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 <linux/sched/mm.h>
  43 #include <linux/sched/task.h>
  44
  45 #include <net/ipv6.h>
  46 #include <net/addrconf.h>
  47 #include <net/devlink.h>
  48
  49 #include <rdma/ib_smi.h>
  50 #include <rdma/ib_user_verbs.h>
  51 #include <rdma/ib_addr.h>
  52 #include <rdma/ib_cache.h>
  53
  54 #include <net/bonding.h>
  55
  56 #include <linux/mlx4/driver.h>
  57 #include <linux/mlx4/cmd.h>
  58 #include <linux/mlx4/qp.h>
  59
  60 #include "mlx4_ib.h"
  61 #include <rdma/mlx4-abi.h>
  62
  63 #define DRV_NAME        MLX4_IB_DRV_NAME
  64 #define DRV_VERSION     "4.0-0"
  65
  66 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  67 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  68 #define MLX4_IB_CARD_REV_A0   0xA0
  69
  70 MODULE_AUTHOR("Roland Dreier");
  71 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  72 MODULE_LICENSE("Dual BSD/GPL");
  73
  74 int mlx4_ib_sm_guid_assign = 0;
  75 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  76 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
  77
  78 static const char mlx4_ib_version[] =
  79         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  80         DRV_VERSION "\n";
  81
  82 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  83 static enum rdma_link_layer mlx4_ib_port_link_layer(struct ib_device *device,
  84                                                     u8 port_num);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static 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 num_ib_ports(struct mlx4_dev *dev)
 122 {
 123         int ib_ports = 0;
 124         int i;
 125
 126         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 127                 ib_ports++;
 128
 129         return ib_ports;
 130 }
 131
 132 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
 133 {
 134         struct mlx4_ib_dev *ibdev = to_mdev(device);
 135         struct net_device *dev;
 136
 137         rcu_read_lock();
 138         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
 139
 140         if (dev) {
 141                 if (mlx4_is_bonded(ibdev->dev)) {
 142                         struct net_device *upper = NULL;
 143
 144                         upper = netdev_master_upper_dev_get_rcu(dev);
 145                         if (upper) {
 146                                 struct net_device *active;
 147
 148                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
 149                                 if (active)
 150                                         dev = active;
 151                         }
 152                 }
 153         }
 154         if (dev)
 155                 dev_hold(dev);
 156
 157         rcu_read_unlock();
 158         return dev;
 159 }
 160
 161 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
 162                                   struct mlx4_ib_dev *ibdev,
 163                                   u8 port_num)
 164 {
 165         struct mlx4_cmd_mailbox *mailbox;
 166         int err;
 167         struct mlx4_dev *dev = ibdev->dev;
 168         int i;
 169         union ib_gid *gid_tbl;
 170
 171         mailbox = mlx4_alloc_cmd_mailbox(dev);
 172         if (IS_ERR(mailbox))
 173                 return -ENOMEM;
 174
 175         gid_tbl = mailbox->buf;
 176
 177         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 178                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
 179
 180         err = mlx4_cmd(dev, mailbox->dma,
 181                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
 182                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 183                        MLX4_CMD_WRAPPED);
 184         if (mlx4_is_bonded(dev))
 185                 err += mlx4_cmd(dev, mailbox->dma,
 186                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
 187                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 188                                 MLX4_CMD_WRAPPED);
 189
 190         mlx4_free_cmd_mailbox(dev, mailbox);
 191         return err;
 192 }
 193
 194 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
 195                                      struct mlx4_ib_dev *ibdev,
 196                                      u8 port_num)
 197 {
 198         struct mlx4_cmd_mailbox *mailbox;
 199         int err;
 200         struct mlx4_dev *dev = ibdev->dev;
 201         int i;
 202         struct {
 203                 union ib_gid    gid;
 204                 __be32          rsrvd1[2];
 205                 __be16          rsrvd2;
 206                 u8              type;
 207                 u8              version;
 208                 __be32          rsrvd3;
 209         } *gid_tbl;
 210
 211         mailbox = mlx4_alloc_cmd_mailbox(dev);
 212         if (IS_ERR(mailbox))
 213                 return -ENOMEM;
 214
 215         gid_tbl = mailbox->buf;
 216         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 217                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
 218                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 219                         gid_tbl[i].version = 2;
 220                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
 221                                 gid_tbl[i].type = 1;
 222                 }
 223         }
 224
 225         err = mlx4_cmd(dev, mailbox->dma,
 226                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
 227                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 228                        MLX4_CMD_WRAPPED);
 229         if (mlx4_is_bonded(dev))
 230                 err += mlx4_cmd(dev, mailbox->dma,
 231                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
 232                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 233                                 MLX4_CMD_WRAPPED);
 234
 235         mlx4_free_cmd_mailbox(dev, mailbox);
 236         return err;
 237 }
 238
 239 static int mlx4_ib_update_gids(struct gid_entry *gids,
 240                                struct mlx4_ib_dev *ibdev,
 241                                u8 port_num)
 242 {
 243         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
 244                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
 245
 246         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 247 }
 248
 249 static int mlx4_ib_add_gid(const struct ib_gid_attr *attr, void **context)
 250 {
 251         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 252         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 253         struct mlx4_port_gid_table   *port_gid_table;
 254         int free = -1, found = -1;
 255         int ret = 0;
 256         int hw_update = 0;
 257         int i;
 258         struct gid_entry *gids = NULL;
 259
 260         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 261                 return -EINVAL;
 262
 263         if (attr->port_num > MLX4_MAX_PORTS)
 264                 return -EINVAL;
 265
 266         if (!context)
 267                 return -EINVAL;
 268
 269         port_gid_table = &iboe->gids[attr->port_num - 1];
 270         spin_lock_bh(&iboe->lock);
 271         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 272                 if (!memcmp(&port_gid_table->gids[i].gid,
 273                             &attr->gid, sizeof(attr->gid)) &&
 274                     port_gid_table->gids[i].gid_type == attr->gid_type)  {
 275                         found = i;
 276                         break;
 277                 }
 278                 if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid))
 279                         free = i; /* HW has space */
 280         }
 281
 282         if (found < 0) {
 283                 if (free < 0) {
 284                         ret = -ENOSPC;
 285                 } else {
 286                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 287                         if (!port_gid_table->gids[free].ctx) {
 288                                 ret = -ENOMEM;
 289                         } else {
 290                                 *context = port_gid_table->gids[free].ctx;
 291                                 memcpy(&port_gid_table->gids[free].gid,
 292                                        &attr->gid, sizeof(attr->gid));
 293                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 294                                 port_gid_table->gids[free].ctx->real_index = free;
 295                                 port_gid_table->gids[free].ctx->refcount = 1;
 296                                 hw_update = 1;
 297                         }
 298                 }
 299         } else {
 300                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 301                 *context = ctx;
 302                 ctx->refcount++;
 303         }
 304         if (!ret && hw_update) {
 305                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 306                                      GFP_ATOMIC);
 307                 if (!gids) {
 308                         ret = -ENOMEM;
 309                 } else {
 310                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 311                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 312                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 313                         }
 314                 }
 315         }
 316         spin_unlock_bh(&iboe->lock);
 317
 318         if (!ret && hw_update) {
 319                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 320                 kfree(gids);
 321         }
 322
 323         return ret;
 324 }
 325
 326 static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context)
 327 {
 328         struct gid_cache_context *ctx = *context;
 329         struct mlx4_ib_dev *ibdev = to_mdev(attr->device);
 330         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 331         struct mlx4_port_gid_table   *port_gid_table;
 332         int ret = 0;
 333         int hw_update = 0;
 334         struct gid_entry *gids = NULL;
 335
 336         if (!rdma_cap_roce_gid_table(attr->device, attr->port_num))
 337                 return -EINVAL;
 338
 339         if (attr->port_num > MLX4_MAX_PORTS)
 340                 return -EINVAL;
 341
 342         port_gid_table = &iboe->gids[attr->port_num - 1];
 343         spin_lock_bh(&iboe->lock);
 344         if (ctx) {
 345                 ctx->refcount--;
 346                 if (!ctx->refcount) {
 347                         unsigned int real_index = ctx->real_index;
 348
 349                         memset(&port_gid_table->gids[real_index].gid, 0,
 350                                sizeof(port_gid_table->gids[real_index].gid));
 351                         kfree(port_gid_table->gids[real_index].ctx);
 352                         port_gid_table->gids[real_index].ctx = NULL;
 353                         hw_update = 1;
 354                 }
 355         }
 356         if (!ret && hw_update) {
 357                 int i;
 358
 359                 gids = kmalloc_array(MLX4_MAX_PORT_GIDS, sizeof(*gids),
 360                                      GFP_ATOMIC);
 361                 if (!gids) {
 362                         ret = -ENOMEM;
 363                 } else {
 364                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 365                                 memcpy(&gids[i].gid,
 366                                        &port_gid_table->gids[i].gid,
 367                                        sizeof(union ib_gid));
 368                                 gids[i].gid_type =
 369                                     port_gid_table->gids[i].gid_type;
 370                         }
 371                 }
 372         }
 373         spin_unlock_bh(&iboe->lock);
 374
 375         if (!ret && hw_update) {
 376                 ret = mlx4_ib_update_gids(gids, ibdev, attr->port_num);
 377                 kfree(gids);
 378         }
 379         return ret;
 380 }
 381
 382 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 383                                     const struct ib_gid_attr *attr)
 384 {
 385         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 386         struct gid_cache_context *ctx = NULL;
 387         struct mlx4_port_gid_table   *port_gid_table;
 388         int real_index = -EINVAL;
 389         int i;
 390         unsigned long flags;
 391         u8 port_num = attr->port_num;
 392
 393         if (port_num > MLX4_MAX_PORTS)
 394                 return -EINVAL;
 395
 396         if (mlx4_is_bonded(ibdev->dev))
 397                 port_num = 1;
 398
 399         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 400                 return attr->index;
 401
 402         spin_lock_irqsave(&iboe->lock, flags);
 403         port_gid_table = &iboe->gids[port_num - 1];
 404
 405         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 406                 if (!memcmp(&port_gid_table->gids[i].gid,
 407                             &attr->gid, sizeof(attr->gid)) &&
 408                     attr->gid_type == port_gid_table->gids[i].gid_type) {
 409                         ctx = port_gid_table->gids[i].ctx;
 410                         break;
 411                 }
 412         if (ctx)
 413                 real_index = ctx->real_index;
 414         spin_unlock_irqrestore(&iboe->lock, flags);
 415         return real_index;
 416 }
 417
 418 #define field_avail(type, fld, sz) (offsetof(type, fld) + \
 419                                     sizeof(((type *)0)->fld) <= (sz))
 420
 421 static int mlx4_ib_query_device(struct ib_device *ibdev,
 422                                 struct ib_device_attr *props,
 423                                 struct ib_udata *uhw)
 424 {
 425         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 426         struct ib_smp *in_mad  = NULL;
 427         struct ib_smp *out_mad = NULL;
 428         int err;
 429         int have_ib_ports;
 430         struct mlx4_uverbs_ex_query_device cmd;
 431         struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
 432         struct mlx4_clock_params clock_params;
 433
 434         if (uhw->inlen) {
 435                 if (uhw->inlen < sizeof(cmd))
 436                         return -EINVAL;
 437
 438                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 439                 if (err)
 440                         return err;
 441
 442                 if (cmd.comp_mask)
 443                         return -EINVAL;
 444
 445                 if (cmd.reserved)
 446                         return -EINVAL;
 447         }
 448
 449         resp.response_length = offsetof(typeof(resp), response_length) +
 450                 sizeof(resp.response_length);
 451         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 452         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 453         err = -ENOMEM;
 454         if (!in_mad || !out_mad)
 455                 goto out;
 456
 457         init_query_mad(in_mad);
 458         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 459
