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 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid) 66 { 67 u8 mac_addr[6]; 68 69 memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN); 70 guid[0] = mac_addr[0] ^ 2; 71 guid[1] = mac_addr[1]; 72 guid[2] = mac_addr[2]; 73 guid[3] = 0xff; 74 guid[4] = 0xfe; 75 guid[5] = mac_addr[3]; 76 guid[6] = mac_addr[4]; 77 guid[7] = mac_addr[5]; 78 } 79 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device, 80 u8 port_num) 81 { 82 return IB_LINK_LAYER_ETHERNET; 83 } 84 85 static int ocrdma_port_immutable(struct ib_device *ibdev, u8 port_num, 86 struct ib_port_immutable *immutable) 87 { 88 struct ib_port_attr attr; 89 struct ocrdma_dev *dev; 90 int err; 91 92 dev = get_ocrdma_dev(ibdev); 93 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE; 94 if (ocrdma_is_udp_encap_supported(dev)) 95 immutable->core_cap_flags |= RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 96 97 err = ib_query_port(ibdev, port_num, &attr); 98 if (err) 99 return err; 100 101 immutable->pkey_tbl_len = attr.pkey_tbl_len; 102 immutable->gid_tbl_len = attr.gid_tbl_len; 103 immutable->max_mad_size = IB_MGMT_MAD_SIZE; 104 105 return 0; 106 } 107 108 static void get_dev_fw_str(struct ib_device *device, char *str) 109 { 110 struct ocrdma_dev *dev = get_ocrdma_dev(device); 111 112 snprintf(str, IB_FW_VERSION_NAME_MAX, "%s", &dev->attr.fw_ver[0]); 113 } 114 115 /* OCRDMA sysfs interface */ 116 static ssize_t hw_rev_show(struct device *device, 117 struct device_attribute *attr, char *buf) 118 { 119 struct ocrdma_dev *dev = 120 rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev); 121 122 return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor); 123 } 124 static DEVICE_ATTR_RO(hw_rev); 125 126 static ssize_t hca_type_show(struct device *device, 127 struct device_attribute *attr, char *buf) 128 { 129 struct ocrdma_dev *dev = 130 rdma_device_to_drv_device(device, struct ocrdma_dev, ibdev); 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 const struct ib_device_ops ocrdma_dev_ops = { 147 .owner = THIS_MODULE, 148 .driver_id = RDMA_DRIVER_OCRDMA, 149 .uverbs_abi_ver = OCRDMA_ABI_VERSION, 150 151 .alloc_mr = ocrdma_alloc_mr, 152 .alloc_pd = ocrdma_alloc_pd, 153 .alloc_ucontext = ocrdma_alloc_ucontext, 154 .create_ah = ocrdma_create_ah, 155 .create_cq = ocrdma_create_cq, 156 .create_qp = ocrdma_create_qp, 157 .dealloc_pd = ocrdma_dealloc_pd, 158 .dealloc_ucontext = ocrdma_dealloc_ucontext, 159 .dereg_mr = ocrdma_dereg_mr, 160 .destroy_ah = ocrdma_destroy_ah, 161 .destroy_cq = ocrdma_destroy_cq, 162 .destroy_qp = ocrdma_destroy_qp, 163 .get_dev_fw_str = get_dev_fw_str, 164 .get_dma_mr = ocrdma_get_dma_mr, 165 .get_link_layer = ocrdma_link_layer, 166 .get_port_immutable = ocrdma_port_immutable, 167 .map_mr_sg = ocrdma_map_mr_sg, 