drivers/infiniband/hw/mlx4/main.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/module.h>
  35 #include <linux/init.h>
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/rtnetlink.h>
  41 #include <linux/if_vlan.h>
  42 #include <net/ipv6.h>
  43 #include <net/addrconf.h>
  44 #include <net/devlink.h>
  45
  46 #include <rdma/ib_smi.h>
  47 #include <rdma/ib_user_verbs.h>
  48 #include <rdma/ib_addr.h>
  49 #include <rdma/ib_cache.h>
  50
  51 #include <net/bonding.h>
  52
  53 #include <linux/mlx4/driver.h>
  54 #include <linux/mlx4/cmd.h>
  55 #include <linux/mlx4/qp.h>
  56
  57 #include "mlx4_ib.h"
  58 #include "user.h"
  59
  60 #define DRV_NAME        MLX4_IB_DRV_NAME
  61 #define DRV_VERSION     "2.2-1"
  62 #define DRV_RELDATE     "Feb 2014"
  63
  64 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  65 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  66 #define MLX4_IB_CARD_REV_A0   0xA0
  67
  68 MODULE_AUTHOR("Roland Dreier");
  69 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  70 MODULE_LICENSE("Dual BSD/GPL");
  71 MODULE_VERSION(DRV_VERSION);
  72
  73 int mlx4_ib_sm_guid_assign = 0;
  74 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  75 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
  76
  77 static const char mlx4_ib_version[] =
  78         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  79         DRV_VERSION " (" DRV_RELDATE ")\n";
  80
  81 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  82
  83 static struct workqueue_struct *wq;
  84
  85 static void init_query_mad(struct ib_smp *mad)
  86 {
  87         mad->base_version  = 1;
  88         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  89         mad->class_version = 1;
  90         mad->method        = IB_MGMT_METHOD_GET;
  91 }
  92
  93 static int check_flow_steering_support(struct mlx4_dev *dev)
  94 {
  95         int eth_num_ports = 0;
  96         int ib_num_ports = 0;
  97
  98         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
  99
 100         if (dmfs) {
 101                 int i;
 102                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 103                         eth_num_ports++;
 104                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 105                         ib_num_ports++;
 106                 dmfs &= (!ib_num_ports ||
 107                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 108                         (!eth_num_ports ||
 109                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 110                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 111                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 112                         dmfs = 0;
 113                 }
 114         }
 115         return dmfs;
 116 }
 117
 118 static int num_ib_ports(struct mlx4_dev *dev)
 119 {
 120         int ib_ports = 0;
 121         int i;
 122
 123         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 124                 ib_ports++;
 125
 126         return ib_ports;
 127 }
 128
 129 static struct net_device *mlx4_ib_get_netdev(struct ib_device *device, u8 port_num)
 130 {
 131         struct mlx4_ib_dev *ibdev = to_mdev(device);
 132         struct net_device *dev;
 133
 134         rcu_read_lock();
 135         dev = mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port_num);
 136
 137         if (dev) {
 138                 if (mlx4_is_bonded(ibdev->dev)) {
 139                         struct net_device *upper = NULL;
 140
 141                         upper = netdev_master_upper_dev_get_rcu(dev);
 142                         if (upper) {
 143                                 struct net_device *active;
 144
 145                                 active = bond_option_active_slave_get_rcu(netdev_priv(upper));
 146                                 if (active)
 147                                         dev = active;
 148                         }
 149                 }
 150         }
 151         if (dev)
 152                 dev_hold(dev);
 153
 154         rcu_read_unlock();
 155         return dev;
 156 }
 157
 158 static int mlx4_ib_update_gids_v1(struct gid_entry *gids,
 159                                   struct mlx4_ib_dev *ibdev,
 160                                   u8 port_num)
 161 {
 162         struct mlx4_cmd_mailbox *mailbox;
 163         int err;
 164         struct mlx4_dev *dev = ibdev->dev;
 165         int i;
 166         union ib_gid *gid_tbl;
 167
 168         mailbox = mlx4_alloc_cmd_mailbox(dev);
 169         if (IS_ERR(mailbox))
 170                 return -ENOMEM;
 171
 172         gid_tbl = mailbox->buf;
 173
 174         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 175                 memcpy(&gid_tbl[i], &gids[i].gid, sizeof(union ib_gid));
 176
 177         err = mlx4_cmd(dev, mailbox->dma,
 178                        MLX4_SET_PORT_GID_TABLE << 8 | port_num,
 179                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 180                        MLX4_CMD_WRAPPED);
 181         if (mlx4_is_bonded(dev))
 182                 err += mlx4_cmd(dev, mailbox->dma,
 183                                 MLX4_SET_PORT_GID_TABLE << 8 | 2,
 184                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 185                                 MLX4_CMD_WRAPPED);
 186
 187         mlx4_free_cmd_mailbox(dev, mailbox);
 188         return err;
 189 }
 190
 191 static int mlx4_ib_update_gids_v1_v2(struct gid_entry *gids,
 192                                      struct mlx4_ib_dev *ibdev,
 193                                      u8 port_num)
 194 {
 195         struct mlx4_cmd_mailbox *mailbox;
 196         int err;
 197         struct mlx4_dev *dev = ibdev->dev;
 198         int i;
 199         struct {
 200                 union ib_gid    gid;
 201                 __be32          rsrvd1[2];
 202                 __be16          rsrvd2;
 203                 u8              type;
 204                 u8              version;
 205                 __be32          rsrvd3;
 206         } *gid_tbl;
 207
 208         mailbox = mlx4_alloc_cmd_mailbox(dev);
 209         if (IS_ERR(mailbox))
 210                 return -ENOMEM;
 211
 212         gid_tbl = mailbox->buf;
 213         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 214                 memcpy(&gid_tbl[i].gid, &gids[i].gid, sizeof(union ib_gid));
 215                 if (gids[i].gid_type == IB_GID_TYPE_ROCE_UDP_ENCAP) {
 216                         gid_tbl[i].version = 2;
 217                         if (!ipv6_addr_v4mapped((struct in6_addr *)&gids[i].gid))
 218                                 gid_tbl[i].type = 1;
 219                         else
 220                                 memset(&gid_tbl[i].gid, 0, 12);
 221                 }
 222         }
 223
 224         err = mlx4_cmd(dev, mailbox->dma,
 225                        MLX4_SET_PORT_ROCE_ADDR << 8 | port_num,
 226                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 227                        MLX4_CMD_WRAPPED);
 228         if (mlx4_is_bonded(dev))
 229                 err += mlx4_cmd(dev, mailbox->dma,
 230                                 MLX4_SET_PORT_ROCE_ADDR << 8 | 2,
 231                                 1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 232                                 MLX4_CMD_WRAPPED);
 233
 234         mlx4_free_cmd_mailbox(dev, mailbox);
 235         return err;
 236 }
 237
 238 static int mlx4_ib_update_gids(struct gid_entry *gids,
 239                                struct mlx4_ib_dev *ibdev,
 240                                u8 port_num)
 241 {
 242         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
 243                 return mlx4_ib_update_gids_v1_v2(gids, ibdev, port_num);
 244
 245         return mlx4_ib_update_gids_v1(gids, ibdev, port_num);
 246 }
 247
 248 static int mlx4_ib_add_gid(struct ib_device *device,
 249                            u8 port_num,
 250                            unsigned int index,
 251                            const union ib_gid *gid,
 252                            const struct ib_gid_attr *attr,
 253                            void **context)
 254 {
 255         struct mlx4_ib_dev *ibdev = to_mdev(device);
 256         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 257         struct mlx4_port_gid_table   *port_gid_table;
 258         int free = -1, found = -1;
 259         int ret = 0;
 260         int hw_update = 0;
 261         int i;
 262         struct gid_entry *gids = NULL;
 263
 264         if (!rdma_cap_roce_gid_table(device, port_num))
 265                 return -EINVAL;
 266
 267         if (port_num > MLX4_MAX_PORTS)
 268                 return -EINVAL;
 269
 270         if (!context)
 271                 return -EINVAL;
 272
 273         port_gid_table = &iboe->gids[port_num - 1];
 274         spin_lock_bh(&iboe->lock);
 275         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i) {
 276                 if (!memcmp(&port_gid_table->gids[i].gid, gid, sizeof(*gid)) &&
 277                     (port_gid_table->gids[i].gid_type == attr->gid_type))  {
 278                         found = i;
 279                         break;
 280                 }
 281                 if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid)))
 282                         free = i; /* HW has space */
 283         }
 284
 285         if (found < 0) {
 286                 if (free < 0) {
 287                         ret = -ENOSPC;
 288                 } else {
 289                         port_gid_table->gids[free].ctx = kmalloc(sizeof(*port_gid_table->gids[free].ctx), GFP_ATOMIC);
 290                         if (!port_gid_table->gids[free].ctx) {
 291                                 ret = -ENOMEM;
 292                         } else {
 293                                 *context = port_gid_table->gids[free].ctx;
 294                                 memcpy(&port_gid_table->gids[free].gid, gid, sizeof(*gid));
 295                                 port_gid_table->gids[free].gid_type = attr->gid_type;
 296                                 port_gid_table->gids[free].ctx->real_index = free;
 297                                 port_gid_table->gids[free].ctx->refcount = 1;
 298                                 hw_update = 1;
 299                         }
 300                 }
 301         } else {
 302                 struct gid_cache_context *ctx = port_gid_table->gids[found].ctx;
 303                 *context = ctx;
 304                 ctx->refcount++;
 305         }
 306         if (!ret && hw_update) {
 307                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
 308                 if (!gids) {
 309                         ret = -ENOMEM;
 310                 } else {
 311                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++) {
 312                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 313                                 gids[i].gid_type = port_gid_table->gids[i].gid_type;
 314                         }
 315                 }
 316         }
 317         spin_unlock_bh(&iboe->lock);
 318
 319         if (!ret && hw_update) {
 320                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
 321                 kfree(gids);
 322         }
 323
 324         return ret;
 325 }
 326
 327 static int mlx4_ib_del_gid(struct ib_device *device,
 328                            u8 port_num,
 329                            unsigned int index,
 330                            void **context)
 331 {
 332         struct gid_cache_context *ctx = *context;
 333         struct mlx4_ib_dev *ibdev = to_mdev(device);
 334         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 335         struct mlx4_port_gid_table   *port_gid_table;
 336         int ret = 0;
 337         int hw_update = 0;
 338         struct gid_entry *gids = NULL;
 339
 340         if (!rdma_cap_roce_gid_table(device, port_num))
 341                 return -EINVAL;
 342
 343         if (port_num > MLX4_MAX_PORTS)
 344                 return -EINVAL;
 345
 346         port_gid_table = &iboe->gids[port_num - 1];
 347         spin_lock_bh(&iboe->lock);
 348         if (ctx) {
 349                 ctx->refcount--;
 350                 if (!ctx->refcount) {
 351                         unsigned int real_index = ctx->real_index;
 352
 353                         memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid));
 354                         kfree(port_gid_table->gids[real_index].ctx);
 355                         port_gid_table->gids[real_index].ctx = NULL;
 356                         hw_update = 1;
 357                 }
 358         }
 359         if (!ret && hw_update) {
 360                 int i;
 361
 362                 gids = kmalloc(sizeof(*gids) * MLX4_MAX_PORT_GIDS, GFP_ATOMIC);
 363                 if (!gids) {
 364                         ret = -ENOMEM;
 365                 } else {
 366                         for (i = 0; i < MLX4_MAX_PORT_GIDS; i++)
 367                                 memcpy(&gids[i].gid, &port_gid_table->gids[i].gid, sizeof(union ib_gid));
 368                 }
 369         }
 370         spin_unlock_bh(&iboe->lock);
 371
 372         if (!ret && hw_update) {
 373                 ret = mlx4_ib_update_gids(gids, ibdev, port_num);
 374                 kfree(gids);
 375         }
 376         return ret;
 377 }
 378
 379 int mlx4_ib_gid_index_to_real_index(struct mlx4_ib_dev *ibdev,
 380                                     u8 port_num, int index)
 381 {
 382         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
 383         struct gid_cache_context *ctx = NULL;
 384         union ib_gid gid;
 385         struct mlx4_port_gid_table   *port_gid_table;
 386         int real_index = -EINVAL;
 387         int i;
 388         int ret;
 389         unsigned long flags;
 390         struct ib_gid_attr attr;
 391
 392         if (port_num > MLX4_MAX_PORTS)
 393                 return -EINVAL;
 394
 395         if (mlx4_is_bonded(ibdev->dev))
 396                 port_num = 1;
 397
 398         if (!rdma_cap_roce_gid_table(&ibdev->ib_dev, port_num))
 399                 return index;
 400
 401         ret = ib_get_cached_gid(&ibdev->ib_dev, port_num, index, &gid, &attr);
 402         if (ret)
 403                 return ret;
 404
 405         if (attr.ndev)
 406                 dev_put(attr.ndev);
 407
 408         if (!memcmp(&gid, &zgid, sizeof(gid)))
 409                 return -EINVAL;
 410
 411         spin_lock_irqsave(&iboe->lock, flags);
 412         port_gid_table = &iboe->gids[port_num - 1];
 413
 414         for (i = 0; i < MLX4_MAX_PORT_GIDS; ++i)
 415                 if (!memcmp(&port_gid_table->gids[i].gid, &gid, sizeof(gid)) &&
 416                     attr.gid_type == port_gid_table->gids[i].gid_type) {
 417                         ctx = port_gid_table->gids[i].ctx;
 418                         break;
 419                 }
 420         if (ctx)
 421                 real_index = ctx->real_index;
 422         spin_unlock_irqrestore(&iboe->lock, flags);
 423         return real_index;
 424 }
 425
 426 static int mlx4_ib_query_device(struct ib_device *ibdev,
 427                                 struct ib_device_attr *props,
 428                                 struct ib_udata *uhw)
 429 {
 430         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 431         struct ib_smp *in_mad  = NULL;
 432         struct ib_smp *out_mad = NULL;
 433         int err = -ENOMEM;
 434         int have_ib_ports;
 435         struct mlx4_uverbs_ex_query_device cmd;
 436         struct mlx4_uverbs_ex_query_device_resp resp = {.comp_mask = 0};
 437         struct mlx4_clock_params clock_params;
 438
 439         if (uhw->inlen) {
 440                 if (uhw->inlen < sizeof(cmd))
 441                         return -EINVAL;
 442
 443                 err = ib_copy_from_udata(&cmd, uhw, sizeof(cmd));
 444                 if (err)
 445                         return err;
 446
 447                 if (cmd.comp_mask)
 448                         return -EINVAL;
 449
 450                 if (cmd.reserved)
 451                         return -EINVAL;
 452         }
 453
 454         resp.response_length = offsetof(typeof(resp), response_length) +
 455                 sizeof(resp.response_length);
 456         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 457         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 458         if (!in_mad || !out_mad)
 459                 goto out;
 460
 461         init_query_mad(in_mad);
 462         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 463
 464         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 465                            1, NULL, NULL, in_mad, out_mad);
 466         if (err)
 467                 goto out;
 468
 469         memset(props, 0, sizeof *props);
 470
 471         have_ib_ports = num_ib_ports(dev->dev);
 472
 473         props->fw_ver = dev->dev->caps.fw_ver;
 474         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 475                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 476                 IB_DEVICE_SYS_IMAGE_GUID                |
 477                 IB_DEVICE_RC_RNR_NAK_GEN                |
