SAFE public projects git trees. - safe/jmp/linux-2.6/blob - drivers/net/enic/enic_main.c

   1 /*
   2  * Copyright 2008 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  *
   5  * This program is free software; you may redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; version 2 of the License.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  16  * SOFTWARE.
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/kernel.h>
  22 #include <linux/string.h>
  23 #include <linux/errno.h>
  24 #include <linux/types.h>
  25 #include <linux/init.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/pci.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/etherdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/in.h>
  34 #include <linux/ip.h>
  35 #include <linux/ipv6.h>
  36 #include <linux/tcp.h>
  37 #include <net/ip6_checksum.h>
  38
  39 #include "cq_enet_desc.h"
  40 #include "vnic_dev.h"
  41 #include "vnic_intr.h"
  42 #include "vnic_stats.h"
  43 #include "enic_res.h"
  44 #include "enic.h"
  45
  46 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
  47 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
  48 #define MAX_TSO                         (1 << 16)
  49 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
  50
  51 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
  52
  53 /* Supported devices */
  54 static struct pci_device_id enic_id_table[] = {
  55         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  56         { 0, }  /* end of table */
  57 };
  58
  59 MODULE_DESCRIPTION(DRV_DESCRIPTION);
  60 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
  61 MODULE_LICENSE("GPL");
  62 MODULE_VERSION(DRV_VERSION);
  63 MODULE_DEVICE_TABLE(pci, enic_id_table);
  64
  65 struct enic_stat {
  66         char name[ETH_GSTRING_LEN];
  67         unsigned int offset;
  68 };
  69
  70 #define ENIC_TX_STAT(stat)      \
  71         { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
  72 #define ENIC_RX_STAT(stat)      \
  73         { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
  74
  75 static const struct enic_stat enic_tx_stats[] = {
  76         ENIC_TX_STAT(tx_frames_ok),
  77         ENIC_TX_STAT(tx_unicast_frames_ok),
  78         ENIC_TX_STAT(tx_multicast_frames_ok),
  79         ENIC_TX_STAT(tx_broadcast_frames_ok),
  80         ENIC_TX_STAT(tx_bytes_ok),
  81         ENIC_TX_STAT(tx_unicast_bytes_ok),
  82         ENIC_TX_STAT(tx_multicast_bytes_ok),
  83         ENIC_TX_STAT(tx_broadcast_bytes_ok),
  84         ENIC_TX_STAT(tx_drops),
  85         ENIC_TX_STAT(tx_errors),
  86         ENIC_TX_STAT(tx_tso),
  87 };
  88
  89 static const struct enic_stat enic_rx_stats[] = {
  90         ENIC_RX_STAT(rx_frames_ok),
  91         ENIC_RX_STAT(rx_frames_total),
  92         ENIC_RX_STAT(rx_unicast_frames_ok),
  93         ENIC_RX_STAT(rx_multicast_frames_ok),
  94         ENIC_RX_STAT(rx_broadcast_frames_ok),
  95         ENIC_RX_STAT(rx_bytes_ok),
  96         ENIC_RX_STAT(rx_unicast_bytes_ok),
  97         ENIC_RX_STAT(rx_multicast_bytes_ok),
  98         ENIC_RX_STAT(rx_broadcast_bytes_ok),
  99         ENIC_RX_STAT(rx_drop),
 100         ENIC_RX_STAT(rx_no_bufs),
 101         ENIC_RX_STAT(rx_errors),
 102         ENIC_RX_STAT(rx_rss),
 103         ENIC_RX_STAT(rx_crc_errors),
 104         ENIC_RX_STAT(rx_frames_64),
 105         ENIC_RX_STAT(rx_frames_127),
 106         ENIC_RX_STAT(rx_frames_255),
 107         ENIC_RX_STAT(rx_frames_511),
 108         ENIC_RX_STAT(rx_frames_1023),
 109         ENIC_RX_STAT(rx_frames_1518),
 110         ENIC_RX_STAT(rx_frames_to_max),
 111 };
 112
 113 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
 114 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
 115
 116 static int enic_get_settings(struct net_device *netdev,
 117         struct ethtool_cmd *ecmd)
 118 {
 119         struct enic *enic = netdev_priv(netdev);
 120
 121         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
 122         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
 123         ecmd->port = PORT_FIBRE;
 124         ecmd->transceiver = XCVR_EXTERNAL;
 125
 126         if (netif_carrier_ok(netdev)) {
 127                 ecmd->speed = vnic_dev_port_speed(enic->vdev);
 128                 ecmd->duplex = DUPLEX_FULL;
 129         } else {
 130                 ecmd->speed = -1;
 131                 ecmd->duplex = -1;
 132         }
 133
 134         ecmd->autoneg = AUTONEG_DISABLE;
 135
 136         return 0;
 137 }
 138
 139 static void enic_get_drvinfo(struct net_device *netdev,
 140         struct ethtool_drvinfo *drvinfo)
 141 {
 142         struct enic *enic = netdev_priv(netdev);
 143         struct vnic_devcmd_fw_info *fw_info;
 144
 145         spin_lock(&enic->devcmd_lock);
 146         vnic_dev_fw_info(enic->vdev, &fw_info);
 147         spin_unlock(&enic->devcmd_lock);
 148
 149         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
 150         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
 151         strncpy(drvinfo->fw_version, fw_info->fw_version,
 152                 sizeof(drvinfo->fw_version));
 153         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
 154                 sizeof(drvinfo->bus_info));
 155 }
 156
 157 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
 158 {
 159         unsigned int i;
 160
 161         switch (stringset) {
 162         case ETH_SS_STATS:
 163                 for (i = 0; i < enic_n_tx_stats; i++) {
 164                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
 165                         data += ETH_GSTRING_LEN;
 166                 }
 167                 for (i = 0; i < enic_n_rx_stats; i++) {
 168                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
 169                         data += ETH_GSTRING_LEN;
 170                 }
 171                 break;
 172         }
 173 }
 174
 175 static int enic_get_sset_count(struct net_device *netdev, int sset)
 176 {
 177         switch (sset) {
 178         case ETH_SS_STATS:
 179                 return enic_n_tx_stats + enic_n_rx_stats;
 180         default:
 181                 return -EOPNOTSUPP;
 182         }
 183 }
 184
 185 static void enic_get_ethtool_stats(struct net_device *netdev,
 186         struct ethtool_stats *stats, u64 *data)
 187 {
 188         struct enic *enic = netdev_priv(netdev);
 189         struct vnic_stats *vstats;
 190         unsigned int i;
 191
 192         spin_lock(&enic->devcmd_lock);
 193         vnic_dev_stats_dump(enic->vdev, &vstats);
 194         spin_unlock(&enic->devcmd_lock);
 195
 196         for (i = 0; i < enic_n_tx_stats; i++)
 197                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
 198         for (i = 0; i < enic_n_rx_stats; i++)
 199                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
 200 }
 201
 202 static u32 enic_get_rx_csum(struct net_device *netdev)
 203 {
 204         struct enic *enic = netdev_priv(netdev);
 205         return enic->csum_rx_enabled;
 206 }
 207
 208 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
 209 {
 210         struct enic *enic = netdev_priv(netdev);
 211
 212         if (data && !ENIC_SETTING(enic, RXCSUM))
 213                 return -EINVAL;
 214
 215         enic->csum_rx_enabled = !!data;
 216
 217         return 0;
 218 }
 219
 220 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
 221 {
 222         struct enic *enic = netdev_priv(netdev);
 223
 224         if (data && !ENIC_SETTING(enic, TXCSUM))
 225                 return -EINVAL;
 226
 227         if (data)
 228                 netdev->features |= NETIF_F_HW_CSUM;
 229         else
 230                 netdev->features &= ~NETIF_F_HW_CSUM;
 231
 232         return 0;
 233 }
 234
 235 static int enic_set_tso(struct net_device *netdev, u32 data)
 236 {
 237         struct enic *enic = netdev_priv(netdev);
 238
 239         if (data && !ENIC_SETTING(enic, TSO))
 240                 return -EINVAL;
 241
 242         if (data)
