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 int enic_get_coalesce(struct net_device *netdev,
 265         struct ethtool_coalesce *ecmd)
 266 {
 267         struct enic *enic = netdev_priv(netdev);
 268
 269         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
 270         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
 271
 272         return 0;
 273 }
 274
 275 static int enic_set_coalesce(struct net_device *netdev,
 276         struct ethtool_coalesce *ecmd)
 277 {
 278         struct enic *enic = netdev_priv(netdev);
 279         u32 tx_coalesce_usecs;
 280         u32 rx_coalesce_usecs;
 281
 282         tx_coalesce_usecs = min_t(u32,
 283                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 284                 ecmd->tx_coalesce_usecs);
 285         rx_coalesce_usecs = min_t(u32,
 286                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 287                 ecmd->rx_coalesce_usecs);
 288
 289         switch (vnic_dev_get_intr_mode(enic->vdev)) {
 290         case VNIC_DEV_INTR_MODE_INTX:
 291                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 292                         return -EINVAL;
 293
 294                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_INTX_WQ_RQ],
 295                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 296                 break;
 297         case VNIC_DEV_INTR_MODE_MSI:
 298                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 299                         return -EINVAL;
 300
 301                 vnic_intr_coalescing_timer_set(&enic->intr[0],
 302                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 303                 break;
 304         case VNIC_DEV_INTR_MODE_MSIX:
 305                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_WQ],
 306                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 307                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_RQ],
 308                         INTR_COALESCE_USEC_TO_HW(rx_coalesce_usecs));
 309                 break;
 310         default:
 311                 break;
 312         }
 313
 314         enic->tx_coalesce_usecs = tx_coalesce_usecs;
 315         enic->rx_coalesce_usecs = rx_coalesce_usecs;
 316
 317         return 0;
 318 }
 319
 320 static const struct ethtool_ops enic_ethtool_ops = {
 321         .get_settings = enic_get_settings,
 322         .get_drvinfo = enic_get_drvinfo,
 323         .get_msglevel = enic_get_msglevel,
 324         .set_msglevel = enic_set_msglevel,
 325         .get_link = ethtool_op_get_link,
 326         .get_strings = enic_get_strings,
 327         .get_sset_count = enic_get_sset_count,
 328         .get_ethtool_stats = enic_get_ethtool_stats,
 329         .get_rx_csum = enic_get_rx_csum,
 330         .set_rx_csum = enic_set_rx_csum,
 331         .get_tx_csum = ethtool_op_get_tx_csum,
 332         .set_tx_csum = enic_set_tx_csum,
 333         .get_sg = ethtool_op_get_sg,
 334         .set_sg = ethtool_op_set_sg,
 335         .get_tso = ethtool_op_get_tso,
 336         .set_tso = enic_set_tso,
 337         .get_coalesce = enic_get_coalesce,
 338         .set_coalesce = enic_set_coalesce,
 339         .get_flags = ethtool_op_get_flags,
 340         .set_flags = ethtool_op_set_flags,
 341 };
 342
 343 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 344 {
 345         struct enic *enic = vnic_dev_priv(wq->vdev);
 346
 347         if (buf->sop)
 348                 pci_unmap_single(enic->pdev, buf->dma_addr,
 349                         buf->len, PCI_DMA_TODEVICE);
 350         else
 351                 pci_unmap_page(enic->pdev, buf->dma_addr,
 352                         buf->len, PCI_DMA_TODEVICE);
 353
 354         if (buf->os_buf)
 355                 dev_kfree_skb_any(buf->os_buf);
 356 }
 357
 358 static void enic_wq_free_buf(struct vnic_wq *wq,
 359         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 360 {
 361         enic_free_wq_buf(wq, buf);
 362 }
 363
 364 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 365         u8 type, u16 q_number, u16 completed_index, void *opaque)
 366 {
 367         struct enic *enic = vnic_dev_priv(vdev);
 368
 369         spin_lock(&enic->wq_lock[q_number]);
 370
 371         vnic_wq_service(&enic->wq[q_number], cq_desc,
 372                 completed_index, enic_wq_free_buf,
 373                 opaque);
 374
 375         if (netif_queue_stopped(enic->netdev) &&
 376             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 377             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 378                 netif_wake_queue(enic->netdev);
 379
 380         spin_unlock(&enic->wq_lock[q_number]);
 381
 382         return 0;
 383 }
 384
 385 static void enic_log_q_error(struct enic *enic)
 386 {
 387         unsigned int i;
 388         u32 error_status;
 389
 390         for (i = 0; i < enic->wq_count; i++) {
 391                 error_status = vnic_wq_error_status(&enic->wq[i]);
 392                 if (error_status)
 393                         printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
 394                                 enic->netdev->name, i, error_status);
 395         }
 396
 397         for (i = 0; i < enic->rq_count; i++) {
 398                 error_status = vnic_rq_error_status(&enic->rq[i]);
 399                 if (error_status)
 400                         printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
 401                                 enic->netdev->name, i, error_status);
 402         }
 403 }
 404
 405 static void enic_link_check(struct enic *enic)
 406 {
 407         int link_status = vnic_dev_link_status(enic->vdev);
 408         int carrier_ok = netif_carrier_ok(enic->netdev);
 409
 410         if (link_status && !carrier_ok) {
 411                 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
 412                 netif_carrier_on(enic->netdev);
 413         } else if (!link_status && carrier_ok) {
 414                 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
 415                 netif_carrier_off(enic->netdev);
 416         }
 417 }
 418
 419 static void enic_mtu_check(struct enic *enic)
 420 {
 421         u32 mtu = vnic_dev_mtu(enic->vdev);
 422
 423         if (mtu && mtu != enic->port_mtu) {
 424                 enic->port_mtu = mtu;
 425                 if (mtu < enic->netdev->mtu)
 426                         printk(KERN_WARNING PFX
 427                                 "%s: interface MTU (%d) set higher "
 428                                 "than switch port MTU (%d)\n",
 429                                 enic->netdev->name, enic->netdev->mtu, mtu);
 430         }
 431 }
 432
 433 static void enic_msglvl_check(struct enic *enic)
 434 {
 435         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 436
 437         if (msg_enable != enic->msg_enable) {
 438                 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
 439                         enic->netdev->name, enic->msg_enable, msg_enable);
 440                 enic->msg_enable = msg_enable;
 441         }
 442 }
 443
 444 static void enic_notify_check(struct enic *enic)
 445 {
 446         enic_msglvl_check(enic);
 447         enic_mtu_check(enic);
 448         enic_link_check(enic);
 449 }
 450
 451 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 452
 453 static irqreturn_t enic_isr_legacy(int irq, void *data)
 454 {
 455         struct net_device *netdev = data;
 456         struct enic *enic = netdev_priv(netdev);
 457         u32 pba;
 458
 459         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
 460
 461         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 462         if (!pba) {
 463                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 464                 return IRQ_NONE;        /* not our interrupt */
 465         }
 466
 467         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
 468                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
 469                 enic_notify_check(enic);
 470         }
 471
 472         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
 473                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
 474                 enic_log_q_error(enic);
 475                 /* schedule recovery from WQ/RQ error */
 476                 schedule_work(&enic->reset);
 477                 return IRQ_HANDLED;
 478         }
 479
 480         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
 481                 if (napi_schedule_prep(&enic->napi))
 482                         __napi_schedule(&enic->napi);
 483         } else {
 484                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 485         }
 486
 487         return IRQ_HANDLED;
 488 }
 489
 490 static irqreturn_t enic_isr_msi(int irq, void *data)
 491 {
 492         struct enic *enic = data;
 493
 494         /* With MSI, there is no sharing of interrupts, so this is
 495          * our interrupt and there is no need to ack it.  The device
 496          * is not providing per-vector masking, so the OS will not
 497          * write to PCI config space to mask/unmask the interrupt.
