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 	int ret;
145 	u8 port;
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 	resp.cqe_size = hr_dev->caps.cqe_sz;
327 
328 	ret = ib_copy_to_udata(udata, &resp, sizeof(resp));
329 	if (ret)
330 		goto error_fail_copy_to_udata;
331 
332 	return 0;
333 
334 error_fail_copy_to_udata:
335 	hns_roce_uar_free(hr_dev, &context->uar);
336 
337 error_fail_uar_alloc:
338 	return ret;
339 }
340 
341 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
342 {
343 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
344 
345 	hns_roce_uar_free(to_hr_dev(ibcontext->device), &context->uar);
346 }
347 
348 static int hns_roce_mmap(struct ib_ucontext *context,
349 			 struct vm_area_struct *vma)
350 {
351 	struct hns_roce_dev *hr_dev = to_hr_dev(context->device);
352 
353 	switch (vma->vm_pgoff) {
354 	case 0:
355 		return rdma_user_mmap_io(context, vma,
356 					 to_hr_ucontext(context)->uar.pfn,
357 					 PAGE_SIZE,
358 					 pgprot_noncached(vma->vm_page_prot),
359 					 NULL);
360 
361 	/* vm_pgoff: 1 -- TPTR */
362 	case 1:
363 		if (!hr_dev->tptr_dma_addr || !hr_dev->tptr_size)
364 			return -EINVAL;
365 		/*
366 		 * FIXME: using io_remap_pfn_range on the dma address returned
367 		 * by dma_alloc_coherent is totally wrong.
368 		 */
369 		return rdma_user_mmap_io(context, vma,
370 					 hr_dev->tptr_dma_addr >> PAGE_SHIFT,
371 					 hr_dev->tptr_size,
372 					 vma->vm_page_prot,
373 					 NULL);
374 
375 	default:
376 		return -EINVAL;
377 	}
378 }
379 
380 static int hns_roce_port_immutable(struct ib_device *ib_dev, u8 port_num,
381 				   struct ib_port_immutable *immutable)
382 {
383 	struct ib_port_attr attr;
384 	int ret;
385 
386 	ret = ib_query_port(ib_dev, port_num, &attr);
387 	if (ret)
388 		return ret;
389 
390 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
391 	immutable->gid_tbl_len = attr.gid_tbl_len;
392 
393 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
394 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
395 	if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
396 		immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
397 
398 	return 0;
399 }
400 
401 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
402 {
403 }
404 
405 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
406 {
407 	struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
408 
409 	hr_dev->active = false;
410 	unregister_netdevice_notifier(&iboe->nb);
411 	ib_unregister_device(&hr_dev->ib_dev);
412 }
413 
414 static const struct ib_device_ops hns_roce_dev_ops = {
415 	.owner = THIS_MODULE,
416 	.driver_id = RDMA_DRIVER_HNS,
417 	.uverbs_abi_ver = 1,
418 	.uverbs_no_driver_id_binding = 1,
419 
420 	.add_gid = hns_roce_add_gid,
421 	.alloc_pd = hns_roce_alloc_pd,
422 	.alloc_ucontext = hns_roce_alloc_ucontext,
423 	.create_ah = hns_roce_create_ah,
424 	.create_cq = hns_roce_create_cq,
425 	.create_qp = hns_roce_create_qp,
426 	.dealloc_pd = hns_roce_dealloc_pd,
427 	.dealloc_ucontext = hns_roce_dealloc_ucontext,
428 	.del_gid = hns_roce_del_gid,
429 	.dereg_mr = hns_roce_dereg_mr,
430 	.destroy_ah = hns_roce_destroy_ah,
431 	.destroy_cq = hns_roce_destroy_cq,
432 	.disassociate_ucontext = hns_roce_disassociate_ucontext,
