drivers/infiniband/hw/mlx4/main.c

   1 /*
   2  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
   3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the
   9  * OpenIB.org BSD license below:
  10  *
  11  *     Redistribution and use in source and binary forms, with or
  12  *     without modification, are permitted provided that the following
  13  *     conditions are met:
  14  *
  15  *      - Redistributions of source code must retain the above
  16  *        copyright notice, this list of conditions and the following
  17  *        disclaimer.
  18  *
  19  *      - Redistributions in binary form must reproduce the above
  20  *        copyright notice, this list of conditions and the following
  21  *        disclaimer in the documentation and/or other materials
  22  *        provided with the distribution.
  23  *
  24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  31  * SOFTWARE.
  32  */
  33
  34 #include <linux/module.h>
  35 #include <linux/init.h>
  36 #include <linux/slab.h>
  37 #include <linux/errno.h>
  38 #include <linux/netdevice.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/rtnetlink.h>
  41 #include <linux/if_vlan.h>
  42 #include <net/ipv6.h>
  43 #include <net/addrconf.h>
  44
  45 #include <rdma/ib_smi.h>
  46 #include <rdma/ib_user_verbs.h>
  47 #include <rdma/ib_addr.h>
  48
  49 #include <linux/mlx4/driver.h>
  50 #include <linux/mlx4/cmd.h>
  51
  52 #include "mlx4_ib.h"
  53 #include "user.h"
  54
  55 #define DRV_NAME        MLX4_IB_DRV_NAME
  56 #define DRV_VERSION     "1.0"
  57 #define DRV_RELDATE     "April 4, 2008"
  58
  59 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  60
  61 MODULE_AUTHOR("Roland Dreier");
  62 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  63 MODULE_LICENSE("Dual BSD/GPL");
  64 MODULE_VERSION(DRV_VERSION);
  65
  66 int mlx4_ib_sm_guid_assign = 1;
  67 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  68 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
  69
  70 static const char mlx4_ib_version[] =
  71         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  72         DRV_VERSION " (" DRV_RELDATE ")\n";
  73
  74 struct update_gid_work {
  75         struct work_struct      work;
  76         union ib_gid            gids[128];
  77         struct mlx4_ib_dev     *dev;
  78         int                     port;
  79 };
  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 union ib_gid zgid;
  94
  95 static int check_flow_steering_support(struct mlx4_dev *dev)
  96 {
  97         int ib_num_ports = 0;
  98         int i;
  99
 100         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 101                 ib_num_ports++;
 102
 103         if (dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED) {
 104                 if (ib_num_ports || mlx4_is_mfunc(dev)) {
 105                         pr_warn("Device managed flow steering is unavailable "
 106                                 "for IB ports or in multifunction env.\n");
 107                         return 0;
 108                 }
 109                 return 1;
 110         }
 111         return 0;
 112 }
 113
 114 static int mlx4_ib_query_device(struct ib_device *ibdev,
 115                                 struct ib_device_attr *props)
 116 {
 117         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 118         struct ib_smp *in_mad  = NULL;
 119         struct ib_smp *out_mad = NULL;
 120         int err = -ENOMEM;
 121
 122         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 123         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 124         if (!in_mad || !out_mad)
 125                 goto out;
 126
 127         init_query_mad(in_mad);
 128         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 129
 130         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 131                            1, NULL, NULL, in_mad, out_mad);
 132         if (err)
 133                 goto out;
 134
 135         memset(props, 0, sizeof *props);
 136
 137         props->fw_ver = dev->dev->caps.fw_ver;
 138         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 139                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 140                 IB_DEVICE_SYS_IMAGE_GUID                |
 141                 IB_DEVICE_RC_RNR_NAK_GEN                |
 142                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 143         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 144                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 145         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 146                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 147         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM)
 148                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 149         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 150                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 151         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 152                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 153         if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
 154                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 155         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 156                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 157         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 158             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 159             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 160                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 161         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 162                 props->device_cap_flags |= IB_DEVICE_XRC;
 163         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 164                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 165         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 166                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 167                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 168                 else
 169                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 170         if (check_flow_steering_support(dev->dev))
 171                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 172         }
 173
 174         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 175                 0xffffff;
 176         props->vendor_part_id      = dev->dev->pdev->device;
 177         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 178         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 179
 180         props->max_mr_size         = ~0ull;
 181         props->page_size_cap       = dev->dev->caps.page_size_cap;
 182         props->max_qp              = dev->dev->quotas.qp;
 183         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 184         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 185                                          dev->dev->caps.max_rq_sg);
 186         props->max_cq              = dev->dev->quotas.cq;
 187         props->max_cqe             = dev->dev->caps.max_cqes;
 188         props->max_mr              = dev->dev->quotas.mpt;
 189         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 190         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 191         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 192         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 193         props->max_srq             = dev->dev->quotas.srq;
 194         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 195         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 196         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 197         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 198         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 199                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 200         props->masked_atomic_cap   = props->atomic_cap;
 201         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 202         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 203         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 204         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 205                                            props->max_mcast_grp;
 206         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 207
 208 out:
 209         kfree(in_mad);
 210         kfree(out_mad);
 211
 212         return err;
 213 }
 214
 215 static enum rdma_link_layer
 216 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 217 {
 218         struct mlx4_dev *dev = to_mdev(device)->dev;
 219
 220         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 221                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 222 }
 223
 224 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 225                               struct ib_port_attr *props, int netw_view)
 226 {
 227         struct ib_smp *in_mad  = NULL;
 228         struct ib_smp *out_mad = NULL;
 229         int ext_active_speed;
 230         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 231         int err = -ENOMEM;
 232
 233         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 234         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 235         if (!in_mad || !out_mad)
 236                 goto out;
 237
 238         init_query_mad(in_mad);
 239         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 240         in_mad->attr_mod = cpu_to_be32(port);
 241
 242         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 243                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 244
 245         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 246                                 in_mad, out_mad);
 247         if (err)
 248                 goto out;
 249
 250
 251         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 252         props->lmc              = out_mad->data[34] & 0x7;
 253         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 254         props->sm_sl            = out_mad->data[36] & 0xf;
 255         props->state            = out_mad->data[32] & 0xf;
 256         props->phys_state       = out_mad->data[33] >> 4;
