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 #include <linux/mlx4/qp.h>
  52
  53 #include "mlx4_ib.h"
  54 #include "user.h"
  55
  56 #define DRV_NAME        MLX4_IB_DRV_NAME
  57 #define DRV_VERSION     "2.2-1"
  58 #define DRV_RELDATE     "Feb 2014"
  59
  60 #define MLX4_IB_FLOW_MAX_PRIO 0xFFF
  61 #define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
  62 #define MLX4_IB_CARD_REV_A0   0xA0
  63
  64 MODULE_AUTHOR("Roland Dreier");
  65 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
  66 MODULE_LICENSE("Dual BSD/GPL");
  67 MODULE_VERSION(DRV_VERSION);
  68
  69 int mlx4_ib_sm_guid_assign = 0;
  70 module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
  71 MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 0)");
  72
  73 static const char mlx4_ib_version[] =
  74         DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
  75         DRV_VERSION " (" DRV_RELDATE ")\n";
  76
  77 struct update_gid_work {
  78         struct work_struct      work;
  79         union ib_gid            gids[128];
  80         struct mlx4_ib_dev     *dev;
  81         int                     port;
  82 };
  83
  84 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
  85
  86 static struct workqueue_struct *wq;
  87
  88 static void init_query_mad(struct ib_smp *mad)
  89 {
  90         mad->base_version  = 1;
  91         mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
  92         mad->class_version = 1;
  93         mad->method        = IB_MGMT_METHOD_GET;
  94 }
  95
  96 static union ib_gid zgid;
  97
  98 static int check_flow_steering_support(struct mlx4_dev *dev)
  99 {
 100         int eth_num_ports = 0;
 101         int ib_num_ports = 0;
 102
 103         int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
 104
 105         if (dmfs) {
 106                 int i;
 107                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
 108                         eth_num_ports++;
 109                 mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 110                         ib_num_ports++;
 111                 dmfs &= (!ib_num_ports ||
 112                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
 113                         (!eth_num_ports ||
 114                          (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
 115                 if (ib_num_ports && mlx4_is_mfunc(dev)) {
 116                         pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
 117                         dmfs = 0;
 118                 }
 119         }
 120         return dmfs;
 121 }
 122
 123 static int num_ib_ports(struct mlx4_dev *dev)
 124 {
 125         int ib_ports = 0;
 126         int i;
 127
 128         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
 129                 ib_ports++;
 130
 131         return ib_ports;
 132 }
 133
 134 static int mlx4_ib_query_device(struct ib_device *ibdev,
 135                                 struct ib_device_attr *props)
 136 {
 137         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 138         struct ib_smp *in_mad  = NULL;
 139         struct ib_smp *out_mad = NULL;
 140         int err = -ENOMEM;
 141         int have_ib_ports;
 142
 143         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 144         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 145         if (!in_mad || !out_mad)
 146                 goto out;
 147
 148         init_query_mad(in_mad);
 149         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 150
 151         err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
 152                            1, NULL, NULL, in_mad, out_mad);
 153         if (err)
 154                 goto out;
 155
 156         memset(props, 0, sizeof *props);
 157
 158         have_ib_ports = num_ib_ports(dev->dev);
 159
 160         props->fw_ver = dev->dev->caps.fw_ver;
 161         props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
 162                 IB_DEVICE_PORT_ACTIVE_EVENT             |
 163                 IB_DEVICE_SYS_IMAGE_GUID                |
 164                 IB_DEVICE_RC_RNR_NAK_GEN                |
 165                 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
 166         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_PKEY_CNTR)
 167                 props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
 168         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BAD_QKEY_CNTR)
 169                 props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
 170         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_APM && have_ib_ports)
 171                 props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
 172         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_UD_AV_PORT)
 173                 props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 174         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 175                 props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
 176         if (dev->dev->caps.max_gso_sz &&
 177             (dev->dev->rev_id != MLX4_IB_CARD_REV_A0) &&
 178             (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH))
 179                 props->device_cap_flags |= IB_DEVICE_UD_TSO;
 180         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
 181                 props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
 182         if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
 183             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
 184             (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
 185                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 186         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
 187                 props->device_cap_flags |= IB_DEVICE_XRC;
 188         if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
 189                 props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
 190         if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
 191                 if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
 192                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
 193                 else
 194                         props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
 195         if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
 196                 props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
 197         }
 198
 199         props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 200                 0xffffff;
 201         props->vendor_part_id      = dev->dev->persist->pdev->device;
 202         props->hw_ver              = be32_to_cpup((__be32 *) (out_mad->data + 32));
 203         memcpy(&props->sys_image_guid, out_mad->data +  4, 8);
 204
 205         props->max_mr_size         = ~0ull;
 206         props->page_size_cap       = dev->dev->caps.page_size_cap;
 207         props->max_qp              = dev->dev->quotas.qp;
 208         props->max_qp_wr           = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 209         props->max_sge             = min(dev->dev->caps.max_sq_sg,
 210                                          dev->dev->caps.max_rq_sg);
 211         props->max_cq              = dev->dev->quotas.cq;
 212         props->max_cqe             = dev->dev->caps.max_cqes;
 213         props->max_mr              = dev->dev->quotas.mpt;
 214         props->max_pd              = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 215         props->max_qp_rd_atom      = dev->dev->caps.max_qp_dest_rdma;
 216         props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 217         props->max_res_rd_atom     = props->max_qp_rd_atom * props->max_qp;
 218         props->max_srq             = dev->dev->quotas.srq;
 219         props->max_srq_wr          = dev->dev->caps.max_srq_wqes - 1;
 220         props->max_srq_sge         = dev->dev->caps.max_srq_sge;
 221         props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 222         props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 223         props->atomic_cap          = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 224                 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
 225         props->masked_atomic_cap   = props->atomic_cap;
 226         props->max_pkeys           = dev->dev->caps.pkey_table_len[1];
 227         props->max_mcast_grp       = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 228         props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 229         props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 230                                            props->max_mcast_grp;
 231         props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 232
 233 out:
 234         kfree(in_mad);
 235         kfree(out_mad);
 236
 237         return err;
 238 }
 239
 240 static enum rdma_link_layer
 241 mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
 242 {
 243         struct mlx4_dev *dev = to_mdev(device)->dev;
 244
 245         return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
 246                 IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
 247 }
 248
 249 static int ib_link_query_port(struct ib_device *ibdev, u8 port,
 250                               struct ib_port_attr *props, int netw_view)
 251 {
 252         struct ib_smp *in_mad  = NULL;
 253         struct ib_smp *out_mad = NULL;
 254         int ext_active_speed;
 255         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 256         int err = -ENOMEM;
 257
 258         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 259         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 260         if (!in_mad || !out_mad)
 261                 goto out;
 262
 263         init_query_mad(in_mad);
 264         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 265         in_mad->attr_mod = cpu_to_be32(port);
 266
 267         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 268                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 269
 270         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 271                                 in_mad, out_mad);
 272         if (err)
 273                 goto out;
 274
 275
 276         props->lid              = be16_to_cpup((__be16 *) (out_mad->data + 16));
 277         props->lmc              = out_mad->data[34] & 0x7;
 278         props->sm_lid           = be16_to_cpup((__be16 *) (out_mad->data + 18));
 279         props->sm_sl            = out_mad->data[36] & 0xf;
 280         props->state            = out_mad->data[32] & 0xf;
 281         props->phys_state       = out_mad->data[33] >> 4;
 282         props->port_cap_flags   = be32_to_cpup((__be32 *) (out_mad->data + 20));
 283         if (netw_view)
 284                 props->gid_tbl_len = out_mad->data[50];
 285         else
 286                 props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 287         props->max_msg_sz       = to_mdev(ibdev)->dev->caps.max_msg_sz;
 288         props->pkey_tbl_len     = to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 289         props->bad_pkey_cntr    = be16_to_cpup((__be16 *) (out_mad->data + 46));
 290         props->qkey_viol_cntr   = be16_to_cpup((__be16 *) (out_mad->data + 48));
 291         props->active_width     = out_mad->data[31] & 0xf;
 292         props->active_speed     = out_mad->data[35] >> 4;
 293         props->max_mtu          = out_mad->data[41] & 0xf;
 294         props->active_mtu       = out_mad->data[36] >> 4;
 295         props->subnet_timeout   = out_mad->data[51] & 0x1f;
 296         props->max_vl_num       = out_mad->data[37] >> 4;
 297         props->init_type_reply  = out_mad->data[41] >> 4;
 298
 299         /* Check if extended speeds (EDR/FDR/...) are supported */
 300         if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
 301                 ext_active_speed = out_mad->data[62] >> 4;
 302
 303                 switch (ext_active_speed) {
 304                 case 1:
 305                         props->active_speed = IB_SPEED_FDR;
 306                         break;
 307                 case 2:
 308                         props->active_speed = IB_SPEED_EDR;
 309                         break;
 310                 }
 311         }
 312
 313         /* If reported active speed is QDR, check if is FDR-10 */
 314         if (props->active_speed == IB_SPEED_QDR) {
 315                 init_query_mad(in_mad);
 316                 in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
 317                 in_mad->attr_mod = cpu_to_be32(port);
 318
 319                 err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
 320                                    NULL, NULL, in_mad, out_mad);
 321                 if (err)
 322                         goto out;
 323
 324                 /* Checking LinkSpeedActive for FDR-10 */
 325                 if (out_mad->data[15] & 0x1)
 326                         props->active_speed = IB_SPEED_FDR10;
 327         }
 328
 329         /* Avoid wrong speed value returned by FW if the IB link is down. */
 330         if (props->state == IB_PORT_DOWN)
 331                  props->active_speed = IB_SPEED_SDR;
 332
 333 out:
 334         kfree(in_mad);
 335         kfree(out_mad);
 336         return err;
 337 }
 338
 339 static u8 state_to_phys_state(enum ib_port_state state)
 340 {
 341         return state == IB_PORT_ACTIVE ? 5 : 3;
 342 }
 343
 344 static int eth_link_query_port(struct ib_device *ibdev, u8 port,
 345                                struct ib_port_attr *props, int netw_view)
 346 {
 347
 348         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 349         struct mlx4_ib_iboe *iboe = &mdev->iboe;
 350         struct net_device *ndev;
 351         enum ib_mtu tmp;
 352         struct mlx4_cmd_mailbox *mailbox;
 353         int err = 0;
 354         int is_bonded = mlx4_is_bonded(mdev->dev);
 355
 356         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
 357         if (IS_ERR(mailbox))
 358                 return PTR_ERR(mailbox);
 359
 360         err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
 361                            MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
 362                            MLX4_CMD_WRAPPED);
 363         if (err)
 364                 goto out;
 365
 366         props->active_width     =  (((u8 *)mailbox->buf)[5] == 0x40) ?
 367                                                 IB_WIDTH_4X : IB_WIDTH_1X;
 368         props->active_speed     = IB_SPEED_QDR;
 369         props->port_cap_flags   = IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
 370         props->gid_tbl_len      = mdev->dev->caps.gid_table_len[port];
 371         props->max_msg_sz       = mdev->dev->caps.max_msg_sz;
 372         props->pkey_tbl_len     = 1;
 373         props->max_mtu          = IB_MTU_4096;
 374         props->max_vl_num       = 2;
 375         props->state            = IB_PORT_DOWN;
 376         props->phys_state       = state_to_phys_state(props->state);
 377         props->active_mtu       = IB_MTU_256;
 378         if (is_bonded)
 379                 rtnl_lock(); /* required to get upper dev */
 380         spin_lock_bh(&iboe->lock);
 381         ndev = iboe->netdevs[port - 1];
 382         if (ndev && is_bonded)
 383                 ndev = netdev_master_upper_dev_get(ndev);
 384         if (!ndev)
 385                 goto out_unlock;
 386
 387         tmp = iboe_get_mtu(ndev->mtu);
 388         props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
 389
 390         props->state            = (netif_running(ndev) && netif_carrier_ok(ndev)) ?
