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 
61 static int hns_roce_set_mac(struct hns_roce_dev *hr_dev, u8 port, u8 *addr)
62 {
63 	u8 phy_port;
64 	u32 i = 0;
65 
66 	if (!memcmp(hr_dev->dev_addr[port], addr, ETH_ALEN))
67 		return 0;
68 
69 	for (i = 0; i < ETH_ALEN; i++)
70 		hr_dev->dev_addr[port][i] = addr[i];
71 
72 	phy_port = hr_dev->iboe.phy_port[port];
73 	return hr_dev->hw->set_mac(hr_dev, phy_port, addr);
74 }
75 
76 static int hns_roce_add_gid(const struct ib_gid_attr *attr, void **context)
77 {
78 	struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
79 	u8 port = attr->port_num - 1;
80 	int ret;
81 
82 	if (port >= hr_dev->caps.num_ports)
83 		return -EINVAL;
84 
85 	ret = hr_dev->hw->set_gid(hr_dev, port, attr->index, &attr->gid, attr);
86 
87 	return ret;
88 }
89 
90 static int hns_roce_del_gid(const struct ib_gid_attr *attr, void **context)
91 {
92 	struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
93 	struct ib_gid_attr zattr = {};
94 	u8 port = attr->port_num - 1;
95 	int ret;
96 
97 	if (port >= hr_dev->caps.num_ports)
98 		return -EINVAL;
99 
100 	ret = hr_dev->hw->set_gid(hr_dev, port, attr->index, &zgid, &zattr);
101 
102 	return ret;
103 }
104 
105 static int handle_en_event(struct hns_roce_dev *hr_dev, u8 port,
106 			   unsigned long event)
107 {
108 	struct device *dev = hr_dev->dev;
109 	struct net_device *netdev;
110 	int ret = 0;
111 
112 	netdev = hr_dev->iboe.netdevs[port];
113 	if (!netdev) {
114 		dev_err(dev, "Can't find netdev on port(%u)!\n", port);
115 		return -ENODEV;
116 	}
117 
118 	switch (event) {
119 	case NETDEV_UP:
120 	case NETDEV_CHANGE:
121 	case NETDEV_REGISTER:
122 	case NETDEV_CHANGEADDR:
123 		ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
124 		break;
125 	case NETDEV_DOWN:
126 		/*
127 		 * In v1 engine, only support all ports closed together.
128 		 */
129 		break;
130 	default:
131 		dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
132 		break;
133 	}
134 
135 	return ret;
136 }
137 
138 static int hns_roce_netdev_event(struct notifier_block *self,
139 				 unsigned long event, void *ptr)
140 {
141 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
142 	struct hns_roce_ib_iboe *iboe = NULL;
143 	struct hns_roce_dev *hr_dev = NULL;
144 	u8 port = 0;
145 	int ret = 0;
146 
147 	hr_dev = container_of(self, struct hns_roce_dev, iboe.nb);
148 	iboe = &hr_dev->iboe;
149 
150 	for (port = 0; port < hr_dev->caps.num_ports; port++) {
151 		if (dev == iboe->netdevs[port]) {
152 			ret = handle_en_event(hr_dev, port, event);
153 			if (ret)
154 				return NOTIFY_DONE;
155 			break;
156 		}
157 	}
158 
159 	return NOTIFY_DONE;
160 }
161 
162 static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
163 {
164 	int ret;
165 	u8 i;
166 
167 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
168 		if (hr_dev->hw->set_mtu)
169 			hr_dev->hw->set_mtu(hr_dev, hr_dev->iboe.phy_port[i],
170 					    hr_dev->caps.max_mtu);
171 		ret = hns_roce_set_mac(hr_dev, i,
172 				       hr_dev->iboe.netdevs[i]->dev_addr);
173 		if (ret)
174 			return ret;
175 	}
176 
177 	return 0;
178 }
179 
180 static int hns_roce_query_device(struct ib_device *ib_dev,
181 				 struct ib_device_attr *props,
182 				 struct ib_udata *uhw)
183 {
184 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
185 
186 	memset(props, 0, sizeof(*props));
187 
188 	props->fw_ver = hr_dev->caps.fw_ver;
189 	props->sys_image_guid = cpu_to_be64(hr_dev->sys_image_guid);
190 	props->max_mr_size = (u64)(~(0ULL));
191 	props->page_size_cap = hr_dev->caps.page_size_cap;
192 	props->vendor_id = hr_dev->vendor_id;
193 	props->vendor_part_id = hr_dev->vendor_part_id;
194 	props->hw_ver = hr_dev->hw_rev;
195 	props->max_qp = hr_dev->caps.num_qps;
196 	props->max_qp_wr = hr_dev->caps.max_wqes;
197 	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
198 				  IB_DEVICE_RC_RNR_NAK_GEN;
199 	props->max_send_sge = hr_dev->caps.max_sq_sg;
200 	props->max_recv_sge = hr_dev->caps.max_rq_sg;
201 	props->max_sge_rd = 1;
202 	props->max_cq = hr_dev->caps.num_cqs;
203 	props->max_cqe = hr_dev->caps.max_cqes;
204 	props->max_mr = hr_dev->caps.num_mtpts;
205 	props->max_pd = hr_dev->caps.num_pds;
206 	props->max_qp_rd_atom = hr_dev->caps.max_qp_dest_rdma;
207 	props->max_qp_init_rd_atom = hr_dev->caps.max_qp_init_rdma;
208 	props->atomic_cap = hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_ATOMIC ?