 243                 netdev->features |=
 244                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
 245         else
 246                 netdev->features &=
 247                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
 248
 249         return 0;
 250 }
 251
 252 static u32 enic_get_msglevel(struct net_device *netdev)
 253 {
 254         struct enic *enic = netdev_priv(netdev);
 255         return enic->msg_enable;
 256 }
 257
 258 static void enic_set_msglevel(struct net_device *netdev, u32 value)
 259 {
 260         struct enic *enic = netdev_priv(netdev);
 261         enic->msg_enable = value;
 262 }
 263
 264 static const struct ethtool_ops enic_ethtool_ops = {
 265         .get_settings = enic_get_settings,
 266         .get_drvinfo = enic_get_drvinfo,
 267         .get_msglevel = enic_get_msglevel,
 268         .set_msglevel = enic_set_msglevel,
 269         .get_link = ethtool_op_get_link,
 270         .get_strings = enic_get_strings,
 271         .get_sset_count = enic_get_sset_count,
 272         .get_ethtool_stats = enic_get_ethtool_stats,
 273         .get_rx_csum = enic_get_rx_csum,
 274         .set_rx_csum = enic_set_rx_csum,
 275         .get_tx_csum = ethtool_op_get_tx_csum,
 276         .set_tx_csum = enic_set_tx_csum,
 277         .get_sg = ethtool_op_get_sg,
 278         .set_sg = ethtool_op_set_sg,
 279         .get_tso = ethtool_op_get_tso,
 280         .set_tso = enic_set_tso,
 281         .get_flags = ethtool_op_get_flags,
 282         .set_flags = ethtool_op_set_flags,
 283 };
 284
 285 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 286 {
 287         struct enic *enic = vnic_dev_priv(wq->vdev);
 288
 289         if (buf->sop)
 290                 pci_unmap_single(enic->pdev, buf->dma_addr,
 291                         buf->len, PCI_DMA_TODEVICE);
 292         else
 293                 pci_unmap_page(enic->pdev, buf->dma_addr,
 294                         buf->len, PCI_DMA_TODEVICE);
 295
 296         if (buf->os_buf)
 297                 dev_kfree_skb_any(buf->os_buf);
 298 }
 299
 300 static void enic_wq_free_buf(struct vnic_wq *wq,
 301         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 302 {
 303         enic_free_wq_buf(wq, buf);
 304 }
 305
 306 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 307         u8 type, u16 q_number, u16 completed_index, void *opaque)
 308 {
 309         struct enic *enic = vnic_dev_priv(vdev);
 310
 311         spin_lock(&enic->wq_lock[q_number]);
 312
 313         vnic_wq_service(&enic->wq[q_number], cq_desc,
 314                 completed_index, enic_wq_free_buf,
 315                 opaque);
 316
 317         if (netif_queue_stopped(enic->netdev) &&
 318             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 319             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 320                 netif_wake_queue(enic->netdev);
 321
 322         spin_unlock(&enic->wq_lock[q_number]);
 323
 324         return 0;
 325 }
 326
 327 static void enic_log_q_error(struct enic *enic)
 328 {
 329         unsigned int i;
 330         u32 error_status;
 331
 332         for (i = 0; i < enic->wq_count; i++) {
 333                 error_status = vnic_wq_error_status(&enic->wq[i]);
 334                 if (error_status)
 335                         printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
 336                                 enic->netdev->name, i, error_status);
 337         }
 338
 339         for (i = 0; i < enic->rq_count; i++) {
 340                 error_status = vnic_rq_error_status(&enic->rq[i]);
 341                 if (error_status)
 342                         printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
 343                                 enic->netdev->name, i, error_status);
 344         }
 345 }
 346
 347 static void enic_link_check(struct enic *enic)
 348 {
 349         int link_status = vnic_dev_link_status(enic->vdev);
 350         int carrier_ok = netif_carrier_ok(enic->netdev);
 351
 352         if (link_status && !carrier_ok) {
 353                 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
 354                 netif_carrier_on(enic->netdev);
 355         } else if (!link_status && carrier_ok) {
 356                 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
 357                 netif_carrier_off(enic->netdev);
 358         }
 359 }
 360
 361 static void enic_mtu_check(struct enic *enic)
 362 {
 363         u32 mtu = vnic_dev_mtu(enic->vdev);
 364
 365         if (mtu && mtu != enic->port_mtu) {
 366                 if (mtu < enic->netdev->mtu)
 367                         printk(KERN_WARNING PFX
 368                                 "%s: interface MTU (%d) set higher "
 369                                 "than switch port MTU (%d)\n",
 370                                 enic->netdev->name, enic->netdev->mtu, mtu);
 371                 enic->port_mtu = mtu;
 372         }
 373 }
 374
 375 static void enic_msglvl_check(struct enic *enic)
 376 {
 377         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 378
 379         if (msg_enable != enic->msg_enable) {
 380                 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
 381                         enic->netdev->name, enic->msg_enable, msg_enable);
 382                 enic->msg_enable = msg_enable;
 383         }
 384 }
 385
 386 static void enic_notify_check(struct enic *enic)
 387 {
 388         enic_msglvl_check(enic);
 389         enic_mtu_check(enic);
 390         enic_link_check(enic);
 391 }
 392
 393 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 394
 395 static irqreturn_t enic_isr_legacy(int irq, void *data)
 396 {
 397         struct net_device *netdev = data;
 398         struct enic *enic = netdev_priv(netdev);
 399         u32 pba;
 400
 401         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
 402
 403         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 404         if (!pba) {
 405                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 406                 return IRQ_NONE;        /* not our interrupt */
 407         }
 408
 409         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
 410                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
 411                 enic_notify_check(enic);
 412         }
 413
 414         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
 415                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
 416                 enic_log_q_error(enic);
 417                 /* schedule recovery from WQ/RQ error */
 418                 schedule_work(&enic->reset);
 419                 return IRQ_HANDLED;
 420         }
 421
 422         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
 423                 if (napi_schedule_prep(&enic->napi))
 424                         __napi_schedule(&enic->napi);
 425         } else {
 426                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 427         }
 428
 429         return IRQ_HANDLED;
 430 }
 431
 432 static irqreturn_t enic_isr_msi(int irq, void *data)
 433 {
 434         struct enic *enic = data;
 435
 436         /* With MSI, there is no sharing of interrupts, so this is
 437          * our interrupt and there is no need to ack it.  The device
 438          * is not providing per-vector masking, so the OS will not
 439          * write to PCI config space to mask/unmask the interrupt.
 440          * We're using mask_on_assertion for MSI, so the device
 441          * automatically masks the interrupt when the interrupt is
 442          * generated.  Later, when exiting polling, the interrupt
 443          * will be unmasked (see enic_poll).
 444          *
 445          * Also, the device uses the same PCIe Traffic Class (TC)
 446          * for Memory Write data and MSI, so there are no ordering
 447          * issues; the MSI will always arrive at the Root Complex
 448          * _after_ corresponding Memory Writes (i.e. descriptor
 449          * writes).
