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/if_link.h>
  33 #include <linux/ethtool.h>
  34 #include <linux/in.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <net/ip6_checksum.h>
  39
  40 #include "cq_enet_desc.h"
  41 #include "vnic_dev.h"
  42 #include "vnic_intr.h"
  43 #include "vnic_stats.h"
  44 #include "vnic_vic.h"
  45 #include "enic_res.h"
  46 #include "enic.h"
  47
  48 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
  49 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
  50 #define MAX_TSO                         (1 << 16)
  51 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
  52
  53 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
  54 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
  55
  56 /* Supported devices */
  57 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
  58         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  59         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
  60         { 0, }  /* end of table */
  61 };
  62
  63 MODULE_DESCRIPTION(DRV_DESCRIPTION);
  64 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
  65 MODULE_LICENSE("GPL");
  66 MODULE_VERSION(DRV_VERSION);
  67 MODULE_DEVICE_TABLE(pci, enic_id_table);
  68
  69 struct enic_stat {
  70         char name[ETH_GSTRING_LEN];
  71         unsigned int offset;
  72 };
  73
  74 #define ENIC_TX_STAT(stat)      \
  75         { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
  76 #define ENIC_RX_STAT(stat)      \
  77         { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
  78
  79 static const struct enic_stat enic_tx_stats[] = {
  80         ENIC_TX_STAT(tx_frames_ok),
  81         ENIC_TX_STAT(tx_unicast_frames_ok),
  82         ENIC_TX_STAT(tx_multicast_frames_ok),
  83         ENIC_TX_STAT(tx_broadcast_frames_ok),
  84         ENIC_TX_STAT(tx_bytes_ok),
  85         ENIC_TX_STAT(tx_unicast_bytes_ok),
  86         ENIC_TX_STAT(tx_multicast_bytes_ok),
  87         ENIC_TX_STAT(tx_broadcast_bytes_ok),
  88         ENIC_TX_STAT(tx_drops),
  89         ENIC_TX_STAT(tx_errors),
  90         ENIC_TX_STAT(tx_tso),
  91 };
  92
  93 static const struct enic_stat enic_rx_stats[] = {
  94         ENIC_RX_STAT(rx_frames_ok),
  95         ENIC_RX_STAT(rx_frames_total),
  96         ENIC_RX_STAT(rx_unicast_frames_ok),
  97         ENIC_RX_STAT(rx_multicast_frames_ok),
  98         ENIC_RX_STAT(rx_broadcast_frames_ok),
  99         ENIC_RX_STAT(rx_bytes_ok),
 100         ENIC_RX_STAT(rx_unicast_bytes_ok),
 101         ENIC_RX_STAT(rx_multicast_bytes_ok),
 102         ENIC_RX_STAT(rx_broadcast_bytes_ok),
 103         ENIC_RX_STAT(rx_drop),
 104         ENIC_RX_STAT(rx_no_bufs),
 105         ENIC_RX_STAT(rx_errors),
 106         ENIC_RX_STAT(rx_rss),
 107         ENIC_RX_STAT(rx_crc_errors),
 108         ENIC_RX_STAT(rx_frames_64),
 109         ENIC_RX_STAT(rx_frames_127),
 110         ENIC_RX_STAT(rx_frames_255),
 111         ENIC_RX_STAT(rx_frames_511),
 112         ENIC_RX_STAT(rx_frames_1023),
 113         ENIC_RX_STAT(rx_frames_1518),
 114         ENIC_RX_STAT(rx_frames_to_max),
 115 };
 116
 117 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
 118 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
 119
 120 static int enic_is_dynamic(struct enic *enic)
 121 {
 122         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
 123 }
 124
 125 static int enic_get_settings(struct net_device *netdev,
 126         struct ethtool_cmd *ecmd)
 127 {
 128         struct enic *enic = netdev_priv(netdev);
 129
 130         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
 131         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
 132         ecmd->port = PORT_FIBRE;
 133         ecmd->transceiver = XCVR_EXTERNAL;
 134
 135         if (netif_carrier_ok(netdev)) {
 136                 ecmd->speed = vnic_dev_port_speed(enic->vdev);
 137                 ecmd->duplex = DUPLEX_FULL;
 138         } else {
 139                 ecmd->speed = -1;
 140                 ecmd->duplex = -1;
 141         }
 142
 143         ecmd->autoneg = AUTONEG_DISABLE;
 144
 145         return 0;
 146 }
 147
 148 static void enic_get_drvinfo(struct net_device *netdev,
 149         struct ethtool_drvinfo *drvinfo)
 150 {
 151         struct enic *enic = netdev_priv(netdev);
 152         struct vnic_devcmd_fw_info *fw_info;
 153
 154         spin_lock(&enic->devcmd_lock);
 155         vnic_dev_fw_info(enic->vdev, &fw_info);
 156         spin_unlock(&enic->devcmd_lock);
 157
 158         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
 159         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
 160         strncpy(drvinfo->fw_version, fw_info->fw_version,
 161                 sizeof(drvinfo->fw_version));
 162         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
 163                 sizeof(drvinfo->bus_info));
 164 }
 165
 166 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
 167 {
 168         unsigned int i;
 169
 170         switch (stringset) {
 171         case ETH_SS_STATS:
 172                 for (i = 0; i < enic_n_tx_stats; i++) {
 173                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
 174                         data += ETH_GSTRING_LEN;
 175                 }
 176                 for (i = 0; i < enic_n_rx_stats; i++) {
 177                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
 178                         data += ETH_GSTRING_LEN;
 179                 }
 180                 break;
 181         }
 182 }
 183
 184 static int enic_get_sset_count(struct net_device *netdev, int sset)
 185 {
 186         switch (sset) {
 187         case ETH_SS_STATS:
 188                 return enic_n_tx_stats + enic_n_rx_stats;
 189         default:
 190                 return -EOPNOTSUPP;
 191         }
 192 }
 193
 194 static void enic_get_ethtool_stats(struct net_device *netdev,
 195         struct ethtool_stats *stats, u64 *data)
 196 {
 197         struct enic *enic = netdev_priv(netdev);
 198         struct vnic_stats *vstats;
 199         unsigned int i;
 200
 201         spin_lock(&enic->devcmd_lock);
 202         vnic_dev_stats_dump(enic->vdev, &vstats);
 203         spin_unlock(&enic->devcmd_lock);
 204
 205         for (i = 0; i < enic_n_tx_stats; i++)
 206                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
 207         for (i = 0; i < enic_n_rx_stats; i++)
 208                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
 209 }
 210
 211 static u32 enic_get_rx_csum(struct net_device *netdev)
 212 {
 213         struct enic *enic = netdev_priv(netdev);
 214         return enic->csum_rx_enabled;
 215 }
 216
 217 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
 218 {
 219         struct enic *enic = netdev_priv(netdev);
 220
 221         if (data && !ENIC_SETTING(enic, RXCSUM))
 222                 return -EINVAL;
 223
 224         enic->csum_rx_enabled = !!data;
 225
 226         return 0;
 227 }
 228
 229 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
 230 {
 231         struct enic *enic = netdev_priv(netdev);
 232
 233         if (data && !ENIC_SETTING(enic, TXCSUM))
 234                 return -EINVAL;
 235
 236         if (data)
 237                 netdev->features |= NETIF_F_HW_CSUM;
 238         else
 239                 netdev->features &= ~NETIF_F_HW_CSUM;
 240
 241         return 0;
 242 }
 243
 244 static int enic_set_tso(struct net_device *netdev, u32 data)
 245 {
 246         struct enic *enic = netdev_priv(netdev);
 247
 248         if (data && !ENIC_SETTING(enic, TSO))
 249                 return -EINVAL;
 250
 251         if (data)
 252                 netdev->features |=
 253                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
 254         else
 255                 netdev->features &=
 256                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
 257
 258         return 0;
 259 }
 260
 261 static u32 enic_get_msglevel(struct net_device *netdev)
 262 {
 263         struct enic *enic = netdev_priv(netdev);
 264         return enic->msg_enable;
 265 }
 266
 267 static void enic_set_msglevel(struct net_device *netdev, u32 value)
 268 {
 269         struct enic *enic = netdev_priv(netdev);
 270         enic->msg_enable = value;
 271 }
 272
 273 static int enic_get_coalesce(struct net_device *netdev,
 274         struct ethtool_coalesce *ecmd)
 275 {
 276         struct enic *enic = netdev_priv(netdev);
 277
 278         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
 279         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
 280
 281         return 0;
 282 }
 283
 284 static int enic_set_coalesce(struct net_device *netdev,
 285         struct ethtool_coalesce *ecmd)
 286 {
 287         struct enic *enic = netdev_priv(netdev);
 288         u32 tx_coalesce_usecs;
 289         u32 rx_coalesce_usecs;
 290
 291         tx_coalesce_usecs = min_t(u32,
 292                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 293                 ecmd->tx_coalesce_usecs);
 294         rx_coalesce_usecs = min_t(u32,
 295                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 296                 ecmd->rx_coalesce_usecs);
 297
 298         switch (vnic_dev_get_intr_mode(enic->vdev)) {
 299         case VNIC_DEV_INTR_MODE_INTX:
 300                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 301                         return -EINVAL;
 302
 303                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_INTX_WQ_RQ],
 304                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 305                 break;
 306         case VNIC_DEV_INTR_MODE_MSI:
 307                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 308                         return -EINVAL;
 309
 310                 vnic_intr_coalescing_timer_set(&enic->intr[0],
 311                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 312                 break;
 313         case VNIC_DEV_INTR_MODE_MSIX:
 314                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_WQ],
 315                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 316                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_RQ],
 317                         INTR_COALESCE_USEC_TO_HW(rx_coalesce_usecs));
 318                 break;
 319         default:
 320                 break;
 321         }
 322
 323         enic->tx_coalesce_usecs = tx_coalesce_usecs;
