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 #include <net/ipv6.h>
  43 #include <net/addrconf.h>
  44
  45 #include <rdma/ib_smi.h>
  46 #include <rdma/ib_user_verbs.h>
  47 #include <rdma/ib_addr.h>
  48
  49 #include <linux/mlx4/driver.h>
  50 #include <linux/mlx4/cmd.h>
  51 #include <linux/mlx4/qp.h>
  52
  53 #include "mlx4_ib.h"
  54 #include "user.h"
  55
  56 #define DRV_NAME        MLX4_IB_DRV_NAME
  57 #define DRV_VERSION     "2.2-1"
  58 #define DRV_RELDATE     "Feb 2014"
  59
  60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  62 #define MLX4_IB_CARD_REV_A0   0xA0
  63
  64 MODULE_AUTHOR("Roland Dreier");
  65 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  66 MODULE_LICENSE("Dual BSD/GPL");
  67 MODULE_VERSION(DRV_VERSION);
  68
  69 int mlx4_ib_sm_guid_assign = 1;
  70 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  71 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
  72
  73 static const char mlx4_ib_version[] =
  74         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  75         DRV_VERSION " (" DRV_RELDATE ")\n";
  76
  77 struct update_gid_work {
  78         struct work_struct      work;
  79         union ib_gid            gids[128];
  80         struct mlx4_ib_dev     *dev;
  81         int                     port;
  82 };
  83
  84 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static union ib_gid zgid;
  97
  98 static int check_flow_steering_support(struct mlx4_dev *dev)
  99 {
 100         int eth_num_ports = 0;
 101         int ib_num_ports = 0;
 102
 103         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 104
 105         if (dmfs) {
 106                 int i;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 108                         eth_num_ports++;
 109                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 110                         ib_num_ports++;
 111                 dmfs &= (!ib_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 113                         (!eth_num_ports ||
 114                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 115                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 116                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 117                         dmfs = 0;
 118                 }
 119         }
 120         return dmfs;
 121 }
 122
 123 static int num_ib_ports(struct mlx4_dev *dev)
 124 {
 125         int ib_ports = 0;
 126         int i;
 127
 128         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 129                 ib_ports++;
 130
 131         return ib_ports;
 132 }
 133
 134 static int mlx4_ib_query_device(struct ib_device *ibdev,
 135                                 struct ib_device_attr *props)
 136 {
 137         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 138         struct ib_smp *in_mad  = NULL;
 139         struct ib_smp *out_mad = NULL;
 140         int err = -ENOMEM;
 141         int have_ib_ports;
 142
 143         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 144         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 145         if (!in_mad || !out_mad)
 146                 goto out;
 147
 148         init_query_mad(in_mad);
 149         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 150
 151         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 152                            1, NULL, NULL, in_mad, out_mad);
 153         if (err)
 154                 goto out;
 155
 156         memset(props, 0, sizeof *props);
 157
 158         have_ib_ports = num_ib_ports(dev->dev);
 159
 160         props->fw_ver = dev->dev->caps.fw_ver;
 161         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 162                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 163                 IB_DEVICE_SYS_IMAGE_GUID                |
 164                 IB_DEVICE_RC_RNR_NAK_GEN                |
 165                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 166         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 167                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 168         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 169                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 170         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 171                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 172         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 173                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 174         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 175                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 176         if (dev->dev->caps.max_gso_sz &&
 177             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 178             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 179                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 180         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 181                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 182         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 183             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 184             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 185                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 186         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 187                 props->device_cap_flags |= IB_DEVICE_XRC;
 188         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 189                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 190         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 191                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 192                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 193                 else
 194                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 195         if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
 196                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 197         }
 198
 199         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 200                 0xffffff;
 201         props->vendor_part_id      = dev->dev->persist->pdev->device;
 202         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 203         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 204
 205         props->max_mr_size         = ~0ull;
 206         props->page_size_cap       = dev->dev->caps.page_size_cap;
 207         props->max_qp              = dev->dev->quotas.qp;
 208         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 209         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 210                                          dev->dev->caps.max_rq_sg);
 211         props->max_cq              = dev->dev->quotas.cq;
 212         props->max_cqe             = dev->dev->caps.max_cqes;
 213         props->max_mr              = dev->dev->quotas.mpt;
 214         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 215         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 216         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 217         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 218         props->max_srq             = dev->dev->quotas.srq;
 219         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 220         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 221         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 222         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 223         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 224                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 225         props->masked_atomic_cap   = props->atomic_cap;
 226         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 227         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 228         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 229         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 230                                            props->max_mcast_grp;
 231         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 232
 233 out:
 234         kfree(in_mad);
 235         kfree(out_mad);
 236
 237         return err;
 238 }
 239
 240 static enum rdma_link_layer
 241 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 242 {
 243         struct mlx4_dev *dev = to_mdev(device)->dev;
 244
 245         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 246                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 247 }
 248
 249 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 250                               struct ib_port_attr *props, int netw_view)
 251 {
 252         struct ib_smp *in_mad  = NULL;
 253         struct ib_smp *out_mad = NULL;
 254         int ext_active_speed;
 255         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 256         int err = -ENOMEM;
 257
 258         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 259         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 260         if (!in_mad || !out_mad)
 261                 goto out;
 262
 263         init_query_mad(in_mad);
 264         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 265         in_mad->attr_mod = cpu_to_be32(port);
 266
 267         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 268                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 269
 270         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 271                                 in_mad, out_mad);
 272         if (err)
 273                 goto out;
 274
 275
 276         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 277         props->lmc              = out_mad->data[34] & 0x7;
 278         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 279         props->sm_sl            = out_mad->data[36] & 0xf;
 280         props->state            = out_mad->data[32] & 0xf;
 281         props->phys_state       = out_mad->data[33] >> 4;
 282         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 283         if (netw_view)
 284                 props->gid_tbl_len = out_mad->data[50];
 285         else
 286                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 287         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 288         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 289         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 290         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 291         props->active_width     = out_mad->data[31] & 0xf;
 292         props->active_speed     = out_mad->data[35] >> 4;
 293         props->max_mtu          = out_mad->data[41] & 0xf;
 294         props->active_mtu       = out_mad->data[36] >> 4;
 295         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 296         props->max_vl_num       = out_mad->data[37] >> 4;
 297         props->init_type_reply  = out_mad->data[41] >> 4;
 298
 299         /* Check if extended speeds (EDR/FDR/...) are supported */
 300         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 301                 ext_active_speed = out_mad->data[62] >> 4;
 302
 303                 switch (ext_active_speed) {
 304                 case 1:
 305                         props->active_speed = IB_SPEED_FDR;
 306                         break;
 307                 case 2:
 308                         props->active_speed = IB_SPEED_EDR;
 309                         break;
 310                 }
 311         }
 312
 313         /* If reported active speed is QDR, check if is FDR-10 */
 314         if (props->active_speed == IB_SPEED_QDR) {
 315                 init_query_mad(in_mad);
 316                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 317                 in_mad->attr_mod = cpu_to_be32(port);
 318
 319                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 320                                    NULL, NULL, in_mad, out_mad);
 321                 if (err)
 322                         goto out;
 323
 324                 /* Checking LinkSpeedActive for FDR-10 */
 325                 if (out_mad->data[15] & 0x1)
 326                         props->active_speed = IB_SPEED_FDR10;
 327         }
 328
 329         /* Avoid wrong speed value returned by FW if the IB link is down. */
 330         if (props->state == IB_PORT_DOWN)
 331                  props->active_speed = IB_SPEED_SDR;
 332
 333 out:
 334         kfree(in_mad);
 335         kfree(out_mad);
 336         return err;
 337 }
 338
 339 static u8 state_to_phys_state(enum ib_port_state state)
 340 {
 341         return state == IB_PORT_ACTIVE ? 5 : 3;
 342 }
 343
 344 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 345                                struct ib_port_attr *props, int netw_view)
 346 {
 347
 348         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 349         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 350         struct net_device *ndev;
 351         enum ib_mtu tmp;
 352         struct mlx4_cmd_mailbox *mailbox;
 353         int err = 0;
 354         int is_bonded = mlx4_is_bonded(mdev->dev);
 355
 356         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 357         if (IS_ERR(mailbox))
 358                 return PTR_ERR(mailbox);
 359
 360         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 361                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 362                            MLX4_CMD_WRAPPED);
 363         if (err)
 364                 goto out;
 365
 366         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 367                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 368         props->active_speed     = IB_SPEED_QDR;
 369         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