 450          */
 451
 452         napi_schedule(&enic->napi);
 453
 454         return IRQ_HANDLED;
 455 }
 456
 457 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
 458 {
 459         struct enic *enic = data;
 460
 461         /* schedule NAPI polling for RQ cleanup */
 462         napi_schedule(&enic->napi);
 463
 464         return IRQ_HANDLED;
 465 }
 466
 467 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
 468 {
 469         struct enic *enic = data;
 470         unsigned int wq_work_to_do = -1; /* no limit */
 471         unsigned int wq_work_done;
 472
 473         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
 474                 wq_work_to_do, enic_wq_service, NULL);
 475
 476         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
 477                 wq_work_done,
 478                 1 /* unmask intr */,
 479                 1 /* reset intr timer */);
 480
 481         return IRQ_HANDLED;
 482 }
 483
 484 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 485 {
 486         struct enic *enic = data;
 487
 488         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
 489
 490         enic_log_q_error(enic);
 491
 492         /* schedule recovery from WQ/RQ error */
 493         schedule_work(&enic->reset);
 494
 495         return IRQ_HANDLED;
 496 }
 497
 498 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 499 {
 500         struct enic *enic = data;
 501
 502         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
 503         enic_notify_check(enic);
 504
 505         return IRQ_HANDLED;
 506 }
 507
 508 static inline void enic_queue_wq_skb_cont(struct enic *enic,
 509         struct vnic_wq *wq, struct sk_buff *skb,
 510         unsigned int len_left)
 511 {
 512         skb_frag_t *frag;
 513
 514         /* Queue additional data fragments */
 515         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 516                 len_left -= frag->size;
 517                 enic_queue_wq_desc_cont(wq, skb,
 518                         pci_map_page(enic->pdev, frag->page,
 519                                 frag->page_offset, frag->size,
 520                                 PCI_DMA_TODEVICE),
 521                         frag->size,
 522                         (len_left == 0));       /* EOP? */
 523         }
 524 }
 525
 526 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
 527         struct vnic_wq *wq, struct sk_buff *skb,
 528         int vlan_tag_insert, unsigned int vlan_tag)
 529 {
 530         unsigned int head_len = skb_headlen(skb);
 531         unsigned int len_left = skb->len - head_len;
 532         int eop = (len_left == 0);
 533
 534         /* Queue the main skb fragment. The fragments are no larger
 535          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 536          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 537          * per fragment is queued.
 538          */
 539         enic_queue_wq_desc(wq, skb,
 540                 pci_map_single(enic->pdev, skb->data,
 541                         head_len, PCI_DMA_TODEVICE),
 542                 head_len,
 543                 vlan_tag_insert, vlan_tag,
 544                 eop);
 545
 546         if (!eop)
 547                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 548 }
 549
 550 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
 551         struct vnic_wq *wq, struct sk_buff *skb,
 552         int vlan_tag_insert, unsigned int vlan_tag)
 553 {
 554         unsigned int head_len = skb_headlen(skb);
 555         unsigned int len_left = skb->len - head_len;
 556         unsigned int hdr_len = skb_transport_offset(skb);
 557         unsigned int csum_offset = hdr_len + skb->csum_offset;
 558         int eop = (len_left == 0);
 559
 560         /* Queue the main skb fragment. The fragments are no larger
 561          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 562          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 563          * per fragment is queued.
 564          */
 565         enic_queue_wq_desc_csum_l4(wq, skb,
 566                 pci_map_single(enic->pdev, skb->data,
 567                         head_len, PCI_DMA_TODEVICE),
 568                 head_len,
 569                 csum_offset,
 570                 hdr_len,
 571                 vlan_tag_insert, vlan_tag,
 572                 eop);
 573
 574         if (!eop)
 575                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 576 }
 577
 578 static inline void enic_queue_wq_skb_tso(struct enic *enic,
 579         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
 580         int vlan_tag_insert, unsigned int vlan_tag)
 581 {
 582         unsigned int frag_len_left = skb_headlen(skb);
 583         unsigned int len_left = skb->len - frag_len_left;
 584         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 585         int eop = (len_left == 0);
 586         unsigned int len;
 587         dma_addr_t dma_addr;
 588         unsigned int offset = 0;
 589         skb_frag_t *frag;
 590
 591         /* Preload TCP csum field with IP pseudo hdr calculated
 592          * with IP length set to zero.  HW will later add in length
 593          * to each TCP segment resulting from the TSO.
 594          */
 595
 596         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 597                 ip_hdr(skb)->check = 0;
 598                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 599                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 600         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 601                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 602                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 603         }
 604
 605         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 606          * for the main skb fragment
 607          */
 608         while (frag_len_left) {
 609                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 610                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
 611                                 len, PCI_DMA_TODEVICE);
 612                 enic_queue_wq_desc_tso(wq, skb,
 613                         dma_addr,
 614                         len,
 615                         mss, hdr_len,
 616                         vlan_tag_insert, vlan_tag,
 617                         eop && (len == frag_len_left));
 618                 frag_len_left -= len;
 619                 offset += len;
 620         }
 621
 622         if (eop)
 623                 return;
 624
 625         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 626          * for additional data fragments
 627          */
 628         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 629                 len_left -= frag->size;
 630                 frag_len_left = frag->size;
 631                 offset = frag->page_offset;
 632
 633                 while (frag_len_left) {
 634                         len = min(frag_len_left,
 635                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 636                         dma_addr = pci_map_page(enic->pdev, frag->page,
 637                                 offset, len,
 638                                 PCI_DMA_TODEVICE);
 639                         enic_queue_wq_desc_cont(wq, skb,
 640                                 dma_addr,
 641                                 len,
 642                                 (len_left == 0) &&
 643                                 (len == frag_len_left));        /* EOP? */
 644                         frag_len_left -= len;
 645                         offset += len;
 646                 }
 647         }
 648 }
 649
 650 static inline void enic_queue_wq_skb(struct enic *enic,
 651         struct vnic_wq *wq, struct sk_buff *skb)
 652 {
 653         unsigned int mss = skb_shinfo(skb)->gso_size;
 654         unsigned int vlan_tag = 0;
 655         int vlan_tag_insert = 0;
 656
 657         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
 658                 /* VLAN tag from trunking driver */
 659                 vlan_tag_insert = 1;
 660                 vlan_tag = vlan_tx_tag_get(skb);
 661         }
 662
 663         if (mss)
 664                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
 665                         vlan_tag_insert, vlan_tag);
 666         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 667                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
 668                         vlan_tag_insert, vlan_tag);
 669         else
 670                 enic_queue_wq_skb_vlan(enic, wq, skb,
 671                         vlan_tag_insert, vlan_tag);
 672 }
 673
 674 /* netif_tx_lock held, process context with BHs disabled, or BH */
 675 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 676                                               struct net_device *netdev)
 677 {
 678         struct enic *enic = netdev_priv(netdev);
 679         struct vnic_wq *wq = &enic->wq[0];
 680         unsigned long flags;
 681
 682         if (skb->len <= 0) {
 683                 dev_kfree_skb(skb);
 684                 return NETDEV_TX_OK;
 685         }
 686
 687         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 688          * which is very likely.  In the off chance it's going to take
 689          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 690          */
 691
 692         if (skb_shinfo(skb)->gso_size == 0 &&
 693             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 694             skb_linearize(skb)) {
 695                 dev_kfree_skb(skb);
 696                 return NETDEV_TX_OK;
 697         }
 698
 699         spin_lock_irqsave(&enic->wq_lock[0], flags);
 700
 701         if (vnic_wq_desc_avail(wq) <
 702             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 703                 netif_stop_queue(netdev);
 704                 /* This is a hard error, log it */
 705                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
 706                         "queue awake!\n", netdev->name);
 707                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 708                 return NETDEV_TX_BUSY;
 709         }
 710
 711         enic_queue_wq_skb(enic, wq, skb);
 712
 713         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 714                 netif_stop_queue(netdev);
 715
 716         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 717
 718         return NETDEV_TX_OK;
 719 }
 720
 721 /* dev_base_lock rwlock held, nominally process context */
 722 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
 723 {
 724         struct enic *enic = netdev_priv(netdev);
 725         struct net_device_stats *net_stats = &netdev->stats;
 726         struct vnic_stats *stats;
 727
 728         spin_lock(&enic->devcmd_lock);
 729         vnic_dev_stats_dump(enic->vdev, &stats);
 730         spin_unlock(&enic->devcmd_lock);
 731
 732         net_stats->tx_packets = stats->tx.tx_frames_ok;
 733         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 734         net_stats->tx_errors = stats->tx.tx_errors;
 735         net_stats->tx_dropped = stats->tx.tx_drops;
 736
 737         net_stats->rx_packets = stats->rx.rx_frames_ok;
 738         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 739         net_stats->rx_errors = stats->rx.rx_errors;
 740         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 741         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 742         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 743         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 744
 745         return net_stats;
 746 }
 747
 748 static void enic_reset_mcaddrs(struct enic *enic)
 749 {
 750         enic->mc_count = 0;
 751 }
 752
 753 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 754 {
 755         if (!is_valid_ether_addr(addr))
 756                 return -EADDRNOTAVAIL;
 757
 758         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 759
 760         return 0;
 761 }
 762
 763 /* netif_tx_lock held, BHs disabled */
 764 static void enic_set_multicast_list(struct net_device *netdev)
 765 {
 766         struct enic *enic = netdev_priv(netdev);
 767         struct dev_mc_list *list = netdev->mc_list;
 768         int directed = 1;
 769         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 770         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 771         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
 772         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 773             (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
 774         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
 775         unsigned int mc_count = netdev->mc_count;
 776         unsigned int i, j;
 777
 778         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
 779                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
 780
 781         spin_lock(&enic->devcmd_lock);
 782
 783         vnic_dev_packet_filter(enic->vdev, directed,
 784                 multicast, broadcast, promisc, allmulti);
 785
 786         /* Is there an easier way?  Trying to minimize to
 787          * calls to add/del multicast addrs.  We keep the
 788          * addrs from the last call in enic->mc_addr and
 789          * look for changes to add/del.
