1 /* This file is part of the Emulex RoCE Device Driver for 2 * RoCE (RDMA over Converged Ethernet) adapters. 3 * Copyright (C) 2012-2015 Emulex. All rights reserved. 4 * EMULEX and SLI are trademarks of Emulex. 5 * www.emulex.com 6 * 7 * This software is available to you under a choice of one of two licenses. 8 * You may choose to be licensed under the terms of the GNU General Public 9 * License (GPL) Version 2, available from the file COPYING in the main 10 * directory of this source tree, or the BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 16 * - Redistributions of source code must retain the above copyright notice, 17 * this list of conditions and the following disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above copyright 20 * notice, this list of conditions and the following disclaimer in 21 * the documentation and/or other materials provided with the distribution. 22 * 23 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 24 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,THE 25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 26 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE 27 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 28 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 29 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR 30 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 31 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 32 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 33 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 34 * 35 * Contact Information: 36 * linux-drivers@emulex.com 37 * 38 * Emulex 39 * 3333 Susan Street 40 * Costa Mesa, CA 92626 41 */ 42 43 #include <linux/module.h> 44 #include <linux/idr.h> 45 #include <rdma/ib_verbs.h> 46 #include <rdma/ib_user_verbs.h> 47 #include <rdma/ib_addr.h> 48 #include <rdma/ib_mad.h> 49 50 #include <linux/netdevice.h> 51 #include <net/addrconf.h> 52 53 #include "ocrdma.h" 54 #include "ocrdma_verbs.h" 55 #include "ocrdma_ah.h" 56 #include "be_roce.h" 57 #include "ocrdma_hw.h" 58 #include "ocrdma_stats.h" 59 #include <rdma/ocrdma-abi.h> 60 61 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION); 62 MODULE_AUTHOR("Emulex Corporation"); 63 MODULE_LICENSE("Dual BSD/GPL"); 64 65 static DEFINE_IDR(ocrdma_dev_id); 66 67 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid) 68 { 69 u8 mac_addr[6]; 70 71 memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN); 72 guid[0] = mac_addr[0] ^ 2; 73 guid[1] = mac_addr[1]; 74 guid[2] = mac_addr[2]; 75 guid[3] = 0xff; 76 guid[4] = 0xfe; 77 guid[5] = mac_addr[3]; 78 guid[6] = mac_addr[4]; 79 guid[7] = mac_addr[5]; 80 } 81 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device, 82 u8 port_num) 83 { 84 return IB_LINK_LAYER_ETHERNET; 85 } 86 87 static int ocrdma_port_immutable(struct ib_device *ibdev, u8 port_num, 88 struct ib_port_immutable *immutable) 89 { 90 struct ib_port_attr attr; 91 struct ocrdma_dev *dev; 92 int err; 93 94 dev = get_ocrdma_dev(ibdev); 95 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE; 96 if (ocrdma_is_udp_encap_supported(dev)) 97 immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 98 99 err = ib_query_port(ibdev, port_num, &attr); 100 if (err) 101 return err; 102 103 immutable->pkey_tbl_len = attr.pkey_tbl_len; 104 immutable->gid_tbl_len = attr.gid_tbl_len; 105 immutable->max_mad_size = IB_MGMT_MAD_SIZE; 106 107 return 0; 108 } 109 110 static void get_dev_fw_str(struct