drivers/net/enic/enic_main.c

   1 /*
   2  * Copyright 2008 Cisco Systems, Inc.  All rights reserved.
   3  * Copyright 2007 Nuova Systems, Inc.  All rights reserved.
   4  *
   5  * This program is free software; you may redistribute it and/or modify
   6  * it under the terms of the GNU General Public License as published by
   7  * the Free Software Foundation; version 2 of the License.
   8  *
   9  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  10  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  11  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  12  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  13  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  14  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  15  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  16  * SOFTWARE.
  17  *
  18  */
  19
  20 #include <linux/module.h>
  21 #include <linux/kernel.h>
  22 #include <linux/string.h>
  23 #include <linux/errno.h>
  24 #include <linux/types.h>
  25 #include <linux/init.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/pci.h>
  28 #include <linux/netdevice.h>
  29 #include <linux/etherdevice.h>
  30 #include <linux/if_ether.h>
  31 #include <linux/if_vlan.h>
  32 #include <linux/if_link.h>
  33 #include <linux/ethtool.h>
  34 #include <linux/in.h>
  35 #include <linux/ip.h>
  36 #include <linux/ipv6.h>
  37 #include <linux/tcp.h>
  38 #include <net/ip6_checksum.h>
  39
  40 #include "cq_enet_desc.h"
  41 #include "vnic_dev.h"
  42 #include "vnic_intr.h"
  43 #include "vnic_stats.h"
  44 #include "vnic_vic.h"
  45 #include "enic_res.h"
  46 #include "enic.h"
  47
  48 #define ENIC_NOTIFY_TIMER_PERIOD        (2 * HZ)
  49 #define WQ_ENET_MAX_DESC_LEN            (1 << WQ_ENET_LEN_BITS)
  50 #define MAX_TSO                         (1 << 16)
  51 #define ENIC_DESC_MAX_SPLITS            (MAX_TSO / WQ_ENET_MAX_DESC_LEN + 1)
  52
  53 #define PCI_DEVICE_ID_CISCO_VIC_ENET         0x0043  /* ethernet vnic */
  54 #define PCI_DEVICE_ID_CISCO_VIC_ENET_DYN     0x0044  /* enet dynamic vnic */
  55
  56 /* Supported devices */
  57 static DEFINE_PCI_DEVICE_TABLE(enic_id_table) = {
  58         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET) },
  59         { PCI_VDEVICE(CISCO, PCI_DEVICE_ID_CISCO_VIC_ENET_DYN) },
  60         { 0, }  /* end of table */
  61 };
  62
  63 MODULE_DESCRIPTION(DRV_DESCRIPTION);
  64 MODULE_AUTHOR("Scott Feldman <scofeldm@cisco.com>");
  65 MODULE_LICENSE("GPL");
  66 MODULE_VERSION(DRV_VERSION);
  67 MODULE_DEVICE_TABLE(pci, enic_id_table);
  68
  69 struct enic_stat {
  70         char name[ETH_GSTRING_LEN];
  71         unsigned int offset;
  72 };
  73
  74 #define ENIC_TX_STAT(stat)      \
  75         { .name = #stat, .offset = offsetof(struct vnic_tx_stats, stat) / 8 }
  76 #define ENIC_RX_STAT(stat)      \
  77         { .name = #stat, .offset = offsetof(struct vnic_rx_stats, stat) / 8 }
  78
  79 static const struct enic_stat enic_tx_stats[] = {
  80         ENIC_TX_STAT(tx_frames_ok),
  81         ENIC_TX_STAT(tx_unicast_frames_ok),
  82         ENIC_TX_STAT(tx_multicast_frames_ok),
  83         ENIC_TX_STAT(tx_broadcast_frames_ok),
  84         ENIC_TX_STAT(tx_bytes_ok),
  85         ENIC_TX_STAT(tx_unicast_bytes_ok),
  86         ENIC_TX_STAT(tx_multicast_bytes_ok),
  87         ENIC_TX_STAT(tx_broadcast_bytes_ok),
  88         ENIC_TX_STAT(tx_drops),
  89         ENIC_TX_STAT(tx_errors),
  90         ENIC_TX_STAT(tx_tso),
  91 };
  92
  93 static const struct enic_stat enic_rx_stats[] = {
  94         ENIC_RX_STAT(rx_frames_ok),
  95         ENIC_RX_STAT(rx_frames_total),
  96         ENIC_RX_STAT(rx_unicast_frames_ok),
  97         ENIC_RX_STAT(rx_multicast_frames_ok),
  98         ENIC_RX_STAT(rx_broadcast_frames_ok),
  99         ENIC_RX_STAT(rx_bytes_ok),
 100         ENIC_RX_STAT(rx_unicast_bytes_ok),
 101         ENIC_RX_STAT(rx_multicast_bytes_ok),
 102         ENIC_RX_STAT(rx_broadcast_bytes_ok),
 103         ENIC_RX_STAT(rx_drop),
 104         ENIC_RX_STAT(rx_no_bufs),
 105         ENIC_RX_STAT(rx_errors),
 106         ENIC_RX_STAT(rx_rss),
 107         ENIC_RX_STAT(rx_crc_errors),
 108         ENIC_RX_STAT(rx_frames_64),
 109         ENIC_RX_STAT(rx_frames_127),
 110         ENIC_RX_STAT(rx_frames_255),
 111         ENIC_RX_STAT(rx_frames_511),
 112         ENIC_RX_STAT(rx_frames_1023),
 113         ENIC_RX_STAT(rx_frames_1518),
 114         ENIC_RX_STAT(rx_frames_to_max),
 115 };
 116
 117 static const unsigned int enic_n_tx_stats = ARRAY_SIZE(enic_tx_stats);
 118 static const unsigned int enic_n_rx_stats = ARRAY_SIZE(enic_rx_stats);
 119
 120 static int enic_is_dynamic(struct enic *enic)
 121 {
 122         return enic->pdev->device == PCI_DEVICE_ID_CISCO_VIC_ENET_DYN;
 123 }
 124
 125 static int enic_get_settings(struct net_device *netdev,
 126         struct ethtool_cmd *ecmd)
 127 {
 128         struct enic *enic = netdev_priv(netdev);
 129
 130         ecmd->supported = (SUPPORTED_10000baseT_Full | SUPPORTED_FIBRE);
 131         ecmd->advertising = (ADVERTISED_10000baseT_Full | ADVERTISED_FIBRE);
 132         ecmd->port = PORT_FIBRE;
 133         ecmd->transceiver = XCVR_EXTERNAL;
 134
 135         if (netif_carrier_ok(netdev)) {
 136                 ecmd->speed = vnic_dev_port_speed(enic->vdev);
 137                 ecmd->duplex = DUPLEX_FULL;
 138         } else {
 139                 ecmd->speed = -1;
 140                 ecmd->duplex = -1;
 141         }
 142
 143         ecmd->autoneg = AUTONEG_DISABLE;
 144
 145         return 0;
 146 }
 147
 148 static void enic_get_drvinfo(struct net_device *netdev,
 149         struct ethtool_drvinfo *drvinfo)
 150 {
 151         struct enic *enic = netdev_priv(netdev);
 152         struct vnic_devcmd_fw_info *fw_info;
 153
 154         spin_lock(&enic->devcmd_lock);
 155         vnic_dev_fw_info(enic->vdev, &fw_info);
 156         spin_unlock(&enic->devcmd_lock);
 157
 158         strncpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
 159         strncpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
 160         strncpy(drvinfo->fw_version, fw_info->fw_version,
 161                 sizeof(drvinfo->fw_version));
 162         strncpy(drvinfo->bus_info, pci_name(enic->pdev),
 163                 sizeof(drvinfo->bus_info));
 164 }
 165
 166 static void enic_get_strings(struct net_device *netdev, u32 stringset, u8 *data)
 167 {
 168         unsigned int i;
 169
 170         switch (stringset) {
 171         case ETH_SS_STATS:
 172                 for (i = 0; i < enic_n_tx_stats; i++) {
 173                         memcpy(data, enic_tx_stats[i].name, ETH_GSTRING_LEN);
 174                         data += ETH_GSTRING_LEN;
 175                 }
 176                 for (i = 0; i < enic_n_rx_stats; i++) {
 177                         memcpy(data, enic_rx_stats[i].name, ETH_GSTRING_LEN);
 178                         data += ETH_GSTRING_LEN;
 179                 }
 180                 break;
 181         }
 182 }
 183
 184 static int enic_get_sset_count(struct net_device *netdev, int sset)
 185 {
 186         switch (sset) {
 187         case ETH_SS_STATS:
 188                 return enic_n_tx_stats + enic_n_rx_stats;
 189         default:
 190                 return -EOPNOTSUPP;
 191         }
 192 }
 193
 194 static void enic_get_ethtool_stats(struct net_device *netdev,
 195         struct ethtool_stats *stats, u64 *data)
 196 {
 197         struct enic *enic = netdev_priv(netdev);
 198         struct vnic_stats *vstats;
 199         unsigned int i;
 200
 201         spin_lock(&enic->devcmd_lock);
 202         vnic_dev_stats_dump(enic->vdev, &vstats);
 203         spin_unlock(&enic->devcmd_lock);
 204
 205         for (i = 0; i < enic_n_tx_stats; i++)
 206                 *(data++) = ((u64 *)&vstats->tx)[enic_tx_stats[i].offset];
 207         for (i = 0; i < enic_n_rx_stats; i++)
 208                 *(data++) = ((u64 *)&vstats->rx)[enic_rx_stats[i].offset];
 209 }
 210
 211 static u32 enic_get_rx_csum(struct net_device *netdev)
 212 {
 213         struct enic *enic = netdev_priv(netdev);
 214         return enic->csum_rx_enabled;
 215 }
 216
 217 static int enic_set_rx_csum(struct net_device *netdev, u32 data)
 218 {
 219         struct enic *enic = netdev_priv(netdev);
 220
 221         if (data && !ENIC_SETTING(enic, RXCSUM))
 222                 return -EINVAL;
 223
 224         enic->csum_rx_enabled = !!data;
 225
 226         return 0;
 227 }
 228
 229 static int enic_set_tx_csum(struct net_device *netdev, u32 data)
 230 {
 231         struct enic *enic = netdev_priv(netdev);
 232
 233         if (data && !ENIC_SETTING(enic, TXCSUM))
 234                 return -EINVAL;
 235
 236         if (data)
 237                 netdev->features |= NETIF_F_HW_CSUM;
 238         else
 239                 netdev->features &= ~NETIF_F_HW_CSUM;
 240
 241         return 0;
 242 }
 243
 244 static int enic_set_tso(struct net_device *netdev, u32 data)
 245 {
 246         struct enic *enic = netdev_priv(netdev);
 247
 248         if (data && !ENIC_SETTING(enic, TSO))
 249                 return -EINVAL;
 250
 251         if (data)
 252                 netdev->features |=
 253                         NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN;
 254         else
 255                 netdev->features &=
 256                         ~(NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN);
 257
 258         return 0;
 259 }
 260
 261 static u32 enic_get_msglevel(struct net_device *netdev)
 262 {
 263         struct enic *enic = netdev_priv(netdev);
 264         return enic->msg_enable;
 265 }
 266
 267 static void enic_set_msglevel(struct net_device *netdev, u32 value)
 268 {
 269         struct enic *enic = netdev_priv(netdev);
 270         enic->msg_enable = value;
 271 }
 272
 273 static int enic_get_coalesce(struct net_device *netdev,
 274         struct ethtool_coalesce *ecmd)
 275 {
 276         struct enic *enic = netdev_priv(netdev);
 277
 278         ecmd->tx_coalesce_usecs = enic->tx_coalesce_usecs;
 279         ecmd->rx_coalesce_usecs = enic->rx_coalesce_usecs;
 280
 281         return 0;
 282 }
 283
 284 static int enic_set_coalesce(struct net_device *netdev,
 285         struct ethtool_coalesce *ecmd)
 286 {
 287         struct enic *enic = netdev_priv(netdev);
 288         u32 tx_coalesce_usecs;
 289         u32 rx_coalesce_usecs;
 290
 291         tx_coalesce_usecs = min_t(u32,
 292                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 293                 ecmd->tx_coalesce_usecs);
 294         rx_coalesce_usecs = min_t(u32,
 295                 INTR_COALESCE_HW_TO_USEC(VNIC_INTR_TIMER_MAX),
 296                 ecmd->rx_coalesce_usecs);
 297
 298         switch (vnic_dev_get_intr_mode(enic->vdev)) {
 299         case VNIC_DEV_INTR_MODE_INTX:
 300                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 301                         return -EINVAL;
 302
 303                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_INTX_WQ_RQ],
 304                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 305                 break;
 306         case VNIC_DEV_INTR_MODE_MSI:
 307                 if (tx_coalesce_usecs != rx_coalesce_usecs)
