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
  52 #include "mlx4_ib.h"
  53 #include "user.h"
  54
  55 #define DRV_NAME        MLX4_IB_DRV_NAME
  56 #define DRV_VERSION     "1.0"
  57 #define DRV_RELDATE     "April 4, 2008"
  58
  59 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  60 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  61
  62 MODULE_AUTHOR("Roland Dreier");
  63 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  64 MODULE_LICENSE("Dual BSD/GPL");
  65 MODULE_VERSION(DRV_VERSION);
  66
  67 int mlx4_ib_sm_guid_assign = 1;
  68 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  69 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
  70
  71 static const char mlx4_ib_version[] =
  72         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  73         DRV_VERSION " (" DRV_RELDATE ")\n";
  74
  75 struct update_gid_work {
  76         struct work_struct      work;
  77         union ib_gid            gids[128];
  78         struct mlx4_ib_dev     *dev;
  79         int                     port;
  80 };
  81
  82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  83
  84 static struct workqueue_struct *wq;
  85
  86 static void init_query_mad(struct ib_smp *mad)
  87 {
  88         mad->base_version  = 1;
  89         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  90         mad->class_version = 1;
  91         mad->method        = IB_MGMT_METHOD_GET;
  92 }
  93
  94 static union ib_gid zgid;
  95
  96 static int check_flow_steering_support(struct mlx4_dev *dev)
  97 {
  98         int eth_num_ports = 0;
  99         int ib_num_ports = 0;
 100
 101         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 102
 103         if (dmfs) {
 104                 int i;
 105                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 106                         eth_num_ports++;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 108                         ib_num_ports++;
 109                 dmfs &= (!ib_num_ports ||
 110                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 111                         (!eth_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 113                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 114                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 115                         dmfs = 0;
 116                 }
 117         }
 118         return dmfs;
 119 }
 120
 121 static int mlx4_ib_query_device(struct ib_device *ibdev,
 122                                 struct ib_device_attr *props)
 123 {
 124         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 125         struct ib_smp *in_mad  = NULL;
 126         struct ib_smp *out_mad = NULL;
 127         int err = -ENOMEM;
 128
 129         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 130         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 131         if (!in_mad || !out_mad)
 132                 goto out;
 133
 134         init_query_mad(in_mad);
 135         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 136
 137         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 138                            1, NULL, NULL, in_mad, out_mad);
 139         if (err)
 140                 goto out;
 141
 142         memset(props, 0, sizeof *props);
 143
 144         props->fw_ver = dev->dev->caps.fw_ver;
 145         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 146                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 147                 IB_DEVICE_SYS_IMAGE_GUID                |
 148                 IB_DEVICE_RC_RNR_NAK_GEN                |
 149                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 150         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 151                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 152         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 153                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 154         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
 155                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 156         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 157                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 158         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 159                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 160         if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
 161                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 162         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 163                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 164         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 165             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 166             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 167                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 168         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 169                 props->device_cap_flags |= IB_DEVICE_XRC;
 170         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 171                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 172         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 173                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 174                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 175                 else
 176                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 177         if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
 178                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 179         }
 180
 181         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 182                 0xffffff;
 183         props->vendor_part_id      = dev->dev->pdev->device;
 184         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 185         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 186
 187         props->max_mr_size         = ~0ull;
 188         props->page_size_cap       = dev->dev->caps.page_size_cap;
 189         props->max_qp              = dev->dev->quotas.qp;
 190         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 191         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 192                                          dev->dev->caps.max_rq_sg);
 193         props->max_cq              = dev->dev->quotas.cq;
 194         props->max_cqe             = dev->dev->caps.max_cqes;
 195         props->max_mr              = dev->dev->quotas.mpt;
 196         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 197         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 198         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 199         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 200         props->max_srq             = dev->dev->quotas.srq;
 201         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 202         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 203         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 204         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 205         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 206                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 207         props->masked_atomic_cap   = props->atomic_cap;
 208         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 209         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 210         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 211         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 212                                            props->max_mcast_grp;
 213         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 214
 215 out:
 216         kfree(in_mad);
 217         kfree(out_mad);
 218
 219         return err;
 220 }
 221
 222 static enum rdma_link_layer
 223 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 224 {
 225         struct mlx4_dev *dev = to_mdev(device)->dev;
 226
 227         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 228                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 229 }
 230
 231 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 232                               struct ib_port_attr *props, int netw_view)
 233 {
 234         struct ib_smp *in_mad  = NULL;
 235         struct ib_smp *out_mad = NULL;
 236         int ext_active_speed;
 237         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 238         int err = -ENOMEM;
 239
 240         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 241         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 242         if (!in_mad || !out_mad)
 243                 goto out;
 244
 245         init_query_mad(in_mad);
 246         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 247         in_mad->attr_mod = cpu_to_be32(port);
 248
 249         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 250                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 251
 252         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 253                                 in_mad, out_mad);
 254         if (err)
 255                 goto out;
 256
 257
 258         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 259         props->lmc              = out_mad->data[34] & 0x7;
 260         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 261         props->sm_sl            = out_mad->data[36] & 0xf;
 262         props->state            = out_mad->data[32] & 0xf;
 263         props->phys_state       = out_mad->data[33] >> 4;
 264         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 265         if (netw_view)
 266                 props->gid_tbl_len = out_mad->data[50];
 267         else
 268                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 269         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 270         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 271         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 272         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 273         props->active_width     = out_mad->data[31] & 0xf;
 274         props->active_speed     = out_mad->data[35] >> 4;
 275         props->max_mtu          = out_mad->data[41] & 0xf;
 276         props->active_mtu       = out_mad->data[36] >> 4;
 277         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 278         props->max_vl_num       = out_mad->data[37] >> 4;
 279         props->init_type_reply  = out_mad->data[41] >> 4;
 280
 281         /* Check if extended speeds (EDR/FDR/...) are supported */
 282         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 283                 ext_active_speed = out_mad->data[62] >> 4;
 284
 285                 switch (ext_active_speed) {
 286                 case 1:
 287                         props->active_speed = IB_SPEED_FDR;
 288                         break;
 289                 case 2:
 290                         props->active_speed = IB_SPEED_EDR;
 291                         break;
 292                 }
 293         }
 294
 295         /* If reported active speed is QDR, check if is FDR-10 */
 296         if (props->active_speed == IB_SPEED_QDR) {
 297                 init_query_mad(in_mad);
 298                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 299                 in_mad->attr_mod = cpu_to_be32(port);
 300
 301                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 302                                    NULL, NULL, in_mad, out_mad);
 303                 if (err)
 304                         goto out;
 305
 306                 /* Checking LinkSpeedActive for FDR-10 */
 307                 if (out_mad->data[15] & 0x1)
 308                         props->active_speed = IB_SPEED_FDR10;
 309         }
 310
 311         /* Avoid wrong speed value returned by FW if the IB link is down. */
