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/module.h>
35 #include <linux/pci.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 "hns_roce_hem.h"
43 
44 static int hns_roce_set_mac(struct hns_roce_dev *hr_dev, u32 port,
45 			    const u8 *addr)
46 {
47 	u8 phy_port;
48 	u32 i;
49 
50 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
51 		return 0;
52 
53 	if (!memcmp(hr_dev->dev_addr[port], addr, ETH_ALEN))
54 		return 0;
55 
56 	for (i = 0; i < ETH_ALEN; i++)
57 		hr_dev->dev_addr[port][i] = addr[i];
58 
59 	phy_port = hr_dev->iboe.phy_port[port];
60 	return hr_dev->hw->set_mac(hr_dev, phy_port, addr);
61 }
62 
63 static int hns_roce_add_gid(const struct ib_gid_attr *attr, void **context)
64 {
65 	struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
66 	u32 port = attr->port_num - 1;
67 	int ret;
68 
69 	if (port >= hr_dev->caps.num_ports)
70 		return -EINVAL;
71 
72 	ret = hr_dev->hw->set_gid(hr_dev, attr->index, &attr->gid, attr);
73 
74 	return ret;
75 }
76 
77 static int hns_roce_del_gid(const struct ib_gid_attr *attr, void **context)
78 {
79 	struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
80 	u32 port = attr->port_num - 1;
81 	int ret;
82 
83 	if (port >= hr_dev->caps.num_ports)
84 		return -EINVAL;
85 
86 	ret = hr_dev->hw->set_gid(hr_dev, attr->index, NULL, NULL);
87 
88 	return ret;
89 }
90 
91 static int handle_en_event(struct hns_roce_dev *hr_dev, u32 port,
92 			   unsigned long event)
93 {
94 	struct device *dev = hr_dev->dev;
95 	struct net_device *netdev;
96 	int ret = 0;
97 
98 	netdev = hr_dev->iboe.netdevs[port];
99 	if (!netdev) {
100 		dev_err(dev, "Can't find netdev on port(%u)!\n", port);
101 		return -ENODEV;
102 	}
103 
104 	switch (event) {
105 	case NETDEV_UP:
106 	case NETDEV_CHANGE:
107 	case NETDEV_REGISTER:
108 	case NETDEV_CHANGEADDR:
109 		ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
110 		break;
111 	case NETDEV_DOWN:
112 		/*
113 		 * In v1 engine, only support all ports closed together.
114 		 */
115 		break;
116 	default:
117 		dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
118 		break;
119 	}
120 
121 	return ret;
122 }
123 
124 static int hns_roce_netdev_event(struct notifier_block *self,
125 				 unsigned long event, void *ptr)
126 {
127 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
128 	struct hns_roce_ib_iboe *iboe = NULL;
129 	struct hns_roce_dev *hr_dev = NULL;
130 	int ret;
131 	u32 port;
132 
133 	hr_dev = container_of(self, struct hns_roce_dev, iboe.nb);
134 	iboe = &hr_dev->iboe;
135 
136 	for (port = 0; port < hr_dev->caps.num_ports; port++) {
137 		if (dev == iboe->netdevs[port]) {
138 			ret = handle_en_event(hr_dev, port, event);
139 			if (ret)
140 				return NOTIFY_DONE;
141 			break;
142 		}
143 	}
144 
145 	return NOTIFY_DONE;
146 }
147 
148 static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
149 {
150 	int ret;
151 	u8 i;
152 
153 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
154 		ret = hns_roce_set_mac(hr_dev, i,
155 				       hr_dev->iboe.netdevs[i]->dev_addr);
156 		if (ret)
157 			return ret;
158 	}
159 
160 	return 0;
161 }
162 
163 static int hns_roce_query_device(struct ib_device *ib_dev,
164 				 struct ib_device_attr *props,
165 				 struct ib_udata *uhw)
166 {
