xref: /linux/drivers/infiniband/hw/hns/hns_roce_main.c (revision 48dea9a700c8728cc31a1dd44588b97578de86ee)
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 	port = port_num - 1;
237 
238 	/* props being zeroed by the caller, avoid zeroing it here */
239 
240 	props->max_mtu = hr_dev->caps.max_mtu;
241 	props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
242 	props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
243 				IB_PORT_VENDOR_CLASS_SUP |
244 				IB_PORT_BOOT_MGMT_SUP;
245 	props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
246 	props->pkey_tbl_len = 1;
247 	props->active_width = IB_WIDTH_4X;
248 	props->active_speed = 1;
249 
250 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
251 
252 	net_dev = hr_dev->iboe.netdevs[port];
253 	if (!net_dev) {
254 		spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
255 		dev_err(dev, "Find netdev %u failed!\n", port);
256 		return -EINVAL;
257 	}
258 
259 	mtu = iboe_get_mtu(net_dev->mtu);
260 	props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
261 	props->state = netif_running(net_dev) && netif_carrier_ok(net_dev) ?
262 			       IB_PORT_ACTIVE :
263 			       IB_PORT_DOWN;
264 	props->phys_state = props->state == IB_PORT_ACTIVE ?
265 				    IB_PORT_PHYS_STATE_LINK_UP :
266 				    IB_PORT_PHYS_STATE_DISABLED;
267 
268 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
269 
270 	return 0;
271 }
272 
273 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
274 						    u8 port_num)
275 {
276 	return IB_LINK_LAYER_ETHERNET;
277 }
278 
279 static int hns_roce_query_pkey(struct ib_device *ib_dev, u8 port, u16 index,
280 			       u16 *pkey)
281 {
282 	*pkey = PKEY_ID;
283 
284 	return 0;
285 }
286 
287 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
288 				  struct ib_device_modify *props)
289 {
290 	unsigned long flags;
291 
292 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
293 		return -EOPNOTSUPP;
294 
295 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
296 		spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
297 		memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
298 		spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
299 	}
300 
301 	return 0;
302 }
303 
304 static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx,
305 				   struct ib_udata *udata)
306 {
307 	int ret;
308 	struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
309 	struct hns_roce_ib_alloc_ucontext_resp resp = {};
310 	struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
311 
312 	if (!hr_dev->active)
313 		return -EAGAIN;
314 
315 	resp.qp_tab_size = hr_dev->caps.num_qps;
316 
317 	ret = hns_roce_uar_alloc(hr_dev, &context->uar);
318 	if (ret)
319 		goto error_fail_uar_alloc;
320 
321 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
322 		INIT_LIST_HEAD(&context->page_list);
323 		mutex_init(&context->page_mutex);
324 	}
325 
326 	ret = ib_copy_to_udata(udata, &resp, sizeof(resp));
327 	if (ret)
328 		goto error_fail_copy_to_udata;
329 
330 	return 0;
331 
332 error_fail_copy_to_udata:
333 	hns_roce_uar_free(hr_dev, &context->uar);
334 
335 error_fail_uar_alloc:
336 	return ret;
337 }
338 
339 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
340 {
341 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
342 
343 	hns_roce_uar_free(to_hr_dev(ibcontext->device), &context->uar);
344 }
345 
346 static int hns_roce_mmap(struct ib_ucontext *context,
347 			 struct vm_area_struct *vma)
348 {
349 	struct hns_roce_dev *hr_dev = to_hr_dev(context->device);
350 
351 	switch (vma->vm_pgoff) {
352 	case 0:
353 		return rdma_user_mmap_io(context, vma,
354 					 to_hr_ucontext(context)->uar.pfn,
355 					 PAGE_SIZE,
356 					 pgprot_noncached(vma->vm_page_prot),
357 					 NULL);
358 
359 	/* vm_pgoff: 1 -- TPTR */
360 	case 1:
361 		if (!hr_dev->tptr_dma_addr || !hr_dev->tptr_size)
362 			return -EINVAL;
363 		/*
364 		 * FIXME: using io_remap_pfn_range on the dma address returned
365 		 * by dma_alloc_coherent is totally wrong.
