xref: /openbmc/linux/drivers/infiniband/hw/ocrdma/ocrdma_main.c (revision bbde9fc1824aab58bc78c084163007dd6c03fe5b)
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 "ocrdma_abi.h"
60 
61 MODULE_VERSION(OCRDMA_ROCE_DRV_VERSION);
62 MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION);
63 MODULE_AUTHOR("Emulex Corporation");
64 MODULE_LICENSE("Dual BSD/GPL");
65 
66 static LIST_HEAD(ocrdma_dev_list);
67 static DEFINE_SPINLOCK(ocrdma_devlist_lock);
68 static DEFINE_IDR(ocrdma_dev_id);
69 
70 static union ib_gid ocrdma_zero_sgid;
71 
72 void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
73 {
74 	u8 mac_addr[6];
75 
76 	memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
77 	guid[0] = mac_addr[0] ^ 2;
78 	guid[1] = mac_addr[1];
79 	guid[2] = mac_addr[2];
80 	guid[3] = 0xff;
81 	guid[4] = 0xfe;
82 	guid[5] = mac_addr[3];
83 	guid[6] = mac_addr[4];
84 	guid[7] = mac_addr[5];
85 }
86 
87 static bool ocrdma_add_sgid(struct ocrdma_dev *dev, union ib_gid *new_sgid)
88 {
89 	int i;
90 	unsigned long flags;
91 
92 	memset(&ocrdma_zero_sgid, 0, sizeof(union ib_gid));
93 
94 
95 	spin_lock_irqsave(&dev->sgid_lock, flags);
96 	for (i = 0; i < OCRDMA_MAX_SGID; i++) {
97 		if (!memcmp(&dev->sgid_tbl[i], &ocrdma_zero_sgid,
98 			    sizeof(union ib_gid))) {
99 			/* found free entry */
100 			memcpy(&dev->sgid_tbl[i], new_sgid,
101 			       sizeof(union ib_gid));
102 			spin_unlock_irqrestore(&dev->sgid_lock, flags);
103 			return true;
104 		} else if (!memcmp(&dev->sgid_tbl[i], new_sgid,
105 				   sizeof(union ib_gid))) {
106 			/* entry already present, no addition is required. */
107 			spin_unlock_irqrestore(&dev->sgid_lock, flags);
108 			return false;
109 		}
110 	}
111 	spin_unlock_irqrestore(&dev->sgid_lock, flags);
112 	return false;
113 }
114 
115 static bool ocrdma_del_sgid(struct ocrdma_dev *dev, union ib_gid *sgid)
116 {
117 	int found = false;
118 	int i;
119 	unsigned long flags;
120 
121 
122 	spin_lock_irqsave(&dev->sgid_lock, flags);
123 	/* first is default sgid, which cannot be deleted. */
124 	for (i = 1; i < OCRDMA_MAX_SGID; i++) {
125 		if (!memcmp(&dev->sgid_tbl[i], sgid, sizeof(union ib_gid))) {
126 			/* found matching entry */
127 			memset(&dev->sgid_tbl[i], 0, sizeof(union ib_gid));
128 			found = true;
129 			break;
130 		}
131 	}
132 	spin_unlock_irqrestore(&dev->sgid_lock, flags);
133 	return found;
134 }
135 
136 static int ocrdma_addr_event(unsigned long event, struct net_device *netdev,
137 			     union ib_gid *gid)
138 {
139 	struct ib_event gid_event;
140 	struct ocrdma_dev *dev;
141 	bool found = false;
142 	bool updated = false;
143 	bool is_vlan = false;
144 
145 	is_vlan = netdev->priv_flags & IFF_802_1Q_VLAN;
146 	if (is_vlan)
147 		netdev = rdma_vlan_dev_real_dev(netdev);
148 
149 	rcu_read_lock();
150 	list_for_each_entry_rcu(dev, &ocrdma_dev_list, entry) {
151 		if (dev->nic_info.netdev == netdev) {
152 			found = true;
153 			break;
154 		}
155 	}
156 	rcu_read_unlock();
157 
158 	if (!found)
159 		return NOTIFY_DONE;
160 
161 	mutex_lock(&dev->dev_lock);
162 	switch (event) {
163 	case NETDEV_UP:
164 		updated = ocrdma_add_sgid(dev, gid);
165 		break;
166 	case NETDEV_DOWN:
167 		updated = ocrdma_del_sgid(dev, gid);
168 		break;
169 	default:
170 		break;
171 	}
172 	if (updated) {
