drivers/net/vmxnet3/vmxnet3_drv.c

   1 /*
   2  * Linux driver for VMware's vmxnet3 ethernet NIC.
   3  *
   4  * Copyright (C) 2008-2009, VMware, Inc. All Rights Reserved.
   5  *
   6  * This program is free software; you can redistribute it and/or modify it
   7  * under the terms of the GNU General Public License as published by the
   8  * Free Software Foundation; version 2 of the License and no later version.
   9  *
  10  * This program is distributed in the hope that it will be useful, but
  11  * WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
  13  * NON INFRINGEMENT. See the GNU General Public License for more
  14  * details.
  15  *
  16  * You should have received a copy of the GNU General Public License
  17  * along with this program; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  19  *
  20  * The full GNU General Public License is included in this distribution in
  21  * the file called "COPYING".
  22  *
  23  * Maintained by: Shreyas Bhatewara <pv-drivers@vmware.com>
  24  *
  25  */
  26
  27 #include <linux/module.h>
  28 #include <net/ip6_checksum.h>
  29
  30 #include "vmxnet3_int.h"
  31
  32 char vmxnet3_driver_name[] = "vmxnet3";
  33 #define VMXNET3_DRIVER_DESC "VMware vmxnet3 virtual NIC driver"
  34
  35 /*
  36  * PCI Device ID Table
  37  * Last entry must be all 0s
  38  */
  39 static const struct pci_device_id vmxnet3_pciid_table[] = {
  40         {PCI_VDEVICE(VMWARE, PCI_DEVICE_ID_VMWARE_VMXNET3)},
  41         {0}
  42 };
  43
  44 MODULE_DEVICE_TABLE(pci, vmxnet3_pciid_table);
  45
  46 static int enable_mq = 1;
  47
  48 static void
  49 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac);
  50
  51 /*
  52  *    Enable/Disable the given intr
  53  */
  54 static void
  55 vmxnet3_enable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  56 {
  57         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 0);
  58 }
  59
  60
  61 static void
  62 vmxnet3_disable_intr(struct vmxnet3_adapter *adapter, unsigned intr_idx)
  63 {
  64         VMXNET3_WRITE_BAR0_REG(adapter, VMXNET3_REG_IMR + intr_idx * 8, 1);
  65 }
  66
  67
  68 /*
  69  *    Enable/Disable all intrs used by the device
  70  */
  71 static void
  72 vmxnet3_enable_all_intrs(struct vmxnet3_adapter *adapter)
  73 {
  74         int i;
  75
  76         for (i = 0; i < adapter->intr.num_intrs; i++)
  77                 vmxnet3_enable_intr(adapter, i);
  78         adapter->shared->devRead.intrConf.intrCtrl &=
  79                                         cpu_to_le32(~VMXNET3_IC_DISABLE_ALL);
  80 }
  81
  82
  83 static void
  84 vmxnet3_disable_all_intrs(struct vmxnet3_adapter *adapter)
  85 {
  86         int i;
  87
  88         adapter->shared->devRead.intrConf.intrCtrl |=
  89                                         cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
  90         for (i = 0; i < adapter->intr.num_intrs; i++)
  91                 vmxnet3_disable_intr(adapter, i);
  92 }
  93
  94
  95 static void
  96 vmxnet3_ack_events(struct vmxnet3_adapter *adapter, u32 events)
  97 {
  98         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_ECR, events);
  99 }
 100
 101
 102 static bool
 103 vmxnet3_tq_stopped(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 104 {
 105         return tq->stopped;
 106 }
 107
 108
 109 static void
 110 vmxnet3_tq_start(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 111 {
 112         tq->stopped = false;
 113         netif_start_subqueue(adapter->netdev, tq - adapter->tx_queue);
 114 }
 115
 116
 117 static void
 118 vmxnet3_tq_wake(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 119 {
 120         tq->stopped = false;
 121         netif_wake_subqueue(adapter->netdev, (tq - adapter->tx_queue));
 122 }
 123
 124
 125 static void
 126 vmxnet3_tq_stop(struct vmxnet3_tx_queue *tq, struct vmxnet3_adapter *adapter)
 127 {
 128         tq->stopped = true;
 129         tq->num_stop++;
 130         netif_stop_subqueue(adapter->netdev, (tq - adapter->tx_queue));
 131 }
 132
 133
 134 /*
 135  * Check the link state. This may start or stop the tx queue.
 136  */
 137 static void
 138 vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue)
 139 {
 140         u32 ret;
 141         int i;
 142         unsigned long flags;
 143
 144         spin_lock_irqsave(&adapter->cmd_lock, flags);
 145         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
 146         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
 147         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
 148
 149         adapter->link_speed = ret >> 16;
 150         if (ret & 1) { /* Link is up. */
 151                 netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n",
 152                             adapter->link_speed);
 153                 netif_carrier_on(adapter->netdev);
 154
 155                 if (affectTxQueue) {
 156                         for (i = 0; i < adapter->num_tx_queues; i++)
 157                                 vmxnet3_tq_start(&adapter->tx_queue[i],
 158                                                  adapter);
 159                 }
 160         } else {
 161                 netdev_info(adapter->netdev, "NIC Link is Down\n");
 162                 netif_carrier_off(adapter->netdev);
 163
 164                 if (affectTxQueue) {
 165                         for (i = 0; i < adapter->num_tx_queues; i++)
 166                                 vmxnet3_tq_stop(&adapter->tx_queue[i], adapter);
 167                 }
 168         }
 169 }
 170
 171 static void
 172 vmxnet3_process_events(struct vmxnet3_adapter *adapter)
 173 {
 174         int i;
 175         unsigned long flags;
 176         u32 events = le32_to_cpu(adapter->shared->ecr);
 177         if (!events)
 178                 return;
 179
 180         vmxnet3_ack_events(adapter, events);
 181
 182         /* Check if link state has changed */
 183         if (events & VMXNET3_ECR_LINK)
 184                 vmxnet3_check_link(adapter, true);
 185
 186         /* Check if there is an error on xmit/recv queues */
 187         if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
 188                 spin_lock_irqsave(&adapter->cmd_lock, flags);
 189                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
 190                                        VMXNET3_CMD_GET_QUEUE_STATUS);
 191                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
 192
 193                 for (i = 0; i < adapter->num_tx_queues; i++)
 194                         if (adapter->tqd_start[i].status.stopped)
 195                                 dev_err(&adapter->netdev->dev,
 196                                         "%s: tq[%d] error 0x%x\n",
 197                                         adapter->netdev->name, i, le32_to_cpu(
 198                                         adapter->tqd_start[i].status.error));
 199                 for (i = 0; i < adapter->num_rx_queues; i++)
 200                         if (adapter->rqd_start[i].status.stopped)
 201                                 dev_err(&adapter->netdev->dev,
 202                                         "%s: rq[%d] error 0x%x\n",
 203                                         adapter->netdev->name, i,
 204                                         adapter->rqd_start[i].status.error);
 205
 206                 schedule_work(&adapter->work);
 207         }
 208 }
 209
 210 #ifdef __BIG_ENDIAN_BITFIELD
 211 /*
 212  * The device expects the bitfields in shared structures to be written in
 213  * little endian. When CPU is big endian, the following routines are used to
 214  * correctly read and write into ABI.
 215  * The general technique used here is : double word bitfields are defined in
 216  * opposite order for big endian architecture. Then before reading them in
 217  * driver the complete double word is translated using le32_to_cpu. Similarly
 218  * After the driver writes into bitfields, cpu_to_le32 is used to translate the
 219  * double words into required format.
 220  * In order to avoid touching bits in shared structure more than once, temporary
 221  * descriptors are used. These are passed as srcDesc to following functions.
 222  */
 223 static void vmxnet3_RxDescToCPU(const struct Vmxnet3_RxDesc *srcDesc,
 224                                 struct Vmxnet3_RxDesc *dstDesc)
 225 {
 226         u32 *src = (u32 *)srcDesc + 2;
 227         u32 *dst = (u32 *)dstDesc + 2;
 228         dstDesc->addr = le64_to_cpu(srcDesc->addr);
 229         *dst = le32_to_cpu(*src);
 230         dstDesc->ext1 = le32_to_cpu(srcDesc->ext1);
 231 }
 232
 233 static void vmxnet3_TxDescToLe(const struct Vmxnet3_TxDesc *srcDesc,
 234                                struct Vmxnet3_TxDesc *dstDesc)
 235 {
 236         int i;
 237         u32 *src = (u32 *)(srcDesc + 1);
 238         u32 *dst = (u32 *)(dstDesc + 1);
 239
 240         /* Working backwards so that the gen bit is set at the end. */
 241         for (i = 2; i > 0; i--) {
 242                 src--;
 243                 dst--;
 244                 *dst = cpu_to_le32(*src);
 245         }
 246 }
 247
 248
 249 static void vmxnet3_RxCompToCPU(const struct Vmxnet3_RxCompDesc *srcDesc,
 250                                 struct Vmxnet3_RxCompDesc *dstDesc)
 251 {
 252         int i = 0;
 253         u32 *src = (u32 *)srcDesc;
 254         u32 *dst = (u32 *)dstDesc;
 255         for (i = 0; i < sizeof(struct Vmxnet3_RxCompDesc) / sizeof(u32); i++) {
 256                 *dst = le32_to_cpu(*src);
 257                 src++;
 258                 dst++;
 259         }
 260 }
 261
 262
 263 /* Used to read bitfield values from double words. */
 264 static u32 get_bitfield32(const __le32 *bitfield, u32 pos, u32 size)
 265 {
 266         u32 temp = le32_to_cpu(*bitfield);
 267         u32 mask = ((1 << size) - 1) << pos;
 268         temp &= mask;
 269         temp >>= pos;
 270         return temp;
 271 }
 272
 273
 274
 275 #endif  /* __BIG_ENDIAN_BITFIELD */
 276
 277 #ifdef __BIG_ENDIAN_BITFIELD
 278
 279 #   define VMXNET3_TXDESC_GET_GEN(txdesc) get_bitfield32(((const __le32 *) \
 280                         txdesc) + VMXNET3_TXD_GEN_DWORD_SHIFT, \
 281                         VMXNET3_TXD_GEN_SHIFT, VMXNET3_TXD_GEN_SIZE)
 282 #   define VMXNET3_TXDESC_GET_EOP(txdesc) get_bitfield32(((const __le32 *) \
 283                         txdesc) + VMXNET3_TXD_EOP_DWORD_SHIFT, \
 284                         VMXNET3_TXD_EOP_SHIFT, VMXNET3_TXD_EOP_SIZE)
 285 #   define VMXNET3_TCD_GET_GEN(tcd) get_bitfield32(((const __le32 *)tcd) + \
 286                         VMXNET3_TCD_GEN_DWORD_SHIFT, VMXNET3_TCD_GEN_SHIFT, \
 287                         VMXNET3_TCD_GEN_SIZE)
 288 #   define VMXNET3_TCD_GET_TXIDX(tcd) get_bitfield32((const __le32 *)tcd, \
 289                         VMXNET3_TCD_TXIDX_SHIFT, VMXNET3_TCD_TXIDX_SIZE)
 290 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) do { \
 291                         (dstrcd) = (tmp); \
 292                         vmxnet3_RxCompToCPU((rcd), (tmp)); \
 293                 } while (0)
 294 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) do { \
 295                         (dstrxd) = (tmp); \
 296                         vmxnet3_RxDescToCPU((rxd), (tmp)); \
 297                 } while (0)
 298
 299 #else
 300
 301 #   define VMXNET3_TXDESC_GET_GEN(txdesc) ((txdesc)->gen)
 302 #   define VMXNET3_TXDESC_GET_EOP(txdesc) ((txdesc)->eop)
 303 #   define VMXNET3_TCD_GET_GEN(tcd) ((tcd)->gen)
 304 #   define VMXNET3_TCD_GET_TXIDX(tcd) ((tcd)->txdIdx)
 305 #   define vmxnet3_getRxComp(dstrcd, rcd, tmp) (dstrcd) = (rcd)
 306 #   define vmxnet3_getRxDesc(dstrxd, rxd, tmp) (dstrxd) = (rxd)
 307
 308 #endif /* __BIG_ENDIAN_BITFIELD  */
 309
 310
 311 static void
 312 vmxnet3_unmap_tx_buf(struct vmxnet3_tx_buf_info *tbi,
 313                      struct pci_dev *pdev)
 314 {
 315         if (tbi->map_type == VMXNET3_MAP_SINGLE)
 316                 dma_unmap_single(&pdev->dev, tbi->dma_addr, tbi->len,
 317                                  PCI_DMA_TODEVICE);
 318         else if (tbi->map_type == VMXNET3_MAP_PAGE)
 319                 dma_unmap_page(&pdev->dev, tbi->dma_addr, tbi->len,
 320                                PCI_DMA_TODEVICE);
 321         else
 322                 BUG_ON(tbi->map_type != VMXNET3_MAP_NONE);
 323
 324         tbi->map_type = VMXNET3_MAP_NONE; /* to help debugging */
 325 }
 326
 327
 328 static int
 329 vmxnet3_unmap_pkt(u32 eop_idx, struct vmxnet3_tx_queue *tq,
 330                   struct pci_dev *pdev, struct vmxnet3_adapter *adapter)
 331 {
 332         struct sk_buff *skb;
 333         int entries = 0;
 334
 335         /* no out of order completion */
 336         BUG_ON(tq->buf_info[eop_idx].sop_idx != tq->tx_ring.next2comp);
 337         BUG_ON(VMXNET3_TXDESC_GET_EOP(&(tq->tx_ring.base[eop_idx].txd)) != 1);
 338
 339         skb = tq->buf_info[eop_idx].skb;
 340         BUG_ON(skb == NULL);
 341         tq->buf_info[eop_idx].skb = NULL;
 342
 343         VMXNET3_INC_RING_IDX_ONLY(eop_idx, tq->tx_ring.size);
 344
 345         while (tq->tx_ring.next2comp != eop_idx) {
 346                 vmxnet3_unmap_tx_buf(tq->buf_info + tq->tx_ring.next2comp,
 347                                      pdev);
 348
 349                 /* update next2comp w/o tx_lock. Since we are marking more,
 350                  * instead of less, tx ring entries avail, the worst case is
 351                  * that the tx routine incorrectly re-queues a pkt due to
 352                  * insufficient tx ring entries.