366 		 */
367 		return rdma_user_mmap_io(context, vma,
368 					 hr_dev->tptr_dma_addr >> PAGE_SHIFT,
369 					 hr_dev->tptr_size,
370 					 vma->vm_page_prot,
371 					 NULL);
372 
373 	default:
374 		return -EINVAL;
375 	}
376 }
377 
378 static int hns_roce_port_immutable(struct ib_device *ib_dev, u8 port_num,
379 				   struct ib_port_immutable *immutable)
380 {
381 	struct ib_port_attr attr;
382 	int ret;
383 
384 	ret = ib_query_port(ib_dev, port_num, &attr);
385 	if (ret)
386 		return ret;
387 
388 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
389 	immutable->gid_tbl_len = attr.gid_tbl_len;
390 
391 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
392 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
393 	if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
394 		immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
395 
396 	return 0;
397 }
398 
399 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
400 {
401 }
402 
403 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
404 {
405 	struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
406 
407 	hr_dev->active = false;
408 	unregister_netdevice_notifier(&iboe->nb);
409 	ib_unregister_device(&hr_dev->ib_dev);
410 }
411 
412 static const struct ib_device_ops hns_roce_dev_ops = {
413 	.owner = THIS_MODULE,
414 	.driver_id = RDMA_DRIVER_HNS,
415 	.uverbs_abi_ver = 1,
416 	.uverbs_no_driver_id_binding = 1,
417 
418 	.add_gid = hns_roce_add_gid,
419 	.alloc_pd = hns_roce_alloc_pd,
420 	.alloc_ucontext = hns_roce_alloc_ucontext,
421 	.create_ah = hns_roce_create_ah,
422 	.create_cq = hns_roce_create_cq,
423 	.create_qp = hns_roce_create_qp,
424 	.dealloc_pd = hns_roce_dealloc_pd,
425 	.dealloc_ucontext = hns_roce_dealloc_ucontext,
426 	.del_gid = hns_roce_del_gid,
427 	.dereg_mr = hns_roce_dereg_mr,
428 	.destroy_ah = hns_roce_destroy_ah,
429 	.destroy_cq = hns_roce_destroy_cq,
430 	.disassociate_ucontext = hns_roce_disassociate_ucontext,
431 	.fill_res_cq_entry = hns_roce_fill_res_cq_entry,
432 	.get_dma_mr = hns_roce_get_dma_mr,
433 	.get_link_layer = hns_roce_get_link_layer,
434 	.get_port_immutable = hns_roce_port_immutable,
435 	.mmap = hns_roce_mmap,
436 	.modify_device = hns_roce_modify_device,
437 	.modify_qp = hns_roce_modify_qp,
438 	.query_ah = hns_roce_query_ah,
439 	.query_device = hns_roce_query_device,
440 	.query_pkey = hns_roce_query_pkey,
441 	.query_port = hns_roce_query_port,
442 	.reg_user_mr = hns_roce_reg_user_mr,
443 
444 	INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
445 	INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
446 	INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
447 	INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
448 };
449 
450 static const struct ib_device_ops hns_roce_dev_mr_ops = {
451 	.rereg_user_mr = hns_roce_rereg_user_mr,
452 };
453 
454 static const struct ib_device_ops hns_roce_dev_mw_ops = {
455 	.alloc_mw = hns_roce_alloc_mw,
456 	.dealloc_mw = hns_roce_dealloc_mw,
457 };
458 
459 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
460 	.alloc_mr = hns_roce_alloc_mr,
461 	.map_mr_sg = hns_roce_map_mr_sg,
462 };
463 
464 static const struct ib_device_ops hns_roce_dev_srq_ops = {
465 	.create_srq = hns_roce_create_srq,
466 	.destroy_srq = hns_roce_destroy_srq,
467 
468 	INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
469 };
470 
471 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
472 {
473 	int ret;
474 	struct hns_roce_ib_iboe *iboe = NULL;
475 	struct ib_device *ib_dev = NULL;
476 	struct device *dev = hr_dev->dev;
477 	unsigned int i;
478 
479 	iboe = &hr_dev->iboe;
480 	spin_lock_init(&iboe->lock);
