drivers/net/enic/enic_main.c

   1 /*
   2  * Copyright 2008-2010 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  *
   5  * This program is free software; you may redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; version 2 of the License.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  16  * SOFTWARE.
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/kernel.h>
  22 #include <linux/string.h>
  23 #include <linux/errno.h>
  24 #include <linux/types.h>
  25 #include <linux/init.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/pci.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/etherdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/ethtool.h>
  33 #include <linux/in.h>
  34 #include <linux/ip.h>
  35 #include <linux/ipv6.h>
  36 #include <linux/tcp.h>
  37 #include <linux/rtnetlink.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 inline unsigned int enic_cq_rq(struct enic *enic, unsigned int rq)
 126 {
 127         return rq;
 128 }
 129
 130 static inline unsigned int enic_cq_wq(struct enic *enic, unsigned int wq)
 131 {
 132         return enic->rq_count + wq;
 133 }
 134
 135 static inline unsigned int enic_legacy_io_intr(void)
 136 {
 137         return 0;
 138 }
 139
 140 static inline unsigned int enic_legacy_err_intr(void)
 141 {
 142         return 1;
 143 }
 144
 145 static inline unsigned int enic_legacy_notify_intr(void)
 146 {
 147         return 2;
 148 }
 149
 150 static inline unsigned int enic_msix_rq_intr(struct enic *enic, unsigned int rq)
 151 {
 152         return rq;
 153 }
 154
 155 static inline unsigned int enic_msix_wq_intr(struct enic *enic, unsigned int wq)
 156 {
 157         return enic->rq_count + wq;
 158 }
 159
 160 static inline unsigned int enic_msix_err_intr(struct enic *enic)
 161 {
 162         return enic->rq_count + enic->wq_count;
 163 }
 164
 165 static inline unsigned int enic_msix_notify_intr(struct enic *enic)
 166 {
 167         return enic->rq_count + enic->wq_count + 1;
 168 }
 169
 170 static int enic_get_settings(struct net_device *netdev,
 171         struct ethtool_cmd *ecmd)
 172 {
 173         struct enic *enic = netdev_priv(netdev);
 174
 175         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
 176         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
 177         ecmd->port = PORT_FIBRE;
 178         ecmd->transceiver = XCVR_EXTERNAL;
 179
 180         if (netif_carrier_ok(netdev)) {
 181                 ecmd->speed = vnic_dev_port_speed(enic->vdev);
 182                 ecmd->duplex = DUPLEX_FULL;
 183         } else {
 184                 ecmd->speed = -1;
 185                 ecmd->duplex = -1;
 186         }
 187
 188         ecmd->autoneg = AUTONEG_DISABLE;
 189
 190         return 0;
 191 }
 192
 193 static int enic_dev_fw_info(struct enic *enic,
 194         struct vnic_devcmd_fw_info **fw_info)
 195 {
 196         int err;
 197
 198         spin_lock(&enic->devcmd_lock);
 199         err = vnic_dev_fw_info(enic->vdev, fw_info);
 200         spin_unlock(&enic->devcmd_lock);
 201
 202         return err;
 203 }
 204
 205 static void enic_get_drvinfo(struct net_device *netdev,
 206         struct ethtool_drvinfo *drvinfo)
 207 {
 208         struct enic *enic = netdev_priv(netdev);
 209         struct vnic_devcmd_fw_info *fw_info;
 210
 211         enic_dev_fw_info(enic, &fw_info);
 212
 213         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
 214         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
 215         strncpy(drvinfo->fw_version, fw_info->fw_version,
 216                 sizeof(drvinfo->fw_version));
 217         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
 218                 sizeof(drvinfo->bus_info));
 219 }
 220
 221 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
 222 {
 223         unsigned int i;
 224
 225         switch (stringset) {
 226         case ETH_SS_STATS:
 227                 for (i = 0; i < enic_n_tx_stats; i++) {
 228                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
 229                         data += ETH_GSTRING_LEN;
 230                 }
 231                 for (i = 0; i < enic_n_rx_stats; i++) {
 232                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
 233                         data += ETH_GSTRING_LEN;
 234                 }
 235                 break;
 236         }
 237 }
 238
 239 static int enic_get_sset_count(struct net_device *netdev, int sset)
 240 {
 241         switch (sset) {
 242         case ETH_SS_STATS:
 243                 return enic_n_tx_stats + enic_n_rx_stats;
 244         default:
 245                 return -EOPNOTSUPP;
 246         }
 247 }
 248
 249 static int enic_dev_stats_dump(struct enic *enic, struct vnic_stats **vstats)
 250 {
 251         int err;
 252
 253         spin_lock(&enic->devcmd_lock);
 254         err = vnic_dev_stats_dump(enic->vdev, vstats);
 255         spin_unlock(&enic->devcmd_lock);
 256
 257         return err;
 258 }
 259
 260 static void enic_get_ethtool_stats(struct net_device *netdev,
 261         struct ethtool_stats *stats, u64 *data)
 262 {
 263         struct enic *enic = netdev_priv(netdev);
 264         struct vnic_stats *vstats;
 265         unsigned int i;
 266
 267         enic_dev_stats_dump(enic, &vstats);
 268
 269         for (i = 0; i < enic_n_tx_stats; i++)
 270                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
 271         for (i = 0; i < enic_n_rx_stats; i++)
 272                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
 273 }
 274
 275 static u32 enic_get_rx_csum(struct net_device *netdev)
 276 {
 277         struct enic *enic = netdev_priv(netdev);
 278         return enic->csum_rx_enabled;
 279 }
 280
 281 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
 282 {
 283         struct enic *enic = netdev_priv(netdev);
 284
 285         if (data && !ENIC_SETTING(enic, RXCSUM))
 286                 return -EINVAL;
 287
 288         enic->csum_rx_enabled = !!data;
 289
 290         return 0;
 291 }
 292
 293 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
 294 {
 295         struct enic *enic = netdev_priv(netdev);
 296
 297         if (data && !ENIC_SETTING(enic, TXCSUM))
 298                 return -EINVAL;
 299
 300         if (data)
 301                 netdev->features |= NETIF_F_HW_CSUM;
 302         else
 303                 netdev->features &= ~NETIF_F_HW_CSUM;
 304
 305         return 0;
 306 }
 307
 308 static int enic_set_tso(struct net_device *netdev, u32 data)
 309 {
 310         struct enic *enic = netdev_priv(netdev);
 311
 312         if (data && !ENIC_SETTING(enic, TSO))
 313                 return -EINVAL;
 314
 315         if (data)
 316                 netdev->features |=
 317                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
 318         else
 319                 netdev->features &=
 320                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
 321
 322         return 0;
 323 }
 324
 325 static u32 enic_get_msglevel(struct net_device *netdev)
 326 {
 327         struct enic *enic = netdev_priv(netdev);
 328         return enic->msg_enable;
 329 }
 330
 331 static void enic_set_msglevel(struct net_device *netdev, u32 value)
 332 {
 333         struct enic *enic = netdev_priv(netdev);
 334         enic->msg_enable = value;
 335 }
 336
 337 static int enic_get_coalesce(struct net_device *netdev,
 338         struct ethtool_coalesce *ecmd)
 339 {
 340         struct enic *enic = netdev_priv(netdev);
 341
 342         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
 343         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
 344
 345         return 0;
 346 }
 347
 348 static int enic_set_coalesce(struct net_device *netdev,
 349         struct ethtool_coalesce *ecmd)
 350 {
 351         struct enic *enic = netdev_priv(netdev);
 352         u32 tx_coalesce_usecs;
 353         u32 rx_coalesce_usecs;
 354         unsigned int i, intr;
 355
 356         tx_coalesce_usecs = min_t(u32,
 357                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 358                 ecmd->tx_coalesce_usecs);
 359         rx_coalesce_usecs = min_t(u32,
 360                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 361                 ecmd->rx_coalesce_usecs);
 362
 363         switch (vnic_dev_get_intr_mode(enic->vdev)) {
 364         case VNIC_DEV_INTR_MODE_INTX:
 365                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 366                         return -EINVAL;
 367
 368                 intr = enic_legacy_io_intr();
 369                 vnic_intr_coalescing_timer_set(&enic->intr[intr],
 370                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 371                 break;
 372         case VNIC_DEV_INTR_MODE_MSI:
 373                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 374                         return -EINVAL;
 375
 376                 vnic_intr_coalescing_timer_set(&enic->intr[0],
 377                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 378                 break;
 379         case VNIC_DEV_INTR_MODE_MSIX:
 380                 for (i = 0; i < enic->wq_count; i++) {
 381                         intr = enic_msix_wq_intr(enic, i);
 382                         vnic_intr_coalescing_timer_set(&enic->intr[intr],
 383                                 INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 384                 }
 385
 386                 for (i = 0; i < enic->rq_count; i++) {
 387                         intr = enic_msix_rq_intr(enic, i);
 388                         vnic_intr_coalescing_timer_set(&enic->intr[intr],
 389                                 INTR_COALESCE_USEC_TO_HW(rx_coalesce_usecs));
 390                 }
 391
 392                 break;
 393         default:
 394                 break;
 395         }
 396
 397         enic->tx_coalesce_usecs = tx_coalesce_usecs;
 398         enic->rx_coalesce_usecs = rx_coalesce_usecs;
 399
 400         return 0;
 401 }
 402
 403 static const struct ethtool_ops enic_ethtool_ops = {
 404         .get_settings = enic_get_settings,
 405         .get_drvinfo = enic_get_drvinfo,
 406         .get_msglevel = enic_get_msglevel,
 407         .set_msglevel = enic_set_msglevel,
 408         .get_link = ethtool_op_get_link,
 409         .get_strings = enic_get_strings,
 410         .get_sset_count = enic_get_sset_count,
 411         .get_ethtool_stats = enic_get_ethtool_stats,
 412         .get_rx_csum = enic_get_rx_csum,
 413         .set_rx_csum = enic_set_rx_csum,
 414         .get_tx_csum = ethtool_op_get_tx_csum,
 415         .set_tx_csum = enic_set_tx_csum,
 416         .get_sg = ethtool_op_get_sg,
 417         .set_sg = ethtool_op_set_sg,
 418         .get_tso = ethtool_op_get_tso,
 419         .set_tso = enic_set_tso,
 420         .get_coalesce = enic_get_coalesce,
 421         .set_coalesce = enic_set_coalesce,
 422         .get_flags = ethtool_op_get_flags,
 423 };
 424
 425 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 426 {
 427         struct enic *enic = vnic_dev_priv(wq->vdev);
 428
 429         if (buf->sop)
 430                 pci_unmap_single(enic->pdev, buf->dma_addr,
 431                         buf->len, PCI_DMA_TODEVICE);
 432         else
 433                 pci_unmap_page(enic->pdev, buf->dma_addr,
 434                         buf->len, PCI_DMA_TODEVICE);
 435
 436         if (buf->os_buf)
 437                 dev_kfree_skb_any(buf->os_buf);
 438 }
 439
 440 static void enic_wq_free_buf(struct vnic_wq *wq,
