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                                 if (dma_mapping_error(&adapter->pdev->dev,
 591                                                       rbi->dma_addr)) {
 592                                         dev_kfree_skb_any(rbi->skb);
 593                                         rq->stats.rx_buf_alloc_failure++;
 594                                         break;
 595                                 }
 596                         } else {
 597                                 /* rx buffer skipped by the device */
 598                         }
 599                         val = VMXNET3_RXD_BTYPE_HEAD << VMXNET3_RXD_BTYPE_SHIFT;
 600                 } else {
 601                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE ||
 602                                rbi->len  != PAGE_SIZE);
 603
 604                         if (rbi->page == NULL) {
 605                                 rbi->page = alloc_page(GFP_ATOMIC);
 606                                 if (unlikely(rbi->page == NULL)) {
 607                                         rq->stats.rx_buf_alloc_failure++;
 608                                         break;
 609                                 }
 610                                 rbi->dma_addr = dma_map_page(
 611                                                 &adapter->pdev->dev,
 612                                                 rbi->page, 0, PAGE_SIZE,
 613                                                 PCI_DMA_FROMDEVICE);
 614                                 if (dma_mapping_error(&adapter->pdev->dev,
 615                                                       rbi->dma_addr)) {
 616                                         put_page(rbi->page);
 617                                         rq->stats.rx_buf_alloc_failure++;
 618                                         break;
 619                                 }
 620                         } else {
 621                                 /* rx buffers skipped by the device */
 622                         }
 623                         val = VMXNET3_RXD_BTYPE_BODY << VMXNET3_RXD_BTYPE_SHIFT;
 624                 }
 625
 626                 gd->rxd.addr = cpu_to_le64(rbi->dma_addr);
 627                 gd->dword[2] = cpu_to_le32((!ring->gen << VMXNET3_RXD_GEN_SHIFT)
 628                                            | val | rbi->len);
 629
 630                 /* Fill the last buffer but dont mark it ready, or else the
 631                  * device will think that the queue is full */
 632                 if (num_allocated == num_to_alloc)
 633                         break;
 634
 635                 gd->dword[2] |= cpu_to_le32(ring->gen << VMXNET3_RXD_GEN_SHIFT);
 636                 num_allocated++;
 637                 vmxnet3_cmd_ring_adv_next2fill(ring);
 638         }
 639
 640         netdev_dbg(adapter->netdev,
 641                 "alloc_rx_buf: %d allocated, next2fill %u, next2comp %u\n",
 642                 num_allocated, ring->next2fill, ring->next2comp);
 643
 644         /* so that the device can distinguish a full ring and an empty ring */
 645         BUG_ON(num_allocated != 0 && ring->next2fill == ring->next2comp);
 646
 647         return num_allocated;
 648 }
 649
 650
 651 static void
 652 vmxnet3_append_frag(struct sk_buff *skb, struct Vmxnet3_RxCompDesc *rcd,
 653                     struct vmxnet3_rx_buf_info *rbi)
 654 {
 655         struct skb_frag_struct *frag = skb_shinfo(skb)->frags +
 656                 skb_shinfo(skb)->nr_frags;
 657
 658         BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS);
 659
 660         __skb_frag_set_page(frag, rbi->page);
 661         frag->page_offset = 0;
 662         skb_frag_size_set(frag, rcd->len);
 663         skb->data_len += rcd->len;
 664         skb->truesize += PAGE_SIZE;
 665         skb_shinfo(skb)->nr_frags++;
 666 }
 667
 668
 669 static int
 670 vmxnet3_map_pkt(struct sk_buff *skb, struct vmxnet3_tx_ctx *ctx,
 671                 struct vmxnet3_tx_queue *tq, struct pci_dev *pdev,
 672                 struct vmxnet3_adapter *adapter)
 673 {
 674         u32 dw2, len;
 675         unsigned long buf_offset;
 676         int i;
 677         union Vmxnet3_GenericDesc *gdesc;
 678         struct vmxnet3_tx_buf_info *tbi = NULL;
 679
 680         BUG_ON(ctx->copy_size > skb_headlen(skb));
 681
 682         /* use the previous gen bit for the SOP desc */
 683         dw2 = (tq->tx_ring.gen ^ 0x1) << VMXNET3_TXD_GEN_SHIFT;
 684
 685         ctx->sop_txd = tq->tx_ring.base + tq->tx_ring.next2fill;
 686         gdesc = ctx->sop_txd; /* both loops below can be skipped */
 687
 688         /* no need to map the buffer if headers are copied */
 689         if (ctx->copy_size) {
 690                 ctx->sop_txd->txd.addr = cpu_to_le64(tq->data_ring.basePA +
 691                                         tq->tx_ring.next2fill *
 692                                         sizeof(struct Vmxnet3_TxDataDesc));
 693                 ctx->sop_txd->dword[2] = cpu_to_le32(dw2 | ctx->copy_size);
 694                 ctx->sop_txd->dword[3] = 0;
 695
 696                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 697                 tbi->map_type = VMXNET3_MAP_NONE;
 698
 699                 netdev_dbg(adapter->netdev,
 700                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 701                         tq->tx_ring.next2fill,
 702                         le64_to_cpu(ctx->sop_txd->txd.addr),
 703                         ctx->sop_txd->dword[2], ctx->sop_txd->dword[3]);
 704                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 705
 706                 /* use the right gen for non-SOP desc */
 707                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 708         }
 709
 710         /* linear part can use multiple tx desc if it's big */
 711         len = skb_headlen(skb) - ctx->copy_size;
 712         buf_offset = ctx->copy_size;
 713         while (len) {
 714                 u32 buf_size;
 715
 716                 if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 717                         buf_size = len;
 718                         dw2 |= len;
 719                 } else {
 720                         buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 721                         /* spec says that for TxDesc.len, 0 == 2^14 */
 722                 }
 723
 724                 tbi = tq->buf_info + tq->tx_ring.next2fill;
 725                 tbi->map_type = VMXNET3_MAP_SINGLE;
 726                 tbi->dma_addr = dma_map_single(&adapter->pdev->dev,
 727                                 skb->data + buf_offset, buf_size,
 728                                 PCI_DMA_TODEVICE);
 729                 if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
 730                         return -EFAULT;
 731
 732                 tbi->len = buf_size;
 733
 734                 gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 735                 BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 736
 737                 gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 738                 gdesc->dword[2] = cpu_to_le32(dw2);
 739                 gdesc->dword[3] = 0;
 740
 741                 netdev_dbg(adapter->netdev,
 742                         "txd[%u]: 0x%Lx 0x%x 0x%x\n",
 743                         tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 744                         le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 745                 vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 746                 dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 747
 748                 len -= buf_size;
 749                 buf_offset += buf_size;
 750         }
 751
 752         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 753                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 754                 u32 buf_size;
 755
 756                 buf_offset = 0;
 757                 len = skb_frag_size(frag);
 758                 while (len) {
 759                         tbi = tq->buf_info + tq->tx_ring.next2fill;
 760                         if (len < VMXNET3_MAX_TX_BUF_SIZE) {
 761                                 buf_size = len;
 762                                 dw2 |= len;
 763                         } else {
 764                                 buf_size = VMXNET3_MAX_TX_BUF_SIZE;
 765                                 /* spec says that for TxDesc.len, 0 == 2^14 */
 766                         }
 767                         tbi->map_type = VMXNET3_MAP_PAGE;
 768                         tbi->dma_addr = skb_frag_dma_map(&adapter->pdev->dev, frag,
 769                                                          buf_offset, buf_size,
 770                                                          DMA_TO_DEVICE);
 771                         if (dma_mapping_error(&adapter->pdev->dev, tbi->dma_addr))
 772                                 return -EFAULT;
 773
 774                         tbi->len = buf_size;
 775
 776                         gdesc = tq->tx_ring.base + tq->tx_ring.next2fill;
 777                         BUG_ON(gdesc->txd.gen == tq->tx_ring.gen);
 778
 779                         gdesc->txd.addr = cpu_to_le64(tbi->dma_addr);
 780                         gdesc->dword[2] = cpu_to_le32(dw2);
 781                         gdesc->dword[3] = 0;
 782
 783                         netdev_dbg(adapter->netdev,
 784                                 "txd[%u]: 0x%llx %u %u\n",
 785                                 tq->tx_ring.next2fill, le64_to_cpu(gdesc->txd.addr),
 786                                 le32_to_cpu(gdesc->dword[2]), gdesc->dword[3]);
 787                         vmxnet3_cmd_ring_adv_next2fill(&tq->tx_ring);
 788                         dw2 = tq->tx_ring.gen << VMXNET3_TXD_GEN_SHIFT;
 789
 790                         len -= buf_size;
 791                         buf_offset += buf_size;
 792                 }
 793         }
 794
 795         ctx->eop_txd = gdesc;
 796
 797         /* set the last buf_info for the pkt */
 798         tbi->skb = skb;
 799         tbi->sop_idx = ctx->sop_txd - tq->tx_ring.base;
 800
 801         return 0;
 802 }
 803
 804
 805 /* Init all tx queues */
 806 static void
 807 vmxnet3_tq_init_all(struct vmxnet3_adapter *adapter)
 808 {
 809         int i;
 810
 811         for (i = 0; i < adapter->num_tx_queues; i++)
 812                 vmxnet3_tq_init(&adapter->tx_queue[i], adapter);
 813 }
 814
 815
 816 /*
 817  *    parse relevant protocol headers:
 818  *      For a tso pkt, relevant headers are L2/3/4 including options
 819  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 820  *      if it's a TCP/UDP pkt
 821  *
 822  * Returns:
 823  *    -1:  error happens during parsing
 824  *     0:  protocol headers parsed, but too big to be copied
 825  *     1:  protocol headers parsed and copied
 826  *
 827  * Other effects:
 828  *    1. related *ctx fields are updated.
 829  *    2. ctx->copy_size is # of bytes copied
 830  *    3. the portion to be copied is guaranteed to be in the linear part
 831  *
 832  */
 833 static int
 834 vmxnet3_parse_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 835                   struct vmxnet3_tx_ctx *ctx,
 836                   struct vmxnet3_adapter *adapter)
 837 {
 838         u8 protocol = 0;
 839
 840         if (ctx->mss) { /* TSO */
 841                 ctx->eth_ip_hdr_size = skb_transport_offset(skb);
 842                 ctx->l4_hdr_size = tcp_hdrlen(skb);
 843                 ctx->copy_size = ctx->eth_ip_hdr_size + ctx->l4_hdr_size;
 844         } else {
 845                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
 846                         ctx->eth_ip_hdr_size = skb_checksum_start_offset(skb);
 847
 848                         if (ctx->ipv4) {
 849                                 const struct iphdr *iph = ip_hdr(skb);
 850
 851                                 protocol = iph->protocol;
 852                         } else if (ctx->ipv6) {
 853                                 const struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 854
 855                                 protocol = ipv6h->nexthdr;
 856                         }
 857
 858                         switch (protocol) {
 859                         case IPPROTO_TCP:
 860                                 ctx->l4_hdr_size = tcp_hdrlen(skb);
 861                                 break;
 862                         case IPPROTO_UDP:
 863                                 ctx->l4_hdr_size = sizeof(struct udphdr);
 864                                 break;
 865                         default:
 866                                 ctx->l4_hdr_size = 0;
 867                                 break;
 868                         }
 869
 870                         ctx->copy_size = min(ctx->eth_ip_hdr_size +
 871                                          ctx->l4_hdr_size, skb->len);
 872                 } else {
 873                         ctx->eth_ip_hdr_size = 0;
 874                         ctx->l4_hdr_size = 0;
 875                         /* copy as much as allowed */
 876                         ctx->copy_size = min((unsigned int)VMXNET3_HDR_COPY_SIZE
 877                                              , skb_headlen(skb));
 878                 }
 879
 880                 if (skb->len <= VMXNET3_HDR_COPY_SIZE)
 881                         ctx->copy_size = skb->len;
 882
 883                 /* make sure headers are accessible directly */
 884                 if (unlikely(!pskb_may_pull(skb, ctx->copy_size)))
 885                         goto err;
 886         }
 887
 888         if (unlikely(ctx->copy_size > VMXNET3_HDR_COPY_SIZE)) {
 889                 tq->stats.oversized_hdr++;
 890                 ctx->copy_size = 0;
 891                 return 0;
 892         }
 893
 894         return 1;
 895 err:
 896         return -1;
 897 }
 898
 899 /*
 900  *    copy relevant protocol headers to the transmit ring:
 901  *      For a tso pkt, relevant headers are L2/3/4 including options
 902  *      For a pkt requesting csum offloading, they are L2/3 and may include L4
 903  *      if it's a TCP/UDP pkt
 904  *
 905  *
 906  *    Note that this requires that vmxnet3_parse_hdr be called first to set the
 907  *      appropriate bits in ctx first
 908  */
 909 static void
 910 vmxnet3_copy_hdr(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 911                  struct vmxnet3_tx_ctx *ctx,
 912                  struct vmxnet3_adapter *adapter)
 913 {
 914         struct Vmxnet3_TxDataDesc *tdd;
 915
 916         tdd = tq->data_ring.base + tq->tx_ring.next2fill;
 917
 918         memcpy(tdd->data, skb->data, ctx->copy_size);
 919         netdev_dbg(adapter->netdev,
 920                 "copy %u bytes to dataRing[%u]\n",
 921                 ctx->copy_size, tq->tx_ring.next2fill);
 922 }
 923
 924
 925 static void
 926 vmxnet3_prepare_tso(struct sk_buff *skb,
 927                     struct vmxnet3_tx_ctx *ctx)
 928 {
 929         struct tcphdr *tcph = tcp_hdr(skb);
 930
 931         if (ctx->ipv4) {
 932                 struct iphdr *iph = ip_hdr(skb);
 933
 934                 iph->check = 0;
 935                 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 0,
 936                                                  IPPROTO_TCP, 0);
 937         } else if (ctx->ipv6) {
 938                 struct ipv6hdr *iph = ipv6_hdr(skb);
 939
 940                 tcph->check = ~csum_ipv6_magic(&iph->saddr, &iph->daddr, 0,
 941                                                IPPROTO_TCP, 0);
 942         }
 943 }
 944
 945 static int txd_estimate(const struct sk_buff *skb)
 946 {
 947         int count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
 948         int i;
 949
 950         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 951                 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i];
 952
 953                 count += VMXNET3_TXD_NEEDED(skb_frag_size(frag));
 954         }
 955         return count;
 956 }
 957
 958 /*
 959  * Transmits a pkt thru a given tq
 960  * Returns:
 961  *    NETDEV_TX_OK:      descriptors are setup successfully
 962  *    NETDEV_TX_OK:      error occurred, the pkt is dropped
 963  *    NETDEV_TX_BUSY:    tx ring is full, queue is stopped
 964  *
 965  * Side-effects:
 966  *    1. tx ring may be changed
 967  *    2. tq stats may be updated accordingly
 968  *    3. shared->txNumDeferred may be updated
 969  */
 970
 971 static int
 972 vmxnet3_tq_xmit(struct sk_buff *skb, struct vmxnet3_tx_queue *tq,
 973                 struct vmxnet3_adapter *adapter, struct net_device *netdev)
 974 {
 975         int ret;
 976         u32 count;
 977         unsigned long flags;
 978         struct vmxnet3_tx_ctx ctx;
 979         union Vmxnet3_GenericDesc *gdesc;
 980 #ifdef __BIG_ENDIAN_BITFIELD
 981         /* Use temporary descriptor to avoid touching bits multiple times */
 982         union Vmxnet3_GenericDesc tempTxDesc;
 983 #endif
 984
 985         count = txd_estimate(skb);
 986
 987         ctx.ipv4 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IP));
 988         ctx.ipv6 = (vlan_get_protocol(skb) == cpu_to_be16(ETH_P_IPV6));
 989
 990         ctx.mss = skb_shinfo(skb)->gso_size;
 991         if (ctx.mss) {
 992                 if (skb_header_cloned(skb)) {
 993                         if (unlikely(pskb_expand_head(skb, 0, 0,
 994                                                       GFP_ATOMIC) != 0)) {
 995                                 tq->stats.drop_tso++;
 996                                 goto drop_pkt;
 997                         }
 998                         tq->stats.copy_skb_header++;
