drivers/infiniband/hw/mlx4/main.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/module.h>
  35 #include <linux/init.h>
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/rtnetlink.h>
  41 #include <linux/if_vlan.h>
  42
  43 #include <rdma/ib_smi.h>
  44 #include <rdma/ib_user_verbs.h>
  45 #include <rdma/ib_addr.h>
  46
  47 #include <linux/mlx4/driver.h>
  48 #include <linux/mlx4/cmd.h>
  49
  50 #include "mlx4_ib.h"
  51 #include "user.h"
  52
  53 #define DRV_NAME        MLX4_IB_DRV_NAME
  54 #define DRV_VERSION     "1.0"
  55 #define DRV_RELDATE     "April 4, 2008"
  56
  57 MODULE_AUTHOR("Roland Dreier");
  58 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  59 MODULE_LICENSE("Dual BSD/GPL");
  60 MODULE_VERSION(DRV_VERSION);
  61
  62 int mlx4_ib_sm_guid_assign = 1;
  63 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  64 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
  65
  66 static const char mlx4_ib_version[] =
  67         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  68         DRV_VERSION " (" DRV_RELDATE ")\n";
  69
  70 struct update_gid_work {
  71         struct work_struct      work;
  72         union ib_gid            gids[128];
  73         struct mlx4_ib_dev     *dev;
  74         int                     port;
  75 };
  76
  77 static struct workqueue_struct *wq;
  78
  79 static void init_query_mad(struct ib_smp *mad)
  80 {
  81         mad->base_version  = 1;
  82         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  83         mad->class_version = 1;
  84         mad->method        = IB_MGMT_METHOD_GET;
  85 }
  86
  87 static union ib_gid zgid;
  88
  89 static int mlx4_ib_query_device(struct ib_device *ibdev,
  90                                 struct ib_device_attr *props)
  91 {
  92         struct mlx4_ib_dev *dev = to_mdev(ibdev);
  93         struct ib_smp *in_mad  = NULL;
  94         struct ib_smp *out_mad = NULL;
  95         int err = -ENOMEM;
  96
  97         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
  98         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
  99         if (!in_mad || !out_mad)
 100                 goto out;
 101
 102         init_query_mad(in_mad);
 103         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 104
 105         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 106                            1, NULL, NULL, in_mad, out_mad);
 107         if (err)
 108                 goto out;
 109
 110         memset(props, 0, sizeof *props);
 111
 112         props->fw_ver = dev->dev->caps.fw_ver;
 113         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 114                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 115                 IB_DEVICE_SYS_IMAGE_GUID                |
 116                 IB_DEVICE_RC_RNR_NAK_GEN                |
 117                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 118         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 119                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 120         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 121                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 122         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
 123                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 124         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 125                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 126         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 127                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 128         if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
 129                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 130         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 131                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 132         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 133             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 134             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 135                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 136         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 137                 props->device_cap_flags |= IB_DEVICE_XRC;
 138
 139         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 140                 0xffffff;
 141         props->vendor_part_id      = dev->dev->pdev->device;
 142         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 143         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 144
 145         props->max_mr_size         = ~0ull;
 146         props->page_size_cap       = dev->dev->caps.page_size_cap;
 147         props->max_qp              = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
 148         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 149         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 150                                          dev->dev->caps.max_rq_sg);
 151         props->max_cq              = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
 152         props->max_cqe             = dev->dev->caps.max_cqes;
 153         props->max_mr              = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
 154         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 155         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 156         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 157         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 158         props->max_srq             = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
 159         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 160         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 161         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 162         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 163         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 164                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 165         props->masked_atomic_cap   = props->atomic_cap;
 166         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 167         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 168         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 169         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 170                                            props->max_mcast_grp;
 171         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 172
 173 out:
 174         kfree(in_mad);
 175         kfree(out_mad);
 176
 177         return err;
 178 }
 179
 180 static enum rdma_link_layer
 181 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 182 {
 183         struct mlx4_dev *dev = to_mdev(device)->dev;
 184
 185         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 186                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 187 }
 188
 189 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 190                               struct ib_port_attr *props, int netw_view)
 191 {
 192         struct ib_smp *in_mad  = NULL;
 193         struct ib_smp *out_mad = NULL;
 194         int ext_active_speed;
 195         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 196         int err = -ENOMEM;
 197
 198         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 199         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 200         if (!in_mad || !out_mad)
 201                 goto out;
 202
 203         init_query_mad(in_mad);
 204         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 205         in_mad->attr_mod = cpu_to_be32(port);
 206
 207         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 208                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 209
 210         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 211                                 in_mad, out_mad);
 212         if (err)
 213                 goto out;
 214
 215
 216         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 217         props->lmc              = out_mad->data[34] & 0x7;
 218         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 219         props->sm_sl            = out_mad->data[36] & 0xf;
 220         props->state            = out_mad->data[32] & 0xf;
 221         props->phys_state       = out_mad->data[33] >> 4;
 222         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 223         if (netw_view)
 224                 props->gid_tbl_len = out_mad->data[50];
 225         else
 226                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 227         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 228         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 229         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 230         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 231         props->active_width     = out_mad->data[31] & 0xf;
 232         props->active_speed     = out_mad->data[35] >> 4;
 233         props->max_mtu          = out_mad->data[41] & 0xf;
 234         props->active_mtu       = out_mad->data[36] >> 4;
 235         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 236         props->max_vl_num       = out_mad->data[37] >> 4;
 237         props->init_type_reply  = out_mad->data[41] >> 4;
 238
 239         /* Check if extended speeds (EDR/FDR/...) are supported */