 441         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 442 {
 443         enic_free_wq_buf(wq, buf);
 444 }
 445
 446 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 447         u8 type, u16 q_number, u16 completed_index, void *opaque)
 448 {
 449         struct enic *enic = vnic_dev_priv(vdev);
 450
 451         spin_lock(&enic->wq_lock[q_number]);
 452
 453         vnic_wq_service(&enic->wq[q_number], cq_desc,
 454                 completed_index, enic_wq_free_buf,
 455                 opaque);
 456
 457         if (netif_queue_stopped(enic->netdev) &&
 458             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 459             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 460                 netif_wake_queue(enic->netdev);
 461
 462         spin_unlock(&enic->wq_lock[q_number]);
 463
 464         return 0;
 465 }
 466
 467 static void enic_log_q_error(struct enic *enic)
 468 {
 469         unsigned int i;
 470         u32 error_status;
 471
 472         for (i = 0; i < enic->wq_count; i++) {
 473                 error_status = vnic_wq_error_status(&enic->wq[i]);
 474                 if (error_status)
 475                         netdev_err(enic->netdev, "WQ[%d] error_status %d\n",
 476                                 i, error_status);
 477         }
 478
 479         for (i = 0; i < enic->rq_count; i++) {
 480                 error_status = vnic_rq_error_status(&enic->rq[i]);
 481                 if (error_status)
 482                         netdev_err(enic->netdev, "RQ[%d] error_status %d\n",
 483                                 i, error_status);
 484         }
 485 }
 486
 487 static void enic_msglvl_check(struct enic *enic)
 488 {
 489         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 490
 491         if (msg_enable != enic->msg_enable) {
 492                 netdev_info(enic->netdev, "msg lvl changed from 0x%x to 0x%x\n",
 493                         enic->msg_enable, msg_enable);
 494                 enic->msg_enable = msg_enable;
 495         }
 496 }
 497
 498 static void enic_mtu_check(struct enic *enic)
 499 {
 500         u32 mtu = vnic_dev_mtu(enic->vdev);
 501         struct net_device *netdev = enic->netdev;
 502
 503         if (mtu && mtu != enic->port_mtu) {
 504                 enic->port_mtu = mtu;
 505                 if (mtu < netdev->mtu)
 506                         netdev_warn(netdev,
 507                                 "interface MTU (%d) set higher "
 508                                 "than switch port MTU (%d)\n",
 509                                 netdev->mtu, mtu);
 510         }
 511 }
 512
 513 static void enic_link_check(struct enic *enic)
 514 {
 515         int link_status = vnic_dev_link_status(enic->vdev);
 516         int carrier_ok = netif_carrier_ok(enic->netdev);
 517
 518         if (link_status && !carrier_ok) {
 519                 netdev_info(enic->netdev, "Link UP\n");
 520                 netif_carrier_on(enic->netdev);
 521         } else if (!link_status && carrier_ok) {
 522                 netdev_info(enic->netdev, "Link DOWN\n");
 523                 netif_carrier_off(enic->netdev);
 524         }
 525 }
 526
 527 static void enic_notify_check(struct enic *enic)
 528 {
 529         enic_msglvl_check(enic);
 530         enic_mtu_check(enic);
 531         enic_link_check(enic);
 532 }
 533
 534 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 535
 536 static irqreturn_t enic_isr_legacy(int irq, void *data)
 537 {
 538         struct net_device *netdev = data;
 539         struct enic *enic = netdev_priv(netdev);
 540         unsigned int io_intr = enic_legacy_io_intr();
 541         unsigned int err_intr = enic_legacy_err_intr();
 542         unsigned int notify_intr = enic_legacy_notify_intr();
 543         u32 pba;
 544
 545         vnic_intr_mask(&enic->intr[io_intr]);
 546
 547         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 548         if (!pba) {
 549                 vnic_intr_unmask(&enic->intr[io_intr]);
 550                 return IRQ_NONE;        /* not our interrupt */
 551         }
 552
 553         if (ENIC_TEST_INTR(pba, notify_intr)) {
 554                 vnic_intr_return_all_credits(&enic->intr[notify_intr]);
 555                 enic_notify_check(enic);
 556         }
 557
 558         if (ENIC_TEST_INTR(pba, err_intr)) {
 559                 vnic_intr_return_all_credits(&enic->intr[err_intr]);
 560                 enic_log_q_error(enic);
 561                 /* schedule recovery from WQ/RQ error */
 562                 schedule_work(&enic->reset);
 563                 return IRQ_HANDLED;
 564         }
 565
 566         if (ENIC_TEST_INTR(pba, io_intr)) {
 567                 if (napi_schedule_prep(&enic->napi[0]))
 568                         __napi_schedule(&enic->napi[0]);
 569         } else {
 570                 vnic_intr_unmask(&enic->intr[io_intr]);
 571         }
 572
 573         return IRQ_HANDLED;
 574 }
 575
 576 static irqreturn_t enic_isr_msi(int irq, void *data)
 577 {
 578         struct enic *enic = data;
 579
 580         /* With MSI, there is no sharing of interrupts, so this is
 581          * our interrupt and there is no need to ack it.  The device
 582          * is not providing per-vector masking, so the OS will not
 583          * write to PCI config space to mask/unmask the interrupt.
 584          * We're using mask_on_assertion for MSI, so the device
 585          * automatically masks the interrupt when the interrupt is
 586          * generated.  Later, when exiting polling, the interrupt
 587          * will be unmasked (see enic_poll).
 588          *
 589          * Also, the device uses the same PCIe Traffic Class (TC)
 590          * for Memory Write data and MSI, so there are no ordering
 591          * issues; the MSI will always arrive at the Root Complex
 592          * _after_ corresponding Memory Writes (i.e. descriptor
 593          * writes).
 594          */
 595
 596         napi_schedule(&enic->napi[0]);
 597
 598         return IRQ_HANDLED;
 599 }
 600
 601 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
 602 {
 603         struct napi_struct *napi = data;
 604
 605         /* schedule NAPI polling for RQ cleanup */
 606         napi_schedule(napi);
 607
 608         return IRQ_HANDLED;
 609 }
 610
 611 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
 612 {
 613         struct enic *enic = data;
 614         unsigned int cq = enic_cq_wq(enic, 0);
 615         unsigned int intr = enic_msix_wq_intr(enic, 0);
 616         unsigned int wq_work_to_do = -1; /* no limit */
 617         unsigned int wq_work_done;
 618
 619         wq_work_done = vnic_cq_service(&enic->cq[cq],
 620                 wq_work_to_do, enic_wq_service, NULL);
 621
 622         vnic_intr_return_credits(&enic->intr[intr],
 623                 wq_work_done,
 624                 1 /* unmask intr */,
 625                 1 /* reset intr timer */);
 626
 627         return IRQ_HANDLED;
 628 }
 629
 630 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 631 {
 632         struct enic *enic = data;
 633         unsigned int intr = enic_msix_err_intr(enic);
 634
 635         vnic_intr_return_all_credits(&enic->intr[intr]);
 636
 637         enic_log_q_error(enic);
 638
 639         /* schedule recovery from WQ/RQ error */
 640         schedule_work(&enic->reset);
 641
 642         return IRQ_HANDLED;
 643 }
 644
 645 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 646 {
 647         struct enic *enic = data;
 648         unsigned int intr = enic_msix_notify_intr(enic);
 649
 650         vnic_intr_return_all_credits(&enic->intr[intr]);
 651         enic_notify_check(enic);
 652
 653         return IRQ_HANDLED;
 654 }
 655
 656 static inline void enic_queue_wq_skb_cont(struct enic *enic,
 657         struct vnic_wq *wq, struct sk_buff *skb,
 658         unsigned int len_left, int loopback)
 659 {
 660         skb_frag_t *frag;
 661
 662         /* Queue additional data fragments */
 663         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 664                 len_left -= frag->size;
 665                 enic_queue_wq_desc_cont(wq, skb,
 666                         pci_map_page(enic->pdev, frag->page,
 667                                 frag->page_offset, frag->size,
 668                                 PCI_DMA_TODEVICE),
 669                         frag->size,
 670                         (len_left == 0),        /* EOP? */
 671                         loopback);
 672         }
 673 }
 674
 675 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
 676         struct vnic_wq *wq, struct sk_buff *skb,
 677         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
 678 {
 679         unsigned int head_len = skb_headlen(skb);
 680         unsigned int len_left = skb->len - head_len;
 681         int eop = (len_left == 0);
 682
 683         /* Queue the main skb fragment. The fragments are no larger
 684          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 685          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 686          * per fragment is queued.
 687          */
 688         enic_queue_wq_desc(wq, skb,
 689                 pci_map_single(enic->pdev, skb->data,
 690                         head_len, PCI_DMA_TODEVICE),
 691                 head_len,
 692                 vlan_tag_insert, vlan_tag,
 693                 eop, loopback);
 694
 695         if (!eop)
 696                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 697 }
 698
 699 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
 700         struct vnic_wq *wq, struct sk_buff *skb,
 701         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
 702 {
 703         unsigned int head_len = skb_headlen(skb);
 704         unsigned int len_left = skb->len - head_len;
 705         unsigned int hdr_len = skb_checksum_start_offset(skb);
 706         unsigned int csum_offset = hdr_len + skb->csum_offset;
 707         int eop = (len_left == 0);
 708
 709         /* Queue the main skb fragment. The fragments are no larger
 710          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 711          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 712          * per fragment is queued.
 713          */
 714         enic_queue_wq_desc_csum_l4(wq, skb,
 715                 pci_map_single(enic->pdev, skb->data,
 716                         head_len, PCI_DMA_TODEVICE),
 717                 head_len,
 718                 csum_offset,
 719                 hdr_len,
 720                 vlan_tag_insert, vlan_tag,
 721                 eop, loopback);
 722
 723         if (!eop)
 724                 enic_queue_wq_skb_cont(enic, wq, skb, len_left, loopback);
 725 }
 726
 727 static inline void enic_queue_wq_skb_tso(struct enic *enic,
 728         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
 729         int vlan_tag_insert, unsigned int vlan_tag, int loopback)
 730 {
 731         unsigned int frag_len_left = skb_headlen(skb);
 732         unsigned int len_left = skb->len - frag_len_left;
 733         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 734         int eop = (len_left == 0);
 735         unsigned int len;
 736         dma_addr_t dma_addr;
 737         unsigned int offset = 0;
 738         skb_frag_t *frag;
 739
 740         /* Preload TCP csum field with IP pseudo hdr calculated
 741          * with IP length set to zero.  HW will later add in length
 742          * to each TCP segment resulting from the TSO.