 498          * We're using mask_on_assertion for MSI, so the device
 499          * automatically masks the interrupt when the interrupt is
 500          * generated.  Later, when exiting polling, the interrupt
 501          * will be unmasked (see enic_poll).
 502          *
 503          * Also, the device uses the same PCIe Traffic Class (TC)
 504          * for Memory Write data and MSI, so there are no ordering
 505          * issues; the MSI will always arrive at the Root Complex
 506          * _after_ corresponding Memory Writes (i.e. descriptor
 507          * writes).
 508          */
 509
 510         napi_schedule(&enic->napi);
 511
 512         return IRQ_HANDLED;
 513 }
 514
 515 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
 516 {
 517         struct enic *enic = data;
 518
 519         /* schedule NAPI polling for RQ cleanup */
 520         napi_schedule(&enic->napi);
 521
 522         return IRQ_HANDLED;
 523 }
 524
 525 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
 526 {
 527         struct enic *enic = data;
 528         unsigned int wq_work_to_do = -1; /* no limit */
 529         unsigned int wq_work_done;
 530
 531         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
 532                 wq_work_to_do, enic_wq_service, NULL);
 533
 534         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
 535                 wq_work_done,
 536                 1 /* unmask intr */,
 537                 1 /* reset intr timer */);
 538
 539         return IRQ_HANDLED;
 540 }
 541
 542 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 543 {
 544         struct enic *enic = data;
 545
 546         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
 547
 548         enic_log_q_error(enic);
 549
 550         /* schedule recovery from WQ/RQ error */
 551         schedule_work(&enic->reset);
 552
 553         return IRQ_HANDLED;
 554 }
 555
 556 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 557 {
 558         struct enic *enic = data;
 559
 560         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
 561         enic_notify_check(enic);
 562
 563         return IRQ_HANDLED;
 564 }
 565
 566 static inline void enic_queue_wq_skb_cont(struct enic *enic,
 567         struct vnic_wq *wq, struct sk_buff *skb,
 568         unsigned int len_left)
 569 {
 570         skb_frag_t *frag;
 571
 572         /* Queue additional data fragments */
 573         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 574                 len_left -= frag->size;
 575                 enic_queue_wq_desc_cont(wq, skb,
 576                         pci_map_page(enic->pdev, frag->page,
 577                                 frag->page_offset, frag->size,
 578                                 PCI_DMA_TODEVICE),
 579                         frag->size,
 580                         (len_left == 0));       /* EOP? */
 581         }
 582 }
 583
 584 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
 585         struct vnic_wq *wq, struct sk_buff *skb,
 586         int vlan_tag_insert, unsigned int vlan_tag)
 587 {
 588         unsigned int head_len = skb_headlen(skb);
 589         unsigned int len_left = skb->len - head_len;
 590         int eop = (len_left == 0);
 591
 592         /* Queue the main skb fragment. The fragments are no larger
 593          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 594          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 595          * per fragment is queued.
 596          */
 597         enic_queue_wq_desc(wq, skb,
 598                 pci_map_single(enic->pdev, skb->data,
 599                         head_len, PCI_DMA_TODEVICE),
 600                 head_len,
 601                 vlan_tag_insert, vlan_tag,
 602                 eop);
 603
 604         if (!eop)
 605                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 606 }
 607
 608 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
 609         struct vnic_wq *wq, struct sk_buff *skb,
 610         int vlan_tag_insert, unsigned int vlan_tag)
 611 {
 612         unsigned int head_len = skb_headlen(skb);
 613         unsigned int len_left = skb->len - head_len;
 614         unsigned int hdr_len = skb_transport_offset(skb);
 615         unsigned int csum_offset = hdr_len + skb->csum_offset;
 616         int eop = (len_left == 0);
 617
 618         /* Queue the main skb fragment. The fragments are no larger
 619          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 620          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 621          * per fragment is queued.
 622          */
 623         enic_queue_wq_desc_csum_l4(wq, skb,
 624                 pci_map_single(enic->pdev, skb->data,
 625                         head_len, PCI_DMA_TODEVICE),
 626                 head_len,
 627                 csum_offset,
 628                 hdr_len,
 629                 vlan_tag_insert, vlan_tag,
 630                 eop);
 631
 632         if (!eop)
 633                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 634 }
 635
 636 static inline void enic_queue_wq_skb_tso(struct enic *enic,
 637         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
 638         int vlan_tag_insert, unsigned int vlan_tag)
 639 {
 640         unsigned int frag_len_left = skb_headlen(skb);
 641         unsigned int len_left = skb->len - frag_len_left;
 642         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 643         int eop = (len_left == 0);
 644         unsigned int len;
 645         dma_addr_t dma_addr;
 646         unsigned int offset = 0;
 647         skb_frag_t *frag;
 648
 649         /* Preload TCP csum field with IP pseudo hdr calculated
 650          * with IP length set to zero.  HW will later add in length
 651          * to each TCP segment resulting from the TSO.
