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 != 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 inline struct sk_buff *enic_rq_alloc_skb(struct net_device *netdev,
 874         unsigned int size)
 875 {
 876         struct sk_buff *skb;
 877
 878         skb = netdev_alloc_skb(netdev, size + NET_IP_ALIGN);
 879
 880         if (skb)
 881                 skb_reserve(skb, NET_IP_ALIGN);
 882
 883         return skb;
 884 }
 885
 886 static int enic_rq_alloc_buf(struct vnic_rq *rq)
 887 {
 888         struct enic *enic = vnic_dev_priv(rq->vdev);
 889         struct net_device *netdev = enic->netdev;
 890         struct sk_buff *skb;
 891         unsigned int len = netdev->mtu + ETH_HLEN;
 892         unsigned int os_buf_index = 0;
 893         dma_addr_t dma_addr;
 894
 895         skb = enic_rq_alloc_skb(netdev, len);
 896         if (!skb)
 897                 return -ENOMEM;
 898
 899         dma_addr = pci_map_single(enic->pdev, skb->data,
 900                 len, PCI_DMA_FROMDEVICE);
 901
 902         enic_queue_rq_desc(rq, skb, os_buf_index,
 903                 dma_addr, len);
 904
 905         return 0;
 906 }
 907
 908 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
 909 {
 910         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
 911
 912         if (vnic_rq_posting_soon(rq)) {
 913
 914                 /* SW workaround for A0 HW erratum: if we're just about
 915                  * to write posted_index, insert a dummy desc
 916                  * of type resvd
 917                  */
 918
 919                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
 920                 vnic_rq_post(rq, 0, 0, 0, 0);
 921         } else {
 922                 return enic_rq_alloc_buf(rq);
 923         }
 924
 925         return 0;
 926 }
 927
 928 static int enic_set_rq_alloc_buf(struct enic *enic)
 929 {
 930         enum vnic_dev_hw_version hw_ver;
 931         int err;
 932
 933         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
 934         if (err)
 935                 return err;
 936
 937         switch (hw_ver) {
 938         case VNIC_DEV_HW_VER_A1:
 939                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
 940                 break;
 941         case VNIC_DEV_HW_VER_A2:
 942         case VNIC_DEV_HW_VER_UNKNOWN:
 943                 enic->rq_alloc_buf = enic_rq_alloc_buf;
 944                 break;
 945         default:
 946                 return -ENODEV;
 947         }
 948
 949         return 0;
 950 }
 951
 952 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
 953         void **tcph, u64 *hdr_flags, void *priv)
 954 {
 955         struct cq_enet_rq_desc *cq_desc = priv;
 956         unsigned int ip_len;
 957         struct iphdr *iph;
 958
 959         u8 type, color, eop, sop, ingress_port, vlan_stripped;
 960         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
 961         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
 962         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
 963         u8 packet_error;
 964         u16 q_number, completed_index, bytes_written, vlan, checksum;
 965         u32 rss_hash;
 966
 967         cq_enet_rq_desc_dec(cq_desc,
 968                 &type, &color, &q_number, &completed_index,
 969                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
 970                 &csum_not_calc, &rss_hash, &bytes_written,
 971                 &packet_error, &vlan_stripped, &vlan, &checksum,
 972                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
 973                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
 974                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
 975                 &fcs_ok);
 976
 977         if (!(ipv4 && tcp && !ipv4_fragment))
 978                 return -1;
 979
 980         skb_reset_network_header(skb);
 981         iph = ip_hdr(skb);
 982
 983         ip_len = ip_hdrlen(skb);
 984         skb_set_transport_header(skb, ip_len);
 985
 986         /* check if ip header and tcp header are complete */
 987         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
 988                 return -1;
 989
 990         *hdr_flags = LRO_IPV4 | LRO_TCP;
 991         *tcph = tcp_hdr(skb);
 992         *iphdr = iph;
 993
 994         return 0;
 995 }
 996
 997 static void enic_rq_indicate_buf(struct vnic_rq *rq,
 998         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
 999         int skipped, void *opaque)
1000 {
1001         struct enic *enic = vnic_dev_priv(rq->vdev);
1002         struct net_device *netdev = enic->netdev;
1003         struct sk_buff *skb;
1004
1005         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1006         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1007         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1008         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1009         u8 packet_error;
1010         u16 q_number, completed_index, bytes_written, vlan, checksum;