 240         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 241                 ext_active_speed = out_mad->data[62] >> 4;
 242
 243                 switch (ext_active_speed) {
 244                 case 1:
 245                         props->active_speed = IB_SPEED_FDR;
 246                         break;
 247                 case 2:
 248                         props->active_speed = IB_SPEED_EDR;
 249                         break;
 250                 }
 251         }
 252
 253         /* If reported active speed is QDR, check if is FDR-10 */
 254         if (props->active_speed == IB_SPEED_QDR) {
 255                 init_query_mad(in_mad);
 256                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 257                 in_mad->attr_mod = cpu_to_be32(port);
 258
 259                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 260                                    NULL, NULL, in_mad, out_mad);
 261                 if (err)
 262                         goto out;
 263
 264                 /* Checking LinkSpeedActive for FDR-10 */
 265                 if (out_mad->data[15] & 0x1)
 266                         props->active_speed = IB_SPEED_FDR10;
 267         }
 268
 269         /* Avoid wrong speed value returned by FW if the IB link is down. */
 270         if (props->state == IB_PORT_DOWN)
 271                  props->active_speed = IB_SPEED_SDR;
 272
 273 out:
 274         kfree(in_mad);
 275         kfree(out_mad);
 276         return err;
 277 }
 278
 279 static u8 state_to_phys_state(enum ib_port_state state)
 280 {
 281         return state == IB_PORT_ACTIVE ? 5 : 3;
 282 }
 283
 284 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 285                                struct ib_port_attr *props, int netw_view)
 286 {
 287
 288         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 289         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 290         struct net_device *ndev;
 291         enum ib_mtu tmp;
 292         struct mlx4_cmd_mailbox *mailbox;
 293         int err = 0;
 294
 295         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 296         if (IS_ERR(mailbox))
 297                 return PTR_ERR(mailbox);
 298
 299         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 300                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 301                            MLX4_CMD_WRAPPED);
 302         if (err)
 303                 goto out;
 304
 305         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 306                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 307         props->active_speed     = IB_SPEED_QDR;
 308         props->port_cap_flags   = IB_PORT_CM_SUP;
 309         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 310         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 311         props->pkey_tbl_len     = 1;
 312         props->max_mtu          = IB_MTU_4096;
 313         props->max_vl_num       = 2;
 314         props->state            = IB_PORT_DOWN;
 315         props->phys_state       = state_to_phys_state(props->state);
 316         props->active_mtu       = IB_MTU_256;
 317         spin_lock(&iboe->lock);
 318         ndev = iboe->netdevs[port - 1];
 319         if (!ndev)
 320                 goto out_unlock;
 321
 322         tmp = iboe_get_mtu(ndev->mtu);
 323         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 324
 325         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 326                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 327         props->phys_state       = state_to_phys_state(props->state);
 328 out_unlock:
 329         spin_unlock(&iboe->lock);
 330 out:
 331         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 332         return err;
 333 }
 334
 335 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 336                          struct ib_port_attr *props, int netw_view)
 337 {
 338         int err;
 339
 340         memset(props, 0, sizeof *props);
 341
 342         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 343                 ib_link_query_port(ibdev, port, props, netw_view) :
 344                                 eth_link_query_port(ibdev, port, props, netw_view);
 345
 346         return err;
 347 }
 348
 349 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 350                               struct ib_port_attr *props)
 351 {
 352         /* returns host view */
 353         return __mlx4_ib_query_port(ibdev, port, props, 0);
 354 }
 355
 356 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 357                         union ib_gid *gid, int netw_view)
 358 {
 359         struct ib_smp *in_mad  = NULL;
 360         struct ib_smp *out_mad = NULL;
 361         int err = -ENOMEM;
 362         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 363         int clear = 0;
 364         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 365
 366         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 367         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 368         if (!in_mad || !out_mad)
 369                 goto out;
 370
 371         init_query_mad(in_mad);
 372         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 373         in_mad->attr_mod = cpu_to_be32(port);
 374
 375         if (mlx4_is_mfunc(dev->dev) && netw_view)
 376                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 377
 378         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 379         if (err)
 380                 goto out;
 381
 382         memcpy(gid->raw, out_mad->data + 8, 8);
 383
 384         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 385                 if (index) {
 386                         /* For any index > 0, return the null guid */
 387                         err = 0;
 388                         clear = 1;
 389                         goto out;
 390                 }
 391         }
 392
 393         init_query_mad(in_mad);
 394         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 395         in_mad->attr_mod = cpu_to_be32(index / 8);
 396
 397         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 398                            NULL, NULL, in_mad, out_mad);
 399         if (err)
 400                 goto out;
 401
 402         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 403
 404 out:
 405         if (clear)
 406                 memset(gid->raw + 8, 0, 8);
 407         kfree(in_mad);
 408         kfree(out_mad);
 409         return err;
 410 }
 411
 412 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
 413                           union ib_gid *gid)
 414 {
 415         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 416
 417         *gid = dev->iboe.gid_table[port - 1][index];
 418
 419         return 0;
 420 }
 421
 422 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 423                              union ib_gid *gid)
 424 {
 425         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
 426                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 427         else
 428                 return iboe_query_gid(ibdev, port, index, gid);
 429 }
 430
 431 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 432                          u16 *pkey, int netw_view)
 433 {
 434         struct ib_smp *in_mad  = NULL;
 435         struct ib_smp *out_mad = NULL;
 436         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 437         int err = -ENOMEM;
 438
 439         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 440         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 441         if (!in_mad || !out_mad)
 442                 goto out;
 443
 444         init_query_mad(in_mad);
 445         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 446         in_mad->attr_mod = cpu_to_be32(index / 32);
 447
 448         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 449                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 450
 451         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 452                            in_mad, out_mad);
 453         if (err)
 454                 goto out;
 455
 456         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 457
 458 out:
 459         kfree(in_mad);
 460         kfree(out_mad);
 461         return err;
 462 }
 463
 464 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 465 {
 466         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 467 }
 468
 469 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 470                                  struct ib_device_modify *props)
 471 {
 472         struct mlx4_cmd_mailbox *mailbox;
 473         unsigned long flags;
 474
 475         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 476                 return -EOPNOTSUPP;
 477
 478         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 479                 return 0;
 480
 481         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 482                 return -EOPNOTSUPP;
 483
 484         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 485         memcpy(ibdev->node_desc, props->node_desc, 64);
 486         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 487
 488         /*
 489          * If possible, pass node desc to FW, so it can generate
 490          * a 144 trap.  If cmd fails, just ignore.