209 			    IB_ATOMIC_HCA : IB_ATOMIC_NONE;
210 	props->max_pkeys = 1;
211 	props->local_ca_ack_delay = hr_dev->caps.local_ca_ack_delay;
212 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
213 		props->max_srq = hr_dev->caps.num_srqs;
214 		props->max_srq_wr = hr_dev->caps.max_srq_wrs;
215 		props->max_srq_sge = hr_dev->caps.max_srq_sges;
216 	}
217 
218 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR) {
219 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
220 		props->max_fast_reg_page_list_len = HNS_ROCE_FRMR_MAX_PA;
221 	}
222 
223 	return 0;
224 }
225 
226 static int hns_roce_query_port(struct ib_device *ib_dev, u8 port_num,
227 			       struct ib_port_attr *props)
228 {
229 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
230 	struct device *dev = hr_dev->dev;
231 	struct net_device *net_dev;
232 	unsigned long flags;
233 	enum ib_mtu mtu;
234 	u8 port;
235 
236 	assert(port_num > 0);
237 	port = port_num - 1;
238 
239 	/* props being zeroed by the caller, avoid zeroing it here */
240 
241 	props->max_mtu = hr_dev->caps.max_mtu;
242 	props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
243 	props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
244 				IB_PORT_VENDOR_CLASS_SUP |
245 				IB_PORT_BOOT_MGMT_SUP;
246 	props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
247 	props->pkey_tbl_len = 1;
248 	props->active_width = IB_WIDTH_4X;
249 	props->active_speed = 1;
250 
251 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
252 
253 	net_dev = hr_dev->iboe.netdevs[port];
254 	if (!net_dev) {
255 		spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
256 		dev_err(dev, "Find netdev %u failed!\n", port);
257 		return -EINVAL;
258 	}
259 
260 	mtu = iboe_get_mtu(net_dev->mtu);
261 	props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
262 	props->state = netif_running(net_dev) && netif_carrier_ok(net_dev) ?
263 			       IB_PORT_ACTIVE :
264 			       IB_PORT_DOWN;
265 	props->phys_state = props->state == IB_PORT_ACTIVE ?