167 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
168 
169 	memset(props, 0, sizeof(*props));
170 
171 	props->fw_ver = hr_dev->caps.fw_ver;
172 	props->sys_image_guid = cpu_to_be64(hr_dev->sys_image_guid);
173 	props->max_mr_size = (u64)(~(0ULL));
174 	props->page_size_cap = hr_dev->caps.page_size_cap;
175 	props->vendor_id = hr_dev->vendor_id;
176 	props->vendor_part_id = hr_dev->vendor_part_id;
177 	props->hw_ver = hr_dev->hw_rev;
178 	props->max_qp = hr_dev->caps.num_qps;
179 	props->max_qp_wr = hr_dev->caps.max_wqes;
180 	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
181 				  IB_DEVICE_RC_RNR_NAK_GEN;
182 	props->max_send_sge = hr_dev->caps.max_sq_sg;
183 	props->max_recv_sge = hr_dev->caps.max_rq_sg;
184 	props->max_sge_rd = 1;
185 	props->max_cq = hr_dev->caps.num_cqs;
186 	props->max_cqe = hr_dev->caps.max_cqes;
187 	props->max_mr = hr_dev->caps.num_mtpts;
188 	props->max_pd = hr_dev->caps.num_pds;
189 	props->max_qp_rd_atom = hr_dev->caps.max_qp_dest_rdma;
190 	props->max_qp_init_rd_atom = hr_dev->caps.max_qp_init_rdma;
191 	props->atomic_cap = hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_ATOMIC ?
192 			    IB_ATOMIC_HCA : IB_ATOMIC_NONE;
193 	props->max_pkeys = 1;
194 	props->local_ca_ack_delay = hr_dev->caps.local_ca_ack_delay;
195 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
196 		props->max_srq = hr_dev->caps.num_srqs;
197 		props->max_srq_wr = hr_dev->caps.max_srq_wrs;
198 		props->max_srq_sge = hr_dev->caps.max_srq_sges;
199 	}
200 
201 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR &&
202 	    hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
203 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
204 		props->max_fast_reg_page_list_len = HNS_ROCE_FRMR_MAX_PA;
205 	}
206 
207 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
208 		props->device_cap_flags |= IB_DEVICE_XRC;
209 
210 	return 0;
211 }
212 
213 static int hns_roce_query_port(struct ib_device *ib_dev, u32 port_num,
214 			       struct ib_port_attr *props)
215 {
216 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
217 	struct device *dev = hr_dev->dev;
218 	struct net_device *net_dev;
219 	unsigned long flags;
220 	enum ib_mtu mtu;
221 	u32 port;
222 
223 	port = port_num - 1;
224 
225 	/* props being zeroed by the caller, avoid zeroing it here */
226 
227 	props->max_mtu = hr_dev->caps.max_mtu;
228 	props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
229 	props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
230 				IB_PORT_VENDOR_CLASS_SUP |
231 				IB_PORT_BOOT_MGMT_SUP;
232 	props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
233 	props->pkey_tbl_len = 1;
234 	props->active_width = IB_WIDTH_4X;
235 	props->active_speed = 1;
236 
237 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
238 
239 	net_dev = hr_dev->iboe.netdevs[port];
240 	if (!net_dev) {
241 		spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
242 		dev_err(dev, "Find netdev %u failed!\n", port);
243 		return -EINVAL;
244 	}
245 
246 	mtu = iboe_get_mtu(net_dev->mtu);
247 	props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
248 	props->state = netif_running(net_dev) && netif_carrier_ok(net_dev) ?
249 			       IB_PORT_ACTIVE :
250 			       IB_PORT_DOWN;
251 	props->phys_state = props->state == IB_PORT_ACTIVE ?
