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_multicast_list(struct enic *enic)
 816 {
 817         enic->mc_count = 0;
 818         enic->flags = 0;
 819 }
 820
 821 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 822 {
 823         struct enic *enic = netdev_priv(netdev);
 824
 825         if (enic_is_dynamic(enic)) {
 826                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 827                         return -EADDRNOTAVAIL;
 828         } else {
 829                 if (!is_valid_ether_addr(addr))
 830                         return -EADDRNOTAVAIL;
 831         }
 832
 833         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 834
 835         return 0;
 836 }
 837
 838 static int enic_dev_add_station_addr(struct enic *enic)
 839 {
 840         int err = 0;
 841
 842         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 843                 spin_lock(&enic->devcmd_lock);
 844                 err = vnic_dev_add_addr(enic->vdev, enic->netdev->dev_addr);
 845                 spin_unlock(&enic->devcmd_lock);
 846         }
 847
 848         return err;
 849 }
 850
 851 static int enic_dev_del_station_addr(struct enic *enic)
 852 {
 853         int err = 0;
 854
 855         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 856                 spin_lock(&enic->devcmd_lock);
 857                 err = vnic_dev_del_addr(enic->vdev, enic->netdev->dev_addr);
 858                 spin_unlock(&enic->devcmd_lock);
 859         }
 860
 861         return err;
 862 }
 863
 864 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 865 {
 866         struct enic *enic = netdev_priv(netdev);
 867         struct sockaddr *saddr = p;
 868         char *addr = saddr->sa_data;
 869         int err;
 870
 871         if (netif_running(enic->netdev)) {
 872                 err = enic_dev_del_station_addr(enic);
 873                 if (err)
 874                         return err;
 875         }
 876
 877         err = enic_set_mac_addr(netdev, addr);
 878         if (err)
 879                 return err;
 880
 881         if (netif_running(enic->netdev)) {
 882                 err = enic_dev_add_station_addr(enic);
 883                 if (err)
 884                         return err;
 885         }
 886
 887         return err;
 888 }
 889
 890 static int enic_set_mac_address(struct net_device *netdev, void *p)
 891 {
 892         return -EOPNOTSUPP;
 893 }
 894
 895 /* netif_tx_lock held, BHs disabled */
 896 static void enic_set_multicast_list(struct net_device *netdev)
 897 {
 898         struct enic *enic = netdev_priv(netdev);
 899         struct netdev_hw_addr *ha;
 900         int directed = 1;
 901         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 902         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 903         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
 904         unsigned int mc_count = netdev_mc_count(netdev);
 905         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 906                 mc_count > ENIC_MULTICAST_PERFECT_FILTERS;
 907         unsigned int flags = netdev->flags | (allmulti ? IFF_ALLMULTI : 0);
 908         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
 909         unsigned int i, j;
 910
 911         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
 912                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
 913
 914         spin_lock(&enic->devcmd_lock);
 915
 916         if (enic->flags != flags) {
 917                 enic->flags = flags;
 918                 vnic_dev_packet_filter(enic->vdev, directed,
 919                         multicast, broadcast, promisc, allmulti);
 920         }
 921
 922         /* Is there an easier way?  Trying to minimize to
 923          * calls to add/del multicast addrs.  We keep the
 924          * addrs from the last call in enic->mc_addr and
 925          * look for changes to add/del.
 926          */
 927
 928         i = 0;
 929         netdev_for_each_mc_addr(ha, netdev) {
 930                 if (i == mc_count)
 931                         break;
 932                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
 933         }
 934
 935         for (i = 0; i < enic->mc_count; i++) {
 936                 for (j = 0; j < mc_count; j++)
 937                         if (compare_ether_addr(enic->mc_addr[i],
 938                                 mc_addr[j]) == 0)
 939                                 break;
 940                 if (j == mc_count)
 941                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
 942         }
 943
 944         for (i = 0; i < mc_count; i++) {
 945                 for (j = 0; j < enic->mc_count; j++)
 946                         if (compare_ether_addr(mc_addr[i],
 947                                 enic->mc_addr[j]) == 0)
 948                                 break;
 949                 if (j == enic->mc_count)
 950                         enic_add_multicast_addr(enic, mc_addr[i]);
 951         }
 952
 953         /* Save the list to compare against next time
 954          */
 955
 956         for (i = 0; i < mc_count; i++)
 957                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
 958
 959         enic->mc_count = mc_count;
 960
 961         spin_unlock(&enic->devcmd_lock);
 962 }
 963
 964 /* rtnl lock is held */
 965 static void enic_vlan_rx_register(struct net_device *netdev,
 966         struct vlan_group *vlan_group)
 967 {
 968         struct enic *enic = netdev_priv(netdev);
 969         enic->vlan_group = vlan_group;
 970 }
 971
 972 /* rtnl lock is held */
 973 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
 974 {
 975         struct enic *enic = netdev_priv(netdev);
 976
 977         spin_lock(&enic->devcmd_lock);
 978         enic_add_vlan(enic, vid);
 979         spin_unlock(&enic->devcmd_lock);
 980 }
 981
 982 /* rtnl lock is held */
 983 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