 308                         return -EINVAL;
 309
 310                 vnic_intr_coalescing_timer_set(&enic->intr[0],
 311                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 312                 break;
 313         case VNIC_DEV_INTR_MODE_MSIX:
 314                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_WQ],
 315                         INTR_COALESCE_USEC_TO_HW(tx_coalesce_usecs));
 316                 vnic_intr_coalescing_timer_set(&enic->intr[ENIC_MSIX_RQ],
 317                         INTR_COALESCE_USEC_TO_HW(rx_coalesce_usecs));
 318                 break;
 319         default:
 320                 break;
 321         }
 322
 323         enic->tx_coalesce_usecs = tx_coalesce_usecs;
 324         enic->rx_coalesce_usecs = rx_coalesce_usecs;
 325
 326         return 0;
 327 }
 328
 329 static const struct ethtool_ops enic_ethtool_ops = {
 330         .get_settings = enic_get_settings,
 331         .get_drvinfo = enic_get_drvinfo,
 332         .get_msglevel = enic_get_msglevel,
 333         .set_msglevel = enic_set_msglevel,
 334         .get_link = ethtool_op_get_link,
 335         .get_strings = enic_get_strings,
 336         .get_sset_count = enic_get_sset_count,
 337         .get_ethtool_stats = enic_get_ethtool_stats,
 338         .get_rx_csum = enic_get_rx_csum,
 339         .set_rx_csum = enic_set_rx_csum,
 340         .get_tx_csum = ethtool_op_get_tx_csum,
 341         .set_tx_csum = enic_set_tx_csum,
 342         .get_sg = ethtool_op_get_sg,
 343         .set_sg = ethtool_op_set_sg,
 344         .get_tso = ethtool_op_get_tso,
 345         .set_tso = enic_set_tso,
 346         .get_coalesce = enic_get_coalesce,
 347         .set_coalesce = enic_set_coalesce,
 348         .get_flags = ethtool_op_get_flags,
 349         .set_flags = ethtool_op_set_flags,
 350 };
 351
 352 static void enic_free_wq_buf(struct vnic_wq *wq, struct vnic_wq_buf *buf)
 353 {
 354         struct enic *enic = vnic_dev_priv(wq->vdev);
 355
 356         if (buf->sop)
 357                 pci_unmap_single(enic->pdev, buf->dma_addr,
 358                         buf->len, PCI_DMA_TODEVICE);
 359         else
 360                 pci_unmap_page(enic->pdev, buf->dma_addr,
 361                         buf->len, PCI_DMA_TODEVICE);
 362
 363         if (buf->os_buf)
 364                 dev_kfree_skb_any(buf->os_buf);
 365 }
 366
 367 static void enic_wq_free_buf(struct vnic_wq *wq,
 368         struct cq_desc *cq_desc, struct vnic_wq_buf *buf, void *opaque)
 369 {
 370         enic_free_wq_buf(wq, buf);
 371 }
 372
 373 static int enic_wq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
 374         u8 type, u16 q_number, u16 completed_index, void *opaque)
 375 {
 376         struct enic *enic = vnic_dev_priv(vdev);
 377
 378         spin_lock(&enic->wq_lock[q_number]);
 379
 380         vnic_wq_service(&enic->wq[q_number], cq_desc,
 381                 completed_index, enic_wq_free_buf,
 382                 opaque);
 383
 384         if (netif_queue_stopped(enic->netdev) &&
 385             vnic_wq_desc_avail(&enic->wq[q_number]) >=
 386             (MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS))
 387                 netif_wake_queue(enic->netdev);
 388
 389         spin_unlock(&enic->wq_lock[q_number]);
 390
 391         return 0;
 392 }
 393
 394 static void enic_log_q_error(struct enic *enic)
 395 {
 396         unsigned int i;
 397         u32 error_status;
 398
 399         for (i = 0; i < enic->wq_count; i++) {
 400                 error_status = vnic_wq_error_status(&enic->wq[i]);
 401                 if (error_status)
 402                         printk(KERN_ERR PFX "%s: WQ[%d] error_status %d\n",
 403                                 enic->netdev->name, i, error_status);
 404         }
 405
 406         for (i = 0; i < enic->rq_count; i++) {
 407                 error_status = vnic_rq_error_status(&enic->rq[i]);
 408                 if (error_status)
 409                         printk(KERN_ERR PFX "%s: RQ[%d] error_status %d\n",
 410                                 enic->netdev->name, i, error_status);
 411         }
 412 }
 413
 414 static void enic_link_check(struct enic *enic)
 415 {
 416         int link_status = vnic_dev_link_status(enic->vdev);
 417         int carrier_ok = netif_carrier_ok(enic->netdev);
 418
 419         if (link_status && !carrier_ok) {
 420                 printk(KERN_INFO PFX "%s: Link UP\n", enic->netdev->name);
 421                 netif_carrier_on(enic->netdev);
 422         } else if (!link_status && carrier_ok) {
 423                 printk(KERN_INFO PFX "%s: Link DOWN\n", enic->netdev->name);
 424                 netif_carrier_off(enic->netdev);
 425         }
 426 }
 427
 428 static void enic_mtu_check(struct enic *enic)
 429 {
 430         u32 mtu = vnic_dev_mtu(enic->vdev);
 431
 432         if (mtu && mtu != enic->port_mtu) {
 433                 enic->port_mtu = mtu;
 434                 if (mtu < enic->netdev->mtu)
 435                         printk(KERN_WARNING PFX
 436                                 "%s: interface MTU (%d) set higher "
 437                                 "than switch port MTU (%d)\n",
 438                                 enic->netdev->name, enic->netdev->mtu, mtu);
 439         }
 440 }
 441
 442 static void enic_msglvl_check(struct enic *enic)
 443 {
 444         u32 msg_enable = vnic_dev_msg_lvl(enic->vdev);
 445
 446         if (msg_enable != enic->msg_enable) {
 447                 printk(KERN_INFO PFX "%s: msg lvl changed from 0x%x to 0x%x\n",
 448                         enic->netdev->name, enic->msg_enable, msg_enable);
 449                 enic->msg_enable = msg_enable;
 450         }
 451 }
 452
 453 static void enic_notify_check(struct enic *enic)
 454 {
 455         enic_msglvl_check(enic);
 456         enic_mtu_check(enic);
 457         enic_link_check(enic);
 458 }
 459
 460 #define ENIC_TEST_INTR(pba, i) (pba & (1 << i))
 461
 462 static irqreturn_t enic_isr_legacy(int irq, void *data)
 463 {
 464         struct net_device *netdev = data;
 465         struct enic *enic = netdev_priv(netdev);
 466         u32 pba;
 467
 468         vnic_intr_mask(&enic->intr[ENIC_INTX_WQ_RQ]);
 469
 470         pba = vnic_intr_legacy_pba(enic->legacy_pba);
 471         if (!pba) {
 472                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 473                 return IRQ_NONE;        /* not our interrupt */
 474         }
 475
 476         if (ENIC_TEST_INTR(pba, ENIC_INTX_NOTIFY)) {
 477                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_NOTIFY]);
 478                 enic_notify_check(enic);
 479         }
 480
 481         if (ENIC_TEST_INTR(pba, ENIC_INTX_ERR)) {
 482                 vnic_intr_return_all_credits(&enic->intr[ENIC_INTX_ERR]);
 483                 enic_log_q_error(enic);
 484                 /* schedule recovery from WQ/RQ error */
 485                 schedule_work(&enic->reset);
 486                 return IRQ_HANDLED;
 487         }
 488
 489         if (ENIC_TEST_INTR(pba, ENIC_INTX_WQ_RQ)) {
 490                 if (napi_schedule_prep(&enic->napi))
 491                         __napi_schedule(&enic->napi);
 492         } else {
 493                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
 494         }
 495
 496         return IRQ_HANDLED;
 497 }
 498
 499 static irqreturn_t enic_isr_msi(int irq, void *data)
 500 {
 501         struct enic *enic = data;
 502
 503         /* With MSI, there is no sharing of interrupts, so this is
 504          * our interrupt and there is no need to ack it.  The device
 505          * is not providing per-vector masking, so the OS will not
 506          * write to PCI config space to mask/unmask the interrupt.
 507          * We're using mask_on_assertion for MSI, so the device
 508          * automatically masks the interrupt when the interrupt is
 509          * generated.  Later, when exiting polling, the interrupt
 510          * will be unmasked (see enic_poll).
 511          *
 512          * Also, the device uses the same PCIe Traffic Class (TC)
 513          * for Memory Write data and MSI, so there are no ordering
 514          * issues; the MSI will always arrive at the Root Complex
 515          * _after_ corresponding Memory Writes (i.e. descriptor
 516          * writes).
 517          */
 518
 519         napi_schedule(&enic->napi);
 520
 521         return IRQ_HANDLED;
 522 }
 523
 524 static irqreturn_t enic_isr_msix_rq(int irq, void *data)
 525 {
 526         struct enic *enic = data;
 527
 528         /* schedule NAPI polling for RQ cleanup */
 529         napi_schedule(&enic->napi);
 530
 531         return IRQ_HANDLED;
 532 }
 533
 534 static irqreturn_t enic_isr_msix_wq(int irq, void *data)
 535 {
 536         struct enic *enic = data;
 537         unsigned int wq_work_to_do = -1; /* no limit */
 538         unsigned int wq_work_done;
 539
 540         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
 541                 wq_work_to_do, enic_wq_service, NULL);
 542
 543         vnic_intr_return_credits(&enic->intr[ENIC_MSIX_WQ],
 544                 wq_work_done,
 545                 1 /* unmask intr */,
 546                 1 /* reset intr timer */);
 547
 548         return IRQ_HANDLED;
 549 }
 550
 551 static irqreturn_t enic_isr_msix_err(int irq, void *data)
 552 {
 553         struct enic *enic = data;
 554
 555         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_ERR]);
 556
 557         enic_log_q_error(enic);
 558
 559         /* schedule recovery from WQ/RQ error */
 560         schedule_work(&enic->reset);
 561
 562         return IRQ_HANDLED;
 563 }
 564
 565 static irqreturn_t enic_isr_msix_notify(int irq, void *data)
 566 {
 567         struct enic *enic = data;
 568
 569         vnic_intr_return_all_credits(&enic->intr[ENIC_MSIX_NOTIFY]);
 570         enic_notify_check(enic);
 571
 572         return IRQ_HANDLED;
 573 }
 574
 575 static inline void enic_queue_wq_skb_cont(struct enic *enic,
 576         struct vnic_wq *wq, struct sk_buff *skb,
 577         unsigned int len_left)
 578 {
 579         skb_frag_t *frag;
 580
 581         /* Queue additional data fragments */
 582         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 583                 len_left -= frag->size;
 584                 enic_queue_wq_desc_cont(wq, skb,
 585                         pci_map_page(enic->pdev, frag->page,
 586                                 frag->page_offset, frag->size,
 587                                 PCI_DMA_TODEVICE),
 588                         frag->size,
 589                         (len_left == 0));       /* EOP? */
 590         }
 591 }
 592
 593 static inline void enic_queue_wq_skb_vlan(struct enic *enic,
 594         struct vnic_wq *wq, struct sk_buff *skb,
 595         int vlan_tag_insert, unsigned int vlan_tag)
 596 {
 597         unsigned int head_len = skb_headlen(skb);
 598         unsigned int len_left = skb->len - head_len;
 599         int eop = (len_left == 0);
 600
 601         /* Queue the main skb fragment. The fragments are no larger
 602          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 603          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 604          * per fragment is queued.