 999                 }
1000                 vmxnet3_prepare_tso(skb, &ctx);
1001         } else {
1002                 if (unlikely(count > VMXNET3_MAX_TXD_PER_PKT)) {
1003
1004                         /* non-tso pkts must not use more than
1005                          * VMXNET3_MAX_TXD_PER_PKT entries
1006                          */
1007                         if (skb_linearize(skb) != 0) {
1008                                 tq->stats.drop_too_many_frags++;
1009                                 goto drop_pkt;
1010                         }
1011                         tq->stats.linearized++;
1012
1013                         /* recalculate the # of descriptors to use */
1014                         count = VMXNET3_TXD_NEEDED(skb_headlen(skb)) + 1;
1015                 }
1016         }
1017
1018         ret = vmxnet3_parse_hdr(skb, tq, &ctx, adapter);
1019         if (ret >= 0) {
1020                 BUG_ON(ret <= 0 && ctx.copy_size != 0);
1021                 /* hdrs parsed, check against other limits */
1022                 if (ctx.mss) {
1023                         if (unlikely(ctx.eth_ip_hdr_size + ctx.l4_hdr_size >
1024                                      VMXNET3_MAX_TX_BUF_SIZE)) {
1025                                 tq->stats.drop_oversized_hdr++;
1026                                 goto drop_pkt;
1027                         }
1028                 } else {
1029                         if (skb->ip_summed == CHECKSUM_PARTIAL) {
1030                                 if (unlikely(ctx.eth_ip_hdr_size +
1031                                              skb->csum_offset >
1032                                              VMXNET3_MAX_CSUM_OFFSET)) {
1033                                         tq->stats.drop_oversized_hdr++;
1034                                         goto drop_pkt;
1035                                 }
1036                         }
1037                 }
1038         } else {
1039                 tq->stats.drop_hdr_inspect_err++;
1040                 goto drop_pkt;
1041         }
1042
1043         spin_lock_irqsave(&tq->tx_lock, flags);
1044
1045         if (count > vmxnet3_cmd_ring_desc_avail(&tq->tx_ring)) {
1046                 tq->stats.tx_ring_full++;
1047                 netdev_dbg(adapter->netdev,
1048                         "tx queue stopped on %s, next2comp %u"
1049                         " next2fill %u\n", adapter->netdev->name,
1050                         tq->tx_ring.next2comp, tq->tx_ring.next2fill);
1051
1052                 vmxnet3_tq_stop(tq, adapter);
1053                 spin_unlock_irqrestore(&tq->tx_lock, flags);
1054                 return NETDEV_TX_BUSY;
1055         }
1056
1057
1058         vmxnet3_copy_hdr(skb, tq, &ctx, adapter);
1059
1060         /* fill tx descs related to addr & len */
1061         if (vmxnet3_map_pkt(skb, &ctx, tq, adapter->pdev, adapter))
1062                 goto unlock_drop_pkt;
1063
1064         /* setup the EOP desc */
1065         ctx.eop_txd->dword[3] = cpu_to_le32(VMXNET3_TXD_CQ | VMXNET3_TXD_EOP);
1066
1067         /* setup the SOP desc */
1068 #ifdef __BIG_ENDIAN_BITFIELD
1069         gdesc = &tempTxDesc;
1070         gdesc->dword[2] = ctx.sop_txd->dword[2];
1071         gdesc->dword[3] = ctx.sop_txd->dword[3];
1072 #else
1073         gdesc = ctx.sop_txd;
1074 #endif
1075         if (ctx.mss) {
1076                 gdesc->txd.hlen = ctx.eth_ip_hdr_size + ctx.l4_hdr_size;
1077                 gdesc->txd.om = VMXNET3_OM_TSO;
1078                 gdesc->txd.msscof = ctx.mss;
1079                 le32_add_cpu(&tq->shared->txNumDeferred, (skb->len -
1080                              gdesc->txd.hlen + ctx.mss - 1) / ctx.mss);
1081         } else {
1082                 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1083                         gdesc->txd.hlen = ctx.eth_ip_hdr_size;
1084                         gdesc->txd.om = VMXNET3_OM_CSUM;
1085                         gdesc->txd.msscof = ctx.eth_ip_hdr_size +
1086                                             skb->csum_offset;
1087                 } else {
1088                         gdesc->txd.om = 0;
1089                         gdesc->txd.msscof = 0;
1090                 }
1091                 le32_add_cpu(&tq->shared->txNumDeferred, 1);
1092         }
1093
1094         if (skb_vlan_tag_present(skb)) {
1095                 gdesc->txd.ti = 1;
1096                 gdesc->txd.tci = skb_vlan_tag_get(skb);
1097         }
1098
1099         /* finally flips the GEN bit of the SOP desc. */
1100         gdesc->dword[2] = cpu_to_le32(le32_to_cpu(gdesc->dword[2]) ^
1101                                                   VMXNET3_TXD_GEN);
1102 #ifdef __BIG_ENDIAN_BITFIELD
1103         /* Finished updating in bitfields of Tx Desc, so write them in original
1104          * place.
1105          */
1106         vmxnet3_TxDescToLe((struct Vmxnet3_TxDesc *)gdesc,
1107                            (struct Vmxnet3_TxDesc *)ctx.sop_txd);
1108         gdesc = ctx.sop_txd;
1109 #endif
1110         netdev_dbg(adapter->netdev,
1111                 "txd[%u]: SOP 0x%Lx 0x%x 0x%x\n",
1112                 (u32)(ctx.sop_txd -
1113                 tq->tx_ring.base), le64_to_cpu(gdesc->txd.addr),
1114                 le32_to_cpu(gdesc->dword[2]), le32_to_cpu(gdesc->dword[3]));
1115
1116         spin_unlock_irqrestore(&tq->tx_lock, flags);
1117
1118         if (le32_to_cpu(tq->shared->txNumDeferred) >=
1119                                         le32_to_cpu(tq->shared->txThreshold)) {
1120                 tq->shared->txNumDeferred = 0;
1121                 VMXNET3_WRITE_BAR0_REG(adapter,
1122                                        VMXNET3_REG_TXPROD + tq->qid * 8,
1123                                        tq->tx_ring.next2fill);
1124         }
1125
1126         return NETDEV_TX_OK;
1127
1128 unlock_drop_pkt:
1129         spin_unlock_irqrestore(&tq->tx_lock, flags);
1130 drop_pkt:
1131         tq->stats.drop_total++;
1132         dev_kfree_skb_any(skb);
1133         return NETDEV_TX_OK;
1134 }
1135
1136
1137 static netdev_tx_t
1138 vmxnet3_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
1139 {
1140         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1141
1142         BUG_ON(skb->queue_mapping > adapter->num_tx_queues);
1143         return vmxnet3_tq_xmit(skb,
1144                                &adapter->tx_queue[skb->queue_mapping],
1145                                adapter, netdev);
1146 }
1147
1148
1149 static void
1150 vmxnet3_rx_csum(struct vmxnet3_adapter *adapter,
1151                 struct sk_buff *skb,
1152                 union Vmxnet3_GenericDesc *gdesc)
1153 {
1154         if (!gdesc->rcd.cnc && adapter->netdev->features & NETIF_F_RXCSUM) {
1155                 if (gdesc->rcd.v4 &&
1156                     (le32_to_cpu(gdesc->dword[3]) &
1157                      VMXNET3_RCD_CSUM_OK) == VMXNET3_RCD_CSUM_OK) {
1158                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1159                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1160                         BUG_ON(gdesc->rcd.frg);
1161                 } else if (gdesc->rcd.v6 && (le32_to_cpu(gdesc->dword[3]) &
1162                                              (1 << VMXNET3_RCD_TUC_SHIFT))) {
1163                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1164                         BUG_ON(!(gdesc->rcd.tcp || gdesc->rcd.udp));
1165                         BUG_ON(gdesc->rcd.frg);
1166                 } else {
1167                         if (gdesc->rcd.csum) {
1168                                 skb->csum = htons(gdesc->rcd.csum);
1169                                 skb->ip_summed = CHECKSUM_PARTIAL;
1170                         } else {
1171                                 skb_checksum_none_assert(skb);
1172                         }
1173                 }
1174         } else {
1175                 skb_checksum_none_assert(skb);
1176         }
1177 }
1178
1179
1180 static void
1181 vmxnet3_rx_error(struct vmxnet3_rx_queue *rq, struct Vmxnet3_RxCompDesc *rcd,
1182                  struct vmxnet3_rx_ctx *ctx,  struct vmxnet3_adapter *adapter)
1183 {
1184         rq->stats.drop_err++;
1185         if (!rcd->fcs)
1186                 rq->stats.drop_fcs++;
1187
1188         rq->stats.drop_total++;
1189
1190         /*
1191          * We do not unmap and chain the rx buffer to the skb.
1192          * We basically pretend this buffer is not used and will be recycled
1193          * by vmxnet3_rq_alloc_rx_buf()
1194          */
1195
1196         /*
1197          * ctx->skb may be NULL if this is the first and the only one
1198          * desc for the pkt
1199          */
1200         if (ctx->skb)
1201                 dev_kfree_skb_irq(ctx->skb);
1202
1203         ctx->skb = NULL;
1204 }
1205
1206
1207 static u32
1208 vmxnet3_get_hdr_len(struct vmxnet3_adapter *adapter, struct sk_buff *skb,
1209                     union Vmxnet3_GenericDesc *gdesc)
1210 {
1211         u32 hlen, maplen;
1212         union {
1213                 void *ptr;
1214                 struct ethhdr *eth;
1215                 struct iphdr *ipv4;
1216                 struct ipv6hdr *ipv6;
1217                 struct tcphdr *tcp;
1218         } hdr;
1219         BUG_ON(gdesc->rcd.tcp == 0);
1220
1221         maplen = skb_headlen(skb);
1222         if (unlikely(sizeof(struct iphdr) + sizeof(struct tcphdr) > maplen))
1223                 return 0;
1224
1225         hdr.eth = eth_hdr(skb);
1226         if (gdesc->rcd.v4) {
1227                 BUG_ON(hdr.eth->h_proto != htons(ETH_P_IP));
1228                 hdr.ptr += sizeof(struct ethhdr);
1229                 BUG_ON(hdr.ipv4->protocol != IPPROTO_TCP);
1230                 hlen = hdr.ipv4->ihl << 2;
1231                 hdr.ptr += hdr.ipv4->ihl << 2;
1232         } else if (gdesc->rcd.v6) {
1233                 BUG_ON(hdr.eth->h_proto != htons(ETH_P_IPV6));
1234                 hdr.ptr += sizeof(struct ethhdr);
1235                 /* Use an estimated value, since we also need to handle
1236                  * TSO case.
1237                  */
1238                 if (hdr.ipv6->nexthdr != IPPROTO_TCP)
1239                         return sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
1240                 hlen = sizeof(struct ipv6hdr);
1241                 hdr.ptr += sizeof(struct ipv6hdr);
1242         } else {
1243                 /* Non-IP pkt, dont estimate header length */
1244                 return 0;
1245         }
1246
1247         if (hlen + sizeof(struct tcphdr) > maplen)
1248                 return 0;
1249
1250         return (hlen + (hdr.tcp->doff << 2));
1251 }
1252
1253 static int
1254 vmxnet3_rq_rx_complete(struct vmxnet3_rx_queue *rq,
1255                        struct vmxnet3_adapter *adapter, int quota)
1256 {
1257         static const u32 rxprod_reg[2] = {
1258                 VMXNET3_REG_RXPROD, VMXNET3_REG_RXPROD2
1259         };
1260         u32 num_pkts = 0;
1261         bool skip_page_frags = false;
1262         struct Vmxnet3_RxCompDesc *rcd;
1263         struct vmxnet3_rx_ctx *ctx = &rq->rx_ctx;
1264         u16 segCnt = 0, mss = 0;
1265 #ifdef __BIG_ENDIAN_BITFIELD
1266         struct Vmxnet3_RxDesc rxCmdDesc;
1267         struct Vmxnet3_RxCompDesc rxComp;
1268 #endif
1269         vmxnet3_getRxComp(rcd, &rq->comp_ring.base[rq->comp_ring.next2proc].rcd,
1270                           &rxComp);
1271         while (rcd->gen == rq->comp_ring.gen) {
1272                 struct vmxnet3_rx_buf_info *rbi;
1273                 struct sk_buff *skb, *new_skb = NULL;
1274                 struct page *new_page = NULL;
1275                 dma_addr_t new_dma_addr;
1276                 int num_to_alloc;
1277                 struct Vmxnet3_RxDesc *rxd;
1278                 u32 idx, ring_idx;
1279                 struct vmxnet3_cmd_ring *ring = NULL;
1280                 if (num_pkts >= quota) {
1281                         /* we may stop even before we see the EOP desc of
1282                          * the current pkt
1283                          */
1284                         break;
1285                 }
1286                 BUG_ON(rcd->rqID != rq->qid && rcd->rqID != rq->qid2);
1287                 idx = rcd->rxdIdx;
1288                 ring_idx = rcd->rqID < adapter->num_rx_queues ? 0 : 1;
1289                 ring = rq->rx_ring + ring_idx;
1290                 vmxnet3_getRxDesc(rxd, &rq->rx_ring[ring_idx].base[idx].rxd,
1291                                   &rxCmdDesc);
1292                 rbi = rq->buf_info[ring_idx] + idx;
1293
1294                 BUG_ON(rxd->addr != rbi->dma_addr ||
1295                        rxd->len != rbi->len);
1296
1297                 if (unlikely(rcd->eop && rcd->err)) {
1298                         vmxnet3_rx_error(rq, rcd, ctx, adapter);
1299                         goto rcd_done;
1300                 }
1301
1302                 if (rcd->sop) { /* first buf of the pkt */
1303                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_HEAD ||
1304                                rcd->rqID != rq->qid);
1305
1306                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_SKB);
1307                         BUG_ON(ctx->skb != NULL || rbi->skb == NULL);
1308
1309                         if (unlikely(rcd->len == 0)) {
1310                                 /* Pretend the rx buffer is skipped. */
1311                                 BUG_ON(!(rcd->sop && rcd->eop));
1312                                 netdev_dbg(adapter->netdev,
1313                                         "rxRing[%u][%u] 0 length\n",
1314                                         ring_idx, idx);
1315                                 goto rcd_done;
1316                         }
1317
1318                         skip_page_frags = false;
1319                         ctx->skb = rbi->skb;
1320                         new_skb = netdev_alloc_skb_ip_align(adapter->netdev,
1321                                                             rbi->len);
1322                         if (new_skb == NULL) {
1323                                 /* Skb allocation failed, do not handover this
1324                                  * skb to stack. Reuse it. Drop the existing pkt
1325                                  */
1326                                 rq->stats.rx_buf_alloc_failure++;
1327                                 ctx->skb = NULL;
1328                                 rq->stats.drop_total++;
1329                                 skip_page_frags = true;
1330                                 goto rcd_done;
1331                         }
1332                         new_dma_addr = dma_map_single(&adapter->pdev->dev,
1333                                                       new_skb->data, rbi->len,
1334                                                       PCI_DMA_FROMDEVICE);
1335                         if (dma_mapping_error(&adapter->pdev->dev,
1336                                               new_dma_addr)) {
1337                                 dev_kfree_skb(new_skb);
1338                                 /* Skb allocation failed, do not handover this
1339                                  * skb to stack. Reuse it. Drop the existing pkt
1340                                  */
1341                                 rq->stats.rx_buf_alloc_failure++;
1342                                 ctx->skb = NULL;
1343                                 rq->stats.drop_total++;
1344                                 skip_page_frags = true;
1345                                 goto rcd_done;
1346                         }
1347
1348                         dma_unmap_single(&adapter->pdev->dev, rbi->dma_addr,
1349                                          rbi->len,
1350                                          PCI_DMA_FROMDEVICE);
1351
1352 #ifdef VMXNET3_RSS
1353                         if (rcd->rssType != VMXNET3_RCD_RSS_TYPE_NONE &&
1354                             (adapter->netdev->features & NETIF_F_RXHASH))
1355                                 skb_set_hash(ctx->skb,
1356                                              le32_to_cpu(rcd->rssHash),
1357                                              PKT_HASH_TYPE_L3);
1358 #endif
1359                         skb_put(ctx->skb, rcd->len);
1360
1361                         /* Immediate refill */
1362                         rbi->skb = new_skb;
1363                         rbi->dma_addr = new_dma_addr;
1364                         rxd->addr = cpu_to_le64(rbi->dma_addr);
1365                         rxd->len = rbi->len;
1366                         if (adapter->version == 2 &&
1367                             rcd->type == VMXNET3_CDTYPE_RXCOMP_LRO) {
1368                                 struct Vmxnet3_RxCompDescExt *rcdlro;
1369                                 rcdlro = (struct Vmxnet3_RxCompDescExt *)rcd;
1370
1371                                 segCnt = rcdlro->segCnt;
1372                                 WARN_ON_ONCE(segCnt == 0);
1373                                 mss = rcdlro->mss;
1374                                 if (unlikely(segCnt <= 1))
1375                                         segCnt = 0;
1376                         } else {
1377                                 segCnt = 0;
1378                         }
1379                 } else {
1380                         BUG_ON(ctx->skb == NULL && !skip_page_frags);
1381
1382                         /* non SOP buffer must be type 1 in most cases */
1383                         BUG_ON(rbi->buf_type != VMXNET3_RX_BUF_PAGE);
1384                         BUG_ON(rxd->btype != VMXNET3_RXD_BTYPE_BODY);
1385
1386                         /* If an sop buffer was dropped, skip all
1387                          * following non-sop fragments. They will be reused.