 984 {
 985         struct enic *enic = netdev_priv(netdev);
 986
 987         spin_lock(&enic->devcmd_lock);
 988         enic_del_vlan(enic, vid);
 989         spin_unlock(&enic->devcmd_lock);
 990 }
 991
 992 /* netif_tx_lock held, BHs disabled */
 993 static void enic_tx_timeout(struct net_device *netdev)
 994 {
 995         struct enic *enic = netdev_priv(netdev);
 996         schedule_work(&enic->reset);
 997 }
 998
 999 static int enic_vnic_dev_deinit(struct enic *enic)
1000 {
1001         int err;
1002
1003         spin_lock(&enic->devcmd_lock);
1004         err = vnic_dev_deinit(enic->vdev);
1005         spin_unlock(&enic->devcmd_lock);
1006
1007         return err;
1008 }
1009
1010 static int enic_dev_init_prov(struct enic *enic, struct vic_provinfo *vp)
1011 {
1012         int err;
1013
1014         spin_lock(&enic->devcmd_lock);
1015         err = vnic_dev_init_prov(enic->vdev,
1016                 (u8 *)vp, vic_provinfo_size(vp));
1017         spin_unlock(&enic->devcmd_lock);
1018
1019         return err;
1020 }
1021
1022 static int enic_dev_init_done(struct enic *enic, int *done, int *error)
1023 {
1024         int err;
1025
1026         spin_lock(&enic->devcmd_lock);
1027         err = vnic_dev_init_done(enic->vdev, done, error);
1028         spin_unlock(&enic->devcmd_lock);
1029
1030         return err;
1031 }
1032
1033 static int enic_set_port_profile(struct enic *enic, u8 *mac)
1034 {
1035         struct vic_provinfo *vp;
1036         u8 oui[3] = VIC_PROVINFO_CISCO_OUI;
1037         u8 *uuid;
1038         char uuid_str[38];
1039         static char *uuid_fmt = "%02X%02X%02X%02X-%02X%02X-%02X%02X-"
1040                 "%02X%02X-%02X%02X%02X%02X%0X%02X";
1041         int err;
1042
1043         err = enic_vnic_dev_deinit(enic);
1044         if (err)
1045                 return err;
1046
1047         switch (enic->pp.request) {
1048
1049         case PORT_REQUEST_ASSOCIATE:
1050
1051                 if (!(enic->pp.set & ENIC_SET_NAME) || !strlen(enic->pp.name))
1052                         return -EINVAL;
1053
1054                 if (!is_valid_ether_addr(mac))
1055                         return -EADDRNOTAVAIL;
1056
1057                 vp = vic_provinfo_alloc(GFP_KERNEL, oui,
1058                         VIC_PROVINFO_LINUX_TYPE);
1059                 if (!vp)
1060                         return -ENOMEM;
1061
1062                 vic_provinfo_add_tlv(vp,
1063                         VIC_LINUX_PROV_TLV_PORT_PROFILE_NAME_STR,
1064                         strlen(enic->pp.name) + 1, enic->pp.name);
1065
1066                 vic_provinfo_add_tlv(vp,
1067                         VIC_LINUX_PROV_TLV_CLIENT_MAC_ADDR,
1068                         ETH_ALEN, mac);
1069
1070                 if (enic->pp.set & ENIC_SET_INSTANCE) {
1071                         uuid = enic->pp.instance_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                                 uuid[8],  uuid[9],  uuid[10], uuid[11],
1076                                 uuid[12], uuid[13], uuid[14], uuid[15]);
1077                         vic_provinfo_add_tlv(vp,
1078                                 VIC_LINUX_PROV_TLV_CLIENT_UUID_STR,
1079                                 sizeof(uuid_str), uuid_str);
1080                 }
1081
1082                 if (enic->pp.set & ENIC_SET_HOST) {
1083                         uuid = enic->pp.host_uuid;
1084                         sprintf(uuid_str, uuid_fmt,
1085                                 uuid[0],  uuid[1],  uuid[2],  uuid[3],
1086                                 uuid[4],  uuid[5],  uuid[6],  uuid[7],
1087                                 uuid[8],  uuid[9],  uuid[10], uuid[11],
1088                                 uuid[12], uuid[13], uuid[14], uuid[15]);
1089                         vic_provinfo_add_tlv(vp,
1090                                 VIC_LINUX_PROV_TLV_HOST_UUID_STR,
1091                                 sizeof(uuid_str), uuid_str);
1092                 }
1093
1094                 err = enic_dev_init_prov(enic, vp);
1095                 vic_provinfo_free(vp);
1096                 if (err)
1097                         return err;
1098                 break;
1099
1100         case PORT_REQUEST_DISASSOCIATE:
1101                 break;
1102
1103         default:
1104                 return -EINVAL;
1105         }
1106
1107         enic->pp.set |= ENIC_SET_APPLIED;
1108         return 0;
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
1116         memset(&enic->pp, 0, sizeof(enic->pp));
1117
1118         if (port[IFLA_PORT_REQUEST]) {
1119                 enic->pp.set |= ENIC_SET_REQUEST;
1120                 enic->pp.request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1121         }
1122
1123         if (port[IFLA_PORT_PROFILE]) {
1124                 enic->pp.set |= ENIC_SET_NAME;
1125                 memcpy(enic->pp.name, nla_data(port[IFLA_PORT_PROFILE]),
1126                         PORT_PROFILE_MAX);
1127         }
1128
1129         if (port[IFLA_PORT_INSTANCE_UUID]) {
1130                 enic->pp.set |= ENIC_SET_INSTANCE;
1131                 memcpy(enic->pp.instance_uuid,
1132                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1133         }
1134
1135         if (port[IFLA_PORT_HOST_UUID]) {
1136                 enic->pp.set |= ENIC_SET_HOST;
1137                 memcpy(enic->pp.host_uuid,
1138                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1139         }
1140
1141         /* don't support VFs, yet */
1142         if (vf != PORT_SELF_VF)
1143                 return -EOPNOTSUPP;
1144
1145         if (!(enic->pp.set & ENIC_SET_REQUEST))
1146                 return -EOPNOTSUPP;
1147
1148         if (enic->pp.request == PORT_REQUEST_ASSOCIATE) {
1149
1150                 /* If the interface mac addr hasn't been assigned,
1151                  * assign a random mac addr before setting port-
1152                  * profile.