 324         enic->rx_coalesce_usecs = rx_coalesce_usecs;
 325
 326         return 0;
 327 }
 328
 329 static const struct ethtool_ops enic_ethtool_ops = {
 330         .get_settings = enic_get_settings,
 331         .get_drvinfo = enic_get_drvinfo,
 332         .get_msglevel = enic_get_msglevel,
 333         .set_msglevel = enic_set_msglevel,
 334         .get_link = ethtool_op_get_link,
 335         .get_strings = enic_get_strings,
 336         .get_sset_count = enic_get_sset_count,
 337         .get_ethtool_stats = enic_get_ethtool_stats,
 338         .get_rx_csum = enic_get_rx_csum,
 339         .set_rx_csum = enic_set_rx_csum,
 340         .get_tx_csum = ethtool_op_get_tx_csum,
 341         .set_tx_csum = enic_set_tx_csum,
 342         .get_sg = ethtool_op_get_sg,
 343         .set_sg = ethtool_op_set_sg,
 344         .get_tso = ethtool_op_get_tso,
 345         .set_tso = enic_set_tso,
 346         .get_coalesce = enic_get_coalesce,
 347         .set_coalesce = enic_set_coalesce,
 348         .get_flags = ethtool_op_get_flags,
 349         .set_flags = ethtool_op_set_flags,
 350 };
 351
 352 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 353 {
 354         struct enic *enic = vnic_dev_priv(wq->vdev);
 355
 356         if (buf->sop)
 357                 pci_unmap_single(enic->pdev, buf->dma_addr,
 358                         buf->len, PCI_DMA_TODEVICE);
 359         else
 360                 pci_unmap_page(enic->pdev, buf->dma_addr,
 361                         buf->len, PCI_DMA_TODEVICE);
 362
 363         if (buf->os_buf)
 364                 dev_kfree_skb_any(buf->os_buf);
 365 }
 366
 367 static void enic_wq_free_buf(struct vnic_wq *wq,
 368         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 369 {
 370         enic_free_wq_buf(wq, buf);
 371 }
 372
 373 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 374         u8 type, u16 q_number, u16 completed_index, void *opaque)
 375 {
 376         struct enic *enic = vnic_dev_priv(vdev);
 377
 378         spin_lock(&enic->wq_lock[q_number]);
 379
 380         vnic_wq_service(&enic->wq[q_number], cq_desc,
 381                 completed_index, enic_wq_free_buf,
 382                 opaque);
 383
 384         if (netif_queue_stopped(enic->netdev) &&
 385             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 386             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 387                 netif_wake_queue(enic->netdev);
 388
 389         spin_unlock(&enic->wq_lock[q_number]);
 390
 391         return 0;
 392 }
 393
 394 static void enic_log_q_error(struct enic *enic)
 395 {
 396         unsigned int i;
 397         u32 error_status;
 398
 399         for (i = 0; i < enic->wq_count; i++) {
 400                 error_status = vnic_wq_error_status(&enic->wq[i]);
 401                 if (error_status)
 402                         printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
 403                                 enic->netdev->name, i, error_status);
 404         }
 405
 406         for (i = 0; i < enic->rq_count; i++) {
 407                 error_status = vnic_rq_error_status(&enic->rq[i]);
 408                 if (error_status)
 409                         printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
 410                                 enic->netdev->name, i, error_status);
 411         }
 412 }
 413
 414 static void enic_link_check(struct enic *enic)
 415 {
 416         int link_status = vnic_dev_link_status(enic->vdev);
 417         int carrier_ok = netif_carrier_ok(enic->netdev);
 418
 419         if (link_status && !carrier_ok) {
 420                 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
 421                 netif_carrier_on(enic->netdev);
 422         } else if (!link_status && carrier_ok) {
 423                 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
 424                 netif_carrier_off(enic->netdev);
 425         }
 426 }
 427
 428 static void enic_mtu_check(struct enic *enic)
 429 {
 430         u32 mtu = vnic_dev_mtu(enic->vdev);
 431
 432         if (mtu && mtu != enic->port_mtu) {
 433                 enic->port_mtu = mtu;
 434                 if (mtu < enic->netdev->mtu)
 435                         printk(KERN_WARNING PFX
 436                                 "%s: interface MTU (%d) set higher "
 437                                 "than switch port MTU (%d)\n",
 438                                 enic->netdev->name, enic->netdev->mtu, mtu);
 439         }
 440 }
 441
 442 static void enic_msglvl_check(struct enic *enic)
 443 {
 444         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 445
 446         if (msg_enable != enic->msg_enable) {
 447                 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
 448                         enic->netdev->name, enic->msg_enable, msg_enable);
 449                 enic->msg_enable = msg_enable;
 450         }
 451 }
 452
 453 static void enic_notify_check(struct enic *enic)
 454 {
 455         enic_msglvl_check(enic);
 456         enic_mtu_check(enic);
 457         enic_link_check(enic);
 458 }
 459
 460 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 461
 462 static irqreturn_t enic_isr_legacy(int irq, void *data)
 463 {
 464         struct net_device *netdev = data;
 465         struct enic *enic = netdev_priv(netdev);
 466         u32 pba;
 467
 468         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
 469
 470         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 471         if (!pba) {
 472                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 473                 return IRQ_NONE;        /* not our interrupt */
 474         }
 475
 476         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
 477                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
 478                 enic_notify_check(enic);
 479         }
 480
 481         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
 482                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
 483                 enic_log_q_error(enic);
 484                 /* schedule recovery from WQ/RQ error */
 485                 schedule_work(&enic->reset);
 486                 return IRQ_HANDLED;
 487         }
 488
 489         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
 490                 if (napi_schedule_prep(&enic->napi))
 491                         __napi_schedule(&enic->napi);
 492         } else {
 493                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 494         }
 495
 496         return IRQ_HANDLED;
 497 }
 498
 499 static irqreturn_t enic_isr_msi(int irq, void *data)
 500 {
 501         struct enic *enic = data;
 502
 503         /* With MSI, there is no sharing of interrupts, so this is
 504          * our interrupt and there is no need to ack it.  The device
 505          * is not providing per-vector masking, so the OS will not
 506          * write to PCI config space to mask/unmask the interrupt.
 507          * We're using mask_on_assertion for MSI, so the device
 508          * automatically masks the interrupt when the interrupt is
 509          * generated.  Later, when exiting polling, the interrupt
 510          * will be unmasked (see enic_poll).
 511          *
 512          * Also, the device uses the same PCIe Traffic Class (TC)
 513          * for Memory Write data and MSI, so there are no ordering
 514          * issues; the MSI will always arrive at the Root Complex
 515          * _after_ corresponding Memory Writes (i.e. descriptor
 516          * writes).
 517          */
 518
 519         napi_schedule(&enic->napi);
 520
 521         return IRQ_HANDLED;
 522 }
 523
 524 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
 525 {
 526         struct enic *enic = data;
 527
 528         /* schedule NAPI polling for RQ cleanup */
 529         napi_schedule(&enic->napi);
 530
 531         return IRQ_HANDLED;
 532 }
 533
 534 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
 535 {
 536         struct enic *enic = data;
 537         unsigned int wq_work_to_do = -1; /* no limit */
 538         unsigned int wq_work_done;
 539
 540         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
 541                 wq_work_to_do, enic_wq_service, NULL);
 542
 543         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
 544                 wq_work_done,
 545                 1 /* unmask intr */,
 546                 1 /* reset intr timer */);
 547
 548         return IRQ_HANDLED;
 549 }
 550
 551 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 552 {
 553         struct enic *enic = data;
 554
 555         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
 556
 557         enic_log_q_error(enic);
 558
 559         /* schedule recovery from WQ/RQ error */
 560         schedule_work(&enic->reset);
 561
 562         return IRQ_HANDLED;
 563 }
 564
 565 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 566 {
 567         struct enic *enic = data;
 568
 569         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
 570         enic_notify_check(enic);
 571
 572         return IRQ_HANDLED;
 573 }
 574
 575 static inline void enic_queue_wq_skb_cont(struct enic *enic,
 576         struct vnic_wq *wq, struct sk_buff *skb,
 577         unsigned int len_left)
 578 {
 579         skb_frag_t *frag;
 580
 581         /* Queue additional data fragments */
 582         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 583                 len_left -= frag->size;
 584                 enic_queue_wq_desc_cont(wq, skb,
 585                         pci_map_page(enic->pdev, frag->page,
 586                                 frag->page_offset, frag->size,
 587                                 PCI_DMA_TODEVICE),
 588                         frag->size,
 589                         (len_left == 0));       /* EOP? */
 590         }
 591 }
 592
 593 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
 594         struct vnic_wq *wq, struct sk_buff *skb,
 595         int vlan_tag_insert, unsigned int vlan_tag)
 596 {
 597         unsigned int head_len = skb_headlen(skb);
 598         unsigned int len_left = skb->len - head_len;
 599         int eop = (len_left == 0);
 600
 601         /* Queue the main skb fragment. The fragments are no larger
 602          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 603          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 604          * per fragment is queued.
