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      = be16_to_cpup((__be16 *) (out_mad->data + 30));
 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         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 482         memcpy(ibdev->node_desc, props->node_desc, 64);
 483         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 484
 485         /*
 486          * If possible, pass node desc to FW, so it can generate
 487          * a 144 trap.  If cmd fails, just ignore.
 488          */
 489         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 490         if (IS_ERR(mailbox))
 491                 return 0;
 492
 493         memset(mailbox->buf, 0, 256);
 494         memcpy(mailbox->buf, props->node_desc, 64);
 495         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 496                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_WRAPPED);
 497
 498         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 499
 500         return 0;
 501 }
 502
 503 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 504                          u32 cap_mask)
 505 {
 506         struct mlx4_cmd_mailbox *mailbox;
 507         int err;
 508         u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 509
 510         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 511         if (IS_ERR(mailbox))
 512                 return PTR_ERR(mailbox);
 513
 514         memset(mailbox->buf, 0, 256);
 515
 516         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 517                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 518                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 519         } else {
 520                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 521                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 522         }
 523
 524         err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
 525                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
 526
 527         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 528         return err;
 529 }
 530
 531 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 532                                struct ib_port_modify *props)
 533 {
 534         struct ib_port_attr attr;
 535         u32 cap_mask;
 536         int err;
 537
 538         mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
 539
 540         err = mlx4_ib_query_port(ibdev, port, &attr);
 541         if (err)
 542                 goto out;
 543
 544         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 545                 ~props->clr_port_cap_mask;
 546
 547         err = mlx4_SET_PORT(to_mdev(ibdev), port,
 548                             !!(mask & IB_PORT_RESET_QKEY_CNTR),
 549                             cap_mask);
 550
 551 out:
 552         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 553         return err;
 554 }
 555
 556 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 557                                                   struct ib_udata *udata)
 558 {
 559         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 560         struct mlx4_ib_ucontext *context;
 561         struct mlx4_ib_alloc_ucontext_resp resp;
 562         int err;
 563
 564         if (!dev->ib_active)
 565                 return ERR_PTR(-EAGAIN);
 566
 567         resp.qp_tab_size      = dev->dev->caps.num_qps;
 568         resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 569         resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 570
 571         context = kmalloc(sizeof *context, GFP_KERNEL);
 572         if (!context)
 573                 return ERR_PTR(-ENOMEM);
 574
 575         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 576         if (err) {
 577                 kfree(context);
 578                 return ERR_PTR(err);
 579         }
 580
 581         INIT_LIST_HEAD(&context->db_page_list);
 582         mutex_init(&context->db_page_mutex);
 583
 584         err = ib_copy_to_udata(udata, &resp, sizeof resp);
 585         if (err) {
 586                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 587                 kfree(context);
 588                 return ERR_PTR(-EFAULT);
 589         }
 590
 591         return &context->ibucontext;
 592 }
 593
 594 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 595 {
 596         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 597
 598         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 599         kfree(context);
 600
 601         return 0;
 602 }
 603
 604 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 605 {
 606         struct mlx4_ib_dev *dev = to_mdev(context->device);
 607
 608         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 609                 return -EINVAL;
 610
 611         if (vma->vm_pgoff == 0) {
 612                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 613
 614                 if (io_remap_pfn_range(vma, vma->vm_start,
 615                                        to_mucontext(context)->uar.pfn,
 616                                        PAGE_SIZE, vma->vm_page_prot))
 617                         return -EAGAIN;
 618         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 619                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 620
 621                 if (io_remap_pfn_range(vma, vma->vm_start,
 622                                        to_mucontext(context)->uar.pfn +
 623                                        dev->dev->caps.num_uars,
 624                                        PAGE_SIZE, vma->vm_page_prot))
 625                         return -EAGAIN;
 626         } else
 627                 return -EINVAL;
 628
 629         return 0;
 630 }
 631