 257         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 258         if (netw_view)
 259                 props->gid_tbl_len = out_mad->data[50];
 260         else
 261                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 262         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 263         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 264         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 265         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 266         props->active_width     = out_mad->data[31] & 0xf;
 267         props->active_speed     = out_mad->data[35] >> 4;
 268         props->max_mtu          = out_mad->data[41] & 0xf;
 269         props->active_mtu       = out_mad->data[36] >> 4;
 270         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 271         props->max_vl_num       = out_mad->data[37] >> 4;
 272         props->init_type_reply  = out_mad->data[41] >> 4;
 273
 274         /* Check if extended speeds (EDR/FDR/...) are supported */
 275         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 276                 ext_active_speed = out_mad->data[62] >> 4;
 277
 278                 switch (ext_active_speed) {
 279                 case 1:
 280                         props->active_speed = IB_SPEED_FDR;
 281                         break;
 282                 case 2:
 283                         props->active_speed = IB_SPEED_EDR;
 284                         break;
 285                 }
 286         }
 287
 288         /* If reported active speed is QDR, check if is FDR-10 */
 289         if (props->active_speed == IB_SPEED_QDR) {
 290                 init_query_mad(in_mad);
 291                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 292                 in_mad->attr_mod = cpu_to_be32(port);
 293
 294                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 295                                    NULL, NULL, in_mad, out_mad);
 296                 if (err)
 297                         goto out;
 298
 299                 /* Checking LinkSpeedActive for FDR-10 */
 300                 if (out_mad->data[15] & 0x1)
 301                         props->active_speed = IB_SPEED_FDR10;
 302         }
 303
 304         /* Avoid wrong speed value returned by FW if the IB link is down. */
 305         if (props->state == IB_PORT_DOWN)
 306                  props->active_speed = IB_SPEED_SDR;
 307
 308 out:
 309         kfree(in_mad);
 310         kfree(out_mad);
 311         return err;
 312 }
 313
 314 static u8 state_to_phys_state(enum ib_port_state state)
 315 {
 316         return state == IB_PORT_ACTIVE ? 5 : 3;
 317 }
 318
 319 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 320                                struct ib_port_attr *props, int netw_view)
 321 {
 322
 323         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 324         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 325         struct net_device *ndev;
 326         enum ib_mtu tmp;
 327         struct mlx4_cmd_mailbox *mailbox;
 328         int err = 0;
 329
 330         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 331         if (IS_ERR(mailbox))
 332                 return PTR_ERR(mailbox);
 333
 334         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 335                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 336                            MLX4_CMD_WRAPPED);
 337         if (err)
 338                 goto out;
 339
 340         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 341                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 342         props->active_speed     = IB_SPEED_QDR;
 343         props->port_cap_flags   = IB_PORT_CM_SUP;
 344         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 345         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 346         props->pkey_tbl_len     = 1;
 347         props->max_mtu          = IB_MTU_4096;
 348         props->max_vl_num       = 2;
 349         props->state            = IB_PORT_DOWN;
 350         props->phys_state       = state_to_phys_state(props->state);
 351         props->active_mtu       = IB_MTU_256;
 352         spin_lock(&iboe->lock);
 353         ndev = iboe->netdevs[port - 1];
 354         if (!ndev)
 355                 goto out_unlock;
 356
 357         tmp = iboe_get_mtu(ndev->mtu);
 358         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 359
 360         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 361                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 362         props->phys_state       = state_to_phys_state(props->state);
 363 out_unlock:
 364         spin_unlock(&iboe->lock);
 365 out:
 366         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 367         return err;
 368 }
 369
 370 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 371                          struct ib_port_attr *props, int netw_view)
 372 {
 373         int err;
 374
 375         memset(props, 0, sizeof *props);
 376
 377         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 378                 ib_link_query_port(ibdev, port, props, netw_view) :
 379                                 eth_link_query_port(ibdev, port, props, netw_view);
 380
 381         return err;
 382 }
 383
 384 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 385                               struct ib_port_attr *props)
 386 {
 387         /* returns host view */
 388         return __mlx4_ib_query_port(ibdev, port, props, 0);
 389 }
 390
 391 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 392                         union ib_gid *gid, int netw_view)
 393 {
 394         struct ib_smp *in_mad  = NULL;
 395         struct ib_smp *out_mad = NULL;
 396         int err = -ENOMEM;
 397         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 398         int clear = 0;
 399         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 400
 401         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 402         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 403         if (!in_mad || !out_mad)
 404                 goto out;
 405
 406         init_query_mad(in_mad);
 407         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 408         in_mad->attr_mod = cpu_to_be32(port);
 409
 410         if (mlx4_is_mfunc(dev->dev) && netw_view)
 411                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 412
 413         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 414         if (err)
 415                 goto out;
 416
 417         memcpy(gid->raw, out_mad->data + 8, 8);
 418
 419         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 420                 if (index) {
 421                         /* For any index > 0, return the null guid */
 422                         err = 0;
 423                         clear = 1;
 424                         goto out;
 425                 }
 426         }
 427
 428         init_query_mad(in_mad);
 429         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 430         in_mad->attr_mod = cpu_to_be32(index / 8);
 431
 432         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 433                            NULL, NULL, in_mad, out_mad);
 434         if (err)
 435                 goto out;
 436
 437         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 438
 439 out:
 440         if (clear)
 441                 memset(gid->raw + 8, 0, 8);
 442         kfree(in_mad);
 443         kfree(out_mad);
 444         return err;
 445 }
 446
 447 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
 448                           union ib_gid *gid)
 449 {
 450         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 451
 452         *gid = dev->iboe.gid_table[port - 1][index];
 453
 454         return 0;
 455 }
 456
 457 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 458                              union ib_gid *gid)
 459 {
 460         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
 461                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 462         else
 463                 return iboe_query_gid(ibdev, port, index, gid);
 464 }
 465
 466 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 467                          u16 *pkey, int netw_view)
 468 {
 469         struct ib_smp *in_mad  = NULL;
 470         struct ib_smp *out_mad = NULL;
 471         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 472         int err = -ENOMEM;
 473
 474         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 475         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 476         if (!in_mad || !out_mad)
 477                 goto out;
 478
 479         init_query_mad(in_mad);
 480         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 481         in_mad->attr_mod = cpu_to_be32(index / 32);
 482
 483         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 484                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 485
 486         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 487                            in_mad, out_mad);
 488         if (err)
 489                 goto out;
 490
 491         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 492
 493 out:
 494         kfree(in_mad);
 495         kfree(out_mad);
 496         return err;
 497 }
 498
 499 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 500 {
 501         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 502 }
 503
 504 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 505                                  struct ib_device_modify *props)
 506 {
 507         struct mlx4_cmd_mailbox *mailbox;
 508         unsigned long flags;
 509
 510         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 511                 return -EOPNOTSUPP;
 512
 513         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 514                 return 0;
 515
 516         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 517                 return -EOPNOTSUPP;
 518
 519         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 520         memcpy(ibdev->node_desc, props->node_desc, 64);
 521         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 522
 523         /*
 524          * If possible, pass node desc to FW, so it can generate
 525          * a 144 trap.  If cmd fails, just ignore.