 605          */
 606         enic_queue_wq_desc(wq, skb,
 607                 pci_map_single(enic->pdev, skb->data,
 608                         head_len, PCI_DMA_TODEVICE),
 609                 head_len,
 610                 vlan_tag_insert, vlan_tag,
 611                 eop);
 612
 613         if (!eop)
 614                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 615 }
 616
 617 static inline void enic_queue_wq_skb_csum_l4(struct enic *enic,
 618         struct vnic_wq *wq, struct sk_buff *skb,
 619         int vlan_tag_insert, unsigned int vlan_tag)
 620 {
 621         unsigned int head_len = skb_headlen(skb);
 622         unsigned int len_left = skb->len - head_len;
 623         unsigned int hdr_len = skb_transport_offset(skb);
 624         unsigned int csum_offset = hdr_len + skb->csum_offset;
 625         int eop = (len_left == 0);
 626
 627         /* Queue the main skb fragment. The fragments are no larger
 628          * than max MTU(9000)+ETH_HDR_LEN(14) bytes, which is less
 629          * than WQ_ENET_MAX_DESC_LEN length. So only one descriptor
 630          * per fragment is queued.
 631          */
 632         enic_queue_wq_desc_csum_l4(wq, skb,
 633                 pci_map_single(enic->pdev, skb->data,
 634                         head_len, PCI_DMA_TODEVICE),
 635                 head_len,
 636                 csum_offset,
 637                 hdr_len,
 638                 vlan_tag_insert, vlan_tag,
 639                 eop);
 640
 641         if (!eop)
 642                 enic_queue_wq_skb_cont(enic, wq, skb, len_left);
 643 }
 644
 645 static inline void enic_queue_wq_skb_tso(struct enic *enic,
 646         struct vnic_wq *wq, struct sk_buff *skb, unsigned int mss,
 647         int vlan_tag_insert, unsigned int vlan_tag)
 648 {
 649         unsigned int frag_len_left = skb_headlen(skb);
 650         unsigned int len_left = skb->len - frag_len_left;
 651         unsigned int hdr_len = skb_transport_offset(skb) + tcp_hdrlen(skb);
 652         int eop = (len_left == 0);
 653         unsigned int len;
 654         dma_addr_t dma_addr;
 655         unsigned int offset = 0;
 656         skb_frag_t *frag;
 657
 658         /* Preload TCP csum field with IP pseudo hdr calculated
 659          * with IP length set to zero.  HW will later add in length
 660          * to each TCP segment resulting from the TSO.
 661          */
 662
 663         if (skb->protocol == cpu_to_be16(ETH_P_IP)) {
 664                 ip_hdr(skb)->check = 0;
 665                 tcp_hdr(skb)->check = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
 666                         ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 667         } else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) {
 668                 tcp_hdr(skb)->check = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 669                         &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
 670         }
 671
 672         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 673          * for the main skb fragment
 674          */
 675         while (frag_len_left) {
 676                 len = min(frag_len_left, (unsigned int)WQ_ENET_MAX_DESC_LEN);
 677                 dma_addr = pci_map_single(enic->pdev, skb->data + offset,
 678                                 len, PCI_DMA_TODEVICE);
 679                 enic_queue_wq_desc_tso(wq, skb,
 680                         dma_addr,
 681                         len,
 682                         mss, hdr_len,
 683                         vlan_tag_insert, vlan_tag,
 684                         eop && (len == frag_len_left));
 685                 frag_len_left -= len;
 686                 offset += len;
 687         }
 688
 689         if (eop)
 690                 return;
 691
 692         /* Queue WQ_ENET_MAX_DESC_LEN length descriptors
 693          * for additional data fragments
 694          */
 695         for (frag = skb_shinfo(skb)->frags; len_left; frag++) {
 696                 len_left -= frag->size;
 697                 frag_len_left = frag->size;
 698                 offset = frag->page_offset;
 699
 700                 while (frag_len_left) {
 701                         len = min(frag_len_left,
 702                                 (unsigned int)WQ_ENET_MAX_DESC_LEN);
 703                         dma_addr = pci_map_page(enic->pdev, frag->page,
 704                                 offset, len,
 705                                 PCI_DMA_TODEVICE);
 706                         enic_queue_wq_desc_cont(wq, skb,
 707                                 dma_addr,
 708                                 len,
 709                                 (len_left == 0) &&
 710                                 (len == frag_len_left));        /* EOP? */
 711                         frag_len_left -= len;
 712                         offset += len;
 713                 }
 714         }
 715 }
 716
 717 static inline void enic_queue_wq_skb(struct enic *enic,
 718         struct vnic_wq *wq, struct sk_buff *skb)
 719 {
 720         unsigned int mss = skb_shinfo(skb)->gso_size;
 721         unsigned int vlan_tag = 0;
 722         int vlan_tag_insert = 0;
 723
 724         if (enic->vlan_group && vlan_tx_tag_present(skb)) {
 725                 /* VLAN tag from trunking driver */
 726                 vlan_tag_insert = 1;
 727                 vlan_tag = vlan_tx_tag_get(skb);
 728         }
 729
 730         if (mss)
 731                 enic_queue_wq_skb_tso(enic, wq, skb, mss,
 732                         vlan_tag_insert, vlan_tag);
 733         else if (skb->ip_summed == CHECKSUM_PARTIAL)
 734                 enic_queue_wq_skb_csum_l4(enic, wq, skb,
 735                         vlan_tag_insert, vlan_tag);
 736         else
 737                 enic_queue_wq_skb_vlan(enic, wq, skb,
 738                         vlan_tag_insert, vlan_tag);
 739 }
 740
 741 /* netif_tx_lock held, process context with BHs disabled, or BH */
 742 static netdev_tx_t enic_hard_start_xmit(struct sk_buff *skb,
 743         struct net_device *netdev)
 744 {
 745         struct enic *enic = netdev_priv(netdev);
 746         struct vnic_wq *wq = &enic->wq[0];
 747         unsigned long flags;
 748
 749         if (skb->len <= 0) {
 750                 dev_kfree_skb(skb);
 751                 return NETDEV_TX_OK;
 752         }
 753
 754         /* Non-TSO sends must fit within ENIC_NON_TSO_MAX_DESC descs,
 755          * which is very likely.  In the off chance it's going to take
 756          * more than * ENIC_NON_TSO_MAX_DESC, linearize the skb.
 757          */
 758
 759         if (skb_shinfo(skb)->gso_size == 0 &&
 760             skb_shinfo(skb)->nr_frags + 1 > ENIC_NON_TSO_MAX_DESC &&
 761             skb_linearize(skb)) {
 762                 dev_kfree_skb(skb);
 763                 return NETDEV_TX_OK;
 764         }
 765
 766         spin_lock_irqsave(&enic->wq_lock[0], flags);
 767
 768         if (vnic_wq_desc_avail(wq) <
 769             skb_shinfo(skb)->nr_frags + ENIC_DESC_MAX_SPLITS) {
 770                 netif_stop_queue(netdev);
 771                 /* This is a hard error, log it */
 772                 printk(KERN_ERR PFX "%s: BUG! Tx ring full when "
 773                         "queue awake!\n", netdev->name);
 774                 spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 775                 return NETDEV_TX_BUSY;
 776         }
 777
 778         enic_queue_wq_skb(enic, wq, skb);
 779
 780         if (vnic_wq_desc_avail(wq) < MAX_SKB_FRAGS + ENIC_DESC_MAX_SPLITS)
 781                 netif_stop_queue(netdev);
 782
 783         spin_unlock_irqrestore(&enic->wq_lock[0], flags);
 784
 785         return NETDEV_TX_OK;
 786 }
 787
 788 /* dev_base_lock rwlock held, nominally process context */
 789 static struct net_device_stats *enic_get_stats(struct net_device *netdev)
 790 {
 791         struct enic *enic = netdev_priv(netdev);
 792         struct net_device_stats *net_stats = &netdev->stats;
 793         struct vnic_stats *stats;
 794
 795         spin_lock(&enic->devcmd_lock);
 796         vnic_dev_stats_dump(enic->vdev, &stats);
 797         spin_unlock(&enic->devcmd_lock);
 798
 799         net_stats->tx_packets = stats->tx.tx_frames_ok;
 800         net_stats->tx_bytes = stats->tx.tx_bytes_ok;
 801         net_stats->tx_errors = stats->tx.tx_errors;
 802         net_stats->tx_dropped = stats->tx.tx_drops;
 803
 804         net_stats->rx_packets = stats->rx.rx_frames_ok;
 805         net_stats->rx_bytes = stats->rx.rx_bytes_ok;
 806         net_stats->rx_errors = stats->rx.rx_errors;
 807         net_stats->multicast = stats->rx.rx_multicast_frames_ok;
 808         net_stats->rx_over_errors = enic->rq_truncated_pkts;
 809         net_stats->rx_crc_errors = enic->rq_bad_fcs;
 810         net_stats->rx_dropped = stats->rx.rx_no_bufs + stats->rx.rx_drop;
 811
 812         return net_stats;
 813 }
 814
 815 static void enic_reset_mcaddrs(struct enic *enic)
 816 {
 817         enic->mc_count = 0;
 818 }
 819
 820 static int enic_set_mac_addr(struct net_device *netdev, char *addr)
 821 {
 822         struct enic *enic = netdev_priv(netdev);
 823
 824         if (enic_is_dynamic(enic)) {
 825                 if (!is_valid_ether_addr(addr) && !is_zero_ether_addr(addr))
 826                         return -EADDRNOTAVAIL;
 827         } else {
 828                 if (!is_valid_ether_addr(addr))
 829                         return -EADDRNOTAVAIL;
 830         }
 831
 832         memcpy(netdev->dev_addr, addr, netdev->addr_len);
 833
 834         return 0;
 835 }
 836
 837 static int enic_dev_add_station_addr(struct enic *enic)
 838 {
 839         int err = 0;
 840
 841         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 842                 spin_lock(&enic->devcmd_lock);
 843                 err = vnic_dev_add_addr(enic->vdev, enic->netdev->dev_addr);
 844                 spin_unlock(&enic->devcmd_lock);
 845         }
 846
 847         return err;
 848 }
 849
 850 static int enic_dev_del_station_addr(struct enic *enic)
 851 {
 852         int err = 0;
 853
 854         if (is_valid_ether_addr(enic->netdev->dev_addr)) {
 855                 spin_lock(&enic->devcmd_lock);
 856                 err = vnic_dev_del_addr(enic->vdev, enic->netdev->dev_addr);
 857                 spin_unlock(&enic->devcmd_lock);
 858         }
 859
 860         return err;
 861 }
 862
 863 static int enic_set_mac_address_dynamic(struct net_device *netdev, void *p)
 864 {
 865         struct enic *enic = netdev_priv(netdev);
 866         struct sockaddr *saddr = p;
 867         char *addr = saddr->sa_data;
 868         int err;
 869
 870         if (netif_running(enic->netdev)) {
 871                 err = enic_dev_del_station_addr(enic);
 872                 if (err)
 873                         return err;
 874         }
 875
 876         err = enic_set_mac_addr(netdev, addr);
 877         if (err)
 878                 return err;
 879
 880         if (netif_running(enic->netdev)) {
 881                 err = enic_dev_add_station_addr(enic);
 882                 if (err)
 883                         return err;
 884         }
 885
 886         return err;
 887 }
 888
 889 static int enic_set_mac_address(struct net_device *netdev, void *p)
 890 {
 891         return -EOPNOTSUPP;
 892 }
 893
 894 /* netif_tx_lock held, BHs disabled */
 895 static void enic_set_multicast_list(struct net_device *netdev)
 896 {
 897         struct enic *enic = netdev_priv(netdev);
 898         struct netdev_hw_addr *ha;
 899         int directed = 1;
 900         int multicast = (netdev->flags & IFF_MULTICAST) ? 1 : 0;
 901         int broadcast = (netdev->flags & IFF_BROADCAST) ? 1 : 0;
 902         int promisc = (netdev->flags & IFF_PROMISC) ? 1 : 0;
 903         unsigned int mc_count = netdev_mc_count(netdev);
 904         int allmulti = (netdev->flags & IFF_ALLMULTI) ||
 905                 mc_count > ENIC_MULTICAST_PERFECT_FILTERS;
 906         unsigned int flags = netdev->flags | (allmulti ? IFF_ALLMULTI : 0);
 907         u8 mc_addr[ENIC_MULTICAST_PERFECT_FILTERS][ETH_ALEN];
 908         unsigned int i, j;
 909
 910         if (mc_count > ENIC_MULTICAST_PERFECT_FILTERS)
 911                 mc_count = ENIC_MULTICAST_PERFECT_FILTERS;
 912
 913         spin_lock(&enic->devcmd_lock);
 914
 915         if (enic->flags != flags) {
 916                 enic->flags = flags;
 917                 vnic_dev_packet_filter(enic->vdev, directed,
 918                         multicast, broadcast, promisc, allmulti);
 919         }
 920
 921         /* Is there an easier way?  Trying to minimize to
 922          * calls to add/del multicast addrs.  We keep the
 923          * addrs from the last call in enic->mc_addr and
 924          * look for changes to add/del.