266 				    IB_PORT_PHYS_STATE_LINK_UP :
267 				    IB_PORT_PHYS_STATE_DISABLED;
268 
269 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
270 
271 	return 0;
272 }
273 
274 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
275 						    u8 port_num)
276 {
277 	return IB_LINK_LAYER_ETHERNET;
278 }
279 
280 static int hns_roce_query_pkey(struct ib_device *ib_dev, u8 port, u16 index,
281 			       u16 *pkey)
282 {
283 	*pkey = PKEY_ID;
284 
285 	return 0;
286 }
287 
288 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
289 				  struct ib_device_modify *props)
290 {
291 	unsigned long flags;
292 
293 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
294 		return -EOPNOTSUPP;
295 
296 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
297 		spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
298 		memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
299 		spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
300 	}
301 
302 	return 0;
303 }
304 
305 static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx,
306 				   struct ib_udata *udata)
307 {
308 	int ret;
309 	struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
310 	struct hns_roce_ib_alloc_ucontext_resp resp = {};
311 	struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
312 
313 	if (!hr_dev->active)
314 		return -EAGAIN;
315 
316 	resp.qp_tab_size = hr_dev->caps.num_qps;
317 
318 	ret = hns_roce_uar_alloc(hr_dev, &context->uar);
319 	if (ret)
320 		goto error_fail_uar_alloc;
321 
322 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
323 		INIT_LIST_HEAD(&context->page_list);
324 		mutex_init(&context->page_mutex);
325 	}
326 
327 	ret = ib_copy_to_udata(udata, &resp, sizeof(resp));
328 	if (ret)
329 		goto error_fail_copy_to_udata;
330 
331 	return 0;
332 
333 error_fail_copy_to_udata:
334 	hns_roce_uar_free(hr_dev, &context->uar);
335 
336 error_fail_uar_alloc:
337 	return ret;
338 }
339 
340 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
341 {
342 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
343 
344 	hns_roce_uar_free(to_hr_dev(ibcontext->device), &context->uar);
345 }
346 
347 static int hns_roce_mmap(struct ib_ucontext *context,
348 			 struct vm_area_struct *vma)
349 {
350 	struct hns_roce_dev *hr_dev = to_hr_dev(context->device);
351 
352 	switch (vma->vm_pgoff) {
353 	case 0:
354 		return rdma_user_mmap_io(context, vma,
355 					 to_hr_ucontext(context)->uar.pfn,
356 					 PAGE_SIZE,
357 					 pgprot_noncached(vma->vm_page_prot),
358 					 NULL);
359 
360 	/* vm_pgoff: 1 -- TPTR */
361 	case 1:
362 		if (!hr_dev->tptr_dma_addr || !hr_dev->tptr_size)
363 			return -EINVAL;
364 		/*
365 		 * FIXME: using io_remap_pfn_range on the dma address returned
366 		 * by dma_alloc_coherent is totally wrong.
367 		 */
368 		return rdma_user_mmap_io(context, vma,
369 					 hr_dev->tptr_dma_addr >> PAGE_SHIFT,
370 					 hr_dev->tptr_size,
371 					 vma->vm_page_prot,
372 					 NULL);
373 
374 	default:
375 		return -EINVAL;
376 	}
377 }
378 
379 static int hns_roce_port_immutable(struct ib_device *ib_dev, u8 port_num,
380 				   struct ib_port_immutable *immutable)
381 {
382 	struct ib_port_attr attr;
383 	int ret;
384 
385 	ret = ib_query_port(ib_dev, port_num, &attr);
386 	if (ret)
387 		return ret;
388 
389 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
390 	immutable->gid_tbl_len = attr.gid_tbl_len;
391 
392 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
393 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
394 	if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
395 		immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
396 
397 	return 0;
398 }
399 
400 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
401 {
402 }
403 
404 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
405 {
406 	struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
