xref: /linux/drivers/infiniband/hw/hns/hns_roce_main.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
1 /*
2  * Copyright (c) 2016 Hisilicon Limited.
3  * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the
9  * OpenIB.org BSD license below:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      - Redistributions of source code must retain the above
16  *        copyright notice, this list of conditions and the following
17  *        disclaimer.
18  *
19  *      - Redistributions in binary form must reproduce the above
20  *        copyright notice, this list of conditions and the following
21  *        disclaimer in the documentation and/or other materials
22  *        provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 #include <linux/acpi.h>
34 #include <linux/of_platform.h>
35 #include <linux/module.h>
36 #include <rdma/ib_addr.h>
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_user_verbs.h>
39 #include <rdma/ib_cache.h>
40 #include "hns_roce_common.h"
41 #include "hns_roce_device.h"
42 #include <rdma/hns-abi.h>
43 #include "hns_roce_hem.h"
44 
45 /**
46  * hns_get_gid_index - Get gid index.
47  * @hr_dev: pointer to structure hns_roce_dev.
48  * @port:  port, value range: 0 ~ MAX
49  * @gid_index:  gid_index, value range: 0 ~ MAX
50  * Description:
51  *    N ports shared gids, allocation method as follow:
52  *		GID[0][0], GID[1][0],.....GID[N - 1][0],
53  *		GID[0][0], GID[1][0],.....GID[N - 1][0],
54  *		And so on
55  */
56 int hns_get_gid_index(struct hns_roce_dev *hr_dev, u8 port, int gid_index)
57 {
58 	return gid_index * hr_dev->caps.num_ports + port;
59 }
60 EXPORT_SYMBOL_GPL(hns_get_gid_index);
61 
62 static int hns_roce_set_mac(struct hns_roce_dev *hr_dev, u8 port, u8 *addr)
63 {
64 	u8 phy_port;
65 	u32 i = 0;
66 
67 	if (!memcmp(hr_dev->dev_addr[port], addr, MAC_ADDR_OCTET_NUM))
68 		return 0;
69 
70 	for (i = 0; i < MAC_ADDR_OCTET_NUM; i++)
71 		hr_dev->dev_addr[port][i] = addr[i];
72 
73 	phy_port = hr_dev->iboe.phy_port[port];
74 	return hr_dev->hw->set_mac(hr_dev, phy_port, addr);
75 }
76 
77 static int hns_roce_add_gid(const struct ib_gid_attr *attr, void **context)
78 {
79 	struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
80 	u8 port = attr->port_num - 1;
81 	unsigned long flags;
82 	int ret;
83 
84 	if (port >= hr_dev->caps.num_ports)
85 		return -EINVAL;
86 
87 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
88 
89 	ret = hr_dev->hw->set_gid(hr_dev, port, attr->index, &attr->gid, attr);
90 
91 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
92 
93 	return ret;
94 }
95 
96 static int hns_roce_del_gid(const struct ib_gid_attr *attr, void **context)
97 {
98 	struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
99 	struct ib_gid_attr zattr = { };
100 	u8 port = attr->port_num - 1;
101 	unsigned long flags;
102 	int ret;
103 
104 	if (port >= hr_dev->caps.num_ports)
105 		return -EINVAL;
106 
107 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
108 
109 	ret = hr_dev->hw->set_gid(hr_dev, port, attr->index, &zgid, &zattr);
110 
111 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
112 
113 	return ret;
114 }
115 
116 static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
117 			   unsigned long event)
118 {
119 	struct device *dev = hr_dev->dev;
120 	struct net_device *netdev;
121 	int ret = 0;
122 
123 	netdev = hr_dev->iboe.netdevs[port];
124 	if (!netdev) {
125 		dev_err(dev, "port(%d) can't find netdev\n", port);
126 		return -ENODEV;
127 	}
128 
129 	switch (event) {
130 	case NETDEV_UP:
131 	case NETDEV_CHANGE:
132 	case NETDEV_REGISTER:
133 	case NETDEV_CHANGEADDR:
134 		ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
135 		break;
136 	case NETDEV_DOWN:
137 		/*
138 		 * In v1 engine, only support all ports closed together.
