Pileus Git - ~andy/linux/blob - 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     "2.2-1"
  57 #define DRV_RELDATE     "Feb 2014"
  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 | IB_PORT_IP_BASED_GIDS;
 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         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1511         if (!work)
1512                 return -ENOMEM;
1513
1514         memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1515         memset(work->gids, 0, sizeof(work->gids));
1516         INIT_WORK(&work->work, reset_gids_task);
1517         work->dev = dev;
1518         work->port = port;
1519         queue_work(wq, &work->work);
1520         return 0;
1521 }
1522
1523 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1524                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1525 {
1526         struct mlx4_ib_iboe *iboe;
1527         int port = 0;
1528         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1529                                 rdma_vlan_dev_real_dev(event_netdev) :
1530                                 event_netdev;
1531         union ib_gid default_gid;
1532
1533         mlx4_make_default_gid(real_dev, &default_gid);
1534
1535         if (!memcmp(gid, &default_gid, sizeof(*gid)))
1536                 return 0;
1537
1538         if (event != NETDEV_DOWN && event != NETDEV_UP)
1539                 return 0;
1540
1541         if ((real_dev != event_netdev) &&
1542             (event == NETDEV_DOWN) &&
1543             rdma_link_local_addr((struct in6_addr *)gid))
1544                 return 0;
1545
1546         iboe = &ibdev->iboe;
1547         spin_lock(&iboe->lock);
1548
1549         for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1550                 if ((netif_is_bond_master(real_dev) &&
1551                      (real_dev == iboe->masters[port - 1])) ||
1552                      (!netif_is_bond_master(real_dev) &&
1553                      (real_dev == iboe->netdevs[port - 1])))
1554                         update_gid_table(ibdev, port, gid,
1555                                          event == NETDEV_DOWN, 0);
1556
1557         spin_unlock(&iboe->lock);
1558         return 0;
1559
1560 }
1561
1562 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1563                                struct mlx4_ib_dev *ibdev)
1564 {
1565         u8 port = 0;
1566         struct mlx4_ib_iboe *iboe;
1567         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1568                                 rdma_vlan_dev_real_dev(dev) : dev;
1569
1570         iboe = &ibdev->iboe;
1571
1572         for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1573                 if ((netif_is_bond_master(real_dev) &&
1574                      (real_dev == iboe->masters[port - 1])) ||
1575                      (!netif_is_bond_master(real_dev) &&
1576                      (real_dev == iboe->netdevs[port - 1])))
1577                         break;
1578
1579         if ((port == 0) || (port > MLX4_MAX_PORTS))
1580                 return 0;
1581         else
1582                 return port;
1583 }
1584
1585 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1586                                 void *ptr)
1587 {
1588         struct mlx4_ib_dev *ibdev;
1589         struct in_ifaddr *ifa = ptr;
1590         union ib_gid gid;
1591         struct net_device *event_netdev = ifa->ifa_dev->dev;
1592
1593         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1594
1595         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1596
1597         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1598         return NOTIFY_DONE;
1599 }
1600
1601 #if IS_ENABLED(CONFIG_IPV6)
1602 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1603                                 void *ptr)
1604 {
1605         struct mlx4_ib_dev *ibdev;
1606         struct inet6_ifaddr *ifa = ptr;
1607         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1608         struct net_device *event_netdev = ifa->idev->dev;
1609
1610         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1611
1612         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1613         return NOTIFY_DONE;
1614 }
1615 #endif
1616
1617 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1618                                  struct mlx4_ib_dev *ibdev, u8 port)
1619 {
1620         struct in_device *in_dev;
1621 #if IS_ENABLED(CONFIG_IPV6)
1622         struct inet6_dev *in6_dev;
1623         union ib_gid  *pgid;
1624         struct inet6_ifaddr *ifp;
1625 #endif
1626         union ib_gid gid;
1627
1628
1629         if ((port == 0) || (port > MLX4_MAX_PORTS))