 790          */
 791
 792         for (i = 0; list && i < mc_count; i++) {
 793                 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
 794                 list = list->next;
 795         }
 796
 797         for (i = 0; i < enic->mc_count; i++) {
 798                 for (j = 0; j < mc_count; j++)
 799                         if (compare_ether_addr(enic->mc_addr[i],
 800                                 mc_addr[j]) == 0)
 801                                 break;
 802                 if (j == mc_count)
 803                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
 804         }
 805
 806         for (i = 0; i < mc_count; i++) {
 807                 for (j = 0; j < enic->mc_count; j++)
 808                         if (compare_ether_addr(mc_addr[i],
 809                                 enic->mc_addr[j]) == 0)
 810                                 break;
 811                 if (j == enic->mc_count)
 812                         enic_add_multicast_addr(enic, mc_addr[i]);
 813         }
 814
 815         /* Save the list to compare against next time
 816          */
 817
 818         for (i = 0; i < mc_count; i++)
 819                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
 820
 821         enic->mc_count = mc_count;
 822
 823         spin_unlock(&enic->devcmd_lock);
 824 }
 825
 826 /* rtnl lock is held */
 827 static void enic_vlan_rx_register(struct net_device *netdev,
 828         struct vlan_group *vlan_group)
 829 {
 830         struct enic *enic = netdev_priv(netdev);
 831         enic->vlan_group = vlan_group;
 832 }
 833
 834 /* rtnl lock is held */
 835 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
 836 {
 837         struct enic *enic = netdev_priv(netdev);
 838
 839         spin_lock(&enic->devcmd_lock);
 840         enic_add_vlan(enic, vid);
 841         spin_unlock(&enic->devcmd_lock);
 842 }
 843
 844 /* rtnl lock is held */
 845 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
 846 {
 847         struct enic *enic = netdev_priv(netdev);
 848
 849         spin_lock(&enic->devcmd_lock);
 850         enic_del_vlan(enic, vid);
 851         spin_unlock(&enic->devcmd_lock);
 852 }
 853
 854 /* netif_tx_lock held, BHs disabled */
 855 static void enic_tx_timeout(struct net_device *netdev)
 856 {
 857         struct enic *enic = netdev_priv(netdev);
 858         schedule_work(&enic->reset);
 859 }
 860
 861 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
 862 {
 863         struct enic *enic = vnic_dev_priv(rq->vdev);
 864
 865         if (!buf->os_buf)
 866                 return;
 867
 868         pci_unmap_single(enic->pdev, buf->dma_addr,
 869                 buf->len, PCI_DMA_FROMDEVICE);
 870         dev_kfree_skb_any(buf->os_buf);
 871 }
 872
 873 static int enic_rq_alloc_buf(struct vnic_rq *rq)
 874 {
 875         struct enic *enic = vnic_dev_priv(rq->vdev);
 876         struct net_device *netdev = enic->netdev;
 877         struct sk_buff *skb;
 878         unsigned int len = netdev->mtu + ETH_HLEN;
 879         unsigned int os_buf_index = 0;
 880         dma_addr_t dma_addr;
 881
 882         skb = netdev_alloc_skb_ip_align(netdev, len);
 883         if (!skb)
 884                 return -ENOMEM;
 885
 886         dma_addr = pci_map_single(enic->pdev, skb->data,
 887                 len, PCI_DMA_FROMDEVICE);
 888
 889         enic_queue_rq_desc(rq, skb, os_buf_index,
 890                 dma_addr, len);
 891
 892         return 0;
 893 }
 894
 895 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
 896 {
 897         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
 898
 899         if (vnic_rq_posting_soon(rq)) {
 900
 901                 /* SW workaround for A0 HW erratum: if we're just about
 902                  * to write posted_index, insert a dummy desc
 903                  * of type resvd
 904                  */
 905
 906                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
 907                 vnic_rq_post(rq, 0, 0, 0, 0);
 908         } else {
 909                 return enic_rq_alloc_buf(rq);
 910         }
 911
 912         return 0;
 913 }
 914
 915 static int enic_set_rq_alloc_buf(struct enic *enic)
 916 {
 917         enum vnic_dev_hw_version hw_ver;
 918         int err;
 919
 920         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
 921         if (err)
 922                 return err;
 923
 924         switch (hw_ver) {
 925         case VNIC_DEV_HW_VER_A1:
 926                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
 927                 break;
 928         case VNIC_DEV_HW_VER_A2:
 929         case VNIC_DEV_HW_VER_UNKNOWN:
 930                 enic->rq_alloc_buf = enic_rq_alloc_buf;
 931                 break;
 932         default:
 933                 return -ENODEV;
 934         }
 935
 936         return 0;
 937 }
 938
 939 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
 940         void **tcph, u64 *hdr_flags, void *priv)
 941 {
 942         struct cq_enet_rq_desc *cq_desc = priv;
 943         unsigned int ip_len;
 944         struct iphdr *iph;
 945
 946         u8 type, color, eop, sop, ingress_port, vlan_stripped;
 947         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
 948         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
 949         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
 950         u8 packet_error;
 951         u16 q_number, completed_index, bytes_written, vlan, checksum;
 952         u32 rss_hash;
 953
 954         cq_enet_rq_desc_dec(cq_desc,
 955                 &type, &color, &q_number, &completed_index,
 956                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
 957                 &csum_not_calc, &rss_hash, &bytes_written,
 958                 &packet_error, &vlan_stripped, &vlan, &checksum,
 959                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
 960                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
 961                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
 962                 &fcs_ok);
 963
 964         if (!(ipv4 && tcp && !ipv4_fragment))
 965                 return -1;
 966
 967         skb_reset_network_header(skb);
 968         iph = ip_hdr(skb);
 969
 970         ip_len = ip_hdrlen(skb);
 971         skb_set_transport_header(skb, ip_len);
 972
 973         /* check if ip header and tcp header are complete */
 974         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
 975                 return -1;
 976
 977         *hdr_flags = LRO_IPV4 | LRO_TCP;
 978         *tcph = tcp_hdr(skb);
 979         *iphdr = iph;
 980
 981         return 0;
 982 }
 983
 984 static void enic_rq_indicate_buf(struct vnic_rq *rq,
 985         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
 986         int skipped, void *opaque)
 987 {
 988         struct enic *enic = vnic_dev_priv(rq->vdev);
 989         struct net_device *netdev = enic->netdev;
 990         struct sk_buff *skb;
 991
 992         u8 type, color, eop, sop, ingress_port, vlan_stripped;
 993         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
 994         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
 995         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
 996         u8 packet_error;
 997         u16 q_number, completed_index, bytes_written, vlan, checksum;
 998         u32 rss_hash;
 999
1000         if (skipped)
1001                 return;
1002
1003         skb = buf->os_buf;
1004         prefetch(skb->data - NET_IP_ALIGN);
1005         pci_unmap_single(enic->pdev, buf->dma_addr,
1006                 buf->len, PCI_DMA_FROMDEVICE);
1007
1008         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1009                 &type, &color, &q_number, &completed_index,
1010                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1011                 &csum_not_calc, &rss_hash, &bytes_written,
1012                 &packet_error, &vlan_stripped, &vlan, &checksum,
1013                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1014                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1015                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1016                 &fcs_ok);
1017
1018         if (packet_error) {
1019
1020                 if (!fcs_ok) {
1021                         if (bytes_written > 0)
1022                                 enic->rq_bad_fcs++;
1023                         else if (bytes_written == 0)
1024                                 enic->rq_truncated_pkts++;
1025                 }
1026
1027                 dev_kfree_skb_any(skb);
1028
1029                 return;
1030         }
1031
1032         if (eop && bytes_written > 0) {
1033
1034                 /* Good receive
1035                  */
1036
1037                 skb_put(skb, bytes_written);
1038                 skb->protocol = eth_type_trans(skb, netdev);
1039
1040                 if (enic->csum_rx_enabled && !csum_not_calc) {
1041                         skb->csum = htons(checksum);
1042                         skb->ip_summed = CHECKSUM_COMPLETE;
1043                 }
1044
1045                 skb->dev = netdev;
1046
1047                 if (enic->vlan_group && vlan_stripped) {
1048
1049                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1050                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
1051                                         skb, enic->vlan_group,
1052                                         vlan, cq_desc);
1053                         else
1054                                 vlan_hwaccel_receive_skb(skb,