 605          */
 606         enic_queue_wq_desc(wq, skb,
 607                 pci_map_single(enic->pdev, skb->data,
 608                         head_len, PCI_DMA_TODEVICE),
 609                 head_len,
 610                 vlan_tag_insert, vlan_tag,
 611                 eop);
 612
 613         if (!eop)
 614                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 615 }
 616
 617 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
 618         struct vnic_wq *wq, struct sk_buff *skb,
 619         int vlan_tag_insert, unsigned int vlan_tag)
 620 {
 621         unsigned int head_len = skb_headlen(skb);
 622         unsigned int len_left = skb->len - head_len;
 623         unsigned int hdr_len = skb_transport_offset(skb);
 624         unsigned int csum_offset = hdr_len + skb->csum_offset;
 625         int eop = (len_left == 0);
 626
 627         /* Queue the main skb fragment. The fragments are no larger
 628          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 629          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 630          * per fragment is queued.
 631          */
 632         enic_queue_wq_desc_csum_l4(wq, skb,
 633                 pci_map_single(enic->pdev, skb->data,
 634                         head_len, PCI_DMA_TODEVICE),
 635                 head_len,
 636                 csum_offset,
 637                 hdr_len,
 638                 vlan_tag_insert, vlan_tag,
 639                 eop);
 640
 641         if (!eop)
 642                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 643 }
 644
 645 static inline void enic_queue_wq_skb_tso(struct enic *enic,
 646         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
 647         int vlan_tag_insert, unsigned int vlan_tag)
 648 {
 649         unsigned int frag_len_left = skb_headlen(skb);
 650         unsigned int len_left = skb->len - frag_len_left;
 651         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 652         int eop = (len_left == 0);
 653         unsigned int len;
 654         dma_addr_t dma_addr;
 655         unsigned int offset = 0;
 656         skb_frag_t *frag;
 657
 658         /* Preload TCP csum field with IP pseudo hdr calculated
 659          * with IP length set to zero.  HW will later add in length
 660          * to each TCP segment resulting from the TSO.
 661          */
 662
 663         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 664                 ip_hdr(skb)->check = 0;
 665                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 666                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 667         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 668                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 669                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 670         }
 671
 672         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 673          * for the main skb fragment
 674          */
 675         while (frag_len_left) {
 676                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 677                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
 678                                 len, PCI_DMA_TODEVICE);
 679                 enic_queue_wq_desc_tso(wq, skb,
 680                         dma_addr,
 681                         len,
 682                         mss, hdr_len,
 683                         vlan_tag_insert, vlan_tag,
 684                         eop && (len == frag_len_left));
 685                 frag_len_left -= len;
 686                 offset += len;
 687         }
 688
 689         if (eop)
 690                 return;
 691
 692         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 693          * for additional data fragments
 694          */
 695         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 696                 len_left -= frag->size;
 697                 frag_len_left = frag->size;
 698                 offset = frag->page_offset;
 699
 700                 while (frag_len_left) {
 701                         len = min(frag_len_left,
 702                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 703                         dma_addr = pci_map_page(enic->pdev, frag->page,
 704                                 offset, len,
 705                                 PCI_DMA_TODEVICE);
 706                         enic_queue_wq_desc_cont(wq, skb,
 707                                 dma_addr,
 708                                 len,
 709                                 (len_left == 0) &&
 710                                 (len == frag_len_left));        /* EOP? */
 711                         frag_len_left -= len;
 712                         offset += len;
 713                 }
 714         }
 715 }
 716
 717 static inline void enic_queue_wq_skb(struct enic *enic,
 718         struct vnic_wq *wq, struct sk_buff *skb)
 719 {
 720         unsigned int mss = skb_shinfo(skb)->gso_size;
 721         unsigned int vlan_tag = 0;
 722         int vlan_tag_insert = 0;
 723
 724         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
 725                 /* VLAN tag from trunking driver */
 726                 vlan_tag_insert = 1;
 727                 vlan_tag = vlan_tx_tag_get(skb);
 728         }
 729
 730         if (mss)
 731                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
 732                         vlan_tag_insert, vlan_tag);
 733         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 734                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
 735                         vlan_tag_insert, vlan_tag);
 736         else
 737                 enic_queue_wq_skb_vlan(enic, wq, skb,
 738                         vlan_tag_insert, vlan_tag);
 739 }
 740
 741 /* netif_tx_lock held, process context with BHs disabled, or BH */
 742 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 743         struct net_device *netdev)
 744 {
 745         struct enic *enic = netdev_priv(netdev);
 746         struct vnic_wq *wq = &enic->wq[0];
 747         unsigned long flags;
 748
 749         if (skb->len <= 0) {
 750                 dev_kfree_skb(skb);
 751                 return NETDEV_TX_OK;
 752         }
 753
 754         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 755          * which is very likely.  In the off chance it's going to take
 756          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 757          */
 758
 759         if (skb_shinfo(skb)->gso_size == 0 &&
 760             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 761             skb_linearize(skb)) {
 762                 dev_kfree_skb(skb);
 763                 return NETDEV_TX_OK;
 764         }
 765
 766         spin_lock_irqsave(&enic->wq_lock[0], flags);
 767
 768         if (vnic_wq_desc_avail(wq) <
 769             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 770                 netif_stop_queue(netdev);
 771                 /* This is a hard error, log it */
 772                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
 773                         "queue awake!\n", netdev->name);
 774                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 775                 return NETDEV_TX_BUSY;
 776         }
 777
 778         enic_queue_wq_skb(enic, wq, skb);
 779
 780         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 781                 netif_stop_queue(netdev);
 782
 783         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 784
 785         return NETDEV_TX_OK;
 786 }
 787
 788 /* dev_base_lock rwlock held, nominally process context */
 789 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
 790 {
 791         struct enic *enic = netdev_priv(netdev);
 792         struct net_device_stats *net_stats = &netdev->stats;
 793         struct vnic_stats *stats;
 794
 795         spin_lock(&enic->devcmd_lock);
 796         vnic_dev_stats_dump(enic->vdev, &stats);
 797         spin_unlock(&enic->devcmd_lock);
 798
 799         net_stats->tx_packets = stats->tx.tx_frames_ok;
 800         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 801         net_stats->tx_errors = stats->tx.tx_errors;
 802         net_stats->tx_dropped = stats->tx.tx_drops;
 803
 804         net_stats->rx_packets = stats->rx.rx_frames_ok;
 805         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 806         net_stats->rx_errors = stats->rx.rx_errors;
 807         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 808         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 809         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 810         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 811
 812         return net_stats;
 813 }
 814
 815 static void enic_reset_mcaddrs(struct enic *enic)
 816 {
 817         enic->mc_count = 0;
 818 }
 819
 820 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 821 {
 822         struct enic *enic = netdev_priv(netdev);
 823
 824         if (enic_is_dynamic(enic)) {
 825                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 826                         return -EADDRNOTAVAIL;
 827         } else {
 828                 if (!is_valid_ether_addr(addr))
 829                         return -EADDRNOTAVAIL;
 830         }
 831
 832         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 833
 834         return 0;
 835 }
 836
 837 static int enic_dev_add_station_addr(struct enic *enic)
 838 {
 839         int err = 0;
 840
 841         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 842                 spin_lock(&enic->devcmd_lock);
 843                 err = vnic_dev_add_addr(enic->vdev, enic->netdev->dev_addr);
 844                 spin_unlock(&enic->devcmd_lock);
 845         }
 846
 847         return err;
 848 }
 849
 850 static int enic_dev_del_station_addr(struct enic *enic)
 851 {
 852         int err = 0;
 853
 854         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 855                 spin_lock(&enic->devcmd_lock);
 856                 err = vnic_dev_del_addr(enic->vdev, enic->netdev->dev_addr);
 857                 spin_unlock(&enic->devcmd_lock);
 858         }
 859
 860         return err;
 861 }
 862
 863 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 864 {
 865         struct enic *enic = netdev_priv(netdev);
 866         struct sockaddr *saddr = p;
 867         char *addr = saddr->sa_data;
 868         int err;
 869
 870         if (netif_running(enic->netdev)) {
 871                 err = enic_dev_del_station_addr(enic);
 872                 if (err)
 873                         return err;
 874         }
 875
 876         err = enic_set_mac_addr(netdev, addr);
 877         if (err)
 878                 return err;
 879
 880         if (netif_running(enic->netdev)) {
 881                 err = enic_dev_add_station_addr(enic);
 882                 if (err)
 883                         return err;
 884         }
 885
 886         return err;
 887 }
 888
 889 static int enic_set_mac_address(struct net_device *netdev, void *p)
 890 {
 891         return -EOPNOTSUPP;
 892 }
 893
 894 /* netif_tx_lock held, BHs disabled */
 895 static void enic_set_multicast_list(struct net_device *netdev)
 896 {
 897         struct enic *enic = netdev_priv(netdev);
 898         struct netdev_hw_addr *ha;
 899         int directed = 1;
 900         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 901         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 902         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
 903         unsigned int mc_count = netdev_mc_count(netdev);
 904         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 905                 mc_count > ENIC_MULTICAST_PERFECT_FILTERS;
 906         unsigned int flags = netdev->flags | (allmulti ? IFF_ALLMULTI : 0);
 907         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
 908         unsigned int i, j;
 909
 910         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
 911                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
 912
 913         spin_lock(&enic->devcmd_lock);
 914
 915         if (enic->flags != flags) {
 916                 enic->flags = flags;
 917                 vnic_dev_packet_filter(enic->vdev, directed,
 918                         multicast, broadcast, promisc, allmulti);
 919         }
 920
 921         /* Is there an easier way?  Trying to minimize to
 922          * calls to add/del multicast addrs.  We keep the
 923          * addrs from the last call in enic->mc_addr and
 924          * look for changes to add/del.