 460         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 461                            1, NULL, NULL, in_mad, out_mad);
 462         if (err)
 463                 goto out;
 464
 465         memset(props, 0, sizeof *props);
 466
 467         have_ib_ports = num_ib_ports(dev->dev);
 468
 469         props->fw_ver = dev->dev->caps.fw_ver;
 470         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 471                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 472                 IB_DEVICE_SYS_IMAGE_GUID                |
 473                 IB_DEVICE_RC_RNR_NAK_GEN                |
 474                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 475         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 476                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 477         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 478                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 479         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 480                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 481         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 482                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 483         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 484                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 485         if (dev->dev->caps.max_gso_sz &&
 486             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 487             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 488                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 489         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 490                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 491         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 492             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 493             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 494                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 495         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 496                 props->device_cap_flags |= IB_DEVICE_XRC;
 497         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 498                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 499         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 500                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 501                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 502                 else
 503                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 504         }
 505         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 506                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 507
 508         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 509
 510         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 511                 0xffffff;
 512         props->vendor_part_id      = dev->dev->persist->pdev->device;
 513         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 514         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 515
 516         props->max_mr_size         = ~0ull;
 517         props->page_size_cap       = dev->dev->caps.page_size_cap;
 518         props->max_qp              = dev->dev->quotas.qp;
 519         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 520         props->max_send_sge =
 521                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 522         props->max_recv_sge =
 523                 min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg);
 524         props->max_sge_rd = MLX4_MAX_SGE_RD;
 525         props->max_cq              = dev->dev->quotas.cq;
 526         props->max_cqe             = dev->dev->caps.max_cqes;
 527         props->max_mr              = dev->dev->quotas.mpt;
 528         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 529         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 530         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 531         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 532         props->max_srq             = dev->dev->quotas.srq;
 533         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 534         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 535         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 536         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 537         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 538                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 539         props->masked_atomic_cap   = props->atomic_cap;
 540         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 541         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 542         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 543         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 544                                            props->max_mcast_grp;
 545         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 546         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 547         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 548         props->max_ah = INT_MAX;
 549
 550         if (mlx4_ib_port_link_layer(ibdev, 1) == IB_LINK_LAYER_ETHERNET ||
 551             mlx4_ib_port_link_layer(ibdev, 2) == IB_LINK_LAYER_ETHERNET) {
 552                 if (dev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) {
 553                         props->rss_caps.max_rwq_indirection_tables =
 554                                 props->max_qp;
 555                         props->rss_caps.max_rwq_indirection_table_size =
 556                                 dev->dev->caps.max_rss_tbl_sz;
 557                         props->rss_caps.supported_qpts = 1 << IB_QPT_RAW_PACKET;
 558                         props->max_wq_type_rq = props->max_qp;
 559                 }
 560
 561                 if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP)
 562                         props->raw_packet_caps |= IB_RAW_PACKET_CAP_SCATTER_FCS;
 563         }
 564
 565         props->cq_caps.max_cq_moderation_count = MLX4_MAX_CQ_COUNT;
 566         props->cq_caps.max_cq_moderation_period = MLX4_MAX_CQ_PERIOD;
 567
 568         if (!mlx4_is_slave(dev->dev))
 569                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
 570
 571         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 572                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 573                 if (!err && !mlx4_is_slave(dev->dev)) {
 574                         resp.comp_mask |= MLX4_IB_QUERY_DEV_RESP_MASK_CORE_CLOCK_OFFSET;
 575                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 576                 }
 577         }
 578
 579         if (uhw->outlen >= resp.response_length +
 580             sizeof(resp.max_inl_recv_sz)) {
 581                 resp.response_length += sizeof(resp.max_inl_recv_sz);
 582                 resp.max_inl_recv_sz  = dev->dev->caps.max_rq_sg *
 583                         sizeof(struct mlx4_wqe_data_seg);
 584         }
 585
 586         if (field_avail(typeof(resp), rss_caps, uhw->outlen)) {
 587                 if (props->rss_caps.supported_qpts) {
 588                         resp.rss_caps.rx_hash_function =
 589                                 MLX4_IB_RX_HASH_FUNC_TOEPLITZ;
 590
 591                         resp.rss_caps.rx_hash_fields_mask =
 592                                 MLX4_IB_RX_HASH_SRC_IPV4 |
 593                                 MLX4_IB_RX_HASH_DST_IPV4 |
 594                                 MLX4_IB_RX_HASH_SRC_IPV6 |
 595                                 MLX4_IB_RX_HASH_DST_IPV6 |
 596                                 MLX4_IB_RX_HASH_SRC_PORT_TCP |
 597                                 MLX4_IB_RX_HASH_DST_PORT_TCP |
 598                                 MLX4_IB_RX_HASH_SRC_PORT_UDP |
 599                                 MLX4_IB_RX_HASH_DST_PORT_UDP;
 600
 601                         if (dev->dev->caps.tunnel_offload_mode ==
 602                             MLX4_TUNNEL_OFFLOAD_MODE_VXLAN)
 603                                 resp.rss_caps.rx_hash_fields_mask |=
 604                                         MLX4_IB_RX_HASH_INNER;
 605                 }
 606                 resp.response_length = offsetof(typeof(resp), rss_caps) +
 607                                        sizeof(resp.rss_caps);
 608         }
 609
 610         if (field_avail(typeof(resp), tso_caps, uhw->outlen)) {
 611                 if (dev->dev->caps.max_gso_sz &&
 612                     ((mlx4_ib_port_link_layer(ibdev, 1) ==
 613                     IB_LINK_LAYER_ETHERNET) ||
 614                     (mlx4_ib_port_link_layer(ibdev, 2) ==
 615                     IB_LINK_LAYER_ETHERNET))) {
 616                         resp.tso_caps.max_tso = dev->dev->caps.max_gso_sz;
 617                         resp.tso_caps.supported_qpts |=
 618                                 1 << IB_QPT_RAW_PACKET;
 619                 }
 620                 resp.response_length = offsetof(typeof(resp), tso_caps) +
 621                                        sizeof(resp.tso_caps);
 622         }
 623
 624         if (uhw->outlen) {
 625                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 626                 if (err)
 627                         goto out;
 628         }
 629 out:
 630         kfree(in_mad);
 631         kfree(out_mad);
 632
 633         return err;
 634 }
 635
 636 static enum rdma_link_layer
 637 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 638 {
 639         struct mlx4_dev *dev = to_mdev(device)->dev;
 640
 641         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 642                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 643 }
 644
 645 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 646                               struct ib_port_attr *props, int netw_view)
 647 {
 648         struct ib_smp *in_mad  = NULL;
 649         struct ib_smp *out_mad = NULL;
 650         int ext_active_speed;
 651         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 652         int err = -ENOMEM;
 653
 654         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 655         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 656         if (!in_mad || !out_mad)
 657                 goto out;
 658
 659         init_query_mad(in_mad);
 660         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 661         in_mad->attr_mod = cpu_to_be32(port);
 662
 663         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 664                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 665
 666         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 667                                 in_mad, out_mad);
 668         if (err)
 669                 goto out;
 670
 671
 672         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 673         props->lmc              = out_mad->data[34] & 0x7;
 674         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 675         props->sm_sl            = out_mad->data[36] & 0xf;
 676         props->state            = out_mad->data[32] & 0xf;
 677         props->phys_state       = out_mad->data[33] >> 4;
 678         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 679         if (netw_view)
 680                 props->gid_tbl_len = out_mad->data[50];
 681         else
 682                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 683         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 684         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 685         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 686         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 687         props->active_width     = out_mad->data[31] & 0xf;
 688         props->active_speed     = out_mad->data[35] >> 4;
 689         props->max_mtu          = out_mad->data[41] & 0xf;
 690         props->active_mtu       = out_mad->data[36] >> 4;
 691         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 692         props->max_vl_num       = out_mad->data[37] >> 4;
 693         props->init_type_reply  = out_mad->data[41] >> 4;
 694
 695         /* Check if extended speeds (EDR/FDR/...) are supported */
 696         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 697                 ext_active_speed = out_mad->data[62] >> 4;
 698
 699                 switch (ext_active_speed) {
 700                 case 1:
 701                         props->active_speed = IB_SPEED_FDR;
 702                         break;
 703                 case 2:
 704                         props->active_speed = IB_SPEED_EDR;
 705                         break;
 706                 }
 707         }
 708
 709         /* If reported active speed is QDR, check if is FDR-10 */
 710         if (props->active_speed == IB_SPEED_QDR) {
 711                 init_query_mad(in_mad);
 712                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 713                 in_mad->attr_mod = cpu_to_be32(port);
 714
 715                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 716                                    NULL, NULL, in_mad, out_mad);
 717                 if (err)
 718                         goto out;
 719
 720                 /* Checking LinkSpeedActive for FDR-10 */
 721                 if (out_mad->data[15] & 0x1)
 722                         props->active_speed = IB_SPEED_FDR10;
 723         }
 724
 725         /* Avoid wrong speed value returned by FW if the IB link is down. */
 726         if (props->state == IB_PORT_DOWN)
 727                  props->active_speed = IB_SPEED_SDR;
 728
 729 out:
 730         kfree(in_mad);
 731         kfree(out_mad);
 732         return err;
 733 }
 734
 735 static u8 state_to_phys_state(enum ib_port_state state)
 736 {
 737         return state == IB_PORT_ACTIVE ? 5 : 3;
 738 }
 739
 740 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 741                                struct ib_port_attr *props)
 742 {
 743
 744         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 745         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 746         struct net_device *ndev;
 747         enum ib_mtu tmp;
 748         struct mlx4_cmd_mailbox *mailbox;
 749         int err = 0;
 750         int is_bonded = mlx4_is_bonded(mdev->dev);
 751
 752         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 753         if (IS_ERR(mailbox))
 754                 return PTR_ERR(mailbox);
 755
 756         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 757                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 758                            MLX4_CMD_WRAPPED);
 759         if (err)
 760                 goto out;
 761
 762         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ||
 763                                    (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 764                                            IB_WIDTH_4X : IB_WIDTH_1X;
 765         props->active_speed     =  (((u8 *)mailbox->buf)[5] == 0x20 /*56Gb*/) ?
