drivers/infiniband/sw/rxe/rxe.c

   1 /*
   2  * Copyright (c) 2016 Mellanox Technologies Ltd. All rights reserved.
   3  * Copyright (c) 2015 System Fabric Works, Inc. 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 "rxe.h"
  35 #include "rxe_loc.h"
  36
  37 MODULE_AUTHOR("Bob Pearson, Frank Zago, John Groves, Kamal Heib");
  38 MODULE_DESCRIPTION("Soft RDMA transport");
  39 MODULE_LICENSE("Dual BSD/GPL");
  40 MODULE_VERSION("0.2");
  41
  42 /* free resources for all ports on a device */
  43 static void rxe_cleanup_ports(struct rxe_dev *rxe)
  44 {
  45         kfree(rxe->port.pkey_tbl);
  46         rxe->port.pkey_tbl = NULL;
  47
  48 }
  49
  50 /* free resources for a rxe device all objects created for this device must
  51  * have been destroyed
  52  */
  53 static void rxe_cleanup(struct rxe_dev *rxe)
  54 {
  55         rxe_pool_cleanup(&rxe->uc_pool);
  56         rxe_pool_cleanup(&rxe->pd_pool);
  57         rxe_pool_cleanup(&rxe->ah_pool);
  58         rxe_pool_cleanup(&rxe->srq_pool);
  59         rxe_pool_cleanup(&rxe->qp_pool);
  60         rxe_pool_cleanup(&rxe->cq_pool);
  61         rxe_pool_cleanup(&rxe->mr_pool);
  62         rxe_pool_cleanup(&rxe->mw_pool);
  63         rxe_pool_cleanup(&rxe->mc_grp_pool);
  64         rxe_pool_cleanup(&rxe->mc_elem_pool);
  65
  66         rxe_cleanup_ports(rxe);
  67 }
  68
  69 /* called when all references have been dropped */
  70 void rxe_release(struct kref *kref)
  71 {
  72         struct rxe_dev *rxe = container_of(kref, struct rxe_dev, ref_cnt);
  73
  74         rxe_cleanup(rxe);
  75         ib_dealloc_device(&rxe->ib_dev);
  76 }
  77
  78 void rxe_dev_put(struct rxe_dev *rxe)
  79 {
  80         kref_put(&rxe->ref_cnt, rxe_release);
  81 }
  82 EXPORT_SYMBOL_GPL(rxe_dev_put);
  83
  84 /* initialize rxe device parameters */
  85 static int rxe_init_device_param(struct rxe_dev *rxe)
  86 {
  87         rxe->max_inline_data                    = RXE_MAX_INLINE_DATA;
  88
  89         rxe->attr.fw_ver                        = RXE_FW_VER;
  90         rxe->attr.max_mr_size                   = RXE_MAX_MR_SIZE;
  91         rxe->attr.page_size_cap                 = RXE_PAGE_SIZE_CAP;
  92         rxe->attr.vendor_id                     = RXE_VENDOR_ID;
  93         rxe->attr.vendor_part_id                = RXE_VENDOR_PART_ID;
  94         rxe->attr.hw_ver                        = RXE_HW_VER;
  95         rxe->attr.max_qp                        = RXE_MAX_QP;
  96         rxe->attr.max_qp_wr                     = RXE_MAX_QP_WR;
  97         rxe->attr.device_cap_flags              = RXE_DEVICE_CAP_FLAGS;
  98         rxe->attr.max_sge                       = RXE_MAX_SGE;
  99         rxe->attr.max_sge_rd                    = RXE_MAX_SGE_RD;
 100         rxe->attr.max_cq                        = RXE_MAX_CQ;
 101         rxe->attr.max_cqe                       = (1 << RXE_MAX_LOG_CQE) - 1;
 102         rxe->attr.max_mr                        = RXE_MAX_MR;
 103         rxe->attr.max_pd                        = RXE_MAX_PD;
 104         rxe->attr.max_qp_rd_atom                = RXE_MAX_QP_RD_ATOM;
 105         rxe->attr.max_ee_rd_atom                = RXE_MAX_EE_RD_ATOM;
 106         rxe->attr.max_res_rd_atom               = RXE_MAX_RES_RD_ATOM;
 107         rxe->attr.max_qp_init_rd_atom           = RXE_MAX_QP_INIT_RD_ATOM;
 108         rxe->attr.max_ee_init_rd_atom           = RXE_MAX_EE_INIT_RD_ATOM;