139 		 */
140 		break;
141 	default:
142 		dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
143 		break;
144 	}
145 
146 	return ret;
147 }
148 
149 static int hns_roce_netdev_event(struct notifier_block *self,
150 				 unsigned long event, void *ptr)
151 {
152 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
153 	struct hns_roce_ib_iboe *iboe = NULL;
154 	struct hns_roce_dev *hr_dev = NULL;
155 	u8 port = 0;
156 	int ret = 0;
157 
158 	hr_dev = container_of(self, struct hns_roce_dev, iboe.nb);
159 	iboe = &hr_dev->iboe;
160 
161 	for (port = 0; port < hr_dev->caps.num_ports; port++) {
162 		if (dev == iboe->netdevs[port]) {
163 			ret = handle_en_event(hr_dev, port, event);
164 			if (ret)
165 				return NOTIFY_DONE;
166 			break;
167 		}
168 	}
169 
170 	return NOTIFY_DONE;
171 }
172 
173 static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
174 {
175 	int ret;
176 	u8 i;
177 
178 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
179 		if (hr_dev->hw->set_mtu)
180 			hr_dev->hw->set_mtu(hr_dev, hr_dev->iboe.phy_port[i],
181 					    hr_dev->caps.max_mtu);
182 		ret = hns_roce_set_mac(hr_dev, i,
183 				       hr_dev->iboe.netdevs[i]->dev_addr);
184 		if (ret)
185 			return ret;
186 	}
187 
188 	return 0;
189 }
190 
191 static int hns_roce_query_device(struct ib_device *ib_dev,
192 				 struct ib_device_attr *props,
193 				 struct ib_udata *uhw)
194 {
195 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
196 
197 	memset(props, 0, sizeof(*props));
198 
199 	props->sys_image_guid = cpu_to_be64(hr_dev->sys_image_guid);
200 	props->max_mr_size = (u64)(~(0ULL));
201 	props->page_size_cap = hr_dev->caps.page_size_cap;
202 	props->vendor_id = hr_dev->vendor_id;
203 	props->vendor_part_id = hr_dev->vendor_part_id;
204 	props->hw_ver = hr_dev->hw_rev;
205 	props->max_qp = hr_dev->caps.num_qps;
206 	props->max_qp_wr = hr_dev->caps.max_wqes;
207 	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
208 				  IB_DEVICE_RC_RNR_NAK_GEN;
209 	props->max_send_sge = hr_dev->caps.max_sq_sg;
210 	props->max_recv_sge = hr_dev->caps.max_rq_sg;
211 	props->max_sge_rd = 1;
212 	props->max_cq = hr_dev->caps.num_cqs;
213 	props->max_cqe = hr_dev->caps.max_cqes;
214 	props->max_mr = hr_dev->caps.num_mtpts;
215 	props->max_pd = hr_dev->caps.num_pds;
216 	props->max_qp_rd_atom = hr_dev->caps.max_qp_dest_rdma;
217 	props->max_qp_init_rd_atom = hr_dev->caps.max_qp_init_rdma;
218 	props->atomic_cap = IB_ATOMIC_NONE;
219 	props->max_pkeys = 1;
220 	props->local_ca_ack_delay = hr_dev->caps.local_ca_ack_delay;
221 
222 	return 0;
223 }
224 
225 static struct net_device *hns_roce_get_netdev(struct ib_device *ib_dev,
226 					      u8 port_num)
227 {
228 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
229 	struct net_device *ndev;
230 
231 	if (port_num < 1 || port_num > hr_dev->caps.num_ports)
232 		return NULL;
233 
234 	rcu_read_lock();
235 
236 	ndev = hr_dev->iboe.netdevs[port_num - 1];
237 	if (ndev)
238 		dev_hold(ndev);
239 
240 	rcu_read_unlock();
241 	return ndev;
242 }
243 
244 static int hns_roce_query_port(struct ib_device *ib_dev, u8 port_num,
245 			       struct ib_port_attr *props)
246 {
247 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
248 	struct device *dev = hr_dev->dev;
249 	struct net_device *net_dev;
250 	unsigned long flags;
251 	enum ib_mtu mtu;
252 	u8 port;
253 
254 	assert(port_num > 0);
255 	port = port_num - 1;
256 
257 	/* props being zeroed by the caller, avoid zeroing it here */
258 
259 	props->max_mtu = hr_dev->caps.max_mtu;
260 	props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
261 	props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
262 				IB_PORT_VENDOR_CLASS_SUP |
263 				IB_PORT_BOOT_MGMT_SUP;
264 	props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
265 	props->pkey_tbl_len = 1;
266 	props->active_width = IB_WIDTH_4X;
267 	props->active_speed = 1;
268 
269 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
270 
271 	net_dev = hr_dev->iboe.netdevs[port];
272 	if (!net_dev) {
273 		spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
274 		dev_err(dev, "find netdev %d failed!\r\n", port);
275 		return -EINVAL;
276 	}
277 
278 	mtu = iboe_get_mtu(net_dev->mtu);
279 	props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
280 	props->state = (netif_running(net_dev) && netif_carrier_ok(net_dev)) ?