252 				    IB_PORT_PHYS_STATE_LINK_UP :
253 				    IB_PORT_PHYS_STATE_DISABLED;
254 
255 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
256 
257 	return 0;
258 }
259 
260 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
261 						    u32 port_num)
262 {
263 	return IB_LINK_LAYER_ETHERNET;
264 }
265 
266 static int hns_roce_query_pkey(struct ib_device *ib_dev, u32 port, u16 index,
267 			       u16 *pkey)
268 {
269 	if (index > 0)
270 		return -EINVAL;
271 
272 	*pkey = PKEY_ID;
273 
274 	return 0;
275 }
276 
277 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
278 				  struct ib_device_modify *props)
279 {
280 	unsigned long flags;
281 
282 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
283 		return -EOPNOTSUPP;
284 
285 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
286 		spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
287 		memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
288 		spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
289 	}
290 
291 	return 0;
292 }
293 
294 struct hns_user_mmap_entry *
295 hns_roce_user_mmap_entry_insert(struct ib_ucontext *ucontext, u64 address,
296 				size_t length,
297 				enum hns_roce_mmap_type mmap_type)
298 {
299 	struct hns_user_mmap_entry *entry;
300 	int ret;
301 
302 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
303 	if (!entry)
304 		return NULL;
305 
306 	entry->address = address;
307 	entry->mmap_type = mmap_type;
308 
309 	switch (mmap_type) {
310 	/* pgoff 0 must be used by DB for compatibility */
311 	case HNS_ROCE_MMAP_TYPE_DB:
312 		ret = rdma_user_mmap_entry_insert_exact(
313 				ucontext, &entry->rdma_entry, length, 0);
314 		break;
315 	case HNS_ROCE_MMAP_TYPE_DWQE:
316 		ret = rdma_user_mmap_entry_insert_range(
317 				ucontext, &entry->rdma_entry, length, 1,
318 				U32_MAX);
319 		break;
320 	default:
321 		ret = -EINVAL;
322 		break;
323 	}
324 
325 	if (ret) {
326 		kfree(entry);
327 		return NULL;
328 	}
329 
330 	return entry;
331 }
332 
333 static void hns_roce_dealloc_uar_entry(struct hns_roce_ucontext *context)
334 {
335 	if (context->db_mmap_entry)
336 		rdma_user_mmap_entry_remove(
337 			&context->db_mmap_entry->rdma_entry);
338 }
339 
340 static int hns_roce_alloc_uar_entry(struct ib_ucontext *uctx)
341 {
342 	struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
343 	u64 address;
344 
345 	address = context->uar.pfn << PAGE_SHIFT;
346 	context->db_mmap_entry = hns_roce_user_mmap_entry_insert(
347 		uctx, address, PAGE_SIZE, HNS_ROCE_MMAP_TYPE_DB);
348 	if (!context->db_mmap_entry)
349 		return -ENOMEM;
350 
351 	return 0;
352 }
353 
354 static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx,
355 				   struct ib_udata *udata)
356 {
357 	int ret;
358 	struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
359 	struct hns_roce_ib_alloc_ucontext_resp resp = {};
360 	struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
361 
362 	if (!hr_dev->active)
363 		return -EAGAIN;
364 
365 	resp.qp_tab_size = hr_dev->caps.num_qps;
366 	resp.srq_tab_size = hr_dev->caps.num_srqs;
367 
368 	ret = hns_roce_uar_alloc(hr_dev, &context->uar);
369 	if (ret)
370 		goto error_fail_uar_alloc;
371 
372 	ret = hns_roce_alloc_uar_entry(uctx);
373 	if (ret)
374 		goto error_fail_uar_entry;
375 
376 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
377 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
378 		INIT_LIST_HEAD(&context->page_list);
379 		mutex_init(&context->page_mutex);
380 	}
381 
382 	resp.cqe_size = hr_dev->caps.cqe_sz;
383 
384 	ret = ib_copy_to_udata(udata, &resp,
385 			       min(udata->outlen, sizeof(resp)));
386 	if (ret)
387 		goto error_fail_copy_to_udata;
388 
389 	return 0;
390 
391 error_fail_copy_to_udata:
392 	hns_roce_dealloc_uar_entry(context);
393 
394 error_fail_uar_entry:
395 	ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
396 
397 error_fail_uar_alloc:
398 	return ret;
399 }
400 
401 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
402 {
403 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
404 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcontext->device);
405 
406 	hns_roce_dealloc_uar_entry(context);
407 
408 	ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
409 }
410 
411 static int hns_roce_mmap(struct ib_ucontext *uctx, struct vm_area_struct *vma)
412 {
413 	struct rdma_user_mmap_entry *rdma_entry;
414 	struct hns_user_mmap_entry *entry;
415 	phys_addr_t pfn;
416 	pgprot_t prot;
417 	int ret;
418 
419 	rdma_entry = rdma_user_mmap_entry_get_pgoff(uctx, vma->vm_pgoff);
420 	if (!rdma_entry)
421 		return -EINVAL;
422 
423 	entry = to_hns_mmap(rdma_entry);
424 	pfn = entry->address >> PAGE_SHIFT;
425 
426 	switch (entry->mmap_type) {
427 	case HNS_ROCE_MMAP_TYPE_DB:
428 	case HNS_ROCE_MMAP_TYPE_DWQE:
429 		prot = pgprot_device(vma->vm_page_prot);
430 		break;
431 	default:
432 		return -EINVAL;
433 	}
434 
435 	ret = rdma_user_mmap_io(uctx, vma, pfn, rdma_entry->npages * PAGE_SIZE,
436 				prot, rdma_entry);
437 
438 	rdma_user_mmap_entry_put(rdma_entry);
439 
440 	return ret;
441 }
442 
443 static void hns_roce_free_mmap(struct rdma_user_mmap_entry *rdma_entry)
444 {
445 	struct hns_user_mmap_entry *entry = to_hns_mmap(rdma_entry);
446 
447 	kfree(entry);
448 }
449 
450 static int hns_roce_port_immutable(struct ib_device *ib_dev, u32 port_num,
451 				   struct ib_port_immutable *immutable)
452 {
453 	struct ib_port_attr attr;
454 	int ret;
455 
456 	ret = ib_query_port(ib_dev, port_num, &attr);
457 	if (ret)
458 		return ret;
459 
460 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
461 	immutable->gid_tbl_len = attr.gid_tbl_len;
462 
463 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
464 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
465 	if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
466 		immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
467 
468 	return 0;
469 }
470 
471 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
472 {
473 }
474 
475 static void hns_roce_get_fw_ver(struct ib_device *device, char *str)
476 {
477 	u64 fw_ver = to_hr_dev(device)->caps.fw_ver;
478 	unsigned int major, minor, sub_minor;
479 
480 	major = upper_32_bits(fw_ver);
481 	minor = high_16_bits(lower_32_bits(fw_ver));
482 	sub_minor = low_16_bits(fw_ver);
483 
484 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u.%04u", major, minor,
485 		 sub_minor);
486 }
487 
488 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
489 {
490 	struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
491 
492 	hr_dev->active = false;
493 	unregister_netdevice_notifier(&iboe->nb);
494 	ib_unregister_device(&hr_dev->ib_dev);
495 }
496 
497 static const struct ib_device_ops hns_roce_dev_ops = {
498 	.owner = THIS_MODULE,
499 	.driver_id = RDMA_DRIVER_HNS,
500 	.uverbs_abi_ver = 1,
501 	.uverbs_no_driver_id_binding = 1,
502 
503 	.get_dev_fw_str = hns_roce_get_fw_ver,
504 	.add_gid = hns_roce_add_gid,
505 	.alloc_pd = hns_roce_alloc_pd,
506 	.alloc_ucontext = hns_roce_alloc_ucontext,
507 	.create_ah = hns_roce_create_ah,
508 	.create_user_ah = hns_roce_create_ah,
509 	.create_cq = hns_roce_create_cq,
510 	.create_qp = hns_roce_create_qp,
511 	.dealloc_pd = hns_roce_dealloc_pd,
512 	.dealloc_ucontext = hns_roce_dealloc_ucontext,
513 	.del_gid = hns_roce_del_gid,
514 	.dereg_mr = hns_roce_dereg_mr,
515 	.destroy_ah = hns_roce_destroy_ah,
516 	.destroy_cq = hns_roce_destroy_cq,
517 	.disassociate_ucontext = hns_roce_disassociate_ucontext,
518 	.fill_res_cq_entry = hns_roce_fill_res_cq_entry,
519 	.get_dma_mr = hns_roce_get_dma_mr,
520 	.get_link_layer = hns_roce_get_link_layer,