 766                                            IB_SPEED_FDR : IB_SPEED_QDR;
 767         props->port_cap_flags   = IB_PORT_CM_SUP;
 768         props->ip_gids = true;
 769         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 770         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 771         props->pkey_tbl_len     = 1;
 772         props->max_mtu          = IB_MTU_4096;
 773         props->max_vl_num       = 2;
 774         props->state            = IB_PORT_DOWN;
 775         props->phys_state       = state_to_phys_state(props->state);
 776         props->active_mtu       = IB_MTU_256;
 777         spin_lock_bh(&iboe->lock);
 778         ndev = iboe->netdevs[port - 1];
 779         if (ndev && is_bonded) {
 780                 rcu_read_lock(); /* required to get upper dev */
 781                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 782                 rcu_read_unlock();
 783         }
 784         if (!ndev)
 785                 goto out_unlock;
 786
 787         tmp = iboe_get_mtu(ndev->mtu);
 788         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 789
 790         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 791                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 792         props->phys_state       = state_to_phys_state(props->state);
 793 out_unlock:
 794         spin_unlock_bh(&iboe->lock);
 795 out:
 796         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 797         return err;
 798 }
 799
 800 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 801                          struct ib_port_attr *props, int netw_view)
 802 {
 803         int err;
 804
 805         /* props being zeroed by the caller, avoid zeroing it here */
 806
 807         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 808                 ib_link_query_port(ibdev, port, props, netw_view) :
 809                                 eth_link_query_port(ibdev, port, props);
 810
 811         return err;
 812 }
 813
 814 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 815                               struct ib_port_attr *props)
 816 {
 817         /* returns host view */
 818         return __mlx4_ib_query_port(ibdev, port, props, 0);
 819 }
 820
 821 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 822                         union ib_gid *gid, int netw_view)
 823 {
 824         struct ib_smp *in_mad  = NULL;
 825         struct ib_smp *out_mad = NULL;
 826         int err = -ENOMEM;
 827         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 828         int clear = 0;
 829         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 830
 831         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 832         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 833         if (!in_mad || !out_mad)
 834                 goto out;
 835
 836         init_query_mad(in_mad);
 837         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 838         in_mad->attr_mod = cpu_to_be32(port);
 839
 840         if (mlx4_is_mfunc(dev->dev) && netw_view)
 841                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 842
 843         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 844         if (err)
 845                 goto out;
 846
 847         memcpy(gid->raw, out_mad->data + 8, 8);
 848
 849         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 850                 if (index) {
 851                         /* For any index > 0, return the null guid */
 852                         err = 0;
 853                         clear = 1;
 854                         goto out;
 855                 }
 856         }
 857
 858         init_query_mad(in_mad);
 859         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 860         in_mad->attr_mod = cpu_to_be32(index / 8);
 861
 862         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 863                            NULL, NULL, in_mad, out_mad);
 864         if (err)
 865                 goto out;
 866
 867         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 868
 869 out:
 870         if (clear)
 871                 memset(gid->raw + 8, 0, 8);
 872         kfree(in_mad);
 873         kfree(out_mad);
 874         return err;
 875 }
 876
 877 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 878                              union ib_gid *gid)
 879 {
 880         if (rdma_protocol_ib(ibdev, port))
 881                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 882         return 0;
 883 }
 884
 885 static int mlx4_ib_query_sl2vl(struct ib_device *ibdev, u8 port, u64 *sl2vl_tbl)
 886 {
 887         union sl2vl_tbl_to_u64 sl2vl64;
 888         struct ib_smp *in_mad  = NULL;
 889         struct ib_smp *out_mad = NULL;
 890         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 891         int err = -ENOMEM;
 892         int jj;
 893
 894         if (mlx4_is_slave(to_mdev(ibdev)->dev)) {
 895                 *sl2vl_tbl = 0;
 896                 return 0;
 897         }
 898
 899         in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
 900         out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
 901         if (!in_mad || !out_mad)
 902                 goto out;
 903
 904         init_query_mad(in_mad);
 905         in_mad->attr_id  = IB_SMP_ATTR_SL_TO_VL_TABLE;
 906         in_mad->attr_mod = 0;
 907
 908         if (mlx4_is_mfunc(to_mdev(ibdev)->dev))
 909                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 910
 911         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 912                            in_mad, out_mad);
 913         if (err)
 914                 goto out;
 915
 916         for (jj = 0; jj < 8; jj++)
 917                 sl2vl64.sl8[jj] = ((struct ib_smp *)out_mad)->data[jj];
 918         *sl2vl_tbl = sl2vl64.sl64;
 919
 920 out:
 921         kfree(in_mad);
 922         kfree(out_mad);
 923         return err;
 924 }
 925
 926 static void mlx4_init_sl2vl_tbl(struct mlx4_ib_dev *mdev)
 927 {
 928         u64 sl2vl;
 929         int i;
 930         int err;
 931
 932         for (i = 1; i <= mdev->dev->caps.num_ports; i++) {
 933                 if (mdev->dev->caps.port_type[i] == MLX4_PORT_TYPE_ETH)
 934                         continue;
 935                 err = mlx4_ib_query_sl2vl(&mdev->ib_dev, i, &sl2vl);
 936                 if (err) {
 937                         pr_err("Unable to get default sl to vl mapping for port %d.  Using all zeroes (%d)\n",
 938                                i, err);
 939                         sl2vl = 0;
 940                 }
 941                 atomic64_set(&mdev->sl2vl[i - 1], sl2vl);
 942         }
 943 }
 944
 945 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 946                          u16 *pkey, int netw_view)
 947 {
 948         struct ib_smp *in_mad  = NULL;
 949         struct ib_smp *out_mad = NULL;
 950         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 951         int err = -ENOMEM;
 952
 953         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 954         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 955         if (!in_mad || !out_mad)
 956                 goto out;
 957
 958         init_query_mad(in_mad);
 959         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 960         in_mad->attr_mod = cpu_to_be32(index / 32);
 961
 962         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 963                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 964
 965         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 966                            in_mad, out_mad);
 967         if (err)
 968                 goto out;
 969
 970         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 971
 972 out:
 973         kfree(in_mad);
 974         kfree(out_mad);
 975         return err;
 976 }
 977
 978 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 979 {
 980         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 981 }
 982
 983 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 984                                  struct ib_device_modify *props)
 985 {
 986         struct mlx4_cmd_mailbox *mailbox;
 987         unsigned long flags;
 988
 989         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 990                 return -EOPNOTSUPP;
 991
 992         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 993                 return 0;
 994
 995         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 996                 return -EOPNOTSUPP;
 997
 998         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 999         memcpy(ibdev->node_desc, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1000         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
1001
1002         /*
1003          * If possible, pass node desc to FW, so it can generate
1004          * a 144 trap.  If cmd fails, just ignore.
1005          */
1006         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
1007         if (IS_ERR(mailbox))
1008                 return 0;
1009
1010         memcpy(mailbox->buf, props->node_desc, IB_DEVICE_NODE_DESC_MAX);
1011         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
1012                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
1013
1014         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
1015
1016         return 0;
1017 }
1018
1019 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
1020                             u32 cap_mask)
1021 {
1022         struct mlx4_cmd_mailbox *mailbox;
1023         int err;
1024
1025         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
1026         if (IS_ERR(mailbox))
1027                 return PTR_ERR(mailbox);
1028
1029         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
1030                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
1031                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
1032         } else {
1033                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
1034                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
1035         }
1036
1037         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
1038                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1039                        MLX4_CMD_WRAPPED);
1040
1041         mlx4_free_cmd_mailbox(dev->dev, mailbox);
1042         return err;
1043 }
1044
1045 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
1046                                struct ib_port_modify *props)
1047 {
1048         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
1049         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
1050         struct ib_port_attr attr;
1051         u32 cap_mask;
1052         int err;
1053
1054         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
1055          * of whether port link layer is ETH or IB. For ETH ports, qkey
1056          * violations and port capabilities are not meaningful.
1057          */
1058         if (is_eth)
1059                 return 0;
1060
1061         mutex_lock(&mdev->cap_mask_mutex);
1062
1063         err = ib_query_port(ibdev, port, &attr);
1064         if (err)
1065                 goto out;
1066
1067         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
1068                 ~props->clr_port_cap_mask;
1069
1070         err = mlx4_ib_SET_PORT(mdev, port,
1071                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
1072                                cap_mask);
1073
1074 out:
1075         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
1076         return err;
1077 }
1078
1079 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
1080                                                   struct ib_udata *udata)
1081 {
1082         struct mlx4_ib_dev *dev = to_mdev(ibdev);
1083         struct mlx4_ib_ucontext *context;
1084         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
1085         struct mlx4_ib_alloc_ucontext_resp resp;
1086         int err;
1087
1088         if (!dev->ib_active)
1089                 return ERR_PTR(-EAGAIN);
1090
1091         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
1092                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
1093                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
1094                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1095         } else {
1096                 resp.dev_caps         = dev->dev->caps.userspace_caps;
1097                 resp.qp_tab_size      = dev->dev->caps.num_qps;
1098                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
1099                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
1100                 resp.cqe_size         = dev->dev->caps.cqe_size;
1101         }
1102
1103         context = kzalloc(sizeof(*context), GFP_KERNEL);
1104         if (!context)
1105                 return ERR_PTR(-ENOMEM);
1106
1107         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
1108         if (err) {
1109                 kfree(context);
1110                 return ERR_PTR(err);
1111         }
1112
1113         INIT_LIST_HEAD(&context->db_page_list);
1114         mutex_init(&context->db_page_mutex);
1115
1116         INIT_LIST_HEAD(&context->wqn_ranges_list);
1117         mutex_init(&context->wqn_ranges_mutex);
1118
1119         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1120                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1121         else
1122                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1123
1124         if (err) {
1125                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1126                 kfree(context);
1127                 return ERR_PTR(-EFAULT);
1128         }
1129
1130         return &context->ibucontext;
1131 }
1132
1133 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1134 {
1135         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1136
1137         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1138         kfree(context);
1139
1140         return 0;
1141 }
1142
1143 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1144 {
1145         /* vma_open is called when a new VMA is created on top of our VMA.
1146          * This is done through either mremap flow or split_vma (usually due
1147          * to mlock, madvise, munmap, etc.). We do not support a clone of the
1148          * vma, as this VMA is strongly hardware related. Therefore we set the
1149          * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1150          * calling us again and trying to do incorrect actions. We assume that
1151          * the original vma size is exactly a single page that there will be no
1152          * "splitting" operations on.
1153          */
1154         area->vm_ops = NULL;
1155 }
1156
1157 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1158 {
1159         struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1160
1161         /* It's guaranteed that all VMAs opened on a FD are closed before the
1162          * file itself is closed, therefore no sync is needed with the regular
1163          * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1164          * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1165          * The close operation is usually called under mm->mmap_sem except when
1166          * process is exiting.  The exiting case is handled explicitly as part
1167          * of mlx4_ib_disassociate_ucontext.
1168          */
1169         mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1170                                 area->vm_private_data;
1171
1172         /* set the vma context pointer to null in the mlx4_ib driver's private
1173          * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1174          */
1175         mlx4_ib_vma_priv_data->vma = NULL;
1176 }
1177
1178 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1179         .open = mlx4_ib_vma_open,
1180         .close = mlx4_ib_vma_close
1181 };
1182
1183 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1184 {
1185         int i;
1186         struct vm_area_struct *vma;
1187         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1188
1189         /* need to protect from a race on closing the vma as part of
1190          * mlx4_ib_vma_close().