1388                          */
1389                         if (skip_page_frags)
1390                                 goto rcd_done;
1391
1392                         if (rcd->len) {
1393                                 new_page = alloc_page(GFP_ATOMIC);
1394                                 /* Replacement page frag could not be allocated.
1395                                  * Reuse this page. Drop the pkt and free the
1396                                  * skb which contained this page as a frag. Skip
1397                                  * processing all the following non-sop frags.
1398                                  */
1399                                 if (unlikely(!new_page)) {
1400                                         rq->stats.rx_buf_alloc_failure++;
1401                                         dev_kfree_skb(ctx->skb);
1402                                         ctx->skb = NULL;
1403                                         skip_page_frags = true;
1404                                         goto rcd_done;
1405                                 }
1406                                 new_dma_addr = dma_map_page(&adapter->pdev->dev,
1407                                                             new_page,
1408                                                             0, PAGE_SIZE,
1409                                                             PCI_DMA_FROMDEVICE);
1410                                 if (dma_mapping_error(&adapter->pdev->dev,
1411                                                       new_dma_addr)) {
1412                                         put_page(new_page);
1413                                         rq->stats.rx_buf_alloc_failure++;
1414                                         dev_kfree_skb(ctx->skb);
1415                                         ctx->skb = NULL;
1416                                         skip_page_frags = true;
1417                                         goto rcd_done;
1418                                 }
1419
1420                                 dma_unmap_page(&adapter->pdev->dev,
1421                                                rbi->dma_addr, rbi->len,
1422                                                PCI_DMA_FROMDEVICE);
1423
1424                                 vmxnet3_append_frag(ctx->skb, rcd, rbi);
1425
1426                                 /* Immediate refill */
1427                                 rbi->page = new_page;
1428                                 rbi->dma_addr = new_dma_addr;
1429                                 rxd->addr = cpu_to_le64(rbi->dma_addr);
1430                                 rxd->len = rbi->len;
1431                         }
1432                 }
1433
1434
1435                 skb = ctx->skb;
1436                 if (rcd->eop) {
1437                         u32 mtu = adapter->netdev->mtu;
1438                         skb->len += skb->data_len;
1439
1440                         vmxnet3_rx_csum(adapter, skb,
1441                                         (union Vmxnet3_GenericDesc *)rcd);
1442                         skb->protocol = eth_type_trans(skb, adapter->netdev);
1443                         if (!rcd->tcp || !adapter->lro)
1444                                 goto not_lro;
1445
1446                         if (segCnt != 0 && mss != 0) {
1447                                 skb_shinfo(skb)->gso_type = rcd->v4 ?
1448                                         SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1449                                 skb_shinfo(skb)->gso_size = mss;
1450                                 skb_shinfo(skb)->gso_segs = segCnt;
1451                         } else if (segCnt != 0 || skb->len > mtu) {
1452                                 u32 hlen;
1453
1454                                 hlen = vmxnet3_get_hdr_len(adapter, skb,
1455                                         (union Vmxnet3_GenericDesc *)rcd);
1456                                 if (hlen == 0)
1457                                         goto not_lro;
1458
1459                                 skb_shinfo(skb)->gso_type =
1460                                         rcd->v4 ? SKB_GSO_TCPV4 : SKB_GSO_TCPV6;
1461                                 if (segCnt != 0) {
1462                                         skb_shinfo(skb)->gso_segs = segCnt;
1463                                         skb_shinfo(skb)->gso_size =
1464                                                 DIV_ROUND_UP(skb->len -
1465                                                         hlen, segCnt);
1466                                 } else {
1467                                         skb_shinfo(skb)->gso_size = mtu - hlen;
1468                                 }
1469                         }
1470 not_lro:
1471                         if (unlikely(rcd->ts))
1472                                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), rcd->tci);
1473
1474                         if (adapter->netdev->features & NETIF_F_LRO)
1475                                 netif_receive_skb(skb);
1476                         else
1477                                 napi_gro_receive(&rq->napi, skb);
1478
1479                         ctx->skb = NULL;
1480                         num_pkts++;
1481                 }
1482
1483 rcd_done:
1484                 /* device may have skipped some rx descs */
1485                 ring->next2comp = idx;
1486                 num_to_alloc = vmxnet3_cmd_ring_desc_avail(ring);
1487                 ring = rq->rx_ring + ring_idx;
1488                 while (num_to_alloc) {
1489                         vmxnet3_getRxDesc(rxd, &ring->base[ring->next2fill].rxd,
1490                                           &rxCmdDesc);
1491                         BUG_ON(!rxd->addr);
1492
1493                         /* Recv desc is ready to be used by the device */
1494                         rxd->gen = ring->gen;
1495                         vmxnet3_cmd_ring_adv_next2fill(ring);
1496                         num_to_alloc--;
1497                 }
1498
1499                 /* if needed, update the register */
1500                 if (unlikely(rq->shared->updateRxProd)) {
1501                         VMXNET3_WRITE_BAR0_REG(adapter,
1502                                                rxprod_reg[ring_idx] + rq->qid * 8,
1503                                                ring->next2fill);
1504                 }
1505
1506                 vmxnet3_comp_ring_adv_next2proc(&rq->comp_ring);
1507                 vmxnet3_getRxComp(rcd,
1508                                   &rq->comp_ring.base[rq->comp_ring.next2proc].rcd, &rxComp);
1509         }
1510
1511         return num_pkts;
1512 }
1513
1514
1515 static void
1516 vmxnet3_rq_cleanup(struct vmxnet3_rx_queue *rq,
1517                    struct vmxnet3_adapter *adapter)
1518 {
1519         u32 i, ring_idx;
1520         struct Vmxnet3_RxDesc *rxd;
1521
1522         for (ring_idx = 0; ring_idx < 2; ring_idx++) {
1523                 for (i = 0; i < rq->rx_ring[ring_idx].size; i++) {
1524 #ifdef __BIG_ENDIAN_BITFIELD
1525                         struct Vmxnet3_RxDesc rxDesc;
1526 #endif
1527                         vmxnet3_getRxDesc(rxd,
1528                                 &rq->rx_ring[ring_idx].base[i].rxd, &rxDesc);
1529
1530                         if (rxd->btype == VMXNET3_RXD_BTYPE_HEAD &&
1531                                         rq->buf_info[ring_idx][i].skb) {
1532                                 dma_unmap_single(&adapter->pdev->dev, rxd->addr,
1533                                                  rxd->len, PCI_DMA_FROMDEVICE);
1534                                 dev_kfree_skb(rq->buf_info[ring_idx][i].skb);
1535                                 rq->buf_info[ring_idx][i].skb = NULL;
1536                         } else if (rxd->btype == VMXNET3_RXD_BTYPE_BODY &&
1537                                         rq->buf_info[ring_idx][i].page) {
1538                                 dma_unmap_page(&adapter->pdev->dev, rxd->addr,
1539                                                rxd->len, PCI_DMA_FROMDEVICE);
1540                                 put_page(rq->buf_info[ring_idx][i].page);
1541                                 rq->buf_info[ring_idx][i].page = NULL;
1542                         }
1543                 }
1544
1545                 rq->rx_ring[ring_idx].gen = VMXNET3_INIT_GEN;
1546                 rq->rx_ring[ring_idx].next2fill =
1547                                         rq->rx_ring[ring_idx].next2comp = 0;
1548         }
1549
1550         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1551         rq->comp_ring.next2proc = 0;
1552 }
1553
1554
1555 static void
1556 vmxnet3_rq_cleanup_all(struct vmxnet3_adapter *adapter)
1557 {
1558         int i;
1559
1560         for (i = 0; i < adapter->num_rx_queues; i++)
1561                 vmxnet3_rq_cleanup(&adapter->rx_queue[i], adapter);
1562 }
1563
1564
1565 static void vmxnet3_rq_destroy(struct vmxnet3_rx_queue *rq,
1566                                struct vmxnet3_adapter *adapter)
1567 {
1568         int i;
1569         int j;
1570
1571         /* all rx buffers must have already been freed */
1572         for (i = 0; i < 2; i++) {
1573                 if (rq->buf_info[i]) {
1574                         for (j = 0; j < rq->rx_ring[i].size; j++)
1575                                 BUG_ON(rq->buf_info[i][j].page != NULL);
1576                 }
1577         }
1578
1579
1580         for (i = 0; i < 2; i++) {
1581                 if (rq->rx_ring[i].base) {
1582                         dma_free_coherent(&adapter->pdev->dev,
1583                                           rq->rx_ring[i].size
1584                                           * sizeof(struct Vmxnet3_RxDesc),
1585                                           rq->rx_ring[i].base,
1586                                           rq->rx_ring[i].basePA);
1587                         rq->rx_ring[i].base = NULL;
1588                 }
1589                 rq->buf_info[i] = NULL;
1590         }
1591
1592         if (rq->comp_ring.base) {
1593                 dma_free_coherent(&adapter->pdev->dev, rq->comp_ring.size
1594                                   * sizeof(struct Vmxnet3_RxCompDesc),
1595                                   rq->comp_ring.base, rq->comp_ring.basePA);
1596                 rq->comp_ring.base = NULL;
1597         }
1598
1599         if (rq->buf_info[0]) {
1600                 size_t sz = sizeof(struct vmxnet3_rx_buf_info) *
1601                         (rq->rx_ring[0].size + rq->rx_ring[1].size);
1602                 dma_free_coherent(&adapter->pdev->dev, sz, rq->buf_info[0],
1603                                   rq->buf_info_pa);
1604         }
1605 }
1606
1607
1608 static int
1609 vmxnet3_rq_init(struct vmxnet3_rx_queue *rq,
1610                 struct vmxnet3_adapter  *adapter)
1611 {
1612         int i;
1613
1614         /* initialize buf_info */
1615         for (i = 0; i < rq->rx_ring[0].size; i++) {
1616
1617                 /* 1st buf for a pkt is skbuff */
1618                 if (i % adapter->rx_buf_per_pkt == 0) {
1619                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_SKB;
1620                         rq->buf_info[0][i].len = adapter->skb_buf_size;
1621                 } else { /* subsequent bufs for a pkt is frag */
1622                         rq->buf_info[0][i].buf_type = VMXNET3_RX_BUF_PAGE;
1623                         rq->buf_info[0][i].len = PAGE_SIZE;
1624                 }
1625         }
1626         for (i = 0; i < rq->rx_ring[1].size; i++) {
1627                 rq->buf_info[1][i].buf_type = VMXNET3_RX_BUF_PAGE;
1628                 rq->buf_info[1][i].len = PAGE_SIZE;
1629         }
1630
1631         /* reset internal state and allocate buffers for both rings */
1632         for (i = 0; i < 2; i++) {
1633                 rq->rx_ring[i].next2fill = rq->rx_ring[i].next2comp = 0;
1634
1635                 memset(rq->rx_ring[i].base, 0, rq->rx_ring[i].size *
1636                        sizeof(struct Vmxnet3_RxDesc));
1637                 rq->rx_ring[i].gen = VMXNET3_INIT_GEN;
1638         }
1639         if (vmxnet3_rq_alloc_rx_buf(rq, 0, rq->rx_ring[0].size - 1,
1640                                     adapter) == 0) {
1641                 /* at least has 1 rx buffer for the 1st ring */
1642                 return -ENOMEM;
1643         }
1644         vmxnet3_rq_alloc_rx_buf(rq, 1, rq->rx_ring[1].size - 1, adapter);
1645
1646         /* reset the comp ring */
1647         rq->comp_ring.next2proc = 0;
1648         memset(rq->comp_ring.base, 0, rq->comp_ring.size *
1649                sizeof(struct Vmxnet3_RxCompDesc));
1650         rq->comp_ring.gen = VMXNET3_INIT_GEN;
1651
1652         /* reset rxctx */
1653         rq->rx_ctx.skb = NULL;
1654
1655         /* stats are not reset */
1656         return 0;
1657 }
1658
1659
1660 static int
1661 vmxnet3_rq_init_all(struct vmxnet3_adapter *adapter)
1662 {
1663         int i, err = 0;
1664
1665         for (i = 0; i < adapter->num_rx_queues; i++) {
1666                 err = vmxnet3_rq_init(&adapter->rx_queue[i], adapter);
1667                 if (unlikely(err)) {
1668                         dev_err(&adapter->netdev->dev, "%s: failed to "
1669                                 "initialize rx queue%i\n",
1670                                 adapter->netdev->name, i);
1671                         break;
1672                 }
1673         }
1674         return err;
1675
1676 }
1677
1678
1679 static int
1680 vmxnet3_rq_create(struct vmxnet3_rx_queue *rq, struct vmxnet3_adapter *adapter)
1681 {
1682         int i;
1683         size_t sz;
1684         struct vmxnet3_rx_buf_info *bi;
1685
1686         for (i = 0; i < 2; i++) {
1687
1688                 sz = rq->rx_ring[i].size * sizeof(struct Vmxnet3_RxDesc);
1689                 rq->rx_ring[i].base = dma_alloc_coherent(
1690                                                 &adapter->pdev->dev, sz,
1691                                                 &rq->rx_ring[i].basePA,
1692                                                 GFP_KERNEL);
1693                 if (!rq->rx_ring[i].base) {
1694                         netdev_err(adapter->netdev,
1695                                    "failed to allocate rx ring %d\n", i);
1696                         goto err;
1697                 }
1698         }
1699
1700         sz = rq->comp_ring.size * sizeof(struct Vmxnet3_RxCompDesc);
1701         rq->comp_ring.base = dma_alloc_coherent(&adapter->pdev->dev, sz,
1702                                                 &rq->comp_ring.basePA,
1703                                                 GFP_KERNEL);
1704         if (!rq->comp_ring.base) {
1705                 netdev_err(adapter->netdev, "failed to allocate rx comp ring\n");
1706                 goto err;
1707         }
1708
1709         sz = sizeof(struct vmxnet3_rx_buf_info) * (rq->rx_ring[0].size +
1710                                                    rq->rx_ring[1].size);
1711         bi = dma_zalloc_coherent(&adapter->pdev->dev, sz, &rq->buf_info_pa,
1712                                  GFP_KERNEL);
1713         if (!bi)
1714                 goto err;
1715
1716         rq->buf_info[0] = bi;
1717         rq->buf_info[1] = bi + rq->rx_ring[0].size;
1718
1719         return 0;
1720
1721 err:
1722         vmxnet3_rq_destroy(rq, adapter);
1723         return -ENOMEM;
1724 }
1725
1726
1727 static int
1728 vmxnet3_rq_create_all(struct vmxnet3_adapter *adapter)
1729 {
1730         int i, err = 0;
1731
1732         for (i = 0; i < adapter->num_rx_queues; i++) {
1733                 err = vmxnet3_rq_create(&adapter->rx_queue[i], adapter);
1734                 if (unlikely(err)) {
1735                         dev_err(&adapter->netdev->dev,