 491          */
 492         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 493         if (IS_ERR(mailbox))
 494                 return 0;
 495
 496         memset(mailbox->buf, 0, 256);
 497         memcpy(mailbox->buf, props->node_desc, 64);
 498         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 499                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 500
 501         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 502
 503         return 0;
 504 }
 505
 506 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 507                          u32 cap_mask)
 508 {
 509         struct mlx4_cmd_mailbox *mailbox;
 510         int err;
 511         u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 512
 513         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 514         if (IS_ERR(mailbox))
 515                 return PTR_ERR(mailbox);
 516
 517         memset(mailbox->buf, 0, 256);
 518
 519         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 520                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 521                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 522         } else {
 523                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 524                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 525         }
 526
 527         err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
 528                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
 529
 530         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 531         return err;
 532 }
 533
 534 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 535                                struct ib_port_modify *props)
 536 {
 537         struct ib_port_attr attr;
 538         u32 cap_mask;
 539         int err;
 540
 541         mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
 542
 543         err = mlx4_ib_query_port(ibdev, port, &attr);
 544         if (err)
 545                 goto out;
 546
 547         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 548                 ~props->clr_port_cap_mask;
 549
 550         err = mlx4_SET_PORT(to_mdev(ibdev), port,
 551                             !!(mask & IB_PORT_RESET_QKEY_CNTR),
 552                             cap_mask);
 553
 554 out:
 555         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 556         return err;
 557 }
 558
 559 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 560                                                   struct ib_udata *udata)
 561 {
 562         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 563         struct mlx4_ib_ucontext *context;
 564         struct mlx4_ib_alloc_ucontext_resp resp;
 565         int err;
 566
 567         if (!dev->ib_active)
 568                 return ERR_PTR(-EAGAIN);
 569
 570         resp.qp_tab_size      = dev->dev->caps.num_qps;
 571         resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 572         resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 573
 574         context = kmalloc(sizeof *context, GFP_KERNEL);
 575         if (!context)
 576                 return ERR_PTR(-ENOMEM);
 577
 578         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 579         if (err) {
 580                 kfree(context);
 581                 return ERR_PTR(err);
 582         }
 583
 584         INIT_LIST_HEAD(&context->db_page_list);
 585         mutex_init(&context->db_page_mutex);
 586
 587         err = ib_copy_to_udata(udata, &resp, sizeof resp);
 588         if (err) {
 589                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 590                 kfree(context);
 591                 return ERR_PTR(-EFAULT);
 592         }
 593
 594         return &context->ibucontext;
 595 }
 596
 597 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 598 {
 599         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 600
 601         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 602         kfree(context);
 603
 604         return 0;
 605 }
 606
 607 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 608 {
 609         struct mlx4_ib_dev *dev = to_mdev(context->device);
 610
 611         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 612                 return -EINVAL;
 613
 614         if (vma->vm_pgoff == 0) {
 615                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 616
 617                 if (io_remap_pfn_range(vma, vma->vm_start,
 618                                        to_mucontext(context)->uar.pfn,
 619                                        PAGE_SIZE, vma->vm_page_prot))
 620                         return -EAGAIN;
 621         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 622                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 623
 624                 if (io_remap_pfn_range(vma, vma->vm_start,
 625                                        to_mucontext(context)->uar.pfn +
 626                                        dev->dev->caps.num_uars,
 627                                        PAGE_SIZE, vma->vm_page_prot))
 628                         return -EAGAIN;
 629         } else
 630                 return -EINVAL;
 631
 632         return 0;
 633 }
 634
 635 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 636                                       struct ib_ucontext *context,
 637                                       struct ib_udata *udata)
 638 {
 639         struct mlx4_ib_pd *pd;
 640         int err;
 641
 642         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 643         if (!pd)
 644                 return ERR_PTR(-ENOMEM);
 645
 646         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 647         if (err) {
 648                 kfree(pd);
 649                 return ERR_PTR(err);
 650         }
 651
 652         if (context)
 653                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 654                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 655                         kfree(pd);
 656                         return ERR_PTR(-EFAULT);
 657                 }
 658
 659         return &pd->ibpd;
 660 }
 661
 662 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 663 {
 664         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 665         kfree(pd);
 666
 667         return 0;
 668 }
 669
 670 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 671                                           struct ib_ucontext *context,
 672                                           struct ib_udata *udata)
 673 {
 674         struct mlx4_ib_xrcd *xrcd;
 675         int err;
 676
 677         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 678                 return ERR_PTR(-ENOSYS);
 679
 680         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 681         if (!xrcd)
 682                 return ERR_PTR(-ENOMEM);
 683
 684         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 685         if (err)
 686                 goto err1;
 687
 688         xrcd->pd = ib_alloc_pd(ibdev);
 689         if (IS_ERR(xrcd->pd)) {
 690                 err = PTR_ERR(xrcd->pd);
 691                 goto err2;
 692         }
 693
 694         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
 695         if (IS_ERR(xrcd->cq)) {
 696                 err = PTR_ERR(xrcd->cq);
 697                 goto err3;
 698         }
 699
 700         return &xrcd->ibxrcd;
 701
 702 err3:
 703         ib_dealloc_pd(xrcd->pd);
 704 err2:
 705         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 706 err1:
 707         kfree(xrcd);
 708         return ERR_PTR(err);
 709 }
 710