1055                                         enic->vlan_group, vlan);
1056
1057                 } else {
1058
1059                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1060                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
1061                         else
1062                                 netif_receive_skb(skb);
1063
1064                 }
1065
1066         } else {
1067
1068                 /* Buffer overflow
1069                  */
1070
1071                 dev_kfree_skb_any(skb);
1072         }
1073 }
1074
1075 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1076         u8 type, u16 q_number, u16 completed_index, void *opaque)
1077 {
1078         struct enic *enic = vnic_dev_priv(vdev);
1079
1080         vnic_rq_service(&enic->rq[q_number], cq_desc,
1081                 completed_index, VNIC_RQ_RETURN_DESC,
1082                 enic_rq_indicate_buf, opaque);
1083
1084         return 0;
1085 }
1086
1087 static int enic_poll(struct napi_struct *napi, int budget)
1088 {
1089         struct enic *enic = container_of(napi, struct enic, napi);
1090         struct net_device *netdev = enic->netdev;
1091         unsigned int rq_work_to_do = budget;
1092         unsigned int wq_work_to_do = -1; /* no limit */
1093         unsigned int  work_done, rq_work_done, wq_work_done;
1094         int err;
1095
1096         /* Service RQ (first) and WQ
1097          */
1098
1099         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1100                 rq_work_to_do, enic_rq_service, NULL);
1101
1102         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1103                 wq_work_to_do, enic_wq_service, NULL);
1104
1105         /* Accumulate intr event credits for this polling
1106          * cycle.  An intr event is the completion of a
1107          * a WQ or RQ packet.
1108          */
1109
1110         work_done = rq_work_done + wq_work_done;
1111
1112         if (work_done > 0)
1113                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1114                         work_done,
1115                         0 /* don't unmask intr */,
1116                         0 /* don't reset intr timer */);
1117
1118         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1119
1120         /* Buffer allocation failed. Stay in polling
1121          * mode so we can try to fill the ring again.
1122          */
1123
1124         if (err)
1125                 rq_work_done = rq_work_to_do;
1126
1127         if (rq_work_done < rq_work_to_do) {
1128
1129                 /* Some work done, but not enough to stay in polling,
1130                  * flush all LROs and exit polling
1131                  */
1132
1133                 if (netdev->features & NETIF_F_LRO)
1134                         lro_flush_all(&enic->lro_mgr);
1135
1136                 napi_complete(napi);
1137                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1138         }
1139
1140         return rq_work_done;
1141 }
1142
1143 static int enic_poll_msix(struct napi_struct *napi, int budget)
1144 {
1145         struct enic *enic = container_of(napi, struct enic, napi);
1146         struct net_device *netdev = enic->netdev;
1147         unsigned int work_to_do = budget;
1148         unsigned int work_done;
1149         int err;
1150
1151         /* Service RQ
1152          */
1153
1154         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1155                 work_to_do, enic_rq_service, NULL);
1156
1157         /* Return intr event credits for this polling
1158          * cycle.  An intr event is the completion of a
1159          * RQ packet.
1160          */
1161
1162         if (work_done > 0)
1163                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1164                         work_done,
1165                         0 /* don't unmask intr */,
1166                         0 /* don't reset intr timer */);
1167
1168         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1169
1170         /* Buffer allocation failed. Stay in polling mode
1171          * so we can try to fill the ring again.
1172          */
1173
1174         if (err)
1175                 work_done = work_to_do;
1176
1177         if (work_done < work_to_do) {
1178
1179                 /* Some work done, but not enough to stay in polling,
1180                  * flush all LROs and exit polling
1181                  */
1182
1183                 if (netdev->features & NETIF_F_LRO)
1184                         lro_flush_all(&enic->lro_mgr);
1185
1186                 napi_complete(napi);
1187                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1188         }
1189
1190         return work_done;
1191 }
1192
1193 static void enic_notify_timer(unsigned long data)
1194 {
1195         struct enic *enic = (struct enic *)data;
1196
1197         enic_notify_check(enic);
1198
1199         mod_timer(&enic->notify_timer,
1200                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1201 }
1202
1203 static void enic_free_intr(struct enic *enic)
1204 {
1205         struct net_device *netdev = enic->netdev;
1206         unsigned int i;
1207
1208         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1209         case VNIC_DEV_INTR_MODE_INTX:
1210                 free_irq(enic->pdev->irq, netdev);
1211                 break;
1212         case VNIC_DEV_INTR_MODE_MSI:
1213                 free_irq(enic->pdev->irq, enic);
1214                 break;
1215         case VNIC_DEV_INTR_MODE_MSIX:
1216                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1217                         if (enic->msix[i].requested)
1218                                 free_irq(enic->msix_entry[i].vector,
1219                                         enic->msix[i].devid);
1220                 break;
1221         default:
1222                 break;
1223         }
1224 }
1225
1226 static int enic_request_intr(struct enic *enic)
1227 {
1228         struct net_device *netdev = enic->netdev;
1229         unsigned int i;
1230         int err = 0;
1231
1232         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1233
1234         case VNIC_DEV_INTR_MODE_INTX:
1235
1236                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1237                         IRQF_SHARED, netdev->name, netdev);
1238                 break;
1239
1240         case VNIC_DEV_INTR_MODE_MSI:
1241
1242                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1243                         0, netdev->name, enic);
1244                 break;
1245
1246         case VNIC_DEV_INTR_MODE_MSIX:
1247
1248                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1249                         "%.11s-rx-0", netdev->name);
1250                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1251                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1252
1253                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1254                         "%.11s-tx-0", netdev->name);
1255                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1256                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1257
1258                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1259                         "%.11s-err", netdev->name);
1260                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1261                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1262
1263                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1264                         "%.11s-notify", netdev->name);
1265                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1266                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1267
1268                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1269                         err = request_irq(enic->msix_entry[i].vector,
1270                                 enic->msix[i].isr, 0,
1271                                 enic->msix[i].devname,
1272                                 enic->msix[i].devid);
1273                         if (err) {
1274                                 enic_free_intr(enic);
1275                                 break;
1276                         }
1277                         enic->msix[i].requested = 1;
1278                 }
1279
1280                 break;
1281
1282         default:
1283                 break;
1284         }
1285
1286         return err;
1287 }
1288
1289 static void enic_synchronize_irqs(struct enic *enic)
1290 {
1291         unsigned int i;
1292
1293         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1294         case VNIC_DEV_INTR_MODE_INTX:
1295         case VNIC_DEV_INTR_MODE_MSI:
1296                 synchronize_irq(enic->pdev->irq);
1297                 break;
1298         case VNIC_DEV_INTR_MODE_MSIX:
1299                 for (i = 0; i < enic->intr_count; i++)
1300                         synchronize_irq(enic->msix_entry[i].vector);
1301                 break;
1302         default:
1303                 break;
1304         }
1305 }
1306
1307 static int enic_notify_set(struct enic *enic)
1308 {
1309         int err;
1310
1311         spin_lock(&enic->devcmd_lock);
1312         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1313         case VNIC_DEV_INTR_MODE_INTX:
1314                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1315                 break;
1316         case VNIC_DEV_INTR_MODE_MSIX:
1317                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1318                 break;
1319         default:
1320                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1321                 break;