281 			IB_PORT_ACTIVE : IB_PORT_DOWN;
282 	props->phys_state = (props->state == IB_PORT_ACTIVE) ? 5 : 3;
283 
284 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
285 
286 	return 0;
287 }
288 
289 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
290 						    u8 port_num)
291 {
292 	return IB_LINK_LAYER_ETHERNET;
293 }
294 
295 static int hns_roce_query_pkey(struct ib_device *ib_dev, u8 port, u16 index,
296 			       u16 *pkey)
297 {
298 	*pkey = PKEY_ID;
299 
300 	return 0;
301 }
302 
303 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
304 				  struct ib_device_modify *props)
305 {
306 	unsigned long flags;
307 
308 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
309 		return -EOPNOTSUPP;
310 
311 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
312 		spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
313 		memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
314 		spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
315 	}
316 
317 	return 0;
318 }
319 
320 static int hns_roce_modify_port(struct ib_device *ib_dev, u8 port_num, int mask,
321 				struct ib_port_modify *props)
322 {
323 	return 0;
324 }
325 
326 static struct ib_ucontext *hns_roce_alloc_ucontext(struct ib_device *ib_dev,
327 						   struct ib_udata *udata)
328 {
329 	int ret = 0;
330 	struct hns_roce_ucontext *context;
331 	struct hns_roce_ib_alloc_ucontext_resp resp = {};
332 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
333 
334 	if (!hr_dev->active)
335 		return ERR_PTR(-EAGAIN);
336 
337 	resp.qp_tab_size = hr_dev->caps.num_qps;
338 
339 	context = kmalloc(sizeof(*context), GFP_KERNEL);
340 	if (!context)
341 		return ERR_PTR(-ENOMEM);
342 
343 	ret = hns_roce_uar_alloc(hr_dev, &context->uar);
344 	if (ret)
345 		goto error_fail_uar_alloc;
346 
347 	INIT_LIST_HEAD(&context->vma_list);
348 	mutex_init(&context->vma_list_mutex);
349 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
350 		INIT_LIST_HEAD(&context->page_list);
351 		mutex_init(&context->page_mutex);
352 	}
353 
354 	ret = ib_copy_to_udata(udata, &resp, sizeof(resp));
355 	if (ret)
356 		goto error_fail_copy_to_udata;
357 
358 	return &context->ibucontext;
359 
360 error_fail_copy_to_udata:
361 	hns_roce_uar_free(hr_dev, &context->uar);
362 
363 error_fail_uar_alloc:
364 	kfree(context);
365 
366 	return ERR_PTR(ret);
367 }
368 
369 static int hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
370 {
371 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
372 
373 	hns_roce_uar_free(to_hr_dev(ibcontext->device), &context->uar);
374 	kfree(context);
375 
376 	return 0;
377 }
378 
379 static void hns_roce_vma_open(struct vm_area_struct *vma)
380 {
381 	vma->vm_ops = NULL;
382 }
383 
384 static void hns_roce_vma_close(struct vm_area_struct *vma)
385 {
386 	struct hns_roce_vma_data *vma_data;
387 
388 	vma_data = (struct hns_roce_vma_data *)vma->vm_private_data;
389 	vma_data->vma = NULL;
390 	mutex_lock(vma_data->vma_list_mutex);
391 	list_del(&vma_data->list);
392 	mutex_unlock(vma_data->vma_list_mutex);
393 	kfree(vma_data);
394 }
395 
396 static const struct vm_operations_struct hns_roce_vm_ops = {
397 	.open = hns_roce_vma_open,