 370         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 371         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 372         props->pkey_tbl_len     = 1;
 373         props->max_mtu          = IB_MTU_4096;
 374         props->max_vl_num       = 2;
 375         props->state            = IB_PORT_DOWN;
 376         props->phys_state       = state_to_phys_state(props->state);
 377         props->active_mtu       = IB_MTU_256;
 378         if (is_bonded)
 379                 rtnl_lock(); /* required to get upper dev */
 380         spin_lock_bh(&iboe->lock);
 381         ndev = iboe->netdevs[port - 1];
 382         if (ndev && is_bonded)
 383                 ndev = netdev_master_upper_dev_get(ndev);
 384         if (!ndev)
 385                 goto out_unlock;
 386
 387         tmp = iboe_get_mtu(ndev->mtu);
 388         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 389
 390         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 391                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 392         props->phys_state       = state_to_phys_state(props->state);
 393 out_unlock:
 394         spin_unlock_bh(&iboe->lock);
 395         if (is_bonded)
 396                 rtnl_unlock();
 397 out:
 398         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 399         return err;
 400 }
 401
 402 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 403                          struct ib_port_attr *props, int netw_view)
 404 {
 405         int err;
 406
 407         memset(props, 0, sizeof *props);
 408
 409         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 410                 ib_link_query_port(ibdev, port, props, netw_view) :
 411                                 eth_link_query_port(ibdev, port, props, netw_view);
 412
 413         return err;
 414 }
 415
 416 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 417                               struct ib_port_attr *props)
 418 {
 419         /* returns host view */
 420         return __mlx4_ib_query_port(ibdev, port, props, 0);
 421 }
 422
 423 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 424                         union ib_gid *gid, int netw_view)
 425 {
 426         struct ib_smp *in_mad  = NULL;
 427         struct ib_smp *out_mad = NULL;
 428         int err = -ENOMEM;
 429         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 430         int clear = 0;
 431         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 432
 433         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 434         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 435         if (!in_mad || !out_mad)
 436                 goto out;
 437
 438         init_query_mad(in_mad);
 439         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 440         in_mad->attr_mod = cpu_to_be32(port);
 441
 442         if (mlx4_is_mfunc(dev->dev) && netw_view)
 443                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 444
 445         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 446         if (err)
 447                 goto out;
 448
 449         memcpy(gid->raw, out_mad->data + 8, 8);
 450
 451         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 452                 if (index) {
 453                         /* For any index > 0, return the null guid */
 454                         err = 0;
 455                         clear = 1;
 456                         goto out;
 457                 }
 458         }
 459
 460         init_query_mad(in_mad);
 461         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 462         in_mad->attr_mod = cpu_to_be32(index / 8);
 463
 464         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 465                            NULL, NULL, in_mad, out_mad);
 466         if (err)
 467                 goto out;
 468
 469         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 470
 471 out:
 472         if (clear)
 473                 memset(gid->raw + 8, 0, 8);
 474         kfree(in_mad);
 475         kfree(out_mad);
 476         return err;
 477 }
 478
 479 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
 480                           union ib_gid *gid)
 481 {
 482         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 483
 484         *gid = dev->iboe.gid_table[port - 1][index];
 485
 486         return 0;
 487 }
 488
 489 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 490                              union ib_gid *gid)
 491 {
 492         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
 493                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 494         else
 495                 return iboe_query_gid(ibdev, port, index, gid);
 496 }
 497
 498 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 499                          u16 *pkey, int netw_view)
 500 {
 501         struct ib_smp *in_mad  = NULL;
 502         struct ib_smp *out_mad = NULL;
 503         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 504         int err = -ENOMEM;
 505
 506         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 507         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 508         if (!in_mad || !out_mad)
 509                 goto out;
 510
 511         init_query_mad(in_mad);
 512         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 513         in_mad->attr_mod = cpu_to_be32(index / 32);
 514
 515         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 516                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 517
 518         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 519                            in_mad, out_mad);
 520         if (err)
 521                 goto out;
 522
 523         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 524
 525 out:
 526         kfree(in_mad);
 527         kfree(out_mad);
 528         return err;
 529 }
 530
 531 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 532 {
 533         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 534 }
 535
 536 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 537                                  struct ib_device_modify *props)
 538 {
 539         struct mlx4_cmd_mailbox *mailbox;
 540         unsigned long flags;
 541
 542         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 543                 return -EOPNOTSUPP;
 544
 545         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 546                 return 0;
 547
 548         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 549                 return -EOPNOTSUPP;
 550
 551         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 552         memcpy(ibdev->node_desc, props->node_desc, 64);
 553         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 554
 555         /*
 556          * If possible, pass node desc to FW, so it can generate
 557          * a 144 trap.  If cmd fails, just ignore.
 558          */
 559         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 560         if (IS_ERR(mailbox))
 561                 return 0;
 562
 563         memcpy(mailbox->buf, props->node_desc, 64);
 564         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 565                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 566
 567         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 568
 569         return 0;
 570 }
 571
 572 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 573                             u32 cap_mask)
 574 {
 575         struct mlx4_cmd_mailbox *mailbox;
 576         int err;
 577
 578         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 579         if (IS_ERR(mailbox))
 580                 return PTR_ERR(mailbox);
 581
 582         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 583                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 584                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 585         } else {
 586                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 587                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 588         }
 589
 590         err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
 591                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
 592
 593         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 594         return err;
 595 }
 596
 597 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 598                                struct ib_port_modify *props)
 599 {
 600         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 601         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 602         struct ib_port_attr attr;
 603         u32 cap_mask;
 604         int err;
 605
 606         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
 607          * of whether port link layer is ETH or IB. For ETH ports, qkey
 608          * violations and port capabilities are not meaningful.
 609          */
 610         if (is_eth)
 611                 return 0;
 612
 613         mutex_lock(&mdev->cap_mask_mutex);
 614
 615         err = mlx4_ib_query_port(ibdev, port, &attr);
 616         if (err)
 617                 goto out;
 618
 619         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 620                 ~props->clr_port_cap_mask;
 621
 622         err = mlx4_ib_SET_PORT(mdev, port,
 623                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
 624                                cap_mask);
 625
 626 out:
 627         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 628         return err;
 629 }
 630
 631 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 632                                                   struct ib_udata *udata)
 633 {
 634         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 635         struct mlx4_ib_ucontext *context;
 636         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 637         struct mlx4_ib_alloc_ucontext_resp resp;
 638         int err;
 639
 640         if (!dev->ib_active)
 641                 return ERR_PTR(-EAGAIN);
 642
 643         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
 644                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
 645                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
 646                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 647         } else {
 648                 resp.dev_caps         = dev->dev->caps.userspace_caps;
 649                 resp.qp_tab_size      = dev->dev->caps.num_qps;
 650                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 651                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 652                 resp.cqe_size         = dev->dev->caps.cqe_size;
 653         }
 654
 655         context = kmalloc(sizeof *context, GFP_KERNEL);
 656         if (!context)
 657                 return ERR_PTR(-ENOMEM);
 658
 659         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 660         if (err) {
 661                 kfree(context);
 662                 return ERR_PTR(err);
 663         }
 664
 665         INIT_LIST_HEAD(&context->db_page_list);
 666         mutex_init(&context->db_page_mutex);
 667
 668         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
 669                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
 670         else
 671                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
 672
 673         if (err) {
 674                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 675                 kfree(context);
 676                 return ERR_PTR(-EFAULT);
 677         }
 678
 679         return &context->ibucontext;
 680 }
 681
 682 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 683 {
 684         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 685
 686         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 687         kfree(context);
 688
 689         return 0;
 690 }
 691
 692 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 693 {
 694         struct mlx4_ib_dev *dev = to_mdev(context->device);
 695
 696         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 697                 return -EINVAL;
 698
 699         if (vma->vm_pgoff == 0) {
 700                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 701
 702                 if (io_remap_pfn_range(vma, vma->vm_start,
 703                                        to_mucontext(context)->uar.pfn,
 704                                        PAGE_SIZE, vma->vm_page_prot))
 705                         return -EAGAIN;
 706         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 707                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 708
 709                 if (io_remap_pfn_range(vma, vma->vm_start,
 710                                        to_mucontext(context)->uar.pfn +
 711                                        dev->dev->caps.num_uars,
 712                                        PAGE_SIZE, vma->vm_page_prot))
 713                         return -EAGAIN;
 714         } else
 715                 return -EINVAL;
 716
 717         return 0;
 718 }
 719
 720 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 721                                       struct ib_ucontext *context,
 722                                       struct ib_udata *udata)
 723 {
 724         struct mlx4_ib_pd *pd;
 725         int err;
 726
 727         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 728         if (!pd)
 729                 return ERR_PTR(-ENOMEM);
 730
 731         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 732         if (err) {
 733                 kfree(pd);
 734                 return ERR_PTR(err);
 735         }
 736
 737         if (context)