407 
408 	hr_dev->active = false;
409 	unregister_netdevice_notifier(&iboe->nb);
410 	ib_unregister_device(&hr_dev->ib_dev);
411 }
412 
413 static const struct ib_device_ops hns_roce_dev_ops = {
414 	.owner = THIS_MODULE,
415 	.driver_id = RDMA_DRIVER_HNS,
416 	.uverbs_abi_ver = 1,
417 	.uverbs_no_driver_id_binding = 1,
418 
419 	.add_gid = hns_roce_add_gid,
420 	.alloc_pd = hns_roce_alloc_pd,
421 	.alloc_ucontext = hns_roce_alloc_ucontext,
422 	.create_ah = hns_roce_create_ah,
423 	.create_cq = hns_roce_create_cq,
424 	.create_qp = hns_roce_create_qp,
425 	.dealloc_pd = hns_roce_dealloc_pd,
426 	.dealloc_ucontext = hns_roce_dealloc_ucontext,
427 	.del_gid = hns_roce_del_gid,
428 	.dereg_mr = hns_roce_dereg_mr,
429 	.destroy_ah = hns_roce_destroy_ah,
430 	.destroy_cq = hns_roce_destroy_cq,
431 	.disassociate_ucontext = hns_roce_disassociate_ucontext,
432 	.fill_res_entry = hns_roce_fill_res_entry,
433 	.get_dma_mr = hns_roce_get_dma_mr,
434 	.get_link_layer = hns_roce_get_link_layer,
435 	.get_port_immutable = hns_roce_port_immutable,
436 	.mmap = hns_roce_mmap,
437 	.modify_device = hns_roce_modify_device,
438 	.modify_qp = hns_roce_modify_qp,
439 	.query_ah = hns_roce_query_ah,
440 	.query_device = hns_roce_query_device,
441 	.query_pkey = hns_roce_query_pkey,
442 	.query_port = hns_roce_query_port,
443 	.reg_user_mr = hns_roce_reg_user_mr,
444 
445 	INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
446 	INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
447 	INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
448 	INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
449 };
450 
451 static const struct ib_device_ops hns_roce_dev_mr_ops = {
452 	.rereg_user_mr = hns_roce_rereg_user_mr,
453 };
454 
455 static const struct ib_device_ops hns_roce_dev_mw_ops = {
456 	.alloc_mw = hns_roce_alloc_mw,
457 	.dealloc_mw = hns_roce_dealloc_mw,
458 };
459 
460 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
461 	.alloc_mr = hns_roce_alloc_mr,
462 	.map_mr_sg = hns_roce_map_mr_sg,
463 };
464 
465 static const struct ib_device_ops hns_roce_dev_srq_ops = {
466 	.create_srq = hns_roce_create_srq,
467 	.destroy_srq = hns_roce_destroy_srq,
468 
469 	INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
470 };
471 
472 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
473 {
474 	int ret;
475 	struct hns_roce_ib_iboe *iboe = NULL;
476 	struct ib_device *ib_dev = NULL;
477 	struct device *dev = hr_dev->dev;
478 	unsigned int i;
479 
480 	iboe = &hr_dev->iboe;
481 	spin_lock_init(&iboe->lock);
482 
483 	ib_dev = &hr_dev->ib_dev;
484 
485 	ib_dev->node_type = RDMA_NODE_IB_CA;
486 	ib_dev->dev.parent = dev;
487 
488 	ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
489 	ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
490 	ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
491 	ib_dev->uverbs_cmd_mask =
492 		(1ULL << IB_USER_VERBS_CMD_GET_CONTEXT) |
493 		(1ULL << IB_USER_VERBS_CMD_QUERY_DEVICE) |
494 		(1ULL << IB_USER_VERBS_CMD_QUERY_PORT) |
495 		(1ULL << IB_USER_VERBS_CMD_ALLOC_PD) |
496 		(1ULL << IB_USER_VERBS_CMD_DEALLOC_PD) |
497 		(1ULL << IB_USER_VERBS_CMD_REG_MR) |
498 		(1ULL << IB_USER_VERBS_CMD_DEREG_MR) |
499 		(1ULL << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
500 		(1ULL << IB_USER_VERBS_CMD_CREATE_CQ) |
501 		(1ULL << IB_USER_VERBS_CMD_DESTROY_CQ) |
502 		(1ULL << IB_USER_VERBS_CMD_CREATE_QP) |
503 		(1ULL << IB_USER_VERBS_CMD_MODIFY_QP) |
504 		(1ULL << IB_USER_VERBS_CMD_QUERY_QP) |
505 		(1ULL << IB_USER_VERBS_CMD_DESTROY_QP);
506 
507 	ib_dev->uverbs_ex_cmd_mask |= (1ULL << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
508 
509 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR) {
510 		ib_dev->uverbs_cmd_mask |= (1ULL << IB_USER_VERBS_CMD_REREG_MR);
511 		ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