 526          */
 527         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 528         if (IS_ERR(mailbox))
 529                 return 0;
 530
 531         memcpy(mailbox->buf, props->node_desc, 64);
 532         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 533                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 534
 535         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 536
 537         return 0;
 538 }
 539
 540 static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 541                          u32 cap_mask)
 542 {
 543         struct mlx4_cmd_mailbox *mailbox;
 544         int err;
 545         u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 546
 547         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 548         if (IS_ERR(mailbox))
 549                 return PTR_ERR(mailbox);
 550
 551         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 552                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 553                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 554         } else {
 555                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 556                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 557         }
 558
 559         err = mlx4_cmd(dev->dev, mailbox->dma, port, is_eth, MLX4_CMD_SET_PORT,
 560                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_NATIVE);
 561
 562         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 563         return err;
 564 }
 565
 566 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 567                                struct ib_port_modify *props)
 568 {
 569         struct ib_port_attr attr;
 570         u32 cap_mask;
 571         int err;
 572
 573         mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
 574
 575         err = mlx4_ib_query_port(ibdev, port, &attr);
 576         if (err)
 577                 goto out;
 578
 579         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 580                 ~props->clr_port_cap_mask;
 581
 582         err = mlx4_SET_PORT(to_mdev(ibdev), port,
 583                             !!(mask & IB_PORT_RESET_QKEY_CNTR),
 584                             cap_mask);
 585
 586 out:
 587         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 588         return err;
 589 }
 590
 591 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 592                                                   struct ib_udata *udata)
 593 {
 594         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 595         struct mlx4_ib_ucontext *context;
 596         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 597         struct mlx4_ib_alloc_ucontext_resp resp;
 598         int err;
 599
 600         if (!dev->ib_active)
 601                 return ERR_PTR(-EAGAIN);
 602
 603         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
 604                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
 605                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
 606                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 607         } else {
 608                 resp.dev_caps         = dev->dev->caps.userspace_caps;
 609                 resp.qp_tab_size      = dev->dev->caps.num_qps;
 610                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 611                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 612                 resp.cqe_size         = dev->dev->caps.cqe_size;
 613         }
 614
 615         context = kmalloc(sizeof *context, GFP_KERNEL);
 616         if (!context)
 617                 return ERR_PTR(-ENOMEM);
 618
 619         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 620         if (err) {
 621                 kfree(context);
 622                 return ERR_PTR(err);
 623         }
 624
 625         INIT_LIST_HEAD(&context->db_page_list);
 626         mutex_init(&context->db_page_mutex);
 627
 628         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
 629                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
 630         else
 631                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
 632
 633         if (err) {
 634                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 635                 kfree(context);
 636                 return ERR_PTR(-EFAULT);
 637         }
 638
 639         return &context->ibucontext;
 640 }
 641
 642 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 643 {
 644         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 645
 646         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 647         kfree(context);
 648
 649         return 0;
 650 }
 651
 652 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 653 {
 654         struct mlx4_ib_dev *dev = to_mdev(context->device);
 655
 656         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 657                 return -EINVAL;
 658
 659         if (vma->vm_pgoff == 0) {
 660                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 661
 662                 if (io_remap_pfn_range(vma, vma->vm_start,
 663                                        to_mucontext(context)->uar.pfn,
 664                                        PAGE_SIZE, vma->vm_page_prot))
 665                         return -EAGAIN;
 666         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 667                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 668
 669                 if (io_remap_pfn_range(vma, vma->vm_start,
 670                                        to_mucontext(context)->uar.pfn +
 671                                        dev->dev->caps.num_uars,
 672                                        PAGE_SIZE, vma->vm_page_prot))
 673                         return -EAGAIN;
 674         } else
 675                 return -EINVAL;
 676
 677         return 0;
 678 }
 679
 680 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 681                                       struct ib_ucontext *context,
 682                                       struct ib_udata *udata)
 683 {
 684         struct mlx4_ib_pd *pd;
 685         int err;
 686
 687         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 688         if (!pd)
 689                 return ERR_PTR(-ENOMEM);
 690
 691         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 692         if (err) {
 693                 kfree(pd);
 694                 return ERR_PTR(err);
 695         }
 696
 697         if (context)
 698                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 699                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 700                         kfree(pd);
 701                         return ERR_PTR(-EFAULT);
 702                 }
 703
 704         return &pd->ibpd;
 705 }
 706
 707 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 708 {
 709         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 710         kfree(pd);
 711
 712         return 0;
 713 }
 714
 715 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 716                                           struct ib_ucontext *context,
 717                                           struct ib_udata *udata)
 718 {
 719         struct mlx4_ib_xrcd *xrcd;
 720         int err;
 721
 722         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 723                 return ERR_PTR(-ENOSYS);
 724
 725         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 726         if (!xrcd)
 727                 return ERR_PTR(-ENOMEM);
 728
 729         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 730         if (err)
 731                 goto err1;
 732
 733         xrcd->pd = ib_alloc_pd(ibdev);
 734         if (IS_ERR(xrcd->pd)) {
 735                 err = PTR_ERR(xrcd->pd);
 736                 goto err2;
 737         }
 738
 739         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
 740         if (IS_ERR(xrcd->cq)) {
 741                 err = PTR_ERR(xrcd->cq);
 742                 goto err3;
 743         }
 744
 745         return &xrcd->ibxrcd;
 746
 747 err3:
 748         ib_dealloc_pd(xrcd->pd);
 749 err2:
 750         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 751 err1:
 752         kfree(xrcd);
 753         return ERR_PTR(err);
 754 }
 755
 756 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 757 {
 758         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 759         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 760         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 761         kfree(xrcd);
 762
 763         return 0;
 764 }
 765
 766 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 767 {
 768         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 769         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 770         struct mlx4_ib_gid_entry *ge;
 771
 772         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 773         if (!ge)
 774                 return -ENOMEM;
 775
 776         ge->gid = *gid;
 777         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 778                 ge->port = mqp->port;
 779                 ge->added = 1;
 780         }
 781
 782         mutex_lock(&mqp->mutex);
 783         list_add_tail(&ge->list, &mqp->gid_list);
 784         mutex_unlock(&mqp->mutex);
 785
 786         return 0;
 787 }
 788
 789 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 790                    union ib_gid *gid)
 791 {
 792         struct net_device *ndev;
 793         int ret = 0;
 794
 795         if (!mqp->port)
 796                 return 0;
 797
 798         spin_lock(&mdev->iboe.lock);
 799         ndev = mdev->iboe.netdevs[mqp->port - 1];
 800         if (ndev)
 801                 dev_hold(ndev);
 802         spin_unlock(&mdev->iboe.lock);
 803
 804         if (ndev) {
 805                 ret = 1;
 806                 dev_put(ndev);
 807         }
 808
 809         return ret;
 810 }
 811
 812 struct mlx4_ib_steering {
 813         struct list_head list;
 814         u64 reg_id;
 815         union ib_gid gid;
 816 };
 817
 818 static int parse_flow_attr(struct mlx4_dev *dev,
 819                            union ib_flow_spec *ib_spec,
 820                            struct _rule_hw *mlx4_spec)
 821 {
 822         enum mlx4_net_trans_rule_id type;
 823
 824         switch (ib_spec->type) {
 825         case IB_FLOW_SPEC_ETH:
 826                 type = MLX4_NET_TRANS_RULE_ID_ETH;
 827                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
 828                        ETH_ALEN);
 829                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