 925          */
 926
 927         i = 0;
 928         netdev_for_each_mc_addr(ha, netdev) {
 929                 if (i == mc_count)
 930                         break;
 931                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
 932         }
 933
 934         for (i = 0; i < enic->mc_count; i++) {
 935                 for (j = 0; j < mc_count; j++)
 936                         if (compare_ether_addr(enic->mc_addr[i],
 937                                 mc_addr[j]) == 0)
 938                                 break;
 939                 if (j == mc_count)
 940                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
 941         }
 942
 943         for (i = 0; i < mc_count; i++) {
 944                 for (j = 0; j < enic->mc_count; j++)
 945                         if (compare_ether_addr(mc_addr[i],
 946                                 enic->mc_addr[j]) == 0)
 947                                 break;
 948                 if (j == enic->mc_count)
 949                         enic_add_multicast_addr(enic, mc_addr[i]);
 950         }
 951
 952         /* Save the list to compare against next time
 953          */
 954
 955         for (i = 0; i < mc_count; i++)
 956                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
 957
 958         enic->mc_count = mc_count;
 959
 960         spin_unlock(&enic->devcmd_lock);
 961 }
 962
 963 /* rtnl lock is held */
 964 static void enic_vlan_rx_register(struct net_device *netdev,
 965         struct vlan_group *vlan_group)
 966 {
 967         struct enic *enic = netdev_priv(netdev);
 968         enic->vlan_group = vlan_group;
 969 }
 970
 971 /* rtnl lock is held */
 972 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
 973 {
 974         struct enic *enic = netdev_priv(netdev);
 975
 976         spin_lock(&enic->devcmd_lock);
 977         enic_add_vlan(enic, vid);
 978         spin_unlock(&enic->devcmd_lock);
 979 }
 980
 981 /* rtnl lock is held */
 982 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
 983 {
 984         struct enic *enic = netdev_priv(netdev);
 985
 986         spin_lock(&enic->devcmd_lock);
 987         enic_del_vlan(enic, vid);
 988         spin_unlock(&enic->devcmd_lock);
 989 }
 990
 991 /* netif_tx_lock held, BHs disabled */
 992 static void enic_tx_timeout(struct net_device *netdev)
 993 {
 994         struct enic *enic = netdev_priv(netdev);
 995         schedule_work(&enic->reset);
 996 }
 997
 998 static int enic_vnic_dev_deinit(struct enic *enic)
 999 {
1000         int err;
1001
1002         spin_lock(&enic->devcmd_lock);
1003         err = vnic_dev_deinit(enic->vdev);
1004         spin_unlock(&enic->devcmd_lock);
1005
1006         return err;
1007 }
1008
1009 static int enic_dev_init_prov(struct enic *enic, struct vic_provinfo *vp)
1010 {
1011         int err;
1012
1013         spin_lock(&enic->devcmd_lock);
1014         err = vnic_dev_init_prov(enic->vdev,
1015                 (u8 *)vp, vic_provinfo_size(vp));
1016         spin_unlock(&enic->devcmd_lock);
1017
1018         return err;
1019 }
1020
1021 static int enic_dev_init_done(struct enic *enic, int *done, int *error)
1022 {
1023         int err;
1024
1025         spin_lock(&enic->devcmd_lock);
1026         err = vnic_dev_init_done(enic->vdev, done, error);
1027         spin_unlock(&enic->devcmd_lock);
1028
1029         return err;
1030 }
1031
1032 static int enic_set_port_profile(struct enic *enic, u8 request, u8 *mac,
1033         char *name, u8 *instance_uuid, u8 *host_uuid)
1034 {
1035         struct vic_provinfo *vp;
1036         u8 oui[3] = VIC_PROVINFO_CISCO_OUI;
1037         unsigned short *uuid;
1038         char uuid_str[38];
1039         static char *uuid_fmt = "%04X%04X-%04X-%04X-%04X-%04X%04X%04X";
1040         int err;
1041
1042         if (!name)
1043                 return -EINVAL;
1044
1045         if (!is_valid_ether_addr(mac))
1046                 return -EADDRNOTAVAIL;
1047
1048         vp = vic_provinfo_alloc(GFP_KERNEL, oui, VIC_PROVINFO_LINUX_TYPE);
1049         if (!vp)
1050                 return -ENOMEM;
1051
1052         vic_provinfo_add_tlv(vp,
1053                 VIC_LINUX_PROV_TLV_PORT_PROFILE_NAME_STR,
1054                 strlen(name) + 1, name);
1055
1056         vic_provinfo_add_tlv(vp,
1057                 VIC_LINUX_PROV_TLV_CLIENT_MAC_ADDR,
1058                 ETH_ALEN, mac);
1059
1060         if (instance_uuid) {
1061                 uuid = (unsigned short *)instance_uuid;
1062                 sprintf(uuid_str, uuid_fmt,
1063                         uuid[0], uuid[1], uuid[2], uuid[3],
1064                         uuid[4], uuid[5], uuid[6], uuid[7]);
1065                 vic_provinfo_add_tlv(vp,
1066                         VIC_LINUX_PROV_TLV_CLIENT_UUID_STR,
1067                         sizeof(uuid_str), uuid_str);
1068         }
1069
1070         if (host_uuid) {
1071                 uuid = (unsigned short *)host_uuid;
1072                 sprintf(uuid_str, uuid_fmt,
1073                         uuid[0], uuid[1], uuid[2], uuid[3],
1074                         uuid[4], uuid[5], uuid[6], uuid[7]);
1075                 vic_provinfo_add_tlv(vp,
1076                         VIC_LINUX_PROV_TLV_HOST_UUID_STR,
1077                         sizeof(uuid_str), uuid_str);
1078         }
1079
1080         err = enic_vnic_dev_deinit(enic);
1081         if (err)
1082                 goto err_out;
1083
1084         memset(&enic->pp, 0, sizeof(enic->pp));
1085
1086         err = enic_dev_init_prov(enic, vp);
1087         if (err)
1088                 goto err_out;
1089
1090         enic->pp.request = request;
1091         memcpy(enic->pp.name, name, PORT_PROFILE_MAX);
1092         if (instance_uuid)
1093                 memcpy(enic->pp.instance_uuid,
1094                         instance_uuid, PORT_UUID_MAX);
1095         if (host_uuid)
1096                 memcpy(enic->pp.host_uuid,
1097                         host_uuid, PORT_UUID_MAX);
1098
1099 err_out:
1100         vic_provinfo_free(vp);
1101
1102         return err;
1103 }
1104
1105 static int enic_unset_port_profile(struct enic *enic)
1106 {
1107         memset(&enic->pp, 0, sizeof(enic->pp));
1108         return enic_vnic_dev_deinit(enic);
1109 }
1110
1111 static int enic_set_vf_port(struct net_device *netdev, int vf,
1112         struct nlattr *port[])
1113 {
1114         struct enic *enic = netdev_priv(netdev);
1115         char *name = NULL;
1116         u8 *instance_uuid = NULL;
1117         u8 *host_uuid = NULL;
1118         u8 request = PORT_REQUEST_DISASSOCIATE;
1119
1120         /* don't support VFs, yet */
1121         if (vf != PORT_SELF_VF)
1122                 return -EOPNOTSUPP;
1123
1124         if (port[IFLA_PORT_REQUEST])
1125                 request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1126
1127         switch (request) {
1128         case PORT_REQUEST_ASSOCIATE:
1129
1130                 if (port[IFLA_PORT_PROFILE])
1131                         name = nla_data(port[IFLA_PORT_PROFILE]);
1132
1133                 if (port[IFLA_PORT_INSTANCE_UUID])
1134                         instance_uuid =
1135                                 nla_data(port[IFLA_PORT_INSTANCE_UUID]);
1136
1137                 if (port[IFLA_PORT_HOST_UUID])
1138                         host_uuid = nla_data(port[IFLA_PORT_HOST_UUID]);
1139
1140                 return enic_set_port_profile(enic, request,
1141                         netdev->dev_addr, name,
1142                         instance_uuid, host_uuid);
1143
1144         case PORT_REQUEST_DISASSOCIATE:
1145
1146                 return enic_unset_port_profile(enic);
1147
1148         default:
1149                 break;
1150         }
1151
1152         return -EOPNOTSUPP;
1153 }
1154
1155 static int enic_get_vf_port(struct net_device *netdev, int vf,
1156         struct sk_buff *skb)
1157 {
1158         struct enic *enic = netdev_priv(netdev);
1159         int err, error, done;
1160         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1161
1162         /* don't support VFs, yet */
1163         if (vf != PORT_SELF_VF)
1164                 return -EOPNOTSUPP;
1165
1166         err = enic_dev_init_done(enic, &done, &error);
1167
1168         if (err)
1169                 return err;
1170
1171         switch (error) {
1172         case ERR_SUCCESS:
1173                 if (!done)
1174                         response = PORT_PROFILE_RESPONSE_INPROGRESS;
1175                 break;
1176         case ERR_EINVAL:
1177                 response = PORT_PROFILE_RESPONSE_INVALID;
1178                 break;
1179         case ERR_EBADSTATE:
1180                 response = PORT_PROFILE_RESPONSE_BADSTATE;
1181                 break;
1182         case ERR_ENOMEM:
1183                 response = PORT_PROFILE_RESPONSE_INSUFFICIENT_RESOURCES;
1184                 break;
1185         default:
1186                 response = PORT_PROFILE_RESPONSE_ERROR;
1187                 break;
1188         }
1189
1190         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1191         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1192         NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1193                 enic->pp.name);
1194         NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1195                 enic->pp.instance_uuid);
1196         NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1197                 enic->pp.host_uuid);
1198
1199         return 0;
1200
1201 nla_put_failure:
1202         return -EMSGSIZE;
1203 }
1204
1205 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1206 {
1207         struct enic *enic = vnic_dev_priv(rq->vdev);
1208
1209         if (!buf->os_buf)
1210                 return;
1211
1212         pci_unmap_single(enic->pdev, buf->dma_addr,
1213                 buf->len, PCI_DMA_FROMDEVICE);
1214         dev_kfree_skb_any(buf->os_buf);
1215 }
1216
1217 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1218 {
1219         struct enic *enic = vnic_dev_priv(rq->vdev);
1220         struct net_device *netdev = enic->netdev;
1221         struct sk_buff *skb;
1222         unsigned int len = netdev->mtu + ETH_HLEN;
1223         unsigned int os_buf_index = 0;
1224         dma_addr_t dma_addr;
1225
1226         skb = netdev_alloc_skb_ip_align(netdev, len);
1227         if (!skb)
1228                 return -ENOMEM;
1229
1230         dma_addr = pci_map_single(enic->pdev, skb->data,