1011         u32 rss_hash;
1012
1013         if (skipped)
1014                 return;
1015
1016         skb = buf->os_buf;
1017         prefetch(skb->data - NET_IP_ALIGN);
1018         pci_unmap_single(enic->pdev, buf->dma_addr,
1019                 buf->len, PCI_DMA_FROMDEVICE);
1020
1021         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1022                 &type, &color, &q_number, &completed_index,
1023                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1024                 &csum_not_calc, &rss_hash, &bytes_written,
1025                 &packet_error, &vlan_stripped, &vlan, &checksum,
1026                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1027                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1028                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1029                 &fcs_ok);
1030
1031         if (packet_error) {
1032
1033                 if (!fcs_ok) {
1034                         if (bytes_written > 0)
1035                                 enic->rq_bad_fcs++;
1036                         else if (bytes_written == 0)
1037                                 enic->rq_truncated_pkts++;
1038                 }
1039
1040                 dev_kfree_skb_any(skb);
1041
1042                 return;
1043         }
1044
1045         if (eop && bytes_written > 0) {
1046
1047                 /* Good receive
1048                  */
1049
1050                 skb_put(skb, bytes_written);
1051                 skb->protocol = eth_type_trans(skb, netdev);
1052
1053                 if (enic->csum_rx_enabled && !csum_not_calc) {
1054                         skb->csum = htons(checksum);
1055                         skb->ip_summed = CHECKSUM_COMPLETE;
1056                 }
1057
1058                 skb->dev = netdev;
1059
1060                 if (enic->vlan_group && vlan_stripped) {
1061
1062                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1063                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
1064                                         skb, enic->vlan_group,
1065                                         vlan, cq_desc);
1066                         else
1067                                 vlan_hwaccel_receive_skb(skb,
1068                                         enic->vlan_group, vlan);
1069
1070                 } else {
1071
1072                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1073                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
1074                         else
1075                                 netif_receive_skb(skb);
1076
1077                 }
1078
1079         } else {
1080
1081                 /* Buffer overflow
1082                  */
1083
1084                 dev_kfree_skb_any(skb);
1085         }
1086 }
1087
1088 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1089         u8 type, u16 q_number, u16 completed_index, void *opaque)
1090 {
1091         struct enic *enic = vnic_dev_priv(vdev);
1092
1093         vnic_rq_service(&enic->rq[q_number], cq_desc,
1094                 completed_index, VNIC_RQ_RETURN_DESC,
1095                 enic_rq_indicate_buf, opaque);
1096
1097         return 0;
1098 }
1099
1100 static void enic_rq_drop_buf(struct vnic_rq *rq,
1101         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1102         int skipped, void *opaque)
1103 {
1104         struct enic *enic = vnic_dev_priv(rq->vdev);
1105         struct sk_buff *skb = buf->os_buf;
1106
1107         if (skipped)
1108                 return;
1109
1110         pci_unmap_single(enic->pdev, buf->dma_addr,
1111                 buf->len, PCI_DMA_FROMDEVICE);
1112
1113         dev_kfree_skb_any(skb);
1114 }
1115
1116 static int enic_rq_service_drop(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1117         u8 type, u16 q_number, u16 completed_index, void *opaque)
1118 {
1119         struct enic *enic = vnic_dev_priv(vdev);
1120
1121         vnic_rq_service(&enic->rq[q_number], cq_desc,
1122                 completed_index, VNIC_RQ_RETURN_DESC,
1123                 enic_rq_drop_buf, opaque);
1124
1125         return 0;
1126 }
1127
1128 static int enic_poll(struct napi_struct *napi, int budget)
1129 {
1130         struct enic *enic = container_of(napi, struct enic, napi);
1131         struct net_device *netdev = enic->netdev;
1132         unsigned int rq_work_to_do = budget;
1133         unsigned int wq_work_to_do = -1; /* no limit */
1134         unsigned int  work_done, rq_work_done, wq_work_done;
1135
1136         /* Service RQ (first) and WQ
1137          */
1138
1139         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1140                 rq_work_to_do, enic_rq_service, NULL);
1141
1142         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1143                 wq_work_to_do, enic_wq_service, NULL);
1144
1145         /* Accumulate intr event credits for this polling
1146          * cycle.  An intr event is the completion of a
1147          * a WQ or RQ packet.
