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