 738                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 739                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 740                         kfree(pd);
 741                         return ERR_PTR(-EFAULT);
 742                 }
 743
 744         return &pd->ibpd;
 745 }
 746
 747 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 748 {
 749         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 750         kfree(pd);
 751
 752         return 0;
 753 }
 754
 755 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 756                                           struct ib_ucontext *context,
 757                                           struct ib_udata *udata)
 758 {
 759         struct mlx4_ib_xrcd *xrcd;
 760         int err;
 761
 762         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 763                 return ERR_PTR(-ENOSYS);
 764
 765         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 766         if (!xrcd)
 767                 return ERR_PTR(-ENOMEM);
 768
 769         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 770         if (err)
 771                 goto err1;
 772
 773         xrcd->pd = ib_alloc_pd(ibdev);
 774         if (IS_ERR(xrcd->pd)) {
 775                 err = PTR_ERR(xrcd->pd);
 776                 goto err2;
 777         }
 778
 779         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
 780         if (IS_ERR(xrcd->cq)) {
 781                 err = PTR_ERR(xrcd->cq);
 782                 goto err3;
 783         }
 784
 785         return &xrcd->ibxrcd;
 786
 787 err3:
 788         ib_dealloc_pd(xrcd->pd);
 789 err2:
 790         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 791 err1:
 792         kfree(xrcd);
 793         return ERR_PTR(err);
 794 }
 795
 796 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 797 {
 798         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 799         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 800         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 801         kfree(xrcd);
 802
 803         return 0;
 804 }
 805
 806 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 807 {
 808         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 809         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 810         struct mlx4_ib_gid_entry *ge;
 811
 812         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 813         if (!ge)
 814                 return -ENOMEM;
 815
 816         ge->gid = *gid;
 817         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 818                 ge->port = mqp->port;
 819                 ge->added = 1;
 820         }
 821
 822         mutex_lock(&mqp->mutex);
 823         list_add_tail(&ge->list, &mqp->gid_list);
 824         mutex_unlock(&mqp->mutex);
 825
 826         return 0;
 827 }
 828
 829 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 830                    union ib_gid *gid)
 831 {
 832         struct net_device *ndev;
 833         int ret = 0;
 834
 835         if (!mqp->port)
 836                 return 0;
 837
 838         spin_lock_bh(&mdev->iboe.lock);
 839         ndev = mdev->iboe.netdevs[mqp->port - 1];
 840         if (ndev)
 841                 dev_hold(ndev);
 842         spin_unlock_bh(&mdev->iboe.lock);
 843
 844         if (ndev) {
 845                 ret = 1;
 846                 dev_put(ndev);
 847         }
 848
 849         return ret;
 850 }
 851
 852 struct mlx4_ib_steering {
 853         struct list_head list;
 854         struct mlx4_flow_reg_id reg_id;
 855         union ib_gid gid;
 856 };
 857
 858 static int parse_flow_attr(struct mlx4_dev *dev,
 859                            u32 qp_num,
 860                            union ib_flow_spec *ib_spec,
 861                            struct _rule_hw *mlx4_spec)
 862 {
 863         enum mlx4_net_trans_rule_id type;
 864
 865         switch (ib_spec->type) {
 866         case IB_FLOW_SPEC_ETH:
 867                 type = MLX4_NET_TRANS_RULE_ID_ETH;
 868                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
 869                        ETH_ALEN);
 870                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
 871                        ETH_ALEN);
 872                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
 873                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
 874                 break;
 875         case IB_FLOW_SPEC_IB:
 876                 type = MLX4_NET_TRANS_RULE_ID_IB;
 877                 mlx4_spec->ib.l3_qpn =
 878                         cpu_to_be32(qp_num);
 879                 mlx4_spec->ib.qpn_mask =
 880                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
 881                 break;
 882
 883
 884         case IB_FLOW_SPEC_IPV4:
 885                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
 886                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
 887                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
 888                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
 889                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
 890                 break;
 891
 892         case IB_FLOW_SPEC_TCP:
 893         case IB_FLOW_SPEC_UDP:
 894                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
 895                                         MLX4_NET_TRANS_RULE_ID_TCP :
 896                                         MLX4_NET_TRANS_RULE_ID_UDP;
 897                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
 898                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
 899                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
 900                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
 901                 break;
 902
 903         default:
 904                 return -EINVAL;
 905         }
 906         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
 907             mlx4_hw_rule_sz(dev, type) < 0)
 908                 return -EINVAL;
 909         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
 910         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
 911         return mlx4_hw_rule_sz(dev, type);
 912 }
 913
 914 struct default_rules {
 915         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 916         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 917         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
 918         __u8  link_layer;
 919 };
 920 static const struct default_rules default_table[] = {
 921         {
 922                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
 923                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
 924                 .rules_create_list = {IB_FLOW_SPEC_IB},
 925                 .link_layer = IB_LINK_LAYER_INFINIBAND
 926         }
 927 };
 928
 929 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
 930                                          struct ib_flow_attr *flow_attr)
 931 {
 932         int i, j, k;
 933         void *ib_flow;
 934         const struct default_rules *pdefault_rules = default_table;
 935         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
 936
 937         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
 938                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
 939                 memset(&field_types, 0, sizeof(field_types));
 940
 941                 if (link_layer != pdefault_rules->link_layer)
 942                         continue;
 943
 944                 ib_flow = flow_attr + 1;
 945                 /* we assume the specs are sorted */
 946                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
 947                      j < flow_attr->num_of_specs; k++) {
 948                         union ib_flow_spec *current_flow =
 949                                 (union ib_flow_spec *)ib_flow;
 950
 951                         /* same layer but different type */
 952                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
 953                              (pdefault_rules->mandatory_fields[k] &
 954                               IB_FLOW_SPEC_LAYER_MASK)) &&
 955                             (current_flow->type !=
 956                              pdefault_rules->mandatory_fields[k]))
 957                                 goto out;
 958
 959                         /* same layer, try match next one */
 960                         if (current_flow->type ==
 961                             pdefault_rules->mandatory_fields[k]) {
 962                                 j++;
 963                                 ib_flow +=
 964                                         ((union ib_flow_spec *)ib_flow)->size;
 965                         }
 966                 }
 967
 968                 ib_flow = flow_attr + 1;
 969                 for (j = 0; j < flow_attr->num_of_specs;
 970                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
 971                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
 972                                 /* same layer and same type */
 973                                 if (((union ib_flow_spec *)ib_flow)->type ==
 974                                     pdefault_rules->mandatory_not_fields[k])
 975                                         goto out;
 976
 977                 return i;
 978         }
 979 out:
 980         return -1;
 981 }
 982
 983 static int __mlx4_ib_create_default_rules(
 984                 struct mlx4_ib_dev *mdev,
 985                 struct ib_qp *qp,
 986                 const struct default_rules *pdefault_rules,
 987                 struct _rule_hw *mlx4_spec) {
 988         int size = 0;
 989         int i;
 990
 991         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
 992                 int ret;
 993                 union ib_flow_spec ib_spec;
 994                 switch (pdefault_rules->rules_create_list[i]) {
 995                 case 0:
 996                         /* no rule */
 997                         continue;
 998                 case IB_FLOW_SPEC_IB:
 999                         ib_spec.type = IB_FLOW_SPEC_IB;
1000                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1001
1002                         break;
1003                 default:
1004                         /* invalid rule */
1005                         return -EINVAL;
1006                 }
1007                 /* We must put empty rule, qpn is being ignored */
1008                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1009                                       mlx4_spec);
1010                 if (ret < 0) {
1011                         pr_info("invalid parsing\n");
1012                         return -EINVAL;
1013                 }
1014
1015                 mlx4_spec = (void *)mlx4_spec + ret;
1016                 size += ret;
1017         }
1018         return size;
1019 }
1020
1021 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1022                           int domain,
1023                           enum mlx4_net_trans_promisc_mode flow_type,
1024                           u64 *reg_id)
1025 {
1026         int ret, i;
1027         int size = 0;
1028         void *ib_flow;
1029         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1030         struct mlx4_cmd_mailbox *mailbox;
1031         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1032         int default_flow;
1033
1034         static const u16 __mlx4_domain[] = {
1035                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1036                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1037                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1038                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1039         };
1040
1041         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1042                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1043                 return -EINVAL;
1044         }
1045
1046         if (domain >= IB_FLOW_DOMAIN_NUM) {
1047                 pr_err("Invalid domain value %d\n", domain);
1048                 return -EINVAL;
1049         }
1050
1051         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1052                 return -EINVAL;
1053
1054         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1055         if (IS_ERR(mailbox))
1056                 return PTR_ERR(mailbox);
1057         ctrl = mailbox->buf;
1058
1059         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1060                                  flow_attr->priority);
1061         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1062         ctrl->port = flow_attr->port;
1063         ctrl->qpn = cpu_to_be32(qp->qp_num);
1064
1065         ib_flow = flow_attr + 1;
1066         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1067         /* Add default flows */
1068         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1069         if (default_flow >= 0) {
1070                 ret = __mlx4_ib_create_default_rules(
1071                                 mdev, qp, default_table + default_flow,
1072                                 mailbox->buf + size);
1073                 if (ret < 0) {
1074                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1075                         return -EINVAL;
1076                 }
1077                 size += ret;
1078         }
1079         for (i = 0; i < flow_attr->num_of_specs; i++) {
1080                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1081                                       mailbox->buf + size);
1082                 if (ret < 0) {
1083                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1084                         return -EINVAL;
1085                 }
