 0xff;
2721 }
2722
2723 #ifdef CONFIG_PCI_MSI
2724
2725 /*
2726  * Enable MSIx vectors.
2727  * Returns :
2728  *      0 on successful enabling of required vectors,
2729  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2730  *       could be enabled.
2731  *      number of vectors which can be enabled otherwise (this number is smaller
2732  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2733  */
2734
2735 static int
2736 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter,
2737                              int vectors)
2738 {
2739         int err = 0, vector_threshold;
2740         vector_threshold = VMXNET3_LINUX_MIN_MSIX_VECT;
2741
2742         while (vectors >= vector_threshold) {
2743                 err = pci_enable_msix(adapter->pdev, adapter->intr.msix_entries,
2744                                       vectors);
2745                 if (!err) {
2746                         adapter->intr.num_intrs = vectors;
2747                         return 0;
2748                 } else if (err < 0) {
2749                         printk(KERN_ERR "Failed to enable MSI-X for %s, error"
2750                                " %d\n", adapter->netdev->name, err);
2751                         vectors = 0;
2752                 } else if (err < vector_threshold) {
2753                         break;
2754                 } else {
2755                         /* If fails to enable required number of MSI-x vectors
2756                          * try enabling minimum number of vectors required.
2757                          */
2758                         vectors = vector_threshold;
2759                         printk(KERN_ERR "Failed to enable %d MSI-X for %s, try"
2760                                " %d instead\n", vectors, adapter->netdev->name,
2761                                vector_threshold);
2762                 }
2763         }
2764
2765         printk(KERN_INFO "Number of MSI-X interrupts which can be allocatedi"
2766                " are lower than min threshold required.\n");
2767         return err;
2768 }
2769
2770
2771 #endif /* CONFIG_PCI_MSI */
2772
2773 static void
2774 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2775 {
2776         u32 cfg;
2777         unsigned long flags;
2778
2779         /* intr settings */
2780         spin_lock_irqsave(&adapter->cmd_lock, flags);
2781         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2782                                VMXNET3_CMD_GET_CONF_INTR);
2783         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2784         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2785         adapter->intr.type = cfg & 0x3;
2786         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2787
2788         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2789                 adapter->intr.type = VMXNET3_IT_MSIX;
2790         }
2791
2792 #ifdef CONFIG_PCI_MSI
2793         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2794                 int vector, err = 0;
2795
2796                 adapter->intr.num_intrs = (adapter->share_intr ==
2797                                            VMXNET3_INTR_TXSHARE) ? 1 :
2798                                            adapter->num_tx_queues;
2799                 adapter->intr.num_intrs += (adapter->share_intr ==
2800                                            VMXNET3_INTR_BUDDYSHARE) ? 0 :
2801                                            adapter->num_rx_queues;
2802                 adapter->intr.num_intrs += 1;           /* for link event */
2803
2804                 adapter->intr.num_intrs = (adapter->intr.num_intrs >
2805                                            VMXNET3_LINUX_MIN_MSIX_VECT
2806                                            ? adapter->intr.num_intrs :
2807                                            VMXNET3_LINUX_MIN_MSIX_VECT);
2808
2809                 for (vector = 0; vector < adapter->intr.num_intrs; vector++)
2810                         adapter->intr.msix_entries[vector].entry = vector;
2811
2812                 err = vmxnet3_acquire_msix_vectors(adapter,
2813                                                    adapter->intr.num_intrs);
2814                 /* If we cannot allocate one MSIx vector per queue
2815                  * then limit the number of rx queues to 1
2816                  */
2817                 if (err == VMXNET3_LINUX_MIN_MSIX_VECT) {
2818                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2819                             || adapter->num_rx_queues != 1) {
2820                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2821                                 printk(KERN_ERR "Number of rx queues : 1\n");
2822                                 adapter->num_rx_queues = 1;
2823                                 adapter->intr.num_intrs =
2824                                                 VMXNET3_LINUX_MIN_MSIX_VECT;
2825                         }
2826                         return;
2827                 }
2828                 if (!err)
2829                         return;
2830
2831                 /* If we cannot allocate MSIx vectors use only one rx queue */
2832                 printk(KERN_INFO "Failed to enable MSI-X for %s, error %d."