 830                        ETH_ALEN);
 831                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
 832                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
 833                 break;
 834
 835         case IB_FLOW_SPEC_IPV4:
 836                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
 837                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
 838                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
 839                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
 840                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
 841                 break;
 842
 843         case IB_FLOW_SPEC_TCP:
 844         case IB_FLOW_SPEC_UDP:
 845                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
 846                                         MLX4_NET_TRANS_RULE_ID_TCP :
 847                                         MLX4_NET_TRANS_RULE_ID_UDP;
 848                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
 849                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
 850                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
 851                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
 852                 break;
 853
 854         default:
 855                 return -EINVAL;
 856         }
 857         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
 858             mlx4_hw_rule_sz(dev, type) < 0)
 859                 return -EINVAL;
 860         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
 861         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
 862         return mlx4_hw_rule_sz(dev, type);
 863 }
 864
 865 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
 866                           int domain,
 867                           enum mlx4_net_trans_promisc_mode flow_type,
 868                           u64 *reg_id)
 869 {
 870         int ret, i;
 871         int size = 0;
 872         void *ib_flow;
 873         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
 874         struct mlx4_cmd_mailbox *mailbox;
 875         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
 876
 877         static const u16 __mlx4_domain[] = {
 878                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
 879                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
 880                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
 881                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
 882         };
 883
 884         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
 885                 pr_err("Invalid priority value %d\n", flow_attr->priority);
 886                 return -EINVAL;
 887         }
 888
 889         if (domain >= IB_FLOW_DOMAIN_NUM) {
 890                 pr_err("Invalid domain value %d\n", domain);
 891                 return -EINVAL;
 892         }
 893
 894         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
 895                 return -EINVAL;
 896
 897         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 898         if (IS_ERR(mailbox))
 899                 return PTR_ERR(mailbox);
 900         ctrl = mailbox->buf;
 901
 902         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
 903                                  flow_attr->priority);
 904         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
 905         ctrl->port = flow_attr->port;
 906         ctrl->qpn = cpu_to_be32(qp->qp_num);
 907
 908         ib_flow = flow_attr + 1;
 909         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
 910         for (i = 0; i < flow_attr->num_of_specs; i++) {
 911                 ret = parse_flow_attr(mdev->dev, ib_flow, mailbox->buf + size);
 912                 if (ret < 0) {
 913                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 914                         return -EINVAL;
 915                 }
 916                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
 917                 size += ret;
 918         }
 919
 920         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
 921                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
 922                            MLX4_CMD_NATIVE);
 923         if (ret == -ENOMEM)
 924                 pr_err("mcg table is full. Fail to register network rule.\n");
 925         else if (ret == -ENXIO)
 926                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
 927         else if (ret)
 928                 pr_err("Invalid argumant. Fail to register network rule.\n");
 929
 930         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 931         return ret;
 932 }
 933
 934 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
 935 {
 936         int err;
 937         err = mlx4_cmd(dev, reg_id, 0, 0,
 938                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
 939                        MLX4_CMD_NATIVE);
 940         if (err)
 941                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
 942                        reg_id);
 943         return err;
 944 }
 945
 946 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
 947                                     struct ib_flow_attr *flow_attr,
 948                                     int domain)
 949 {
 950         int err = 0, i = 0;
 951         struct mlx4_ib_flow *mflow;
 952         enum mlx4_net_trans_promisc_mode type[2];
 953
 954         memset(type, 0, sizeof(type));
 955
 956         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
 957         if (!mflow) {
 958                 err = -ENOMEM;
 959                 goto err_free;
 960         }
 961
 962         switch (flow_attr->type) {
 963         case IB_FLOW_ATTR_NORMAL:
 964                 type[0] = MLX4_FS_REGULAR;
 965                 break;
 966
 967         case IB_FLOW_ATTR_ALL_DEFAULT:
 968                 type[0] = MLX4_FS_ALL_DEFAULT;
 969                 break;
 970
 971         case IB_FLOW_ATTR_MC_DEFAULT:
 972                 type[0] = MLX4_FS_MC_DEFAULT;
 973                 break;
 974
 975         case IB_FLOW_ATTR_SNIFFER:
 976                 type[0] = MLX4_FS_UC_SNIFFER;
 977                 type[1] = MLX4_FS_MC_SNIFFER;
 978                 break;
 979
 980         default:
 981                 err = -EINVAL;
 982                 goto err_free;
 983         }
 984
 985         while (i < ARRAY_SIZE(type) && type[i]) {
 986                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
 987                                             &mflow->reg_id[i]);
 988                 if (err)
 989                         goto err_free;
 990                 i++;
 991         }
 992
 993         return &mflow->ibflow;
 994
 995 err_free:
 996         kfree(mflow);
 997         return ERR_PTR(err);
 998 }
 999
1000 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1001 {
1002         int err, ret = 0;
1003         int i = 0;
1004         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1005         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1006
1007         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
1008                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
1009                 if (err)
1010                         ret = err;
1011                 i++;
1012         }
1013
1014         kfree(mflow);
1015         return ret;
1016 }
1017
1018 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1019 {
1020         int err;
1021         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1022         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1023         u64 reg_id;
1024         struct mlx4_ib_steering *ib_steering = NULL;
1025         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1026                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1027
1028         if (mdev->dev->caps.steering_mode ==
1029             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1030                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1031                 if (!ib_steering)
1032                         return -ENOMEM;
1033         }
1034
1035         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1036                                     !!(mqp->flags &
1037                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1038                                     prot, &reg_id);
1039         if (err)
1040                 goto err_malloc;
1041
1042         err = add_gid_entry(ibqp, gid);
1043         if (err)
1044                 goto err_add;
1045
1046         if (ib_steering) {
1047                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1048                 ib_steering->reg_id = reg_id;
1049                 mutex_lock(&mqp->mutex);
1050                 list_add(&ib_steering->list, &mqp->steering_rules);
1051                 mutex_unlock(&mqp->mutex);
1052         }
1053         return 0;
1054
1055 err_add:
1056         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1057                               prot, reg_id);
1058 err_malloc:
1059         kfree(ib_steering);
1060
1061         return err;
1062 }
1063
1064 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1065 {
1066         struct mlx4_ib_gid_entry *ge;
1067         struct mlx4_ib_gid_entry *tmp;
1068         struct mlx4_ib_gid_entry *ret = NULL;
1069
1070         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1071                 if (!memcmp(raw, ge->gid.raw, 16)) {
1072                         ret = ge;
1073                         break;
1074                 }
1075         }
1076
1077         return ret;
1078 }
1079
1080 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1081 {
1082         int err;
1083         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1084         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1085         struct net_device *ndev;
1086         struct mlx4_ib_gid_entry *ge;
1087         u64 reg_id = 0;
1088         enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
1089                 MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
1090
1091         if (mdev->dev->caps.steering_mode ==
1092             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1093                 struct mlx4_ib_steering *ib_steering;
1094
1095                 mutex_lock(&mqp->mutex);
1096                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1097                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1098                                 list_del(&ib_steering->list);
1099                                 break;