 353                  */
 354                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 355                 entries++;
 356         }
 357
 358         dev_kfree_skb_any(skb);
 359         return entries;
 360 }
 361
 362
 363 static int
 364 vmxnet3_tq_tx_complete(struct vmxnet3_tx_queue *tq,
 365                         struct vmxnet3_adapter *adapter)
 366 {
 367         int completed = 0;
 368         union Vmxnet3_GenericDesc *gdesc;
 369
 370         gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 371         while (VMXNET3_TCD_GET_GEN(&gdesc->tcd) == tq->comp_ring.gen) {
 372                 completed += vmxnet3_unmap_pkt(VMXNET3_TCD_GET_TXIDX(
 373                                                &gdesc->tcd), tq, adapter->pdev,
 374                                                adapter);
 375
 376                 vmxnet3_comp_ring_adv_next2proc(&tq->comp_ring);
 377                 gdesc = tq->comp_ring.base + tq->comp_ring.next2proc;
 378         }
 379
 380         if (completed) {
 381                 spin_lock(&tq->tx_lock);
 382                 if (unlikely(vmxnet3_tq_stopped(tq, adapter) &&
 383                              vmxnet3_cmd_ring_desc_avail(&tq->tx_ring) >
 384                              VMXNET3_WAKE_QUEUE_THRESHOLD(tq) &&
 385                              netif_carrier_ok(adapter->netdev))) {
 386                         vmxnet3_tq_wake(tq, adapter);
 387                 }
 388                 spin_unlock(&tq->tx_lock);
 389         }
 390         return completed;
 391 }
 392
 393
 394 static void
 395 vmxnet3_tq_cleanup(struct vmxnet3_tx_queue *tq,
 396                    struct vmxnet3_adapter *adapter)
 397 {
 398         int i;
 399
 400         while (tq->tx_ring.next2comp != tq->tx_ring.next2fill) {
 401                 struct vmxnet3_tx_buf_info *tbi;
 402
 403                 tbi = tq->buf_info + tq->tx_ring.next2comp;
 404
 405                 vmxnet3_unmap_tx_buf(tbi, adapter->pdev);
 406                 if (tbi->skb) {
 407                         dev_kfree_skb_any(tbi->skb);
 408                         tbi->skb = NULL;
 409                 }
 410                 vmxnet3_cmd_ring_adv_next2comp(&tq->tx_ring);
 411         }
 412
 413         /* sanity check, verify all buffers are indeed unmapped and freed */
 414         for (i = 0; i < tq->tx_ring.size; i++) {
 415                 BUG_ON(tq->buf_info[i].skb != NULL ||
 416                        tq->buf_info[i].map_type != VMXNET3_MAP_NONE);
 417         }
 418
 419         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 420         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 421
 422         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 423         tq->comp_ring.next2proc = 0;
 424 }
 425
 426
 427 static void
 428 vmxnet3_tq_destroy(struct vmxnet3_tx_queue *tq,
 429                    struct vmxnet3_adapter *adapter)
 430 {
 431         if (tq->tx_ring.base) {
 432                 dma_free_coherent(&adapter->pdev->dev, tq->tx_ring.size *
 433                                   sizeof(struct Vmxnet3_TxDesc),
 434                                   tq->tx_ring.base, tq->tx_ring.basePA);
 435                 tq->tx_ring.base = NULL;
 436         }
 437         if (tq->data_ring.base) {
 438                 dma_free_coherent(&adapter->pdev->dev, tq->data_ring.size *
 439                                   sizeof(struct Vmxnet3_TxDataDesc),
 440                                   tq->data_ring.base, tq->data_ring.basePA);
 441                 tq->data_ring.base = NULL;
 442         }
 443         if (tq->comp_ring.base) {
 444                 dma_free_coherent(&adapter->pdev->dev, tq->comp_ring.size *
 445                                   sizeof(struct Vmxnet3_TxCompDesc),
 446                                   tq->comp_ring.base, tq->comp_ring.basePA);
 447                 tq->comp_ring.base = NULL;
 448         }
 449         if (tq->buf_info) {
 450                 dma_free_coherent(&adapter->pdev->dev,
 451                                   tq->tx_ring.size * sizeof(tq->buf_info[0]),
 452                                   tq->buf_info, tq->buf_info_pa);
 453                 tq->buf_info = NULL;
 454         }
 455 }
 456
 457
 458 /* Destroy all tx queues */
 459 void
 460 vmxnet3_tq_destroy_all(struct vmxnet3_adapter *adapter)
 461 {
 462         int i;
 463
 464         for (i = 0; i < adapter->num_tx_queues; i++)
 465                 vmxnet3_tq_destroy(&adapter->tx_queue[i], adapter);
 466 }
 467
 468
 469 static void
 470 vmxnet3_tq_init(struct vmxnet3_tx_queue *tq,
 471                 struct vmxnet3_adapter *adapter)
 472 {
 473         int i;
 474
 475         /* reset the tx ring contents to 0 and reset the tx ring states */
 476         memset(tq->tx_ring.base, 0, tq->tx_ring.size *
 477                sizeof(struct Vmxnet3_TxDesc));
 478         tq->tx_ring.next2fill = tq->tx_ring.next2comp = 0;
 479         tq->tx_ring.gen = VMXNET3_INIT_GEN;
 480
 481         memset(tq->data_ring.base, 0, tq->data_ring.size *
 482                sizeof(struct Vmxnet3_TxDataDesc));
 483
 484         /* reset the tx comp ring contents to 0 and reset comp ring states */
 485         memset(tq->comp_ring.base, 0, tq->comp_ring.size *
 486                sizeof(struct Vmxnet3_TxCompDesc));
 487         tq->comp_ring.next2proc = 0;
 488         tq->comp_ring.gen = VMXNET3_INIT_GEN;
 489
 490         /* reset the bookkeeping data */
 491         memset(tq->buf_info, 0, sizeof(tq->buf_info[0]) * tq->tx_ring.size);
 492         for (i = 0; i < tq->tx_ring.size; i++)
 493                 tq->buf_info[i].map_type = VMXNET3_MAP_NONE;
 494
 495         /* stats are not reset */
 496 }
 497
 498
 499 static int
 500 vmxnet3_tq_create(struct vmxnet3_tx_queue *tq,
 501                   struct vmxnet3_adapter *adapter)
 502 {
 503         size_t sz;
 504
 505         BUG_ON(tq->tx_ring.base || tq->data_ring.base ||
 506                tq->comp_ring.base || tq->buf_info);
 507
 508         tq->tx_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
 509                         tq->tx_ring.size * sizeof(struct Vmxnet3_TxDesc),
 510                         &tq->tx_ring.basePA, GFP_KERNEL);
 511         if (!tq->tx_ring.base) {
 512                 netdev_err(adapter->netdev, "failed to allocate tx ring\n");
 513                 goto err;
 514         }
 515
 516         tq->data_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
 517                         tq->data_ring.size * sizeof(struct Vmxnet3_TxDataDesc),
 518                         &tq->data_ring.basePA, GFP_KERNEL);
 519         if (!tq->data_ring.base) {
 520                 netdev_err(adapter->netdev, "failed to allocate data ring\n");
 521                 goto err;
 522         }
 523
 524         tq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev,
 525                         tq->comp_ring.size * sizeof(struct Vmxnet3_TxCompDesc),
 526                         &tq->comp_ring.basePA, GFP_KERNEL);
 527         if (!tq->comp_ring.base) {
 528                 netdev_err(adapter->netdev, "failed to allocate tx comp ring\n");
 529                 goto err;
 530         }
 531
 532         sz = tq->tx_ring.size * sizeof(tq->buf_info[0]);
 533         tq->buf_info = dma_zalloc_coherent(&adapter->pdev->dev, sz,
 534                                            &tq->buf_info_pa, GFP_KERNEL);
 535         if (!tq->buf_info)
 536                 goto err;
 537
 538         return 0;
 539
 540 err:
 541         vmxnet3_tq_destroy(tq, adapter);
 542         return -ENOMEM;
 543 }
 544
 545 static void
 546 vmxnet3_tq_cleanup_all(struct vmxnet3_adapter *adapter)
 547 {
 548         int i;
 549
 550         for (i = 0; i < adapter->num_tx_queues; i++)
 551                 vmxnet3_tq_cleanup(&adapter->tx_queue[i], adapter);
 552 }
 553
 554 /*
 555  *    starting from ring->next2fill, allocate rx buffers for the given ring
 556  *    of the rx queue and update the rx desc. stop after @num_to_alloc buffers
 557  *    are allocated or allocation fails
 558  */
 559
 560 static int
 561 vmxnet3_rq_alloc_rx_buf(struct vmxnet3_rx_queue *rq, u32 ring_idx,
 562                         int num_to_alloc, struct vmxnet3_adapter *adapter)
 563 {
 564         int num_allocated = 0;
 565         struct vmxnet3_rx_buf_info *rbi_base = rq->buf_info[ring_idx];
 566         struct vmxnet3_cmd_ring *ring = &rq->rx_ring[ring_idx];
 567         u32 val;
 568
 569         while (num_allocated <= num_to_alloc) {
 570                 struct vmxnet3_rx_buf_info *rbi;
 571                 union Vmxnet3_GenericDesc *gd;
 572
 573                 rbi = rbi_base + ring->next2fill;
 574                 gd = ring->base + ring->next2fill;
 575
 576                 if (rbi->buf_type == VMXNET3_RX_BUF_SKB) {
 577                         if (rbi->skb == NULL) {
 578                                 rbi->skb = __netdev_alloc_skb_ip_align(adapter->netdev,
 579                                                                        rbi->len,
 580                                                                        GFP_KERNEL);
 581                                 if (unlikely(rbi->skb == NULL)) {
 582                                         rq->stats.rx_buf_alloc_failure++;
 583                                         break;
 584                                 }
 585
 586                                 rbi->dma_addr = dma_map_single(
 587                                                 &adapter->pdev->dev,
 588                                                 rbi->skb->data, rbi->len,
 589                                                 PCI_DMA_FROMDEVICE);
 590                         } else {
 591                                 /* rx buffer skipped by the device */
 592                         }
 593                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 594                 } else {
 595                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 596                                rbi->len  != PAGE_SIZE);
 597
 598                         if (rbi->page == NULL) {
 599                                 rbi->page = alloc_page(GFP_ATOMIC);
 600                                 if (unlikely(rbi->page == NULL)) {
 601                                         rq->stats.rx_buf_alloc_failure++;
 602                                         break;
 603                                 }
 604                                 rbi->dma_addr = dma_map_page(
 605                                                 &adapter->pdev->dev,
 606                                                 rbi->page, 0, PAGE_SIZE,
 607                                                 PCI_DMA_FROMDEVICE);
 608                         } else {
 609                                 /* rx buffers skipped by the device */
 610                         }
 611                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 612                 }
 613
 614                 BUG_ON(rbi->dma_addr == 0);
 615                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 616                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 617                                            | val | rbi->len);
 618
 619                 /* Fill the last buffer but dont mark it ready, or else the
 620                  * device will think that the queue is full */
 621                 if (num_allocated == num_to_alloc)
 622                         break;
 623
 624                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 625                 num_allocated++;
 626                 vmxnet3_cmd_ring_adv_next2fill(ring);
 627         }
 628
 629         netdev_dbg(adapter->netdev,
 630                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
 631                 num_allocated, ring->next2fill, ring->next2comp);
 632
 633         /* so that the device can distinguish a full ring and an empty ring */
 634         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 635
 636         return num_allocated;
 637 }
 638
 639
 640 static void
 641 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 642                     struct vmxnet3_rx_buf_info *rbi)
 643 {
 644         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 645                 skb_shinfo(skb)->nr_frags;
 646
 647         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 648
 649         __skb_frag_set_page(frag, rbi->page);
 650         frag->page_offset = 0;
 651         skb_frag_size_set(frag, rcd->len);
 652         skb->data_len += rcd->len;
 653         skb->truesize += PAGE_SIZE;
 654         skb_shinfo(skb)->nr_frags++;
 655 }
 656
 657
 658 static void
 659 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 660                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 661                 struct vmxnet3_adapter *adapter)
 662 {
 663         u32 dw2, len;
 664         unsigned long buf_offset;
 665         int i;
 666         union Vmxnet3_GenericDesc *gdesc;
 667         struct vmxnet3_tx_buf_info *tbi = NULL;
 668
 669         BUG_ON(ctx->copy_size > skb_headlen(skb));
 670
 671         /* use the previous gen bit for the SOP desc */
 672         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 673
 674         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 675         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 676
 677         /* no need to map the buffer if headers are copied */
 678         if (ctx->copy_size) {
 679                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 680                                         tq->tx_ring.next2fill *
 681                                         sizeof(struct Vmxnet3_TxDataDesc));
 682                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 683                 ctx->sop_txd->dword[3] = 0;
 684
 685                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 686                 tbi->map_type = VMXNET3_MAP_NONE;
 687
 688                 netdev_dbg(adapter->netdev,
 689                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 690                         tq->tx_ring.next2fill,
 691                         le64_to_cpu(ctx->sop_txd->txd.addr),
 692                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 693                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 694
 695                 /* use the right gen for non-SOP desc */
 696                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 697         }
 698
 699         /* linear part can use multiple tx desc if it's big */
 700         len = skb_headlen(skb) - ctx->copy_size;
 701         buf_offset = ctx->copy_size;
 702         while (len) {
 703                 u32 buf_size;
 704
 705                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 706                         buf_size = len;
 707                         dw2 |= len;
 708                 } else {
 709                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 710                         /* spec says that for TxDesc.len, 0 == 2^14 */
 711                 }
 712
 713                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 714                 tbi->map_type = VMXNET3_MAP_SINGLE;
 715                 tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
 716                                 skb->data + buf_offset, buf_size,
 717                                 PCI_DMA_TODEVICE);
 718
 719                 tbi->len = buf_size;
 720
 721                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 722                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 723
 724                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 725                 gdesc->dword[2] = cpu_to_le32(dw2);
 726                 gdesc->dword[3] = 0;
 727
 728                 netdev_dbg(adapter->netdev,
 729                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 730                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 731                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 732                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 733                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 734
 735                 len -= buf_size;
 736                 buf_offset += buf_size;
 737         }
 738