1736                                 "%s: failed to create rx queue%i\n",
1737                                 adapter->netdev->name, i);
1738                         goto err_out;
1739                 }
1740         }
1741         return err;
1742 err_out:
1743         vmxnet3_rq_destroy_all(adapter);
1744         return err;
1745
1746 }
1747
1748 /* Multiple queue aware polling function for tx and rx */
1749
1750 static int
1751 vmxnet3_do_poll(struct vmxnet3_adapter *adapter, int budget)
1752 {
1753         int rcd_done = 0, i;
1754         if (unlikely(adapter->shared->ecr))
1755                 vmxnet3_process_events(adapter);
1756         for (i = 0; i < adapter->num_tx_queues; i++)
1757                 vmxnet3_tq_tx_complete(&adapter->tx_queue[i], adapter);
1758
1759         for (i = 0; i < adapter->num_rx_queues; i++)
1760                 rcd_done += vmxnet3_rq_rx_complete(&adapter->rx_queue[i],
1761                                                    adapter, budget);
1762         return rcd_done;
1763 }
1764
1765
1766 static int
1767 vmxnet3_poll(struct napi_struct *napi, int budget)
1768 {
1769         struct vmxnet3_rx_queue *rx_queue = container_of(napi,
1770                                           struct vmxnet3_rx_queue, napi);
1771         int rxd_done;
1772
1773         rxd_done = vmxnet3_do_poll(rx_queue->adapter, budget);
1774
1775         if (rxd_done < budget) {
1776                 napi_complete(napi);
1777                 vmxnet3_enable_all_intrs(rx_queue->adapter);
1778         }
1779         return rxd_done;
1780 }
1781
1782 /*
1783  * NAPI polling function for MSI-X mode with multiple Rx queues
1784  * Returns the # of the NAPI credit consumed (# of rx descriptors processed)
1785  */
1786
1787 static int
1788 vmxnet3_poll_rx_only(struct napi_struct *napi, int budget)
1789 {
1790         struct vmxnet3_rx_queue *rq = container_of(napi,
1791                                                 struct vmxnet3_rx_queue, napi);
1792         struct vmxnet3_adapter *adapter = rq->adapter;
1793         int rxd_done;
1794
1795         /* When sharing interrupt with corresponding tx queue, process
1796          * tx completions in that queue as well
1797          */
1798         if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE) {
1799                 struct vmxnet3_tx_queue *tq =
1800                                 &adapter->tx_queue[rq - adapter->rx_queue];
1801                 vmxnet3_tq_tx_complete(tq, adapter);
1802         }
1803
1804         rxd_done = vmxnet3_rq_rx_complete(rq, adapter, budget);
1805
1806         if (rxd_done < budget) {
1807                 napi_complete(napi);
1808                 vmxnet3_enable_intr(adapter, rq->comp_ring.intr_idx);
1809         }
1810         return rxd_done;
1811 }
1812
1813
1814 #ifdef CONFIG_PCI_MSI
1815
1816 /*
1817  * Handle completion interrupts on tx queues
1818  * Returns whether or not the intr is handled
1819  */
1820
1821 static irqreturn_t
1822 vmxnet3_msix_tx(int irq, void *data)
1823 {
1824         struct vmxnet3_tx_queue *tq = data;
1825         struct vmxnet3_adapter *adapter = tq->adapter;
1826
1827         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1828                 vmxnet3_disable_intr(adapter, tq->comp_ring.intr_idx);
1829
1830         /* Handle the case where only one irq is allocate for all tx queues */
1831         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1832                 int i;
1833                 for (i = 0; i < adapter->num_tx_queues; i++) {
1834                         struct vmxnet3_tx_queue *txq = &adapter->tx_queue[i];
1835                         vmxnet3_tq_tx_complete(txq, adapter);
1836                 }
1837         } else {
1838                 vmxnet3_tq_tx_complete(tq, adapter);
1839         }
1840         vmxnet3_enable_intr(adapter, tq->comp_ring.intr_idx);
1841
1842         return IRQ_HANDLED;
1843 }
1844
1845
1846 /*
1847  * Handle completion interrupts on rx queues. Returns whether or not the
1848  * intr is handled
1849  */
1850
1851 static irqreturn_t
1852 vmxnet3_msix_rx(int irq, void *data)
1853 {
1854         struct vmxnet3_rx_queue *rq = data;
1855         struct vmxnet3_adapter *adapter = rq->adapter;
1856
1857         /* disable intr if needed */
1858         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1859                 vmxnet3_disable_intr(adapter, rq->comp_ring.intr_idx);
1860         napi_schedule(&rq->napi);
1861
1862         return IRQ_HANDLED;
1863 }
1864
1865 /*
1866  *----------------------------------------------------------------------------
1867  *
1868  * vmxnet3_msix_event --
1869  *
1870  *    vmxnet3 msix event intr handler
1871  *
1872  * Result:
1873  *    whether or not the intr is handled
1874  *
1875  *----------------------------------------------------------------------------
1876  */
1877
1878 static irqreturn_t
1879 vmxnet3_msix_event(int irq, void *data)
1880 {
1881         struct net_device *dev = data;
1882         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1883
1884         /* disable intr if needed */
1885         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1886                 vmxnet3_disable_intr(adapter, adapter->intr.event_intr_idx);
1887
1888         if (adapter->shared->ecr)
1889                 vmxnet3_process_events(adapter);
1890
1891         vmxnet3_enable_intr(adapter, adapter->intr.event_intr_idx);
1892
1893         return IRQ_HANDLED;
1894 }
1895
1896 #endif /* CONFIG_PCI_MSI  */
1897
1898
1899 /* Interrupt handler for vmxnet3  */
1900 static irqreturn_t
1901 vmxnet3_intr(int irq, void *dev_id)
1902 {
1903         struct net_device *dev = dev_id;
1904         struct vmxnet3_adapter *adapter = netdev_priv(dev);
1905
1906         if (adapter->intr.type == VMXNET3_IT_INTX) {
1907                 u32 icr = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_ICR);
1908                 if (unlikely(icr == 0))
1909                         /* not ours */
1910                         return IRQ_NONE;
1911         }
1912
1913
1914         /* disable intr if needed */
1915         if (adapter->intr.mask_mode == VMXNET3_IMM_ACTIVE)
1916                 vmxnet3_disable_all_intrs(adapter);
1917
1918         napi_schedule(&adapter->rx_queue[0].napi);
1919
1920         return IRQ_HANDLED;
1921 }
1922
1923 #ifdef CONFIG_NET_POLL_CONTROLLER
1924
1925 /* netpoll callback. */
1926 static void
1927 vmxnet3_netpoll(struct net_device *netdev)
1928 {
1929         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
1930
1931         switch (adapter->intr.type) {
1932 #ifdef CONFIG_PCI_MSI
1933         case VMXNET3_IT_MSIX: {
1934                 int i;
1935                 for (i = 0; i < adapter->num_rx_queues; i++)
1936                         vmxnet3_msix_rx(0, &adapter->rx_queue[i]);
1937                 break;
1938         }
1939 #endif
1940         case VMXNET3_IT_MSI:
1941         default:
1942                 vmxnet3_intr(0, adapter->netdev);
1943                 break;
1944         }
1945
1946 }
1947 #endif  /* CONFIG_NET_POLL_CONTROLLER */
1948
1949 static int
1950 vmxnet3_request_irqs(struct vmxnet3_adapter *adapter)
1951 {
1952         struct vmxnet3_intr *intr = &adapter->intr;
1953         int err = 0, i;
1954         int vector = 0;
1955
1956 #ifdef CONFIG_PCI_MSI
1957         if (adapter->intr.type == VMXNET3_IT_MSIX) {
1958                 for (i = 0; i < adapter->num_tx_queues; i++) {
1959                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
1960                                 sprintf(adapter->tx_queue[i].name, "%s-tx-%d",
1961                                         adapter->netdev->name, vector);
1962                                 err = request_irq(
1963                                               intr->msix_entries[vector].vector,
1964                                               vmxnet3_msix_tx, 0,
1965                                               adapter->tx_queue[i].name,
1966                                               &adapter->tx_queue[i]);
1967                         } else {
1968                                 sprintf(adapter->tx_queue[i].name, "%s-rxtx-%d",
1969                                         adapter->netdev->name, vector);
1970                         }
1971                         if (err) {
1972                                 dev_err(&adapter->netdev->dev,
1973                                         "Failed to request irq for MSIX, %s, "
1974                                         "error %d\n",
1975                                         adapter->tx_queue[i].name, err);
1976                                 return err;
1977                         }
1978
1979                         /* Handle the case where only 1 MSIx was allocated for
1980                          * all tx queues */
1981                         if (adapter->share_intr == VMXNET3_INTR_TXSHARE) {
1982                                 for (; i < adapter->num_tx_queues; i++)
1983                                         adapter->tx_queue[i].comp_ring.intr_idx
1984                                                                 = vector;
1985                                 vector++;
1986                                 break;
1987                         } else {
1988                                 adapter->tx_queue[i].comp_ring.intr_idx
1989                                                                 = vector++;
1990                         }
1991                 }
1992                 if (adapter->share_intr == VMXNET3_INTR_BUDDYSHARE)
1993                         vector = 0;
1994
1995                 for (i = 0; i < adapter->num_rx_queues; i++) {
1996                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE)
1997                                 sprintf(adapter->rx_queue[i].name, "%s-rx-%d",
1998                                         adapter->netdev->name, vector);
1999                         else
2000                                 sprintf(adapter->rx_queue[i].name, "%s-rxtx-%d",
2001                                         adapter->netdev->name, vector);
2002                         err = request_irq(intr->msix_entries[vector].vector,
2003                                           vmxnet3_msix_rx, 0,
2004                                           adapter->rx_queue[i].name,
2005                                           &(adapter->rx_queue[i]));
2006                         if (err) {
2007                                 netdev_err(adapter->netdev,
2008                                            "Failed to request irq for MSIX, "
2009                                            "%s, error %d\n",
2010                                            adapter->rx_queue[i].name, err);
2011                                 return err;
2012                         }
2013
2014                         adapter->rx_queue[i].comp_ring.intr_idx = vector++;
2015                 }
2016
2017                 sprintf(intr->event_msi_vector_name, "%s-event-%d",
2018                         adapter->netdev->name, vector);
2019                 err = request_irq(intr->msix_entries[vector].vector,
2020                                   vmxnet3_msix_event, 0,
2021                                   intr->event_msi_vector_name, adapter->netdev);
2022                 intr->event_intr_idx = vector;
2023
2024         } else if (intr->type == VMXNET3_IT_MSI) {
2025                 adapter->num_rx_queues = 1;
2026                 err = request_irq(adapter->pdev->irq, vmxnet3_intr, 0,
2027                                   adapter->netdev->name, adapter->netdev);
2028         } else {
2029 #endif
2030                 adapter->num_rx_queues = 1;
2031                 err = request_irq(adapter->pdev->irq, vmxnet3_intr,
2032                                   IRQF_SHARED, adapter->netdev->name,
2033                                   adapter->netdev);
2034 #ifdef CONFIG_PCI_MSI
2035         }
2036 #endif
2037         intr->num_intrs = vector + 1;
2038         if (err) {
2039                 netdev_err(adapter->netdev,
2040                            "Failed to request irq (intr type:%d), error %d\n",
2041                            intr->type, err);
2042         } else {
2043                 /* Number of rx queues will not change after this */
2044                 for (i = 0; i < adapter->num_rx_queues; i++) {
2045                         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2046                         rq->qid = i;
2047                         rq->qid2 = i + adapter->num_rx_queues;
2048                 }
2049
2050
2051
2052                 /* init our intr settings */
2053                 for (i = 0; i < intr->num_intrs; i++)
2054                         intr->mod_levels[i] = UPT1_IML_ADAPTIVE;
2055                 if (adapter->intr.type != VMXNET3_IT_MSIX) {
2056                         adapter->intr.event_intr_idx = 0;
2057                         for (i = 0; i < adapter->num_tx_queues; i++)
2058                                 adapter->tx_queue[i].comp_ring.intr_idx = 0;
2059                         adapter->rx_queue[0].comp_ring.intr_idx = 0;
2060                 }
2061
2062                 netdev_info(adapter->netdev,
2063                             "intr type %u, mode %u, %u vectors allocated\n",
2064                             intr->type, intr->mask_mode, intr->num_intrs);
2065         }
2066
2067         return err;
2068 }
2069
2070
2071 static void
2072 vmxnet3_free_irqs(struct vmxnet3_adapter *adapter)
2073 {
2074         struct vmxnet3_intr *intr = &adapter->intr;
2075         BUG_ON(intr->type == VMXNET3_IT_AUTO || intr->num_intrs <= 0);
2076
2077         switch (intr->type) {
2078 #ifdef CONFIG_PCI_MSI
2079         case VMXNET3_IT_MSIX:
2080         {
2081                 int i, vector = 0;
2082
2083                 if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE) {
2084                         for (i = 0; i < adapter->num_tx_queues; i++) {
2085                                 free_irq(intr->msix_entries[vector++].vector,
2086                                          &(adapter->tx_queue[i]));
2087                                 if (adapter->share_intr == VMXNET3_INTR_TXSHARE)
2088                                         break;
2089                         }
2090                 }
2091
2092                 for (i = 0; i < adapter->num_rx_queues; i++) {
2093                         free_irq(intr->msix_entries[vector++].vector,
2094                                  &(adapter->rx_queue[i]));
2095                 }
2096
2097                 free_irq(intr->msix_entries[vector].vector,
2098                          adapter->netdev);
2099                 BUG_ON(vector >= intr->num_intrs);
2100                 break;
2101         }
2102 #endif
2103         case VMXNET3_IT_MSI:
2104                 free_irq(adapter->pdev->irq, adapter->netdev);
2105                 break;
2106         case VMXNET3_IT_INTX:
2107                 free_irq(adapter->pdev->irq, adapter->netdev);
2108                 break;
2109         default:
2110                 BUG();
2111         }
2112 }
2113
2114
2115 static void
2116 vmxnet3_restore_vlan(struct vmxnet3_adapter *adapter)
2117 {
2118         u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2119         u16 vid;
2120
2121         /* allow untagged pkts */
2122         VMXNET3_SET_VFTABLE_ENTRY(vfTable, 0);
2123
2124         for_each_set_bit(vid, adapter->active_vlans, VLAN_N_VID)
2125                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2126 }
2127
2128
2129 static int
2130 vmxnet3_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
2131 {
2132         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2133