 312         if (props->state == IB_PORT_DOWN)
 313                  props->active_speed = IB_SPEED_SDR;
 314
 315 out:
 316         kfree(in_mad);
 317         kfree(out_mad);
 318         return err;
 319 }
 320
 321 static u8 state_to_phys_state(enum ib_port_state state)
 322 {
 323         return state == IB_PORT_ACTIVE ? 5 : 3;
 324 }
 325
 326 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 327                                struct ib_port_attr *props, int netw_view)
 328 {
 329
 330         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 331         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 332         struct net_device *ndev;
 333         enum ib_mtu tmp;
 334         struct mlx4_cmd_mailbox *mailbox;
 335         int err = 0;
 336
 337         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 338         if (IS_ERR(mailbox))
 339                 return PTR_ERR(mailbox);
 340
 341         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 342                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 343                            MLX4_CMD_WRAPPED);
 344         if (err)
 345                 goto out;
 346
 347         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 348                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 349         props->active_speed     = IB_SPEED_QDR;
 350         props->port_cap_flags   = IB_PORT_CM_SUP;
 351         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 352         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 353         props->pkey_tbl_len     = 1;
 354         props->max_mtu          = IB_MTU_4096;
 355         props->max_vl_num       = 2;
 356         props->state            = IB_PORT_DOWN;
 357         props->phys_state       = state_to_phys_state(props->state);
 358         props->active_mtu       = IB_MTU_256;
 359         spin_lock(&iboe->lock);
 360         ndev = iboe->netdevs[port - 1];
 361         if (!ndev)
 362                 goto out_unlock;
 363
 364         tmp = iboe_get_mtu(ndev->mtu);
 365         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 366
 367         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 368                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 369         props->phys_state       = state_to_phys_state(props->state);
 370 out_unlock:
 371         spin_unlock(&iboe->lock);
 372 out:
 373         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 374         return err;
 375 }
 376
 377 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 378                          struct ib_port_attr *props, int netw_view)
 379 {
 380         int err;
 381
 382         memset(props, 0, sizeof *props);
 383
 384         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 385                 ib_link_query_port(ibdev, port, props, netw_view) :
 386                                 eth_link_query_port(ibdev, port, props, netw_view);
 387
 388         return err;
 389 }
 390
 391 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 392                               struct ib_port_attr *props)
 393 {
 394         /* returns host view */
 395         return __mlx4_ib_query_port(ibdev, port, props, 0);
 396 }
 397
 398 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 399                         union ib_gid *gid, int netw_view)
 400 {
 401         struct ib_smp *in_mad  = NULL;
 402         struct ib_smp *out_mad = NULL;
 403         int err = -ENOMEM;
 404         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 405         int clear = 0;
 406         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 407
 408         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 409         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 410         if (!in_mad || !out_mad)
 411                 goto out;
 412
 413         init_query_mad(in_mad);
 414         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 415         in_mad->attr_mod = cpu_to_be32(port);
 416
 417         if (mlx4_is_mfunc(dev->dev) && netw_view)
 418                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 419
 420         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 421         if (err)
 422                 goto out;
 423
 424         memcpy(gid->raw, out_mad->data + 8, 8);
 425
 426         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 427                 if (index) {
 428                         /* For any index > 0, return the null guid */
 429                         err = 0;
 430                         clear = 1;
 431                         goto out;
 432                 }
 433         }
 434
 435         init_query_mad(in_mad);
 436         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 437         in_mad->attr_mod = cpu_to_be32(index / 8);
 438
 439         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 440                            NULL, NULL, in_mad, out_mad);
 441         if (err)
 442                 goto out;
 443
 444         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 445
 446 out:
 447         if (clear)
 448                 memset(gid->raw + 8, 0, 8);
 449         kfree(in_mad);
 450         kfree(out_mad);
 451         return err;
 452 }
 453
 454 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
 455                           union ib_gid *gid)
 456 {
 457         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 458
 459         *gid = dev->iboe.gid_table[port - 1][index];
 460
 461         return 0;
 462 }
 463
 464 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 465                              union ib_gid *gid)
 466 {
 467         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
 468                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 469         else
 470                 return iboe_query_gid(ibdev, port, index, gid);
 471 }
 472
 473 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 474                          u16 *pkey, int netw_view)
 475 {
 476         struct ib_smp *in_mad  = NULL;
 477         struct ib_smp *out_mad = NULL;
 478         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 479         int err = -ENOMEM;
 480
 481         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 482         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 483         if (!in_mad || !out_mad)
 484                 goto out;
 485
 486         init_query_mad(in_mad);
 487         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 488         in_mad->attr_mod = cpu_to_be32(index / 32);
 489
 490         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 491                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 492
 493         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 494                            in_mad, out_mad);
 495         if (err)
 496                 goto out;
 497
 498         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 499
 500 out:
 501         kfree(in_mad);
 502         kfree(out_mad);
 503         return err;
 504 }
 505
 506 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 507 {
 508         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 509 }
 510
 511 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 512                                  struct ib_device_modify *props)
 513 {
 514         struct mlx4_cmd_mailbox *mailbox;
 515         unsigned long flags;
 516
 517         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 518                 return -EOPNOTSUPP;
 519
 520         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 521                 return 0;
 522
 523         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 524                 return -EOPNOTSUPP;
 525
 526         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 527         memcpy(ibdev->node_desc, props->node_desc, 64);
 528         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 529
 530         /*
 531          * If possible, pass node desc to FW, so it can generate
 532          * a 144 trap.  If cmd fails, just ignore.
 533          */
 534         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 535         if (IS_ERR(mailbox))
 536                 return 0;
 537
 538         memcpy(mailbox->buf, props->node_desc, 64);
 539         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 540                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 541
 542         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 543
 544         return 0;
 545 }
 546
 547 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 548                          u32 cap_mask)
 549 {
 550         struct mlx4_cmd_mailbox *mailbox;
 551         int err;
 552         u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 553
 554         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 555         if (IS_ERR(mailbox))
 556                 return PTR_ERR(mailbox);
 557
 558         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 559                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 560                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 561         } else {
 562                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 563                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 564         }
 565
 566         err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
 567                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
 568
 569         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 570         return err;
 571 }
 572
 573 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 574                                struct ib_port_modify *props)
 575 {
 576         struct ib_port_attr attr;
 577         u32 cap_mask;
 578         int err;
 579
 580         mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
 581
 582         err = mlx4_ib_query_port(ibdev, port, &attr);
 583         if (err)
 584                 goto out;
 585
 586         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 587                 ~props->clr_port_cap_mask;
 588
 589         err = mlx4_SET_PORT(to_mdev(ibdev), port,
 590                             !!(mask & IB_PORT_RESET_QKEY_CNTR),
 591                             cap_mask);
 592
 593 out:
 594         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 595         return err;
 596 }
 597
 598 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 599                                                   struct ib_udata *udata)
 600 {
 601         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 602         struct mlx4_ib_ucontext *context;
 603         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 604         struct mlx4_ib_alloc_ucontext_resp resp;
 605         int err;
 606
 607         if (!dev->ib_active)
 608                 return ERR_PTR(-EAGAIN);
 609
 610         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
 611                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
 612                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
 613                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 614         } else {
 615                 resp.dev_caps         = dev->dev->caps.userspace_caps;
 616                 resp.qp_tab_size      = dev->dev->caps.num_qps;
 617                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 618                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 619                 resp.cqe_size         = dev->dev->caps.cqe_size;
 620         }
 621
 622         context = kmalloc(sizeof *context, GFP_KERNEL);
 623         if (!context)
 624                 return ERR_PTR(-ENOMEM);
 625
 626         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 627         if (err) {
 628                 kfree(context);
 629                 return ERR_PTR(err);
 630         }
 631
 632         INIT_LIST_HEAD(&context->db_page_list);
