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 addrconf_addr_eui48((unsigned char *)&rxe->attr.sys_image_guid, 107 rxe->ndev->dev_addr); 108 109 rxe->max_ucontext = RXE_MAX_UCONTEXT; 110 } 111 112 /* initialize port attributes */ 113 static int rxe_init_port_param(struct rxe_port *port) 114 { 115 port->attr.state = IB_PORT_DOWN; 116 port->attr.max_mtu = IB_MTU_4096; 117 port->attr.active_mtu = IB_MTU_256; 118 port->attr.gid_tbl_len = RXE_PORT_GID_TBL_LEN; 119 port->attr.port_cap_flags = RXE_PORT_PORT_CAP_FLAGS; 120 port->attr.max_msg_sz = RXE_PORT_MAX_MSG_SZ; 121 port->attr.bad_pkey_cntr = RXE_PORT_BAD_PKEY_CNTR; 122 port->attr.qkey_viol_cntr = RXE_PORT_QKEY_VIOL_CNTR; 123 port->attr.pkey_tbl_len = RXE_PORT_PKEY_TBL_LEN; 124 port->attr.lid = RXE_PORT_LID; 125 port->attr.sm_lid = RXE_PORT_SM_LID; 126 port->attr.lmc = RXE_PORT_LMC; 127 port->attr.max_vl_num = RXE_PORT_MAX_VL_NUM; 128 port->attr.sm_sl = RXE_PORT_SM_SL; 129 port->attr.subnet_timeout = RXE_PORT_SUBNET_TIMEOUT; 130 port->attr.init_type_reply = RXE_PORT_INIT_TYPE_REPLY; 131 port->attr.active_width = RXE_PORT_ACTIVE_WIDTH; 132 port->attr.active_speed = RXE_PORT_ACTIVE_SPEED; 133 port->attr.phys_state = RXE_PORT_PHYS_STATE; 134 port->mtu_cap = ib_mtu_enum_to_int(IB_MTU_256); 135 port->subnet_prefix = cpu_to_be64(RXE_PORT_SUBNET_PREFIX); 136 137 return 0; 138 } 139 140 /* initialize port state, note IB convention that HCA ports are always 141 * numbered from 1 142 */ 143 static int rxe_init_ports(struct rxe_dev *rxe) 144 { 145 struct rxe_port *port = &rxe->port; 146 147 rxe_init_port_param(port); 148 149 if (!port->attr.pkey_tbl_len || !port->attr.gid_tbl_len) 150 return -EINVAL; 151 152 port->pkey_tbl = kcalloc(port->attr.pkey_tbl_len, 153 sizeof(*port->pkey_tbl), GFP_KERNEL); 154 155 if (!port->pkey_tbl) 156 return -ENOMEM; 157 158 port->pkey_tbl[0] = 0xffff; 159 addrconf_addr_eui48((unsigned char *)&port->port_guid, 160 rxe->ndev->dev_addr); 161 162 spin_lock_init(&port->port_lock); 163 164 return 0; 165 } 166 167 /* init pools of managed objects */ 168 static int rxe_init_pools(struct rxe_dev *rxe) 169 { 170 int err; 171 172 err = rxe_pool_init(rxe, &rxe->uc_pool, RXE_TYPE_UC, 173 rxe->max_ucontext); 174 if (err) 175 goto err1; 176 177 err = rxe_pool_init(rxe, &rxe->pd_pool, RXE_TYPE_PD, 178 rxe->attr.max_pd); 179 if (err) 180 goto err2; 181 182 err = rxe_pool_init(rxe, &rxe->ah_pool, RXE_TYPE_AH, 183 rxe->attr.max_ah); 184 if (err) 185 goto err3; 186 187 err = rxe_pool_init(rxe, &rxe->srq_pool, RXE_TYPE_SRQ, 188 rxe->attr.max_srq); 189 if (err) 190 goto err4; 191 192 err = rxe_pool_init(rxe, &rxe->qp_pool, RXE_TYPE_QP, 193 rxe->attr.max_qp); 194 if (err) 195 goto err5; 196 197 err = rxe_pool_init(rxe, &rxe->cq_pool, RXE_TYPE_CQ, 198 rxe->attr.max_cq); 199 if (err) 200 goto err6; 201 202 err = rxe_pool_init(rxe, &rxe->mr_pool, RXE_TYPE_MR, 203 rxe->attr.max_mr); 204 if (err) 205 goto err7; 206 207 err = rxe_pool_init(rxe, &rxe->mw_pool, RXE_TYPE_MW, 208 rxe->attr.max_mw); 209 if (err) 210 goto err8; 211 212 err = rxe_pool_init(rxe, &rxe->mc_grp_pool, RXE_TYPE_MC_GRP, 213 rxe->attr.max_mcast_grp); 214 if (err) 215 goto err9; 216 217 err = rxe_pool_init(rxe, &rxe->mc_elem_pool, RXE_TYPE_MC_ELEM, 218 