 711 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 712 {
 713         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 714         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 715         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 716         kfree(xrcd);
 717
 718         return 0;
 719 }
 720
 721 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 722 {
 723         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 724         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 725         struct mlx4_ib_gid_entry *ge;
 726
 727         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 728         if (!ge)
 729                 return -ENOMEM;
 730
 731         ge->gid = *gid;
 732         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 733                 ge->port = mqp->port;
 734                 ge->added = 1;
 735         }
 736
 737         mutex_lock(&mqp->mutex);
 738         list_add_tail(&ge->list, &mqp->gid_list);
 739         mutex_unlock(&mqp->mutex);
 740
 741         return 0;
 742 }
 743
 744 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 745                    union ib_gid *gid)
 746 {
 747         u8 mac[6];
 748         struct net_device *ndev;
 749         int ret = 0;
 750
 751         if (!mqp->port)
 752                 return 0;
 753
 754         spin_lock(&mdev->iboe.lock);
 755         ndev = mdev->iboe.netdevs[mqp->port - 1];
 756         if (ndev)
 757                 dev_hold(ndev);
 758         spin_unlock(&mdev->iboe.lock);
 759
 760         if (ndev) {
 761                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
 762                 rtnl_lock();
 763                 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
 764                 ret = 1;
 765                 rtnl_unlock();
 766                 dev_put(ndev);
 767         }
 768
 769         return ret;
 770 }
 771
 772 struct mlx4_ib_steering {
 773         struct list_head list;
 774         u64 reg_id;
 775         union ib_gid gid;
 776 };
 777
 778 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 779 {
 780         int err;
 781         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 782         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 783         u64 reg_id;
 784         struct mlx4_ib_steering *ib_steering = NULL;
 785
 786         if (mdev->dev->caps.steering_mode ==
 787             MLX4_STEERING_MODE_DEVICE_MANAGED) {
 788                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
 789                 if (!ib_steering)
 790                         return -ENOMEM;
 791         }
 792
 793         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
 794                                     !!(mqp->flags &
 795                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
 796                                     MLX4_PROT_IB_IPV6, &reg_id);
 797         if (err)
 798                 goto err_malloc;
 799
 800         err = add_gid_entry(ibqp, gid);
 801         if (err)
 802                 goto err_add;
 803
 804         if (ib_steering) {
 805                 memcpy(ib_steering->gid.raw, gid->raw, 16);
 806                 ib_steering->reg_id = reg_id;
 807                 mutex_lock(&mqp->mutex);
 808                 list_add(&ib_steering->list, &mqp->steering_rules);
 809                 mutex_unlock(&mqp->mutex);
 810         }
 811         return 0;
 812
 813 err_add:
 814         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
 815                               MLX4_PROT_IB_IPV6, reg_id);
 816 err_malloc:
 817         kfree(ib_steering);
 818
 819         return err;
 820 }
 821
 822 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
 823 {
 824         struct mlx4_ib_gid_entry *ge;
 825         struct mlx4_ib_gid_entry *tmp;
 826         struct mlx4_ib_gid_entry *ret = NULL;
 827
 828         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
 829                 if (!memcmp(raw, ge->gid.raw, 16)) {
 830                         ret = ge;
 831                         break;
 832                 }
 833         }
 834
 835         return ret;
 836 }
 837
 838 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 839 {
 840         int err;
 841         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 842         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 843         u8 mac[6];
 844         struct net_device *ndev;
 845         struct mlx4_ib_gid_entry *ge;
 846         u64 reg_id = 0;
 847
 848         if (mdev->dev->caps.steering_mode ==
 849             MLX4_STEERING_MODE_DEVICE_MANAGED) {
 850                 struct mlx4_ib_steering *ib_steering;
 851
 852                 mutex_lock(&mqp->mutex);
 853                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
 854                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
 855                                 list_del(&ib_steering->list);
 856                                 break;
 857                         }
 858                 }
 859                 mutex_unlock(&mqp->mutex);
 860                 if (&ib_steering->list == &mqp->steering_rules) {
 861                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
 862                         return -EINVAL;
 863                 }
 864                 reg_id = ib_steering->reg_id;
 865                 kfree(ib_steering);
 866         }
 867
 868         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
 869                                     MLX4_PROT_IB_IPV6, reg_id);
 870         if (err)
 871                 return err;
 872
 873         mutex_lock(&mqp->mutex);
 874         ge = find_gid_entry(mqp, gid->raw);
 875         if (ge) {
 876                 spin_lock(&mdev->iboe.lock);
 877                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
 878                 if (ndev)
 879                         dev_hold(ndev);
 880                 spin_unlock(&mdev->iboe.lock);
 881                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
 882                 if (ndev) {
 883                         rtnl_lock();
 884                         dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
 885                         rtnl_unlock();
 886                         dev_put(ndev);
 887                 }
 888                 list_del(&ge->list);
 889                 kfree(ge);
 890         } else
 891                 pr_warn("could not find mgid entry\n");
 892
 893         mutex_unlock(&mqp->mutex);
 894
 895         return 0;
 896 }
 897
 898 static int init_node_data(struct mlx4_ib_dev *dev)
 899 {
 900         struct ib_smp *in_mad  = NULL;
 901         struct ib_smp *out_mad = NULL;
 902         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 903         int err = -ENOMEM;
 904
 905         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 906         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 907         if (!in_mad || !out_mad)
 908                 goto out;
 909
 910         init_query_mad(in_mad);
 911         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
 912         if (mlx4_is_master(dev->dev))
 913                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 914
 915         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
 916         if (err)
 917                 goto out;
 918
 919         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
 920
 921         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 922
 923         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
 924         if (err)
 925                 goto out;
 926
 927         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