 633         mutex_init(&context->db_page_mutex);
 634
 635         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
 636                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
 637         else
 638                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
 639
 640         if (err) {
 641                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 642                 kfree(context);
 643                 return ERR_PTR(-EFAULT);
 644         }
 645
 646         return &context->ibucontext;
 647 }
 648
 649 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 650 {
 651         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 652
 653         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 654         kfree(context);
 655
 656         return 0;
 657 }
 658
 659 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 660 {
 661         struct mlx4_ib_dev *dev = to_mdev(context->device);
 662
 663         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 664                 return -EINVAL;
 665
 666         if (vma->vm_pgoff == 0) {
 667                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 668
 669                 if (io_remap_pfn_range(vma, vma->vm_start,
 670                                        to_mucontext(context)->uar.pfn,
 671                                        PAGE_SIZE, vma->vm_page_prot))
 672                         return -EAGAIN;
 673         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 674                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 675
 676                 if (io_remap_pfn_range(vma, vma->vm_start,
 677                                        to_mucontext(context)->uar.pfn +
 678                                        dev->dev->caps.num_uars,
 679                                        PAGE_SIZE, vma->vm_page_prot))
 680                         return -EAGAIN;
 681         } else
 682                 return -EINVAL;
 683
 684         return 0;
 685 }
 686
 687 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 688                                       struct ib_ucontext *context,
 689                                       struct ib_udata *udata)
 690 {
 691         struct mlx4_ib_pd *pd;
 692         int err;
 693
 694         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 695         if (!pd)
 696                 return ERR_PTR(-ENOMEM);
 697
 698         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 699         if (err) {
 700                 kfree(pd);
 701                 return ERR_PTR(err);
 702         }
 703
 704         if (context)
 705                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 706                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 707                         kfree(pd);
 708                         return ERR_PTR(-EFAULT);
 709                 }
 710
 711         return &pd->ibpd;
 712 }
 713
 714 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 715 {
 716         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 717         kfree(pd);
 718
 719         return 0;
 720 }
 721
 722 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 723                                           struct ib_ucontext *context,
 724                                           struct ib_udata *udata)
 725 {
 726         struct mlx4_ib_xrcd *xrcd;
 727         int err;
 728
 729         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 730                 return ERR_PTR(-ENOSYS);
 731
 732         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 733         if (!xrcd)
 734                 return ERR_PTR(-ENOMEM);
 735
 736         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 737         if (err)
 738                 goto err1;
 739
 740         xrcd->pd = ib_alloc_pd(ibdev);
 741         if (IS_ERR(xrcd->pd)) {
 742                 err = PTR_ERR(xrcd->pd);
 743                 goto err2;
 744         }
 745
 746         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
 747         if (IS_ERR(xrcd->cq)) {
 748                 err = PTR_ERR(xrcd->cq);
 749                 goto err3;
 750         }
 751
 752         return &xrcd->ibxrcd;
 753
 754 err3:
 755         ib_dealloc_pd(xrcd->pd);
 756 err2:
 757         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 758 err1:
 759         kfree(xrcd);
 760         return ERR_PTR(err);
 761 }
 762
 763 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 764 {
 765         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 766         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 767         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 768         kfree(xrcd);
 769
 770         return 0;
 771 }
 772
 773 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 774 {
 775         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 776         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 777         struct mlx4_ib_gid_entry *ge;
 778
 779         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 780         if (!ge)
 781                 return -ENOMEM;
 782
 783         ge->gid = *gid;
 784         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 785                 ge->port = mqp->port;
 786                 ge->added = 1;
 787         }
 788
 789         mutex_lock(&mqp->mutex);
 790         list_add_tail(&ge->list, &mqp->gid_list);
 791         mutex_unlock(&mqp->mutex);
 792
 793         return 0;
 794 }
 795
 796 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 797                    union ib_gid *gid)
 798 {
 799         struct net_device *ndev;
 800         int ret = 0;
 801
 802         if (!mqp->port)
 803                 return 0;
 804
 805         spin_lock(&mdev->iboe.lock);
 806         ndev = mdev->iboe.netdevs[mqp->port - 1];
 807         if (ndev)
 808                 dev_hold(ndev);
 809         spin_unlock(&mdev->iboe.lock);
 810
 811         if (ndev) {
 812                 ret = 1;
 813                 dev_put(ndev);
 814         }
 815
 816         return ret;
 817 }
 818
 819 struct mlx4_ib_steering {
 820         struct list_head list;
 821         u64 reg_id;
 822         union ib_gid gid;
 823 };
 824
 825 static int parse_flow_attr(struct mlx4_dev *dev,
 826                            u32 qp_num,
 827                            union ib_flow_spec *ib_spec,
 828                            struct _rule_hw *mlx4_spec)
 829 {
 830         enum mlx4_net_trans_rule_id type;
 831
 832         switch (ib_spec->type) {
 833         case IB_FLOW_SPEC_ETH:
 834                 type = MLX4_NET_TRANS_RULE_ID_ETH;
 835                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
 836                        ETH_ALEN);
 837                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
 838                        ETH_ALEN);
 839                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
 840                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
 841                 break;
 842         case IB_FLOW_SPEC_IB:
 843                 type = MLX4_NET_TRANS_RULE_ID_IB;
 844                 mlx4_spec->ib.l3_qpn =
 845                         cpu_to_be32(qp_num);
 846                 mlx4_spec->ib.qpn_mask =
 847                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
 848                 break;
 849
 850
 851         case IB_FLOW_SPEC_IPV4:
 852                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
 853                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
 854                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
 855                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
 856                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
 857                 break;
 858
 859         case IB_FLOW_SPEC_TCP:
 860         case IB_FLOW_SPEC_UDP:
 861                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
 862                                         MLX4_NET_TRANS_RULE_ID_TCP :
 863                                         MLX4_NET_TRANS_RULE_ID_UDP;
 864                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
 865                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
 866                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
 867                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
 868                 break;
 869
 870         default:
 871                 return -EINVAL;
 872         }
 873         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
 874             mlx4_hw_rule_sz(dev, type) < 0)
 875                 return -EINVAL;
 876         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
 877         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
 878         return mlx4_hw_rule_sz(dev, type);
 879 }
 880
 881 struct default_rules {
 882         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 883         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 884         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
 885         __u8  link_layer;
 886 };
 887 static const struct default_rules default_table[] = {
 888         {
 889                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
 890                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
 891                 .rules_create_list = {IB_FLOW_SPEC_IB},
 892                 .link_layer = IB_LINK_LAYER_INFINIBAND
 893         }
 894 };
 895
 896 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
 897                                          struct ib_flow_attr *flow_attr)
 898 {
 899         int i, j, k;
 900         void *ib_flow;
 901         const struct default_rules *pdefault_rules = default_table;
 902         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
 903
 904         for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
 905              pdefault_rules++) {
 906                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
 907                 memset(&field_types, 0, sizeof(field_types));
 908
 909                 if (link_layer != pdefault_rules->link_layer)
 910                         continue;
 911
 912                 ib_flow = flow_attr + 1;
 913                 /* we assume the specs are sorted */
 914                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
 915                      j < flow_attr->num_of_specs; k++) {
 916                         union ib_flow_spec *current_flow =
 917                                 (union ib_flow_spec *)ib_flow;
 918
 919                         /* same layer but different type */
 920                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
 921                              (pdefault_rules->mandatory_fields[k] &
 922                               IB_FLOW_SPEC_LAYER_MASK)) &&
 923                             (current_flow->type !=
 924                              pdefault_rules->mandatory_fields[k]))
 925                                 goto out;
 926
 927                         /* same layer, try match next one */
 928                         if (current_flow->type ==
 929                             pdefault_rules->mandatory_fields[k]) {
 930                                 j++;
 931                                 ib_flow +=
 932                                         ((union ib_flow_spec *)ib_flow)->size;
 933                         }
 934                 }
 935
 936                 ib_flow = flow_attr + 1;
 937                 for (j = 0; j < flow_attr->num_of_specs;
 938                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
 939                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
 940                                 /* same layer and same type */
 941                                 if (((union ib_flow_spec *)ib_flow)->type ==
 942                                     pdefault_rules->mandatory_not_fields[k])
 943                                         goto out;
 944