1191          */
1192         for (i = 0; i < HW_BAR_COUNT; i++) {
1193                 vma = context->hw_bar_info[i].vma;
1194                 if (!vma)
1195                         continue;
1196
1197                 zap_vma_ptes(context->hw_bar_info[i].vma,
1198                              context->hw_bar_info[i].vma->vm_start, PAGE_SIZE);
1199
1200                 context->hw_bar_info[i].vma->vm_flags &=
1201                         ~(VM_SHARED | VM_MAYSHARE);
1202                 /* context going to be destroyed, should not access ops any more */
1203                 context->hw_bar_info[i].vma->vm_ops = NULL;
1204         }
1205 }
1206
1207 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1208                                  struct mlx4_ib_vma_private_data *vma_private_data)
1209 {
1210         vma_private_data->vma = vma;
1211         vma->vm_private_data = vma_private_data;
1212         vma->vm_ops =  &mlx4_ib_vm_ops;
1213 }
1214
1215 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1216 {
1217         struct mlx4_ib_dev *dev = to_mdev(context->device);
1218         struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1219
1220         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1221                 return -EINVAL;
1222
1223         if (vma->vm_pgoff == 0) {
1224                 /* We prevent double mmaping on same context */
1225                 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1226                         return -EINVAL;
1227
1228                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1229
1230                 if (io_remap_pfn_range(vma, vma->vm_start,
1231                                        to_mucontext(context)->uar.pfn,
1232                                        PAGE_SIZE, vma->vm_page_prot))
1233                         return -EAGAIN;
1234
1235                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1236
1237         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1238                 /* We prevent double mmaping on same context */
1239                 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1240                         return -EINVAL;
1241
1242                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1243
1244                 if (io_remap_pfn_range(vma, vma->vm_start,
1245                                        to_mucontext(context)->uar.pfn +
1246                                        dev->dev->caps.num_uars,
1247                                        PAGE_SIZE, vma->vm_page_prot))
1248                         return -EAGAIN;
1249
1250                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1251
1252         } else if (vma->vm_pgoff == 3) {
1253                 struct mlx4_clock_params params;
1254                 int ret;
1255
1256                 /* We prevent double mmaping on same context */
1257                 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1258                         return -EINVAL;
1259
1260                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1261
1262                 if (ret)
1263                         return ret;
1264
1265                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1266                 if (io_remap_pfn_range(vma, vma->vm_start,
1267                                        (pci_resource_start(dev->dev->persist->pdev,
1268                                                            params.bar) +
1269                                         params.offset)
1270                                        >> PAGE_SHIFT,
1271                                        PAGE_SIZE, vma->vm_page_prot))
1272                         return -EAGAIN;
1273
1274                 mlx4_ib_set_vma_data(vma,
1275                                      &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1276         } else {
1277                 return -EINVAL;
1278         }
1279
1280         return 0;
1281 }
1282
1283 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1284                                       struct ib_ucontext *context,
1285                                       struct ib_udata *udata)
1286 {
1287         struct mlx4_ib_pd *pd;
1288         int err;
1289
1290         pd = kzalloc(sizeof(*pd), GFP_KERNEL);
1291         if (!pd)
1292                 return ERR_PTR(-ENOMEM);
1293
1294         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1295         if (err) {
1296                 kfree(pd);
1297                 return ERR_PTR(err);
1298         }
1299
1300         if (context)
1301                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1302                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1303                         kfree(pd);
1304                         return ERR_PTR(-EFAULT);
1305                 }
1306         return &pd->ibpd;
1307 }
1308
1309 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1310 {
1311         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1312         kfree(pd);
1313
1314         return 0;
1315 }
1316
1317 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1318                                           struct ib_ucontext *context,
1319                                           struct ib_udata *udata)
1320 {
1321         struct mlx4_ib_xrcd *xrcd;
1322         struct ib_cq_init_attr cq_attr = {};
1323         int err;
1324
1325         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1326                 return ERR_PTR(-ENOSYS);
1327
1328         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1329         if (!xrcd)
1330                 return ERR_PTR(-ENOMEM);
1331
1332         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1333         if (err)
1334                 goto err1;
1335
1336         xrcd->pd = ib_alloc_pd(ibdev, 0);
1337         if (IS_ERR(xrcd->pd)) {
1338                 err = PTR_ERR(xrcd->pd);
1339                 goto err2;
1340         }
1341
1342         cq_attr.cqe = 1;
1343         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1344         if (IS_ERR(xrcd->cq)) {
1345                 err = PTR_ERR(xrcd->cq);
1346                 goto err3;
1347         }
1348
1349         return &xrcd->ibxrcd;
1350
1351 err3:
1352         ib_dealloc_pd(xrcd->pd);
1353 err2:
1354         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1355 err1:
1356         kfree(xrcd);
1357         return ERR_PTR(err);
1358 }
1359
1360 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1361 {
1362         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1363         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1364         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1365         kfree(xrcd);
1366
1367         return 0;
1368 }
1369
1370 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1371 {
1372         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1373         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1374         struct mlx4_ib_gid_entry *ge;
1375
1376         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1377         if (!ge)
1378                 return -ENOMEM;
1379
1380         ge->gid = *gid;
1381         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1382                 ge->port = mqp->port;
1383                 ge->added = 1;
1384         }
1385
1386         mutex_lock(&mqp->mutex);
1387         list_add_tail(&ge->list, &mqp->gid_list);
1388         mutex_unlock(&mqp->mutex);
1389
1390         return 0;
1391 }
1392
1393 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1394                                           struct mlx4_ib_counters *ctr_table)
1395 {
1396         struct counter_index *counter, *tmp_count;
1397
1398         mutex_lock(&ctr_table->mutex);
1399         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1400                                  list) {
1401                 if (counter->allocated)
1402                         mlx4_counter_free(ibdev->dev, counter->index);
1403                 list_del(&counter->list);
1404                 kfree(counter);
1405         }
1406         mutex_unlock(&ctr_table->mutex);
1407 }
1408
1409 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1410                    union ib_gid *gid)
1411 {
1412         struct net_device *ndev;
1413         int ret = 0;
1414
1415         if (!mqp->port)
1416                 return 0;
1417
1418         spin_lock_bh(&mdev->iboe.lock);
1419         ndev = mdev->iboe.netdevs[mqp->port - 1];
1420         if (ndev)
1421                 dev_hold(ndev);
1422         spin_unlock_bh(&mdev->iboe.lock);
1423
1424         if (ndev) {
1425                 ret = 1;
1426                 dev_put(ndev);
1427         }
1428
1429         return ret;
1430 }
1431
1432 struct mlx4_ib_steering {
1433         struct list_head list;
1434         struct mlx4_flow_reg_id reg_id;
1435         union ib_gid gid;
1436 };
1437
1438 #define LAST_ETH_FIELD vlan_tag
1439 #define LAST_IB_FIELD sl
1440 #define LAST_IPV4_FIELD dst_ip
1441 #define LAST_TCP_UDP_FIELD src_port
1442
1443 /* Field is the last supported field */
1444 #define FIELDS_NOT_SUPPORTED(filter, field)\
1445         memchr_inv((void *)&filter.field  +\
1446                    sizeof(filter.field), 0,\
1447                    sizeof(filter) -\
1448                    offsetof(typeof(filter), field) -\
1449                    sizeof(filter.field))
1450
1451 static int parse_flow_attr(struct mlx4_dev *dev,
1452                            u32 qp_num,
1453                            union ib_flow_spec *ib_spec,
1454                            struct _rule_hw *mlx4_spec)
1455 {
1456         enum mlx4_net_trans_rule_id type;
1457
1458         switch (ib_spec->type) {
1459         case IB_FLOW_SPEC_ETH:
1460                 if (FIELDS_NOT_SUPPORTED(ib_spec->eth.mask, LAST_ETH_FIELD))
1461                         return -ENOTSUPP;
1462
1463                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1464                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1465                        ETH_ALEN);
1466                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1467                        ETH_ALEN);
1468                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1469                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1470                 break;
1471         case IB_FLOW_SPEC_IB:
1472                 if (FIELDS_NOT_SUPPORTED(ib_spec->ib.mask, LAST_IB_FIELD))
1473                         return -ENOTSUPP;
1474
1475                 type = MLX4_NET_TRANS_RULE_ID_IB;
1476                 mlx4_spec->ib.l3_qpn =
1477                         cpu_to_be32(qp_num);
1478                 mlx4_spec->ib.qpn_mask =
1479                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1480                 break;
1481
1482
1483         case IB_FLOW_SPEC_IPV4:
1484                 if (FIELDS_NOT_SUPPORTED(ib_spec->ipv4.mask, LAST_IPV4_FIELD))
1485                         return -ENOTSUPP;
1486
1487                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1488                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1489                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1490                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1491                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1492                 break;
1493
1494         case IB_FLOW_SPEC_TCP:
1495         case IB_FLOW_SPEC_UDP:
1496                 if (FIELDS_NOT_SUPPORTED(ib_spec->tcp_udp.mask, LAST_TCP_UDP_FIELD))
1497                         return -ENOTSUPP;
1498
1499                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1500                                         MLX4_NET_TRANS_RULE_ID_TCP :
1501                                         MLX4_NET_TRANS_RULE_ID_UDP;
1502                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1503                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1504                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1505                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1506                 break;
1507
1508         default:
1509                 return -EINVAL;
1510         }
1511         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1512             mlx4_hw_rule_sz(dev, type) < 0)
1513                 return -EINVAL;
1514         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1515         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1516         return mlx4_hw_rule_sz(dev, type);
1517 }
1518
1519 struct default_rules {
1520         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1521         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1522         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1523         __u8  link_layer;
1524 };
1525 static const struct default_rules default_table[] = {
1526         {
1527                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1528                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1529                 .rules_create_list = {IB_FLOW_SPEC_IB},
1530                 .link_layer = IB_LINK_LAYER_INFINIBAND
1531         }
1532 };
1533
1534 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1535                                          struct ib_flow_attr *flow_attr)
1536 {
1537         int i, j, k;
1538         void *ib_flow;
1539         const struct default_rules *pdefault_rules = default_table;
1540         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1541
1542         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1543                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1544                 memset(&field_types, 0, sizeof(field_types));
1545
1546                 if (link_layer != pdefault_rules->link_layer)
1547                         continue;
1548
1549                 ib_flow = flow_attr + 1;
1550                 /* we assume the specs are sorted */
1551                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1552                      j < flow_attr->num_of_specs; k++) {
1553                         union ib_flow_spec *current_flow =
1554                                 (union ib_flow_spec *)ib_flow;
1555
1556                         /* same layer but different type */
1557                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1558                              (pdefault_rules->mandatory_fields[k] &
1559                               IB_FLOW_SPEC_LAYER_MASK)) &&
1560                             (current_flow->type !=
1561                              pdefault_rules->mandatory_fields[k]))
1562                                 goto out;
1563
1564                         /* same layer, try match next one */
1565                         if (current_flow->type ==
1566                             pdefault_rules->mandatory_fields[k]) {
1567                                 j++;
1568                                 ib_flow +=
1569                                         ((union ib_flow_spec *)ib_flow)->size;
1570                         }
1571                 }
1572
1573                 ib_flow = flow_attr + 1;
1574                 for (j = 0; j < flow_attr->num_of_specs;
1575                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1576                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1577                                 /* same layer and same type */
1578                                 if (((union ib_flow_spec *)ib_flow)->type ==
1579                                     pdefault_rules->mandatory_not_fields[k])
1580                                         goto out;
1581
1582                 return i;
1583         }
1584 out:
1585         return -1;
1586 }
1587
1588 static int __mlx4_ib_create_default_rules(
1589                 struct mlx4_ib_dev *mdev,
1590                 struct ib_qp *qp,
1591                 const struct default_rules *pdefault_rules,
1592                 struct _rule_hw *mlx4_spec) {
1593         int size = 0;
1594         int i;
1595
1596         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1597                 int ret;
1598                 union ib_flow_spec ib_spec;
1599                 switch (pdefault_rules->rules_create_list[i]) {
1600                 case 0:
1601                         /* no rule */
1602                         continue;
1603                 case IB_FLOW_SPEC_IB:
1604                         ib_spec.type = IB_FLOW_SPEC_IB;
1605                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1606
1607                         break;
1608                 default:
1609                         /* invalid rule */
1610                         return -EINVAL;
1611                 }
1612                 /* We must put empty rule, qpn is being ignored */
1613                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1614                                       mlx4_spec);
1615                 if (ret < 0) {
1616                         pr_info("invalid parsing\n");
1617                         return -EINVAL;
1618                 }
1619
1620                 mlx4_spec = (void *)mlx4_spec + ret;
1621                 size += ret;
1622         }
1623         return size;
1624 }
1625
1626 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1627                           int domain,
1628                           enum mlx4_net_trans_promisc_mode flow_type,
1629                           u64 *reg_id)
1630 {
1631         int ret, i;
1632         int size = 0;
1633         void *ib_flow;
1634         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1635         struct mlx4_cmd_mailbox *mailbox;
1636         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1637         int default_flow;
1638
1639         static const u16 __mlx4_domain[] = {
1640                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1641                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1642                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1643                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1644         };
1645
1646         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1647                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1648                 return -EINVAL;
1649         }
1650
1651         if (domain >= IB_FLOW_DOMAIN_NUM) {
1652                 pr_err("Invalid domain value %d\n", domain);
1653                 return -EINVAL;
1654         }
1655
1656         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1657                 return -EINVAL;
1658
1659         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1660         if (IS_ERR(mailbox))
1661                 return PTR_ERR(mailbox);
1662         ctrl = mailbox->buf;
1663
1664         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1665                                  flow_attr->priority);
1666         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1667         ctrl->port = flow_attr->port;
1668         ctrl->qpn = cpu_to_be32(qp->qp_num);
1669
1670         ib_flow = flow_attr + 1;
1671         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1672         /* Add default flows */
1673         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1674         if (default_flow >= 0) {
1675                 ret = __mlx4_ib_create_default_rules(
1676                                 mdev, qp, default_table + default_flow,
1677                                 mailbox->buf + size);
1678                 if (ret < 0) {
1679                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1680                         return -EINVAL;
1681                 }
1682                 size += ret;
1683         }
1684         for (i = 0; i < flow_attr->num_of_specs; i++) {
1685                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1686                                       mailbox->buf + size);
1687                 if (ret < 0) {
1688                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1689                         return -EINVAL;
1690                 }
1691                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1692                 size += ret;
1693         }
1694
1695         if (mlx4_is_master(mdev->dev) && flow_type == MLX4_FS_REGULAR &&
1696             flow_attr->num_of_specs == 1) {
1697                 struct _rule_hw *rule_header = (struct _rule_hw *)(ctrl + 1);
1698                 enum ib_flow_spec_type header_spec =
1699                         ((union ib_flow_spec *)(flow_attr + 1))->type;
1700
1701                 if (header_spec == IB_FLOW_SPEC_ETH)
1702                         mlx4_handle_eth_header_mcast_prio(ctrl, rule_header);
1703         }
1704
1705         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1706                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1707                            MLX4_CMD_NATIVE);
1708         if (ret == -ENOMEM)
1709                 pr_err("mcg table is full. Fail to register network rule.\n");
1710         else if (ret == -ENXIO)
1711                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1712         else if (ret)
1713                 pr_err("Invalid argument. Fail to register network rule.\n");
1714
1715         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1716         return ret;
1717 }
1718
1719 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1720 {
1721         int err;
1722         err = mlx4_cmd(dev, reg_id, 0, 0,
1723                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1724                        MLX4_CMD_NATIVE);
1725         if (err)
1726                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1727                        reg_id);
1728         return err;
1729 }
1730
1731 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1732                                     u64 *reg_id)
1733 {
1734         void *ib_flow;
1735         union ib_flow_spec *ib_spec;
1736         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1737         int err = 0;
1738
1739         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1740             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1741                 return 0; /* do nothing */
1742
1743         ib_flow = flow_attr + 1;
1744         ib_spec = (union ib_flow_spec *)ib_flow;
1745
1746         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1747                 return 0; /* do nothing */
1748
1749         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1750                                     flow_attr->port, qp->qp_num,
1751                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1752                                     reg_id);
1753         return err;
1754 }
1755
1756 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1757                                       struct ib_flow_attr *flow_attr,
1758                                       enum mlx4_net_trans_promisc_mode *type)
1759 {
1760         int err = 0;
1761
1762         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1763             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1764             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1765                 return -EOPNOTSUPP;
1766         }
1767
1768         if (flow_attr->num_of_specs == 0) {
1769                 type[0] = MLX4_FS_MC_SNIFFER;
1770                 type[1] = MLX4_FS_UC_SNIFFER;
1771         } else {
1772                 union ib_flow_spec *ib_spec;
1773
1774                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1775                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1776                         return -EINVAL;
1777
1778                 /* if all is zero than MC and UC */
1779                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1780                         type[0] = MLX4_FS_MC_SNIFFER;
1781                         type[1] = MLX4_FS_UC_SNIFFER;
1782                 } else {
1783                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1784                                             ib_spec->eth.mask.dst_mac[1],
1785                                             ib_spec->eth.mask.dst_mac[2],
1786                                             ib_spec->eth.mask.dst_mac[3],
1787                                             ib_spec->eth.mask.dst_mac[4],
1788                                             ib_spec->eth.mask.dst_mac[5]};
1789
1790                         /* Above xor was only on MC bit, non empty mask is valid
1791                          * only if this bit is set and rest are zero.