168 .mmap = ocrdma_mmap, 169 .modify_port = ocrdma_modify_port, 170 .modify_qp = ocrdma_modify_qp, 171 .poll_cq = ocrdma_poll_cq, 172 .post_recv = ocrdma_post_recv, 173 .post_send = ocrdma_post_send, 174 .process_mad = ocrdma_process_mad, 175 .query_ah = ocrdma_query_ah, 176 .query_device = ocrdma_query_device, 177 .query_pkey = ocrdma_query_pkey, 178 .query_port = ocrdma_query_port, 179 .query_qp = ocrdma_query_qp, 180 .reg_user_mr = ocrdma_reg_user_mr, 181 .req_notify_cq = ocrdma_arm_cq, 182 .resize_cq = ocrdma_resize_cq, 183 184 INIT_RDMA_OBJ_SIZE(ib_ah, ocrdma_ah, ibah), 185 INIT_RDMA_OBJ_SIZE(ib_cq, ocrdma_cq, ibcq), 186 INIT_RDMA_OBJ_SIZE(ib_pd, ocrdma_pd, ibpd), 187 INIT_RDMA_OBJ_SIZE(ib_ucontext, ocrdma_ucontext, ibucontext), 188 }; 189 190 static const struct ib_device_ops ocrdma_dev_srq_ops = { 191 .create_srq = ocrdma_create_srq, 192 .destroy_srq = ocrdma_destroy_srq, 193 .modify_srq = ocrdma_modify_srq, 194 .post_srq_recv = ocrdma_post_srq_recv, 195 .query_srq = ocrdma_query_srq, 196 197 INIT_RDMA_OBJ_SIZE(ib_srq, ocrdma_srq, ibsrq), 198 }; 199 200 static int ocrdma_register_device(struct ocrdma_dev *dev) 201 { 202 int ret; 203 204 ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid); 205 BUILD_BUG_ON(sizeof(OCRDMA_NODE_DESC) > IB_DEVICE_NODE_DESC_MAX); 206 memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC, 207 sizeof(OCRDMA_NODE_DESC)); 208 dev->ibdev.uverbs_cmd_mask = 209 OCRDMA_UVERBS(GET_CONTEXT) | 210 OCRDMA_UVERBS(QUERY_DEVICE) | 211 OCRDMA_UVERBS(QUERY_PORT) | 212 OCRDMA_UVERBS(ALLOC_PD) | 213 OCRDMA_UVERBS(DEALLOC_PD) | 214 OCRDMA_UVERBS(REG_MR) | 215 OCRDMA_UVERBS(DEREG_MR) | 216 OCRDMA_UVERBS(CREATE_COMP_CHANNEL) | 217 OCRDMA_UVERBS(CREATE_CQ) | 218 OCRDMA_UVERBS(RESIZE_CQ) | 219 OCRDMA_UVERBS(DESTROY_CQ) | 220 OCRDMA_UVERBS(REQ_NOTIFY_CQ) | 221 OCRDMA_UVERBS(CREATE_QP) | 222 OCRDMA_UVERBS(MODIFY_QP) | 223 OCRDMA_UVERBS(QUERY_QP) | 224 OCRDMA_UVERBS(DESTROY_QP) | 225 OCRDMA_UVERBS(POLL_CQ) | 226 OCRDMA_UVERBS(POST_SEND) | 227 OCRDMA_UVERBS(POST_RECV); 228 229 dev->ibdev.uverbs_cmd_mask |= 230 OCRDMA_UVERBS(CREATE_AH) | 231 OCRDMA_UVERBS(MODIFY_AH) | 232 OCRDMA_UVERBS(QUERY_AH) | 233 OCRDMA_UVERBS(DESTROY_AH); 234 235 dev->ibdev.node_type = RDMA_NODE_IB_CA; 236 dev->ibdev.phys_port_cnt = 1; 237 dev->ibdev.num_comp_vectors = dev->eq_cnt; 238 239 /* mandatory to support user space verbs consumer. */ 240 dev->ibdev.dev.parent = &dev->nic_info.pdev->dev; 241 242 ib_set_device_ops(&dev->ibdev, &ocrdma_dev_ops); 243 244 if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) { 245 dev->ibdev.uverbs_cmd_mask |= 246 OCRDMA_UVERBS(CREATE_SRQ) | 247 OCRDMA_UVERBS(MODIFY_SRQ) | 248 OCRDMA_UVERBS(QUERY_SRQ) | 249 OCRDMA_UVERBS(DESTROY_SRQ) | 250 OCRDMA_UVERBS(POST_SRQ_RECV); 251 252 ib_set_device_ops(&dev->ibdev, &ocrdma_dev_srq_ops); 