 925          */
 926
 927         i = 0;
 928         netdev_for_each_mc_addr(ha, netdev) {
 929                 if (i == mc_count)
 930                         break;
 931                 memcpy(mc_addr[i++], ha->addr, ETH_ALEN);
 932         }
 933
 934         for (i = 0; i < enic->mc_count; i++) {
 935                 for (j = 0; j < mc_count; j++)
 936                         if (compare_ether_addr(enic->mc_addr[i],
 937                                 mc_addr[j]) == 0)
 938                                 break;
 939                 if (j == mc_count)
 940                         enic_del_multicast_addr(enic, enic->mc_addr[i]);
 941         }
 942
 943         for (i = 0; i < mc_count; i++) {
 944                 for (j = 0; j < enic->mc_count; j++)
 945                         if (compare_ether_addr(mc_addr[i],
 946                                 enic->mc_addr[j]) == 0)
 947                                 break;
 948                 if (j == enic->mc_count)
 949                         enic_add_multicast_addr(enic, mc_addr[i]);
 950         }
 951
 952         /* Save the list to compare against next time
 953          */
 954
 955         for (i = 0; i < mc_count; i++)
 956                 memcpy(enic->mc_addr[i], mc_addr[i], ETH_ALEN);
 957
 958         enic->mc_count = mc_count;
 959
 960         spin_unlock(&enic->devcmd_lock);
 961 }
 962
 963 /* rtnl lock is held */
 964 static void enic_vlan_rx_register(struct net_device *netdev,
 965         struct vlan_group *vlan_group)
 966 {
 967         struct enic *enic = netdev_priv(netdev);
 968         enic->vlan_group = vlan_group;
 969 }
 970
 971 /* rtnl lock is held */
 972 static void enic_vlan_rx_add_vid(struct net_device *netdev, u16 vid)
 973 {
 974         struct enic *enic = netdev_priv(netdev);
 975
 976         spin_lock(&enic->devcmd_lock);
 977         enic_add_vlan(enic, vid);
 978         spin_unlock(&enic->devcmd_lock);
 979 }
 980
 981 /* rtnl lock is held */
 982 static void enic_vlan_rx_kill_vid(struct net_device *netdev, u16 vid)
 983 {
 984         struct enic *enic = netdev_priv(netdev);
 985
 986         spin_lock(&enic->devcmd_lock);
 987         enic_del_vlan(enic, vid);
 988         spin_unlock(&enic->devcmd_lock);
 989 }
 990
 991 /* netif_tx_lock held, BHs disabled */
 992 static void enic_tx_timeout(struct net_device *netdev)
 993 {
 994         struct enic *enic = netdev_priv(netdev);
 995         schedule_work(&enic->reset);
 996 }
 997
 998 static int enic_vnic_dev_deinit(struct enic *enic)
 999 {
1000         int err;
1001
1002         spin_lock(&enic->devcmd_lock);
1003         err = vnic_dev_deinit(enic->vdev);
1004         spin_unlock(&enic->devcmd_lock);
1005
1006         return err;
1007 }
1008
1009 static int enic_dev_init_prov(struct enic *enic, struct vic_provinfo *vp)
1010 {
1011         int err;
1012
1013         spin_lock(&enic->devcmd_lock);
1014         err = vnic_dev_init_prov(enic->vdev,
1015                 (u8 *)vp, vic_provinfo_size(vp));
1016         spin_unlock(&enic->devcmd_lock);
1017
1018         return err;
1019 }
1020
1021 static int enic_dev_init_done(struct enic *enic, int *done, int *error)
1022 {
1023         int err;
1024
1025         spin_lock(&enic->devcmd_lock);
1026         err = vnic_dev_init_done(enic->vdev, done, error);
1027         spin_unlock(&enic->devcmd_lock);
1028
1029         return err;
1030 }
1031
1032 static int enic_set_port_profile(struct enic *enic, u8 request, u8 *mac,
1033         char *name, u8 *instance_uuid, u8 *host_uuid)
1034 {
1035         struct vic_provinfo *vp;
1036         u8 oui[3] = VIC_PROVINFO_CISCO_OUI;
1037         u8 *uuid;
1038         char uuid_str[38];
1039         static char *uuid_fmt = "%02X%02X%02X%02X-%02X%02X-%02X%02X-"
1040                 "%02X%02X-%02X%02X%02X%02X%0X%02X";
1041         int err;
1042
1043         if (!name)
1044                 return -EINVAL;
1045
1046         if (!is_valid_ether_addr(mac))
1047                 return -EADDRNOTAVAIL;
1048
1049         vp = vic_provinfo_alloc(GFP_KERNEL, oui, VIC_PROVINFO_LINUX_TYPE);
1050         if (!vp)
1051                 return -ENOMEM;
1052
1053         vic_provinfo_add_tlv(vp,
1054                 VIC_LINUX_PROV_TLV_PORT_PROFILE_NAME_STR,
1055                 strlen(name) + 1, name);
1056
1057         vic_provinfo_add_tlv(vp,
1058                 VIC_LINUX_PROV_TLV_CLIENT_MAC_ADDR,
1059                 ETH_ALEN, mac);
1060
1061         if (instance_uuid) {
1062                 uuid = instance_uuid;
1063                 sprintf(uuid_str, uuid_fmt,
1064                         uuid[0],  uuid[1],  uuid[2],  uuid[3],
1065                         uuid[4],  uuid[5],  uuid[6],  uuid[7],
1066                         uuid[8],  uuid[9],  uuid[10], uuid[11],
1067                         uuid[12], uuid[13], uuid[14], uuid[15]);
1068                 vic_provinfo_add_tlv(vp,
1069                         VIC_LINUX_PROV_TLV_CLIENT_UUID_STR,
1070                         sizeof(uuid_str), uuid_str);
1071         }
1072
1073         if (host_uuid) {
1074                 uuid = host_uuid;
1075                 sprintf(uuid_str, uuid_fmt,
1076                         uuid[0],  uuid[1],  uuid[2],  uuid[3],
1077                         uuid[4],  uuid[5],  uuid[6],  uuid[7],
1078                         uuid[8],  uuid[9],  uuid[10], uuid[11],
1079                         uuid[12], uuid[13], uuid[14], uuid[15]);
1080                 vic_provinfo_add_tlv(vp,
1081                         VIC_LINUX_PROV_TLV_HOST_UUID_STR,
1082                         sizeof(uuid_str), uuid_str);
1083         }
1084
1085         err = enic_vnic_dev_deinit(enic);
1086         if (err)
1087                 goto err_out;
1088
1089         memset(&enic->pp, 0, sizeof(enic->pp));
1090
1091         err = enic_dev_init_prov(enic, vp);
1092         if (err)
1093                 goto err_out;
1094
1095         enic->pp.request = request;
1096         memcpy(enic->pp.name, name, PORT_PROFILE_MAX);
1097         if (instance_uuid)
1098                 memcpy(enic->pp.instance_uuid,
1099                         instance_uuid, PORT_UUID_MAX);
1100         if (host_uuid)
1101                 memcpy(enic->pp.host_uuid,
1102                         host_uuid, PORT_UUID_MAX);
1103
1104 err_out:
1105         vic_provinfo_free(vp);
1106
1107         return err;
1108 }
1109
1110 static int enic_unset_port_profile(struct enic *enic)
1111 {
1112         memset(&enic->pp, 0, sizeof(enic->pp));
1113         return enic_vnic_dev_deinit(enic);
1114 }
1115
1116 static int enic_set_vf_port(struct net_device *netdev, int vf,
1117         struct nlattr *port[])
1118 {
1119         struct enic *enic = netdev_priv(netdev);
1120         char *name = NULL;
1121         u8 *instance_uuid = NULL;
1122         u8 *host_uuid = NULL;
1123         u8 request = PORT_REQUEST_DISASSOCIATE;
1124
1125         /* don't support VFs, yet */
1126         if (vf != PORT_SELF_VF)
1127                 return -EOPNOTSUPP;
1128
1129         if (port[IFLA_PORT_REQUEST])
1130                 request = nla_get_u8(port[IFLA_PORT_REQUEST]);
1131
1132         switch (request) {
1133         case PORT_REQUEST_ASSOCIATE:
1134
1135                 /* If the interface mac addr hasn't been assigned,
1136                  * assign a random mac addr before setting port-
1137                  * profile.