 391                                         IB_PORT_ACTIVE : IB_PORT_DOWN;
 392         props->phys_state       = state_to_phys_state(props->state);
 393 out_unlock:
 394         spin_unlock_bh(&iboe->lock);
 395         if (is_bonded)
 396                 rtnl_unlock();
 397 out:
 398         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 399         return err;
 400 }
 401
 402 int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 403                          struct ib_port_attr *props, int netw_view)
 404 {
 405         int err;
 406
 407         memset(props, 0, sizeof *props);
 408
 409         err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
 410                 ib_link_query_port(ibdev, port, props, netw_view) :
 411                                 eth_link_query_port(ibdev, port, props, netw_view);
 412
 413         return err;
 414 }
 415
 416 static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 417                               struct ib_port_attr *props)
 418 {
 419         /* returns host view */
 420         return __mlx4_ib_query_port(ibdev, port, props, 0);
 421 }
 422
 423 int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 424                         union ib_gid *gid, int netw_view)
 425 {
 426         struct ib_smp *in_mad  = NULL;
 427         struct ib_smp *out_mad = NULL;
 428         int err = -ENOMEM;
 429         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 430         int clear = 0;
 431         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 432
 433         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 434         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 435         if (!in_mad || !out_mad)
 436                 goto out;
 437
 438         init_query_mad(in_mad);
 439         in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 440         in_mad->attr_mod = cpu_to_be32(port);
 441
 442         if (mlx4_is_mfunc(dev->dev) && netw_view)
 443                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 444
 445         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 446         if (err)
 447                 goto out;
 448
 449         memcpy(gid->raw, out_mad->data + 8, 8);
 450
 451         if (mlx4_is_mfunc(dev->dev) && !netw_view) {
 452                 if (index) {
 453                         /* For any index > 0, return the null guid */
 454                         err = 0;
 455                         clear = 1;
 456                         goto out;
 457                 }
 458         }
 459
 460         init_query_mad(in_mad);
 461         in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 462         in_mad->attr_mod = cpu_to_be32(index / 8);
 463
 464         err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
 465                            NULL, NULL, in_mad, out_mad);
 466         if (err)
 467                 goto out;
 468
 469         memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 470
 471 out:
 472         if (clear)
 473                 memset(gid->raw + 8, 0, 8);
 474         kfree(in_mad);
 475         kfree(out_mad);
 476         return err;
 477 }
 478
 479 static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
 480                           union ib_gid *gid)
 481 {
 482         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 483
 484         *gid = dev->iboe.gid_table[port - 1][index];
 485
 486         return 0;
 487 }
 488
 489 static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 490                              union ib_gid *gid)
 491 {
 492         if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
 493                 return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
 494         else
 495                 return iboe_query_gid(ibdev, port, index, gid);
 496 }
 497
 498 int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 499                          u16 *pkey, int netw_view)
 500 {
 501         struct ib_smp *in_mad  = NULL;
 502         struct ib_smp *out_mad = NULL;
 503         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 504         int err = -ENOMEM;
 505
 506         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 507         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 508         if (!in_mad || !out_mad)
 509                 goto out;
 510
 511         init_query_mad(in_mad);
 512         in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 513         in_mad->attr_mod = cpu_to_be32(index / 32);
 514
 515         if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
 516                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 517
 518         err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
 519                            in_mad, out_mad);
 520         if (err)
 521                 goto out;
 522
 523         *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 524
 525 out:
 526         kfree(in_mad);
 527         kfree(out_mad);
 528         return err;
 529 }
 530
 531 static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
 532 {
 533         return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
 534 }
 535
 536 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 537                                  struct ib_device_modify *props)
 538 {
 539         struct mlx4_cmd_mailbox *mailbox;
 540         unsigned long flags;
 541
 542         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 543                 return -EOPNOTSUPP;
 544
 545         if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
 546                 return 0;
 547
 548         if (mlx4_is_slave(to_mdev(ibdev)->dev))
 549                 return -EOPNOTSUPP;
 550
 551         spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
 552         memcpy(ibdev->node_desc, props->node_desc, 64);
 553         spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
 554
 555         /*
 556          * If possible, pass node desc to FW, so it can generate
 557          * a 144 trap.  If cmd fails, just ignore.
 558          */
 559         mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
 560         if (IS_ERR(mailbox))
 561                 return 0;
 562
 563         memcpy(mailbox->buf, props->node_desc, 64);
 564         mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
 565                  MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
 566
 567         mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 568
 569         return 0;
 570 }
 571
 572 static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 573                             u32 cap_mask)
 574 {
 575         struct mlx4_cmd_mailbox *mailbox;
 576         int err;
 577
 578         mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
 579         if (IS_ERR(mailbox))
 580                 return PTR_ERR(mailbox);
 581
 582         if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 583                 *(u8 *) mailbox->buf         = !!reset_qkey_viols << 6;
 584                 ((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
 585         } else {
 586                 ((u8 *) mailbox->buf)[3]     = !!reset_qkey_viols;
 587                 ((__be32 *) mailbox->buf)[1] = cpu_to_be32(cap_mask);
 588         }
 589
 590         err = mlx4_cmd(dev->dev, mailbox->dma, port, MLX4_SET_PORT_IB_OPCODE,
 591                        MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
 592                        MLX4_CMD_WRAPPED);
 593
 594         mlx4_free_cmd_mailbox(dev->dev, mailbox);
 595         return err;
 596 }
 597
 598 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 599                                struct ib_port_modify *props)
 600 {
 601         struct mlx4_ib_dev *mdev = to_mdev(ibdev);
 602         u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 603         struct ib_port_attr attr;
 604         u32 cap_mask;
 605         int err;
 606
 607         /* return OK if this is RoCE. CM calls ib_modify_port() regardless
 608          * of whether port link layer is ETH or IB. For ETH ports, qkey
 609          * violations and port capabilities are not meaningful.
 610          */
 611         if (is_eth)
 612                 return 0;
 613
 614         mutex_lock(&mdev->cap_mask_mutex);
 615
 616         err = mlx4_ib_query_port(ibdev, port, &attr);
 617         if (err)
 618                 goto out;
 619
 620         cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 621                 ~props->clr_port_cap_mask;
 622
 623         err = mlx4_ib_SET_PORT(mdev, port,
 624                                !!(mask & IB_PORT_RESET_QKEY_CNTR),
 625                                cap_mask);
 626
 627 out:
 628         mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 629         return err;
 630 }
 631
 632 static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 633                                                   struct ib_udata *udata)
 634 {
 635         struct mlx4_ib_dev *dev = to_mdev(ibdev);
 636         struct mlx4_ib_ucontext *context;
 637         struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 638         struct mlx4_ib_alloc_ucontext_resp resp;
 639         int err;
 640
 641         if (!dev->ib_active)
 642                 return ERR_PTR(-EAGAIN);
 643
 644         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
 645                 resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
 646                 resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
 647                 resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 648         } else {
 649                 resp.dev_caps         = dev->dev->caps.userspace_caps;
 650                 resp.qp_tab_size      = dev->dev->caps.num_qps;
 651                 resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
 652                 resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
 653                 resp.cqe_size         = dev->dev->caps.cqe_size;
 654         }
 655
 656         context = kmalloc(sizeof *context, GFP_KERNEL);
 657         if (!context)
 658                 return ERR_PTR(-ENOMEM);
 659
 660         err = mlx4_uar_alloc(to_mdev(ibdev)->dev, &context->uar);
 661         if (err) {
 662                 kfree(context);
 663                 return ERR_PTR(err);
 664         }
 665
 666         INIT_LIST_HEAD(&context->db_page_list);
 667         mutex_init(&context->db_page_mutex);
 668
 669         if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
 670                 err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
 671         else
 672                 err = ib_copy_to_udata(udata, &resp, sizeof(resp));
 673
 674         if (err) {
 675                 mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 676                 kfree(context);
 677                 return ERR_PTR(-EFAULT);
 678         }
 679
 680         return &context->ibucontext;
 681 }
 682
 683 static int mlx4_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
 684 {
 685         struct mlx4_ib_ucontext *context = to_mucontext(ibcontext);
 686
 687         mlx4_uar_free(to_mdev(ibcontext->device)->dev, &context->uar);
 688         kfree(context);
 689
 690         return 0;
 691 }
 692
 693 static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 694 {
 695         struct mlx4_ib_dev *dev = to_mdev(context->device);
 696
 697         if (vma->vm_end - vma->vm_start != PAGE_SIZE)
 698                 return -EINVAL;
 699
 700         if (vma->vm_pgoff == 0) {
 701                 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 702
 703                 if (io_remap_pfn_range(vma, vma->vm_start,
 704                                        to_mucontext(context)->uar.pfn,
 705                                        PAGE_SIZE, vma->vm_page_prot))
 706                         return -EAGAIN;
 707         } else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
 708                 vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 709
 710                 if (io_remap_pfn_range(vma, vma->vm_start,
 711                                        to_mucontext(context)->uar.pfn +
 712                                        dev->dev->caps.num_uars,
 713                                        PAGE_SIZE, vma->vm_page_prot))
 714                         return -EAGAIN;
 715         } else
 716                 return -EINVAL;
 717
 718         return 0;
 719 }
 720
 721 static struct ib_pd *mlx4_ib_alloc_pd(struct ib_device *ibdev,
 722                                       struct ib_ucontext *context,
 723                                       struct ib_udata *udata)
 724 {
 725         struct mlx4_ib_pd *pd;
 726         int err;
 727
 728         pd = kmalloc(sizeof *pd, GFP_KERNEL);
 729         if (!pd)
 730                 return ERR_PTR(-ENOMEM);
 731
 732         err = mlx4_pd_alloc(to_mdev(ibdev)->dev, &pd->pdn);
 733         if (err) {
 734                 kfree(pd);
 735                 return ERR_PTR(err);
 736         }
 737
 738         if (context)
 739                 if (ib_copy_to_udata(udata, &pd->pdn, sizeof (__u32))) {
 740                         mlx4_pd_free(to_mdev(ibdev)->dev, pd->pdn);
 741                         kfree(pd);
 742                         return ERR_PTR(-EFAULT);
 743                 }
 744
 745         return &pd->ibpd;
 746 }
 747
 748 static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 749 {
 750         mlx4_pd_free(to_mdev(pd->device)->dev, to_mpd(pd)->pdn);
 751         kfree(pd);
 752
 753         return 0;
 754 }
 755
 756 static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
 757                                           struct ib_ucontext *context,
 758                                           struct ib_udata *udata)
 759 {
 760         struct mlx4_ib_xrcd *xrcd;
 761         struct ib_cq_init_attr cq_attr = {};
 762         int err;
 763
 764         if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
 765                 return ERR_PTR(-ENOSYS);
 766
 767         xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
 768         if (!xrcd)
 769                 return ERR_PTR(-ENOMEM);
 770
 771         err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
 772         if (err)
 773                 goto err1;
 774
 775         xrcd->pd = ib_alloc_pd(ibdev);
 776         if (IS_ERR(xrcd->pd)) {
 777                 err = PTR_ERR(xrcd->pd);
 778                 goto err2;
 779         }
 780
 781         cq_attr.cqe = 1;
 782         xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, &cq_attr);
 783         if (IS_ERR(xrcd->cq)) {
 784                 err = PTR_ERR(xrcd->cq);
 785                 goto err3;
 786         }
 787
 788         return &xrcd->ibxrcd;
 789
 790 err3:
 791         ib_dealloc_pd(xrcd->pd);
 792 err2:
 793         mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
 794 err1:
 795         kfree(xrcd);
 796         return ERR_PTR(err);
 797 }
 798
 799 static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
 800 {
 801         ib_destroy_cq(to_mxrcd(xrcd)->cq);
 802         ib_dealloc_pd(to_mxrcd(xrcd)->pd);
 803         mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
 804         kfree(xrcd);
 805
 806         return 0;
 807 }
 808
 809 static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
 810 {
 811         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
 812         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
 813         struct mlx4_ib_gid_entry *ge;
 814
 815         ge = kzalloc(sizeof *ge, GFP_KERNEL);
 816         if (!ge)
 817                 return -ENOMEM;
 818
 819         ge->gid = *gid;
 820         if (mlx4_ib_add_mc(mdev, mqp, gid)) {
 821                 ge->port = mqp->port;
 822                 ge->added = 1;
 823         }
 824
 825         mutex_lock(&mqp->mutex);
 826         list_add_tail(&ge->list, &mqp->gid_list);
 827         mutex_unlock(&mqp->mutex);
 828
 829         return 0;
 830 }
 831
 832 int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
 833                    union ib_gid *gid)
 834 {
 835         struct net_device *ndev;
 836         int ret = 0;
 837
 838         if (!mqp->port)
 839                 return 0;
 840
 841         spin_lock_bh(&mdev->iboe.lock);
 842         ndev = mdev->iboe.netdevs[mqp->port - 1];
 843         if (ndev)
 844                 dev_hold(ndev);
 845         spin_unlock_bh(&mdev->iboe.lock);
 846
 847         if (ndev) {
 848                 ret = 1;
 849                 dev_put(ndev);
 850         }
 851
 852         return ret;
 853 }
 854
 855 struct mlx4_ib_steering {
 856         struct list_head list;
 857         struct mlx4_flow_reg_id reg_id;
 858         union ib_gid gid;
 859 };
 860
 861 static int parse_flow_attr(struct mlx4_dev *dev,
 862                            u32 qp_num,
 863                            union ib_flow_spec *ib_spec,
 864                            struct _rule_hw *mlx4_spec)
 865 {
 866         enum mlx4_net_trans_rule_id type;
 867
 868         switch (ib_spec->type) {
 869         case IB_FLOW_SPEC_ETH:
 870                 type = MLX4_NET_TRANS_RULE_ID_ETH;
 871                 memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
 872                        ETH_ALEN);
 873                 memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
 874                        ETH_ALEN);
 875                 mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
 876                 mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
 877                 break;
 878         case IB_FLOW_SPEC_IB:
 879                 type = MLX4_NET_TRANS_RULE_ID_IB;
 880                 mlx4_spec->ib.l3_qpn =
 881                         cpu_to_be32(qp_num);
 882                 mlx4_spec->ib.qpn_mask =
 883                         cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
 884                 break;
 885
 886
 887         case IB_FLOW_SPEC_IPV4:
 888                 type = MLX4_NET_TRANS_RULE_ID_IPV4;
 889                 mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
 890                 mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
 891                 mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
 892                 mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
 893                 break;
 894
 895         case IB_FLOW_SPEC_TCP:
 896         case IB_FLOW_SPEC_UDP:
 897                 type = ib_spec->type == IB_FLOW_SPEC_TCP ?