521 	.get_port_immutable = hns_roce_port_immutable,
522 	.mmap = hns_roce_mmap,
523 	.mmap_free = hns_roce_free_mmap,
524 	.modify_device = hns_roce_modify_device,
525 	.modify_qp = hns_roce_modify_qp,
526 	.query_ah = hns_roce_query_ah,
527 	.query_device = hns_roce_query_device,
528 	.query_pkey = hns_roce_query_pkey,
529 	.query_port = hns_roce_query_port,
530 	.reg_user_mr = hns_roce_reg_user_mr,
531 
532 	INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
533 	INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
534 	INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
535 	INIT_RDMA_OBJ_SIZE(ib_qp, hns_roce_qp, ibqp),
536 	INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
537 };
538 
539 static const struct ib_device_ops hns_roce_dev_mr_ops = {
540 	.rereg_user_mr = hns_roce_rereg_user_mr,
541 };
542 
543 static const struct ib_device_ops hns_roce_dev_mw_ops = {
544 	.alloc_mw = hns_roce_alloc_mw,
545 	.dealloc_mw = hns_roce_dealloc_mw,
546 
547 	INIT_RDMA_OBJ_SIZE(ib_mw, hns_roce_mw, ibmw),
548 };
549 
550 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
551 	.alloc_mr = hns_roce_alloc_mr,
552 	.map_mr_sg = hns_roce_map_mr_sg,
553 };
554 
555 static const struct ib_device_ops hns_roce_dev_srq_ops = {
556 	.create_srq = hns_roce_create_srq,
557 	.destroy_srq = hns_roce_destroy_srq,
558 
559 	INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
560 };
561 
562 static const struct ib_device_ops hns_roce_dev_xrcd_ops = {
563 	.alloc_xrcd = hns_roce_alloc_xrcd,
564 	.dealloc_xrcd = hns_roce_dealloc_xrcd,
565 
566 	INIT_RDMA_OBJ_SIZE(ib_xrcd, hns_roce_xrcd, ibxrcd),
567 };
568 
569 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
570 {
571 	int ret;
572 	struct hns_roce_ib_iboe *iboe = NULL;
573 	struct ib_device *ib_dev = NULL;
574 	struct device *dev = hr_dev->dev;
575 	unsigned int i;
576 
577 	iboe = &hr_dev->iboe;
578 	spin_lock_init(&iboe->lock);
579 
580 	ib_dev = &hr_dev->ib_dev;
581 
582 	ib_dev->node_type = RDMA_NODE_IB_CA;
583 	ib_dev->dev.parent = dev;
584 
585 	ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
586 	ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
587 	ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
588 
589 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR)
590 		ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
591 
592 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_MW)
593 		ib_set_device_ops(ib_dev, &hns_roce_dev_mw_ops);
594 
595 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
596 		ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
597 
598 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
599 		ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
600 		ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
601 	}
602 
603 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
604 		ib_set_device_ops(ib_dev, &hns_roce_dev_xrcd_ops);
605 
606 	ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
607 	ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
608 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
609 		if (!hr_dev->iboe.netdevs[i])
610 			continue;
611 
612 		ret = ib_device_set_netdev(ib_dev, hr_dev->iboe.netdevs[i],
613 					   i + 1);
614 		if (ret)
615 			return ret;
616 	}
617 	dma_set_max_seg_size(dev, UINT_MAX);
618 	ret = ib_register_device(ib_dev, "hns_%d", dev);
619 	if (ret) {
620 		dev_err(dev, "ib_register_device failed!\n");
621 		return ret;
622 	}
623 
624 	ret = hns_roce_setup_mtu_mac(hr_dev);
625 	if (ret) {
626 		dev_err(dev, "setup_mtu_mac failed!\n");
627 		goto error_failed_setup_mtu_mac;
628 	}
629 
630 	iboe->nb.notifier_call = hns_roce_netdev_event;
631 	ret = register_netdevice_notifier(&iboe->nb);
632 	if (ret) {
633 		dev_err(dev, "register_netdevice_notifier failed!\n");