 478                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 479         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 480                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 481         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 482                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 483         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 484                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 485         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 486                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 487         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 488                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 489         if (dev->dev->caps.max_gso_sz &&
 490             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 491             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 492                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 493         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 494                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 495         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 496             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 497             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 498                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 499         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 500                 props->device_cap_flags |= IB_DEVICE_XRC;
 501         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 502                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 503         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 504                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 505                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 506                 else
 507                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 508         }
 509         if (dev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
 510                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 511
 512         props->device_cap_flags |= IB_DEVICE_RAW_IP_CSUM;
 513
 514         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 515                 0xffffff;
 516         props->vendor_part_id      = dev->dev->persist->pdev->device;
 517         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 518         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 519
 520         props->max_mr_size         = ~0ull;
 521         props->page_size_cap       = dev->dev->caps.page_size_cap;
 522         props->max_qp              = dev->dev->quotas.qp;
 523         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 524         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 525                                          dev->dev->caps.max_rq_sg);
 526         props->max_sge_rd          = MLX4_MAX_SGE_RD;
 527         props->max_cq              = dev->dev->quotas.cq;
 528         props->max_cqe             = dev->dev->caps.max_cqes;
 529         props->max_mr              = dev->dev->quotas.mpt;
 530         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 531         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 532         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 533         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 534         props->max_srq             = dev->dev->quotas.srq;
 535         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 536         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 537         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 538         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 539         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 540                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 541         props->masked_atomic_cap   = props->atomic_cap;
 542         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 543         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 544         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 545         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 546                                            props->max_mcast_grp;
 547         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 548         props->hca_core_clock = dev->dev->caps.hca_core_clock * 1000UL;
 549         props->timestamp_mask = 0xFFFFFFFFFFFFULL;
 550
 551         if (!mlx4_is_slave(dev->dev))
 552                 err = mlx4_get_internal_clock_params(dev->dev, &clock_params);
 553
 554         if (uhw->outlen >= resp.response_length + sizeof(resp.hca_core_clock_offset)) {
 555                 resp.response_length += sizeof(resp.hca_core_clock_offset);
 556                 if (!err && !mlx4_is_slave(dev->dev)) {
 557                         resp.comp_mask |= QUERY_DEVICE_RESP_MASK_TIMESTAMP;
 558                         resp.hca_core_clock_offset = clock_params.offset % PAGE_SIZE;
 559                 }
 560         }
 561
 562         if (uhw->outlen) {
 563                 err = ib_copy_to_udata(uhw, &resp, resp.response_length);
 564                 if (err)
 565                         goto out;
 566         }
 567 out:
 568         kfree(in_mad);
 569         kfree(out_mad);
 570
 571         return err;
 572 }
 573
 574 static enum rdma_link_layer
 575 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 576 {
 577         struct mlx4_dev *dev = to_mdev(device)->dev;
 578
 579         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 580                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 581 }
 582
 583 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 584                               struct ib_port_attr *props, int netw_view)
 585 {
 586         struct ib_smp *in_mad  = NULL;
 587         struct ib_smp *out_mad = NULL;
 588         int ext_active_speed;
 589         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 590         int err = -ENOMEM;
 591
 592         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 593         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 594         if (!in_mad || !out_mad)
 595                 goto out;
 596
 597         init_query_mad(in_mad);
 598         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 599         in_mad->attr_mod = cpu_to_be32(port);
 600
 601         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 602                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 603
 604         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 605                                 in_mad, out_mad);
 606         if (err)
 607                 goto out;
 608
 609
 610         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 611         props->lmc              = out_mad->data[34] & 0x7;
 612         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 613         props->sm_sl            = out_mad->data[36] & 0xf;
 614         props->state            = out_mad->data[32] & 0xf;
 615         props->phys_state       = out_mad->data[33] >> 4;
 616         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 617         if (netw_view)
 618                 props->gid_tbl_len = out_mad->data[50];
 619         else
 620                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 621         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 622         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 623         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 624         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 625         props->active_width     = out_mad->data[31] & 0xf;
 626         props->active_speed     = out_mad->data[35] >> 4;
 627         props->max_mtu          = out_mad->data[41] & 0xf;
 628         props->active_mtu       = out_mad->data[36] >> 4;
 629         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 630         props->max_vl_num       = out_mad->data[37] >> 4;
 631         props->init_type_reply  = out_mad->data[41] >> 4;
 632
 633         /* Check if extended speeds (EDR/FDR/...) are supported */
 634         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 635                 ext_active_speed = out_mad->data[62] >> 4;
 636
 637                 switch (ext_active_speed) {
 638                 case 1:
 639                         props->active_speed = IB_SPEED_FDR;
 640                         break;
 641                 case 2:
 642                         props->active_speed = IB_SPEED_EDR;
 643                         break;
 644                 }
 645         }
 646
 647         /* If reported active speed is QDR, check if is FDR-10 */
 648         if (props->active_speed == IB_SPEED_QDR) {
 649                 init_query_mad(in_mad);
 650                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 651                 in_mad->attr_mod = cpu_to_be32(port);
 652
 653                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 654                                    NULL, NULL, in_mad, out_mad);
 655                 if (err)
 656                         goto out;
 657
 658                 /* Checking LinkSpeedActive for FDR-10 */
 659                 if (out_mad->data[15] & 0x1)
 660                         props->active_speed = IB_SPEED_FDR10;
 661         }
 662
 663         /* Avoid wrong speed value returned by FW if the IB link is down. */
 664         if (props->state == IB_PORT_DOWN)
 665                  props->active_speed = IB_SPEED_SDR;
 666
 667 out:
 668         kfree(in_mad);
 669         kfree(out_mad);
 670         return err;
 671 }
 672
 673 static u8 state_to_phys_state(enum ib_port_state state)
 674 {
 675         return state == IB_PORT_ACTIVE ? 5 : 3;
 676 }
 677
 678 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 679                                struct ib_port_attr *props, int netw_view)
 680 {
 681
 682         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 683         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 684         struct net_device *ndev;
 685         enum ib_mtu tmp;
 686         struct mlx4_cmd_mailbox *mailbox;
 687         int err = 0;
 688         int is_bonded = mlx4_is_bonded(mdev->dev);
 689
 690         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 691         if (IS_ERR(mailbox))
 692                 return PTR_ERR(mailbox);
 693
 694         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 695                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 696                            MLX4_CMD_WRAPPED);
 697         if (err)
 698                 goto out;
 699
 700         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 701                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 702         props->active_speed     = IB_SPEED_QDR;
 703         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
 704         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 705         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 706         props->pkey_tbl_len     = 1;
 707         props->max_mtu          = IB_MTU_4096;
 708         props->max_vl_num       = 2;
 709         props->state            = IB_PORT_DOWN;
 710         props->phys_state       = state_to_phys_state(props->state);
 711         props->active_mtu       = IB_MTU_256;
 712         spin_lock_bh(&iboe->lock);
 713         ndev = iboe->netdevs[port - 1];
 714         if (ndev && is_bonded) {
 715                 rcu_read_lock(); /* required to get upper dev */
 716                 ndev = netdev_master_upper_dev_get_rcu(ndev);
 717                 rcu_read_unlock();
 718         }
 719         if (!ndev)
 720                 goto out_unlock;
 721
 722         tmp = iboe_get_mtu(ndev->mtu);
 723         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 724
 725         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 726                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 727         props->phys_state       = state_to_phys_state(props->state);
 728 out_unlock:
 729         spin_unlock_bh(&iboe->lock);
 730 out:
 731         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 732         return err;
 733 }
 734
 735 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 736                          struct ib_port_attr *props, int netw_view)
 737 {
 738         int err;
 739
 740         memset(props, 0, sizeof *props);
 741
 742         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 743                 ib_link_query_port(ibdev, port, props, netw_view) :
 744                                 eth_link_query_port(ibdev, port, props, netw_view);
 745
 746         return err;
 747 }
 748
 749 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 750                               struct ib_port_attr *props)
 751 {
 752         /* returns host view */
 753         return __mlx4_ib_query_port(ibdev, port, props, 0);
 754 }
 755
 756 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 757                         union ib_gid *gid, int netw_view)
 758 {
 759         struct ib_smp *in_mad  = NULL;
 760         struct ib_smp *out_mad = NULL;
 761         int err = -ENOMEM;
 762         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 763         int clear = 0;
 764         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 765
 766         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 767         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 768         if (!in_mad || !out_mad)
 769                 goto out;
 770
 771         init_query_mad(in_mad);
 772         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 773         in_mad->attr_mod = cpu_to_be32(port);
 774
 775         if (mlx4_is_mfunc(dev->dev) && netw_view)
 776                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 777
 778         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 779         if (err)
 780                 goto out;
 781
 782         memcpy(gid->raw, out_mad->data + 8, 8);
 783
 784         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 785                 if (index) {
 786                         /* For any index > 0, return the null guid */
 787                         err = 0;
 788                         clear = 1;
 789                         goto out;
 790                 }
 791         }
 792
 793         init_query_mad(in_mad);
 794         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 795         in_mad->attr_mod = cpu_to_be32(index / 8);
 796
 797         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 798                            NULL, NULL, in_mad, out_mad);
 799         if (err)
 800                 goto out;
 801
 802         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 803
 804 out:
 805         if (clear)
 806                 memset(gid->raw + 8, 0, 8);
 807         kfree(in_mad);
 808         kfree(out_mad);
 809         return err;
 810 }
 811
 812 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 813                              union ib_gid *gid)
 814 {
 815         int ret;
 816
 817         if (rdma_protocol_ib(ibdev, port))
 818                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 819
 820         if (!rdma_protocol_roce(ibdev, port))
 821                 return -ENODEV;
 822
 823         if (!rdma_cap_roce_gid_table(ibdev, port))
 824                 return -ENODEV;
 825
 826         ret = ib_get_cached_gid(ibdev, port, index, gid, NULL);
 827         if (ret == -EAGAIN) {
 828                 memcpy(gid, &zgid, sizeof(*gid));
 829                 return 0;
 830         }
 831
 832         return ret;
 833 }
 834
 835 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 836                          u16 *pkey, int netw_view)
 837 {
 838         struct ib_smp *in_mad  = NULL;
 839         struct ib_smp *out_mad = NULL;
 840         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 841         int err = -ENOMEM;
 842
 843         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 844         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 845         if (!in_mad || !out_mad)
 846                 goto out;
 847
 848         init_query_mad(in_mad);
 849         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 850         in_mad->attr_mod = cpu_to_be32(index / 32);
 851
 852         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 853                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 854
 855         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 856                            in_mad, out_mad);
 857         if (err)
 858                 goto out;
 859
 860         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 861
 862 out:
 863         kfree(in_mad);
 864         kfree(out_mad);
 865         return err;
 866 }
 867
 868 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 869 {
 870         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 871 }
 872
 873 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 874                                  struct ib_device_modify *props)
 875 {
 876         struct mlx4_cmd_mailbox *mailbox;
 877         unsigned long flags;
 878
 879         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 880                 return -EOPNOTSUPP;
 881
 882         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 883                 return 0;
 884
 885         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 886                 return -EOPNOTSUPP;
 887
 888         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 889         memcpy(ibdev->node_desc, props->node_desc, 64);
 890         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 891
 892         /*
 893          * If possible, pass node desc to FW, so it can generate
 894          * a 144 trap.  If cmd fails, just ignore.