512 	}
513 
514 	/* MW */
515 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_MW) {
516 		ib_dev->uverbs_cmd_mask |=
517 					(1ULL << IB_USER_VERBS_CMD_ALLOC_MW) |
518 					(1ULL << IB_USER_VERBS_CMD_DEALLOC_MW);
519 		ib_set_device_ops(ib_dev, &hns_roce_dev_mw_ops);
520 	}
521 
522 	/* FRMR */
523 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
524 		ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
525 
526 	/* SRQ */
527 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
528 		ib_dev->uverbs_cmd_mask |=
529 				(1ULL << IB_USER_VERBS_CMD_CREATE_SRQ) |
530 				(1ULL << IB_USER_VERBS_CMD_MODIFY_SRQ) |
531 				(1ULL << IB_USER_VERBS_CMD_QUERY_SRQ) |
532 				(1ULL << IB_USER_VERBS_CMD_DESTROY_SRQ) |
533 				(1ULL << IB_USER_VERBS_CMD_POST_SRQ_RECV);
534 		ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
535 		ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
536 	}
537 
538 	ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
539 	ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
540 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
541 		if (!hr_dev->iboe.netdevs[i])
542 			continue;
543 
544 		ret = ib_device_set_netdev(ib_dev, hr_dev->iboe.netdevs[i],
545 					   i + 1);
546 		if (ret)
547 			return ret;
548 	}
549 	ret = ib_register_device(ib_dev, "hns_%d");
550 	if (ret) {
551 		dev_err(dev, "ib_register_device failed!\n");
552 		return ret;
553 	}
554 
555 	ret = hns_roce_setup_mtu_mac(hr_dev);
556 	if (ret) {
557 		dev_err(dev, "setup_mtu_mac failed!\n");
558 		goto error_failed_setup_mtu_mac;
559 	}
560 
561 	iboe->nb.notifier_call = hns_roce_netdev_event;
562 	ret = register_netdevice_notifier(&iboe->nb);
563 	if (ret) {
564 		dev_err(dev, "register_netdevice_notifier failed!\n");
565 		goto error_failed_setup_mtu_mac;
566 	}
567 
568 	hr_dev->active = true;
569 	return 0;
570 
571 error_failed_setup_mtu_mac:
572 	ib_unregister_device(ib_dev);
573 
574 	return ret;
575 }
576 
577 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
578 {
579 	int ret;
580 	struct device *dev = hr_dev->dev;
581 
582 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtt_table,
583 				      HEM_TYPE_MTT, hr_dev->caps.mtt_entry_sz,
584 				      hr_dev->caps.num_mtt_segs, 1);
585 	if (ret) {
586 		dev_err(dev, "Failed to init MTT context memory, aborting.\n");
587 		return ret;
588 	}
589 
590 	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE)) {
591 		ret = hns_roce_init_hem_table(hr_dev,
592 					      &hr_dev->mr_table.mtt_cqe_table,
593 					      HEM_TYPE_CQE,
594 					      hr_dev->caps.mtt_entry_sz,
595 					      hr_dev->caps.num_cqe_segs, 1);
596 		if (ret) {
597 			dev_err(dev,
598 				"Failed to init CQE context memory, aborting.\n");
599 			goto err_unmap_cqe;
600 		}
601 	}
602 
603 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
604 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
605 				      hr_dev->caps.num_mtpts, 1);
606 	if (ret) {
607 		dev_err(dev, "Failed to init MTPT context memory, aborting.\n");
608 		goto err_unmap_mtt;
609 	}
610 
611 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
612 				      HEM_TYPE_QPC, hr_dev->caps.qpc_entry_sz,
613 				      hr_dev->caps.num_qps, 1);
614 	if (ret) {
615 		dev_err(dev, "Failed to init QP context memory, aborting.\n");
616 		goto err_unmap_dmpt;
617 	}
618 
619 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
620 				      HEM_TYPE_IRRL,
621 				      hr_dev->caps.irrl_entry_sz *
622 				      hr_dev->caps.max_qp_init_rdma,
623 				      hr_dev->caps.num_qps, 1);
624 	if (ret) {
625 		dev_err(dev, "Failed to init irrl_table memory, aborting.\n");
626 		goto err_unmap_qp;
627 	}
628 
629 	if (hr_dev->caps.trrl_entry_sz) {
630 		ret = hns_roce_init_hem_table(hr_dev,
631 					      &hr_dev->qp_table.trrl_table,
632 					      HEM_TYPE_TRRL,