433 	.fill_res_cq_entry = hns_roce_fill_res_cq_entry,
434 	.get_dma_mr = hns_roce_get_dma_mr,
435 	.get_link_layer = hns_roce_get_link_layer,
436 	.get_port_immutable = hns_roce_port_immutable,
437 	.mmap = hns_roce_mmap,
438 	.modify_device = hns_roce_modify_device,
439 	.modify_qp = hns_roce_modify_qp,
440 	.query_ah = hns_roce_query_ah,
441 	.query_device = hns_roce_query_device,
442 	.query_pkey = hns_roce_query_pkey,
443 	.query_port = hns_roce_query_port,
444 	.reg_user_mr = hns_roce_reg_user_mr,
445 
446 	INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
447 	INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
448 	INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
449 	INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
450 };
451 
452 static const struct ib_device_ops hns_roce_dev_mr_ops = {
453 	.rereg_user_mr = hns_roce_rereg_user_mr,
454 };
455 
456 static const struct ib_device_ops hns_roce_dev_mw_ops = {
457 	.alloc_mw = hns_roce_alloc_mw,
458 	.dealloc_mw = hns_roce_dealloc_mw,
459 
460 	INIT_RDMA_OBJ_SIZE(ib_mw, hns_roce_mw, ibmw),
461 };
462 
463 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
464 	.alloc_mr = hns_roce_alloc_mr,
465 	.map_mr_sg = hns_roce_map_mr_sg,
466 };
467 
468 static const struct ib_device_ops hns_roce_dev_srq_ops = {
469 	.create_srq = hns_roce_create_srq,
470 	.destroy_srq = hns_roce_destroy_srq,
471 
472 	INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
473 };
474 
475 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
476 {
477 	int ret;
478 	struct hns_roce_ib_iboe *iboe = NULL;
479 	struct ib_device *ib_dev = NULL;
480 	struct device *dev = hr_dev->dev;
481 	unsigned int i;
482 
483 	iboe = &hr_dev->iboe;
484 	spin_lock_init(&iboe->lock);
485 
486 	ib_dev = &hr_dev->ib_dev;
487 
488 	ib_dev->node_type = RDMA_NODE_IB_CA;
489 	ib_dev->dev.parent = dev;
490 
491 	ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
492 	ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
493 	ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
494 	ib_dev->uverbs_cmd_mask =
495 		(1ULL << IB_USER_VERBS_CMD_GET_CONTEXT) |
496 		(1ULL << IB_USER_VERBS_CMD_QUERY_DEVICE) |
497 		(1ULL << IB_USER_VERBS_CMD_QUERY_PORT) |
498 		(1ULL << IB_USER_VERBS_CMD_ALLOC_PD) |
499 		(1ULL << IB_USER_VERBS_CMD_DEALLOC_PD) |
500 		(1ULL << IB_USER_VERBS_CMD_REG_MR) |
501 		(1ULL << IB_USER_VERBS_CMD_DEREG_MR) |
502 		(1ULL << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
503 		(1ULL << IB_USER_VERBS_CMD_CREATE_CQ) |
504 		(1ULL << IB_USER_VERBS_CMD_DESTROY_CQ) |
505 		(1ULL << IB_USER_VERBS_CMD_CREATE_QP) |
506 		(1ULL << IB_USER_VERBS_CMD_MODIFY_QP) |
507 		(1ULL << IB_USER_VERBS_CMD_QUERY_QP) |
508 		(1ULL << IB_USER_VERBS_CMD_DESTROY_QP);
509 
510 	ib_dev->uverbs_ex_cmd_mask |= (1ULL << IB_USER_VERBS_EX_CMD_MODIFY_CQ);
511 
512 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR) {
513 		ib_dev->uverbs_cmd_mask |= (1ULL << IB_USER_VERBS_CMD_REREG_MR);
514 		ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
515 	}
516 
517 	/* MW */
518 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_MW) {
519 		ib_dev->uverbs_cmd_mask |=
520 					(1ULL << IB_USER_VERBS_CMD_ALLOC_MW) |
521 					(1ULL << IB_USER_VERBS_CMD_DEALLOC_MW);
522 		ib_set_device_ops(ib_dev, &hns_roce_dev_mw_ops);