 652          */
 653
 654         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 655                 ip_hdr(skb)->check = 0;
 656                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 657                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 658         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 659                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 660                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 661         }
 662
 663         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 664          * for the main skb fragment
 665          */
 666         while (frag_len_left) {
 667                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 668                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
 669                                 len, PCI_DMA_TODEVICE);
 670                 enic_queue_wq_desc_tso(wq, skb,
 671                         dma_addr,
 672                         len,
 673                         mss, hdr_len,
 674                         vlan_tag_insert, vlan_tag,
 675                         eop && (len == frag_len_left));
 676                 frag_len_left -= len;
 677                 offset += len;
 678         }
 679
 680         if (eop)
 681                 return;
 682
 683         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 684          * for additional data fragments
 685          */
 686         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 687                 len_left -= frag->size;
 688                 frag_len_left = frag->size;
 689                 offset = frag->page_offset;
 690
 691                 while (frag_len_left) {
 692                         len = min(frag_len_left,
 693                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 694                         dma_addr = pci_map_page(enic->pdev, frag->page,
 695                                 offset, len,
 696                                 PCI_DMA_TODEVICE);
 697                         enic_queue_wq_desc_cont(wq, skb,
 698                                 dma_addr,
 699                                 len,
 700                                 (len_left == 0) &&
 701                                 (len == frag_len_left));        /* EOP? */
 702                         frag_len_left -= len;
 703                         offset += len;
 704                 }
 705         }
 706 }
 707
 708 static inline void enic_queue_wq_skb(struct enic *enic,
 709         struct vnic_wq *wq, struct sk_buff *skb)
 710 {
 711         unsigned int mss = skb_shinfo(skb)->gso_size;
 712         unsigned int vlan_tag = 0;
 713         int vlan_tag_insert = 0;
 714
 715         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
 716                 /* VLAN tag from trunking driver */
 717                 vlan_tag_insert = 1;
 718                 vlan_tag = vlan_tx_tag_get(skb);
 719         }
 720
 721         if (mss)
 722                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
 723                         vlan_tag_insert, vlan_tag);
 724         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 725                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
 726                         vlan_tag_insert, vlan_tag);
 727         else
 728                 enic_queue_wq_skb_vlan(enic, wq, skb,
 729                         vlan_tag_insert, vlan_tag);
 730 }
 731
 732 /* netif_tx_lock held, process context with BHs disabled, or BH */
 733 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 734         struct net_device *netdev)
 735 {
 736         struct enic *enic = netdev_priv(netdev);
 737         struct vnic_wq *wq = &enic->wq[0];
 738         unsigned long flags;
 739
 740         if (skb->len <= 0) {
 741                 dev_kfree_skb(skb);
 742                 return NETDEV_TX_OK;
 743         }
 744
 745         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 746          * which is very likely.  In the off chance it's going to take
 747          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 748          */
 749
 750         if (skb_shinfo(skb)->gso_size == 0 &&
 751             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 752             skb_linearize(skb)) {
 753                 dev_kfree_skb(skb);
 754                 return NETDEV_TX_OK;
 755         }
 756
 757         spin_lock_irqsave(&enic->wq_lock[0], flags);
 758
 759         if (vnic_wq_desc_avail(wq) <
 760             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 761                 netif_stop_queue(netdev);
 762                 /* This is a hard error, log it */
 763                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
 764                         "queue awake!\n", netdev->name);
 765                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 766                 return NETDEV_TX_BUSY;
 767         }
 768
 769         enic_queue_wq_skb(enic, wq, skb);
 770
 771         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 772                 netif_stop_queue(netdev);
 773
 774         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 775
 776         return NETDEV_TX_OK;
 777 }
 778
 779 /* dev_base_lock rwlock held, nominally process context */
 780 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
 781 {
 782         struct enic *enic = netdev_priv(netdev);
 783         struct net_device_stats *net_stats = &netdev->stats;
 784         struct vnic_stats *stats;
 785
 786         spin_lock(&enic->devcmd_lock);
 787         vnic_dev_stats_dump(enic->vdev, &stats);
 788         spin_unlock(&enic->devcmd_lock);
 789
 790         net_stats->tx_packets = stats->tx.tx_frames_ok;
 791         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 792         net_stats->tx_errors = stats->tx.tx_errors;
 793         net_stats->tx_dropped = stats->tx.tx_drops;
 794
 795         net_stats->rx_packets = stats->rx.rx_frames_ok;
 796         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 797         net_stats->rx_errors = stats->rx.rx_errors;
 798         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 799         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 800         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 801         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 802
 803         return net_stats;
 804 }
 805
 806 static void enic_reset_mcaddrs(struct enic *enic)
 807 {
 808         enic->mc_count = 0;
 809 }
 810
 811 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 812 {
 813         if (!is_valid_ether_addr(addr))
 814                 return -EADDRNOTAVAIL;
 815
 816         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 817
 818         return 0;
 819 }
 820
 821 /* netif_tx_lock held, BHs disabled */
 822 static void enic_set_multicast_list(struct net_device *netdev)
 823 {
 824         struct enic *enic = netdev_priv(netdev);
 825         struct dev_mc_list *list = netdev->mc_list;
 826         int directed = 1;
 827         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 828         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 829         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
 830         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 831             (netdev->mc_count > ENIC_MULTICAST_PERFECT_FILTERS);
 832         unsigned int flags = netdev->flags | (allmulti ? IFF_ALLMULTI : 0);
 833         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
 834         unsigned int mc_count = netdev->mc_count;
 835         unsigned int i, j;
 836
 837         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
 838                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
 839
 840         spin_lock(&enic->devcmd_lock);
 841
 842         if (enic->flags != flags) {
 843                 enic->flags = flags;
 844                 vnic_dev_packet_filter(enic->vdev, directed,
 845                         multicast, broadcast, promisc, allmulti);
 846         }
 847
 848         /* Is there an easier way?  Trying to minimize to
 849          * calls to add/del multicast addrs.  We keep the
 850          * addrs from the last call in enic->mc_addr and
 851          * look for changes to add/del.
 852          */
 853
 854         for (i = 0; list && i < mc_count; i++) {
 855                 memcpy(mc_addr[i], list->dmi_addr, ETH_ALEN);
 856                 list = list->next;
 857         }
 858
 859         for (i = 0; i < enic->mc_count; i++) {
 860                 for (j = 0; j < mc_count; j++)
 861                         if (compare_ether_addr(enic->mc_addr[i],
 862                                 mc_addr[j]) == 0)
 863                                 break;
 864                 if (j == mc_count)
 865                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
 866         }
 867
 868         for (i = 0; i < mc_count; i++) {
 869                 for (j = 0; j < enic->mc_count; j++)
 870                         if (compare_ether_addr(mc_addr[i],
 871                                 enic->mc_addr[j]) == 0)
 872                                 break;
 873                 if (j == enic->mc_count)
 874                         enic_add_multicast_addr(enic, mc_addr[i]);
 875         }
 876
 877         /* Save the list to compare against next time
 878          */
 879
 880         for (i = 0; i < mc_count; i++)
 881                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
 882
 883         enic->mc_count = mc_count;
 884
 885         spin_unlock(&enic->devcmd_lock);
 886 }
 887
 888 /* rtnl lock is held */
 889 static void enic_vlan_rx_register(struct net_device *netdev,
 890         struct vlan_group *vlan_group)
 891 {
 892         struct enic *enic = netdev_priv(netdev);
 893         enic->vlan_group = vlan_group;
 894 }
 895
 896 /* rtnl lock is held */
 897 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
 898 {
 899         struct enic *enic = netdev_priv(netdev);
 900
 901         spin_lock(&enic->devcmd_lock);
 902         enic_add_vlan(enic, vid);