 898                                         MLX4_NET_TRANS_RULE_ID_TCP :
 899                                         MLX4_NET_TRANS_RULE_ID_UDP;
 900                 mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
 901                 mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
 902                 mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
 903                 mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
 904                 break;
 905
 906         default:
 907                 return -EINVAL;
 908         }
 909         if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
 910             mlx4_hw_rule_sz(dev, type) < 0)
 911                 return -EINVAL;
 912         mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
 913         mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
 914         return mlx4_hw_rule_sz(dev, type);
 915 }
 916
 917 struct default_rules {
 918         __u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 919         __u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
 920         __u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
 921         __u8  link_layer;
 922 };
 923 static const struct default_rules default_table[] = {
 924         {
 925                 .mandatory_fields = {IB_FLOW_SPEC_IPV4},
 926                 .mandatory_not_fields = {IB_FLOW_SPEC_ETH},
 927                 .rules_create_list = {IB_FLOW_SPEC_IB},
 928                 .link_layer = IB_LINK_LAYER_INFINIBAND
 929         }
 930 };
 931
 932 static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
 933                                          struct ib_flow_attr *flow_attr)
 934 {
 935         int i, j, k;
 936         void *ib_flow;
 937         const struct default_rules *pdefault_rules = default_table;
 938         u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
 939
 940         for (i = 0; i < ARRAY_SIZE(default_table); i++, pdefault_rules++) {
 941                 __u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
 942                 memset(&field_types, 0, sizeof(field_types));
 943
 944                 if (link_layer != pdefault_rules->link_layer)
 945                         continue;
 946
 947                 ib_flow = flow_attr + 1;
 948                 /* we assume the specs are sorted */
 949                 for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
 950                      j < flow_attr->num_of_specs; k++) {
 951                         union ib_flow_spec *current_flow =
 952                                 (union ib_flow_spec *)ib_flow;
 953
 954                         /* same layer but different type */
 955                         if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
 956                              (pdefault_rules->mandatory_fields[k] &
 957                               IB_FLOW_SPEC_LAYER_MASK)) &&
 958                             (current_flow->type !=
 959                              pdefault_rules->mandatory_fields[k]))
 960                                 goto out;
 961
 962                         /* same layer, try match next one */
 963                         if (current_flow->type ==
 964                             pdefault_rules->mandatory_fields[k]) {
 965                                 j++;
 966                                 ib_flow +=
 967                                         ((union ib_flow_spec *)ib_flow)->size;
 968                         }
 969                 }
 970
 971                 ib_flow = flow_attr + 1;
 972                 for (j = 0; j < flow_attr->num_of_specs;
 973                      j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
 974                         for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
 975                                 /* same layer and same type */
 976                                 if (((union ib_flow_spec *)ib_flow)->type ==
 977                                     pdefault_rules->mandatory_not_fields[k])
 978                                         goto out;
 979
 980                 return i;
 981         }
 982 out:
 983         return -1;
 984 }
 985
 986 static int __mlx4_ib_create_default_rules(
 987                 struct mlx4_ib_dev *mdev,
 988                 struct ib_qp *qp,
 989                 const struct default_rules *pdefault_rules,
 990                 struct _rule_hw *mlx4_spec) {
 991         int size = 0;
 992         int i;
 993
 994         for (i = 0; i < ARRAY_SIZE(pdefault_rules->rules_create_list); i++) {
 995                 int ret;
 996                 union ib_flow_spec ib_spec;
 997                 switch (pdefault_rules->rules_create_list[i]) {
 998                 case 0:
 999                         /* no rule */
1000                         continue;
1001                 case IB_FLOW_SPEC_IB:
1002                         ib_spec.type = IB_FLOW_SPEC_IB;
1003                         ib_spec.size = sizeof(struct ib_flow_spec_ib);
1004
1005                         break;
1006                 default:
1007                         /* invalid rule */
1008                         return -EINVAL;
1009                 }
1010                 /* We must put empty rule, qpn is being ignored */
1011                 ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
1012                                       mlx4_spec);
1013                 if (ret < 0) {
1014                         pr_info("invalid parsing\n");
1015                         return -EINVAL;
1016                 }
1017
1018                 mlx4_spec = (void *)mlx4_spec + ret;
1019                 size += ret;
1020         }
1021         return size;
1022 }
1023
1024 static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1025                           int domain,
1026                           enum mlx4_net_trans_promisc_mode flow_type,
1027                           u64 *reg_id)
1028 {
1029         int ret, i;
1030         int size = 0;
1031         void *ib_flow;
1032         struct mlx4_ib_dev *mdev = to_mdev(qp->device);
1033         struct mlx4_cmd_mailbox *mailbox;
1034         struct mlx4_net_trans_rule_hw_ctrl *ctrl;
1035         int default_flow;
1036
1037         static const u16 __mlx4_domain[] = {
1038                 [IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
1039                 [IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
1040                 [IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
1041                 [IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
1042         };
1043
1044         if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
1045                 pr_err("Invalid priority value %d\n", flow_attr->priority);
1046                 return -EINVAL;
1047         }
1048
1049         if (domain >= IB_FLOW_DOMAIN_NUM) {
1050                 pr_err("Invalid domain value %d\n", domain);
1051                 return -EINVAL;
1052         }
1053
1054         if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
1055                 return -EINVAL;
1056
1057         mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
1058         if (IS_ERR(mailbox))
1059                 return PTR_ERR(mailbox);
1060         ctrl = mailbox->buf;
1061
1062         ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
1063                                  flow_attr->priority);
1064         ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
1065         ctrl->port = flow_attr->port;
1066         ctrl->qpn = cpu_to_be32(qp->qp_num);
1067
1068         ib_flow = flow_attr + 1;
1069         size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
1070         /* Add default flows */
1071         default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
1072         if (default_flow >= 0) {
1073                 ret = __mlx4_ib_create_default_rules(
1074                                 mdev, qp, default_table + default_flow,
1075                                 mailbox->buf + size);
1076                 if (ret < 0) {
1077                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1078                         return -EINVAL;
1079                 }
1080                 size += ret;
1081         }
1082         for (i = 0; i < flow_attr->num_of_specs; i++) {
1083                 ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
1084                                       mailbox->buf + size);
1085                 if (ret < 0) {
1086                         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1087                         return -EINVAL;
1088                 }
1089                 ib_flow += ((union ib_flow_spec *) ib_flow)->size;
1090                 size += ret;
1091         }
1092
1093         ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
1094                            MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
1095                            MLX4_CMD_NATIVE);
1096         if (ret == -ENOMEM)
1097                 pr_err("mcg table is full. Fail to register network rule.\n");
1098         else if (ret == -ENXIO)
1099                 pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
1100         else if (ret)
1101                 pr_err("Invalid argumant. Fail to register network rule.\n");
1102
1103         mlx4_free_cmd_mailbox(mdev->dev, mailbox);
1104         return ret;
1105 }
1106
1107 static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
1108 {
1109         int err;
1110         err = mlx4_cmd(dev, reg_id, 0, 0,
1111                        MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
1112                        MLX4_CMD_NATIVE);
1113         if (err)
1114                 pr_err("Fail to detach network rule. registration id = 0x%llx\n",
1115                        reg_id);
1116         return err;
1117 }
1118
1119 static int mlx4_ib_tunnel_steer_add(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
1120                                     u64 *reg_id)
1121 {
1122         void *ib_flow;
1123         union ib_flow_spec *ib_spec;
1124         struct mlx4_dev *dev = to_mdev(qp->device)->dev;
1125         int err = 0;
1126
1127         if (dev->caps.tunnel_offload_mode != MLX4_TUNNEL_OFFLOAD_MODE_VXLAN ||
1128             dev->caps.dmfs_high_steer_mode == MLX4_STEERING_DMFS_A0_STATIC)
1129                 return 0; /* do nothing */
1130
1131         ib_flow = flow_attr + 1;
1132         ib_spec = (union ib_flow_spec *)ib_flow;
1133
1134         if (ib_spec->type !=  IB_FLOW_SPEC_ETH || flow_attr->num_of_specs != 1)
1135                 return 0; /* do nothing */
1136
1137         err = mlx4_tunnel_steer_add(to_mdev(qp->device)->dev, ib_spec->eth.val.dst_mac,
1138                                     flow_attr->port, qp->qp_num,
1139                                     MLX4_DOMAIN_UVERBS | (flow_attr->priority & 0xff),
1140                                     reg_id);
1141         return err;
1142 }
1143
1144 static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
1145                                     struct ib_flow_attr *flow_attr,
1146                                     int domain)
1147 {
1148         int err = 0, i = 0, j = 0;
1149         struct mlx4_ib_flow *mflow;
1150         enum mlx4_net_trans_promisc_mode type[2];
1151         struct mlx4_dev *dev = (to_mdev(qp->device))->dev;
1152         int is_bonded = mlx4_is_bonded(dev);
1153
1154         memset(type, 0, sizeof(type));
1155
1156         mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
1157         if (!mflow) {
1158                 err = -ENOMEM;
1159                 goto err_free;
1160         }
1161
1162         switch (flow_attr->type) {
1163         case IB_FLOW_ATTR_NORMAL:
1164                 type[0] = MLX4_FS_REGULAR;
1165                 break;
1166
1167         case IB_FLOW_ATTR_ALL_DEFAULT:
1168                 type[0] = MLX4_FS_ALL_DEFAULT;
1169                 break;
1170
1171         case IB_FLOW_ATTR_MC_DEFAULT:
1172                 type[0] = MLX4_FS_MC_DEFAULT;
1173                 break;
1174
1175         case IB_FLOW_ATTR_SNIFFER:
1176                 type[0] = MLX4_FS_UC_SNIFFER;
1177                 type[1] = MLX4_FS_MC_SNIFFER;
1178                 break;
1179
1180         default:
1181                 err = -EINVAL;
1182                 goto err_free;
1183         }
1184
1185         while (i < ARRAY_SIZE(type) && type[i]) {
1186                 err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
1187                                             &mflow->reg_id[i].id);
1188                 if (err)
1189                         goto err_create_flow;
1190                 if (is_bonded) {
1191                         /* Application always sees one port so the mirror rule
1192                          * must be on port #2
1193                          */
1194                         flow_attr->port = 2;
1195                         err = __mlx4_ib_create_flow(qp, flow_attr,
1196                                                     domain, type[j],
1197                                                     &mflow->reg_id[j].mirror);
1198                         flow_attr->port = 1;
1199                         if (err)
1200                                 goto err_create_flow;
1201                         j++;
1202                 }
1203
1204                 i++;
1205         }
1206
1207         if (i < ARRAY_SIZE(type) && flow_attr->type == IB_FLOW_ATTR_NORMAL) {
1208                 err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1209                                                &mflow->reg_id[i].id);
1210                 if (err)
1211                         goto err_create_flow;
1212
1213                 if (is_bonded) {
1214                         flow_attr->port = 2;
1215                         err = mlx4_ib_tunnel_steer_add(qp, flow_attr,
1216                                                        &mflow->reg_id[j].mirror);
1217                         flow_attr->port = 1;
1218                         if (err)
1219                                 goto err_create_flow;
1220                         j++;
1221                 }
1222                 /* function to create mirror rule */