1086                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1087                 size += ret;
1088         }
1089
1090         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1091                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1092                            MLX4_CMD_NATIVE);
1093         if (ret == -ENOMEM)
1094                 pr_err("mcg table is full. Fail to register network rule.\n");
1095         else if (ret == -ENXIO)
1096                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1097         else if (ret)
1098                 pr_err("Invalid argumant. Fail to register network rule.\n");
1099
1100         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1101         return ret;
1102 }
1103
1104 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1105 {
1106         int err;
1107         err = mlx4_cmd(dev, reg_id, 0, 0,
1108                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1109                        MLX4_CMD_NATIVE);
1110         if (err)
1111                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1112                        reg_id);
1113         return err;
1114 }
1115
1116 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1117                                     u64 *reg_id)
1118 {
1119         void *ib_flow;
1120         union ib_flow_spec *ib_spec;
1121         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1122         int err = 0;
1123
1124         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1125             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1126                 return 0; /* do nothing */
1127
1128         ib_flow = flow_attr + 1;
1129         ib_spec = (union ib_flow_spec *)ib_flow;
1130
1131         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1132                 return 0; /* do nothing */
1133
1134         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1135                                     flow_attr->port, qp->qp_num,
1136                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1137                                     reg_id);
1138         return err;
1139 }
1140
1141 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1142                                     struct ib_flow_attr *flow_attr,
1143                                     int domain)
1144 {
1145         int err = 0, i = 0, j = 0;
1146         struct mlx4_ib_flow *mflow;
1147         enum mlx4_net_trans_promisc_mode type[2];
1148         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1149         int is_bonded = mlx4_is_bonded(dev);
1150
1151         memset(type, 0, sizeof(type));
1152
1153         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1154         if (!mflow) {
1155                 err = -ENOMEM;
1156                 goto err_free;
1157         }
1158
1159         switch (flow_attr->type) {
1160         case IB_FLOW_ATTR_NORMAL:
1161                 type[0] = MLX4_FS_REGULAR;
1162                 break;
1163
1164         case IB_FLOW_ATTR_ALL_DEFAULT:
1165                 type[0] = MLX4_FS_ALL_DEFAULT;
1166                 break;
1167
1168         case IB_FLOW_ATTR_MC_DEFAULT:
1169                 type[0] = MLX4_FS_MC_DEFAULT;
1170                 break;
1171
1172         case IB_FLOW_ATTR_SNIFFER:
1173                 type[0] = MLX4_FS_UC_SNIFFER;
1174                 type[1] = MLX4_FS_MC_SNIFFER;
1175                 break;
1176
1177         default:
1178                 err = -EINVAL;
1179                 goto err_free;
1180         }
1181
1182         while (i < ARRAY_SIZE(type) && type[i]) {
1183                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1184                                             &mflow->reg_id[i].id);
1185                 if (err)
1186                         goto err_create_flow;
1187                 i++;
1188                 if (is_bonded) {
1189                         flow_attr->port = 2;
1190                         err = __mlx4_ib_create_flow(qp, flow_attr,
1191                                                     domain, type[j],
1192                                                     &mflow->reg_id[j].mirror);
1193                         flow_attr->port = 1;
1194                         if (err)
1195                                 goto err_create_flow;
1196                         j++;
1197                 }
1198
1199         }
1200
1201         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1202                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1203                                                &mflow->reg_id[i].id);
1204                 if (err)
1205                         goto err_create_flow;
1206                 i++;
1207                 if (is_bonded) {
1208                         flow_attr->port = 2;
1209                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1210                                                        &mflow->reg_id[j].mirror);
1211                         flow_attr->port = 1;
1212                         if (err)
1213                                 goto err_create_flow;
1214                         j++;
1215                 }
1216                 /* function to create mirror rule */
1217         }
1218
1219         return &mflow->ibflow;
1220
1221 err_create_flow:
1222         while (i) {
1223                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1224                                              mflow->reg_id[i].id);
1225                 i--;
1226         }
1227
1228         while (j) {
1229                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1230                                              mflow->reg_id[j].mirror);
1231                 j--;
1232         }
1233 err_free:
1234         kfree(mflow);
1235         return ERR_PTR(err);
1236 }
1237
1238 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1239 {
1240         int err, ret = 0;
1241         int i = 0;
1242         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1243         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1244
1245         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1246                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1247                 if (err)
1248                         ret = err;
1249                 if (mflow->reg_id[i].mirror) {
1250                         err = __mlx4_ib_destroy_flow(mdev->dev,
1251                                                      mflow->reg_id[i].mirror);
1252                         if (err)
1253                                 ret = err;
1254                 }
1255                 i++;
1256         }
1257
1258         kfree(mflow);
1259         return ret;
1260 }
1261
1262 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1263 {
1264         int err;
1265         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1266         struct mlx4_dev *dev = mdev->dev;
1267         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1268         struct mlx4_ib_steering *ib_steering = NULL;
1269         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1270                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1271         struct mlx4_flow_reg_id reg_id;
1272
1273         if (mdev->dev->caps.steering_mode ==
1274             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1275                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1276                 if (!ib_steering)
1277                         return -ENOMEM;
1278         }
1279
1280         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1281                                     !!(mqp->flags &
1282                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1283                                     prot, &reg_id.id);
1284         if (err)
1285                 goto err_malloc;
1286
1287         reg_id.mirror = 0;
1288         if (mlx4_is_bonded(dev)) {
1289                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, 2,
1290                                             !!(mqp->flags &
1291                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1292                                             prot, &reg_id.mirror);
1293                 if (err)
1294                         goto err_add;
1295         }
1296
1297         err = add_gid_entry(ibqp, gid);
1298         if (err)
1299                 goto err_add;
1300
1301         if (ib_steering) {
1302                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1303                 ib_steering->reg_id = reg_id;
1304                 mutex_lock(&mqp->mutex);
1305                 list_add(&ib_steering->list, &mqp->steering_rules);
1306                 mutex_unlock(&mqp->mutex);
1307         }
1308         return 0;
1309
1310 err_add:
1311         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1312                               prot, reg_id.id);
1313         if (reg_id.mirror)
1314                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1315                                       prot, reg_id.mirror);
1316 err_malloc:
1317         kfree(ib_steering);
1318
1319         return err;
1320 }
1321
1322 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1323 {
1324         struct mlx4_ib_gid_entry *ge;
1325         struct mlx4_ib_gid_entry *tmp;
1326         struct mlx4_ib_gid_entry *ret = NULL;
1327
1328         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1329                 if (!memcmp(raw, ge->gid.raw, 16)) {
1330                         ret = ge;
1331                         break;
1332                 }
1333         }
1334
1335         return ret;
1336 }
1337
1338 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1339 {
1340         int err;
1341         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1342         struct mlx4_dev *dev = mdev->dev;
1343         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1344         struct net_device *ndev;
1345         struct mlx4_ib_gid_entry *ge;
1346         struct mlx4_flow_reg_id reg_id = {0, 0};
1347
1348         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1349                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1350
1351         if (mdev->dev->caps.steering_mode ==
1352             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1353                 struct mlx4_ib_steering *ib_steering;
1354
1355                 mutex_lock(&mqp->mutex);
1356                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1357                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1358                                 list_del(&ib_steering->list);
1359                                 break;
1360                         }
1361                 }
1362                 mutex_unlock(&mqp->mutex);
1363                 if (&ib_steering->list == &mqp->steering_rules) {
1364                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1365                         return -EINVAL;
1366                 }
1367                 reg_id = ib_steering->reg_id;
1368                 kfree(ib_steering);
1369         }
1370
1371         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1372                                     prot, reg_id.id);
1373         if (err)
1374                 return err;
1375
1376         if (mlx4_is_bonded(dev)) {
1377                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1378                                             prot, reg_id.mirror);
1379                 if (err)
1380                         return err;
1381         }
1382
1383         mutex_lock(&mqp->mutex);
1384         ge = find_gid_entry(mqp, gid->raw);
1385         if (ge) {
1386                 spin_lock_bh(&mdev->iboe.lock);
1387                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1388                 if (ndev)
1389                         dev_hold(ndev);
1390                 spin_unlock_bh(&mdev->iboe.lock);
1391                 if (ndev)
1392                         dev_put(ndev);
1393                 list_del(&ge->list);
1394                 kfree(ge);
1395         } else
1396                 pr_warn("could not find mgid entry\n");
1397
1398         mutex_unlock(&mqp->mutex);
1399
1400         return 0;
1401 }
1402
1403 static int init_node_data(struct mlx4_ib_dev *dev)
1404 {
1405         struct ib_smp *in_mad  = NULL;
1406         struct ib_smp *out_mad = NULL;
1407         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1408         int err = -ENOMEM;
1409
1410         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1411         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1412         if (!in_mad || !out_mad)
1413                 goto out;
1414
1415         init_query_mad(in_mad);
1416         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1417         if (mlx4_is_master(dev->dev))
1418                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1419
1420         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1421         if (err)
1422                 goto out;
1423
1424         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1425
1426         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1427
1428         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1429         if (err)
1430                 goto out;
1431
1432         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1433         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1434
1435 out:
1436         kfree(in_mad);
1437         kfree(out_mad);
1438         return err;
1439 }
1440
1441 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1442                         char *buf)
1443 {
1444         struct mlx4_ib_dev *dev =
1445                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1446         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1447 }
