1 /******************************************************************* 2 * This file is part of the Emulex RoCE Device Driver for * 3 * RoCE (RDMA over Converged Ethernet) adapters. * 4 * Copyright (C) 2008-2012 Emulex. All rights reserved. * 5 * EMULEX and SLI are trademarks of Emulex. * 6 * www.emulex.com * 7 * * 8 * This program is free software; you can redistribute it and/or * 9 * modify it under the terms of version 2 of the GNU General * 10 * Public License as published by the Free Software Foundation. * 11 * This program is distributed in the hope that it will be useful. * 12 * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND * 13 * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY, * 14 * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE * 15 * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD * 16 * TO BE LEGALLY INVALID. See the GNU General Public License for * 17 * more details, a copy of which can be found in the file COPYING * 18 * included with this package. * 19 * 20 * Contact Information: 21 * linux-drivers@emulex.com 22 * 23 * Emulex 24 * 3333 Susan Street 25 * Costa Mesa, CA 92626 26 *******************************************************************/ 27 28 #include <linux/module.h> 29 #include <linux/idr.h> 30 #include <rdma/ib_verbs.h> 31 #include <rdma/ib_user_verbs.h> 32 #include <rdma/ib_addr.h> 33 34 #include <linux/netdevice.h> 35 #include <net/addrconf.h> 36 37 #include "ocrdma.h" 38 #include "ocrdma_verbs.h" 39 #include "ocrdma_ah.h" 40 #include "be_roce.h" 41 #include "ocrdma_hw.h" 42 #include "ocrdma_stats.h" 43 #include "ocrdma_abi.h" 44 45 MODULE_VERSION(OCRDMA_ROCE_DRV_VERSION); 46 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION); 47 MODULE_AUTHOR("Emulex Corporation"); 48 MODULE_LICENSE("GPL"); 49 50 static LIST_HEAD(ocrdma_dev_list); 51 static DEFINE_SPINLOCK(ocrdma_devlist_lock); 52 static DEFINE_IDR(ocrdma_dev_id); 53 54 static union ib_gid ocrdma_zero_sgid; 55 56 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid) 57 { 58 u8 mac_addr[6]; 59 60 memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN); 61 guid[0] = mac_addr[0] ^ 2; 62 guid[1] = mac_addr[1]; 63 guid[2] = mac_addr[2]; 64 guid[3] = 0xff; 65 guid[4] = 0xfe; 66 guid[5] = mac_addr[3]; 67 guid[6] = mac_addr[4]; 68 guid[7] = mac_addr[5]; 69 } 70 71 static bool ocrdma_add_sgid(struct ocrdma_dev *dev, union ib_gid *new_sgid) 72 { 73 int i; 74 unsigned long flags; 75 76 memset(&ocrdma_zero_sgid, 0, sizeof(union ib_gid)); 77 78 79 spin_lock_irqsave(&dev->sgid_lock, flags); 80 for (i = 0; i < OCRDMA_MAX_SGID; i++) { 81 if (!memcmp(&dev->sgid_tbl[i], &ocrdma_zero_sgid, 82 sizeof(union ib_gid))) { 83 /* found free entry */ 84 memcpy(&dev->sgid_tbl[i], new_sgid, 85 sizeof(union ib_gid)); 86 spin_unlock_irqrestore(&dev->sgid_lock, flags); 87 return true; 88 } else if (!memcmp(&dev->sgid_tbl[i], new_sgid, 89 sizeof(union ib_gid))) { 90 /* entry already present, no addition is required. */ 91 spin_unlock_irqrestore(&dev->sgid_lock, flags); 92 return false; 93 } 94 } 95 spin_unlock_irqrestore(&dev->sgid_lock, flags); 96 return false; 97 } 98 99 static bool ocrdma_del_sgid(struct ocrdma_dev *dev, union ib_gid *sgid) 100 { 101 int found = false; 102 int i; 103 unsigned long flags; 104 105 106 spin_lock_irqsave(&dev->sgid_lock, flags); 107 /* first is default sgid, which cannot be deleted. */ 108 for (i = 1; i < OCRDMA_MAX_SGID; i++) { 