 895          */
 896         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 897         if (IS_ERR(mailbox))
 898                 return 0;
 899
 900         memcpy(mailbox->buf, props->node_desc, 64);
 901         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 902                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 903
 904         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 905
 906         return 0;
 907 }
 908
 909 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 910                             u32 cap_mask)
 911 {
 912         struct mlx4_cmd_mailbox *mailbox;
 913         int err;
 914
 915         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 916         if (IS_ERR(mailbox))
 917                 return PTR_ERR(mailbox);
 918
 919         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 920                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 921                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 922         } else {
 923                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 924                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 925         }
 926
 927         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
 928                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 929                        MLX4_CMD_WRAPPED);
 930
 931         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 932         return err;
 933 }
 934
 935 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 936                                struct ib_port_modify *props)
 937 {
 938         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 939         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 940         struct ib_port_attr attr;
 941         u32 cap_mask;
 942         int err;
 943
 944         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
 945          * of whether port link layer is ETH or IB. For ETH ports, qkey
 946          * violations and port capabilities are not meaningful.
 947          */
 948         if (is_eth)
 949                 return 0;
 950
 951         mutex_lock(&mdev->cap_mask_mutex);
 952
 953         err = mlx4_ib_query_port(ibdev, port, &attr);
 954         if (err)
 955                 goto out;
 956
 957         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 958                 ~props->clr_port_cap_mask;
 959
 960         err = mlx4_ib_SET_PORT(mdev, port,
 961                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
 962                                cap_mask);
 963
 964 out:
 965         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 966         return err;
 967 }
 968
 969 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 970                                                   struct ib_udata *udata)
 971 {
 972         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 973         struct mlx4_ib_ucontext *context;
 974         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 975         struct mlx4_ib_alloc_ucontext_resp resp;
 976         int err;
 977
 978         if (!dev->ib_active)
 979                 return ERR_PTR(-EAGAIN);
 980
 981         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
 982                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
 983                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
 984                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 985         } else {
 986                 resp.dev_caps         = dev->dev->caps.userspace_caps;
 987                 resp.qp_tab_size      = dev->dev->caps.num_qps;
 988                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 989                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 990                 resp.cqe_size         = dev->dev->caps.cqe_size;
 991         }
 992
 993         context = kzalloc(sizeof(*context), GFP_KERNEL);
 994         if (!context)
 995                 return ERR_PTR(-ENOMEM);
 996
 997         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 998         if (err) {
 999                 kfree(context);
1000                 return ERR_PTR(err);
1001         }
1002
1003         INIT_LIST_HEAD(&context->db_page_list);
1004         mutex_init(&context->db_page_mutex);
1005
1006         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
1007                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
1008         else
1009                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
1010
1011         if (err) {
1012                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
1013                 kfree(context);
1014                 return ERR_PTR(-EFAULT);
1015         }
1016
1017         return &context->ibucontext;
1018 }
1019
1020 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
1021 {
1022         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1023
1024         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
1025         kfree(context);
1026
1027         return 0;
1028 }
1029
1030 static void  mlx4_ib_vma_open(struct vm_area_struct *area)
1031 {
1032         /* vma_open is called when a new VMA is created on top of our VMA.
1033          * This is done through either mremap flow or split_vma (usually due
1034          * to mlock, madvise, munmap, etc.). We do not support a clone of the
1035          * vma, as this VMA is strongly hardware related. Therefore we set the
1036          * vm_ops of the newly created/cloned VMA to NULL, to prevent it from
1037          * calling us again and trying to do incorrect actions. We assume that
1038          * the original vma size is exactly a single page that there will be no
1039          * "splitting" operations on.
1040          */
1041         area->vm_ops = NULL;
1042 }
1043
1044 static void  mlx4_ib_vma_close(struct vm_area_struct *area)
1045 {
1046         struct mlx4_ib_vma_private_data *mlx4_ib_vma_priv_data;
1047
1048         /* It's guaranteed that all VMAs opened on a FD are closed before the
1049          * file itself is closed, therefore no sync is needed with the regular
1050          * closing flow. (e.g. mlx4_ib_dealloc_ucontext) However need a sync
1051          * with accessing the vma as part of mlx4_ib_disassociate_ucontext.
1052          * The close operation is usually called under mm->mmap_sem except when
1053          * process is exiting.  The exiting case is handled explicitly as part
1054          * of mlx4_ib_disassociate_ucontext.
1055          */
1056         mlx4_ib_vma_priv_data = (struct mlx4_ib_vma_private_data *)
1057                                 area->vm_private_data;
1058
1059         /* set the vma context pointer to null in the mlx4_ib driver's private
1060          * data to protect against a race condition in mlx4_ib_dissassociate_ucontext().
1061          */
1062         mlx4_ib_vma_priv_data->vma = NULL;
1063 }
1064
1065 static const struct vm_operations_struct mlx4_ib_vm_ops = {
1066         .open = mlx4_ib_vma_open,
1067         .close = mlx4_ib_vma_close
1068 };
1069
1070 static void mlx4_ib_disassociate_ucontext(struct ib_ucontext *ibcontext)
1071 {
1072         int i;
1073         int ret = 0;
1074         struct vm_area_struct *vma;
1075         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
1076         struct task_struct *owning_process  = NULL;
1077         struct mm_struct   *owning_mm       = NULL;
1078
1079         owning_process = get_pid_task(ibcontext->tgid, PIDTYPE_PID);
1080         if (!owning_process)
1081                 return;
1082
1083         owning_mm = get_task_mm(owning_process);
1084         if (!owning_mm) {
1085                 pr_info("no mm, disassociate ucontext is pending task termination\n");
1086                 while (1) {
1087                         /* make sure that task is dead before returning, it may
1088                          * prevent a rare case of module down in parallel to a
1089                          * call to mlx4_ib_vma_close.
1090                          */
1091                         put_task_struct(owning_process);
1092                         msleep(1);
1093                         owning_process = get_pid_task(ibcontext->tgid,
1094                                                       PIDTYPE_PID);
1095                         if (!owning_process ||
1096                             owning_process->state == TASK_DEAD) {
1097                                 pr_info("disassociate ucontext done, task was terminated\n");
1098                                 /* in case task was dead need to release the task struct */
1099                                 if (owning_process)
1100                                         put_task_struct(owning_process);
1101                                 return;
1102                         }
1103                 }
1104         }
1105
1106         /* need to protect from a race on closing the vma as part of
1107          * mlx4_ib_vma_close().
1108          */
1109         down_read(&owning_mm->mmap_sem);
1110         for (i = 0; i < HW_BAR_COUNT; i++) {
1111                 vma = context->hw_bar_info[i].vma;
1112                 if (!vma)
1113                         continue;
1114
1115                 ret = zap_vma_ptes(context->hw_bar_info[i].vma,
1116                                    context->hw_bar_info[i].vma->vm_start,
1117                                    PAGE_SIZE);
1118                 if (ret) {
1119                         pr_err("Error: zap_vma_ptes failed for index=%d, ret=%d\n", i, ret);
1120                         BUG_ON(1);
1121                 }
1122
1123                 /* context going to be destroyed, should not access ops any more */
1124                 context->hw_bar_info[i].vma->vm_ops = NULL;
1125         }
1126
1127         up_read(&owning_mm->mmap_sem);
1128         mmput(owning_mm);
1129         put_task_struct(owning_process);
1130 }
1131
1132 static void mlx4_ib_set_vma_data(struct vm_area_struct *vma,
1133                                  struct mlx4_ib_vma_private_data *vma_private_data)
1134 {
1135         vma_private_data->vma = vma;
1136         vma->vm_private_data = vma_private_data;
1137         vma->vm_ops =  &mlx4_ib_vm_ops;
1138 }
1139
1140 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
1141 {
1142         struct mlx4_ib_dev *dev = to_mdev(context->device);
1143         struct mlx4_ib_ucontext *mucontext = to_mucontext(context);
1144
1145         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
1146                 return -EINVAL;
1147
1148         if (vma->vm_pgoff == 0) {
1149                 /* We prevent double mmaping on same context */
1150                 if (mucontext->hw_bar_info[HW_BAR_DB].vma)
1151                         return -EINVAL;
1152
1153                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1154
1155                 if (io_remap_pfn_range(vma, vma->vm_start,
1156                                        to_mucontext(context)->uar.pfn,
1157                                        PAGE_SIZE, vma->vm_page_prot))
1158                         return -EAGAIN;
1159
1160                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_DB]);
1161
1162         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
1163                 /* We prevent double mmaping on same context */
1164                 if (mucontext->hw_bar_info[HW_BAR_BF].vma)
1165                         return -EINVAL;
1166
1167                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
1168
1169                 if (io_remap_pfn_range(vma, vma->vm_start,
1170                                        to_mucontext(context)->uar.pfn +
1171                                        dev->dev->caps.num_uars,
1172                                        PAGE_SIZE, vma->vm_page_prot))
1173                         return -EAGAIN;
1174
1175                 mlx4_ib_set_vma_data(vma, &mucontext->hw_bar_info[HW_BAR_BF]);
1176
1177         } else if (vma->vm_pgoff == 3) {
1178                 struct mlx4_clock_params params;
1179                 int ret;
1180
1181                 /* We prevent double mmaping on same context */
1182                 if (mucontext->hw_bar_info[HW_BAR_CLOCK].vma)
1183                         return -EINVAL;
1184
1185                 ret = mlx4_get_internal_clock_params(dev->dev, &params);
1186
1187                 if (ret)
1188                         return ret;
1189
1190                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
1191                 if (io_remap_pfn_range(vma, vma->vm_start,
1192                                        (pci_resource_start(dev->dev->persist->pdev,
1193                                                            params.bar) +
1194                                         params.offset)
1195                                        >> PAGE_SHIFT,
1196                                        PAGE_SIZE, vma->vm_page_prot))
1197                         return -EAGAIN;
1198
1199                 mlx4_ib_set_vma_data(vma,
1200                                      &mucontext->hw_bar_info[HW_BAR_CLOCK]);
1201         } else {
1202                 return -EINVAL;
1203         }
1204
1205         return 0;
1206 }
1207
1208 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
1209                                       struct ib_ucontext *context,
1210                                       struct ib_udata *udata)
1211 {
1212         struct mlx4_ib_pd *pd;
1213         int err;
1214
1215         pd = kmalloc(sizeof *pd, GFP_KERNEL);
1216         if (!pd)
1217                 return ERR_PTR(-ENOMEM);
1218
1219         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
1220         if (err) {
1221                 kfree(pd);
1222                 return ERR_PTR(err);
1223         }
1224
1225         if (context)
1226                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
1227                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
1228                         kfree(pd);
1229                         return ERR_PTR(-EFAULT);
1230                 }
1231
1232         return &pd->ibpd;
1233 }
1234
1235 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
1236 {
1237         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
1238         kfree(pd);
1239
1240         return 0;
1241 }
1242
1243 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
1244                                           struct ib_ucontext *context,
1245                                           struct ib_udata *udata)
1246 {
1247         struct mlx4_ib_xrcd *xrcd;
1248         struct ib_cq_init_attr cq_attr = {};
1249         int err;
1250
1251         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
1252                 return ERR_PTR(-ENOSYS);
1253
1254         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
1255         if (!xrcd)
1256                 return ERR_PTR(-ENOMEM);
1257
1258         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
1259         if (err)
1260                 goto err1;
1261
1262         xrcd->pd = ib_alloc_pd(ibdev);
1263         if (IS_ERR(xrcd->pd)) {
1264                 err = PTR_ERR(xrcd->pd);
1265                 goto err2;
1266         }
1267
1268         cq_attr.cqe = 1;
1269         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
1270         if (IS_ERR(xrcd->cq)) {
1271                 err = PTR_ERR(xrcd->cq);
1272                 goto err3;
1273         }
1274
1275         return &xrcd->ibxrcd;
1276
1277 err3:
1278         ib_dealloc_pd(xrcd->pd);
1279 err2:
1280         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
1281 err1:
1282         kfree(xrcd);
1283         return ERR_PTR(err);
1284 }
1285
1286 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1287 {
1288         ib_destroy_cq(to_mxrcd(xrcd)->cq);
1289         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
1290         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
1291         kfree(xrcd);
1292
1293         return 0;
1294 }
1295
1296 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
1297 {
1298         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1299         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1300         struct mlx4_ib_gid_entry *ge;