1448
1449 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1450                            char *buf)
1451 {
1452         struct mlx4_ib_dev *dev =
1453                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1454         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1455                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1456                        (int) dev->dev->caps.fw_ver & 0xffff);
1457 }
1458
1459 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1460                         char *buf)
1461 {
1462         struct mlx4_ib_dev *dev =
1463                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1464         return sprintf(buf, "%x\n", dev->dev->rev_id);
1465 }
1466
1467 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1468                           char *buf)
1469 {
1470         struct mlx4_ib_dev *dev =
1471                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1472         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1473                        dev->dev->board_id);
1474 }
1475
1476 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1477 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1478 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1479 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1480
1481 static struct device_attribute *mlx4_class_attributes[] = {
1482         &dev_attr_hw_rev,
1483         &dev_attr_fw_ver,
1484         &dev_attr_hca_type,
1485         &dev_attr_board_id
1486 };
1487
1488 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1489                                      struct net_device *dev)
1490 {
1491         memcpy(eui, dev->dev_addr, 3);
1492         memcpy(eui + 5, dev->dev_addr + 3, 3);
1493         if (vlan_id < 0x1000) {
1494                 eui[3] = vlan_id >> 8;
1495                 eui[4] = vlan_id & 0xff;
1496         } else {
1497                 eui[3] = 0xff;
1498                 eui[4] = 0xfe;
1499         }
1500         eui[0] ^= 2;
1501 }
1502
1503 static void update_gids_task(struct work_struct *work)
1504 {
1505         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1506         struct mlx4_cmd_mailbox *mailbox;
1507         union ib_gid *gids;
1508         int err;
1509         struct mlx4_dev *dev = gw->dev->dev;
1510         int is_bonded = mlx4_is_bonded(dev);
1511
1512         if (!gw->dev->ib_active)
1513                 return;
1514
1515         mailbox = mlx4_alloc_cmd_mailbox(dev);
1516         if (IS_ERR(mailbox)) {
1517                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1518                 return;
1519         }
1520
1521         gids = mailbox->buf;
1522         memcpy(gids, gw->gids, sizeof gw->gids);
1523
1524         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1525                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1526                        MLX4_CMD_WRAPPED);
1527         if (err)
1528                 pr_warn("set port command failed\n");
1529         else
1530                 if ((gw->port == 1) || !is_bonded)
1531                         mlx4_ib_dispatch_event(gw->dev,
1532                                                is_bonded ? 1 : gw->port,
1533                                                IB_EVENT_GID_CHANGE);
1534
1535         mlx4_free_cmd_mailbox(dev, mailbox);
1536         kfree(gw);
1537 }
1538
1539 static void reset_gids_task(struct work_struct *work)
1540 {
1541         struct update_gid_work *gw =
1542                         container_of(work, struct update_gid_work, work);
1543         struct mlx4_cmd_mailbox *mailbox;
1544         union ib_gid *gids;
1545         int err;
1546         struct mlx4_dev *dev = gw->dev->dev;
1547
1548         if (!gw->dev->ib_active)
1549                 return;
1550
1551         mailbox = mlx4_alloc_cmd_mailbox(dev);
1552         if (IS_ERR(mailbox)) {
1553                 pr_warn("reset gid table failed\n");
1554                 goto free;
1555         }
1556
1557         gids = mailbox->buf;
1558         memcpy(gids, gw->gids, sizeof(gw->gids));
1559
1560         if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1561                                     IB_LINK_LAYER_ETHERNET) {
1562                 err = mlx4_cmd(dev, mailbox->dma,
1563                                MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1564                                1, MLX4_CMD_SET_PORT,
1565                                MLX4_CMD_TIME_CLASS_B,
1566                                MLX4_CMD_WRAPPED);
1567                 if (err)
1568                         pr_warn(KERN_WARNING
1569                                 "set port %d command failed\n", gw->port);
1570         }
1571
1572         mlx4_free_cmd_mailbox(dev, mailbox);
1573 free:
1574         kfree(gw);
1575 }
1576
1577 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1578                             union ib_gid *gid, int clear,
1579                             int default_gid)
1580 {
1581         struct update_gid_work *work;
1582         int i;
1583         int need_update = 0;
1584         int free = -1;
1585         int found = -1;
1586         int max_gids;
1587
1588         if (default_gid) {
1589                 free = 0;
1590         } else {
1591                 max_gids = dev->dev->caps.gid_table_len[port];
1592                 for (i = 1; i < max_gids; ++i) {
1593                         if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1594                                     sizeof(*gid)))
1595                                 found = i;
1596
1597                         if (clear) {
1598                                 if (found >= 0) {
1599                                         need_update = 1;
1600                                         dev->iboe.gid_table[port - 1][found] =
1601                                                 zgid;
1602                                         break;
1603                                 }
1604                         } else {
1605                                 if (found >= 0)
1606                                         break;
1607
1608                                 if (free < 0 &&
1609                                     !memcmp(&dev->iboe.gid_table[port - 1][i],
1610                                             &zgid, sizeof(*gid)))
1611                                         free = i;
1612                         }
1613                 }
1614         }
1615
1616         if (found == -1 && !clear && free >= 0) {
1617                 dev->iboe.gid_table[port - 1][free] = *gid;
1618                 need_update = 1;
1619         }
1620
1621         if (!need_update)
1622                 return 0;
1623
1624         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1625         if (!work)
1626                 return -ENOMEM;
1627
1628         memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1629         INIT_WORK(&work->work, update_gids_task);
1630         work->port = port;
1631         work->dev = dev;
1632         queue_work(wq, &work->work);
1633
1634         return 0;
1635 }
1636
1637 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1638 {
1639         gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1640         mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1641 }
1642
1643
1644 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1645 {
1646         struct update_gid_work *work;
1647
1648         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1649         if (!work)
1650                 return -ENOMEM;
1651
1652         memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1653         memset(work->gids, 0, sizeof(work->gids));
1654         INIT_WORK(&work->work, reset_gids_task);
1655         work->dev = dev;
1656         work->port = port;
1657         queue_work(wq, &work->work);
1658         return 0;
1659 }
1660
1661 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1662                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1663 {
1664         struct mlx4_ib_iboe *iboe;
1665         int port = 0;
1666         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1667                                 rdma_vlan_dev_real_dev(event_netdev) :
1668                                 event_netdev;
1669         union ib_gid default_gid;
1670
1671         mlx4_make_default_gid(real_dev, &default_gid);
1672
1673         if (!memcmp(gid, &default_gid, sizeof(*gid)))
1674                 return 0;
1675
1676         if (event != NETDEV_DOWN && event != NETDEV_UP)
1677                 return 0;
1678
1679         if ((real_dev != event_netdev) &&
1680             (event == NETDEV_DOWN) &&
1681             rdma_link_local_addr((struct in6_addr *)gid))
1682                 return 0;
1683
1684         iboe = &ibdev->iboe;
1685         spin_lock_bh(&iboe->lock);
1686
1687         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1688                 if ((netif_is_bond_master(real_dev) &&
1689                      (real_dev == iboe->masters[port - 1])) ||
1690                      (!netif_is_bond_master(real_dev) &&
1691                      (real_dev == iboe->netdevs[port - 1])))
1692                         update_gid_table(ibdev, port, gid,
1693                                          event == NETDEV_DOWN, 0);
1694
1695         spin_unlock_bh(&iboe->lock);
1696         return 0;
1697
1698 }
1699
1700 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1701                                struct mlx4_ib_dev *ibdev)
1702 {
1703         u8 port = 0;
1704         struct mlx4_ib_iboe *iboe;
1705         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1706                                 rdma_vlan_dev_real_dev(dev) : dev;
1707
1708         iboe = &ibdev->iboe;
1709
1710         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1711                 if ((netif_is_bond_master(real_dev) &&
1712                      (real_dev == iboe->masters[port - 1])) ||
1713                      (!netif_is_bond_master(real_dev) &&
1714                      (real_dev == iboe->netdevs[port - 1])))
1715                         break;
1716
1717         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1718                 return 0;
1719         else
1720                 return port;
1721 }
1722
1723 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1724                                 void *ptr)
1725 {
1726         struct mlx4_ib_dev *ibdev;
1727         struct in_ifaddr *ifa = ptr;
1728         union ib_gid gid;
1729         struct net_device *event_netdev = ifa->ifa_dev->dev;
1730
1731         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1732
1733         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1734
1735         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1736         return NOTIFY_DONE;
1737 }
1738
1739 #if IS_ENABLED(CONFIG_IPV6)
1740 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1741                                 void *ptr)
1742 {
1743         struct mlx4_ib_dev *ibdev;
1744         struct inet6_ifaddr *ifa = ptr;
1745         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1746         struct net_device *event_netdev = ifa->idev->dev;
1747
1748         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1749
1750         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1751         return NOTIFY_DONE;
1752 }
1753 #endif
1754
1755 #define MLX4_IB_INVALID_MAC     ((u64)-1)
1756 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1757                                struct net_device *dev,
1758                                int port)
1759 {
1760         u64 new_smac = 0;
1761         u64 release_mac = MLX4_IB_INVALID_MAC;
1762         struct mlx4_ib_qp *qp;
1763
1764         read_lock(&dev_base_lock);
1765         new_smac = mlx4_mac_to_u64(dev->dev_addr);
1766         read_unlock(&dev_base_lock);
1767
1768         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1769
1770         /* no need for update QP1 and mac registration in non-SRIOV */
1771         if (!mlx4_is_mfunc(ibdev->dev))
1772                 return;
1773
1774         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1775         qp = ibdev->qp1_proxy[port - 1];
1776         if (qp) {
1777                 int new_smac_index;
1778                 u64 old_smac;
1779                 struct mlx4_update_qp_params update_params;
1780
1781                 mutex_lock(&qp->mutex);
1782                 old_smac = qp->pri.smac;
1783                 if (new_smac == old_smac)
1784                         goto unlock;
1785
1786                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1787
1788                 if (new_smac_index < 0)
1789                         goto unlock;
1790
1791                 update_params.smac_index = new_smac_index;
1792                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1793                                    &update_params)) {
1794                         release_mac = new_smac;
1795                         goto unlock;
1796                 }
1797                 /* if old port was zero, no mac was yet registered for this QP */
1798                 if (qp->pri.smac_port)
1799                         release_mac = old_smac;
1800                 qp->pri.smac = new_smac;
1801                 qp->pri.smac_port = port;
1802                 qp->pri.smac_index = new_smac_index;
1803         }
1804
1805 unlock:
1806         if (release_mac != MLX4_IB_INVALID_MAC)
1807                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1808         if (qp)
1809                 mutex_unlock(&qp->mutex);
1810         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1811 }
1812
1813 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1814                                  struct mlx4_ib_dev *ibdev, u8 port)