1630                 return;
1631
1632         /* IPv4 gids */
1633         in_dev = in_dev_get(dev);
1634         if (in_dev) {
1635                 for_ifa(in_dev) {
1636                         /*ifa->ifa_address;*/
1637                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1638                                                (struct in6_addr *)&gid);
1639                         update_gid_table(ibdev, port, &gid, 0, 0);
1640                 }
1641                 endfor_ifa(in_dev);
1642                 in_dev_put(in_dev);
1643         }
1644 #if IS_ENABLED(CONFIG_IPV6)
1645         /* IPv6 gids */
1646         in6_dev = in6_dev_get(dev);
1647         if (in6_dev) {
1648                 read_lock_bh(&in6_dev->lock);
1649                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1650                         pgid = (union ib_gid *)&ifp->addr;
1651                         update_gid_table(ibdev, port, pgid, 0, 0);
1652                 }
1653                 read_unlock_bh(&in6_dev->lock);
1654                 in6_dev_put(in6_dev);
1655         }
1656 #endif
1657 }
1658
1659 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1660                                  struct  net_device *dev, u8 port)
1661 {
1662         union ib_gid gid;
1663         mlx4_make_default_gid(dev, &gid);
1664         update_gid_table(ibdev, port, &gid, 0, 1);
1665 }
1666
1667 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1668 {
1669         struct  net_device *dev;
1670         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1671         int i;
1672
1673         for (i = 1; i <= ibdev->num_ports; ++i)
1674                 if (reset_gid_table(ibdev, i))
1675                         return -1;
1676
1677         read_lock(&dev_base_lock);
1678         spin_lock(&iboe->lock);
1679
1680         for_each_netdev(&init_net, dev) {
1681                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1682                 if (port)
1683                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1684         }
1685
1686         spin_unlock(&iboe->lock);
1687         read_unlock(&dev_base_lock);
1688
1689         return 0;
1690 }
1691
1692 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev)
1693 {
1694         struct mlx4_ib_iboe *iboe;
1695         int port;
1696
1697         iboe = &ibdev->iboe;
1698
1699         spin_lock(&iboe->lock);
1700         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1701                 enum ib_port_state      port_state = IB_PORT_NOP;
1702                 struct net_device *old_master = iboe->masters[port - 1];
1703                 struct net_device *curr_netdev;
1704                 struct net_device *curr_master;
1705
1706                 iboe->netdevs[port - 1] =
1707                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1708                 if (iboe->netdevs[port - 1])
1709                         mlx4_ib_set_default_gid(ibdev,
1710                                                 iboe->netdevs[port - 1], port);
1711                 curr_netdev = iboe->netdevs[port - 1];
1712
1713                 if (iboe->netdevs[port - 1] &&
1714                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1715                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1716                                 iboe->netdevs[port - 1]);
1717                 } else {
1718                         iboe->masters[port - 1] = NULL;
1719                 }
1720                 curr_master = iboe->masters[port - 1];
1721
1722                 if (curr_netdev) {
1723                         port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1724                                                 IB_PORT_ACTIVE : IB_PORT_DOWN;
1725                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1726                 } else {
1727                         reset_gid_table(ibdev, port);
1728                 }
1729                 /* if using bonding/team and a slave port is down, we don't the bond IP
1730                  * based gids in the table since flows that select port by gid may get
1731                  * the down port.
1732                  */
1733                 if (curr_master && (port_state == IB_PORT_DOWN)) {
1734                         reset_gid_table(ibdev, port);
1735                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1736                 }
1737                 /* if bonding is used it is possible that we add it to masters
1738                  * only after IP address is assigned to the net bonding
1739                  * interface.