 903         spin_unlock(&enic->devcmd_lock);
 904 }
 905
 906 /* rtnl lock is held */
 907 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
 908 {
 909         struct enic *enic = netdev_priv(netdev);
 910
 911         spin_lock(&enic->devcmd_lock);
 912         enic_del_vlan(enic, vid);
 913         spin_unlock(&enic->devcmd_lock);
 914 }
 915
 916 /* netif_tx_lock held, BHs disabled */
 917 static void enic_tx_timeout(struct net_device *netdev)
 918 {
 919         struct enic *enic = netdev_priv(netdev);
 920         schedule_work(&enic->reset);
 921 }
 922
 923 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
 924 {
 925         struct enic *enic = vnic_dev_priv(rq->vdev);
 926
 927         if (!buf->os_buf)
 928                 return;
 929
 930         pci_unmap_single(enic->pdev, buf->dma_addr,
 931                 buf->len, PCI_DMA_FROMDEVICE);
 932         dev_kfree_skb_any(buf->os_buf);
 933 }
 934
 935 static int enic_rq_alloc_buf(struct vnic_rq *rq)
 936 {
 937         struct enic *enic = vnic_dev_priv(rq->vdev);
 938         struct net_device *netdev = enic->netdev;
 939         struct sk_buff *skb;
 940         unsigned int len = netdev->mtu + ETH_HLEN;
 941         unsigned int os_buf_index = 0;
 942         dma_addr_t dma_addr;
 943
 944         skb = netdev_alloc_skb_ip_align(netdev, len);
 945         if (!skb)
 946                 return -ENOMEM;
 947
 948         dma_addr = pci_map_single(enic->pdev, skb->data,
 949                 len, PCI_DMA_FROMDEVICE);
 950
 951         enic_queue_rq_desc(rq, skb, os_buf_index,
 952                 dma_addr, len);
 953
 954         return 0;
 955 }
 956
 957 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
 958 {
 959         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
 960
 961         if (vnic_rq_posting_soon(rq)) {
 962
 963                 /* SW workaround for A0 HW erratum: if we're just about
 964                  * to write posted_index, insert a dummy desc
 965                  * of type resvd
 966                  */
 967
 968                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
 969                 vnic_rq_post(rq, 0, 0, 0, 0);
 970         } else {
 971                 return enic_rq_alloc_buf(rq);
 972         }
 973
 974         return 0;
 975 }
 976
 977 static int enic_set_rq_alloc_buf(struct enic *enic)
 978 {
 979         enum vnic_dev_hw_version hw_ver;
 980         int err;
 981
 982         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
 983         if (err)
 984                 return err;
 985
 986         switch (hw_ver) {
 987         case VNIC_DEV_HW_VER_A1:
 988                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
 989                 break;
 990         case VNIC_DEV_HW_VER_A2:
 991         case VNIC_DEV_HW_VER_UNKNOWN:
 992                 enic->rq_alloc_buf = enic_rq_alloc_buf;
 993                 break;
 994         default:
 995                 return -ENODEV;
 996         }
 997
 998         return 0;
 999 }
1000
1001 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
1002         void **tcph, u64 *hdr_flags, void *priv)
1003 {
1004         struct cq_enet_rq_desc *cq_desc = priv;
1005         unsigned int ip_len;
1006         struct iphdr *iph;
1007
1008         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1009         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1010         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1011         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1012         u8 packet_error;
1013         u16 q_number, completed_index, bytes_written, vlan, checksum;
1014         u32 rss_hash;
1015
1016         cq_enet_rq_desc_dec(cq_desc,
1017                 &type, &color, &q_number, &completed_index,
1018                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1019                 &csum_not_calc, &rss_hash, &bytes_written,
1020                 &packet_error, &vlan_stripped, &vlan, &checksum,
1021                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1022                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1023                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1024                 &fcs_ok);
1025
1026         if (!(ipv4 && tcp && !ipv4_fragment))
1027                 return -1;
1028
1029         skb_reset_network_header(skb);
1030         iph = ip_hdr(skb);
1031
1032         ip_len = ip_hdrlen(skb);
1033         skb_set_transport_header(skb, ip_len);
1034
1035         /* check if ip header and tcp header are complete */
1036         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
1037                 return -1;
1038
1039         *hdr_flags = LRO_IPV4 | LRO_TCP;
1040         *tcph = tcp_hdr(skb);
1041         *iphdr = iph;
1042
1043         return 0;
1044 }
1045
1046 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1047         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1048         int skipped, void *opaque)
1049 {
1050         struct enic *enic = vnic_dev_priv(rq->vdev);
1051         struct net_device *netdev = enic->netdev;
1052         struct sk_buff *skb;
1053
1054         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1055         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1056         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1057         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1058         u8 packet_error;
1059         u16 q_number, completed_index, bytes_written, vlan, checksum;
1060         u32 rss_hash;
1061
1062         if (skipped)
1063                 return;
1064
1065         skb = buf->os_buf;
1066         prefetch(skb->data - NET_IP_ALIGN);
1067         pci_unmap_single(enic->pdev, buf->dma_addr,
1068                 buf->len, PCI_DMA_FROMDEVICE);
1069
1070         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1071                 &type, &color, &q_number, &completed_index,
1072                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1073                 &csum_not_calc, &rss_hash, &bytes_written,
1074                 &packet_error, &vlan_stripped, &vlan, &checksum,
1075                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1076                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1077                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1078                 &fcs_ok);
1079
1080         if (packet_error) {
1081
1082                 if (!fcs_ok) {
1083                         if (bytes_written > 0)
1084                                 enic->rq_bad_fcs++;
1085                         else if (bytes_written == 0)
1086                                 enic->rq_truncated_pkts++;
1087                 }
1088
1089                 dev_kfree_skb_any(skb);
1090
1091                 return;
1092         }
1093
1094         if (eop && bytes_written > 0) {
1095
1096                 /* Good receive
1097                  */
1098
1099                 skb_put(skb, bytes_written);
1100                 skb->protocol = eth_type_trans(skb, netdev);
1101
1102                 if (enic->csum_rx_enabled && !csum_not_calc) {
1103                         skb->csum = htons(checksum);
1104                         skb->ip_summed = CHECKSUM_COMPLETE;
1105                 }
1106
1107                 skb->dev = netdev;
1108
1109                 if (enic->vlan_group && vlan_stripped) {
1110
1111                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1112                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
1113                                         skb, enic->vlan_group,
1114                                         vlan, cq_desc);
1115                         else
1116                                 vlan_hwaccel_receive_skb(skb,
1117                                         enic->vlan_group, vlan);
1118
1119                 } else {
1120
1121                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1122                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
1123                         else
1124                                 netif_receive_skb(skb);
1125
1126                 }
1127
1128         } else {
1129
1130                 /* Buffer overflow
1131                  */
1132
1133                 dev_kfree_skb_any(skb);
1134         }
1135 }
1136
1137 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1138         u8 type, u16 q_number, u16 completed_index, void *opaque)
1139 {
1140         struct enic *enic = vnic_dev_priv(vdev);
1141
1142         vnic_rq_service(&enic->rq[q_number], cq_desc,
1143                 completed_index, VNIC_RQ_RETURN_DESC,
1144                 enic_rq_indicate_buf, opaque);
1145
1146         return 0;
1147 }
1148
1149 static int enic_poll(struct napi_struct *napi, int budget)
1150 {
1151         struct enic *enic = container_of(napi, struct enic, napi);
1152         struct net_device *netdev = enic->netdev;
1153         unsigned int rq_work_to_do = budget;
1154         unsigned int wq_work_to_do = -1; /* no limit */
1155         unsigned int  work_done, rq_work_done, wq_work_done;
1156         int err;
1157
1158         /* Service RQ (first) and WQ
1159          */
1160
1161         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1162                 rq_work_to_do, enic_rq_service, NULL);
1163
1164         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1165                 wq_work_to_do, enic_wq_service, NULL);
1166
1167         /* Accumulate intr event credits for this polling
1168          * cycle.  An intr event is the completion of a
1169          * a WQ or RQ packet.
1170          */
1171
1172         work_done = rq_work_done + wq_work_done;
1173
1174         if (work_done > 0)
1175                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1176                         work_done,
1177                         0 /* don't unmask intr */,
1178                         0 /* don't reset intr timer */);
1179
1180         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1181
1182         /* Buffer allocation failed. Stay in polling
1183          * mode so we can try to fill the ring again.