1231                 len, PCI_DMA_FROMDEVICE);
1232
1233         enic_queue_rq_desc(rq, skb, os_buf_index,
1234                 dma_addr, len);
1235
1236         return 0;
1237 }
1238
1239 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
1240 {
1241         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
1242
1243         if (vnic_rq_posting_soon(rq)) {
1244
1245                 /* SW workaround for A0 HW erratum: if we're just about
1246                  * to write posted_index, insert a dummy desc
1247                  * of type resvd
1248                  */
1249
1250                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
1251                 vnic_rq_post(rq, 0, 0, 0, 0);
1252         } else {
1253                 return enic_rq_alloc_buf(rq);
1254         }
1255
1256         return 0;
1257 }
1258
1259 static int enic_set_rq_alloc_buf(struct enic *enic)
1260 {
1261         enum vnic_dev_hw_version hw_ver;
1262         int err;
1263
1264         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
1265         if (err)
1266                 return err;
1267
1268         switch (hw_ver) {
1269         case VNIC_DEV_HW_VER_A1:
1270                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
1271                 break;
1272         case VNIC_DEV_HW_VER_A2:
1273         case VNIC_DEV_HW_VER_UNKNOWN:
1274                 enic->rq_alloc_buf = enic_rq_alloc_buf;
1275                 break;
1276         default:
1277                 return -ENODEV;
1278         }
1279
1280         return 0;
1281 }
1282
1283 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
1284         void **tcph, u64 *hdr_flags, void *priv)
1285 {
1286         struct cq_enet_rq_desc *cq_desc = priv;
1287         unsigned int ip_len;
1288         struct iphdr *iph;
1289
1290         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1291         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1292         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1293         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1294         u8 packet_error;
1295         u16 q_number, completed_index, bytes_written, vlan, checksum;
1296         u32 rss_hash;
1297
1298         cq_enet_rq_desc_dec(cq_desc,
1299                 &type, &color, &q_number, &completed_index,
1300                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1301                 &csum_not_calc, &rss_hash, &bytes_written,
1302                 &packet_error, &vlan_stripped, &vlan, &checksum,
1303                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1304                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1305                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1306                 &fcs_ok);
1307
1308         if (!(ipv4 && tcp && !ipv4_fragment))
1309                 return -1;
1310
1311         skb_reset_network_header(skb);
1312         iph = ip_hdr(skb);
1313
1314         ip_len = ip_hdrlen(skb);
1315         skb_set_transport_header(skb, ip_len);
1316
1317         /* check if ip header and tcp header are complete */
1318         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
1319                 return -1;
1320
1321         *hdr_flags = LRO_IPV4 | LRO_TCP;
1322         *tcph = tcp_hdr(skb);
1323         *iphdr = iph;
1324
1325         return 0;
1326 }
1327
1328 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1329         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1330         int skipped, void *opaque)
1331 {
1332         struct enic *enic = vnic_dev_priv(rq->vdev);
1333         struct net_device *netdev = enic->netdev;
1334         struct sk_buff *skb;
1335
1336         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1337         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1338         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1339         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1340         u8 packet_error;
1341         u16 q_number, completed_index, bytes_written, vlan, checksum;
1342         u32 rss_hash;
1343
1344         if (skipped)
1345                 return;
1346
1347         skb = buf->os_buf;
1348         prefetch(skb->data - NET_IP_ALIGN);
1349         pci_unmap_single(enic->pdev, buf->dma_addr,
1350                 buf->len, PCI_DMA_FROMDEVICE);
1351
1352         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1353                 &type, &color, &q_number, &completed_index,
1354                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1355                 &csum_not_calc, &rss_hash, &bytes_written,
1356                 &packet_error, &vlan_stripped, &vlan, &checksum,
1357                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1358                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1359                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1360                 &fcs_ok);
1361
1362         if (packet_error) {
1363
1364                 if (!fcs_ok) {
1365                         if (bytes_written > 0)
1366                                 enic->rq_bad_fcs++;
1367                         else if (bytes_written == 0)
1368                                 enic->rq_truncated_pkts++;
1369                 }
1370
1371                 dev_kfree_skb_any(skb);
1372
1373                 return;
1374         }
1375
1376         if (eop && bytes_written > 0) {
1377
1378                 /* Good receive
1379                  */
1380
1381                 skb_put(skb, bytes_written);
1382                 skb->protocol = eth_type_trans(skb, netdev);
1383
1384                 if (enic->csum_rx_enabled && !csum_not_calc) {
1385                         skb->csum = htons(checksum);
1386                         skb->ip_summed = CHECKSUM_COMPLETE;
1387                 }
1388
1389                 skb->dev = netdev;
1390
1391                 if (enic->vlan_group && vlan_stripped) {
1392
1393                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1394                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
1395                                         skb, enic->vlan_group,
1396                                         vlan, cq_desc);
1397                         else
1398                                 vlan_hwaccel_receive_skb(skb,
1399                                         enic->vlan_group, vlan);
1400
1401                 } else {
1402
1403                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1404                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
1405                         else
1406                                 netif_receive_skb(skb);
1407
1408                 }
1409
1410         } else {
1411
1412                 /* Buffer overflow
1413                  */
1414
1415                 dev_kfree_skb_any(skb);
1416         }
1417 }
1418
1419 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1420         u8 type, u16 q_number, u16 completed_index, void *opaque)
1421 {
1422         struct enic *enic = vnic_dev_priv(vdev);
1423
1424         vnic_rq_service(&enic->rq[q_number], cq_desc,
1425                 completed_index, VNIC_RQ_RETURN_DESC,
1426                 enic_rq_indicate_buf, opaque);
1427
1428         return 0;
1429 }
1430
1431 static int enic_poll(struct napi_struct *napi, int budget)
1432 {
1433         struct enic *enic = container_of(napi, struct enic, napi);
1434         struct net_device *netdev = enic->netdev;
1435         unsigned int rq_work_to_do = budget;
1436         unsigned int wq_work_to_do = -1; /* no limit */
1437         unsigned int  work_done, rq_work_done, wq_work_done;
1438         int err;
1439
1440         /* Service RQ (first) and WQ
1441          */
1442
1443         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1444                 rq_work_to_do, enic_rq_service, NULL);
1445
1446         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1447                 wq_work_to_do, enic_wq_service, NULL);
1448
1449         /* Accumulate intr event credits for this polling
1450          * cycle.  An intr event is the completion of a
1451          * a WQ or RQ packet.
1452          */
1453
1454         work_done = rq_work_done + wq_work_done;
1455
1456         if (work_done > 0)
1457                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1458                         work_done,
1459                         0 /* don't unmask intr */,
1460                         0 /* don't reset intr timer */);
1461
1462         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1463
1464         /* Buffer allocation failed. Stay in polling
1465          * mode so we can try to fill the ring again.
1466          */
1467
1468         if (err)
1469                 rq_work_done = rq_work_to_do;
1470
1471         if (rq_work_done < rq_work_to_do) {
1472
1473                 /* Some work done, but not enough to stay in polling,
1474                  * flush all LROs and exit polling
1475                  */
1476
1477                 if (netdev->features & NETIF_F_LRO)
1478                         lro_flush_all(&enic->lro_mgr);
1479
1480                 napi_complete(napi);
1481                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1482         }
1483
1484         return rq_work_done;
1485 }
1486
1487 static int enic_poll_msix(struct napi_struct *napi, int budget)
1488 {
1489         struct enic *enic = container_of(napi, struct enic, napi);
1490         struct net_device *netdev = enic->netdev;
1491         unsigned int work_to_do = budget;
1492         unsigned int work_done;
1493         int err;
1494
1495         /* Service RQ
1496          */
1497
1498         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1499                 work_to_do, enic_rq_service, NULL);
1500
1501         /* Return intr event credits for this polling
1502          * cycle.  An intr event is the completion of a
1503          * RQ packet.
1504          */
1505
1506         if (work_done > 0)
1507                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1508                         work_done,
1509                         0 /* don't unmask intr */,
1510                         0 /* don't reset intr timer */);
1511
1512         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1513
1514         /* Buffer allocation failed. Stay in polling mode
1515          * so we can try to fill the ring again.