633 					      hr_dev->caps.trrl_entry_sz *
634 					      hr_dev->caps.max_qp_dest_rdma,
635 					      hr_dev->caps.num_qps, 1);
636 		if (ret) {
637 			dev_err(dev,
638 				"Failed to init trrl_table memory, aborting.\n");
639 			goto err_unmap_irrl;
640 		}
641 	}
642 
643 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
644 				      HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
645 				      hr_dev->caps.num_cqs, 1);
646 	if (ret) {
647 		dev_err(dev, "Failed to init CQ context memory, aborting.\n");
648 		goto err_unmap_trrl;
649 	}
650 
651 	if (hr_dev->caps.srqc_entry_sz) {
652 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
653 					      HEM_TYPE_SRQC,
654 					      hr_dev->caps.srqc_entry_sz,
655 					      hr_dev->caps.num_srqs, 1);
656 		if (ret) {
657 			dev_err(dev,
658 				"Failed to init SRQ context memory, aborting.\n");
659 			goto err_unmap_cq;
660 		}
661 	}
662 
663 	if (hr_dev->caps.num_srqwqe_segs) {
664 		ret = hns_roce_init_hem_table(hr_dev,
665 					     &hr_dev->mr_table.mtt_srqwqe_table,
666 					     HEM_TYPE_SRQWQE,
667 					     hr_dev->caps.mtt_entry_sz,
668 					     hr_dev->caps.num_srqwqe_segs, 1);
669 		if (ret) {
670 			dev_err(dev,
671 				"Failed to init MTT srqwqe memory, aborting.\n");
672 			goto err_unmap_srq;
673 		}
674 	}
675 
676 	if (hr_dev->caps.num_idx_segs) {
677 		ret = hns_roce_init_hem_table(hr_dev,
678 					      &hr_dev->mr_table.mtt_idx_table,
679 					      HEM_TYPE_IDX,
680 					      hr_dev->caps.idx_entry_sz,
681 					      hr_dev->caps.num_idx_segs, 1);
682 		if (ret) {
683 			dev_err(dev,
684 				"Failed to init MTT idx memory, aborting.\n");
685 			goto err_unmap_srqwqe;
686 		}
687 	}
688 
689 	if (hr_dev->caps.sccc_entry_sz) {
690 		ret = hns_roce_init_hem_table(hr_dev,
691 					      &hr_dev->qp_table.sccc_table,
692 					      HEM_TYPE_SCCC,
693 					      hr_dev->caps.sccc_entry_sz,
694 					      hr_dev->caps.num_qps, 1);
695 		if (ret) {
696 			dev_err(dev,
697 				"Failed to init SCC context memory, aborting.\n");
698 			goto err_unmap_idx;
699 		}
700 	}
701 
702 	if (hr_dev->caps.qpc_timer_entry_sz) {
703 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
704 					      HEM_TYPE_QPC_TIMER,
705 					      hr_dev->caps.qpc_timer_entry_sz,
706 					      hr_dev->caps.num_qpc_timer, 1);
707 		if (ret) {
708 			dev_err(dev,
709 				"Failed to init QPC timer memory, aborting.\n");
710 			goto err_unmap_ctx;
711 		}
712 	}
713 
714 	if (hr_dev->caps.cqc_timer_entry_sz) {
715 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
716 					      HEM_TYPE_CQC_TIMER,
717 					      hr_dev->caps.cqc_timer_entry_sz,
718 					      hr_dev->caps.num_cqc_timer, 1);
719 		if (ret) {
720 			dev_err(dev,
721 				"Failed to init CQC timer memory, aborting.\n");
722 			goto err_unmap_qpc_timer;
723 		}
724 	}
725 
726 	return 0;
727 
728 err_unmap_qpc_timer:
729 	if (hr_dev->caps.qpc_timer_entry_sz)
730 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
731 
732 err_unmap_ctx:
733 	if (hr_dev->caps.sccc_entry_sz)
734 		hns_roce_cleanup_hem_table(hr_dev,
735 					   &hr_dev->qp_table.sccc_table);
736 
737 err_unmap_idx:
738 	if (hr_dev->caps.num_idx_segs)
739 		hns_roce_cleanup_hem_table(hr_dev,
740 					   &hr_dev->mr_table.mtt_idx_table);
741 
742 err_unmap_srqwqe:
743 	if (hr_dev->caps.num_srqwqe_segs)
744 		hns_roce_cleanup_hem_table(hr_dev,
745 					   &hr_dev->mr_table.mtt_srqwqe_table);
746 
747 err_unmap_srq:
748 	if (hr_dev->caps.srqc_entry_sz)
749 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
750 
751 err_unmap_cq:
752 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
753 
754 err_unmap_trrl:
755 	if (hr_dev->caps.trrl_entry_sz)
756 		hns_roce_cleanup_hem_table(hr_dev,
757 					   &hr_dev->qp_table.trrl_table);
758 
759 err_unmap_irrl:
760 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
761 