1100                         }
1101                 }
1102                 mutex_unlock(&mqp->mutex);
1103                 if (&ib_steering->list == &mqp->steering_rules) {
1104                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1105                         return -EINVAL;
1106                 }
1107                 reg_id = ib_steering->reg_id;
1108                 kfree(ib_steering);
1109         }
1110
1111         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1112                                     prot, reg_id);
1113         if (err)
1114                 return err;
1115
1116         mutex_lock(&mqp->mutex);
1117         ge = find_gid_entry(mqp, gid->raw);
1118         if (ge) {
1119                 spin_lock(&mdev->iboe.lock);
1120                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1121                 if (ndev)
1122                         dev_hold(ndev);
1123                 spin_unlock(&mdev->iboe.lock);
1124                 if (ndev)
1125                         dev_put(ndev);
1126                 list_del(&ge->list);
1127                 kfree(ge);
1128         } else
1129                 pr_warn("could not find mgid entry\n");
1130
1131         mutex_unlock(&mqp->mutex);
1132
1133         return 0;
1134 }
1135
1136 static int init_node_data(struct mlx4_ib_dev *dev)
1137 {
1138         struct ib_smp *in_mad  = NULL;
1139         struct ib_smp *out_mad = NULL;
1140         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1141         int err = -ENOMEM;
1142
1143         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1144         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1145         if (!in_mad || !out_mad)
1146                 goto out;
1147
1148         init_query_mad(in_mad);
1149         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1150         if (mlx4_is_master(dev->dev))
1151                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1152
1153         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1154         if (err)
1155                 goto out;
1156
1157         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1158
1159         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1160
1161         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1162         if (err)
1163                 goto out;
1164
1165         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1166         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1167
1168 out:
1169         kfree(in_mad);
1170         kfree(out_mad);
1171         return err;
1172 }
1173
1174 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1175                         char *buf)
1176 {
1177         struct mlx4_ib_dev *dev =
1178                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1179         return sprintf(buf, "MT%d\n", dev->dev->pdev->device);
1180 }
1181
1182 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1183                            char *buf)
1184 {
1185         struct mlx4_ib_dev *dev =
1186                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1187         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1188                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1189                        (int) dev->dev->caps.fw_ver & 0xffff);
1190 }
1191
1192 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1193                         char *buf)
1194 {
1195         struct mlx4_ib_dev *dev =
1196                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1197         return sprintf(buf, "%x\n", dev->dev->rev_id);
1198 }
1199
1200 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1201                           char *buf)
1202 {
1203         struct mlx4_ib_dev *dev =
1204                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1205         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1206                        dev->dev->board_id);
1207 }
1208
1209 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1210 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1211 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1212 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1213
1214 static struct device_attribute *mlx4_class_attributes[] = {
1215         &dev_attr_hw_rev,
1216         &dev_attr_fw_ver,
1217         &dev_attr_hca_type,
1218         &dev_attr_board_id
1219 };
1220
1221 static void update_gids_task(struct work_struct *work)
1222 {
1223         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1224         struct mlx4_cmd_mailbox *mailbox;
1225         union ib_gid *gids;
1226         int err;
1227         struct mlx4_dev *dev = gw->dev->dev;
1228
1229         mailbox = mlx4_alloc_cmd_mailbox(dev);
1230         if (IS_ERR(mailbox)) {
1231                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1232                 return;
1233         }
1234
1235         gids = mailbox->buf;
1236         memcpy(gids, gw->gids, sizeof gw->gids);
1237
1238         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1239                        1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
1240                        MLX4_CMD_WRAPPED);
1241         if (err)
1242                 pr_warn("set port command failed\n");
1243         else
1244                 mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
1245
1246         mlx4_free_cmd_mailbox(dev, mailbox);
1247         kfree(gw);
1248 }
1249
1250 static void reset_gids_task(struct work_struct *work)
1251 {
1252         struct update_gid_work *gw =
1253                         container_of(work, struct update_gid_work, work);
1254         struct mlx4_cmd_mailbox *mailbox;
1255         union ib_gid *gids;
1256         int err;
1257         int i;
1258         struct mlx4_dev *dev = gw->dev->dev;
1259
1260         mailbox = mlx4_alloc_cmd_mailbox(dev);
1261         if (IS_ERR(mailbox)) {
1262                 pr_warn("reset gid table failed\n");
1263                 goto free;
1264         }
1265
1266         gids = mailbox->buf;
1267         memcpy(gids, gw->gids, sizeof(gw->gids));
1268
1269         for (i = 1; i < gw->dev->num_ports + 1; i++) {
1270                 if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, i) ==
1271                                             IB_LINK_LAYER_ETHERNET) {
1272                         err = mlx4_cmd(dev, mailbox->dma,
1273                                        MLX4_SET_PORT_GID_TABLE << 8 | i,
1274                                        1, MLX4_CMD_SET_PORT,
1275                                        MLX4_CMD_TIME_CLASS_B,
1276                                        MLX4_CMD_WRAPPED);
1277                         if (err)
1278                                 pr_warn(KERN_WARNING
1279                                         "set port %d command failed\n", i);
1280                 }
1281         }
1282
1283         mlx4_free_cmd_mailbox(dev, mailbox);
1284 free:
1285         kfree(gw);
1286 }
1287
1288 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1289                             union ib_gid *gid, int clear)
1290 {
1291         struct update_gid_work *work;
1292         int i;
1293         int need_update = 0;
1294         int free = -1;
1295         int found = -1;
1296         int max_gids;
1297
1298         max_gids = dev->dev->caps.gid_table_len[port];
1299         for (i = 0; i < max_gids; ++i) {
1300                 if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1301                             sizeof(*gid)))
1302                         found = i;
1303
1304                 if (clear) {
1305                         if (found >= 0) {
1306                                 need_update = 1;
1307                                 dev->iboe.gid_table[port - 1][found] = zgid;
1308                                 break;
1309                         }
1310                 } else {
1311                         if (found >= 0)
1312                                 break;
1313
1314                         if (free < 0 &&
1315                             !memcmp(&dev->iboe.gid_table[port - 1][i], &zgid,
1316                                     sizeof(*gid)))
1317                                 free = i;
1318                 }
1319         }
1320
1321         if (found == -1 && !clear && free >= 0) {
1322                 dev->iboe.gid_table[port - 1][free] = *gid;
1323                 need_update = 1;
1324         }
1325
1326         if (!need_update)
1327                 return 0;
1328
1329         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1330         if (!work)
1331                 return -ENOMEM;
1332
1333         memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1334         INIT_WORK(&work->work, update_gids_task);
1335         work->port = port;
1336         work->dev = dev;
1337         queue_work(wq, &work->work);
1338
1339         return 0;
1340 }
1341
1342 static int reset_gid_table(struct mlx4_ib_dev *dev)
1343 {
1344         struct update_gid_work *work;
1345
1346
1347         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1348         if (!work)
1349                 return -ENOMEM;
1350         memset(dev->iboe.gid_table, 0, sizeof(dev->iboe.gid_table));
1351         memset(work->gids, 0, sizeof(work->gids));
1352         INIT_WORK(&work->work, reset_gids_task);
1353         work->dev = dev;
1354         queue_work(wq, &work->work);
1355         return 0;
1356 }
1357
1358 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1359                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1360 {
1361         struct mlx4_ib_iboe *iboe;
1362         int port = 0;
1363         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1364                                 rdma_vlan_dev_real_dev(event_netdev) :
1365                                 event_netdev;
1366
1367         if (event != NETDEV_DOWN && event != NETDEV_UP)
1368                 return 0;
1369
1370         if ((real_dev != event_netdev) &&
1371             (event == NETDEV_DOWN) &&
1372             rdma_link_local_addr((struct in6_addr *)gid))
1373                 return 0;
1374
1375         iboe = &ibdev->iboe;
1376         spin_lock(&iboe->lock);
1377
1378         for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1379                 if ((netif_is_bond_master(real_dev) &&
1380                      (real_dev == iboe->masters[port - 1])) ||
1381                      (!netif_is_bond_master(real_dev) &&
1382                      (real_dev == iboe->netdevs[port - 1])))
1383                         update_gid_table(ibdev, port, gid,
1384                                          event == NETDEV_DOWN);