2134         if (!(netdev->flags & IFF_PROMISC)) {
2135                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2136                 unsigned long flags;
2137
2138                 VMXNET3_SET_VFTABLE_ENTRY(vfTable, vid);
2139                 spin_lock_irqsave(&adapter->cmd_lock, flags);
2140                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2141                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2142                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2143         }
2144
2145         set_bit(vid, adapter->active_vlans);
2146
2147         return 0;
2148 }
2149
2150
2151 static int
2152 vmxnet3_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
2153 {
2154         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2155
2156         if (!(netdev->flags & IFF_PROMISC)) {
2157                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2158                 unsigned long flags;
2159
2160                 VMXNET3_CLEAR_VFTABLE_ENTRY(vfTable, vid);
2161                 spin_lock_irqsave(&adapter->cmd_lock, flags);
2162                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2163                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2164                 spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2165         }
2166
2167         clear_bit(vid, adapter->active_vlans);
2168
2169         return 0;
2170 }
2171
2172
2173 static u8 *
2174 vmxnet3_copy_mc(struct net_device *netdev)
2175 {
2176         u8 *buf = NULL;
2177         u32 sz = netdev_mc_count(netdev) * ETH_ALEN;
2178
2179         /* struct Vmxnet3_RxFilterConf.mfTableLen is u16. */
2180         if (sz <= 0xffff) {
2181                 /* We may be called with BH disabled */
2182                 buf = kmalloc(sz, GFP_ATOMIC);
2183                 if (buf) {
2184                         struct netdev_hw_addr *ha;
2185                         int i = 0;
2186
2187                         netdev_for_each_mc_addr(ha, netdev)
2188                                 memcpy(buf + i++ * ETH_ALEN, ha->addr,
2189                                        ETH_ALEN);
2190                 }
2191         }
2192         return buf;
2193 }
2194
2195
2196 static void
2197 vmxnet3_set_mc(struct net_device *netdev)
2198 {
2199         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2200         unsigned long flags;
2201         struct Vmxnet3_RxFilterConf *rxConf =
2202                                         &adapter->shared->devRead.rxFilterConf;
2203         u8 *new_table = NULL;
2204         dma_addr_t new_table_pa = 0;
2205         u32 new_mode = VMXNET3_RXM_UCAST;
2206
2207         if (netdev->flags & IFF_PROMISC) {
2208                 u32 *vfTable = adapter->shared->devRead.rxFilterConf.vfTable;
2209                 memset(vfTable, 0, VMXNET3_VFT_SIZE * sizeof(*vfTable));
2210
2211                 new_mode |= VMXNET3_RXM_PROMISC;
2212         } else {
2213                 vmxnet3_restore_vlan(adapter);
2214         }
2215
2216         if (netdev->flags & IFF_BROADCAST)
2217                 new_mode |= VMXNET3_RXM_BCAST;
2218
2219         if (netdev->flags & IFF_ALLMULTI)
2220                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2221         else
2222                 if (!netdev_mc_empty(netdev)) {
2223                         new_table = vmxnet3_copy_mc(netdev);
2224                         if (new_table) {
2225                                 size_t sz = netdev_mc_count(netdev) * ETH_ALEN;
2226
2227                                 rxConf->mfTableLen = cpu_to_le16(sz);
2228                                 new_table_pa = dma_map_single(
2229                                                         &adapter->pdev->dev,
2230                                                         new_table,
2231                                                         sz,
2232                                                         PCI_DMA_TODEVICE);
2233                         }
2234
2235                         if (!dma_mapping_error(&adapter->pdev->dev,
2236                                                new_table_pa)) {
2237                                 new_mode |= VMXNET3_RXM_MCAST;
2238                                 rxConf->mfTablePA = cpu_to_le64(new_table_pa);
2239                         } else {
2240                                 netdev_info(netdev,
2241                                             "failed to copy mcast list, setting ALL_MULTI\n");
2242                                 new_mode |= VMXNET3_RXM_ALL_MULTI;
2243                         }
2244                 }
2245
2246         if (!(new_mode & VMXNET3_RXM_MCAST)) {
2247                 rxConf->mfTableLen = 0;
2248                 rxConf->mfTablePA = 0;
2249         }
2250
2251         spin_lock_irqsave(&adapter->cmd_lock, flags);
2252         if (new_mode != rxConf->rxMode) {
2253                 rxConf->rxMode = cpu_to_le32(new_mode);
2254                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2255                                        VMXNET3_CMD_UPDATE_RX_MODE);
2256                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2257                                        VMXNET3_CMD_UPDATE_VLAN_FILTERS);
2258         }
2259
2260         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2261                                VMXNET3_CMD_UPDATE_MAC_FILTERS);
2262         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2263
2264         if (new_table_pa)
2265                 dma_unmap_single(&adapter->pdev->dev, new_table_pa,
2266                                  rxConf->mfTableLen, PCI_DMA_TODEVICE);
2267         kfree(new_table);
2268 }
2269
2270 void
2271 vmxnet3_rq_destroy_all(struct vmxnet3_adapter *adapter)
2272 {
2273         int i;
2274
2275         for (i = 0; i < adapter->num_rx_queues; i++)
2276                 vmxnet3_rq_destroy(&adapter->rx_queue[i], adapter);
2277 }
2278
2279
2280 /*
2281  *   Set up driver_shared based on settings in adapter.
2282  */
2283
2284 static void
2285 vmxnet3_setup_driver_shared(struct vmxnet3_adapter *adapter)
2286 {
2287         struct Vmxnet3_DriverShared *shared = adapter->shared;
2288         struct Vmxnet3_DSDevRead *devRead = &shared->devRead;
2289         struct Vmxnet3_TxQueueConf *tqc;
2290         struct Vmxnet3_RxQueueConf *rqc;
2291         int i;
2292
2293         memset(shared, 0, sizeof(*shared));
2294
2295         /* driver settings */
2296         shared->magic = cpu_to_le32(VMXNET3_REV1_MAGIC);
2297         devRead->misc.driverInfo.version = cpu_to_le32(
2298                                                 VMXNET3_DRIVER_VERSION_NUM);
2299         devRead->misc.driverInfo.gos.gosBits = (sizeof(void *) == 4 ?
2300                                 VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64);
2301         devRead->misc.driverInfo.gos.gosType = VMXNET3_GOS_TYPE_LINUX;
2302         *((u32 *)&devRead->misc.driverInfo.gos) = cpu_to_le32(
2303                                 *((u32 *)&devRead->misc.driverInfo.gos));
2304         devRead->misc.driverInfo.vmxnet3RevSpt = cpu_to_le32(1);
2305         devRead->misc.driverInfo.uptVerSpt = cpu_to_le32(1);
2306
2307         devRead->misc.ddPA = cpu_to_le64(adapter->adapter_pa);
2308         devRead->misc.ddLen = cpu_to_le32(sizeof(struct vmxnet3_adapter));
2309
2310         /* set up feature flags */
2311         if (adapter->netdev->features & NETIF_F_RXCSUM)
2312                 devRead->misc.uptFeatures |= UPT1_F_RXCSUM;
2313
2314         if (adapter->netdev->features & NETIF_F_LRO) {
2315                 devRead->misc.uptFeatures |= UPT1_F_LRO;
2316                 devRead->misc.maxNumRxSG = cpu_to_le16(1 + MAX_SKB_FRAGS);
2317         }
2318         if (adapter->netdev->features & NETIF_F_HW_VLAN_CTAG_RX)
2319                 devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
2320
2321         devRead->misc.mtu = cpu_to_le32(adapter->netdev->mtu);
2322         devRead->misc.queueDescPA = cpu_to_le64(adapter->queue_desc_pa);
2323         devRead->misc.queueDescLen = cpu_to_le32(
2324                 adapter->num_tx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
2325                 adapter->num_rx_queues * sizeof(struct Vmxnet3_RxQueueDesc));
2326
2327         /* tx queue settings */
2328         devRead->misc.numTxQueues =  adapter->num_tx_queues;
2329         for (i = 0; i < adapter->num_tx_queues; i++) {
2330                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2331                 BUG_ON(adapter->tx_queue[i].tx_ring.base == NULL);
2332                 tqc = &adapter->tqd_start[i].conf;
2333                 tqc->txRingBasePA   = cpu_to_le64(tq->tx_ring.basePA);
2334                 tqc->dataRingBasePA = cpu_to_le64(tq->data_ring.basePA);
2335                 tqc->compRingBasePA = cpu_to_le64(tq->comp_ring.basePA);
2336                 tqc->ddPA           = cpu_to_le64(tq->buf_info_pa);
2337                 tqc->txRingSize     = cpu_to_le32(tq->tx_ring.size);
2338                 tqc->dataRingSize   = cpu_to_le32(tq->data_ring.size);
2339                 tqc->compRingSize   = cpu_to_le32(tq->comp_ring.size);
2340                 tqc->ddLen          = cpu_to_le32(
2341                                         sizeof(struct vmxnet3_tx_buf_info) *
2342                                         tqc->txRingSize);
2343                 tqc->intrIdx        = tq->comp_ring.intr_idx;
2344         }
2345
2346         /* rx queue settings */
2347         devRead->misc.numRxQueues = adapter->num_rx_queues;
2348         for (i = 0; i < adapter->num_rx_queues; i++) {
2349                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2350                 rqc = &adapter->rqd_start[i].conf;
2351                 rqc->rxRingBasePA[0] = cpu_to_le64(rq->rx_ring[0].basePA);
2352                 rqc->rxRingBasePA[1] = cpu_to_le64(rq->rx_ring[1].basePA);
2353                 rqc->compRingBasePA  = cpu_to_le64(rq->comp_ring.basePA);
2354                 rqc->ddPA            = cpu_to_le64(rq->buf_info_pa);
2355                 rqc->rxRingSize[0]   = cpu_to_le32(rq->rx_ring[0].size);
2356                 rqc->rxRingSize[1]   = cpu_to_le32(rq->rx_ring[1].size);
2357                 rqc->compRingSize    = cpu_to_le32(rq->comp_ring.size);
2358                 rqc->ddLen           = cpu_to_le32(
2359                                         sizeof(struct vmxnet3_rx_buf_info) *
2360                                         (rqc->rxRingSize[0] +
2361                                          rqc->rxRingSize[1]));
2362                 rqc->intrIdx         = rq->comp_ring.intr_idx;
2363         }
2364
2365 #ifdef VMXNET3_RSS
2366         memset(adapter->rss_conf, 0, sizeof(*adapter->rss_conf));
2367
2368         if (adapter->rss) {
2369                 struct UPT1_RSSConf *rssConf = adapter->rss_conf;
2370
2371                 devRead->misc.uptFeatures |= UPT1_F_RSS;
2372                 devRead->misc.numRxQueues = adapter->num_rx_queues;
2373                 rssConf->hashType = UPT1_RSS_HASH_TYPE_TCP_IPV4 |
2374                                     UPT1_RSS_HASH_TYPE_IPV4 |
2375                                     UPT1_RSS_HASH_TYPE_TCP_IPV6 |
2376                                     UPT1_RSS_HASH_TYPE_IPV6;
2377                 rssConf->hashFunc = UPT1_RSS_HASH_FUNC_TOEPLITZ;
2378                 rssConf->hashKeySize = UPT1_RSS_MAX_KEY_SIZE;
2379                 rssConf->indTableSize = VMXNET3_RSS_IND_TABLE_SIZE;
2380                 netdev_rss_key_fill(rssConf->hashKey, sizeof(rssConf->hashKey));
2381
2382                 for (i = 0; i < rssConf->indTableSize; i++)
2383                         rssConf->indTable[i] = ethtool_rxfh_indir_default(
2384                                 i, adapter->num_rx_queues);
2385
2386                 devRead->rssConfDesc.confVer = 1;
2387                 devRead->rssConfDesc.confLen = cpu_to_le32(sizeof(*rssConf));
2388                 devRead->rssConfDesc.confPA =
2389                         cpu_to_le64(adapter->rss_conf_pa);
2390         }
2391
2392 #endif /* VMXNET3_RSS */
2393
2394         /* intr settings */
2395         devRead->intrConf.autoMask = adapter->intr.mask_mode ==
2396                                      VMXNET3_IMM_AUTO;
2397         devRead->intrConf.numIntrs = adapter->intr.num_intrs;
2398         for (i = 0; i < adapter->intr.num_intrs; i++)
2399                 devRead->intrConf.modLevels[i] = adapter->intr.mod_levels[i];
2400
2401         devRead->intrConf.eventIntrIdx = adapter->intr.event_intr_idx;
2402         devRead->intrConf.intrCtrl |= cpu_to_le32(VMXNET3_IC_DISABLE_ALL);
2403
2404         /* rx filter settings */
2405         devRead->rxFilterConf.rxMode = 0;
2406         vmxnet3_restore_vlan(adapter);
2407         vmxnet3_write_mac_addr(adapter, adapter->netdev->dev_addr);
2408
2409         /* the rest are already zeroed */
2410 }
2411
2412
2413 int
2414 vmxnet3_activate_dev(struct vmxnet3_adapter *adapter)
2415 {
2416         int err, i;
2417         u32 ret;
2418         unsigned long flags;
2419
2420         netdev_dbg(adapter->netdev, "%s: skb_buf_size %d, rx_buf_per_pkt %d,"
2421                 " ring sizes %u %u %u\n", adapter->netdev->name,
2422                 adapter->skb_buf_size, adapter->rx_buf_per_pkt,
2423                 adapter->tx_queue[0].tx_ring.size,
2424                 adapter->rx_queue[0].rx_ring[0].size,
2425                 adapter->rx_queue[0].rx_ring[1].size);
2426
2427         vmxnet3_tq_init_all(adapter);
2428         err = vmxnet3_rq_init_all(adapter);
2429         if (err) {
2430                 netdev_err(adapter->netdev,
2431                            "Failed to init rx queue error %d\n", err);
2432                 goto rq_err;
2433         }
2434
2435         err = vmxnet3_request_irqs(adapter);
2436         if (err) {
2437                 netdev_err(adapter->netdev,
2438                            "Failed to setup irq for error %d\n", err);
2439                 goto irq_err;
2440         }
2441
2442         vmxnet3_setup_driver_shared(adapter);
2443
2444         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, VMXNET3_GET_ADDR_LO(
2445                                adapter->shared_pa));
2446         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, VMXNET3_GET_ADDR_HI(
2447                                adapter->shared_pa));
2448         spin_lock_irqsave(&adapter->cmd_lock, flags);
2449         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2450                                VMXNET3_CMD_ACTIVATE_DEV);
2451         ret = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2452         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2453
2454         if (ret != 0) {
2455                 netdev_err(adapter->netdev,
2456                            "Failed to activate dev: error %u\n", ret);
2457                 err = -EINVAL;
2458                 goto activate_err;
2459         }
2460
2461         for (i = 0; i < adapter->num_rx_queues; i++) {
2462                 VMXNET3_WRITE_BAR0_REG(adapter,
2463                                 VMXNET3_REG_RXPROD + i * VMXNET3_REG_ALIGN,
2464                                 adapter->rx_queue[i].rx_ring[0].next2fill);
2465                 VMXNET3_WRITE_BAR0_REG(adapter, (VMXNET3_REG_RXPROD2 +
2466                                 (i * VMXNET3_REG_ALIGN)),
2467                                 adapter->rx_queue[i].rx_ring[1].next2fill);
2468         }
2469
2470         /* Apply the rx filter settins last. */
2471         vmxnet3_set_mc(adapter->netdev);
2472
2473         /*
2474          * Check link state when first activating device. It will start the
2475          * tx queue if the link is up.