1301
1302         ge = kzalloc(sizeof *ge, GFP_KERNEL);
1303         if (!ge)
1304                 return -ENOMEM;
1305
1306         ge->gid = *gid;
1307         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
1308                 ge->port = mqp->port;
1309                 ge->added = 1;
1310         }
1311
1312         mutex_lock(&mqp->mutex);
1313         list_add_tail(&ge->list, &mqp->gid_list);
1314         mutex_unlock(&mqp->mutex);
1315
1316         return 0;
1317 }
1318
1319 static void mlx4_ib_delete_counters_table(struct mlx4_ib_dev *ibdev,
1320                                           struct mlx4_ib_counters *ctr_table)
1321 {
1322         struct counter_index *counter, *tmp_count;
1323
1324         mutex_lock(&ctr_table->mutex);
1325         list_for_each_entry_safe(counter, tmp_count, &ctr_table->counters_list,
1326                                  list) {
1327                 if (counter->allocated)
1328                         mlx4_counter_free(ibdev->dev, counter->index);
1329                 list_del(&counter->list);
1330                 kfree(counter);
1331         }
1332         mutex_unlock(&ctr_table->mutex);
1333 }
1334
1335 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
1336                    union ib_gid *gid)
1337 {
1338         struct net_device *ndev;
1339         int ret = 0;
1340
1341         if (!mqp->port)
1342                 return 0;
1343
1344         spin_lock_bh(&mdev->iboe.lock);
1345         ndev = mdev->iboe.netdevs[mqp->port - 1];
1346         if (ndev)
1347                 dev_hold(ndev);
1348         spin_unlock_bh(&mdev->iboe.lock);
1349
1350         if (ndev) {
1351                 ret = 1;
1352                 dev_put(ndev);
1353         }
1354
1355         return ret;
1356 }
1357
1358 struct mlx4_ib_steering {
1359         struct list_head list;
1360         struct mlx4_flow_reg_id reg_id;
1361         union ib_gid gid;
1362 };
1363
1364 static int parse_flow_attr(struct mlx4_dev *dev,
1365                            u32 qp_num,
1366                            union ib_flow_spec *ib_spec,
1367                            struct _rule_hw *mlx4_spec)
1368 {
1369         enum mlx4_net_trans_rule_id type;
1370
1371         switch (ib_spec->type) {
1372         case IB_FLOW_SPEC_ETH:
1373                 type = MLX4_NET_TRANS_RULE_ID_ETH;
1374                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
1375                        ETH_ALEN);
1376                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
1377                        ETH_ALEN);
1378                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
1379                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
1380                 break;
1381         case IB_FLOW_SPEC_IB:
1382                 type = MLX4_NET_TRANS_RULE_ID_IB;
1383                 mlx4_spec->ib.l3_qpn =
1384                         cpu_to_be32(qp_num);
1385                 mlx4_spec->ib.qpn_mask =
1386                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
1387                 break;
1388
1389
1390         case IB_FLOW_SPEC_IPV4:
1391                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
1392                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
1393                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
1394                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
1395                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
1396                 break;
1397
1398         case IB_FLOW_SPEC_TCP:
1399         case IB_FLOW_SPEC_UDP:
1400                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
1401                                         MLX4_NET_TRANS_RULE_ID_TCP :
1402                                         MLX4_NET_TRANS_RULE_ID_UDP;
1403                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
1404                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
1405                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
1406                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
1407                 break;
1408
1409         default:
1410                 return -EINVAL;
1411         }
1412         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
1413             mlx4_hw_rule_sz(dev, type) < 0)
1414                 return -EINVAL;
1415         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
1416         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
1417         return mlx4_hw_rule_sz(dev, type);
1418 }
1419
1420 struct default_rules {
1421         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1422         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
1423         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
1424         __u8  link_layer;
1425 };
1426 static const struct default_rules default_table[] = {
1427         {
1428                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
1429                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
1430                 .rules_create_list = {IB_FLOW_SPEC_IB},
1431                 .link_layer = IB_LINK_LAYER_INFINIBAND
1432         }
1433 };
1434
1435 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
1436                                          struct ib_flow_attr *flow_attr)
1437 {
1438         int i, j, k;
1439         void *ib_flow;
1440         const struct default_rules *pdefault_rules = default_table;
1441         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
1442
1443         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
1444                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
1445                 memset(&field_types, 0, sizeof(field_types));
1446
1447                 if (link_layer != pdefault_rules->link_layer)
1448                         continue;
1449
1450                 ib_flow = flow_attr + 1;
1451                 /* we assume the specs are sorted */
1452                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
1453                      j < flow_attr->num_of_specs; k++) {
1454                         union ib_flow_spec *current_flow =
1455                                 (union ib_flow_spec *)ib_flow;
1456
1457                         /* same layer but different type */
1458                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
1459                              (pdefault_rules->mandatory_fields[k] &
1460                               IB_FLOW_SPEC_LAYER_MASK)) &&
1461                             (current_flow->type !=
1462                              pdefault_rules->mandatory_fields[k]))
1463                                 goto out;
1464
1465                         /* same layer, try match next one */
1466                         if (current_flow->type ==
1467                             pdefault_rules->mandatory_fields[k]) {
1468                                 j++;
1469                                 ib_flow +=
1470                                         ((union ib_flow_spec *)ib_flow)->size;
1471                         }
1472                 }
1473
1474                 ib_flow = flow_attr + 1;
1475                 for (j = 0; j < flow_attr->num_of_specs;
1476                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
1477                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
1478                                 /* same layer and same type */
1479                                 if (((union ib_flow_spec *)ib_flow)->type ==
1480                                     pdefault_rules->mandatory_not_fields[k])
1481                                         goto out;
1482
1483                 return i;
1484         }
1485 out:
1486         return -1;
1487 }
1488
1489 static int __mlx4_ib_create_default_rules(
1490                 struct mlx4_ib_dev *mdev,
1491                 struct ib_qp *qp,
1492                 const struct default_rules *pdefault_rules,
1493                 struct _rule_hw *mlx4_spec) {
1494         int size = 0;
1495         int i;
1496
1497         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
1498                 int ret;
1499                 union ib_flow_spec ib_spec;
1500                 switch (pdefault_rules->rules_create_list[i]) {
1501                 case 0:
1502                         /* no rule */
1503                         continue;
1504                 case IB_FLOW_SPEC_IB:
1505                         ib_spec.type = IB_FLOW_SPEC_IB;
1506                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1507
1508                         break;
1509                 default:
1510                         /* invalid rule */
1511                         return -EINVAL;
1512                 }
1513                 /* We must put empty rule, qpn is being ignored */
1514                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1515                                       mlx4_spec);
1516                 if (ret < 0) {
1517                         pr_info("invalid parsing\n");
1518                         return -EINVAL;
1519                 }
1520
1521                 mlx4_spec = (void *)mlx4_spec + ret;
1522                 size += ret;
1523         }
1524         return size;
1525 }
1526
1527 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1528                           int domain,
1529                           enum mlx4_net_trans_promisc_mode flow_type,
1530                           u64 *reg_id)
1531 {
1532         int ret, i;
1533         int size = 0;
1534         void *ib_flow;
1535         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1536         struct mlx4_cmd_mailbox *mailbox;
1537         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1538         int default_flow;
1539
1540         static const u16 __mlx4_domain[] = {
1541                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1542                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1543                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1544                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1545         };
1546
1547         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1548                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1549                 return -EINVAL;
1550         }
1551
1552         if (domain >= IB_FLOW_DOMAIN_NUM) {
1553                 pr_err("Invalid domain value %d\n", domain);
1554                 return -EINVAL;
1555         }
1556
1557         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1558                 return -EINVAL;
1559
1560         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1561         if (IS_ERR(mailbox))
1562                 return PTR_ERR(mailbox);
1563         ctrl = mailbox->buf;
1564
1565         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1566                                  flow_attr->priority);
1567         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1568         ctrl->port = flow_attr->port;
1569         ctrl->qpn = cpu_to_be32(qp->qp_num);
1570
1571         ib_flow = flow_attr + 1;
1572         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1573         /* Add default flows */
1574         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1575         if (default_flow >= 0) {
1576                 ret = __mlx4_ib_create_default_rules(
1577                                 mdev, qp, default_table + default_flow,
1578                                 mailbox->buf + size);
1579                 if (ret < 0) {
1580                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1581                         return -EINVAL;
1582                 }
1583                 size += ret;
1584         }
1585         for (i = 0; i < flow_attr->num_of_specs; i++) {
1586                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1587                                       mailbox->buf + size);
1588                 if (ret < 0) {
1589                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1590                         return -EINVAL;
1591                 }
1592                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1593                 size += ret;
1594         }
1595
1596         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1597                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1598                            MLX4_CMD_WRAPPED);
1599         if (ret == -ENOMEM)
1600                 pr_err("mcg table is full. Fail to register network rule.\n");
1601         else if (ret == -ENXIO)
1602                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1603         else if (ret)
1604                 pr_err("Invalid argument. Fail to register network rule.\n");
1605
1606         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1607         return ret;
1608 }
1609
1610 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1611 {
1612         int err;
1613         err = mlx4_cmd(dev, reg_id, 0, 0,
1614                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1615                        MLX4_CMD_WRAPPED);
1616         if (err)
1617                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1618                        reg_id);
1619         return err;
1620 }
1621
1622 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1623                                     u64 *reg_id)
1624 {
1625         void *ib_flow;
1626         union ib_flow_spec *ib_spec;
1627         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1628         int err = 0;
1629
1630         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1631             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1632                 return 0; /* do nothing */
1633
1634         ib_flow = flow_attr + 1;
1635         ib_spec = (union ib_flow_spec *)ib_flow;
1636
1637         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1638                 return 0; /* do nothing */
1639
1640         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1641                                     flow_attr->port, qp->qp_num,
1642                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1643                                     reg_id);
1644         return err;
1645 }
1646
1647 static int mlx4_ib_add_dont_trap_rule(struct mlx4_dev *dev,
1648                                       struct ib_flow_attr *flow_attr,
1649                                       enum mlx4_net_trans_promisc_mode *type)
1650 {
1651         int err = 0;
1652
1653         if (!(dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_UC_MC_SNIFFER) ||
1654             (dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC) ||
1655             (flow_attr->num_of_specs > 1) || (flow_attr->priority != 0)) {
1656                 return -EOPNOTSUPP;
1657         }
1658
1659         if (flow_attr->num_of_specs == 0) {
1660                 type[0] = MLX4_FS_MC_SNIFFER;
1661                 type[1] = MLX4_FS_UC_SNIFFER;
1662         } else {
1663                 union ib_flow_spec *ib_spec;
1664
1665                 ib_spec = (union ib_flow_spec *)(flow_attr + 1);
1666                 if (ib_spec->type !=  IB_FLOW_SPEC_ETH)
1667                         return -EINVAL;
1668
1669                 /* if all is zero than MC and UC */
1670                 if (is_zero_ether_addr(ib_spec->eth.mask.dst_mac)) {
1671                         type[0] = MLX4_FS_MC_SNIFFER;
1672                         type[1] = MLX4_FS_UC_SNIFFER;
1673                 } else {
1674                         u8 mac[ETH_ALEN] = {ib_spec->eth.mask.dst_mac[0] ^ 0x01,
1675                                             ib_spec->eth.mask.dst_mac[1],
1676                                             ib_spec->eth.mask.dst_mac[2],
1677                                             ib_spec->eth.mask.dst_mac[3],
1678                                             ib_spec->eth.mask.dst_mac[4],
1679                                             ib_spec->eth.mask.dst_mac[5]};
1680
1681                         /* Above xor was only on MC bit, non empty mask is valid
1682                          * only if this bit is set and rest are zero.
