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 <rdma/rdma_netlink.h>
  35 #include <net/addrconf.h>
  36 #include "rxe.h"
  37 #include "rxe_loc.h"
  38
  39 MODULE_AUTHOR("Bob Pearson, Frank Zago, John Groves, Kamal Heib");
  40 MODULE_DESCRIPTION("Soft RDMA transport");
  41 MODULE_LICENSE("Dual BSD/GPL");
  42
  43 /* free resources for all ports on a device */
  44 static void rxe_cleanup_ports(struct rxe_dev *rxe)
  45 {
  46         kfree(rxe->port.pkey_tbl);
  47         rxe->port.pkey_tbl = NULL;
  48
  49 }
  50
  51 /* free resources for a rxe device all objects created for this device must
  52  * have been destroyed
  53  */
  54 void rxe_dealloc(struct ib_device *ib_dev)
  55 {
  56         struct rxe_dev *rxe = container_of(ib_dev, struct rxe_dev, ib_dev);
  57
  58         rxe_pool_cleanup(&rxe->uc_pool);
  59         rxe_pool_cleanup(&rxe->pd_pool);
  60         rxe_pool_cleanup(&rxe->ah_pool);
  61         rxe_pool_cleanup(&rxe->srq_pool);
  62         rxe_pool_cleanup(&rxe->qp_pool);
  63         rxe_pool_cleanup(&rxe->cq_pool);
  64         rxe_pool_cleanup(&rxe->mr_pool);
  65         rxe_pool_cleanup(&rxe->mw_pool);
  66         rxe_pool_cleanup(&rxe->mc_grp_pool);
  67         rxe_pool_cleanup(&rxe->mc_elem_pool);
  68
  69         rxe_cleanup_ports(rxe);
  70
  71         if (rxe->tfm)
  72                 crypto_free_shash(rxe->tfm);
  73 }
  74
  75 /* initialize rxe device parameters */
  76 static void rxe_init_device_param(struct rxe_dev *rxe)
  77 {
  78         rxe->max_inline_data                    = RXE_MAX_INLINE_DATA;
  79
  80         rxe->attr.max_mr_size                   = RXE_MAX_MR_SIZE;
  81         rxe->attr.page_size_cap                 = RXE_PAGE_SIZE_CAP;
  82         rxe->attr.max_qp                        = RXE_MAX_QP;
  83         rxe->attr.max_qp_wr                     = RXE_MAX_QP_WR;
  84         rxe->attr.device_cap_flags              = RXE_DEVICE_CAP_FLAGS;
  85         rxe->attr.max_send_sge                  = RXE_MAX_SGE;
  86         rxe->attr.max_recv_sge                  = RXE_MAX_SGE;
  87         rxe->attr.max_sge_rd                    = RXE_MAX_SGE_RD;
  88         rxe->attr.max_cq                        = RXE_MAX_CQ;
  89         rxe->attr.max_cqe                       = (1 << RXE_MAX_LOG_CQE) - 1;
  90         rxe->attr.max_mr                        = RXE_MAX_MR;
  91         rxe->attr.max_pd                        = RXE_MAX_PD;
  92         rxe->attr.max_qp_rd_atom                = RXE_MAX_QP_RD_ATOM;
  93         rxe->attr.max_res_rd_atom               = RXE_MAX_RES_RD_ATOM;
  94         rxe->attr.max_qp_init_rd_atom           = RXE_MAX_QP_INIT_RD_ATOM;
  95         rxe->attr.atomic_cap                    = IB_ATOMIC_HCA;
  96         rxe->attr.max_mcast_grp                 = RXE_MAX_MCAST_GRP;
  97         rxe->attr.max_mcast_qp_attach           = RXE_MAX_MCAST_QP_ATTACH;
  98         rxe->attr.max_total_mcast_qp_attach     = RXE_MAX_TOT_MCAST_QP_ATTACH;
  99         rxe->attr.max_ah                        = RXE_MAX_AH;
 100         rxe->attr.max_srq                       = RXE_MAX_SRQ;
 101         rxe->attr.max_srq_wr                    = RXE_MAX_SRQ_WR;
 102         rxe->attr.max_srq_sge                   = RXE_MAX_SRQ_SGE;
 103         rxe->attr.max_fast_reg_page_list_len    = RXE_MAX_FMR_PAGE_LIST_LEN;
 104         rxe->attr.max_pkeys                     = RXE_MAX_PKEYS;
 105         rxe->attr.local_ca_ack_delay            = RXE_LOCAL_CA_ACK_DELAY;
 106
 107         rxe->max_ucontext                       = RXE_MAX_UCONTEXT;
 108 }
 109
 110 /* initialize port attributes */
 111 static int rxe_init_port_param(struct rxe_port *port)
 112 {
 113         port->attr.state                = IB_PORT_DOWN;
 114         port->attr.max_mtu              = IB_MTU_4096;
 115         port->attr.active_mtu           = IB_MTU_256;
 116         port->attr.gid_tbl_len          = RXE_PORT_GID_TBL_LEN;
 117         port->attr.port_cap_flags       = RXE_PORT_PORT_CAP_FLAGS;