1223                 i++;
1224         }
1225
1226         return &mflow->ibflow;
1227
1228 err_create_flow:
1229         while (i) {
1230                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1231                                              mflow->reg_id[i].id);
1232                 i--;
1233         }
1234
1235         while (j) {
1236                 (void)__mlx4_ib_destroy_flow(to_mdev(qp->device)->dev,
1237                                              mflow->reg_id[j].mirror);
1238                 j--;
1239         }
1240 err_free:
1241         kfree(mflow);
1242         return ERR_PTR(err);
1243 }
1244
1245 static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
1246 {
1247         int err, ret = 0;
1248         int i = 0;
1249         struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
1250         struct mlx4_ib_flow *mflow = to_mflow(flow_id);
1251
1252         while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i].id) {
1253                 err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i].id);
1254                 if (err)
1255                         ret = err;
1256                 if (mflow->reg_id[i].mirror) {
1257                         err = __mlx4_ib_destroy_flow(mdev->dev,
1258                                                      mflow->reg_id[i].mirror);
1259                         if (err)
1260                                 ret = err;
1261                 }
1262                 i++;
1263         }
1264
1265         kfree(mflow);
1266         return ret;
1267 }
1268
1269 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1270 {
1271         int err;
1272         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1273         struct mlx4_dev *dev = mdev->dev;
1274         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1275         struct mlx4_ib_steering *ib_steering = NULL;
1276         enum mlx4_protocol prot = MLX4_PROT_IB_IPV6;
1277         struct mlx4_flow_reg_id reg_id;
1278
1279         if (mdev->dev->caps.steering_mode ==
1280             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1281                 ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
1282                 if (!ib_steering)
1283                         return -ENOMEM;
1284         }
1285
1286         err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
1287                                     !!(mqp->flags &
1288                                        MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1289                                     prot, &reg_id.id);
1290         if (err) {
1291                 pr_err("multicast attach op failed, err %d\n", err);
1292                 goto err_malloc;
1293         }
1294
1295         reg_id.mirror = 0;
1296         if (mlx4_is_bonded(dev)) {
1297                 err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw,
1298                                             (mqp->port == 1) ? 2 : 1,
1299                                             !!(mqp->flags &
1300                                             MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
1301                                             prot, &reg_id.mirror);
1302                 if (err)
1303                         goto err_add;
1304         }
1305
1306         err = add_gid_entry(ibqp, gid);
1307         if (err)
1308                 goto err_add;
1309
1310         if (ib_steering) {
1311                 memcpy(ib_steering->gid.raw, gid->raw, 16);
1312                 ib_steering->reg_id = reg_id;
1313                 mutex_lock(&mqp->mutex);
1314                 list_add(&ib_steering->list, &mqp->steering_rules);
1315                 mutex_unlock(&mqp->mutex);
1316         }
1317         return 0;
1318
1319 err_add:
1320         mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1321                               prot, reg_id.id);
1322         if (reg_id.mirror)
1323                 mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1324                                       prot, reg_id.mirror);
1325 err_malloc:
1326         kfree(ib_steering);
1327
1328         return err;
1329 }
1330
1331 static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
1332 {
1333         struct mlx4_ib_gid_entry *ge;
1334         struct mlx4_ib_gid_entry *tmp;
1335         struct mlx4_ib_gid_entry *ret = NULL;
1336
1337         list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
1338                 if (!memcmp(raw, ge->gid.raw, 16)) {
1339                         ret = ge;
1340                         break;
1341                 }
1342         }
1343
1344         return ret;
1345 }
1346
1347 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
1348 {
1349         int err;
1350         struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
1351         struct mlx4_dev *dev = mdev->dev;
1352         struct mlx4_ib_qp *mqp = to_mqp(ibqp);
1353         struct net_device *ndev;
1354         struct mlx4_ib_gid_entry *ge;
1355         struct mlx4_flow_reg_id reg_id = {0, 0};
1356         enum mlx4_protocol prot =  MLX4_PROT_IB_IPV6;
1357
1358         if (mdev->dev->caps.steering_mode ==
1359             MLX4_STEERING_MODE_DEVICE_MANAGED) {
1360                 struct mlx4_ib_steering *ib_steering;
1361
1362                 mutex_lock(&mqp->mutex);
1363                 list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
1364                         if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
1365                                 list_del(&ib_steering->list);
1366                                 break;
1367                         }
1368                 }
1369                 mutex_unlock(&mqp->mutex);
1370                 if (&ib_steering->list == &mqp->steering_rules) {
1371                         pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
1372                         return -EINVAL;
1373                 }
1374                 reg_id = ib_steering->reg_id;
1375                 kfree(ib_steering);
1376         }
1377
1378         err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1379                                     prot, reg_id.id);
1380         if (err)
1381                 return err;
1382
1383         if (mlx4_is_bonded(dev)) {
1384                 err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
1385                                             prot, reg_id.mirror);
1386                 if (err)
1387                         return err;
1388         }
1389
1390         mutex_lock(&mqp->mutex);
1391         ge = find_gid_entry(mqp, gid->raw);
1392         if (ge) {
1393                 spin_lock_bh(&mdev->iboe.lock);
1394                 ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
1395                 if (ndev)
1396                         dev_hold(ndev);
1397                 spin_unlock_bh(&mdev->iboe.lock);
1398                 if (ndev)
1399                         dev_put(ndev);
1400                 list_del(&ge->list);
1401                 kfree(ge);
1402         } else
1403                 pr_warn("could not find mgid entry\n");
1404
1405         mutex_unlock(&mqp->mutex);
1406
1407         return 0;
1408 }
1409
1410 static int init_node_data(struct mlx4_ib_dev *dev)
1411 {
1412         struct ib_smp *in_mad  = NULL;
1413         struct ib_smp *out_mad = NULL;
1414         int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
1415         int err = -ENOMEM;
1416
1417         in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
1418         out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1419         if (!in_mad || !out_mad)
1420                 goto out;
1421
1422         init_query_mad(in_mad);
1423         in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1424         if (mlx4_is_master(dev->dev))
1425                 mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
1426
1427         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1428         if (err)
1429                 goto out;
1430
1431         memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
1432
1433         in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1434
1435         err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
1436         if (err)
1437                 goto out;
1438
1439         dev->dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1440         memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1441
1442 out:
1443         kfree(in_mad);
1444         kfree(out_mad);
1445         return err;
1446 }
1447
1448 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1449                         char *buf)
1450 {
1451         struct mlx4_ib_dev *dev =
1452                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1453         return sprintf(buf, "MT%d\n", dev->dev->persist->pdev->device);
1454 }
1455
1456 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
1457                            char *buf)
1458 {
1459         struct mlx4_ib_dev *dev =
1460                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1461         return sprintf(buf, "%d.%d.%d\n", (int) (dev->dev->caps.fw_ver >> 32),
1462                        (int) (dev->dev->caps.fw_ver >> 16) & 0xffff,
1463                        (int) dev->dev->caps.fw_ver & 0xffff);
1464 }
1465
1466 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1467                         char *buf)
1468 {
1469         struct mlx4_ib_dev *dev =
1470                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1471         return sprintf(buf, "%x\n", dev->dev->rev_id);
1472 }
1473
1474 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1475                           char *buf)
1476 {
1477         struct mlx4_ib_dev *dev =
1478                 container_of(device, struct mlx4_ib_dev, ib_dev.dev);
1479         return sprintf(buf, "%.*s\n", MLX4_BOARD_ID_LEN,
1480                        dev->dev->board_id);
1481 }
1482
1483 static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
1484 static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
1485 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
1486 static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
1487
1488 static struct device_attribute *mlx4_class_attributes[] = {
1489         &dev_attr_hw_rev,
1490         &dev_attr_fw_ver,
1491         &dev_attr_hca_type,
1492         &dev_attr_board_id
1493 };
1494
1495 static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
1496                                      struct net_device *dev)
1497 {
1498         memcpy(eui, dev->dev_addr, 3);
1499         memcpy(eui + 5, dev->dev_addr + 3, 3);
1500         if (vlan_id < 0x1000) {
1501                 eui[3] = vlan_id >> 8;
1502                 eui[4] = vlan_id & 0xff;
1503         } else {
1504                 eui[3] = 0xff;
1505                 eui[4] = 0xfe;
1506         }
1507         eui[0] ^= 2;
1508 }
1509
1510 static void update_gids_task(struct work_struct *work)
1511 {
1512         struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
1513         struct mlx4_cmd_mailbox *mailbox;
1514         union ib_gid *gids;
1515         int err;
1516         struct mlx4_dev *dev = gw->dev->dev;
1517         int is_bonded = mlx4_is_bonded(dev);
1518
1519         if (!gw->dev->ib_active)
1520                 return;
1521
1522         mailbox = mlx4_alloc_cmd_mailbox(dev);
1523         if (IS_ERR(mailbox)) {
1524                 pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
1525                 return;
1526         }
1527
1528         gids = mailbox->buf;
1529         memcpy(gids, gw->gids, sizeof gw->gids);
1530
1531         err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1532                        MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1533                        MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
1534         if (err)
1535                 pr_warn("set port command failed\n");
1536         else
1537                 if ((gw->port == 1) || !is_bonded)
1538                         mlx4_ib_dispatch_event(gw->dev,
1539                                                is_bonded ? 1 : gw->port,
1540                                                IB_EVENT_GID_CHANGE);
1541
1542         mlx4_free_cmd_mailbox(dev, mailbox);
1543         kfree(gw);
1544 }
1545
1546 static void reset_gids_task(struct work_struct *work)
1547 {
1548         struct update_gid_work *gw =
1549                         container_of(work, struct update_gid_work, work);
1550         struct mlx4_cmd_mailbox *mailbox;
1551         union ib_gid *gids;
1552         int err;
1553         struct mlx4_dev *dev = gw->dev->dev;
1554
1555         if (!gw->dev->ib_active)
1556                 return;
1557
1558         mailbox = mlx4_alloc_cmd_mailbox(dev);
1559         if (IS_ERR(mailbox)) {
1560                 pr_warn("reset gid table failed\n");
1561                 goto free;
1562         }
1563
1564         gids = mailbox->buf;
1565         memcpy(gids, gw->gids, sizeof(gw->gids));
1566
1567         if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
1568                                     IB_LINK_LAYER_ETHERNET) {
1569                 err = mlx4_cmd(dev, mailbox->dma,
1570                                MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
1571                                MLX4_SET_PORT_ETH_OPCODE, MLX4_CMD_SET_PORT,
1572                                MLX4_CMD_TIME_CLASS_B,
1573                                MLX4_CMD_WRAPPED);
1574                 if (err)
1575                         pr_warn("set port %d command failed\n", gw->port);
1576         }
1577
1578         mlx4_free_cmd_mailbox(dev, mailbox);
1579 free:
1580         kfree(gw);
1581 }
1582
1583 static int update_gid_table(struct mlx4_ib_dev *dev, int port,
1584                             union ib_gid *gid, int clear,
1585                             int default_gid)
1586 {
1587         struct update_gid_work *work;
1588         int i;
1589         int need_update = 0;
1590         int free = -1;
1591         int found = -1;
1592         int max_gids;
1593
1594         if (default_gid) {
1595                 free = 0;
1596         } else {