1740                 */
1741                 if (curr_master && (old_master != curr_master)) {
1742                         reset_gid_table(ibdev, port);
1743                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1744                         mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1745                 }
1746
1747                 if (!curr_master && (old_master != curr_master)) {
1748                         reset_gid_table(ibdev, port);
1749                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1750                         mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1751                 }
1752         }
1753
1754         spin_unlock(&iboe->lock);
1755 }
1756
1757 static int mlx4_ib_netdev_event(struct notifier_block *this,
1758                                 unsigned long event, void *ptr)
1759 {
1760         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1761         struct mlx4_ib_dev *ibdev;
1762
1763         if (!net_eq(dev_net(dev), &init_net))
1764                 return NOTIFY_DONE;
1765
1766         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1767         mlx4_ib_scan_netdevs(ibdev);
1768
1769         return NOTIFY_DONE;
1770 }
1771
1772 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1773 {
1774         int port;
1775         int slave;
1776         int i;
1777
1778         if (mlx4_is_master(ibdev->dev)) {
1779                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1780                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1781                                 for (i = 0;
1782                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1783                                      ++i) {
1784                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1785                                         /* master has the identity virt2phys pkey mapping */
1786                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1787                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1788                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1789                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1790                                 }
1791                         }
1792                 }
1793                 /* initialize pkey cache */
1794                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1795                         for (i = 0;
1796                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1797                              ++i)
1798                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1799                                         (i) ? 0 : 0xFFFF;
1800                 }
1801         }
1802 }
1803
1804 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1805 {
1806         char name[32];
1807         int eq_per_port = 0;
1808         int added_eqs = 0;
1809         int total_eqs = 0;
1810         int i, j, eq;
1811
1812         /* Legacy mode or comp_pool is not large enough */
1813         if (dev->caps.comp_pool == 0 ||
1814             dev->caps.num_ports > dev->caps.comp_pool)
1815                 return;
1816
1817         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1818                                         dev->caps.num_ports);
1819
1820         /* Init eq table */
1821         added_eqs = 0;
1822         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1823                 added_eqs += eq_per_port;
1824
1825         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1826
1827         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1828         if (!ibdev->eq_table)
1829                 return;
1830
1831         ibdev->eq_added = added_eqs;
1832
1833         eq = 0;
1834         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1835                 for (j = 0; j < eq_per_port; j++) {
1836                         sprintf(name, "mlx4-ib-%d-%d@%s",
1837                                 i, j, dev->pdev->bus->name);
1838                         /* Set IRQ for specific name (per ring) */
1839                         if (mlx4_assign_eq(dev, name, NULL,
1840                                            &ibdev->eq_table[eq])) {
1841                                 /* Use legacy (same as mlx4_en driver) */
1842                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1843                                 ibdev->eq_table[eq] =
1844                                         (eq % dev->caps.num_comp_vectors);
1845                         }
1846                         eq++;
1847                 }
1848         }
1849
1850         /* Fill the reset of the vector with legacy EQ */
1851         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1852                 ibdev->eq_table[eq++] = i;
1853
1854         /* Advertise the new number of EQs to clients */
1855         ibdev->ib_dev.num_comp_vectors = total_eqs;
1856 }
1857
1858 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1859 {
1860         int i;
1861
1862         /* no additional eqs were added */
1863         if (!ibdev->eq_table)
1864                 return;
1865
1866         /* Reset the advertised EQ number */
1867         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1868
1869         /* Free only the added eqs */
1870         for (i = 0; i < ibdev->eq_added; i++) {
1871                 /* Don't free legacy eqs if used */
1872                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1873                         continue;
1874                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1875         }
1876
1877         kfree(ibdev->eq_table);
1878 }
1879
1880 static void *mlx4_ib_add(struct mlx4_dev *dev)
1881 {
1882         struct mlx4_ib_dev *ibdev;
1883         int num_ports = 0;
1884         int i, j;
1885         int err;
1886         struct mlx4_ib_iboe *iboe;
1887         int ib_num_ports = 0;
1888
1889         pr_info_once("%s", mlx4_ib_version);
1890
1891         mlx4_foreach_non_ib_transport_port(i, dev)
1892                 num_ports++;
1893
1894         if (mlx4_is_mfunc(dev) && num_ports) {
1895                 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1896                 return NULL;
1897         }
1898
1899         num_ports = 0;
1900         mlx4_foreach_ib_transport_port(i, dev)