481 
482 	ib_dev = &hr_dev->ib_dev;
483 
484 	ib_dev->node_type = RDMA_NODE_IB_CA;
485 	ib_dev->dev.parent = dev;
486 
487 	ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
488 	ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
489 	ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
490 	ib_dev->uverbs_cmd_mask =
491 		(1ULL << IB_USER_VERBS_CMD_GET_CONTEXT) |
492 		(1ULL << IB_USER_VERBS_CMD_QUERY_DEVICE) |
493 		(1ULL << IB_USER_VERBS_CMD_QUERY_PORT) |
494 		(1ULL << IB_USER_VERBS_CMD_ALLOC_PD) |
495 		(1ULL << IB_USER_VERBS_CMD_DEALLOC_PD) |
496 		(1ULL << IB_USER_VERBS_CMD_REG_MR) |
497 		(1ULL << IB_USER_VERBS_CMD_DEREG_MR) |
498 		(1ULL << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
499 		(1ULL << IB_USER_VERBS_CMD_CREATE_CQ) |
500 		(1ULL << IB_USER_VERBS_CMD_DESTROY_CQ) |
501 		(1ULL << IB_USER_VERBS_CMD_CREATE_QP) |
502 		(1ULL << IB_USER_VERBS_CMD_MODIFY_QP) |
503 		(1ULL << IB_USER_VERBS_CMD_QUERY_QP) |
504 		(1ULL << IB_USER_VERBS_CMD_DESTROY_QP);
505 
506 	ib_dev->uverbs_ex_cmd_mask |= (1ULL << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
507 
508 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR) {
509 		ib_dev->uverbs_cmd_mask |= (1ULL << IB_USER_VERBS_CMD_REREG_MR);
510 		ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
511 	}
512 
513 	/* MW */
514 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_MW) {
515 		ib_dev->uverbs_cmd_mask |=
516 					(1ULL << IB_USER_VERBS_CMD_ALLOC_MW) |
517 					(1ULL << IB_USER_VERBS_CMD_DEALLOC_MW);
518 		ib_set_device_ops(ib_dev, &hns_roce_dev_mw_ops);
519 	}
520 
521 	/* FRMR */
522 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
523 		ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
524 
525 	/* SRQ */
526 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
527 		ib_dev->uverbs_cmd_mask |=
528 				(1ULL << IB_USER_VERBS_CMD_CREATE_SRQ) |
529 				(1ULL << IB_USER_VERBS_CMD_MODIFY_SRQ) |
530 				(1ULL << IB_USER_VERBS_CMD_QUERY_SRQ) |
531 				(1ULL << IB_USER_VERBS_CMD_DESTROY_SRQ) |
532 				(1ULL << IB_USER_VERBS_CMD_POST_SRQ_RECV);
533 		ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
534 		ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
535 	}
536 
537 	ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
538 	ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
539 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
540 		if (!hr_dev->iboe.netdevs[i])
541 			continue;
542 
543 		ret = ib_device_set_netdev(ib_dev, hr_dev->iboe.netdevs[i],
544 					   i + 1);
545 		if (ret)
546 			return ret;
547 	}
548 	ret = ib_register_device(ib_dev, "hns_%d");
549 	if (ret) {
550 		dev_err(dev, "ib_register_device failed!\n");
551 		return ret;
552 	}
553 
554 	ret = hns_roce_setup_mtu_mac(hr_dev);
555 	if (ret) {
556 		dev_err(dev, "setup_mtu_mac failed!\n");
557 		goto error_failed_setup_mtu_mac;
558 	}
559 
560 	iboe->nb.notifier_call = hns_roce_netdev_event;
561 	ret = register_netdevice_notifier(&iboe->nb);
562 	if (ret) {
563 		dev_err(dev, "register_netdevice_notifier failed!\n");
564 		goto error_failed_setup_mtu_mac;
565 	}
566 
567 	hr_dev->active = true;
568 	return 0;
569 
570 error_failed_setup_mtu_mac:
571 	ib_unregister_device(ib_dev);
572 
573 	return ret;
574 }
575 
576 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
577 {
578 	int ret;
579 	struct device *dev = hr_dev->dev;
580 
581 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
582 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
583 				      hr_dev->caps.num_mtpts, 1);
584 	if (ret) {
585 		dev_err(dev, "Failed to init MTPT context memory, aborting.\n");
586 		return ret;
587 	}
588 