1597                 max_gids = dev->dev->caps.gid_table_len[port];
1598                 for (i = 1; i < max_gids; ++i) {
1599                         if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
1600                                     sizeof(*gid)))
1601                                 found = i;
1602
1603                         if (clear) {
1604                                 if (found >= 0) {
1605                                         need_update = 1;
1606                                         dev->iboe.gid_table[port - 1][found] =
1607                                                 zgid;
1608                                         break;
1609                                 }
1610                         } else {
1611                                 if (found >= 0)
1612                                         break;
1613
1614                                 if (free < 0 &&
1615                                     !memcmp(&dev->iboe.gid_table[port - 1][i],
1616                                             &zgid, sizeof(*gid)))
1617                                         free = i;
1618                         }
1619                 }
1620         }
1621
1622         if (found == -1 && !clear && free >= 0) {
1623                 dev->iboe.gid_table[port - 1][free] = *gid;
1624                 need_update = 1;
1625         }
1626
1627         if (!need_update)
1628                 return 0;
1629
1630         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1631         if (!work)
1632                 return -ENOMEM;
1633
1634         memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
1635         INIT_WORK(&work->work, update_gids_task);
1636         work->port = port;
1637         work->dev = dev;
1638         queue_work(wq, &work->work);
1639
1640         return 0;
1641 }
1642
1643 static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
1644 {
1645         gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
1646         mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
1647 }
1648
1649
1650 static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
1651 {
1652         struct update_gid_work *work;
1653
1654         work = kzalloc(sizeof(*work), GFP_ATOMIC);
1655         if (!work)
1656                 return -ENOMEM;
1657
1658         memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
1659         memset(work->gids, 0, sizeof(work->gids));
1660         INIT_WORK(&work->work, reset_gids_task);
1661         work->dev = dev;
1662         work->port = port;
1663         queue_work(wq, &work->work);
1664         return 0;
1665 }
1666
1667 static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
1668                               struct mlx4_ib_dev *ibdev, union ib_gid *gid)
1669 {
1670         struct mlx4_ib_iboe *iboe;
1671         int port = 0;
1672         struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
1673                                 rdma_vlan_dev_real_dev(event_netdev) :
1674                                 event_netdev;
1675         union ib_gid default_gid;
1676
1677         mlx4_make_default_gid(real_dev, &default_gid);
1678
1679         if (!memcmp(gid, &default_gid, sizeof(*gid)))
1680                 return 0;
1681
1682         if (event != NETDEV_DOWN && event != NETDEV_UP)
1683                 return 0;
1684
1685         if ((real_dev != event_netdev) &&
1686             (event == NETDEV_DOWN) &&
1687             rdma_link_local_addr((struct in6_addr *)gid))
1688                 return 0;
1689
1690         iboe = &ibdev->iboe;
1691         spin_lock_bh(&iboe->lock);
1692
1693         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1694                 if ((netif_is_bond_master(real_dev) &&
1695                      (real_dev == iboe->masters[port - 1])) ||
1696                      (!netif_is_bond_master(real_dev) &&
1697                      (real_dev == iboe->netdevs[port - 1])))
1698                         update_gid_table(ibdev, port, gid,
1699                                          event == NETDEV_DOWN, 0);
1700
1701         spin_unlock_bh(&iboe->lock);
1702         return 0;
1703
1704 }
1705
1706 static u8 mlx4_ib_get_dev_port(struct net_device *dev,
1707                                struct mlx4_ib_dev *ibdev)
1708 {
1709         u8 port = 0;
1710         struct mlx4_ib_iboe *iboe;
1711         struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
1712                                 rdma_vlan_dev_real_dev(dev) : dev;
1713
1714         iboe = &ibdev->iboe;
1715
1716         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
1717                 if ((netif_is_bond_master(real_dev) &&
1718                      (real_dev == iboe->masters[port - 1])) ||
1719                      (!netif_is_bond_master(real_dev) &&
1720                      (real_dev == iboe->netdevs[port - 1])))
1721                         break;
1722
1723         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1724                 return 0;
1725         else
1726                 return port;
1727 }
1728
1729 static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
1730                                 void *ptr)
1731 {
1732         struct mlx4_ib_dev *ibdev;
1733         struct in_ifaddr *ifa = ptr;
1734         union ib_gid gid;
1735         struct net_device *event_netdev = ifa->ifa_dev->dev;
1736
1737         ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
1738
1739         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
1740
1741         mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
1742         return NOTIFY_DONE;
1743 }
1744
1745 #if IS_ENABLED(CONFIG_IPV6)
1746 static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
1747                                 void *ptr)
1748 {
1749         struct mlx4_ib_dev *ibdev;
1750         struct inet6_ifaddr *ifa = ptr;
1751         union  ib_gid *gid = (union ib_gid *)&ifa->addr;
1752         struct net_device *event_netdev = ifa->idev->dev;
1753
1754         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
1755
1756         mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
1757         return NOTIFY_DONE;
1758 }
1759 #endif
1760
1761 #define MLX4_IB_INVALID_MAC     ((u64)-1)
1762 static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
1763                                struct net_device *dev,
1764                                int port)
1765 {
1766         u64 new_smac = 0;
1767         u64 release_mac = MLX4_IB_INVALID_MAC;
1768         struct mlx4_ib_qp *qp;
1769
1770         read_lock(&dev_base_lock);
1771         new_smac = mlx4_mac_to_u64(dev->dev_addr);
1772         read_unlock(&dev_base_lock);
1773
1774         atomic64_set(&ibdev->iboe.mac[port - 1], new_smac);
1775
1776         /* no need for update QP1 and mac registration in non-SRIOV */
1777         if (!mlx4_is_mfunc(ibdev->dev))
1778                 return;
1779
1780         mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
1781         qp = ibdev->qp1_proxy[port - 1];
1782         if (qp) {
1783                 int new_smac_index;
1784                 u64 old_smac;
1785                 struct mlx4_update_qp_params update_params;
1786
1787                 mutex_lock(&qp->mutex);
1788                 old_smac = qp->pri.smac;
1789                 if (new_smac == old_smac)
1790                         goto unlock;
1791
1792                 new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
1793
1794                 if (new_smac_index < 0)
1795                         goto unlock;
1796
1797                 update_params.smac_index = new_smac_index;
1798                 if (mlx4_update_qp(ibdev->dev, qp->mqp.qpn, MLX4_UPDATE_QP_SMAC,
1799                                    &update_params)) {
1800                         release_mac = new_smac;
1801                         goto unlock;
1802                 }
1803                 /* if old port was zero, no mac was yet registered for this QP */
1804                 if (qp->pri.smac_port)
1805                         release_mac = old_smac;
1806                 qp->pri.smac = new_smac;
1807                 qp->pri.smac_port = port;
1808                 qp->pri.smac_index = new_smac_index;
1809         }
1810
1811 unlock:
1812         if (release_mac != MLX4_IB_INVALID_MAC)
1813                 mlx4_unregister_mac(ibdev->dev, port, release_mac);
1814         if (qp)
1815                 mutex_unlock(&qp->mutex);
1816         mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
1817 }
1818
1819 static void mlx4_ib_get_dev_addr(struct net_device *dev,
1820                                  struct mlx4_ib_dev *ibdev, u8 port)
1821 {
1822         struct in_device *in_dev;
1823 #if IS_ENABLED(CONFIG_IPV6)
1824         struct inet6_dev *in6_dev;
1825         union ib_gid  *pgid;
1826         struct inet6_ifaddr *ifp;
1827         union ib_gid default_gid;
1828 #endif
1829         union ib_gid gid;
1830
1831
1832         if ((port == 0) || (port > ibdev->dev->caps.num_ports))
1833                 return;
1834
1835         /* IPv4 gids */
1836         in_dev = in_dev_get(dev);
1837         if (in_dev) {
1838                 for_ifa(in_dev) {
1839                         /*ifa->ifa_address;*/
1840                         ipv6_addr_set_v4mapped(ifa->ifa_address,
1841                                                (struct in6_addr *)&gid);
1842                         update_gid_table(ibdev, port, &gid, 0, 0);
1843                 }
1844                 endfor_ifa(in_dev);
1845                 in_dev_put(in_dev);
1846         }
1847 #if IS_ENABLED(CONFIG_IPV6)
1848         mlx4_make_default_gid(dev, &default_gid);
1849         /* IPv6 gids */
1850         in6_dev = in6_dev_get(dev);
1851         if (in6_dev) {
1852                 read_lock_bh(&in6_dev->lock);
1853                 list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
1854                         pgid = (union ib_gid *)&ifp->addr;
1855                         if (!memcmp(pgid, &default_gid, sizeof(*pgid)))
1856                                 continue;
1857                         update_gid_table(ibdev, port, pgid, 0, 0);
1858                 }
1859                 read_unlock_bh(&in6_dev->lock);
1860                 in6_dev_put(in6_dev);
1861         }
1862 #endif
1863 }
1864
1865 static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
1866                                  struct  net_device *dev, u8 port)
1867 {
1868         union ib_gid gid;
1869         mlx4_make_default_gid(dev, &gid);
1870         update_gid_table(ibdev, port, &gid, 0, 1);
1871 }
1872
1873 static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
1874 {
1875         struct  net_device *dev;
1876         struct mlx4_ib_iboe *iboe = &ibdev->iboe;
1877         int i;
1878         int err = 0;
1879
1880         for (i = 1; i <= ibdev->num_ports; ++i) {
1881                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i) ==
1882                     IB_LINK_LAYER_ETHERNET) {
1883                         err = reset_gid_table(ibdev, i);
1884                         if (err)
1885                                 goto out;
1886                 }
1887         }
1888
1889         read_lock(&dev_base_lock);
1890         spin_lock_bh(&iboe->lock);
1891
1892         for_each_netdev(&init_net, dev) {
1893                 u8 port = mlx4_ib_get_dev_port(dev, ibdev);
1894                 /* port will be non-zero only for ETH ports */
1895                 if (port) {
1896                         mlx4_ib_set_default_gid(ibdev, dev, port);
1897                         mlx4_ib_get_dev_addr(dev, ibdev, port);
1898                 }
1899         }
1900
1901         spin_unlock_bh(&iboe->lock);
1902         read_unlock(&dev_base_lock);
1903 out:
1904         return err;
1905 }
1906
1907 static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
1908                                  struct net_device *dev,
1909                                  unsigned long event)
1910
1911 {
1912         struct mlx4_ib_iboe *iboe;
1913         int update_qps_port = -1;
1914         int port;
1915
1916         iboe = &ibdev->iboe;
1917
1918         spin_lock_bh(&iboe->lock);
1919         mlx4_foreach_ib_transport_port(port, ibdev->dev) {
1920                 enum ib_port_state      port_state = IB_PORT_NOP;
1921                 struct net_device *old_master = iboe->masters[port - 1];
1922                 struct net_device *curr_netdev;
1923                 struct net_device *curr_master;
1924
1925                 iboe->netdevs[port - 1] =
1926                         mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
1927                 if (iboe->netdevs[port - 1])
1928                         mlx4_ib_set_default_gid(ibdev,
1929                                                 iboe->netdevs[port - 1], port);
1930                 curr_netdev = iboe->netdevs[port - 1];
1931
1932                 if (iboe->netdevs[port - 1] &&
1933                     netif_is_bond_slave(iboe->netdevs[port - 1])) {
1934                         iboe->masters[port - 1] = netdev_master_upper_dev_get(
1935                                 iboe->netdevs[port - 1]);
1936                 } else {
1937                         iboe->masters[port - 1] = NULL;
1938                 }
1939                 curr_master = iboe->masters[port - 1];
1940
1941                 if (dev == iboe->netdevs[port - 1] &&
1942                     (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
1943                      event == NETDEV_UP || event == NETDEV_CHANGE))
1944                         update_qps_port = port;
1945
1946                 if (curr_netdev) {
1947                         port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
1948                                                 IB_PORT_ACTIVE : IB_PORT_DOWN;
1949                         mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
1950                         if (curr_master) {
1951                                 /* if using bonding/team and a slave port is down, we
1952                                  * don't want the bond IP based gids in the table since
1953                                  * flows that select port by gid may get the down port.