1792                          */
1793                         if (!is_zero_ether_addr(&mac[0]))
1794                                 return -EINVAL;
1795
1796                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1797                                 type[0] = MLX4_FS_MC_SNIFFER;
1798                         else
1799                                 type[0] = MLX4_FS_UC_SNIFFER;
1800                 }
1801         }
1802
1803         return err;
1804 }
1805
1806 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1807                                     struct ib_flow_attr *flow_attr,
1808                                     int domain, struct ib_udata *udata)
1809 {
1810         int err = 0, i = 0, j = 0;
1811         struct mlx4_ib_flow *mflow;
1812         enum mlx4_net_trans_promisc_mode type[2];
1813         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1814         int is_bonded = mlx4_is_bonded(dev);
1815
1816         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1817                 return ERR_PTR(-EINVAL);
1818
1819         if (flow_attr->flags & ~IB_FLOW_ATTR_FLAGS_DONT_TRAP)
1820                 return ERR_PTR(-EOPNOTSUPP);
1821
1822         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1823             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1824                 return ERR_PTR(-EOPNOTSUPP);
1825
1826         if (udata &&
1827             udata->inlen && !ib_is_udata_cleared(udata, 0, udata->inlen))
1828                 return ERR_PTR(-EOPNOTSUPP);
1829
1830         memset(type, 0, sizeof(type));
1831
1832         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1833         if (!mflow) {
1834                 err = -ENOMEM;
1835                 goto err_free;
1836         }
1837
1838         switch (flow_attr->type) {
1839         case IB_FLOW_ATTR_NORMAL:
1840                 /* If dont trap flag (continue match) is set, under specific
1841                  * condition traffic be replicated to given qp,
1842                  * without stealing it
1843                  */
1844                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1845                         err = mlx4_ib_add_dont_trap_rule(dev,
1846                                                          flow_attr,
1847                                                          type);
1848                         if (err)
1849                                 goto err_free;
1850                 } else {
1851                         type[0] = MLX4_FS_REGULAR;
1852                 }
1853                 break;
1854
1855         case IB_FLOW_ATTR_ALL_DEFAULT:
1856                 type[0] = MLX4_FS_ALL_DEFAULT;
1857                 break;
1858
1859         case IB_FLOW_ATTR_MC_DEFAULT:
1860                 type[0] = MLX4_FS_MC_DEFAULT;
1861                 break;
1862
1863         case IB_FLOW_ATTR_SNIFFER:
1864                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1865                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1866                 break;
1867
1868         default:
1869                 err = -EINVAL;
1870                 goto err_free;
1871         }
1872
1873         while (i < ARRAY_SIZE(type) && type[i]) {
1874                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1875                                             &mflow->reg_id[i].id);
1876                 if (err)
1877                         goto err_create_flow;
1878                 if (is_bonded) {
1879                         /* Application always sees one port so the mirror rule
1880                          * must be on port #2
1881                          */
1882                         flow_attr->port = 2;
1883                         err = __mlx4_ib_create_flow(qp, flow_attr,
1884                                                     domain, type[j],
1885                                                     &mflow->reg_id[j].mirror);
1886                         flow_attr->port = 1;
1887                         if (err)
1888                                 goto err_create_flow;
1889                         j++;
1890                 }
1891
1892                 i++;
1893         }
1894
1895         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1896                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1897                                                &mflow->reg_id[i].id);
1898                 if (err)
1899                         goto err_create_flow;
1900
1901                 if (is_bonded) {
1902                         flow_attr->port = 2;
1903                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1904                                                        &mflow->reg_id[j].mirror);
1905                         flow_attr->port = 1;
1906                         if (err)
1907                                 goto err_create_flow;
1908                         j++;
1909                 }
1910                 /* function to create mirror rule */
1911                 i++;
1912         }
1913
1914         return &mflow->ibflow;
1915
1916 err_create_flow:
1917         while (i) {
1918                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1919                                              mflow->reg_id[i].id);
1920                 i--;
1921         }
1922
1923         while (j) {
1924                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1925                                              mflow->reg_id[j].mirror);
1926                 j--;
1927         }
1928 err_free:
1929         kfree(mflow);
1930         return ERR_PTR(err);
1931 }
1932
1933 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1934 {
1935         int err, ret = 0;
1936         int i = 0;
1937         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1938         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1939
1940         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1941                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1942                 if (err)
1943                         ret = err;
1944                 if (mflow->reg_id[i].mirror) {
1945                         err = __mlx4_ib_destroy_flow(mdev->dev,
1946                                                      mflow->reg_id[i].mirror);
1947                         if (err)
1948                                 ret = err;
1949                 }
1950                 i++;
1951         }
1952
1953         kfree(mflow);
1954         return ret;
1955 }
1956
1957 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1958 {
1959         int err;
1960         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1961         struct mlx4_dev *dev = mdev->dev;
1962         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1963         struct mlx4_ib_steering *ib_steering = NULL;
1964         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1965         struct mlx4_flow_reg_id reg_id;
1966
1967         if (mdev->dev->caps.steering_mode ==
1968             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1969                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1970                 if (!ib_steering)
1971                         return -ENOMEM;
1972         }
1973
1974         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1975                                     !!(mqp->flags &
1976                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1977                                     prot, &reg_id.id);
1978         if (err) {
1979                 pr_err("multicast attach op failed, err %d\n", err);
1980                 goto err_malloc;
1981         }
1982
1983         reg_id.mirror = 0;
1984         if (mlx4_is_bonded(dev)) {
1985                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1986                                             (mqp->port == 1) ? 2 : 1,
1987                                             !!(mqp->flags &
1988                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1989                                             prot, &reg_id.mirror);
1990                 if (err)
1991                         goto err_add;
1992         }
1993
1994         err = add_gid_entry(ibqp, gid);
1995         if (err)
1996                 goto err_add;
1997
1998         if (ib_steering) {
1999                 memcpy(ib_steering->gid.raw, gid->raw, 16);
2000                 ib_steering->reg_id = reg_id;
2001                 mutex_lock(&mqp->mutex);
2002                 list_add(&ib_steering->list, &mqp->steering_rules);
2003                 mutex_unlock(&mqp->mutex);
2004         }
2005         return 0;
2006
2007 err_add:
2008         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2009                               prot, reg_id.id);
2010         if (reg_id.mirror)
2011                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2012                                       prot, reg_id.mirror);
2013 err_malloc:
2014         kfree(ib_steering);
2015
2016         return err;
2017 }
2018
2019 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
2020 {
2021         struct mlx4_ib_gid_entry *ge;
2022         struct mlx4_ib_gid_entry *tmp;
2023         struct mlx4_ib_gid_entry *ret = NULL;
2024
2025         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
2026                 if (!memcmp(raw, ge->gid.raw, 16)) {
2027                         ret = ge;
2028                         break;
2029                 }
2030         }
2031
2032         return ret;
2033 }
2034
2035 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
2036 {
2037         int err;
2038         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
2039         struct mlx4_dev *dev = mdev->dev;
2040         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
2041         struct net_device *ndev;
2042         struct mlx4_ib_gid_entry *ge;
2043         struct mlx4_flow_reg_id reg_id = {0, 0};
2044         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
2045
2046         if (mdev->dev->caps.steering_mode ==
2047             MLX4_STEERING_MODE_DEVICE_MANAGED) {
2048                 struct mlx4_ib_steering *ib_steering;
2049
2050                 mutex_lock(&mqp->mutex);
2051                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
2052                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
2053                                 list_del(&ib_steering->list);
2054                                 break;
2055                         }
2056                 }
2057                 mutex_unlock(&mqp->mutex);
2058                 if (&ib_steering->list == &mqp->steering_rules) {
2059                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
2060                         return -EINVAL;
2061                 }
2062                 reg_id = ib_steering->reg_id;
2063                 kfree(ib_steering);
2064         }
2065
2066         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2067                                     prot, reg_id.id);
2068         if (err)
2069                 return err;
2070
2071         if (mlx4_is_bonded(dev)) {
2072                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
2073                                             prot, reg_id.mirror);
2074                 if (err)
2075                         return err;
2076         }
2077
2078         mutex_lock(&mqp->mutex);
2079         ge = find_gid_entry(mqp, gid->raw);
2080         if (ge) {
2081                 spin_lock_bh(&mdev->iboe.lock);
2082                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
2083                 if (ndev)
2084                         dev_hold(ndev);
2085                 spin_unlock_bh(&mdev->iboe.lock);
2086                 if (ndev)
2087                         dev_put(ndev);
2088                 list_del(&ge->list);
2089                 kfree(ge);
2090         } else
2091                 pr_warn("could not find mgid entry\n");
2092
2093         mutex_unlock(&mqp->mutex);
2094
2095         return 0;
2096 }
2097
2098 static int init_node_data(struct mlx4_ib_dev *dev)
2099 {
2100         struct ib_smp *in_mad  = NULL;
2101         struct ib_smp *out_mad = NULL;
2102         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
2103         int err = -ENOMEM;
2104
2105         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
2106         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
2107         if (!in_mad || !out_mad)
2108                 goto out;
2109
2110         init_query_mad(in_mad);
2111         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
2112         if (mlx4_is_master(dev->dev))
2113                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
2114
2115         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2116         if (err)
2117                 goto out;
2118
2119         memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
2120
2121         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2122
2123         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2124         if (err)
2125                 goto out;
2126
2127         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2128         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2129
2130 out:
2131         kfree(in_mad);
2132         kfree(out_mad);
2133         return err;
2134 }
2135
2136 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2137                         char *buf)
2138 {
2139         struct mlx4_ib_dev *dev =
2140                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2141         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2142 }
2143
2144 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2145                         char *buf)
2146 {
2147         struct mlx4_ib_dev *dev =
2148                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2149         return sprintf(buf, "%x\n", dev->dev->rev_id);
2150 }
2151
2152 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2153                           char *buf)
2154 {
2155         struct mlx4_ib_dev *dev =
2156                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2157         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2158                        dev->dev->board_id);
2159 }
2160
2161 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2162 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2163 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2164
2165 static struct device_attribute *mlx4_class_attributes[] = {
2166         &dev_attr_hw_rev,
2167         &dev_attr_hca_type,
2168         &dev_attr_board_id
2169 };
2170
2171 struct diag_counter {
2172         const char *name;
2173         u32 offset;
2174 };
2175
2176 #define DIAG_COUNTER(_name, _offset)                    \
2177         { .name = #_name, .offset = _offset }
2178
2179 static const struct diag_counter diag_basic[] = {
2180         DIAG_COUNTER(rq_num_lle, 0x00),
2181         DIAG_COUNTER(sq_num_lle, 0x04),
2182         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2183         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2184         DIAG_COUNTER(rq_num_lpe, 0x18),
2185         DIAG_COUNTER(sq_num_lpe, 0x1C),
2186         DIAG_COUNTER(rq_num_wrfe, 0x20),
2187         DIAG_COUNTER(sq_num_wrfe, 0x24),
2188         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2189         DIAG_COUNTER(sq_num_bre, 0x34),
2190         DIAG_COUNTER(sq_num_rire, 0x44),
2191         DIAG_COUNTER(rq_num_rire, 0x48),
2192         DIAG_COUNTER(sq_num_rae, 0x4C),
2193         DIAG_COUNTER(rq_num_rae, 0x50),
2194         DIAG_COUNTER(sq_num_roe, 0x54),
2195         DIAG_COUNTER(sq_num_tree, 0x5C),
2196         DIAG_COUNTER(sq_num_rree, 0x64),
2197         DIAG_COUNTER(rq_num_rnr, 0x68),
2198         DIAG_COUNTER(sq_num_rnr, 0x6C),
2199         DIAG_COUNTER(rq_num_oos, 0x100),
2200         DIAG_COUNTER(sq_num_oos, 0x104),
2201 };
2202
2203 static const struct diag_counter diag_ext[] = {
2204         DIAG_COUNTER(rq_num_dup, 0x130),
2205         DIAG_COUNTER(sq_num_to, 0x134),
2206 };
2207
2208 static const struct diag_counter diag_device_only[] = {
2209         DIAG_COUNTER(num_cqovf, 0x1A0),
2210         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2211 };
2212
2213 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2214                                                     u8 port_num)
2215 {
2216         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2217         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2218
2219         if (!diag[!!port_num].name)
2220                 return NULL;
2221
2222         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2223                                           diag[!!port_num].num_counters,
2224                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2225 }
2226
2227 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2228                                 struct rdma_hw_stats *stats,
2229                                 u8 port, int index)
2230 {
2231         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2232         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2233         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2234                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2235         int ret;