1153                  */
1154
1155                 if (is_zero_ether_addr(netdev->dev_addr))
1156                         random_ether_addr(netdev->dev_addr);
1157         }
1158
1159         return enic_set_port_profile(enic, netdev->dev_addr);
1160 }
1161
1162 static int enic_get_vf_port(struct net_device *netdev, int vf,
1163         struct sk_buff *skb)
1164 {
1165         struct enic *enic = netdev_priv(netdev);
1166         int err, error, done;
1167         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1168
1169         if (!(enic->pp.set & ENIC_SET_APPLIED))
1170                 return -ENODATA;
1171
1172         err = enic_dev_init_done(enic, &done, &error);
1173         if (err)
1174                 error = err;
1175
1176         switch (error) {
1177         case ERR_SUCCESS:
1178                 if (!done)
1179                         response = PORT_PROFILE_RESPONSE_INPROGRESS;
1180                 break;
1181         case ERR_EINVAL:
1182                 response = PORT_PROFILE_RESPONSE_INVALID;
1183                 break;
1184         case ERR_EBADSTATE:
1185                 response = PORT_PROFILE_RESPONSE_BADSTATE;
1186                 break;
1187         case ERR_ENOMEM:
1188                 response = PORT_PROFILE_RESPONSE_INSUFFICIENT_RESOURCES;
1189                 break;
1190         default:
1191                 response = PORT_PROFILE_RESPONSE_ERROR;
1192                 break;
1193         }
1194
1195         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1196         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1197         if (enic->pp.set & ENIC_SET_NAME)
1198                 NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1199                         enic->pp.name);
1200         if (enic->pp.set & ENIC_SET_INSTANCE)
1201                 NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1202                         enic->pp.instance_uuid);
1203         if (enic->pp.set & ENIC_SET_HOST)
1204                 NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1205                         enic->pp.host_uuid);
1206
1207         return 0;
1208
1209 nla_put_failure:
1210         return -EMSGSIZE;
1211 }
1212
1213 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1214 {
1215         struct enic *enic = vnic_dev_priv(rq->vdev);
1216
1217         if (!buf->os_buf)
1218                 return;
1219
1220         pci_unmap_single(enic->pdev, buf->dma_addr,
1221                 buf->len, PCI_DMA_FROMDEVICE);
1222         dev_kfree_skb_any(buf->os_buf);
1223 }
1224
1225 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1226 {
1227         struct enic *enic = vnic_dev_priv(rq->vdev);
1228         struct net_device *netdev = enic->netdev;
1229         struct sk_buff *skb;
1230         unsigned int len = netdev->mtu + ETH_HLEN;
1231         unsigned int os_buf_index = 0;
1232         dma_addr_t dma_addr;
1233
1234         skb = netdev_alloc_skb_ip_align(netdev, len);
1235         if (!skb)
1236                 return -ENOMEM;
1237
1238         dma_addr = pci_map_single(enic->pdev, skb->data,
1239                 len, PCI_DMA_FROMDEVICE);
1240
1241         enic_queue_rq_desc(rq, skb, os_buf_index,
1242                 dma_addr, len);
1243
1244         return 0;
1245 }
1246
1247 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
1248 {
1249         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
1250
1251         if (vnic_rq_posting_soon(rq)) {
1252
1253                 /* SW workaround for A0 HW erratum: if we're just about
1254                  * to write posted_index, insert a dummy desc
1255                  * of type resvd
1256                  */
1257
1258                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
1259                 vnic_rq_post(rq, 0, 0, 0, 0);
1260         } else {
1261                 return enic_rq_alloc_buf(rq);
1262         }
1263
1264         return 0;
1265 }
1266
1267 static int enic_set_rq_alloc_buf(struct enic *enic)
1268 {
1269         enum vnic_dev_hw_version hw_ver;
1270         int err;
1271
1272         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
1273         if (err)
1274                 return err;
1275
1276         switch (hw_ver) {
1277         case VNIC_DEV_HW_VER_A1:
1278                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
1279                 break;
1280         case VNIC_DEV_HW_VER_A2:
1281         case VNIC_DEV_HW_VER_UNKNOWN:
1282                 enic->rq_alloc_buf = enic_rq_alloc_buf;
1283                 break;
1284         default:
1285                 return -ENODEV;
1286         }
1287
1288         return 0;
1289 }
1290
1291 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1292         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1293         int skipped, void *opaque)
1294 {
1295         struct enic *enic = vnic_dev_priv(rq->vdev);
1296         struct net_device *netdev = enic->netdev;
1297         struct sk_buff *skb;
1298
1299         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1300         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1301         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1302         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1303         u8 packet_error;
1304         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1305         u32 rss_hash;
1306
1307         if (skipped)
1308                 return;
1309
1310         skb = buf->os_buf;
1311         prefetch(skb->data - NET_IP_ALIGN);
1312         pci_unmap_single(enic->pdev, buf->dma_addr,
1313                 buf->len, PCI_DMA_FROMDEVICE);
1314
1315         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1316                 &type, &color, &q_number, &completed_index,
1317                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1318                 &csum_not_calc, &rss_hash, &bytes_written,
1319                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1320                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1321                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1322                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1323                 &fcs_ok);
1324
1325         if (packet_error) {
1326
1327                 if (!fcs_ok) {
1328                         if (bytes_written > 0)
1329                                 enic->rq_bad_fcs++;
1330                         else if (bytes_written == 0)
1331                                 enic->rq_truncated_pkts++;
1332                 }
1333
1334                 dev_kfree_skb_any(skb);
1335
1336                 return;
1337         }
1338
1339         if (eop && bytes_written > 0) {
1340
1341                 /* Good receive
1342                  */
1343
1344                 skb_put(skb, bytes_written);
1345                 skb->protocol = eth_type_trans(skb, netdev);
1346
1347                 if (enic->csum_rx_enabled && !csum_not_calc) {
1348                         skb->csum = htons(checksum);
1349                         skb->ip_summed = CHECKSUM_COMPLETE;
1350                 }
1351
1352                 skb->dev = netdev;
1353
1354                 if (enic->vlan_group && vlan_stripped &&
1355                         (vlan_tci & CQ_ENET_RQ_DESC_VLAN_TCI_VLAN_MASK)) {
1356
1357                         if (netdev->features & NETIF_F_GRO)
1358                                 vlan_gro_receive(&enic->napi, enic->vlan_group,
1359                                         vlan_tci, skb);
1360                         else
1361                                 vlan_hwaccel_receive_skb(skb,
1362                                         enic->vlan_group, vlan_tci);
1363
1364                 } else {
1365
1366                         if (netdev->features & NETIF_F_GRO)
1367                                 napi_gro_receive(&enic->napi, skb);
1368                         else
1369                                 netif_receive_skb(skb);
1370
1371                 }
1372
1373         } else {
1374
1375                 /* Buffer overflow
1376                  */
1377
1378                 dev_kfree_skb_any(skb);
1379         }
1380 }
1381
1382 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1383         u8 type, u16 q_number, u16 completed_index, void *opaque)
1384 {
1385         struct enic *enic = vnic_dev_priv(vdev);
1386
1387         vnic_rq_service(&enic->rq[q_number], cq_desc,
1388                 completed_index, VNIC_RQ_RETURN_DESC,
1389                 enic_rq_indicate_buf, opaque);
1390
1391         return 0;
1392 }
1393
1394 static int enic_poll(struct napi_struct *napi, int budget)
1395 {
1396         struct enic *enic = container_of(napi, struct enic, napi);
1397         unsigned int rq_work_to_do = budget;
1398         unsigned int wq_work_to_do = -1; /* no limit */
1399         unsigned int  work_done, rq_work_done, wq_work_done;
1400         int err;
1401
1402         /* Service RQ (first) and WQ
1403          */
1404
1405         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1406                 rq_work_to_do, enic_rq_service, NULL);
1407
1408         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1409                 wq_work_to_do, enic_wq_service, NULL);
1410
1411         /* Accumulate intr event credits for this polling
1412          * cycle.  An intr event is the completion of a
1413          * a WQ or RQ packet.