 739         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 740                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 741                 u32 buf_size;
 742
 743                 buf_offset = 0;
 744                 len = skb_frag_size(frag);
 745                 while (len) {
 746                         tbi = tq->buf_info + tq->tx_ring.next2fill;
 747                         if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 748                                 buf_size = len;
 749                                 dw2 |= len;
 750                         } else {
 751                                 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 752                                 /* spec says that for TxDesc.len, 0 == 2^14 */
 753                         }
 754                         tbi->map_type = VMXNET3_MAP_PAGE;
 755                         tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
 756                                                          buf_offset, buf_size,
 757                                                          DMA_TO_DEVICE);
 758
 759                         tbi->len = buf_size;
 760
 761                         gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 762                         BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 763
 764                         gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 765                         gdesc->dword[2] = cpu_to_le32(dw2);
 766                         gdesc->dword[3] = 0;
 767
 768                         netdev_dbg(adapter->netdev,
 769                                 "txd[%u]: 0x%llx %u %u\n",
 770                                 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 771                                 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 772                         vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 773                         dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 774
 775                         len -= buf_size;
 776                         buf_offset += buf_size;
 777                 }
 778         }
 779
 780         ctx->eop_txd = gdesc;
 781
 782         /* set the last buf_info for the pkt */
 783         tbi->skb = skb;
 784         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 785 }
 786
 787
 788 /* Init all tx queues */
 789 static void
 790 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 791 {
 792         int i;
 793
 794         for (i = 0; i < adapter->num_tx_queues; i++)
 795                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 796 }
 797
 798
 799 /*
 800  *    parse and copy relevant protocol headers:
 801  *      For a tso pkt, relevant headers are L2/3/4 including options
 802  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 803  *      if it's a TCP/UDP pkt
 804  *
 805  * Returns:
 806  *    -1:  error happens during parsing
 807  *     0:  protocol headers parsed, but too big to be copied
 808  *     1:  protocol headers parsed and copied
 809  *
 810  * Other effects:
 811  *    1. related *ctx fields are updated.
 812  *    2. ctx->copy_size is # of bytes copied
 813  *    3. the portion copied is guaranteed to be in the linear part
 814  *
 815  */
 816 static int
 817 vmxnet3_parse_and_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 818                            struct vmxnet3_tx_ctx *ctx,
 819                            struct vmxnet3_adapter *adapter)
 820 {
 821         struct Vmxnet3_TxDataDesc *tdd;
 822
 823         if (ctx->mss) { /* TSO */
 824                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 825                 ctx->l4_hdr_size = tcp_hdrlen(skb);
 826                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 827         } else {
 828                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 829                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 830
 831                         if (ctx->ipv4) {
 832                                 const struct iphdr *iph = ip_hdr(skb);
 833
 834                                 if (iph->protocol == IPPROTO_TCP)
 835                                         ctx->l4_hdr_size = tcp_hdrlen(skb);
 836                                 else if (iph->protocol == IPPROTO_UDP)
 837                                         ctx->l4_hdr_size = sizeof(struct udphdr);
 838                                 else
 839                                         ctx->l4_hdr_size = 0;
 840                         } else {
 841                                 /* for simplicity, don't copy L4 headers */
 842                                 ctx->l4_hdr_size = 0;
 843                         }
 844                         ctx->copy_size = min(ctx->eth_ip_hdr_size +
 845                                          ctx->l4_hdr_size, skb->len);
 846                 } else {
 847                         ctx->eth_ip_hdr_size = 0;
 848                         ctx->l4_hdr_size = 0;
 849                         /* copy as much as allowed */
 850                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 851                                              , skb_headlen(skb));
 852                 }
 853
 854                 /* make sure headers are accessible directly */
 855                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 856                         goto err;
 857         }
 858
 859         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 860                 tq->stats.oversized_hdr++;
 861                 ctx->copy_size = 0;
 862                 return 0;
 863         }
 864
 865         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 866
 867         memcpy(tdd->data, skb->data, ctx->copy_size);
 868         netdev_dbg(adapter->netdev,
 869                 "copy %u bytes to dataRing[%u]\n",
 870                 ctx->copy_size, tq->tx_ring.next2fill);
 871         return 1;
 872
 873 err:
 874         return -1;
 875 }
 876
 877
 878 static void
 879 vmxnet3_prepare_tso(struct sk_buff *skb,
 880                     struct vmxnet3_tx_ctx *ctx)
 881 {
 882         struct tcphdr *tcph = tcp_hdr(skb);
 883
 884         if (ctx->ipv4) {
 885                 struct iphdr *iph = ip_hdr(skb);
 886
 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 = ipv6_hdr(skb);
 892
 893                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 894                                                IPPROTO_TCP, 0);
 895         }
 896 }
 897
 898 static int txd_estimate(const struct sk_buff *skb)
 899 {
 900         int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 901         int i;
 902
 903         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 904                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 905
 906                 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
 907         }
 908         return count;
 909 }
 910
 911 /*
 912  * Transmits a pkt thru a given tq
 913  * Returns:
 914  *    NETDEV_TX_OK:      descriptors are setup successfully
 915  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 916  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 917  *
 918  * Side-effects:
 919  *    1. tx ring may be changed
 920  *    2. tq stats may be updated accordingly
 921  *    3. shared->txNumDeferred may be updated
 922  */
 923
 924 static int
 925 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 926                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 927 {
 928         int ret;
 929         u32 count;
 930         unsigned long flags;
 931         struct vmxnet3_tx_ctx ctx;
 932         union Vmxnet3_GenericDesc *gdesc;
 933 #ifdef __BIG_ENDIAN_BITFIELD
 934         /* Use temporary descriptor to avoid touching bits multiple times */
 935         union Vmxnet3_GenericDesc tempTxDesc;
 936 #endif
 937
 938         count = txd_estimate(skb);
 939
 940         ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
 941
 942         ctx.mss = skb_shinfo(skb)->gso_size;
 943         if (ctx.mss) {
 944                 if (skb_header_cloned(skb)) {
 945                         if (unlikely(pskb_expand_head(skb, 0, 0,
 946                                                       GFP_ATOMIC) != 0)) {
 947                                 tq->stats.drop_tso++;
 948                                 goto drop_pkt;
 949                         }
 950                         tq->stats.copy_skb_header++;
 951                 }
 952                 vmxnet3_prepare_tso(skb, &ctx);
 953         } else {
 954                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
 955
 956                         /* non-tso pkts must not use more than
 957                          * VMXNET3_MAX_TXD_PER_PKT entries
 958                          */
 959                         if (skb_linearize(skb) != 0) {
 960                                 tq->stats.drop_too_many_frags++;
 961                                 goto drop_pkt;
 962                         }
 963                         tq->stats.linearized++;
 964
 965                         /* recalculate the # of descriptors to use */
 966                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 967                 }
 968         }
 969
 970         spin_lock_irqsave(&tq->tx_lock, flags);
 971
 972         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
 973                 tq->stats.tx_ring_full++;
 974                 netdev_dbg(adapter->netdev,
 975                         "tx queue stopped on %s, next2comp %u"
 976                         " next2fill %u\n", adapter->netdev->name,
 977                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
 978
 979                 vmxnet3_tq_stop(tq, adapter);
 980                 spin_unlock_irqrestore(&tq->tx_lock, flags);
 981                 return NETDEV_TX_BUSY;
 982         }
 983
 984
 985         ret = vmxnet3_parse_and_copy_hdr(skb, tq, &ctx, adapter);
 986         if (ret >= 0) {
 987                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
 988                 /* hdrs parsed, check against other limits */
 989                 if (ctx.mss) {
 990                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
 991                                      VMXNET3_MAX_TX_BUF_SIZE)) {
 992                                 goto hdr_too_big;
 993                         }
 994                 } else {
 995                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
 996                                 if (unlikely(ctx.eth_ip_hdr_size +
 997                                              skb->csum_offset >
 998                                              VMXNET3_MAX_CSUM_OFFSET)) {
 999                                         goto hdr_too_big;
1000                                 }
1001                         }
1002                 }
1003         } else {
1004                 tq->stats.drop_hdr_inspect_err++;
1005                 goto unlock_drop_pkt;
1006         }
1007
1008         /* fill tx descs related to addr & len */
1009         vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter);
1010
1011         /* setup the EOP desc */
1012         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1013
1014         /* setup the SOP desc */
1015 #ifdef __BIG_ENDIAN_BITFIELD
1016         gdesc = &tempTxDesc;
1017         gdesc->dword[2] = ctx.sop_txd->dword[2];
1018         gdesc->dword[3] = ctx.sop_txd->dword[3];
1019 #else
1020         gdesc = ctx.sop_txd;
1021 #endif
1022         if (ctx.mss) {
1023                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1024                 gdesc->txd.om = VMXNET3_OM_TSO;
1025                 gdesc->txd.msscof = ctx.mss;
1026                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1027                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1028         } else {
1029                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1030                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1031                         gdesc->txd.om = VMXNET3_OM_CSUM;
1032                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1033                                             skb->csum_offset;
1034                 } else {
1035                         gdesc->txd.om = 0;
1036                         gdesc->txd.msscof = 0;
1037                 }
1038                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1039         }
1040
1041         if (vlan_tx_tag_present(skb)) {
1042                 gdesc->txd.ti = 1;
1043                 gdesc->txd.tci = vlan_tx_tag_get(skb);
1044         }
1045
1046         /* finally flips the GEN bit of the SOP desc. */
1047         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1048                                                   VMXNET3_TXD_GEN);
1049 #ifdef __BIG_ENDIAN_BITFIELD
1050         /* Finished updating in bitfields of Tx Desc, so write them in original
1051          * place.
1052          */
1053         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1054                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1055         gdesc = ctx.sop_txd;
1056 #endif
1057         netdev_dbg(adapter->netdev,
1058                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1059                 (u32)(ctx.sop_txd -
1060                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1061                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1062
1063         spin_unlock_irqrestore(&tq->tx_lock, flags);
1064
1065         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1066                                         le32_to_cpu(tq->shared->txThreshold)) {
1067                 tq->shared->txNumDeferred = 0;
1068                 VMXNET3_WRITE_BAR0_REG(adapter,
1069                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1070                                        tq->tx_ring.next2fill);
1071         }
1072
1073         return NETDEV_TX_OK;
1074
1075 hdr_too_big:
1076         tq->stats.drop_oversized_hdr++;
1077 unlock_drop_pkt:
1078         spin_unlock_irqrestore(&tq->tx_lock, flags);
1079 drop_pkt:
1080         tq->stats.drop_total++;
1081         dev_kfree_skb_any(skb);
1082         return NETDEV_TX_OK;
1083 }
1084
1085
1086 static netdev_tx_t
1087 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1088 {
1089         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1090
1091         BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1092         return vmxnet3_tq_xmit(skb,
1093                                &adapter->tx_queue[skb->queue_mapping],
1094                                adapter, netdev);
1095 }
1096
1097
1098 static void
1099 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1100                 struct sk_buff *skb,
1101                 union Vmxnet3_GenericDesc *gdesc)
1102 {
1103         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1104                 /* typical case: TCP/UDP over IP and both csums are correct */
1105                 if ((le32_to_cpu(gdesc->dword[3]) & VMXNET3_RCD_CSUM_OK) ==
1106                                                         VMXNET3_RCD_CSUM_OK) {
1107                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1108                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1109                         BUG_ON(!(gdesc->rcd.v4  || gdesc->rcd.v6));
1110                         BUG_ON(gdesc->rcd.frg);
1111                 } else {
1112                         if (gdesc->rcd.csum) {
1113                                 skb->csum = htons(gdesc->rcd.csum);
1114                                 skb->ip_summed = CHECKSUM_PARTIAL;
1115                         } else {
1116                                 skb_checksum_none_assert(skb);
1117                         }
1118                 }
1119         } else {
1120                 skb_checksum_none_assert(skb);
1121         }
1122 }
1123
1124
1125 static void
1126 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1127                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1128 {
1129         rq->stats.drop_err++;
1130         if (!rcd->fcs)
1131                 rq->stats.drop_fcs++;
1132
1133         rq->stats.drop_total++;
1134
1135         /*
1136          * We do not unmap and chain the rx buffer to the skb.