 945                 return i;
 946         }
 947 out:
 948         return -1;
 949 }
 950
 951 static int __mlx4_ib_create_default_rules(
 952                 struct mlx4_ib_dev *mdev,
 953                 struct ib_qp *qp,
 954                 const struct default_rules *pdefault_rules,
 955                 struct _rule_hw *mlx4_spec) {
 956         int size = 0;
 957         int i;
 958
 959         for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
 960                         sizeof(pdefault_rules->rules_create_list[0]); i++) {
 961                 int ret;
 962                 union ib_flow_spec ib_spec;
 963                 switch (pdefault_rules->rules_create_list[i]) {
 964                 case 0:
 965                         /* no rule */
 966                         continue;
 967                 case IB_FLOW_SPEC_IB:
 968                         ib_spec.type = IB_FLOW_SPEC_IB;
 969                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
 970
 971                         break;
 972                 default:
 973                         /* invalid rule */
 974                         return -EINVAL;
 975                 }
 976                 /* We must put empty rule, qpn is being ignored */
 977                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
 978                                       mlx4_spec);
 979                 if (ret < 0) {
 980                         pr_info("invalid parsing\n");
 981                         return -EINVAL;
 982                 }
 983
 984                 mlx4_spec = (void *)mlx4_spec + ret;
 985                 size += ret;
 986         }
 987         return size;
 988 }
 989
 990 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
 991                           int domain,
 992                           enum mlx4_net_trans_promisc_mode flow_type,
 993                           u64 *reg_id)
 994 {
 995         int ret, i;
 996         int size = 0;
 997         void *ib_flow;
 998         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
 999         struct mlx4_cmd_mailbox *mailbox;
1000         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1001         int default_flow;
1002
1003         static const u16 __mlx4_domain[] = {
1004                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1005                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1006                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1007                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1008         };
1009
1010         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1011                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1012                 return -EINVAL;
1013         }
1014
1015         if (domain >= IB_FLOW_DOMAIN_NUM) {
1016                 pr_err("Invalid domain value %d\n", domain);
1017                 return -EINVAL;
1018         }
1019
1020         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1021                 return -EINVAL;
1022
1023         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1024         if (IS_ERR(mailbox))
1025                 return PTR_ERR(mailbox);
1026         ctrl = mailbox->buf;
1027
1028         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1029                                  flow_attr->priority);
1030         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1031         ctrl->port = flow_attr->port;
1032         ctrl->qpn = cpu_to_be32(qp->qp_num);
1033
1034         ib_flow = flow_attr + 1;
1035         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1036         /* Add default flows */
1037         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1038         if (default_flow >= 0) {
1039                 ret = __mlx4_ib_create_default_rules(
1040                                 mdev, qp, default_table + default_flow,
1041                                 mailbox->buf + size);
1042                 if (ret < 0) {
1043                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1044                         return -EINVAL;
1045                 }
1046                 size += ret;
1047         }
1048         for (i = 0; i < flow_attr->num_of_specs; i++) {
1049                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1050                                       mailbox->buf + size);
1051                 if (ret < 0) {
1052                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1053                         return -EINVAL;
1054                 }
1055                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1056                 size += ret;
1057         }
1058
1059         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1060                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1061                            MLX4_CMD_NATIVE);
1062         if (ret == -ENOMEM)
1063                 pr_err("mcg table is full. Fail to register network rule.\n");
1064         else if (ret == -ENXIO)
1065                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1066         else if (ret)
1067                 pr_err("Invalid argumant. Fail to register network rule.\n");
1068
1069         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1070         return ret;
1071 }
1072
1073 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1074 {
1075         int err;
1076         err = mlx4_cmd(dev, reg_id, 0, 0,
1077                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1078                        MLX4_CMD_NATIVE);
1079         if (err)
1080                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1081                        reg_id);
1082         return err;
1083 }
1084
1085 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1086                                     struct ib_flow_attr *flow_attr,
1087                                     int domain)
1088 {
1089         int err = 0, i = 0;
1090         struct mlx4_ib_flow *mflow;
1091         enum mlx4_net_trans_promisc_mode type[2];
1092
1093         memset(type, 0, sizeof(type));
1094
1095         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1096         if (!mflow) {
1097                 err = -ENOMEM;
1098                 goto err_free;
1099         }
1100
1101         switch (flow_attr->type) {
1102         case IB_FLOW_ATTR_NORMAL:
1103                 type[0] = MLX4_FS_REGULAR;
1104                 break;
1105
1106         case IB_FLOW_ATTR_ALL_DEFAULT:
1107                 type[0] = MLX4_FS_ALL_DEFAULT;
1108                 break;
1109
1110         case IB_FLOW_ATTR_MC_DEFAULT:
1111                 type[0] = MLX4_FS_MC_DEFAULT;
1112                 break;
1113
1114         case IB_FLOW_ATTR_SNIFFER:
1115                 type[0] = MLX4_FS_UC_SNIFFER;
1116                 type[1] = MLX4_FS_MC_SNIFFER;
1117                 break;
1118
1119         default:
1120                 err = -EINVAL;
1121                 goto err_free;
1122         }
1123
1124         while (i < ARRAY_SIZE(type) && type[i]) {
1125                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1126                                             &mflow->reg_id[i]);
1127                 if (err)
1128                         goto err_free;
1129                 i++;
1130         }
1131
1132         return &mflow->ibflow;
1133
1134 err_free:
1135         kfree(mflow);
1136         return ERR_PTR(err);
1137 }
1138
1139 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1140 {
1141         int err, ret = 0;
1142         int i = 0;
1143         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1144         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1145
1146         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1147                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1148                 if (err)
1149                         ret = err;
1150                 i++;
1151         }
1152
1153         kfree(mflow);
1154         return ret;
1155 }
1156
1157 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1158 {
1159         int err;
1160         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1161         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1162         u64 reg_id;
1163         struct mlx4_ib_steering *ib_steering = NULL;
1164         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1165                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1166
1167         if (mdev->dev->caps.steering_mode ==
1168             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1169                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1170                 if (!ib_steering)
1171                         return -ENOMEM;
1172         }
1173
1174         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1175                                     !!(mqp->flags &
1176                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1177                                     prot, &reg_id);
1178         if (err)
1179                 goto err_malloc;
1180
1181         err = add_gid_entry(ibqp, gid);
1182         if (err)
1183                 goto err_add;
1184
1185         if (ib_steering) {
1186                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1187                 ib_steering->reg_id = reg_id;
1188                 mutex_lock(&mqp->mutex);
1189                 list_add(&ib_steering->list, &mqp->steering_rules);
1190                 mutex_unlock(&mqp->mutex);
1191         }
1192         return 0;
1193
1194 err_add:
1195         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1196                               prot, reg_id);
1197 err_malloc:
1198         kfree(ib_steering);
1199
1200         return err;
1201 }
1202
1203 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1204 {
1205         struct mlx4_ib_gid_entry *ge;
1206         struct mlx4_ib_gid_entry *tmp;
1207         struct mlx4_ib_gid_entry *ret = NULL;
1208
1209         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1210                 if (!memcmp(raw, ge->gid.raw, 16)) {
1211                         ret = ge;
1212                         break;
1213                 }
1214         }
1215
1216         return ret;
1217 }
1218
1219 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1220 {
1221         int err;
1222         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1223         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1224         struct net_device *ndev;
1225         struct mlx4_ib_gid_entry *ge;
1226         u64 reg_id = 0;
1227         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1228                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1229
1230         if (mdev->dev->caps.steering_mode ==
1231             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1232                 struct mlx4_ib_steering *ib_steering;
1233
1234                 mutex_lock(&mqp->mutex);
1235                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1236                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1237                                 list_del(&ib_steering->list);
1238                                 break;
1239                         }
1240                 }
1241                 mutex_unlock(&mqp->mutex);
1242                 if (&ib_steering->list == &mqp->steering_rules) {
1243                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1244                         return -EINVAL;
1245                 }
1246                 reg_id = ib_steering->reg_id;
1247                 kfree(ib_steering);
1248         }
1249
1250         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1251                                     prot, reg_id);
1252         if (err)
1253                 return err;
1254
1255         mutex_lock(&mqp->mutex);
1256         ge = find_gid_entry(mqp, gid->raw);
1257         if (ge) {
1258                 spin_lock(&mdev->iboe.lock);