1148          */
1149
1150         work_done = rq_work_done + wq_work_done;
1151
1152         if (work_done > 0)
1153                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1154                         work_done,
1155                         0 /* don't unmask intr */,
1156                         0 /* don't reset intr timer */);
1157
1158         if (rq_work_done > 0) {
1159
1160                 /* Replenish RQ
1161                  */
1162
1163                 vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1164
1165         } else {
1166
1167                 /* If no work done, flush all LROs and exit polling
1168                  */
1169
1170                 if (netdev->features & NETIF_F_LRO)
1171                         lro_flush_all(&enic->lro_mgr);
1172
1173                 napi_complete(napi);
1174                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1175         }
1176
1177         return rq_work_done;
1178 }
1179
1180 static int enic_poll_msix(struct napi_struct *napi, int budget)
1181 {
1182         struct enic *enic = container_of(napi, struct enic, napi);
1183         struct net_device *netdev = enic->netdev;
1184         unsigned int work_to_do = budget;
1185         unsigned int work_done;
1186
1187         /* Service RQ
1188          */
1189
1190         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1191                 work_to_do, enic_rq_service, NULL);
1192
1193         if (work_done > 0) {
1194
1195                 /* Replenish RQ
1196                  */
1197
1198                 vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1199
1200                 /* Return intr event credits for this polling
1201                  * cycle.  An intr event is the completion of a
1202                  * RQ packet.
1203                  */
1204
1205                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1206                         work_done,
1207                         0 /* don't unmask intr */,
1208                         0 /* don't reset intr timer */);
1209         } else {
1210
1211                 /* If no work done, flush all LROs and exit polling
1212                  */
1213
1214                 if (netdev->features & NETIF_F_LRO)
1215                         lro_flush_all(&enic->lro_mgr);
1216
1217                 napi_complete(napi);
1218                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1219         }
1220
1221         return work_done;
1222 }
1223
1224 static void enic_notify_timer(unsigned long data)
1225 {
1226         struct enic *enic = (struct enic *)data;
1227
1228         enic_notify_check(enic);
1229
1230         mod_timer(&enic->notify_timer,
1231                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1232 }
1233
1234 static void enic_free_intr(struct enic *enic)
1235 {
1236         struct net_device *netdev = enic->netdev;
1237         unsigned int i;
1238
1239         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1240         case VNIC_DEV_INTR_MODE_INTX:
1241                 free_irq(enic->pdev->irq, netdev);
1242                 break;
1243         case VNIC_DEV_INTR_MODE_MSI:
1244                 free_irq(enic->pdev->irq, enic);
1245                 break;
1246         case VNIC_DEV_INTR_MODE_MSIX:
1247                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1248                         if (enic->msix[i].requested)
1249                                 free_irq(enic->msix_entry[i].vector,
1250                                         enic->msix[i].devid);
1251                 break;
1252         default:
1253                 break;
1254         }
1255 }
1256
1257 static int enic_request_intr(struct enic *enic)
1258 {
1259         struct net_device *netdev = enic->netdev;
1260         unsigned int i;
1261         int err = 0;
1262
1263         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1264
1265         case VNIC_DEV_INTR_MODE_INTX:
1266
1267                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1268                         IRQF_SHARED, netdev->name, netdev);
1269                 break;
1270
1271         case VNIC_DEV_INTR_MODE_MSI:
1272
1273                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1274                         0, netdev->name, enic);
1275                 break;
1276
1277         case VNIC_DEV_INTR_MODE_MSIX:
1278
1279                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1280                         "%.11s-rx-0", netdev->name);
1281                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1282                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1283
1284                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1285                         "%.11s-tx-0", netdev->name);
1286                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1287                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1288
1289                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1290                         "%.11s-err", netdev->name);
1291                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1292                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1293
1294                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1295                         "%.11s-notify", netdev->name);
1296                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1297                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1298