1137          * We basically pretend this buffer is not used and will be recycled
1138          * by vmxnet3_rq_alloc_rx_buf()
1139          */
1140
1141         /*
1142          * ctx->skb may be NULL if this is the first and the only one
1143          * desc for the pkt
1144          */
1145         if (ctx->skb)
1146                 dev_kfree_skb_irq(ctx->skb);
1147
1148         ctx->skb = NULL;
1149 }
1150
1151
1152 static int
1153 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1154                        struct vmxnet3_adapter *adapter, int quota)
1155 {
1156         static const u32 rxprod_reg[2] = {
1157                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1158         };
1159         u32 num_rxd = 0;
1160         bool skip_page_frags = false;
1161         struct Vmxnet3_RxCompDesc *rcd;
1162         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1163 #ifdef __BIG_ENDIAN_BITFIELD
1164         struct Vmxnet3_RxDesc rxCmdDesc;
1165         struct Vmxnet3_RxCompDesc rxComp;
1166 #endif
1167         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1168                           &rxComp);
1169         while (rcd->gen == rq->comp_ring.gen) {
1170                 struct vmxnet3_rx_buf_info *rbi;
1171                 struct sk_buff *skb, *new_skb = NULL;
1172                 struct page *new_page = NULL;
1173                 int num_to_alloc;
1174                 struct Vmxnet3_RxDesc *rxd;
1175                 u32 idx, ring_idx;
1176                 struct vmxnet3_cmd_ring *ring = NULL;
1177                 if (num_rxd >= quota) {
1178                         /* we may stop even before we see the EOP desc of
1179                          * the current pkt
1180                          */
1181                         break;
1182                 }
1183                 num_rxd++;
1184                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1185                 idx = rcd->rxdIdx;
1186                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1187                 ring = rq->rx_ring + ring_idx;
1188                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1189                                   &rxCmdDesc);
1190                 rbi = rq->buf_info[ring_idx] + idx;
1191
1192                 BUG_ON(rxd->addr != rbi->dma_addr ||
1193                        rxd->len != rbi->len);
1194
1195                 if (unlikely(rcd->eop && rcd->err)) {
1196                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1197                         goto rcd_done;
1198                 }
1199
1200                 if (rcd->sop) { /* first buf of the pkt */
1201                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1202                                rcd->rqID != rq->qid);
1203
1204                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1205                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1206
1207                         if (unlikely(rcd->len == 0)) {
1208                                 /* Pretend the rx buffer is skipped. */
1209                                 BUG_ON(!(rcd->sop && rcd->eop));
1210                                 netdev_dbg(adapter->netdev,
1211                                         "rxRing[%u][%u] 0 length\n",
1212                                         ring_idx, idx);
1213                                 goto rcd_done;
1214                         }
1215
1216                         skip_page_frags = false;
1217                         ctx->skb = rbi->skb;
1218                         new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1219                                                             rbi->len);
1220                         if (new_skb == NULL) {
1221                                 /* Skb allocation failed, do not handover this
1222                                  * skb to stack. Reuse it. Drop the existing pkt
1223                                  */
1224                                 rq->stats.rx_buf_alloc_failure++;
1225                                 ctx->skb = NULL;
1226                                 rq->stats.drop_total++;
1227                                 skip_page_frags = true;
1228                                 goto rcd_done;
1229                         }
1230
1231                         dma_unmap_single(&adapter->pdev->dev, rbi->dma_addr,
1232                                          rbi->len,
1233                                          PCI_DMA_FROMDEVICE);
1234
1235 #ifdef VMXNET3_RSS
1236                         if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1237                             (adapter->netdev->features & NETIF_F_RXHASH))
1238                                 skb_set_hash(ctx->skb,
1239                                              le32_to_cpu(rcd->rssHash),
1240                                              PKT_HASH_TYPE_L3);
1241 #endif
1242                         skb_put(ctx->skb, rcd->len);
1243
1244                         /* Immediate refill */
1245                         rbi->skb = new_skb;
1246                         rbi->dma_addr = dma_map_single(&adapter->pdev->dev,
1247                                                        rbi->skb->data, rbi->len,
1248                                                        PCI_DMA_FROMDEVICE);
1249                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1250                         rxd->len = rbi->len;
1251
1252                 } else {
1253                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1254
1255                         /* non SOP buffer must be type 1 in most cases */
1256                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1257                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1258
1259                         /* If an sop buffer was dropped, skip all
1260                          * following non-sop fragments. They will be reused.
1261                          */
1262                         if (skip_page_frags)
1263                                 goto rcd_done;
1264
1265                         new_page = alloc_page(GFP_ATOMIC);
1266                         if (unlikely(new_page == NULL)) {
1267                                 /* Replacement page frag could not be allocated.
1268                                  * Reuse this page. Drop the pkt and free the
1269                                  * skb which contained this page as a frag. Skip
1270                                  * processing all the following non-sop frags.
1271                                  */
1272                                 rq->stats.rx_buf_alloc_failure++;
1273                                 dev_kfree_skb(ctx->skb);
1274                                 ctx->skb = NULL;
1275                                 skip_page_frags = true;
1276                                 goto rcd_done;
1277                         }
1278
1279                         if (rcd->len) {
1280                                 dma_unmap_page(&adapter->pdev->dev,
1281                                                rbi->dma_addr, rbi->len,
1282                                                PCI_DMA_FROMDEVICE);
1283
1284                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1285                         }
1286
1287                         /* Immediate refill */
1288                         rbi->page = new_page;
1289                         rbi->dma_addr = dma_map_page(&adapter->pdev->dev,
1290                                                      rbi->page,
1291                                                      0, PAGE_SIZE,
1292                                                      PCI_DMA_FROMDEVICE);
1293                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1294                         rxd->len = rbi->len;
1295                 }
1296
1297
1298                 skb = ctx->skb;
1299                 if (rcd->eop) {
1300                         skb->len += skb->data_len;
1301
1302                         vmxnet3_rx_csum(adapter, skb,
1303                                         (union Vmxnet3_GenericDesc *)rcd);
1304                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1305
1306                         if (unlikely(rcd->ts))
1307                                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1308
1309                         if (adapter->netdev->features & NETIF_F_LRO)
1310                                 netif_receive_skb(skb);
1311                         else
1312                                 napi_gro_receive(&rq->napi, skb);
1313
1314                         ctx->skb = NULL;
1315                 }
1316
1317 rcd_done:
1318                 /* device may have skipped some rx descs */
1319                 ring->next2comp = idx;
1320                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1321                 ring = rq->rx_ring + ring_idx;
1322                 while (num_to_alloc) {
1323                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1324                                           &rxCmdDesc);
1325                         BUG_ON(!rxd->addr);
1326
1327                         /* Recv desc is ready to be used by the device */
1328                         rxd->gen = ring->gen;
1329                         vmxnet3_cmd_ring_adv_next2fill(ring);
1330                         num_to_alloc--;
1331                 }
1332
1333                 /* if needed, update the register */
1334                 if (unlikely(rq->shared->updateRxProd)) {
1335                         VMXNET3_WRITE_BAR0_REG(adapter,
1336                                                rxprod_reg[ring_idx] + rq->qid * 8,
1337                                                ring->next2fill);
1338                 }
1339
1340                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1341                 vmxnet3_getRxComp(rcd,
1342                                   &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1343         }
1344
1345         return num_rxd;
1346 }
1347
1348
1349 static void
1350 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1351                    struct vmxnet3_adapter *adapter)
1352 {
1353         u32 i, ring_idx;
1354         struct Vmxnet3_RxDesc *rxd;
1355
1356         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1357                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1358 #ifdef __BIG_ENDIAN_BITFIELD
1359                         struct Vmxnet3_RxDesc rxDesc;
1360 #endif
1361                         vmxnet3_getRxDesc(rxd,
1362                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1363
1364                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1365                                         rq->buf_info[ring_idx][i].skb) {
1366                                 dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1367                                                  rxd->len, PCI_DMA_FROMDEVICE);
1368                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1369                                 rq->buf_info[ring_idx][i].skb = NULL;
1370                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1371                                         rq->buf_info[ring_idx][i].page) {
1372                                 dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1373                                                rxd->len, PCI_DMA_FROMDEVICE);
1374                                 put_page(rq->buf_info[ring_idx][i].page);
1375                                 rq->buf_info[ring_idx][i].page = NULL;
1376                         }
1377                 }
1378
1379                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1380                 rq->rx_ring[ring_idx].next2fill =
1381                                         rq->rx_ring[ring_idx].next2comp = 0;
1382         }
1383
1384         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1385         rq->comp_ring.next2proc = 0;
1386 }
1387
1388
1389 static void
1390 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1391 {
1392         int i;
1393
1394         for (i = 0; i < adapter->num_rx_queues; i++)
1395                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1396 }
1397
1398
1399 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1400                                struct vmxnet3_adapter *adapter)
1401 {
1402         int i;
1403         int j;
1404
1405         /* all rx buffers must have already been freed */
1406         for (i = 0; i < 2; i++) {
1407                 if (rq->buf_info[i]) {
1408                         for (j = 0; j < rq->rx_ring[i].size; j++)
1409                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1410                 }
1411         }
1412
1413
1414         for (i = 0; i < 2; i++) {
1415                 if (rq->rx_ring[i].base) {
1416                         dma_free_coherent(&adapter->pdev->dev,
1417                                           rq->rx_ring[i].size
1418                                           * sizeof(struct Vmxnet3_RxDesc),
1419                                           rq->rx_ring[i].base,
1420                                           rq->rx_ring[i].basePA);
1421                         rq->rx_ring[i].base = NULL;
1422                 }
1423                 rq->buf_info[i] = NULL;
1424         }
1425
1426         if (rq->comp_ring.base) {
1427                 dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1428                                   * sizeof(struct Vmxnet3_RxCompDesc),
1429                                   rq->comp_ring.base, rq->comp_ring.basePA);
1430                 rq->comp_ring.base = NULL;
1431         }
1432
1433         if (rq->buf_info[0]) {
1434                 size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1435                         (rq->rx_ring[0].size + rq->rx_ring[1].size);
1436                 dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1437                                   rq->buf_info_pa);
1438         }
1439 }
1440
1441
1442 static int
1443 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1444                 struct vmxnet3_adapter  *adapter)
1445 {
1446         int i;
1447
1448         /* initialize buf_info */
1449         for (i = 0; i < rq->rx_ring[0].size; i++) {
1450
1451                 /* 1st buf for a pkt is skbuff */
1452                 if (i % adapter->rx_buf_per_pkt == 0) {
1453                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1454                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1455                 } else { /* subsequent bufs for a pkt is frag */
1456                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1457                         rq->buf_info[0][i].len = PAGE_SIZE;
1458                 }
1459         }
1460         for (i = 0; i < rq->rx_ring[1].size; i++) {
1461                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1462                 rq->buf_info[1][i].len = PAGE_SIZE;
1463         }
1464
1465         /* reset internal state and allocate buffers for both rings */
1466         for (i = 0; i < 2; i++) {
1467                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1468
1469                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1470                        sizeof(struct Vmxnet3_RxDesc));
1471                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1472         }
1473         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1474                                     adapter) == 0) {
1475                 /* at least has 1 rx buffer for the 1st ring */
1476                 return -ENOMEM;
1477         }
1478         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1479
1480         /* reset the comp ring */
1481         rq->comp_ring.next2proc = 0;
1482         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1483                sizeof(struct Vmxnet3_RxCompDesc));
1484         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1485
1486         /* reset rxctx */
1487         rq->rx_ctx.skb = NULL;
1488
1489         /* stats are not reset */
1490         return 0;
1491 }
1492
1493
1494 static int
1495 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1496 {
1497         int i, err = 0;
1498
1499         for (i = 0; i < adapter->num_rx_queues; i++) {
1500                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1501                 if (unlikely(err)) {
1502                         dev_err(&adapter->netdev->dev, "%s: failed to "
1503                                 "initialize rx queue%i\n",
1504                                 adapter->netdev->name, i);
1505                         break;
1506                 }
1507         }
1508         return err;
1509
1510 }
1511
1512
1513 static int
1514 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1515 {
1516         int i;
1517         size_t sz;
1518         struct vmxnet3_rx_buf_info *bi;
1519
1520         for (i = 0; i < 2; i++) {
1521
1522                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1523                 rq->rx_ring[i].base = dma_alloc_coherent(
1524                                                 &adapter->pdev->dev, sz,
1525                                                 &rq->rx_ring[i].basePA,
1526                                                 GFP_KERNEL);
1527                 if (!rq->rx_ring[i].base) {
1528                         netdev_err(adapter->netdev,
1529                                    "failed to allocate rx ring %d\n", i);
1530                         goto err;
1531                 }
1532         }
1533
1534         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1535         rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1536                                                 &rq->comp_ring.basePA,
1537                                                 GFP_KERNEL);
1538         if (!rq->comp_ring.base) {
1539                 netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1540                 goto err;
1541         }
1542
1543         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1544                                                    rq->rx_ring[1].size);
1545         bi = dma_zalloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1546                                  GFP_KERNEL);
1547         if (!bi)
1548                 goto err;
1549
1550         rq->buf_info[0] = bi;
1551         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1552
1553         return 0;
1554
1555 err:
1556         vmxnet3_rq_destroy(rq, adapter);
1557         return -ENOMEM;
1558 }
1559
1560
1561 static int
1562 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1563 {
1564         int i, err = 0;
1565
1566         for (i = 0; i < adapter->num_rx_queues; i++) {
1567                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1568                 if (unlikely(err)) {
1569                         dev_err(&adapter->netdev->dev,
1570                                 "%s: failed to create rx queue%i\n",
1571                                 adapter->netdev->name, i);
1572                         goto err_out;
1573                 }
1574         }
1575         return err;
1576 err_out:
1577         vmxnet3_rq_destroy_all(adapter);
1578         return err;
1579
1580 }
1581
1582 /* Multiple queue aware polling function for tx and rx */
1583
1584 static int
1585 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1586 {
1587         int rcd_done = 0, i;
1588         if (unlikely(adapter->shared->ecr))
1589                 vmxnet3_process_events(adapter);
1590         for (i = 0; i < adapter->num_tx_queues; i++)
1591                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1592
1593         for (i = 0; i < adapter->num_rx_queues; i++)
1594                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1595                                                    adapter, budget);