1322         }
1323         spin_unlock(&enic->devcmd_lock);
1324
1325         return err;
1326 }
1327
1328 static void enic_notify_timer_start(struct enic *enic)
1329 {
1330         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1331         case VNIC_DEV_INTR_MODE_MSI:
1332                 mod_timer(&enic->notify_timer, jiffies);
1333                 break;
1334         default:
1335                 /* Using intr for notification for INTx/MSI-X */
1336                 break;
1337         };
1338 }
1339
1340 /* rtnl lock is held, process context */
1341 static int enic_open(struct net_device *netdev)
1342 {
1343         struct enic *enic = netdev_priv(netdev);
1344         unsigned int i;
1345         int err;
1346
1347         err = enic_request_intr(enic);
1348         if (err) {
1349                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1350                         netdev->name);
1351                 return err;
1352         }
1353
1354         err = enic_notify_set(enic);
1355         if (err) {
1356                 printk(KERN_ERR PFX
1357                         "%s: Failed to alloc notify buffer, aborting.\n",
1358                         netdev->name);
1359                 goto err_out_free_intr;
1360         }
1361
1362         for (i = 0; i < enic->rq_count; i++) {
1363                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1364                 /* Need at least one buffer on ring to get going */
1365                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1366                         printk(KERN_ERR PFX
1367                                 "%s: Unable to alloc receive buffers.\n",
1368                                 netdev->name);
1369                         err = -ENOMEM;
1370                         goto err_out_notify_unset;
1371                 }
1372         }
1373
1374         for (i = 0; i < enic->wq_count; i++)
1375                 vnic_wq_enable(&enic->wq[i]);
1376         for (i = 0; i < enic->rq_count; i++)
1377                 vnic_rq_enable(&enic->rq[i]);
1378
1379         spin_lock(&enic->devcmd_lock);
1380         enic_add_station_addr(enic);
1381         spin_unlock(&enic->devcmd_lock);
1382         enic_set_multicast_list(netdev);
1383
1384         netif_wake_queue(netdev);
1385         napi_enable(&enic->napi);
1386         spin_lock(&enic->devcmd_lock);
1387         vnic_dev_enable(enic->vdev);
1388         spin_unlock(&enic->devcmd_lock);
1389
1390         for (i = 0; i < enic->intr_count; i++)
1391                 vnic_intr_unmask(&enic->intr[i]);
1392
1393         enic_notify_timer_start(enic);
1394
1395         return 0;
1396
1397 err_out_notify_unset:
1398         spin_lock(&enic->devcmd_lock);
1399         vnic_dev_notify_unset(enic->vdev);
1400         spin_unlock(&enic->devcmd_lock);
1401 err_out_free_intr:
1402         enic_free_intr(enic);
1403
1404         return err;
1405 }
1406
1407 /* rtnl lock is held, process context */
1408 static int enic_stop(struct net_device *netdev)
1409 {
1410         struct enic *enic = netdev_priv(netdev);
1411         unsigned int i;
1412         int err;
1413
1414         for (i = 0; i < enic->intr_count; i++)
1415                 vnic_intr_mask(&enic->intr[i]);
1416
1417         enic_synchronize_irqs(enic);
1418
1419         del_timer_sync(&enic->notify_timer);
1420
1421         spin_lock(&enic->devcmd_lock);
1422         vnic_dev_disable(enic->vdev);
1423         spin_unlock(&enic->devcmd_lock);
1424         napi_disable(&enic->napi);
1425         netif_carrier_off(netdev);
1426         netif_tx_disable(netdev);
1427
1428         for (i = 0; i < enic->wq_count; i++) {
1429                 err = vnic_wq_disable(&enic->wq[i]);
1430                 if (err)
1431                         return err;
1432         }
1433         for (i = 0; i < enic->rq_count; i++) {
1434                 err = vnic_rq_disable(&enic->rq[i]);
1435                 if (err)
1436                         return err;
1437         }
1438
1439         spin_lock(&enic->devcmd_lock);
1440         vnic_dev_notify_unset(enic->vdev);
1441         spin_unlock(&enic->devcmd_lock);
1442         enic_free_intr(enic);
1443
1444         for (i = 0; i < enic->wq_count; i++)
1445                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1446         for (i = 0; i < enic->rq_count; i++)
1447                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1448         for (i = 0; i < enic->cq_count; i++)
1449                 vnic_cq_clean(&enic->cq[i]);
1450         for (i = 0; i < enic->intr_count; i++)
1451                 vnic_intr_clean(&enic->intr[i]);
1452
1453         return 0;
1454 }
1455
1456 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1457 {
1458         struct enic *enic = netdev_priv(netdev);
1459         int running = netif_running(netdev);
1460
1461         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1462                 return -EINVAL;
1463
1464         if (running)
1465                 enic_stop(netdev);
1466
1467         netdev->mtu = new_mtu;
1468
1469         if (netdev->mtu > enic->port_mtu)
1470                 printk(KERN_WARNING PFX
1471                         "%s: interface MTU (%d) set higher "
1472                         "than port MTU (%d)\n",
1473                         netdev->name, netdev->mtu, enic->port_mtu);
1474
1475         if (running)
1476                 enic_open(netdev);
1477
1478         return 0;
1479 }
1480
1481 #ifdef CONFIG_NET_POLL_CONTROLLER
1482 static void enic_poll_controller(struct net_device *netdev)
1483 {
1484         struct enic *enic = netdev_priv(netdev);
1485         struct vnic_dev *vdev = enic->vdev;
1486
1487         switch (vnic_dev_get_intr_mode(vdev)) {
1488         case VNIC_DEV_INTR_MODE_MSIX:
1489                 enic_isr_msix_rq(enic->pdev->irq, enic);
1490                 enic_isr_msix_wq(enic->pdev->irq, enic);
1491                 break;
1492         case VNIC_DEV_INTR_MODE_MSI:
1493                 enic_isr_msi(enic->pdev->irq, enic);
1494                 break;
1495         case VNIC_DEV_INTR_MODE_INTX:
1496                 enic_isr_legacy(enic->pdev->irq, netdev);
1497                 break;
1498         default:
1499                 break;
1500         }
1501 }
1502 #endif
1503
1504 static int enic_dev_wait(struct vnic_dev *vdev,
1505         int (*start)(struct vnic_dev *, int),
1506         int (*finished)(struct vnic_dev *, int *),
1507         int arg)
1508 {
1509         unsigned long time;
1510         int done;
1511         int err;
1512
1513         BUG_ON(in_interrupt());
1514
1515         err = start(vdev, arg);
1516         if (err)
1517                 return err;
1518
1519         /* Wait for func to complete...2 seconds max
1520          */
1521
1522         time = jiffies + (HZ * 2);
1523         do {
1524
1525                 err = finished(vdev, &done);
1526                 if (err)
1527                         return err;
1528
1529                 if (done)
1530                         return 0;
1531
1532                 schedule_timeout_uninterruptible(HZ / 10);
1533
1534         } while (time_after(time, jiffies));
1535
1536         return -ETIMEDOUT;
1537 }
1538
1539 static int enic_dev_open(struct enic *enic)
1540 {
1541         int err;
1542
1543         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1544                 vnic_dev_open_done, 0);
1545         if (err)
1546                 printk(KERN_ERR PFX
1547                         "vNIC device open failed, err %d.\n", err);
1548
1549         return err;
1550 }
1551
1552 static int enic_dev_soft_reset(struct enic *enic)
1553 {
1554         int err;
1555
1556         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1557                 vnic_dev_soft_reset_done, 0);
1558         if (err)
1559                 printk(KERN_ERR PFX
1560                         "vNIC soft reset failed, err %d.\n", err);
1561
1562         return err;
1563 }
1564
1565 static int enic_set_niccfg(struct enic *enic)
1566 {
1567         const u8 rss_default_cpu = 0;
1568         const u8 rss_hash_type = 0;
1569         const u8 rss_hash_bits = 0;
1570         const u8 rss_base_cpu = 0;
1571         const u8 rss_enable = 0;
1572         const u8 tso_ipid_split_en = 0;
1573         const u8 ig_vlan_strip_en = 1;
1574
1575         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1576          */
1577
1578         return enic_set_nic_cfg(enic,
1579                 rss_default_cpu, rss_hash_type,
1580                 rss_hash_bits, rss_base_cpu,
1581                 rss_enable, tso_ipid_split_en,
1582                 ig_vlan_strip_en);
1583 }
1584
1585 static void enic_reset(struct work_struct *work)
1586 {
1587         struct enic *enic = container_of(work, struct enic, reset);
1588
1589         if (!netif_running(enic->netdev))
1590                 return;
1591
1592         rtnl_lock();
1593
1594         spin_lock(&enic->devcmd_lock);
1595         vnic_dev_hang_notify(enic->vdev);
1596         spin_unlock(&enic->devcmd_lock);
1597
1598         enic_stop(enic->netdev);
1599         enic_dev_soft_reset(enic);
1600         vnic_dev_init(enic->vdev, 0);
1601         enic_reset_mcaddrs(enic);
1602         enic_init_vnic_resources(enic);
1603         enic_set_niccfg(enic);
1604         enic_open(enic->netdev);
1605
1606         rtnl_unlock();
1607 }
1608
1609 static int enic_set_intr_mode(struct enic *enic)
1610 {
1611         unsigned int n = 1;
1612         unsigned int m = 1;
1613         unsigned int i;
1614
1615         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1616          * system capabilities.