173 		/* GID table updated, notify the consumers about it */
174 		gid_event.device = &dev->ibdev;
175 		gid_event.element.port_num = 1;
176 		gid_event.event = IB_EVENT_GID_CHANGE;
177 		ib_dispatch_event(&gid_event);
178 	}
179 	mutex_unlock(&dev->dev_lock);
180 	return NOTIFY_OK;
181 }
182 
183 static int ocrdma_inetaddr_event(struct notifier_block *notifier,
184 				  unsigned long event, void *ptr)
185 {
186 	struct in_ifaddr *ifa = ptr;
187 	union ib_gid gid;
188 	struct net_device *netdev = ifa->ifa_dev->dev;
189 
190 	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
191 	return ocrdma_addr_event(event, netdev, &gid);
192 }
193 
194 static struct notifier_block ocrdma_inetaddr_notifier = {
195 	.notifier_call = ocrdma_inetaddr_event
196 };
197 
198 #if IS_ENABLED(CONFIG_IPV6)
199 
200 static int ocrdma_inet6addr_event(struct notifier_block *notifier,
201 				  unsigned long event, void *ptr)
202 {
203 	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
204 	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
205 	struct net_device *netdev = ifa->idev->dev;
206 	return ocrdma_addr_event(event, netdev, gid);
207 }
208 
209 static struct notifier_block ocrdma_inet6addr_notifier = {
210 	.notifier_call = ocrdma_inet6addr_event
211 };
212 
213 #endif /* IPV6 and VLAN */
214 
215 static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
216 					      u8 port_num)
217 {
218 	return IB_LINK_LAYER_ETHERNET;
219 }
220 
221 static int ocrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
222 			         struct ib_port_immutable *immutable)
223 {
224 	struct ib_port_attr attr;
225 	int err;
226 
227 	err = ocrdma_query_port(ibdev, port_num, &attr);
228 	if (err)
229 		return err;
230 
231 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
232 	immutable->gid_tbl_len = attr.gid_tbl_len;
233 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
234 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
235 
236 	return 0;
237 }
238 
239 static int ocrdma_register_device(struct ocrdma_dev *dev)
240 {
241 	strlcpy(dev->ibdev.name, "ocrdma%d", IB_DEVICE_NAME_MAX);
242 	ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
243 	memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
244 	       sizeof(OCRDMA_NODE_DESC));
245 	dev->ibdev.owner = THIS_MODULE;
246 	dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION;
247 	dev->ibdev.uverbs_cmd_mask =
248 	    OCRDMA_UVERBS(GET_CONTEXT) |
249 	    OCRDMA_UVERBS(QUERY_DEVICE) |
250 	    OCRDMA_UVERBS(QUERY_PORT) |
251 	    OCRDMA_UVERBS(ALLOC_PD) |
252 	    OCRDMA_UVERBS(DEALLOC_PD) |
253 	    OCRDMA_UVERBS(REG_MR) |
254 	    OCRDMA_UVERBS(DEREG_MR) |
255 	    OCRDMA_UVERBS(CREATE_COMP_CHANNEL) |
256 	    OCRDMA_UVERBS(CREATE_CQ) |
257 	    OCRDMA_UVERBS(RESIZE_CQ) |
258 	    OCRDMA_UVERBS(DESTROY_CQ) |
259 	    OCRDMA_UVERBS(REQ_NOTIFY_CQ) |
260 	    OCRDMA_UVERBS(CREATE_QP) |
261 	    OCRDMA_UVERBS(MODIFY_QP) |
262 	    OCRDMA_UVERBS(QUERY_QP) |
263 	    OCRDMA_UVERBS(DESTROY_QP) |
264 	    OCRDMA_UVERBS(POLL_CQ) |
265 	    OCRDMA_UVERBS(POST_SEND) |
266 	    OCRDMA_UVERBS(POST_RECV);
267 
268 	dev->ibdev.uverbs_cmd_mask |=
269 	    OCRDMA_UVERBS(CREATE_AH) |
270 	     OCRDMA_UVERBS(MODIFY_AH) |
271 	     OCRDMA_UVERBS(QUERY_AH) |
272 	     OCRDMA_UVERBS(DESTROY_AH);
273 
274 	dev->ibdev.node_type = RDMA_NODE_IB_CA;