1138                  */
1139
1140                 if (is_zero_ether_addr(netdev->dev_addr))
1141                         random_ether_addr(netdev->dev_addr);
1142
1143                 if (port[IFLA_PORT_PROFILE])
1144                         name = nla_data(port[IFLA_PORT_PROFILE]);
1145
1146                 if (port[IFLA_PORT_INSTANCE_UUID])
1147                         instance_uuid =
1148                                 nla_data(port[IFLA_PORT_INSTANCE_UUID]);
1149
1150                 if (port[IFLA_PORT_HOST_UUID])
1151                         host_uuid = nla_data(port[IFLA_PORT_HOST_UUID]);
1152
1153                 return enic_set_port_profile(enic, request,
1154                         netdev->dev_addr, name,
1155                         instance_uuid, host_uuid);
1156
1157         case PORT_REQUEST_DISASSOCIATE:
1158
1159                 return enic_unset_port_profile(enic);
1160
1161         default:
1162                 break;
1163         }
1164
1165         return -EOPNOTSUPP;
1166 }
1167
1168 static int enic_get_vf_port(struct net_device *netdev, int vf,
1169         struct sk_buff *skb)
1170 {
1171         struct enic *enic = netdev_priv(netdev);
1172         int err, error, done;
1173         u16 response = PORT_PROFILE_RESPONSE_SUCCESS;
1174
1175         /* don't support VFs, yet */
1176         if (vf != PORT_SELF_VF)
1177                 return -EOPNOTSUPP;
1178
1179         err = enic_dev_init_done(enic, &done, &error);
1180
1181         if (err)
1182                 return err;
1183
1184         switch (error) {
1185         case ERR_SUCCESS:
1186                 if (!done)
1187                         response = PORT_PROFILE_RESPONSE_INPROGRESS;
1188                 break;
1189         case ERR_EINVAL:
1190                 response = PORT_PROFILE_RESPONSE_INVALID;
1191                 break;
1192         case ERR_EBADSTATE:
1193                 response = PORT_PROFILE_RESPONSE_BADSTATE;
1194                 break;
1195         case ERR_ENOMEM:
1196                 response = PORT_PROFILE_RESPONSE_INSUFFICIENT_RESOURCES;
1197                 break;
1198         default:
1199                 response = PORT_PROFILE_RESPONSE_ERROR;
1200                 break;
1201         }
1202
1203         NLA_PUT_U16(skb, IFLA_PORT_REQUEST, enic->pp.request);
1204         NLA_PUT_U16(skb, IFLA_PORT_RESPONSE, response);
1205         NLA_PUT(skb, IFLA_PORT_PROFILE, PORT_PROFILE_MAX,
1206                 enic->pp.name);
1207         NLA_PUT(skb, IFLA_PORT_INSTANCE_UUID, PORT_UUID_MAX,
1208                 enic->pp.instance_uuid);
1209         NLA_PUT(skb, IFLA_PORT_HOST_UUID, PORT_UUID_MAX,
1210                 enic->pp.host_uuid);
1211
1212         return 0;
1213
1214 nla_put_failure:
1215         return -EMSGSIZE;
1216 }
1217
1218 static void enic_free_rq_buf(struct vnic_rq *rq, struct vnic_rq_buf *buf)
1219 {
1220         struct enic *enic = vnic_dev_priv(rq->vdev);
1221
1222         if (!buf->os_buf)
1223                 return;
1224
1225         pci_unmap_single(enic->pdev, buf->dma_addr,
1226                 buf->len, PCI_DMA_FROMDEVICE);
1227         dev_kfree_skb_any(buf->os_buf);
1228 }
1229
1230 static int enic_rq_alloc_buf(struct vnic_rq *rq)
1231 {
1232         struct enic *enic = vnic_dev_priv(rq->vdev);
1233         struct net_device *netdev = enic->netdev;
1234         struct sk_buff *skb;
1235         unsigned int len = netdev->mtu + ETH_HLEN;
1236         unsigned int os_buf_index = 0;
1237         dma_addr_t dma_addr;
1238
1239         skb = netdev_alloc_skb_ip_align(netdev, len);
1240         if (!skb)
1241                 return -ENOMEM;
1242
1243         dma_addr = pci_map_single(enic->pdev, skb->data,
1244                 len, PCI_DMA_FROMDEVICE);
1245
1246         enic_queue_rq_desc(rq, skb, os_buf_index,
1247                 dma_addr, len);
1248
1249         return 0;
1250 }
1251
1252 static int enic_rq_alloc_buf_a1(struct vnic_rq *rq)
1253 {
1254         struct rq_enet_desc *desc = vnic_rq_next_desc(rq);
1255
1256         if (vnic_rq_posting_soon(rq)) {
1257
1258                 /* SW workaround for A0 HW erratum: if we're just about
1259                  * to write posted_index, insert a dummy desc
1260                  * of type resvd
1261                  */
1262
1263                 rq_enet_desc_enc(desc, 0, RQ_ENET_TYPE_RESV2, 0);
1264                 vnic_rq_post(rq, 0, 0, 0, 0);
1265         } else {
1266                 return enic_rq_alloc_buf(rq);
1267         }
1268
1269         return 0;
1270 }
1271
1272 static int enic_set_rq_alloc_buf(struct enic *enic)
1273 {
1274         enum vnic_dev_hw_version hw_ver;
1275         int err;
1276
1277         err = vnic_dev_hw_version(enic->vdev, &hw_ver);
1278         if (err)
1279                 return err;
1280
1281         switch (hw_ver) {
1282         case VNIC_DEV_HW_VER_A1:
1283                 enic->rq_alloc_buf = enic_rq_alloc_buf_a1;
1284                 break;
1285         case VNIC_DEV_HW_VER_A2:
1286         case VNIC_DEV_HW_VER_UNKNOWN:
1287                 enic->rq_alloc_buf = enic_rq_alloc_buf;
1288                 break;
1289         default:
1290                 return -ENODEV;
1291         }
1292
1293         return 0;
1294 }
1295
1296 static int enic_get_skb_header(struct sk_buff *skb, void **iphdr,
1297         void **tcph, u64 *hdr_flags, void *priv)
1298 {
1299         struct cq_enet_rq_desc *cq_desc = priv;
1300         unsigned int ip_len;
1301         struct iphdr *iph;
1302
1303         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1304         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1305         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1306         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1307         u8 packet_error;
1308         u16 q_number, completed_index, bytes_written, vlan, checksum;
1309         u32 rss_hash;
1310
1311         cq_enet_rq_desc_dec(cq_desc,
1312                 &type, &color, &q_number, &completed_index,
1313                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1314                 &csum_not_calc, &rss_hash, &bytes_written,
1315                 &packet_error, &vlan_stripped, &vlan, &checksum,
1316                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1317                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1318                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1319                 &fcs_ok);
1320
1321         if (!(ipv4 && tcp && !ipv4_fragment))
1322                 return -1;
1323
1324         skb_reset_network_header(skb);
1325         iph = ip_hdr(skb);
1326
1327         ip_len = ip_hdrlen(skb);
1328         skb_set_transport_header(skb, ip_len);
1329
1330         /* check if ip header and tcp header are complete */
1331         if (ntohs(iph->tot_len) < ip_len + tcp_hdrlen(skb))
1332                 return -1;
1333
1334         *hdr_flags = LRO_IPV4 | LRO_TCP;
1335         *tcph = tcp_hdr(skb);
1336         *iphdr = iph;
1337
1338         return 0;
1339 }
1340
1341 static void enic_rq_indicate_buf(struct vnic_rq *rq,
1342         struct cq_desc *cq_desc, struct vnic_rq_buf *buf,
1343         int skipped, void *opaque)
1344 {
1345         struct enic *enic = vnic_dev_priv(rq->vdev);
1346         struct net_device *netdev = enic->netdev;
1347         struct sk_buff *skb;
1348
1349         u8 type, color, eop, sop, ingress_port, vlan_stripped;
1350         u8 fcoe, fcoe_sof, fcoe_fc_crc_ok, fcoe_enc_error, fcoe_eof;
1351         u8 tcp_udp_csum_ok, udp, tcp, ipv4_csum_ok;
1352         u8 ipv6, ipv4, ipv4_fragment, fcs_ok, rss_type, csum_not_calc;
1353         u8 packet_error;
1354         u16 q_number, completed_index, bytes_written, vlan, checksum;
1355         u32 rss_hash;
1356
1357         if (skipped)
1358                 return;
1359
1360         skb = buf->os_buf;
1361         prefetch(skb->data - NET_IP_ALIGN);
1362         pci_unmap_single(enic->pdev, buf->dma_addr,
1363                 buf->len, PCI_DMA_FROMDEVICE);
1364
1365         cq_enet_rq_desc_dec((struct cq_enet_rq_desc *)cq_desc,
1366                 &type, &color, &q_number, &completed_index,
1367                 &ingress_port, &fcoe, &eop, &sop, &rss_type,
1368                 &csum_not_calc, &rss_hash, &bytes_written,
1369                 &packet_error, &vlan_stripped, &vlan, &checksum,
1370                 &fcoe_sof, &fcoe_fc_crc_ok, &fcoe_enc_error,
1371                 &fcoe_eof, &tcp_udp_csum_ok, &udp, &tcp,
1372                 &ipv4_csum_ok, &ipv6, &ipv4, &ipv4_fragment,
1373                 &fcs_ok);
1374
1375         if (packet_error) {
1376
1377                 if (!fcs_ok) {
1378                         if (bytes_written > 0)
1379                                 enic->rq_bad_fcs++;
1380                         else if (bytes_written == 0)
1381                                 enic->rq_truncated_pkts++;
1382                 }
1383
1384                 dev_kfree_skb_any(skb);
1385
1386                 return;
1387         }
1388
1389         if (eop && bytes_written > 0) {
1390
1391                 /* Good receive
1392                  */
1393
1394                 skb_put(skb, bytes_written);
1395                 skb->protocol = eth_type_trans(skb, netdev);
1396
1397                 if (enic->csum_rx_enabled && !csum_not_calc) {
1398                         skb->csum = htons(checksum);
1399                         skb->ip_summed = CHECKSUM_COMPLETE;
1400                 }
1401
1402                 skb->dev = netdev;
1403
1404                 if (enic->vlan_group && vlan_stripped) {
1405
1406                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1407                                 lro_vlan_hwaccel_receive_skb(&enic->lro_mgr,
1408                                         skb, enic->vlan_group,
1409                                         vlan, cq_desc);
1410                         else
1411                                 vlan_hwaccel_receive_skb(skb,
1412                                         enic->vlan_group, vlan);
1413
1414                 } else {
1415
1416                         if ((netdev->features & NETIF_F_LRO) && ipv4)
1417                                 lro_receive_skb(&enic->lro_mgr, skb, cq_desc);
1418                         else
1419                                 netif_receive_skb(skb);
1420
1421                 }
1422
1423         } else {
1424
1425                 /* Buffer overflow
1426                  */
1427
1428                 dev_kfree_skb_any(skb);
1429         }
1430 }
1431
1432 static int enic_rq_service(struct vnic_dev *vdev, struct cq_desc *cq_desc,
1433         u8 type, u16 q_number, u16 completed_index, void *opaque)
1434 {
1435         struct enic *enic = vnic_dev_priv(vdev);
1436
1437         vnic_rq_service(&enic->rq[q_number], cq_desc,
1438                 completed_index, VNIC_RQ_RETURN_DESC,
1439                 enic_rq_indicate_buf, opaque);
1440
1441         return 0;
1442 }
1443
1444 static int enic_poll(struct napi_struct *napi, int budget)
1445 {
1446         struct enic *enic = container_of(napi, struct enic, napi);
1447         struct net_device *netdev = enic->netdev;
1448         unsigned int rq_work_to_do = budget;
1449         unsigned int wq_work_to_do = -1; /* no limit */
1450         unsigned int  work_done, rq_work_done, wq_work_done;
1451         int err;
1452
1453         /* Service RQ (first) and WQ
1454          */
1455
1456         rq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1457                 rq_work_to_do, enic_rq_service, NULL);
1458
1459         wq_work_done = vnic_cq_service(&enic->cq[ENIC_CQ_WQ],
1460                 wq_work_to_do, enic_wq_service, NULL);
1461
1462         /* Accumulate intr event credits for this polling
1463          * cycle.  An intr event is the completion of a
1464          * a WQ or RQ packet.
1465          */
1466
1467         work_done = rq_work_done + wq_work_done;
1468
1469         if (work_done > 0)
1470                 vnic_intr_return_credits(&enic->intr[ENIC_INTX_WQ_RQ],
1471                         work_done,
1472                         0 /* don't unmask intr */,
1473                         0 /* don't reset intr timer */);
1474
1475         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1476
1477         /* Buffer allocation failed. Stay in polling
1478          * mode so we can try to fill the ring again.