398 	.close = hns_roce_vma_close,
399 };
400 
401 static int hns_roce_set_vma_data(struct vm_area_struct *vma,
402 				 struct hns_roce_ucontext *context)
403 {
404 	struct list_head *vma_head = &context->vma_list;
405 	struct hns_roce_vma_data *vma_data;
406 
407 	vma_data = kzalloc(sizeof(*vma_data), GFP_KERNEL);
408 	if (!vma_data)
409 		return -ENOMEM;
410 
411 	vma_data->vma = vma;
412 	vma_data->vma_list_mutex = &context->vma_list_mutex;
413 	vma->vm_private_data = vma_data;
414 	vma->vm_ops = &hns_roce_vm_ops;
415 
416 	mutex_lock(&context->vma_list_mutex);
417 	list_add(&vma_data->list, vma_head);
418 	mutex_unlock(&context->vma_list_mutex);
419 
420 	return 0;
421 }
422 
423 static int hns_roce_mmap(struct ib_ucontext *context,
424 			 struct vm_area_struct *vma)
425 {
426 	struct hns_roce_dev *hr_dev = to_hr_dev(context->device);
427 
428 	if (((vma->vm_end - vma->vm_start) % PAGE_SIZE) != 0)
429 		return -EINVAL;
430 
431 	if (vma->vm_pgoff == 0) {
432 		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
433 		if (io_remap_pfn_range(vma, vma->vm_start,
434 				       to_hr_ucontext(context)->uar.pfn,
435 				       PAGE_SIZE, vma->vm_page_prot))
436 			return -EAGAIN;
437 	} else if (vma->vm_pgoff == 1 && hr_dev->tptr_dma_addr &&
438 		   hr_dev->tptr_size) {
439 		/* vm_pgoff: 1 -- TPTR */
440 		if (io_remap_pfn_range(vma, vma->vm_start,
441 				       hr_dev->tptr_dma_addr >> PAGE_SHIFT,
442 				       hr_dev->tptr_size,
443 				       vma->vm_page_prot))
444 			return -EAGAIN;
445 	} else
446 		return -EINVAL;
447 
448 	return hns_roce_set_vma_data(vma, to_hr_ucontext(context));
449 }
450 
451 static int hns_roce_port_immutable(struct ib_device *ib_dev, u8 port_num,
452 				   struct ib_port_immutable *immutable)
453 {
454 	struct ib_port_attr attr;
455 	int ret;
456 
457 	ret = ib_query_port(ib_dev, port_num, &attr);
458 	if (ret)
459 		return ret;
460 
461 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
462 	immutable->gid_tbl_len = attr.gid_tbl_len;
463 
464 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
465 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
466 	if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
467 		immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
468 
469 	return 0;
470 }
471 
472 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
473 {
474 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
475 	struct hns_roce_vma_data *vma_data, *n;
476 	struct vm_area_struct *vma;
477 
478 	mutex_lock(&context->vma_list_mutex);
479 	list_for_each_entry_safe(vma_data, n, &context->vma_list, list) {
480 		vma = vma_data->vma;
481 		zap_vma_ptes(vma, vma->vm_start, PAGE_SIZE);
482 
483 		vma->vm_flags &= ~(VM_SHARED | VM_MAYSHARE);
484 		vma->vm_ops = NULL;
485 		list_del(&vma_data->list);
486 		kfree(vma_data);
487 	}
488 	mutex_unlock(&context->vma_list_mutex);
489 }
490 
491 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
492 {
493 	struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
494 
495 	hr_dev->active = false;
496 	unregister_netdevice_notifier(&iboe->nb);
497 	ib_unregister_device(&hr_dev->ib_dev);
498 }
499 