275 	dev->ibdev.phys_port_cnt = 1;
276 	dev->ibdev.num_comp_vectors = dev->eq_cnt;
277 
278 	/* mandatory verbs. */
279 	dev->ibdev.query_device = ocrdma_query_device;
280 	dev->ibdev.query_port = ocrdma_query_port;
281 	dev->ibdev.modify_port = ocrdma_modify_port;
282 	dev->ibdev.query_gid = ocrdma_query_gid;
283 	dev->ibdev.get_link_layer = ocrdma_link_layer;
284 	dev->ibdev.alloc_pd = ocrdma_alloc_pd;
285 	dev->ibdev.dealloc_pd = ocrdma_dealloc_pd;
286 
287 	dev->ibdev.create_cq = ocrdma_create_cq;
288 	dev->ibdev.destroy_cq = ocrdma_destroy_cq;
289 	dev->ibdev.resize_cq = ocrdma_resize_cq;
290 
291 	dev->ibdev.create_qp = ocrdma_create_qp;
292 	dev->ibdev.modify_qp = ocrdma_modify_qp;
293 	dev->ibdev.query_qp = ocrdma_query_qp;
294 	dev->ibdev.destroy_qp = ocrdma_destroy_qp;
295 
296 	dev->ibdev.query_pkey = ocrdma_query_pkey;
297 	dev->ibdev.create_ah = ocrdma_create_ah;
298 	dev->ibdev.destroy_ah = ocrdma_destroy_ah;
299 	dev->ibdev.query_ah = ocrdma_query_ah;
300 	dev->ibdev.modify_ah = ocrdma_modify_ah;
301 
302 	dev->ibdev.poll_cq = ocrdma_poll_cq;
303 	dev->ibdev.post_send = ocrdma_post_send;
304 	dev->ibdev.post_recv = ocrdma_post_recv;
305 	dev->ibdev.req_notify_cq = ocrdma_arm_cq;
306 
307 	dev->ibdev.get_dma_mr = ocrdma_get_dma_mr;
308 	dev->ibdev.reg_phys_mr = ocrdma_reg_kernel_mr;
309 	dev->ibdev.dereg_mr = ocrdma_dereg_mr;
310 	dev->ibdev.reg_user_mr = ocrdma_reg_user_mr;
311 
312 	dev->ibdev.alloc_fast_reg_mr = ocrdma_alloc_frmr;
313 	dev->ibdev.alloc_fast_reg_page_list = ocrdma_alloc_frmr_page_list;
314 	dev->ibdev.free_fast_reg_page_list = ocrdma_free_frmr_page_list;
315 
316 	/* mandatory to support user space verbs consumer. */
317 	dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext;
318 	dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext;
319 	dev->ibdev.mmap = ocrdma_mmap;
320 	dev->ibdev.dma_device = &dev->nic_info.pdev->dev;
321 
322 	dev->ibdev.process_mad = ocrdma_process_mad;
323 	dev->ibdev.get_port_immutable = ocrdma_port_immutable;
324 
325 	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
326 		dev->ibdev.uverbs_cmd_mask |=
327 		     OCRDMA_UVERBS(CREATE_SRQ) |
328 		     OCRDMA_UVERBS(MODIFY_SRQ) |
329 		     OCRDMA_UVERBS(QUERY_SRQ) |
330 		     OCRDMA_UVERBS(DESTROY_SRQ) |
331 		     OCRDMA_UVERBS(POST_SRQ_RECV);
332 
333 		dev->ibdev.create_srq = ocrdma_create_srq;
334 		dev->ibdev.modify_srq = ocrdma_modify_srq;
335 		dev->ibdev.query_srq = ocrdma_query_srq;
336 		dev->ibdev.destroy_srq = ocrdma_destroy_srq;
337 		dev->ibdev.post_srq_recv = ocrdma_post_srq_recv;
338 	}
339 	return ib_register_device(&dev->ibdev, NULL);
340 }
341 
342 static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
343 {
344 	mutex_init(&dev->dev_lock);
345 	dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
346 				OCRDMA_MAX_SGID, GFP_KERNEL);
347 	if (!dev->sgid_tbl)
348 		goto alloc_err;
349 	spin_lock_init(&dev->sgid_lock);
350 
351 	dev->cq_tbl = kzalloc(sizeof(struct ocrdma_cq *) *
352 			      OCRDMA_MAX_CQ, GFP_KERNEL);
353 	if (!dev->cq_tbl)
354 		goto alloc_err;
355 
356 	if (dev->attr.max_qp) {
357 		dev->qp_tbl = kzalloc(sizeof(struct ocrdma_qp *) *
358 				      OCRDMA_MAX_QP, GFP_KERNEL);
359 		if (!dev->qp_tbl)
360 			goto alloc_err;
361 	}
362 
363 	dev->stag_arr = kzalloc(sizeof(u64) * OCRDMA_MAX_STAG, GFP_KERNEL);