 632 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 633                                       struct ib_ucontext *context,
 634                                       struct ib_udata *udata)
 635 {
 636         struct mlx4_ib_pd *pd;
 637         int err;
 638
 639         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 640         if (!pd)
 641                 return ERR_PTR(-ENOMEM);
 642
 643         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 644         if (err) {
 645                 kfree(pd);
 646                 return ERR_PTR(err);
 647         }
 648
 649         if (context)
 650                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 651                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 652                         kfree(pd);
 653                         return ERR_PTR(-EFAULT);
 654                 }
 655
 656         return &pd->ibpd;
 657 }
 658
 659 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 660 {
 661         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 662         kfree(pd);
 663
 664         return 0;
 665 }
 666
 667 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 668                                           struct ib_ucontext *context,
 669                                           struct ib_udata *udata)
 670 {
 671         struct mlx4_ib_xrcd *xrcd;
 672         int err;
 673
 674         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 675                 return ERR_PTR(-ENOSYS);
 676
 677         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 678         if (!xrcd)
 679                 return ERR_PTR(-ENOMEM);
 680
 681         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 682         if (err)
 683                 goto err1;
 684
 685         xrcd->pd = ib_alloc_pd(ibdev);
 686         if (IS_ERR(xrcd->pd)) {
 687                 err = PTR_ERR(xrcd->pd);
 688                 goto err2;
 689         }
 690
 691         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
 692         if (IS_ERR(xrcd->cq)) {
 693                 err = PTR_ERR(xrcd->cq);
 694                 goto err3;
 695         }
 696
 697         return &xrcd->ibxrcd;
 698
 699 err3:
 700         ib_dealloc_pd(xrcd->pd);
 701 err2:
 702         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 703 err1:
 704         kfree(xrcd);
 705         return ERR_PTR(err);
 706 }
 707
 708 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 709 {
 710         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 711         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 712         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 713         kfree(xrcd);
 714
 715         return 0;
 716 }
 717
 718 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 719 {
 720         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 721         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 722         struct mlx4_ib_gid_entry *ge;
 723
 724         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 725         if (!ge)
 726                 return -ENOMEM;
 727
 728         ge->gid = *gid;
 729         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 730                 ge->port = mqp->port;
 731                 ge->added = 1;
 732         }
 733
 734         mutex_lock(&mqp->mutex);
 735         list_add_tail(&ge->list, &mqp->gid_list);
 736         mutex_unlock(&mqp->mutex);
 737
 738         return 0;
 739 }
 740
 741 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 742                    union ib_gid *gid)
 743 {
 744         u8 mac[6];
 745         struct net_device *ndev;
 746         int ret = 0;
 747
 748         if (!mqp->port)
 749                 return 0;
 750
 751         spin_lock(&mdev->iboe.lock);
 752         ndev = mdev->iboe.netdevs[mqp->port - 1];
 753         if (ndev)
 754                 dev_hold(ndev);
 755         spin_unlock(&mdev->iboe.lock);
 756
 757         if (ndev) {
 758                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
 759                 rtnl_lock();
 760                 dev_mc_add(mdev->iboe.netdevs[mqp->port - 1], mac);
 761                 ret = 1;
 762                 rtnl_unlock();
 763                 dev_put(ndev);
 764         }
 765
 766         return ret;
 767 }
 768
 769 struct mlx4_ib_steering {
 770         struct list_head list;
 771         u64 reg_id;
 772         union ib_gid gid;
 773 };
 774
 775 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 776 {
 777         int err;
 778         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 779         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 780         u64 reg_id;
 781         struct mlx4_ib_steering *ib_steering = NULL;
 782
 783         if (mdev->dev->caps.steering_mode ==
 784             MLX4_STEERING_MODE_DEVICE_MANAGED) {
 785                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
 786                 if (!ib_steering)
 787                         return -ENOMEM;
 788         }
 789
 790         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
 791                                     !!(mqp->flags &
 792                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
 793                                     MLX4_PROT_IB_IPV6, &reg_id);
 794         if (err)
 795                 goto err_malloc;
 796
 797         err = add_gid_entry(ibqp, gid);
 798         if (err)
 799                 goto err_add;
 800
 801         if (ib_steering) {
 802                 memcpy(ib_steering->gid.raw, gid->raw, 16);
 803                 ib_steering->reg_id = reg_id;
 804                 mutex_lock(&mqp->mutex);
 805                 list_add(&ib_steering->list, &mqp->steering_rules);
 806                 mutex_unlock(&mqp->mutex);
 807         }
 808         return 0;