762 err_unmap_qp:
763 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
764 
765 err_unmap_dmpt:
766 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
767 
768 err_unmap_mtt:
769 	if (hns_roce_check_whether_mhop(hr_dev, HEM_TYPE_CQE))
770 		hns_roce_cleanup_hem_table(hr_dev,
771 					   &hr_dev->mr_table.mtt_cqe_table);
772 
773 err_unmap_cqe:
774 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtt_table);
775 
776 	return ret;
777 }
778 
779 /**
780  * hns_roce_setup_hca - setup host channel adapter
781  * @hr_dev: pointer to hns roce device
782  * Return : int
783  */
784 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
785 {
786 	int ret;
787 	struct device *dev = hr_dev->dev;
788 
789 	spin_lock_init(&hr_dev->sm_lock);
790 	spin_lock_init(&hr_dev->bt_cmd_lock);
791 
792 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
793 		INIT_LIST_HEAD(&hr_dev->pgdir_list);
794 		mutex_init(&hr_dev->pgdir_mutex);
795 	}
796 
797 	ret = hns_roce_init_uar_table(hr_dev);
798 	if (ret) {
799 		dev_err(dev, "Failed to initialize uar table. aborting\n");
800 		return ret;
801 	}
802 
803 	ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
804 	if (ret) {
805 		dev_err(dev, "Failed to allocate priv_uar.\n");
806 		goto err_uar_table_free;
807 	}
808 
809 	ret = hns_roce_init_pd_table(hr_dev);
810 	if (ret) {
811 		dev_err(dev, "Failed to init protected domain table.\n");
812 		goto err_uar_alloc_free;
813 	}
814 
815 	ret = hns_roce_init_mr_table(hr_dev);
816 	if (ret) {
817 		dev_err(dev, "Failed to init memory region table.\n");
818 		goto err_pd_table_free;
819 	}
820 
821 	ret = hns_roce_init_cq_table(hr_dev);
822 	if (ret) {
823 		dev_err(dev, "Failed to init completion queue table.\n");
824 		goto err_mr_table_free;
825 	}
826 
827 	ret = hns_roce_init_qp_table(hr_dev);
828 	if (ret) {
829 		dev_err(dev, "Failed to init queue pair table.\n");
830 		goto err_cq_table_free;
831 	}
832 
833 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
834 		ret = hns_roce_init_srq_table(hr_dev);
835 		if (ret) {
836 			dev_err(dev,
837 				"Failed to init share receive queue table.\n");
838 			goto err_qp_table_free;
839 		}
840 	}
841 
842 	return 0;
843 
844 err_qp_table_free:
845 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
846 		hns_roce_cleanup_qp_table(hr_dev);
847 
848 err_cq_table_free:
849 	hns_roce_cleanup_cq_table(hr_dev);
850 
851 err_mr_table_free:
852 	hns_roce_cleanup_mr_table(hr_dev);
853 
854 err_pd_table_free:
855 	hns_roce_cleanup_pd_table(hr_dev);
856 
857 err_uar_alloc_free:
858 	hns_roce_uar_free(hr_dev, &hr_dev->priv_uar);
859 
860 err_uar_table_free:
861 	hns_roce_cleanup_uar_table(hr_dev);
862 	return ret;
863 }
864 
865 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
866 {
867 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
868 	unsigned long flags;
869 
870 	spin_lock_irqsave(&hr_cq->lock, flags);
871 	if (cq->comp_handler) {
872 		if (!hr_cq->is_armed) {
873 			hr_cq->is_armed = 1;
874 			list_add_tail(&hr_cq->node, cq_list);
875 		}
876 	}
877 	spin_unlock_irqrestore(&hr_cq->lock, flags);
878 }
879 
880 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
881 {
882 	struct hns_roce_qp *hr_qp;
883 	struct hns_roce_cq *hr_cq;
884 	struct list_head cq_list;
885 	unsigned long flags_qp;
886 	unsigned long flags;
887 
888 	INIT_LIST_HEAD(&cq_list);
889 
890 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
891 	list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
892 		spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
893 		if (hr_qp->sq.tail != hr_qp->sq.head)
894 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
895 		spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
896 
897 		spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
898 		if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