364 	if (dev->stag_arr == NULL)
365 		goto alloc_err;
366 
367 	ocrdma_alloc_pd_pool(dev);
368 
369 	spin_lock_init(&dev->av_tbl.lock);
370 	spin_lock_init(&dev->flush_q_lock);
371 	return 0;
372 alloc_err:
373 	pr_err("%s(%d) error.\n", __func__, dev->id);
374 	return -ENOMEM;
375 }
376 
377 static void ocrdma_free_resources(struct ocrdma_dev *dev)
378 {
379 	kfree(dev->stag_arr);
380 	kfree(dev->qp_tbl);
381 	kfree(dev->cq_tbl);
382 	kfree(dev->sgid_tbl);
383 }
384 
385 /* OCRDMA sysfs interface */
386 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
387 			char *buf)
388 {
389 	struct ocrdma_dev *dev = dev_get_drvdata(device);
390 
391 	return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor);
392 }
393 
394 static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
395 			char *buf)
396 {
397 	struct ocrdma_dev *dev = dev_get_drvdata(device);
398 
399 	return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->attr.fw_ver[0]);
400 }
401 
402 static ssize_t show_hca_type(struct device *device,
403 			     struct device_attribute *attr, char *buf)
404 {
405 	struct ocrdma_dev *dev = dev_get_drvdata(device);
406 
407 	return scnprintf(buf, PAGE_SIZE, "%s\n", &dev->model_number[0]);
408 }
409 
410 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
411 static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
412 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca_type, NULL);
413 
414 static struct device_attribute *ocrdma_attributes[] = {
415 	&dev_attr_hw_rev,
416 	&dev_attr_fw_ver,
417 	&dev_attr_hca_type
418 };
419 
420 static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev)
421 {
422 	int i;
423 
424 	for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
425 		device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]);
426 }
427 
428 static void ocrdma_add_default_sgid(struct ocrdma_dev *dev)
429 {
430 	/* GID Index 0 - Invariant manufacturer-assigned EUI-64 */
431 	union ib_gid *sgid = &dev->sgid_tbl[0];
432 
433 	sgid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
434 	ocrdma_get_guid(dev, &sgid->raw[8]);
435 }
436 
437 static void ocrdma_init_ipv4_gids(struct ocrdma_dev *dev,
438 				  struct net_device *net)
439 {
440 	struct in_device *in_dev;
441 	union ib_gid gid;
442 	in_dev = in_dev_get(net);
443 	if (in_dev) {
444 		for_ifa(in_dev) {
445 			ipv6_addr_set_v4mapped(ifa->ifa_address,
446 					       (struct in6_addr *)&gid);
447 			ocrdma_add_sgid(dev, &gid);
448 		}
449 		endfor_ifa(in_dev);
450 		in_dev_put(in_dev);
451 	}
452 }
453 
454 static void ocrdma_init_ipv6_gids(struct ocrdma_dev *dev,
455 				  struct net_device *net)
456 {
457 #if IS_ENABLED(CONFIG_IPV6)
458 	struct inet6_dev *in6_dev;
459 	union ib_gid  *pgid;
460 	struct inet6_ifaddr *ifp;
461 	in6_dev = in6_dev_get(net);
462 	if (in6_dev) {
463 		read_lock_bh(&in6_dev->lock);
464 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
465 			pgid = (union ib_gid *)&ifp->addr;
466 			ocrdma_add_sgid(dev, pgid);
467 		}
468 		read_unlock_bh(&in6_dev->lock);
469 		in6_dev_put(in6_dev);
470 	}
471 #endif
472 }
473 
474 static void ocrdma_init_gid_table(struct ocrdma_dev *dev)
475 {
476 	struct  net_device *net_dev;
477 
478 	for_each_netdev(&init_net, net_dev) {
479 		struct net_device *real_dev = rdma_vlan_dev_real_dev(net_dev) ?