1954                                 */
1955                                 if (port_state == IB_PORT_DOWN &&
1956                                     !mlx4_is_bonded(ibdev->dev)) {
1957                                         reset_gid_table(ibdev, port);
1958                                         mlx4_ib_set_default_gid(ibdev,
1959                                                                 curr_netdev,
1960                                                                 port);
1961                                 } else {
1962                                         /* gids from the upper dev (bond/team)
1963                                          * should appear in port's gid table
1964                                         */
1965                                         mlx4_ib_get_dev_addr(curr_master,
1966                                                              ibdev, port);
1967                                 }
1968                         }
1969                         /* if bonding is used it is possible that we add it to
1970                          * masters only after IP address is assigned to the
1971                          * net bonding interface.
1972                         */
1973                         if (curr_master && (old_master != curr_master)) {
1974                                 reset_gid_table(ibdev, port);
1975                                 mlx4_ib_set_default_gid(ibdev,
1976                                                         curr_netdev, port);
1977                                 mlx4_ib_get_dev_addr(curr_master, ibdev, port);
1978                         }
1979
1980                         if (!curr_master && (old_master != curr_master)) {
1981                                 reset_gid_table(ibdev, port);
1982                                 mlx4_ib_set_default_gid(ibdev,
1983                                                         curr_netdev, port);
1984                                 mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
1985                         }
1986                 } else {
1987                         reset_gid_table(ibdev, port);
1988                 }
1989         }
1990
1991         spin_unlock_bh(&iboe->lock);
1992
1993         if (update_qps_port > 0)
1994                 mlx4_ib_update_qps(ibdev, dev, update_qps_port);
1995 }
1996
1997 static int mlx4_ib_netdev_event(struct notifier_block *this,
1998                                 unsigned long event, void *ptr)
1999 {
2000         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
2001         struct mlx4_ib_dev *ibdev;
2002
2003         if (!net_eq(dev_net(dev), &init_net))
2004                 return NOTIFY_DONE;
2005
2006         ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
2007         mlx4_ib_scan_netdevs(ibdev, dev, event);
2008
2009         return NOTIFY_DONE;
2010 }
2011
2012 static void init_pkeys(struct mlx4_ib_dev *ibdev)
2013 {
2014         int port;
2015         int slave;
2016         int i;
2017
2018         if (mlx4_is_master(ibdev->dev)) {
2019                 for (slave = 0; slave <= ibdev->dev->persist->num_vfs;
2020                      ++slave) {
2021                         for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2022                                 for (i = 0;
2023                                      i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2024                                      ++i) {
2025                                         ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
2026                                         /* master has the identity virt2phys pkey mapping */
2027                                                 (slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
2028                                                         ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
2029                                         mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
2030                                                              ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
2031                                 }
2032                         }
2033                 }
2034                 /* initialize pkey cache */
2035                 for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
2036                         for (i = 0;
2037                              i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
2038                              ++i)
2039                                 ibdev->pkeys.phys_pkey_cache[port-1][i] =
2040                                         (i) ? 0 : 0xFFFF;
2041                 }
2042         }
2043 }
2044
2045 static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2046 {
2047         char name[80];
2048         int eq_per_port = 0;
2049         int added_eqs = 0;
2050         int total_eqs = 0;
2051         int i, j, eq;
2052
2053         /* Legacy mode or comp_pool is not large enough */
2054         if (dev->caps.comp_pool == 0 ||
2055             dev->caps.num_ports > dev->caps.comp_pool)
2056                 return;
2057
2058         eq_per_port = dev->caps.comp_pool / dev->caps.num_ports;
2059
2060         /* Init eq table */
2061         added_eqs = 0;
2062         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2063                 added_eqs += eq_per_port;
2064
2065         total_eqs = dev->caps.num_comp_vectors + added_eqs;
2066
2067         ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
2068         if (!ibdev->eq_table)
2069                 return;
2070
2071         ibdev->eq_added = added_eqs;
2072
2073         eq = 0;
2074         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
2075                 for (j = 0; j < eq_per_port; j++) {
2076                         snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
2077                                  i, j, dev->persist->pdev->bus->name);
2078                         /* Set IRQ for specific name (per ring) */
2079                         if (mlx4_assign_eq(dev, name, NULL,
2080                                            &ibdev->eq_table[eq])) {
2081                                 /* Use legacy (same as mlx4_en driver) */
2082                                 pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
2083                                 ibdev->eq_table[eq] =
2084                                         (eq % dev->caps.num_comp_vectors);
2085                         }
2086                         eq++;
2087                 }
2088         }
2089
2090         /* Fill the reset of the vector with legacy EQ */
2091         for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
2092                 ibdev->eq_table[eq++] = i;
2093
2094         /* Advertise the new number of EQs to clients */
2095         ibdev->ib_dev.num_comp_vectors = total_eqs;
2096 }
2097
2098 static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
2099 {
2100         int i;
2101
2102         /* no additional eqs were added */
2103         if (!ibdev->eq_table)
2104                 return;
2105
2106         /* Reset the advertised EQ number */
2107         ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
2108
2109         /* Free only the added eqs */
2110         for (i = 0; i < ibdev->eq_added; i++) {
2111                 /* Don't free legacy eqs if used */
2112                 if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
2113                         continue;
2114                 mlx4_release_eq(dev, ibdev->eq_table[i]);
2115         }
2116
2117         kfree(ibdev->eq_table);
2118 }
2119
2120 static int mlx4_port_immutable(struct ib_device *ibdev, u8 port_num,
2121                                struct ib_port_immutable *immutable)
2122 {
2123         struct ib_port_attr attr;
2124         int err;
2125
2126         err = mlx4_ib_query_port(ibdev, port_num, &attr);
2127         if (err)
2128                 return err;
2129
2130         immutable->pkey_tbl_len = attr.pkey_tbl_len;
2131         immutable->gid_tbl_len = attr.gid_tbl_len;
2132
2133         if (mlx4_ib_port_link_layer(ibdev, port_num) == IB_LINK_LAYER_INFINIBAND)
2134                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
2135         else
2136                 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
2137
2138         return 0;
2139 }
2140
2141 static void *mlx4_ib_add(struct mlx4_dev *dev)
2142 {
2143         struct mlx4_ib_dev *ibdev;
2144         int num_ports = 0;
2145         int i, j;
2146         int err;
2147         struct mlx4_ib_iboe *iboe;
2148         int ib_num_ports = 0;
2149         int num_req_counters;
2150
2151         pr_info_once("%s", mlx4_ib_version);
2152
2153         num_ports = 0;
2154         mlx4_foreach_ib_transport_port(i, dev)
2155                 num_ports++;
2156
2157         /* No point in registering a device with no ports... */
2158         if (num_ports == 0)
2159                 return NULL;
2160
2161         ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
2162         if (!ibdev) {
2163                 dev_err(&dev->persist->pdev->dev,
2164                         "Device struct alloc failed\n");
2165                 return NULL;
2166         }
2167
2168         iboe = &ibdev->iboe;
2169
2170         if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
2171                 goto err_dealloc;
2172
2173         if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
2174                 goto err_pd;
2175
2176         ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
2177                                  PAGE_SIZE);
2178         if (!ibdev->uar_map)
2179                 goto err_uar;
2180         MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
2181
2182         ibdev->dev = dev;
2183         ibdev->bond_next_port   = 0;
2184
2185         strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
2186         ibdev->ib_dev.owner             = THIS_MODULE;
2187         ibdev->ib_dev.node_type         = RDMA_NODE_IB_CA;
2188         ibdev->ib_dev.local_dma_lkey    = dev->caps.reserved_lkey;
2189         ibdev->num_ports                = num_ports;
2190         ibdev->ib_dev.phys_port_cnt     = mlx4_is_bonded(dev) ?