2236         int i;
2237
2238         ret = mlx4_query_diag_counters(dev->dev,
2239                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2240                                        diag[!!port].offset, hw_value,
2241                                        diag[!!port].num_counters, port);
2242
2243         if (ret)
2244                 return ret;
2245
2246         for (i = 0; i < diag[!!port].num_counters; i++)
2247                 stats->value[i] = hw_value[i];
2248
2249         return diag[!!port].num_counters;
2250 }
2251
2252 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2253                                          const char ***name,
2254                                          u32 **offset,
2255                                          u32 *num,
2256                                          bool port)
2257 {
2258         u32 num_counters;
2259
2260         num_counters = ARRAY_SIZE(diag_basic);
2261
2262         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2263                 num_counters += ARRAY_SIZE(diag_ext);
2264
2265         if (!port)
2266                 num_counters += ARRAY_SIZE(diag_device_only);
2267
2268         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2269         if (!*name)
2270                 return -ENOMEM;
2271
2272         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2273         if (!*offset)
2274                 goto err_name;
2275
2276         *num = num_counters;
2277
2278         return 0;
2279
2280 err_name:
2281         kfree(*name);
2282         return -ENOMEM;
2283 }
2284
2285 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2286                                        const char **name,
2287                                        u32 *offset,
2288                                        bool port)
2289 {
2290         int i;
2291         int j;
2292
2293         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2294                 name[i] = diag_basic[i].name;
2295                 offset[i] = diag_basic[i].offset;
2296         }
2297
2298         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2299                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2300                         name[j] = diag_ext[i].name;
2301                         offset[j] = diag_ext[i].offset;
2302                 }
2303         }
2304
2305         if (!port) {
2306                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2307                         name[j] = diag_device_only[i].name;
2308                         offset[j] = diag_device_only[i].offset;
2309                 }
2310         }
2311 }
2312
2313 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2314 {
2315         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2316         int i;
2317         int ret;
2318         bool per_port = !!(ibdev->dev->caps.flags2 &
2319                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2320
2321         if (mlx4_is_slave(ibdev->dev))
2322                 return 0;
2323
2324         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2325                 /* i == 1 means we are building port counters */
2326                 if (i && !per_port)
2327                         continue;
2328
2329                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2330                                                     &diag[i].offset,
2331                                                     &diag[i].num_counters, i);
2332                 if (ret)
2333                         goto err_alloc;
2334
2335                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2336                                            diag[i].offset, i);
2337         }
2338
2339         ibdev->ib_dev.get_hw_stats      = mlx4_ib_get_hw_stats;
2340         ibdev->ib_dev.alloc_hw_stats    = mlx4_ib_alloc_hw_stats;
2341
2342         return 0;
2343
2344 err_alloc:
2345         if (i) {
2346                 kfree(diag[i - 1].name);
2347                 kfree(diag[i - 1].offset);
2348         }
2349
2350         return ret;
2351 }
2352
2353 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2354 {
2355         int i;
2356
2357         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2358                 kfree(ibdev->diag_counters[i].offset);
2359                 kfree(ibdev->diag_counters[i].name);
2360         }
2361 }
2362
2363 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2364 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2365                                struct net_device *dev,
2366                                int port)
2367 {
2368         u64 new_smac = 0;
2369         u64 release_mac = MLX4_IB_INVALID_MAC;
2370         struct mlx4_ib_qp *qp;
2371
2372         read_lock(&dev_base_lock);
2373         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2374         read_unlock(&dev_base_lock);
2375
2376         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2377
2378         /* no need for update QP1 and mac registration in non-SRIOV */
2379         if (!mlx4_is_mfunc(ibdev->dev))
2380                 return;
2381
2382         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2383         qp = ibdev->qp1_proxy[port - 1];
2384         if (qp) {
2385                 int new_smac_index;
2386                 u64 old_smac;
2387                 struct mlx4_update_qp_params update_params;
2388
2389                 mutex_lock(&qp->mutex);
2390                 old_smac = qp->pri.smac;
2391                 if (new_smac == old_smac)
2392                         goto unlock;
2393
2394                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2395
2396                 if (new_smac_index < 0)
2397                         goto unlock;
2398
2399                 update_params.smac_index = new_smac_index;
2400                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2401                                    &update_params)) {
2402                         release_mac = new_smac;
2403                         goto unlock;
2404                 }
2405                 /* if old port was zero, no mac was yet registered for this QP */
2406                 if (qp->pri.smac_port)
2407                         release_mac = old_smac;
2408                 qp->pri.smac = new_smac;
2409                 qp->pri.smac_port = port;
2410                 qp->pri.smac_index = new_smac_index;
2411         }
2412
2413 unlock:
2414         if (release_mac != MLX4_IB_INVALID_MAC)
2415                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2416         if (qp)
2417                 mutex_unlock(&qp->mutex);
2418         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2419 }
2420
2421 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2422                                  struct net_device *dev,
2423                                  unsigned long event)
2424
2425 {
2426         struct mlx4_ib_iboe *iboe;
2427         int update_qps_port = -1;
2428         int port;
2429
2430         ASSERT_RTNL();
2431
2432         iboe = &ibdev->iboe;
2433
2434         spin_lock_bh(&iboe->lock);
2435         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2436
2437                 iboe->netdevs[port - 1] =
2438                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2439
2440                 if (dev == iboe->netdevs[port - 1] &&
2441                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2442                      event == NETDEV_UP || event == NETDEV_CHANGE))
2443                         update_qps_port = port;
2444
2445         }
2446         spin_unlock_bh(&iboe->lock);
2447
2448         if (update_qps_port > 0)
2449                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2450 }
2451
2452 static int mlx4_ib_netdev_event(struct notifier_block *this,
2453                                 unsigned long event, void *ptr)
2454 {
2455         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2456         struct mlx4_ib_dev *ibdev;
2457
2458         if (!net_eq(dev_net(dev), &init_net))
2459                 return NOTIFY_DONE;
2460
2461         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2462         mlx4_ib_scan_netdevs(ibdev, dev, event);
2463
2464         return NOTIFY_DONE;
2465 }
2466
2467 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2468 {
2469         int port;
2470         int slave;
2471         int i;
2472
2473         if (mlx4_is_master(ibdev->dev)) {
2474                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2475                      ++slave) {
2476                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2477                                 for (i = 0;
2478                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2479                                      ++i) {
2480                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2481                                         /* master has the identity virt2phys pkey mapping */
2482                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2483                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2484                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2485                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2486                                 }
2487                         }
2488                 }
2489                 /* initialize pkey cache */
2490                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2491                         for (i = 0;
2492                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2493                              ++i)
2494                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2495                                         (i) ? 0 : 0xFFFF;
2496                 }
2497         }
2498 }
2499
2500 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2501 {
2502         int i, j, eq = 0, total_eqs = 0;
2503
2504         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2505                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2506         if (!ibdev->eq_table)
2507                 return;
2508
2509         for (i = 1; i <= dev->caps.num_ports; i++) {
2510                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2511                      j++, total_eqs++) {
2512                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2513                                 continue;
2514                         ibdev->eq_table[eq] = total_eqs;
2515                         if (!mlx4_assign_eq(dev, i,
2516                                             &ibdev->eq_table[eq]))
2517                                 eq++;
2518                         else
2519                                 ibdev->eq_table[eq] = -1;
2520                 }
2521         }
2522
2523         for (i = eq; i < dev->caps.num_comp_vectors;
2524              ibdev->eq_table[i++] = -1)
2525                 ;
2526
2527         /* Advertise the new number of EQs to clients */
2528         ibdev->ib_dev.num_comp_vectors = eq;
2529 }
2530
2531 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2532 {
2533         int i;
2534         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2535
2536         /* no eqs were allocated */
2537         if (!ibdev->eq_table)
2538                 return;
2539
2540         /* Reset the advertised EQ number */
2541         ibdev->ib_dev.num_comp_vectors = 0;
2542
2543         for (i = 0; i < total_eqs; i++)
2544                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2545
2546         kfree(ibdev->eq_table);
2547         ibdev->eq_table = NULL;
2548 }
2549
2550 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2551                                struct ib_port_immutable *immutable)
2552 {
2553         struct ib_port_attr attr;
2554         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2555         int err;
2556
2557         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2558                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2559                 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2560         } else {
2561                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2562                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2563                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2564                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2565                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2566                 immutable->core_cap_flags |= RDMA_CORE_PORT_RAW_PACKET;
2567                 if (immutable->core_cap_flags & (RDMA_CORE_PORT_IBA_ROCE |
2568                     RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP))
2569                         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2570         }
2571
2572         err = ib_query_port(ibdev, port_num, &attr);
2573         if (err)
2574                 return err;
2575
2576         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2577         immutable->gid_tbl_len = attr.gid_tbl_len;
2578
2579         return 0;
2580 }
2581
2582 static void get_fw_ver_str(struct ib_device *device, char *str)
2583 {
2584         struct mlx4_ib_dev *dev =
2585                 container_of(device, struct mlx4_ib_dev, ib_dev);
2586         snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d",
2587                  (int) (dev->dev->caps.fw_ver >> 32),
2588                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2589                  (int) dev->dev->caps.fw_ver & 0xffff);
2590 }
2591
2592 static void *mlx4_ib_add(struct mlx4_dev *dev)
2593 {
2594         struct mlx4_ib_dev *ibdev;
2595         int num_ports = 0;
2596         int i, j;
2597         int err;
2598         struct mlx4_ib_iboe *iboe;
2599         int ib_num_ports = 0;
2600         int num_req_counters;
2601         int allocated;
2602         u32 counter_index;
2603         struct counter_index *new_counter_index = NULL;
2604
2605         pr_info_once("%s", mlx4_ib_version);
2606
2607         num_ports = 0;
2608         mlx4_foreach_ib_transport_port(i, dev)
2609                 num_ports++;
2610
2611         /* No point in registering a device with no ports... */
2612         if (num_ports == 0)
2613                 return NULL;
2614
2615         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2616         if (!ibdev) {
2617                 dev_err(&dev->persist->pdev->dev,
2618                         "Device struct alloc failed\n");
2619                 return NULL;
2620         }
2621
2622         iboe = &ibdev->iboe;
2623
2624         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2625                 goto err_dealloc;
2626
2627         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2628                 goto err_pd;
2629
2630         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2631                                  PAGE_SIZE);
2632         if (!ibdev->uar_map)
2633                 goto err_uar;
2634         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2635
2636         ibdev->dev = dev;
2637         ibdev->bond_next_port   = 0;
2638
2639         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2640         ibdev->ib_dev.owner             = THIS_MODULE;
2641         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2642         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2643         ibdev->num_ports                = num_ports;
2644         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2645                                                 1 : ibdev->num_ports;
2646         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2647         ibdev->ib_dev.dev.parent        = &dev->persist->pdev->dev;
2648         ibdev->ib_dev.get_netdev        = mlx4_ib_get_netdev;
2649         ibdev->ib_dev.add_gid           = mlx4_ib_add_gid;
2650         ibdev->ib_dev.del_gid           = mlx4_ib_del_gid;
2651
2652         if (dev->caps.userspace_caps)
2653                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2654         else
2655                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2656
2657         ibdev->ib_dev.uverbs_cmd_mask   =
2658                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2659                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2660                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2661                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2662                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2663                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2664                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2665                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2666                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2667                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2668                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2669                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2670                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2671                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2672                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2673                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2674                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2675                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2676                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2677                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2678                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2679                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2680                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2681                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2682