2476          */
2477         vmxnet3_check_link(adapter, true);
2478         for (i = 0; i < adapter->num_rx_queues; i++)
2479                 napi_enable(&adapter->rx_queue[i].napi);
2480         vmxnet3_enable_all_intrs(adapter);
2481         clear_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
2482         return 0;
2483
2484 activate_err:
2485         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAL, 0);
2486         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_DSAH, 0);
2487         vmxnet3_free_irqs(adapter);
2488 irq_err:
2489 rq_err:
2490         /* free up buffers we allocated */
2491         vmxnet3_rq_cleanup_all(adapter);
2492         return err;
2493 }
2494
2495
2496 void
2497 vmxnet3_reset_dev(struct vmxnet3_adapter *adapter)
2498 {
2499         unsigned long flags;
2500         spin_lock_irqsave(&adapter->cmd_lock, flags);
2501         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
2502         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2503 }
2504
2505
2506 int
2507 vmxnet3_quiesce_dev(struct vmxnet3_adapter *adapter)
2508 {
2509         int i;
2510         unsigned long flags;
2511         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state))
2512                 return 0;
2513
2514
2515         spin_lock_irqsave(&adapter->cmd_lock, flags);
2516         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2517                                VMXNET3_CMD_QUIESCE_DEV);
2518         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2519         vmxnet3_disable_all_intrs(adapter);
2520
2521         for (i = 0; i < adapter->num_rx_queues; i++)
2522                 napi_disable(&adapter->rx_queue[i].napi);
2523         netif_tx_disable(adapter->netdev);
2524         adapter->link_speed = 0;
2525         netif_carrier_off(adapter->netdev);
2526
2527         vmxnet3_tq_cleanup_all(adapter);
2528         vmxnet3_rq_cleanup_all(adapter);
2529         vmxnet3_free_irqs(adapter);
2530         return 0;
2531 }
2532
2533
2534 static void
2535 vmxnet3_write_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2536 {
2537         u32 tmp;
2538
2539         tmp = *(u32 *)mac;
2540         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACL, tmp);
2541
2542         tmp = (mac[5] << 8) | mac[4];
2543         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_MACH, tmp);
2544 }
2545
2546
2547 static int
2548 vmxnet3_set_mac_addr(struct net_device *netdev, void *p)
2549 {
2550         struct sockaddr *addr = p;
2551         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2552
2553         memcpy(netdev->dev_addr, addr->sa_data, netdev->addr_len);
2554         vmxnet3_write_mac_addr(adapter, addr->sa_data);
2555
2556         return 0;
2557 }
2558
2559
2560 /* ==================== initialization and cleanup routines ============ */
2561
2562 static int
2563 vmxnet3_alloc_pci_resources(struct vmxnet3_adapter *adapter, bool *dma64)
2564 {
2565         int err;
2566         unsigned long mmio_start, mmio_len;
2567         struct pci_dev *pdev = adapter->pdev;
2568
2569         err = pci_enable_device(pdev);
2570         if (err) {
2571                 dev_err(&pdev->dev, "Failed to enable adapter: error %d\n", err);
2572                 return err;
2573         }
2574
2575         if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
2576                 if (pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)) != 0) {
2577                         dev_err(&pdev->dev,
2578                                 "pci_set_consistent_dma_mask failed\n");
2579                         err = -EIO;
2580                         goto err_set_mask;
2581                 }
2582                 *dma64 = true;
2583         } else {
2584                 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(32)) != 0) {
2585                         dev_err(&pdev->dev,
2586                                 "pci_set_dma_mask failed\n");
2587                         err = -EIO;
2588                         goto err_set_mask;
2589                 }
2590                 *dma64 = false;
2591         }
2592
2593         err = pci_request_selected_regions(pdev, (1 << 2) - 1,
2594                                            vmxnet3_driver_name);
2595         if (err) {
2596                 dev_err(&pdev->dev,
2597                         "Failed to request region for adapter: error %d\n", err);
2598                 goto err_set_mask;
2599         }
2600
2601         pci_set_master(pdev);
2602
2603         mmio_start = pci_resource_start(pdev, 0);
2604         mmio_len = pci_resource_len(pdev, 0);
2605         adapter->hw_addr0 = ioremap(mmio_start, mmio_len);
2606         if (!adapter->hw_addr0) {
2607                 dev_err(&pdev->dev, "Failed to map bar0\n");
2608                 err = -EIO;
2609                 goto err_ioremap;
2610         }
2611
2612         mmio_start = pci_resource_start(pdev, 1);
2613         mmio_len = pci_resource_len(pdev, 1);
2614         adapter->hw_addr1 = ioremap(mmio_start, mmio_len);
2615         if (!adapter->hw_addr1) {
2616                 dev_err(&pdev->dev, "Failed to map bar1\n");
2617                 err = -EIO;
2618                 goto err_bar1;
2619         }
2620         return 0;
2621
2622 err_bar1:
2623         iounmap(adapter->hw_addr0);
2624 err_ioremap:
2625         pci_release_selected_regions(pdev, (1 << 2) - 1);
2626 err_set_mask:
2627         pci_disable_device(pdev);
2628         return err;
2629 }
2630
2631
2632 static void
2633 vmxnet3_free_pci_resources(struct vmxnet3_adapter *adapter)
2634 {
2635         BUG_ON(!adapter->pdev);
2636
2637         iounmap(adapter->hw_addr0);
2638         iounmap(adapter->hw_addr1);
2639         pci_release_selected_regions(adapter->pdev, (1 << 2) - 1);
2640         pci_disable_device(adapter->pdev);
2641 }
2642
2643
2644 static void
2645 vmxnet3_adjust_rx_ring_size(struct vmxnet3_adapter *adapter)
2646 {
2647         size_t sz, i, ring0_size, ring1_size, comp_size;
2648         struct vmxnet3_rx_queue *rq = &adapter->rx_queue[0];
2649
2650
2651         if (adapter->netdev->mtu <= VMXNET3_MAX_SKB_BUF_SIZE -
2652                                     VMXNET3_MAX_ETH_HDR_SIZE) {
2653                 adapter->skb_buf_size = adapter->netdev->mtu +
2654                                         VMXNET3_MAX_ETH_HDR_SIZE;
2655                 if (adapter->skb_buf_size < VMXNET3_MIN_T0_BUF_SIZE)
2656                         adapter->skb_buf_size = VMXNET3_MIN_T0_BUF_SIZE;
2657
2658                 adapter->rx_buf_per_pkt = 1;
2659         } else {
2660                 adapter->skb_buf_size = VMXNET3_MAX_SKB_BUF_SIZE;
2661                 sz = adapter->netdev->mtu - VMXNET3_MAX_SKB_BUF_SIZE +
2662                                             VMXNET3_MAX_ETH_HDR_SIZE;
2663                 adapter->rx_buf_per_pkt = 1 + (sz + PAGE_SIZE - 1) / PAGE_SIZE;
2664         }
2665
2666         /*
2667          * for simplicity, force the ring0 size to be a multiple of
2668          * rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN
2669          */
2670         sz = adapter->rx_buf_per_pkt * VMXNET3_RING_SIZE_ALIGN;
2671         ring0_size = adapter->rx_queue[0].rx_ring[0].size;
2672         ring0_size = (ring0_size + sz - 1) / sz * sz;
2673         ring0_size = min_t(u32, ring0_size, VMXNET3_RX_RING_MAX_SIZE /
2674                            sz * sz);
2675         ring1_size = adapter->rx_queue[0].rx_ring[1].size;
2676         ring1_size = (ring1_size + sz - 1) / sz * sz;
2677         ring1_size = min_t(u32, ring1_size, VMXNET3_RX_RING2_MAX_SIZE /
2678                            sz * sz);
2679         comp_size = ring0_size + ring1_size;
2680
2681         for (i = 0; i < adapter->num_rx_queues; i++) {
2682                 rq = &adapter->rx_queue[i];
2683                 rq->rx_ring[0].size = ring0_size;
2684                 rq->rx_ring[1].size = ring1_size;
2685                 rq->comp_ring.size = comp_size;
2686         }
2687 }
2688
2689
2690 int
2691 vmxnet3_create_queues(struct vmxnet3_adapter *adapter, u32 tx_ring_size,
2692                       u32 rx_ring_size, u32 rx_ring2_size)
2693 {
2694         int err = 0, i;
2695
2696         for (i = 0; i < adapter->num_tx_queues; i++) {
2697                 struct vmxnet3_tx_queue *tq = &adapter->tx_queue[i];
2698                 tq->tx_ring.size   = tx_ring_size;
2699                 tq->data_ring.size = tx_ring_size;
2700                 tq->comp_ring.size = tx_ring_size;
2701                 tq->shared = &adapter->tqd_start[i].ctrl;
2702                 tq->stopped = true;
2703                 tq->adapter = adapter;
2704                 tq->qid = i;
2705                 err = vmxnet3_tq_create(tq, adapter);
2706                 /*
2707                  * Too late to change num_tx_queues. We cannot do away with
2708                  * lesser number of queues than what we asked for
2709                  */
2710                 if (err)
2711                         goto queue_err;
2712         }
2713
2714         adapter->rx_queue[0].rx_ring[0].size = rx_ring_size;
2715         adapter->rx_queue[0].rx_ring[1].size = rx_ring2_size;
2716         vmxnet3_adjust_rx_ring_size(adapter);
2717         for (i = 0; i < adapter->num_rx_queues; i++) {
2718                 struct vmxnet3_rx_queue *rq = &adapter->rx_queue[i];
2719                 /* qid and qid2 for rx queues will be assigned later when num
2720                  * of rx queues is finalized after allocating intrs */
2721                 rq->shared = &adapter->rqd_start[i].ctrl;
2722                 rq->adapter = adapter;
2723                 err = vmxnet3_rq_create(rq, adapter);
2724                 if (err) {
2725                         if (i == 0) {
2726                                 netdev_err(adapter->netdev,
2727                                            "Could not allocate any rx queues. "
2728                                            "Aborting.\n");
2729                                 goto queue_err;
2730                         } else {
2731                                 netdev_info(adapter->netdev,
2732                                             "Number of rx queues changed "
2733                                             "to : %d.\n", i);
2734                                 adapter->num_rx_queues = i;
2735                                 err = 0;
2736                                 break;
2737                         }
2738                 }
2739         }
2740         return err;
2741 queue_err:
2742         vmxnet3_tq_destroy_all(adapter);
2743         return err;
2744 }
2745
2746 static int
2747 vmxnet3_open(struct net_device *netdev)
2748 {
2749         struct vmxnet3_adapter *adapter;
2750         int err, i;
2751
2752         adapter = netdev_priv(netdev);
2753
2754         for (i = 0; i < adapter->num_tx_queues; i++)
2755                 spin_lock_init(&adapter->tx_queue[i].tx_lock);
2756
2757         err = vmxnet3_create_queues(adapter, adapter->tx_ring_size,
2758                                     adapter->rx_ring_size,
2759                                     adapter->rx_ring2_size);
2760         if (err)
2761                 goto queue_err;
2762
2763         err = vmxnet3_activate_dev(adapter);
2764         if (err)
2765                 goto activate_err;
2766
2767         return 0;
2768
2769 activate_err:
2770         vmxnet3_rq_destroy_all(adapter);
2771         vmxnet3_tq_destroy_all(adapter);
2772 queue_err:
2773         return err;
2774 }
2775
2776
2777 static int
2778 vmxnet3_close(struct net_device *netdev)
2779 {
2780         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2781
2782         /*
2783          * Reset_work may be in the middle of resetting the device, wait for its
2784          * completion.
2785          */
2786         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2787                 msleep(1);
2788
2789         vmxnet3_quiesce_dev(adapter);
2790
2791         vmxnet3_rq_destroy_all(adapter);
2792         vmxnet3_tq_destroy_all(adapter);
2793
2794         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2795
2796
2797         return 0;
2798 }
2799
2800
2801 void
2802 vmxnet3_force_close(struct vmxnet3_adapter *adapter)
2803 {
2804         int i;
2805
2806         /*
2807          * we must clear VMXNET3_STATE_BIT_RESETTING, otherwise
2808          * vmxnet3_close() will deadlock.
2809          */
2810         BUG_ON(test_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state));
2811
2812         /* we need to enable NAPI, otherwise dev_close will deadlock */
2813         for (i = 0; i < adapter->num_rx_queues; i++)
2814                 napi_enable(&adapter->rx_queue[i].napi);
2815         dev_close(adapter->netdev);
2816 }
2817
2818
2819 static int
2820 vmxnet3_change_mtu(struct net_device *netdev, int new_mtu)
2821 {
2822         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
2823         int err = 0;
2824
2825         if (new_mtu < VMXNET3_MIN_MTU || new_mtu > VMXNET3_MAX_MTU)
2826                 return -EINVAL;
2827
2828         netdev->mtu = new_mtu;
2829
2830         /*
2831          * Reset_work may be in the middle of resetting the device, wait for its
2832          * completion.