589 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
590 				      HEM_TYPE_QPC, hr_dev->caps.qpc_entry_sz,
591 				      hr_dev->caps.num_qps, 1);
592 	if (ret) {
593 		dev_err(dev, "Failed to init QP context memory, aborting.\n");
594 		goto err_unmap_dmpt;
595 	}
596 
597 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
598 				      HEM_TYPE_IRRL,
599 				      hr_dev->caps.irrl_entry_sz *
600 				      hr_dev->caps.max_qp_init_rdma,
601 				      hr_dev->caps.num_qps, 1);
602 	if (ret) {
603 		dev_err(dev, "Failed to init irrl_table memory, aborting.\n");
604 		goto err_unmap_qp;
605 	}
606 
607 	if (hr_dev->caps.trrl_entry_sz) {
608 		ret = hns_roce_init_hem_table(hr_dev,
609 					      &hr_dev->qp_table.trrl_table,
610 					      HEM_TYPE_TRRL,
611 					      hr_dev->caps.trrl_entry_sz *
612 					      hr_dev->caps.max_qp_dest_rdma,
613 					      hr_dev->caps.num_qps, 1);
614 		if (ret) {
615 			dev_err(dev,
616 				"Failed to init trrl_table memory, aborting.\n");
617 			goto err_unmap_irrl;
618 		}
619 	}
620 
621 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
622 				      HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
623 				      hr_dev->caps.num_cqs, 1);
624 	if (ret) {
625 		dev_err(dev, "Failed to init CQ context memory, aborting.\n");
626 		goto err_unmap_trrl;
627 	}
628 
629 	if (hr_dev->caps.srqc_entry_sz) {
630 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
631 					      HEM_TYPE_SRQC,
632 					      hr_dev->caps.srqc_entry_sz,
633 					      hr_dev->caps.num_srqs, 1);
634 		if (ret) {
635 			dev_err(dev,
636 				"Failed to init SRQ context memory, aborting.\n");
637 			goto err_unmap_cq;
638 		}
639 	}
640 
641 	if (hr_dev->caps.sccc_entry_sz) {
642 		ret = hns_roce_init_hem_table(hr_dev,
643 					      &hr_dev->qp_table.sccc_table,
644 					      HEM_TYPE_SCCC,
645 					      hr_dev->caps.sccc_entry_sz,
646 					      hr_dev->caps.num_qps, 1);
647 		if (ret) {
648 			dev_err(dev,
649 				"Failed to init SCC context memory, aborting.\n");
650 			goto err_unmap_srq;
651 		}
652 	}
653 
654 	if (hr_dev->caps.qpc_timer_entry_sz) {
655 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
656 					      HEM_TYPE_QPC_TIMER,
657 					      hr_dev->caps.qpc_timer_entry_sz,
658 					      hr_dev->caps.num_qpc_timer, 1);
659 		if (ret) {
660 			dev_err(dev,
661 				"Failed to init QPC timer memory, aborting.\n");
662 			goto err_unmap_ctx;
663 		}
664 	}
665 
666 	if (hr_dev->caps.cqc_timer_entry_sz) {
667 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
668 					      HEM_TYPE_CQC_TIMER,
669 					      hr_dev->caps.cqc_timer_entry_sz,
670 					      hr_dev->caps.num_cqc_timer, 1);
671 		if (ret) {
672 			dev_err(dev,
673 				"Failed to init CQC timer memory, aborting.\n");
674 			goto err_unmap_qpc_timer;
675 		}
676 	}
677 
678 	return 0;
679 
680 err_unmap_qpc_timer:
681 	if (hr_dev->caps.qpc_timer_entry_sz)
682 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
683 
684 err_unmap_ctx:
685 	if (hr_dev->caps.sccc_entry_sz)
686 		hns_roce_cleanup_hem_table(hr_dev,
687 					   &hr_dev->qp_table.sccc_table);
688 err_unmap_srq:
689 	if (hr_dev->caps.srqc_entry_sz)
690 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
691 
692 err_unmap_cq:
693 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
694 
695 err_unmap_trrl:
696 	if (hr_dev->caps.trrl_entry_sz)
697 		hns_roce_cleanup_hem_table(hr_dev,
698 					   &hr_dev->qp_table.trrl_table);
699 
700 err_unmap_irrl:
701 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
702 
703 err_unmap_qp:
704 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
705 
706 err_unmap_dmpt:
707 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
708 
709 	return ret;
710 }
711 
712 /**
713  * hns_roce_setup_hca - setup host channel adapter
714  * @hr_dev: pointer to hns roce device
715  * Return : int
716  */
717 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
718 {
719 	int ret;
720 	struct device *dev = hr_dev->dev;
721 
722 	spin_lock_init(&hr_dev->sm_lock);
723 	spin_lock_init(&hr_dev->bt_cmd_lock);
724 
725 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
726 		INIT_LIST_HEAD(&hr_dev->pgdir_list);
727 		mutex_init(&hr_dev->pgdir_mutex);
728 	}
729 
730 	ret = hns_roce_init_uar_table(hr_dev);
731 	if (ret) {
732 		dev_err(dev, "Failed to initialize uar table. aborting\n");
733 		return ret;
734 	}
735 
736 	ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
737 	if (ret) {
738 		dev_err(dev, "Failed to allocate priv_uar.\n");
739 		goto err_uar_table_free;
740 	}
741 
742 	ret = hns_roce_init_pd_table(hr_dev);
743 	if (ret) {
744 		dev_err(dev, "Failed to init protected domain table.\n");
745 		goto err_uar_alloc_free;
746 	}
747 
748 	ret = hns_roce_init_mr_table(hr_dev);
749 	if (ret) {
750 		dev_err(dev, "Failed to init memory region table.\n");
751 		goto err_pd_table_free;
752 	}
753 
754 	ret = hns_roce_init_cq_table(hr_dev);
755 	if (ret) {
756 		dev_err(dev, "Failed to init completion queue table.\n");
757 		goto err_mr_table_free;
758 	}
759 
760 	ret = hns_roce_init_qp_table(hr_dev);
761 	if (ret) {
762 		dev_err(dev, "Failed to init queue pair table.\n");
763 		goto err_cq_table_free;
764 	}
765 
766 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
767 		ret = hns_roce_init_srq_table(hr_dev);
768 		if (ret) {
769 			dev_err(dev,
770 				"Failed to init share receive queue table.\n");
771 			goto err_qp_table_free;
772 		}
773 	}
774 
775 	return 0;
776 
777 err_qp_table_free:
778 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
779 		hns_roce_cleanup_qp_table(hr_dev);
780 
781 err_cq_table_free:
782 	hns_roce_cleanup_cq_table(hr_dev);
783 
784 err_mr_table_free:
785 	hns_roce_cleanup_mr_table(hr_dev);
786 
787 err_pd_table_free:
788 	hns_roce_cleanup_pd_table(hr_dev);
789 
790 err_uar_alloc_free:
791 	hns_roce_uar_free(hr_dev, &hr_dev->priv_uar);
792 
793 err_uar_table_free:
794 	hns_roce_cleanup_uar_table(hr_dev);
795 	return ret;
796 }
797 
798 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
799 {
800 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
801 	unsigned long flags;
802 
803 	spin_lock_irqsave(&hr_cq->lock, flags);
804 	if (cq->comp_handler) {
805 		if (!hr_cq->is_armed) {
806 			hr_cq->is_armed = 1;
807 			list_add_tail(&hr_cq->node, cq_list);
808 		}
809 	}
810 	spin_unlock_irqrestore(&hr_cq->lock, flags);
811 }
812 
813 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
814 {
815 	struct hns_roce_qp *hr_qp;
816 	struct hns_roce_cq *hr_cq;
817 	struct list_head cq_list;
818 	unsigned long flags_qp;
819 	unsigned long flags;
820 
821 	INIT_LIST_HEAD(&cq_list);
822 
823 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
824 	list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
825 		spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
826 		if (hr_qp->sq.tail != hr_qp->sq.head)
827 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
828 		spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
829 
830 		spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
831 		if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
832 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
833 		spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
834 	}
835 
836 	list_for_each_entry(hr_cq, &cq_list, node)