 743          */
 744
 745         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 746                 ip_hdr(skb)->check = 0;
 747                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 748                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 749         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 750                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 751                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 752         }
 753
 754         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 755          * for the main skb fragment
 756          */
 757         while (frag_len_left) {
 758                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 759                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
 760                                 len, PCI_DMA_TODEVICE);
 761                 enic_queue_wq_desc_tso(wq, skb,
 762                         dma_addr,
 763                         len,
 764                         mss, hdr_len,
 765                         vlan_tag_insert, vlan_tag,
 766                         eop && (len == frag_len_left), loopback);
 767                 frag_len_left -= len;
 768                 offset += len;
 769         }
 770
 771         if (eop)
 772                 return;
 773
 774         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 775          * for additional data fragments
 776          */
 777         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 778                 len_left -= frag->size;
 779                 frag_len_left = frag->size;
 780                 offset = frag->page_offset;
 781
 782                 while (frag_len_left) {
 783                         len = min(frag_len_left,
 784                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 785                         dma_addr = pci_map_page(enic->pdev, frag->page,
 786                                 offset, len,
 787                                 PCI_DMA_TODEVICE);
 788                         enic_queue_wq_desc_cont(wq, skb,
 789                                 dma_addr,
 790                                 len,
 791                                 (len_left == 0) &&
 792                                 (len == frag_len_left),         /* EOP? */
 793                                 loopback);
 794                         frag_len_left -= len;
 795                         offset += len;
 796                 }
 797         }
 798 }
 799
 800 static inline void enic_queue_wq_skb(struct enic *enic,
 801         struct vnic_wq *wq, struct sk_buff *skb)
 802 {
 803         unsigned int mss = skb_shinfo(skb)->gso_size;
 804         unsigned int vlan_tag = 0;
 805         int vlan_tag_insert = 0;
 806         int loopback = 0;
 807
 808         if (vlan_tx_tag_present(skb)) {
 809                 /* VLAN tag from trunking driver */
 810                 vlan_tag_insert = 1;
 811                 vlan_tag = vlan_tx_tag_get(skb);
 812         } else if (enic->loop_enable) {
 813                 vlan_tag = enic->loop_tag;
 814                 loopback = 1;
 815         }
 816
 817         if (mss)
 818                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
 819                         vlan_tag_insert, vlan_tag, loopback);
 820         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 821                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
 822                         vlan_tag_insert, vlan_tag, loopback);
 823         else
 824                 enic_queue_wq_skb_vlan(enic, wq, skb,
 825                         vlan_tag_insert, vlan_tag, loopback);
 826 }
 827
 828 /* netif_tx_lock held, process context with BHs disabled, or BH */
 829 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 830         struct net_device *netdev)
 831 {
 832         struct enic *enic = netdev_priv(netdev);
 833         struct vnic_wq *wq = &enic->wq[0];
 834         unsigned long flags;
 835
 836         if (skb->len <= 0) {
 837                 dev_kfree_skb(skb);
 838                 return NETDEV_TX_OK;
 839         }
 840
 841         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 842          * which is very likely.  In the off chance it's going to take
 843          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 844          */
 845
 846         if (skb_shinfo(skb)->gso_size == 0 &&
 847             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 848             skb_linearize(skb)) {
 849                 dev_kfree_skb(skb);
 850                 return NETDEV_TX_OK;
 851         }
 852
 853         spin_lock_irqsave(&enic->wq_lock[0], flags);
 854
 855         if (vnic_wq_desc_avail(wq) <
 856             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 857                 netif_stop_queue(netdev);
 858                 /* This is a hard error, log it */
 859                 netdev_err(netdev, "BUG! Tx ring full when queue awake!\n");
 860                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 861                 return NETDEV_TX_BUSY;
 862         }
 863
 864         enic_queue_wq_skb(enic, wq, skb);
 865
 866         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 867                 netif_stop_queue(netdev);
 868
 869         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 870
 871         return NETDEV_TX_OK;
 872 }
 873
 874 /* dev_base_lock rwlock held, nominally process context */
 875 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
 876 {
 877         struct enic *enic = netdev_priv(netdev);
 878         struct net_device_stats *net_stats = &netdev->stats;
 879         struct vnic_stats *stats;
 880
 881         enic_dev_stats_dump(enic, &stats);
 882
 883         net_stats->tx_packets = stats->tx.tx_frames_ok;
 884         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 885         net_stats->tx_errors = stats->tx.tx_errors;
 886         net_stats->tx_dropped = stats->tx.tx_drops;
 887
 888         net_stats->rx_packets = stats->rx.rx_frames_ok;
 889         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 890         net_stats->rx_errors = stats->rx.rx_errors;
 891         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 892         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 893         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 894         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 895
 896         return net_stats;
 897 }
 898
 899 static void enic_reset_multicast_list(struct enic *enic)
 900 {
 901         enic->mc_count = 0;
 902         enic->flags = 0;
 903 }
 904
 905 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 906 {
 907         struct enic *enic = netdev_priv(netdev);
 908
 909         if (enic_is_dynamic(enic)) {
 910                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 911                         return -EADDRNOTAVAIL;
 912         } else {
 913                 if (!is_valid_ether_addr(addr))
 914                         return -EADDRNOTAVAIL;
 915         }
 916
 917         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 918
 919         return 0;
 920 }
 921
 922 static int enic_dev_add_station_addr(struct enic *enic)
 923 {
 924         int err = 0;
 925
 926         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 927                 spin_lock(&enic->devcmd_lock);
 928                 err = vnic_dev_add_addr(enic->vdev, enic->netdev->dev_addr);
 929                 spin_unlock(&enic->devcmd_lock);
 930         }
 931
 932         return err;
 933 }
 934
 935 static int enic_dev_del_station_addr(struct enic *enic)
 936 {
 937         int err = 0;
 938
 939         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 940                 spin_lock(&enic->devcmd_lock);
 941                 err = vnic_dev_del_addr(enic->vdev, enic->netdev->dev_addr);
 942                 spin_unlock(&enic->devcmd_lock);
 943         }
 944
 945         return err;
 946 }
 947
 948 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 949 {
 950         struct enic *enic = netdev_priv(netdev);
 951         struct sockaddr *saddr = p;
 952         char *addr = saddr->sa_data;
 953         int err;
 954
 955         if (netif_running(enic->netdev)) {
 956                 err = enic_dev_del_station_addr(enic);
 957                 if (err)
 958                         return err;
 959         }
 960
 961         err = enic_set_mac_addr(netdev, addr);
 962         if (err)
 963                 return err;
 964
 965         if (netif_running(enic->netdev)) {
 966                 err = enic_dev_add_station_addr(enic);
 967                 if (err)
 968                         return err;
 969         }
 970
 971         return err;
 972 }
 973
 974 static int enic_set_mac_address(struct net_device *netdev, void *p)
 975 {
 976         struct sockaddr *saddr = p;
 977         char *addr = saddr->sa_data;
 978         struct enic *enic = netdev_priv(netdev);
 979         int err;
 980
 981         err = enic_dev_del_station_addr(enic);
 982         if (err)
 983                 return err;
 984
 985         err = enic_set_mac_addr(netdev, addr);
 986         if (err)
 987                 return err;
 988
 989         return enic_dev_add_station_addr(enic);
 990 }
 991
 992 static int enic_dev_packet_filter(struct enic *enic, int directed,
 993         int multicast, int broadcast, int promisc, int allmulti)
 994 {
 995         int err;
 996
 997         spin_lock(&enic->devcmd_lock);
 998         err = vnic_dev_packet_filter(enic->vdev, directed,
 999                 multicast, broadcast, promisc, allmulti);
1000         spin_unlock(&enic->devcmd_lock);
1001
1002         return err;
1003 }
1004
1005 static int enic_dev_add_addr(struct enic *enic, u8 *addr)
1006 {
1007         int err;
1008
1009         spin_lock(&enic->devcmd_lock);
1010         err = vnic_dev_add_addr(enic->vdev, addr);
1011         spin_unlock(&enic->devcmd_lock);
1012
1013         return err;
1014 }
1015
1016 static int enic_dev_del_addr(struct enic *enic, u8 *addr)
1017 {
1018         int err;
1019
1020         spin_lock(&enic->devcmd_lock);
1021         err = vnic_dev_del_addr(enic->vdev, addr);
1022         spin_unlock(&enic->devcmd_lock);
1023
1024         return err;
1025 }
1026
1027 static void enic_add_multicast_addr_list(struct enic *enic)
1028 {
1029         struct net_device *netdev = enic->netdev;
1030         struct netdev_hw_addr *ha;
1031         unsigned int mc_count = netdev_mc_count(netdev);
1032         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
1033         unsigned int i, j;
1034
1035         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS) {
1036                 netdev_warn(netdev, "Registering only %d out of %d "
1037                         "multicast addresses\n",
1038                         ENIC_MULTICAST_PERFECT_FILTERS, mc_count);
1039                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
1040         }
1041
1042         /* Is there an easier way?  Trying to minimize to
1043          * calls to add/del multicast addrs.  We keep the
1044          * addrs from the last call in enic->mc_addr and
1045          * look for changes to add/del.
1046          */
1047
1048         i = 0;
1049         netdev_for_each_mc_addr(ha, netdev) {
1050                 if (i == mc_count)
1051                         break;
1052                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
1053         }
1054
1055         for (i = 0; i < enic->mc_count; i++) {
1056                 for (j = 0; j < mc_count; j++)
1057                         if (compare_ether_addr(enic->mc_addr[i],
1058                                 mc_addr[j]) == 0)
1059                                 break;
1060                 if (j == mc_count)
1061                         enic_dev_del_addr(enic, enic->mc_addr[i]);
1062         }
1063
1064         for (i = 0; i < mc_count; i++) {
1065                 for (j = 0; j < enic->mc_count; j++)
1066                         if (compare_ether_addr(mc_addr[i],
1067                                 enic->mc_addr[j]) == 0)
1068                                 break;
1069                 if (j == enic->mc_count)
1070                         enic_dev_add_addr(enic, mc_addr[i]);
1071         }
1072
1073         /* Save the list to compare against next time
1074          */
1075
1076         for (i = 0; i < mc_count; i++)
1077                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
1078
1079         enic->mc_count = mc_count;
1080 }
1081
1082 static void enic_add_unicast_addr_list(struct enic *enic)
1083 {
1084         struct net_device *netdev = enic->netdev;
1085         struct netdev_hw_addr *ha;
1086         unsigned int uc_count = netdev_uc_count(netdev);
1087         u8 uc_addr[ENIC_UNICAST_PERFECT_FILTERS][ETH_ALEN];
1088         unsigned int i, j;
1089
1090         if (uc_count > ENIC_UNICAST_PERFECT_FILTERS) {
1091                 netdev_warn(netdev, "Registering only %d out of %d "
1092                         "unicast addresses\n",
1093                         ENIC_UNICAST_PERFECT_FILTERS, uc_count);
1094                 uc_count = ENIC_UNICAST_PERFECT_FILTERS;
1095         }
1096
1097         /* Is there an easier way?  Trying to minimize to
1098          * calls to add/del unicast addrs.  We keep the
1099          * addrs from the last call in enic->uc_addr and
1100          * look for changes to add/del.