2833          */
2834         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
2835                 msleep(1);
2836
2837         if (netif_running(netdev)) {
2838                 vmxnet3_quiesce_dev(adapter);
2839                 vmxnet3_reset_dev(adapter);
2840
2841                 /* we need to re-create the rx queue based on the new mtu */
2842                 vmxnet3_rq_destroy_all(adapter);
2843                 vmxnet3_adjust_rx_ring_size(adapter);
2844                 err = vmxnet3_rq_create_all(adapter);
2845                 if (err) {
2846                         netdev_err(netdev,
2847                                    "failed to re-create rx queues, "
2848                                    " error %d. Closing it.\n", err);
2849                         goto out;
2850                 }
2851
2852                 err = vmxnet3_activate_dev(adapter);
2853                 if (err) {
2854                         netdev_err(netdev,
2855                                    "failed to re-activate, error %d. "
2856                                    "Closing it\n", err);
2857                         goto out;
2858                 }
2859         }
2860
2861 out:
2862         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
2863         if (err)
2864                 vmxnet3_force_close(adapter);
2865
2866         return err;
2867 }
2868
2869
2870 static void
2871 vmxnet3_declare_features(struct vmxnet3_adapter *adapter, bool dma64)
2872 {
2873         struct net_device *netdev = adapter->netdev;
2874
2875         netdev->hw_features = NETIF_F_SG | NETIF_F_RXCSUM |
2876                 NETIF_F_HW_CSUM | NETIF_F_HW_VLAN_CTAG_TX |
2877                 NETIF_F_HW_VLAN_CTAG_RX | NETIF_F_TSO | NETIF_F_TSO6 |
2878                 NETIF_F_LRO;
2879         if (dma64)
2880                 netdev->hw_features |= NETIF_F_HIGHDMA;
2881         netdev->vlan_features = netdev->hw_features &
2882                                 ~(NETIF_F_HW_VLAN_CTAG_TX |
2883                                   NETIF_F_HW_VLAN_CTAG_RX);
2884         netdev->features = netdev->hw_features | NETIF_F_HW_VLAN_CTAG_FILTER;
2885 }
2886
2887
2888 static void
2889 vmxnet3_read_mac_addr(struct vmxnet3_adapter *adapter, u8 *mac)
2890 {
2891         u32 tmp;
2892
2893         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACL);
2894         *(u32 *)mac = tmp;
2895
2896         tmp = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_MACH);
2897         mac[4] = tmp & 0xff;
2898         mac[5] = (tmp >> 8) & 0xff;
2899 }
2900
2901 #ifdef CONFIG_PCI_MSI
2902
2903 /*
2904  * Enable MSIx vectors.
2905  * Returns :
2906  *      VMXNET3_LINUX_MIN_MSIX_VECT when only minimum number of vectors required
2907  *       were enabled.
2908  *      number of vectors which were enabled otherwise (this number is greater
2909  *       than VMXNET3_LINUX_MIN_MSIX_VECT)
2910  */
2911
2912 static int
2913 vmxnet3_acquire_msix_vectors(struct vmxnet3_adapter *adapter, int nvec)
2914 {
2915         int ret = pci_enable_msix_range(adapter->pdev,
2916                                         adapter->intr.msix_entries, nvec, nvec);
2917
2918         if (ret == -ENOSPC && nvec > VMXNET3_LINUX_MIN_MSIX_VECT) {
2919                 dev_err(&adapter->netdev->dev,
2920                         "Failed to enable %d MSI-X, trying %d\n",
2921                         nvec, VMXNET3_LINUX_MIN_MSIX_VECT);
2922
2923                 ret = pci_enable_msix_range(adapter->pdev,
2924                                             adapter->intr.msix_entries,
2925                                             VMXNET3_LINUX_MIN_MSIX_VECT,
2926                                             VMXNET3_LINUX_MIN_MSIX_VECT);
2927         }
2928
2929         if (ret < 0) {
2930                 dev_err(&adapter->netdev->dev,
2931                         "Failed to enable MSI-X, error: %d\n", ret);
2932         }
2933
2934         return ret;
2935 }
2936
2937
2938 #endif /* CONFIG_PCI_MSI */
2939
2940 static void
2941 vmxnet3_alloc_intr_resources(struct vmxnet3_adapter *adapter)
2942 {
2943         u32 cfg;
2944         unsigned long flags;
2945
2946         /* intr settings */
2947         spin_lock_irqsave(&adapter->cmd_lock, flags);
2948         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
2949                                VMXNET3_CMD_GET_CONF_INTR);
2950         cfg = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_CMD);
2951         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
2952         adapter->intr.type = cfg & 0x3;
2953         adapter->intr.mask_mode = (cfg >> 2) & 0x3;
2954
2955         if (adapter->intr.type == VMXNET3_IT_AUTO) {
2956                 adapter->intr.type = VMXNET3_IT_MSIX;
2957         }
2958
2959 #ifdef CONFIG_PCI_MSI
2960         if (adapter->intr.type == VMXNET3_IT_MSIX) {
2961                 int i, nvec;
2962
2963                 nvec  = adapter->share_intr == VMXNET3_INTR_TXSHARE ?
2964                         1 : adapter->num_tx_queues;
2965                 nvec += adapter->share_intr == VMXNET3_INTR_BUDDYSHARE ?
2966                         0 : adapter->num_rx_queues;
2967                 nvec += 1;      /* for link event */
2968                 nvec = nvec > VMXNET3_LINUX_MIN_MSIX_VECT ?
2969                        nvec : VMXNET3_LINUX_MIN_MSIX_VECT;
2970
2971                 for (i = 0; i < nvec; i++)
2972                         adapter->intr.msix_entries[i].entry = i;
2973
2974                 nvec = vmxnet3_acquire_msix_vectors(adapter, nvec);
2975                 if (nvec < 0)
2976                         goto msix_err;
2977
2978                 /* If we cannot allocate one MSIx vector per queue
2979                  * then limit the number of rx queues to 1
2980                  */
2981                 if (nvec == VMXNET3_LINUX_MIN_MSIX_VECT) {
2982                         if (adapter->share_intr != VMXNET3_INTR_BUDDYSHARE
2983                             || adapter->num_rx_queues != 1) {
2984                                 adapter->share_intr = VMXNET3_INTR_TXSHARE;
2985                                 netdev_err(adapter->netdev,
2986                                            "Number of rx queues : 1\n");
2987                                 adapter->num_rx_queues = 1;
2988                         }
2989                 }
2990
2991                 adapter->intr.num_intrs = nvec;
2992                 return;
2993
2994 msix_err:
2995                 /* If we cannot allocate MSIx vectors use only one rx queue */
2996                 dev_info(&adapter->pdev->dev,
2997                          "Failed to enable MSI-X, error %d. "
2998                          "Limiting #rx queues to 1, try MSI.\n", nvec);
2999
3000                 adapter->intr.type = VMXNET3_IT_MSI;
3001         }
3002
3003         if (adapter->intr.type == VMXNET3_IT_MSI) {
3004                 if (!pci_enable_msi(adapter->pdev)) {
3005                         adapter->num_rx_queues = 1;
3006                         adapter->intr.num_intrs = 1;
3007                         return;
3008                 }
3009         }
3010 #endif /* CONFIG_PCI_MSI */
3011
3012         adapter->num_rx_queues = 1;
3013         dev_info(&adapter->netdev->dev,
3014                  "Using INTx interrupt, #Rx queues: 1.\n");
3015         adapter->intr.type = VMXNET3_IT_INTX;
3016
3017         /* INT-X related setting */
3018         adapter->intr.num_intrs = 1;
3019 }
3020
3021
3022 static void
3023 vmxnet3_free_intr_resources(struct vmxnet3_adapter *adapter)
3024 {
3025         if (adapter->intr.type == VMXNET3_IT_MSIX)
3026                 pci_disable_msix(adapter->pdev);
3027         else if (adapter->intr.type == VMXNET3_IT_MSI)
3028                 pci_disable_msi(adapter->pdev);
3029         else
3030                 BUG_ON(adapter->intr.type != VMXNET3_IT_INTX);
3031 }
3032
3033
3034 static void
3035 vmxnet3_tx_timeout(struct net_device *netdev)
3036 {
3037         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3038         adapter->tx_timeout_count++;
3039
3040         netdev_err(adapter->netdev, "tx hang\n");
3041         schedule_work(&adapter->work);
3042         netif_wake_queue(adapter->netdev);
3043 }
3044
3045
3046 static void
3047 vmxnet3_reset_work(struct work_struct *data)
3048 {
3049         struct vmxnet3_adapter *adapter;
3050
3051         adapter = container_of(data, struct vmxnet3_adapter, work);
3052
3053         /* if another thread is resetting the device, no need to proceed */
3054         if (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3055                 return;
3056
3057         /* if the device is closed, we must leave it alone */
3058         rtnl_lock();
3059         if (netif_running(adapter->netdev)) {
3060                 netdev_notice(adapter->netdev, "resetting\n");
3061                 vmxnet3_quiesce_dev(adapter);
3062                 vmxnet3_reset_dev(adapter);
3063                 vmxnet3_activate_dev(adapter);
3064         } else {
3065                 netdev_info(adapter->netdev, "already closed\n");
3066         }
3067         rtnl_unlock();
3068
3069         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3070 }
3071
3072
3073 static int
3074 vmxnet3_probe_device(struct pci_dev *pdev,
3075                      const struct pci_device_id *id)
3076 {
3077         static const struct net_device_ops vmxnet3_netdev_ops = {
3078                 .ndo_open = vmxnet3_open,
3079                 .ndo_stop = vmxnet3_close,
3080                 .ndo_start_xmit = vmxnet3_xmit_frame,
3081                 .ndo_set_mac_address = vmxnet3_set_mac_addr,
3082                 .ndo_change_mtu = vmxnet3_change_mtu,
3083                 .ndo_set_features = vmxnet3_set_features,
3084                 .ndo_get_stats64 = vmxnet3_get_stats64,
3085                 .ndo_tx_timeout = vmxnet3_tx_timeout,
3086                 .ndo_set_rx_mode = vmxnet3_set_mc,
3087                 .ndo_vlan_rx_add_vid = vmxnet3_vlan_rx_add_vid,
3088                 .ndo_vlan_rx_kill_vid = vmxnet3_vlan_rx_kill_vid,
3089 #ifdef CONFIG_NET_POLL_CONTROLLER
3090                 .ndo_poll_controller = vmxnet3_netpoll,
3091 #endif
3092         };
3093         int err;
3094         bool dma64 = false; /* stupid gcc */
3095         u32 ver;
3096         struct net_device *netdev;
3097         struct vmxnet3_adapter *adapter;
3098         u8 mac[ETH_ALEN];
3099         int size;
3100         int num_tx_queues;
3101         int num_rx_queues;
3102
3103         if (!pci_msi_enabled())
3104                 enable_mq = 0;
3105
3106 #ifdef VMXNET3_RSS
3107         if (enable_mq)
3108                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3109                                     (int)num_online_cpus());
3110         else
3111 #endif
3112                 num_rx_queues = 1;
3113         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3114
3115         if (enable_mq)
3116                 num_tx_queues = min(VMXNET3_DEVICE_MAX_TX_QUEUES,
3117                                     (int)num_online_cpus());
3118         else
3119                 num_tx_queues = 1;
3120
3121         num_tx_queues = rounddown_pow_of_two(num_tx_queues);
3122         netdev = alloc_etherdev_mq(sizeof(struct vmxnet3_adapter),
3123                                    max(num_tx_queues, num_rx_queues));
3124         dev_info(&pdev->dev,
3125                  "# of Tx queues : %d, # of Rx queues : %d\n",
3126                  num_tx_queues, num_rx_queues);
3127
3128         if (!netdev)
3129                 return -ENOMEM;
3130
3131         pci_set_drvdata(pdev, netdev);
3132         adapter = netdev_priv(netdev);
3133         adapter->netdev = netdev;
3134         adapter->pdev = pdev;
3135
3136         adapter->tx_ring_size = VMXNET3_DEF_TX_RING_SIZE;
3137         adapter->rx_ring_size = VMXNET3_DEF_RX_RING_SIZE;
3138         adapter->rx_ring2_size = VMXNET3_DEF_RX_RING2_SIZE;
3139
3140         spin_lock_init(&adapter->cmd_lock);
3141         adapter->adapter_pa = dma_map_single(&adapter->pdev->dev, adapter,
3142                                              sizeof(struct vmxnet3_adapter),
3143                                              PCI_DMA_TODEVICE);
3144         if (dma_mapping_error(&adapter->pdev->dev, adapter->adapter_pa)) {
3145                 dev_err(&pdev->dev, "Failed to map dma\n");
3146                 err = -EFAULT;
3147                 goto err_dma_map;
3148         }
3149         adapter->shared = dma_alloc_coherent(
3150                                 &adapter->pdev->dev,
3151                                 sizeof(struct Vmxnet3_DriverShared),
3152                                 &adapter->shared_pa, GFP_KERNEL);
3153         if (!adapter->shared) {
3154                 dev_err(&pdev->dev, "Failed to allocate memory\n");
3155                 err = -ENOMEM;
3156                 goto err_alloc_shared;
3157         }
3158
3159         adapter->num_rx_queues = num_rx_queues;
3160         adapter->num_tx_queues = num_tx_queues;
3161         adapter->rx_buf_per_pkt = 1;
3162
3163         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3164         size += sizeof(struct Vmxnet3_RxQueueDesc) * adapter->num_rx_queues;
3165         adapter->tqd_start = dma_alloc_coherent(&adapter->pdev->dev, size,
3166                                                 &adapter->queue_desc_pa,
3167                                                 GFP_KERNEL);
3168
3169         if (!adapter->tqd_start) {
3170                 dev_err(&pdev->dev, "Failed to allocate memory\n");
3171                 err = -ENOMEM;
3172                 goto err_alloc_queue_desc;
3173         }
3174         adapter->rqd_start = (struct Vmxnet3_RxQueueDesc *)(adapter->tqd_start +
3175                                                             adapter->num_tx_queues);
3176
3177         adapter->pm_conf = dma_alloc_coherent(&adapter->pdev->dev,
3178                                               sizeof(struct Vmxnet3_PMConf),
3179                                               &adapter->pm_conf_pa,
3180                                               GFP_KERNEL);
3181         if (adapter->pm_conf == NULL) {
3182                 err = -ENOMEM;
3183                 goto err_alloc_pm;
3184         }
3185
3186 #ifdef VMXNET3_RSS
3187
3188         adapter->rss_conf = dma_alloc_coherent(&adapter->pdev->dev,
3189                                                sizeof(struct UPT1_RSSConf),
3190                                                &adapter->rss_conf_pa,
3191                                                GFP_KERNEL);
3192         if (adapter->rss_conf == NULL) {
3193                 err = -ENOMEM;
3194                 goto err_alloc_rss;
3195         }
3196 #endif /* VMXNET3_RSS */
3197
3198         err = vmxnet3_alloc_pci_resources(adapter, &dma64);
3199         if (err < 0)
3200                 goto err_alloc_pci;
3201
3202         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_VRRS);
3203         if (ver & 2) {
3204                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 2);