1901                 num_ports++;
1902
1903         /* No point in registering a device with no ports... */
1904         if (num_ports == 0)
1905                 return NULL;
1906
1907         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1908         if (!ibdev) {
1909                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1910                 return NULL;
1911         }
1912
1913         iboe = &ibdev->iboe;
1914
1915         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1916                 goto err_dealloc;
1917
1918         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1919                 goto err_pd;
1920
1921         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1922                                  PAGE_SIZE);
1923         if (!ibdev->uar_map)
1924                 goto err_uar;
1925         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1926
1927         ibdev->dev = dev;
1928
1929         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1930         ibdev->ib_dev.owner             = THIS_MODULE;
1931         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1932         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1933         ibdev->num_ports                = num_ports;
1934         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1935         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1936         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1937
1938         if (dev->caps.userspace_caps)
1939                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1940         else
1941                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1942
1943         ibdev->ib_dev.uverbs_cmd_mask   =
1944                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1945                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1946                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1947                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1948                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1949                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1950                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1951                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1952                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1953                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1954                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1955                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1956                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1957                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1958                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1959                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1960                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1961                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1962                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1963                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1964                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1965                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
1966                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1967
1968         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1969         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1970         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1971         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1972         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1973         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1974         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1975         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1976         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1977         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1978         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1979         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1980         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1981         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1982         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1983         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1984         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1985         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1986         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1987         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1988         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1989         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1990         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1991         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1992         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1993         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1994         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1995         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1996         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1997         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1998         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1999         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2000         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2001         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2002         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2003         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2004         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2005         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2006         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2007         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2008         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2009
2010         if (!mlx4_is_slave(ibdev->dev)) {