 109         rxe->attr.atomic_cap                    = RXE_ATOMIC_CAP;
 110         rxe->attr.max_ee                        = RXE_MAX_EE;
 111         rxe->attr.max_rdd                       = RXE_MAX_RDD;
 112         rxe->attr.max_mw                        = RXE_MAX_MW;
 113         rxe->attr.max_raw_ipv6_qp               = RXE_MAX_RAW_IPV6_QP;
 114         rxe->attr.max_raw_ethy_qp               = RXE_MAX_RAW_ETHY_QP;
 115         rxe->attr.max_mcast_grp                 = RXE_MAX_MCAST_GRP;
 116         rxe->attr.max_mcast_qp_attach           = RXE_MAX_MCAST_QP_ATTACH;
 117         rxe->attr.max_total_mcast_qp_attach     = RXE_MAX_TOT_MCAST_QP_ATTACH;
 118         rxe->attr.max_ah                        = RXE_MAX_AH;
 119         rxe->attr.max_fmr                       = RXE_MAX_FMR;
 120         rxe->attr.max_map_per_fmr               = RXE_MAX_MAP_PER_FMR;
 121         rxe->attr.max_srq                       = RXE_MAX_SRQ;
 122         rxe->attr.max_srq_wr                    = RXE_MAX_SRQ_WR;
 123         rxe->attr.max_srq_sge                   = RXE_MAX_SRQ_SGE;
 124         rxe->attr.max_fast_reg_page_list_len    = RXE_MAX_FMR_PAGE_LIST_LEN;
 125         rxe->attr.max_pkeys                     = RXE_MAX_PKEYS;
 126         rxe->attr.local_ca_ack_delay            = RXE_LOCAL_CA_ACK_DELAY;
 127
 128         rxe->max_ucontext                       = RXE_MAX_UCONTEXT;
 129
 130         return 0;
 131 }
 132
 133 /* initialize port attributes */
 134 static int rxe_init_port_param(struct rxe_port *port)
 135 {
 136         port->attr.state                = RXE_PORT_STATE;
 137         port->attr.max_mtu              = RXE_PORT_MAX_MTU;
 138         port->attr.active_mtu           = RXE_PORT_ACTIVE_MTU;
 139         port->attr.gid_tbl_len          = RXE_PORT_GID_TBL_LEN;
 140         port->attr.port_cap_flags       = RXE_PORT_PORT_CAP_FLAGS;
 141         port->attr.max_msg_sz           = RXE_PORT_MAX_MSG_SZ;
 142         port->attr.bad_pkey_cntr        = RXE_PORT_BAD_PKEY_CNTR;
 143         port->attr.qkey_viol_cntr       = RXE_PORT_QKEY_VIOL_CNTR;
 144         port->attr.pkey_tbl_len         = RXE_PORT_PKEY_TBL_LEN;
 145         port->attr.lid                  = RXE_PORT_LID;
 146         port->attr.sm_lid               = RXE_PORT_SM_LID;
 147         port->attr.lmc                  = RXE_PORT_LMC;
 148         port->attr.max_vl_num           = RXE_PORT_MAX_VL_NUM;
 149         port->attr.sm_sl                = RXE_PORT_SM_SL;
 150         port->attr.subnet_timeout       = RXE_PORT_SUBNET_TIMEOUT;
 151         port->attr.init_type_reply      = RXE_PORT_INIT_TYPE_REPLY;
 152         port->attr.active_width         = RXE_PORT_ACTIVE_WIDTH;
 153         port->attr.active_speed         = RXE_PORT_ACTIVE_SPEED;
 154         port->attr.phys_state           = RXE_PORT_PHYS_STATE;
 155         port->mtu_cap                   =
 156                                 ib_mtu_enum_to_int(RXE_PORT_ACTIVE_MTU);
 157         port->subnet_prefix             = cpu_to_be64(RXE_PORT_SUBNET_PREFIX);
 158
 159         return 0;
 160 }
 161
 162 /* initialize port state, note IB convention that HCA ports are always
 163  * numbered from 1
 164  */
 165 static int rxe_init_ports(struct rxe_dev *rxe)
 166 {
 167         struct rxe_port *port = &rxe->port;
 168
 169         rxe_init_port_param(port);
 170