1101          */
1102
1103         i = 0;
1104         netdev_for_each_uc_addr(ha, netdev) {
1105                 if (i == uc_count)
1106                         break;
1107                 memcpy(uc_addr[i++], ha->addr, ETH_ALEN);
1108         }
1109
1110         for (i = 0; i < enic->uc_count; i++) {
1111                 for (j = 0; j < uc_count; j++)
1112                         if (compare_ether_addr(enic->uc_addr[i],
1113                                 uc_addr[j]) == 0)
1114                                 break;
1115                 if (j == uc_count)
1116                         enic_dev_del_addr(enic, enic->uc_addr[i]);
1117         }
1118
1119         for (i = 0; i < uc_count; i++) {
1120                 for (j = 0; j < enic->uc_count; j++)
1121                         if (compare_ether_addr(uc_addr[i],
1122                                 enic->uc_addr[j]) == 0)
1123                                 break;
1124                 if (j == enic->uc_count)
1125                         enic_dev_add_addr(enic, uc_addr[i]);
1126         }
1127
1128         /* Save the list to compare against next time
1129          */
1130
1131         for (i = 0; i < uc_count; i++)
1132                 memcpy(enic->uc_addr[i], uc_addr[i], ETH_ALEN);
1133
1134         enic->uc_count = uc_count;
1135 }
1136
1137 /* netif_tx_lock held, BHs disabled */
1138 static void enic_set_rx_mode(struct net_device *netdev)
1139 {
1140         struct enic *enic = netdev_priv(netdev);
1141         int directed = 1;
1142         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
1143         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
1144         int promisc = (netdev->flags & IFF_PROMISC) ||
1145                 netdev_uc_count(netdev) > ENIC_UNICAST_PERFECT_FILTERS;
1146         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
1147                 netdev_mc_count(netdev) > ENIC_MULTICAST_PERFECT_FILTERS;
1148         unsigned int flags = netdev->flags |
1149                 (allmulti ? IFF_ALLMULTI : 0) |
1150                 (promisc ? IFF_PROMISC : 0);
1151
1152         if (enic->flags != flags) {
1153                 enic->flags = flags;
1154                 enic_dev_packet_filter(enic, directed,
1155                         multicast, broadcast, promisc, allmulti);
1156         }
1157
1158         if (!promisc) {
1159                 enic_add_unicast_addr_list(enic);
1160                 if (!allmulti)
1161                         enic_add_multicast_addr_list(enic);
1162         }
1163 }
1164
1165 /* rtnl lock is held */
1166 static void enic_vlan_rx_register(struct net_device *netdev,
1167         struct vlan_group *vlan_group)
1168 {
1169         struct enic *enic = netdev_priv(netdev);
1170         enic->vlan_group = vlan_group;
1171 }
1172
1173 /* rtnl lock is held */
1174 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
1175 {
1176         struct enic *enic = netdev_priv(netdev);
1177
1178         spin_lock(&enic->devcmd_lock);
1179         enic_add_vlan(enic, vid);
1180         spin_unlock(&enic->devcmd_lock);
1181 }
1182
1183 /* rtnl lock is held */
1184 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
1185 {
1186         struct enic *enic = netdev_priv(netdev);
1187
1188         spin_lock(&enic->devcmd_lock);
1189         enic_del_vlan(enic, vid);
1190         spin_unlock(&enic->devcmd_lock);
1191 }
1192
1193 /* netif_tx_lock held, BHs disabled */
1194 static void enic_tx_timeout(struct net_device *netdev)
1195 {
1196         struct enic *enic = netdev_priv(netdev);
1197         schedule_work(&enic->reset);
1198 }
1199
1200 static int enic_vnic_dev_deinit(struct enic *enic)
1201 {
1202         int err;
1203
1204         spin_lock(&enic->devcmd_lock);
1205         err = vnic_dev_deinit(enic->vdev);
1206         spin_unlock(&enic->devcmd_lock);
1207
1208         return err;
1209 }
1210
1211 static int enic_dev_init_prov(struct enic *enic, struct vic_provinfo *vp)
1212 {
1213         int err;
1214
1215         spin_lock(&enic->devcmd_lock);
1216         err = vnic_dev_init_prov(enic->vdev,
1217                 (u8 *)vp, vic_provinfo_size(vp));
1218         spin_unlock(&enic->devcmd_lock);
1219
1220         return err;
1221 }
1222
1223 static int enic_dev_init_done(struct enic *enic, int *done, int *error)
1224 {
1225         int err;
1226
1227         spin_lock(&enic->devcmd_lock);
1228         err = vnic_dev_init_done(enic->vdev, done, error);
1229         spin_unlock(&enic->devcmd_lock);
1230
1231         return err;
1232 }
1233
1234 static int enic_set_vf_mac(struct net_device *netdev, int vf, u8 *mac)
1235 {
1236         struct enic *enic = netdev_priv(netdev);
1237
1238         if (vf != PORT_SELF_VF)
1239                 return -EOPNOTSUPP;
1240
1241         /* Ignore the vf argument for now. We can assume the request
1242          * is coming on a vf.
1243          */
1244         if (is_valid_ether_addr(mac)) {
1245                 memcpy(enic->pp.vf_mac, mac, ETH_ALEN);
1246                 return 0;
1247         } else
1248                 return -EINVAL;
1249 }
1250
1251 static int enic_set_port_profile(struct enic *enic, u8 *mac)
1252 {
1253         struct vic_provinfo *vp;
1254         u8 oui[3] = VIC_PROVINFO_CISCO_OUI;
1255         u16 os_type = VIC_GENERIC_PROV_OS_TYPE_LINUX;
1256         char uuid_str[38];
1257         char client_mac_str[18];
1258         u8 *client_mac;
1259         int err;
1260
1261         err = enic_vnic_dev_deinit(enic);
1262         if (err)
1263                 return err;
1264
1265         switch (enic->pp.request) {
1266
1267         case PORT_REQUEST_ASSOCIATE:
1268
1269                 if (!(enic->pp.set & ENIC_SET_NAME) || !strlen(enic->pp.name))
1270                         return -EINVAL;
1271
1272                 if (!is_valid_ether_addr(mac))
1273                         return -EADDRNOTAVAIL;
1274
1275                 vp = vic_provinfo_alloc(GFP_KERNEL, oui,
1276                         VIC_PROVINFO_GENERIC_TYPE);
1277                 if (!vp)
1278                         return -ENOMEM;
1279
1280                 vic_provinfo_add_tlv(vp,
1281                         VIC_GENERIC_PROV_TLV_PORT_PROFILE_NAME_STR,
1282                         strlen(enic->pp.name) + 1, enic->pp.name);
1283
1284                 if (!is_zero_ether_addr(enic->pp.mac_addr))
1285                         client_mac = enic->pp.mac_addr;
1286                 else
1287                         client_mac = mac;
1288
1289                 vic_provinfo_add_tlv(vp,
1290                         VIC_GENERIC_PROV_TLV_CLIENT_MAC_ADDR,
1291                         ETH_ALEN, client_mac);
1292
1293                 sprintf(client_mac_str, "%pM", client_mac);
1294                 vic_provinfo_add_tlv(vp,
1295                         VIC_GENERIC_PROV_TLV_CLUSTER_PORT_UUID_STR,
1296                         sizeof(client_mac_str), client_mac_str);
1297
1298                 if (enic->pp.set & ENIC_SET_INSTANCE) {
1299                         sprintf(uuid_str, "%pUB", enic->pp.instance_uuid);
1300                         vic_provinfo_add_tlv(vp,
1301                                 VIC_GENERIC_PROV_TLV_CLIENT_UUID_STR,
1302                                 sizeof(uuid_str), uuid_str);
1303                 }
1304
1305                 if (enic->pp.set & ENIC_SET_HOST) {
1306                         sprintf(uuid_str, "%pUB", enic->pp.host_uuid);
1307                         vic_provinfo_add_tlv(vp,
1308                                 VIC_GENERIC_PROV_TLV_HOST_UUID_STR,
1309                                 sizeof(uuid_str), uuid_str);
1310                 }
1311
1312                 os_type = htons(os_type);
1313                 vic_provinfo_add_tlv(vp,
1314                         VIC_GENERIC_PROV_TLV_OS_TYPE,
1315                         sizeof(os_type), &os_type);
1316
1317                 err = enic_dev_init_prov(enic, vp);
1318                 vic_provinfo_free(vp);
1319                 if (err)
1320                         return err;
1321
1322                 enic->pp.set |= ENIC_SET_APPLIED;
1323                 break;
1324
1325         case PORT_REQUEST_DISASSOCIATE:
1326                 enic->pp.set &= ~ENIC_SET_APPLIED;
1327                 break;
1328
1329         default:
1330                 return -EINVAL;
1331         }
1332
1333         return 0;
1334 }
1335
1336 static int enic_set_vf_port(struct net_device *netdev, int vf,
1337         struct nlattr *port[])
1338 {
1339         struct enic *enic = netdev_priv(netdev);
1340         struct enic_port_profile new_pp;
1341         int err = 0;
1342
1343         memset(&new_pp, 0, sizeof(new_pp));
1344
1345         if (port[IFLA_PORT_REQUEST]) {
1346                 new_pp.set |= ENIC_SET_REQUEST;
1347                 new_pp.request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1348         }
1349
1350         if (port[IFLA_PORT_PROFILE]) {
1351                 new_pp.set |= ENIC_SET_NAME;
1352                 memcpy(new_pp.name, nla_data(port[IFLA_PORT_PROFILE]),
1353                         PORT_PROFILE_MAX);
1354         }
1355
1356         if (port[IFLA_PORT_INSTANCE_UUID]) {
1357                 new_pp.set |= ENIC_SET_INSTANCE;
1358                 memcpy(new_pp.instance_uuid,
1359                         nla_data(port[IFLA_PORT_INSTANCE_UUID]), PORT_UUID_MAX);
1360         }
1361
1362         if (port[IFLA_PORT_HOST_UUID]) {
1363                 new_pp.set |= ENIC_SET_HOST;
1364                 memcpy(new_pp.host_uuid,
1365                         nla_data(port[IFLA_PORT_HOST_UUID]), PORT_UUID_MAX);
1366         }
1367
1368         /* don't support VFs, yet */
1369         if (vf != PORT_SELF_VF)
1370                 return -EOPNOTSUPP;
1371
1372         if (!(new_pp.set & ENIC_SET_REQUEST))
1373                 return -EOPNOTSUPP;
1374
1375         if (new_pp.request == PORT_REQUEST_ASSOCIATE) {
1376                 /* Special case handling */
1377                 if (!is_zero_ether_addr(enic->pp.vf_mac))
1378                         memcpy(new_pp.mac_addr, enic->pp.vf_mac, ETH_ALEN);
1379
1380                 if (is_zero_ether_addr(netdev->dev_addr))
1381                         random_ether_addr(netdev->dev_addr);
1382         } else if (new_pp.request == PORT_REQUEST_DISASSOCIATE) {
1383                 if (!is_zero_ether_addr(enic->pp.mac_addr))
1384                         enic_dev_del_addr(enic, enic->pp.mac_addr);
1385         }
1386
1387         memcpy(&enic->pp, &new_pp, sizeof(struct enic_port_profile));
1388
1389         err = enic_set_port_profile(enic, netdev->dev_addr);
1390         if (err)
1391                 goto set_port_profile_cleanup;
1392
1393         if (!is_zero_ether_addr(enic->pp.mac_addr))
1394                 enic_dev_add_addr(enic, enic->pp.mac_addr);
1395
1396 set_port_profile_cleanup:
1397         memset(enic->pp.vf_mac, 0, ETH_ALEN);
1398
1399         if (err || enic->pp.request == PORT_REQUEST_DISASSOCIATE) {
1400                 memset(netdev->dev_addr, 0, ETH_ALEN);
1401                 memset(enic->pp.mac_addr, 0, ETH_ALEN);
1402         }
1403
1404         return err;
1405 }
1406
1407 static int enic_get_vf_port(struct net_device *netdev, int vf,
1408         struct sk_buff *skb)
1409 {
1410         struct enic *enic = netdev_priv(netdev);
1411         int err, error, done;
1412         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1413
1414         if (!(enic->pp.set & ENIC_SET_APPLIED))
1415                 return -ENODATA;
1416
1417         err = enic_dev_init_done(enic, &done, &error);