109 if (!memcmp(&dev->sgid_tbl[i], sgid, sizeof(union ib_gid))) { 110 /* found matching entry */ 111 memset(&dev->sgid_tbl[i], 0, sizeof(union ib_gid)); 112 found = true; 113 break; 114 } 115 } 116 spin_unlock_irqrestore(&dev->sgid_lock, flags); 117 return found; 118 } 119 120 static int ocrdma_addr_event(unsigned long event, struct net_device *netdev, 121 union ib_gid *gid) 122 { 123 struct ib_event gid_event; 124 struct ocrdma_dev *dev; 125 bool found = false; 126 bool updated = false; 127 bool is_vlan = false; 128 129 is_vlan = netdev->priv_flags & IFF_802_1Q_VLAN; 130 if (is_vlan) 131 netdev = rdma_vlan_dev_real_dev(netdev); 132 133 rcu_read_lock(); 134 list_for_each_entry_rcu(dev, &ocrdma_dev_list, entry) { 135 if (dev->nic_info.netdev == netdev) { 136 found = true; 137 break; 138 } 139 } 140 rcu_read_unlock(); 141 142 if (!found) 143 return NOTIFY_DONE; 144 145 mutex_lock(&dev->dev_lock); 146 switch (event) { 147 case NETDEV_UP: 148 updated = ocrdma_add_sgid(dev, gid); 149 break; 150 case NETDEV_DOWN: 151 updated = ocrdma_del_sgid(dev, gid); 152 break; 153 default: 154 break; 155 } 156 if (updated) { 157 /* GID table updated, notify the consumers about it */ 158 gid_event.device = &dev->ibdev; 159 gid_event.element.port_num = 1; 160 gid_event.event = IB_EVENT_GID_CHANGE; 161 ib_dispatch_event(&gid_event); 162 } 163 mutex_unlock(&dev->dev_lock); 164 return NOTIFY_OK; 165 } 166 167 static int ocrdma_inetaddr_event(struct notifier_block *notifier, 168 unsigned long event, void *ptr) 169 { 170 struct in_ifaddr *ifa = ptr; 171 union ib_gid gid; 172 struct net_device *netdev = ifa->ifa_dev->dev; 173 174 ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid); 175 return ocrdma_addr_event(event, netdev, &gid); 176 } 177 178 static struct notifier_block ocrdma_inetaddr_notifier = { 179 .notifier_call = ocrdma_inetaddr_event 180 }; 181 182 #if IS_ENABLED(CONFIG_IPV6) 183 184 static int ocrdma_inet6addr_event(struct notifier_block *notifier, 185 unsigned long event, void *ptr) 186 { 187 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr; 188 union ib_gid *gid = (union ib_gid *)&ifa->addr; 189 struct net_device *netdev = ifa->idev->dev; 190 return ocrdma_addr_event(event, netdev, gid); 191 } 192 193 static struct notifier_block ocrdma_inet6addr_notifier = { 194 .notifier_call = ocrdma_inet6addr_event 195 }; 196 197 #endif /* IPV6 and VLAN */ 198 199 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device, 200 u8 port_num) 201 { 202 return IB_LINK_LAYER_ETHERNET; 203 } 204 205 static int ocrdma_register_device(struct ocrdma_dev *dev) 206 { 207 strlcpy(dev->ibdev.name, "ocrdma%d", IB_DEVICE_NAME_MAX); 208 ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid); 209 memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC, 210 sizeof(OCRDMA_NODE_DESC)); 211 dev->ibdev.owner = THIS_MODULE; 212 dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION; 213 dev->ibdev.uverbs_cmd_mask = 214 OCRDMA_UVERBS(GET_CONTEXT) | 215 OCRDMA_UVERBS(QUERY_DEVICE) | 216 OCRDMA_UVERBS(QUERY_PORT) | 217 OCRDMA_UVERBS(ALLOC_PD) | 218 OCRDMA_UVERBS(DEALLOC_PD) | 219 OCRDMA_UVERBS(REG_MR) | 220 OCRDMA_UVERBS(DEREG_MR) | 221 OCRDMA_UVERBS(CREATE_COMP_CHANNEL) | 222 OCRDMA_UVERBS(CREATE_CQ) | 223 OCRDMA_UVERBS(RESIZE_CQ) | 224 OCRDMA_UVERBS(DESTROY_CQ) | 225 OCRDMA_UVERBS(REQ_NOTIFY_CQ) | 226 OCRDMA_UVERBS(CREATE_QP) | 227 OCRDMA_UVERBS(MODIFY_QP) | 228 