1683                          */
1684                         if (!is_zero_ether_addr(&mac[0]))
1685                                 return -EINVAL;
1686
1687                         if (is_multicast_ether_addr(ib_spec->eth.val.dst_mac))
1688                                 type[0] = MLX4_FS_MC_SNIFFER;
1689                         else
1690                                 type[0] = MLX4_FS_UC_SNIFFER;
1691                 }
1692         }
1693
1694         return err;
1695 }
1696
1697 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1698                                     struct ib_flow_attr *flow_attr,
1699                                     int domain)
1700 {
1701         int err = 0, i = 0, j = 0;
1702         struct mlx4_ib_flow *mflow;
1703         enum mlx4_net_trans_promisc_mode type[2];
1704         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1705         int is_bonded = mlx4_is_bonded(dev);
1706
1707         if (flow_attr->port < 1 || flow_attr->port > qp->device->phys_port_cnt)
1708                 return ERR_PTR(-EINVAL);
1709
1710         if ((flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP) &&
1711             (flow_attr->type != IB_FLOW_ATTR_NORMAL))
1712                 return ERR_PTR(-EOPNOTSUPP);
1713
1714         memset(type, 0, sizeof(type));
1715
1716         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1717         if (!mflow) {
1718                 err = -ENOMEM;
1719                 goto err_free;
1720         }
1721
1722         switch (flow_attr->type) {
1723         case IB_FLOW_ATTR_NORMAL:
1724                 /* If dont trap flag (continue match) is set, under specific
1725                  * condition traffic be replicated to given qp,
1726                  * without stealing it
1727                  */
1728                 if (unlikely(flow_attr->flags & IB_FLOW_ATTR_FLAGS_DONT_TRAP)) {
1729                         err = mlx4_ib_add_dont_trap_rule(dev,
1730                                                          flow_attr,
1731                                                          type);
1732                         if (err)
1733                                 goto err_free;
1734                 } else {
1735                         type[0] = MLX4_FS_REGULAR;
1736                 }
1737                 break;
1738
1739         case IB_FLOW_ATTR_ALL_DEFAULT:
1740                 type[0] = MLX4_FS_ALL_DEFAULT;
1741                 break;
1742
1743         case IB_FLOW_ATTR_MC_DEFAULT:
1744                 type[0] = MLX4_FS_MC_DEFAULT;
1745                 break;
1746
1747         case IB_FLOW_ATTR_SNIFFER:
1748                 type[0] = MLX4_FS_MIRROR_RX_PORT;
1749                 type[1] = MLX4_FS_MIRROR_SX_PORT;
1750                 break;
1751
1752         default:
1753                 err = -EINVAL;
1754                 goto err_free;
1755         }
1756
1757         while (i < ARRAY_SIZE(type) && type[i]) {
1758                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1759                                             &mflow->reg_id[i].id);
1760                 if (err)
1761                         goto err_create_flow;
1762                 if (is_bonded) {
1763                         /* Application always sees one port so the mirror rule
1764                          * must be on port #2
1765                          */
1766                         flow_attr->port = 2;
1767                         err = __mlx4_ib_create_flow(qp, flow_attr,
1768                                                     domain, type[j],
1769                                                     &mflow->reg_id[j].mirror);
1770                         flow_attr->port = 1;
1771                         if (err)
1772                                 goto err_create_flow;
1773                         j++;
1774                 }
1775
1776                 i++;
1777         }
1778
1779         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1780                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1781                                                &mflow->reg_id[i].id);
1782                 if (err)
1783                         goto err_create_flow;
1784
1785                 if (is_bonded) {
1786                         flow_attr->port = 2;
1787                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1788                                                        &mflow->reg_id[j].mirror);
1789                         flow_attr->port = 1;
1790                         if (err)
1791                                 goto err_create_flow;
1792                         j++;
1793                 }
1794                 /* function to create mirror rule */
1795                 i++;
1796         }
1797
1798         return &mflow->ibflow;
1799
1800 err_create_flow:
1801         while (i) {
1802                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1803                                              mflow->reg_id[i].id);
1804                 i--;
1805         }
1806
1807         while (j) {
1808                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1809                                              mflow->reg_id[j].mirror);
1810                 j--;
1811         }
1812 err_free:
1813         kfree(mflow);
1814         return ERR_PTR(err);
1815 }
1816
1817 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1818 {
1819         int err, ret = 0;
1820         int i = 0;
1821         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1822         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1823
1824         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1825                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1826                 if (err)
1827                         ret = err;
1828                 if (mflow->reg_id[i].mirror) {
1829                         err = __mlx4_ib_destroy_flow(mdev->dev,
1830                                                      mflow->reg_id[i].mirror);
1831                         if (err)
1832                                 ret = err;
1833                 }
1834                 i++;
1835         }
1836
1837         kfree(mflow);
1838         return ret;
1839 }
1840
1841 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1842 {
1843         int err;
1844         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1845         struct mlx4_dev *dev = mdev->dev;
1846         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1847         struct mlx4_ib_steering *ib_steering = NULL;
1848         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1849         struct mlx4_flow_reg_id reg_id;
1850
1851         if (mdev->dev->caps.steering_mode ==
1852             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1853                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1854                 if (!ib_steering)
1855                         return -ENOMEM;
1856         }
1857
1858         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1859                                     !!(mqp->flags &
1860                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1861                                     prot, &reg_id.id);
1862         if (err) {
1863                 pr_err("multicast attach op failed, err %d\n", err);
1864                 goto err_malloc;
1865         }
1866
1867         reg_id.mirror = 0;
1868         if (mlx4_is_bonded(dev)) {
1869                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1870                                             (mqp->port == 1) ? 2 : 1,
1871                                             !!(mqp->flags &
1872                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1873                                             prot, &reg_id.mirror);
1874                 if (err)
1875                         goto err_add;
1876         }
1877
1878         err = add_gid_entry(ibqp, gid);
1879         if (err)
1880                 goto err_add;
1881
1882         if (ib_steering) {
1883                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1884                 ib_steering->reg_id = reg_id;
1885                 mutex_lock(&mqp->mutex);
1886                 list_add(&ib_steering->list, &mqp->steering_rules);
1887                 mutex_unlock(&mqp->mutex);
1888         }
1889         return 0;
1890
1891 err_add:
1892         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1893                               prot, reg_id.id);
1894         if (reg_id.mirror)
1895                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1896                                       prot, reg_id.mirror);
1897 err_malloc:
1898         kfree(ib_steering);
1899
1900         return err;
1901 }
1902
1903 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1904 {
1905         struct mlx4_ib_gid_entry *ge;
1906         struct mlx4_ib_gid_entry *tmp;
1907         struct mlx4_ib_gid_entry *ret = NULL;
1908
1909         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1910                 if (!memcmp(raw, ge->gid.raw, 16)) {
1911                         ret = ge;
1912                         break;
1913                 }
1914         }
1915
1916         return ret;
1917 }
1918
1919 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1920 {
1921         int err;
1922         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1923         struct mlx4_dev *dev = mdev->dev;
1924         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1925         struct net_device *ndev;
1926         struct mlx4_ib_gid_entry *ge;
1927         struct mlx4_flow_reg_id reg_id = {0, 0};
1928         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1929
1930         if (mdev->dev->caps.steering_mode ==
1931             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1932                 struct mlx4_ib_steering *ib_steering;
1933
1934                 mutex_lock(&mqp->mutex);
1935                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1936                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1937                                 list_del(&ib_steering->list);
1938                                 break;
1939                         }
1940                 }
1941                 mutex_unlock(&mqp->mutex);
1942                 if (&ib_steering->list == &mqp->steering_rules) {
1943                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1944                         return -EINVAL;
1945                 }
1946                 reg_id = ib_steering->reg_id;
1947                 kfree(ib_steering);
1948         }
1949
1950         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1951                                     prot, reg_id.id);
1952         if (err)
1953                 return err;
1954
1955         if (mlx4_is_bonded(dev)) {
1956                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1957                                             prot, reg_id.mirror);
1958                 if (err)
1959                         return err;
1960         }
1961
1962         mutex_lock(&mqp->mutex);
1963         ge = find_gid_entry(mqp, gid->raw);
1964         if (ge) {
1965                 spin_lock_bh(&mdev->iboe.lock);
1966                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1967                 if (ndev)
1968                         dev_hold(ndev);
1969                 spin_unlock_bh(&mdev->iboe.lock);
1970                 if (ndev)
1971                         dev_put(ndev);
1972                 list_del(&ge->list);
1973                 kfree(ge);
1974         } else
1975                 pr_warn("could not find mgid entry\n");
1976
1977         mutex_unlock(&mqp->mutex);
1978
1979         return 0;
1980 }
1981
1982 static int init_node_data(struct mlx4_ib_dev *dev)
1983 {
1984         struct ib_smp *in_mad  = NULL;
1985         struct ib_smp *out_mad = NULL;
1986         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1987         int err = -ENOMEM;
1988
1989         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1990         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1991         if (!in_mad || !out_mad)
1992                 goto out;
1993
1994         init_query_mad(in_mad);
1995         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1996         if (mlx4_is_master(dev->dev))
1997                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1998
1999         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2000         if (err)
2001                 goto out;
2002
2003         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
2004
2005         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
2006
2007         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
2008         if (err)
2009                 goto out;
2010
2011         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
2012         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
2013
2014 out:
2015         kfree(in_mad);
2016         kfree(out_mad);
2017         return err;
2018 }
2019
2020 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
2021                         char *buf)
2022 {
2023         struct mlx4_ib_dev *dev =
2024                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2025         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
2026 }
2027
2028 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
2029                         char *buf)
2030 {
2031         struct mlx4_ib_dev *dev =
2032                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2033         return sprintf(buf, "%x\n", dev->dev->rev_id);
2034 }
2035
2036 static ssize_t show_board(struct device *device, struct device_attribute *attr,
2037                           char *buf)
2038 {
2039         struct mlx4_ib_dev *dev =
2040                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
2041         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
2042                        dev->dev->board_id);
2043 }
2044
2045 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
2046 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
2047 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
2048
2049 static struct device_attribute *mlx4_class_attributes[] = {
2050         &dev_attr_hw_rev,
2051         &dev_attr_hca_type,
2052         &dev_attr_board_id
2053 };
2054
2055 struct diag_counter {
2056         const char *name;
2057         u32 offset;
2058 };
2059
2060 #define DIAG_COUNTER(_name, _offset)                    \
2061         { .name = #_name, .offset = _offset }
2062
2063 static const struct diag_counter diag_basic[] = {
2064         DIAG_COUNTER(rq_num_lle, 0x00),
2065         DIAG_COUNTER(sq_num_lle, 0x04),
2066         DIAG_COUNTER(rq_num_lqpoe, 0x08),
2067         DIAG_COUNTER(sq_num_lqpoe, 0x0C),
2068         DIAG_COUNTER(rq_num_lpe, 0x18),
2069         DIAG_COUNTER(sq_num_lpe, 0x1C),
2070         DIAG_COUNTER(rq_num_wrfe, 0x20),
2071         DIAG_COUNTER(sq_num_wrfe, 0x24),
2072         DIAG_COUNTER(sq_num_mwbe, 0x2C),
2073         DIAG_COUNTER(sq_num_bre, 0x34),
2074         DIAG_COUNTER(sq_num_rire, 0x44),
2075         DIAG_COUNTER(rq_num_rire, 0x48),
2076         DIAG_COUNTER(sq_num_rae, 0x4C),
2077         DIAG_COUNTER(rq_num_rae, 0x50),
2078         DIAG_COUNTER(sq_num_roe, 0x54),
2079         DIAG_COUNTER(sq_num_tree, 0x5C),
2080         DIAG_COUNTER(sq_num_rree, 0x64),
2081         DIAG_COUNTER(rq_num_rnr, 0x68),
2082         DIAG_COUNTER(sq_num_rnr, 0x6C),
2083         DIAG_COUNTER(rq_num_oos, 0x100),
2084         DIAG_COUNTER(sq_num_oos, 0x104),
2085 };
2086
2087 static const struct diag_counter diag_ext[] = {
2088         DIAG_COUNTER(rq_num_dup, 0x130),
2089         DIAG_COUNTER(sq_num_to, 0x134),
2090 };
2091
2092 static const struct diag_counter diag_device_only[] = {
2093         DIAG_COUNTER(num_cqovf, 0x1A0),
2094         DIAG_COUNTER(rq_num_udsdprd, 0x118),
2095 };
2096
2097 static struct rdma_hw_stats *mlx4_ib_alloc_hw_stats(struct ib_device *ibdev,
2098                                                     u8 port_num)
2099 {
2100         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2101         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2102
2103         if (!diag[!!port_num].name)
2104                 return NULL;
2105
2106         return rdma_alloc_hw_stats_struct(diag[!!port_num].name,