1418         if (err)
1419                 error = err;
1420
1421         switch (error) {
1422         case ERR_SUCCESS:
1423                 if (!done)
1424                         response = PORT_PROFILE_RESPONSE_INPROGRESS;
1425                 break;
1426         case ERR_EINVAL:
1427                 response = PORT_PROFILE_RESPONSE_INVALID;
1428                 break;
1429         case ERR_EBADSTATE:
1430                 response = PORT_PROFILE_RESPONSE_BADSTATE;
1431                 break;
1432         case ERR_ENOMEM:
1433                 response = PORT_PROFILE_RESPONSE_INSUFFICIENT_RESOURCES;
1434                 break;
1435         default:
1436                 response = PORT_PROFILE_RESPONSE_ERROR;
1437                 break;
1438         }
1439
1440         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1441         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1442         if (enic->pp.set & ENIC_SET_NAME)
1443                 NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1444                         enic->pp.name);
1445         if (enic->pp.set & ENIC_SET_INSTANCE)
1446                 NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1447                         enic->pp.instance_uuid);
1448         if (enic->pp.set & ENIC_SET_HOST)
1449                 NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1450                         enic->pp.host_uuid);
1451
1452         return 0;
1453
1454 nla_put_failure:
1455         return -EMSGSIZE;
1456 }
1457
1458 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1459 {
1460         struct enic *enic = vnic_dev_priv(rq->vdev);
1461
1462         if (!buf->os_buf)
1463                 return;
1464
1465         pci_unmap_single(enic->pdev, buf->dma_addr,
1466                 buf->len, PCI_DMA_FROMDEVICE);
1467         dev_kfree_skb_any(buf->os_buf);
1468 }
1469
1470 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1471 {
1472         struct enic *enic = vnic_dev_priv(rq->vdev);
1473         struct net_device *netdev = enic->netdev;
1474         struct sk_buff *skb;
1475         unsigned int len = netdev->mtu + VLAN_ETH_HLEN;
1476         unsigned int os_buf_index = 0;
1477         dma_addr_t dma_addr;
1478
1479         skb = netdev_alloc_skb_ip_align(netdev, len);
1480         if (!skb)
1481                 return -ENOMEM;
1482
1483         dma_addr = pci_map_single(enic->pdev, skb->data,
1484                 len, PCI_DMA_FROMDEVICE);
1485
1486         enic_queue_rq_desc(rq, skb, os_buf_index,
1487                 dma_addr, len);
1488
1489         return 0;
1490 }
1491
1492 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
1493 {
1494         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
1495
1496         if (vnic_rq_posting_soon(rq)) {
1497
1498                 /* SW workaround for A0 HW erratum: if we're just about
1499                  * to write posted_index, insert a dummy desc
1500                  * of type resvd
1501                  */
1502
1503                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
1504                 vnic_rq_post(rq, 0, 0, 0, 0);
1505         } else {
1506                 return enic_rq_alloc_buf(rq);
1507         }
1508
1509         return 0;
1510 }
1511
1512 static int enic_dev_hw_version(struct enic *enic,
1513         enum vnic_dev_hw_version *hw_ver)
1514 {
1515         int err;
1516
1517         spin_lock(&enic->devcmd_lock);
1518         err = vnic_dev_hw_version(enic->vdev, hw_ver);
1519         spin_unlock(&enic->devcmd_lock);
1520
1521         return err;
1522 }
1523
1524 static int enic_set_rq_alloc_buf(struct enic *enic)
1525 {
1526         enum vnic_dev_hw_version hw_ver;
1527         int err;
1528
1529         err = enic_dev_hw_version(enic, &hw_ver);
1530         if (err)
1531                 return err;
1532
1533         switch (hw_ver) {
1534         case VNIC_DEV_HW_VER_A1:
1535                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
1536                 break;
1537         case VNIC_DEV_HW_VER_A2:
1538         case VNIC_DEV_HW_VER_UNKNOWN:
1539                 enic->rq_alloc_buf = enic_rq_alloc_buf;
1540                 break;
1541         default:
1542                 return -ENODEV;
1543         }
1544
1545         return 0;
1546 }
1547
1548 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1549         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1550         int skipped, void *opaque)
1551 {
1552         struct enic *enic = vnic_dev_priv(rq->vdev);
1553         struct net_device *netdev = enic->netdev;
1554         struct sk_buff *skb;
1555
1556         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1557         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1558         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1559         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1560         u8 packet_error;
1561         u16 q_number, completed_index, bytes_written, vlan_tci, checksum;
1562         u32 rss_hash;
1563
1564         if (skipped)
1565                 return;
1566
1567         skb = buf->os_buf;
1568         prefetch(skb->data - NET_IP_ALIGN);
1569         pci_unmap_single(enic->pdev, buf->dma_addr,
1570                 buf->len, PCI_DMA_FROMDEVICE);
1571
1572         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1573                 &type, &color, &q_number, &completed_index,
1574                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1575                 &csum_not_calc, &rss_hash, &bytes_written,
1576                 &packet_error, &vlan_stripped, &vlan_tci, &checksum,
1577                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1578                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1579                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1580                 &fcs_ok);
1581
1582         if (packet_error) {
1583
1584                 if (!fcs_ok) {
1585                         if (bytes_written > 0)
1586                                 enic->rq_bad_fcs++;
1587                         else if (bytes_written == 0)
1588                                 enic->rq_truncated_pkts++;
1589                 }
1590
1591                 dev_kfree_skb_any(skb);
1592
1593                 return;
1594         }
1595
1596         if (eop && bytes_written > 0) {
1597
1598                 /* Good receive
1599                  */
1600
1601                 skb_put(skb, bytes_written);
1602                 skb->protocol = eth_type_trans(skb, netdev);
1603
1604                 if (enic->csum_rx_enabled && !csum_not_calc) {
1605                         skb->csum = htons(checksum);
1606                         skb->ip_summed = CHECKSUM_COMPLETE;
1607                 }
1608
1609                 skb->dev = netdev;
1610
1611                 if (enic->vlan_group && vlan_stripped &&
1612                         (vlan_tci & CQ_ENET_RQ_DESC_VLAN_TCI_VLAN_MASK)) {
1613
1614                         if (netdev->features & NETIF_F_GRO)
1615                                 vlan_gro_receive(&enic->napi[q_number],
1616                                         enic->vlan_group, vlan_tci, skb);
1617                         else
1618                                 vlan_hwaccel_receive_skb(skb,
1619                                         enic->vlan_group, vlan_tci);
1620
1621                 } else {
1622
1623                         if (netdev->features & NETIF_F_GRO)
1624                                 napi_gro_receive(&enic->napi[q_number], skb);
1625                         else
1626                                 netif_receive_skb(skb);
1627
1628                 }
1629         } else {
1630
1631                 /* Buffer overflow
1632                  */
1633
1634                 dev_kfree_skb_any(skb);
1635         }
1636 }
1637
1638 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1639         u8 type, u16 q_number, u16 completed_index, void *opaque)
1640 {
1641         struct enic *enic = vnic_dev_priv(vdev);
1642
1643         vnic_rq_service(&enic->rq[q_number], cq_desc,
1644                 completed_index, VNIC_RQ_RETURN_DESC,
1645                 enic_rq_indicate_buf, opaque);
1646
1647         return 0;
1648 }
1649
1650 static int enic_poll(struct napi_struct *napi, int budget)
1651 {
1652         struct net_device *netdev = napi->dev;
1653         struct enic *enic = netdev_priv(netdev);
1654         unsigned int cq_rq = enic_cq_rq(enic, 0);
1655         unsigned int cq_wq = enic_cq_wq(enic, 0);
1656         unsigned int intr = enic_legacy_io_intr();
1657         unsigned int rq_work_to_do = budget;
1658         unsigned int wq_work_to_do = -1; /* no limit */
1659         unsigned int  work_done, rq_work_done, wq_work_done;
1660         int err;
1661
1662         /* Service RQ (first) and WQ
1663          */
1664
1665         rq_work_done = vnic_cq_service(&enic->cq[cq_rq],
1666                 rq_work_to_do, enic_rq_service, NULL);
1667
1668         wq_work_done = vnic_cq_service(&enic->cq[cq_wq],
1669                 wq_work_to_do, enic_wq_service, NULL);
1670
1671         /* Accumulate intr event credits for this polling
1672          * cycle.  An intr event is the completion of a
1673          * a WQ or RQ packet.
1674          */
1675
1676         work_done = rq_work_done + wq_work_done;
1677
1678         if (work_done > 0)
1679                 vnic_intr_return_credits(&enic->intr[intr],
1680                         work_done,
1681                         0 /* don't unmask intr */,
1682                         0 /* don't reset intr timer */);
1683
1684         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1685
1686         /* Buffer allocation failed. Stay in polling
1687          * mode so we can try to fill the ring again.
1688          */
1689
1690         if (err)
1691                 rq_work_done = rq_work_to_do;
1692
1693         if (rq_work_done < rq_work_to_do) {
1694
1695                 /* Some work done, but not enough to stay in polling,
1696                  * exit polling
1697                  */
1698
1699                 napi_complete(napi);
1700                 vnic_intr_unmask(&enic->intr[intr]);
1701         }
1702
1703         return rq_work_done;
1704 }
1705
1706 static int enic_poll_msix(struct napi_struct *napi, int budget)
1707 {
1708         struct net_device *netdev = napi->dev;
1709         struct enic *enic = netdev_priv(netdev);
1710         unsigned int rq = (napi - &enic->napi[0]);
1711         unsigned int cq = enic_cq_rq(enic, rq);
1712         unsigned int intr = enic_msix_rq_intr(enic, rq);
1713         unsigned int work_to_do = budget;
1714         unsigned int work_done;
1715         int err;
1716
1717         /* Service RQ
1718          */
1719
1720         work_done = vnic_cq_service(&enic->cq[cq],
1721                 work_to_do, enic_rq_service, NULL);
1722
1723         /* Return intr event credits for this polling
1724          * cycle.  An intr event is the completion of a
1725          * RQ packet.
1726          */
1727
1728         if (work_done > 0)
1729                 vnic_intr_return_credits(&enic->intr[intr],
1730                         work_done,
1731                         0 /* don't unmask intr */,
1732                         0 /* don't reset intr timer */);
1733
1734         err = vnic_rq_fill(&enic->rq[rq], enic->rq_alloc_buf);
1735
1736         /* Buffer allocation failed. Stay in polling mode
1737          * so we can try to fill the ring again.