1479          */
1480
1481         if (err)
1482                 rq_work_done = rq_work_to_do;
1483
1484         if (rq_work_done < rq_work_to_do) {
1485
1486                 /* Some work done, but not enough to stay in polling,
1487                  * flush all LROs and exit polling
1488                  */
1489
1490                 if (netdev->features & NETIF_F_LRO)
1491                         lro_flush_all(&enic->lro_mgr);
1492
1493                 napi_complete(napi);
1494                 vnic_intr_unmask(&enic->intr[ENIC_INTX_WQ_RQ]);
1495         }
1496
1497         return rq_work_done;
1498 }
1499
1500 static int enic_poll_msix(struct napi_struct *napi, int budget)
1501 {
1502         struct enic *enic = container_of(napi, struct enic, napi);
1503         struct net_device *netdev = enic->netdev;
1504         unsigned int work_to_do = budget;
1505         unsigned int work_done;
1506         int err;
1507
1508         /* Service RQ
1509          */
1510
1511         work_done = vnic_cq_service(&enic->cq[ENIC_CQ_RQ],
1512                 work_to_do, enic_rq_service, NULL);
1513
1514         /* Return intr event credits for this polling
1515          * cycle.  An intr event is the completion of a
1516          * RQ packet.
1517          */
1518
1519         if (work_done > 0)
1520                 vnic_intr_return_credits(&enic->intr[ENIC_MSIX_RQ],
1521                         work_done,
1522                         0 /* don't unmask intr */,
1523                         0 /* don't reset intr timer */);
1524
1525         err = vnic_rq_fill(&enic->rq[0], enic->rq_alloc_buf);
1526
1527         /* Buffer allocation failed. Stay in polling mode
1528          * so we can try to fill the ring again.
1529          */
1530
1531         if (err)
1532                 work_done = work_to_do;
1533
1534         if (work_done < work_to_do) {
1535
1536                 /* Some work done, but not enough to stay in polling,
1537                  * flush all LROs and exit polling
1538                  */
1539
1540                 if (netdev->features & NETIF_F_LRO)
1541                         lro_flush_all(&enic->lro_mgr);
1542
1543                 napi_complete(napi);
1544                 vnic_intr_unmask(&enic->intr[ENIC_MSIX_RQ]);
1545         }
1546
1547         return work_done;
1548 }
1549
1550 static void enic_notify_timer(unsigned long data)
1551 {
1552         struct enic *enic = (struct enic *)data;
1553
1554         enic_notify_check(enic);
1555
1556         mod_timer(&enic->notify_timer,
1557                 round_jiffies(jiffies + ENIC_NOTIFY_TIMER_PERIOD));
1558 }
1559
1560 static void enic_free_intr(struct enic *enic)
1561 {
1562         struct net_device *netdev = enic->netdev;
1563         unsigned int i;
1564
1565         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1566         case VNIC_DEV_INTR_MODE_INTX:
1567                 free_irq(enic->pdev->irq, netdev);
1568                 break;
1569         case VNIC_DEV_INTR_MODE_MSI:
1570                 free_irq(enic->pdev->irq, enic);
1571                 break;
1572         case VNIC_DEV_INTR_MODE_MSIX:
1573                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++)
1574                         if (enic->msix[i].requested)
1575                                 free_irq(enic->msix_entry[i].vector,
1576                                         enic->msix[i].devid);
1577                 break;
1578         default:
1579                 break;
1580         }
1581 }
1582
1583 static int enic_request_intr(struct enic *enic)
1584 {
1585         struct net_device *netdev = enic->netdev;
1586         unsigned int i;
1587         int err = 0;
1588
1589         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1590
1591         case VNIC_DEV_INTR_MODE_INTX:
1592
1593                 err = request_irq(enic->pdev->irq, enic_isr_legacy,
1594                         IRQF_SHARED, netdev->name, netdev);
1595                 break;
1596
1597         case VNIC_DEV_INTR_MODE_MSI:
1598
1599                 err = request_irq(enic->pdev->irq, enic_isr_msi,
1600                         0, netdev->name, enic);
1601                 break;
1602
1603         case VNIC_DEV_INTR_MODE_MSIX:
1604
1605                 sprintf(enic->msix[ENIC_MSIX_RQ].devname,
1606                         "%.11s-rx-0", netdev->name);
1607                 enic->msix[ENIC_MSIX_RQ].isr = enic_isr_msix_rq;
1608                 enic->msix[ENIC_MSIX_RQ].devid = enic;
1609
1610                 sprintf(enic->msix[ENIC_MSIX_WQ].devname,
1611                         "%.11s-tx-0", netdev->name);
1612                 enic->msix[ENIC_MSIX_WQ].isr = enic_isr_msix_wq;
1613                 enic->msix[ENIC_MSIX_WQ].devid = enic;
1614
1615                 sprintf(enic->msix[ENIC_MSIX_ERR].devname,
1616                         "%.11s-err", netdev->name);
1617                 enic->msix[ENIC_MSIX_ERR].isr = enic_isr_msix_err;
1618                 enic->msix[ENIC_MSIX_ERR].devid = enic;
1619
1620                 sprintf(enic->msix[ENIC_MSIX_NOTIFY].devname,
1621                         "%.11s-notify", netdev->name);
1622                 enic->msix[ENIC_MSIX_NOTIFY].isr = enic_isr_msix_notify;
1623                 enic->msix[ENIC_MSIX_NOTIFY].devid = enic;
1624
1625                 for (i = 0; i < ARRAY_SIZE(enic->msix); i++) {
1626                         err = request_irq(enic->msix_entry[i].vector,
1627                                 enic->msix[i].isr, 0,
1628                                 enic->msix[i].devname,
1629                                 enic->msix[i].devid);
1630                         if (err) {
1631                                 enic_free_intr(enic);
1632                                 break;
1633                         }
1634                         enic->msix[i].requested = 1;
1635                 }
1636
1637                 break;
1638
1639         default:
1640                 break;
1641         }
1642
1643         return err;
1644 }
1645
1646 static void enic_synchronize_irqs(struct enic *enic)
1647 {
1648         unsigned int i;
1649
1650         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1651         case VNIC_DEV_INTR_MODE_INTX:
1652         case VNIC_DEV_INTR_MODE_MSI:
1653                 synchronize_irq(enic->pdev->irq);
1654                 break;
1655         case VNIC_DEV_INTR_MODE_MSIX:
1656                 for (i = 0; i < enic->intr_count; i++)
1657                         synchronize_irq(enic->msix_entry[i].vector);
1658                 break;
1659         default:
1660                 break;
1661         }
1662 }
1663
1664 static int enic_notify_set(struct enic *enic)
1665 {
1666         int err;
1667
1668         spin_lock(&enic->devcmd_lock);
1669         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1670         case VNIC_DEV_INTR_MODE_INTX:
1671                 err = vnic_dev_notify_set(enic->vdev, ENIC_INTX_NOTIFY);
1672                 break;
1673         case VNIC_DEV_INTR_MODE_MSIX:
1674                 err = vnic_dev_notify_set(enic->vdev, ENIC_MSIX_NOTIFY);
1675                 break;
1676         default:
1677                 err = vnic_dev_notify_set(enic->vdev, -1 /* no intr */);
1678                 break;
1679         }
1680         spin_unlock(&enic->devcmd_lock);
1681
1682         return err;
1683 }
1684
1685 static void enic_notify_timer_start(struct enic *enic)
1686 {
1687         switch (vnic_dev_get_intr_mode(enic->vdev)) {
1688         case VNIC_DEV_INTR_MODE_MSI:
1689                 mod_timer(&enic->notify_timer, jiffies);
1690                 break;
1691         default:
1692                 /* Using intr for notification for INTx/MSI-X */
1693                 break;
1694         };
1695 }
1696
1697 /* rtnl lock is held, process context */
1698 static int enic_open(struct net_device *netdev)
1699 {
1700         struct enic *enic = netdev_priv(netdev);
1701         unsigned int i;
1702         int err;
1703
1704         err = enic_request_intr(enic);
1705         if (err) {
1706                 printk(KERN_ERR PFX "%s: Unable to request irq.\n",
1707                         netdev->name);
1708                 return err;
1709         }
1710
1711         err = enic_notify_set(enic);
1712         if (err) {
1713                 printk(KERN_ERR PFX
1714                         "%s: Failed to alloc notify buffer, aborting.\n",
1715                         netdev->name);
1716                 goto err_out_free_intr;
1717         }
1718
1719         for (i = 0; i < enic->rq_count; i++) {
1720                 vnic_rq_fill(&enic->rq[i], enic->rq_alloc_buf);
1721                 /* Need at least one buffer on ring to get going */
1722                 if (vnic_rq_desc_used(&enic->rq[i]) == 0) {
1723                         printk(KERN_ERR PFX
1724                                 "%s: Unable to alloc receive buffers.\n",
1725                                 netdev->name);
1726                         err = -ENOMEM;
1727                         goto err_out_notify_unset;
1728                 }
1729         }
1730
1731         for (i = 0; i < enic->wq_count; i++)
1732                 vnic_wq_enable(&enic->wq[i]);
1733         for (i = 0; i < enic->rq_count; i++)
1734                 vnic_rq_enable(&enic->rq[i]);
1735
1736         enic_dev_add_station_addr(enic);
1737         enic_set_multicast_list(netdev);
1738
1739         netif_wake_queue(netdev);
1740         napi_enable(&enic->napi);
1741         spin_lock(&enic->devcmd_lock);
1742         vnic_dev_enable(enic->vdev);
1743         spin_unlock(&enic->devcmd_lock);
1744
1745         for (i = 0; i < enic->intr_count; i++)
1746                 vnic_intr_unmask(&enic->intr[i]);
1747
1748         enic_notify_timer_start(enic);
1749
1750         return 0;
1751
1752 err_out_notify_unset:
1753         spin_lock(&enic->devcmd_lock);
1754         vnic_dev_notify_unset(enic->vdev);
1755         spin_unlock(&enic->devcmd_lock);
1756 err_out_free_intr:
1757         enic_free_intr(enic);
1758
1759         return err;
1760 }
1761
1762 /* rtnl lock is held, process context */
1763 static int enic_stop(struct net_device *netdev)
1764 {
1765         struct enic *enic = netdev_priv(netdev);
1766         unsigned int i;
1767         int err;
1768
1769         for (i = 0; i < enic->intr_count; i++)
1770                 vnic_intr_mask(&enic->intr[i]);
1771
1772         enic_synchronize_irqs(enic);
1773
1774         del_timer_sync(&enic->notify_timer);
1775
1776         spin_lock(&enic->devcmd_lock);
1777         vnic_dev_disable(enic->vdev);
1778         spin_unlock(&enic->devcmd_lock);
1779         napi_disable(&enic->napi);
1780         netif_carrier_off(netdev);
1781         netif_tx_disable(netdev);
1782
1783         enic_dev_del_station_addr(enic);
1784
1785         for (i = 0; i < enic->wq_count; i++) {
1786                 err = vnic_wq_disable(&enic->wq[i]);
1787                 if (err)
1788                         return err;
1789         }
1790         for (i = 0; i < enic->rq_count; i++) {
1791                 err = vnic_rq_disable(&enic->rq[i]);
1792                 if (err)
1793                         return err;
1794         }
1795
1796         spin_lock(&enic->devcmd_lock);
1797         vnic_dev_notify_unset(enic->vdev);
1798         spin_unlock(&enic->devcmd_lock);
1799         enic_free_intr(enic);
1800
1801         for (i = 0; i < enic->wq_count; i++)
1802                 vnic_wq_clean(&enic->wq[i], enic_free_wq_buf);
1803         for (i = 0; i < enic->rq_count; i++)
1804                 vnic_rq_clean(&enic->rq[i], enic_free_rq_buf);
1805         for (i = 0; i < enic->cq_count; i++)
1806                 vnic_cq_clean(&enic->cq[i]);
1807         for (i = 0; i < enic->intr_count; i++)
1808                 vnic_intr_clean(&enic->intr[i]);
1809
1810         return 0;
1811 }
1812
1813 static int enic_change_mtu(struct net_device *netdev, int new_mtu)
1814 {
1815         struct enic *enic = netdev_priv(netdev);
1816         int running = netif_running(netdev);
1817
1818         if (new_mtu < ENIC_MIN_MTU || new_mtu > ENIC_MAX_MTU)
1819                 return -EINVAL;
1820
1821         if (running)
1822                 enic_stop(netdev);
1823
1824         netdev->mtu = new_mtu;
1825
1826         if (netdev->mtu > enic->port_mtu)
1827                 printk(KERN_WARNING PFX
1828                         "%s: interface MTU (%d) set higher "
1829                         "than port MTU (%d)\n",
1830                         netdev->name, netdev->mtu, enic->port_mtu);
1831
1832         if (running)
1833                 enic_open(netdev);
1834
