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. *
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. *
20 * Contact Information:
21 * linux-drivers@emulex.com
25 * Costa Mesa, CA 92626
26 *******************************************************************/
28 #include <linux/module.h>
29 #include <linux/version.h>
30 #include <linux/idr.h>
31 #include <rdma/ib_verbs.h>
32 #include <rdma/ib_user_verbs.h>
33 #include <rdma/ib_addr.h>
35 #include <linux/netdevice.h>
36 #include <net/addrconf.h>
39 #include "ocrdma_verbs.h"
40 #include "ocrdma_ah.h"
42 #include "ocrdma_hw.h"
44 MODULE_VERSION(OCRDMA_ROCE_DEV_VERSION);
45 MODULE_DESCRIPTION("Emulex RoCE HCA Driver");
46 MODULE_AUTHOR("Emulex Corporation");
47 MODULE_LICENSE("GPL");
49 static LIST_HEAD(ocrdma_dev_list);
50 static DEFINE_SPINLOCK(ocrdma_devlist_lock);
51 static DEFINE_IDR(ocrdma_dev_id);
53 static union ib_gid ocrdma_zero_sgid;
55 static int ocrdma_get_instance(void)
59 /* Assign an unused number */
60 if (!idr_pre_get(&ocrdma_dev_id, GFP_KERNEL))
62 if (idr_get_new(&ocrdma_dev_id, NULL, &instance))
67 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
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];
77 guid[5] = mac_addr[3];
78 guid[6] = mac_addr[4];
79 guid[7] = mac_addr[5];
82 static void ocrdma_build_sgid_mac(union ib_gid *sgid, unsigned char *mac_addr,
83 bool is_vlan, u16 vlan_id)
85 sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
86 sgid->raw[8] = mac_addr[0] ^ 2;
87 sgid->raw[9] = mac_addr[1];
88 sgid->raw[10] = mac_addr[2];
90 sgid->raw[11] = vlan_id >> 8;
91 sgid->raw[12] = vlan_id & 0xff;
96 sgid->raw[13] = mac_addr[3];
97 sgid->raw[14] = mac_addr[4];
98 sgid->raw[15] = mac_addr[5];
101 static void ocrdma_add_sgid(struct ocrdma_dev *dev, unsigned char *mac_addr,
102 bool is_vlan, u16 vlan_id)
106 union ib_gid new_sgid;
107 int free_idx = OCRDMA_MAX_SGID;
110 memset(&ocrdma_zero_sgid, 0, sizeof(union ib_gid));
112 ocrdma_build_sgid_mac(&new_sgid, mac_addr, is_vlan, vlan_id);
114 spin_lock_irqsave(&dev->sgid_lock, flags);
115 for (i = 0; i < OCRDMA_MAX_SGID; i++) {
116 if (!memcmp(&dev->sgid_tbl[i], &ocrdma_zero_sgid,
117 sizeof(union ib_gid))) {
118 /* found free entry */
124 } else if (!memcmp(&dev->sgid_tbl[i], &new_sgid,
125 sizeof(union ib_gid))) {
126 /* entry already present, no addition is required. */
127 spin_unlock_irqrestore(&dev->sgid_lock, flags);
131 /* if entry doesn't exist and if table has some space, add entry */
133 memcpy(&dev->sgid_tbl[free_idx], &new_sgid,
134 sizeof(union ib_gid));
135 spin_unlock_irqrestore(&dev->sgid_lock, flags);
138 static bool ocrdma_del_sgid(struct ocrdma_dev *dev, unsigned char *mac_addr,
139 bool is_vlan, u16 vlan_id)
146 ocrdma_build_sgid_mac(&sgid, mac_addr, is_vlan, vlan_id);
148 spin_lock_irqsave(&dev->sgid_lock, flags);
149 /* first is default sgid, which cannot be deleted. */
150 for (i = 1; i < OCRDMA_MAX_SGID; i++) {
151 if (!memcmp(&dev->sgid_tbl[i], &sgid, sizeof(union ib_gid))) {
152 /* found matching entry */
153 memset(&dev->sgid_tbl[i], 0, sizeof(union ib_gid));
158 spin_unlock_irqrestore(&dev->sgid_lock, flags);