1738          */
1739
1740         if (err)
1741                 work_done = work_to_do;
1742
1743         if (work_done < work_to_do) {
1744
1745                 /* Some work done, but not enough to stay in polling,
1746                  * exit polling
1747                  */
1748
1749                 napi_complete(napi);
1750                 vnic_intr_unmask(&enic->intr[intr]);
1751         }
1752
1753         return work_done;
1754 }
1755
1756 static void enic_notify_timer(unsigned long data)
1757 {
1758         struct enic *enic = (struct enic *)data;
1759
1760         enic_notify_check(enic);
1761
1762         mod_timer(&enic->notify_timer,
1763                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1764 }
1765
1766 static void enic_free_intr(struct enic *enic)
1767 {
1768         struct net_device *netdev = enic->netdev;
1769         unsigned int i;
1770
1771         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1772         case VNIC_DEV_INTR_MODE_INTX:
1773                 free_irq(enic->pdev->irq, netdev);
1774                 break;
1775         case VNIC_DEV_INTR_MODE_MSI:
1776                 free_irq(enic->pdev->irq, enic);
1777                 break;
1778         case VNIC_DEV_INTR_MODE_MSIX:
1779                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1780                         if (enic->msix[i].requested)
1781                                 free_irq(enic->msix_entry[i].vector,
1782                                         enic->msix[i].devid);
1783                 break;
1784         default:
1785                 break;
1786         }
1787 }
1788
1789 static int enic_request_intr(struct enic *enic)
1790 {
1791         struct net_device *netdev = enic->netdev;
1792         unsigned int i, intr;
1793         int err = 0;
1794
1795         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1796
1797         case VNIC_DEV_INTR_MODE_INTX:
1798
1799                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1800                         IRQF_SHARED, netdev->name, netdev);
1801                 break;
1802
1803         case VNIC_DEV_INTR_MODE_MSI:
1804
1805                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1806                         0, netdev->name, enic);
1807                 break;
1808
1809         case VNIC_DEV_INTR_MODE_MSIX:
1810
1811                 for (i = 0; i < enic->rq_count; i++) {
1812                         intr = enic_msix_rq_intr(enic, i);
1813                         sprintf(enic->msix[intr].devname,
1814                                 "%.11s-rx-%d", netdev->name, i);
1815                         enic->msix[intr].isr = enic_isr_msix_rq;
1816                         enic->msix[intr].devid = &enic->napi[i];
1817                 }
1818
1819                 for (i = 0; i < enic->wq_count; i++) {
1820                         intr = enic_msix_wq_intr(enic, i);
1821                         sprintf(enic->msix[intr].devname,
1822                                 "%.11s-tx-%d", netdev->name, i);
1823                         enic->msix[intr].isr = enic_isr_msix_wq;
1824                         enic->msix[intr].devid = enic;
1825                 }
1826
1827                 intr = enic_msix_err_intr(enic);
1828                 sprintf(enic->msix[intr].devname,
1829                         "%.11s-err", netdev->name);
1830                 enic->msix[intr].isr = enic_isr_msix_err;
1831                 enic->msix[intr].devid = enic;
1832
1833                 intr = enic_msix_notify_intr(enic);
1834                 sprintf(enic->msix[intr].devname,
1835                         "%.11s-notify", netdev->name);
1836                 enic->msix[intr].isr = enic_isr_msix_notify;
1837                 enic->msix[intr].devid = enic;
1838
1839                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1840                         enic->msix[i].requested = 0;
1841
1842                 for (i = 0; i < enic->intr_count; i++) {
1843                         err = request_irq(enic->msix_entry[i].vector,
1844                                 enic->msix[i].isr, 0,
1845                                 enic->msix[i].devname,
1846                                 enic->msix[i].devid);
1847                         if (err) {
1848                                 enic_free_intr(enic);
1849                                 break;
1850                         }
1851                         enic->msix[i].requested = 1;
1852                 }
1853
1854                 break;
1855
1856         default:
1857                 break;
1858         }
1859
1860         return err;
1861 }
1862
1863 static void enic_synchronize_irqs(struct enic *enic)
1864 {
1865         unsigned int i;
1866
1867         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1868         case VNIC_DEV_INTR_MODE_INTX:
1869         case VNIC_DEV_INTR_MODE_MSI:
1870                 synchronize_irq(enic->pdev->irq);
1871                 break;
1872         case VNIC_DEV_INTR_MODE_MSIX:
1873                 for (i = 0; i < enic->intr_count; i++)
1874                         synchronize_irq(enic->msix_entry[i].vector);
1875                 break;
1876         default:
1877                 break;
1878         }
1879 }
1880
1881 static int enic_dev_notify_set(struct enic *enic)
1882 {
1883         int err;
1884
1885         spin_lock(&enic->devcmd_lock);
1886         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1887         case VNIC_DEV_INTR_MODE_INTX:
1888                 err = vnic_dev_notify_set(enic->vdev,
1889                         enic_legacy_notify_intr());
1890                 break;
1891         case VNIC_DEV_INTR_MODE_MSIX:
1892                 err = vnic_dev_notify_set(enic->vdev,
1893                         enic_msix_notify_intr(enic));
1894                 break;
1895         default:
1896                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1897                 break;
1898         }
1899         spin_unlock(&enic->devcmd_lock);
1900
1901         return err;
1902 }
1903
1904 static int enic_dev_notify_unset(struct enic *enic)
1905 {
1906         int err;
1907
1908         spin_lock(&enic->devcmd_lock);
1909         err = vnic_dev_notify_unset(enic->vdev);
1910         spin_unlock(&enic->devcmd_lock);
1911
1912         return err;
1913 }
1914
1915 static int enic_dev_enable(struct enic *enic)
1916 {
1917         int err;
1918
1919         spin_lock(&enic->devcmd_lock);
1920         err = vnic_dev_enable_wait(enic->vdev);
1921         spin_unlock(&enic->devcmd_lock);
1922
1923         return err;
1924 }
1925
1926 static int enic_dev_disable(struct enic *enic)
1927 {
1928         int err;
1929
1930         spin_lock(&enic->devcmd_lock);
1931         err = vnic_dev_disable(enic->vdev);
1932         spin_unlock(&enic->devcmd_lock);
1933
1934         return err;
1935 }
1936
1937 static void enic_notify_timer_start(struct enic *enic)
1938 {
1939         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1940         case VNIC_DEV_INTR_MODE_MSI:
1941                 mod_timer(&enic->notify_timer, jiffies);
1942                 break;
1943         default:
1944                 /* Using intr for notification for INTx/MSI-X */
1945                 break;
1946         };
1947 }
1948
1949 /* rtnl lock is held, process context */
1950 static int enic_open(struct net_device *netdev)
1951 {
1952         struct enic *enic = netdev_priv(netdev);
1953         unsigned int i;
1954         int err;
1955
1956         err = enic_request_intr(enic);
1957         if (err) {
1958                 netdev_err(netdev, "Unable to request irq.\n");
1959                 return err;
1960         }
1961
1962         err = enic_dev_notify_set(enic);
1963         if (err) {
1964                 netdev_err(netdev,
1965                         "Failed to alloc notify buffer, aborting.\n");
1966                 goto err_out_free_intr;
1967         }
1968
1969         for (i = 0; i < enic->rq_count; i++) {
1970                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1971                 /* Need at least one buffer on ring to get going */
1972                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1973                         netdev_err(netdev, "Unable to alloc receive buffers\n");
1974                         err = -ENOMEM;
1975                         goto err_out_notify_unset;
1976                 }
1977         }
1978
1979         for (i = 0; i < enic->wq_count; i++)
1980                 vnic_wq_enable(&enic->wq[i]);
1981         for (i = 0; i < enic->rq_count; i++)
1982                 vnic_rq_enable(&enic->rq[i]);
1983
1984         if (enic_is_dynamic(enic) && !is_zero_ether_addr(enic->pp.mac_addr))
1985                 enic_dev_add_addr(enic, enic->pp.mac_addr);
1986         else
1987                 enic_dev_add_station_addr(enic);
1988         enic_set_rx_mode(netdev);
1989
1990         netif_wake_queue(netdev);
1991
1992         for (i = 0; i < enic->rq_count; i++)
1993                 napi_enable(&enic->napi[i]);
1994
1995         enic_dev_enable(enic);
1996
1997         for (i = 0; i < enic->intr_count; i++)
1998                 vnic_intr_unmask(&enic->intr[i]);
1999
2000         enic_notify_timer_start(enic);
2001
2002         return 0;
2003
2004 err_out_notify_unset:
2005         enic_dev_notify_unset(enic);
2006 err_out_free_intr:
2007         enic_free_intr(enic);
2008
2009         return err;
2010 }
2011
2012 /* rtnl lock is held, process context */
2013 static int enic_stop(struct net_device *netdev)
2014 {
2015         struct enic *enic = netdev_priv(netdev);
2016         unsigned int i;
2017         int err;
2018
2019         for (i = 0; i < enic->intr_count; i++) {
2020                 vnic_intr_mask(&enic->intr[i]);
2021                 (void)vnic_intr_masked(&enic->intr[i]); /* flush write */
2022         }
2023
2024         enic_synchronize_irqs(enic);
2025
2026         del_timer_sync(&enic->notify_timer);
2027
2028         enic_dev_disable(enic);
2029
2030         for (i = 0; i < enic->rq_count; i++)
2031                 napi_disable(&enic->napi[i]);
2032
2033         netif_carrier_off(netdev);
2034         netif_tx_disable(netdev);
2035         if (enic_is_dynamic(enic) && !is_zero_ether_addr(enic->pp.mac_addr))
2036                 enic_dev_del_addr(enic, enic->pp.mac_addr);
2037         else
2038                 enic_dev_del_station_addr(enic);
2039
2040         for (i = 0; i < enic->wq_count; i++) {
2041                 err = vnic_wq_disable(&enic->wq[i]);
2042                 if (err)
2043                         return err;
2044         }
2045         for (i = 0; i < enic->rq_count; i++) {
2046                 err = vnic_rq_disable(&enic->rq[i]);
2047                 if (err)
2048                         return err;
2049         }
2050
2051         enic_dev_notify_unset(enic);
2052         enic_free_intr(enic);
2053
2054         for (i = 0; i < enic->wq_count; i++)
2055                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
2056         for (i = 0; i < enic->rq_count; i++)
2057                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
2058         for (i = 0; i < enic->cq_count; i++)
2059                 vnic_cq_clean(&enic->cq[i]);
2060         for (i = 0; i < enic->intr_count; i++)
2061                 vnic_intr_clean(&enic->intr[i]);
2062
2063         return 0;
2064 }
2065
2066 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
2067 {
2068         struct enic *enic = netdev_priv(netdev);
2069         int running = netif_running(netdev);
2070
2071         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
2072                 return -EINVAL;
2073
2074         if (running)
2075                 enic_stop(netdev);
2076
2077         netdev->mtu = new_mtu;
2078
2079         if (netdev->mtu > enic->port_mtu)
2080                 netdev_warn(netdev,
2081                         "interface MTU (%d) set higher than port MTU (%d)\n",
2082                         netdev->mtu, enic->port_mtu);
2083
2084         if (running)
2085                 enic_open(netdev);
2086
2087         return 0;
2088 }
2089
2090 #ifdef CONFIG_NET_POLL_CONTROLLER
2091 static void enic_poll_controller(struct net_device *netdev)
2092 {
2093         struct enic *enic = netdev_priv(netdev);
2094         struct vnic_dev *vdev = enic->vdev;
2095         unsigned int i, intr;
2096
2097         switch (vnic_dev_get_intr_mode(vdev)) {
2098         case VNIC_DEV_INTR_MODE_MSIX:
2099                 for (i = 0; i < enic->rq_count; i++) {
2100                         intr = enic_msix_rq_intr(enic, i);
2101                         enic_isr_msix_rq(enic->msix_entry[intr].vector,
2102                                 &enic->napi[i]);
2103                 }
2104                 intr = enic_msix_wq_intr(enic, i);
2105                 enic_isr_msix_wq(enic->msix_entry[intr].vector, enic);
2106                 break;
2107         case VNIC_DEV_INTR_MODE_MSI:
2108                 enic_isr_msi(enic->pdev->irq, enic);
2109                 break;
2110         case VNIC_DEV_INTR_MODE_INTX:
2111                 enic_isr_legacy(enic->pdev->irq, netdev);
2112                 break;
2113         default:
2114                 break;
2115         }
2116 }
2117 #endif
2118
2119 static int enic_dev_wait(struct vnic_dev *vdev,
2120         int (*start)(struct vnic_dev *, int),
2121         int (*finished)(struct vnic_dev *, int *),
2122         int arg)
2123 {
2124         unsigned long time;
2125         int done;
2126         int err;
2127
2128         BUG_ON(in_interrupt());
2129
2130         err = start(vdev, arg);
2131         if (err)
2132                 return err;
2133
2134         /* Wait for func to complete...2 seconds max
2135          */
2136
2137         time = jiffies + (HZ * 2);
2138         do {
2139
2140                 err = finished(vdev, &done);
2141                 if (err)
2142                         return err;
2143
2144                 if (done)
2145                         return 0;
2146
2147                 schedule_timeout_uninterruptible(HZ / 10);
2148
2149         } while (time_after(time, jiffies));
2150
2151         return -ETIMEDOUT;
2152 }
2153
2154 static int enic_dev_open(struct enic *enic)
2155 {
2156         int err;
2157
2158         err = enic_dev_wait(enic->vdev, vnic_dev_open,
2159                 vnic_dev_open_done, 0);
2160         if (err)
2161                 dev_err(enic_get_dev(enic), "vNIC device open failed, err %d\n",
2162                         err);
2163
2164         return err;
2165 }
2166
2167 static int enic_dev_hang_reset(struct enic *enic)
2168 {
2169         int err;
2170
2171         err = enic_dev_wait(enic->vdev, vnic_dev_hang_reset,
2172                 vnic_dev_hang_reset_done, 0);
2173         if (err)
2174                 netdev_err(enic->netdev, "vNIC hang reset failed, err %d\n",
2175                         err);
2176
2177         return err;
2178 }
2179
2180 static int enic_set_rsskey(struct enic *enic)
2181 {
2182         dma_addr_t rss_key_buf_pa;
2183         union vnic_rss_key *rss_key_buf_va = NULL;
2184         union vnic_rss_key rss_key = {
2185                 .key[0].b = {85, 67, 83, 97, 119, 101, 115, 111, 109, 101},
2186                 .key[1].b = {80, 65, 76, 79, 117, 110, 105, 113, 117, 101},
2187                 .key[2].b = {76, 73, 78, 85, 88, 114, 111, 99, 107, 115},
2188                 .key[3].b = {69, 78, 73, 67, 105, 115, 99, 111, 111, 108},
2189         };
2190         int err;
2191
2192         rss_key_buf_va = pci_alloc_consistent(enic->pdev,
2193                 sizeof(union vnic_rss_key), &rss_key_buf_pa);
2194         if (!rss_key_buf_va)
2195                 return -ENOMEM;
2196
2197         memcpy(rss_key_buf_va, &rss_key, sizeof(union vnic_rss_key));
2198
2199         spin_lock(&enic->devcmd_lock);
2200         err = enic_set_rss_key(enic,
2201                 rss_key_buf_pa,
2202                 sizeof(union vnic_rss_key));
2203         spin_unlock(&enic->devcmd_lock);
2204
2205         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_key),
2206                 rss_key_buf_va, rss_key_buf_pa);
2207
2208         return err;
2209 }
2210
2211 static int enic_set_rsscpu(struct enic *enic, u8 rss_hash_bits)
2212 {
2213         dma_addr_t rss_cpu_buf_pa;
2214         union vnic_rss_cpu *rss_cpu_buf_va = NULL;
2215         unsigned int i;
2216         int err;
2217
2218         rss_cpu_buf_va = pci_alloc_consistent(enic->pdev,
2219                 sizeof(union vnic_rss_cpu), &rss_cpu_buf_pa);
2220         if (!rss_cpu_buf_va)
2221                 return -ENOMEM;
2222
2223         for (i = 0; i < (1 << rss_hash_bits); i++)
2224                 (*rss_cpu_buf_va).cpu[i/4].b[i%4] = i % enic->rq_count;
2225
2226         spin_lock(&enic->devcmd_lock);
2227         err = enic_set_rss_cpu(enic,
2228                 rss_cpu_buf_pa,
2229                 sizeof(union vnic_rss_cpu));
2230         spin_unlock(&enic->devcmd_lock);
2231
2232         pci_free_consistent(enic->pdev, sizeof(union vnic_rss_cpu),
2233                 rss_cpu_buf_va, rss_cpu_buf_pa);
2234
2235         return err;
2236 }
2237
2238 static int enic_set_niccfg(struct enic *enic, u8 rss_default_cpu,
2239         u8 rss_hash_type, u8 rss_hash_bits, u8 rss_base_cpu, u8 rss_enable)
2240 {
2241         const u8 tso_ipid_split_en = 0;
2242         const u8 ig_vlan_strip_en = 1;
2243         int err;
2244
2245         /* Enable VLAN tag stripping.