1815 {
1816         struct in_device *in_dev;
1817 #if IS_ENABLED(CONFIG_IPV6)
1818         struct inet6_dev *in6_dev;
1819         union ib_gid  *pgid;
1820         struct inet6_ifaddr *ifp;
1821         union ib_gid default_gid;
1822 #endif
1823         union ib_gid gid;
1824
1825
1826         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1827                 return;
1828
1829         /* IPv4 gids */
1830         in_dev = in_dev_get(dev);
1831         if (in_dev) {
1832                 for_ifa(in_dev) {
1833                         /*ifa->ifa_address;*/
1834                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1835                                                (struct in6_addr *)&gid);
1836                         update_gid_table(ibdev, port, &gid, 0, 0);
1837                 }
1838                 endfor_ifa(in_dev);
1839                 in_dev_put(in_dev);
1840         }
1841 #if IS_ENABLED(CONFIG_IPV6)
1842         mlx4_make_default_gid(dev, &default_gid);
1843         /* IPv6 gids */
1844         in6_dev = in6_dev_get(dev);
1845         if (in6_dev) {
1846                 read_lock_bh(&in6_dev->lock);
1847                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1848                         pgid = (union ib_gid *)&ifp->addr;
1849                         if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1850                                 continue;
1851                         update_gid_table(ibdev, port, pgid, 0, 0);
1852                 }
1853                 read_unlock_bh(&in6_dev->lock);
1854                 in6_dev_put(in6_dev);
1855         }
1856 #endif
1857 }
1858
1859 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1860                                  struct  net_device *dev, u8 port)
1861 {
1862         union ib_gid gid;
1863         mlx4_make_default_gid(dev, &gid);
1864         update_gid_table(ibdev, port, &gid, 0, 1);
1865 }
1866
1867 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1868 {
1869         struct  net_device *dev;
1870         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1871         int i;
1872         int err = 0;
1873
1874         for (i = 1; i <= ibdev->num_ports; ++i) {
1875                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1876                     IB_LINK_LAYER_ETHERNET) {
1877                         err = reset_gid_table(ibdev, i);
1878                         if (err)
1879                                 goto out;
1880                 }
1881         }
1882
1883         read_lock(&dev_base_lock);
1884         spin_lock_bh(&iboe->lock);
1885
1886         for_each_netdev(&init_net, dev) {
1887                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1888                 /* port will be non-zero only for ETH ports */
1889                 if (port) {
1890                         mlx4_ib_set_default_gid(ibdev, dev, port);
1891                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1892                 }
1893         }
1894
1895         spin_unlock_bh(&iboe->lock);
1896         read_unlock(&dev_base_lock);
1897 out:
1898         return err;
1899 }
1900
1901 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1902                                  struct net_device *dev,
1903                                  unsigned long event)
1904
1905 {
1906         struct mlx4_ib_iboe *iboe;
1907         int update_qps_port = -1;
1908         int port;
1909
1910         iboe = &ibdev->iboe;
1911
1912         spin_lock_bh(&iboe->lock);
1913         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1914                 enum ib_port_state      port_state = IB_PORT_NOP;
1915                 struct net_device *old_master = iboe->masters[port - 1];
1916                 struct net_device *curr_netdev;
1917                 struct net_device *curr_master;
1918
1919                 iboe->netdevs[port - 1] =
1920                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1921                 if (iboe->netdevs[port - 1])
1922                         mlx4_ib_set_default_gid(ibdev,
1923                                                 iboe->netdevs[port - 1], port);
1924                 curr_netdev = iboe->netdevs[port - 1];
1925
1926                 if (iboe->netdevs[port - 1] &&
1927                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1928                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1929                                 iboe->netdevs[port - 1]);
1930                 } else {
1931                         iboe->masters[port - 1] = NULL;
1932                 }
1933                 curr_master = iboe->masters[port - 1];
1934
1935                 if (dev == iboe->netdevs[port - 1] &&
1936                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1937                      event == NETDEV_UP || event == NETDEV_CHANGE))
1938                         update_qps_port = port;
1939
1940                 if (curr_netdev) {
1941                         port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1942                                                 IB_PORT_ACTIVE : IB_PORT_DOWN;
1943                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1944                         if (curr_master) {
1945                                 /* if using bonding/team and a slave port is down, we
1946                                  * don't want the bond IP based gids in the table since
1947                                  * flows that select port by gid may get the down port.
1948                                 */
1949                                 if (port_state == IB_PORT_DOWN &&
1950                                     !mlx4_is_bonded(ibdev->dev)) {
1951                                         reset_gid_table(ibdev, port);
1952                                         mlx4_ib_set_default_gid(ibdev,
1953                                                                 curr_netdev,
1954                                                                 port);
1955                                 } else {
1956                                         /* gids from the upper dev (bond/team)
1957                                          * should appear in port's gid table
1958                                         */
1959                                         mlx4_ib_get_dev_addr(curr_master,
1960                                                              ibdev, port);
1961                                 }
1962                         }
1963                         /* if bonding is used it is possible that we add it to
1964                          * masters only after IP address is assigned to the
1965                          * net bonding interface.
1966                         */
1967                         if (curr_master && (old_master != curr_master)) {
1968                                 reset_gid_table(ibdev, port);
1969                                 mlx4_ib_set_default_gid(ibdev,
1970                                                         curr_netdev, port);
1971                                 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1972                         }
1973
1974                         if (!curr_master && (old_master != curr_master)) {
1975                                 reset_gid_table(ibdev, port);
1976                                 mlx4_ib_set_default_gid(ibdev,
1977                                                         curr_netdev, port);
1978                                 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1979                         }
1980                 } else {
1981                         reset_gid_table(ibdev, port);
1982                 }
1983         }
1984
1985         spin_unlock_bh(&iboe->lock);
1986
1987         if (update_qps_port > 0)
1988                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1989 }
1990
1991 static int mlx4_ib_netdev_event(struct notifier_block *this,
1992                                 unsigned long event, void *ptr)
1993 {
1994         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1995         struct mlx4_ib_dev *ibdev;
1996
1997         if (!net_eq(dev_net(dev), &init_net))
1998                 return NOTIFY_DONE;
1999
2000         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2001         mlx4_ib_scan_netdevs(ibdev, dev, event);
2002
2003         return NOTIFY_DONE;
2004 }
2005
2006 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2007 {
2008         int port;
2009         int slave;
2010         int i;
2011
2012         if (mlx4_is_master(ibdev->dev)) {
2013                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2014                      ++slave) {
2015                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2016                                 for (i = 0;
2017                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2018                                      ++i) {
2019                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2020                                         /* master has the identity virt2phys pkey mapping */
2021                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2022                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2023                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2024                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2025                                 }
2026                         }
2027                 }
2028                 /* initialize pkey cache */
2029                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2030                         for (i = 0;
2031                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2032                              ++i)
2033                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2034                                         (i) ? 0 : 0xFFFF;
2035                 }
2036         }
2037 }
2038
2039 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2040 {
2041         char name[80];
2042         int eq_per_port = 0;
2043         int added_eqs = 0;
2044         int total_eqs = 0;
2045         int i, j, eq;
2046
2047         /* Legacy mode or comp_pool is not large enough */
2048         if (dev->caps.comp_pool == 0 ||
2049             dev->caps.num_ports > dev->caps.comp_pool)
2050                 return;
2051
2052         eq_per_port = dev->caps.comp_pool / dev->caps.num_ports;
2053
2054         /* Init eq table */
2055         added_eqs = 0;
2056         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2057                 added_eqs += eq_per_port;
2058
2059         total_eqs = dev->caps.num_comp_vectors + added_eqs;
2060
2061         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
2062         if (!ibdev->eq_table)
2063                 return;
2064
2065         ibdev->eq_added = added_eqs;
2066
2067         eq = 0;
2068         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2069                 for (j = 0; j < eq_per_port; j++) {
2070                         snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
2071                                  i, j, dev->persist->pdev->bus->name);
2072                         /* Set IRQ for specific name (per ring) */
2073                         if (mlx4_assign_eq(dev, name, NULL,
2074                                            &ibdev->eq_table[eq])) {
2075                                 /* Use legacy (same as mlx4_en driver) */
2076                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2077                                 ibdev->eq_table[eq] =
2078                                         (eq % dev->caps.num_comp_vectors);
2079                         }
2080                         eq++;
2081                 }
2082         }
2083
2084         /* Fill the reset of the vector with legacy EQ */
2085         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2086                 ibdev->eq_table[eq++] = i;
2087
2088         /* Advertise the new number of EQs to clients */
2089         ibdev->ib_dev.num_comp_vectors = total_eqs;
2090 }
2091
2092 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2093 {
2094         int i;
2095
2096         /* no additional eqs were added */
2097         if (!ibdev->eq_table)
2098                 return;
2099
2100         /* Reset the advertised EQ number */
2101         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2102
2103         /* Free only the added eqs */
2104         for (i = 0; i < ibdev->eq_added; i++) {
2105                 /* Don't free legacy eqs if used */
2106                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2107                         continue;
2108                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2109         }
2110
2111         kfree(ibdev->eq_table);
2112 }
2113
2114 static void *mlx4_ib_add(struct mlx4_dev *dev)
2115 {
2116         struct mlx4_ib_dev *ibdev;
2117         int num_ports = 0;
2118         int i, j;
2119         int err;
2120         struct mlx4_ib_iboe *iboe;
2121         int ib_num_ports = 0;
2122         int num_req_counters;
2123
2124         pr_info_once("%s", mlx4_ib_version);
2125
2126         num_ports = 0;
2127         mlx4_foreach_ib_transport_port(i, dev)
2128                 num_ports++;
2129
2130         /* No point in registering a device with no ports... */
2131         if (num_ports == 0)
2132                 return NULL;
2133
2134         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2135         if (!ibdev) {
2136                 dev_err(&dev->persist->pdev->dev,
2137                         "Device struct alloc failed\n");
2138                 return NULL;
2139         }
2140
2141         iboe = &ibdev->iboe;
2142