480 				rdma_vlan_dev_real_dev(net_dev) : net_dev;
481 
482 		if (real_dev == dev->nic_info.netdev) {
483 			ocrdma_add_default_sgid(dev);
484 			ocrdma_init_ipv4_gids(dev, net_dev);
485 			ocrdma_init_ipv6_gids(dev, net_dev);
486 		}
487 	}
488 }
489 
490 static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
491 {
492 	int status = 0, i;
493 	struct ocrdma_dev *dev;
494 
495 	dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev));
496 	if (!dev) {
497 		pr_err("Unable to allocate ib device\n");
498 		return NULL;
499 	}
500 	dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
501 	if (!dev->mbx_cmd)
502 		goto idr_err;
503 
504 	memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
505 	dev->id = idr_alloc(&ocrdma_dev_id, NULL, 0, 0, GFP_KERNEL);
506 	if (dev->id < 0)
507 		goto idr_err;
508 
509 	status = ocrdma_init_hw(dev);
510 	if (status)
511 		goto init_err;
512 
513 	status = ocrdma_alloc_resources(dev);
514 	if (status)
515 		goto alloc_err;
516 
517 	ocrdma_init_service_level(dev);
518 	ocrdma_init_gid_table(dev);
519 	status = ocrdma_register_device(dev);
520 	if (status)
521 		goto alloc_err;
522 
523 	for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
524 		if (device_create_file(&dev->ibdev.dev, ocrdma_attributes[i]))
525 			goto sysfs_err;
526 	spin_lock(&ocrdma_devlist_lock);
527 	list_add_tail_rcu(&dev->entry, &ocrdma_dev_list);
528 	spin_unlock(&ocrdma_devlist_lock);
529 	/* Init stats */
530 	ocrdma_add_port_stats(dev);
531 	/* Interrupt Moderation */
532 	INIT_DELAYED_WORK(&dev->eqd_work, ocrdma_eqd_set_task);
533 	schedule_delayed_work(&dev->eqd_work, msecs_to_jiffies(1000));
534 
535 	pr_info("%s %s: %s \"%s\" port %d\n",
536 		dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
537 		port_speed_string(dev), dev->model_number,
538 		dev->hba_port_num);
539 	pr_info("%s ocrdma%d driver loaded successfully\n",
540 		dev_name(&dev->nic_info.pdev->dev), dev->id);
541 	return dev;
542 
543 sysfs_err:
544 	ocrdma_remove_sysfiles(dev);
545 alloc_err:
546 	ocrdma_free_resources(dev);
547 	ocrdma_cleanup_hw(dev);
548 init_err:
549 	idr_remove(&ocrdma_dev_id, dev->id);
550 idr_err:
551 	kfree(dev->mbx_cmd);
552 	ib_dealloc_device(&dev->ibdev);
553 	pr_err("%s() leaving. ret=%d\n", __func__, status);
554 	return NULL;
555 }
556 
557 static void ocrdma_remove_free(struct rcu_head *rcu)
558 {
559 	struct ocrdma_dev *dev = container_of(rcu, struct ocrdma_dev, rcu);
560 
561 	idr_remove(&ocrdma_dev_id, dev->id);
562 	kfree(dev->mbx_cmd);
563 	ib_dealloc_device(&dev->ibdev);
564 }
565 
566 static void ocrdma_remove(struct ocrdma_dev *dev)
567 {
568 	/* first unregister with stack to stop all the active traffic
569 	 * of the registered clients.