 809
 810 err_add:
 811         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
 812                               MLX4_PROT_IB_IPV6, reg_id);
 813 err_malloc:
 814         kfree(ib_steering);
 815
 816         return err;
 817 }
 818
 819 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
 820 {
 821         struct mlx4_ib_gid_entry *ge;
 822         struct mlx4_ib_gid_entry *tmp;
 823         struct mlx4_ib_gid_entry *ret = NULL;
 824
 825         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
 826                 if (!memcmp(raw, ge->gid.raw, 16)) {
 827                         ret = ge;
 828                         break;
 829                 }
 830         }
 831
 832         return ret;
 833 }
 834
 835 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 836 {
 837         int err;
 838         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 839         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 840         u8 mac[6];
 841         struct net_device *ndev;
 842         struct mlx4_ib_gid_entry *ge;
 843         u64 reg_id = 0;
 844
 845         if (mdev->dev->caps.steering_mode ==
 846             MLX4_STEERING_MODE_DEVICE_MANAGED) {
 847                 struct mlx4_ib_steering *ib_steering;
 848
 849                 mutex_lock(&mqp->mutex);
 850                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
 851                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
 852                                 list_del(&ib_steering->list);
 853                                 break;
 854                         }
 855                 }
 856                 mutex_unlock(&mqp->mutex);
 857                 if (&ib_steering->list == &mqp->steering_rules) {
 858                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
 859                         return -EINVAL;
 860                 }
 861                 reg_id = ib_steering->reg_id;
 862                 kfree(ib_steering);
 863         }
 864
 865         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
 866                                     MLX4_PROT_IB_IPV6, reg_id);
 867         if (err)
 868                 return err;
 869
 870         mutex_lock(&mqp->mutex);
 871         ge = find_gid_entry(mqp, gid->raw);
 872         if (ge) {
 873                 spin_lock(&mdev->iboe.lock);
 874                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
 875                 if (ndev)
 876                         dev_hold(ndev);
 877                 spin_unlock(&mdev->iboe.lock);
 878                 rdma_get_mcast_mac((struct in6_addr *)gid, mac);
 879                 if (ndev) {
 880                         rtnl_lock();
 881                         dev_mc_del(mdev->iboe.netdevs[ge->port - 1], mac);
 882                         rtnl_unlock();
 883                         dev_put(ndev);
 884                 }
 885                 list_del(&ge->list);
 886                 kfree(ge);
 887         } else
 888                 pr_warn("could not find mgid entry\n");
 889
 890         mutex_unlock(&mqp->mutex);
 891
 892         return 0;
 893 }
 894
 895 static int init_node_data(struct mlx4_ib_dev *dev)
 896 {
 897         struct ib_smp *in_mad  = NULL;
 898         struct ib_smp *out_mad = NULL;
 899         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 900         int err = -ENOMEM;
 901
 902         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 903         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 904         if (!in_mad || !out_mad)
 905                 goto out;
 906
 907         init_query_mad(in_mad);
 908         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
 909         if (mlx4_is_master(dev->dev))
 910                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 911
 912         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
 913         if (err)
 914                 goto out;
 915
 916         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
 917
 918         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 919
 920         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
 921         if (err)
 922                 goto out;
 923
 924         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
 925
 926 out:
 927         kfree(in_mad);
 928         kfree(out_mad);
 929         return err;
 930 }
 931
 932 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
 933                         char *buf)
 934 {
 935         struct mlx4_ib_dev *dev =
 936                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 937         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
 938 }
 939
 940 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
 941                            char *buf)
 942 {
 943         struct mlx4_ib_dev *dev =
 944                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 945         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
 946                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
 947                        (int) dev->dev->caps.fw_ver & 0xffff);
 948 }
 949
 950 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
 951                         char *buf)
 952 {
 953         struct mlx4_ib_dev *dev =
 954                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 955         return sprintf(buf, "%x\n", dev->dev->rev_id);
 956 }
 957
 958 static ssize_t show_board(struct device *device, struct device_attribute *attr,
 959                           char *buf)
 960 {
 961         struct mlx4_ib_dev *dev =
 962                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
 963         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
 964                        dev->dev->board_id);
 965 }
 966