2191                                                 1 : ibdev->num_ports;
2192         ibdev->ib_dev.num_comp_vectors  = dev->caps.num_comp_vectors;
2193         ibdev->ib_dev.dma_device        = &dev->persist->pdev->dev;
2194
2195         if (dev->caps.userspace_caps)
2196                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
2197         else
2198                 ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
2199
2200         ibdev->ib_dev.uverbs_cmd_mask   =
2201                 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
2202                 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
2203                 (1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
2204                 (1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
2205                 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
2206                 (1ull << IB_USER_VERBS_CMD_REG_MR)              |
2207                 (1ull << IB_USER_VERBS_CMD_REREG_MR)            |
2208                 (1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
2209                 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
2210                 (1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
2211                 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
2212                 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
2213                 (1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
2214                 (1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
2215                 (1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
2216                 (1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
2217                 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
2218                 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
2219                 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
2220                 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
2221                 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
2222                 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
2223                 (1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)         |
2224                 (1ull << IB_USER_VERBS_CMD_OPEN_QP);
2225
2226         ibdev->ib_dev.query_device      = mlx4_ib_query_device;
2227         ibdev->ib_dev.query_port        = mlx4_ib_query_port;
2228         ibdev->ib_dev.get_link_layer    = mlx4_ib_port_link_layer;
2229         ibdev->ib_dev.query_gid         = mlx4_ib_query_gid;
2230         ibdev->ib_dev.query_pkey        = mlx4_ib_query_pkey;
2231         ibdev->ib_dev.modify_device     = mlx4_ib_modify_device;
2232         ibdev->ib_dev.modify_port       = mlx4_ib_modify_port;
2233         ibdev->ib_dev.alloc_ucontext    = mlx4_ib_alloc_ucontext;
2234         ibdev->ib_dev.dealloc_ucontext  = mlx4_ib_dealloc_ucontext;
2235         ibdev->ib_dev.mmap              = mlx4_ib_mmap;
2236         ibdev->ib_dev.alloc_pd          = mlx4_ib_alloc_pd;
2237         ibdev->ib_dev.dealloc_pd        = mlx4_ib_dealloc_pd;
2238         ibdev->ib_dev.create_ah         = mlx4_ib_create_ah;
2239         ibdev->ib_dev.query_ah          = mlx4_ib_query_ah;
2240         ibdev->ib_dev.destroy_ah        = mlx4_ib_destroy_ah;
2241         ibdev->ib_dev.create_srq        = mlx4_ib_create_srq;
2242         ibdev->ib_dev.modify_srq        = mlx4_ib_modify_srq;
2243         ibdev->ib_dev.query_srq         = mlx4_ib_query_srq;
2244         ibdev->ib_dev.destroy_srq       = mlx4_ib_destroy_srq;
2245         ibdev->ib_dev.post_srq_recv     = mlx4_ib_post_srq_recv;
2246         ibdev->ib_dev.create_qp         = mlx4_ib_create_qp;
2247         ibdev->ib_dev.modify_qp         = mlx4_ib_modify_qp;
2248         ibdev->ib_dev.query_qp          = mlx4_ib_query_qp;
2249         ibdev->ib_dev.destroy_qp        = mlx4_ib_destroy_qp;
2250         ibdev->ib_dev.post_send         = mlx4_ib_post_send;
2251         ibdev->ib_dev.post_recv         = mlx4_ib_post_recv;
2252         ibdev->ib_dev.create_cq         = mlx4_ib_create_cq;
2253         ibdev->ib_dev.modify_cq         = mlx4_ib_modify_cq;
2254         ibdev->ib_dev.resize_cq         = mlx4_ib_resize_cq;
2255         ibdev->ib_dev.destroy_cq        = mlx4_ib_destroy_cq;
2256         ibdev->ib_dev.poll_cq           = mlx4_ib_poll_cq;
2257         ibdev->ib_dev.req_notify_cq     = mlx4_ib_arm_cq;
2258         ibdev->ib_dev.get_dma_mr        = mlx4_ib_get_dma_mr;
2259         ibdev->ib_dev.reg_user_mr       = mlx4_ib_reg_user_mr;
2260         ibdev->ib_dev.rereg_user_mr     = mlx4_ib_rereg_user_mr;
2261         ibdev->ib_dev.dereg_mr          = mlx4_ib_dereg_mr;
2262         ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
2263         ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
2264         ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
2265         ibdev->ib_dev.attach_mcast      = mlx4_ib_mcg_attach;
2266         ibdev->ib_dev.detach_mcast      = mlx4_ib_mcg_detach;
2267         ibdev->ib_dev.process_mad       = mlx4_ib_process_mad;
2268         ibdev->ib_dev.get_port_immutable = mlx4_port_immutable;
2269
2270         if (!mlx4_is_slave(ibdev->dev)) {
2271                 ibdev->ib_dev.alloc_fmr         = mlx4_ib_fmr_alloc;
2272                 ibdev->ib_dev.map_phys_fmr      = mlx4_ib_map_phys_fmr;
2273                 ibdev->ib_dev.unmap_fmr         = mlx4_ib_unmap_fmr;
2274                 ibdev->ib_dev.dealloc_fmr       = mlx4_ib_fmr_dealloc;
2275         }
2276
2277         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
2278             dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
2279                 ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
2280                 ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
2281                 ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
2282
2283                 ibdev->ib_dev.uverbs_cmd_mask |=
2284                         (1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
2285                         (1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
2286         }
2287
2288         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
2289                 ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
2290                 ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
2291                 ibdev->ib_dev.uverbs_cmd_mask |=
2292                         (1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
2293                         (1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
2294         }
2295
2296         if (check_flow_steering_support(dev)) {
2297                 ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
2298                 ibdev->ib_dev.create_flow       = mlx4_ib_create_flow;
2299                 ibdev->ib_dev.destroy_flow      = mlx4_ib_destroy_flow;
2300
2301                 ibdev->ib_dev.uverbs_ex_cmd_mask        |=
2302                         (1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
2303                         (1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
2304         }
2305
2306         mlx4_ib_alloc_eqs(dev, ibdev);
2307
2308         spin_lock_init(&iboe->lock);
2309
2310         if (init_node_data(ibdev))
2311                 goto err_map;
2312
2313         num_req_counters = mlx4_is_bonded(dev) ? 1 : ibdev->num_ports;
2314         for (i = 0; i < num_req_counters; ++i) {
2315                 mutex_init(&ibdev->qp1_proxy_lock[i]);
2316                 if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
2317                                                 IB_LINK_LAYER_ETHERNET) {
2318                         err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
2319                         if (err)
2320                                 ibdev->counters[i] = -1;
2321                 } else {
2322                         ibdev->counters[i] = -1;
2323                 }
2324         }
2325         if (mlx4_is_bonded(dev))
2326                 for (i = 1; i < ibdev->num_ports ; ++i)
2327                         ibdev->counters[i] = ibdev->counters[0];
2328
2329
2330         mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
2331                 ib_num_ports++;
2332
2333         spin_lock_init(&ibdev->sm_lock);
2334         mutex_init(&ibdev->cap_mask_mutex);
2335         INIT_LIST_HEAD(&ibdev->qp_list);
2336         spin_lock_init(&ibdev->reset_flow_resource_lock);
2337
2338         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
2339             ib_num_ports) {
2340                 ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
2341                 err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
2342                                             MLX4_IB_UC_STEER_QPN_ALIGN,
2343                                             &ibdev->steer_qpn_base, 0);
2344                 if (err)
2345                         goto err_counter;
2346
2347                 ibdev->ib_uc_qpns_bitmap =
2348                         kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
2349                                 sizeof(long),
2350                                 GFP_KERNEL);
2351                 if (!ibdev->ib_uc_qpns_bitmap) {
2352                         dev_err(&dev->persist->pdev->dev,
2353                                 "bit map alloc failed\n");
2354                         goto err_steer_qp_release;
2355                 }
2356
2357                 bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
2358
2359                 err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
2360                                 dev, ibdev->steer_qpn_base,
2361                                 ibdev->steer_qpn_base +
2362                                 ibdev->steer_qpn_count - 1);
2363                 if (err)
2364                         goto err_steer_free_bitmap;
2365         }
2366
2367         for (j = 1; j <= ibdev->dev->caps.num_ports; j++)
2368                 atomic64_set(&iboe->mac[j - 1], ibdev->dev->caps.def_mac[j]);
2369
2370         if (ib_register_device(&ibdev->ib_dev, NULL))
2371                 goto err_steer_free_bitmap;
2372
2373         if (mlx4_ib_mad_init(ibdev))
2374                 goto err_reg;
2375
2376         if (mlx4_ib_init_sriov(ibdev))
2377                 goto err_mad;
2378
2379         if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
2380                 if (!iboe->nb.notifier_call) {
2381                         iboe->nb.notifier_call = mlx4_ib_netdev_event;
2382                         err = register_netdevice_notifier(&iboe->nb);
2383                         if (err) {
2384                                 iboe->nb.notifier_call = NULL;
2385                                 goto err_notif;
2386                         }
2387                 }
2388                 if (!iboe->nb_inet.notifier_call) {
2389                         iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
2390                         err = register_inetaddr_notifier(&iboe->nb_inet);
2391                         if (err) {
2392                                 iboe->nb_inet.notifier_call = NULL;
2393                                 goto err_notif;
2394                         }
2395                 }
2396 #if IS_ENABLED(CONFIG_IPV6)
2397                 if (!iboe->nb_inet6.notifier_call) {
2398                         iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
2399                         err = register_inet6addr_notifier(&iboe->nb_inet6);
2400                         if (err) {
2401                                 iboe->nb_inet6.notifier_call = NULL;
2402                                 goto err_notif;
2403                         }
2404                 }
2405 #endif
2406                 if (mlx4_ib_init_gid_table(ibdev))
2407                         goto err_notif;
2408         }
2409
2410         for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
2411                 if (device_create_file(&ibdev->ib_dev.dev,
2412                                        mlx4_class_attributes[j]))
2413                         goto err_notif;
2414         }
2415
2416         ibdev->ib_active = true;
2417
2418         if (mlx4_is_mfunc(ibdev->dev))
2419                 init_pkeys(ibdev);
2420
2421         /* create paravirt contexts for any VFs which are active */
2422         if (mlx4_is_master(ibdev->dev)) {
2423                 for (j = 0; j < MLX4_MFUNC_MAX; j++) {
2424                         if (j == mlx4_master_func_num(ibdev->dev))
2425                                 continue;
2426                         if (mlx4_is_slave_active(ibdev->dev, j))
2427                                 do_slave_init(ibdev, j, 1);
2428                 }
2429         }
2430         return ibdev;
2431
2432 err_notif:
2433         if (ibdev->iboe.nb.notifier_call) {
2434                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2435                         pr_warn("failure unregistering notifier\n");
2436                 ibdev->iboe.nb.notifier_call = NULL;
2437         }
2438         if (ibdev->iboe.nb_inet.notifier_call) {
2439                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2440                         pr_warn("failure unregistering notifier\n");
2441                 ibdev->iboe.nb_inet.notifier_call = NULL;
2442         }
2443 #if IS_ENABLED(CONFIG_IPV6)
2444         if (ibdev->iboe.nb_inet6.notifier_call) {
2445                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2446                         pr_warn("failure unregistering notifier\n");
2447                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2448         }
2449 #endif
2450         flush_workqueue(wq);
2451
2452         mlx4_ib_close_sriov(ibdev);
2453
2454 err_mad:
2455         mlx4_ib_mad_cleanup(ibdev);
2456
2457 err_reg:
2458         ib_unregister_device(&ibdev->ib_dev);
2459
2460 err_steer_free_bitmap:
2461         kfree(ibdev->ib_uc_qpns_bitmap);
2462
2463 err_steer_qp_release:
2464         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
2465                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2466                                       ibdev->steer_qpn_count);
2467 err_counter:
2468         for (; i; --i)
2469                 if (ibdev->counters[i - 1] != -1)
2470                         mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
2471
2472 err_map:
2473         iounmap(ibdev->uar_map);
2474
2475 err_uar:
2476         mlx4_uar_free(dev, &ibdev->priv_uar);
2477
2478 err_pd:
2479         mlx4_pd_free(dev, ibdev->priv_pdn);
2480
2481 err_dealloc:
2482         ib_dealloc_device(&ibdev->ib_dev);
2483
2484         return NULL;
2485 }
2486
2487 int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
2488 {
2489         int offset;
2490
2491         WARN_ON(!dev->ib_uc_qpns_bitmap);
2492
2493         offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
2494                                          dev->steer_qpn_count,
2495                                          get_count_order(count));
2496         if (offset < 0)
2497                 return offset;
2498
2499         *qpn = dev->steer_qpn_base + offset;
2500         return 0;
2501 }
2502
2503 void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
2504 {
2505         if (!qpn ||
2506             dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
2507                 return;
2508
2509         BUG_ON(qpn < dev->steer_qpn_base);
2510
2511         bitmap_release_region(dev->ib_uc_qpns_bitmap,
2512                               qpn - dev->steer_qpn_base,
2513                               get_count_order(count));
2514 }
2515
2516 int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
2517                          int is_attach)
2518 {
2519         int err;
2520         size_t flow_size;
2521         struct ib_flow_attr *flow = NULL;
2522         struct ib_flow_spec_ib *ib_spec;
2523
2524         if (is_attach) {