2246         */
2247
2248         spin_lock(&enic->devcmd_lock);
2249         err = enic_set_nic_cfg(enic,
2250                 rss_default_cpu, rss_hash_type,
2251                 rss_hash_bits, rss_base_cpu,
2252                 rss_enable, tso_ipid_split_en,
2253                 ig_vlan_strip_en);
2254         spin_unlock(&enic->devcmd_lock);
2255
2256         return err;
2257 }
2258
2259 static int enic_set_rss_nic_cfg(struct enic *enic)
2260 {
2261         struct device *dev = enic_get_dev(enic);
2262         const u8 rss_default_cpu = 0;
2263         const u8 rss_hash_type = NIC_CFG_RSS_HASH_TYPE_IPV4 |
2264                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV4 |
2265                 NIC_CFG_RSS_HASH_TYPE_IPV6 |
2266                 NIC_CFG_RSS_HASH_TYPE_TCP_IPV6;
2267         const u8 rss_hash_bits = 7;
2268         const u8 rss_base_cpu = 0;
2269         u8 rss_enable = ENIC_SETTING(enic, RSS) && (enic->rq_count > 1);
2270
2271         if (rss_enable) {
2272                 if (!enic_set_rsskey(enic)) {
2273                         if (enic_set_rsscpu(enic, rss_hash_bits)) {
2274                                 rss_enable = 0;
2275                                 dev_warn(dev, "RSS disabled, "
2276                                         "Failed to set RSS cpu indirection table.");
2277                         }
2278                 } else {
2279                         rss_enable = 0;
2280                         dev_warn(dev, "RSS disabled, Failed to set RSS key.\n");
2281                 }
2282         }
2283
2284         return enic_set_niccfg(enic, rss_default_cpu, rss_hash_type,
2285                 rss_hash_bits, rss_base_cpu, rss_enable);
2286 }
2287
2288 static int enic_dev_hang_notify(struct enic *enic)
2289 {
2290         int err;
2291
2292         spin_lock(&enic->devcmd_lock);
2293         err = vnic_dev_hang_notify(enic->vdev);
2294         spin_unlock(&enic->devcmd_lock);
2295
2296         return err;
2297 }
2298
2299 static int enic_dev_set_ig_vlan_rewrite_mode(struct enic *enic)
2300 {
2301         int err;
2302
2303         spin_lock(&enic->devcmd_lock);
2304         err = vnic_dev_set_ig_vlan_rewrite_mode(enic->vdev,
2305                 IG_VLAN_REWRITE_MODE_PRIORITY_TAG_DEFAULT_VLAN);
2306         spin_unlock(&enic->devcmd_lock);
2307
2308         return err;
2309 }
2310
2311 static void enic_reset(struct work_struct *work)
2312 {
2313         struct enic *enic = container_of(work, struct enic, reset);
2314
2315         if (!netif_running(enic->netdev))
2316                 return;
2317
2318         rtnl_lock();
2319
2320         enic_dev_hang_notify(enic);
2321         enic_stop(enic->netdev);
2322         enic_dev_hang_reset(enic);
2323         enic_reset_multicast_list(enic);
2324         enic_init_vnic_resources(enic);
2325         enic_set_rss_nic_cfg(enic);
2326         enic_dev_set_ig_vlan_rewrite_mode(enic);
2327         enic_open(enic->netdev);
2328
2329         rtnl_unlock();
2330 }
2331
2332 static int enic_set_intr_mode(struct enic *enic)
2333 {
2334         unsigned int n = min_t(unsigned int, enic->rq_count, ENIC_RQ_MAX);
2335         unsigned int m = 1;
2336         unsigned int i;
2337
2338         /* Set interrupt mode (INTx, MSI, MSI-X) depending
2339          * on system capabilities.
2340          *
2341          * Try MSI-X first
2342          *
2343          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
2344          * (the second to last INTR is used for WQ/RQ errors)
2345          * (the last INTR is used for notifications)
2346          */
2347
2348         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
2349         for (i = 0; i < n + m + 2; i++)
2350                 enic->msix_entry[i].entry = i;
2351
2352         /* Use multiple RQs if RSS is enabled
2353          */
2354
2355         if (ENIC_SETTING(enic, RSS) &&
2356             enic->config.intr_mode < 1 &&
2357             enic->rq_count >= n &&
2358             enic->wq_count >= m &&
2359             enic->cq_count >= n + m &&
2360             enic->intr_count >= n + m + 2) {
2361
2362                 if (!pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
2363
2364                         enic->rq_count = n;
2365                         enic->wq_count = m;
2366                         enic->cq_count = n + m;
2367                         enic->intr_count = n + m + 2;
2368
2369                         vnic_dev_set_intr_mode(enic->vdev,
2370                                 VNIC_DEV_INTR_MODE_MSIX);
2371
2372                         return 0;
2373                 }
2374         }
2375
2376         if (enic->config.intr_mode < 1 &&
2377             enic->rq_count >= 1 &&
2378             enic->wq_count >= m &&
2379             enic->cq_count >= 1 + m &&
2380             enic->intr_count >= 1 + m + 2) {
2381                 if (!pci_enable_msix(enic->pdev, enic->msix_entry, 1 + m + 2)) {
2382
2383                         enic->rq_count = 1;
2384                         enic->wq_count = m;
2385                         enic->cq_count = 1 + m;
2386                         enic->intr_count = 1 + m + 2;
2387
2388                         vnic_dev_set_intr_mode(enic->vdev,
2389                                 VNIC_DEV_INTR_MODE_MSIX);
2390
2391                         return 0;
2392                 }
2393         }
2394
2395         /* Next try MSI
2396          *
2397          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2398          */
2399
2400         if (enic->config.intr_mode < 2 &&
2401             enic->rq_count >= 1 &&
2402             enic->wq_count >= 1 &&
2403             enic->cq_count >= 2 &&
2404             enic->intr_count >= 1 &&
2405             !pci_enable_msi(enic->pdev)) {
2406
2407                 enic->rq_count = 1;
2408                 enic->wq_count = 1;
2409                 enic->cq_count = 2;
2410                 enic->intr_count = 1;
2411
2412                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2413
2414                 return 0;
2415         }
2416
2417         /* Next try INTx
2418          *
2419          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2420          * (the first INTR is used for WQ/RQ)
2421          * (the second INTR is used for WQ/RQ errors)
2422          * (the last INTR is used for notifications)
2423          */
2424
2425         if (enic->config.intr_mode < 3 &&
2426             enic->rq_count >= 1 &&
2427             enic->wq_count >= 1 &&
2428             enic->cq_count >= 2 &&
2429             enic->intr_count >= 3) {
2430
2431                 enic->rq_count = 1;
2432                 enic->wq_count = 1;
2433                 enic->cq_count = 2;
2434                 enic->intr_count = 3;
2435
2436                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2437
2438                 return 0;
2439         }
2440
2441         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2442
2443         return -EINVAL;
2444 }
2445
2446 static void enic_clear_intr_mode(struct enic *enic)
2447 {
2448         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2449         case VNIC_DEV_INTR_MODE_MSIX:
2450                 pci_disable_msix(enic->pdev);
2451                 break;
2452         case VNIC_DEV_INTR_MODE_MSI:
2453                 pci_disable_msi(enic->pdev);
2454                 break;
2455         default:
2456                 break;
2457         }
2458
2459         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2460 }
2461
2462 static const struct net_device_ops enic_netdev_dynamic_ops = {
2463         .ndo_open               = enic_open,
2464         .ndo_stop               = enic_stop,
2465         .ndo_start_xmit         = enic_hard_start_xmit,
2466         .ndo_get_stats          = enic_get_stats,
2467         .ndo_validate_addr      = eth_validate_addr,
2468         .ndo_set_rx_mode        = enic_set_rx_mode,
2469         .ndo_set_multicast_list = enic_set_rx_mode,
2470         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2471         .ndo_change_mtu         = enic_change_mtu,
2472         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2473         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2474         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2475         .ndo_tx_timeout         = enic_tx_timeout,
2476         .ndo_set_vf_port        = enic_set_vf_port,
2477         .ndo_get_vf_port        = enic_get_vf_port,
2478 #ifdef IFLA_VF_MAX
2479         .ndo_set_vf_mac         = enic_set_vf_mac,
2480 #endif
2481 #ifdef CONFIG_NET_POLL_CONTROLLER
2482         .ndo_poll_controller    = enic_poll_controller,
2483 #endif
2484 };
2485
2486 static const struct net_device_ops enic_netdev_ops = {
2487         .ndo_open               = enic_open,
2488         .ndo_stop               = enic_stop,
2489         .ndo_start_xmit         = enic_hard_start_xmit,
2490         .ndo_get_stats          = enic_get_stats,
2491         .ndo_validate_addr      = eth_validate_addr,
2492         .ndo_set_mac_address    = enic_set_mac_address,
2493         .ndo_set_rx_mode        = enic_set_rx_mode,
2494         .ndo_set_multicast_list = enic_set_rx_mode,
2495         .ndo_change_mtu         = enic_change_mtu,
2496         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2497         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2498         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2499         .ndo_tx_timeout         = enic_tx_timeout,
2500 #ifdef CONFIG_NET_POLL_CONTROLLER
2501         .ndo_poll_controller    = enic_poll_controller,
2502 #endif
2503 };
2504
2505 static void enic_dev_deinit(struct enic *enic)
2506 {
2507         unsigned int i;
2508
2509         for (i = 0; i < enic->rq_count; i++)
2510                 netif_napi_del(&enic->napi[i]);
2511
2512         enic_free_vnic_resources(enic);
2513         enic_clear_intr_mode(enic);
2514 }
2515
2516 static int enic_dev_init(struct enic *enic)
2517 {
2518         struct device *dev = enic_get_dev(enic);
2519         struct net_device *netdev = enic->netdev;
2520         unsigned int i;
2521         int err;
2522
2523         /* Get vNIC configuration
2524          */
2525
2526         err = enic_get_vnic_config(enic);
2527         if (err) {
2528                 dev_err(dev, "Get vNIC configuration failed, aborting\n");
2529                 return err;
2530         }
2531
2532         /* Get available resource counts
2533          */
2534
2535         enic_get_res_counts(enic);
2536
2537         /* Set interrupt mode based on resource counts and system
2538          * capabilities
2539          */
2540
2541         err = enic_set_intr_mode(enic);
2542         if (err) {
2543                 dev_err(dev, "Failed to set intr mode based on resource "
2544                         "counts and system capabilities, aborting\n");
2545                 return err;
2546         }
2547
2548         /* Allocate and configure vNIC resources
2549          */
2550
2551         err = enic_alloc_vnic_resources(enic);
2552         if (err) {
2553                 dev_err(dev, "Failed to alloc vNIC resources, aborting\n");
2554                 goto err_out_free_vnic_resources;
2555         }
2556
2557         enic_init_vnic_resources(enic);
2558
2559         err = enic_set_rq_alloc_buf(enic);
2560         if (err) {
2561                 dev_err(dev, "Failed to set RQ buffer allocator, aborting\n");
2562                 goto err_out_free_vnic_resources;
2563         }
2564
2565         err = enic_set_rss_nic_cfg(enic);
2566         if (err) {
2567                 dev_err(dev, "Failed to config nic, aborting\n");
2568                 goto err_out_free_vnic_resources;
2569         }
2570
2571         err = enic_dev_set_ig_vlan_rewrite_mode(enic);
2572         if (err) {
2573                 dev_err(dev,
2574                         "Failed to set ingress vlan rewrite mode, aborting.\n");
2575                 goto err_out_free_vnic_resources;
2576         }
2577
2578         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2579         default:
2580                 netif_napi_add(netdev, &enic->napi[0], enic_poll, 64);
2581                 break;
2582         case VNIC_DEV_INTR_MODE_MSIX:
2583                 for (i = 0; i < enic->rq_count; i++)
2584                         netif_napi_add(netdev, &enic->napi[i],
2585                                 enic_poll_msix, 64);
2586                 break;
2587         }
2588
2589         return 0;
2590
2591 err_out_free_vnic_resources:
2592         enic_clear_intr_mode(enic);
2593         enic_free_vnic_resources(enic);
2594
2595         return err;
2596 }
2597
2598 static void enic_iounmap(struct enic *enic)
2599 {
2600         unsigned int i;
2601
2602         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2603                 if (enic->bar[i].vaddr)
2604                         iounmap(enic->bar[i].vaddr);
2605 }
2606
2607 static int __devinit enic_probe(struct pci_dev *pdev,
2608         const struct pci_device_id *ent)
2609 {
2610         struct device *dev = &pdev->dev;
2611         struct net_device *netdev;
2612         struct enic *enic;
2613         int using_dac = 0;
2614         unsigned int i;
2615         int err;
2616
2617         /* Allocate net device structure and initialize.  Private
2618          * instance data is initialized to zero.