1835         return 0;
1836 }
1837
1838 #ifdef CONFIG_NET_POLL_CONTROLLER
1839 static void enic_poll_controller(struct net_device *netdev)
1840 {
1841         struct enic *enic = netdev_priv(netdev);
1842         struct vnic_dev *vdev = enic->vdev;
1843
1844         switch (vnic_dev_get_intr_mode(vdev)) {
1845         case VNIC_DEV_INTR_MODE_MSIX:
1846                 enic_isr_msix_rq(enic->pdev->irq, enic);
1847                 enic_isr_msix_wq(enic->pdev->irq, enic);
1848                 break;
1849         case VNIC_DEV_INTR_MODE_MSI:
1850                 enic_isr_msi(enic->pdev->irq, enic);
1851                 break;
1852         case VNIC_DEV_INTR_MODE_INTX:
1853                 enic_isr_legacy(enic->pdev->irq, netdev);
1854                 break;
1855         default:
1856                 break;
1857         }
1858 }
1859 #endif
1860
1861 static int enic_dev_wait(struct vnic_dev *vdev,
1862         int (*start)(struct vnic_dev *, int),
1863         int (*finished)(struct vnic_dev *, int *),
1864         int arg)
1865 {
1866         unsigned long time;
1867         int done;
1868         int err;
1869
1870         BUG_ON(in_interrupt());
1871
1872         err = start(vdev, arg);
1873         if (err)
1874                 return err;
1875
1876         /* Wait for func to complete...2 seconds max
1877          */
1878
1879         time = jiffies + (HZ * 2);
1880         do {
1881
1882                 err = finished(vdev, &done);
1883                 if (err)
1884                         return err;
1885
1886                 if (done)
1887                         return 0;
1888
1889                 schedule_timeout_uninterruptible(HZ / 10);
1890
1891         } while (time_after(time, jiffies));
1892
1893         return -ETIMEDOUT;
1894 }
1895
1896 static int enic_dev_open(struct enic *enic)
1897 {
1898         int err;
1899
1900         err = enic_dev_wait(enic->vdev, vnic_dev_open,
1901                 vnic_dev_open_done, 0);
1902         if (err)
1903                 printk(KERN_ERR PFX
1904                         "vNIC device open failed, err %d.\n", err);
1905
1906         return err;
1907 }
1908
1909 static int enic_dev_soft_reset(struct enic *enic)
1910 {
1911         int err;
1912
1913         err = enic_dev_wait(enic->vdev, vnic_dev_soft_reset,
1914                 vnic_dev_soft_reset_done, 0);
1915         if (err)
1916                 printk(KERN_ERR PFX
1917                         "vNIC soft reset failed, err %d.\n", err);
1918
1919         return err;
1920 }
1921
1922 static int enic_set_niccfg(struct enic *enic)
1923 {
1924         const u8 rss_default_cpu = 0;
1925         const u8 rss_hash_type = 0;
1926         const u8 rss_hash_bits = 0;
1927         const u8 rss_base_cpu = 0;
1928         const u8 rss_enable = 0;
1929         const u8 tso_ipid_split_en = 0;
1930         const u8 ig_vlan_strip_en = 1;
1931
1932         /* Enable VLAN tag stripping.  RSS not enabled (yet).
1933          */
1934
1935         return enic_set_nic_cfg(enic,
1936                 rss_default_cpu, rss_hash_type,
1937                 rss_hash_bits, rss_base_cpu,
1938                 rss_enable, tso_ipid_split_en,
1939                 ig_vlan_strip_en);
1940 }
1941
1942 static void enic_reset(struct work_struct *work)
1943 {
1944         struct enic *enic = container_of(work, struct enic, reset);
1945
1946         if (!netif_running(enic->netdev))
1947                 return;
1948
1949         rtnl_lock();
1950
1951         spin_lock(&enic->devcmd_lock);
1952         vnic_dev_hang_notify(enic->vdev);
1953         spin_unlock(&enic->devcmd_lock);
1954
1955         enic_stop(enic->netdev);
1956         enic_dev_soft_reset(enic);
1957         vnic_dev_init(enic->vdev, 0);
1958         enic_reset_mcaddrs(enic);
1959         enic_init_vnic_resources(enic);
1960         enic_set_niccfg(enic);
1961         enic_open(enic->netdev);
1962
1963         rtnl_unlock();
1964 }
1965
1966 static int enic_set_intr_mode(struct enic *enic)
1967 {
1968         unsigned int n = 1;
1969         unsigned int m = 1;
1970         unsigned int i;
1971
1972         /* Set interrupt mode (INTx, MSI, MSI-X) depending
1973          * system capabilities.
1974          *
1975          * Try MSI-X first
1976          *
1977          * We need n RQs, m WQs, n+m CQs, and n+m+2 INTRs
1978          * (the second to last INTR is used for WQ/RQ errors)
1979          * (the last INTR is used for notifications)
1980          */
1981
1982         BUG_ON(ARRAY_SIZE(enic->msix_entry) < n + m + 2);
1983         for (i = 0; i < n + m + 2; i++)
1984                 enic->msix_entry[i].entry = i;
1985
1986         if (enic->config.intr_mode < 1 &&
1987             enic->rq_count >= n &&
1988             enic->wq_count >= m &&
1989             enic->cq_count >= n + m &&
1990             enic->intr_count >= n + m + 2 &&
1991             !pci_enable_msix(enic->pdev, enic->msix_entry, n + m + 2)) {
1992
1993                 enic->rq_count = n;
1994                 enic->wq_count = m;
1995                 enic->cq_count = n + m;
1996                 enic->intr_count = n + m + 2;
1997
1998                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSIX);
1999
2000                 return 0;
2001         }
2002
2003         /* Next try MSI
2004          *
2005          * We need 1 RQ, 1 WQ, 2 CQs, and 1 INTR
2006          */
2007
2008         if (enic->config.intr_mode < 2 &&
2009             enic->rq_count >= 1 &&
2010             enic->wq_count >= 1 &&
2011             enic->cq_count >= 2 &&
2012             enic->intr_count >= 1 &&
2013             !pci_enable_msi(enic->pdev)) {
2014
2015                 enic->rq_count = 1;
2016                 enic->wq_count = 1;
2017                 enic->cq_count = 2;
2018                 enic->intr_count = 1;
2019
2020                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_MSI);
2021
2022                 return 0;
2023         }
2024
2025         /* Next try INTx
2026          *
2027          * We need 1 RQ, 1 WQ, 2 CQs, and 3 INTRs
2028          * (the first INTR is used for WQ/RQ)
2029          * (the second INTR is used for WQ/RQ errors)
2030          * (the last INTR is used for notifications)
2031          */
2032
2033         if (enic->config.intr_mode < 3 &&
2034             enic->rq_count >= 1 &&
2035             enic->wq_count >= 1 &&
2036             enic->cq_count >= 2 &&
2037             enic->intr_count >= 3) {
2038
2039                 enic->rq_count = 1;
2040                 enic->wq_count = 1;
2041                 enic->cq_count = 2;
2042                 enic->intr_count = 3;
2043
2044                 vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_INTX);
2045
2046                 return 0;
2047         }
2048
2049         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2050
2051         return -EINVAL;
2052 }
2053
2054 static void enic_clear_intr_mode(struct enic *enic)
2055 {
2056         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2057         case VNIC_DEV_INTR_MODE_MSIX:
2058                 pci_disable_msix(enic->pdev);
2059                 break;
2060         case VNIC_DEV_INTR_MODE_MSI:
2061                 pci_disable_msi(enic->pdev);
2062                 break;
2063         default:
2064                 break;
2065         }
2066
2067         vnic_dev_set_intr_mode(enic->vdev, VNIC_DEV_INTR_MODE_UNKNOWN);
2068 }
2069
2070 static const struct net_device_ops enic_netdev_dynamic_ops = {
2071         .ndo_open               = enic_open,
2072         .ndo_stop               = enic_stop,
2073         .ndo_start_xmit         = enic_hard_start_xmit,
2074         .ndo_get_stats          = enic_get_stats,
2075         .ndo_validate_addr      = eth_validate_addr,
2076         .ndo_set_multicast_list = enic_set_multicast_list,
2077         .ndo_set_mac_address    = enic_set_mac_address_dynamic,
2078         .ndo_change_mtu         = enic_change_mtu,
2079         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2080         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2081         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2082         .ndo_tx_timeout         = enic_tx_timeout,
2083         .ndo_set_vf_port        = enic_set_vf_port,
2084         .ndo_get_vf_port        = enic_get_vf_port,
2085 #ifdef CONFIG_NET_POLL_CONTROLLER
2086         .ndo_poll_controller    = enic_poll_controller,
2087 #endif
2088 };
2089
2090 static const struct net_device_ops enic_netdev_ops = {
2091         .ndo_open               = enic_open,
2092         .ndo_stop               = enic_stop,
2093         .ndo_start_xmit         = enic_hard_start_xmit,
2094         .ndo_get_stats          = enic_get_stats,
2095         .ndo_validate_addr      = eth_validate_addr,
2096         .ndo_set_multicast_list = enic_set_multicast_list,
2097         .ndo_set_mac_address    = enic_set_mac_address,
2098         .ndo_change_mtu         = enic_change_mtu,
2099         .ndo_vlan_rx_register   = enic_vlan_rx_register,
2100         .ndo_vlan_rx_add_vid    = enic_vlan_rx_add_vid,
2101         .ndo_vlan_rx_kill_vid   = enic_vlan_rx_kill_vid,
2102         .ndo_tx_timeout         = enic_tx_timeout,
2103 #ifdef CONFIG_NET_POLL_CONTROLLER
2104         .ndo_poll_controller    = enic_poll_controller,
2105 #endif
2106 };
2107
2108 void enic_dev_deinit(struct enic *enic)
2109 {
2110         netif_napi_del(&enic->napi);
2111         enic_free_vnic_resources(enic);
2112         enic_clear_intr_mode(enic);
2113 }
2114
2115 int enic_dev_init(struct enic *enic)
2116 {
2117         struct net_device *netdev = enic->netdev;
2118         int err;
2119
2120         /* Get vNIC configuration
2121          */
2122
2123         err = enic_get_vnic_config(enic);
2124         if (err) {
2125                 printk(KERN_ERR PFX
2126                         "Get vNIC configuration failed, aborting.\n");
2127                 return err;
2128         }
2129
2130         /* Get available resource counts
2131          */
2132
2133         enic_get_res_counts(enic);
2134
2135         /* Set interrupt mode based on resource counts and system
2136          * capabilities
2137          */
2138
2139         err = enic_set_intr_mode(enic);
2140         if (err) {
2141                 printk(KERN_ERR PFX
2142                         "Failed to set intr mode based on resource "
2143                         "counts and system capabilities, aborting.\n");
2144                 return err;
2145         }
2146
2147         /* Allocate and configure vNIC resources
2148          */
2149
2150         err = enic_alloc_vnic_resources(enic);
2151         if (err) {
2152                 printk(KERN_ERR PFX
2153                         "Failed to alloc vNIC resources, aborting.\n");
2154                 goto err_out_free_vnic_resources;
2155         }
2156
2157         enic_init_vnic_resources(enic);
2158
2159         err = enic_set_rq_alloc_buf(enic);
2160         if (err) {
2161                 printk(KERN_ERR PFX
2162                         "Failed to set RQ buffer allocator, aborting.\n");
2163                 goto err_out_free_vnic_resources;
2164         }
2165
2166         err = enic_set_niccfg(enic);
2167         if (err) {
2168                 printk(KERN_ERR PFX
2169                         "Failed to config nic, aborting.\n");
2170                 goto err_out_free_vnic_resources;
2171         }
2172
2173         switch (vnic_dev_get_intr_mode(enic->vdev)) {
2174         default:
2175                 netif_napi_add(netdev, &enic->napi, enic_poll, 64);
2176                 break;
2177         case VNIC_DEV_INTR_MODE_MSIX:
2178                 netif_napi_add(netdev, &enic->napi, enic_poll_msix, 64);
2179                 break;
2180         }
2181
2182         return 0;
2183
2184 err_out_free_vnic_resources:
2185         enic_clear_intr_mode(enic);
2186         enic_free_vnic_resources(enic);
2187
2188         return err;
2189 }
2190
2191 static void enic_iounmap(struct enic *enic)
2192 {
2193         unsigned int i;
2194
2195         for (i = 0; i < ARRAY_SIZE(enic->bar); i++)
2196                 if (enic->bar[i].vaddr)
2197                         iounmap(enic->bar[i].vaddr);
2198 }
2199
2200 static int __devinit enic_probe(struct pci_dev *pdev,
2201         const struct pci_device_id *ent)
2202 {
2203         struct net_device *netdev;
2204         struct enic *enic;
2205         int using_dac = 0;
2206         unsigned int i;
2207         int err;
2208
2209         /* Allocate net device structure and initialize.  Private
2210          * instance data is initialized to zero.