1259                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1260                 if (ndev)
1261                         dev_hold(ndev);
1262                 spin_unlock(&mdev->iboe.lock);
1263                 if (ndev)
1264                         dev_put(ndev);
1265                 list_del(&ge->list);
1266                 kfree(ge);
1267         } else
1268                 pr_warn("could not find mgid entry\n");
1269
1270         mutex_unlock(&mqp->mutex);
1271
1272         return 0;
1273 }
1274
1275 static int init_node_data(struct mlx4_ib_dev *dev)
1276 {
1277         struct ib_smp *in_mad  = NULL;
1278         struct ib_smp *out_mad = NULL;
1279         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1280         int err = -ENOMEM;
1281
1282         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1283         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1284         if (!in_mad || !out_mad)
1285                 goto out;
1286
1287         init_query_mad(in_mad);
1288         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1289         if (mlx4_is_master(dev->dev))
1290                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1291
1292         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1293         if (err)
1294                 goto out;
1295
1296         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1297
1298         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1299
1300         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1301         if (err)
1302                 goto out;
1303
1304         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1305         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1306
1307 out:
1308         kfree(in_mad);
1309         kfree(out_mad);
1310         return err;
1311 }
1312
1313 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1314                         char *buf)
1315 {
1316         struct mlx4_ib_dev *dev =
1317                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1318         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1319 }
1320
1321 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1322                            char *buf)
1323 {
1324         struct mlx4_ib_dev *dev =
1325                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1326         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1327                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1328                        (int) dev->dev->caps.fw_ver & 0xffff);
1329 }
1330
1331 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1332                         char *buf)
1333 {
1334         struct mlx4_ib_dev *dev =
1335                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1336         return sprintf(buf, "%x\n", dev->dev->rev_id);
1337 }
1338
1339 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1340                           char *buf)
1341 {
1342         struct mlx4_ib_dev *dev =
1343                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1344         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1345                        dev->dev->board_id);
1346 }
1347
1348 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1349 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1350 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1351 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1352
1353 static struct device_attribute *mlx4_class_attributes[] = {
1354         &dev_attr_hw_rev,
1355         &dev_attr_fw_ver,
1356         &dev_attr_hca_type,
1357         &dev_attr_board_id
1358 };
1359
1360 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1361                                      struct net_device *dev)
1362 {
1363         memcpy(eui, dev->dev_addr, 3);
1364         memcpy(eui + 5, dev->dev_addr + 3, 3);
1365         if (vlan_id < 0x1000) {
1366                 eui[3] = vlan_id >> 8;
1367                 eui[4] = vlan_id & 0xff;
1368         } else {
1369                 eui[3] = 0xff;
1370                 eui[4] = 0xfe;
1371         }
1372         eui[0] ^= 2;
1373 }
1374
1375 static void update_gids_task(struct work_struct *work)
1376 {
1377         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1378         struct mlx4_cmd_mailbox *mailbox;
1379         union ib_gid *gids;
1380         int err;
1381         struct mlx4_dev *dev = gw->dev->dev;
1382
1383         mailbox = mlx4_alloc_cmd_mailbox(dev);
1384         if (IS_ERR(mailbox)) {
1385                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1386                 return;
1387         }
1388
1389         gids = mailbox->buf;
1390         memcpy(gids, gw->gids, sizeof gw->gids);
1391
1392         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1393                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1394                        MLX4_CMD_WRAPPED);
1395         if (err)
1396                 pr_warn("set port command failed\n");
1397         else
1398                 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1399
1400         mlx4_free_cmd_mailbox(dev, mailbox);
1401         kfree(gw);
1402 }
1403
1404 static void reset_gids_task(struct work_struct *work)
1405 {
1406         struct update_gid_work *gw =
1407                         container_of(work, struct update_gid_work, work);
1408         struct mlx4_cmd_mailbox *mailbox;
1409         union ib_gid *gids;
1410         int err;
1411         struct mlx4_dev *dev = gw->dev->dev;
1412
1413         mailbox = mlx4_alloc_cmd_mailbox(dev);
1414         if (IS_ERR(mailbox)) {
1415                 pr_warn("reset gid table failed\n");
1416                 goto free;
1417         }
1418
1419         gids = mailbox->buf;
1420         memcpy(gids, gw->gids, sizeof(gw->gids));
1421
1422         if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1423                                     IB_LINK_LAYER_ETHERNET) {
1424                 err = mlx4_cmd(dev, mailbox->dma,
1425                                MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1426                                1, MLX4_CMD_SET_PORT,
1427                                MLX4_CMD_TIME_CLASS_B,
1428                                MLX4_CMD_WRAPPED);
1429                 if (err)
1430                         pr_warn(KERN_WARNING
1431                                 "set port %d command failed\n", gw->port);
1432         }
1433
1434         mlx4_free_cmd_mailbox(dev, mailbox);
1435 free:
1436         kfree(gw);
1437 }
1438
1439 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1440                             union ib_gid *gid, int clear,
1441                             int default_gid)
1442 {
1443         struct update_gid_work *work;
1444         int i;
1445         int need_update = 0;
1446         int free = -1;
1447         int found = -1;
1448         int max_gids;
1449
1450         if (default_gid) {
1451                 free = 0;
1452         } else {
1453                 max_gids = dev->dev->caps.gid_table_len[port];
1454                 for (i = 1; i < max_gids; ++i) {
1455                         if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1456                                     sizeof(*gid)))
1457                                 found = i;
1458
1459                         if (clear) {
1460                                 if (found >= 0) {
1461                                         need_update = 1;
1462                                         dev->iboe.gid_table[port - 1][found] =
1463                                                 zgid;
1464                                         break;
1465                                 }
1466                         } else {
1467                                 if (found >= 0)
1468                                         break;
1469
1470                                 if (free < 0 &&
1471                                     !memcmp(&dev->iboe.gid_table[port - 1][i],
1472                                             &zgid, sizeof(*gid)))
1473                                         free = i;
1474                         }
1475                 }
1476         }
1477
1478         if (found == -1 && !clear && free >= 0) {
1479                 dev->iboe.gid_table[port - 1][free] = *gid;
1480                 need_update = 1;
1481         }
1482
1483         if (!need_update)
1484                 return 0;
1485
1486         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1487         if (!work)
1488                 return -ENOMEM;
1489
1490         memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1491         INIT_WORK(&work->work, update_gids_task);
1492         work->port = port;
1493         work->dev = dev;
1494         queue_work(wq, &work->work);
1495
1496         return 0;
1497 }
1498
1499 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1500 {
1501         gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1502         mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1503 }
1504
1505
1506 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1507 {
1508         struct update_gid_work *work;
1509
1510
1511         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1512         if (!work)
1513                 return -ENOMEM;
1514
1515         memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1516         memset(work->gids, 0, sizeof(work->gids));
1517         INIT_WORK(&work->work, reset_gids_task);
1518         work->dev = dev;
1519         work->port = port;
1520         queue_work(wq, &work->work);
1521         return 0;
1522 }
1523
1524 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1525                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1526 {
1527         struct mlx4_ib_iboe *iboe;
1528         int port = 0;
1529         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1530                                 rdma_vlan_dev_real_dev(event_netdev) :
1531                                 event_netdev;
1532         union ib_gid default_gid;
1533
1534         mlx4_make_default_gid(real_dev, &default_gid);
1535
1536         if (!memcmp(gid, &default_gid, sizeof(*gid)))
1537                 return 0;
1538
1539         if (event != NETDEV_DOWN && event != NETDEV_UP)
1540                 return 0;
1541
1542         if ((real_dev != event_netdev) &&
1543             (event == NETDEV_DOWN) &&
1544             rdma_link_local_addr((struct in6_addr *)gid))
1545                 return 0;
1546
1547         iboe = &ibdev->iboe;
1548         spin_lock(&iboe->lock);
1549
1550         for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1551                 if ((netif_is_bond_master(real_dev) &&
1552                      (real_dev == iboe->masters[port - 1])) ||
1553                      (!netif_is_bond_master(real_dev) &&
1554                      (real_dev == iboe->netdevs[port - 1])))
1555                         update_gid_table(ibdev, port, gid,
1556                                          event == NETDEV_DOWN, 0);
1557
1558         spin_unlock(&iboe->lock);
1559         return 0;
1560
1561 }
1562
1563 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1564                                struct mlx4_ib_dev *ibdev)
1565 {
1566         u8 port = 0;
1567         struct mlx4_ib_iboe *iboe;
1568         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1569                                 rdma_vlan_dev_real_dev(dev) : dev;
1570
1571         iboe = &ibdev->iboe;
1572
1573         for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1574                 if ((netif_is_bond_master(real_dev) &&
1575                      (real_dev == iboe->masters[port - 1])) ||
1576                      (!netif_is_bond_master(real_dev) &&
1577                      (real_dev == iboe->netdevs[port - 1])))
1578                         break;
1579
1580         if ((port == 0) || (port > MLX4_MAX_PORTS))
1581                 return 0;
1582         else
1583                 return port;