1414          */
1415
1416         work_done = rq_work_done + wq_work_done;
1417
1418         if (work_done > 0)
1419                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1420                         work_done,
1421                         0 /* don't unmask intr */,
1422                         0 /* don't reset intr timer */);
1423
1424         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1425
1426         /* Buffer allocation failed. Stay in polling
1427          * mode so we can try to fill the ring again.
1428          */
1429
1430         if (err)
1431                 rq_work_done = rq_work_to_do;
1432
1433         if (rq_work_done < rq_work_to_do) {
1434
1435                 /* Some work done, but not enough to stay in polling,
1436                  * exit polling
1437                  */
1438
1439                 napi_complete(napi);
1440                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1441         }
1442
1443         return rq_work_done;
1444 }
1445
1446 static int enic_poll_msix(struct napi_struct *napi, int budget)
1447 {
1448         struct enic *enic = container_of(napi, struct enic, napi);
1449         unsigned int work_to_do = budget;
1450         unsigned int work_done;
1451         int err;
1452
1453         /* Service RQ
1454          */
1455
1456         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1457                 work_to_do, enic_rq_service, NULL);
1458
1459         /* Return intr event credits for this polling
1460          * cycle.  An intr event is the completion of a
1461          * RQ packet.
1462          */
1463
1464         if (work_done > 0)
1465                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1466                         work_done,
1467                         0 /* don't unmask intr */,
1468                         0 /* don't reset intr timer */);
1469
1470         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1471
1472         /* Buffer allocation failed. Stay in polling mode
1473          * so we can try to fill the ring again.
1474          */
1475
1476         if (err)
1477                 work_done = work_to_do;
1478
1479         if (work_done < work_to_do) {
1480
1481                 /* Some work done, but not enough to stay in polling,
1482                  * exit polling
1483                  */
1484
1485                 napi_complete(napi);
1486                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1487         }
1488
1489         return work_done;
1490 }
1491
1492 static void enic_notify_timer(unsigned long data)
1493 {
1494         struct enic *enic = (struct enic *)data;
1495
1496         enic_notify_check(enic);
1497
1498         mod_timer(&enic->notify_timer,
1499                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1500 }
1501
1502 static void enic_free_intr(struct enic *enic)
1503 {
1504         struct net_device *netdev = enic->netdev;
1505         unsigned int i;
1506
1507         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1508         case VNIC_DEV_INTR_MODE_INTX:
1509                 free_irq(enic->pdev->irq, netdev);
1510                 break;
1511         case VNIC_DEV_INTR_MODE_MSI:
1512                 free_irq(enic->pdev->irq, enic);
1513                 break;
1514         case VNIC_DEV_INTR_MODE_MSIX:
1515                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1516                         if (enic->msix[i].requested)
1517                                 free_irq(enic->msix_entry[i].vector,
1518                                         enic->msix[i].devid);
1519                 break;
1520         default:
1521                 break;
1522         }
1523 }
1524
1525 static int enic_request_intr(struct enic *enic)
1526 {
1527         struct net_device *netdev = enic->netdev;
1528         unsigned int i;
1529         int err = 0;
1530
1531         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1532
1533         case VNIC_DEV_INTR_MODE_INTX:
1534
1535                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1536                         IRQF_SHARED, netdev->name, netdev);
1537                 break;
1538
1539         case VNIC_DEV_INTR_MODE_MSI:
1540
1541                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1542                         0, netdev->name, enic);
1543                 break;
1544
1545         case VNIC_DEV_INTR_MODE_MSIX:
1546
1547                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1548                         "%.11s-rx-0", netdev->name);
1549                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1550                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1551
1552                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1553                         "%.11s-tx-0", netdev->name);
1554                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1555                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1556
1557                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1558                         "%.11s-err", netdev->name);
1559                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1560                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1561
1562                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1563                         "%.11s-notify", netdev->name);
1564                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1565                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1566
1567                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1568                         err = request_irq(enic->msix_entry[i].vector,
1569                                 enic->msix[i].isr, 0,
1570                                 enic->msix[i].devname,
1571                                 enic->msix[i].devid);
1572                         if (err) {
1573                                 enic_free_intr(enic);
1574                                 break;
1575                         }
1576                         enic->msix[i].requested = 1;
1577                 }
1578
1579                 break;
1580
1581         default:
1582                 break;
1583         }
1584
1585         return err;
1586 }
1587
1588 static void enic_synchronize_irqs(struct enic *enic)
1589 {
1590         unsigned int i;
1591
1592         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1593         case VNIC_DEV_INTR_MODE_INTX:
1594         case VNIC_DEV_INTR_MODE_MSI:
1595                 synchronize_irq(enic->pdev->irq);
1596                 break;
1597         case VNIC_DEV_INTR_MODE_MSIX:
1598                 for (i = 0; i < enic->intr_count; i++)
1599                         synchronize_irq(enic->msix_entry[i].vector);
1600                 break;
1601         default:
1602                 break;
1603         }
1604 }
1605
1606 static int enic_notify_set(struct enic *enic)
1607 {
1608         int err;
1609
1610         spin_lock(&enic->devcmd_lock);
1611         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1612         case VNIC_DEV_INTR_MODE_INTX:
1613                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1614                 break;
1615         case VNIC_DEV_INTR_MODE_MSIX:
1616                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1617                 break;
1618         default:
1619                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1620                 break;
1621         }
1622         spin_unlock(&enic->devcmd_lock);
1623
1624         return err;
1625 }
1626
1627 static void enic_notify_timer_start(struct enic *enic)
1628 {
1629         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1630         case VNIC_DEV_INTR_MODE_MSI:
1631                 mod_timer(&enic->notify_timer, jiffies);
1632                 break;
1633         default:
1634                 /* Using intr for notification for INTx/MSI-X */
1635                 break;
1636         };
1637 }
1638
1639 /* rtnl lock is held, process context */
1640 static int enic_open(struct net_device *netdev)
1641 {
1642         struct enic *enic = netdev_priv(netdev);
1643         unsigned int i;
1644         int err;
1645
1646         err = enic_request_intr(enic);
1647         if (err) {
1648                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1649                         netdev->name);
1650                 return err;
1651         }
1652
1653         err = enic_notify_set(enic);
1654         if (err) {
1655                 printk(KERN_ERR PFX
1656                         "%s: Failed to alloc notify buffer, aborting.\n",
1657                         netdev->name);
1658                 goto err_out_free_intr;
1659         }
1660
1661         for (i = 0; i < enic->rq_count; i++) {
1662                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1663                 /* Need at least one buffer on ring to get going */
1664                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1665                         printk(KERN_ERR PFX
1666                                 "%s: Unable to alloc receive buffers.\n",
1667                                 netdev->name);
1668                         err = -ENOMEM;
1669                         goto err_out_notify_unset;
1670                 }
1671         }
1672
1673         for (i = 0; i < enic->wq_count; i++)
1674                 vnic_wq_enable(&enic->wq[i]);
1675         for (i = 0; i < enic->rq_count; i++)
1676                 vnic_rq_enable(&enic->rq[i]);
1677
1678         enic_dev_add_station_addr(enic);
1679         enic_set_multicast_list(netdev);
1680
1681         netif_wake_queue(netdev);
1682         napi_enable(&enic->napi);
1683         spin_lock(&enic->devcmd_lock);
1684         vnic_dev_enable(enic->vdev);
1685         spin_unlock(&enic->devcmd_lock);
1686
1687         for (i = 0; i < enic->intr_count; i++)
1688                 vnic_intr_unmask(&enic->intr[i]);
1689
1690         enic_notify_timer_start(enic);
1691
1692         return 0;
1693
1694 err_out_notify_unset:
1695         spin_lock(&enic->devcmd_lock);
1696         vnic_dev_notify_unset(enic->vdev);
1697         spin_unlock(&enic->devcmd_lock);
1698 err_out_free_intr:
1699         enic_free_intr(enic);
1700
1701         return err;
1702 }
1703
1704 /* rtnl lock is held, process context */
1705 static int enic_stop(struct net_device *netdev)
1706 {
1707         struct enic *enic = netdev_priv(netdev);
1708         unsigned int i;
1709         int err;
1710
1711         for (i = 0; i < enic->intr_count; i++)
1712                 vnic_intr_mask(&enic->intr[i]);
1713
1714         enic_synchronize_irqs(enic);
1715
1716         del_timer_sync(&enic->notify_timer);
1717
1718         spin_lock(&enic->devcmd_lock);
1719         vnic_dev_disable(enic->vdev);
1720         spin_unlock(&enic->devcmd_lock);
1721         napi_disable(&enic->napi);
1722         netif_carrier_off(netdev);
1723         netif_tx_disable(netdev);
1724
1725         enic_dev_del_station_addr(enic);
1726
1727         for (i = 0; i < enic->wq_count; i++) {
1728                 err = vnic_wq_disable(&enic->wq[i]);
1729                 if (err)
1730                         return err;
1731         }
1732         for (i = 0; i < enic->rq_count; i++) {
1733                 err = vnic_rq_disable(&enic->rq[i]);
1734                 if (err)
1735                         return err;
1736         }
1737
1738         spin_lock(&enic->devcmd_lock);
1739         vnic_dev_notify_unset(enic->vdev);
1740         spin_unlock(&enic->devcmd_lock);
1741         enic_free_intr(enic);
1742
1743         for (i = 0; i < enic->wq_count; i++)
1744                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1745         for (i = 0; i < enic->rq_count; i++)
1746                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1747         for (i = 0; i < enic->cq_count; i++)
1748                 vnic_cq_clean(&enic->cq[i]);
1749         for (i = 0; i < enic->intr_count; i++)
1750                 vnic_intr_clean(&enic->intr[i]);
1751
1752         return 0;
1753 }
1754
1755 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1756 {
1757         struct enic *enic = netdev_priv(netdev);
1758         int running = netif_running(netdev);
1759
1760         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1761                 return -EINVAL;
1762
1763         if (running)
1764                 enic_stop(netdev);
1765
1766         netdev->mtu = new_mtu;
1767
1768         if (netdev->mtu > enic->port_mtu)
1769                 printk(KERN_WARNING PFX
1770                         "%s: interface MTU (%d) set higher "
1771                         "than port MTU (%d)\n",
1772                         netdev->name, netdev->mtu, enic->port_mtu);
1773
1774         if (running)
1775                 enic_open(netdev);
1776
1777         return 0;
1778 }
1779
1780 #ifdef CONFIG_NET_POLL_CONTROLLER
1781 static void enic_poll_controller(struct net_device *netdev)
1782 {
1783         struct enic *enic = netdev_priv(netdev);
1784         struct vnic_dev *vdev = enic->vdev;
1785
1786         switch (vnic_dev_get_intr_mode(vdev)) {
1787         case VNIC_DEV_INTR_MODE_MSIX:
1788                 enic_isr_msix_rq(enic->pdev->irq, enic);
1789                 enic_isr_msix_wq(enic->pdev->irq, enic);
1790                 break;
1791         case VNIC_DEV_INTR_MODE_MSI:
1792                 enic_isr_msi(enic->pdev->irq, enic);
1793                 break;
1794         case VNIC_DEV_INTR_MODE_INTX:
1795                 enic_isr_legacy(enic->pdev->irq, netdev);
1796                 break;
1797         default:
1798                 break;