2833                        "#rx queues : 1, try MSI\n", adapter->netdev->name, err);
2834
2835                 adapter->intr.type = VMXNET3_IT_MSI;
2836         }
2837
2838         if (adapter->intr.type == VMXNET3_IT_MSI) {
2839                 int err;
2840                 err = pci_enable_msi(adapter->pdev);
2841                 if (!err) {
2842                         adapter->num_rx_queues = 1;
2843                         adapter->intr.num_intrs = 1;
2844                         return;
2845                 }
2846         }
2847 #endif /* CONFIG_PCI_MSI */
2848
2849         adapter->num_rx_queues = 1;
2850         printk(KERN_INFO "Using INTx interrupt, #Rx queues: 1.\n");
2851         adapter->intr.type = VMXNET3_IT_INTX;
2852
2853         /* INT-X related setting */
2854         adapter->intr.num_intrs = 1;
2855 }
2856
2857
2858 static void
2859 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2860 {
2861         if (adapter->intr.type == VMXNET3_IT_MSIX)
2862                 pci_disable_msix(adapter->pdev);
2863         else if (adapter->intr.type == VMXNET3_IT_MSI)
2864                 pci_disable_msi(adapter->pdev);
2865         else
2866                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2867 }
2868
2869
2870 static void
2871 vmxnet3_tx_timeout(struct net_device *netdev)
2872 {
2873         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2874         adapter->tx_timeout_count++;
2875
2876         printk(KERN_ERR "%s: tx hang\n", adapter->netdev->name);
2877         schedule_work(&adapter->work);
2878         netif_wake_queue(adapter->netdev);
2879 }
2880
2881
2882 static void
2883 vmxnet3_reset_work(struct work_struct *data)
2884 {
2885         struct vmxnet3_adapter *adapter;
2886
2887         adapter = container_of(data, struct vmxnet3_adapter, work);
2888
2889         /* if another thread is resetting the device, no need to proceed */
2890         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2891                 return;
2892
2893         /* if the device is closed, we must leave it alone */
2894         rtnl_lock();
2895         if (netif_running(adapter->netdev)) {
2896                 printk(KERN_INFO "%s: resetting\n", adapter->netdev->name);
2897                 vmxnet3_quiesce_dev(adapter);
2898                 vmxnet3_reset_dev(adapter);
2899                 vmxnet3_activate_dev(adapter);
2900         } else {
2901                 printk(KERN_INFO "%s: already closed\n", adapter->netdev->name);
2902         }
2903         rtnl_unlock();
2904
2905         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2906 }
2907
2908
2909 static int __devinit
2910 vmxnet3_probe_device(struct pci_dev *pdev,
2911                      const struct pci_device_id *id)
2912 {
2913         static const struct net_device_ops vmxnet3_netdev_ops = {
2914                 .ndo_open = vmxnet3_open,
2915                 .ndo_stop = vmxnet3_close,
2916                 .ndo_start_xmit = vmxnet3_xmit_frame,
2917                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2918                 .ndo_change_mtu = vmxnet3_change_mtu,
2919                 .ndo_set_features = vmxnet3_set_features,
2920                 .ndo_get_stats = vmxnet3_get_stats,
2921                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2922                 .ndo_set_multicast_list = vmxnet3_set_mc,
2923                 .ndo_vlan_rx_register = vmxnet3_vlan_rx_register,
2924                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2925                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2926 #ifdef CONFIG_NET_POLL_CONTROLLER
2927                 .ndo_poll_controller = vmxnet3_netpoll,
2928 #endif
2929         };
2930         int err;
2931         bool dma64 = false; /* stupid gcc */
2932         u32 ver;
2933         struct net_device *netdev;
2934         struct vmxnet3_adapter *adapter;
2935         u8 mac[ETH_ALEN];
2936         int size;
2937         int num_tx_queues;
2938         int num_rx_queues;
2939
2940         if (!pci_msi_enabled())
2941                 enable_mq = 0;
2942
2943 #ifdef VMXNET3_RSS
2944         if (enable_mq)
2945                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2946                                     (int)num_online_cpus());
2947         else
2948 #endif
2949                 num_rx_queues = 1;
2950         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2951
2952         if (enable_mq)
2953                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2954                                     (int)num_online_cpus());
2955         else
2956                 num_tx_queues = 1;
2957
2958         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2959         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2960                                    max(num_tx_queues, num_rx_queues));
2961         printk(KERN_INFO "# of Tx queues : %d, # of Rx queues : %d\n",
2962                num_tx_queues, num_rx_queues);
2963
2964         if (!netdev) {
2965                 printk(KERN_ERR "Failed to alloc ethernet device for adapter "