2619          */
2620
2621         netdev = alloc_etherdev(sizeof(struct enic));
2622         if (!netdev) {
2623                 pr_err("Etherdev alloc failed, aborting\n");
2624                 return -ENOMEM;
2625         }
2626
2627         pci_set_drvdata(pdev, netdev);
2628
2629         SET_NETDEV_DEV(netdev, &pdev->dev);
2630
2631         enic = netdev_priv(netdev);
2632         enic->netdev = netdev;
2633         enic->pdev = pdev;
2634
2635         /* Setup PCI resources
2636          */
2637
2638         err = pci_enable_device_mem(pdev);
2639         if (err) {
2640                 dev_err(dev, "Cannot enable PCI device, aborting\n");
2641                 goto err_out_free_netdev;
2642         }
2643
2644         err = pci_request_regions(pdev, DRV_NAME);
2645         if (err) {
2646                 dev_err(dev, "Cannot request PCI regions, aborting\n");
2647                 goto err_out_disable_device;
2648         }
2649
2650         pci_set_master(pdev);
2651
2652         /* Query PCI controller on system for DMA addressing
2653          * limitation for the device.  Try 40-bit first, and
2654          * fail to 32-bit.
2655          */
2656
2657         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2658         if (err) {
2659                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2660                 if (err) {
2661                         dev_err(dev, "No usable DMA configuration, aborting\n");
2662                         goto err_out_release_regions;
2663                 }
2664                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2665                 if (err) {
2666                         dev_err(dev, "Unable to obtain %u-bit DMA "
2667                                 "for consistent allocations, aborting\n", 32);
2668                         goto err_out_release_regions;
2669                 }
2670         } else {
2671                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2672                 if (err) {
2673                         dev_err(dev, "Unable to obtain %u-bit DMA "
2674                                 "for consistent allocations, aborting\n", 40);
2675                         goto err_out_release_regions;
2676                 }
2677                 using_dac = 1;
2678         }
2679
2680         /* Map vNIC resources from BAR0-5
2681          */
2682
2683         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2684                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2685                         continue;
2686                 enic->bar[i].len = pci_resource_len(pdev, i);
2687                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2688                 if (!enic->bar[i].vaddr) {
2689                         dev_err(dev, "Cannot memory-map BAR %d, aborting\n", i);
2690                         err = -ENODEV;
2691                         goto err_out_iounmap;
2692                 }
2693                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2694         }
2695
2696         /* Register vNIC device
2697          */
2698
2699         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2700                 ARRAY_SIZE(enic->bar));
2701         if (!enic->vdev) {
2702                 dev_err(dev, "vNIC registration failed, aborting\n");
2703                 err = -ENODEV;
2704                 goto err_out_iounmap;
2705         }
2706
2707         /* Issue device open to get device in known state
2708          */
2709
2710         err = enic_dev_open(enic);
2711         if (err) {
2712                 dev_err(dev, "vNIC dev open failed, aborting\n");
2713                 goto err_out_vnic_unregister;
2714         }
2715
2716         /* Issue device init to initialize the vnic-to-switch link.
2717          * We'll start with carrier off and wait for link UP
2718          * notification later to turn on carrier.  We don't need
2719          * to wait here for the vnic-to-switch link initialization
2720          * to complete; link UP notification is the indication that
2721          * the process is complete.
2722          */
2723
2724         netif_carrier_off(netdev);
2725
2726         /* Do not call dev_init for a dynamic vnic.
2727          * For a dynamic vnic, init_prov_info will be
2728          * called later by an upper layer.
2729          */
2730
2731         if (!enic_is_dynamic(enic)) {
2732                 err = vnic_dev_init(enic->vdev, 0);
2733                 if (err) {
2734                         dev_err(dev, "vNIC dev init failed, aborting\n");
2735                         goto err_out_dev_close;
2736                 }
2737         }
2738
2739         /* Setup devcmd lock
2740          */
2741
2742         spin_lock_init(&enic->devcmd_lock);
2743
2744         err = enic_dev_init(enic);
2745         if (err) {
2746                 dev_err(dev, "Device initialization failed, aborting\n");
2747                 goto err_out_dev_close;
2748         }
2749
2750         /* Setup notification timer, HW reset task, and wq locks
2751          */
2752
2753         init_timer(&enic->notify_timer);
2754         enic->notify_timer.function = enic_notify_timer;
2755         enic->notify_timer.data = (unsigned long)enic;
2756
2757         INIT_WORK(&enic->reset, enic_reset);
2758
2759         for (i = 0; i < enic->wq_count; i++)
2760                 spin_lock_init(&enic->wq_lock[i]);
2761
2762         /* Register net device
2763          */
2764
2765         enic->port_mtu = enic->config.mtu;
2766         (void)enic_change_mtu(netdev, enic->port_mtu);
2767
2768         err = enic_set_mac_addr(netdev, enic->mac_addr);
2769         if (err) {
2770                 dev_err(dev, "Invalid MAC address, aborting\n");
2771                 goto err_out_dev_deinit;
2772         }
2773
2774         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2775         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2776
2777         if (enic_is_dynamic(enic))
2778                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2779         else
2780                 netdev->netdev_ops = &enic_netdev_ops;
2781
2782         netdev->watchdog_timeo = 2 * HZ;
2783         netdev->ethtool_ops = &enic_ethtool_ops;
2784
2785         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2786         if (ENIC_SETTING(enic, LOOP)) {
2787                 netdev->features &= ~NETIF_F_HW_VLAN_TX;
2788                 enic->loop_enable = 1;
2789                 enic->loop_tag = enic->config.loop_tag;
2790                 dev_info(dev, "loopback tag=0x%04x\n", enic->loop_tag);
2791         }
2792         if (ENIC_SETTING(enic, TXCSUM))
2793                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2794         if (ENIC_SETTING(enic, TSO))
2795                 netdev->features |= NETIF_F_TSO |
2796                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2797         if (ENIC_SETTING(enic, LRO))
2798                 netdev->features |= NETIF_F_GRO;
2799         if (using_dac)
2800                 netdev->features |= NETIF_F_HIGHDMA;
2801
2802         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2803
2804         err = register_netdev(netdev);
2805         if (err) {
2806                 dev_err(dev, "Cannot register net device, aborting\n");
2807                 goto err_out_dev_deinit;
2808         }
2809
2810         return 0;
2811
2812 err_out_dev_deinit:
2813         enic_dev_deinit(enic);
2814 err_out_dev_close:
2815         vnic_dev_close(enic->vdev);
2816 err_out_vnic_unregister:
2817         vnic_dev_unregister(enic->vdev);
2818 err_out_iounmap:
2819         enic_iounmap(enic);
2820 err_out_release_regions:
2821         pci_release_regions(pdev);
2822 err_out_disable_device:
2823         pci_disable_device(pdev);
2824 err_out_free_netdev:
2825         pci_set_drvdata(pdev, NULL);
2826         free_netdev(netdev);
2827
2828         return err;
2829 }
2830
2831 static void __devexit enic_remove(struct pci_dev *pdev)
2832 {
2833         struct net_device *netdev = pci_get_drvdata(pdev);
2834
2835         if (netdev) {
2836                 struct enic *enic = netdev_priv(netdev);
2837
2838                 cancel_work_sync(&enic->reset);
2839                 unregister_netdev(netdev);
2840                 enic_dev_deinit(enic);
2841                 vnic_dev_close(enic->vdev);
2842                 vnic_dev_unregister(enic->vdev);
2843                 enic_iounmap(enic);
2844                 pci_release_regions(pdev);
2845                 pci_disable_device(pdev);
2846                 pci_set_drvdata(pdev, NULL);
2847                 free_netdev(netdev);
2848         }
2849 }
2850
2851 static struct pci_driver enic_driver = {
2852         .name = DRV_NAME,
2853         .id_table = enic_id_table,
2854         .probe = enic_probe,
2855         .remove = __devexit_p(enic_remove),
2856 };
2857
2858 static int __init enic_init_module(void)
2859 {
2860         pr_info("%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2861
2862         return pci_register_driver(&enic_driver);
2863 }
2864
2865 static void __exit enic_cleanup_module(void)
2866 {
2867         pci_unregister_driver(&enic_driver);
2868 }
2869
2870 module_init(enic_init_module);
2871 module_exit(enic_cleanup_module);