1184          */
1185
1186         if (err)
1187                 rq_work_done = rq_work_to_do;
1188
1189         if (rq_work_done < rq_work_to_do) {
1190
1191                 /* Some work done, but not enough to stay in polling,
1192                  * flush all LROs and exit polling
1193                  */
1194
1195                 if (netdev->features & NETIF_F_LRO)
1196                         lro_flush_all(&enic->lro_mgr);
1197
1198                 napi_complete(napi);
1199                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1200         }
1201
1202         return rq_work_done;
1203 }
1204
1205 static int enic_poll_msix(struct napi_struct *napi, int budget)
1206 {
1207         struct enic *enic = container_of(napi, struct enic, napi);
1208         struct net_device *netdev = enic->netdev;
1209         unsigned int work_to_do = budget;
1210         unsigned int work_done;
1211         int err;
1212
1213         /* Service RQ
1214          */
1215
1216         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1217                 work_to_do, enic_rq_service, NULL);
1218
1219         /* Return intr event credits for this polling
1220          * cycle.  An intr event is the completion of a
1221          * RQ packet.
1222          */
1223
1224         if (work_done > 0)
1225                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1226                         work_done,
1227                         0 /* don't unmask intr */,
1228                         0 /* don't reset intr timer */);
1229
1230         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1231
1232         /* Buffer allocation failed. Stay in polling mode
1233          * so we can try to fill the ring again.
1234          */
1235
1236         if (err)
1237                 work_done = work_to_do;
1238
1239         if (work_done < work_to_do) {
1240
1241                 /* Some work done, but not enough to stay in polling,
1242                  * flush all LROs and exit polling
1243                  */
1244
1245                 if (netdev->features & NETIF_F_LRO)
1246                         lro_flush_all(&enic->lro_mgr);
1247
1248                 napi_complete(napi);
1249                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1250         }
1251
1252         return work_done;
1253 }
1254
1255 static void enic_notify_timer(unsigned long data)
1256 {
1257         struct enic *enic = (struct enic *)data;
1258
1259         enic_notify_check(enic);
1260
1261         mod_timer(&enic->notify_timer,
1262                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1263 }
1264
1265 static void enic_free_intr(struct enic *enic)
1266 {
1267         struct net_device *netdev = enic->netdev;
1268         unsigned int i;
1269
1270         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1271         case VNIC_DEV_INTR_MODE_INTX:
1272                 free_irq(enic->pdev->irq, netdev);
1273                 break;
1274         case VNIC_DEV_INTR_MODE_MSI:
1275                 free_irq(enic->pdev->irq, enic);
1276                 break;
1277         case VNIC_DEV_INTR_MODE_MSIX:
1278                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1279                         if (enic->msix[i].requested)
1280                                 free_irq(enic->msix_entry[i].vector,
1281                                         enic->msix[i].devid);
1282                 break;
1283         default:
1284                 break;
1285         }
1286 }
1287
1288 static int enic_request_intr(struct enic *enic)
1289 {
1290         struct net_device *netdev = enic->netdev;
1291         unsigned int i;
1292         int err = 0;
1293
1294         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1295
1296         case VNIC_DEV_INTR_MODE_INTX:
1297
1298                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1299                         IRQF_SHARED, netdev->name, netdev);
1300                 break;
1301
1302         case VNIC_DEV_INTR_MODE_MSI:
1303
1304                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1305                         0, netdev->name, enic);
1306                 break;
1307
1308         case VNIC_DEV_INTR_MODE_MSIX:
1309
1310                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1311                         "%.11s-rx-0", netdev->name);
1312                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1313                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1314
1315                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1316                         "%.11s-tx-0", netdev->name);
1317                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1318                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1319
1320                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1321                         "%.11s-err", netdev->name);
1322                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1323                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1324
1325                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1326                         "%.11s-notify", netdev->name);
1327                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1328                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1329
1330                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1331                         err = request_irq(enic->msix_entry[i].vector,
1332                                 enic->msix[i].isr, 0,
1333                                 enic->msix[i].devname,
1334                                 enic->msix[i].devid);
1335                         if (err) {
1336                                 enic_free_intr(enic);
1337                                 break;
1338                         }
1339                         enic->msix[i].requested = 1;
1340                 }
1341
1342                 break;
1343
1344         default:
1345                 break;
1346         }
1347
1348         return err;
1349 }
1350
1351 static void enic_synchronize_irqs(struct enic *enic)
1352 {
1353         unsigned int i;
1354
1355         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1356         case VNIC_DEV_INTR_MODE_INTX:
1357         case VNIC_DEV_INTR_MODE_MSI:
1358                 synchronize_irq(enic->pdev->irq);
1359                 break;
1360         case VNIC_DEV_INTR_MODE_MSIX:
1361                 for (i = 0; i < enic->intr_count; i++)
1362                         synchronize_irq(enic->msix_entry[i].vector);
1363                 break;
1364         default:
1365                 break;
1366         }
1367 }
1368
1369 static int enic_notify_set(struct enic *enic)
1370 {
1371         int err;
1372
1373         spin_lock(&enic->devcmd_lock);
1374         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1375         case VNIC_DEV_INTR_MODE_INTX:
1376                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1377                 break;
1378         case VNIC_DEV_INTR_MODE_MSIX:
1379                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1380                 break;
1381         default:
1382                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1383                 break;
1384         }
1385         spin_unlock(&enic->devcmd_lock);
1386
1387         return err;
1388 }
1389
1390 static void enic_notify_timer_start(struct enic *enic)
1391 {
1392         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1393         case VNIC_DEV_INTR_MODE_MSI:
1394                 mod_timer(&enic->notify_timer, jiffies);
1395                 break;
1396         default:
1397                 /* Using intr for notification for INTx/MSI-X */
1398                 break;
1399         };
1400 }
1401
1402 /* rtnl lock is held, process context */
1403 static int enic_open(struct net_device *netdev)
1404 {
1405         struct enic *enic = netdev_priv(netdev);
1406         unsigned int i;
1407         int err;
1408
1409         err = enic_request_intr(enic);
1410         if (err) {
1411                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1412                         netdev->name);
1413                 return err;
1414         }
1415
1416         err = enic_notify_set(enic);
1417         if (err) {
1418                 printk(KERN_ERR PFX
1419                         "%s: Failed to alloc notify buffer, aborting.\n",