253 } 254 rdma_set_device_sysfs_group(&dev->ibdev, &ocrdma_attr_group); 255 ret = ib_device_set_netdev(&dev->ibdev, dev->nic_info.netdev, 1); 256 if (ret) 257 return ret; 258 259 return ib_register_device(&dev->ibdev, "ocrdma%d"); 260 } 261 262 static int ocrdma_alloc_resources(struct ocrdma_dev *dev) 263 { 264 mutex_init(&dev->dev_lock); 265 dev->cq_tbl = kcalloc(OCRDMA_MAX_CQ, sizeof(struct ocrdma_cq *), 266 GFP_KERNEL); 267 if (!dev->cq_tbl) 268 goto alloc_err; 269 270 if (dev->attr.max_qp) { 271 dev->qp_tbl = kcalloc(OCRDMA_MAX_QP, 272 sizeof(struct ocrdma_qp *), 273 GFP_KERNEL); 274 if (!dev->qp_tbl) 275 goto alloc_err; 276 } 277 278 dev->stag_arr = kcalloc(OCRDMA_MAX_STAG, sizeof(u64), GFP_KERNEL); 279 if (dev->stag_arr == NULL) 280 goto alloc_err; 281 282 ocrdma_alloc_pd_pool(dev); 283 284 if (!ocrdma_alloc_stats_resources(dev)) { 285 pr_err("%s: stats resource allocation failed\n", __func__); 286 goto alloc_err; 287 } 288 289 spin_lock_init(&dev->av_tbl.lock); 290 spin_lock_init(&dev->flush_q_lock); 291 return 0; 292 alloc_err: 293 pr_err("%s(%d) error.\n", __func__, dev->id); 294 return -ENOMEM; 295 } 296 297 static void ocrdma_free_resources(struct ocrdma_dev *dev) 298 { 299 ocrdma_release_stats_resources(dev); 300 kfree(dev->stag_arr); 301 kfree(dev->qp_tbl); 302 kfree(dev->cq_tbl); 303 } 304 305 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info) 306 { 307 int status = 0; 308 u8 lstate = 0; 309 struct ocrdma_dev *dev; 310 311 dev = ib_alloc_device(ocrdma_dev, ibdev); 312 if (!dev) { 313 pr_err("Unable to allocate ib device\n"); 314 return NULL; 315 } 316 317 dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL); 318 if (!dev->mbx_cmd) 319 goto init_err; 320 321 memcpy(&dev->nic_info, dev_info, sizeof(*dev_info)); 322 dev->id = PCI_FUNC(dev->nic_info.pdev->devfn); 323 status = ocrdma_init_hw(dev); 324 if (status) 325 goto init_err; 326 327 status = ocrdma_alloc_resources(dev); 328 if (status) 329 goto alloc_err; 330 331 ocrdma_init_service_level(dev); 332 status = ocrdma_register_device(dev); 333 if (status) 334 goto alloc_err; 335 336 /* Query Link state and update */ 337 status = ocrdma_mbx_get_link_speed(dev, NULL, &lstate); 338 if (!status) 339 ocrdma_update_link_state(dev, lstate); 340 341 /* Init stats */ 342 ocrdma_add_port_stats(dev); 343 /* Interrupt Moderation */ 344 INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task); 345 schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000)); 346 347 pr_info("%s %s: %s \"%s\" port %d\n", 348 dev_name(&dev->nic_info.pdev->dev), hca_name(dev), 349 port_speed_string(dev), dev->model_number, 350 dev->hba_port_num); 351 pr_info("%s ocrdma%d driver loaded successfully\n", 352 dev_name(&dev->nic_info.pdev->dev), dev->id); 353 return dev; 354 355 alloc_err: 356 ocrdma_free_resources(dev); 