1584 }
1585
1586 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1587                                 void *ptr)
1588 {
1589         struct mlx4_ib_dev *ibdev;
1590         struct in_ifaddr *ifa = ptr;
1591         union ib_gid gid;
1592         struct net_device *event_netdev = ifa->ifa_dev->dev;
1593
1594         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1595
1596         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1597
1598         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1599         return NOTIFY_DONE;
1600 }
1601
1602 #if IS_ENABLED(CONFIG_IPV6)
1603 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1604                                 void *ptr)
1605 {
1606         struct mlx4_ib_dev *ibdev;
1607         struct inet6_ifaddr *ifa = ptr;
1608         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1609         struct net_device *event_netdev = ifa->idev->dev;
1610
1611         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1612
1613         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1614         return NOTIFY_DONE;
1615 }
1616 #endif
1617
1618 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1619                                  struct mlx4_ib_dev *ibdev, u8 port)
1620 {
1621         struct in_device *in_dev;
1622 #if IS_ENABLED(CONFIG_IPV6)
1623         struct inet6_dev *in6_dev;
1624         union ib_gid  *pgid;
1625         struct inet6_ifaddr *ifp;
1626 #endif
1627         union ib_gid gid;
1628
1629
1630         if ((port == 0) || (port > MLX4_MAX_PORTS))
1631                 return;
1632
1633         /* IPv4 gids */
1634         in_dev = in_dev_get(dev);
1635         if (in_dev) {
1636                 for_ifa(in_dev) {
1637                         /*ifa->ifa_address;*/
1638                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1639                                                (struct in6_addr *)&gid);
1640                         update_gid_table(ibdev, port, &gid, 0, 0);
1641                 }
1642                 endfor_ifa(in_dev);
1643                 in_dev_put(in_dev);
1644         }
1645 #if IS_ENABLED(CONFIG_IPV6)
1646         /* IPv6 gids */
1647         in6_dev = in6_dev_get(dev);
1648         if (in6_dev) {
1649                 read_lock_bh(&in6_dev->lock);
1650                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1651                         pgid = (union ib_gid *)&ifp->addr;
1652                         update_gid_table(ibdev, port, pgid, 0, 0);
1653                 }
1654                 read_unlock_bh(&in6_dev->lock);
1655                 in6_dev_put(in6_dev);
1656         }
1657 #endif
1658 }
1659
1660 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1661                                  struct  net_device *dev, u8 port)
1662 {
1663         union ib_gid gid;
1664         mlx4_make_default_gid(dev, &gid);
1665         update_gid_table(ibdev, port, &gid, 0, 1);
1666 }
1667
1668 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1669 {
1670         struct  net_device *dev;
1671         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1672         int i;
1673
1674         for (i = 1; i <= ibdev->num_ports; ++i)
1675                 if (reset_gid_table(ibdev, i))
1676                         return -1;
1677
1678         read_lock(&dev_base_lock);
1679         spin_lock(&iboe->lock);
1680
1681         for_each_netdev(&init_net, dev) {
1682                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1683                 if (port)
1684                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1685         }
1686
1687         spin_unlock(&iboe->lock);
1688         read_unlock(&dev_base_lock);
1689
1690         return 0;
1691 }
1692
1693 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev)
1694 {
1695         struct mlx4_ib_iboe *iboe;
1696         int port;
1697
1698         iboe = &ibdev->iboe;
1699
1700         spin_lock(&iboe->lock);
1701         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1702                 struct net_device *old_master = iboe->masters[port - 1];
1703                 struct net_device *curr_master;
1704                 iboe->netdevs[port - 1] =
1705                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1706                 if (iboe->netdevs[port - 1])
1707                         mlx4_ib_set_default_gid(ibdev,
1708                                                 iboe->netdevs[port - 1], port);
1709                 if (iboe->netdevs[port - 1] &&
1710                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1711                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1712                                 iboe->netdevs[port - 1]);
1713                 }
1714                 curr_master = iboe->masters[port - 1];
1715
1716                 /* if bonding is used it is possible that we add it to masters
1717                     only after IP address is assigned to the net bonding
1718                     interface */
1719                 if (curr_master && (old_master != curr_master))
1720                         mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1721         }
1722
1723         spin_unlock(&iboe->lock);
1724 }
1725
1726 static int mlx4_ib_netdev_event(struct notifier_block *this,
1727                                 unsigned long event, void *ptr)
1728 {
1729         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1730         struct mlx4_ib_dev *ibdev;
1731
1732         if (!net_eq(dev_net(dev), &init_net))
1733                 return NOTIFY_DONE;
1734
1735         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1736         mlx4_ib_scan_netdevs(ibdev);
1737
1738         return NOTIFY_DONE;
1739 }
1740
1741 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1742 {
1743         int port;
1744         int slave;
1745         int i;
1746
1747         if (mlx4_is_master(ibdev->dev)) {
1748                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1749                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1750                                 for (i = 0;
1751                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1752                                      ++i) {
1753                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1754                                         /* master has the identity virt2phys pkey mapping */
1755                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1756                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1757                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1758                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1759                                 }
1760                         }
1761                 }
1762                 /* initialize pkey cache */
1763                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1764                         for (i = 0;
1765                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1766                              ++i)
1767                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1768                                         (i) ? 0 : 0xFFFF;
1769                 }
1770         }
1771 }
1772
1773 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1774 {
1775         char name[32];
1776         int eq_per_port = 0;
1777         int added_eqs = 0;
1778         int total_eqs = 0;
1779         int i, j, eq;
1780
1781         /* Legacy mode or comp_pool is not large enough */
1782         if (dev->caps.comp_pool == 0 ||
1783             dev->caps.num_ports > dev->caps.comp_pool)
1784                 return;
1785
1786         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1787                                         dev->caps.num_ports);
1788
1789         /* Init eq table */
1790         added_eqs = 0;
1791         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1792                 added_eqs += eq_per_port;
1793
1794         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1795
1796         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1797         if (!ibdev->eq_table)
1798                 return;
1799
1800         ibdev->eq_added = added_eqs;
1801
1802         eq = 0;
1803         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1804                 for (j = 0; j < eq_per_port; j++) {
1805                         sprintf(name, "mlx4-ib-%d-%d@%s",
1806                                 i, j, dev->pdev->bus->name);
1807                         /* Set IRQ for specific name (per ring) */
1808                         if (mlx4_assign_eq(dev, name, NULL,
1809                                            &ibdev->eq_table[eq])) {
1810                                 /* Use legacy (same as mlx4_en driver) */
1811                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1812                                 ibdev->eq_table[eq] =
1813                                         (eq % dev->caps.num_comp_vectors);
1814                         }
1815                         eq++;
1816                 }
1817         }
1818
1819         /* Fill the reset of the vector with legacy EQ */
1820         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1821                 ibdev->eq_table[eq++] = i;
1822
1823         /* Advertise the new number of EQs to clients */
1824         ibdev->ib_dev.num_comp_vectors = total_eqs;
1825 }
1826
1827 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1828 {
1829         int i;
1830
1831         /* no additional eqs were added */
1832         if (!ibdev->eq_table)
1833                 return;
1834
1835         /* Reset the advertised EQ number */
1836         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1837
1838         /* Free only the added eqs */
1839         for (i = 0; i < ibdev->eq_added; i++) {
1840                 /* Don't free legacy eqs if used */
1841                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1842                         continue;
1843                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1844         }
1845
1846         kfree(ibdev->eq_table);
1847 }
1848
1849 static void *mlx4_ib_add(struct mlx4_dev *dev)
1850 {
1851         struct mlx4_ib_dev *ibdev;
1852         int num_ports = 0;
1853         int i, j;
1854         int err;
1855         struct mlx4_ib_iboe *iboe;
1856         int ib_num_ports = 0;
1857
1858         pr_info_once("%s", mlx4_ib_version);
1859
1860         mlx4_foreach_non_ib_transport_port(i, dev)
1861                 num_ports++;
1862
1863         if (mlx4_is_mfunc(dev) && num_ports) {
1864                 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1865                 return NULL;
1866         }
1867
1868         num_ports = 0;
1869         mlx4_foreach_ib_transport_port(i, dev)
1870                 num_ports++;
1871
1872         /* No point in registering a device with no ports... */
1873         if (num_ports == 0)
1874                 return NULL;
1875
1876         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1877         if (!ibdev) {
1878                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1879                 return NULL;
1880         }
1881
1882         iboe = &ibdev->iboe;
1883
1884         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1885                 goto err_dealloc;
1886
1887         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1888                 goto err_pd;
1889
1890         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1891                                  PAGE_SIZE);
1892         if (!ibdev->uar_map)
1893                 goto err_uar;
1894         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1895
1896         ibdev->dev = dev;
1897
1898         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1899         ibdev->ib_dev.owner             = THIS_MODULE;