 171         if (!port->attr.pkey_tbl_len || !port->attr.gid_tbl_len)
 172                 return -EINVAL;
 173
 174         port->pkey_tbl = kcalloc(port->attr.pkey_tbl_len,
 175                         sizeof(*port->pkey_tbl), GFP_KERNEL);
 176
 177         if (!port->pkey_tbl)
 178                 return -ENOMEM;
 179
 180         port->pkey_tbl[0] = 0xffff;
 181         port->port_guid = rxe->ifc_ops->port_guid(rxe);
 182
 183         spin_lock_init(&port->port_lock);
 184
 185         return 0;
 186 }
 187
 188 /* init pools of managed objects */
 189 static int rxe_init_pools(struct rxe_dev *rxe)
 190 {
 191         int err;
 192
 193         err = rxe_pool_init(rxe, &rxe->uc_pool, RXE_TYPE_UC,
 194                             rxe->max_ucontext);
 195         if (err)
 196                 goto err1;
 197
 198         err = rxe_pool_init(rxe, &rxe->pd_pool, RXE_TYPE_PD,
 199                             rxe->attr.max_pd);
 200         if (err)
 201                 goto err2;
 202
 203         err = rxe_pool_init(rxe, &rxe->ah_pool, RXE_TYPE_AH,
 204                             rxe->attr.max_ah);
 205         if (err)
 206                 goto err3;
 207
 208         err = rxe_pool_init(rxe, &rxe->srq_pool, RXE_TYPE_SRQ,
 209                             rxe->attr.max_srq);
 210         if (err)
 211                 goto err4;
 212
 213         err = rxe_pool_init(rxe, &rxe->qp_pool, RXE_TYPE_QP,
 214                             rxe->attr.max_qp);
 215         if (err)
 216                 goto err5;
 217
 218         err = rxe_pool_init(rxe, &rxe->cq_pool, RXE_TYPE_CQ,
 219                             rxe->attr.max_cq);
 220         if (err)
 221                 goto err6;
 222
 223         err = rxe_pool_init(rxe, &rxe->mr_pool, RXE_TYPE_MR,
 224                             rxe->attr.max_mr);
 225         if (err)
 226                 goto err7;
 227
 228         err = rxe_pool_init(rxe, &rxe->mw_pool, RXE_TYPE_MW,
 229                             rxe->attr.max_mw);
 230         if (err)
 231                 goto err8;
 232
 233         err = rxe_pool_init(rxe, &rxe->mc_grp_pool, RXE_TYPE_MC_GRP,
 234                             rxe->attr.max_mcast_grp);
 235         if (err)
 236                 goto err9;
 237
 238         err = rxe_pool_init(rxe, &rxe->mc_elem_pool, RXE_TYPE_MC_ELEM,
 239                             rxe->attr.max_total_mcast_qp_attach);
 240         if (err)
 241                 goto err10;
 242
 243         return 0;
 244
 245 err10:
 246         rxe_pool_cleanup(&rxe->mc_grp_pool);
 247 err9:
 248         rxe_pool_cleanup(&rxe->mw_pool);
 249 err8:
 250         rxe_pool_cleanup(&rxe->mr_pool);
 251 err7:
 252         rxe_pool_cleanup(&rxe->cq_pool);
 253 err6:
 254         rxe_pool_cleanup(&rxe->qp_pool);
 255 err5:
 256         rxe_pool_cleanup(&rxe->srq_pool);
 257 err4:
 258         rxe_pool_cleanup(&rxe->ah_pool);
 259 err3:
 260         rxe_pool_cleanup(&rxe->pd_pool);
 261 err2:
 262         rxe_pool_cleanup(&rxe->uc_pool);
 263 err1:
 264         return err;
 265 }
 266
 267 /* initialize rxe device state */
 268 static int rxe_init(struct rxe_dev *rxe)
 269 {
 270         int err;
 271
 272         /* init default device parameters */
 273         rxe_init_device_param(rxe);
 274
 275         err = rxe_init_ports(rxe);
 276         if (err)
 277                 goto err1;
 278
 279         err = rxe_init_pools(rxe);