OCRDMA_UVERBS(QUERY_QP) | 229 OCRDMA_UVERBS(DESTROY_QP) | 230 OCRDMA_UVERBS(POLL_CQ) | 231 OCRDMA_UVERBS(POST_SEND) | 232 OCRDMA_UVERBS(POST_RECV); 233 234 dev->ibdev.uverbs_cmd_mask |= 235 OCRDMA_UVERBS(CREATE_AH) | 236 OCRDMA_UVERBS(MODIFY_AH) | 237 OCRDMA_UVERBS(QUERY_AH) | 238 OCRDMA_UVERBS(DESTROY_AH); 239 240 dev->ibdev.node_type = RDMA_NODE_IB_CA; 241 dev->ibdev.phys_port_cnt = 1; 242 dev->ibdev.num_comp_vectors = 1; 243 244 /* mandatory verbs. */ 245 dev->ibdev.query_device = ocrdma_query_device; 246 dev->ibdev.query_port = ocrdma_query_port; 247 dev->ibdev.modify_port = ocrdma_modify_port; 248 dev->ibdev.query_gid = ocrdma_query_gid; 249 dev->ibdev.get_link_layer = ocrdma_link_layer; 250 dev->ibdev.alloc_pd = ocrdma_alloc_pd; 251 dev->ibdev.dealloc_pd = ocrdma_dealloc_pd; 252 253 dev->ibdev.create_cq = ocrdma_create_cq; 254 dev->ibdev.destroy_cq = ocrdma_destroy_cq; 255 dev->ibdev.resize_cq = ocrdma_resize_cq; 256 257 dev->ibdev.create_qp = ocrdma_create_qp; 258 dev->ibdev.modify_qp = ocrdma_modify_qp; 259 dev->ibdev.query_qp = ocrdma_query_qp; 260 dev->ibdev.destroy_qp = ocrdma_destroy_qp; 261 262 dev->ibdev.query_pkey = ocrdma_query_pkey; 263 dev->ibdev.create_ah = ocrdma_create_ah; 264 dev->ibdev.destroy_ah = ocrdma_destroy_ah; 265 dev->ibdev.query_ah = ocrdma_query_ah; 266 dev->ibdev.modify_ah = ocrdma_modify_ah; 267 268 dev->ibdev.poll_cq = ocrdma_poll_cq; 269 dev->ibdev.post_send = ocrdma_post_send; 270 dev->ibdev.post_recv = ocrdma_post_recv; 271 dev->ibdev.req_notify_cq = ocrdma_arm_cq; 272 273 dev->ibdev.get_dma_mr = ocrdma_get_dma_mr; 274 dev->ibdev.reg_phys_mr = ocrdma_reg_kernel_mr; 275 dev->ibdev.dereg_mr = ocrdma_dereg_mr; 276 dev->ibdev.reg_user_mr = ocrdma_reg_user_mr; 277 278 dev->ibdev.alloc_fast_reg_mr = ocrdma_alloc_frmr; 279 dev->ibdev.alloc_fast_reg_page_list = ocrdma_alloc_frmr_page_list; 280 dev->ibdev.free_fast_reg_page_list = ocrdma_free_frmr_page_list; 281 282 /* mandatory to support user space verbs consumer. */ 283 dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext; 284 dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext; 285 dev->ibdev.mmap = ocrdma_mmap; 286 dev->ibdev.dma_device = &dev->nic_info.pdev->dev; 287 288 dev->ibdev.process_mad = ocrdma_process_mad; 289 290 if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) { 291 dev->ibdev.uverbs_cmd_mask |= 292 OCRDMA_UVERBS(CREATE_SRQ) | 293 OCRDMA_UVERBS(MODIFY_SRQ) | 294 OCRDMA_UVERBS(QUERY_SRQ) | 295 OCRDMA_UVERBS(DESTROY_SRQ) | 296 OCRDMA_UVERBS(POST_SRQ_RECV); 297 298 dev->ibdev.create_srq = ocrdma_create_srq; 299 dev->ibdev.modify_srq = ocrdma_modify_srq; 300 dev->ibdev.query_srq = ocrdma_query_srq; 301 dev->ibdev.destroy_srq = ocrdma_destroy_srq; 302 dev->ibdev.post_srq_recv = ocrdma_post_srq_recv; 303 } 304 return ib_register_device(&dev->ibdev, NULL); 305 } 306 307 static int ocrdma_alloc_resources(struct ocrdma_dev *dev) 308 { 309 mutex_init(&dev->dev_lock); 310 dev->sgid_tbl = kzalloc(sizeof(union ib_gid) * 311 OCRDMA_MAX_SGID, GFP_KERNEL); 312 if (!dev->sgid_tbl) 313 goto alloc_err; 314 spin_lock_init(&dev->sgid_lock); 315 316 dev->cq_tbl = kzalloc(sizeof(struct ocrdma_cq *) * 317 OCRDMA_MAX_CQ, GFP_KERNEL); 318 if (!dev->cq_tbl) 319 goto alloc_err; 320 321 if (dev->attr.max_qp) { 322 dev->qp_tbl = kzalloc(sizeof(struct ocrdma_qp *) * 323 OCRDMA_MAX_QP, GFP_KERNEL); 324 if (!dev->qp_tbl) 325 goto alloc_err; 326 } 327 spin_lock_init(&dev->av_tbl.lock); 328 spin_lock_init(&dev->flush_q_lock); 329 return 0; 330 alloc_err: 331 pr_err("%s(%d) error.