570 	 */
571 	cancel_delayed_work_sync(&dev->eqd_work);
572 	ocrdma_remove_sysfiles(dev);
573 	ib_unregister_device(&dev->ibdev);
574 
575 	ocrdma_rem_port_stats(dev);
576 
577 	spin_lock(&ocrdma_devlist_lock);
578 	list_del_rcu(&dev->entry);
579 	spin_unlock(&ocrdma_devlist_lock);
580 
581 	ocrdma_free_resources(dev);
582 	ocrdma_cleanup_hw(dev);
583 
584 	call_rcu(&dev->rcu, ocrdma_remove_free);
585 }
586 
587 static int ocrdma_open(struct ocrdma_dev *dev)
588 {
589 	struct ib_event port_event;
590 
591 	port_event.event = IB_EVENT_PORT_ACTIVE;
592 	port_event.element.port_num = 1;
593 	port_event.device = &dev->ibdev;
594 	ib_dispatch_event(&port_event);
595 	return 0;
596 }
597 
598 static int ocrdma_close(struct ocrdma_dev *dev)
599 {
600 	int i;
601 	struct ocrdma_qp *qp, **cur_qp;
602 	struct ib_event err_event;
603 	struct ib_qp_attr attrs;
604 	int attr_mask = IB_QP_STATE;
605 
606 	attrs.qp_state = IB_QPS_ERR;
607 	mutex_lock(&dev->dev_lock);
608 	if (dev->qp_tbl) {
609 		cur_qp = dev->qp_tbl;
610 		for (i = 0; i < OCRDMA_MAX_QP; i++) {
611 			qp = cur_qp[i];
612 			if (qp && qp->ibqp.qp_type != IB_QPT_GSI) {
613 				/* change the QP state to ERROR */
614 				_ocrdma_modify_qp(&qp->ibqp, &attrs, attr_mask);
615 
616 				err_event.event = IB_EVENT_QP_FATAL;
617 				err_event.element.qp = &qp->ibqp;
618 				err_event.device = &dev->ibdev;
619 				ib_dispatch_event(&err_event);
620 			}
621 		}
622 	}
623 	mutex_unlock(&dev->dev_lock);
624 
625 	err_event.event = IB_EVENT_PORT_ERR;
626 	err_event.element.port_num = 1;
627 	err_event.device = &dev->ibdev;
628 	ib_dispatch_event(&err_event);
629 	return 0;
630 }
631 
632 static void ocrdma_shutdown(struct ocrdma_dev *dev)
633 {
634 	ocrdma_close(dev);
635 	ocrdma_remove(dev);
636 }
637 
638 /* event handling via NIC driver ensures that all the NIC specific
639  * initialization done before RoCE driver notifies
640  * event to stack.
641  */
642 static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
643 {
644 	switch (event) {
645 	case BE_DEV_UP:
646 		ocrdma_open(dev);
647 		break;
648 	case BE_DEV_DOWN:
649 		ocrdma_close(dev);
650 		break;
651 	case BE_DEV_SHUTDOWN:
652 		ocrdma_shutdown(dev);
653 		break;
654 	}
655 }
656 
657 static struct ocrdma_driver ocrdma_drv = {
658 	.name			= "ocrdma_driver",
659 	.add			= ocrdma_add,
660 	.remove			= ocrdma_remove,
661 	.state_change_handler	= ocrdma_event_handler,
662 	.be_abi_version		= OCRDMA_BE_ROCE_ABI_VERSION,
663 };
664 
665 static void ocrdma_unregister_inet6addr_notifier(void)
666 {
667 #if IS_ENABLED(CONFIG_IPV6)
668 	unregister_inet6addr_notifier(&ocrdma_inet6addr_notifier);
669 #endif
670 }
671 
672 static void ocrdma_unregister_inetaddr_notifier(void)
673 {
674 	unregister_inetaddr_notifier(&ocrdma_inetaddr_notifier);
675 }
676 
677 static int __init ocrdma_init_module(void)
678 {
679 	int status;
680 
681 	ocrdma_init_debugfs();
682 
683 	status = register_inetaddr_notifier(&ocrdma_inetaddr_notifier);
684 	if (status)
685 		return status;
686 
687 #if IS_ENABLED(CONFIG_IPV6)
688 	status = register_inet6addr_notifier(&ocrdma_inet6addr_notifier);
689 	if (status)
690 		goto err_notifier6;
691 #endif
692 
693 	status = be_roce_register_driver(&ocrdma_drv);
694 	if (status)
695 		goto err_be_reg;
696 
697 	return 0;
698 
699 err_be_reg:
700 #if IS_ENABLED(CONFIG_IPV6)
701 	ocrdma_unregister_inet6addr_notifier();
702 err_notifier6:
703 #endif
704 	ocrdma_unregister_inetaddr_notifier();
705 	return status;
706 }
707 
708 static void __exit ocrdma_exit_module(void)
709 {
710 	be_roce_unregister_driver(&ocrdma_drv);
711 	ocrdma_unregister_inet6addr_notifier();
712 	ocrdma_unregister_inetaddr_notifier();
713 	ocrdma_rem_debugfs();
714 	idr_destroy(&ocrdma_dev_id);
715 }
716 
717 module_init(ocrdma_init_module);
718 module_exit(ocrdma_exit_module);
719