2107                                           diag[!!port_num].num_counters,
2108                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
2109 }
2110
2111 static int mlx4_ib_get_hw_stats(struct ib_device *ibdev,
2112                                 struct rdma_hw_stats *stats,
2113                                 u8 port, int index)
2114 {
2115         struct mlx4_ib_dev *dev = to_mdev(ibdev);
2116         struct mlx4_ib_diag_counters *diag = dev->diag_counters;
2117         u32 hw_value[ARRAY_SIZE(diag_device_only) +
2118                 ARRAY_SIZE(diag_ext) + ARRAY_SIZE(diag_basic)] = {};
2119         int ret;
2120         int i;
2121
2122         ret = mlx4_query_diag_counters(dev->dev,
2123                                        MLX4_OP_MOD_QUERY_TRANSPORT_CI_ERRORS,
2124                                        diag[!!port].offset, hw_value,
2125                                        diag[!!port].num_counters, port);
2126
2127         if (ret)
2128                 return ret;
2129
2130         for (i = 0; i < diag[!!port].num_counters; i++)
2131                 stats->value[i] = hw_value[i];
2132
2133         return diag[!!port].num_counters;
2134 }
2135
2136 static int __mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev,
2137                                          const char ***name,
2138                                          u32 **offset,
2139                                          u32 *num,
2140                                          bool port)
2141 {
2142         u32 num_counters;
2143
2144         num_counters = ARRAY_SIZE(diag_basic);
2145
2146         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT)
2147                 num_counters += ARRAY_SIZE(diag_ext);
2148
2149         if (!port)
2150                 num_counters += ARRAY_SIZE(diag_device_only);
2151
2152         *name = kcalloc(num_counters, sizeof(**name), GFP_KERNEL);
2153         if (!*name)
2154                 return -ENOMEM;
2155
2156         *offset = kcalloc(num_counters, sizeof(**offset), GFP_KERNEL);
2157         if (!*offset)
2158                 goto err_name;
2159
2160         *num = num_counters;
2161
2162         return 0;
2163
2164 err_name:
2165         kfree(*name);
2166         return -ENOMEM;
2167 }
2168
2169 static void mlx4_ib_fill_diag_counters(struct mlx4_ib_dev *ibdev,
2170                                        const char **name,
2171                                        u32 *offset,
2172                                        bool port)
2173 {
2174         int i;
2175         int j;
2176
2177         for (i = 0, j = 0; i < ARRAY_SIZE(diag_basic); i++, j++) {
2178                 name[i] = diag_basic[i].name;
2179                 offset[i] = diag_basic[i].offset;
2180         }
2181
2182         if (ibdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT) {
2183                 for (i = 0; i < ARRAY_SIZE(diag_ext); i++, j++) {
2184                         name[j] = diag_ext[i].name;
2185                         offset[j] = diag_ext[i].offset;
2186                 }
2187         }
2188
2189         if (!port) {
2190                 for (i = 0; i < ARRAY_SIZE(diag_device_only); i++, j++) {
2191                         name[j] = diag_device_only[i].name;
2192                         offset[j] = diag_device_only[i].offset;
2193                 }
2194         }
2195 }
2196
2197 static int mlx4_ib_alloc_diag_counters(struct mlx4_ib_dev *ibdev)
2198 {
2199         struct mlx4_ib_diag_counters *diag = ibdev->diag_counters;
2200         int i;
2201         int ret;
2202         bool per_port = !!(ibdev->dev->caps.flags2 &
2203                 MLX4_DEV_CAP_FLAG2_DIAG_PER_PORT);
2204
2205         if (mlx4_is_slave(ibdev->dev))
2206                 return 0;
2207
2208         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2209                 /* i == 1 means we are building port counters */
2210                 if (i && !per_port)
2211                         continue;
2212
2213                 ret = __mlx4_ib_alloc_diag_counters(ibdev, &diag[i].name,
2214                                                     &diag[i].offset,
2215                                                     &diag[i].num_counters, i);
2216                 if (ret)
2217                         goto err_alloc;
2218
2219                 mlx4_ib_fill_diag_counters(ibdev, diag[i].name,
2220                                            diag[i].offset, i);
2221         }
2222
2223         ibdev->ib_dev.get_hw_stats      = mlx4_ib_get_hw_stats;
2224         ibdev->ib_dev.alloc_hw_stats    = mlx4_ib_alloc_hw_stats;
2225
2226         return 0;
2227
2228 err_alloc:
2229         if (i) {
2230                 kfree(diag[i - 1].name);
2231                 kfree(diag[i - 1].offset);
2232         }
2233
2234         return ret;
2235 }
2236
2237 static void mlx4_ib_diag_cleanup(struct mlx4_ib_dev *ibdev)
2238 {
2239         int i;
2240
2241         for (i = 0; i < MLX4_DIAG_COUNTERS_TYPES; i++) {
2242                 kfree(ibdev->diag_counters[i].offset);
2243                 kfree(ibdev->diag_counters[i].name);
2244         }
2245 }
2246
2247 #define MLX4_IB_INVALID_MAC     ((u64)-1)
2248 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
2249                                struct net_device *dev,
2250                                int port)
2251 {
2252         u64 new_smac = 0;
2253         u64 release_mac = MLX4_IB_INVALID_MAC;
2254         struct mlx4_ib_qp *qp;
2255
2256         read_lock(&dev_base_lock);
2257         new_smac = mlx4_mac_to_u64(dev->dev_addr);
2258         read_unlock(&dev_base_lock);
2259
2260         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
2261
2262         /* no need for update QP1 and mac registration in non-SRIOV */
2263         if (!mlx4_is_mfunc(ibdev->dev))
2264                 return;
2265
2266         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
2267         qp = ibdev->qp1_proxy[port - 1];
2268         if (qp) {
2269                 int new_smac_index;
2270                 u64 old_smac;
2271                 struct mlx4_update_qp_params update_params;
2272
2273                 mutex_lock(&qp->mutex);
2274                 old_smac = qp->pri.smac;
2275                 if (new_smac == old_smac)
2276                         goto unlock;
2277
2278                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
2279
2280                 if (new_smac_index < 0)
2281                         goto unlock;
2282
2283                 update_params.smac_index = new_smac_index;
2284                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
2285                                    &update_params)) {
2286                         release_mac = new_smac;
2287                         goto unlock;
2288                 }
2289                 /* if old port was zero, no mac was yet registered for this QP */
2290                 if (qp->pri.smac_port)
2291                         release_mac = old_smac;
2292                 qp->pri.smac = new_smac;
2293                 qp->pri.smac_port = port;
2294                 qp->pri.smac_index = new_smac_index;
2295         }
2296
2297 unlock:
2298         if (release_mac != MLX4_IB_INVALID_MAC)
2299                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
2300         if (qp)
2301                 mutex_unlock(&qp->mutex);
2302         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
2303 }
2304
2305 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
2306                                  struct net_device *dev,
2307                                  unsigned long event)
2308
2309 {
2310         struct mlx4_ib_iboe *iboe;
2311         int update_qps_port = -1;
2312         int port;
2313
2314         ASSERT_RTNL();
2315
2316         iboe = &ibdev->iboe;
2317
2318         spin_lock_bh(&iboe->lock);
2319         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
2320
2321                 iboe->netdevs[port - 1] =
2322                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
2323
2324                 if (dev == iboe->netdevs[port - 1] &&
2325                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
2326                      event == NETDEV_UP || event == NETDEV_CHANGE))
2327                         update_qps_port = port;
2328
2329         }
2330         spin_unlock_bh(&iboe->lock);
2331
2332         if (update_qps_port > 0)
2333                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
2334 }
2335
2336 static int mlx4_ib_netdev_event(struct notifier_block *this,
2337                                 unsigned long event, void *ptr)
2338 {
2339         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2340         struct mlx4_ib_dev *ibdev;
2341
2342         if (!net_eq(dev_net(dev), &init_net))
2343                 return NOTIFY_DONE;
2344
2345         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2346         mlx4_ib_scan_netdevs(ibdev, dev, event);
2347
2348         return NOTIFY_DONE;
2349 }
2350
2351 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2352 {
2353         int port;
2354         int slave;
2355         int i;
2356
2357         if (mlx4_is_master(ibdev->dev)) {
2358                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2359                      ++slave) {
2360                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2361                                 for (i = 0;
2362                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2363                                      ++i) {
2364                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2365                                         /* master has the identity virt2phys pkey mapping */
2366                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2367                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2368                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2369                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2370                                 }
2371                         }
2372                 }
2373                 /* initialize pkey cache */
2374                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2375                         for (i = 0;
2376                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2377                              ++i)
2378                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2379                                         (i) ? 0 : 0xFFFF;
2380                 }
2381         }
2382 }
2383
2384 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2385 {
2386         int i, j, eq = 0, total_eqs = 0;
2387
2388         ibdev->eq_table = kcalloc(dev->caps.num_comp_vectors,
2389                                   sizeof(ibdev->eq_table[0]), GFP_KERNEL);
2390         if (!ibdev->eq_table)
2391                 return;
2392
2393         for (i = 1; i <= dev->caps.num_ports; i++) {
2394                 for (j = 0; j < mlx4_get_eqs_per_port(dev, i);
2395                      j++, total_eqs++) {
2396                         if (i > 1 &&  mlx4_is_eq_shared(dev, total_eqs))
2397                                 continue;
2398                         ibdev->eq_table[eq] = total_eqs;
2399                         if (!mlx4_assign_eq(dev, i,
2400                                             &ibdev->eq_table[eq]))
2401                                 eq++;
2402                         else
2403                                 ibdev->eq_table[eq] = -1;
2404                 }
2405         }
2406
2407         for (i = eq; i < dev->caps.num_comp_vectors;
2408              ibdev->eq_table[i++] = -1)
2409                 ;
2410
2411         /* Advertise the new number of EQs to clients */
2412         ibdev->ib_dev.num_comp_vectors = eq;
2413 }
2414
2415 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2416 {
2417         int i;
2418         int total_eqs = ibdev->ib_dev.num_comp_vectors;
2419
2420         /* no eqs were allocated */
2421         if (!ibdev->eq_table)
2422                 return;
2423
2424         /* Reset the advertised EQ number */
2425         ibdev->ib_dev.num_comp_vectors = 0;
2426
2427         for (i = 0; i < total_eqs; i++)
2428                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2429
2430         kfree(ibdev->eq_table);
2431         ibdev->eq_table = NULL;
2432 }
2433
2434 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2435                                struct ib_port_immutable *immutable)
2436 {
2437         struct ib_port_attr attr;
2438         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
2439         int err;
2440
2441         err = mlx4_ib_query_port(ibdev, port_num, &attr);
2442         if (err)
2443                 return err;
2444
2445         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2446         immutable->gid_tbl_len = attr.gid_tbl_len;
2447
2448         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND) {
2449                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2450         } else {
2451                 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE)
2452                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2453                 if (mdev->dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2)
2454                         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE |
2455                                 RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
2456         }
2457
2458         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
2459
2460         return 0;
2461 }
2462
2463 static void get_fw_ver_str(struct ib_device *device, char *str,
2464                            size_t str_len)
2465 {
2466         struct mlx4_ib_dev *dev =
2467                 container_of(device, struct mlx4_ib_dev, ib_dev);
2468         snprintf(str, str_len, "%d.%d.%d",
2469                  (int) (dev->dev->caps.fw_ver >> 32),
2470                  (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
2471                  (int) dev->dev->caps.fw_ver & 0xffff);
2472 }
2473
2474 static void *mlx4_ib_add(struct mlx4_dev *dev)
2475 {
2476         struct mlx4_ib_dev *ibdev;
2477         int num_ports = 0;
2478         int i, j;
2479         int err;
2480         struct mlx4_ib_iboe *iboe;
2481         int ib_num_ports = 0;
2482         int num_req_counters;
2483         int allocated;
2484         u32 counter_index;
2485         struct counter_index *new_counter_index = NULL;
2486
2487         pr_info_once("%s", mlx4_ib_version);
2488
2489         num_ports = 0;
2490         mlx4_foreach_ib_transport_port(i, dev)
2491                 num_ports++;
2492
2493         /* No point in registering a device with no ports... */
2494         if (num_ports == 0)
2495                 return NULL;
2496
2497         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2498         if (!ibdev) {
2499                 dev_err(&dev->persist->pdev->dev,
2500                         "Device struct alloc failed\n");
2501                 return NULL;
2502         }
2503
2504         iboe = &ibdev->iboe;
2505
2506         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2507                 goto err_dealloc;
2508
2509         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2510                 goto err_pd;
2511
2512         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2513                                  PAGE_SIZE);
2514         if (!ibdev->uar_map)
2515                 goto err_uar;
2516         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2517
2518         ibdev->dev = dev;
2519         ibdev->bond_next_port   = 0;
2520
2521         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2522         ibdev->ib_dev.owner             = THIS_MODULE;
2523         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2524         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2525         ibdev->num_ports                = num_ports;
2526         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2527                                                 1 : ibdev->num_ports;
2528         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2529         ibdev->ib_dev.dma_device        = &dev->persist->pdev->dev;
2530         ibdev->ib_dev.get_netdev        = mlx4_ib_get_netdev;