1299                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1300                         err = request_irq(enic->msix_entry[i].vector,
1301                                 enic->msix[i].isr, 0,
1302                                 enic->msix[i].devname,
1303                                 enic->msix[i].devid);
1304                         if (err) {
1305                                 enic_free_intr(enic);
1306                                 break;
1307                         }
1308                         enic->msix[i].requested = 1;
1309                 }
1310
1311                 break;
1312
1313         default:
1314                 break;
1315         }
1316
1317         return err;
1318 }
1319
1320 static int enic_notify_set(struct enic *enic)
1321 {
1322         int err;
1323
1324         spin_lock(&enic->devcmd_lock);
1325         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1326         case VNIC_DEV_INTR_MODE_INTX:
1327                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1328                 break;
1329         case VNIC_DEV_INTR_MODE_MSIX:
1330                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1331                 break;
1332         default:
1333                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1334                 break;
1335         }
1336         spin_unlock(&enic->devcmd_lock);
1337
1338         return err;
1339 }
1340
1341 static void enic_notify_timer_start(struct enic *enic)
1342 {
1343         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1344         case VNIC_DEV_INTR_MODE_MSI:
1345                 mod_timer(&enic->notify_timer, jiffies);
1346                 break;
1347         default:
1348                 /* Using intr for notification for INTx/MSI-X */
1349                 break;
1350         };
1351 }
1352
1353 /* rtnl lock is held, process context */
1354 static int enic_open(struct net_device *netdev)
1355 {
1356         struct enic *enic = netdev_priv(netdev);
1357         unsigned int i;
1358         int err;
1359
1360         err = enic_request_intr(enic);
1361         if (err) {
1362                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1363                         netdev->name);
1364                 return err;
1365         }
1366
1367         err = enic_notify_set(enic);
1368         if (err) {
1369                 printk(KERN_ERR PFX
1370                         "%s: Failed to alloc notify buffer, aborting.\n",
1371                         netdev->name);
1372                 goto err_out_free_intr;
1373         }
1374
1375         for (i = 0; i < enic->rq_count; i++) {
1376                 err = vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1377                 if (err) {
1378                         printk(KERN_ERR PFX
1379                                 "%s: Unable to alloc receive buffers.\n",
1380                                 netdev->name);
1381                         goto err_out_notify_unset;
1382                 }
1383         }
1384
1385         for (i = 0; i < enic->wq_count; i++)
1386                 vnic_wq_enable(&enic->wq[i]);
1387         for (i = 0; i < enic->rq_count; i++)
1388                 vnic_rq_enable(&enic->rq[i]);
1389
1390         spin_lock(&enic->devcmd_lock);
1391         enic_add_station_addr(enic);
1392         spin_unlock(&enic->devcmd_lock);
1393         enic_set_multicast_list(netdev);
1394
1395         netif_wake_queue(netdev);
1396         napi_enable(&enic->napi);
1397         spin_lock(&enic->devcmd_lock);
1398         vnic_dev_enable(enic->vdev);
1399         spin_unlock(&enic->devcmd_lock);
1400
1401         for (i = 0; i < enic->intr_count; i++)
1402                 vnic_intr_unmask(&enic->intr[i]);
1403
1404         enic_notify_timer_start(enic);
1405
1406         return 0;
1407
1408 err_out_notify_unset:
1409         spin_lock(&enic->devcmd_lock);
1410         vnic_dev_notify_unset(enic->vdev);
1411         spin_unlock(&enic->devcmd_lock);
1412 err_out_free_intr:
1413         enic_free_intr(enic);
1414
1415         return err;
1416 }
1417
1418 /* rtnl lock is held, process context */
1419 static int enic_stop(struct net_device *netdev)
1420 {
1421         struct enic *enic = netdev_priv(netdev);
1422         unsigned int i;
1423         int err;
1424
1425         del_timer_sync(&enic->notify_timer);
1426
1427         spin_lock(&enic->devcmd_lock);
1428         vnic_dev_disable(enic->vdev);
1429         spin_unlock(&enic->devcmd_lock);
1430         napi_disable(&enic->napi);
1431         netif_stop_queue(netdev);
1432
1433         for (i = 0; i < enic->intr_count; i++)
1434                 vnic_intr_mask(&enic->intr[i]);
1435
1436         for (i = 0; i < enic->wq_count; i++) {
1437                 err = vnic_wq_disable(&enic->wq[i]);
1438                 if (err)
1439                         return err;
1440         }
1441         for (i = 0; i < enic->rq_count; i++) {
1442                 err = vnic_rq_disable(&enic->rq[i]);
1443                 if (err)
1444                         return err;
1445         }
1446
1447         spin_lock(&enic->devcmd_lock);
1448         vnic_dev_notify_unset(enic->vdev);
1449         spin_unlock(&enic->devcmd_lock);
1450         enic_free_intr(enic);
1451
1452         (void)vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1453                 -1, enic_rq_service_drop, NULL);
1454         (void)vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1455                 -1, enic_wq_service, NULL);