1420                         netdev->name);
1421                 goto err_out_free_intr;
1422         }
1423
1424         for (i = 0; i < enic->rq_count; i++) {
1425                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1426                 /* Need at least one buffer on ring to get going */
1427                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1428                         printk(KERN_ERR PFX
1429                                 "%s: Unable to alloc receive buffers.\n",
1430                                 netdev->name);
1431                         err = -ENOMEM;
1432                         goto err_out_notify_unset;
1433                 }
1434         }
1435
1436         for (i = 0; i < enic->wq_count; i++)
1437                 vnic_wq_enable(&enic->wq[i]);
1438         for (i = 0; i < enic->rq_count; i++)
1439                 vnic_rq_enable(&enic->rq[i]);
1440
1441         spin_lock(&enic->devcmd_lock);
1442         enic_add_station_addr(enic);
1443         spin_unlock(&enic->devcmd_lock);
1444         enic_set_multicast_list(netdev);
1445
1446         netif_wake_queue(netdev);
1447         napi_enable(&enic->napi);
1448         spin_lock(&enic->devcmd_lock);
1449         vnic_dev_enable(enic->vdev);
1450         spin_unlock(&enic->devcmd_lock);
1451
1452         for (i = 0; i < enic->intr_count; i++)
1453                 vnic_intr_unmask(&enic->intr[i]);
1454
1455         enic_notify_timer_start(enic);
1456
1457         return 0;
1458
1459 err_out_notify_unset:
1460         spin_lock(&enic->devcmd_lock);
1461         vnic_dev_notify_unset(enic->vdev);
1462         spin_unlock(&enic->devcmd_lock);
1463 err_out_free_intr:
1464         enic_free_intr(enic);
1465
1466         return err;
1467 }
1468
1469 /* rtnl lock is held, process context */
1470 static int enic_stop(struct net_device *netdev)
1471 {
1472         struct enic *enic = netdev_priv(netdev);
1473         unsigned int i;
1474         int err;
1475
1476         for (i = 0; i < enic->intr_count; i++)
1477                 vnic_intr_mask(&enic->intr[i]);
1478
1479         enic_synchronize_irqs(enic);
1480
1481         del_timer_sync(&enic->notify_timer);
1482
1483         spin_lock(&enic->devcmd_lock);
1484         vnic_dev_disable(enic->vdev);
1485         spin_unlock(&enic->devcmd_lock);
1486         napi_disable(&enic->napi);
1487         netif_carrier_off(netdev);
1488         netif_tx_disable(netdev);
1489
1490         for (i = 0; i < enic->wq_count; i++) {
1491                 err = vnic_wq_disable(&enic->wq[i]);
1492                 if (err)
1493                         return err;
1494         }
1495         for (i = 0; i < enic->rq_count; i++) {
1496                 err = vnic_rq_disable(&enic->rq[i]);
1497                 if (err)
1498                         return err;
1499         }
1500
1501         spin_lock(&enic->devcmd_lock);
1502         vnic_dev_notify_unset(enic->vdev);
1503         spin_unlock(&enic->devcmd_lock);
1504         enic_free_intr(enic);
1505
1506         for (i = 0; i < enic->wq_count; i++)
1507                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1508         for (i = 0; i < enic->rq_count; i++)
1509                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1510         for (i = 0; i < enic->cq_count; i++)
1511                 vnic_cq_clean(&enic->cq[i]);
1512         for (i = 0; i < enic->intr_count; i++)
1513                 vnic_intr_clean(&enic->intr[i]);
1514
1515         return 0;
1516 }
1517
1518 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1519 {
1520         struct enic *enic = netdev_priv(netdev);
1521         int running = netif_running(netdev);
1522
1523         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1524                 return -EINVAL;
1525
1526         if (running)
1527                 enic_stop(netdev);
1528
1529         netdev->mtu = new_mtu;
1530
1531         if (netdev->mtu > enic->port_mtu)
1532                 printk(KERN_WARNING PFX
1533                         "%s: interface MTU (%d) set higher "
1534                         "than port MTU (%d)\n",
1535                         netdev->name, netdev->mtu, enic->port_mtu);
1536
1537         if (running)
1538                 enic_open(netdev);
1539
1540         return 0;
1541 }
1542
1543 #ifdef CONFIG_NET_POLL_CONTROLLER
1544 static void enic_poll_controller(struct net_device *netdev)
1545 {
1546         struct enic *enic = netdev_priv(netdev);
1547         struct vnic_dev *vdev = enic->vdev;
1548
1549         switch (vnic_dev_get_intr_mode(vdev)) {
1550         case VNIC_DEV_INTR_MODE_MSIX:
1551                 enic_isr_msix_rq(enic->pdev->irq, enic);
1552                 enic_isr_msix_wq(enic->pdev->irq, enic);
1553                 break;
1554         case VNIC_DEV_INTR_MODE_MSI:
1555                 enic_isr_msi(enic->pdev->irq, enic);
1556                 break;
1557         case VNIC_DEV_INTR_MODE_INTX:
1558                 enic_isr_legacy(enic->pdev->irq, netdev);
1559                 break;
1560         default:
1561                 break;
1562         }
1563 }
1564 #endif
1565
1566 static int enic_dev_wait(struct vnic_dev *vdev,
1567         int (*start)(struct vnic_dev *, int),
1568         int (*finished)(struct vnic_dev *, int *),
1569         int arg)
1570 {
1571         unsigned long time;
1572         int done;
1573         int err;
1574
1575         BUG_ON(in_interrupt());
1576
1577         err = start(vdev, arg);
1578         if (err)
1579                 return err;
1580
1581         /* Wait for func to complete...2 seconds max
1582          */
1583
1584         time = jiffies + (HZ * 2);
1585         do {
1586
1587                 err = finished(vdev, &done);
1588                 if (err)
1589                         return err;
1590
1591                 if (done)
1592                         return 0;
1593
1594                 schedule_timeout_uninterruptible(HZ / 10);
1595
1596         } while (time_after(time, jiffies));
1597
1598         return -ETIMEDOUT;
1599 }
1600
1601 static int enic_dev_open(struct enic *enic)
1602 {
1603         int err;
1604
1605         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1606                 vnic_dev_open_done, 0);
1607         if (err)
1608                 printk(KERN_ERR PFX
1609                         "vNIC device open failed, err %d.\n", err);
1610
1611         return err;
1612 }
1613
1614 static int enic_dev_soft_reset(struct enic *enic)
1615 {
1616         int err;
1617
1618         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1619                 vnic_dev_soft_reset_done, 0);
1620         if (err)
1621                 printk(KERN_ERR PFX
1622                         "vNIC soft reset failed, err %d.\n", err);
1623
1624         return err;
1625 }
1626
1627 static int enic_set_niccfg(struct enic *enic)
1628 {
1629         const u8 rss_default_cpu = 0;
1630         const u8 rss_hash_type = 0;
1631         const u8 rss_hash_bits = 0;
1632         const u8 rss_base_cpu = 0;
1633         const u8 rss_enable = 0;
1634         const u8 tso_ipid_split_en = 0;
1635         const u8 ig_vlan_strip_en = 1;
1636
1637         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1638          */
1639
1640         return enic_set_nic_cfg(enic,
1641                 rss_default_cpu, rss_hash_type,
1642                 rss_hash_bits, rss_base_cpu,
1643                 rss_enable, tso_ipid_split_en,
1644                 ig_vlan_strip_en);
1645 }
1646
1647 static void enic_reset(struct work_struct *work)
1648 {
1649         struct enic *enic = container_of(work, struct enic, reset);
1650
1651         if (!netif_running(enic->netdev))
1652                 return;
1653
1654         rtnl_lock();
1655
1656         spin_lock(&enic->devcmd_lock);
1657         vnic_dev_hang_notify(enic->vdev);
1658         spin_unlock(&enic->devcmd_lock);
1659
1660         enic_stop(enic->netdev);
1661         enic_dev_soft_reset(enic);
1662         vnic_dev_init(enic->vdev, 0);
1663         enic_reset_mcaddrs(enic);
1664         enic_init_vnic_resources(enic);
1665         enic_set_niccfg(enic);
1666         enic_open(enic->netdev);
1667
1668         rtnl_unlock();
1669 }
1670
1671 static int enic_set_intr_mode(struct enic *enic)
1672 {
1673         unsigned int n = 1;
1674         unsigned int m = 1;
1675         unsigned int i;
1676
1677         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1678          * system capabilities.