2011                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2012                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2013                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2014                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2015         }
2016
2017         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2018             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2019                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2020                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2021                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2022
2023                 ibdev->ib_dev.uverbs_cmd_mask |=
2024                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2025                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2026         }
2027
2028         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2029                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2030                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2031                 ibdev->ib_dev.uverbs_cmd_mask |=
2032                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2033                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2034         }
2035
2036         if (check_flow_steering_support(dev)) {
2037                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2038                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2039                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2040
2041                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2042                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2043                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2044         }
2045
2046         mlx4_ib_alloc_eqs(dev, ibdev);
2047
2048         spin_lock_init(&iboe->lock);
2049
2050         if (init_node_data(ibdev))
2051                 goto err_map;
2052
2053         for (i = 0; i < ibdev->num_ports; ++i) {
2054                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2055                                                 IB_LINK_LAYER_ETHERNET) {
2056                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2057                         if (err)
2058                                 ibdev->counters[i] = -1;
2059                 } else
2060                                 ibdev->counters[i] = -1;
2061         }
2062
2063         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2064                 ib_num_ports++;
2065
2066         spin_lock_init(&ibdev->sm_lock);
2067         mutex_init(&ibdev->cap_mask_mutex);
2068
2069         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2070             ib_num_ports) {
2071                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2072                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2073                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2074                                             &ibdev->steer_qpn_base);
2075                 if (err)
2076                         goto err_counter;
2077
2078                 ibdev->ib_uc_qpns_bitmap =
2079                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2080                                 sizeof(long),
2081                                 GFP_KERNEL);
2082                 if (!ibdev->ib_uc_qpns_bitmap) {
2083                         dev_err(&dev->pdev->dev, "bit map alloc failed\n");
2084                         goto err_steer_qp_release;
2085                 }
2086
2087                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2088
2089                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2090                                 dev, ibdev->steer_qpn_base,
2091                                 ibdev->steer_qpn_base +
2092                                 ibdev->steer_qpn_count - 1);
2093                 if (err)
2094                         goto err_steer_free_bitmap;
2095         }
2096
2097         if (ib_register_device(&ibdev->ib_dev, NULL))
2098                 goto err_steer_free_bitmap;
2099
2100         if (mlx4_ib_mad_init(ibdev))
2101                 goto err_reg;
2102
2103         if (mlx4_ib_init_sriov(ibdev))
2104                 goto err_mad;
2105
2106         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2107                 if (!iboe->nb.notifier_call) {
2108                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2109                         err = register_netdevice_notifier(&iboe->nb);
2110                         if (err) {
2111                                 iboe->nb.notifier_call = NULL;
2112                                 goto err_notif;
2113                         }
2114                 }
2115                 if (!iboe->nb_inet.notifier_call) {
2116                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2117                         err = register_inetaddr_notifier(&iboe->nb_inet);
2118                         if (err) {
2119                                 iboe->nb_inet.notifier_call = NULL;
2120                                 goto err_notif;
2121                         }
2122                 }
2123 #if IS_ENABLED(CONFIG_IPV6)
2124                 if (!iboe->nb_inet6.notifier_call) {
2125                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2126                         err = register_inet6addr_notifier(&iboe->nb_inet6);
2127                         if (err) {
2128                                 iboe->nb_inet6.notifier_call = NULL;
2129                                 goto err_notif;
2130                         }
2131                 }
2132 #endif
2133                 for (i = 1 ; i <= ibdev->num_ports ; ++i)
2134                         reset_gid_table(ibdev, i);
2135                 rtnl_lock();
2136                 mlx4_ib_scan_netdevs(ibdev);
2137                 rtnl_unlock();
2138                 mlx4_ib_init_gid_table(ibdev);
2139         }
2140
2141         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2142                 if (device_create_file(&ibdev->ib_dev.dev,
2143                                        mlx4_class_attributes[j]))
2144                         goto err_notif;
2145         }
2146
2147         ibdev->ib_active = true;
2148
2149         if (mlx4_is_mfunc(ibdev->dev))
2150                 init_pkeys(ibdev);
2151