1617          *
1618          * Try MSI-X first
1619          *
1620          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1621          * (the second to last INTR is used for WQ/RQ errors)
1622          * (the last INTR is used for notifications)
1623          */
1624
1625         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1626         for (i = 0; i < n + m + 2; i++)
1627                 enic->msix_entry[i].entry = i;
1628
1629         if (enic->config.intr_mode < 1 &&
1630             enic->rq_count >= n &&
1631             enic->wq_count >= m &&
1632             enic->cq_count >= n + m &&
1633             enic->intr_count >= n + m + 2 &&
1634             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1635
1636                 enic->rq_count = n;
1637                 enic->wq_count = m;
1638                 enic->cq_count = n + m;
1639                 enic->intr_count = n + m + 2;
1640
1641                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1642
1643                 return 0;
1644         }
1645
1646         /* Next try MSI
1647          *
1648          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1649          */
1650
1651         if (enic->config.intr_mode < 2 &&
1652             enic->rq_count >= 1 &&
1653             enic->wq_count >= 1 &&
1654             enic->cq_count >= 2 &&
1655             enic->intr_count >= 1 &&
1656             !pci_enable_msi(enic->pdev)) {
1657
1658                 enic->rq_count = 1;
1659                 enic->wq_count = 1;
1660                 enic->cq_count = 2;
1661                 enic->intr_count = 1;
1662
1663                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1664
1665                 return 0;
1666         }
1667
1668         /* Next try INTx
1669          *
1670          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1671          * (the first INTR is used for WQ/RQ)
1672          * (the second INTR is used for WQ/RQ errors)
1673          * (the last INTR is used for notifications)
1674          */
1675
1676         if (enic->config.intr_mode < 3 &&
1677             enic->rq_count >= 1 &&
1678             enic->wq_count >= 1 &&
1679             enic->cq_count >= 2 &&
1680             enic->intr_count >= 3) {
1681
1682                 enic->rq_count = 1;
1683                 enic->wq_count = 1;
1684                 enic->cq_count = 2;
1685                 enic->intr_count = 3;
1686
1687                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1688
1689                 return 0;
1690         }
1691
1692         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1693
1694         return -EINVAL;
1695 }
1696
1697 static void enic_clear_intr_mode(struct enic *enic)
1698 {
1699         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1700         case VNIC_DEV_INTR_MODE_MSIX:
1701                 pci_disable_msix(enic->pdev);
1702                 break;
1703         case VNIC_DEV_INTR_MODE_MSI:
1704                 pci_disable_msi(enic->pdev);
1705                 break;
1706         default:
1707                 break;
1708         }
1709
1710         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1711 }
1712
1713 static const struct net_device_ops enic_netdev_ops = {
1714         .ndo_open               = enic_open,
1715         .ndo_stop               = enic_stop,
1716         .ndo_start_xmit         = enic_hard_start_xmit,
1717         .ndo_get_stats          = enic_get_stats,
1718         .ndo_validate_addr      = eth_validate_addr,
1719         .ndo_set_mac_address    = eth_mac_addr,
1720         .ndo_set_multicast_list = enic_set_multicast_list,
1721         .ndo_change_mtu         = enic_change_mtu,
1722         .ndo_vlan_rx_register   = enic_vlan_rx_register,
1723         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
1724         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
1725         .ndo_tx_timeout         = enic_tx_timeout,
1726 #ifdef CONFIG_NET_POLL_CONTROLLER
1727         .ndo_poll_controller    = enic_poll_controller,
1728 #endif
1729 };
1730
1731 void enic_dev_deinit(struct enic *enic)
1732 {
1733         netif_napi_del(&enic->napi);
1734         enic_free_vnic_resources(enic);
1735         enic_clear_intr_mode(enic);
1736 }
1737
1738 int enic_dev_init(struct enic *enic)
1739 {
1740         struct net_device *netdev = enic->netdev;
1741         int err;
1742
1743         /* Get vNIC configuration
1744          */
1745
1746         err = enic_get_vnic_config(enic);
1747         if (err) {
1748                 printk(KERN_ERR PFX
1749                         "Get vNIC configuration failed, aborting.\n");
1750                 return err;
1751         }
1752
1753         /* Get available resource counts
1754          */
1755
1756         enic_get_res_counts(enic);
1757
1758         /* Set interrupt mode based on resource counts and system
1759          * capabilities
1760          */
1761
1762         err = enic_set_intr_mode(enic);
1763         if (err) {
1764                 printk(KERN_ERR PFX
1765                         "Failed to set intr mode, aborting.\n");
1766                 return err;
1767         }
1768
1769         /* Allocate and configure vNIC resources
1770          */
1771
1772         err = enic_alloc_vnic_resources(enic);
1773         if (err) {
1774                 printk(KERN_ERR PFX
1775                         "Failed to alloc vNIC resources, aborting.\n");
1776                 goto err_out_free_vnic_resources;
1777         }
1778
1779         enic_init_vnic_resources(enic);
1780
1781         err = enic_set_rq_alloc_buf(enic);
1782         if (err) {
1783                 printk(KERN_ERR PFX
1784                         "Failed to set RQ buffer allocator, aborting.\n");
1785                 goto err_out_free_vnic_resources;
1786         }
1787
1788         err = enic_set_niccfg(enic);
1789         if (err) {
1790                 printk(KERN_ERR PFX
1791                         "Failed to config nic, aborting.\n");
1792                 goto err_out_free_vnic_resources;
1793         }
1794
1795         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1796         default:
1797                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1798                 break;
1799         case VNIC_DEV_INTR_MODE_MSIX:
1800                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1801                 break;
1802         }
1803
1804         return 0;
1805
1806 err_out_free_vnic_resources:
1807         enic_clear_intr_mode(enic);
1808         enic_free_vnic_resources(enic);
1809
1810         return err;
1811 }
1812
1813 static void enic_iounmap(struct enic *enic)
1814 {
1815         unsigned int i;
1816
1817         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
1818                 if (enic->bar[i].vaddr)
1819                         iounmap(enic->bar[i].vaddr);
1820 }
1821
1822 static int __devinit enic_probe(struct pci_dev *pdev,
1823         const struct pci_device_id *ent)
1824 {
1825         struct net_device *netdev;
1826         struct enic *enic;
1827         int using_dac = 0;
1828         unsigned int i;
1829         int err;
1830
1831         /* Allocate net device structure and initialize.  Private
1832          * instance data is initialized to zero.