1679          *
1680          * Try MSI-X first
1681          *
1682          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1683          * (the second to last INTR is used for WQ/RQ errors)
1684          * (the last INTR is used for notifications)
1685          */
1686
1687         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1688         for (i = 0; i < n + m + 2; i++)
1689                 enic->msix_entry[i].entry = i;
1690
1691         if (enic->config.intr_mode < 1 &&
1692             enic->rq_count >= n &&
1693             enic->wq_count >= m &&
1694             enic->cq_count >= n + m &&
1695             enic->intr_count >= n + m + 2 &&
1696             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1697
1698                 enic->rq_count = n;
1699                 enic->wq_count = m;
1700                 enic->cq_count = n + m;
1701                 enic->intr_count = n + m + 2;
1702
1703                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1704
1705                 return 0;
1706         }
1707
1708         /* Next try MSI
1709          *
1710          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1711          */
1712
1713         if (enic->config.intr_mode < 2 &&
1714             enic->rq_count >= 1 &&
1715             enic->wq_count >= 1 &&
1716             enic->cq_count >= 2 &&
1717             enic->intr_count >= 1 &&
1718             !pci_enable_msi(enic->pdev)) {
1719
1720                 enic->rq_count = 1;
1721                 enic->wq_count = 1;
1722                 enic->cq_count = 2;
1723                 enic->intr_count = 1;
1724
1725                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1726
1727                 return 0;
1728         }
1729
1730         /* Next try INTx
1731          *
1732          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1733          * (the first INTR is used for WQ/RQ)
1734          * (the second INTR is used for WQ/RQ errors)
1735          * (the last INTR is used for notifications)
1736          */
1737
1738         if (enic->config.intr_mode < 3 &&
1739             enic->rq_count >= 1 &&
1740             enic->wq_count >= 1 &&
1741             enic->cq_count >= 2 &&
1742             enic->intr_count >= 3) {
1743
1744                 enic->rq_count = 1;
1745                 enic->wq_count = 1;
1746                 enic->cq_count = 2;
1747                 enic->intr_count = 3;
1748
1749                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1750
1751                 return 0;
1752         }
1753
1754         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1755
1756         return -EINVAL;
1757 }
1758
1759 static void enic_clear_intr_mode(struct enic *enic)
1760 {
1761         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1762         case VNIC_DEV_INTR_MODE_MSIX:
1763                 pci_disable_msix(enic->pdev);
1764                 break;
1765         case VNIC_DEV_INTR_MODE_MSI:
1766                 pci_disable_msi(enic->pdev);
1767                 break;
1768         default:
1769                 break;
1770         }
1771
1772         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1773 }
1774
1775 static const struct net_device_ops enic_netdev_ops = {
1776         .ndo_open               = enic_open,
1777         .ndo_stop               = enic_stop,
1778         .ndo_start_xmit         = enic_hard_start_xmit,
1779         .ndo_get_stats          = enic_get_stats,
1780         .ndo_validate_addr      = eth_validate_addr,
1781         .ndo_set_mac_address    = eth_mac_addr,
1782         .ndo_set_multicast_list = enic_set_multicast_list,
1783         .ndo_change_mtu         = enic_change_mtu,
1784         .ndo_vlan_rx_register   = enic_vlan_rx_register,
1785         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
1786         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
1787         .ndo_tx_timeout         = enic_tx_timeout,
1788 #ifdef CONFIG_NET_POLL_CONTROLLER
1789         .ndo_poll_controller    = enic_poll_controller,
1790 #endif
1791 };
1792
1793 void enic_dev_deinit(struct enic *enic)
1794 {
1795         netif_napi_del(&enic->napi);
1796         enic_free_vnic_resources(enic);
1797         enic_clear_intr_mode(enic);
1798 }
1799
1800 int enic_dev_init(struct enic *enic)
1801 {
1802         struct net_device *netdev = enic->netdev;
1803         int err;
1804
1805         /* Get vNIC configuration
1806          */
1807
1808         err = enic_get_vnic_config(enic);
1809         if (err) {
1810                 printk(KERN_ERR PFX
1811                         "Get vNIC configuration failed, aborting.\n");
1812                 return err;
1813         }
1814
1815         /* Get available resource counts
1816          */
1817
1818         enic_get_res_counts(enic);
1819
1820         /* Set interrupt mode based on resource counts and system
1821          * capabilities
1822          */
1823
1824         err = enic_set_intr_mode(enic);
1825         if (err) {
1826                 printk(KERN_ERR PFX
1827                         "Failed to set intr mode based on resource "
1828                         "counts and system capabilities, aborting.\n");
1829                 return err;
1830         }
1831
1832         /* Allocate and configure vNIC resources
1833          */
1834
1835         err = enic_alloc_vnic_resources(enic);
1836         if (err) {
1837                 printk(KERN_ERR PFX
1838                         "Failed to alloc vNIC resources, aborting.\n");
1839                 goto err_out_free_vnic_resources;
1840         }
1841
1842         enic_init_vnic_resources(enic);
1843
1844         err = enic_set_rq_alloc_buf(enic);
1845         if (err) {
1846                 printk(KERN_ERR PFX
1847                         "Failed to set RQ buffer allocator, aborting.\n");
1848                 goto err_out_free_vnic_resources;
1849         }
1850
1851         err = enic_set_niccfg(enic);
1852         if (err) {
1853                 printk(KERN_ERR PFX
1854                         "Failed to config nic, aborting.\n");
1855                 goto err_out_free_vnic_resources;
1856         }
1857
1858         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1859         default:
1860                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1861                 break;
1862         case VNIC_DEV_INTR_MODE_MSIX:
1863                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1864                 break;
1865         }
1866
1867         return 0;
1868
1869 err_out_free_vnic_resources:
1870         enic_clear_intr_mode(enic);
1871         enic_free_vnic_resources(enic);
1872
1873         return err;
1874 }
1875
1876 static void enic_iounmap(struct enic *enic)
1877 {
1878         unsigned int i;
1879
1880         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
1881                 if (enic->bar[i].vaddr)
1882                         iounmap(enic->bar[i].vaddr);
1883 }
1884
1885 static int __devinit enic_probe(struct pci_dev *pdev,
1886         const struct pci_device_id *ent)
1887 {
1888         struct net_device *netdev;
1889         struct enic *enic;
1890         int using_dac = 0;
1891         unsigned int i;
1892         int err;
1893
1894         /* Allocate net device structure and initialize.  Private
1895          * instance data is initialized to zero.
1896          */
1897
1898         netdev = alloc_etherdev(sizeof(struct enic));
1899         if (!netdev) {
1900                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1901                 return -ENOMEM;
1902         }
1903
1904         pci_set_drvdata(pdev, netdev);
1905
1906         SET_NETDEV_DEV(netdev, &pdev->dev);
1907
1908         enic = netdev_priv(netdev);
1909         enic->netdev = netdev;
1910         enic->pdev = pdev;
1911
1912         /* Setup PCI resources
1913          */
1914
1915         err = pci_enable_device(pdev);
1916         if (err) {
1917                 printk(KERN_ERR PFX
1918                         "Cannot enable PCI device, aborting.\n");
1919                 goto err_out_free_netdev;
1920         }
1921
1922         err = pci_request_regions(pdev, DRV_NAME);
1923         if (err) {
1924                 printk(KERN_ERR PFX
1925                         "Cannot request PCI regions, aborting.\n");
1926                 goto err_out_disable_device;
1927         }
1928
1929         pci_set_master(pdev);
1930
1931         /* Query PCI controller on system for DMA addressing
1932          * limitation for the device.  Try 40-bit first, and
1933          * fail to 32-bit.