357 ocrdma_cleanup_hw(dev); 358 init_err: 359 kfree(dev->mbx_cmd); 360 ib_dealloc_device(&dev->ibdev); 361 pr_err("%s() leaving. ret=%d\n", __func__, status); 362 return NULL; 363 } 364 365 static void ocrdma_remove_free(struct ocrdma_dev *dev) 366 { 367 368 kfree(dev->mbx_cmd); 369 ib_dealloc_device(&dev->ibdev); 370 } 371 372 static void ocrdma_remove(struct ocrdma_dev *dev) 373 { 374 /* first unregister with stack to stop all the active traffic 375 * of the registered clients. 376 */ 377 cancel_delayed_work_sync(&dev->eqd_work); 378 ib_unregister_device(&dev->ibdev); 379 380 ocrdma_rem_port_stats(dev); 381 ocrdma_free_resources(dev); 382 ocrdma_cleanup_hw(dev); 383 ocrdma_remove_free(dev); 384 } 385 386 static int ocrdma_dispatch_port_active(struct ocrdma_dev *dev) 387 { 388 struct ib_event port_event; 389 390 port_event.event = IB_EVENT_PORT_ACTIVE; 391 port_event.element.port_num = 1; 392 port_event.device = &dev->ibdev; 393 ib_dispatch_event(&port_event); 394 return 0; 395 } 396 397 static int ocrdma_dispatch_port_error(struct ocrdma_dev *dev) 398 { 399 struct ib_event err_event; 400 401 err_event.event = IB_EVENT_PORT_ERR; 402 err_event.element.port_num = 1; 403 err_event.device = &dev->ibdev; 404 ib_dispatch_event(&err_event); 405 return 0; 406 } 407 408 static void ocrdma_shutdown(struct ocrdma_dev *dev) 409 { 410 ocrdma_dispatch_port_error(dev); 411 ocrdma_remove(dev); 412 } 413 414 /* event handling via NIC driver ensures that all the NIC specific 415 * initialization done before RoCE driver notifies 416 * event to stack. 417 */ 418 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event) 419 { 420 switch (event) { 421 case BE_DEV_SHUTDOWN: 422 ocrdma_shutdown(dev); 423 break; 424 default: 425 break; 426 } 427 } 428 429 void ocrdma_update_link_state(struct ocrdma_dev *dev, u8 lstate) 430 { 431 if (!(dev->flags & OCRDMA_FLAGS_LINK_STATUS_INIT)) { 432 dev->flags |= OCRDMA_FLAGS_LINK_STATUS_INIT; 433 if (!lstate) 434 return; 435 } 436 437 if (!lstate) 438 ocrdma_dispatch_port_error(dev); 439 else 440 ocrdma_dispatch_port_active(dev); 441 } 442 443 static struct ocrdma_driver ocrdma_drv = { 444 .name = "ocrdma_driver", 445 .add = ocrdma_add, 446 .remove = ocrdma_remove, 447 .state_change_handler = ocrdma_event_handler, 448 .be_abi_version = OCRDMA_BE_ROCE_ABI_VERSION, 449 }; 450 451 static int __init ocrdma_init_module(void) 452 { 453 int status; 454 455 ocrdma_init_debugfs(); 456 457 status = be_roce_register_driver(&ocrdma_drv); 458 if (status) 459 goto err_be_reg; 460 461 return 0; 462 463 err_be_reg: 464 465 return status; 466 } 467 468 static void __exit ocrdma_exit_module(void) 469 { 470 be_roce_unregister_driver(&ocrdma_drv); 471 ocrdma_rem_debugfs(); 472 } 473 474 module_init(ocrdma_init_module); 475 module_exit(ocrdma_exit_module); 476