1900         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1901         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1902         ibdev->num_ports                = num_ports;
1903         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1904         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1905         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1906
1907         if (dev->caps.userspace_caps)
1908                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1909         else
1910                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1911
1912         ibdev->ib_dev.uverbs_cmd_mask   =
1913                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1914                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1915                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1916                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1917                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1918                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1919                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1920                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1921                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1922                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1923                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1924                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1925                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1926                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1927                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1928                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1929                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1930                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1931                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1932                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1933                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1934                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
1935                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1936
1937         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1938         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1939         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1940         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1941         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1942         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1943         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1944         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1945         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1946         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1947         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1948         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1949         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1950         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1951         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1952         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1953         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1954         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1955         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1956         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1957         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1958         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1959         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1960         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1961         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1962         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1963         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1964         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1965         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1966         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1967         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1968         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
1969         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
1970         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
1971         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
1972         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1973         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1974         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1975         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
1976         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
1977         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
1978
1979         if (!mlx4_is_slave(ibdev->dev)) {
1980                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
1981                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
1982                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
1983                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
1984         }
1985
1986         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1987             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1988                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1989                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1990                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1991
1992                 ibdev->ib_dev.uverbs_cmd_mask |=
1993                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1994                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1995         }
1996
1997         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1998                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1999                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2000                 ibdev->ib_dev.uverbs_cmd_mask |=
2001                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2002                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2003         }
2004
2005         if (check_flow_steering_support(dev)) {
2006                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2007                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2008                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2009
2010                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2011                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2012                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2013         }
2014
2015         mlx4_ib_alloc_eqs(dev, ibdev);
2016
2017         spin_lock_init(&iboe->lock);
2018
2019         if (init_node_data(ibdev))
2020                 goto err_map;
2021
2022         for (i = 0; i < ibdev->num_ports; ++i) {
2023                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2024                                                 IB_LINK_LAYER_ETHERNET) {
2025                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2026                         if (err)
2027                                 ibdev->counters[i] = -1;
2028                 } else
2029                                 ibdev->counters[i] = -1;
2030         }
2031
2032         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2033                 ib_num_ports++;
2034
2035         spin_lock_init(&ibdev->sm_lock);
2036         mutex_init(&ibdev->cap_mask_mutex);
2037
2038         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2039             ib_num_ports) {
2040                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2041                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2042                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2043                                             &ibdev->steer_qpn_base);
2044                 if (err)
2045                         goto err_counter;
2046
2047                 ibdev->ib_uc_qpns_bitmap =
2048                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2049                                 sizeof(long),
2050                                 GFP_KERNEL);
2051                 if (!ibdev->ib_uc_qpns_bitmap) {
2052                         dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2053                         goto err_steer_qp_release;
2054                 }
2055
2056                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2057
2058                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2059                                 dev, ibdev->steer_qpn_base,
2060                                 ibdev->steer_qpn_base +
2061                                 ibdev->steer_qpn_count - 1);
2062                 if (err)
2063                         goto err_steer_free_bitmap;
2064         }
2065
2066         if (ib_register_device(&ibdev->ib_dev, NULL))
2067                 goto err_steer_free_bitmap;
2068
2069         if (mlx4_ib_mad_init(ibdev))
2070                 goto err_reg;
2071
2072         if (mlx4_ib_init_sriov(ibdev))
2073                 goto err_mad;
2074
2075         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2076                 if (!iboe->nb.notifier_call) {
2077                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2078                         err = register_netdevice_notifier(&iboe->nb);
2079                         if (err) {
2080                                 iboe->nb.notifier_call = NULL;
2081                                 goto err_notif;
2082                         }
2083                 }
2084                 if (!iboe->nb_inet.notifier_call) {
2085                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2086                         err = register_inetaddr_notifier(&iboe->nb_inet);
2087                         if (err) {
2088                                 iboe->nb_inet.notifier_call = NULL;
2089                                 goto err_notif;
2090                         }
2091                 }
2092 #if IS_ENABLED(CONFIG_IPV6)
2093                 if (!iboe->nb_inet6.notifier_call) {
2094                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2095                         err = register_inet6addr_notifier(&iboe->nb_inet6);
2096                         if (err) {
2097                                 iboe->nb_inet6.notifier_call = NULL;
2098                                 goto err_notif;
2099                         }
2100                 }
2101 #endif
2102                 rtnl_lock();
2103                 mlx4_ib_scan_netdevs(ibdev);
2104                 rtnl_unlock();
2105                 mlx4_ib_init_gid_table(ibdev);
2106         }
2107
2108         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2109                 if (device_create_file(&ibdev->ib_dev.dev,
2110                                        mlx4_class_attributes[j]))
2111                         goto err_notif;
2112         }
2113
2114         ibdev->ib_active = true;
2115
2116         if (mlx4_is_mfunc(ibdev->dev))
2117                 init_pkeys(ibdev);
2118
2119         /* create paravirt contexts for any VFs which are active */
2120         if (mlx4_is_master(ibdev->dev)) {
2121                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2122                         if (j == mlx4_master_func_num(ibdev->dev))
2123                                 continue;
2124                         if (mlx4_is_slave_active(ibdev->dev, j))
2125                                 do_slave_init(ibdev, j, 1);
2126                 }
2127         }
2128         return ibdev;
2129
2130 err_notif:
2131         if (ibdev->iboe.nb.notifier_call) {