523 	}
524 
525 	/* FRMR */
526 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
527 		ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
528 
529 	/* SRQ */
530 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
531 		ib_dev->uverbs_cmd_mask |=
532 				(1ULL << IB_USER_VERBS_CMD_CREATE_SRQ) |
533 				(1ULL << IB_USER_VERBS_CMD_MODIFY_SRQ) |
534 				(1ULL << IB_USER_VERBS_CMD_QUERY_SRQ) |
535 				(1ULL << IB_USER_VERBS_CMD_DESTROY_SRQ) |
536 				(1ULL << IB_USER_VERBS_CMD_POST_SRQ_RECV);
537 		ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
538 		ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
539 	}
540 
541 	ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
542 	ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
543 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
544 		if (!hr_dev->iboe.netdevs[i])
545 			continue;
546 
547 		ret = ib_device_set_netdev(ib_dev, hr_dev->iboe.netdevs[i],
548 					   i + 1);
549 		if (ret)
550 			return ret;
551 	}
552 	dma_set_max_seg_size(dev, UINT_MAX);
553 	ret = ib_register_device(ib_dev, "hns_%d", dev);
554 	if (ret) {
555 		dev_err(dev, "ib_register_device failed!\n");
556 		return ret;
557 	}
558 
559 	ret = hns_roce_setup_mtu_mac(hr_dev);
560 	if (ret) {
561 		dev_err(dev, "setup_mtu_mac failed!\n");
562 		goto error_failed_setup_mtu_mac;
563 	}
564 
565 	iboe->nb.notifier_call = hns_roce_netdev_event;
566 	ret = register_netdevice_notifier(&iboe->nb);
567 	if (ret) {
568 		dev_err(dev, "register_netdevice_notifier failed!\n");
569 		goto error_failed_setup_mtu_mac;
570 	}
571 
572 	hr_dev->active = true;
573 	return 0;
574 
575 error_failed_setup_mtu_mac:
576 	ib_unregister_device(ib_dev);
577 
578 	return ret;
579 }
580 
581 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
582 {
583 	int ret;
584 	struct device *dev = hr_dev->dev;
585 
586 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
587 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
588 				      hr_dev->caps.num_mtpts, 1);
589 	if (ret) {
590 		dev_err(dev, "Failed to init MTPT context memory, aborting.\n");
591 		return ret;
592 	}
593 
594 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
595 				      HEM_TYPE_QPC, hr_dev->caps.qpc_sz,
596 				      hr_dev->caps.num_qps, 1);
597 	if (ret) {
598 		dev_err(dev, "Failed to init QP context memory, aborting.\n");
599 		goto err_unmap_dmpt;
600 	}
601 
602 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
603 				      HEM_TYPE_IRRL,
604 				      hr_dev->caps.irrl_entry_sz *
605 				      hr_dev->caps.max_qp_init_rdma,
606 				      hr_dev->caps.num_qps, 1);
607 	if (ret) {
608 		dev_err(dev, "Failed to init irrl_table memory, aborting.\n");
609 		goto err_unmap_qp;
610 	}
611 
612 	if (hr_dev->caps.trrl_entry_sz) {
613 		ret = hns_roce_init_hem_table(hr_dev,
614 					      &hr_dev->qp_table.trrl_table,
615 					      HEM_TYPE_TRRL,
616 					      hr_dev->caps.trrl_entry_sz *
617 					      hr_dev->caps.max_qp_dest_rdma,
618 					      hr_dev->caps.num_qps, 1);
619 		if (ret) {
620 			dev_err(dev,
621 				"Failed to init trrl_table memory, aborting.\n");
622 			goto err_unmap_irrl;
623 		}
624 	}
625 
626 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