1596         return rcd_done;
1597 }
1598
1599
1600 static int
1601 vmxnet3_poll(struct napi_struct *napi, int budget)
1602 {
1603         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1604                                           struct vmxnet3_rx_queue, napi);
1605         int rxd_done;
1606
1607         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1608
1609         if (rxd_done < budget) {
1610                 napi_complete(napi);
1611                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1612         }
1613         return rxd_done;
1614 }
1615
1616 /*
1617  * NAPI polling function for MSI-X mode with multiple Rx queues
1618  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1619  */
1620
1621 static int
1622 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1623 {
1624         struct vmxnet3_rx_queue *rq = container_of(napi,
1625                                                 struct vmxnet3_rx_queue, napi);
1626         struct vmxnet3_adapter *adapter = rq->adapter;
1627         int rxd_done;
1628
1629         /* When sharing interrupt with corresponding tx queue, process
1630          * tx completions in that queue as well
1631          */
1632         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1633                 struct vmxnet3_tx_queue *tq =
1634                                 &adapter->tx_queue[rq - adapter->rx_queue];
1635                 vmxnet3_tq_tx_complete(tq, adapter);
1636         }
1637
1638         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1639
1640         if (rxd_done < budget) {
1641                 napi_complete(napi);
1642                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1643         }
1644         return rxd_done;
1645 }
1646
1647
1648 #ifdef CONFIG_PCI_MSI
1649
1650 /*
1651  * Handle completion interrupts on tx queues
1652  * Returns whether or not the intr is handled
1653  */
1654
1655 static irqreturn_t
1656 vmxnet3_msix_tx(int irq, void *data)
1657 {
1658         struct vmxnet3_tx_queue *tq = data;
1659         struct vmxnet3_adapter *adapter = tq->adapter;
1660
1661         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1662                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1663
1664         /* Handle the case where only one irq is allocate for all tx queues */
1665         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1666                 int i;
1667                 for (i = 0; i < adapter->num_tx_queues; i++) {
1668                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1669                         vmxnet3_tq_tx_complete(txq, adapter);
1670                 }
1671         } else {
1672                 vmxnet3_tq_tx_complete(tq, adapter);
1673         }
1674         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1675
1676         return IRQ_HANDLED;
1677 }
1678
1679
1680 /*
1681  * Handle completion interrupts on rx queues. Returns whether or not the
1682  * intr is handled
1683  */
1684
1685 static irqreturn_t
1686 vmxnet3_msix_rx(int irq, void *data)
1687 {
1688         struct vmxnet3_rx_queue *rq = data;
1689         struct vmxnet3_adapter *adapter = rq->adapter;
1690
1691         /* disable intr if needed */
1692         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1693                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1694         napi_schedule(&rq->napi);
1695
1696         return IRQ_HANDLED;
1697 }
1698
1699 /*
1700  *----------------------------------------------------------------------------
1701  *
1702  * vmxnet3_msix_event --
1703  *
1704  *    vmxnet3 msix event intr handler
1705  *
1706  * Result:
1707  *    whether or not the intr is handled
1708  *
1709  *----------------------------------------------------------------------------
1710  */
1711
1712 static irqreturn_t
1713 vmxnet3_msix_event(int irq, void *data)
1714 {
1715         struct net_device *dev = data;
1716         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1717
1718         /* disable intr if needed */
1719         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1720                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1721
1722         if (adapter->shared->ecr)
1723                 vmxnet3_process_events(adapter);
1724
1725         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1726
1727         return IRQ_HANDLED;
1728 }
1729
1730 #endif /* CONFIG_PCI_MSI  */
1731
1732
1733 /* Interrupt handler for vmxnet3  */
1734 static irqreturn_t
1735 vmxnet3_intr(int irq, void *dev_id)
1736 {
1737         struct net_device *dev = dev_id;
1738         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1739
1740         if (adapter->intr.type == VMXNET3_IT_INTX) {
1741                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1742                 if (unlikely(icr == 0))
1743                         /* not ours */
1744                         return IRQ_NONE;
1745         }
1746
1747
1748         /* disable intr if needed */
1749         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1750                 vmxnet3_disable_all_intrs(adapter);
1751
1752         napi_schedule(&adapter->rx_queue[0].napi);
1753
1754         return IRQ_HANDLED;
1755 }
1756
1757 #ifdef CONFIG_NET_POLL_CONTROLLER
1758
1759 /* netpoll callback. */
1760 static void
1761 vmxnet3_netpoll(struct net_device *netdev)
1762 {
1763         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1764
1765         switch (adapter->intr.type) {
1766 #ifdef CONFIG_PCI_MSI
1767         case VMXNET3_IT_MSIX: {
1768                 int i;
1769                 for (i = 0; i < adapter->num_rx_queues; i++)
1770                         vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
1771                 break;
1772         }
1773 #endif
1774         case VMXNET3_IT_MSI:
1775         default:
1776                 vmxnet3_intr(0, adapter->netdev);
1777                 break;
1778         }
1779
1780 }
1781 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1782
1783 static int
1784 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1785 {
1786         struct vmxnet3_intr *intr = &adapter->intr;
1787         int err = 0, i;
1788         int vector = 0;
1789
1790 #ifdef CONFIG_PCI_MSI
1791         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1792                 for (i = 0; i < adapter->num_tx_queues; i++) {
1793                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1794                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1795                                         adapter->netdev->name, vector);
1796                                 err = request_irq(
1797                                               intr->msix_entries[vector].vector,
1798                                               vmxnet3_msix_tx, 0,
1799                                               adapter->tx_queue[i].name,
1800                                               &adapter->tx_queue[i]);
1801                         } else {
1802                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1803                                         adapter->netdev->name, vector);
1804                         }
1805                         if (err) {
1806                                 dev_err(&adapter->netdev->dev,
1807                                         "Failed to request irq for MSIX, %s, "
1808                                         "error %d\n",
1809                                         adapter->tx_queue[i].name, err);
1810                                 return err;
1811                         }
1812
1813                         /* Handle the case where only 1 MSIx was allocated for
1814                          * all tx queues */
1815                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1816                                 for (; i < adapter->num_tx_queues; i++)
1817                                         adapter->tx_queue[i].comp_ring.intr_idx
1818                                                                 = vector;
1819                                 vector++;
1820                                 break;
1821                         } else {
1822                                 adapter->tx_queue[i].comp_ring.intr_idx
1823                                                                 = vector++;
1824                         }
1825                 }
1826                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1827                         vector = 0;
1828
1829                 for (i = 0; i < adapter->num_rx_queues; i++) {
1830                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1831                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1832                                         adapter->netdev->name, vector);
1833                         else
1834                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
1835                                         adapter->netdev->name, vector);
1836                         err = request_irq(intr->msix_entries[vector].vector,
1837                                           vmxnet3_msix_rx, 0,
1838                                           adapter->rx_queue[i].name,
1839                                           &(adapter->rx_queue[i]));
1840                         if (err) {
1841                                 netdev_err(adapter->netdev,
1842                                            "Failed to request irq for MSIX, "
1843                                            "%s, error %d\n",
1844                                            adapter->rx_queue[i].name, err);
1845                                 return err;
1846                         }
1847
1848                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
1849                 }
1850
1851                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
1852                         adapter->netdev->name, vector);
1853                 err = request_irq(intr->msix_entries[vector].vector,
1854                                   vmxnet3_msix_event, 0,
1855                                   intr->event_msi_vector_name, adapter->netdev);
1856                 intr->event_intr_idx = vector;
1857
1858         } else if (intr->type == VMXNET3_IT_MSI) {
1859                 adapter->num_rx_queues = 1;
1860                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
1861                                   adapter->netdev->name, adapter->netdev);
1862         } else {
1863 #endif
1864                 adapter->num_rx_queues = 1;
1865                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
1866                                   IRQF_SHARED, adapter->netdev->name,
1867                                   adapter->netdev);
1868 #ifdef CONFIG_PCI_MSI
1869         }
1870 #endif
1871         intr->num_intrs = vector + 1;
1872         if (err) {
1873                 netdev_err(adapter->netdev,
1874                            "Failed to request irq (intr type:%d), error %d\n",
1875                            intr->type, err);
1876         } else {
1877                 /* Number of rx queues will not change after this */
1878                 for (i = 0; i < adapter->num_rx_queues; i++) {
1879                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
1880                         rq->qid = i;
1881                         rq->qid2 = i + adapter->num_rx_queues;
1882                 }
1883
1884
1885
1886                 /* init our intr settings */
1887                 for (i = 0; i < intr->num_intrs; i++)
1888                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
1889                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
1890                         adapter->intr.event_intr_idx = 0;
1891                         for (i = 0; i < adapter->num_tx_queues; i++)
1892                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
1893                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
1894                 }
1895
1896                 netdev_info(adapter->netdev,
1897                             "intr type %u, mode %u, %u vectors allocated\n",
1898                             intr->type, intr->mask_mode, intr->num_intrs);
1899         }
1900
1901         return err;
1902 }
1903
1904
1905 static void
1906 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
1907 {
1908         struct vmxnet3_intr *intr = &adapter->intr;
1909         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
1910
1911         switch (intr->type) {
1912 #ifdef CONFIG_PCI_MSI
1913         case VMXNET3_IT_MSIX:
1914         {
1915                 int i, vector = 0;
1916
1917                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1918                         for (i = 0; i < adapter->num_tx_queues; i++) {
1919                                 free_irq(intr->msix_entries[vector++].vector,
1920                                          &(adapter->tx_queue[i]));
1921                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
1922                                         break;
1923                         }
1924                 }
1925
1926                 for (i = 0; i < adapter->num_rx_queues; i++) {
1927                         free_irq(intr->msix_entries[vector++].vector,
1928                                  &(adapter->rx_queue[i]));
1929                 }
1930
1931                 free_irq(intr->msix_entries[vector].vector,
1932                          adapter->netdev);
1933                 BUG_ON(vector >= intr->num_intrs);
1934                 break;
1935         }
1936 #endif
1937         case VMXNET3_IT_MSI:
1938                 free_irq(adapter->pdev->irq, adapter->netdev);
1939                 break;
1940         case VMXNET3_IT_INTX:
1941                 free_irq(adapter->pdev->irq, adapter->netdev);
1942                 break;
1943         default:
1944                 BUG();
1945         }
1946 }
1947
1948
1949 static void
1950 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
1951 {
1952         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1953         u16 vid;
1954
1955         /* allow untagged pkts */
1956         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
1957
1958         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
1959                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1960 }
1961
1962
1963 static int
1964 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1965 {
1966         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1967
1968         if (!(netdev->flags & IFF_PROMISC)) {
1969                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1970                 unsigned long flags;
1971
1972                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
1973                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1974                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1975                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1976                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1977         }
1978
1979         set_bit(vid, adapter->active_vlans);
1980
1981         return 0;
1982 }
1983
1984
1985 static int
1986 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
1987 {
1988         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1989
1990         if (!(netdev->flags & IFF_PROMISC)) {
1991                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
1992                 unsigned long flags;
1993
1994                 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
1995                 spin_lock_irqsave(&adapter->cmd_lock, flags);
1996                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
1997                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1998                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
1999         }
2000
2001         clear_bit(vid, adapter->active_vlans);
2002
2003         return 0;
2004 }
2005
2006
2007 static u8 *
2008 vmxnet3_copy_mc(struct net_device *netdev)
2009 {
2010         u8 *buf = NULL;
2011         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2012
2013         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2014         if (sz <= 0xffff) {
2015                 /* We may be called with BH disabled */
2016                 buf = kmalloc(sz, GFP_ATOMIC);
2017                 if (buf) {
2018                         struct netdev_hw_addr *ha;
2019                         int i = 0;
2020
2021                         netdev_for_each_mc_addr(ha, netdev)
2022                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2023                                        ETH_ALEN);
2024                 }
2025         }
2026         return buf;
2027 }
2028
2029
2030 static void
2031 vmxnet3_set_mc(struct net_device *netdev)
2032 {
2033         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2034         unsigned long flags;
2035         struct Vmxnet3_RxFilterConf *rxConf =
2036                                         &adapter->shared->devRead.rxFilterConf;
2037         u8 *new_table = NULL;
2038         dma_addr_t new_table_pa = 0;
2039         u32 new_mode = VMXNET3_RXM_UCAST;
2040
2041         if (netdev->flags & IFF_PROMISC) {
2042                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2043                 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2044
2045                 new_mode |= VMXNET3_RXM_PROMISC;
2046         } else {
2047                 vmxnet3_restore_vlan(adapter);
2048         }
2049
2050         if (netdev->flags & IFF_BROADCAST)
2051                 new_mode |= VMXNET3_RXM_BCAST;
2052
2053         if (netdev->flags & IFF_ALLMULTI)
2054                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2055         else
2056                 if (!netdev_mc_empty(netdev)) {
2057                         new_table = vmxnet3_copy_mc(netdev);
2058                         if (new_table) {
2059                                 rxConf->mfTableLen = cpu_to_le16(
2060                                         netdev_mc_count(netdev) * ETH_ALEN);
2061                                 new_table_pa = dma_map_single(
2062                                                         &adapter->pdev->dev,
2063                                                         new_table,
2064                                                         rxConf->mfTableLen,
2065                                                         PCI_DMA_TODEVICE);
2066                         }
2067
2068                         if (new_table_pa) {
2069                                 new_mode |= VMXNET3_RXM_MCAST;
2070                                 rxConf->mfTablePA = cpu_to_le64(new_table_pa);
2071                         } else {
2072                                 netdev_info(netdev,
2073                                             "failed to copy mcast list, setting ALL_MULTI\n");
2074                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
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                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2089                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2090         }
2091
2092         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2093                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2094         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2095
2096         if (new_table_pa)
2097                 dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2098                                  rxConf->mfTableLen, PCI_DMA_TODEVICE);
2099         kfree(new_table);
2100 }
2101
2102 void
2103 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2104 {
2105         int i;
2106
2107         for (i = 0; i < adapter->num_rx_queues; i++)
2108                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2109 }
2110
2111
2112 /*
2113  *   Set up driver_shared based on settings in adapter.
