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