500 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
501 {
502 	int ret;
503 	struct hns_roce_ib_iboe *iboe = NULL;
504 	struct ib_device *ib_dev = NULL;
505 	struct device *dev = hr_dev->dev;
506 
507 	iboe = &hr_dev->iboe;
508 	spin_lock_init(&iboe->lock);
509 
510 	ib_dev = &hr_dev->ib_dev;
511 	strlcpy(ib_dev->name, "hns_%d", IB_DEVICE_NAME_MAX);
512 
513 	ib_dev->owner			= THIS_MODULE;
514 	ib_dev->node_type		= RDMA_NODE_IB_CA;
515 	ib_dev->dev.parent		= dev;
516 
517 	ib_dev->phys_port_cnt		= hr_dev->caps.num_ports;
518 	ib_dev->local_dma_lkey		= hr_dev->caps.reserved_lkey;
519 	ib_dev->num_comp_vectors	= hr_dev->caps.num_comp_vectors;
520 	ib_dev->uverbs_abi_ver		= 1;
521 	ib_dev->uverbs_cmd_mask		=
522 		(1ULL << IB_USER_VERBS_CMD_GET_CONTEXT) |
523 		(1ULL << IB_USER_VERBS_CMD_QUERY_DEVICE) |
524 		(1ULL << IB_USER_VERBS_CMD_QUERY_PORT) |
525 		(1ULL << IB_USER_VERBS_CMD_ALLOC_PD) |
526 		(1ULL << IB_USER_VERBS_CMD_DEALLOC_PD) |
527 		(1ULL << IB_USER_VERBS_CMD_REG_MR) |
528 		(1ULL << IB_USER_VERBS_CMD_DEREG_MR) |
529 		(1ULL << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
530 		(1ULL << IB_USER_VERBS_CMD_CREATE_CQ) |
531 		(1ULL << IB_USER_VERBS_CMD_DESTROY_CQ) |
532 		(1ULL << IB_USER_VERBS_CMD_CREATE_QP) |
533 		(1ULL << IB_USER_VERBS_CMD_MODIFY_QP) |
534 		(1ULL << IB_USER_VERBS_CMD_QUERY_QP) |
535 		(1ULL << IB_USER_VERBS_CMD_DESTROY_QP);
536 
537 	ib_dev->uverbs_ex_cmd_mask |=
538 		(1ULL << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
539 
540 	/* HCA||device||port */
541 	ib_dev->modify_device		= hns_roce_modify_device;
542 	ib_dev->query_device		= hns_roce_query_device;
543 	ib_dev->query_port		= hns_roce_query_port;
544 	ib_dev->modify_port		= hns_roce_modify_port;
545 	ib_dev->get_link_layer		= hns_roce_get_link_layer;
546 	ib_dev->get_netdev		= hns_roce_get_netdev;
547 	ib_dev->add_gid			= hns_roce_add_gid;
548 	ib_dev->del_gid			= hns_roce_del_gid;
549 	ib_dev->query_pkey		= hns_roce_query_pkey;
550 	ib_dev->alloc_ucontext		= hns_roce_alloc_ucontext;
551 	ib_dev->dealloc_ucontext	= hns_roce_dealloc_ucontext;
552 	ib_dev->mmap			= hns_roce_mmap;
553 
554 	/* PD */
555 	ib_dev->alloc_pd		= hns_roce_alloc_pd;
556 	ib_dev->dealloc_pd		= hns_roce_dealloc_pd;
557 
558 	/* AH */
559 	ib_dev->create_ah		= hns_roce_create_ah;
560 	ib_dev->query_ah		= hns_roce_query_ah;
561 	ib_dev->destroy_ah		= hns_roce_destroy_ah;
562 
563 	/* QP */
564 	ib_dev->create_qp		= hns_roce_create_qp;
565 	ib_dev->modify_qp		= hns_roce_modify_qp;
566 	ib_dev->query_qp		= hr_dev->hw->query_qp;
567 	ib_dev->destroy_qp		= hr_dev->hw->destroy_qp;
568 	ib_dev->post_send		= hr_dev->hw->post_send;
569 	ib_dev->post_recv		= hr_dev->hw->post_recv;
570 
571 	/* CQ */
572 	ib_dev->create_cq		= hns_roce_ib_create_cq;
573 	ib_dev->modify_cq		= hr_dev->hw->modify_cq;
574 	ib_dev->destroy_cq		= hns_roce_ib_destroy_cq;
575 	ib_dev->req_notify_cq		= hr_dev->hw->req_notify_cq;
576 	ib_dev->poll_cq			= hr_dev->hw->poll_cq;
577 
578 	/* MR */
579 	ib_dev->get_dma_mr		= hns_roce_get_dma_mr;
580 	ib_dev->reg_user_mr		= hns_roce_reg_user_mr;
581 	ib_dev->dereg_mr		= hns_roce_dereg_mr;
582 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR) {