 280         if (err)
 281                 goto err2;
 282
 283         /* init pending mmap list */
 284         spin_lock_init(&rxe->mmap_offset_lock);
 285         spin_lock_init(&rxe->pending_lock);
 286         INIT_LIST_HEAD(&rxe->pending_mmaps);
 287         INIT_LIST_HEAD(&rxe->list);
 288
 289         mutex_init(&rxe->usdev_lock);
 290
 291         return 0;
 292
 293 err2:
 294         rxe_cleanup_ports(rxe);
 295 err1:
 296         return err;
 297 }
 298
 299 int rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu)
 300 {
 301         struct rxe_port *port = &rxe->port;
 302         enum ib_mtu mtu;
 303
 304         mtu = eth_mtu_int_to_enum(ndev_mtu);
 305
 306         /* Make sure that new MTU in range */
 307         mtu = mtu ? min_t(enum ib_mtu, mtu, RXE_PORT_MAX_MTU) : IB_MTU_256;
 308
 309         port->attr.active_mtu = mtu;
 310         port->mtu_cap = ib_mtu_enum_to_int(mtu);
 311
 312         return 0;
 313 }
 314 EXPORT_SYMBOL(rxe_set_mtu);
 315
 316 /* called by ifc layer to create new rxe device.
 317  * The caller should allocate memory for rxe by calling ib_alloc_device.
 318  */
 319 int rxe_add(struct rxe_dev *rxe, unsigned int mtu)
 320 {
 321         int err;
 322
 323         kref_init(&rxe->ref_cnt);
 324
 325         err = rxe_init(rxe);
 326         if (err)
 327                 goto err1;
 328
 329         err = rxe_set_mtu(rxe, mtu);
 330         if (err)
 331                 goto err1;
 332
 333         err = rxe_register_device(rxe);
 334         if (err)
 335                 goto err1;
 336
 337         return 0;
 338
 339 err1:
 340         rxe_dev_put(rxe);
 341         return err;
 342 }
 343 EXPORT_SYMBOL(rxe_add);
 344
 345 /* called by the ifc layer to remove a device */
 346 void rxe_remove(struct rxe_dev *rxe)
 347 {
 348         rxe_unregister_device(rxe);
 349
 350         rxe_dev_put(rxe);
 351 }
 352 EXPORT_SYMBOL(rxe_remove);
 353
 354 static int __init rxe_module_init(void)
 355 {
 356         int err;
 357
 358         /* initialize slab caches for managed objects */
 359         err = rxe_cache_init();
 360         if (err) {
 361                 pr_err("rxe: unable to init object pools\n");
 362                 return err;
 363         }
 364
 365         err = rxe_net_ipv4_init();
 366         if (err) {
 367                 pr_err("rxe: unable to init ipv4 tunnel\n");
 368                 rxe_cache_exit();
 369                 goto exit;
 370         }
 371
 372         err = rxe_net_ipv6_init();
 373         if (err) {
 374                 pr_err("rxe: unable to init ipv6 tunnel\n");
 375                 rxe_cache_exit();
 376                 goto exit;
 377         }
 378
 379         err = register_netdevice_notifier(&rxe_net_notifier);
 380         if (err) {
 381                 pr_err("rxe: Failed to rigister netdev notifier\n");
 382                 goto exit;
 383         }
 384
 385         pr_info("rxe: loaded\n");
 386
 387         return 0;
 388
 389 exit:
 390         rxe_release_udp_tunnel(recv_sockets.sk4);
 391         rxe_release_udp_tunnel(recv_sockets.sk6);
 392         return err;
 393 }
 394
 395 static void __exit rxe_module_exit(void)
 396 {
 397         rxe_remove_all();
 398         rxe_net_exit();
 399         rxe_cache_exit();
 400
 401         pr_info("rxe: unloaded\n");
 402 }
 403
 404 module_init(rxe_module_init);
 405 module_exit(rxe_module_exit);