2683         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2684         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2685         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2686         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2687         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2688         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2689         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2690         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2691         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2692         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2693         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2694         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2695         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2696         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2697         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2698         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2699         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2700         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2701         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2702         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2703         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2704         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2705         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2706         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2707         ibdev->ib_dev.drain_sq          = mlx4_ib_drain_sq;
2708         ibdev->ib_dev.drain_rq          = mlx4_ib_drain_rq;
2709         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2710         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2711         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2712         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2713         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2714         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2715         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2716         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2717         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2718         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2719         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2720         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2721         ibdev->ib_dev.alloc_mr          = mlx4_ib_alloc_mr;
2722         ibdev->ib_dev.map_mr_sg         = mlx4_ib_map_mr_sg;
2723         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2724         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2725         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2726         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2727         ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2728         ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2729
2730         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2731                 (1ull << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
2732
2733         if ((dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_RSS) &&
2734             ((mlx4_ib_port_link_layer(&ibdev->ib_dev, 1) ==
2735             IB_LINK_LAYER_ETHERNET) ||
2736             (mlx4_ib_port_link_layer(&ibdev->ib_dev, 2) ==
2737             IB_LINK_LAYER_ETHERNET))) {
2738                 ibdev->ib_dev.create_wq         = mlx4_ib_create_wq;
2739                 ibdev->ib_dev.modify_wq         = mlx4_ib_modify_wq;
2740                 ibdev->ib_dev.destroy_wq        = mlx4_ib_destroy_wq;
2741                 ibdev->ib_dev.create_rwq_ind_table  =
2742                         mlx4_ib_create_rwq_ind_table;
2743                 ibdev->ib_dev.destroy_rwq_ind_table =
2744                         mlx4_ib_destroy_rwq_ind_table;
2745                 ibdev->ib_dev.uverbs_ex_cmd_mask |=
2746                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_WQ)          |
2747                         (1ull << IB_USER_VERBS_EX_CMD_MODIFY_WQ)          |
2748                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_WQ)         |
2749                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_RWQ_IND_TBL) |
2750                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_RWQ_IND_TBL);
2751         }
2752
2753         if (!mlx4_is_slave(ibdev->dev)) {
2754                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2755                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2756                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2757                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2758         }
2759
2760         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2761             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2762                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2763                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2764
2765                 ibdev->ib_dev.uverbs_cmd_mask |=
2766                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2767                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2768         }
2769
2770         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2771                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2772                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2773                 ibdev->ib_dev.uverbs_cmd_mask |=
2774                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2775                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2776         }
2777
2778         if (check_flow_steering_support(dev)) {
2779                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2780                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2781                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2782
2783                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2784                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2785                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2786         }
2787
2788         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2789                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2790                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2791                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2792
2793         mlx4_ib_alloc_eqs(dev, ibdev);
2794
2795         spin_lock_init(&iboe->lock);
2796
2797         if (init_node_data(ibdev))
2798                 goto err_map;
2799         mlx4_init_sl2vl_tbl(ibdev);
2800
2801         for (i = 0; i < ibdev->num_ports; ++i) {
2802                 mutex_init(&ibdev->counters_table[i].mutex);
2803                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2804         }
2805
2806         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2807         for (i = 0; i < num_req_counters; ++i) {
2808                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2809                 allocated = 0;
2810                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2811                                                 IB_LINK_LAYER_ETHERNET) {
2812                         err = mlx4_counter_alloc(ibdev->dev, &counter_index,
2813                                                  MLX4_RES_USAGE_DRIVER);
2814                         /* if failed to allocate a new counter, use default */
2815                         if (err)
2816                                 counter_index =
2817                                         mlx4_get_default_counter_index(dev,
2818                                                                        i + 1);
2819                         else
2820                                 allocated = 1;
2821                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2822                         counter_index = mlx4_get_default_counter_index(dev,
2823                                                                        i + 1);
2824                 }
2825                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2826                                             GFP_KERNEL);
2827                 if (!new_counter_index) {
2828                         if (allocated)
2829                                 mlx4_counter_free(ibdev->dev, counter_index);
2830                         goto err_counter;
2831                 }
2832                 new_counter_index->index = counter_index;
2833                 new_counter_index->allocated = allocated;
2834                 list_add_tail(&new_counter_index->list,
2835                               &ibdev->counters_table[i].counters_list);
2836                 ibdev->counters_table[i].default_counter = counter_index;
2837                 pr_info("counter index %d for port %d allocated %d\n",
2838                         counter_index, i + 1, allocated);
2839         }
2840         if (mlx4_is_bonded(dev))
2841                 for (i = 1; i < ibdev->num_ports ; ++i) {
2842                         new_counter_index =
2843                                         kmalloc(sizeof(struct counter_index),
2844                                                 GFP_KERNEL);
2845                         if (!new_counter_index)
2846                                 goto err_counter;
2847                         new_counter_index->index = counter_index;
2848                         new_counter_index->allocated = 0;
2849                         list_add_tail(&new_counter_index->list,
2850                                       &ibdev->counters_table[i].counters_list);
2851                         ibdev->counters_table[i].default_counter =
2852                                                                 counter_index;
2853                 }
2854
2855         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2856                 ib_num_ports++;
2857
2858         spin_lock_init(&ibdev->sm_lock);
2859         mutex_init(&ibdev->cap_mask_mutex);
2860         INIT_LIST_HEAD(&ibdev->qp_list);
2861         spin_lock_init(&ibdev->reset_flow_resource_lock);
2862
2863         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2864             ib_num_ports) {
2865                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2866                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2867                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2868                                             &ibdev->steer_qpn_base, 0,
2869                                             MLX4_RES_USAGE_DRIVER);
2870                 if (err)
2871                         goto err_counter;
2872
2873                 ibdev->ib_uc_qpns_bitmap =
2874                         kmalloc_array(BITS_TO_LONGS(ibdev->steer_qpn_count),
2875                                       sizeof(long),
2876                                       GFP_KERNEL);
2877                 if (!ibdev->ib_uc_qpns_bitmap)
2878                         goto err_steer_qp_release;
2879
2880                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB) {
2881                         bitmap_zero(ibdev->ib_uc_qpns_bitmap,
2882                                     ibdev->steer_qpn_count);
2883                         err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2884                                         dev, ibdev->steer_qpn_base,
2885                                         ibdev->steer_qpn_base +
2886                                         ibdev->steer_qpn_count - 1);
2887                         if (err)
2888                                 goto err_steer_free_bitmap;
2889                 } else {
2890                         bitmap_fill(ibdev->ib_uc_qpns_bitmap,
2891                                     ibdev->steer_qpn_count);
2892                 }
2893         }
2894
2895         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2896                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2897
2898         if (mlx4_ib_alloc_diag_counters(ibdev))
2899                 goto err_steer_free_bitmap;
2900
2901         ibdev->ib_dev.driver_id = RDMA_DRIVER_MLX4;
2902         if (ib_register_device(&ibdev->ib_dev, NULL))
2903                 goto err_diag_counters;
2904
2905         if (mlx4_ib_mad_init(ibdev))
2906                 goto err_reg;
2907
2908         if (mlx4_ib_init_sriov(ibdev))
2909                 goto err_mad;
2910
2911         if (!iboe->nb.notifier_call) {
2912                 iboe->nb.notifier_call = mlx4_ib_netdev_event;
2913                 err = register_netdevice_notifier(&iboe->nb);
2914                 if (err) {
2915                         iboe->nb.notifier_call = NULL;
2916                         goto err_notif;
2917                 }
2918         }
2919         if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2920                 err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2921                 if (err)
2922                         goto err_notif;
2923         }
2924
2925         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2926                 if (device_create_file(&ibdev->ib_dev.dev,
2927                                        mlx4_class_attributes[j]))
2928                         goto err_notif;
2929         }
2930
2931         ibdev->ib_active = true;
2932         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2933                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2934                                          &ibdev->ib_dev);
2935
2936         if (mlx4_is_mfunc(ibdev->dev))
2937                 init_pkeys(ibdev);
2938
2939         /* create paravirt contexts for any VFs which are active */
2940         if (mlx4_is_master(ibdev->dev)) {
2941                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2942                         if (j == mlx4_master_func_num(ibdev->dev))
2943                                 continue;
2944                         if (mlx4_is_slave_active(ibdev->dev, j))
2945                                 do_slave_init(ibdev, j, 1);
2946                 }
2947         }
2948         return ibdev;
2949
2950 err_notif:
2951         if (ibdev->iboe.nb.notifier_call) {
2952                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2953                         pr_warn("failure unregistering notifier\n");
2954                 ibdev->iboe.nb.notifier_call = NULL;
2955         }
2956         flush_workqueue(wq);
2957
2958         mlx4_ib_close_sriov(ibdev);
2959
2960 err_mad:
2961         mlx4_ib_mad_cleanup(ibdev);
2962
2963 err_reg:
2964         ib_unregister_device(&ibdev->ib_dev);
2965
2966 err_diag_counters:
2967         mlx4_ib_diag_cleanup(ibdev);
2968
2969 err_steer_free_bitmap:
2970         kfree(ibdev->ib_uc_qpns_bitmap);
2971
2972 err_steer_qp_release:
2973         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2974                               ibdev->steer_qpn_count);
2975 err_counter:
2976         for (i = 0; i < ibdev->num_ports; ++i)
2977                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2978
2979 err_map:
2980         mlx4_ib_free_eqs(dev, ibdev);
2981         iounmap(ibdev->uar_map);
2982
2983 err_uar:
2984         mlx4_uar_free(dev, &ibdev->priv_uar);
2985
2986 err_pd:
2987         mlx4_pd_free(dev, ibdev->priv_pdn);
2988
2989 err_dealloc:
2990         ib_dealloc_device(&ibdev->ib_dev);
2991
2992         return NULL;
2993 }
2994
2995 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2996 {
2997         int offset;
2998
2999         WARN_ON(!dev->ib_uc_qpns_bitmap);
3000
3001         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
3002                                          dev->steer_qpn_count,
3003                                          get_count_order(count));
3004         if (offset < 0)
3005                 return offset;
3006
3007         *qpn = dev->steer_qpn_base + offset;
3008         return 0;
3009 }
3010
3011 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
3012 {
3013         if (!qpn ||
3014             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
3015                 return;
3016
3017         if (WARN(qpn < dev->steer_qpn_base, "qpn = %u, steer_qpn_base = %u\n",
3018                  qpn, dev->steer_qpn_base))