3205                 adapter->version = 2;
3206         } else if (ver & 1) {
3207                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_VRRS, 1);
3208                 adapter->version = 1;
3209         } else {
3210                 dev_err(&pdev->dev,
3211                         "Incompatible h/w version (0x%x) for adapter\n", ver);
3212                 err = -EBUSY;
3213                 goto err_ver;
3214         }
3215         dev_dbg(&pdev->dev, "Using device version %d\n", adapter->version);
3216
3217         ver = VMXNET3_READ_BAR1_REG(adapter, VMXNET3_REG_UVRS);
3218         if (ver & 1) {
3219                 VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_UVRS, 1);
3220         } else {
3221                 dev_err(&pdev->dev,
3222                         "Incompatible upt version (0x%x) for adapter\n", ver);
3223                 err = -EBUSY;
3224                 goto err_ver;
3225         }
3226
3227         SET_NETDEV_DEV(netdev, &pdev->dev);
3228         vmxnet3_declare_features(adapter, dma64);
3229
3230         if (adapter->num_tx_queues == adapter->num_rx_queues)
3231                 adapter->share_intr = VMXNET3_INTR_BUDDYSHARE;
3232         else
3233                 adapter->share_intr = VMXNET3_INTR_DONTSHARE;
3234
3235         vmxnet3_alloc_intr_resources(adapter);
3236
3237 #ifdef VMXNET3_RSS
3238         if (adapter->num_rx_queues > 1 &&
3239             adapter->intr.type == VMXNET3_IT_MSIX) {
3240                 adapter->rss = true;
3241                 netdev->hw_features |= NETIF_F_RXHASH;
3242                 netdev->features |= NETIF_F_RXHASH;
3243                 dev_dbg(&pdev->dev, "RSS is enabled.\n");
3244         } else {
3245                 adapter->rss = false;
3246         }
3247 #endif
3248
3249         vmxnet3_read_mac_addr(adapter, mac);
3250         memcpy(netdev->dev_addr,  mac, netdev->addr_len);
3251
3252         netdev->netdev_ops = &vmxnet3_netdev_ops;
3253         vmxnet3_set_ethtool_ops(netdev);
3254         netdev->watchdog_timeo = 5 * HZ;
3255
3256         INIT_WORK(&adapter->work, vmxnet3_reset_work);
3257         set_bit(VMXNET3_STATE_BIT_QUIESCED, &adapter->state);
3258
3259         if (adapter->intr.type == VMXNET3_IT_MSIX) {
3260                 int i;
3261                 for (i = 0; i < adapter->num_rx_queues; i++) {
3262                         netif_napi_add(adapter->netdev,
3263                                        &adapter->rx_queue[i].napi,
3264                                        vmxnet3_poll_rx_only, 64);
3265                 }
3266         } else {
3267                 netif_napi_add(adapter->netdev, &adapter->rx_queue[0].napi,
3268                                vmxnet3_poll, 64);
3269         }
3270
3271         netif_set_real_num_tx_queues(adapter->netdev, adapter->num_tx_queues);
3272         netif_set_real_num_rx_queues(adapter->netdev, adapter->num_rx_queues);
3273
3274         netif_carrier_off(netdev);
3275         err = register_netdev(netdev);
3276
3277         if (err) {
3278                 dev_err(&pdev->dev, "Failed to register adapter\n");
3279                 goto err_register;
3280         }
3281
3282         vmxnet3_check_link(adapter, false);
3283         return 0;
3284
3285 err_register:
3286         vmxnet3_free_intr_resources(adapter);
3287 err_ver:
3288         vmxnet3_free_pci_resources(adapter);
3289 err_alloc_pci:
3290 #ifdef VMXNET3_RSS
3291         dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3292                           adapter->rss_conf, adapter->rss_conf_pa);
3293 err_alloc_rss:
3294 #endif
3295         dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3296                           adapter->pm_conf, adapter->pm_conf_pa);
3297 err_alloc_pm:
3298         dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3299                           adapter->queue_desc_pa);
3300 err_alloc_queue_desc:
3301         dma_free_coherent(&adapter->pdev->dev,
3302                           sizeof(struct Vmxnet3_DriverShared),
3303                           adapter->shared, adapter->shared_pa);
3304 err_alloc_shared:
3305         dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3306                          sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3307 err_dma_map:
3308         free_netdev(netdev);
3309         return err;
3310 }
3311
3312
3313 static void
3314 vmxnet3_remove_device(struct pci_dev *pdev)
3315 {
3316         struct net_device *netdev = pci_get_drvdata(pdev);
3317         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3318         int size = 0;
3319         int num_rx_queues;
3320
3321 #ifdef VMXNET3_RSS
3322         if (enable_mq)
3323                 num_rx_queues = min(VMXNET3_DEVICE_MAX_RX_QUEUES,
3324                                     (int)num_online_cpus());
3325         else
3326 #endif
3327                 num_rx_queues = 1;
3328         num_rx_queues = rounddown_pow_of_two(num_rx_queues);
3329
3330         cancel_work_sync(&adapter->work);
3331
3332         unregister_netdev(netdev);
3333
3334         vmxnet3_free_intr_resources(adapter);
3335         vmxnet3_free_pci_resources(adapter);
3336 #ifdef VMXNET3_RSS
3337         dma_free_coherent(&adapter->pdev->dev, sizeof(struct UPT1_RSSConf),
3338                           adapter->rss_conf, adapter->rss_conf_pa);
3339 #endif
3340         dma_free_coherent(&adapter->pdev->dev, sizeof(struct Vmxnet3_PMConf),
3341                           adapter->pm_conf, adapter->pm_conf_pa);
3342
3343         size = sizeof(struct Vmxnet3_TxQueueDesc) * adapter->num_tx_queues;
3344         size += sizeof(struct Vmxnet3_RxQueueDesc) * num_rx_queues;
3345         dma_free_coherent(&adapter->pdev->dev, size, adapter->tqd_start,
3346                           adapter->queue_desc_pa);
3347         dma_free_coherent(&adapter->pdev->dev,
3348                           sizeof(struct Vmxnet3_DriverShared),
3349                           adapter->shared, adapter->shared_pa);
3350         dma_unmap_single(&adapter->pdev->dev, adapter->adapter_pa,
3351                          sizeof(struct vmxnet3_adapter), PCI_DMA_TODEVICE);
3352         free_netdev(netdev);
3353 }
3354
3355 static void vmxnet3_shutdown_device(struct pci_dev *pdev)
3356 {
3357         struct net_device *netdev = pci_get_drvdata(pdev);
3358         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3359         unsigned long flags;
3360
3361         /* Reset_work may be in the middle of resetting the device, wait for its
3362          * completion.
3363          */
3364         while (test_and_set_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state))
3365                 msleep(1);
3366
3367         if (test_and_set_bit(VMXNET3_STATE_BIT_QUIESCED,
3368                              &adapter->state)) {
3369                 clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3370                 return;
3371         }
3372         spin_lock_irqsave(&adapter->cmd_lock, flags);
3373         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3374                                VMXNET3_CMD_QUIESCE_DEV);
3375         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3376         vmxnet3_disable_all_intrs(adapter);
3377
3378         clear_bit(VMXNET3_STATE_BIT_RESETTING, &adapter->state);
3379 }
3380
3381
3382 #ifdef CONFIG_PM
3383
3384 static int
3385 vmxnet3_suspend(struct device *device)
3386 {
3387         struct pci_dev *pdev = to_pci_dev(device);
3388         struct net_device *netdev = pci_get_drvdata(pdev);
3389         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3390         struct Vmxnet3_PMConf *pmConf;
3391         struct ethhdr *ehdr;
3392         struct arphdr *ahdr;
3393         u8 *arpreq;
3394         struct in_device *in_dev;
3395         struct in_ifaddr *ifa;
3396         unsigned long flags;
3397         int i = 0;
3398
3399         if (!netif_running(netdev))
3400                 return 0;
3401
3402         for (i = 0; i < adapter->num_rx_queues; i++)
3403                 napi_disable(&adapter->rx_queue[i].napi);
3404
3405         vmxnet3_disable_all_intrs(adapter);
3406         vmxnet3_free_irqs(adapter);
3407         vmxnet3_free_intr_resources(adapter);
3408
3409         netif_device_detach(netdev);
3410         netif_tx_stop_all_queues(netdev);
3411
3412         /* Create wake-up filters. */
3413         pmConf = adapter->pm_conf;
3414         memset(pmConf, 0, sizeof(*pmConf));
3415
3416         if (adapter->wol & WAKE_UCAST) {
3417                 pmConf->filters[i].patternSize = ETH_ALEN;
3418                 pmConf->filters[i].maskSize = 1;
3419                 memcpy(pmConf->filters[i].pattern, netdev->dev_addr, ETH_ALEN);
3420                 pmConf->filters[i].mask[0] = 0x3F; /* LSB ETH_ALEN bits */
3421
3422                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3423                 i++;
3424         }
3425
3426         if (adapter->wol & WAKE_ARP) {
3427                 in_dev = in_dev_get(netdev);
3428                 if (!in_dev)
3429                         goto skip_arp;
3430
3431                 ifa = (struct in_ifaddr *)in_dev->ifa_list;
3432                 if (!ifa)
3433                         goto skip_arp;
3434
3435                 pmConf->filters[i].patternSize = ETH_HLEN + /* Ethernet header*/
3436                         sizeof(struct arphdr) +         /* ARP header */
3437                         2 * ETH_ALEN +          /* 2 Ethernet addresses*/
3438                         2 * sizeof(u32);        /*2 IPv4 addresses */
3439                 pmConf->filters[i].maskSize =
3440                         (pmConf->filters[i].patternSize - 1) / 8 + 1;
3441
3442                 /* ETH_P_ARP in Ethernet header. */
3443                 ehdr = (struct ethhdr *)pmConf->filters[i].pattern;
3444                 ehdr->h_proto = htons(ETH_P_ARP);
3445
3446                 /* ARPOP_REQUEST in ARP header. */
3447                 ahdr = (struct arphdr *)&pmConf->filters[i].pattern[ETH_HLEN];
3448                 ahdr->ar_op = htons(ARPOP_REQUEST);
3449                 arpreq = (u8 *)(ahdr + 1);
3450
3451                 /* The Unicast IPv4 address in 'tip' field. */
3452                 arpreq += 2 * ETH_ALEN + sizeof(u32);
3453                 *(u32 *)arpreq = ifa->ifa_address;
3454
3455                 /* The mask for the relevant bits. */
3456                 pmConf->filters[i].mask[0] = 0x00;
3457                 pmConf->filters[i].mask[1] = 0x30; /* ETH_P_ARP */
3458                 pmConf->filters[i].mask[2] = 0x30; /* ARPOP_REQUEST */
3459                 pmConf->filters[i].mask[3] = 0x00;
3460                 pmConf->filters[i].mask[4] = 0xC0; /* IPv4 TIP */
3461                 pmConf->filters[i].mask[5] = 0x03; /* IPv4 TIP */
3462                 in_dev_put(in_dev);
3463
3464                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_FILTER;
3465                 i++;
3466         }
3467
3468 skip_arp:
3469         if (adapter->wol & WAKE_MAGIC)
3470                 pmConf->wakeUpEvents |= VMXNET3_PM_WAKEUP_MAGIC;
3471
3472         pmConf->numFilters = i;
3473
3474         adapter->shared->devRead.pmConfDesc.confVer = cpu_to_le32(1);
3475         adapter->shared->devRead.pmConfDesc.confLen = cpu_to_le32(sizeof(
3476                                                                   *pmConf));
3477         adapter->shared->devRead.pmConfDesc.confPA =
3478                 cpu_to_le64(adapter->pm_conf_pa);
3479
3480         spin_lock_irqsave(&adapter->cmd_lock, flags);
3481         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3482                                VMXNET3_CMD_UPDATE_PMCFG);
3483         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3484
3485         pci_save_state(pdev);
3486         pci_enable_wake(pdev, pci_choose_state(pdev, PMSG_SUSPEND),
3487                         adapter->wol);
3488         pci_disable_device(pdev);
3489         pci_set_power_state(pdev, pci_choose_state(pdev, PMSG_SUSPEND));
3490
3491         return 0;
3492 }
3493
3494
3495 static int
3496 vmxnet3_resume(struct device *device)
3497 {
3498         int err;
3499         unsigned long flags;
3500         struct pci_dev *pdev = to_pci_dev(device);
3501         struct net_device *netdev = pci_get_drvdata(pdev);
3502         struct vmxnet3_adapter *adapter = netdev_priv(netdev);
3503
3504         if (!netif_running(netdev))
3505                 return 0;
3506
3507         pci_set_power_state(pdev, PCI_D0);
3508         pci_restore_state(pdev);
3509         err = pci_enable_device_mem(pdev);
3510         if (err != 0)
3511                 return err;
3512
3513         pci_enable_wake(pdev, PCI_D0, 0);
3514
3515         vmxnet3_alloc_intr_resources(adapter);
3516
3517         /* During hibernate and suspend, device has to be reinitialized as the
3518          * device state need not be preserved.
3519          */
3520
3521         /* Need not check adapter state as other reset tasks cannot run during
3522          * device resume.
3523          */
3524         spin_lock_irqsave(&adapter->cmd_lock, flags);
3525         VMXNET3_WRITE_BAR1_REG(adapter, VMXNET3_REG_CMD,
3526                                VMXNET3_CMD_QUIESCE_DEV);
3527         spin_unlock_irqrestore(&adapter->cmd_lock, flags);
3528         vmxnet3_tq_cleanup_all(adapter);
3529         vmxnet3_rq_cleanup_all(adapter);
3530
3531         vmxnet3_reset_dev(adapter);
3532         err = vmxnet3_activate_dev(adapter);
3533         if (err != 0) {
3534                 netdev_err(netdev,
3535                            "failed to re-activate on resume, error: %d", err);
3536                 vmxnet3_force_close(adapter);
3537                 return err;
3538         }
3539         netif_device_attach(netdev);
3540
3541         return 0;
3542 }
3543
3544 static const struct dev_pm_ops vmxnet3_pm_ops = {
3545         .suspend = vmxnet3_suspend,
3546         .resume = vmxnet3_resume,
3547         .freeze = vmxnet3_suspend,
3548         .restore = vmxnet3_resume,
3549 };
3550 #endif
3551
3552 static struct pci_driver vmxnet3_driver = {
3553         .name           = vmxnet3_driver_name,
3554         .id_table       = vmxnet3_pciid_table,
3555         .probe          = vmxnet3_probe_device,
3556         .remove         = vmxnet3_remove_device,
3557         .shutdown       = vmxnet3_shutdown_device,
3558 #ifdef CONFIG_PM
3559         .driver.pm      = &vmxnet3_pm_ops,
3560 #endif
3561 };
3562
3563
3564 static int __init
3565 vmxnet3_init_module(void)
3566 {
3567         pr_info("%s - version %s\n", VMXNET3_DRIVER_DESC,
3568                 VMXNET3_DRIVER_VERSION_REPORT);
3569         return pci_register_driver(&vmxnet3_driver);
3570 }
3571
3572 module_init(vmxnet3_init_module);
3573
3574
3575 static void
3576 vmxnet3_exit_module(void)
3577 {
3578         pci_unregister_driver(&vmxnet3_driver);
3579 }
3580
3581 module_exit(vmxnet3_exit_module);
3582
3583 MODULE_AUTHOR("VMware, Inc.");
3584 MODULE_DESCRIPTION(VMXNET3_DRIVER_DESC);
3585 MODULE_LICENSE("GPL v2");
3586 MODULE_VERSION(VMXNET3_DRIVER_VERSION_STRING);