1456
1457         for (i = 0; i < enic->wq_count; i++)
1458                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1459         for (i = 0; i < enic->rq_count; i++)
1460                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1461         for (i = 0; i < enic->cq_count; i++)
1462                 vnic_cq_clean(&enic->cq[i]);
1463         for (i = 0; i < enic->intr_count; i++)
1464                 vnic_intr_clean(&enic->intr[i]);
1465
1466         return 0;
1467 }
1468
1469 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1470 {
1471         struct enic *enic = netdev_priv(netdev);
1472         int running = netif_running(netdev);
1473
1474         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1475                 return -EINVAL;
1476
1477         if (running)
1478                 enic_stop(netdev);
1479
1480         netdev->mtu = new_mtu;
1481
1482         if (netdev->mtu > enic->port_mtu)
1483                 printk(KERN_WARNING PFX
1484                         "%s: interface MTU (%d) set higher "
1485                         "than port MTU (%d)\n",
1486                         netdev->name, netdev->mtu, enic->port_mtu);
1487
1488         if (running)
1489                 enic_open(netdev);
1490
1491         return 0;
1492 }
1493
1494 #ifdef CONFIG_NET_POLL_CONTROLLER
1495 static void enic_poll_controller(struct net_device *netdev)
1496 {
1497         struct enic *enic = netdev_priv(netdev);
1498         struct vnic_dev *vdev = enic->vdev;
1499
1500         switch (vnic_dev_get_intr_mode(vdev)) {
1501         case VNIC_DEV_INTR_MODE_MSIX:
1502                 enic_isr_msix_rq(enic->pdev->irq, enic);
1503                 enic_isr_msix_wq(enic->pdev->irq, enic);
1504                 break;
1505         case VNIC_DEV_INTR_MODE_MSI:
1506                 enic_isr_msi(enic->pdev->irq, enic);
1507                 break;
1508         case VNIC_DEV_INTR_MODE_INTX:
1509                 enic_isr_legacy(enic->pdev->irq, netdev);
1510                 break;
1511         default:
1512                 break;
1513         }
1514 }
1515 #endif
1516
1517 static int enic_dev_wait(struct vnic_dev *vdev,
1518         int (*start)(struct vnic_dev *, int),
1519         int (*finished)(struct vnic_dev *, int *),
1520         int arg)
1521 {
1522         unsigned long time;
1523         int done;
1524         int err;
1525
1526         BUG_ON(in_interrupt());
1527
1528         err = start(vdev, arg);
1529         if (err)
1530                 return err;
1531
1532         /* Wait for func to complete...2 seconds max
1533          */
1534
1535         time = jiffies + (HZ * 2);
1536         do {
1537
1538                 err = finished(vdev, &done);
1539                 if (err)
1540                         return err;
1541
1542                 if (done)
1543                         return 0;
1544
1545                 schedule_timeout_uninterruptible(HZ / 10);
1546
1547         } while (time_after(time, jiffies));
1548
1549         return -ETIMEDOUT;
1550 }
1551
1552 static int enic_dev_open(struct enic *enic)
1553 {
1554         int err;
1555
1556         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1557                 vnic_dev_open_done, 0);
1558         if (err)
1559                 printk(KERN_ERR PFX
1560                         "vNIC device open failed, err %d.\n", err);
1561
1562         return err;
1563 }
1564
1565 static int enic_dev_soft_reset(struct enic *enic)
1566 {
1567         int err;
1568
1569         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1570                 vnic_dev_soft_reset_done, 0);
1571         if (err)
1572                 printk(KERN_ERR PFX
1573                         "vNIC soft reset failed, err %d.\n", err);
1574
1575         return err;
1576 }
1577
1578 static int enic_set_niccfg(struct enic *enic)
1579 {
1580         const u8 rss_default_cpu = 0;
1581         const u8 rss_hash_type = 0;
1582         const u8 rss_hash_bits = 0;
1583         const u8 rss_base_cpu = 0;
1584         const u8 rss_enable = 0;
1585         const u8 tso_ipid_split_en = 0;
1586         const u8 ig_vlan_strip_en = 1;
1587
1588         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1589          */
1590
1591         return enic_set_nic_cfg(enic,
1592                 rss_default_cpu, rss_hash_type,
1593                 rss_hash_bits, rss_base_cpu,
1594                 rss_enable, tso_ipid_split_en,
1595                 ig_vlan_strip_en);
1596 }
1597
1598 static void enic_reset(struct work_struct *work)
1599 {
1600         struct enic *enic = container_of(work, struct enic, reset);
1601
1602         if (!netif_running(enic->netdev))
1603                 return;
1604
1605         rtnl_lock();
1606
1607         spin_lock(&enic->devcmd_lock);
1608         vnic_dev_hang_notify(enic->vdev);
1609         spin_unlock(&enic->devcmd_lock);
1610
1611         enic_stop(enic->netdev);
1612         enic_dev_soft_reset(enic);
1613         vnic_dev_init(enic->vdev, 0);
1614         enic_reset_mcaddrs(enic);
1615         enic_init_vnic_resources(enic);
1616         enic_set_niccfg(enic);
1617         enic_open(enic->netdev);
1618
1619         rtnl_unlock();
1620 }
1621
1622 static int enic_set_intr_mode(struct enic *enic)
1623 {
1624         unsigned int n = 1;
1625         unsigned int m = 1;
1626         unsigned int i;
1627
1628         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1629          * system capabilities.