627 				      HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
628 				      hr_dev->caps.num_cqs, 1);
629 	if (ret) {
630 		dev_err(dev, "Failed to init CQ context memory, aborting.\n");
631 		goto err_unmap_trrl;
632 	}
633 
634 	if (hr_dev->caps.srqc_entry_sz) {
635 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
636 					      HEM_TYPE_SRQC,
637 					      hr_dev->caps.srqc_entry_sz,
638 					      hr_dev->caps.num_srqs, 1);
639 		if (ret) {
640 			dev_err(dev,
641 				"Failed to init SRQ context memory, aborting.\n");
642 			goto err_unmap_cq;
643 		}
644 	}
645 
646 	if (hr_dev->caps.sccc_sz) {
647 		ret = hns_roce_init_hem_table(hr_dev,
648 					      &hr_dev->qp_table.sccc_table,
649 					      HEM_TYPE_SCCC,
650 					      hr_dev->caps.sccc_sz,
651 					      hr_dev->caps.num_qps, 1);
652 		if (ret) {
653 			dev_err(dev,
654 				"Failed to init SCC context memory, aborting.\n");
655 			goto err_unmap_srq;
656 		}
657 	}
658 
659 	if (hr_dev->caps.qpc_timer_entry_sz) {
660 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
661 					      HEM_TYPE_QPC_TIMER,
662 					      hr_dev->caps.qpc_timer_entry_sz,
663 					      hr_dev->caps.num_qpc_timer, 1);
664 		if (ret) {
665 			dev_err(dev,
666 				"Failed to init QPC timer memory, aborting.\n");
667 			goto err_unmap_ctx;
668 		}
669 	}
670 
671 	if (hr_dev->caps.cqc_timer_entry_sz) {
672 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
673 					      HEM_TYPE_CQC_TIMER,
674 					      hr_dev->caps.cqc_timer_entry_sz,
675 					      hr_dev->caps.num_cqc_timer, 1);
676 		if (ret) {
677 			dev_err(dev,
678 				"Failed to init CQC timer memory, aborting.\n");
679 			goto err_unmap_qpc_timer;
680 		}
681 	}
682 
683 	return 0;
684 
685 err_unmap_qpc_timer:
686 	if (hr_dev->caps.qpc_timer_entry_sz)
687 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
688 
689 err_unmap_ctx:
690 	if (hr_dev->caps.sccc_sz)
691 		hns_roce_cleanup_hem_table(hr_dev,
692 					   &hr_dev->qp_table.sccc_table);
693 err_unmap_srq:
694 	if (hr_dev->caps.srqc_entry_sz)
695 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
696 
697 err_unmap_cq:
698 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
699 
700 err_unmap_trrl:
701 	if (hr_dev->caps.trrl_entry_sz)
702 		hns_roce_cleanup_hem_table(hr_dev,
703 					   &hr_dev->qp_table.trrl_table);
704 
705 err_unmap_irrl:
706 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
707 
708 err_unmap_qp:
709 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
710 
711 err_unmap_dmpt:
712 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
713 
714 	return ret;
715 }
716 
717 /**
718  * hns_roce_setup_hca - setup host channel adapter
719  * @hr_dev: pointer to hns roce device
720  * Return : int
721  */
722 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
723 {
724 	int ret;
725 	struct device *dev = hr_dev->dev;
726 
727 	spin_lock_init(&hr_dev->sm_lock);
728 	spin_lock_init(&hr_dev->bt_cmd_lock);
729 
730 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) {
731 		INIT_LIST_HEAD(&hr_dev->pgdir_list);
732 		mutex_init(&hr_dev->pgdir_mutex);
733 	}
734 
735 	ret = hns_roce_init_uar_table(hr_dev);
736 	if (ret) {
737 		dev_err(dev, "Failed to initialize uar table. aborting\n");
738 		return ret;
739 	}
740 