1799         }
1800 }
1801 #endif
1802
1803 static int enic_dev_wait(struct vnic_dev *vdev,
1804         int (*start)(struct vnic_dev *, int),
1805         int (*finished)(struct vnic_dev *, int *),
1806         int arg)
1807 {
1808         unsigned long time;
1809         int done;
1810         int err;
1811
1812         BUG_ON(in_interrupt());
1813
1814         err = start(vdev, arg);
1815         if (err)
1816                 return err;
1817
1818         /* Wait for func to complete...2 seconds max
1819          */
1820
1821         time = jiffies + (HZ * 2);
1822         do {
1823
1824                 err = finished(vdev, &done);
1825                 if (err)
1826                         return err;
1827
1828                 if (done)
1829                         return 0;
1830
1831                 schedule_timeout_uninterruptible(HZ / 10);
1832
1833         } while (time_after(time, jiffies));
1834
1835         return -ETIMEDOUT;
1836 }
1837
1838 static int enic_dev_open(struct enic *enic)
1839 {
1840         int err;
1841
1842         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1843                 vnic_dev_open_done, 0);
1844         if (err)
1845                 printk(KERN_ERR PFX
1846                         "vNIC device open failed, err %d.\n", err);
1847
1848         return err;
1849 }
1850
1851 static int enic_dev_hang_reset(struct enic *enic)
1852 {
1853         int err;
1854
1855         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
1856                 vnic_dev_hang_reset_done, 0);
1857         if (err)
1858                 printk(KERN_ERR PFX
1859                         "vNIC hang reset failed, err %d.\n", err);
1860
1861         return err;
1862 }
1863
1864 static int enic_set_niccfg(struct enic *enic)
1865 {
1866         const u8 rss_default_cpu = 0;
1867         const u8 rss_hash_type = 0;
1868         const u8 rss_hash_bits = 0;
1869         const u8 rss_base_cpu = 0;
1870         const u8 rss_enable = 0;
1871         const u8 tso_ipid_split_en = 0;
1872         const u8 ig_vlan_strip_en = 1;
1873
1874         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1875          */
1876
1877         return enic_set_nic_cfg(enic,
1878                 rss_default_cpu, rss_hash_type,
1879                 rss_hash_bits, rss_base_cpu,
1880                 rss_enable, tso_ipid_split_en,
1881                 ig_vlan_strip_en);
1882 }
1883
1884 int enic_dev_set_ig_vlan_rewrite_mode(struct enic *enic)
1885 {
1886         int err;
1887
1888         spin_lock(&enic->devcmd_lock);
1889         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
1890                 IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN);
1891         spin_unlock(&enic->devcmd_lock);
1892
1893         return err;
1894 }
1895
1896 static void enic_reset(struct work_struct *work)
1897 {
1898         struct enic *enic = container_of(work, struct enic, reset);
1899
1900         if (!netif_running(enic->netdev))
1901                 return;
1902
1903         rtnl_lock();
1904
1905         spin_lock(&enic->devcmd_lock);
1906         vnic_dev_hang_notify(enic->vdev);
1907         spin_unlock(&enic->devcmd_lock);
1908
1909         enic_stop(enic->netdev);
1910         enic_dev_hang_reset(enic);
1911         enic_reset_multicast_list(enic);
1912         enic_init_vnic_resources(enic);
1913         enic_set_niccfg(enic);
1914         enic_dev_set_ig_vlan_rewrite_mode(enic);
1915         enic_open(enic->netdev);
1916
1917         rtnl_unlock();
1918 }
1919
1920 static int enic_set_intr_mode(struct enic *enic)
1921 {
1922         unsigned int n = 1;
1923         unsigned int m = 1;
1924         unsigned int i;
1925
1926         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1927          * system capabilities.
1928          *
1929          * Try MSI-X first
1930          *
1931          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1932          * (the second to last INTR is used for WQ/RQ errors)
1933          * (the last INTR is used for notifications)
1934          */
1935
1936         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1937         for (i = 0; i < n + m + 2; i++)
1938                 enic->msix_entry[i].entry = i;
1939
1940         if (enic->config.intr_mode < 1 &&
1941             enic->rq_count >= n &&
1942             enic->wq_count >= m &&
1943             enic->cq_count >= n + m &&
1944             enic->intr_count >= n + m + 2 &&
1945             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1946
1947                 enic->rq_count = n;
1948                 enic->wq_count = m;
1949                 enic->cq_count = n + m;
1950                 enic->intr_count = n + m + 2;
1951
1952                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1953
1954                 return 0;
1955         }
1956
1957         /* Next try MSI
1958          *
1959          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
1960          */
1961
1962         if (enic->config.intr_mode < 2 &&
1963             enic->rq_count >= 1 &&
1964             enic->wq_count >= 1 &&
1965             enic->cq_count >= 2 &&
1966             enic->intr_count >= 1 &&
1967             !pci_enable_msi(enic->pdev)) {
1968
1969                 enic->rq_count = 1;
1970                 enic->wq_count = 1;
1971                 enic->cq_count = 2;
1972                 enic->intr_count = 1;
1973
1974                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
1975
1976                 return 0;
1977         }
1978
1979         /* Next try INTx
1980          *
1981          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
1982          * (the first INTR is used for WQ/RQ)
1983          * (the second INTR is used for WQ/RQ errors)
1984          * (the last INTR is used for notifications)
1985          */
1986
1987         if (enic->config.intr_mode < 3 &&
1988             enic->rq_count >= 1 &&
1989             enic->wq_count >= 1 &&
1990             enic->cq_count >= 2 &&
1991             enic->intr_count >= 3) {
1992
1993                 enic->rq_count = 1;
1994                 enic->wq_count = 1;
1995                 enic->cq_count = 2;
1996                 enic->intr_count = 3;
1997
1998                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
1999
2000                 return 0;
2001         }
2002
2003         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2004
2005         return -EINVAL;
2006 }
2007
2008 static void enic_clear_intr_mode(struct enic *enic)
2009 {
2010         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2011         case VNIC_DEV_INTR_MODE_MSIX:
2012                 pci_disable_msix(enic->pdev);
2013                 break;
2014         case VNIC_DEV_INTR_MODE_MSI:
2015                 pci_disable_msi(enic->pdev);
2016                 break;
2017         default:
2018                 break;
2019         }
2020
2021         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2022 }
2023
2024 static const struct net_device_ops enic_netdev_dynamic_ops = {
2025         .ndo_open               = enic_open,
2026         .ndo_stop               = enic_stop,
2027         .ndo_start_xmit         = enic_hard_start_xmit,
2028         .ndo_get_stats          = enic_get_stats,
2029         .ndo_validate_addr      = eth_validate_addr,
2030         .ndo_set_multicast_list = enic_set_multicast_list,
2031         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2032         .ndo_change_mtu         = enic_change_mtu,
2033         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2034         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2035         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2036         .ndo_tx_timeout         = enic_tx_timeout,
2037         .ndo_set_vf_port        = enic_set_vf_port,
2038         .ndo_get_vf_port        = enic_get_vf_port,
2039 #ifdef CONFIG_NET_POLL_CONTROLLER
2040         .ndo_poll_controller    = enic_poll_controller,
2041 #endif
2042 };
2043
2044 static const struct net_device_ops enic_netdev_ops = {
2045         .ndo_open               = enic_open,
2046         .ndo_stop               = enic_stop,
2047         .ndo_start_xmit         = enic_hard_start_xmit,
2048         .ndo_get_stats          = enic_get_stats,
2049         .ndo_validate_addr      = eth_validate_addr,
2050         .ndo_set_multicast_list = enic_set_multicast_list,
2051         .ndo_set_mac_address    = enic_set_mac_address,
2052         .ndo_change_mtu         = enic_change_mtu,
2053         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2054         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2055         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2056         .ndo_tx_timeout         = enic_tx_timeout,
2057 #ifdef CONFIG_NET_POLL_CONTROLLER
2058         .ndo_poll_controller    = enic_poll_controller,
2059 #endif
2060 };
2061
2062 void enic_dev_deinit(struct enic *enic)
2063 {
2064         netif_napi_del(&enic->napi);
2065         enic_free_vnic_resources(enic);
2066         enic_clear_intr_mode(enic);
2067 }
2068
2069 int enic_dev_init(struct enic *enic)
2070 {
2071         struct net_device *netdev = enic->netdev;
2072         int err;
2073
2074         /* Get vNIC configuration
2075          */
2076
2077         err = enic_get_vnic_config(enic);
2078         if (err) {
2079                 printk(KERN_ERR PFX
2080                         "Get vNIC configuration failed, aborting.\n");
2081                 return err;
2082         }
2083
2084         /* Get available resource counts
2085          */
2086
2087         enic_get_res_counts(enic);
2088
2089         /* Set interrupt mode based on resource counts and system
2090          * capabilities
2091          */
2092
2093         err = enic_set_intr_mode(enic);
2094         if (err) {
2095                 printk(KERN_ERR PFX
2096                         "Failed to set intr mode based on resource "
2097                         "counts and system capabilities, aborting.\n");
2098                 return err;
2099         }
2100
2101         /* Allocate and configure vNIC resources
2102          */
2103
2104         err = enic_alloc_vnic_resources(enic);
2105         if (err) {
2106                 printk(KERN_ERR PFX
2107                         "Failed to alloc vNIC resources, aborting.\n");
2108                 goto err_out_free_vnic_resources;
2109         }
2110
2111         enic_init_vnic_resources(enic);
2112
2113         err = enic_set_rq_alloc_buf(enic);
2114         if (err) {
2115                 printk(KERN_ERR PFX
2116                         "Failed to set RQ buffer allocator, aborting.\n");
2117                 goto err_out_free_vnic_resources;
2118         }
2119
2120         err = enic_set_niccfg(enic);
2121         if (err) {
2122                 printk(KERN_ERR PFX
2123                         "Failed to config nic, aborting.\n");
2124                 goto err_out_free_vnic_resources;
2125         }
2126
2127         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2128         if (err) {
2129                 printk(KERN_ERR PFX
2130                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2131                 goto err_out_free_vnic_resources;
2132         }
2133
2134         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2135         default:
2136                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
2137                 break;
2138         case VNIC_DEV_INTR_MODE_MSIX:
2139                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
2140                 break;
2141         }
2142
2143         return 0;
2144
2145 err_out_free_vnic_resources:
2146         enic_clear_intr_mode(enic);
2147         enic_free_vnic_resources(enic);
2148
2149         return err;
2150 }
2151
2152 static void enic_iounmap(struct enic *enic)
2153 {
2154         unsigned int i;
2155
2156         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2157                 if (enic->bar[i].vaddr)
2158                         iounmap(enic->bar[i].vaddr);
2159 }
2160
2161 static int __devinit enic_probe(struct pci_dev *pdev,
2162         const struct pci_device_id *ent)
2163 {
2164         struct net_device *netdev;
2165         struct enic *enic;
2166         int using_dac = 0;
2167         unsigned int i;
2168         int err;
2169
2170         /* Allocate net device structure and initialize.  Private
2171          * instance data is initialized to zero.
2172          */
2173
2174         netdev = alloc_etherdev(sizeof(struct enic));
2175         if (!netdev) {
2176                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
2177                 return -ENOMEM;
2178         }
2179
2180         pci_set_drvdata(pdev, netdev);
2181
2182         SET_NETDEV_DEV(netdev, &pdev->dev);
2183
2184         enic = netdev_priv(netdev);
2185         enic->netdev = netdev;
2186         enic->pdev = pdev;
2187
2188         /* Setup PCI resources
2189          */
2190
2191         err = pci_enable_device(pdev);
2192         if (err) {
2193                 printk(KERN_ERR PFX
2194                         "Cannot enable PCI device, aborting.\n");
2195                 goto err_out_free_netdev;
2196         }
2197
2198         err = pci_request_regions(pdev, DRV_NAME);
2199         if (err) {
2200                 printk(KERN_ERR PFX
2201                         "Cannot request PCI regions, aborting.\n");
2202                 goto err_out_disable_device;
2203         }
2204
2205         pci_set_master(pdev);
2206
2207         /* Query PCI controller on system for DMA addressing
2208          * limitation for the device.  Try 40-bit first, and
2209          * fail to 32-bit.