 928         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
 929
 930 out:
 931         kfree(in_mad);
 932         kfree(out_mad);
 933         return err;
 934 }
 935
 936 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
 937                         char *buf)
 938 {
 939         struct mlx4_ib_dev *dev =
 940                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 941         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
 942 }
 943
 944 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
 945                            char *buf)
 946 {
 947         struct mlx4_ib_dev *dev =
 948                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 949         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
 950                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
 951                        (int) dev->dev->caps.fw_ver & 0xffff);
 952 }
 953
 954 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
 955                         char *buf)
 956 {
 957         struct mlx4_ib_dev *dev =
 958                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 959         return sprintf(buf, "%x\n", dev->dev->rev_id);
 960 }
 961
 962 static ssize_t show_board(struct device *device, struct device_attribute *attr,
 963                           char *buf)
 964 {
 965         struct mlx4_ib_dev *dev =
 966                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 967         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
 968                        dev->dev->board_id);
 969 }
 970
 971 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
 972 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
 973 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
 974 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
 975
 976 static struct device_attribute *mlx4_class_attributes[] = {
 977         &dev_attr_hw_rev,
 978         &dev_attr_fw_ver,
 979         &dev_attr_hca_type,
 980         &dev_attr_board_id
 981 };
 982
 983 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
 984 {
 985         memcpy(eui, dev->dev_addr, 3);
 986         memcpy(eui + 5, dev->dev_addr + 3, 3);
 987         if (vlan_id < 0x1000) {
 988                 eui[3] = vlan_id >> 8;
 989                 eui[4] = vlan_id & 0xff;
 990         } else {
 991                 eui[3] = 0xff;
 992                 eui[4] = 0xfe;
 993         }
 994         eui[0] ^= 2;
 995 }
 996
 997 static void update_gids_task(struct work_struct *work)
 998 {
 999         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1000         struct mlx4_cmd_mailbox *mailbox;
1001         union ib_gid *gids;
1002         int err;
1003         struct mlx4_dev *dev = gw->dev->dev;
1004
1005         mailbox = mlx4_alloc_cmd_mailbox(dev);
1006         if (IS_ERR(mailbox)) {
1007                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1008                 return;
1009         }
1010
1011         gids = mailbox->buf;
1012         memcpy(gids, gw->gids, sizeof gw->gids);
1013
1014         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1015                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1016                        MLX4_CMD_WRAPPED);
1017         if (err)
1018                 pr_warn("set port command failed\n");
1019         else {
1020                 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1021                 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1022         }
1023
1024         mlx4_free_cmd_mailbox(dev, mailbox);
1025         kfree(gw);
1026 }
1027
1028 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1029 {
1030         struct net_device *ndev = dev->iboe.netdevs[port - 1];
1031         struct update_gid_work *work;
1032         struct net_device *tmp;
1033         int i;
1034         u8 *hits;
1035         int ret;
1036         union ib_gid gid;
1037         int free;
1038         int found;
1039         int need_update = 0;
1040         u16 vid;
1041
1042         work = kzalloc(sizeof *work, GFP_ATOMIC);
1043         if (!work)
1044                 return -ENOMEM;
1045
1046         hits = kzalloc(128, GFP_ATOMIC);
1047         if (!hits) {
1048                 ret = -ENOMEM;
1049                 goto out;
1050         }
1051
1052         rcu_read_lock();
1053         for_each_netdev_rcu(&init_net, tmp) {
1054                 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1055                         gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1056                         vid = rdma_vlan_dev_vlan_id(tmp);
1057                         mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1058                         found = 0;
1059                         free = -1;
1060                         for (i = 0; i < 128; ++i) {
1061                                 if (free < 0 &&
1062                                     !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1063                                         free = i;
1064                                 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1065                                         hits[i] = 1;
1066                                         found = 1;
1067                                         break;
1068                                 }
1069                         }
1070
1071                         if (!found) {
1072                                 if (tmp == ndev &&
1073                                     (memcmp(&dev->iboe.gid_table[port - 1][0],
1074                                             &gid, sizeof gid) ||
1075                                      !memcmp(&dev->iboe.gid_table[port - 1][0],
1076                                              &zgid, sizeof gid))) {
1077                                         dev->iboe.gid_table[port - 1][0] = gid;
1078                                         ++need_update;
1079                                         hits[0] = 1;
1080                                 } else if (free >= 0) {
1081                                         dev->iboe.gid_table[port - 1][free] = gid;
1082                                         hits[free] = 1;
1083                                         ++need_update;
1084                                 }
1085                         }
1086                 }
1087         }
1088         rcu_read_unlock();
1089
1090         for (i = 0; i < 128; ++i)
1091                 if (!hits[i]) {
1092                         if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1093                                 ++need_update;
1094                         dev->iboe.gid_table[port - 1][i] = zgid;
1095                 }
1096
1097         if (need_update) {
1098                 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1099                 INIT_WORK(&work->work, update_gids_task);
1100                 work->port = port;
1101                 work->dev = dev;
1102                 queue_work(wq, &work->work);
1103         } else
1104                 kfree(work);
1105
1106         kfree(hits);
1107         return 0;
1108
1109 out:
1110         kfree(work);
1111         return ret;
1112 }
1113
1114 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1115 {
1116         switch (event) {
1117         case NETDEV_UP:
1118         case NETDEV_CHANGEADDR:
1119                 update_ipv6_gids(dev, port, 0);
1120                 break;
1121
1122         case NETDEV_DOWN:
1123                 update_ipv6_gids(dev, port, 1);
1124                 dev->iboe.netdevs[port - 1] = NULL;
1125         }
1126 }
1127
1128 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1129 {
1130         update_ipv6_gids(dev, port, 0);