2525                 flow_size = sizeof(struct ib_flow_attr) +
2526                             sizeof(struct ib_flow_spec_ib);
2527                 flow = kzalloc(flow_size, GFP_KERNEL);
2528                 if (!flow)
2529                         return -ENOMEM;
2530                 flow->port = mqp->port;
2531                 flow->num_of_specs = 1;
2532                 flow->size = flow_size;
2533                 ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
2534                 ib_spec->type = IB_FLOW_SPEC_IB;
2535                 ib_spec->size = sizeof(struct ib_flow_spec_ib);
2536                 /* Add an empty rule for IB L2 */
2537                 memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
2538
2539                 err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
2540                                             IB_FLOW_DOMAIN_NIC,
2541                                             MLX4_FS_REGULAR,
2542                                             &mqp->reg_id);
2543         } else {
2544                 err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
2545         }
2546         kfree(flow);
2547         return err;
2548 }
2549
2550 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
2551 {
2552         struct mlx4_ib_dev *ibdev = ibdev_ptr;
2553         int p;
2554
2555         ibdev->ib_active = false;
2556         flush_workqueue(wq);
2557
2558         mlx4_ib_close_sriov(ibdev);
2559         mlx4_ib_mad_cleanup(ibdev);
2560         ib_unregister_device(&ibdev->ib_dev);
2561         if (ibdev->iboe.nb.notifier_call) {
2562                 if (unregister_netdevice_notifier(&ibdev->iboe.nb))
2563                         pr_warn("failure unregistering notifier\n");
2564                 ibdev->iboe.nb.notifier_call = NULL;
2565         }
2566
2567         if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
2568                 mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
2569                                       ibdev->steer_qpn_count);
2570                 kfree(ibdev->ib_uc_qpns_bitmap);
2571         }
2572
2573         if (ibdev->iboe.nb_inet.notifier_call) {
2574                 if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
2575                         pr_warn("failure unregistering notifier\n");
2576                 ibdev->iboe.nb_inet.notifier_call = NULL;
2577         }
2578 #if IS_ENABLED(CONFIG_IPV6)
2579         if (ibdev->iboe.nb_inet6.notifier_call) {
2580                 if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
2581                         pr_warn("failure unregistering notifier\n");
2582                 ibdev->iboe.nb_inet6.notifier_call = NULL;
2583         }
2584 #endif
2585
2586         iounmap(ibdev->uar_map);
2587         for (p = 0; p < ibdev->num_ports; ++p)
2588                 if (ibdev->counters[p] != -1)
2589                         mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
2590         mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
2591                 mlx4_CLOSE_PORT(dev, p);
2592
2593         mlx4_ib_free_eqs(dev, ibdev);
2594
2595         mlx4_uar_free(dev, &ibdev->priv_uar);
2596         mlx4_pd_free(dev, ibdev->priv_pdn);
2597         ib_dealloc_device(&ibdev->ib_dev);
2598 }
2599
2600 static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
2601 {
2602         struct mlx4_ib_demux_work **dm = NULL;
2603         struct mlx4_dev *dev = ibdev->dev;
2604         int i;
2605         unsigned long flags;
2606         struct mlx4_active_ports actv_ports;
2607         unsigned int ports;
2608         unsigned int first_port;
2609
2610         if (!mlx4_is_master(dev))
2611                 return;
2612
2613         actv_ports = mlx4_get_active_ports(dev, slave);
2614         ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
2615         first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
2616
2617         dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
2618         if (!dm) {
2619                 pr_err("failed to allocate memory for tunneling qp update\n");
2620                 goto out;
2621         }
2622
2623         for (i = 0; i < ports; i++) {
2624                 dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
2625                 if (!dm[i]) {
2626                         pr_err("failed to allocate memory for tunneling qp update work struct\n");
2627                         for (i = 0; i < dev->caps.num_ports; i++) {
2628                                 if (dm[i])
2629                                         kfree(dm[i]);
2630                         }
2631                         goto out;
2632                 }
2633         }
2634         /* initialize or tear down tunnel QPs for the slave */
2635         for (i = 0; i < ports; i++) {
2636                 INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
2637                 dm[i]->port = first_port + i + 1;
2638                 dm[i]->slave = slave;
2639                 dm[i]->do_init = do_init;
2640                 dm[i]->dev = ibdev;
2641                 spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
2642                 if (!ibdev->sriov.is_going_down)
2643                         queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
2644                 spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
2645         }
2646 out:
2647         kfree(dm);
2648         return;
2649 }
2650
2651 static void mlx4_ib_handle_catas_error(struct mlx4_ib_dev *ibdev)
2652 {
2653         struct mlx4_ib_qp *mqp;
2654         unsigned long flags_qp;
2655         unsigned long flags_cq;
2656         struct mlx4_ib_cq *send_mcq, *recv_mcq;
2657         struct list_head    cq_notify_list;
2658         struct mlx4_cq *mcq;
2659         unsigned long flags;
2660
2661         pr_warn("mlx4_ib_handle_catas_error was started\n");
2662         INIT_LIST_HEAD(&cq_notify_list);
2663
2664         /* Go over qp list reside on that ibdev, sync with create/destroy qp.*/
2665         spin_lock_irqsave(&ibdev->reset_flow_resource_lock, flags);
2666
2667         list_for_each_entry(mqp, &ibdev->qp_list, qps_list) {
2668                 spin_lock_irqsave(&mqp->sq.lock, flags_qp);
2669                 if (mqp->sq.tail != mqp->sq.head) {
2670                         send_mcq = to_mcq(mqp->ibqp.send_cq);
2671                         spin_lock_irqsave(&send_mcq->lock, flags_cq);
2672                         if (send_mcq->mcq.comp &&
2673                             mqp->ibqp.send_cq->comp_handler) {
2674                                 if (!send_mcq->mcq.reset_notify_added) {
2675                                         send_mcq->mcq.reset_notify_added = 1;
2676                                         list_add_tail(&send_mcq->mcq.reset_notify,
2677                                                       &cq_notify_list);
2678                                 }
2679                         }
2680                         spin_unlock_irqrestore(&send_mcq->lock, flags_cq);
2681                 }
2682                 spin_unlock_irqrestore(&mqp->sq.lock, flags_qp);
2683                 /* Now, handle the QP's receive queue */
2684                 spin_lock_irqsave(&mqp->rq.lock, flags_qp);
2685                 /* no handling is needed for SRQ */
2686                 if (!mqp->ibqp.srq) {
2687                         if (mqp->rq.tail != mqp->rq.head) {
2688                                 recv_mcq = to_mcq(mqp->ibqp.recv_cq);
2689                                 spin_lock_irqsave(&recv_mcq->lock, flags_cq);
2690                                 if (recv_mcq->mcq.comp &&
2691                                     mqp->ibqp.recv_cq->comp_handler) {
2692                                         if (!recv_mcq->mcq.reset_notify_added) {
2693                                                 recv_mcq->mcq.reset_notify_added = 1;
2694                                                 list_add_tail(&recv_mcq->mcq.reset_notify,
2695                                                               &cq_notify_list);
2696                                         }
2697                                 }
2698                                 spin_unlock_irqrestore(&recv_mcq->lock,
2699                                                        flags_cq);
2700                         }
2701                 }
2702                 spin_unlock_irqrestore(&mqp->rq.lock, flags_qp);
2703         }
2704
2705         list_for_each_entry(mcq, &cq_notify_list, reset_notify) {
2706                 mcq->comp(mcq);
2707         }
2708         spin_unlock_irqrestore(&ibdev->reset_flow_resource_lock, flags);
2709         pr_warn("mlx4_ib_handle_catas_error ended\n");
2710 }
2711
2712 static void handle_bonded_port_state_event(struct work_struct *work)
2713 {
2714         struct ib_event_work *ew =
2715                 container_of(work, struct ib_event_work, work);
2716         struct mlx4_ib_dev *ibdev = ew->ib_dev;
2717         enum ib_port_state bonded_port_state = IB_PORT_NOP;
2718         int i;
2719         struct ib_event ibev;
2720
2721         kfree(ew);
2722         spin_lock_bh(&ibdev->iboe.lock);
2723         for (i = 0; i < MLX4_MAX_PORTS; ++i) {
2724                 struct net_device *curr_netdev = ibdev->iboe.netdevs[i];
2725                 enum ib_port_state curr_port_state;
2726
2727                 if (!curr_netdev)
2728                         continue;
2729
2730                 curr_port_state =
2731                         (netif_running(curr_netdev) &&
2732                          netif_carrier_ok(curr_netdev)) ?
2733                         IB_PORT_ACTIVE : IB_PORT_DOWN;
2734
2735                 bonded_port_state = (bonded_port_state != IB_PORT_ACTIVE) ?
2736                         curr_port_state : IB_PORT_ACTIVE;
2737         }
2738         spin_unlock_bh(&ibdev->iboe.lock);
2739
2740         ibev.device = &ibdev->ib_dev;
2741         ibev.element.port_num = 1;
2742         ibev.event = (bonded_port_state == IB_PORT_ACTIVE) ?
2743                 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
2744
2745         ib_dispatch_event(&ibev);
2746 }
2747
2748 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
2749                           enum mlx4_dev_event event, unsigned long param)
2750 {
2751         struct ib_event ibev;
2752         struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
2753         struct mlx4_eqe *eqe = NULL;
2754         struct ib_event_work *ew;
2755         int p = 0;
2756
2757         if (mlx4_is_bonded(dev) &&
2758             ((event == MLX4_DEV_EVENT_PORT_UP) ||
2759             (event == MLX4_DEV_EVENT_PORT_DOWN))) {
2760                 ew = kmalloc(sizeof(*ew), GFP_ATOMIC);
2761                 if (!ew)
2762                         return;
2763                 INIT_WORK(&ew->work, handle_bonded_port_state_event);
2764                 ew->ib_dev = ibdev;
2765                 queue_work(wq, &ew->work);
2766                 return;
2767         }
2768
2769         if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
2770                 eqe = (struct mlx4_eqe *)param;
2771         else
2772                 p = (int) param;
2773
2774         switch (event) {
2775         case MLX4_DEV_EVENT_PORT_UP:
2776                 if (p > ibdev->num_ports)
2777                         return;
2778                 if (mlx4_is_master(dev) &&
2779                     rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
2780                         IB_LINK_LAYER_INFINIBAND) {
2781                         mlx4_ib_invalidate_all_guid_record(ibdev, p);
2782                 }
2783                 ibev.event = IB_EVENT_PORT_ACTIVE;
2784                 break;
2785
2786         case MLX4_DEV_EVENT_PORT_DOWN:
2787                 if (p > ibdev->num_ports)
2788                         return;
2789                 ibev.event = IB_EVENT_PORT_ERR;
2790                 break;
2791
2792         case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
2793                 ibdev->ib_active = false;
2794                 ibev.event = IB_EVENT_DEVICE_FATAL;
2795                 mlx4_ib_handle_catas_error(ibdev);
2796                 break;
2797
2798         case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
2799                 ew = kmalloc(sizeof *ew, GFP_ATOMIC);
2800                 if (!ew) {
2801                         pr_err("failed to allocate memory for events work\n");
2802                         break;
2803                 }
2804
2805                 INIT_WORK(&ew->work, handle_port_mgmt_change_event);
2806                 memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
2807                 ew->ib_dev = ibdev;
2808                 /* need to queue only for port owner, which uses GEN_EQE */
2809                 if (mlx4_is_master(dev))
2810                         queue_work(wq, &ew->work);
2811                 else
2812                         handle_port_mgmt_change_event(&ew->work);
2813                 return;
2814
2815         case MLX4_DEV_EVENT_SLAVE_INIT:
2816                 /* here, p is the slave id */
2817                 do_slave_init(ibdev, p, 1);
2818                 if (mlx4_is_master(dev)) {
2819                         int i;
2820
2821                         for (i = 1; i <= ibdev->num_ports; i++) {
2822                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2823                                         == IB_LINK_LAYER_INFINIBAND)
2824                                         mlx4_ib_slave_alias_guid_event(ibdev,
2825                                                                        p, i,
2826                                                                        1);
2827                         }
2828                 }
2829                 return;
2830
2831         case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
2832                 if (mlx4_is_master(dev)) {
2833                         int i;
2834
2835                         for (i = 1; i <= ibdev->num_ports; i++) {
2836                                 if (rdma_port_get_link_layer(&ibdev->ib_dev, i)
2837                                         == IB_LINK_LAYER_INFINIBAND)
2838                                         mlx4_ib_slave_alias_guid_event(ibdev,
2839                                                                        p, i,
2840                                                                        0);
2841                         }
2842                 }
2843                 /* here, p is the slave id */
2844                 do_slave_init(ibdev, p, 0);
2845                 return;
2846
2847         default:
2848                 return;
2849         }
2850
2851         ibev.device           = ibdev_ptr;
2852         ibev.element.port_num = mlx4_is_bonded(ibdev->dev) ? 1 : (u8)p;
2853
2854         ib_dispatch_event(&ibev);
2855 }
2856
2857 static struct mlx4_interface mlx4_ib_interface = {
2858         .add            = mlx4_ib_add,
2859         .remove         = mlx4_ib_remove,
2860         .event          = mlx4_ib_event,
2861         .protocol       = MLX4_PROT_IB_IPV6,
2862         .flags          = MLX4_INTFF_BONDING
2863 };
2864
2865 static int __init mlx4_ib_init(void)
2866 {
2867         int err;
2868
2869         wq = create_singlethread_workqueue("mlx4_ib");
2870         if (!wq)
2871                 return -ENOMEM;
2872
2873         err = mlx4_ib_mcg_init();
2874         if (err)
2875                 goto clean_wq;
2876
2877         err = mlx4_register_interface(&mlx4_ib_interface);
2878         if (err)
2879                 goto clean_mcg;
2880
2881         return 0;
2882
2883 clean_mcg:
2884         mlx4_ib_mcg_destroy();
2885
2886 clean_wq:
2887         destroy_workqueue(wq);
2888         return err;
2889 }
2890
2891 static void __exit mlx4_ib_cleanup(void)
2892 {
2893         mlx4_unregister_interface(&mlx4_ib_interface);
2894         mlx4_ib_mcg_destroy();
2895         destroy_workqueue(wq);
2896 }
2897
2898 module_init(mlx4_ib_init);
2899 module_exit(mlx4_ib_cleanup);