1516          */
1517
1518         if (err)
1519                 work_done = work_to_do;
1520
1521         if (work_done < work_to_do) {
1522
1523                 /* Some work done, but not enough to stay in polling,
1524                  * flush all LROs and exit polling
1525                  */
1526
1527                 if (netdev->features & NETIF_F_LRO)
1528                         lro_flush_all(&enic->lro_mgr);
1529
1530                 napi_complete(napi);
1531                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1532         }
1533
1534         return work_done;
1535 }
1536
1537 static void enic_notify_timer(unsigned long data)
1538 {
1539         struct enic *enic = (struct enic *)data;
1540
1541         enic_notify_check(enic);
1542
1543         mod_timer(&enic->notify_timer,
1544                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1545 }
1546
1547 static void enic_free_intr(struct enic *enic)
1548 {
1549         struct net_device *netdev = enic->netdev;
1550         unsigned int i;
1551
1552         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1553         case VNIC_DEV_INTR_MODE_INTX:
1554                 free_irq(enic->pdev->irq, netdev);
1555                 break;
1556         case VNIC_DEV_INTR_MODE_MSI:
1557                 free_irq(enic->pdev->irq, enic);
1558                 break;
1559         case VNIC_DEV_INTR_MODE_MSIX:
1560                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1561                         if (enic->msix[i].requested)
1562                                 free_irq(enic->msix_entry[i].vector,
1563                                         enic->msix[i].devid);
1564                 break;
1565         default:
1566                 break;
1567         }
1568 }
1569
1570 static int enic_request_intr(struct enic *enic)
1571 {
1572         struct net_device *netdev = enic->netdev;
1573         unsigned int i;
1574         int err = 0;
1575
1576         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1577
1578         case VNIC_DEV_INTR_MODE_INTX:
1579
1580                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1581                         IRQF_SHARED, netdev->name, netdev);
1582                 break;
1583
1584         case VNIC_DEV_INTR_MODE_MSI:
1585
1586                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1587                         0, netdev->name, enic);
1588                 break;
1589
1590         case VNIC_DEV_INTR_MODE_MSIX:
1591
1592                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1593                         "%.11s-rx-0", netdev->name);
1594                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1595                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1596
1597                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1598                         "%.11s-tx-0", netdev->name);
1599                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1600                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1601
1602                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1603                         "%.11s-err", netdev->name);
1604                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1605                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1606
1607                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1608                         "%.11s-notify", netdev->name);
1609                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1610                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1611
1612                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1613                         err = request_irq(enic->msix_entry[i].vector,
1614                                 enic->msix[i].isr, 0,
1615                                 enic->msix[i].devname,
1616                                 enic->msix[i].devid);
1617                         if (err) {
1618                                 enic_free_intr(enic);
1619                                 break;
1620                         }
1621                         enic->msix[i].requested = 1;
1622                 }
1623
1624                 break;
1625
1626         default:
1627                 break;
1628         }
1629
1630         return err;
1631 }
1632
1633 static void enic_synchronize_irqs(struct enic *enic)
1634 {
1635         unsigned int i;
1636
1637         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1638         case VNIC_DEV_INTR_MODE_INTX:
1639         case VNIC_DEV_INTR_MODE_MSI:
1640                 synchronize_irq(enic->pdev->irq);
1641                 break;
1642         case VNIC_DEV_INTR_MODE_MSIX:
1643                 for (i = 0; i < enic->intr_count; i++)
1644                         synchronize_irq(enic->msix_entry[i].vector);
1645                 break;
1646         default:
1647                 break;
1648         }
1649 }
1650
1651 static int enic_notify_set(struct enic *enic)
1652 {
1653         int err;
1654
1655         spin_lock(&enic->devcmd_lock);
1656         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1657         case VNIC_DEV_INTR_MODE_INTX:
1658                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1659                 break;
1660         case VNIC_DEV_INTR_MODE_MSIX:
1661                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1662                 break;
1663         default:
1664                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1665                 break;
1666         }
1667         spin_unlock(&enic->devcmd_lock);
1668
1669         return err;
1670 }
1671
1672 static void enic_notify_timer_start(struct enic *enic)
1673 {
1674         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1675         case VNIC_DEV_INTR_MODE_MSI:
1676                 mod_timer(&enic->notify_timer, jiffies);
1677                 break;
1678         default:
1679                 /* Using intr for notification for INTx/MSI-X */
1680                 break;
1681         };
1682 }
1683
1684 /* rtnl lock is held, process context */
1685 static int enic_open(struct net_device *netdev)
1686 {
1687         struct enic *enic = netdev_priv(netdev);
1688         unsigned int i;
1689         int err;
1690
1691         err = enic_request_intr(enic);
1692         if (err) {
1693                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1694                         netdev->name);
1695                 return err;
1696         }
1697
1698         err = enic_notify_set(enic);
1699         if (err) {
1700                 printk(KERN_ERR PFX
1701                         "%s: Failed to alloc notify buffer, aborting.\n",
1702                         netdev->name);
1703                 goto err_out_free_intr;
1704         }
1705
1706         for (i = 0; i < enic->rq_count; i++) {
1707                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1708                 /* Need at least one buffer on ring to get going */
1709                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1710                         printk(KERN_ERR PFX
1711                                 "%s: Unable to alloc receive buffers.\n",
1712                                 netdev->name);
1713                         err = -ENOMEM;
1714                         goto err_out_notify_unset;
1715                 }
1716         }
1717
1718         for (i = 0; i < enic->wq_count; i++)
1719                 vnic_wq_enable(&enic->wq[i]);
1720         for (i = 0; i < enic->rq_count; i++)
1721                 vnic_rq_enable(&enic->rq[i]);
1722
1723         enic_dev_add_station_addr(enic);
1724         enic_set_multicast_list(netdev);
1725
1726         netif_wake_queue(netdev);
1727         napi_enable(&enic->napi);
1728         spin_lock(&enic->devcmd_lock);
1729         vnic_dev_enable(enic->vdev);
1730         spin_unlock(&enic->devcmd_lock);
1731
1732         for (i = 0; i < enic->intr_count; i++)
1733                 vnic_intr_unmask(&enic->intr[i]);
1734
1735         enic_notify_timer_start(enic);
1736
1737         return 0;
1738
1739 err_out_notify_unset:
1740         spin_lock(&enic->devcmd_lock);
1741         vnic_dev_notify_unset(enic->vdev);
1742         spin_unlock(&enic->devcmd_lock);
1743 err_out_free_intr:
1744         enic_free_intr(enic);
1745
1746         return err;
1747 }
1748
1749 /* rtnl lock is held, process context */
1750 static int enic_stop(struct net_device *netdev)
1751 {
1752         struct enic *enic = netdev_priv(netdev);
1753         unsigned int i;
1754         int err;
1755
1756         for (i = 0; i < enic->intr_count; i++)
1757                 vnic_intr_mask(&enic->intr[i]);
1758
1759         enic_synchronize_irqs(enic);
1760
1761         del_timer_sync(&enic->notify_timer);
1762
1763         spin_lock(&enic->devcmd_lock);
1764         vnic_dev_disable(enic->vdev);
1765         spin_unlock(&enic->devcmd_lock);
1766         napi_disable(&enic->napi);
1767         netif_carrier_off(netdev);
1768         netif_tx_disable(netdev);
1769
1770         enic_dev_del_station_addr(enic);
1771
1772         for (i = 0; i < enic->wq_count; i++) {
1773                 err = vnic_wq_disable(&enic->wq[i]);
1774                 if (err)
1775                         return err;
1776         }
1777         for (i = 0; i < enic->rq_count; i++) {
1778                 err = vnic_rq_disable(&enic->rq[i]);
1779                 if (err)
1780                         return err;
1781         }
1782
1783         spin_lock(&enic->devcmd_lock);
1784         vnic_dev_notify_unset(enic->vdev);
1785         spin_unlock(&enic->devcmd_lock);
1786         enic_free_intr(enic);
1787
1788         for (i = 0; i < enic->wq_count; i++)
1789                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1790         for (i = 0; i < enic->rq_count; i++)
1791                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1792         for (i = 0; i < enic->cq_count; i++)
1793                 vnic_cq_clean(&enic->cq[i]);
1794         for (i = 0; i < enic->intr_count; i++)
1795                 vnic_intr_clean(&enic->intr[i]);
1796
1797         return 0;
1798 }
1799
1800 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1801 {
1802         struct enic *enic = netdev_priv(netdev);
1803         int running = netif_running(netdev);
1804
1805         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1806                 return -EINVAL;
1807
1808         if (running)
1809                 enic_stop(netdev);
1810
1811         netdev->mtu = new_mtu;
1812
1813         if (netdev->mtu > enic->port_mtu)
1814                 printk(KERN_WARNING PFX
1815                         "%s: interface MTU (%d) set higher "
1816                         "than port MTU (%d)\n",
1817                         netdev->name, netdev->mtu, enic->port_mtu);
1818
1819         if (running)
1820                 enic_open(netdev);
1821
1822         return 0;
1823 }
1824
1825 #ifdef CONFIG_NET_POLL_CONTROLLER
1826 static void enic_poll_controller(struct net_device *netdev)
1827 {
1828         struct enic *enic = netdev_priv(netdev);
1829         struct vnic_dev *vdev = enic->vdev;
1830
1831         switch (vnic_dev_get_intr_mode(vdev)) {
1832         case VNIC_DEV_INTR_MODE_MSIX:
1833                 enic_isr_msix_rq(enic->pdev->irq, enic);
1834                 enic_isr_msix_wq(enic->pdev->irq, enic);
1835                 break;
1836         case VNIC_DEV_INTR_MODE_MSI:
1837                 enic_isr_msi(enic->pdev->irq, enic);
1838                 break;
1839         case VNIC_DEV_INTR_MODE_INTX:
1840                 enic_isr_legacy(enic->pdev->irq, netdev);
1841                 break;
1842         default:
1843                 break;
1844         }