2143         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2144                 goto err_dealloc;
2145
2146         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2147                 goto err_pd;
2148
2149         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2150                                  PAGE_SIZE);
2151         if (!ibdev->uar_map)
2152                 goto err_uar;
2153         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2154
2155         ibdev->dev = dev;
2156         ibdev->bond_next_port   = 0;
2157
2158         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2159         ibdev->ib_dev.owner             = THIS_MODULE;
2160         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2161         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2162         ibdev->num_ports                = num_ports;
2163         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2164                                                 1 : ibdev->num_ports;
2165         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2166         ibdev->ib_dev.dma_device        = &dev->persist->pdev->dev;
2167
2168         if (dev->caps.userspace_caps)
2169                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2170         else
2171                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2172
2173         ibdev->ib_dev.uverbs_cmd_mask   =
2174                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2175                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2176                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2177                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2178                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2179                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2180                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2181                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2182                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2183                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2184                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2185                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2186                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2187                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2188                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2189                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2190                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2191                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2192                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2193                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2194                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2195                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2196                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2197                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2198
2199         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2200         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2201         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2202         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2203         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2204         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2205         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2206         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2207         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2208         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2209         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2210         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2211         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2212         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2213         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2214         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2215         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2216         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2217         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2218         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2219         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2220         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2221         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2222         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2223         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2224         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2225         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2226         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2227         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2228         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2229         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2230         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2231         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2232         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2233         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2234         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2235         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2236         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2237         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2238         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2239         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2240         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2241
2242         if (!mlx4_is_slave(ibdev->dev)) {
2243                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2244                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2245                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2246                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2247         }
2248
2249         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2250             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2251                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2252                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2253                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2254
2255                 ibdev->ib_dev.uverbs_cmd_mask |=
2256                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2257                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2258         }
2259
2260         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2261                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2262                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2263                 ibdev->ib_dev.uverbs_cmd_mask |=
2264                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2265                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2266         }
2267
2268         if (check_flow_steering_support(dev)) {
2269                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2270                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2271                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2272
2273                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2274                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2275                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2276         }
2277
2278         mlx4_ib_alloc_eqs(dev, ibdev);
2279
2280         spin_lock_init(&iboe->lock);
2281
2282         if (init_node_data(ibdev))
2283                 goto err_map;
2284
2285         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2286         for (i = 0; i < num_req_counters; ++i) {
2287                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2288                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2289                                                 IB_LINK_LAYER_ETHERNET) {
2290                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2291                         if (err)
2292                                 ibdev->counters[i] = -1;
2293                 } else {
2294                         ibdev->counters[i] = -1;
2295                 }
2296         }
2297         if (mlx4_is_bonded(dev))
2298                 for (i = 1; i < ibdev->num_ports ; ++i)
2299                         ibdev->counters[i] = ibdev->counters[0];
2300
2301
2302         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2303                 ib_num_ports++;
2304
2305         spin_lock_init(&ibdev->sm_lock);
2306         mutex_init(&ibdev->cap_mask_mutex);
2307
2308         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2309             ib_num_ports) {
2310                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2311                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2312                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2313                                             &ibdev->steer_qpn_base, 0);
2314                 if (err)
2315                         goto err_counter;
2316
2317                 ibdev->ib_uc_qpns_bitmap =
2318                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2319                                 sizeof(long),
2320                                 GFP_KERNEL);
2321                 if (!ibdev->ib_uc_qpns_bitmap) {
2322                         dev_err(&dev->persist->pdev->dev,
2323                                 "bit map alloc failed\n");
2324                         goto err_steer_qp_release;
2325                 }
2326
2327                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2328
2329                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2330                                 dev, ibdev->steer_qpn_base,
2331                                 ibdev->steer_qpn_base +
2332                                 ibdev->steer_qpn_count - 1);
2333                 if (err)
2334                         goto err_steer_free_bitmap;
2335         }
2336
2337         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2338                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2339
2340         if (ib_register_device(&ibdev->ib_dev, NULL))
2341                 goto err_steer_free_bitmap;
2342
2343         if (mlx4_ib_mad_init(ibdev))
2344                 goto err_reg;
2345
2346         if (mlx4_ib_init_sriov(ibdev))
2347                 goto err_mad;
2348
2349         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2350                 if (!iboe->nb.notifier_call) {
2351                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2352                         err = register_netdevice_notifier(&iboe->nb);
2353                         if (err) {
2354                                 iboe->nb.notifier_call = NULL;
2355                                 goto err_notif;
2356                         }
2357                 }
2358                 if (!iboe->nb_inet.notifier_call) {
2359                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2360                         err = register_inetaddr_notifier(&iboe->nb_inet);
2361                         if (err) {
2362                                 iboe->nb_inet.notifier_call = NULL;
2363                                 goto err_notif;
2364                         }
2365                 }
2366 #if IS_ENABLED(CONFIG_IPV6)
2367                 if (!iboe->nb_inet6.notifier_call) {
2368                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2369                         err = register_inet6addr_notifier(&iboe->nb_inet6);
2370                         if (err) {
2371                                 iboe->nb_inet6.notifier_call = NULL;
2372                                 goto err_notif;
2373                         }
2374                 }
2375 #endif
2376                 if (mlx4_ib_init_gid_table(ibdev))
2377                         goto err_notif;
2378         }
2379
2380         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2381                 if (device_create_file(&ibdev->ib_dev.dev,
2382                                        mlx4_class_attributes[j]))
2383                         goto err_notif;
2384         }
2385
2386         ibdev->ib_active = true;
2387
2388         if (mlx4_is_mfunc(ibdev->dev))
2389                 init_pkeys(ibdev);
2390
2391         /* create paravirt contexts for any VFs which are active */
2392         if (mlx4_is_master(ibdev->dev)) {
2393                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2394                         if (j == mlx4_master_func_num(ibdev->dev))
2395                                 continue;
2396                         if (mlx4_is_slave_active(ibdev->dev, j))
2397                                 do_slave_init(ibdev, j, 1);
2398                 }
2399         }
2400         return ibdev;
2401
2402 err_notif:
2403         if (ibdev->iboe.nb.notifier_call) {
2404                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2405                         pr_warn("failure unregistering notifier\n");
2406                 ibdev->iboe.nb.notifier_call = NULL;
2407         }
2408         if (ibdev->iboe.nb_inet.notifier_call) {
2409                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2410                         pr_warn("failure unregistering notifier\n");
2411                 ibdev->iboe.nb_inet.notifier_call = NULL;
2412         }