 118         port->attr.max_msg_sz           = RXE_PORT_MAX_MSG_SZ;
 119         port->attr.bad_pkey_cntr        = RXE_PORT_BAD_PKEY_CNTR;
 120         port->attr.qkey_viol_cntr       = RXE_PORT_QKEY_VIOL_CNTR;
 121         port->attr.pkey_tbl_len         = RXE_PORT_PKEY_TBL_LEN;
 122         port->attr.lid                  = RXE_PORT_LID;
 123         port->attr.sm_lid               = RXE_PORT_SM_LID;
 124         port->attr.lmc                  = RXE_PORT_LMC;
 125         port->attr.max_vl_num           = RXE_PORT_MAX_VL_NUM;
 126         port->attr.sm_sl                = RXE_PORT_SM_SL;
 127         port->attr.subnet_timeout       = RXE_PORT_SUBNET_TIMEOUT;
 128         port->attr.init_type_reply      = RXE_PORT_INIT_TYPE_REPLY;
 129         port->attr.active_width         = RXE_PORT_ACTIVE_WIDTH;
 130         port->attr.active_speed         = RXE_PORT_ACTIVE_SPEED;
 131         port->attr.phys_state           = RXE_PORT_PHYS_STATE;
 132         port->mtu_cap                   = ib_mtu_enum_to_int(IB_MTU_256);
 133         port->subnet_prefix             = cpu_to_be64(RXE_PORT_SUBNET_PREFIX);
 134
 135         return 0;
 136 }
 137
 138 /* initialize port state, note IB convention that HCA ports are always
 139  * numbered from 1
 140  */
 141 static int rxe_init_ports(struct rxe_dev *rxe)
 142 {
 143         struct rxe_port *port = &rxe->port;
 144
 145         rxe_init_port_param(port);
 146
 147         if (!port->attr.pkey_tbl_len || !port->attr.gid_tbl_len)
 148                 return -EINVAL;
 149
 150         port->pkey_tbl = kcalloc(port->attr.pkey_tbl_len,
 151                         sizeof(*port->pkey_tbl), GFP_KERNEL);
 152
 153         if (!port->pkey_tbl)
 154                 return -ENOMEM;
 155
 156         port->pkey_tbl[0] = 0xffff;
 157         addrconf_addr_eui48((unsigned char *)&port->port_guid,
 158                             rxe->ndev->dev_addr);
 159
 160         spin_lock_init(&port->port_lock);
 161
 162         return 0;
 163 }
 164
 165 /* init pools of managed objects */
 166 static int rxe_init_pools(struct rxe_dev *rxe)
 167 {
 168         int err;
 169
 170         err = rxe_pool_init(rxe, &rxe->uc_pool, RXE_TYPE_UC,
 171                             rxe->max_ucontext);
 172         if (err)
 173                 goto err1;
 174
 175         err = rxe_pool_init(rxe, &rxe->pd_pool, RXE_TYPE_PD,
 176                             rxe->attr.max_pd);
 177         if (err)
 178                 goto err2;
 179
 180         err = rxe_pool_init(rxe, &rxe->ah_pool, RXE_TYPE_AH,
 181                             rxe->attr.max_ah);
 182         if (err)
 183                 goto err3;
 184
 185         err = rxe_pool_init(rxe, &rxe->srq_pool, RXE_TYPE_SRQ,
 186                             rxe->attr.max_srq);
 187         if (err)
 188                 goto err4;
 189
 190         err = rxe_pool_init(rxe, &rxe->qp_pool, RXE_TYPE_QP,
 191                             rxe->attr.max_qp);
 192         if (err)
 193                 goto err5;
 194
 195         err = rxe_pool_init(rxe, &rxe->cq_pool, RXE_TYPE_CQ,
 196                             rxe->attr.max_cq);
 197         if (err)
 198                 goto err6;
 199
 200         err = rxe_pool_init(rxe, &rxe->mr_pool, RXE_TYPE_MR,
 201                             rxe->attr.max_mr);
 202         if (err)
 203                 goto err7;
 204
 205         err = rxe_pool_init(rxe, &rxe->mw_pool, RXE_TYPE_MW,
 206                             rxe->attr.max_mw);
 207         if (err)
 208                 goto err8;
 209
 210         err = rxe_pool_init(rxe, &rxe->mc_grp_pool, RXE_TYPE_MC_GRP,
 211                             rxe->attr.max_mcast_grp);
 212         if (err)
 213                 goto err9;
 214
 215         err = rxe_pool_init(rxe, &rxe->mc_elem_pool, RXE_TYPE_MC_ELEM,
 216                             rxe->attr.max_total_mcast_qp_attach);
 217         if (err)
 218                 goto err10;
 219