583 		ib_dev->rereg_user_mr	= hns_roce_rereg_user_mr;
584 		ib_dev->uverbs_cmd_mask |= (1ULL << IB_USER_VERBS_CMD_REREG_MR);
585 	}
586 
587 	/* OTHERS */
588 	ib_dev->get_port_immutable	= hns_roce_port_immutable;
589 	ib_dev->disassociate_ucontext	= hns_roce_disassociate_ucontext;
590 
591 	ib_dev->driver_id = RDMA_DRIVER_HNS;
592 	ret = ib_register_device(ib_dev, NULL);
593 	if (ret) {
594 		dev_err(dev, "ib_register_device failed!\n");
595 		return ret;
596 	}
597 
598 	ret = hns_roce_setup_mtu_mac(hr_dev);
599 	if (ret) {
600 		dev_err(dev, "setup_mtu_mac failed!\n");
601 		goto error_failed_setup_mtu_mac;
602 	}
603 
604 	iboe->nb.notifier_call = hns_roce_netdev_event;
605 	ret = register_netdevice_notifier(&iboe->nb);
606 	if (ret) {
607 		dev_err(dev, "register_netdevice_notifier failed!\n");
608 		goto error_failed_setup_mtu_mac;
609 	}
610 
611 	hr_dev->active = true;
612 	return 0;
613 
614 error_failed_setup_mtu_mac:
615 	ib_unregister_device(ib_dev);
616 
617 	return ret;
618 }
619 
620 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
621 {
622 	int ret;
623 	struct device *dev = hr_dev->dev;
624 
625 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtt_table,
626 				      HEM_TYPE_MTT, hr_dev->caps.mtt_entry_sz,
627 				      hr_dev->caps.num_mtt_segs, 1);
628 	if (ret) {
629 		dev_err(dev, "Failed to init MTT context memory, aborting.\n");
630 		return ret;
631 	}
632 
633 	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE)) {
634 		ret = hns_roce_init_hem_table(hr_dev,
635 				      &hr_dev->mr_table.mtt_cqe_table,
636 				      HEM_TYPE_CQE, hr_dev->caps.mtt_entry_sz,
637 				      hr_dev->caps.num_cqe_segs, 1);
638 		if (ret) {
639 			dev_err(dev, "Failed to init MTT CQE context memory, aborting.\n");
640 			goto err_unmap_cqe;
641 		}
642 	}
643 
644 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
645 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
646 				      hr_dev->caps.num_mtpts, 1);
647 	if (ret) {
648 		dev_err(dev, "Failed to init MTPT context memory, aborting.\n");
649 		goto err_unmap_mtt;
650 	}
651 
652 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
653 				      HEM_TYPE_QPC, hr_dev->caps.qpc_entry_sz,
654 				      hr_dev->caps.num_qps, 1);
655 	if (ret) {
656 		dev_err(dev, "Failed to init QP context memory, aborting.\n");
657 		goto err_unmap_dmpt;
658 	}
659 
660 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
661 				      HEM_TYPE_IRRL,
662 				      hr_dev->caps.irrl_entry_sz *
663 				      hr_dev->caps.max_qp_init_rdma,
664 				      hr_dev->caps.num_qps, 1);
665 	if (ret) {
666 		dev_err(dev, "Failed to init irrl_table memory, aborting.\n");
667 		goto err_unmap_qp;
668 	}
669 
670 	if (hr_dev->caps.trrl_entry_sz) {
671 		ret = hns_roce_init_hem_table(hr_dev,
672 					      &hr_dev->qp_table.trrl_table,
673 					      HEM_TYPE_TRRL,
674 					      hr_dev->caps.trrl_entry_sz *
675 					      hr_dev->caps.max_qp_dest_rdma,
676 					      hr_dev->caps.num_qps, 1);
677 		if (ret) {
678 			dev_err(dev,
679 			       "Failed to init trrl_table memory, aborting.\n");
680 			goto err_unmap_irrl;
681 		}