 967 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
 968 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
 969 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
 970 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
 971
 972 static struct device_attribute *mlx4_class_attributes[] = {
 973         &dev_attr_hw_rev,
 974         &dev_attr_fw_ver,
 975         &dev_attr_hca_type,
 976         &dev_attr_board_id
 977 };
 978
 979 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id, struct net_device *dev)
 980 {
 981         memcpy(eui, dev->dev_addr, 3);
 982         memcpy(eui + 5, dev->dev_addr + 3, 3);
 983         if (vlan_id < 0x1000) {
 984                 eui[3] = vlan_id >> 8;
 985                 eui[4] = vlan_id & 0xff;
 986         } else {
 987                 eui[3] = 0xff;
 988                 eui[4] = 0xfe;
 989         }
 990         eui[0] ^= 2;
 991 }
 992
 993 static void update_gids_task(struct work_struct *work)
 994 {
 995         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
 996         struct mlx4_cmd_mailbox *mailbox;
 997         union ib_gid *gids;
 998         int err;
 999         struct mlx4_dev *dev = gw->dev->dev;
1000
1001         mailbox = mlx4_alloc_cmd_mailbox(dev);
1002         if (IS_ERR(mailbox)) {
1003                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1004                 return;
1005         }
1006
1007         gids = mailbox->buf;
1008         memcpy(gids, gw->gids, sizeof gw->gids);
1009
1010         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1011                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1012                        MLX4_CMD_NATIVE);
1013         if (err)
1014                 pr_warn("set port command failed\n");
1015         else {
1016                 memcpy(gw->dev->iboe.gid_table[gw->port - 1], gw->gids, sizeof gw->gids);
1017                 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1018         }
1019
1020         mlx4_free_cmd_mailbox(dev, mailbox);
1021         kfree(gw);
1022 }
1023
1024 static int update_ipv6_gids(struct mlx4_ib_dev *dev, int port, int clear)
1025 {
1026         struct net_device *ndev = dev->iboe.netdevs[port - 1];
1027         struct update_gid_work *work;
1028         struct net_device *tmp;
1029         int i;
1030         u8 *hits;
1031         int ret;
1032         union ib_gid gid;
1033         int free;
1034         int found;
1035         int need_update = 0;
1036         u16 vid;
1037
1038         work = kzalloc(sizeof *work, GFP_ATOMIC);
1039         if (!work)
1040                 return -ENOMEM;
1041
1042         hits = kzalloc(128, GFP_ATOMIC);
1043         if (!hits) {
1044                 ret = -ENOMEM;
1045                 goto out;
1046         }
1047
1048         rcu_read_lock();
1049         for_each_netdev_rcu(&init_net, tmp) {
1050                 if (ndev && (tmp == ndev || rdma_vlan_dev_real_dev(tmp) == ndev)) {
1051                         gid.global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1052                         vid = rdma_vlan_dev_vlan_id(tmp);
1053                         mlx4_addrconf_ifid_eui48(&gid.raw[8], vid, ndev);
1054                         found = 0;
1055                         free = -1;
1056                         for (i = 0; i < 128; ++i) {
1057                                 if (free < 0 &&
1058                                     !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1059                                         free = i;
1060                                 if (!memcmp(&dev->iboe.gid_table[port - 1][i], &gid, sizeof gid)) {
1061                                         hits[i] = 1;
1062                                         found = 1;
1063                                         break;
1064                                 }
1065                         }
1066
1067                         if (!found) {
1068                                 if (tmp == ndev &&
1069                                     (memcmp(&dev->iboe.gid_table[port - 1][0],
1070                                             &gid, sizeof gid) ||
1071                                      !memcmp(&dev->iboe.gid_table[port - 1][0],
1072                                              &zgid, sizeof gid))) {
1073                                         dev->iboe.gid_table[port - 1][0] = gid;
1074                                         ++need_update;
1075                                         hits[0] = 1;
1076                                 } else if (free >= 0) {
1077                                         dev->iboe.gid_table[port - 1][free] = gid;
1078                                         hits[free] = 1;
1079                                         ++need_update;
1080                                 }
1081                         }
1082                 }
1083         }
1084         rcu_read_unlock();
1085
1086         for (i = 0; i < 128; ++i)
1087                 if (!hits[i]) {
1088                         if (memcmp(&dev->iboe.gid_table[port - 1][i], &zgid, sizeof zgid))
1089                                 ++need_update;
1090                         dev->iboe.gid_table[port - 1][i] = zgid;
1091                 }
1092
1093         if (need_update) {
1094                 memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof work->gids);
1095                 INIT_WORK(&work->work, update_gids_task);
1096                 work->port = port;
1097                 work->dev = dev;
1098                 queue_work(wq, &work->work);
1099         } else
1100                 kfree(work);
1101
1102         kfree(hits);
1103         return 0;
1104
1105 out:
1106         kfree(work);
1107         return ret;
1108 }
1109
1110 static void handle_en_event(struct mlx4_ib_dev *dev, int port, unsigned long event)
1111 {