1385
1386         spin_unlock(&iboe->lock);
1387         return 0;
1388
1389 }
1390
1391 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1392                                struct mlx4_ib_dev *ibdev)
1393 {
1394         u8 port = 0;
1395         struct mlx4_ib_iboe *iboe;
1396         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1397                                 rdma_vlan_dev_real_dev(dev) : dev;
1398
1399         iboe = &ibdev->iboe;
1400         spin_lock(&iboe->lock);
1401
1402         for (port = 1; port <= MLX4_MAX_PORTS; ++port)
1403                 if ((netif_is_bond_master(real_dev) &&
1404                      (real_dev == iboe->masters[port - 1])) ||
1405                      (!netif_is_bond_master(real_dev) &&
1406                      (real_dev == iboe->netdevs[port - 1])))
1407                         break;
1408
1409         spin_unlock(&iboe->lock);
1410
1411         if ((port == 0) || (port > MLX4_MAX_PORTS))
1412                 return 0;
1413         else
1414                 return port;
1415 }
1416
1417 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1418                                 void *ptr)
1419 {
1420         struct mlx4_ib_dev *ibdev;
1421         struct in_ifaddr *ifa = ptr;
1422         union ib_gid gid;
1423         struct net_device *event_netdev = ifa->ifa_dev->dev;
1424
1425         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1426
1427         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1428
1429         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1430         return NOTIFY_DONE;
1431 }
1432
1433 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1434 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1435                                 void *ptr)
1436 {
1437         struct mlx4_ib_dev *ibdev;
1438         struct inet6_ifaddr *ifa = ptr;
1439         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1440         struct net_device *event_netdev = ifa->idev->dev;
1441
1442         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1443
1444         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1445         return NOTIFY_DONE;
1446 }
1447 #endif
1448
1449 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1450                                  struct mlx4_ib_dev *ibdev, u8 port)
1451 {
1452         struct in_device *in_dev;
1453 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1454         struct inet6_dev *in6_dev;
1455         union ib_gid  *pgid;
1456         struct inet6_ifaddr *ifp;
1457 #endif
1458         union ib_gid gid;
1459
1460
1461         if ((port == 0) || (port > MLX4_MAX_PORTS))
1462                 return;
1463
1464         /* IPv4 gids */
1465         in_dev = in_dev_get(dev);
1466         if (in_dev) {
1467                 for_ifa(in_dev) {
1468                         /*ifa->ifa_address;*/
1469                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1470                                                (struct in6_addr *)&gid);
1471                         update_gid_table(ibdev, port, &gid, 0);
1472                 }
1473                 endfor_ifa(in_dev);
1474                 in_dev_put(in_dev);
1475         }
1476 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1477         /* IPv6 gids */
1478         in6_dev = in6_dev_get(dev);
1479         if (in6_dev) {
1480                 read_lock_bh(&in6_dev->lock);
1481                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1482                         pgid = (union ib_gid *)&ifp->addr;
1483                         update_gid_table(ibdev, port, pgid, 0);
1484                 }
1485                 read_unlock_bh(&in6_dev->lock);
1486                 in6_dev_put(in6_dev);
1487         }
1488 #endif
1489 }
1490
1491 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1492 {
1493         struct  net_device *dev;
1494
1495         if (reset_gid_table(ibdev))
1496                 return -1;
1497
1498         read_lock(&dev_base_lock);
1499
1500         for_each_netdev(&init_net, dev) {
1501                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1502                 if (port)
1503                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1504         }
1505
1506         read_unlock(&dev_base_lock);
1507
1508         return 0;
1509 }
1510
1511 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev)
1512 {
1513         struct mlx4_ib_iboe *iboe;
1514         int port;
1515
1516         iboe = &ibdev->iboe;
1517
1518         spin_lock(&iboe->lock);
1519         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1520                 struct net_device *old_master = iboe->masters[port - 1];
1521                 struct net_device *curr_master;
1522                 iboe->netdevs[port - 1] =
1523                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1524
1525                 if (iboe->netdevs[port - 1] &&
1526                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1527                         rtnl_lock();
1528                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1529                                 iboe->netdevs[port - 1]);
1530                         rtnl_unlock();
1531                 }
1532                 curr_master = iboe->masters[port - 1];
1533
1534                 /* if bonding is used it is possible that we add it to masters
1535                     only after IP address is assigned to the net bonding
1536                     interface */
1537                 if (curr_master && (old_master != curr_master))
1538                         mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1539         }
1540
1541         spin_unlock(&iboe->lock);
1542 }
1543
1544 static int mlx4_ib_netdev_event(struct notifier_block *this,
1545                                 unsigned long event, void *ptr)
1546 {
1547         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
1548         struct mlx4_ib_dev *ibdev;
1549
1550         if (!net_eq(dev_net(dev), &init_net))
1551                 return NOTIFY_DONE;
1552
1553         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
1554         mlx4_ib_scan_netdevs(ibdev);
1555
1556         return NOTIFY_DONE;
1557 }
1558
1559 static void init_pkeys(struct mlx4_ib_dev *ibdev)
1560 {
1561         int port;
1562         int slave;
1563         int i;
1564
1565         if (mlx4_is_master(ibdev->dev)) {
1566                 for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
1567                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1568                                 for (i = 0;
1569                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1570                                      ++i) {
1571                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
1572                                         /* master has the identity virt2phys pkey mapping */
1573                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
1574                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
1575                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
1576                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
1577                                 }
1578                         }
1579                 }
1580                 /* initialize pkey cache */
1581                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
1582                         for (i = 0;
1583                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
1584                              ++i)
1585                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
1586                                         (i) ? 0 : 0xFFFF;
1587                 }
1588         }
1589 }
1590
1591 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1592 {
1593         char name[32];
1594         int eq_per_port = 0;
1595         int added_eqs = 0;
1596         int total_eqs = 0;
1597         int i, j, eq;
1598
1599         /* Legacy mode or comp_pool is not large enough */
1600         if (dev->caps.comp_pool == 0 ||
1601             dev->caps.num_ports > dev->caps.comp_pool)
1602                 return;
1603
1604         eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
1605                                         dev->caps.num_ports);
1606
1607         /* Init eq table */
1608         added_eqs = 0;
1609         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
1610                 added_eqs += eq_per_port;
1611
1612         total_eqs = dev->caps.num_comp_vectors + added_eqs;
1613
1614         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
1615         if (!ibdev->eq_table)
1616                 return;
1617
1618         ibdev->eq_added = added_eqs;
1619
1620         eq = 0;
1621         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
1622                 for (j = 0; j < eq_per_port; j++) {
1623                         sprintf(name, "mlx4-ib-%d-%d@%s",
1624                                 i, j, dev->pdev->bus->name);
1625                         /* Set IRQ for specific name (per ring) */
1626                         if (mlx4_assign_eq(dev, name, NULL,
1627                                            &ibdev->eq_table[eq])) {
1628                                 /* Use legacy (same as mlx4_en driver) */
1629                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
1630                                 ibdev->eq_table[eq] =
1631                                         (eq % dev->caps.num_comp_vectors);
1632                         }
1633                         eq++;
1634                 }
1635         }
1636
1637         /* Fill the reset of the vector with legacy EQ */
1638         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
1639                 ibdev->eq_table[eq++] = i;
1640
1641         /* Advertise the new number of EQs to clients */
1642         ibdev->ib_dev.num_comp_vectors = total_eqs;
1643 }
1644
1645 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
1646 {
1647         int i;
1648
1649         /* no additional eqs were added */
1650         if (!ibdev->eq_table)
1651                 return;
1652
1653         /* Reset the advertised EQ number */
1654         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
1655
1656         /* Free only the added eqs */
1657         for (i = 0; i < ibdev->eq_added; i++) {
1658                 /* Don't free legacy eqs if used */
1659                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
1660                         continue;
1661                 mlx4_release_eq(dev, ibdev->eq_table[i]);
1662         }
1663