2966                         "%s\n", pci_name(pdev));
2967                 return -ENOMEM;
2968         }
2969
2970         pci_set_drvdata(pdev, netdev);
2971         adapter = netdev_priv(netdev);
2972         adapter->netdev = netdev;
2973         adapter->pdev = pdev;
2974
2975         spin_lock_init(&adapter->cmd_lock);
2976         adapter->shared = pci_alloc_consistent(adapter->pdev,
2977                           sizeof(struct Vmxnet3_DriverShared),
2978                           &adapter->shared_pa);
2979         if (!adapter->shared) {
2980                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2981                         pci_name(pdev));
2982                 err = -ENOMEM;
2983                 goto err_alloc_shared;
2984         }
2985
2986         adapter->num_rx_queues = num_rx_queues;
2987         adapter->num_tx_queues = num_tx_queues;
2988
2989         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2990         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2991         adapter->tqd_start = pci_alloc_consistent(adapter->pdev, size,
2992                              &adapter->queue_desc_pa);
2993
2994         if (!adapter->tqd_start) {
2995                 printk(KERN_ERR "Failed to allocate memory for %s\n",
2996                         pci_name(pdev));
2997                 err = -ENOMEM;
2998                 goto err_alloc_queue_desc;
2999         }
3000         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3001                                                         adapter->num_tx_queues);
3002
3003         adapter->pm_conf = kmalloc(sizeof(struct Vmxnet3_PMConf), GFP_KERNEL);
3004         if (adapter->pm_conf == NULL) {
3005                 printk(KERN_ERR "Failed to allocate memory for %s\n",
3006                         pci_name(pdev));
3007                 err = -ENOMEM;
3008                 goto err_alloc_pm;
3009         }
3010
3011 #ifdef VMXNET3_RSS
3012
3013         adapter->rss_conf = kmalloc(sizeof(struct UPT1_RSSConf), GFP_KERNEL);
3014         if (adapter->rss_conf == NULL) {
3015                 printk(KERN_ERR "Failed to allocate memory for %s\n",
3016                        pci_name(pdev));
3017                 err = -ENOMEM;
3018                 goto err_alloc_rss;
3019         }
3020 #endif /* VMXNET3_RSS */
3021
3022         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
3023         if (err < 0)
3024                 goto err_alloc_pci;
3025
3026         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3027         if (ver & 1) {
3028                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
3029         } else {
3030                 printk(KERN_ERR "Incompatible h/w version (0x%x) for adapter"
3031                        " %s\n", ver, pci_name(pdev));
3032                 err = -EBUSY;
3033                 goto err_ver;
3034         }
3035
3036         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3037         if (ver & 1) {
3038                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3039         } else {
3040                 printk(KERN_ERR "Incompatible upt version (0x%x) for "
3041                        "adapter %s\n", ver, pci_name(pdev));
3042                 err = -EBUSY;
3043                 goto err_ver;
3044         }
3045
3046         vmxnet3_declare_features(adapter, dma64);
3047
3048         adapter->dev_number = atomic_read(&devices_found);
3049
3050          adapter->share_intr = irq_share_mode;
3051         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE &&
3052             adapter->num_tx_queues != adapter->num_rx_queues)
3053                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3054
3055         vmxnet3_alloc_intr_resources(adapter);
3056
3057 #ifdef VMXNET3_RSS
3058         if (adapter->num_rx_queues > 1 &&
3059             adapter->intr.type == VMXNET3_IT_MSIX) {
3060                 adapter->rss = true;
3061                 printk(KERN_INFO "RSS is enabled.\n");
3062         } else {
3063                 adapter->rss = false;
3064         }
3065 #endif
3066
3067         vmxnet3_read_mac_addr(adapter, mac);
3068         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3069
3070         netdev->netdev_ops = &vmxnet3_netdev_ops;
3071         vmxnet3_set_ethtool_ops(netdev);
3072         netdev->watchdog_timeo = 5 * HZ;
3073
3074         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3075
3076         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3077                 int i;
3078                 for (i = 0; i < adapter->num_rx_queues; i++) {
3079                         netif_napi_add(adapter->netdev,
3080                                        &adapter->rx_queue[i].napi,
3081                                        vmxnet3_poll_rx_only, 64);
3082                 }
3083         } else {
3084                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3085                                vmxnet3_poll, 64);
3086         }
3087
3088         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3089         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3090
3091         SET_NETDEV_DEV(netdev, &pdev->dev);