837 		hns_roce_cq_completion(hr_dev, hr_cq->cqn);
838 
839 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
840 }
841 
842 int hns_roce_init(struct hns_roce_dev *hr_dev)
843 {
844 	int ret;
845 	struct device *dev = hr_dev->dev;
846 
847 	if (hr_dev->hw->reset) {
848 		ret = hr_dev->hw->reset(hr_dev, true);
849 		if (ret) {
850 			dev_err(dev, "Reset RoCE engine failed!\n");
851 			return ret;
852 		}
853 	}
854 	hr_dev->is_reset = false;
855 
856 	if (hr_dev->hw->cmq_init) {
857 		ret = hr_dev->hw->cmq_init(hr_dev);
858 		if (ret) {
859 			dev_err(dev, "Init RoCE Command Queue failed!\n");
860 			goto error_failed_cmq_init;
861 		}
862 	}
863 
864 	ret = hr_dev->hw->hw_profile(hr_dev);
865 	if (ret) {
866 		dev_err(dev, "Get RoCE engine profile failed!\n");
867 		goto error_failed_cmd_init;
868 	}
869 
870 	ret = hns_roce_cmd_init(hr_dev);
871 	if (ret) {
872 		dev_err(dev, "cmd init failed!\n");
873 		goto error_failed_cmd_init;
874 	}
875 
876 	/* EQ depends on poll mode, event mode depends on EQ */
877 	ret = hr_dev->hw->init_eq(hr_dev);
878 	if (ret) {
879 		dev_err(dev, "eq init failed!\n");
880 		goto error_failed_eq_table;
881 	}
882 
883 	if (hr_dev->cmd_mod) {
884 		ret = hns_roce_cmd_use_events(hr_dev);
885 		if (ret) {
886 			dev_warn(dev,
887 				 "Cmd event  mode failed, set back to poll!\n");
888 			hns_roce_cmd_use_polling(hr_dev);
889 		}
890 	}
891 
892 	ret = hns_roce_init_hem(hr_dev);
893 	if (ret) {
894 		dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
895 		goto error_failed_init_hem;
896 	}
897 
898 	ret = hns_roce_setup_hca(hr_dev);
899 	if (ret) {
900 		dev_err(dev, "setup hca failed!\n");
901 		goto error_failed_setup_hca;
902 	}
903 
904 	if (hr_dev->hw->hw_init) {
905 		ret = hr_dev->hw->hw_init(hr_dev);
906 		if (ret) {
907 			dev_err(dev, "hw_init failed!\n");
908 			goto error_failed_engine_init;
909 		}
910 	}
911 
912 	INIT_LIST_HEAD(&hr_dev->qp_list);
913 	spin_lock_init(&hr_dev->qp_list_lock);
914 
915 	ret = hns_roce_register_device(hr_dev);
916 	if (ret)
917 		goto error_failed_register_device;
918 
919 	return 0;
920 
921 error_failed_register_device:
922 	if (hr_dev->hw->hw_exit)
923 		hr_dev->hw->hw_exit(hr_dev);
924 
925 error_failed_engine_init:
926 	hns_roce_cleanup_bitmap(hr_dev);
927 
928 error_failed_setup_hca:
929 	hns_roce_cleanup_hem(hr_dev);
930 
931 error_failed_init_hem:
932 	if (hr_dev->cmd_mod)
933 		hns_roce_cmd_use_polling(hr_dev);
934 	hr_dev->hw->cleanup_eq(hr_dev);
935 
936 error_failed_eq_table:
937 	hns_roce_cmd_cleanup(hr_dev);
938 
939 error_failed_cmd_init:
940 	if (hr_dev->hw->cmq_exit)
941 		hr_dev->hw->cmq_exit(hr_dev);
942 
943 error_failed_cmq_init:
944 	if (hr_dev->hw->reset) {
945 		if (hr_dev->hw->reset(hr_dev, false))
946 			dev_err(dev, "Dereset RoCE engine failed!\n");
947 	}
948 
949 	return ret;
950 }
951 
952 void hns_roce_exit(struct hns_roce_dev *hr_dev)
953 {
954 	hns_roce_unregister_device(hr_dev);
955 
956 	if (hr_dev->hw->hw_exit)
957 		hr_dev->hw->hw_exit(hr_dev);
958 	hns_roce_cleanup_bitmap(hr_dev);
959 	hns_roce_cleanup_hem(hr_dev);
960 
961 	if (hr_dev->cmd_mod)
962 		hns_roce_cmd_use_polling(hr_dev);
963 
964 	hr_dev->hw->cleanup_eq(hr_dev);
965 	hns_roce_cmd_cleanup(hr_dev);
966 	if (hr_dev->hw->cmq_exit)
967 		hr_dev->hw->cmq_exit(hr_dev);
968 	if (hr_dev->hw->reset)
969 		hr_dev->hw->reset(hr_dev, false);
970 }
971 
972 MODULE_LICENSE("Dual BSD/GPL");
973 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
974 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
975 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
976 MODULE_DESCRIPTION("HNS RoCE Driver");
977