1112         switch (event) {
1113         case NETDEV_UP:
1114         case NETDEV_CHANGEADDR:
1115                 update_ipv6_gids(dev, port, 0);
1116                 break;
1117
1118         case NETDEV_DOWN:
1119                 update_ipv6_gids(dev, port, 1);
1120                 dev->iboe.netdevs[port - 1] = NULL;
1121         }
1122 }
1123
1124 static void netdev_added(struct mlx4_ib_dev *dev, int port)
1125 {
1126         update_ipv6_gids(dev, port, 0);
1127 }
1128
1129 static void netdev_removed(struct mlx4_ib_dev *dev, int port)
1130 {
1131         update_ipv6_gids(dev, port, 1);
1132 }
1133
1134 static int mlx4_ib_netdev_event(struct notifier_block *this, unsigned long event,
1135                                 void *ptr)
1136 {
1137         struct net_device *dev = ptr;
1138         struct mlx4_ib_dev *ibdev;
1139         struct net_device *oldnd;
1140         struct mlx4_ib_iboe *iboe;
1141         int port;
1142
1143         if (!net_eq(dev_net(dev), &init_net))
1144                 return NOTIFY_DONE;
1145
1146         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1147         iboe = &ibdev->iboe;
1148
1149         spin_lock(&iboe->lock);
1150         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1151                 oldnd = iboe->netdevs[port - 1];
1152                 iboe->netdevs[port - 1] =
1153                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1154                 if (oldnd != iboe->netdevs[port - 1]) {
1155                         if (iboe->netdevs[port - 1])
1156                                 netdev_added(ibdev, port);
1157                         else
1158                                 netdev_removed(ibdev, port);
1159                 }
1160         }
1161
1162         if (dev == iboe->netdevs[0] ||
1163             (iboe->netdevs[0] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[0]))
1164                 handle_en_event(ibdev, 1, event);
1165         else if (dev == iboe->netdevs[1]
1166                  || (iboe->netdevs[1] && rdma_vlan_dev_real_dev(dev) == iboe->netdevs[1]))
1167                 handle_en_event(ibdev, 2, event);
1168
1169         spin_unlock(&iboe->lock);
1170
1171         return NOTIFY_DONE;
1172 }
1173
1174 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1175 {
1176         int port;
1177         int slave;
1178         int i;
1179
1180         if (mlx4_is_master(ibdev->dev)) {
1181                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1182                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1183                                 for (i = 0;
1184                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1185                                      ++i) {
1186                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1187                                         /* master has the identity virt2phys pkey mapping */
1188                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1189                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1190                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1191                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1192                                 }
1193                         }
1194                 }
1195                 /* initialize pkey cache */
1196                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1197                         for (i = 0;
1198                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1199                              ++i)
1200                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1201                                         (i) ? 0 : 0xFFFF;
1202                 }
1203         }
1204 }
1205
1206 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1207 {
1208         char name[32];
1209         int eq_per_port = 0;
1210         int added_eqs = 0;
1211         int total_eqs = 0;
1212         int i, j, eq;
1213
1214         /* Legacy mode or comp_pool is not large enough */
1215         if (dev->caps.comp_pool == 0 ||
1216             dev->caps.num_ports > dev->caps.comp_pool)
1217                 return;
1218
1219         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1220                                         dev->caps.num_ports);
1221
1222         /* Init eq table */
1223         added_eqs = 0;
1224         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1225                 added_eqs += eq_per_port;
1226
1227         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1228
1229         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1230         if (!ibdev->eq_table)
1231                 return;
1232
1233         ibdev->eq_added = added_eqs;
1234
1235         eq = 0;
1236         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1237                 for (j = 0; j < eq_per_port; j++) {
1238                         sprintf(name, "mlx4-ib-%d-%d@%s",
1239                                 i, j, dev->pdev->bus->name);
1240                         /* Set IRQ for specific name (per ring) */
1241                         if (mlx4_assign_eq(dev, name, NULL,
1242                                            &ibdev->eq_table[eq])) {
1243                                 /* Use legacy (same as mlx4_en driver) */
1244                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1245                                 ibdev->eq_table[eq] =
1246                                         (eq % dev->caps.num_comp_vectors);
1247                         }
1248                         eq++;
1249                 }
1250         }
1251
1252         /* Fill the reset of the vector with legacy EQ */
1253         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1254                 ibdev->eq_table[eq++] = i;