162 static void ocrdma_add_default_sgid(struct ocrdma_dev *dev)
164 /* GID Index 0 - Invariant manufacturer-assigned EUI-64 */
165 union ib_gid *sgid = &dev->sgid_tbl[0];
167 sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
168 ocrdma_get_guid(dev, &sgid->raw[8]);
171 static int ocrdma_build_sgid_tbl(struct ocrdma_dev *dev)
173 struct net_device *netdev, *tmp;
177 netdev = dev->nic_info.netdev;
179 ocrdma_add_default_sgid(dev);
182 for_each_netdev_rcu(&init_net, tmp) {
183 if (netdev == tmp || vlan_dev_real_dev(tmp) == netdev) {
184 if (!netif_running(tmp) || !netif_oper_up(tmp))
187 vlan_id = vlan_dev_vlan_id(tmp);
194 ocrdma_add_sgid(dev, tmp->dev_addr, is_vlan, vlan_id);
201 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
203 static int ocrdma_inet6addr_event(struct notifier_block *notifier,
204 unsigned long event, void *ptr)
206 struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
207 struct net_device *event_netdev = ifa->idev->dev;
208 struct net_device *netdev = NULL;
209 struct ib_event gid_event;
210 struct ocrdma_dev *dev;
212 bool is_vlan = false;
215 netdev = vlan_dev_real_dev(event_netdev);
216 if (netdev != event_netdev) {
218 vid = vlan_dev_vlan_id(event_netdev);
221 list_for_each_entry_rcu(dev, &ocrdma_dev_list, entry) {
222 if (dev->nic_info.netdev == netdev) {
231 if (!rdma_link_local_addr((struct in6_addr *)&ifa->addr))
234 mutex_lock(&dev->dev_lock);
237 ocrdma_add_sgid(dev, netdev->dev_addr, is_vlan, vid);
240 found = ocrdma_del_sgid(dev, netdev->dev_addr, is_vlan, vid);
242 /* found the matching entry, notify
243 * the consumers about it
245 gid_event.device = &dev->ibdev;
246 gid_event.element.port_num = 1;
247 gid_event.event = IB_EVENT_GID_CHANGE;
248 ib_dispatch_event(&gid_event);
254 mutex_unlock(&dev->dev_lock);
258 static struct notifier_block ocrdma_inet6addr_notifier = {
259 .notifier_call = ocrdma_inet6addr_event
264 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
267 return IB_LINK_LAYER_ETHERNET;
270 static int ocrdma_register_device(struct ocrdma_dev *dev)
272 strlcpy(dev->ibdev.name, "ocrdma%d", IB_DEVICE_NAME_MAX);
273 ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
274 memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
275 sizeof(OCRDMA_NODE_DESC));
276 dev->ibdev.owner = THIS_MODULE;
277 dev->ibdev.uverbs_cmd_mask =
278 OCRDMA_UVERBS(GET_CONTEXT) |
279 OCRDMA_UVERBS(QUERY_DEVICE) |
280 OCRDMA_UVERBS(QUERY_PORT) |
281 OCRDMA_UVERBS(ALLOC_PD) |
282 OCRDMA_UVERBS(DEALLOC_PD) |
283 OCRDMA_UVERBS(REG_MR) |
284 OCRDMA_UVERBS(DEREG_MR) |
285 OCRDMA_UVERBS(CREATE_COMP_CHANNEL) |
286 OCRDMA_UVERBS(CREATE_CQ) |
287 OCRDMA_UVERBS(RESIZE_CQ) |
288 OCRDMA_UVERBS(DESTROY_CQ) |
289 OCRDMA_UVERBS(REQ_NOTIFY_CQ) |
290 OCRDMA_UVERBS(CREATE_QP) |
291 OCRDMA_UVERBS(MODIFY_QP) |
292 OCRDMA_UVERBS(QUERY_QP) |
293 OCRDMA_UVERBS(DESTROY_QP) |
294 OCRDMA_UVERBS(POLL_CQ) |
295 OCRDMA_UVERBS(POST_SEND) |
296 OCRDMA_UVERBS(POST_RECV);
298 dev->ibdev.uverbs_cmd_mask |=
299 OCRDMA_UVERBS(CREATE_AH) |
300 OCRDMA_UVERBS(MODIFY_AH) |
301 OCRDMA_UVERBS(QUERY_AH) |
302 OCRDMA_UVERBS(DESTROY_AH);
304 dev->ibdev.node_type = RDMA_NODE_IB_CA;