2531         ibdev->ib_dev.add_gid           = mlx4_ib_add_gid;
2532         ibdev->ib_dev.del_gid           = mlx4_ib_del_gid;
2533
2534         if (dev->caps.userspace_caps)
2535                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2536         else
2537                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2538
2539         ibdev->ib_dev.uverbs_cmd_mask   =
2540                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2541                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2542                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2543                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2544                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2545                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2546                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2547                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2548                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2549                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2550                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2551                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2552                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2553                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2554                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2555                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2556                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2557                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2558                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2559                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2560                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2561                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2562                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2563                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2564
2565         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2566         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2567         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2568         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2569         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2570         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2571         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2572         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2573         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2574         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2575         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2576         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2577         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2578         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2579         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2580         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2581         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2582         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2583         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2584         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2585         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2586         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2587         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2588         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2589         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2590         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2591         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2592         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2593         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2594         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2595         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2596         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2597         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2598         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2599         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2600         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2601         ibdev->ib_dev.alloc_mr          = mlx4_ib_alloc_mr;
2602         ibdev->ib_dev.map_mr_sg         = mlx4_ib_map_mr_sg;
2603         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2604         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2605         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2606         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2607         ibdev->ib_dev.get_dev_fw_str    = get_fw_ver_str;
2608         ibdev->ib_dev.disassociate_ucontext = mlx4_ib_disassociate_ucontext;
2609
2610         if (!mlx4_is_slave(ibdev->dev)) {
2611                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2612                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2613                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2614                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2615         }
2616
2617         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2618             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2619                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2620                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2621
2622                 ibdev->ib_dev.uverbs_cmd_mask |=
2623                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2624                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2625         }
2626
2627         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2628                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2629                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2630                 ibdev->ib_dev.uverbs_cmd_mask |=
2631                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2632                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2633         }
2634
2635         if (check_flow_steering_support(dev)) {
2636                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2637                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2638                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2639
2640                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2641                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2642                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2643         }
2644
2645         ibdev->ib_dev.uverbs_ex_cmd_mask |=
2646                 (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE) |
2647                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_CQ) |
2648                 (1ull << IB_USER_VERBS_EX_CMD_CREATE_QP);
2649
2650         mlx4_ib_alloc_eqs(dev, ibdev);
2651
2652         spin_lock_init(&iboe->lock);
2653
2654         if (init_node_data(ibdev))
2655                 goto err_map;
2656
2657         for (i = 0; i < ibdev->num_ports; ++i) {
2658                 mutex_init(&ibdev->counters_table[i].mutex);
2659                 INIT_LIST_HEAD(&ibdev->counters_table[i].counters_list);
2660         }
2661
2662         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2663         for (i = 0; i < num_req_counters; ++i) {
2664                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2665                 allocated = 0;
2666                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2667                                                 IB_LINK_LAYER_ETHERNET) {
2668                         err = mlx4_counter_alloc(ibdev->dev, &counter_index);
2669                         /* if failed to allocate a new counter, use default */
2670                         if (err)
2671                                 counter_index =
2672                                         mlx4_get_default_counter_index(dev,
2673                                                                        i + 1);
2674                         else
2675                                 allocated = 1;
2676                 } else { /* IB_LINK_LAYER_INFINIBAND use the default counter */
2677                         counter_index = mlx4_get_default_counter_index(dev,
2678                                                                        i + 1);
2679                 }
2680                 new_counter_index = kmalloc(sizeof(*new_counter_index),
2681                                             GFP_KERNEL);
2682                 if (!new_counter_index) {
2683                         if (allocated)
2684                                 mlx4_counter_free(ibdev->dev, counter_index);
2685                         goto err_counter;
2686                 }
2687                 new_counter_index->index = counter_index;
2688                 new_counter_index->allocated = allocated;
2689                 list_add_tail(&new_counter_index->list,
2690                               &ibdev->counters_table[i].counters_list);
2691                 ibdev->counters_table[i].default_counter = counter_index;
2692                 pr_info("counter index %d for port %d allocated %d\n",
2693                         counter_index, i + 1, allocated);
2694         }
2695         if (mlx4_is_bonded(dev))
2696                 for (i = 1; i < ibdev->num_ports ; ++i) {
2697                         new_counter_index =
2698                                         kmalloc(sizeof(struct counter_index),
2699                                                 GFP_KERNEL);
2700                         if (!new_counter_index)
2701                                 goto err_counter;
2702                         new_counter_index->index = counter_index;
2703                         new_counter_index->allocated = 0;
2704                         list_add_tail(&new_counter_index->list,
2705                                       &ibdev->counters_table[i].counters_list);
2706                         ibdev->counters_table[i].default_counter =
2707                                                                 counter_index;
2708                 }
2709
2710         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2711                 ib_num_ports++;
2712
2713         spin_lock_init(&ibdev->sm_lock);
2714         mutex_init(&ibdev->cap_mask_mutex);
2715         INIT_LIST_HEAD(&ibdev->qp_list);
2716         spin_lock_init(&ibdev->reset_flow_resource_lock);
2717
2718         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2719             ib_num_ports) {
2720                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2721                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2722                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2723                                             &ibdev->steer_qpn_base, 0);
2724                 if (err)
2725                         goto err_counter;
2726
2727                 ibdev->ib_uc_qpns_bitmap =
2728                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2729                                 sizeof(long),
2730                                 GFP_KERNEL);
2731                 if (!ibdev->ib_uc_qpns_bitmap) {
2732                         dev_err(&dev->persist->pdev->dev,
2733                                 "bit map alloc failed\n");
2734                         goto err_steer_qp_release;
2735                 }
2736
2737                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2738
2739                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2740                                 dev, ibdev->steer_qpn_base,
2741                                 ibdev->steer_qpn_base +
2742                                 ibdev->steer_qpn_count - 1);
2743                 if (err)
2744                         goto err_steer_free_bitmap;
2745         }
2746
2747         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2748                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2749
2750         if (mlx4_ib_alloc_diag_counters(ibdev))
2751                 goto err_steer_free_bitmap;
2752
2753         if (ib_register_device(&ibdev->ib_dev, NULL))
2754                 goto err_diag_counters;
2755
2756         if (mlx4_ib_mad_init(ibdev))
2757                 goto err_reg;
2758
2759         if (mlx4_ib_init_sriov(ibdev))
2760                 goto err_mad;
2761
2762         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE ||
2763             dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2764                 if (!iboe->nb.notifier_call) {
2765                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2766                         err = register_netdevice_notifier(&iboe->nb);
2767                         if (err) {
2768                                 iboe->nb.notifier_call = NULL;
2769                                 goto err_notif;
2770                         }
2771                 }
2772                 if (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_ROCE_V1_V2) {
2773                         err = mlx4_config_roce_v2_port(dev, ROCE_V2_UDP_DPORT);
2774                         if (err) {
2775                                 goto err_notif;
2776                         }
2777                 }
2778         }
2779
2780         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2781                 if (device_create_file(&ibdev->ib_dev.dev,
2782                                        mlx4_class_attributes[j]))
2783                         goto err_notif;
2784         }
2785
2786         ibdev->ib_active = true;
2787         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2788                 devlink_port_type_ib_set(mlx4_get_devlink_port(dev, i),
2789                                          &ibdev->ib_dev);
2790
2791         if (mlx4_is_mfunc(ibdev->dev))
2792                 init_pkeys(ibdev);
2793
2794         /* create paravirt contexts for any VFs which are active */
2795         if (mlx4_is_master(ibdev->dev)) {
2796                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2797                         if (j == mlx4_master_func_num(ibdev->dev))
2798                                 continue;
2799                         if (mlx4_is_slave_active(ibdev->dev, j))
2800                                 do_slave_init(ibdev, j, 1);
2801                 }
2802         }
2803         return ibdev;
2804
2805 err_notif:
2806         if (ibdev->iboe.nb.notifier_call) {
2807                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2808                         pr_warn("failure unregistering notifier\n");
2809                 ibdev->iboe.nb.notifier_call = NULL;
2810         }
2811         flush_workqueue(wq);
2812
2813         mlx4_ib_close_sriov(ibdev);
2814
2815 err_mad:
2816         mlx4_ib_mad_cleanup(ibdev);
2817
2818 err_reg:
2819         ib_unregister_device(&ibdev->ib_dev);
2820
2821 err_diag_counters:
2822         mlx4_ib_diag_cleanup(ibdev);
2823
2824 err_steer_free_bitmap:
2825         kfree(ibdev->ib_uc_qpns_bitmap);
2826
2827 err_steer_qp_release:
2828         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2829                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2830                                       ibdev->steer_qpn_count);
2831 err_counter:
2832         for (i = 0; i < ibdev->num_ports; ++i)
2833                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[i]);
2834
2835 err_map:
2836         iounmap(ibdev->uar_map);
2837
2838 err_uar:
2839         mlx4_uar_free(dev, &ibdev->priv_uar);
2840
2841 err_pd:
2842         mlx4_pd_free(dev, ibdev->priv_pdn);
2843
2844 err_dealloc:
2845         ib_dealloc_device(&ibdev->ib_dev);
2846
2847         return NULL;
2848 }
2849
2850 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2851 {
2852         int offset;
2853
2854         WARN_ON(!dev->ib_uc_qpns_bitmap);
2855
2856         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2857                                          dev->steer_qpn_count,
2858                                          get_count_order(count));