2413 #if IS_ENABLED(CONFIG_IPV6)
2414         if (ibdev->iboe.nb_inet6.notifier_call) {
2415                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2416                         pr_warn("failure unregistering notifier\n");
2417                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2418         }
2419 #endif
2420         flush_workqueue(wq);
2421
2422         mlx4_ib_close_sriov(ibdev);
2423
2424 err_mad:
2425         mlx4_ib_mad_cleanup(ibdev);
2426
2427 err_reg:
2428         ib_unregister_device(&ibdev->ib_dev);
2429
2430 err_steer_free_bitmap:
2431         kfree(ibdev->ib_uc_qpns_bitmap);
2432
2433 err_steer_qp_release:
2434         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2435                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2436                                       ibdev->steer_qpn_count);
2437 err_counter:
2438         for (; i; --i)
2439                 if (ibdev->counters[i - 1] != -1)
2440                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2441
2442 err_map:
2443         iounmap(ibdev->uar_map);
2444
2445 err_uar:
2446         mlx4_uar_free(dev, &ibdev->priv_uar);
2447
2448 err_pd:
2449         mlx4_pd_free(dev, ibdev->priv_pdn);
2450
2451 err_dealloc:
2452         ib_dealloc_device(&ibdev->ib_dev);
2453
2454         return NULL;
2455 }
2456
2457 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2458 {
2459         int offset;
2460
2461         WARN_ON(!dev->ib_uc_qpns_bitmap);
2462
2463         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2464                                          dev->steer_qpn_count,
2465                                          get_count_order(count));
2466         if (offset < 0)
2467                 return offset;
2468
2469         *qpn = dev->steer_qpn_base + offset;
2470         return 0;
2471 }
2472
2473 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2474 {
2475         if (!qpn ||
2476             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2477                 return;
2478
2479         BUG_ON(qpn < dev->steer_qpn_base);
2480
2481         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2482                               qpn - dev->steer_qpn_base,
2483                               get_count_order(count));
2484 }
2485
2486 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2487                          int is_attach)
2488 {
2489         int err;
2490         size_t flow_size;
2491         struct ib_flow_attr *flow = NULL;
2492         struct ib_flow_spec_ib *ib_spec;
2493
2494         if (is_attach) {
2495                 flow_size = sizeof(struct ib_flow_attr) +
2496                             sizeof(struct ib_flow_spec_ib);
2497                 flow = kzalloc(flow_size, GFP_KERNEL);
2498                 if (!flow)
2499                         return -ENOMEM;
2500                 flow->port = mqp->port;
2501                 flow->num_of_specs = 1;
2502                 flow->size = flow_size;
2503                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2504                 ib_spec->type = IB_FLOW_SPEC_IB;
2505                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2506                 /* Add an empty rule for IB L2 */
2507                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2508
2509                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2510                                             IB_FLOW_DOMAIN_NIC,
2511                                             MLX4_FS_REGULAR,
2512                                             &mqp->reg_id);
2513         } else {
2514                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2515         }
2516         kfree(flow);
2517         return err;
2518 }
2519
2520 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2521 {
2522         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2523         int p;
2524
2525         ibdev->ib_active = false;
2526         flush_workqueue(wq);
2527
2528         mlx4_ib_close_sriov(ibdev);
2529         mlx4_ib_mad_cleanup(ibdev);
2530         ib_unregister_device(&ibdev->ib_dev);
2531         if (ibdev->iboe.nb.notifier_call) {
2532                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2533                         pr_warn("failure unregistering notifier\n");
2534                 ibdev->iboe.nb.notifier_call = NULL;
2535         }
2536
2537         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2538                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2539                                       ibdev->steer_qpn_count);
2540                 kfree(ibdev->ib_uc_qpns_bitmap);
2541         }
2542
2543         if (ibdev->iboe.nb_inet.notifier_call) {
2544                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2545                         pr_warn("failure unregistering notifier\n");
2546                 ibdev->iboe.nb_inet.notifier_call = NULL;
2547         }
2548 #if IS_ENABLED(CONFIG_IPV6)
2549         if (ibdev->iboe.nb_inet6.notifier_call) {
2550                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2551                         pr_warn("failure unregistering notifier\n");
2552                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2553         }
2554 #endif
2555
2556         iounmap(ibdev->uar_map);
2557         for (p = 0; p < ibdev->num_ports; ++p)
2558                 if (ibdev->counters[p] != -1)
2559                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2560         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2561                 mlx4_CLOSE_PORT(dev, p);
2562
2563         mlx4_ib_free_eqs(dev, ibdev);
2564
2565         mlx4_uar_free(dev, &ibdev->priv_uar);
2566         mlx4_pd_free(dev, ibdev->priv_pdn);
2567         ib_dealloc_device(&ibdev->ib_dev);
2568 }
2569
2570 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2571 {
2572         struct mlx4_ib_demux_work **dm = NULL;
2573         struct mlx4_dev *dev = ibdev->dev;
2574         int i;
2575         unsigned long flags;
2576         struct mlx4_active_ports actv_ports;
2577         unsigned int ports;
2578         unsigned int first_port;
2579
2580         if (!mlx4_is_master(dev))
2581                 return;
2582
2583         actv_ports = mlx4_get_active_ports(dev, slave);
2584         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2585         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2586
2587         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2588         if (!dm) {
2589                 pr_err("failed to allocate memory for tunneling qp update\n");
2590                 goto out;
2591         }
2592
2593         for (i = 0; i < ports; i++) {
2594                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2595                 if (!dm[i]) {
2596                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2597                         for (i = 0; i < dev->caps.num_ports; i++) {
2598                                 if (dm[i])
2599                                         kfree(dm[i]);
2600                         }
2601                         goto out;
2602                 }
2603         }
2604         /* initialize or tear down tunnel QPs for the slave */
2605         for (i = 0; i < ports; i++) {
2606                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2607                 dm[i]->port = first_port + i + 1;
2608                 dm[i]->slave = slave;
2609                 dm[i]->do_init = do_init;
2610                 dm[i]->dev = ibdev;
2611                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2612                 if (!ibdev->sriov.is_going_down)
2613                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2614                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2615         }
2616 out:
2617         kfree(dm);
2618         return;
2619 }
2620
2621 static void handle_bonded_port_state_event(struct work_struct *work)
2622 {
2623         struct ib_event_work *ew =
2624                 container_of(work, struct ib_event_work, work);
2625         struct mlx4_ib_dev *ibdev = ew->ib_dev;
2626         enum ib_port_state bonded_port_state = IB_PORT_NOP;
2627         int i;
2628         struct ib_event ibev;
2629
2630         kfree(ew);
2631         spin_lock_bh(&ibdev->iboe.lock);
2632         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2633                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2634
2635                 enum ib_port_state curr_port_state =
2636                         (netif_running(curr_netdev) &&
2637                          netif_carrier_ok(curr_netdev)) ?
2638                         IB_PORT_ACTIVE : IB_PORT_DOWN;
2639
2640                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2641                         curr_port_state : IB_PORT_ACTIVE;
2642         }
2643         spin_unlock_bh(&ibdev->iboe.lock);
2644
2645         ibev.device = &ibdev->ib_dev;
2646         ibev.element.port_num = 1;
2647         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2648                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2649
2650         ib_dispatch_event(&ibev);
2651 }
2652
2653 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2654                           enum mlx4_dev_event event, unsigned long param)
2655 {
2656         struct ib_event ibev;
2657         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2658         struct mlx4_eqe *eqe = NULL;
2659         struct ib_event_work *ew;
2660         int p = 0;
2661
2662         if (mlx4_is_bonded(dev) &&
2663             ((event == MLX4_DEV_EVENT_PORT_UP) ||
2664             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2665                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2666                 if (!ew)
2667                         return;
2668                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
2669                 ew->ib_dev = ibdev;
2670                 queue_work(wq, &ew->work);
2671                 return;
2672         }
2673
2674         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2675                 eqe = (struct mlx4_eqe *)param;
2676         else
2677                 p = (int) param;
2678
2679         switch (event) {
2680         case MLX4_DEV_EVENT_PORT_UP:
2681                 if (p > ibdev->num_ports)
2682                         return;
2683                 if (mlx4_is_master(dev) &&
2684                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2685                         IB_LINK_LAYER_INFINIBAND) {
2686                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2687                 }
2688                 ibev.event = IB_EVENT_PORT_ACTIVE;
2689                 break;
2690
2691         case MLX4_DEV_EVENT_PORT_DOWN:
2692                 if (p > ibdev->num_ports)
2693                         return;
2694                 ibev.event = IB_EVENT_PORT_ERR;
2695                 break;
2696
2697         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2698                 ibdev->ib_active = false;
2699                 ibev.event = IB_EVENT_DEVICE_FATAL;
2700                 break;
2701
2702         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2703                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2704                 if (!ew) {
2705                         pr_err("failed to allocate memory for events work\n");
2706                         break;
2707                 }
2708
2709                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2710                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2711                 ew->ib_dev = ibdev;
2712                 /* need to queue only for port owner, which uses GEN_EQE */
2713                 if (mlx4_is_master(dev))
2714                         queue_work(wq, &ew->work);
2715                 else
2716                         handle_port_mgmt_change_event(&ew->work);
2717                 return;
2718
2719         case MLX4_DEV_EVENT_SLAVE_INIT:
2720                 /* here, p is the slave id */
2721                 do_slave_init(ibdev, p, 1);
2722                 return;
2723
2724         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2725                 /* here, p is the slave id */
2726                 do_slave_init(ibdev, p, 0);
2727                 return;
2728
2729         default:
2730                 return;
2731         }
2732
2733         ibev.device           = ibdev_ptr;
2734         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2735
2736         ib_dispatch_event(&ibev);
2737 }
2738
2739 static struct mlx4_interface mlx4_ib_interface = {
2740         .add            = mlx4_ib_add,
2741         .remove         = mlx4_ib_remove,
2742         .event          = mlx4_ib_event,
2743         .protocol       = MLX4_PROT_IB_IPV6,
2744         .flags          = MLX4_INTFF_BONDING
2745 };
2746
2747 static int __init mlx4_ib_init(void)
2748 {
2749         int err;
2750
2751         wq = create_singlethread_workqueue("mlx4_ib");
2752         if (!wq)
2753                 return -ENOMEM;
2754
2755         err = mlx4_ib_mcg_init();
2756         if (err)
2757                 goto clean_wq;
2758
2759         err = mlx4_register_interface(&mlx4_ib_interface);
2760         if (err)
2761                 goto clean_mcg;
2762
2763         return 0;
2764
2765 clean_mcg:
2766         mlx4_ib_mcg_destroy();
2767
2768 clean_wq:
2769         destroy_workqueue(wq);
2770         return err;
2771 }
2772
2773 static void __exit mlx4_ib_cleanup(void)
2774 {
2775         mlx4_unregister_interface(&mlx4_ib_interface);
2776         mlx4_ib_mcg_destroy();
2777         destroy_workqueue(wq);
2778 }
2779
2780 module_init(mlx4_ib_init);
2781 module_exit(mlx4_ib_cleanup);