1255
1256         /* Advertise the new number of EQs to clients */
1257         ibdev->ib_dev.num_comp_vectors = total_eqs;
1258 }
1259
1260 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1261 {
1262         int i;
1263
1264         /* no additional eqs were added */
1265         if (!ibdev->eq_table)
1266                 return;
1267
1268         /* Reset the advertised EQ number */
1269         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1270
1271         /* Free only the added eqs */
1272         for (i = 0; i < ibdev->eq_added; i++) {
1273                 /* Don't free legacy eqs if used */
1274                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1275                         continue;
1276                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1277         }
1278
1279         kfree(ibdev->eq_table);
1280 }
1281
1282 static void *mlx4_ib_add(struct mlx4_dev *dev)
1283 {
1284         struct mlx4_ib_dev *ibdev;
1285         int num_ports = 0;
1286         int i, j;
1287         int err;
1288         struct mlx4_ib_iboe *iboe;
1289
1290         pr_info_once("%s", mlx4_ib_version);
1291
1292         if (mlx4_is_mfunc(dev)) {
1293                 pr_warn("IB not yet supported in SRIOV\n");
1294                 return NULL;
1295         }
1296
1297         mlx4_foreach_ib_transport_port(i, dev)
1298                 num_ports++;
1299
1300         /* No point in registering a device with no ports... */
1301         if (num_ports == 0)
1302                 return NULL;
1303
1304         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1305         if (!ibdev) {
1306                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1307                 return NULL;
1308         }
1309
1310         iboe = &ibdev->iboe;
1311
1312         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1313                 goto err_dealloc;
1314
1315         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1316                 goto err_pd;
1317
1318         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1319                                  PAGE_SIZE);
1320         if (!ibdev->uar_map)
1321                 goto err_uar;
1322         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1323
1324         ibdev->dev = dev;
1325
1326         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1327         ibdev->ib_dev.owner             = THIS_MODULE;
1328         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1329         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1330         ibdev->num_ports                = num_ports;
1331         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1332         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1333         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1334
1335         ibdev->ib_dev.uverbs_abi_ver    = MLX4_IB_UVERBS_ABI_VERSION;
1336         ibdev->ib_dev.uverbs_cmd_mask   =
1337                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1338                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1339                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1340                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1341                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1342                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1343                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1344                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1345                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1346                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1347                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1348                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1349                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1350                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1351                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1352                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1353                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1354                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1355                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1356                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1357                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1358                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
1359                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1360
1361         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1362         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1363         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1364         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1365         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1366         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1367         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1368         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1369         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1370         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1371         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1372         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1373         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1374         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1375         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1376         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1377         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1378         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1379         