305 dev->ibdev.phys_port_cnt = 1;
306 dev->ibdev.num_comp_vectors = 1;
308 /* mandatory verbs. */
309 dev->ibdev.query_device = ocrdma_query_device;
310 dev->ibdev.query_port = ocrdma_query_port;
311 dev->ibdev.modify_port = ocrdma_modify_port;
312 dev->ibdev.query_gid = ocrdma_query_gid;
313 dev->ibdev.get_link_layer = ocrdma_link_layer;
314 dev->ibdev.alloc_pd = ocrdma_alloc_pd;
315 dev->ibdev.dealloc_pd = ocrdma_dealloc_pd;
317 dev->ibdev.create_cq = ocrdma_create_cq;
318 dev->ibdev.destroy_cq = ocrdma_destroy_cq;
319 dev->ibdev.resize_cq = ocrdma_resize_cq;
321 dev->ibdev.create_qp = ocrdma_create_qp;
322 dev->ibdev.modify_qp = ocrdma_modify_qp;
323 dev->ibdev.query_qp = ocrdma_query_qp;
324 dev->ibdev.destroy_qp = ocrdma_destroy_qp;
326 dev->ibdev.query_pkey = ocrdma_query_pkey;
327 dev->ibdev.create_ah = ocrdma_create_ah;
328 dev->ibdev.destroy_ah = ocrdma_destroy_ah;
329 dev->ibdev.query_ah = ocrdma_query_ah;
330 dev->ibdev.modify_ah = ocrdma_modify_ah;
332 dev->ibdev.poll_cq = ocrdma_poll_cq;
333 dev->ibdev.post_send = ocrdma_post_send;
334 dev->ibdev.post_recv = ocrdma_post_recv;
335 dev->ibdev.req_notify_cq = ocrdma_arm_cq;
337 dev->ibdev.get_dma_mr = ocrdma_get_dma_mr;
338 dev->ibdev.dereg_mr = ocrdma_dereg_mr;
339 dev->ibdev.reg_user_mr = ocrdma_reg_user_mr;
341 /* mandatory to support user space verbs consumer. */
342 dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext;
343 dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext;
344 dev->ibdev.mmap = ocrdma_mmap;
345 dev->ibdev.dma_device = &dev->nic_info.pdev->dev;
347 dev->ibdev.process_mad = ocrdma_process_mad;
349 if (dev->nic_info.dev_family == OCRDMA_GEN2_FAMILY) {
350 dev->ibdev.uverbs_cmd_mask |=
351 OCRDMA_UVERBS(CREATE_SRQ) |
352 OCRDMA_UVERBS(MODIFY_SRQ) |
353 OCRDMA_UVERBS(QUERY_SRQ) |
354 OCRDMA_UVERBS(DESTROY_SRQ) |
355 OCRDMA_UVERBS(POST_SRQ_RECV);
357 dev->ibdev.create_srq = ocrdma_create_srq;
358 dev->ibdev.modify_srq = ocrdma_modify_srq;
359 dev->ibdev.query_srq = ocrdma_query_srq;
360 dev->ibdev.destroy_srq = ocrdma_destroy_srq;
361 dev->ibdev.post_srq_recv = ocrdma_post_srq_recv;
363 return ib_register_device(&dev->ibdev, NULL);
366 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
368 mutex_init(&dev->dev_lock);
369 dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
370 OCRDMA_MAX_SGID, GFP_KERNEL);
373 spin_lock_init(&dev->sgid_lock);
375 dev->cq_tbl = kzalloc(sizeof(struct ocrdma_cq *) *
376 OCRDMA_MAX_CQ, GFP_KERNEL);
380 if (dev->attr.max_qp) {
381 dev->qp_tbl = kzalloc(sizeof(struct ocrdma_qp *) *
382 OCRDMA_MAX_QP, GFP_KERNEL);
386 spin_lock_init(&dev->av_tbl.lock);
387 spin_lock_init(&dev->flush_q_lock);
390 ocrdma_err("%s(%d) error.\n", __func__, dev->id);
394 static void ocrdma_free_resources(struct ocrdma_dev *dev)
398 kfree(dev->sgid_tbl);
401 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
404 struct ocrdma_dev *dev;
406 dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev));
408 ocrdma_err("Unable to allocate ib device\n");
411 dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
415 memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