2152         /* create paravirt contexts for any VFs which are active */
2153         if (mlx4_is_master(ibdev->dev)) {
2154                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2155                         if (j == mlx4_master_func_num(ibdev->dev))
2156                                 continue;
2157                         if (mlx4_is_slave_active(ibdev->dev, j))
2158                                 do_slave_init(ibdev, j, 1);
2159                 }
2160         }
2161         return ibdev;
2162
2163 err_notif:
2164         if (ibdev->iboe.nb.notifier_call) {
2165                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2166                         pr_warn("failure unregistering notifier\n");
2167                 ibdev->iboe.nb.notifier_call = NULL;
2168         }
2169         if (ibdev->iboe.nb_inet.notifier_call) {
2170                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2171                         pr_warn("failure unregistering notifier\n");
2172                 ibdev->iboe.nb_inet.notifier_call = NULL;
2173         }
2174 #if IS_ENABLED(CONFIG_IPV6)
2175         if (ibdev->iboe.nb_inet6.notifier_call) {
2176                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2177                         pr_warn("failure unregistering notifier\n");
2178                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2179         }
2180 #endif
2181         flush_workqueue(wq);
2182
2183         mlx4_ib_close_sriov(ibdev);
2184
2185 err_mad:
2186         mlx4_ib_mad_cleanup(ibdev);
2187
2188 err_reg:
2189         ib_unregister_device(&ibdev->ib_dev);
2190
2191 err_steer_free_bitmap:
2192         kfree(ibdev->ib_uc_qpns_bitmap);
2193
2194 err_steer_qp_release:
2195         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2196                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2197                                       ibdev->steer_qpn_count);
2198 err_counter:
2199         for (; i; --i)
2200                 if (ibdev->counters[i - 1] != -1)
2201                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2202
2203 err_map:
2204         iounmap(ibdev->uar_map);
2205
2206 err_uar:
2207         mlx4_uar_free(dev, &ibdev->priv_uar);
2208
2209 err_pd:
2210         mlx4_pd_free(dev, ibdev->priv_pdn);
2211
2212 err_dealloc:
2213         ib_dealloc_device(&ibdev->ib_dev);
2214
2215         return NULL;
2216 }
2217
2218 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2219 {
2220         int offset;
2221
2222         WARN_ON(!dev->ib_uc_qpns_bitmap);
2223
2224         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2225                                          dev->steer_qpn_count,
2226                                          get_count_order(count));
2227         if (offset < 0)
2228                 return offset;
2229
2230         *qpn = dev->steer_qpn_base + offset;
2231         return 0;
2232 }
2233
2234 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2235 {
2236         if (!qpn ||
2237             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2238                 return;
2239
2240         BUG_ON(qpn < dev->steer_qpn_base);
2241
2242         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2243                               qpn - dev->steer_qpn_base,
2244                               get_count_order(count));
2245 }
2246
2247 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2248                          int is_attach)
2249 {
2250         int err;
2251         size_t flow_size;
2252         struct ib_flow_attr *flow = NULL;
2253         struct ib_flow_spec_ib *ib_spec;
2254
2255         if (is_attach) {
2256                 flow_size = sizeof(struct ib_flow_attr) +
2257                             sizeof(struct ib_flow_spec_ib);
2258                 flow = kzalloc(flow_size, GFP_KERNEL);
2259                 if (!flow)
2260                         return -ENOMEM;
2261                 flow->port = mqp->port;
2262                 flow->num_of_specs = 1;
2263                 flow->size = flow_size;
2264                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2265                 ib_spec->type = IB_FLOW_SPEC_IB;
2266                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2267                 /* Add an empty rule for IB L2 */
2268                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2269
2270                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2271                                             IB_FLOW_DOMAIN_NIC,
2272                                             MLX4_FS_REGULAR,
2273                                             &mqp->reg_id);
2274         } else {
2275                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2276         }
2277         kfree(flow);
2278         return err;
2279 }
2280
2281 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2282 {
2283         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2284         int p;
2285
2286         mlx4_ib_close_sriov(ibdev);
2287         mlx4_ib_mad_cleanup(ibdev);
2288         ib_unregister_device(&ibdev->ib_dev);
2289         if (ibdev->iboe.nb.notifier_call) {
2290                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2291                         pr_warn("failure unregistering notifier\n");
2292                 ibdev->iboe.nb.notifier_call = NULL;
2293         }
2294
2295         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2296                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2297                                       ibdev->steer_qpn_count);
2298                 kfree(ibdev->ib_uc_qpns_bitmap);
2299         }
2300
2301         if (ibdev->iboe.nb_inet.notifier_call) {
2302                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2303                         pr_warn("failure unregistering notifier\n");
2304                 ibdev->iboe.nb_inet.notifier_call = NULL;
2305         }
2306 #if IS_ENABLED(CONFIG_IPV6)
2307         if (ibdev->iboe.nb_inet6.notifier_call) {
2308                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2309                         pr_warn("failure unregistering notifier\n");
2310                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2311         }
2312 #endif
2313