1131 }
1132
1133 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1134 {
1135         update_ipv6_gids(dev, port, 1);
1136 }
1137
1138 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1139                                 void *ptr)
1140 {
1141         struct net_device *dev = ptr;
1142         struct mlx4_ib_dev *ibdev;
1143         struct net_device *oldnd;
1144         struct mlx4_ib_iboe *iboe;
1145         int port;
1146
1147         if (!net_eq(dev_net(dev), &init_net))
1148                 return NOTIFY_DONE;
1149
1150         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1151         iboe = &ibdev->iboe;
1152
1153         spin_lock(&iboe->lock);
1154         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1155                 oldnd = iboe->netdevs[port - 1];
1156                 iboe->netdevs[port - 1] =
1157                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1158                 if (oldnd != iboe->netdevs[port - 1]) {
1159                         if (iboe->netdevs[port - 1])
1160                                 netdev_added(ibdev, port);
1161                         else
1162                                 netdev_removed(ibdev, port);
1163                 }
1164         }
1165
1166         if (dev == iboe->netdevs[0] ||
1167             (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1168                 handle_en_event(ibdev, 1, event);
1169         else if (dev == iboe->netdevs[1]
1170                  || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1171                 handle_en_event(ibdev, 2, event);
1172
1173         spin_unlock(&iboe->lock);
1174
1175         return NOTIFY_DONE;
1176 }
1177
1178 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1179 {
1180         int port;
1181         int slave;
1182         int i;
1183
1184         if (mlx4_is_master(ibdev->dev)) {
1185                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1186                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1187                                 for (i = 0;
1188                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1189                                      ++i) {
1190                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1191                                         /* master has the identity virt2phys pkey mapping */
1192                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1193                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1194                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1195                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1196                                 }
1197                         }
1198                 }
1199                 /* initialize pkey cache */
1200                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1201                         for (i = 0;
1202                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1203                              ++i)
1204                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1205                                         (i) ? 0 : 0xFFFF;
1206                 }
1207         }
1208 }
1209
1210 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1211 {
1212         char name[32];
1213         int eq_per_port = 0;
1214         int added_eqs = 0;
1215         int total_eqs = 0;
1216         int i, j, eq;
1217
1218         /* Legacy mode or comp_pool is not large enough */
1219         if (dev->caps.comp_pool == 0 ||
1220             dev->caps.num_ports > dev->caps.comp_pool)
1221                 return;
1222
1223         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1224                                         dev->caps.num_ports);
1225
1226         /* Init eq table */
1227         added_eqs = 0;
1228         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1229                 added_eqs += eq_per_port;
1230
1231         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1232
1233         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1234         if (!ibdev->eq_table)
1235                 return;
1236
1237         ibdev->eq_added = added_eqs;
1238
1239         eq = 0;
1240         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1241                 for (j = 0; j < eq_per_port; j++) {
1242                         sprintf(name, "mlx4-ib-%d-%d@%s",
1243                                 i, j, dev->pdev->bus->name);
1244                         /* Set IRQ for specific name (per ring) */
1245                         if (mlx4_assign_eq(dev, name, NULL,
1246                                            &ibdev->eq_table[eq])) {
1247                                 /* Use legacy (same as mlx4_en driver) */
1248                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1249                                 ibdev->eq_table[eq] =
1250                                         (eq % dev->caps.num_comp_vectors);
1251                         }
1252                         eq++;
1253                 }
1254         }
1255
1256         /* Fill the reset of the vector with legacy EQ */
1257         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1258                 ibdev->eq_table[eq++] = i;
1259
1260         /* Advertise the new number of EQs to clients */
1261         ibdev->ib_dev.num_comp_vectors = total_eqs;
1262 }
1263
1264 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1265 {
1266         int i;
1267
1268         /* no additional eqs were added */
1269         if (!ibdev->eq_table)
1270                 return;
1271
1272         /* Reset the advertised EQ number */
1273         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1274
1275         /* Free only the added eqs */
1276         for (i = 0; i < ibdev->eq_added; i++) {
1277                 /* Don't free legacy eqs if used */
1278                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1279                         continue;
1280                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1281         }
1282
1283         kfree(ibdev->eq_table);
1284 }
1285
1286 static void *mlx4_ib_add(struct mlx4_dev *dev)
1287 {
1288         struct mlx4_ib_dev *ibdev;
1289         int num_ports = 0;
1290         int i, j;
1291         int err;
1292         struct mlx4_ib_iboe *iboe;
1293
1294         pr_info_once("%s", mlx4_ib_version);
1295
1296         mlx4_foreach_non_ib_transport_port(i, dev)
1297                 num_ports++;
1298
1299         if (mlx4_is_mfunc(dev) && num_ports) {
1300                 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1301                 return NULL;
1302         }
1303
1304         num_ports = 0;
1305         mlx4_foreach_ib_transport_port(i, dev)
1306                 num_ports++;
1307
1308         /* No point in registering a device with no ports... */
1309         if (num_ports == 0)
1310                 return NULL;
1311
1312         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1313         if (!ibdev) {
1314                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1315                 return NULL;
1316         }
1317
1318         iboe = &ibdev->iboe;
1319
1320         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1321                 goto err_dealloc;
1322
1323         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1324                 goto err_pd;
1325
1326         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1327                                  PAGE_SIZE);
1328         if (!ibdev->uar_map)
1329                 goto err_uar;
1330         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1331
1332         ibdev->dev = dev;