416 dev->id = ocrdma_get_instance();
420 status = ocrdma_init_hw(dev);
424 status = ocrdma_alloc_resources(dev);
428 status = ocrdma_build_sgid_tbl(dev);
432 status = ocrdma_register_device(dev);
436 spin_lock(&ocrdma_devlist_lock);
437 list_add_tail_rcu(&dev->entry, &ocrdma_dev_list);
438 spin_unlock(&ocrdma_devlist_lock);
442 ocrdma_free_resources(dev);
443 ocrdma_cleanup_hw(dev);
445 idr_remove(&ocrdma_dev_id, dev->id);
448 ib_dealloc_device(&dev->ibdev);
449 ocrdma_err("%s() leaving. ret=%d\n", __func__, status);
453 static void ocrdma_remove_free(struct rcu_head *rcu)
455 struct ocrdma_dev *dev = container_of(rcu, struct ocrdma_dev, rcu);
457 ocrdma_free_resources(dev);
458 ocrdma_cleanup_hw(dev);
460 idr_remove(&ocrdma_dev_id, dev->id);
462 ib_dealloc_device(&dev->ibdev);
465 static void ocrdma_remove(struct ocrdma_dev *dev)
467 /* first unregister with stack to stop all the active traffic
468 * of the registered clients.
470 ib_unregister_device(&dev->ibdev);
472 spin_lock(&ocrdma_devlist_lock);
473 list_del_rcu(&dev->entry);
474 spin_unlock(&ocrdma_devlist_lock);
475 call_rcu(&dev->rcu, ocrdma_remove_free);
478 static int ocrdma_open(struct ocrdma_dev *dev)
480 struct ib_event port_event;
482 port_event.event = IB_EVENT_PORT_ACTIVE;
483 port_event.element.port_num = 1;
484 port_event.device = &dev->ibdev;
485 ib_dispatch_event(&port_event);
489 static int ocrdma_close(struct ocrdma_dev *dev)
492 struct ocrdma_qp *qp, **cur_qp;
493 struct ib_event err_event;
494 struct ib_qp_attr attrs;
495 int attr_mask = IB_QP_STATE;
497 attrs.qp_state = IB_QPS_ERR;
498 mutex_lock(&dev->dev_lock);
500 cur_qp = dev->qp_tbl;
501 for (i = 0; i < OCRDMA_MAX_QP; i++) {
504 /* change the QP state to ERROR */
505 _ocrdma_modify_qp(&qp->ibqp, &attrs, attr_mask);
507 err_event.event = IB_EVENT_QP_FATAL;
508 err_event.element.qp = &qp->ibqp;
509 err_event.device = &dev->ibdev;
510 ib_dispatch_event(&err_event);
514 mutex_unlock(&dev->dev_lock);
516 err_event.event = IB_EVENT_PORT_ERR;
517 err_event.element.port_num = 1;
518 err_event.device = &dev->ibdev;
519 ib_dispatch_event(&err_event);
523 /* event handling via NIC driver ensures that all the NIC specific
524 * initialization done before RoCE driver notifies
527 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
539 static struct ocrdma_driver ocrdma_drv = {
540 .name = "ocrdma_driver",
542 .remove = ocrdma_remove,
543 .state_change_handler = ocrdma_event_handler,
546 static void ocrdma_unregister_inet6addr_notifier(void)
548 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
549 unregister_inet6addr_notifier(&ocrdma_inet6addr_notifier);
553 static int __init ocrdma_init_module(void)
557 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
558 status = register_inet6addr_notifier(&ocrdma_inet6addr_notifier);
563 status = be_roce_register_driver(&ocrdma_drv);
565 ocrdma_unregister_inet6addr_notifier();
570 static void __exit ocrdma_exit_module(void)
572 be_roce_unregister_driver(&ocrdma_drv);
573 ocrdma_unregister_inet6addr_notifier();
576 module_init(ocrdma_init_module);
577 module_exit(ocrdma_exit_module);