ib_device *device, char *str) 111 { 112 struct ocrdma_dev *dev = get_ocrdma_dev(device); 113 114 snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", &dev->attr.fw_ver[0]); 115 } 116 117 /* OCRDMA sysfs interface */ 118 static ssize_t hw_rev_show(struct device *device, 119 struct device_attribute *attr, char *buf) 120 { 121 struct ocrdma_dev *dev = dev_get_drvdata(device); 122 123 return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor); 124 } 125 static DEVICE_ATTR_RO(hw_rev); 126 127 static ssize_t hca_type_show(struct device *device, 128 struct device_attribute *attr, char *buf) 129 { 130 struct ocrdma_dev *dev = dev_get_drvdata(device); 131 132 return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]); 133 } 134 static DEVICE_ATTR_RO(hca_type); 135 136 static struct attribute *ocrdma_attributes[] = { 137 &dev_attr_hw_rev.attr, 138 &dev_attr_hca_type.attr, 139 NULL 140 }; 141 142 static const struct attribute_group ocrdma_attr_group = { 143 .attrs = ocrdma_attributes, 144 }; 145 146 static int ocrdma_register_device(struct ocrdma_dev *dev) 147 { 148 ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid); 149 BUILD_BUG_ON(sizeof(OCRDMA_NODE_DESC) > IB_DEVICE_NODE_DESC_MAX); 150 memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC, 151 sizeof(OCRDMA_NODE_DESC)); 152 dev->ibdev.owner = THIS_MODULE; 153 dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION; 154 dev->ibdev.uverbs_cmd_mask = 155 OCRDMA_UVERBS(GET_CONTEXT) | 156 OCRDMA_UVERBS(QUERY_DEVICE) | 157 OCRDMA_UVERBS(QUERY_PORT) | 158 OCRDMA_UVERBS(ALLOC_PD) | 159 OCRDMA_UVERBS(DEALLOC_PD) | 160 OCRDMA_UVERBS(REG_MR) | 161 OCRDMA_UVERBS(DEREG_MR) | 162 OCRDMA_UVERBS(CREATE_COMP_CHANNEL) | 163 OCRDMA_UVERBS(CREATE_CQ) | 164 OCRDMA_UVERBS(RESIZE_CQ) | 165 OCRDMA_UVERBS(DESTROY_CQ) | 166 OCRDMA_UVERBS(REQ_NOTIFY_CQ) | 167 OCRDMA_UVERBS(CREATE_QP) | 168 OCRDMA_UVERBS(MODIFY_QP) | 169 OCRDMA_UVERBS(QUERY_QP) | 170 OCRDMA_UVERBS(DESTROY_QP) | 171 OCRDMA_UVERBS(POLL_CQ) | 172 OCRDMA_UVERBS(POST_SEND) | 173 OCRDMA_UVERBS(POST_RECV); 174 175 dev->ibdev.uverbs_cmd_mask |= 176 OCRDMA_UVERBS(CREATE_AH) | 177 OCRDMA_UVERBS(MODIFY_AH) | 178 OCRDMA_UVERBS(QUERY_AH) | 179 OCRDMA_UVERBS(DESTROY_AH); 180 181 dev->ibdev.node_type = RDMA_NODE_IB_CA; 182 dev->ibdev.phys_port_cnt = 1; 183 dev->ibdev.num_comp_vectors = dev->eq_cnt; 184 185 /* mandatory verbs. */ 186 dev->ibdev.query_device = ocrdma_query_device; 187 dev->ibdev.query_port = ocrdma_query_port; 188 dev->ibdev.modify_port = ocrdma_modify_port; 189 dev->ibdev.get_netdev = ocrdma_get_netdev; 190 dev->ibdev.get_link_layer = ocrdma_link_layer; 191 dev->ibdev.alloc_pd = ocrdma_alloc_pd; 192 dev->ibdev.dealloc_pd = ocrdma_dealloc_pd; 193 194 dev->ibdev.create_cq = ocrdma_create_cq; 195 dev->ibdev.destroy_cq = ocrdma_destroy_cq; 196 dev->ibdev.resize_cq = ocrdma_resize_cq; 197 198 dev->ibdev.create_qp = ocrdma_create_qp; 199 dev->ibdev.modify_qp = ocrdma_modify_qp; 200 dev->ibdev.query_qp = ocrdma_query_qp; 201 dev->ibdev.destroy_qp = ocrdma_destroy_qp; 202 203 dev->ibdev.query_pkey = ocrdma_query_pkey; 204 dev->ibdev.create_ah = ocrdma_create_ah; 205 dev->ibdev.destroy_ah = ocrdma_destroy_ah; 206 dev->ibdev.query_ah = ocrdma_query_ah; 207 208 dev->ibdev.poll_cq = ocrdma_poll_cq; 209 dev->ibdev.post_send = ocrdma_post_send; 210 dev->ibdev.post_recv = ocrdma_post_recv; 211 dev->ibdev.req_notify_cq = ocrdma_arm_cq; 212 213 dev->ibdev.get_dma_mr = ocrdma_get_dma_mr; 214 