2211          */
2212
2213         netdev = alloc_etherdev(sizeof(struct enic));
2214         if (!netdev) {
2215                 printk(KERN_ERR PFX "Etherdev alloc failed, aborting.\n");
2216                 return -ENOMEM;
2217         }
2218
2219         pci_set_drvdata(pdev, netdev);
2220
2221         SET_NETDEV_DEV(netdev, &pdev->dev);
2222
2223         enic = netdev_priv(netdev);
2224         enic->netdev = netdev;
2225         enic->pdev = pdev;
2226
2227         /* Setup PCI resources
2228          */
2229
2230         err = pci_enable_device(pdev);
2231         if (err) {
2232                 printk(KERN_ERR PFX
2233                         "Cannot enable PCI device, aborting.\n");
2234                 goto err_out_free_netdev;
2235         }
2236
2237         err = pci_request_regions(pdev, DRV_NAME);
2238         if (err) {
2239                 printk(KERN_ERR PFX
2240                         "Cannot request PCI regions, aborting.\n");
2241                 goto err_out_disable_device;
2242         }
2243
2244         pci_set_master(pdev);
2245
2246         /* Query PCI controller on system for DMA addressing
2247          * limitation for the device.  Try 40-bit first, and
2248          * fail to 32-bit.
2249          */
2250
2251         err = pci_set_dma_mask(pdev, DMA_BIT_MASK(40));
2252         if (err) {
2253                 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
2254                 if (err) {
2255                         printk(KERN_ERR PFX
2256                                 "No usable DMA configuration, aborting.\n");
2257                         goto err_out_release_regions;
2258                 }
2259                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
2260                 if (err) {
2261                         printk(KERN_ERR PFX
2262                                 "Unable to obtain 32-bit DMA "
2263                                 "for consistent allocations, aborting.\n");
2264                         goto err_out_release_regions;
2265                 }
2266         } else {
2267                 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(40));
2268                 if (err) {
2269                         printk(KERN_ERR PFX
2270                                 "Unable to obtain 40-bit DMA "
2271                                 "for consistent allocations, aborting.\n");
2272                         goto err_out_release_regions;
2273                 }
2274                 using_dac = 1;
2275         }
2276
2277         /* Map vNIC resources from BAR0-5
2278          */
2279
2280         for (i = 0; i < ARRAY_SIZE(enic->bar); i++) {
2281                 if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
2282                         continue;
2283                 enic->bar[i].len = pci_resource_len(pdev, i);
2284                 enic->bar[i].vaddr = pci_iomap(pdev, i, enic->bar[i].len);
2285                 if (!enic->bar[i].vaddr) {
2286                         printk(KERN_ERR PFX
2287                                 "Cannot memory-map BAR %d, aborting.\n", i);
2288                         err = -ENODEV;
2289                         goto err_out_iounmap;
2290                 }
2291                 enic->bar[i].bus_addr = pci_resource_start(pdev, i);
2292         }
2293
2294         /* Register vNIC device
2295          */
2296
2297         enic->vdev = vnic_dev_register(NULL, enic, pdev, enic->bar,
2298                 ARRAY_SIZE(enic->bar));
2299         if (!enic->vdev) {
2300                 printk(KERN_ERR PFX
2301                         "vNIC registration failed, aborting.\n");
2302                 err = -ENODEV;
2303                 goto err_out_iounmap;
2304         }
2305
2306         /* Issue device open to get device in known state
2307          */
2308
2309         err = enic_dev_open(enic);
2310         if (err) {
2311                 printk(KERN_ERR PFX
2312                         "vNIC dev open failed, aborting.\n");
2313                 goto err_out_vnic_unregister;
2314         }
2315
2316         /* Issue device init to initialize the vnic-to-switch link.
2317          * We'll start with carrier off and wait for link UP
2318          * notification later to turn on carrier.  We don't need
2319          * to wait here for the vnic-to-switch link initialization
2320          * to complete; link UP notification is the indication that
2321          * the process is complete.
2322          */
2323
2324         netif_carrier_off(netdev);
2325
2326         if (!enic_is_dynamic(enic)) {
2327                 err = vnic_dev_init(enic->vdev, 0);
2328                 if (err) {
2329                         printk(KERN_ERR PFX
2330                                 "vNIC dev init failed, aborting.\n");
2331                         goto err_out_dev_close;
2332                 }
2333         }
2334
2335         err = enic_dev_init(enic);
2336         if (err) {
2337                 printk(KERN_ERR PFX
2338                         "Device initialization failed, aborting.\n");
2339                 goto err_out_dev_close;
2340         }
2341
2342         /* Setup notification timer, HW reset task, and locks
2343          */
2344
2345         init_timer(&enic->notify_timer);
2346         enic->notify_timer.function = enic_notify_timer;
2347         enic->notify_timer.data = (unsigned long)enic;
2348
2349         INIT_WORK(&enic->reset, enic_reset);
2350
2351         for (i = 0; i < enic->wq_count; i++)
2352                 spin_lock_init(&enic->wq_lock[i]);
2353
2354         spin_lock_init(&enic->devcmd_lock);
2355
2356         /* Register net device
2357          */
2358
2359         enic->port_mtu = enic->config.mtu;
2360         (void)enic_change_mtu(netdev, enic->port_mtu);
2361
2362         err = enic_set_mac_addr(netdev, enic->mac_addr);
2363         if (err) {
2364                 printk(KERN_ERR PFX
2365                         "Invalid MAC address, aborting.\n");
2366                 goto err_out_dev_deinit;
2367         }
2368
2369         enic->tx_coalesce_usecs = enic->config.intr_timer_usec;
2370         enic->rx_coalesce_usecs = enic->tx_coalesce_usecs;
2371
2372         if (enic_is_dynamic(enic))
2373                 netdev->netdev_ops = &enic_netdev_dynamic_ops;
2374         else
2375                 netdev->netdev_ops = &enic_netdev_ops;
2376
2377         netdev->watchdog_timeo = 2 * HZ;
2378         netdev->ethtool_ops = &enic_ethtool_ops;
2379
2380         netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
2381         if (ENIC_SETTING(enic, TXCSUM))
2382                 netdev->features |= NETIF_F_SG | NETIF_F_HW_CSUM;
2383         if (ENIC_SETTING(enic, TSO))
2384                 netdev->features |= NETIF_F_TSO |
2385                         NETIF_F_TSO6 | NETIF_F_TSO_ECN;
2386         if (ENIC_SETTING(enic, LRO))
2387                 netdev->features |= NETIF_F_LRO;
2388         if (using_dac)
2389                 netdev->features |= NETIF_F_HIGHDMA;
2390
2391         enic->csum_rx_enabled = ENIC_SETTING(enic, RXCSUM);
2392
2393         enic->lro_mgr.max_aggr = ENIC_LRO_MAX_AGGR;
2394         enic->lro_mgr.max_desc = ENIC_LRO_MAX_DESC;
2395         enic->lro_mgr.lro_arr = enic->lro_desc;
2396         enic->lro_mgr.get_skb_header = enic_get_skb_header;
2397         enic->lro_mgr.features  = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
2398         enic->lro_mgr.dev = netdev;
2399         enic->lro_mgr.ip_summed = CHECKSUM_COMPLETE;
2400         enic->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
2401
2402         err = register_netdev(netdev);
2403         if (err) {
2404                 printk(KERN_ERR PFX
2405                         "Cannot register net device, aborting.\n");
2406                 goto err_out_dev_deinit;
2407         }
2408
2409         return 0;
2410
2411 err_out_dev_deinit:
2412         enic_dev_deinit(enic);
2413 err_out_dev_close:
2414         vnic_dev_close(enic->vdev);
2415 err_out_vnic_unregister:
2416         vnic_dev_unregister(enic->vdev);
2417 err_out_iounmap:
2418         enic_iounmap(enic);
2419 err_out_release_regions:
2420         pci_release_regions(pdev);
2421 err_out_disable_device:
2422         pci_disable_device(pdev);
2423 err_out_free_netdev:
2424         pci_set_drvdata(pdev, NULL);
2425         free_netdev(netdev);
2426
2427         return err;
2428 }
2429
2430 static void __devexit enic_remove(struct pci_dev *pdev)
2431 {
2432         struct net_device *netdev = pci_get_drvdata(pdev);
2433
2434         if (netdev) {
2435                 struct enic *enic = netdev_priv(netdev);
2436
2437                 flush_scheduled_work();
2438                 unregister_netdev(netdev);
2439                 enic_dev_deinit(enic);
2440                 vnic_dev_close(enic->vdev);
2441                 vnic_dev_unregister(enic->vdev);
2442                 enic_iounmap(enic);
2443                 pci_release_regions(pdev);
2444                 pci_disable_device(pdev);
2445                 pci_set_drvdata(pdev, NULL);
2446                 free_netdev(netdev);
2447         }
2448 }
2449
2450 static struct pci_driver enic_driver = {
2451         .name = DRV_NAME,
2452         .id_table = enic_id_table,
2453         .probe = enic_probe,
2454         .remove = __devexit_p(enic_remove),
2455 };
2456
2457 static int __init enic_init_module(void)
2458 {
2459         printk(KERN_INFO PFX "%s, ver %s\n", DRV_DESCRIPTION, DRV_VERSION);
2460
2461         return pci_register_driver(&enic_driver);
2462 }
2463
2464 static void __exit enic_cleanup_module(void)
2465 {
2466         pci_unregister_driver(&enic_driver);
2467 }
2468
2469 module_init(enic_init_module);
2470 module_exit(enic_cleanup_module);