1845 }
1846 #endif
1847
1848 static int enic_dev_wait(struct vnic_dev *vdev,
1849         int (*start)(struct vnic_dev *, int),
1850         int (*finished)(struct vnic_dev *, int *),
1851         int arg)
1852 {
1853         unsigned long time;
1854         int done;
1855         int err;
1856
1857         BUG_ON(in_interrupt());
1858
1859         err = start(vdev, arg);
1860         if (err)
1861                 return err;
1862
1863         /* Wait for func to complete...2 seconds max
1864          */
1865
1866         time = jiffies + (HZ * 2);
1867         do {
1868
1869                 err = finished(vdev, &done);
1870                 if (err)
1871                         return err;
1872
1873                 if (done)
1874                         return 0;
1875
1876                 schedule_timeout_uninterruptible(HZ / 10);
1877
1878         } while (time_after(time, jiffies));
1879
1880         return -ETIMEDOUT;
1881 }
1882
1883 static int enic_dev_open(struct enic *enic)
1884 {
1885         int err;
1886
1887         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1888                 vnic_dev_open_done, 0);
1889         if (err)
1890                 printk(KERN_ERR PFX
1891                         "vNIC device open failed, err %d.\n", err);
1892
1893         return err;
1894 }
1895
1896 static int enic_dev_soft_reset(struct enic *enic)
1897 {
1898         int err;
1899
1900         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1901                 vnic_dev_soft_reset_done, 0);
1902         if (err)
1903                 printk(KERN_ERR PFX
1904                         "vNIC soft reset failed, err %d.\n", err);
1905
1906         return err;
1907 }
1908
1909 static int enic_set_niccfg(struct enic *enic)
1910 {
1911         const u8 rss_default_cpu = 0;
1912         const u8 rss_hash_type = 0;
1913         const u8 rss_hash_bits = 0;
1914         const u8 rss_base_cpu = 0;
1915         const u8 rss_enable = 0;
1916         const u8 tso_ipid_split_en = 0;
1917         const u8 ig_vlan_strip_en = 1;
1918
1919         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1920          */
1921
1922         return enic_set_nic_cfg(enic,
1923                 rss_default_cpu, rss_hash_type,
1924                 rss_hash_bits, rss_base_cpu,
1925                 rss_enable, tso_ipid_split_en,
1926                 ig_vlan_strip_en);
1927 }
1928
1929 static void enic_reset(struct work_struct *work)
1930 {
1931         struct enic *enic = container_of(work, struct enic, reset);
1932
1933         if (!netif_running(enic->netdev))
1934                 return;
1935
1936         rtnl_lock();
1937
1938         spin_lock(&enic->devcmd_lock);
1939         vnic_dev_hang_notify(enic->vdev);
1940         spin_unlock(&enic->devcmd_lock);
1941
1942         enic_stop(enic->netdev);
1943         enic_dev_soft_reset(enic);
1944         vnic_dev_init(enic->vdev, 0);
1945         enic_reset_mcaddrs(enic);
1946         enic_init_vnic_resources(enic);
1947         enic_set_niccfg(enic);
1948         enic_open(enic->netdev);
1949
1950         rtnl_unlock();
1951 }
1952
1953 static int enic_set_intr_mode(struct enic *enic)
1954 {
1955         unsigned int n = 1;
1956         unsigned int m = 1;
1957         unsigned int i;
1958
1959         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1960          * system capabilities.
1961          *
1962          * Try MSI-X first
1963          *
1964          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1965          * (the second to last INTR is used for WQ/RQ errors)
1966          * (the last INTR is used for notifications)
1967          */
1968
1969         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1970         for (i = 0; i < n + m + 2; i++)
1971                 enic->msix_entry[i].entry = i;
1972
1973         if (enic->config.intr_mode < 1 &&
1974             enic->rq_count >= n &&
1975             enic->wq_count >= m &&
1976             enic->cq_count >= n + m &&
1977             enic->intr_count >= n + m + 2 &&
1978             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1979
1980                 enic->rq_count = n;
1981                 enic->wq_count = m;
1982                 enic->cq_count = n + m;
1983                 enic->intr_count = n + m + 2;
1984
1985                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1986
1987                 return 0;
1988         }
1989
1990         /* Next try MSI
1991          *
1992          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1993          */
1994
1995         if (enic->config.intr_mode < 2 &&
1996             enic->rq_count >= 1 &&
1997             enic->wq_count >= 1 &&
1998             enic->cq_count >= 2 &&
1999             enic->intr_count >= 1 &&
2000             !pci_enable_msi(enic->pdev)) {
2001
2002                 enic->rq_count = 1;
2003                 enic->wq_count = 1;
2004                 enic->cq_count = 2;
2005                 enic->intr_count = 1;
2006
2007                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2008
2009                 return 0;
2010         }
2011
2012         /* Next try INTx
2013          *
2014          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2015          * (the first INTR is used for WQ/RQ)
2016          * (the second INTR is used for WQ/RQ errors)
2017          * (the last INTR is used for notifications)
2018          */
2019
2020         if (enic->config.intr_mode < 3 &&
2021             enic->rq_count >= 1 &&
2022             enic->wq_count >= 1 &&
2023             enic->cq_count >= 2 &&
2024             enic->intr_count >= 3) {
2025
2026                 enic->rq_count = 1;
2027                 enic->wq_count = 1;
2028                 enic->cq_count = 2;
2029                 enic->intr_count = 3;
2030
2031                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2032
2033                 return 0;
2034         }
2035
2036         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2037
2038         return -EINVAL;
2039 }
2040
2041 static void enic_clear_intr_mode(struct enic *enic)
2042 {
2043         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2044         case VNIC_DEV_INTR_MODE_MSIX:
2045                 pci_disable_msix(enic->pdev);
2046                 break;
2047         case VNIC_DEV_INTR_MODE_MSI:
2048                 pci_disable_msi(enic->pdev);
2049                 break;
2050         default:
2051                 break;
2052         }
2053
2054         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2055 }
2056
2057 static const struct net_device_ops enic_netdev_dynamic_ops = {
2058         .ndo_open               = enic_open,
2059         .ndo_stop               = enic_stop,
2060         .ndo_start_xmit         = enic_hard_start_xmit,
2061         .ndo_get_stats          = enic_get_stats,
2062         .ndo_validate_addr      = eth_validate_addr,
2063         .ndo_set_multicast_list = enic_set_multicast_list,
2064         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2065         .ndo_change_mtu         = enic_change_mtu,
2066         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2067         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2068         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2069         .ndo_tx_timeout         = enic_tx_timeout,
2070         .ndo_set_vf_port        = enic_set_vf_port,
2071         .ndo_get_vf_port        = enic_get_vf_port,
2072 #ifdef CONFIG_NET_POLL_CONTROLLER
2073         .ndo_poll_controller    = enic_poll_controller,
2074 #endif
2075 };
2076
2077 static const struct net_device_ops enic_netdev_ops = {
2078         .ndo_open               = enic_open,
2079         .ndo_stop               = enic_stop,
2080         .ndo_start_xmit         = enic_hard_start_xmit,
2081         .ndo_get_stats          = enic_get_stats,
2082         .ndo_validate_addr      = eth_validate_addr,
2083         .ndo_set_multicast_list = enic_set_multicast_list,
2084         .ndo_set_mac_address    = enic_set_mac_address,
2085         .ndo_change_mtu         = enic_change_mtu,
2086         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2087         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2088         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2089         .ndo_tx_timeout         = enic_tx_timeout,
2090 #ifdef CONFIG_NET_POLL_CONTROLLER
2091         .ndo_poll_controller    = enic_poll_controller,
2092 #endif
2093 };
2094
2095 void enic_dev_deinit(struct enic *enic)
2096 {
2097         netif_napi_del(&enic->napi);
2098         enic_free_vnic_resources(enic);
2099         enic_clear_intr_mode(enic);
2100 }
2101
2102 int enic_dev_init(struct enic *enic)
2103 {
2104         struct net_device *netdev = enic->netdev;
2105         int err;
2106
2107         /* Get vNIC configuration
2108          */
2109
2110         err = enic_get_vnic_config(enic);
2111         if (err) {
2112                 printk(KERN_ERR PFX
2113                         "Get vNIC configuration failed, aborting.\n");
2114                 return err;
2115         }
2116
2117         /* Get available resource counts
2118          */
2119
2120         enic_get_res_counts(enic);
2121
2122         /* Set interrupt mode based on resource counts and system
2123          * capabilities
2124          */
2125
2126         err = enic_set_intr_mode(enic);
2127         if (err) {
2128                 printk(KERN_ERR PFX
2129                         "Failed to set intr mode based on resource "
2130                         "counts and system capabilities, aborting.\n");
2131                 return err;
2132         }
2133
2134         /* Allocate and configure vNIC resources
2135          */
2136
2137         err = enic_alloc_vnic_resources(enic);
2138         if (err) {
2139                 printk(KERN_ERR PFX
2140                         "Failed to alloc vNIC resources, aborting.\n");
2141                 goto err_out_free_vnic_resources;
2142         }
2143
2144         enic_init_vnic_resources(enic);
2145
2146         err = enic_set_rq_alloc_buf(enic);
2147         if (err) {
2148                 printk(KERN_ERR PFX
2149                         "Failed to set RQ buffer allocator, aborting.\n");
2150                 goto err_out_free_vnic_resources;
2151         }
2152
2153         err = enic_set_niccfg(enic);
2154         if (err) {
2155                 printk(KERN_ERR PFX
2156                         "Failed to config nic, aborting.\n");
2157                 goto err_out_free_vnic_resources;
2158         }
2159
2160         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2161         default:
2162                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
2163                 break;
2164         case VNIC_DEV_INTR_MODE_MSIX:
2165                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
2166                 break;
2167         }
2168
2169         return 0;
2170
2171 err_out_free_vnic_resources:
2172         enic_clear_intr_mode(enic);
2173         enic_free_vnic_resources(enic);
2174
2175         return err;
2176 }
2177
2178 static void enic_iounmap(struct enic *enic)
2179 {
2180         unsigned int i;
2181
2182         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2183                 if (enic->bar[i].vaddr)
2184                         iounmap(enic->bar[i].vaddr);
2185 }
2186
2187 static int __devinit enic_probe(struct pci_dev *pdev,
2188         const struct pci_device_id *ent)
2189 {
2190         struct net_device *netdev;
2191         struct enic *enic;
2192         int using_dac = 0;
2193         unsigned int i;
2194         int err;
2195
2196         /* Allocate net device structure and initialize.  Private
2197          * instance data is initialized to zero.