2114  */
2115
2116 static void
2117 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2118 {
2119         struct Vmxnet3_DriverShared *shared = adapter->shared;
2120         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2121         struct Vmxnet3_TxQueueConf *tqc;
2122         struct Vmxnet3_RxQueueConf *rqc;
2123         int i;
2124
2125         memset(shared, 0, sizeof(*shared));
2126
2127         /* driver settings */
2128         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2129         devRead->misc.driverInfo.version = cpu_to_le32(
2130                                                 VMXNET3_DRIVER_VERSION_NUM);
2131         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2132                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2133         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2134         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2135                                 *((u32 *)&devRead->misc.driverInfo.gos));
2136         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2137         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2138
2139         devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2140         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2141
2142         /* set up feature flags */
2143         if (adapter->netdev->features & NETIF_F_RXCSUM)
2144                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2145
2146         if (adapter->netdev->features & NETIF_F_LRO) {
2147                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2148                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2149         }
2150         if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2151                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2152
2153         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2154         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2155         devRead->misc.queueDescLen = cpu_to_le32(
2156                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2157                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2158
2159         /* tx queue settings */
2160         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2161         for (i = 0; i < adapter->num_tx_queues; i++) {
2162                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2163                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2164                 tqc = &adapter->tqd_start[i].conf;
2165                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2166                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2167                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2168                 tqc->ddPA           = cpu_to_le64(tq->buf_info_pa);
2169                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2170                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2171                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2172                 tqc->ddLen          = cpu_to_le32(
2173                                         sizeof(struct vmxnet3_tx_buf_info) *
2174                                         tqc->txRingSize);
2175                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2176         }
2177
2178         /* rx queue settings */
2179         devRead->misc.numRxQueues = adapter->num_rx_queues;
2180         for (i = 0; i < adapter->num_rx_queues; i++) {
2181                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2182                 rqc = &adapter->rqd_start[i].conf;
2183                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2184                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2185                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2186                 rqc->ddPA            = cpu_to_le64(rq->buf_info_pa);
2187                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2188                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2189                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2190                 rqc->ddLen           = cpu_to_le32(
2191                                         sizeof(struct vmxnet3_rx_buf_info) *
2192                                         (rqc->rxRingSize[0] +
2193                                          rqc->rxRingSize[1]));
2194                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2195         }
2196
2197 #ifdef VMXNET3_RSS
2198         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2199
2200         if (adapter->rss) {
2201                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2202
2203                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2204                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2205                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2206                                     UPT1_RSS_HASH_TYPE_IPV4 |
2207                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2208                                     UPT1_RSS_HASH_TYPE_IPV6;
2209                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2210                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2211                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2212                 netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2213
2214                 for (i = 0; i < rssConf->indTableSize; i++)
2215                         rssConf->indTable[i] = ethtool_rxfh_indir_default(
2216                                 i, adapter->num_rx_queues);
2217
2218                 devRead->rssConfDesc.confVer = 1;
2219                 devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2220                 devRead->rssConfDesc.confPA =
2221                         cpu_to_le64(adapter->rss_conf_pa);
2222         }
2223
2224 #endif /* VMXNET3_RSS */
2225
2226         /* intr settings */
2227         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2228                                      VMXNET3_IMM_AUTO;
2229         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2230         for (i = 0; i < adapter->intr.num_intrs; i++)
2231                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2232
2233         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2234         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2235
2236         /* rx filter settings */
2237         devRead->rxFilterConf.rxMode = 0;
2238         vmxnet3_restore_vlan(adapter);
2239         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2240
2241         /* the rest are already zeroed */
2242 }
2243
2244
2245 int
2246 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2247 {
2248         int err, i;
2249         u32 ret;
2250         unsigned long flags;
2251
2252         netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2253                 " ring sizes %u %u %u\n", adapter->netdev->name,
2254                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2255                 adapter->tx_queue[0].tx_ring.size,
2256                 adapter->rx_queue[0].rx_ring[0].size,
2257                 adapter->rx_queue[0].rx_ring[1].size);
2258
2259         vmxnet3_tq_init_all(adapter);
2260         err = vmxnet3_rq_init_all(adapter);
2261         if (err) {
2262                 netdev_err(adapter->netdev,
2263                            "Failed to init rx queue error %d\n", err);
2264                 goto rq_err;
2265         }
2266
2267         err = vmxnet3_request_irqs(adapter);
2268         if (err) {
2269                 netdev_err(adapter->netdev,
2270                            "Failed to setup irq for error %d\n", err);
2271                 goto irq_err;
2272         }
2273
2274         vmxnet3_setup_driver_shared(adapter);
2275
2276         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2277                                adapter->shared_pa));
2278         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2279                                adapter->shared_pa));
2280         spin_lock_irqsave(&adapter->cmd_lock, flags);
2281         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2282                                VMXNET3_CMD_ACTIVATE_DEV);
2283         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2284         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2285
2286         if (ret != 0) {
2287                 netdev_err(adapter->netdev,
2288                            "Failed to activate dev: error %u\n", ret);
2289                 err = -EINVAL;
2290                 goto activate_err;
2291         }
2292
2293         for (i = 0; i < adapter->num_rx_queues; i++) {
2294                 VMXNET3_WRITE_BAR0_REG(adapter,
2295                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2296                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2297                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2298                                 (i * VMXNET3_REG_ALIGN)),
2299                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2300         }
2301
2302         /* Apply the rx filter settins last. */
2303         vmxnet3_set_mc(adapter->netdev);
2304
2305         /*
2306          * Check link state when first activating device. It will start the
2307          * tx queue if the link is up.
2308          */
2309         vmxnet3_check_link(adapter, true);
2310         for (i = 0; i < adapter->num_rx_queues; i++)
2311                 napi_enable(&adapter->rx_queue[i].napi);
2312         vmxnet3_enable_all_intrs(adapter);
2313         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2314         return 0;
2315
2316 activate_err:
2317         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2318         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2319         vmxnet3_free_irqs(adapter);
2320 irq_err:
2321 rq_err:
2322         /* free up buffers we allocated */
2323         vmxnet3_rq_cleanup_all(adapter);
2324         return err;
2325 }
2326
2327
2328 void
2329 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2330 {
2331         unsigned long flags;
2332         spin_lock_irqsave(&adapter->cmd_lock, flags);
2333         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2334         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2335 }
2336
2337
2338 int
2339 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2340 {
2341         int i;
2342         unsigned long flags;
2343         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2344                 return 0;
2345
2346
2347         spin_lock_irqsave(&adapter->cmd_lock, flags);
2348         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2349                                VMXNET3_CMD_QUIESCE_DEV);
2350         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2351         vmxnet3_disable_all_intrs(adapter);
2352
2353         for (i = 0; i < adapter->num_rx_queues; i++)
2354                 napi_disable(&adapter->rx_queue[i].napi);
2355         netif_tx_disable(adapter->netdev);
2356         adapter->link_speed = 0;
2357         netif_carrier_off(adapter->netdev);
2358
2359         vmxnet3_tq_cleanup_all(adapter);
2360         vmxnet3_rq_cleanup_all(adapter);
2361         vmxnet3_free_irqs(adapter);
2362         return 0;
2363 }
2364
2365
2366 static void
2367 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2368 {
2369         u32 tmp;
2370
2371         tmp = *(u32 *)mac;
2372         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2373
2374         tmp = (mac[5] << 8) | mac[4];
2375         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2376 }
2377
2378
2379 static int
2380 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2381 {
2382         struct sockaddr *addr = p;
2383         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2384
2385         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2386         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2387
2388         return 0;
2389 }
2390
2391
2392 /* ==================== initialization and cleanup routines ============ */
2393
2394 static int
2395 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2396 {
2397         int err;
2398         unsigned long mmio_start, mmio_len;
2399         struct pci_dev *pdev = adapter->pdev;
2400
2401         err = pci_enable_device(pdev);
2402         if (err) {
2403                 dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2404                 return err;
2405         }
2406
2407         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2408                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2409                         dev_err(&pdev->dev,
2410                                 "pci_set_consistent_dma_mask failed\n");
2411                         err = -EIO;
2412                         goto err_set_mask;
2413                 }
2414                 *dma64 = true;
2415         } else {
2416                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2417                         dev_err(&pdev->dev,
2418                                 "pci_set_dma_mask failed\n");
2419                         err = -EIO;
2420                         goto err_set_mask;
2421                 }
2422                 *dma64 = false;
2423         }
2424
2425         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2426                                            vmxnet3_driver_name);
2427         if (err) {
2428                 dev_err(&pdev->dev,
2429                         "Failed to request region for adapter: error %d\n", err);
2430                 goto err_set_mask;
2431         }
2432
2433         pci_set_master(pdev);
2434
2435         mmio_start = pci_resource_start(pdev, 0);
2436         mmio_len = pci_resource_len(pdev, 0);
2437         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2438         if (!adapter->hw_addr0) {
2439                 dev_err(&pdev->dev, "Failed to map bar0\n");
2440                 err = -EIO;
2441                 goto err_ioremap;
2442         }
2443
2444         mmio_start = pci_resource_start(pdev, 1);
2445         mmio_len = pci_resource_len(pdev, 1);
2446         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2447         if (!adapter->hw_addr1) {
2448                 dev_err(&pdev->dev, "Failed to map bar1\n");
2449                 err = -EIO;
2450                 goto err_bar1;
2451         }
2452         return 0;
2453
2454 err_bar1:
2455         iounmap(adapter->hw_addr0);
2456 err_ioremap:
2457         pci_release_selected_regions(pdev, (1 << 2) - 1);
2458 err_set_mask:
2459         pci_disable_device(pdev);
2460         return err;
2461 }
2462
2463
2464 static void
2465 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2466 {
2467         BUG_ON(!adapter->pdev);
2468
2469         iounmap(adapter->hw_addr0);
2470         iounmap(adapter->hw_addr1);
2471         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2472         pci_disable_device(adapter->pdev);
2473 }
2474
2475
2476 static void
2477 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2478 {
2479         size_t sz, i, ring0_size, ring1_size, comp_size;
2480         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2481
2482
2483         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2484                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2485                 adapter->skb_buf_size = adapter->netdev->mtu +
2486                                         VMXNET3_MAX_ETH_HDR_SIZE;
2487                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2488                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2489
2490                 adapter->rx_buf_per_pkt = 1;
2491         } else {
2492                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2493                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2494                                             VMXNET3_MAX_ETH_HDR_SIZE;
2495                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2496         }
2497
2498         /*
2499          * for simplicity, force the ring0 size to be a multiple of
2500          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2501          */
2502         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2503         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2504         ring0_size = (ring0_size + sz - 1) / sz * sz;
2505         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2506                            sz * sz);
2507         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2508         comp_size = ring0_size + ring1_size;
2509
2510         for (i = 0; i < adapter->num_rx_queues; i++) {
2511                 rq = &adapter->rx_queue[i];
2512                 rq->rx_ring[0].size = ring0_size;
2513                 rq->rx_ring[1].size = ring1_size;
2514                 rq->comp_ring.size = comp_size;
2515         }
2516 }
2517
2518
2519 int
2520 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2521                       u32 rx_ring_size, u32 rx_ring2_size)
2522 {
2523         int err = 0, i;
2524
2525         for (i = 0; i < adapter->num_tx_queues; i++) {
2526                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2527                 tq->tx_ring.size   = tx_ring_size;
2528                 tq->data_ring.size = tx_ring_size;
2529                 tq->comp_ring.size = tx_ring_size;
2530                 tq->shared = &adapter->tqd_start[i].ctrl;
2531                 tq->stopped = true;
2532                 tq->adapter = adapter;
2533                 tq->qid = i;
2534                 err = vmxnet3_tq_create(tq, adapter);
2535                 /*
2536                  * Too late to change num_tx_queues. We cannot do away with
2537                  * lesser number of queues than what we asked for
2538                  */
2539                 if (err)
2540                         goto queue_err;
2541         }
2542
2543         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2544         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2545         vmxnet3_adjust_rx_ring_size(adapter);
2546         for (i = 0; i < adapter->num_rx_queues; i++) {
2547                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2548                 /* qid and qid2 for rx queues will be assigned later when num
2549                  * of rx queues is finalized after allocating intrs */
2550                 rq->shared = &adapter->rqd_start[i].ctrl;
2551                 rq->adapter = adapter;
2552                 err = vmxnet3_rq_create(rq, adapter);
2553                 if (err) {
2554                         if (i == 0) {
2555                                 netdev_err(adapter->netdev,
2556                                            "Could not allocate any rx queues. "
2557                                            "Aborting.\n");
2558                                 goto queue_err;
2559                         } else {
2560                                 netdev_info(adapter->netdev,
2561                                             "Number of rx queues changed "
2562                                             "to : %d.\n", i);
2563                                 adapter->num_rx_queues = i;
2564                                 err = 0;
2565                                 break;
2566                         }
2567                 }
2568         }
2569         return err;
2570 queue_err:
2571         vmxnet3_tq_destroy_all(adapter);
2572         return err;
2573 }
2574
2575 static int
2576 vmxnet3_open(struct net_device *netdev)
2577 {
2578         struct vmxnet3_adapter *adapter;
2579         int err, i;
2580
2581         adapter = netdev_priv(netdev);
2582
2583         for (i = 0; i < adapter->num_tx_queues; i++)
2584                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2585
2586         err = vmxnet3_create_queues(adapter, adapter->tx_ring_size,
2587                                     adapter->rx_ring_size,
2588                                     VMXNET3_DEF_RX_RING_SIZE);
2589         if (err)
2590                 goto queue_err;
2591
2592         err = vmxnet3_activate_dev(adapter);
2593         if (err)
2594                 goto activate_err;
2595
2596         return 0;
2597
2598 activate_err:
2599         vmxnet3_rq_destroy_all(adapter);
2600         vmxnet3_tq_destroy_all(adapter);
2601 queue_err:
2602         return err;
2603 }
2604
2605
2606 static int
2607 vmxnet3_close(struct net_device *netdev)
2608 {
2609         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2610
2611         /*
2612          * Reset_work may be in the middle of resetting the device, wait for its
2613          * completion.