\n", __func__, dev->id); 332 return -ENOMEM; 333 } 334 335 static void ocrdma_free_resources(struct ocrdma_dev *dev) 336 { 337 kfree(dev->qp_tbl); 338 kfree(dev->cq_tbl); 339 kfree(dev->sgid_tbl); 340 } 341 342 /* OCRDMA sysfs interface */ 343 static ssize_t show_rev(struct device *device, struct device_attribute *attr, 344 char *buf) 345 { 346 struct ocrdma_dev *dev = dev_get_drvdata(device); 347 348 return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor); 349 } 350 351 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr, 352 char *buf) 353 { 354 struct ocrdma_dev *dev = dev_get_drvdata(device); 355 356 return scnprintf(buf, PAGE_SIZE, "%s", &dev->attr.fw_ver[0]); 357 } 358 359 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL); 360 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL); 361 362 static struct device_attribute *ocrdma_attributes[] = { 363 &dev_attr_hw_rev, 364 &dev_attr_fw_ver 365 }; 366 367 static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev) 368 { 369 int i; 370 371 for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++) 372 device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]); 373 } 374 375 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info) 376 { 377 int status = 0, i; 378 struct ocrdma_dev *dev; 379 380 dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev)); 381 if (!dev) { 382 pr_err("Unable to allocate ib device\n"); 383 return NULL; 384 } 385 dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL); 386 if (!dev->mbx_cmd) 387 goto idr_err; 388 389 memcpy(&dev->nic_info, dev_info, sizeof(*dev_info)); 390 dev->id = idr_alloc(&ocrdma_dev_id, NULL, 0, 0, GFP_KERNEL); 391 if (dev->id < 0) 392 goto idr_err; 393 394 status = ocrdma_init_hw(dev); 395 if (status) 396 goto init_err; 397 398 status = ocrdma_alloc_resources(dev); 399 if (status) 400 goto alloc_err; 401 402 status = ocrdma_register_device(dev); 403 if (status) 404 goto alloc_err; 405 406 for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++) 407 if (device_create_file(&dev->ibdev.dev, ocrdma_attributes[i])) 408 goto sysfs_err; 409 spin_lock(&ocrdma_devlist_lock); 410 list_add_tail_rcu(&dev->entry, &ocrdma_dev_list); 411 spin_unlock(&ocrdma_devlist_lock); 412 /* Init stats */ 413 ocrdma_add_port_stats(dev); 414 415 pr_info("%s %s: %s \"%s\" port %d\n", 416 dev_name(&dev->nic_info.pdev->dev), hca_name(dev), 417 port_speed_string(dev), dev->model_number, 418 dev->hba_port_num); 419 pr_info("%s ocrdma%d driver loaded successfully\n", 420 dev_name(&dev->nic_info.pdev->dev), dev->id); 421 return dev; 422 423 sysfs_err: 424 ocrdma_remove_sysfiles(dev); 425 alloc_err: 426 ocrdma_free_resources(dev); 427 ocrdma_cleanup_hw(dev); 428 init_err: 429 idr_remove(&ocrdma_dev_id, dev->id); 430 idr_err: 431 kfree(dev->mbx_cmd); 432 ib_dealloc_device(&dev->ibdev); 433 pr_err("%s() leaving. ret=%d\n", __func__, status); 434 return NULL; 435 } 436 437 static void ocrdma_remove_free(struct rcu_head *rcu) 438 { 439 struct ocrdma_dev *dev = container_of(rcu, struct ocrdma_dev, rcu); 440 441 idr_remove(&ocrdma_dev_id, dev->id); 442 kfree(dev->mbx_cmd); 443 ib_dealloc_device(&dev->ibdev); 444 } 445 446 static void ocrdma_remove(struct ocrdma_dev *dev) 447 { 448 /* first unregister with stack to stop all the active traffic 449 * of the registered clients. 