2314         iounmap(ibdev->uar_map);
2315         for (p = 0; p < ibdev->num_ports; ++p)
2316                 if (ibdev->counters[p] != -1)
2317                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2318         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2319                 mlx4_CLOSE_PORT(dev, p);
2320
2321         mlx4_ib_free_eqs(dev, ibdev);
2322
2323         mlx4_uar_free(dev, &ibdev->priv_uar);
2324         mlx4_pd_free(dev, ibdev->priv_pdn);
2325         ib_dealloc_device(&ibdev->ib_dev);
2326 }
2327
2328 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2329 {
2330         struct mlx4_ib_demux_work **dm = NULL;
2331         struct mlx4_dev *dev = ibdev->dev;
2332         int i;
2333         unsigned long flags;
2334
2335         if (!mlx4_is_master(dev))
2336                 return;
2337
2338         dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2339         if (!dm) {
2340                 pr_err("failed to allocate memory for tunneling qp update\n");
2341                 goto out;
2342         }
2343
2344         for (i = 0; i < dev->caps.num_ports; i++) {
2345                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2346                 if (!dm[i]) {
2347                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2348                         for (i = 0; i < dev->caps.num_ports; i++) {
2349                                 if (dm[i])
2350                                         kfree(dm[i]);
2351                         }
2352                         goto out;
2353                 }
2354         }
2355         /* initialize or tear down tunnel QPs for the slave */
2356         for (i = 0; i < dev->caps.num_ports; i++) {
2357                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2358                 dm[i]->port = i + 1;
2359                 dm[i]->slave = slave;
2360                 dm[i]->do_init = do_init;
2361                 dm[i]->dev = ibdev;
2362                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2363                 if (!ibdev->sriov.is_going_down)
2364                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2365                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2366         }
2367 out:
2368         kfree(dm);
2369         return;
2370 }
2371
2372 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2373                           enum mlx4_dev_event event, unsigned long param)
2374 {
2375         struct ib_event ibev;
2376         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2377         struct mlx4_eqe *eqe = NULL;
2378         struct ib_event_work *ew;
2379         int p = 0;
2380
2381         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2382                 eqe = (struct mlx4_eqe *)param;
2383         else
2384                 p = (int) param;
2385
2386         switch (event) {
2387         case MLX4_DEV_EVENT_PORT_UP:
2388                 if (p > ibdev->num_ports)
2389                         return;
2390                 if (mlx4_is_master(dev) &&
2391                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2392                         IB_LINK_LAYER_INFINIBAND) {
2393                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2394                 }
2395                 ibev.event = IB_EVENT_PORT_ACTIVE;
2396                 break;
2397
2398         case MLX4_DEV_EVENT_PORT_DOWN:
2399                 if (p > ibdev->num_ports)
2400                         return;
2401                 ibev.event = IB_EVENT_PORT_ERR;
2402                 break;
2403
2404         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2405                 ibdev->ib_active = false;
2406                 ibev.event = IB_EVENT_DEVICE_FATAL;
2407                 break;
2408
2409         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2410                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2411                 if (!ew) {
2412                         pr_err("failed to allocate memory for events work\n");
2413                         break;
2414                 }
2415
2416                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2417                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2418                 ew->ib_dev = ibdev;
2419                 /* need to queue only for port owner, which uses GEN_EQE */
2420                 if (mlx4_is_master(dev))
2421                         queue_work(wq, &ew->work);
2422                 else
2423                         handle_port_mgmt_change_event(&ew->work);
2424                 return;
2425
2426         case MLX4_DEV_EVENT_SLAVE_INIT:
2427                 /* here, p is the slave id */
2428                 do_slave_init(ibdev, p, 1);
2429                 return;
2430
2431         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2432                 /* here, p is the slave id */
2433                 do_slave_init(ibdev, p, 0);
2434                 return;
2435
2436         default:
2437                 return;
2438         }
2439
2440         ibev.device           = ibdev_ptr;
2441         ibev.element.port_num = (u8) p;
2442
2443         ib_dispatch_event(&ibev);
2444 }
2445
2446 static struct mlx4_interface mlx4_ib_interface = {
2447         .add            = mlx4_ib_add,
2448         .remove         = mlx4_ib_remove,
2449         .event          = mlx4_ib_event,
2450         .protocol       = MLX4_PROT_IB_IPV6
2451 };
2452
2453 static int __init mlx4_ib_init(void)
2454 {
2455         int err;
2456
2457         wq = create_singlethread_workqueue("mlx4_ib");
2458         if (!wq)
2459                 return -ENOMEM;
2460
2461         err = mlx4_ib_mcg_init();
2462         if (err)
2463                 goto clean_wq;
2464
2465         err = mlx4_register_interface(&mlx4_ib_interface);
2466         if (err)
2467                 goto clean_mcg;
2468
2469         return 0;
2470
2471 clean_mcg:
2472         mlx4_ib_mcg_destroy();
2473
2474 clean_wq:
2475         destroy_workqueue(wq);
2476         return err;
2477 }
2478
2479 static void __exit mlx4_ib_cleanup(void)
2480 {
2481         mlx4_unregister_interface(&mlx4_ib_interface);
2482         mlx4_ib_mcg_destroy();
2483         destroy_workqueue(wq);
2484 }
2485
2486 module_init(mlx4_ib_init);
2487 module_exit(mlx4_ib_cleanup);