1934          */
1935
1936         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
1937         if (err) {
1938                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1939                 if (err) {
1940                         printk(KERN_ERR PFX
1941                                 "No usable DMA configuration, aborting.\n");
1942                         goto err_out_release_regions;
1943                 }
1944                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1945                 if (err) {
1946                         printk(KERN_ERR PFX
1947                                 "Unable to obtain 32-bit DMA "
1948                                 "for consistent allocations, aborting.\n");
1949                         goto err_out_release_regions;
1950                 }
1951         } else {
1952                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
1953                 if (err) {
1954                         printk(KERN_ERR PFX
1955                                 "Unable to obtain 40-bit DMA "
1956                                 "for consistent allocations, aborting.\n");
1957                         goto err_out_release_regions;
1958                 }
1959                 using_dac = 1;
1960         }
1961
1962         /* Map vNIC resources from BAR0-5
1963          */
1964
1965         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
1966                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
1967                         continue;
1968                 enic->bar[i].len = pci_resource_len(pdev, i);
1969                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
1970                 if (!enic->bar[i].vaddr) {
1971                         printk(KERN_ERR PFX
1972                                 "Cannot memory-map BAR %d, aborting.\n", i);
1973                         err = -ENODEV;
1974                         goto err_out_iounmap;
1975                 }
1976                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
1977         }
1978
1979         /* Register vNIC device
1980          */
1981
1982         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
1983                 ARRAY_SIZE(enic->bar));
1984         if (!enic->vdev) {
1985                 printk(KERN_ERR PFX
1986                         "vNIC registration failed, aborting.\n");
1987                 err = -ENODEV;
1988                 goto err_out_iounmap;
1989         }
1990
1991         /* Issue device open to get device in known state
1992          */
1993
1994         err = enic_dev_open(enic);
1995         if (err) {
1996                 printk(KERN_ERR PFX
1997                         "vNIC dev open failed, aborting.\n");
1998                 goto err_out_vnic_unregister;
1999         }
2000
2001         /* Issue device init to initialize the vnic-to-switch link.
2002          * We'll start with carrier off and wait for link UP
2003          * notification later to turn on carrier.  We don't need
2004          * to wait here for the vnic-to-switch link initialization
2005          * to complete; link UP notification is the indication that
2006          * the process is complete.
2007          */
2008
2009         netif_carrier_off(netdev);
2010
2011         err = vnic_dev_init(enic->vdev, 0);
2012         if (err) {
2013                 printk(KERN_ERR PFX
2014                         "vNIC dev init failed, aborting.\n");
2015                 goto err_out_dev_close;
2016         }
2017
2018         err = enic_dev_init(enic);
2019         if (err) {
2020                 printk(KERN_ERR PFX
2021                         "Device initialization failed, aborting.\n");
2022                 goto err_out_dev_close;
2023         }
2024
2025         /* Setup notification timer, HW reset task, and locks
2026          */
2027
2028         init_timer(&enic->notify_timer);
2029         enic->notify_timer.function = enic_notify_timer;
2030         enic->notify_timer.data = (unsigned long)enic;
2031
2032         INIT_WORK(&enic->reset, enic_reset);
2033
2034         for (i = 0; i < enic->wq_count; i++)
2035                 spin_lock_init(&enic->wq_lock[i]);
2036
2037         spin_lock_init(&enic->devcmd_lock);
2038
2039         /* Register net device
2040          */
2041
2042         enic->port_mtu = enic->config.mtu;
2043         (void)enic_change_mtu(netdev, enic->port_mtu);
2044
2045         err = enic_set_mac_addr(netdev, enic->mac_addr);
2046         if (err) {
2047                 printk(KERN_ERR PFX
2048                         "Invalid MAC address, aborting.\n");
2049                 goto err_out_dev_deinit;
2050         }
2051
2052         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2053         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2054
2055         netdev->netdev_ops = &enic_netdev_ops;
2056         netdev->watchdog_timeo = 2 * HZ;
2057         netdev->ethtool_ops = &enic_ethtool_ops;
2058
2059         netdev->features |= NETIF_F_HW_VLAN_TX |
2060                 NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_FILTER;
2061         if (ENIC_SETTING(enic, TXCSUM))
2062                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2063         if (ENIC_SETTING(enic, TSO))
2064                 netdev->features |= NETIF_F_TSO |
2065                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2066         if (ENIC_SETTING(enic, LRO))
2067                 netdev->features |= NETIF_F_LRO;
2068         if (using_dac)
2069                 netdev->features |= NETIF_F_HIGHDMA;
2070
2071         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2072
2073         enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
2074         enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
2075         enic->lro_mgr.lro_arr = enic->lro_desc;
2076         enic->lro_mgr.get_skb_header = enic_get_skb_header;
2077         enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
2078         enic->lro_mgr.dev = netdev;
2079         enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
2080         enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
2081
2082         err = register_netdev(netdev);
2083         if (err) {
2084                 printk(KERN_ERR PFX
2085                         "Cannot register net device, aborting.\n");
2086                 goto err_out_dev_deinit;
2087         }
2088
2089         return 0;
2090
2091 err_out_dev_deinit:
2092         enic_dev_deinit(enic);
2093 err_out_dev_close:
2094         vnic_dev_close(enic->vdev);
2095 err_out_vnic_unregister:
2096         vnic_dev_unregister(enic->vdev);
2097 err_out_iounmap:
2098         enic_iounmap(enic);
2099 err_out_release_regions:
2100         pci_release_regions(pdev);
2101 err_out_disable_device:
2102         pci_disable_device(pdev);
2103 err_out_free_netdev:
2104         pci_set_drvdata(pdev, NULL);
2105         free_netdev(netdev);
2106
2107         return err;
2108 }
2109
2110 static void __devexit enic_remove(struct pci_dev *pdev)
2111 {
2112         struct net_device *netdev = pci_get_drvdata(pdev);
2113
2114         if (netdev) {
2115                 struct enic *enic = netdev_priv(netdev);
2116
2117                 flush_scheduled_work();
2118                 unregister_netdev(netdev);
2119                 enic_dev_deinit(enic);
2120                 vnic_dev_close(enic->vdev);
2121                 vnic_dev_unregister(enic->vdev);
2122                 enic_iounmap(enic);
2123                 pci_release_regions(pdev);
2124                 pci_disable_device(pdev);
2125                 pci_set_drvdata(pdev, NULL);
2126                 free_netdev(netdev);
2127         }
2128 }
2129
2130 static struct pci_driver enic_driver = {
2131         .name = DRV_NAME,
2132         .id_table = enic_id_table,
2133         .probe = enic_probe,
2134         .remove = __devexit_p(enic_remove),
2135 };
2136
2137 static int __init enic_init_module(void)
2138 {
2139         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2140
2141         return pci_register_driver(&enic_driver);
2142 }
2143
2144 static void __exit enic_cleanup_module(void)
2145 {
2146         pci_unregister_driver(&enic_driver);
2147 }
2148
2149 module_init(enic_init_module);
2150 module_exit(enic_cleanup_module);