682 	}
683 
684 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
685 				      HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
686 				      hr_dev->caps.num_cqs, 1);
687 	if (ret) {
688 		dev_err(dev, "Failed to init CQ context memory, aborting.\n");
689 		goto err_unmap_trrl;
690 	}
691 
692 	return 0;
693 
694 err_unmap_trrl:
695 	if (hr_dev->caps.trrl_entry_sz)
696 		hns_roce_cleanup_hem_table(hr_dev,
697 					   &hr_dev->qp_table.trrl_table);
698 
699 err_unmap_irrl:
700 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
701 
702 err_unmap_qp:
703 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
704 
705 err_unmap_dmpt:
706 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
707 
708 err_unmap_mtt:
709 	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
710 		hns_roce_cleanup_hem_table(hr_dev,
711 					   &hr_dev->mr_table.mtt_cqe_table);
712 
713 err_unmap_cqe:
714 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table);
715 
716 	return ret;
717 }
718 
719 /**
720  * hns_roce_setup_hca - setup host channel adapter
721  * @hr_dev: pointer to hns roce device
722  * Return : int
723  */
724 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
725 {
726 	int ret;
727 	struct device *dev = hr_dev->dev;
728 
729 	spin_lock_init(&hr_dev->sm_lock);
730 	spin_lock_init(&hr_dev->bt_cmd_lock);
731 
732 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
733 		INIT_LIST_HEAD(&hr_dev->pgdir_list);
734 		mutex_init(&hr_dev->pgdir_mutex);
735 	}
736 
737 	ret = hns_roce_init_uar_table(hr_dev);
738 	if (ret) {
739 		dev_err(dev, "Failed to initialize uar table. aborting\n");
740 		return ret;
741 	}
742 
743 	ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
744 	if (ret) {
745 		dev_err(dev, "Failed to allocate priv_uar.\n");
746 		goto err_uar_table_free;
747 	}
748 
749 	ret = hns_roce_init_pd_table(hr_dev);
750 	if (ret) {
751 		dev_err(dev, "Failed to init protected domain table.\n");
752 		goto err_uar_alloc_free;
753 	}
754 
755 	ret = hns_roce_init_mr_table(hr_dev);
756 	if (ret) {
757 		dev_err(dev, "Failed to init memory region table.\n");
758 		goto err_pd_table_free;
759 	}
760 
761 	ret = hns_roce_init_cq_table(hr_dev);
762 	if (ret) {
763 		dev_err(dev, "Failed to init completion queue table.\n");
764 		goto err_mr_table_free;
765 	}
766 
767 	ret = hns_roce_init_qp_table(hr_dev);
768 	if (ret) {
769 		dev_err(dev, "Failed to init queue pair table.\n");
770 		goto err_cq_table_free;
771 	}
772 
773 	return 0;
774 
775 err_cq_table_free:
776 	hns_roce_cleanup_cq_table(hr_dev);
777 
778 err_mr_table_free:
779 	hns_roce_cleanup_mr_table(hr_dev);
780 
781 err_pd_table_free:
782 	hns_roce_cleanup_pd_table(hr_dev);
783 
784 err_uar_alloc_free:
785 	hns_roce_uar_free(hr_dev, &hr_dev->priv_uar);
786 
787 err_uar_table_free:
788 	hns_roce_cleanup_uar_table(hr_dev);
789 	return ret;
790 }
791 
792 int hns_roce_init(struct hns_roce_dev *hr_dev)
793 {
794 	int ret;
795 	struct device *dev = hr_dev->dev;
796 
797 	if (hr_dev->hw->reset) {
798 		ret = hr_dev->hw->reset(hr_dev, true);
799 		if (ret) {
800 			dev_err(dev, "Reset RoCE engine failed!\n");
801 			return ret;
802 		}