2614          */
2615         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2616                 msleep(1);
2617
2618         vmxnet3_quiesce_dev(adapter);
2619
2620         vmxnet3_rq_destroy_all(adapter);
2621         vmxnet3_tq_destroy_all(adapter);
2622
2623         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2624
2625
2626         return 0;
2627 }
2628
2629
2630 void
2631 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2632 {
2633         int i;
2634
2635         /*
2636          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2637          * vmxnet3_close() will deadlock.
2638          */
2639         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2640
2641         /* we need to enable NAPI, otherwise dev_close will deadlock */
2642         for (i = 0; i < adapter->num_rx_queues; i++)
2643                 napi_enable(&adapter->rx_queue[i].napi);
2644         dev_close(adapter->netdev);
2645 }
2646
2647
2648 static int
2649 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2650 {
2651         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2652         int err = 0;
2653
2654         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2655                 return -EINVAL;
2656
2657         netdev->mtu = new_mtu;
2658
2659         /*
2660          * Reset_work may be in the middle of resetting the device, wait for its
2661          * completion.
2662          */
2663         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2664                 msleep(1);
2665
2666         if (netif_running(netdev)) {
2667                 vmxnet3_quiesce_dev(adapter);
2668                 vmxnet3_reset_dev(adapter);
2669
2670                 /* we need to re-create the rx queue based on the new mtu */
2671                 vmxnet3_rq_destroy_all(adapter);
2672                 vmxnet3_adjust_rx_ring_size(adapter);
2673                 err = vmxnet3_rq_create_all(adapter);
2674                 if (err) {
2675                         netdev_err(netdev,
2676                                    "failed to re-create rx queues, "
2677                                    " error %d. Closing it.\n", err);
2678                         goto out;
2679                 }
2680
2681                 err = vmxnet3_activate_dev(adapter);
2682                 if (err) {
2683                         netdev_err(netdev,
2684                                    "failed to re-activate, error %d. "
2685                                    "Closing it\n", err);
2686                         goto out;
2687                 }
2688         }
2689
2690 out:
2691         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2692         if (err)
2693                 vmxnet3_force_close(adapter);
2694
2695         return err;
2696 }
2697
2698
2699 static void
2700 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2701 {
2702         struct net_device *netdev = adapter->netdev;
2703
2704         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2705                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2706                 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2707                 NETIF_F_LRO;
2708         if (dma64)
2709                 netdev->hw_features |= NETIF_F_HIGHDMA;
2710         netdev->vlan_features = netdev->hw_features &
2711                                 ~(NETIF_F_HW_VLAN_CTAG_TX |
2712                                   NETIF_F_HW_VLAN_CTAG_RX);
2713         netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
2714 }
2715
2716
2717 static void
2718 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2719 {
2720         u32 tmp;
2721
2722         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2723         *(u32 *)mac = tmp;
2724
2725         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2726         mac[4] = tmp & 0xff;
2727         mac[5] = (tmp >> 8) & 0xff;
2728 }
2729
2730 #ifdef CONFIG_PCI_MSI
2731
2732 /*
2733  * Enable MSIx vectors.
2734  * Returns :
2735  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2736  *       were enabled.
2737  *      number of vectors which were enabled otherwise (this number is greater
2738  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2739  */
2740
2741 static int
2742 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
2743 {
2744         int ret = pci_enable_msix_range(adapter->pdev,
2745                                         adapter->intr.msix_entries, nvec, nvec);
2746
2747         if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
2748                 dev_err(&adapter->netdev->dev,
2749                         "Failed to enable %d MSI-X, trying %d\n",
2750                         nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
2751
2752                 ret = pci_enable_msix_range(adapter->pdev,
2753                                             adapter->intr.msix_entries,
2754                                             VMXNET3_LINUX_MIN_MSIX_VECT,
2755                                             VMXNET3_LINUX_MIN_MSIX_VECT);
2756         }
2757
2758         if (ret < 0) {
2759                 dev_err(&adapter->netdev->dev,
2760                         "Failed to enable MSI-X, error: %d\n", ret);
2761         }
2762
2763         return ret;
2764 }
2765
2766
2767 #endif /* CONFIG_PCI_MSI */
2768
2769 static void
2770 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2771 {
2772         u32 cfg;
2773         unsigned long flags;
2774
2775         /* intr settings */
2776         spin_lock_irqsave(&adapter->cmd_lock, flags);
2777         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2778                                VMXNET3_CMD_GET_CONF_INTR);
2779         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2780         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2781         adapter->intr.type = cfg & 0x3;
2782         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2783
2784         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2785                 adapter->intr.type = VMXNET3_IT_MSIX;
2786         }
2787
2788 #ifdef CONFIG_PCI_MSI
2789         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2790                 int i, nvec;
2791
2792                 nvec  = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
2793                         1 : adapter->num_tx_queues;
2794                 nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
2795                         0 : adapter->num_rx_queues;
2796                 nvec += 1;      /* for link event */
2797                 nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
2798                        nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
2799
2800                 for (i = 0; i < nvec; i++)
2801                         adapter->intr.msix_entries[i].entry = i;
2802
2803                 nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
2804                 if (nvec < 0)
2805                         goto msix_err;
2806
2807                 /* If we cannot allocate one MSIx vector per queue
2808                  * then limit the number of rx queues to 1
2809                  */
2810                 if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
2811                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2812                             || adapter->num_rx_queues != 1) {
2813                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2814                                 netdev_err(adapter->netdev,
2815                                            "Number of rx queues : 1\n");
2816                                 adapter->num_rx_queues = 1;
2817                         }
2818                 }
2819
2820                 adapter->intr.num_intrs = nvec;
2821                 return;
2822
2823 msix_err:
2824                 /* If we cannot allocate MSIx vectors use only one rx queue */
2825                 dev_info(&adapter->pdev->dev,
2826                          "Failed to enable MSI-X, error %d. "
2827                          "Limiting #rx queues to 1, try MSI.\n", nvec);
2828
2829                 adapter->intr.type = VMXNET3_IT_MSI;
2830         }
2831
2832         if (adapter->intr.type == VMXNET3_IT_MSI) {
2833                 if (!pci_enable_msi(adapter->pdev)) {
2834                         adapter->num_rx_queues = 1;
2835                         adapter->intr.num_intrs = 1;
2836                         return;
2837                 }
2838         }
2839 #endif /* CONFIG_PCI_MSI */
2840
2841         adapter->num_rx_queues = 1;
2842         dev_info(&adapter->netdev->dev,
2843                  "Using INTx interrupt, #Rx queues: 1.\n");
2844         adapter->intr.type = VMXNET3_IT_INTX;
2845
2846         /* INT-X related setting */
2847         adapter->intr.num_intrs = 1;
2848 }
2849
2850
2851 static void
2852 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
2853 {
2854         if (adapter->intr.type == VMXNET3_IT_MSIX)
2855                 pci_disable_msix(adapter->pdev);
2856         else if (adapter->intr.type == VMXNET3_IT_MSI)
2857                 pci_disable_msi(adapter->pdev);
2858         else
2859                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
2860 }
2861
2862
2863 static void
2864 vmxnet3_tx_timeout(struct net_device *netdev)
2865 {
2866         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2867         adapter->tx_timeout_count++;
2868
2869         netdev_err(adapter->netdev, "tx hang\n");
2870         schedule_work(&adapter->work);
2871         netif_wake_queue(adapter->netdev);
2872 }
2873
2874
2875 static void
2876 vmxnet3_reset_work(struct work_struct *data)
2877 {
2878         struct vmxnet3_adapter *adapter;
2879
2880         adapter = container_of(data, struct vmxnet3_adapter, work);
2881
2882         /* if another thread is resetting the device, no need to proceed */
2883         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2884                 return;
2885
2886         /* if the device is closed, we must leave it alone */
2887         rtnl_lock();
2888         if (netif_running(adapter->netdev)) {
2889                 netdev_notice(adapter->netdev, "resetting\n");
2890                 vmxnet3_quiesce_dev(adapter);
2891                 vmxnet3_reset_dev(adapter);
2892                 vmxnet3_activate_dev(adapter);
2893         } else {
2894                 netdev_info(adapter->netdev, "already closed\n");
2895         }
2896         rtnl_unlock();
2897
2898         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2899 }
2900
2901
2902 static int
2903 vmxnet3_probe_device(struct pci_dev *pdev,
2904                      const struct pci_device_id *id)
2905 {
2906         static const struct net_device_ops vmxnet3_netdev_ops = {
2907                 .ndo_open = vmxnet3_open,
2908                 .ndo_stop = vmxnet3_close,
2909                 .ndo_start_xmit = vmxnet3_xmit_frame,
2910                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
2911                 .ndo_change_mtu = vmxnet3_change_mtu,
2912                 .ndo_set_features = vmxnet3_set_features,
2913                 .ndo_get_stats64 = vmxnet3_get_stats64,
2914                 .ndo_tx_timeout = vmxnet3_tx_timeout,
2915                 .ndo_set_rx_mode = vmxnet3_set_mc,
2916                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
2917                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
2918 #ifdef CONFIG_NET_POLL_CONTROLLER
2919                 .ndo_poll_controller = vmxnet3_netpoll,
2920 #endif
2921         };
2922         int err;
2923         bool dma64 = false; /* stupid gcc */
2924         u32 ver;
2925         struct net_device *netdev;
2926         struct vmxnet3_adapter *adapter;
2927         u8 mac[ETH_ALEN];
2928         int size;
2929         int num_tx_queues;
2930         int num_rx_queues;
2931
2932         if (!pci_msi_enabled())
2933                 enable_mq = 0;
2934
2935 #ifdef VMXNET3_RSS
2936         if (enable_mq)
2937                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
2938                                     (int)num_online_cpus());
2939         else
2940 #endif
2941                 num_rx_queues = 1;
2942         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
2943
2944         if (enable_mq)
2945                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
2946                                     (int)num_online_cpus());
2947         else
2948                 num_tx_queues = 1;
2949
2950         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
2951         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
2952                                    max(num_tx_queues, num_rx_queues));
2953         dev_info(&pdev->dev,
2954                  "# of Tx queues : %d, # of Rx queues : %d\n",
2955                  num_tx_queues, num_rx_queues);
2956
2957         if (!netdev)
2958                 return -ENOMEM;
2959
2960         pci_set_drvdata(pdev, netdev);
2961         adapter = netdev_priv(netdev);
2962         adapter->netdev = netdev;
2963         adapter->pdev = pdev;
2964
2965         adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
2966         adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
2967
2968         spin_lock_init(&adapter->cmd_lock);
2969         adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
2970                                              sizeof(struct vmxnet3_adapter),
2971                                              PCI_DMA_TODEVICE);
2972         adapter->shared = dma_alloc_coherent(
2973                                 &adapter->pdev->dev,
2974                                 sizeof(struct Vmxnet3_DriverShared),
2975                                 &adapter->shared_pa, GFP_KERNEL);
2976         if (!adapter->shared) {
2977                 dev_err(&pdev->dev, "Failed to allocate memory\n");
2978                 err = -ENOMEM;
2979                 goto err_alloc_shared;
2980         }
2981
2982         adapter->num_rx_queues = num_rx_queues;
2983         adapter->num_tx_queues = num_tx_queues;
2984         adapter->rx_buf_per_pkt = 1;
2985
2986         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
2987         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
2988         adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
2989                                                 &adapter->queue_desc_pa,
2990                                                 GFP_KERNEL);
2991
2992         if (!adapter->tqd_start) {