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1380         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1381         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1382         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1383         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1384         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1385         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1386         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1387         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1388         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1389         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1390         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1391         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1392         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
1393         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
1394         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
1395         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
1396         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1397         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1398         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1399         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
1400         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
1401         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
1402
1403         ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
1404         ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
1405         ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
1406         ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
1407
1408         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1409                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1410                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1411                 ibdev->ib_dev.uverbs_cmd_mask |=
1412                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1413                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1414         }
1415
1416         mlx4_ib_alloc_eqs(dev, ibdev);
1417
1418         spin_lock_init(&iboe->lock);
1419
1420         if (init_node_data(ibdev))
1421                 goto err_map;
1422
1423         for (i = 0; i < ibdev->num_ports; ++i) {
1424                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1425                                                 IB_LINK_LAYER_ETHERNET) {
1426                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1427                         if (err)
1428                                 ibdev->counters[i] = -1;
1429                 } else
1430                                 ibdev->counters[i] = -1;
1431         }
1432
1433         spin_lock_init(&ibdev->sm_lock);
1434         mutex_init(&ibdev->cap_mask_mutex);
1435
1436         if (ib_register_device(&ibdev->ib_dev, NULL))
1437                 goto err_counter;
1438
1439         if (mlx4_ib_mad_init(ibdev))
1440                 goto err_reg;
1441
1442         if (mlx4_ib_init_sriov(ibdev))
1443                 goto err_mad;
1444
1445         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE && !iboe->nb.notifier_call) {
1446                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
1447                 err = register_netdevice_notifier(&iboe->nb);
1448                 if (err)
1449                         goto err_sriov;
1450         }
1451
1452         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1453                 if (device_create_file(&ibdev->ib_dev.dev,
1454                                        mlx4_class_attributes[j]))
1455                         goto err_notif;
1456         }
1457
1458         ibdev->ib_active = true;
1459
1460         if (mlx4_is_mfunc(ibdev->dev))
1461                 init_pkeys(ibdev);
1462
1463         return ibdev;
1464
1465 err_notif:
1466         if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1467                 pr_warn("failure unregistering notifier\n");
1468         flush_workqueue(wq);
1469
1470 err_sriov:
1471         mlx4_ib_close_sriov(ibdev);
1472
1473 err_mad:
1474         mlx4_ib_mad_cleanup(ibdev);
1475
1476 err_reg:
1477         ib_unregister_device(&ibdev->ib_dev);
1478
1479 err_counter:
1480         for (; i; --i)
1481                 if (ibdev->counters[i - 1] != -1)
1482                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1483
1484 err_map:
1485         iounmap(ibdev->uar_map);
1486
1487 err_uar:
1488         mlx4_uar_free(dev, &ibdev->priv_uar);
1489
1490 err_pd:
1491         mlx4_pd_free(dev, ibdev->priv_pdn);
1492
1493 err_dealloc:
1494         ib_dealloc_device(&ibdev->ib_dev);
1495
1496         return NULL;
1497 }
1498
1499 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1500 {
1501         struct mlx4_ib_dev *ibdev = ibdev_ptr;
1502         int p;
1503
1504         mlx4_ib_close_sriov(ibdev);
1505         mlx4_ib_mad_cleanup(ibdev);
1506         ib_unregister_device(&ibdev->ib_dev);
1507         if (ibdev->iboe.nb.notifier_call) {
1508                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1509                         pr_warn("failure unregistering notifier\n");
1510                 ibdev->iboe.nb.notifier_call = NULL;
1511         }
1512         iounmap(ibdev->uar_map);
1513         for (p = 0; p < ibdev->num_ports; ++p)
1514                 if (ibdev->counters[p] != -1)
1515                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1516         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1517                 mlx4_CLOSE_PORT(dev, p);
1518