2859         if (offset < 0)
2860                 return offset;
2861
2862         *qpn = dev->steer_qpn_base + offset;
2863         return 0;
2864 }
2865
2866 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2867 {
2868         if (!qpn ||
2869             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2870                 return;
2871
2872         BUG_ON(qpn < dev->steer_qpn_base);
2873
2874         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2875                               qpn - dev->steer_qpn_base,
2876                               get_count_order(count));
2877 }
2878
2879 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2880                          int is_attach)
2881 {
2882         int err;
2883         size_t flow_size;
2884         struct ib_flow_attr *flow = NULL;
2885         struct ib_flow_spec_ib *ib_spec;
2886
2887         if (is_attach) {
2888                 flow_size = sizeof(struct ib_flow_attr) +
2889                             sizeof(struct ib_flow_spec_ib);
2890                 flow = kzalloc(flow_size, GFP_KERNEL);
2891                 if (!flow)
2892                         return -ENOMEM;
2893                 flow->port = mqp->port;
2894                 flow->num_of_specs = 1;
2895                 flow->size = flow_size;
2896                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2897                 ib_spec->type = IB_FLOW_SPEC_IB;
2898                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2899                 /* Add an empty rule for IB L2 */
2900                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2901
2902                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2903                                             IB_FLOW_DOMAIN_NIC,
2904                                             MLX4_FS_REGULAR,
2905                                             &mqp->reg_id);
2906         } else {
2907                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2908         }
2909         kfree(flow);
2910         return err;
2911 }
2912
2913 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2914 {
2915         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2916         int p;
2917         int i;
2918
2919         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2920                 devlink_port_type_clear(mlx4_get_devlink_port(dev, i));
2921         ibdev->ib_active = false;
2922         flush_workqueue(wq);
2923
2924         mlx4_ib_close_sriov(ibdev);
2925         mlx4_ib_mad_cleanup(ibdev);
2926         ib_unregister_device(&ibdev->ib_dev);
2927         mlx4_ib_diag_cleanup(ibdev);
2928         if (ibdev->iboe.nb.notifier_call) {
2929                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2930                         pr_warn("failure unregistering notifier\n");
2931                 ibdev->iboe.nb.notifier_call = NULL;
2932         }
2933
2934         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2935                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2936                                       ibdev->steer_qpn_count);
2937                 kfree(ibdev->ib_uc_qpns_bitmap);
2938         }
2939
2940         iounmap(ibdev->uar_map);
2941         for (p = 0; p < ibdev->num_ports; ++p)
2942                 mlx4_ib_delete_counters_table(ibdev, &ibdev->counters_table[p]);
2943
2944         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2945                 mlx4_CLOSE_PORT(dev, p);
2946
2947         mlx4_ib_free_eqs(dev, ibdev);
2948
2949         mlx4_uar_free(dev, &ibdev->priv_uar);
2950         mlx4_pd_free(dev, ibdev->priv_pdn);
2951         ib_dealloc_device(&ibdev->ib_dev);
2952 }
2953
2954 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2955 {
2956         struct mlx4_ib_demux_work **dm = NULL;
2957         struct mlx4_dev *dev = ibdev->dev;
2958         int i;
2959         unsigned long flags;
2960         struct mlx4_active_ports actv_ports;
2961         unsigned int ports;
2962         unsigned int first_port;
2963
2964         if (!mlx4_is_master(dev))
2965                 return;
2966
2967         actv_ports = mlx4_get_active_ports(dev, slave);
2968         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2969         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2970
2971         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2972         if (!dm) {
2973                 pr_err("failed to allocate memory for tunneling qp update\n");
2974                 return;
2975         }
2976
2977         for (i = 0; i < ports; i++) {
2978                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2979                 if (!dm[i]) {
2980                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2981                         while (--i >= 0)
2982                                 kfree(dm[i]);
2983                         goto out;
2984                 }
2985                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2986                 dm[i]->port = first_port + i + 1;
2987                 dm[i]->slave = slave;
2988                 dm[i]->do_init = do_init;
2989                 dm[i]->dev = ibdev;
2990         }
2991         /* initialize or tear down tunnel QPs for the slave */
2992         spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2993         if (!ibdev->sriov.is_going_down) {
2994                 for (i = 0; i < ports; i++)
2995                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2996                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2997         } else {
2998                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2999                 for (i = 0; i < ports; i++)
3000                         kfree(dm[i]);
3001         }
3002 out:
3003         kfree(dm);
3004         return;
3005 }
3006
3007 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
3008 {
3009         struct mlx4_ib_qp *mqp;
3010         unsigned long flags_qp;
3011         unsigned long flags_cq;
3012         struct mlx4_ib_cq *send_mcq, *recv_mcq;
3013         struct list_head    cq_notify_list;
3014         struct mlx4_cq *mcq;
3015         unsigned long flags;
3016
3017         pr_warn("mlx4_ib_handle_catas_error was started\n");
3018         INIT_LIST_HEAD(&cq_notify_list);
3019
3020         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
3021         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
3022
3023         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
3024                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
3025                 if (mqp->sq.tail != mqp->sq.head) {
3026                         send_mcq = to_mcq(mqp->ibqp.send_cq);
3027                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
3028                         if (send_mcq->mcq.comp &&
3029                             mqp->ibqp.send_cq->comp_handler) {
3030                                 if (!send_mcq->mcq.reset_notify_added) {
3031                                         send_mcq->mcq.reset_notify_added = 1;
3032                                         list_add_tail(&send_mcq->mcq.reset_notify,
3033                                                       &cq_notify_list);
3034                                 }
3035                         }
3036                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
3037                 }
3038                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
3039                 /* Now, handle the QP's receive queue */
3040                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
3041                 /* no handling is needed for SRQ */
3042                 if (!mqp->ibqp.srq) {
3043                         if (mqp->rq.tail != mqp->rq.head) {
3044                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
3045                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
3046                                 if (recv_mcq->mcq.comp &&
3047                                     mqp->ibqp.recv_cq->comp_handler) {
3048                                         if (!recv_mcq->mcq.reset_notify_added) {
3049                                                 recv_mcq->mcq.reset_notify_added = 1;
3050                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
3051                                                               &cq_notify_list);
3052                                         }
3053                                 }
3054                                 spin_unlock_irqrestore(&recv_mcq->lock,
3055                                                        flags_cq);
3056                         }
3057                 }
3058                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
3059         }
3060
3061         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
3062                 mcq->comp(mcq);
3063         }
3064         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
3065         pr_warn("mlx4_ib_handle_catas_error ended\n");
3066 }
3067
3068 static void handle_bonded_port_state_event(struct work_struct *work)
3069 {
3070         struct ib_event_work *ew =
3071                 container_of(work, struct ib_event_work, work);
3072         struct mlx4_ib_dev *ibdev = ew->ib_dev;
3073         enum ib_port_state bonded_port_state = IB_PORT_NOP;
3074         int i;
3075         struct ib_event ibev;
3076
3077         kfree(ew);
3078         spin_lock_bh(&ibdev->iboe.lock);
3079         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
3080                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
3081                 enum ib_port_state curr_port_state;
3082
3083                 if (!curr_netdev)
3084                         continue;
3085
3086                 curr_port_state =
3087                         (netif_running(curr_netdev) &&
3088                          netif_carrier_ok(curr_netdev)) ?
3089                         IB_PORT_ACTIVE : IB_PORT_DOWN;
3090
3091                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
3092                         curr_port_state : IB_PORT_ACTIVE;
3093         }
3094         spin_unlock_bh(&ibdev->iboe.lock);
3095
3096         ibev.device = &ibdev->ib_dev;
3097         ibev.element.port_num = 1;
3098         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
3099                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
3100
3101         ib_dispatch_event(&ibev);
3102 }
3103
3104 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
3105                           enum mlx4_dev_event event, unsigned long param)
3106 {
3107         struct ib_event ibev;
3108         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
3109         struct mlx4_eqe *eqe = NULL;
3110         struct ib_event_work *ew;
3111         int p = 0;
3112
3113         if (mlx4_is_bonded(dev) &&
3114             ((event == MLX4_DEV_EVENT_PORT_UP) ||
3115             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
3116                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
3117                 if (!ew)
3118                         return;
3119                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
3120                 ew->ib_dev = ibdev;
3121                 queue_work(wq, &ew->work);
3122                 return;
3123         }
3124
3125         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
3126                 eqe = (struct mlx4_eqe *)param;
3127         else
3128                 p = (int) param;
3129
3130         switch (event) {
3131         case MLX4_DEV_EVENT_PORT_UP:
3132                 if (p > ibdev->num_ports)
3133                         return;
3134                 if (mlx4_is_master(dev) &&
3135                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
3136                         IB_LINK_LAYER_INFINIBAND) {
3137                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
3138                 }
3139                 ibev.event = IB_EVENT_PORT_ACTIVE;
3140                 break;
3141
3142         case MLX4_DEV_EVENT_PORT_DOWN:
3143                 if (p > ibdev->num_ports)
3144                         return;
3145                 ibev.event = IB_EVENT_PORT_ERR;
3146                 break;
3147
3148         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
3149                 ibdev->ib_active = false;
3150                 ibev.event = IB_EVENT_DEVICE_FATAL;
3151                 mlx4_ib_handle_catas_error(ibdev);
3152                 break;
3153
3154         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
3155                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
3156                 if (!ew) {
3157                         pr_err("failed to allocate memory for events work\n");
3158                         break;
3159                 }
3160
3161                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
3162                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
3163                 ew->ib_dev = ibdev;
3164                 /* need to queue only for port owner, which uses GEN_EQE */
3165                 if (mlx4_is_master(dev))
3166                         queue_work(wq, &ew->work);
3167                 else
3168                         handle_port_mgmt_change_event(&ew->work);
3169                 return;
3170
3171         case MLX4_DEV_EVENT_SLAVE_INIT:
3172                 /* here, p is the slave id */
3173                 do_slave_init(ibdev, p, 1);
3174                 if (mlx4_is_master(dev)) {
3175                         int i;
3176
3177                         for (i = 1; i <= ibdev->num_ports; i++) {
3178                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3179                                         == IB_LINK_LAYER_INFINIBAND)
3180                                         mlx4_ib_slave_alias_guid_event(ibdev,
3181                                                                        p, i,
3182                                                                        1);
3183                         }
3184                 }
3185                 return;
3186
3187         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
3188                 if (mlx4_is_master(dev)) {
3189                         int i;
3190
3191                         for (i = 1; i <= ibdev->num_ports; i++) {
3192                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
3193                                         == IB_LINK_LAYER_INFINIBAND)
3194                                         mlx4_ib_slave_alias_guid_event(ibdev,
3195                                                                        p, i,
3196                                                                        0);
3197                         }
3198                 }
3199                 /* here, p is the slave id */
3200                 do_slave_init(ibdev, p, 0);
3201                 return;
3202
3203         default:
3204                 return;
3205         }
3206
3207         ibev.device           = ibdev_ptr;
3208         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
3209
3210         ib_dispatch_event(&ibev);
3211 }
3212
3213 static struct mlx4_interface mlx4_ib_interface = {
3214         .add            = mlx4_ib_add,
3215         .remove         = mlx4_ib_remove,
3216         .event          = mlx4_ib_event,
3217         .protocol       = MLX4_PROT_IB_IPV6,
3218         .flags          = MLX4_INTFF_BONDING
3219 };
3220
3221 static int __init mlx4_ib_init(void)
3222 {
3223         int err;
3224
3225         wq = create_singlethread_workqueue("mlx4_ib");
3226         if (!wq)
3227                 return -ENOMEM;
3228
3229         err = mlx4_ib_mcg_init();
3230         if (err)
3231                 goto clean_wq;
3232
3233         err = mlx4_register_interface(&mlx4_ib_interface);
3234         if (err)
3235                 goto clean_mcg;
3236
3237         return 0;
3238
3239 clean_mcg:
3240         mlx4_ib_mcg_destroy();
3241
3242 clean_wq:
3243         destroy_workqueue(wq);
3244         return err;
3245 }
3246
3247 static void __exit mlx4_ib_cleanup(void)
3248 {
3249         mlx4_unregister_interface(&mlx4_ib_interface);
3250         mlx4_ib_mcg_destroy();
3251         destroy_workqueue(wq);
3252 }
3253
3254 module_init(mlx4_ib_init);
3255 module_exit(mlx4_ib_cleanup);