1333
1334         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1335         ibdev->ib_dev.owner             = THIS_MODULE;
1336         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1337         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1338         ibdev->num_ports                = num_ports;
1339         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1340         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1341         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1342
1343         ibdev->ib_dev.uverbs_abi_ver    = MLX4_IB_UVERBS_ABI_VERSION;
1344         ibdev->ib_dev.uverbs_cmd_mask   =
1345                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1346                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1347                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1348                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1349                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1350                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1351                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1352                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1353                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1354                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1355                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1356                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1357                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1358                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1359                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1360                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1361                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1362                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1363                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1364                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1365                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1366                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
1367                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1368
1369         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1370         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1371         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1372         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1373         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1374         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1375         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1376         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1377         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1378         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1379         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1380         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1381         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1382         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1383         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1384         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1385         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1386         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1387         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1388         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1389         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1390         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1391         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1392         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1393         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1394         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1395         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1396         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1397         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1398         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1399         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1400         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
1401         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
1402         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
1403         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
1404         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1405         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1406         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1407         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
1408         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
1409         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
1410
1411         if (!mlx4_is_slave(ibdev->dev)) {
1412                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
1413                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
1414                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
1415                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
1416         }
1417
1418         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1419                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1420                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1421                 ibdev->ib_dev.uverbs_cmd_mask |=
1422                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1423                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1424         }
1425
1426         mlx4_ib_alloc_eqs(dev, ibdev);
1427
1428         spin_lock_init(&iboe->lock);
1429
1430         if (init_node_data(ibdev))
1431                 goto err_map;
1432
1433         for (i = 0; i < ibdev->num_ports; ++i) {
1434                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1435                                                 IB_LINK_LAYER_ETHERNET) {
1436                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1437                         if (err)
1438                                 ibdev->counters[i] = -1;
1439                 } else
1440                                 ibdev->counters[i] = -1;
1441         }
1442
1443         spin_lock_init(&ibdev->sm_lock);
1444         mutex_init(&ibdev->cap_mask_mutex);
1445
1446         if (ib_register_device(&ibdev->ib_dev, NULL))
1447                 goto err_counter;
1448
1449         if (mlx4_ib_mad_init(ibdev))
1450                 goto err_reg;
1451
1452         if (mlx4_ib_init_sriov(ibdev))
1453                 goto err_mad;
1454
1455         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1456                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1457                 err = register_netdevice_notifier(&iboe->nb);
1458                 if (err)
1459                         goto err_sriov;
1460         }
1461
1462         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1463                 if (device_create_file(&ibdev->ib_dev.dev,
1464                                        mlx4_class_attributes[j]))
1465                         goto err_notif;
1466         }
1467
1468         ibdev->ib_active = true;
1469
1470         if (mlx4_is_mfunc(ibdev->dev))
1471                 init_pkeys(ibdev);
1472
1473         return ibdev;
1474
1475 err_notif:
1476         if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1477                 pr_warn("failure unregistering notifier\n");
1478         flush_workqueue(wq);
1479
1480 err_sriov:
1481         mlx4_ib_close_sriov(ibdev);
1482
1483 err_mad:
1484         mlx4_ib_mad_cleanup(ibdev);
1485
1486 err_reg:
1487         ib_unregister_device(&ibdev->ib_dev);
1488
1489 err_counter:
1490         for (; i; --i)
1491                 if (ibdev->counters[i - 1] != -1)
1492                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1493
1494 err_map:
1495         iounmap(ibdev->uar_map);
1496
1497 err_uar:
1498         mlx4_uar_free(dev, &ibdev->priv_uar);
1499
1500 err_pd:
1501         mlx4_pd_free(dev, ibdev->priv_pdn);
1502
1503 err_dealloc:
1504         ib_dealloc_device(&ibdev->ib_dev);
1505
1506         return NULL;
1507 }
1508
1509 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1510 {
1511         struct mlx4_ib_dev *ibdev = ibdev_ptr;
1512         int p;
1513
1514         mlx4_ib_close_sriov(ibdev);
1515         mlx4_ib_mad_cleanup(ibdev);
1516         ib_unregister_device(&ibdev->ib_dev);
1517         if (ibdev->iboe.nb.notifier_call) {
1518                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1519                         pr_warn("failure unregistering notifier\n");
1520                 ibdev->iboe.nb.notifier_call = NULL;
1521         }
1522         iounmap(ibdev->uar_map);
1523         for (p = 0; p < ibdev->num_ports; ++p)
1524                 if (ibdev->counters[p] != -1)
1525                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1526         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1527                 mlx4_CLOSE_PORT(dev, p);
1528
1529         mlx4_ib_free_eqs(dev, ibdev);
1530
1531         mlx4_uar_free(dev, &ibdev->priv_uar);
1532         mlx4_pd_free(dev, ibdev->priv_pdn);
1533         ib_dealloc_device(&ibdev->ib_dev);
1534 }
1535
1536 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1537 {
1538         struct mlx4_ib_demux_work **dm = NULL;
1539         struct mlx4_dev *dev = ibdev->dev;
1540         int i;
1541         unsigned long flags;
1542
1543         if (!mlx4_is_master(dev))
1544                 return;
1545
1546         dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1547         if (!dm) {
1548                 pr_err("failed to allocate memory for tunneling qp update\n");
1549                 goto out;
1550         }
1551
1552         for (i = 0; i < dev->caps.num_ports; i++) {
1553                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1554                 if (!dm[i]) {
1555                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
1556                         for (i = 0; i < dev->caps.num_ports; i++) {
1557                                 if (dm[i])
1558                                         kfree(dm[i]);
1559                         }
1560                         goto out;
1561                 }
1562         }
1563         /* initialize or tear down tunnel QPs for the slave */
1564         for (i = 0; i < dev->caps.num_ports; i++) {
1565                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1566                 dm[i]->port = i + 1;
1567                 dm[i]->slave = slave;
1568                 dm[i]->do_init = do_init;
1569                 dm[i]->dev = ibdev;
1570                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1571                 if (!ibdev->sriov.is_going_down)
1572                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1573                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1574         }
1575 out:
1576         if (dm)
1577                 kfree(dm);
1578         return;
1579 }
1580
1581 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1582                           enum mlx4_dev_event event, unsigned long param)
1583 {
1584         struct ib_event ibev;
1585         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1586         struct mlx4_eqe *eqe = NULL;
1587         struct ib_event_work *ew;
1588         int p = 0;
1589
1590         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1591                 eqe = (struct mlx4_eqe *)param;
1592         else
1593                 p = (int) param;
1594
1595         switch (event) {
1596         case MLX4_DEV_EVENT_PORT_UP:
1597                 if (p > ibdev->num_ports)
1598                         return;
1599                 if (mlx4_is_master(dev) &&
1600                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1601                         IB_LINK_LAYER_INFINIBAND) {
1602                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
1603                 }
1604                 ibev.event = IB_EVENT_PORT_ACTIVE;
1605                 break;
1606
1607         case MLX4_DEV_EVENT_PORT_DOWN:
1608                 if (p > ibdev->num_ports)
1609                         return;
1610                 ibev.event = IB_EVENT_PORT_ERR;
1611                 break;
1612
1613         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1614                 ibdev->ib_active = false;
1615                 ibev.event = IB_EVENT_DEVICE_FATAL;
1616                 break;
1617
1618         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1619                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1620                 if (!ew) {
1621                         pr_err("failed to allocate memory for events work\n");
1622                         break;
1623                 }
1624
1625                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1626                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1627                 ew->ib_dev = ibdev;
1628                 /* need to queue only for port owner, which uses GEN_EQE */
1629                 if (mlx4_is_master(dev))
1630                         queue_work(wq, &ew->work);
1631                 else
1632                         handle_port_mgmt_change_event(&ew->work);
1633                 return;
1634
1635         case MLX4_DEV_EVENT_SLAVE_INIT:
1636                 /* here, p is the slave id */
1637                 do_slave_init(ibdev, p, 1);
1638                 return;
1639
1640         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1641                 /* here, p is the slave id */
1642                 do_slave_init(ibdev, p, 0);
1643                 return;
1644
1645         default:
1646                 return;
1647         }
1648
1649         ibev.device           = ibdev_ptr;
1650         ibev.element.port_num = (u8) p;
1651
1652         ib_dispatch_event(&ibev);
1653 }
1654
1655 static struct mlx4_interface mlx4_ib_interface = {
1656         .add            = mlx4_ib_add,
1657         .remove         = mlx4_ib_remove,
1658         .event          = mlx4_ib_event,
1659         .protocol       = MLX4_PROT_IB_IPV6
1660 };
1661
1662 static int __init mlx4_ib_init(void)
1663 {
1664         int err;
1665
1666         wq = create_singlethread_workqueue("mlx4_ib");
1667         if (!wq)
1668                 return -ENOMEM;
1669
1670         err = mlx4_ib_mcg_init();
1671         if (err)
1672                 goto clean_wq;
1673
1674         err = mlx4_register_interface(&mlx4_ib_interface);
1675         if (err)
1676                 goto clean_mcg;
1677
1678         return 0;
1679
1680 clean_mcg:
1681         mlx4_ib_mcg_destroy();
1682
1683 clean_wq:
1684         destroy_workqueue(wq);
1685         return err;
1686 }
1687
1688 static void __exit mlx4_ib_cleanup(void)
1689 {
1690         mlx4_unregister_interface(&mlx4_ib_interface);
1691         mlx4_ib_mcg_destroy();
1692         destroy_workqueue(wq);
1693 }
1694
1695 module_init(mlx4_ib_init);
1696 module_exit(mlx4_ib_cleanup);