3092         err = register_netdev(netdev);
3093
3094         if (err) {
3095                 printk(KERN_ERR "Failed to register adapter %s\n",
3096                         pci_name(pdev));
3097                 goto err_register;
3098         }
3099
3100         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3101         vmxnet3_check_link(adapter, false);
3102         atomic_inc(&devices_found);
3103         return 0;
3104
3105 err_register:
3106         vmxnet3_free_intr_resources(adapter);
3107 err_ver:
3108         vmxnet3_free_pci_resources(adapter);
3109 err_alloc_pci:
3110 #ifdef VMXNET3_RSS
3111         kfree(adapter->rss_conf);
3112 err_alloc_rss:
3113 #endif
3114         kfree(adapter->pm_conf);
3115 err_alloc_pm:
3116         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3117                             adapter->queue_desc_pa);
3118 err_alloc_queue_desc:
3119         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3120                             adapter->shared, adapter->shared_pa);
3121 err_alloc_shared:
3122         pci_set_drvdata(pdev, NULL);
3123         free_netdev(netdev);
3124         return err;
3125 }
3126
3127
3128 static void __devexit
3129 vmxnet3_remove_device(struct pci_dev *pdev)
3130 {
3131         struct net_device *netdev = pci_get_drvdata(pdev);
3132         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3133         int size = 0;
3134         int num_rx_queues;
3135
3136 #ifdef VMXNET3_RSS
3137         if (enable_mq)
3138                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3139                                     (int)num_online_cpus());
3140         else
3141 #endif
3142                 num_rx_queues = 1;
3143         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3144
3145         cancel_work_sync(&adapter->work);
3146
3147         unregister_netdev(netdev);
3148
3149         vmxnet3_free_intr_resources(adapter);
3150         vmxnet3_free_pci_resources(adapter);
3151 #ifdef VMXNET3_RSS
3152         kfree(adapter->rss_conf);
3153 #endif
3154         kfree(adapter->pm_conf);
3155
3156         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3157         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3158         pci_free_consistent(adapter->pdev, size, adapter->tqd_start,
3159                             adapter->queue_desc_pa);
3160         pci_free_consistent(adapter->pdev, sizeof(struct Vmxnet3_DriverShared),
3161                             adapter->shared, adapter->shared_pa);
3162         free_netdev(netdev);
3163 }
3164
3165
3166 #ifdef CONFIG_PM
3167
3168 static int
3169 vmxnet3_suspend(struct device *device)
3170 {
3171         struct pci_dev *pdev = to_pci_dev(device);
3172         struct net_device *netdev = pci_get_drvdata(pdev);
3173         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3174         struct Vmxnet3_PMConf *pmConf;
3175         struct ethhdr *ehdr;
3176         struct arphdr *ahdr;
3177         u8 *arpreq;
3178         struct in_device *in_dev;
3179         struct in_ifaddr *ifa;
3180         unsigned long flags;
3181         int i = 0;
3182
3183         if (!netif_running(netdev))
3184                 return 0;
3185
3186         for (i = 0; i < adapter->num_rx_queues; i++)
3187                 napi_disable(&adapter->rx_queue[i].napi);
3188
3189         vmxnet3_disable_all_intrs(adapter);
3190         vmxnet3_free_irqs(adapter);
3191         vmxnet3_free_intr_resources(adapter);
3192
3193         netif_device_detach(netdev);
3194         netif_tx_stop_all_queues(netdev);
3195
3196         /* Create wake-up filters. */
3197         pmConf = adapter->pm_conf;
3198         memset(pmConf, 0, sizeof(*pmConf));
3199
3200         if (adapter->wol & WAKE_UCAST) {
3201                 pmConf->filters[i].patternSize = ETH_ALEN;
3202                 pmConf->filters[i].maskSize = 1;
3203                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3204                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3205
3206                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3207                 i++;
3208         }
3209
3210         if (adapter->wol & WAKE_ARP) {
3211                 in_dev = in_dev_get(netdev);
3212                 if (!in_dev)
3213                         goto skip_arp;
3214
3215                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3216                 if (!ifa)
3217                         goto skip_arp;
3218
3219                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3220                         sizeof(struct arphdr) +         /* ARP header */
3221                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3222                         2 * sizeof(u32);        /*2 IPv4 addresses */
3223                 pmConf->filters[i].maskSize =
3224                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3225
3226                 /* ETH_P_ARP in Ethernet header. */
3227                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3228                 ehdr->h_proto = htons(ETH_P_ARP);
3229
3230                 /* ARPOP_REQUEST in ARP header. */