1630          *
1631          * Try MSI-X first
1632          *
1633          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1634          * (the second to last INTR is used for WQ/RQ errors)
1635          * (the last INTR is used for notifications)
1636          */
1637
1638         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1639         for (i = 0; i < n + m + 2; i++)
1640                 enic->msix_entry[i].entry = i;
1641
1642         if (enic->config.intr_mode < 1 &&
1643             enic->rq_count >= n &&
1644             enic->wq_count >= m &&
1645             enic->cq_count >= n + m &&
1646             enic->intr_count >= n + m + 2 &&
1647             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1648
1649                 enic->rq_count = n;
1650                 enic->wq_count = m;
1651                 enic->cq_count = n + m;
1652                 enic->intr_count = n + m + 2;
1653
1654                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1655
1656                 return 0;
1657         }
1658
1659         /* Next try MSI
1660          *
1661          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1662          */
1663
1664         if (enic->config.intr_mode < 2 &&
1665             enic->rq_count >= 1 &&
1666             enic->wq_count >= 1 &&
1667             enic->cq_count >= 2 &&
1668             enic->intr_count >= 1 &&
1669             !pci_enable_msi(enic->pdev)) {
1670
1671                 enic->rq_count = 1;
1672                 enic->wq_count = 1;
1673                 enic->cq_count = 2;
1674                 enic->intr_count = 1;
1675
1676                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1677
1678                 return 0;
1679         }
1680
1681         /* Next try INTx
1682          *
1683          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1684          * (the first INTR is used for WQ/RQ)
1685          * (the second INTR is used for WQ/RQ errors)
1686          * (the last INTR is used for notifications)
1687          */
1688
1689         if (enic->config.intr_mode < 3 &&
1690             enic->rq_count >= 1 &&
1691             enic->wq_count >= 1 &&
1692             enic->cq_count >= 2 &&
1693             enic->intr_count >= 3) {
1694
1695                 enic->rq_count = 1;
1696                 enic->wq_count = 1;
1697                 enic->cq_count = 2;
1698                 enic->intr_count = 3;
1699
1700                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1701
1702                 return 0;
1703         }
1704
1705         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1706
1707         return -EINVAL;
1708 }
1709
1710 static void enic_clear_intr_mode(struct enic *enic)
1711 {
1712         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1713         case VNIC_DEV_INTR_MODE_MSIX:
1714                 pci_disable_msix(enic->pdev);
1715                 break;
1716         case VNIC_DEV_INTR_MODE_MSI:
1717                 pci_disable_msi(enic->pdev);
1718                 break;
1719         default:
1720                 break;
1721         }
1722
1723         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
1724 }
1725
1726 static const struct net_device_ops enic_netdev_ops = {
1727         .ndo_open               = enic_open,
1728         .ndo_stop               = enic_stop,
1729         .ndo_start_xmit         = enic_hard_start_xmit,
1730         .ndo_get_stats          = enic_get_stats,
1731         .ndo_validate_addr      = eth_validate_addr,
1732         .ndo_set_mac_address    = eth_mac_addr,
1733         .ndo_set_multicast_list = enic_set_multicast_list,
1734         .ndo_change_mtu         = enic_change_mtu,
1735         .ndo_vlan_rx_register   = enic_vlan_rx_register,
1736         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
1737         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
1738         .ndo_tx_timeout         = enic_tx_timeout,
1739 #ifdef CONFIG_NET_POLL_CONTROLLER
1740         .ndo_poll_controller    = enic_poll_controller,
1741 #endif
1742 };
1743
1744 static void enic_iounmap(struct enic *enic)
1745 {
1746         unsigned int i;
1747
1748         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
1749                 if (enic->bar[i].vaddr)
1750                         iounmap(enic->bar[i].vaddr);
1751 }
1752
1753 static int __devinit enic_probe(struct pci_dev *pdev,
1754         const struct pci_device_id *ent)
1755 {
1756         struct net_device *netdev;
1757         struct enic *enic;
1758         int using_dac = 0;
1759         unsigned int i;
1760         int err;
1761
1762         /* Allocate net device structure and initialize.  Private
1763          * instance data is initialized to zero.