rxe->attr.max_total_mcast_qp_attach); 219 if (err) 220 goto err10; 221 222 return 0; 223 224 err10: 225 rxe_pool_cleanup(&rxe->mc_grp_pool); 226 err9: 227 rxe_pool_cleanup(&rxe->mw_pool); 228 err8: 229 rxe_pool_cleanup(&rxe->mr_pool); 230 err7: 231 rxe_pool_cleanup(&rxe->cq_pool); 232 err6: 233 rxe_pool_cleanup(&rxe->qp_pool); 234 err5: 235 rxe_pool_cleanup(&rxe->srq_pool); 236 err4: 237 rxe_pool_cleanup(&rxe->ah_pool); 238 err3: 239 rxe_pool_cleanup(&rxe->pd_pool); 240 err2: 241 rxe_pool_cleanup(&rxe->uc_pool); 242 err1: 243 return err; 244 } 245 246 /* initialize rxe device state */ 247 static int rxe_init(struct rxe_dev *rxe) 248 { 249 int err; 250 251 /* init default device parameters */ 252 rxe_init_device_param(rxe); 253 254 err = rxe_init_ports(rxe); 255 if (err) 256 goto err1; 257 258 err = rxe_init_pools(rxe); 259 if (err) 260 goto err2; 261 262 /* init pending mmap list */ 263 spin_lock_init(&rxe->mmap_offset_lock); 264 spin_lock_init(&rxe->pending_lock); 265 INIT_LIST_HEAD(&rxe->pending_mmaps); 266 267 mutex_init(&rxe->usdev_lock); 268 269 return 0; 270 271 err2: 272 rxe_cleanup_ports(rxe); 273 err1: 274 return err; 275 } 276 277 void rxe_set_mtu(struct rxe_dev *rxe, unsigned int ndev_mtu) 278 { 279 struct rxe_port *port = &rxe->port; 280 enum ib_mtu mtu; 281 282 mtu = eth_mtu_int_to_enum(ndev_mtu); 283 284 /* Make sure that new MTU in range */ 285 mtu = mtu ? min_t(enum ib_mtu, mtu, IB_MTU_4096) : IB_MTU_256; 286 287 port->attr.active_mtu = mtu; 288 port->mtu_cap = ib_mtu_enum_to_int(mtu); 289 } 290 291 /* called by ifc layer to create new rxe device. 292 * The caller should allocate memory for rxe by calling ib_alloc_device. 293 */ 294 int rxe_add(struct rxe_dev *rxe, unsigned int mtu, const char *ibdev_name) 295 { 296 int err; 297 298 err = rxe_init(rxe); 299 if (err) 300 return err; 301 302 rxe_set_mtu(rxe, mtu); 303 304 return rxe_register_device(rxe, ibdev_name); 305 } 306 307 static int rxe_newlink(const char *ibdev_name, struct net_device *ndev) 308 { 309 struct rxe_dev *exists; 310 int err = 0; 311 312 exists = rxe_get_dev_from_net(ndev); 313 if (exists) { 314 ib_device_put(&exists->ib_dev); 315 pr_err("already configured on %s\n", ndev->name); 316 err = -EEXIST; 317 goto err; 318 } 319 320 err = rxe_net_add(ibdev_name, ndev); 321 if (err) { 322 pr_err("failed to add %s\n", ndev->name); 323 goto err; 324 } 325 err: 326 return err; 327 } 328 329 static struct rdma_link_ops rxe_link_ops = { 330 .type = "rxe", 331 .newlink = rxe_newlink, 332 }; 333 334 static int __init rxe_module_init(void) 335 { 336 int err; 337 338 /* initialize slab caches for managed objects */ 339 err = rxe_cache_init(); 340 if (err) { 341 pr_err("unable to init object pools\n"); 342 return err; 343 } 344 345 err = rxe_net_init(); 346 if (err) 347 return err; 348 349 rdma_link_register(&rxe_link_ops); 350 pr_info("loaded\n"); 351 return 0; 352 } 353 354 static void __exit rxe_module_exit(void) 355 { 356 rdma_link_unregister(&rxe_link_ops); 357 ib_unregister_driver(RDMA_DRIVER_RXE); 358 rxe_net_exit(); 359 rxe_cache_exit(); 360 361 pr_info("unloaded\n"); 362 } 363 364 late_initcall(rxe_module_init); 365 module_exit(rxe_module_exit); 366 367 MODULE_ALIAS_RDMA_LINK("rxe"); 368