741 	ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
742 	if (ret) {
743 		dev_err(dev, "Failed to allocate priv_uar.\n");
744 		goto err_uar_table_free;
745 	}
746 
747 	ret = hns_roce_init_pd_table(hr_dev);
748 	if (ret) {
749 		dev_err(dev, "Failed to init protected domain table.\n");
750 		goto err_uar_alloc_free;
751 	}
752 
753 	ret = hns_roce_init_mr_table(hr_dev);
754 	if (ret) {
755 		dev_err(dev, "Failed to init memory region table.\n");
756 		goto err_pd_table_free;
757 	}
758 
759 	ret = hns_roce_init_cq_table(hr_dev);
760 	if (ret) {
761 		dev_err(dev, "Failed to init completion queue table.\n");
762 		goto err_mr_table_free;
763 	}
764 
765 	ret = hns_roce_init_qp_table(hr_dev);
766 	if (ret) {
767 		dev_err(dev, "Failed to init queue pair table.\n");
768 		goto err_cq_table_free;
769 	}
770 
771 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
772 		ret = hns_roce_init_srq_table(hr_dev);
773 		if (ret) {
774 			dev_err(dev,
775 				"Failed to init share receive queue table.\n");
776 			goto err_qp_table_free;
777 		}
778 	}
779 
780 	return 0;
781 
782 err_qp_table_free:
783 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
784 		hns_roce_cleanup_qp_table(hr_dev);
785 
786 err_cq_table_free:
787 	hns_roce_cleanup_cq_table(hr_dev);
788 
789 err_mr_table_free:
790 	hns_roce_cleanup_mr_table(hr_dev);
791 
792 err_pd_table_free:
793 	hns_roce_cleanup_pd_table(hr_dev);
794 
795 err_uar_alloc_free:
796 	hns_roce_uar_free(hr_dev, &hr_dev->priv_uar);
797 
798 err_uar_table_free:
799 	hns_roce_cleanup_uar_table(hr_dev);
800 	return ret;
801 }
802 
803 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
804 {
805 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
806 	unsigned long flags;
807 
808 	spin_lock_irqsave(&hr_cq->lock, flags);
809 	if (cq->comp_handler) {
810 		if (!hr_cq->is_armed) {
811 			hr_cq->is_armed = 1;
812 			list_add_tail(&hr_cq->node, cq_list);
813 		}
814 	}
815 	spin_unlock_irqrestore(&hr_cq->lock, flags);
816 }
817 
818 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
819 {
820 	struct hns_roce_qp *hr_qp;
821 	struct hns_roce_cq *hr_cq;
822 	struct list_head cq_list;
823 	unsigned long flags_qp;
824 	unsigned long flags;
825 
826 	INIT_LIST_HEAD(&cq_list);
827 
828 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
829 	list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
830 		spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
831 		if (hr_qp->sq.tail != hr_qp->sq.head)
832 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
833 		spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
834 
835 		spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
836 		if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
837 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
838 		spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
839 	}
840 
841 	list_for_each_entry(hr_cq, &cq_list, node)
842 		hns_roce_cq_completion(hr_dev, hr_cq->cqn);
843 
844 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
845 }
846 
847 int hns_roce_init(struct hns_roce_dev *hr_dev)
848 {
849 	int ret;
850 	struct device *dev = hr_dev->dev;
851 
852 	if (hr_dev->hw->reset) {
853 		ret = hr_dev->hw->reset(hr_dev, true);
854 		if (ret) {
855 			dev_err(dev, "Reset RoCE engine failed!\n");