1664         kfree(ibdev->eq_table);
1665 }
1666
1667 static void *mlx4_ib_add(struct mlx4_dev *dev)
1668 {
1669         struct mlx4_ib_dev *ibdev;
1670         int num_ports = 0;
1671         int i, j;
1672         int err;
1673         struct mlx4_ib_iboe *iboe;
1674
1675         pr_info_once("%s", mlx4_ib_version);
1676
1677         mlx4_foreach_non_ib_transport_port(i, dev)
1678                 num_ports++;
1679
1680         if (mlx4_is_mfunc(dev) && num_ports) {
1681                 dev_err(&dev->pdev->dev, "RoCE is not supported over SRIOV as yet\n");
1682                 return NULL;
1683         }
1684
1685         num_ports = 0;
1686         mlx4_foreach_ib_transport_port(i, dev)
1687                 num_ports++;
1688
1689         /* No point in registering a device with no ports... */
1690         if (num_ports == 0)
1691                 return NULL;
1692
1693         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
1694         if (!ibdev) {
1695                 dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
1696                 return NULL;
1697         }
1698
1699         iboe = &ibdev->iboe;
1700
1701         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
1702                 goto err_dealloc;
1703
1704         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
1705                 goto err_pd;
1706
1707         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
1708                                  PAGE_SIZE);
1709         if (!ibdev->uar_map)
1710                 goto err_uar;
1711         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
1712
1713         ibdev->dev = dev;
1714
1715         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
1716         ibdev->ib_dev.owner             = THIS_MODULE;
1717         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
1718         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
1719         ibdev->num_ports                = num_ports;
1720         ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
1721         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
1722         ibdev->ib_dev.dma_device        = &dev->pdev->dev;
1723
1724         if (dev->caps.userspace_caps)
1725                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
1726         else
1727                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
1728
1729         ibdev->ib_dev.uverbs_cmd_mask   =
1730                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
1731                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
1732                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
1733                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
1734                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
1735                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
1736                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
1737                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1738                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
1739                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
1740                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
1741                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
1742                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
1743                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
1744                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
1745                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
1746                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
1747                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
1748                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
1749                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
1750                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
1751                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
1752                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
1753
1754         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
1755         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
1756         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
1757         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
1758         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
1759         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
1760         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
1761         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
1762         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
1763         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
1764         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
1765         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
1766         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
1767         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
1768         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
1769         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
1770         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
1771         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
1772         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
1773         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
1774         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
1775         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
1776         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
1777         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
1778         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
1779         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
1780         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
1781         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
1782         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
1783         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
1784         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
1785         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
1786         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
1787         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
1788         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
1789         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
1790         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
1791         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
1792         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
1793         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
1794         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
1795
1796         if (!mlx4_is_slave(ibdev->dev)) {
1797                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
1798                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
1799                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
1800                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
1801         }
1802
1803         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
1804             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
1805                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
1806                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
1807                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
1808
1809                 ibdev->ib_dev.uverbs_cmd_mask |=
1810                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
1811                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
1812         }
1813
1814         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
1815                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
1816                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
1817                 ibdev->ib_dev.uverbs_cmd_mask |=
1818                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
1819                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
1820         }
1821
1822         if (check_flow_steering_support(dev)) {
1823                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
1824                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
1825
1826                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
1827                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
1828                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
1829         }
1830
1831         mlx4_ib_alloc_eqs(dev, ibdev);
1832
1833         spin_lock_init(&iboe->lock);
1834
1835         if (init_node_data(ibdev))
1836                 goto err_map;
1837
1838         for (i = 0; i < ibdev->num_ports; ++i) {
1839                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
1840                                                 IB_LINK_LAYER_ETHERNET) {
1841                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
1842                         if (err)
1843                                 ibdev->counters[i] = -1;
1844                 } else
1845                                 ibdev->counters[i] = -1;
1846         }
1847
1848         spin_lock_init(&ibdev->sm_lock);
1849         mutex_init(&ibdev->cap_mask_mutex);
1850
1851         if (ib_register_device(&ibdev->ib_dev, NULL))
1852                 goto err_counter;
1853
1854         if (mlx4_ib_mad_init(ibdev))
1855                 goto err_reg;
1856
1857         if (mlx4_ib_init_sriov(ibdev))
1858                 goto err_mad;
1859
1860         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
1861                 if (!iboe->nb.notifier_call) {
1862                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
1863                         err = register_netdevice_notifier(&iboe->nb);
1864                         if (err) {
1865                                 iboe->nb.notifier_call = NULL;
1866                                 goto err_notif;
1867                         }
1868                 }
1869                 if (!iboe->nb_inet.notifier_call) {
1870                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
1871                         err = register_inetaddr_notifier(&iboe->nb_inet);
1872                         if (err) {
1873                                 iboe->nb_inet.notifier_call = NULL;
1874                                 goto err_notif;
1875                         }
1876                 }