803 	}
804 	hr_dev->is_reset = false;
805 
806 	if (hr_dev->hw->cmq_init) {
807 		ret = hr_dev->hw->cmq_init(hr_dev);
808 		if (ret) {
809 			dev_err(dev, "Init RoCE Command Queue failed!\n");
810 			goto error_failed_cmq_init;
811 		}
812 	}
813 
814 	ret = hr_dev->hw->hw_profile(hr_dev);
815 	if (ret) {
816 		dev_err(dev, "Get RoCE engine profile failed!\n");
817 		goto error_failed_cmd_init;
818 	}
819 
820 	ret = hns_roce_cmd_init(hr_dev);
821 	if (ret) {
822 		dev_err(dev, "cmd init failed!\n");
823 		goto error_failed_cmd_init;
824 	}
825 
826 	ret = hr_dev->hw->init_eq(hr_dev);
827 	if (ret) {
828 		dev_err(dev, "eq init failed!\n");
829 		goto error_failed_eq_table;
830 	}
831 
832 	if (hr_dev->cmd_mod) {
833 		ret = hns_roce_cmd_use_events(hr_dev);
834 		if (ret) {
835 			dev_err(dev, "Switch to event-driven cmd failed!\n");
836 			goto error_failed_use_event;
837 		}
838 	}
839 
840 	ret = hns_roce_init_hem(hr_dev);
841 	if (ret) {
842 		dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
843 		goto error_failed_init_hem;
844 	}
845 
846 	ret = hns_roce_setup_hca(hr_dev);
847 	if (ret) {
848 		dev_err(dev, "setup hca failed!\n");
849 		goto error_failed_setup_hca;
850 	}
851 
852 	if (hr_dev->hw->hw_init) {
853 		ret = hr_dev->hw->hw_init(hr_dev);
854 		if (ret) {
855 			dev_err(dev, "hw_init failed!\n");
856 			goto error_failed_engine_init;
857 		}
858 	}
859 
860 	ret = hns_roce_register_device(hr_dev);
861 	if (ret)
862 		goto error_failed_register_device;
863 
864 	return 0;
865 
866 error_failed_register_device:
867 	if (hr_dev->hw->hw_exit)
868 		hr_dev->hw->hw_exit(hr_dev);
869 
870 error_failed_engine_init:
871 	hns_roce_cleanup_bitmap(hr_dev);
872 
873 error_failed_setup_hca:
874 	hns_roce_cleanup_hem(hr_dev);
875 
876 error_failed_init_hem:
877 	if (hr_dev->cmd_mod)
878 		hns_roce_cmd_use_polling(hr_dev);
879 
880 error_failed_use_event:
881 	hr_dev->hw->cleanup_eq(hr_dev);
882 
883 error_failed_eq_table:
884 	hns_roce_cmd_cleanup(hr_dev);
885 
886 error_failed_cmd_init:
887 	if (hr_dev->hw->cmq_exit)
888 		hr_dev->hw->cmq_exit(hr_dev);
889 
890 error_failed_cmq_init:
891 	if (hr_dev->hw->reset) {
892 		if (hr_dev->hw->reset(hr_dev, false))
893 			dev_err(dev, "Dereset RoCE engine failed!\n");
894 	}
895 
896 	return ret;
897 }
898 EXPORT_SYMBOL_GPL(hns_roce_init);
899 
900 void hns_roce_exit(struct hns_roce_dev *hr_dev)
901 {
902 	hns_roce_unregister_device(hr_dev);
903 
904 	if (hr_dev->hw->hw_exit)
905 		hr_dev->hw->hw_exit(hr_dev);
906 	hns_roce_cleanup_bitmap(hr_dev);
907 	hns_roce_cleanup_hem(hr_dev);
908 
909 	if (hr_dev->cmd_mod)
910 		hns_roce_cmd_use_polling(hr_dev);
911 
912 	hr_dev->hw->cleanup_eq(hr_dev);
913 	hns_roce_cmd_cleanup(hr_dev);
914 	if (hr_dev->hw->cmq_exit)
915 		hr_dev->hw->cmq_exit(hr_dev);
916 	if (hr_dev->hw->reset)
917 		hr_dev->hw->reset(hr_dev, false);
918 }
919 EXPORT_SYMBOL_GPL(hns_roce_exit);
920 
921 MODULE_LICENSE("Dual BSD/GPL");
922 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
923 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
924 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
925 MODULE_DESCRIPTION("HNS RoCE Driver");
926