1764          */
1765
1766         netdev = alloc_etherdev(sizeof(struct enic));
1767         if (!netdev) {
1768                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
1769                 return -ENOMEM;
1770         }
1771
1772         pci_set_drvdata(pdev, netdev);
1773
1774         SET_NETDEV_DEV(netdev, &pdev->dev);
1775
1776         enic = netdev_priv(netdev);
1777         enic->netdev = netdev;
1778         enic->pdev = pdev;
1779
1780         /* Setup PCI resources
1781          */
1782
1783         err = pci_enable_device(pdev);
1784         if (err) {
1785                 printk(KERN_ERR PFX
1786                         "Cannot enable PCI device, aborting.\n");
1787                 goto err_out_free_netdev;
1788         }
1789
1790         err = pci_request_regions(pdev, DRV_NAME);
1791         if (err) {
1792                 printk(KERN_ERR PFX
1793                         "Cannot request PCI regions, aborting.\n");
1794                 goto err_out_disable_device;
1795         }
1796
1797         pci_set_master(pdev);
1798
1799         /* Query PCI controller on system for DMA addressing
1800          * limitation for the device.  Try 40-bit first, and
1801          * fail to 32-bit.
1802          */
1803
1804         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
1805         if (err) {
1806                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
1807                 if (err) {
1808                         printk(KERN_ERR PFX
1809                                 "No usable DMA configuration, aborting.\n");
1810                         goto err_out_release_regions;
1811                 }
1812                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
1813                 if (err) {
1814                         printk(KERN_ERR PFX
1815                                 "Unable to obtain 32-bit DMA "
1816                                 "for consistent allocations, aborting.\n");
1817                         goto err_out_release_regions;
1818                 }
1819         } else {
1820                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
1821                 if (err) {
1822                         printk(KERN_ERR PFX
1823                                 "Unable to obtain 40-bit DMA "
1824                                 "for consistent allocations, aborting.\n");
1825                         goto err_out_release_regions;
1826                 }
1827                 using_dac = 1;
1828         }
1829
1830         /* Map vNIC resources from BAR0-5
1831          */
1832
1833         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
1834                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
1835                         continue;
1836                 enic->bar[i].len = pci_resource_len(pdev, i);
1837                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
1838                 if (!enic->bar[i].vaddr) {
1839                         printk(KERN_ERR PFX
1840                                 "Cannot memory-map BAR %d, aborting.\n", i);
1841                         err = -ENODEV;
1842                         goto err_out_iounmap;
1843                 }
1844                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
1845         }
1846
1847         /* Register vNIC device
1848          */
1849
1850         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
1851                 ARRAY_SIZE(enic->bar));
1852         if (!enic->vdev) {
1853                 printk(KERN_ERR PFX
1854                         "vNIC registration failed, aborting.\n");
1855                 err = -ENODEV;
1856                 goto err_out_iounmap;
1857         }
1858
1859         /* Issue device open to get device in known state
1860          */
1861
1862         err = enic_dev_open(enic);
1863         if (err) {
1864                 printk(KERN_ERR PFX
1865                         "vNIC dev open failed, aborting.\n");
1866                 goto err_out_vnic_unregister;
1867         }
1868
1869         /* Issue device init to initialize the vnic-to-switch link.
1870          * We'll start with carrier off and wait for link UP
1871          * notification later to turn on carrier.  We don't need
1872          * to wait here for the vnic-to-switch link initialization
1873          * to complete; link UP notification is the indication that
1874          * the process is complete.
1875          */
1876
1877         netif_carrier_off(netdev);
1878
1879         err = vnic_dev_init(enic->vdev, 0);
1880         if (err) {
1881                 printk(KERN_ERR PFX
1882                         "vNIC dev init failed, aborting.\n");
1883                 goto err_out_dev_close;
1884         }
1885
1886         /* Get vNIC configuration
1887          */
1888
1889         err = enic_get_vnic_config(enic);
1890         if (err) {
1891                 printk(KERN_ERR PFX
1892                         "Get vNIC configuration failed, aborting.\n");
1893                 goto err_out_dev_close;
1894         }
1895
1896         /* Get available resource counts
1897          */
1898
1899         enic_get_res_counts(enic);
1900
1901         /* Set interrupt mode based on resource counts and system
1902          * capabilities
1903          */
1904
1905         err = enic_set_intr_mode(enic);
1906         if (err) {
1907                 printk(KERN_ERR PFX
1908                         "Failed to set intr mode, aborting.\n");
1909                 goto err_out_dev_close;
1910         }
1911