2198          */
2199
2200         netdev = alloc_etherdev(sizeof(struct enic));
2201         if (!netdev) {
2202                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
2203                 return -ENOMEM;
2204         }
2205
2206         pci_set_drvdata(pdev, netdev);
2207
2208         SET_NETDEV_DEV(netdev, &pdev->dev);
2209
2210         enic = netdev_priv(netdev);
2211         enic->netdev = netdev;
2212         enic->pdev = pdev;
2213
2214         /* Setup PCI resources
2215          */
2216
2217         err = pci_enable_device(pdev);
2218         if (err) {
2219                 printk(KERN_ERR PFX
2220                         "Cannot enable PCI device, aborting.\n");
2221                 goto err_out_free_netdev;
2222         }
2223
2224         err = pci_request_regions(pdev, DRV_NAME);
2225         if (err) {
2226                 printk(KERN_ERR PFX
2227                         "Cannot request PCI regions, aborting.\n");
2228                 goto err_out_disable_device;
2229         }
2230
2231         pci_set_master(pdev);
2232
2233         /* Query PCI controller on system for DMA addressing
2234          * limitation for the device.  Try 40-bit first, and
2235          * fail to 32-bit.
2236          */
2237
2238         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2239         if (err) {
2240                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2241                 if (err) {
2242                         printk(KERN_ERR PFX
2243                                 "No usable DMA configuration, aborting.\n");
2244                         goto err_out_release_regions;
2245                 }
2246                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2247                 if (err) {
2248                         printk(KERN_ERR PFX
2249                                 "Unable to obtain 32-bit DMA "
2250                                 "for consistent allocations, aborting.\n");
2251                         goto err_out_release_regions;
2252                 }
2253         } else {
2254                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2255                 if (err) {
2256                         printk(KERN_ERR PFX
2257                                 "Unable to obtain 40-bit DMA "
2258                                 "for consistent allocations, aborting.\n");
2259                         goto err_out_release_regions;
2260                 }
2261                 using_dac = 1;
2262         }
2263
2264         /* Map vNIC resources from BAR0-5
2265          */
2266
2267         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2268                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2269                         continue;
2270                 enic->bar[i].len = pci_resource_len(pdev, i);
2271                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2272                 if (!enic->bar[i].vaddr) {
2273                         printk(KERN_ERR PFX
2274                                 "Cannot memory-map BAR %d, aborting.\n", i);
2275                         err = -ENODEV;
2276                         goto err_out_iounmap;
2277                 }
2278                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2279         }
2280
2281         /* Register vNIC device
2282          */
2283
2284         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2285                 ARRAY_SIZE(enic->bar));
2286         if (!enic->vdev) {
2287                 printk(KERN_ERR PFX
2288                         "vNIC registration failed, aborting.\n");
2289                 err = -ENODEV;
2290                 goto err_out_iounmap;
2291         }
2292
2293         /* Issue device open to get device in known state
2294          */
2295
2296         err = enic_dev_open(enic);
2297         if (err) {
2298                 printk(KERN_ERR PFX
2299                         "vNIC dev open failed, aborting.\n");
2300                 goto err_out_vnic_unregister;
2301         }
2302
2303         /* Issue device init to initialize the vnic-to-switch link.
2304          * We'll start with carrier off and wait for link UP
2305          * notification later to turn on carrier.  We don't need
2306          * to wait here for the vnic-to-switch link initialization
2307          * to complete; link UP notification is the indication that
2308          * the process is complete.
2309          */
2310
2311         netif_carrier_off(netdev);
2312
2313         if (!enic_is_dynamic(enic)) {
2314                 err = vnic_dev_init(enic->vdev, 0);
2315                 if (err) {
2316                         printk(KERN_ERR PFX
2317                                 "vNIC dev init failed, aborting.\n");
2318                         goto err_out_dev_close;
2319                 }
2320         }
2321
2322         err = enic_dev_init(enic);
2323         if (err) {
2324                 printk(KERN_ERR PFX
2325                         "Device initialization failed, aborting.\n");
2326                 goto err_out_dev_close;
2327         }
2328
2329         /* Setup notification timer, HW reset task, and locks
2330          */
2331
2332         init_timer(&enic->notify_timer);
2333         enic->notify_timer.function = enic_notify_timer;
2334         enic->notify_timer.data = (unsigned long)enic;
2335
2336         INIT_WORK(&enic->reset, enic_reset);
2337
2338         for (i = 0; i < enic->wq_count; i++)
2339                 spin_lock_init(&enic->wq_lock[i]);
2340
2341         spin_lock_init(&enic->devcmd_lock);
2342
2343         /* Register net device
2344          */
2345
2346         enic->port_mtu = enic->config.mtu;
2347         (void)enic_change_mtu(netdev, enic->port_mtu);
2348
2349         err = enic_set_mac_addr(netdev, enic->mac_addr);
2350         if (err) {
2351                 printk(KERN_ERR PFX
2352                         "Invalid MAC address, aborting.\n");
2353                 goto err_out_dev_deinit;
2354         }
2355
2356         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2357         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2358
2359         if (enic_is_dynamic(enic))
2360                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2361         else
2362                 netdev->netdev_ops = &enic_netdev_ops;
2363
2364         netdev->watchdog_timeo = 2 * HZ;
2365         netdev->ethtool_ops = &enic_ethtool_ops;
2366
2367         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2368         if (ENIC_SETTING(enic, TXCSUM))
2369                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2370         if (ENIC_SETTING(enic, TSO))
2371                 netdev->features |= NETIF_F_TSO |
2372                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2373         if (ENIC_SETTING(enic, LRO))
2374                 netdev->features |= NETIF_F_LRO;
2375         if (using_dac)
2376                 netdev->features |= NETIF_F_HIGHDMA;
2377
2378         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2379
2380         enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
2381         enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
2382         enic->lro_mgr.lro_arr = enic->lro_desc;
2383         enic->lro_mgr.get_skb_header = enic_get_skb_header;
2384         enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
2385         enic->lro_mgr.dev = netdev;
2386         enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
2387         enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
2388
2389         err = register_netdev(netdev);
2390         if (err) {
2391                 printk(KERN_ERR PFX
2392                         "Cannot register net device, aborting.\n");
2393                 goto err_out_dev_deinit;
2394         }
2395
2396         return 0;
2397
2398 err_out_dev_deinit:
2399         enic_dev_deinit(enic);
2400 err_out_dev_close:
2401         vnic_dev_close(enic->vdev);
2402 err_out_vnic_unregister:
2403         vnic_dev_unregister(enic->vdev);
2404 err_out_iounmap:
2405         enic_iounmap(enic);
2406 err_out_release_regions:
2407         pci_release_regions(pdev);
2408 err_out_disable_device:
2409         pci_disable_device(pdev);
2410 err_out_free_netdev:
2411         pci_set_drvdata(pdev, NULL);
2412         free_netdev(netdev);
2413
2414         return err;
2415 }
2416
2417 static void __devexit enic_remove(struct pci_dev *pdev)
2418 {
2419         struct net_device *netdev = pci_get_drvdata(pdev);
2420
2421         if (netdev) {
2422                 struct enic *enic = netdev_priv(netdev);
2423
2424                 flush_scheduled_work();
2425                 unregister_netdev(netdev);
2426                 enic_dev_deinit(enic);
2427                 vnic_dev_close(enic->vdev);
2428                 vnic_dev_unregister(enic->vdev);
2429                 enic_iounmap(enic);
2430                 pci_release_regions(pdev);
2431                 pci_disable_device(pdev);
2432                 pci_set_drvdata(pdev, NULL);
2433                 free_netdev(netdev);
2434         }
2435 }
2436
2437 static struct pci_driver enic_driver = {
2438         .name = DRV_NAME,
2439         .id_table = enic_id_table,
2440         .probe = enic_probe,
2441         .remove = __devexit_p(enic_remove),
2442 };
2443
2444 static int __init enic_init_module(void)
2445 {
2446         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2447
2448         return pci_register_driver(&enic_driver);
2449 }
2450
2451 static void __exit enic_cleanup_module(void)
2452 {
2453         pci_unregister_driver(&enic_driver);
2454 }
2455
2456 module_init(enic_init_module);
2457 module_exit(enic_cleanup_module);