1519         mlx4_ib_free_eqs(dev, ibdev);
1520
1521         mlx4_uar_free(dev, &ibdev->priv_uar);
1522         mlx4_pd_free(dev, ibdev->priv_pdn);
1523         ib_dealloc_device(&ibdev->ib_dev);
1524 }
1525
1526 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
1527 {
1528         struct mlx4_ib_demux_work **dm = NULL;
1529         struct mlx4_dev *dev = ibdev->dev;
1530         int i;
1531         unsigned long flags;
1532
1533         if (!mlx4_is_master(dev))
1534                 return;
1535
1536         dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
1537         if (!dm) {
1538                 pr_err("failed to allocate memory for tunneling qp update\n");
1539                 goto out;
1540         }
1541
1542         for (i = 0; i < dev->caps.num_ports; i++) {
1543                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
1544                 if (!dm[i]) {
1545                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
1546                         for (i = 0; i < dev->caps.num_ports; i++) {
1547                                 if (dm[i])
1548                                         kfree(dm[i]);
1549                         }
1550                         goto out;
1551                 }
1552         }
1553         /* initialize or tear down tunnel QPs for the slave */
1554         for (i = 0; i < dev->caps.num_ports; i++) {
1555                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
1556                 dm[i]->port = i + 1;
1557                 dm[i]->slave = slave;
1558                 dm[i]->do_init = do_init;
1559                 dm[i]->dev = ibdev;
1560                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
1561                 if (!ibdev->sriov.is_going_down)
1562                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
1563                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
1564         }
1565 out:
1566         if (dm)
1567                 kfree(dm);
1568         return;
1569 }
1570
1571 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
1572                           enum mlx4_dev_event event, unsigned long param)
1573 {
1574         struct ib_event ibev;
1575         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
1576         struct mlx4_eqe *eqe = NULL;
1577         struct ib_event_work *ew;
1578         int p = 0;
1579
1580         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
1581                 eqe = (struct mlx4_eqe *)param;
1582         else
1583                 p = (int) param;
1584
1585         switch (event) {
1586         case MLX4_DEV_EVENT_PORT_UP:
1587                 if (p > ibdev->num_ports)
1588                         return;
1589                 if (mlx4_is_master(dev) &&
1590                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
1591                         IB_LINK_LAYER_INFINIBAND) {
1592                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
1593                 }
1594                 ibev.event = IB_EVENT_PORT_ACTIVE;
1595                 break;
1596
1597         case MLX4_DEV_EVENT_PORT_DOWN:
1598                 if (p > ibdev->num_ports)
1599                         return;
1600                 ibev.event = IB_EVENT_PORT_ERR;
1601                 break;
1602
1603         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
1604                 ibdev->ib_active = false;
1605                 ibev.event = IB_EVENT_DEVICE_FATAL;
1606                 break;
1607
1608         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
1609                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
1610                 if (!ew) {
1611                         pr_err("failed to allocate memory for events work\n");
1612                         break;
1613                 }
1614
1615                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
1616                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
1617                 ew->ib_dev = ibdev;
1618                 handle_port_mgmt_change_event(&ew->work);
1619                 return;
1620
1621         case MLX4_DEV_EVENT_SLAVE_INIT:
1622                 /* here, p is the slave id */
1623                 do_slave_init(ibdev, p, 1);
1624                 return;
1625
1626         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
1627                 /* here, p is the slave id */
1628                 do_slave_init(ibdev, p, 0);
1629                 return;
1630
1631         default:
1632                 return;
1633         }
1634
1635         ibev.device           = ibdev_ptr;
1636         ibev.element.port_num = (u8) p;
1637
1638         ib_dispatch_event(&ibev);
1639 }
1640
1641 static struct mlx4_interface mlx4_ib_interface = {
1642         .add            = mlx4_ib_add,
1643         .remove         = mlx4_ib_remove,
1644         .event          = mlx4_ib_event,
1645         .protocol       = MLX4_PROT_IB_IPV6
1646 };
1647
1648 static int __init mlx4_ib_init(void)
1649 {
1650         int err;
1651
1652         wq = create_singlethread_workqueue("mlx4_ib");
1653         if (!wq)
1654                 return -ENOMEM;
1655
1656         err = mlx4_ib_mcg_init();
1657         if (err)
1658                 goto clean_wq;
1659
1660         err = mlx4_register_interface(&mlx4_ib_interface);
1661         if (err)
1662                 goto clean_mcg;
1663
1664         return 0;
1665
1666 clean_mcg:
1667         mlx4_ib_mcg_destroy();
1668
1669 clean_wq:
1670         destroy_workqueue(wq);
1671         return err;
1672 }
1673
1674 static void __exit mlx4_ib_cleanup(void)
1675 {
1676         mlx4_unregister_interface(&mlx4_ib_interface);
1677         mlx4_ib_mcg_destroy();
1678         destroy_workqueue(wq);
1679 }
1680
1681 module_init(mlx4_ib_init);
1682 module_exit(mlx4_ib_cleanup);