450 */ 451 ocrdma_rem_port_stats(dev); 452 ocrdma_remove_sysfiles(dev); 453 454 ib_unregister_device(&dev->ibdev); 455 456 spin_lock(&ocrdma_devlist_lock); 457 list_del_rcu(&dev->entry); 458 spin_unlock(&ocrdma_devlist_lock); 459 460 ocrdma_free_resources(dev); 461 ocrdma_cleanup_hw(dev); 462 463 call_rcu(&dev->rcu, ocrdma_remove_free); 464 } 465 466 static int ocrdma_open(struct ocrdma_dev *dev) 467 { 468 struct ib_event port_event; 469 470 port_event.event = IB_EVENT_PORT_ACTIVE; 471 port_event.element.port_num = 1; 472 port_event.device = &dev->ibdev; 473 ib_dispatch_event(&port_event); 474 return 0; 475 } 476 477 static int ocrdma_close(struct ocrdma_dev *dev) 478 { 479 int i; 480 struct ocrdma_qp *qp, **cur_qp; 481 struct ib_event err_event; 482 struct ib_qp_attr attrs; 483 int attr_mask = IB_QP_STATE; 484 485 attrs.qp_state = IB_QPS_ERR; 486 mutex_lock(&dev->dev_lock); 487 if (dev->qp_tbl) { 488 cur_qp = dev->qp_tbl; 489 for (i = 0; i < OCRDMA_MAX_QP; i++) { 490 qp = cur_qp[i]; 491 if (qp && qp->ibqp.qp_type != IB_QPT_GSI) { 492 /* change the QP state to ERROR */ 493 _ocrdma_modify_qp(&qp->ibqp, &attrs, attr_mask); 494 495 err_event.event = IB_EVENT_QP_FATAL; 496 err_event.element.qp = &qp->ibqp; 497 err_event.device = &dev->ibdev; 498 ib_dispatch_event(&err_event); 499 } 500 } 501 } 502 mutex_unlock(&dev->dev_lock); 503 504 err_event.event = IB_EVENT_PORT_ERR; 505 err_event.element.port_num = 1; 506 err_event.device = &dev->ibdev; 507 ib_dispatch_event(&err_event); 508 return 0; 509 } 510 511 /* event handling via NIC driver ensures that all the NIC specific 512 * initialization done before RoCE driver notifies 513 * event to stack. 514 */ 515 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event) 516 { 517 switch (event) { 518 case BE_DEV_UP: 519 ocrdma_open(dev); 520 break; 521 case BE_DEV_DOWN: 522 ocrdma_close(dev); 523 break; 524 } 525 } 526 527 static struct ocrdma_driver ocrdma_drv = { 528 .name = "ocrdma_driver", 529 .add = ocrdma_add, 530 .remove = ocrdma_remove, 531 .state_change_handler = ocrdma_event_handler, 532 .be_abi_version = OCRDMA_BE_ROCE_ABI_VERSION, 533 }; 534 535 static void ocrdma_unregister_inet6addr_notifier(void) 536 { 537 #if IS_ENABLED(CONFIG_IPV6) 538 unregister_inet6addr_notifier(&ocrdma_inet6addr_notifier); 539 #endif 540 } 541 542 static void ocrdma_unregister_inetaddr_notifier(void) 543 { 544 unregister_inetaddr_notifier(&ocrdma_inetaddr_notifier); 545 } 546 547 static int __init ocrdma_init_module(void) 548 { 549 int status; 550 551 ocrdma_init_debugfs(); 552 553 status = register_inetaddr_notifier(&ocrdma_inetaddr_notifier); 554 if (status) 555 return status; 556 557 #if IS_ENABLED(CONFIG_IPV6) 558 status = register_inet6addr_notifier(&ocrdma_inet6addr_notifier); 559 if (status) 560 goto err_notifier6; 561 #endif 562 563 status = be_roce_register_driver(&ocrdma_drv); 564 if (status) 565 goto err_be_reg; 566 567 return 0; 568 569 err_be_reg: 570 ocrdma_unregister_inet6addr_notifier(); 571 err_notifier6: 572 ocrdma_unregister_inetaddr_notifier(); 573 return status; 574 } 575 576 static void __exit ocrdma_exit_module(void) 577 { 578 be_roce_unregister_driver(&ocrdma_drv); 579 ocrdma_unregister_inet6addr_notifier(); 580 ocrdma_unregister_inetaddr_notifier(); 581 ocrdma_rem_debugfs(); 582 } 583 584 module_init(ocrdma_init_module); 585 module_exit(ocrdma_exit_module); 586