3231                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3232                 ahdr->ar_op = htons(ARPOP_REQUEST);
3233                 arpreq = (u8 *)(ahdr + 1);
3234
3235                 /* The Unicast IPv4 address in 'tip' field. */
3236                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3237                 *(u32 *)arpreq = ifa->ifa_address;
3238
3239                 /* The mask for the relevant bits. */
3240                 pmConf->filters[i].mask[0] = 0x00;
3241                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3242                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3243                 pmConf->filters[i].mask[3] = 0x00;
3244                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3245                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3246                 in_dev_put(in_dev);
3247
3248                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3249                 i++;
3250         }
3251
3252 skip_arp:
3253         if (adapter->wol & WAKE_MAGIC)
3254                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3255
3256         pmConf->numFilters = i;
3257
3258         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3259         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3260                                                                   *pmConf));
3261         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3262                                                                  pmConf));
3263
3264         spin_lock_irqsave(&adapter->cmd_lock, flags);
3265         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3266                                VMXNET3_CMD_UPDATE_PMCFG);
3267         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3268
3269         pci_save_state(pdev);
3270         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3271                         adapter->wol);
3272         pci_disable_device(pdev);
3273         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3274
3275         return 0;
3276 }
3277
3278
3279 static int
3280 vmxnet3_resume(struct device *device)
3281 {
3282         int err, i = 0;
3283         unsigned long flags;
3284         struct pci_dev *pdev = to_pci_dev(device);
3285         struct net_device *netdev = pci_get_drvdata(pdev);
3286         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3287         struct Vmxnet3_PMConf *pmConf;
3288
3289         if (!netif_running(netdev))
3290                 return 0;
3291
3292         /* Destroy wake-up filters. */
3293         pmConf = adapter->pm_conf;
3294         memset(pmConf, 0, sizeof(*pmConf));
3295
3296         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3297         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3298                                                                   *pmConf));
3299         adapter->shared->devRead.pmConfDesc.confPA = cpu_to_le64(virt_to_phys(
3300                                                                  pmConf));
3301
3302         netif_device_attach(netdev);
3303         pci_set_power_state(pdev, PCI_D0);
3304         pci_restore_state(pdev);
3305         err = pci_enable_device_mem(pdev);
3306         if (err != 0)
3307                 return err;
3308
3309         pci_enable_wake(pdev, PCI_D0, 0);
3310
3311         spin_lock_irqsave(&adapter->cmd_lock, flags);
3312         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3313                                VMXNET3_CMD_UPDATE_PMCFG);
3314         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3315         vmxnet3_alloc_intr_resources(adapter);
3316         vmxnet3_request_irqs(adapter);
3317         for (i = 0; i < adapter->num_rx_queues; i++)
3318                 napi_enable(&adapter->rx_queue[i].napi);
3319         vmxnet3_enable_all_intrs(adapter);
3320
3321         return 0;
3322 }
3323
3324 static const struct dev_pm_ops vmxnet3_pm_ops = {
3325         .suspend = vmxnet3_suspend,
3326         .resume = vmxnet3_resume,
3327 };
3328 #endif
3329
3330 static struct pci_driver vmxnet3_driver = {
3331         .name           = vmxnet3_driver_name,
3332         .id_table       = vmxnet3_pciid_table,
3333         .probe          = vmxnet3_probe_device,
3334         .remove         = __devexit_p(vmxnet3_remove_device),
3335 #ifdef CONFIG_PM
3336         .driver.pm      = &vmxnet3_pm_ops,
3337 #endif
3338 };
3339
3340
3341 static int __init
3342 vmxnet3_init_module(void)
3343 {
3344         printk(KERN_INFO "%s - version %s\n", VMXNET3_DRIVER_DESC,
3345                 VMXNET3_DRIVER_VERSION_REPORT);
3346         return pci_register_driver(&vmxnet3_driver);
3347 }
3348
3349 module_init(vmxnet3_init_module);
3350
3351
3352 static void
3353 vmxnet3_exit_module(void)
3354 {
3355         pci_unregister_driver(&vmxnet3_driver);
3356 }
3357
3358 module_exit(vmxnet3_exit_module);
3359
3360 MODULE_AUTHOR("VMware, Inc.");
3361 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3362 MODULE_LICENSE("GPL v2");
3363 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);