1833          */
1834
1835         netdev = alloc_etherdev(sizeof(struct enic));
1836         if (!netdev) {
1837                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1838                 return -ENOMEM;
1839         }
1840
1841         pci_set_drvdata(pdev, netdev);
1842
1843         SET_NETDEV_DEV(netdev, &pdev->dev);
1844
1845         enic = netdev_priv(netdev);
1846         enic->netdev = netdev;
1847         enic->pdev = pdev;
1848
1849         /* Setup PCI resources
1850          */
1851
1852         err = pci_enable_device(pdev);
1853         if (err) {
1854                 printk(KERN_ERR PFX
1855                         "Cannot enable PCI device, aborting.\n");
1856                 goto err_out_free_netdev;
1857         }
1858
1859         err = pci_request_regions(pdev, DRV_NAME);
1860         if (err) {
1861                 printk(KERN_ERR PFX
1862                         "Cannot request PCI regions, aborting.\n");
1863                 goto err_out_disable_device;
1864         }
1865
1866         pci_set_master(pdev);
1867
1868         /* Query PCI controller on system for DMA addressing
1869          * limitation for the device.  Try 40-bit first, and
1870          * fail to 32-bit.
1871          */
1872
1873         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
1874         if (err) {
1875                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1876                 if (err) {
1877                         printk(KERN_ERR PFX
1878                                 "No usable DMA configuration, aborting.\n");
1879                         goto err_out_release_regions;
1880                 }
1881                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1882                 if (err) {
1883                         printk(KERN_ERR PFX
1884                                 "Unable to obtain 32-bit DMA "
1885                                 "for consistent allocations, aborting.\n");
1886                         goto err_out_release_regions;
1887                 }
1888         } else {
1889                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
1890                 if (err) {
1891                         printk(KERN_ERR PFX
1892                                 "Unable to obtain 40-bit DMA "
1893                                 "for consistent allocations, aborting.\n");
1894                         goto err_out_release_regions;
1895                 }
1896                 using_dac = 1;
1897         }
1898
1899         /* Map vNIC resources from BAR0-5
1900          */
1901
1902         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
1903                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
1904                         continue;
1905                 enic->bar[i].len = pci_resource_len(pdev, i);
1906                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
1907                 if (!enic->bar[i].vaddr) {
1908                         printk(KERN_ERR PFX
1909                                 "Cannot memory-map BAR %d, aborting.\n", i);
1910                         err = -ENODEV;
1911                         goto err_out_iounmap;
1912                 }
1913                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
1914         }
1915
1916         /* Register vNIC device
1917          */
1918
1919         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
1920                 ARRAY_SIZE(enic->bar));
1921         if (!enic->vdev) {
1922                 printk(KERN_ERR PFX
1923                         "vNIC registration failed, aborting.\n");
1924                 err = -ENODEV;
1925                 goto err_out_iounmap;
1926         }
1927
1928         /* Issue device open to get device in known state
1929          */
1930
1931         err = enic_dev_open(enic);
1932         if (err) {
1933                 printk(KERN_ERR PFX
1934                         "vNIC dev open failed, aborting.\n");
1935                 goto err_out_vnic_unregister;
1936         }
1937
1938         /* Issue device init to initialize the vnic-to-switch link.
1939          * We'll start with carrier off and wait for link UP
1940          * notification later to turn on carrier.  We don't need
1941          * to wait here for the vnic-to-switch link initialization
1942          * to complete; link UP notification is the indication that
1943          * the process is complete.
1944          */
1945
1946         netif_carrier_off(netdev);
1947
1948         err = vnic_dev_init(enic->vdev, 0);
1949         if (err) {
1950                 printk(KERN_ERR PFX
1951                         "vNIC dev init failed, aborting.\n");
1952                 goto err_out_dev_close;
1953         }
1954
1955         err = enic_dev_init(enic);
1956         if (err) {
1957                 printk(KERN_ERR PFX
1958                         "Device initialization failed, aborting.\n");
1959                 goto err_out_dev_close;
1960         }
1961
1962         /* Setup notification timer, HW reset task, and locks
1963          */
1964
1965         init_timer(&enic->notify_timer);
1966         enic->notify_timer.function = enic_notify_timer;
1967         enic->notify_timer.data = (unsigned long)enic;
1968
1969         INIT_WORK(&enic->reset, enic_reset);
1970
1971         for (i = 0; i < enic->wq_count; i++)
1972                 spin_lock_init(&enic->wq_lock[i]);
1973
1974         spin_lock_init(&enic->devcmd_lock);
1975
1976         /* Register net device
1977          */
1978
1979         enic->port_mtu = enic->config.mtu;
1980         (void)enic_change_mtu(netdev, enic->port_mtu);
1981
1982         err = enic_set_mac_addr(netdev, enic->mac_addr);
1983         if (err) {
1984                 printk(KERN_ERR PFX
1985                         "Invalid MAC address, aborting.\n");
1986                 goto err_out_dev_deinit;
1987         }
1988
1989         netdev->netdev_ops = &enic_netdev_ops;
1990         netdev->watchdog_timeo = 2 * HZ;
1991         netdev->ethtool_ops = &enic_ethtool_ops;
1992
1993         netdev->features |= NETIF_F_HW_VLAN_TX |
1994                 NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_FILTER;
1995         if (ENIC_SETTING(enic, TXCSUM))
1996                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1997         if (ENIC_SETTING(enic, TSO))
1998                 netdev->features |= NETIF_F_TSO |
1999                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2000         if (ENIC_SETTING(enic, LRO))
2001                 netdev->features |= NETIF_F_LRO;
2002         if (using_dac)
2003                 netdev->features |= NETIF_F_HIGHDMA;
2004
2005         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2006
2007         enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
2008         enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
2009         enic->lro_mgr.lro_arr = enic->lro_desc;
2010         enic->lro_mgr.get_skb_header = enic_get_skb_header;
2011         enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
2012         enic->lro_mgr.dev = netdev;
2013         enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
2014         enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
2015
2016         err = register_netdev(netdev);
2017         if (err) {
2018                 printk(KERN_ERR PFX
2019                         "Cannot register net device, aborting.\n");
2020                 goto err_out_dev_deinit;
2021         }
2022
2023         return 0;
2024
2025 err_out_dev_deinit:
2026         enic_dev_deinit(enic);
2027 err_out_dev_close:
2028         vnic_dev_close(enic->vdev);
2029 err_out_vnic_unregister:
2030         vnic_dev_unregister(enic->vdev);
2031 err_out_iounmap:
2032         enic_iounmap(enic);
2033 err_out_release_regions:
2034         pci_release_regions(pdev);
2035 err_out_disable_device:
2036         pci_disable_device(pdev);
2037 err_out_free_netdev:
2038         pci_set_drvdata(pdev, NULL);
2039         free_netdev(netdev);
2040
2041         return err;
2042 }
2043
2044 static void __devexit enic_remove(struct pci_dev *pdev)
2045 {
2046         struct net_device *netdev = pci_get_drvdata(pdev);
2047
2048         if (netdev) {
2049                 struct enic *enic = netdev_priv(netdev);
2050
2051                 flush_scheduled_work();
2052                 unregister_netdev(netdev);
2053                 enic_dev_deinit(enic);
2054                 vnic_dev_close(enic->vdev);
2055                 vnic_dev_unregister(enic->vdev);
2056                 enic_iounmap(enic);
2057                 pci_release_regions(pdev);
2058                 pci_disable_device(pdev);
2059                 pci_set_drvdata(pdev, NULL);
2060                 free_netdev(netdev);
2061         }
2062 }
2063
2064 static struct pci_driver enic_driver = {
2065         .name = DRV_NAME,
2066         .id_table = enic_id_table,
2067         .probe = enic_probe,
2068         .remove = __devexit_p(enic_remove),
2069 };
2070
2071 static int __init enic_init_module(void)
2072 {
2073         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2074
2075         return pci_register_driver(&enic_driver);
2076 }
2077
2078 static void __exit enic_cleanup_module(void)
2079 {
2080         pci_unregister_driver(&enic_driver);
2081 }
2082
2083 module_init(enic_init_module);
2084 module_exit(enic_cleanup_module);