 220         return 0;
 221
 222 err10:
 223         rxe_pool_cleanup(&rxe->mc_grp_pool);
 224 err9:
 225         rxe_pool_cleanup(&rxe->mw_pool);
 226 err8:
 227         rxe_pool_cleanup(&rxe->mr_pool);
 228 err7:
 229         rxe_pool_cleanup(&rxe->cq_pool);
 230 err6:
 231         rxe_pool_cleanup(&rxe->qp_pool);
 232 err5:
 233         rxe_pool_cleanup(&rxe->srq_pool);
 234 err4:
 235         rxe_pool_cleanup(&rxe->ah_pool);
 236 err3:
 237         rxe_pool_cleanup(&rxe->pd_pool);
 238 err2:
 239         rxe_pool_cleanup(&rxe->uc_pool);
 240 err1:
 241         return err;
 242 }
 243
 244 /* initialize rxe device state */
 245 static int rxe_init(struct rxe_dev *rxe)
 246 {
 247         int err;
 248
 249         /* init default device parameters */
 250         rxe_init_device_param(rxe);
 251
 252         err = rxe_init_ports(rxe);
 253         if (err)
 254                 goto err1;
 255
 256         err = rxe_init_pools(rxe);
 257         if (err)
 258                 goto err2;
 259
 260         /* init pending mmap list */
 261         spin_lock_init(&rxe->mmap_offset_lock);
 262         spin_lock_init(&rxe->pending_lock);
 263         INIT_LIST_HEAD(&rxe->pending_mmaps);
 264
 265         mutex_init(&rxe->usdev_lock);
 266
 267         return 0;
 268
 269 err2:
 270         rxe_cleanup_ports(rxe);
 271 err1:
 272         return err;
 273 }
 274
 275 void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu)
 276 {
 277         struct rxe_port *port = &rxe->port;
 278         enum ib_mtu mtu;
 279
 280         mtu = eth_mtu_int_to_enum(ndev_mtu);
 281
 282         /* Make sure that new MTU in range */
 283         mtu = mtu ? min_t(enum ib_mtu, mtu, IB_MTU_4096) : IB_MTU_256;
 284
 285         port->attr.active_mtu = mtu;
 286         port->mtu_cap = ib_mtu_enum_to_int(mtu);
 287 }
 288
 289 /* called by ifc layer to create new rxe device.
 290  * The caller should allocate memory for rxe by calling ib_alloc_device.
 291  */
 292 int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name)
 293 {
 294         int err;
 295
 296         err = rxe_init(rxe);
 297         if (err)
 298                 return err;
 299
 300         rxe_set_mtu(rxe, mtu);
 301
 302         return rxe_register_device(rxe, ibdev_name);
 303 }
 304
 305 static int rxe_newlink(const char *ibdev_name, struct net_device *ndev)
 306 {
 307         struct rxe_dev *exists;
 308         int err = 0;
 309
 310         exists = rxe_get_dev_from_net(ndev);
 311         if (exists) {
 312                 ib_device_put(&exists->ib_dev);
 313                 pr_err("already configured on %s\n", ndev->name);
 314                 err = -EEXIST;
 315                 goto err;
 316         }
 317
 318         err = rxe_net_add(ibdev_name, ndev);
 319         if (err) {
 320                 pr_err("failed to add %s\n", ndev->name);
 321                 goto err;
 322         }
 323 err:
 324         return err;
 325 }
 326
 327 static struct rdma_link_ops rxe_link_ops = {
 328         .type = "rxe",
 329         .newlink = rxe_newlink,
 330 };
 331
 332 static int __init rxe_module_init(void)
 333 {
 334         int err;
 335
 336         /* initialize slab caches for managed objects */
 337         err = rxe_cache_init();
 338         if (err) {
 339                 pr_err("unable to init object pools\n");
 340                 return err;
 341         }
 342
 343         err = rxe_net_init();
 344         if (err)
 345                 return err;
 346
 347         rdma_link_register(&rxe_link_ops);
 348         pr_info("loaded\n");
 349         return 0;
 350 }
 351
 352 static void __exit rxe_module_exit(void)
 353 {
 354         rdma_link_unregister(&rxe_link_ops);
 355         ib_unregister_driver(RDMA_DRIVER_RXE);
 356         rxe_net_exit();
 357         rxe_cache_exit();
 358
 359         pr_info("unloaded\n");
 360 }
 361
 362 late_initcall(rxe_module_init);
 363 module_exit(rxe_module_exit);
 364
 365 MODULE_ALIAS_RDMA_LINK("rxe");