856 			return ret;
857 		}
858 	}
859 	hr_dev->is_reset = false;
860 
861 	if (hr_dev->hw->cmq_init) {
862 		ret = hr_dev->hw->cmq_init(hr_dev);
863 		if (ret) {
864 			dev_err(dev, "Init RoCE Command Queue failed!\n");
865 			goto error_failed_cmq_init;
866 		}
867 	}
868 
869 	ret = hr_dev->hw->hw_profile(hr_dev);
870 	if (ret) {
871 		dev_err(dev, "Get RoCE engine profile failed!\n");
872 		goto error_failed_cmd_init;
873 	}
874 
875 	ret = hns_roce_cmd_init(hr_dev);
876 	if (ret) {
877 		dev_err(dev, "cmd init failed!\n");
878 		goto error_failed_cmd_init;
879 	}
880 
881 	/* EQ depends on poll mode, event mode depends on EQ */
882 	ret = hr_dev->hw->init_eq(hr_dev);
883 	if (ret) {
884 		dev_err(dev, "eq init failed!\n");
885 		goto error_failed_eq_table;
886 	}
887 
888 	if (hr_dev->cmd_mod) {
889 		ret = hns_roce_cmd_use_events(hr_dev);
890 		if (ret) {
891 			dev_warn(dev,
892 				 "Cmd event  mode failed, set back to poll!\n");
893 			hns_roce_cmd_use_polling(hr_dev);
894 		}
895 	}
896 
897 	ret = hns_roce_init_hem(hr_dev);
898 	if (ret) {
899 		dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
900 		goto error_failed_init_hem;
901 	}
902 
903 	ret = hns_roce_setup_hca(hr_dev);
904 	if (ret) {
905 		dev_err(dev, "setup hca failed!\n");
906 		goto error_failed_setup_hca;
907 	}
908 
909 	if (hr_dev->hw->hw_init) {
910 		ret = hr_dev->hw->hw_init(hr_dev);
911 		if (ret) {
912 			dev_err(dev, "hw_init failed!\n");
913 			goto error_failed_engine_init;
914 		}
915 	}
916 
917 	INIT_LIST_HEAD(&hr_dev->qp_list);
918 	spin_lock_init(&hr_dev->qp_list_lock);
919 
920 	ret = hns_roce_register_device(hr_dev);
921 	if (ret)
922 		goto error_failed_register_device;
923 
924 	return 0;
925 
926 error_failed_register_device:
927 	if (hr_dev->hw->hw_exit)
928 		hr_dev->hw->hw_exit(hr_dev);
929 
930 error_failed_engine_init:
931 	hns_roce_cleanup_bitmap(hr_dev);
932 
933 error_failed_setup_hca:
934 	hns_roce_cleanup_hem(hr_dev);
935 
936 error_failed_init_hem:
937 	if (hr_dev->cmd_mod)
938 		hns_roce_cmd_use_polling(hr_dev);
939 	hr_dev->hw->cleanup_eq(hr_dev);
940 
941 error_failed_eq_table:
942 	hns_roce_cmd_cleanup(hr_dev);
943 
944 error_failed_cmd_init:
945 	if (hr_dev->hw->cmq_exit)
946 		hr_dev->hw->cmq_exit(hr_dev);
947 
948 error_failed_cmq_init:
949 	if (hr_dev->hw->reset) {
950 		if (hr_dev->hw->reset(hr_dev, false))
951 			dev_err(dev, "Dereset RoCE engine failed!\n");
952 	}
953 
954 	return ret;
955 }
956 
957 void hns_roce_exit(struct hns_roce_dev *hr_dev)
958 {
959 	hns_roce_unregister_device(hr_dev);
960 
961 	if (hr_dev->hw->hw_exit)
962 		hr_dev->hw->hw_exit(hr_dev);
963 	hns_roce_cleanup_bitmap(hr_dev);
964 	hns_roce_cleanup_hem(hr_dev);
965 
966 	if (hr_dev->cmd_mod)
967 		hns_roce_cmd_use_polling(hr_dev);
968 
969 	hr_dev->hw->cleanup_eq(hr_dev);
970 	hns_roce_cmd_cleanup(hr_dev);
971 	if (hr_dev->hw->cmq_exit)
972 		hr_dev->hw->cmq_exit(hr_dev);
973 	if (hr_dev->hw->reset)
974 		hr_dev->hw->reset(hr_dev, false);
975 }
976 
977 MODULE_LICENSE("Dual BSD/GPL");
978 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
979 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
980 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
981 MODULE_DESCRIPTION("HNS RoCE Driver");
982