899 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
900 		spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
901 	}
902 
903 	list_for_each_entry(hr_cq, &cq_list, node)
904 		hns_roce_cq_completion(hr_dev, hr_cq->cqn);
905 
906 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
907 }
908 
909 int hns_roce_init(struct hns_roce_dev *hr_dev)
910 {
911 	int ret;
912 	struct device *dev = hr_dev->dev;
913 
914 	if (hr_dev->hw->reset) {
915 		ret = hr_dev->hw->reset(hr_dev, true);
916 		if (ret) {
917 			dev_err(dev, "Reset RoCE engine failed!\n");
918 			return ret;
919 		}
920 	}
921 	hr_dev->is_reset = false;
922 
923 	if (hr_dev->hw->cmq_init) {
924 		ret = hr_dev->hw->cmq_init(hr_dev);
925 		if (ret) {
926 			dev_err(dev, "Init RoCE Command Queue failed!\n");
927 			goto error_failed_cmq_init;
928 		}
929 	}
930 
931 	ret = hr_dev->hw->hw_profile(hr_dev);
932 	if (ret) {
933 		dev_err(dev, "Get RoCE engine profile failed!\n");
934 		goto error_failed_cmd_init;
935 	}
936 
937 	ret = hns_roce_cmd_init(hr_dev);
938 	if (ret) {
939 		dev_err(dev, "cmd init failed!\n");
940 		goto error_failed_cmd_init;
941 	}
942 
943 	/* EQ depends on poll mode, event mode depends on EQ */
944 	ret = hr_dev->hw->init_eq(hr_dev);
945 	if (ret) {
946 		dev_err(dev, "eq init failed!\n");
947 		goto error_failed_eq_table;
948 	}
949 
950 	if (hr_dev->cmd_mod) {
951 		ret = hns_roce_cmd_use_events(hr_dev);
952 		if (ret) {
953 			dev_warn(dev,
954 				 "Cmd event  mode failed, set back to poll!\n");
955 			hns_roce_cmd_use_polling(hr_dev);
956 		}
957 	}
958 
959 	ret = hns_roce_init_hem(hr_dev);
960 	if (ret) {
961 		dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
962 		goto error_failed_init_hem;
963 	}
964 
965 	ret = hns_roce_setup_hca(hr_dev);
966 	if (ret) {
967 		dev_err(dev, "setup hca failed!\n");
968 		goto error_failed_setup_hca;
969 	}
970 
971 	if (hr_dev->hw->hw_init) {
972 		ret = hr_dev->hw->hw_init(hr_dev);
973 		if (ret) {
974 			dev_err(dev, "hw_init failed!\n");
975 			goto error_failed_engine_init;
976 		}
977 	}
978 
979 	INIT_LIST_HEAD(&hr_dev->qp_list);
980 	spin_lock_init(&hr_dev->qp_list_lock);
981 
982 	ret = hns_roce_register_device(hr_dev);
983 	if (ret)
984 		goto error_failed_register_device;
985 
986 	return 0;
987 
988 error_failed_register_device:
989 	if (hr_dev->hw->hw_exit)
990 		hr_dev->hw->hw_exit(hr_dev);
991 
992 error_failed_engine_init:
993 	hns_roce_cleanup_bitmap(hr_dev);
994 
995 error_failed_setup_hca:
996 	hns_roce_cleanup_hem(hr_dev);
997 
998 error_failed_init_hem:
999 	if (hr_dev->cmd_mod)
1000 		hns_roce_cmd_use_polling(hr_dev);
1001 	hr_dev->hw->cleanup_eq(hr_dev);
1002 
1003 error_failed_eq_table:
1004 	hns_roce_cmd_cleanup(hr_dev);
1005 
1006 error_failed_cmd_init:
1007 	if (hr_dev->hw->cmq_exit)
1008 		hr_dev->hw->cmq_exit(hr_dev);
1009 
1010 error_failed_cmq_init:
1011 	if (hr_dev->hw->reset) {
1012 		if (hr_dev->hw->reset(hr_dev, false))
1013 			dev_err(dev, "Dereset RoCE engine failed!\n");
1014 	}
1015 
1016 	return ret;
1017 }
1018 
1019 void hns_roce_exit(struct hns_roce_dev *hr_dev)
1020 {
1021 	hns_roce_unregister_device(hr_dev);
1022 
1023 	if (hr_dev->hw->hw_exit)
1024 		hr_dev->hw->hw_exit(hr_dev);
1025 	hns_roce_cleanup_bitmap(hr_dev);
1026 	hns_roce_cleanup_hem(hr_dev);
1027 
1028 	if (hr_dev->cmd_mod)
1029 		hns_roce_cmd_use_polling(hr_dev);
1030 
1031 	hr_dev->hw->cleanup_eq(hr_dev);
1032 	hns_roce_cmd_cleanup(hr_dev);
1033 	if (hr_dev->hw->cmq_exit)
1034 		hr_dev->hw->cmq_exit(hr_dev);
1035 	if (hr_dev->hw->reset)
1036 		hr_dev->hw->reset(hr_dev, false);
1037 }
1038 
1039 MODULE_LICENSE("Dual BSD/GPL");
1040 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
1041 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
1042 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
1043 MODULE_DESCRIPTION("HNS RoCE Driver");
1044