3019                 /* not supposed to be here */
3020                 return;
3021
3022         bitmap_release_region(dev->ib_uc_qpns_bitmap,
3023                               qpn - dev->steer_qpn_base,
3024                               get_count_order(count));
3025 }
3026
3027 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
3028                          int is_attach)
3029 {
3030         int err;
3031         size_t flow_size;
3032         struct ib_flow_attr *flow = NULL;
3033         struct ib_flow_spec_ib *ib_spec;
3034
3035         if (is_attach) {
3036                 flow_size = sizeof(struct ib_flow_attr) +
3037                             sizeof(struct ib_flow_spec_ib);
3038                 flow = kzalloc(flow_size, GFP_KERNEL);
3039                 if (!flow)
3040                         return -ENOMEM;
3041                 flow->port = mqp->port;
3042                 flow->num_of_specs = 1;
3043                 flow->size = flow_size;
3044                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
3045                 ib_spec->type = IB_FLOW_SPEC_IB;
3046                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
3047                 /* Add an empty rule for IB L2 */
3048                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
3049
3050                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
3051                                             IB_FLOW_DOMAIN_NIC,
3052                                             MLX4_FS_REGULAR,
3053                                             &mqp->reg_id);
3054         } else {
3055                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
3056         }
3057         kfree(flow);
3058         return err;
3059 }
3060
3061 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
3062 {
3063         struct mlx4_ib_dev *ibdev = ibdev_ptr;
3064         int p;
3065         int i;
3066
3067         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
3068                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
3069         ibdev->ib_active = false;
3070         flush_workqueue(wq);
3071
3072         mlx4_ib_close_sriov(ibdev);
3073         mlx4_ib_mad_cleanup(ibdev);
3074         ib_unregister_device(&ibdev->ib_dev);
3075         mlx4_ib_diag_cleanup(ibdev);
3076         if (ibdev->iboe.nb.notifier_call) {
3077                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
3078                         pr_warn("failure unregistering notifier\n");
3079                 ibdev->iboe.nb.notifier_call = NULL;
3080         }
3081
3082         mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
3083                               ibdev->steer_qpn_count);
3084         kfree(ibdev->ib_uc_qpns_bitmap);
3085
3086         iounmap(ibdev->uar_map);
3087         for (p = 0; p < ibdev->num_ports; ++p)
3088                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
3089
3090         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
3091                 mlx4_CLOSE_PORT(dev, p);
3092
3093         mlx4_ib_free_eqs(dev, ibdev);
3094
3095         mlx4_uar_free(dev, &ibdev->priv_uar);
3096         mlx4_pd_free(dev, ibdev->priv_pdn);
3097         ib_dealloc_device(&ibdev->ib_dev);
3098 }
3099
3100 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
3101 {
3102         struct mlx4_ib_demux_work **dm = NULL;
3103         struct mlx4_dev *dev = ibdev->dev;
3104         int i;
3105         unsigned long flags;
3106         struct mlx4_active_ports actv_ports;
3107         unsigned int ports;
3108         unsigned int first_port;
3109
3110         if (!mlx4_is_master(dev))
3111                 return;
3112
3113         actv_ports = mlx4_get_active_ports(dev, slave);
3114         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
3115         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
3116
3117         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
3118         if (!dm)
3119                 return;
3120
3121         for (i = 0; i < ports; i++) {
3122                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
3123                 if (!dm[i]) {
3124                         while (--i >= 0)
3125                                 kfree(dm[i]);
3126                         goto out;
3127                 }
3128                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
3129                 dm[i]->port = first_port + i + 1;
3130                 dm[i]->slave = slave;
3131                 dm[i]->do_init = do_init;
3132                 dm[i]->dev = ibdev;
3133         }
3134         /* initialize or tear down tunnel QPs for the slave */
3135         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
3136         if (!ibdev->sriov.is_going_down) {
3137                 for (i = 0; i < ports; i++)
3138                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
3139                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3140         } else {
3141                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
3142                 for (i = 0; i < ports; i++)
3143                         kfree(dm[i]);
3144         }
3145 out:
3146         kfree(dm);
3147         return;
3148 }
3149
3150 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3151 {
3152         struct mlx4_ib_qp *mqp;
3153         unsigned long flags_qp;
3154         unsigned long flags_cq;
3155         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3156         struct list_head    cq_notify_list;
3157         struct mlx4_cq *mcq;
3158         unsigned long flags;
3159
3160         pr_warn("mlx4_ib_handle_catas_error was started\n");
3161         INIT_LIST_HEAD(&cq_notify_list);
3162
3163         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3164         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3165
3166         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3167                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3168                 if (mqp->sq.tail != mqp->sq.head) {
3169                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3170                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3171                         if (send_mcq->mcq.comp &&
3172                             mqp->ibqp.send_cq->comp_handler) {
3173                                 if (!send_mcq->mcq.reset_notify_added) {
3174                                         send_mcq->mcq.reset_notify_added = 1;
3175                                         list_add_tail(&send_mcq->mcq.reset_notify,
3176                                                       &cq_notify_list);
3177                                 }
3178                         }
3179                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3180                 }
3181                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3182                 /* Now, handle the QP's receive queue */
3183                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3184                 /* no handling is needed for SRQ */
3185                 if (!mqp->ibqp.srq) {
3186                         if (mqp->rq.tail != mqp->rq.head) {
3187                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3188                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3189                                 if (recv_mcq->mcq.comp &&
3190                                     mqp->ibqp.recv_cq->comp_handler) {
3191                                         if (!recv_mcq->mcq.reset_notify_added) {
3192                                                 recv_mcq->mcq.reset_notify_added = 1;
3193                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3194                                                               &cq_notify_list);
3195                                         }
3196                                 }
3197                                 spin_unlock_irqrestore(&recv_mcq->lock,
3198                                                        flags_cq);
3199                         }
3200                 }
3201                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3202         }
3203
3204         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3205                 mcq->comp(mcq);
3206         }
3207         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3208         pr_warn("mlx4_ib_handle_catas_error ended\n");
3209 }
3210
3211 static void handle_bonded_port_state_event(struct work_struct *work)
3212 {
3213         struct ib_event_work *ew =
3214                 container_of(work, struct ib_event_work, work);
3215         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3216         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3217         int i;
3218         struct ib_event ibev;
3219
3220         kfree(ew);
3221         spin_lock_bh(&ibdev->iboe.lock);
3222         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3223                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3224                 enum ib_port_state curr_port_state;
3225
3226                 if (!curr_netdev)
3227                         continue;
3228
3229                 curr_port_state =
3230                         (netif_running(curr_netdev) &&
3231                          netif_carrier_ok(curr_netdev)) ?
3232                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3233
3234                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3235                         curr_port_state : IB_PORT_ACTIVE;
3236         }
3237         spin_unlock_bh(&ibdev->iboe.lock);
3238
3239         ibev.device = &ibdev->ib_dev;
3240         ibev.element.port_num = 1;
3241         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3242                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3243
3244         ib_dispatch_event(&ibev);
3245 }
3246
3247 void mlx4_ib_sl2vl_update(struct mlx4_ib_dev *mdev, int port)
3248 {
3249         u64 sl2vl;
3250         int err;
3251
3252         err = mlx4_ib_query_sl2vl(&mdev->ib_dev, port, &sl2vl);
3253         if (err) {
3254                 pr_err("Unable to get current sl to vl mapping for port %d.  Using all zeroes (%d)\n",
3255                        port, err);
3256                 sl2vl = 0;
3257         }
3258         atomic64_set(&mdev->sl2vl[port - 1], sl2vl);
3259 }
3260
3261 static void ib_sl2vl_update_work(struct work_struct *work)
3262 {
3263         struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
3264         struct mlx4_ib_dev *mdev = ew->ib_dev;
3265         int port = ew->port;
3266
3267         mlx4_ib_sl2vl_update(mdev, port);
3268
3269         kfree(ew);
3270 }
3271
3272 void mlx4_sched_ib_sl2vl_update_work(struct mlx4_ib_dev *ibdev,
3273                                      int port)
3274 {
3275         struct ib_event_work *ew;
3276
3277         ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3278         if (ew) {
3279                 INIT_WORK(&ew->work, ib_sl2vl_update_work);
3280                 ew->port = port;
3281                 ew->ib_dev = ibdev;
3282                 queue_work(wq, &ew->work);
3283         }
3284 }
3285
3286 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3287                           enum mlx4_dev_event event, unsigned long param)
3288 {
3289         struct ib_event ibev;
3290         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3291         struct mlx4_eqe *eqe = NULL;
3292         struct ib_event_work *ew;
3293         int p = 0;
3294
3295         if (mlx4_is_bonded(dev) &&
3296             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3297             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3298                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3299                 if (!ew)
3300                         return;
3301                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3302                 ew->ib_dev = ibdev;
3303                 queue_work(wq, &ew->work);
3304                 return;
3305         }
3306
3307         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3308                 eqe = (struct mlx4_eqe *)param;
3309         else
3310                 p = (int) param;
3311
3312         switch (event) {
3313         case MLX4_DEV_EVENT_PORT_UP:
3314                 if (p > ibdev->num_ports)
3315                         return;
3316                 if (!mlx4_is_slave(dev) &&
3317                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3318                         IB_LINK_LAYER_INFINIBAND) {
3319                         if (mlx4_is_master(dev))
3320                                 mlx4_ib_invalidate_all_guid_record(ibdev, p);
3321                         if (ibdev->dev->flags & MLX4_FLAG_SECURE_HOST &&
3322                             !(ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_SL_TO_VL_CHANGE_EVENT))
3323                                 mlx4_sched_ib_sl2vl_update_work(ibdev, p);
3324                 }
3325                 ibev.event = IB_EVENT_PORT_ACTIVE;
3326                 break;
3327
3328         case MLX4_DEV_EVENT_PORT_DOWN:
3329                 if (p > ibdev->num_ports)
3330                         return;
3331                 ibev.event = IB_EVENT_PORT_ERR;
3332                 break;
3333
3334         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3335                 ibdev->ib_active = false;
3336                 ibev.event = IB_EVENT_DEVICE_FATAL;
3337                 mlx4_ib_handle_catas_error(ibdev);
3338                 break;
3339
3340         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3341                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3342                 if (!ew)
3343                         break;
3344
3345                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3346                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3347                 ew->ib_dev = ibdev;
3348                 /* need to queue only for port owner, which uses GEN_EQE */
3349                 if (mlx4_is_master(dev))
3350                         queue_work(wq, &ew->work);
3351                 else
3352                         handle_port_mgmt_change_event(&ew->work);
3353                 return;
3354
3355         case MLX4_DEV_EVENT_SLAVE_INIT:
3356                 /* here, p is the slave id */
3357                 do_slave_init(ibdev, p, 1);
3358                 if (mlx4_is_master(dev)) {
3359                         int i;
3360
3361                         for (i = 1; i <= ibdev->num_ports; i++) {
3362                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3363                                         == IB_LINK_LAYER_INFINIBAND)
3364                                         mlx4_ib_slave_alias_guid_event(ibdev,
3365                                                                        p, i,
3366                                                                        1);
3367                         }
3368                 }
3369                 return;
3370
3371         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3372                 if (mlx4_is_master(dev)) {
3373                         int i;
3374
3375                         for (i = 1; i <= ibdev->num_ports; i++) {
3376                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3377                                         == IB_LINK_LAYER_INFINIBAND)
3378                                         mlx4_ib_slave_alias_guid_event(ibdev,
3379                                                                        p, i,
3380                                                                        0);
3381                         }
3382                 }
3383                 /* here, p is the slave id */
3384                 do_slave_init(ibdev, p, 0);
3385                 return;
3386
3387         default:
3388                 return;
3389         }
3390
3391         ibev.device           = ibdev_ptr;
3392         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3393
3394         ib_dispatch_event(&ibev);
3395 }
3396
3397 static struct mlx4_interface mlx4_ib_interface = {
3398         .add            = mlx4_ib_add,
3399         .remove         = mlx4_ib_remove,
3400         .event          = mlx4_ib_event,
3401         .protocol       = MLX4_PROT_IB_IPV6,
3402         .flags          = MLX4_INTFF_BONDING
3403 };
3404
3405 static int __init mlx4_ib_init(void)
3406 {
3407         int err;
3408
3409         wq = alloc_ordered_workqueue("mlx4_ib", WQ_MEM_RECLAIM);
3410         if (!wq)
3411                 return -ENOMEM;
3412
3413         err = mlx4_ib_mcg_init();
3414         if (err)
3415                 goto clean_wq;
3416
3417         err = mlx4_register_interface(&mlx4_ib_interface);
3418         if (err)
3419                 goto clean_mcg;
3420
3421         return 0;
3422
3423 clean_mcg:
3424         mlx4_ib_mcg_destroy();
3425
3426 clean_wq:
3427         destroy_workqueue(wq);
3428         return err;
3429 }
3430
3431 static void __exit mlx4_ib_cleanup(void)
3432 {
3433         mlx4_unregister_interface(&mlx4_ib_interface);
3434         mlx4_ib_mcg_destroy();
3435         destroy_workqueue(wq);
3436 }
3437
3438 module_init(mlx4_ib_init);
3439 module_exit(mlx4_ib_cleanup);