dev->ibdev.dereg_mr = ocrdma_dereg_mr; 215 dev->ibdev.reg_user_mr = ocrdma_reg_user_mr; 216 217 dev->ibdev.alloc_mr = ocrdma_alloc_mr; 218 dev->ibdev.map_mr_sg = ocrdma_map_mr_sg; 219 220 /* mandatory to support user space verbs consumer. */ 221 dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext; 222 dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext; 223 dev->ibdev.mmap = ocrdma_mmap; 224 dev->ibdev.dev.parent = &dev->nic_info.pdev->dev; 225 226 dev->ibdev.process_mad = ocrdma_process_mad; 227 dev->ibdev.get_port_immutable = ocrdma_port_immutable; 228 dev->ibdev.get_dev_fw_str = get_dev_fw_str; 229 230 if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) { 231 dev->ibdev.uverbs_cmd_mask |= 232 OCRDMA_UVERBS(CREATE_SRQ) | 233 OCRDMA_UVERBS(MODIFY_SRQ) | 234 OCRDMA_UVERBS(QUERY_SRQ) | 235 OCRDMA_UVERBS(DESTROY_SRQ) | 236 OCRDMA_UVERBS(POST_SRQ_RECV); 237 238 dev->ibdev.create_srq = ocrdma_create_srq; 239 dev->ibdev.modify_srq = ocrdma_modify_srq; 240 dev->ibdev.query_srq = ocrdma_query_srq; 241 dev->ibdev.destroy_srq = ocrdma_destroy_srq; 242 dev->ibdev.post_srq_recv = ocrdma_post_srq_recv; 243 } 244 rdma_set_device_sysfs_group(&dev->ibdev, &ocrdma_attr_group); 245 dev->ibdev.driver_id = RDMA_DRIVER_OCRDMA; 246 return ib_register_device(&dev->ibdev, "ocrdma%d", NULL); 247 } 248 249 static int ocrdma_alloc_resources(struct ocrdma_dev *dev) 250 { 251 mutex_init(&dev->dev_lock); 252 dev->cq_tbl = kcalloc(OCRDMA_MAX_CQ, sizeof(struct ocrdma_cq *), 253 GFP_KERNEL); 254 if (!dev->cq_tbl) 255 goto alloc_err; 256 257 if (dev->attr.max_qp) { 258 dev->qp_tbl = kcalloc(OCRDMA_MAX_QP, 259 sizeof(struct ocrdma_qp *), 260 GFP_KERNEL); 261 if (!dev->qp_tbl) 262 goto alloc_err; 263 } 264 265 dev->stag_arr = kcalloc(OCRDMA_MAX_STAG, sizeof(u64), GFP_KERNEL); 266 if (dev->stag_arr == NULL) 267 goto alloc_err; 268 269 ocrdma_alloc_pd_pool(dev); 270 271 if (!ocrdma_alloc_stats_resources(dev)) { 272 pr_err("%s: stats resource allocation failed\n", __func__); 273 goto alloc_err; 274 } 275 276 spin_lock_init(&dev->av_tbl.lock); 277 spin_lock_init(&dev->flush_q_lock); 278 return 0; 279 alloc_err: 280 pr_err("%s(%d) error.\n", __func__, dev->id); 281 return -ENOMEM; 282 } 283 284 static void ocrdma_free_resources(struct ocrdma_dev *dev) 285 { 286 ocrdma_release_stats_resources(dev); 287 kfree(dev->stag_arr); 288 kfree(dev->qp_tbl); 289 kfree(dev->cq_tbl); 290 } 291 292 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info) 293 { 294 int status = 0; 295 u8 lstate = 0; 296 struct ocrdma_dev *dev; 297 298 dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev)); 299 if (!dev) { 300 pr_err("Unable to allocate ib device\n"); 301 return NULL; 302 } 303 dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL); 304 if (!dev->mbx_cmd) 305 goto idr_err; 306 307 memcpy(&dev->nic_info, dev_info, sizeof(*dev_info)); 308 dev->id = idr_alloc(&ocrdma_dev_id, NULL, 0, 0, GFP_KERNEL); 309 if (dev->id < 0) 310 goto idr_err; 311 312 status = ocrdma_init_hw(dev); 313 if (status) 314 goto init_err; 315 316 status = ocrdma_alloc_resources(dev); 317 if (status) 318 goto alloc_err; 319 320 ocrdma_init_service_level(dev); 321 status = ocrdma_register_device(dev); 322 if (status) 323 goto alloc_err; 324 325 /* Query Link state and update */ 326 status = ocrdma_mbx_get_link_speed(dev, NULL, &lstate); 327 if (!status) 328 ocrdma_update_link_state(dev, lstate); 329 330 /* Init stats */ 331 ocrdma_add_port_stats(dev); 332 /* Interrupt Moderation */ 333 INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task); 334 schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000)); 335 336 pr_info("%s %s: %s \"%s\" port %d\n", 337 dev_name(&dev->nic_info.pdev->dev), hca_name(dev), 338 port_speed_string(dev), dev->model_number, 339 dev->hba_port_num); 340 pr_info("%s ocrdma%d driver loaded successfully\n", 341 dev_name(&dev->nic_info.pdev->dev), dev->id); 342 return dev; 343 344 alloc_err: 345 ocrdma_free_resources(dev); 346 ocrdma_cleanup_hw(dev); 347 init_err: 348 idr_remove(&ocrdma_dev_id, dev->id); 349 idr_err: 350 kfree(dev->mbx_cmd); 351 ib_dealloc_device(&dev->ibdev); 352 pr_err("%s() leaving. ret=%d\n", __func__, status); 353 return NULL; 354 } 355 356 static void ocrdma_remove_free(struct ocrdma_dev *dev) 357 { 358 359 idr_remove(&ocrdma_dev_id, dev->id); 360 kfree(dev->mbx_cmd); 361 ib_dealloc_device(&dev->ibdev); 362 } 363 364 static void ocrdma_remove(struct ocrdma_dev *dev) 365 { 366 /* first unregister with stack to stop all the active traffic 367 * of the registered clients. 368 */ 369 cancel_delayed_work_sync(&dev->eqd_work); 370 ib_unregister_device(&dev->ibdev); 371 372 ocrdma_rem_port_stats(dev); 373 ocrdma_free_resources(dev); 374 ocrdma_cleanup_hw(dev); 375 ocrdma_remove_free(dev); 376 } 377 378 static int ocrdma_dispatch_port_active(struct ocrdma_dev *dev) 379 { 380 struct ib_event port_event; 381 382 port_event.event = IB_EVENT_PORT_ACTIVE; 383 port_event.element.port_num = 1; 384 port_event.device = &dev->ibdev; 385 ib_dispatch_event(&port_event); 386 return 0; 387 } 388 389 static int ocrdma_dispatch_port_error(struct ocrdma_dev *dev) 390 { 391 struct ib_event err_event; 392 393 err_event.event = IB_EVENT_PORT_ERR; 394 err_event.element.port_num = 1; 395 err_event.device = &dev->ibdev; 396 ib_dispatch_event(&err_event); 397 return 0; 398 } 399 400 static void ocrdma_shutdown(struct ocrdma_dev *dev) 401 { 402 ocrdma_dispatch_port_error(dev); 403 ocrdma_remove(dev); 404 } 405 406 /* event handling via NIC driver ensures that all the NIC specific 407 * initialization done before RoCE driver notifies 408 * event to stack. 409 */ 410 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event) 411 { 412 switch (event) { 413 case BE_DEV_SHUTDOWN: 414 ocrdma_shutdown(dev); 415 break; 416 default: 417 break; 418 } 419 } 420 421 void ocrdma_update_link_state(struct ocrdma_dev *dev, u8 lstate) 422 { 423 if (!(dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)) { 424 dev->flags |= OCRDMA_FLAGS_LINK_STATUS_INIT; 425 if (!lstate) 426 return; 427 } 428 429 if (!lstate) 430 ocrdma_dispatch_port_error(dev); 431 else 432 ocrdma_dispatch_port_active(dev); 433 } 434 435 static struct ocrdma_driver ocrdma_drv = { 436 .name = "ocrdma_driver", 437 .add = ocrdma_add, 438 .remove = ocrdma_remove, 439 .state_change_handler = ocrdma_event_handler, 440 .be_abi_version = OCRDMA_BE_ROCE_ABI_VERSION, 441 }; 442 443 static int __init ocrdma_init_module(void) 444 { 445 int status; 446 447 ocrdma_init_debugfs(); 448 449 status = be_roce_register_driver(&ocrdma_drv); 450 if (status) 451 goto err_be_reg; 452 453 return 0; 454 455 err_be_reg: 456 457 return status; 458 } 459 460 static void __exit ocrdma_exit_module(void) 461 { 462 be_roce_unregister_driver(&ocrdma_drv); 463 ocrdma_rem_debugfs(); 464 idr_destroy(&ocrdma_dev_id); 465 } 466 467 module_init(ocrdma_init_module); 468 module_exit(ocrdma_exit_module); 469