1912         /* Allocate and configure vNIC resources
1913          */
1914
1915         err = enic_alloc_vnic_resources(enic);
1916         if (err) {
1917                 printk(KERN_ERR PFX
1918                         "Failed to alloc vNIC resources, aborting.\n");
1919                 goto err_out_free_vnic_resources;
1920         }
1921
1922         enic_init_vnic_resources(enic);
1923
1924         err = enic_set_niccfg(enic);
1925         if (err) {
1926                 printk(KERN_ERR PFX
1927                         "Failed to config nic, aborting.\n");
1928                 goto err_out_free_vnic_resources;
1929         }
1930
1931         /* Setup notification timer, HW reset task, and locks
1932          */
1933
1934         init_timer(&enic->notify_timer);
1935         enic->notify_timer.function = enic_notify_timer;
1936         enic->notify_timer.data = (unsigned long)enic;
1937
1938         INIT_WORK(&enic->reset, enic_reset);
1939
1940         for (i = 0; i < enic->wq_count; i++)
1941                 spin_lock_init(&enic->wq_lock[i]);
1942
1943         spin_lock_init(&enic->devcmd_lock);
1944
1945         /* Register net device
1946          */
1947
1948         enic->port_mtu = enic->config.mtu;
1949         (void)enic_change_mtu(netdev, enic->port_mtu);
1950
1951         err = enic_set_mac_addr(netdev, enic->mac_addr);
1952         if (err) {
1953                 printk(KERN_ERR PFX
1954                         "Invalid MAC address, aborting.\n");
1955                 goto err_out_free_vnic_resources;
1956         }
1957
1958         netdev->netdev_ops = &enic_netdev_ops;
1959         netdev->watchdog_timeo = 2 * HZ;
1960         netdev->ethtool_ops = &enic_ethtool_ops;
1961
1962         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1963         default:
1964                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
1965                 break;
1966         case VNIC_DEV_INTR_MODE_MSIX:
1967                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
1968                 break;
1969         }
1970
1971         netdev->features |= NETIF_F_HW_VLAN_TX |
1972                 NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_FILTER;
1973         if (ENIC_SETTING(enic, TXCSUM))
1974                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
1975         if (ENIC_SETTING(enic, TSO))
1976                 netdev->features |= NETIF_F_TSO |
1977                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
1978         if (ENIC_SETTING(enic, LRO))
1979                 netdev->features |= NETIF_F_LRO;
1980         if (using_dac)
1981                 netdev->features |= NETIF_F_HIGHDMA;
1982
1983         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
1984
1985         enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
1986         enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
1987         enic->lro_mgr.lro_arr = enic->lro_desc;
1988         enic->lro_mgr.get_skb_header = enic_get_skb_header;
1989         enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
1990         enic->lro_mgr.dev = netdev;
1991         enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
1992         enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
1993
1994         err = register_netdev(netdev);
1995         if (err) {
1996                 printk(KERN_ERR PFX
1997                         "Cannot register net device, aborting.\n");
1998                 goto err_out_free_vnic_resources;
1999         }
2000
2001         return 0;
2002
2003 err_out_free_vnic_resources:
2004         enic_free_vnic_resources(enic);
2005 err_out_dev_close:
2006         vnic_dev_close(enic->vdev);
2007 err_out_vnic_unregister:
2008         enic_clear_intr_mode(enic);
2009         vnic_dev_unregister(enic->vdev);
2010 err_out_iounmap:
2011         enic_iounmap(enic);
2012 err_out_release_regions:
2013         pci_release_regions(pdev);
2014 err_out_disable_device:
2015         pci_disable_device(pdev);
2016 err_out_free_netdev:
2017         pci_set_drvdata(pdev, NULL);
2018         free_netdev(netdev);
2019
2020         return err;
2021 }
2022
2023 static void __devexit enic_remove(struct pci_dev *pdev)
2024 {
2025         struct net_device *netdev = pci_get_drvdata(pdev);
2026
2027         if (netdev) {
2028                 struct enic *enic = netdev_priv(netdev);
2029
2030                 flush_scheduled_work();
2031                 unregister_netdev(netdev);
2032                 enic_free_vnic_resources(enic);
2033                 vnic_dev_close(enic->vdev);
2034                 enic_clear_intr_mode(enic);
2035                 vnic_dev_unregister(enic->vdev);
2036                 enic_iounmap(enic);
2037                 pci_release_regions(pdev);
2038                 pci_disable_device(pdev);
2039                 pci_set_drvdata(pdev, NULL);
2040                 free_netdev(netdev);
2041         }
2042 }
2043
2044 static struct pci_driver enic_driver = {
2045         .name = DRV_NAME,
2046         .id_table = enic_id_table,
2047         .probe = enic_probe,
2048         .remove = __devexit_p(enic_remove),
2049 };
2050
2051 static int __init enic_init_module(void)
2052 {
2053         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2054
2055         return pci_register_driver(&enic_driver);
2056 }
2057
2058 static void __exit enic_cleanup_module(void)
2059 {
2060         pci_unregister_driver(&enic_driver);
2061 }
2062
2063 module_init(enic_init_module);
2064 module_exit(enic_cleanup_module);