2132                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2133                         pr_warn("failure unregistering notifier\n");
2134                 ibdev->iboe.nb.notifier_call = NULL;
2135         }
2136         if (ibdev->iboe.nb_inet.notifier_call) {
2137                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2138                         pr_warn("failure unregistering notifier\n");
2139                 ibdev->iboe.nb_inet.notifier_call = NULL;
2140         }
2141 #if IS_ENABLED(CONFIG_IPV6)
2142         if (ibdev->iboe.nb_inet6.notifier_call) {
2143                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2144                         pr_warn("failure unregistering notifier\n");
2145                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2146         }
2147 #endif
2148         flush_workqueue(wq);
2149
2150         mlx4_ib_close_sriov(ibdev);
2151
2152 err_mad:
2153         mlx4_ib_mad_cleanup(ibdev);
2154
2155 err_reg:
2156         ib_unregister_device(&ibdev->ib_dev);
2157
2158 err_steer_free_bitmap:
2159         kfree(ibdev->ib_uc_qpns_bitmap);
2160
2161 err_steer_qp_release:
2162         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2163                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2164                                       ibdev->steer_qpn_count);
2165 err_counter:
2166         for (; i; --i)
2167                 if (ibdev->counters[i - 1] != -1)
2168                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2169
2170 err_map:
2171         iounmap(ibdev->uar_map);
2172
2173 err_uar:
2174         mlx4_uar_free(dev, &ibdev->priv_uar);
2175
2176 err_pd:
2177         mlx4_pd_free(dev, ibdev->priv_pdn);
2178
2179 err_dealloc:
2180         ib_dealloc_device(&ibdev->ib_dev);
2181
2182         return NULL;
2183 }
2184
2185 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2186 {
2187         int offset;
2188
2189         WARN_ON(!dev->ib_uc_qpns_bitmap);
2190
2191         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2192                                          dev->steer_qpn_count,
2193                                          get_count_order(count));
2194         if (offset < 0)
2195                 return offset;
2196
2197         *qpn = dev->steer_qpn_base + offset;
2198         return 0;
2199 }
2200
2201 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2202 {
2203         if (!qpn ||
2204             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2205                 return;
2206
2207         BUG_ON(qpn < dev->steer_qpn_base);
2208
2209         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2210                               qpn - dev->steer_qpn_base,
2211                               get_count_order(count));
2212 }
2213
2214 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2215                          int is_attach)
2216 {
2217         int err;
2218         size_t flow_size;
2219         struct ib_flow_attr *flow = NULL;
2220         struct ib_flow_spec_ib *ib_spec;
2221
2222         if (is_attach) {
2223                 flow_size = sizeof(struct ib_flow_attr) +
2224                             sizeof(struct ib_flow_spec_ib);
2225                 flow = kzalloc(flow_size, GFP_KERNEL);
2226                 if (!flow)
2227                         return -ENOMEM;
2228                 flow->port = mqp->port;
2229                 flow->num_of_specs = 1;
2230                 flow->size = flow_size;
2231                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2232                 ib_spec->type = IB_FLOW_SPEC_IB;
2233                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2234                 /* Add an empty rule for IB L2 */
2235                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2236
2237                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2238                                             IB_FLOW_DOMAIN_NIC,
2239                                             MLX4_FS_REGULAR,
2240                                             &mqp->reg_id);
2241         } else {
2242                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2243         }
2244         kfree(flow);
2245         return err;
2246 }
2247
2248 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2249 {
2250         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2251         int p;
2252
2253         mlx4_ib_close_sriov(ibdev);
2254         mlx4_ib_mad_cleanup(ibdev);
2255         ib_unregister_device(&ibdev->ib_dev);
2256         if (ibdev->iboe.nb.notifier_call) {
2257                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2258                         pr_warn("failure unregistering notifier\n");
2259                 ibdev->iboe.nb.notifier_call = NULL;
2260         }
2261
2262         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2263                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2264                                       ibdev->steer_qpn_count);
2265                 kfree(ibdev->ib_uc_qpns_bitmap);
2266         }
2267
2268         if (ibdev->iboe.nb_inet.notifier_call) {
2269                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2270                         pr_warn("failure unregistering notifier\n");
2271                 ibdev->iboe.nb_inet.notifier_call = NULL;
2272         }
2273 #if IS_ENABLED(CONFIG_IPV6)
2274         if (ibdev->iboe.nb_inet6.notifier_call) {
2275                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2276                         pr_warn("failure unregistering notifier\n");
2277                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2278         }
2279 #endif
2280
2281         iounmap(ibdev->uar_map);
2282         for (p = 0; p < ibdev->num_ports; ++p)
2283                 if (ibdev->counters[p] != -1)
2284                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2285         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2286                 mlx4_CLOSE_PORT(dev, p);
2287
2288         mlx4_ib_free_eqs(dev, ibdev);
2289
2290         mlx4_uar_free(dev, &ibdev->priv_uar);
2291         mlx4_pd_free(dev, ibdev->priv_pdn);
2292         ib_dealloc_device(&ibdev->ib_dev);
2293 }
2294
2295 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2296 {
2297         struct mlx4_ib_demux_work **dm = NULL;
2298         struct mlx4_dev *dev = ibdev->dev;
2299         int i;
2300         unsigned long flags;
2301
2302         if (!mlx4_is_master(dev))
2303                 return;
2304
2305         dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2306         if (!dm) {
2307                 pr_err("failed to allocate memory for tunneling qp update\n");
2308                 goto out;
2309         }
2310
2311         for (i = 0; i < dev->caps.num_ports; i++) {
2312                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2313                 if (!dm[i]) {
2314                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2315                         for (i = 0; i < dev->caps.num_ports; i++) {
2316                                 if (dm[i])
2317                                         kfree(dm[i]);
2318                         }
2319                         goto out;
2320                 }
2321         }
2322         /* initialize or tear down tunnel QPs for the slave */
2323         for (i = 0; i < dev->caps.num_ports; i++) {
2324                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2325                 dm[i]->port = i + 1;
2326                 dm[i]->slave = slave;
2327                 dm[i]->do_init = do_init;
2328                 dm[i]->dev = ibdev;
2329                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2330                 if (!ibdev->sriov.is_going_down)
2331                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2332                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2333         }
2334 out:
2335         kfree(dm);
2336         return;
2337 }
2338
2339 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2340                           enum mlx4_dev_event event, unsigned long param)
2341 {
2342         struct ib_event ibev;
2343         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2344         struct mlx4_eqe *eqe = NULL;
2345         struct ib_event_work *ew;
2346         int p = 0;
2347
2348         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2349                 eqe = (struct mlx4_eqe *)param;
2350         else
2351                 p = (int) param;
2352
2353         switch (event) {
2354         case MLX4_DEV_EVENT_PORT_UP:
2355                 if (p > ibdev->num_ports)
2356                         return;
2357                 if (mlx4_is_master(dev) &&
2358                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2359                         IB_LINK_LAYER_INFINIBAND) {
2360                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2361                 }
2362                 ibev.event = IB_EVENT_PORT_ACTIVE;
2363                 break;
2364
2365         case MLX4_DEV_EVENT_PORT_DOWN:
2366                 if (p > ibdev->num_ports)
2367                         return;
2368                 ibev.event = IB_EVENT_PORT_ERR;
2369                 break;
2370
2371         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2372                 ibdev->ib_active = false;
2373                 ibev.event = IB_EVENT_DEVICE_FATAL;
2374                 break;
2375
2376         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2377                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2378                 if (!ew) {
2379                         pr_err("failed to allocate memory for events work\n");
2380                         break;
2381                 }
2382
2383                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2384                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2385                 ew->ib_dev = ibdev;
2386                 /* need to queue only for port owner, which uses GEN_EQE */
2387                 if (mlx4_is_master(dev))
2388                         queue_work(wq, &ew->work);
2389                 else
2390                         handle_port_mgmt_change_event(&ew->work);
2391                 return;
2392
2393         case MLX4_DEV_EVENT_SLAVE_INIT:
2394                 /* here, p is the slave id */
2395                 do_slave_init(ibdev, p, 1);
2396                 return;
2397
2398         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2399                 /* here, p is the slave id */
2400                 do_slave_init(ibdev, p, 0);
2401                 return;
2402
2403         default:
2404                 return;
2405         }
2406
2407         ibev.device           = ibdev_ptr;
2408         ibev.element.port_num = (u8) p;
2409
2410         ib_dispatch_event(&ibev);
2411 }
2412
2413 static struct mlx4_interface mlx4_ib_interface = {
2414         .add            = mlx4_ib_add,
2415         .remove         = mlx4_ib_remove,
2416         .event          = mlx4_ib_event,
2417         .protocol       = MLX4_PROT_IB_IPV6
2418 };
2419
2420 static int __init mlx4_ib_init(void)
2421 {
2422         int err;
2423
2424         wq = create_singlethread_workqueue("mlx4_ib");
2425         if (!wq)
2426                 return -ENOMEM;
2427
2428         err = mlx4_ib_mcg_init();
2429         if (err)
2430                 goto clean_wq;
2431
2432         err = mlx4_register_interface(&mlx4_ib_interface);
2433         if (err)
2434                 goto clean_mcg;
2435
2436         return 0;
2437
2438 clean_mcg:
2439         mlx4_ib_mcg_destroy();
2440
2441 clean_wq:
2442         destroy_workqueue(wq);
2443         return err;
2444 }
2445
2446 static void __exit mlx4_ib_cleanup(void)
2447 {
2448         mlx4_unregister_interface(&mlx4_ib_interface);
2449         mlx4_ib_mcg_destroy();
2450         destroy_workqueue(wq);
2451 }
2452
2453 module_init(mlx4_ib_init);
2454 module_exit(mlx4_ib_cleanup);