1877 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1878                 if (!iboe->nb_inet6.notifier_call) {
1879                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
1880                         err = register_inet6addr_notifier(&iboe->nb_inet6);
1881                         if (err) {
1882                                 iboe->nb_inet6.notifier_call = NULL;
1883                                 goto err_notif;
1884                         }
1885                 }
1886 #endif
1887                 mlx4_ib_scan_netdevs(ibdev);
1888                 mlx4_ib_init_gid_table(ibdev);
1889         }
1890
1891         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
1892                 if (device_create_file(&ibdev->ib_dev.dev,
1893                                        mlx4_class_attributes[j]))
1894                         goto err_notif;
1895         }
1896
1897         ibdev->ib_active = true;
1898
1899         if (mlx4_is_mfunc(ibdev->dev))
1900                 init_pkeys(ibdev);
1901
1902         /* create paravirt contexts for any VFs which are active */
1903         if (mlx4_is_master(ibdev->dev)) {
1904                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
1905                         if (j == mlx4_master_func_num(ibdev->dev))
1906                                 continue;
1907                         if (mlx4_is_slave_active(ibdev->dev, j))
1908                                 do_slave_init(ibdev, j, 1);
1909                 }
1910         }
1911         return ibdev;
1912
1913 err_notif:
1914         if (ibdev->iboe.nb.notifier_call) {
1915                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1916                         pr_warn("failure unregistering notifier\n");
1917                 ibdev->iboe.nb.notifier_call = NULL;
1918         }
1919         if (ibdev->iboe.nb_inet.notifier_call) {
1920                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
1921                         pr_warn("failure unregistering notifier\n");
1922                 ibdev->iboe.nb_inet.notifier_call = NULL;
1923         }
1924 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1925         if (ibdev->iboe.nb_inet6.notifier_call) {
1926                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
1927                         pr_warn("failure unregistering notifier\n");
1928                 ibdev->iboe.nb_inet6.notifier_call = NULL;
1929         }
1930 #endif
1931         flush_workqueue(wq);
1932
1933         mlx4_ib_close_sriov(ibdev);
1934
1935 err_mad:
1936         mlx4_ib_mad_cleanup(ibdev);
1937
1938 err_reg:
1939         ib_unregister_device(&ibdev->ib_dev);
1940
1941 err_counter:
1942         for (; i; --i)
1943                 if (ibdev->counters[i - 1] != -1)
1944                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
1945
1946 err_map:
1947         iounmap(ibdev->uar_map);
1948
1949 err_uar:
1950         mlx4_uar_free(dev, &ibdev->priv_uar);
1951
1952 err_pd:
1953         mlx4_pd_free(dev, ibdev->priv_pdn);
1954
1955 err_dealloc:
1956         ib_dealloc_device(&ibdev->ib_dev);
1957
1958         return NULL;
1959 }
1960
1961 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
1962 {
1963         struct mlx4_ib_dev *ibdev = ibdev_ptr;
1964         int p;
1965
1966         mlx4_ib_close_sriov(ibdev);
1967         mlx4_ib_mad_cleanup(ibdev);
1968         ib_unregister_device(&ibdev->ib_dev);
1969         if (ibdev->iboe.nb.notifier_call) {
1970                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
1971                         pr_warn("failure unregistering notifier\n");
1972                 ibdev->iboe.nb.notifier_call = NULL;
1973         }
1974         if (ibdev->iboe.nb_inet.notifier_call) {
1975                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
1976                         pr_warn("failure unregistering notifier\n");
1977                 ibdev->iboe.nb_inet.notifier_call = NULL;
1978         }
1979 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
1980         if (ibdev->iboe.nb_inet6.notifier_call) {
1981                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
1982                         pr_warn("failure unregistering notifier\n");
1983                 ibdev->iboe.nb_inet6.notifier_call = NULL;
1984         }
1985 #endif
1986         iounmap(ibdev->uar_map);
1987         for (p = 0; p < ibdev->num_ports; ++p)
1988                 if (ibdev->counters[p] != -1)
1989                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
1990         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
1991                 mlx4_CLOSE_PORT(dev, p);
1992
1993         mlx4_ib_free_eqs(dev, ibdev);
1994
1995         mlx4_uar_free(dev, &ibdev->priv_uar);
1996         mlx4_pd_free(dev, ibdev->priv_pdn);
1997         ib_dealloc_device(&ibdev->ib_dev);
1998 }
1999
2000 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2001 {
2002         struct mlx4_ib_demux_work **dm = NULL;
2003         struct mlx4_dev *dev = ibdev->dev;
2004         int i;
2005         unsigned long flags;
2006
2007         if (!mlx4_is_master(dev))
2008                 return;
2009
2010         dm = kcalloc(dev->caps.num_ports, sizeof *dm, GFP_ATOMIC);
2011         if (!dm) {
2012                 pr_err("failed to allocate memory for tunneling qp update\n");
2013                 goto out;
2014         }
2015
2016         for (i = 0; i < dev->caps.num_ports; i++) {
2017                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2018                 if (!dm[i]) {
2019                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2020                         for (i = 0; i < dev->caps.num_ports; i++) {
2021                                 if (dm[i])
2022                                         kfree(dm[i]);
2023                         }
2024                         goto out;
2025                 }
2026         }
2027         /* initialize or tear down tunnel QPs for the slave */
2028         for (i = 0; i < dev->caps.num_ports; i++) {
2029                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2030                 dm[i]->port = i + 1;
2031                 dm[i]->slave = slave;
2032                 dm[i]->do_init = do_init;
2033                 dm[i]->dev = ibdev;
2034                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2035                 if (!ibdev->sriov.is_going_down)
2036                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2037                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2038         }
2039 out:
2040         kfree(dm);
2041         return;
2042 }
2043
2044 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2045                           enum mlx4_dev_event event, unsigned long param)
2046 {
2047         struct ib_event ibev;
2048         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2049         struct mlx4_eqe *eqe = NULL;
2050         struct ib_event_work *ew;
2051         int p = 0;
2052
2053         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2054                 eqe = (struct mlx4_eqe *)param;
2055         else
2056                 p = (int) param;
2057
2058         switch (event) {
2059         case MLX4_DEV_EVENT_PORT_UP:
2060                 if (p > ibdev->num_ports)
2061                         return;
2062                 if (mlx4_is_master(dev) &&
2063                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2064                         IB_LINK_LAYER_INFINIBAND) {
2065                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2066                 }
2067                 ibev.event = IB_EVENT_PORT_ACTIVE;
2068                 break;
2069
2070         case MLX4_DEV_EVENT_PORT_DOWN:
2071                 if (p > ibdev->num_ports)
2072                         return;
2073                 ibev.event = IB_EVENT_PORT_ERR;
2074                 break;
2075
2076         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2077                 ibdev->ib_active = false;
2078                 ibev.event = IB_EVENT_DEVICE_FATAL;
2079                 break;
2080
2081         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2082                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2083                 if (!ew) {
2084                         pr_err("failed to allocate memory for events work\n");
2085                         break;
2086                 }
2087
2088                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2089                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2090                 ew->ib_dev = ibdev;
2091                 /* need to queue only for port owner, which uses GEN_EQE */
2092                 if (mlx4_is_master(dev))
2093                         queue_work(wq, &ew->work);
2094                 else
2095                         handle_port_mgmt_change_event(&ew->work);
2096                 return;
2097
2098         case MLX4_DEV_EVENT_SLAVE_INIT:
2099                 /* here, p is the slave id */
2100                 do_slave_init(ibdev, p, 1);
2101                 return;
2102
2103         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2104                 /* here, p is the slave id */
2105                 do_slave_init(ibdev, p, 0);
2106                 return;
2107
2108         default:
2109                 return;
2110         }
2111
2112         ibev.device           = ibdev_ptr;
2113         ibev.element.port_num = (u8) p;
2114
2115         ib_dispatch_event(&ibev);
2116 }
2117
2118 static struct mlx4_interface mlx4_ib_interface = {
2119         .add            = mlx4_ib_add,
2120         .remove         = mlx4_ib_remove,
2121         .event          = mlx4_ib_event,
2122         .protocol       = MLX4_PROT_IB_IPV6
2123 };
2124
2125 static int __init mlx4_ib_init(void)
2126 {
2127         int err;
2128
2129         wq = create_singlethread_workqueue("mlx4_ib");
2130         if (!wq)
2131                 return -ENOMEM;
2132
2133         err = mlx4_ib_mcg_init();
2134         if (err)
2135                 goto clean_wq;
2136
2137         err = mlx4_register_interface(&mlx4_ib_interface);
2138         if (err)
2139                 goto clean_mcg;
2140
2141         return 0;
2142
2143 clean_mcg:
2144         mlx4_ib_mcg_destroy();
2145
2146 clean_wq:
2147         destroy_workqueue(wq);
2148         return err;
2149 }
2150
2151 static void __exit mlx4_ib_cleanup(void)
2152 {
2153         mlx4_unregister_interface(&mlx4_ib_interface);
2154         mlx4_ib_mcg_destroy();
2155         destroy_workqueue(wq);
2156 }
2157
2158 module_init(mlx4_ib_init);
2159 module_exit(mlx4_ib_cleanup);