2210          */
2211
2212         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2213         if (err) {
2214                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2215                 if (err) {
2216                         printk(KERN_ERR PFX
2217                                 "No usable DMA configuration, aborting.\n");
2218                         goto err_out_release_regions;
2219                 }
2220                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2221                 if (err) {
2222                         printk(KERN_ERR PFX
2223                                 "Unable to obtain 32-bit DMA "
2224                                 "for consistent allocations, aborting.\n");
2225                         goto err_out_release_regions;
2226                 }
2227         } else {
2228                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2229                 if (err) {
2230                         printk(KERN_ERR PFX
2231                                 "Unable to obtain 40-bit DMA "
2232                                 "for consistent allocations, aborting.\n");
2233                         goto err_out_release_regions;
2234                 }
2235                 using_dac = 1;
2236         }
2237
2238         /* Map vNIC resources from BAR0-5
2239          */
2240
2241         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2242                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2243                         continue;
2244                 enic->bar[i].len = pci_resource_len(pdev, i);
2245                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2246                 if (!enic->bar[i].vaddr) {
2247                         printk(KERN_ERR PFX
2248                                 "Cannot memory-map BAR %d, aborting.\n", i);
2249                         err = -ENODEV;
2250                         goto err_out_iounmap;
2251                 }
2252                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2253         }
2254
2255         /* Register vNIC device
2256          */
2257
2258         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2259                 ARRAY_SIZE(enic->bar));
2260         if (!enic->vdev) {
2261                 printk(KERN_ERR PFX
2262                         "vNIC registration failed, aborting.\n");
2263                 err = -ENODEV;
2264                 goto err_out_iounmap;
2265         }
2266
2267         /* Issue device open to get device in known state
2268          */
2269
2270         err = enic_dev_open(enic);
2271         if (err) {
2272                 printk(KERN_ERR PFX
2273                         "vNIC dev open failed, aborting.\n");
2274                 goto err_out_vnic_unregister;
2275         }
2276
2277         /* Issue device init to initialize the vnic-to-switch link.
2278          * We'll start with carrier off and wait for link UP
2279          * notification later to turn on carrier.  We don't need
2280          * to wait here for the vnic-to-switch link initialization
2281          * to complete; link UP notification is the indication that
2282          * the process is complete.
2283          */
2284
2285         netif_carrier_off(netdev);
2286
2287         if (!enic_is_dynamic(enic)) {
2288                 err = vnic_dev_init(enic->vdev, 0);
2289                 if (err) {
2290                         printk(KERN_ERR PFX
2291                                 "vNIC dev init failed, aborting.\n");
2292                         goto err_out_dev_close;
2293                 }
2294         }
2295
2296         err = enic_dev_init(enic);
2297         if (err) {
2298                 printk(KERN_ERR PFX
2299                         "Device initialization failed, aborting.\n");
2300                 goto err_out_dev_close;
2301         }
2302
2303         /* Setup notification timer, HW reset task, and locks
2304          */
2305
2306         init_timer(&enic->notify_timer);
2307         enic->notify_timer.function = enic_notify_timer;
2308         enic->notify_timer.data = (unsigned long)enic;
2309
2310         INIT_WORK(&enic->reset, enic_reset);
2311
2312         for (i = 0; i < enic->wq_count; i++)
2313                 spin_lock_init(&enic->wq_lock[i]);
2314
2315         spin_lock_init(&enic->devcmd_lock);
2316
2317         /* Register net device
2318          */
2319
2320         enic->port_mtu = enic->config.mtu;
2321         (void)enic_change_mtu(netdev, enic->port_mtu);
2322
2323         err = enic_set_mac_addr(netdev, enic->mac_addr);
2324         if (err) {
2325                 printk(KERN_ERR PFX
2326                         "Invalid MAC address, aborting.\n");
2327                 goto err_out_dev_deinit;
2328         }
2329
2330         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2331         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2332
2333         if (enic_is_dynamic(enic))
2334                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2335         else
2336                 netdev->netdev_ops = &enic_netdev_ops;
2337
2338         netdev->watchdog_timeo = 2 * HZ;
2339         netdev->ethtool_ops = &enic_ethtool_ops;
2340
2341         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2342         if (ENIC_SETTING(enic, TXCSUM))
2343                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2344         if (ENIC_SETTING(enic, TSO))
2345                 netdev->features |= NETIF_F_TSO |
2346                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2347         if (ENIC_SETTING(enic, LRO))
2348                 netdev->features |= NETIF_F_GRO;
2349         if (using_dac)
2350                 netdev->features |= NETIF_F_HIGHDMA;
2351
2352         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2353
2354         err = register_netdev(netdev);
2355         if (err) {
2356                 printk(KERN_ERR PFX
2357                         "Cannot register net device, aborting.\n");
2358                 goto err_out_dev_deinit;
2359         }
2360
2361         return 0;
2362
2363 err_out_dev_deinit:
2364         enic_dev_deinit(enic);
2365 err_out_dev_close:
2366         vnic_dev_close(enic->vdev);
2367 err_out_vnic_unregister:
2368         vnic_dev_unregister(enic->vdev);
2369 err_out_iounmap:
2370         enic_iounmap(enic);
2371 err_out_release_regions:
2372         pci_release_regions(pdev);
2373 err_out_disable_device:
2374         pci_disable_device(pdev);
2375 err_out_free_netdev:
2376         pci_set_drvdata(pdev, NULL);
2377         free_netdev(netdev);
2378
2379         return err;
2380 }
2381
2382 static void __devexit enic_remove(struct pci_dev *pdev)
2383 {
2384         struct net_device *netdev = pci_get_drvdata(pdev);
2385
2386         if (netdev) {
2387                 struct enic *enic = netdev_priv(netdev);
2388
2389                 flush_scheduled_work();
2390                 unregister_netdev(netdev);
2391                 enic_dev_deinit(enic);
2392                 vnic_dev_close(enic->vdev);
2393                 vnic_dev_unregister(enic->vdev);
2394                 enic_iounmap(enic);
2395                 pci_release_regions(pdev);
2396                 pci_disable_device(pdev);
2397                 pci_set_drvdata(pdev, NULL);
2398                 free_netdev(netdev);
2399         }
2400 }
2401
2402 static struct pci_driver enic_driver = {
2403         .name = DRV_NAME,
2404         .id_table = enic_id_table,
2405         .probe = enic_probe,
2406         .remove = __devexit_p(enic_remove),
2407 };
2408
2409 static int __init enic_init_module(void)
2410 {
2411         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2412
2413         return pci_register_driver(&enic_driver);
2414 }
2415
2416 static void __exit enic_cleanup_module(void)
2417 {
2418         pci_unregister_driver(&enic_driver);
2419 }
2420
2421 module_init(enic_init_module);
2422 module_exit(enic_cleanup_module);