2993                 dev_err(&pdev->dev, "Failed to allocate memory\n");
2994                 err = -ENOMEM;
2995                 goto err_alloc_queue_desc;
2996         }
2997         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
2998                                                             adapter->num_tx_queues);
2999
3000         adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3001                                               sizeof(struct Vmxnet3_PMConf),
3002                                               &adapter->pm_conf_pa,
3003                                               GFP_KERNEL);
3004         if (adapter->pm_conf == NULL) {
3005                 err = -ENOMEM;
3006                 goto err_alloc_pm;
3007         }
3008
3009 #ifdef VMXNET3_RSS
3010
3011         adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3012                                                sizeof(struct UPT1_RSSConf),
3013                                                &adapter->rss_conf_pa,
3014                                                GFP_KERNEL);
3015         if (adapter->rss_conf == NULL) {
3016                 err = -ENOMEM;
3017                 goto err_alloc_rss;
3018         }
3019 #endif /* VMXNET3_RSS */
3020
3021         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
3022         if (err < 0)
3023                 goto err_alloc_pci;
3024
3025         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3026         if (ver & 1) {
3027                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
3028         } else {
3029                 dev_err(&pdev->dev,
3030                         "Incompatible h/w version (0x%x) for adapter\n", ver);
3031                 err = -EBUSY;
3032                 goto err_ver;
3033         }
3034
3035         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3036         if (ver & 1) {
3037                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3038         } else {
3039                 dev_err(&pdev->dev,
3040                         "Incompatible upt version (0x%x) for adapter\n", ver);
3041                 err = -EBUSY;
3042                 goto err_ver;
3043         }
3044
3045         SET_NETDEV_DEV(netdev, &pdev->dev);
3046         vmxnet3_declare_features(adapter, dma64);
3047
3048         if (adapter->num_tx_queues == adapter->num_rx_queues)
3049                 adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3050         else
3051                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3052
3053         vmxnet3_alloc_intr_resources(adapter);
3054
3055 #ifdef VMXNET3_RSS
3056         if (adapter->num_rx_queues > 1 &&
3057             adapter->intr.type == VMXNET3_IT_MSIX) {
3058                 adapter->rss = true;
3059                 netdev->hw_features |= NETIF_F_RXHASH;
3060                 netdev->features |= NETIF_F_RXHASH;
3061                 dev_dbg(&pdev->dev, "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         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3076
3077         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3078                 int i;
3079                 for (i = 0; i < adapter->num_rx_queues; i++) {
3080                         netif_napi_add(adapter->netdev,
3081                                        &adapter->rx_queue[i].napi,
3082                                        vmxnet3_poll_rx_only, 64);
3083                 }
3084         } else {
3085                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3086                                vmxnet3_poll, 64);
3087         }
3088
3089         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3090         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3091
3092         netif_carrier_off(netdev);
3093         err = register_netdev(netdev);
3094
3095         if (err) {
3096                 dev_err(&pdev->dev, "Failed to register adapter\n");
3097                 goto err_register;
3098         }
3099
3100         vmxnet3_check_link(adapter, false);
3101         return 0;
3102
3103 err_register:
3104         vmxnet3_free_intr_resources(adapter);
3105 err_ver:
3106         vmxnet3_free_pci_resources(adapter);
3107 err_alloc_pci:
3108 #ifdef VMXNET3_RSS
3109         dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3110                           adapter->rss_conf, adapter->rss_conf_pa);
3111 err_alloc_rss:
3112 #endif
3113         dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3114                           adapter->pm_conf, adapter->pm_conf_pa);
3115 err_alloc_pm:
3116         dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3117                           adapter->queue_desc_pa);
3118 err_alloc_queue_desc:
3119         dma_free_coherent(&adapter->pdev->dev,
3120                           sizeof(struct Vmxnet3_DriverShared),
3121                           adapter->shared, adapter->shared_pa);
3122 err_alloc_shared:
3123         dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3124                          sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3125         free_netdev(netdev);
3126         return err;
3127 }
3128
3129
3130 static void
3131 vmxnet3_remove_device(struct pci_dev *pdev)
3132 {
3133         struct net_device *netdev = pci_get_drvdata(pdev);
3134         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3135         int size = 0;
3136         int num_rx_queues;
3137
3138 #ifdef VMXNET3_RSS
3139         if (enable_mq)
3140                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3141                                     (int)num_online_cpus());
3142         else
3143 #endif
3144                 num_rx_queues = 1;
3145         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3146
3147         cancel_work_sync(&adapter->work);
3148
3149         unregister_netdev(netdev);
3150
3151         vmxnet3_free_intr_resources(adapter);
3152         vmxnet3_free_pci_resources(adapter);
3153 #ifdef VMXNET3_RSS
3154         dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3155                           adapter->rss_conf, adapter->rss_conf_pa);
3156 #endif
3157         dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3158                           adapter->pm_conf, adapter->pm_conf_pa);
3159
3160         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3161         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3162         dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3163                           adapter->queue_desc_pa);
3164         dma_free_coherent(&adapter->pdev->dev,
3165                           sizeof(struct Vmxnet3_DriverShared),
3166                           adapter->shared, adapter->shared_pa);
3167         dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3168                          sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3169         free_netdev(netdev);
3170 }
3171
3172
3173 #ifdef CONFIG_PM
3174
3175 static int
3176 vmxnet3_suspend(struct device *device)
3177 {
3178         struct pci_dev *pdev = to_pci_dev(device);
3179         struct net_device *netdev = pci_get_drvdata(pdev);
3180         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3181         struct Vmxnet3_PMConf *pmConf;
3182         struct ethhdr *ehdr;
3183         struct arphdr *ahdr;
3184         u8 *arpreq;
3185         struct in_device *in_dev;
3186         struct in_ifaddr *ifa;
3187         unsigned long flags;
3188         int i = 0;
3189
3190         if (!netif_running(netdev))
3191                 return 0;
3192
3193         for (i = 0; i < adapter->num_rx_queues; i++)
3194                 napi_disable(&adapter->rx_queue[i].napi);
3195
3196         vmxnet3_disable_all_intrs(adapter);
3197         vmxnet3_free_irqs(adapter);
3198         vmxnet3_free_intr_resources(adapter);
3199
3200         netif_device_detach(netdev);
3201         netif_tx_stop_all_queues(netdev);
3202
3203         /* Create wake-up filters. */
3204         pmConf = adapter->pm_conf;
3205         memset(pmConf, 0, sizeof(*pmConf));
3206
3207         if (adapter->wol & WAKE_UCAST) {
3208                 pmConf->filters[i].patternSize = ETH_ALEN;
3209                 pmConf->filters[i].maskSize = 1;
3210                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3211                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3212
3213                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3214                 i++;
3215         }
3216
3217         if (adapter->wol & WAKE_ARP) {
3218                 in_dev = in_dev_get(netdev);
3219                 if (!in_dev)
3220                         goto skip_arp;
3221
3222                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3223                 if (!ifa)
3224                         goto skip_arp;
3225
3226                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3227                         sizeof(struct arphdr) +         /* ARP header */
3228                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3229                         2 * sizeof(u32);        /*2 IPv4 addresses */
3230                 pmConf->filters[i].maskSize =
3231                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3232
3233                 /* ETH_P_ARP in Ethernet header. */
3234                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3235                 ehdr->h_proto = htons(ETH_P_ARP);
3236
3237                 /* ARPOP_REQUEST in ARP header. */
3238                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3239                 ahdr->ar_op = htons(ARPOP_REQUEST);
3240                 arpreq = (u8 *)(ahdr + 1);
3241
3242                 /* The Unicast IPv4 address in 'tip' field. */
3243                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3244                 *(u32 *)arpreq = ifa->ifa_address;
3245
3246                 /* The mask for the relevant bits. */
3247                 pmConf->filters[i].mask[0] = 0x00;
3248                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3249                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3250                 pmConf->filters[i].mask[3] = 0x00;
3251                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3252                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3253                 in_dev_put(in_dev);
3254
3255                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3256                 i++;
3257         }
3258
3259 skip_arp:
3260         if (adapter->wol & WAKE_MAGIC)
3261                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3262
3263         pmConf->numFilters = i;
3264
3265         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3266         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3267                                                                   *pmConf));
3268         adapter->shared->devRead.pmConfDesc.confPA =
3269                 cpu_to_le64(adapter->pm_conf_pa);
3270
3271         spin_lock_irqsave(&adapter->cmd_lock, flags);
3272         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3273                                VMXNET3_CMD_UPDATE_PMCFG);
3274         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3275
3276         pci_save_state(pdev);
3277         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3278                         adapter->wol);
3279         pci_disable_device(pdev);
3280         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3281
3282         return 0;
3283 }
3284
3285
3286 static int
3287 vmxnet3_resume(struct device *device)
3288 {
3289         int err, i = 0;
3290         unsigned long flags;
3291         struct pci_dev *pdev = to_pci_dev(device);
3292         struct net_device *netdev = pci_get_drvdata(pdev);
3293         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3294         struct Vmxnet3_PMConf *pmConf;
3295
3296         if (!netif_running(netdev))
3297                 return 0;
3298
3299         /* Destroy wake-up filters. */
3300         pmConf = adapter->pm_conf;
3301         memset(pmConf, 0, sizeof(*pmConf));
3302
3303         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3304         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3305                                                                   *pmConf));
3306         adapter->shared->devRead.pmConfDesc.confPA =
3307                 cpu_to_le64(adapter->pm_conf_pa);
3308
3309         netif_device_attach(netdev);
3310         pci_set_power_state(pdev, PCI_D0);
3311         pci_restore_state(pdev);
3312         err = pci_enable_device_mem(pdev);
3313         if (err != 0)
3314                 return err;
3315
3316         pci_enable_wake(pdev, PCI_D0, 0);
3317
3318         spin_lock_irqsave(&adapter->cmd_lock, flags);
3319         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3320                                VMXNET3_CMD_UPDATE_PMCFG);
3321         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3322         vmxnet3_alloc_intr_resources(adapter);
3323         vmxnet3_request_irqs(adapter);
3324         for (i = 0; i < adapter->num_rx_queues; i++)
3325                 napi_enable(&adapter->rx_queue[i].napi);
3326         vmxnet3_enable_all_intrs(adapter);
3327
3328         return 0;
3329 }
3330
3331 static const struct dev_pm_ops vmxnet3_pm_ops = {
3332         .suspend = vmxnet3_suspend,
3333         .resume = vmxnet3_resume,
3334 };
3335 #endif
3336
3337 static struct pci_driver vmxnet3_driver = {
3338         .name           = vmxnet3_driver_name,
3339         .id_table       = vmxnet3_pciid_table,
3340         .probe          = vmxnet3_probe_device,
3341         .remove         = vmxnet3_remove_device,
3342 #ifdef CONFIG_PM
3343         .driver.pm      = &vmxnet3_pm_ops,
3344 #endif
3345 };
3346
3347
3348 static int __init
3349 vmxnet3_init_module(void)
3350 {
3351         pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3352                 VMXNET3_DRIVER_VERSION_REPORT);
3353         return pci_register_driver(&vmxnet3_driver);
3354 }
3355
3356 module_init(vmxnet3_init_module);
3357
3358
3359 static void
3360 vmxnet3_exit_module(void)
3361 {
3362         pci_unregister_driver(&vmxnet3_driver);
3363 }
3364
3365 module_exit(vmxnet3_exit_module);
3366
3367 MODULE_AUTHOR("VMware, Inc.");
3368 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3369 MODULE_LICENSE("GPL v2");
3370 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);