634 		goto error_failed_setup_mtu_mac;
635 	}
636 
637 	hr_dev->active = true;
638 	return 0;
639 
640 error_failed_setup_mtu_mac:
641 	ib_unregister_device(ib_dev);
642 
643 	return ret;
644 }
645 
646 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
647 {
648 	struct device *dev = hr_dev->dev;
649 	int ret;
650 
651 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
652 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
653 				      hr_dev->caps.num_mtpts, 1);
654 	if (ret) {
655 		dev_err(dev, "Failed to init MTPT context memory, aborting.\n");
656 		return ret;
657 	}
658 
659 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
660 				      HEM_TYPE_QPC, hr_dev->caps.qpc_sz,
661 				      hr_dev->caps.num_qps, 1);
662 	if (ret) {
663 		dev_err(dev, "Failed to init QP context memory, aborting.\n");
664 		goto err_unmap_dmpt;
665 	}
666 
667 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
668 				      HEM_TYPE_IRRL,
669 				      hr_dev->caps.irrl_entry_sz *
670 				      hr_dev->caps.max_qp_init_rdma,
671 				      hr_dev->caps.num_qps, 1);
672 	if (ret) {
673 		dev_err(dev, "Failed to init irrl_table memory, aborting.\n");
674 		goto err_unmap_qp;
675 	}
676 
677 	if (hr_dev->caps.trrl_entry_sz) {
678 		ret = hns_roce_init_hem_table(hr_dev,
679 					      &hr_dev->qp_table.trrl_table,
680 					      HEM_TYPE_TRRL,
681 					      hr_dev->caps.trrl_entry_sz *
682 					      hr_dev->caps.max_qp_dest_rdma,
683 					      hr_dev->caps.num_qps, 1);
684 		if (ret) {
685 			dev_err(dev,
686 				"Failed to init trrl_table memory, aborting.\n");
687 			goto err_unmap_irrl;
688 		}
689 	}
690 
691 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
692 				      HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
693 				      hr_dev->caps.num_cqs, 1);
694 	if (ret) {
695 		dev_err(dev, "Failed to init CQ context memory, aborting.\n");
696 		goto err_unmap_trrl;
697 	}
698 
699 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
700 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
701 					      HEM_TYPE_SRQC,
702 					      hr_dev->caps.srqc_entry_sz,
703 					      hr_dev->caps.num_srqs, 1);
704 		if (ret) {
705 			dev_err(dev,
706 				"Failed to init SRQ context memory, aborting.\n");
707 			goto err_unmap_cq;
708 		}
709 	}
710 
711 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
712 		ret = hns_roce_init_hem_table(hr_dev,
713 					      &hr_dev->qp_table.sccc_table,
714 					      HEM_TYPE_SCCC,
715 					      hr_dev->caps.sccc_sz,
716 					      hr_dev->caps.num_qps, 1);
717 		if (ret) {
718 			dev_err(dev,
719 				"Failed to init SCC context memory, aborting.\n");
720 			goto err_unmap_srq;
721 		}
722 	}
723 
724 	if (hr_dev->caps.qpc_timer_entry_sz) {
725 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
726 					      HEM_TYPE_QPC_TIMER,
727 					      hr_dev->caps.qpc_timer_entry_sz,
728 					      hr_dev->caps.num_qpc_timer, 1);
729 		if (ret) {
730 			dev_err(dev,
731 				"Failed to init QPC timer memory, aborting.\n");
732 			goto err_unmap_ctx;
733 		}
734 	}
735 
736 	if (hr_dev->caps.cqc_timer_entry_sz) {
737 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
738 					      HEM_TYPE_CQC_TIMER,
739 					      hr_dev->caps.cqc_timer_entry_sz,
740 					      hr_dev->caps.num_cqc_timer, 1);
741 		if (ret) {
742 			dev_err(dev,
743 				"Failed to init CQC timer memory, aborting.\n");
744 			goto err_unmap_qpc_timer;
745 		}
746 	}
747 
748 	if (hr_dev->caps.gmv_entry_sz) {
749 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->gmv_table,
750 					      HEM_TYPE_GMV,
751 					      hr_dev->caps.gmv_entry_sz,
752 					      hr_dev->caps.gmv_entry_num, 1);
753 		if (ret) {
754 			dev_err(dev,
755 				"failed to init gmv table memory, ret = %d\n",
756 				ret);
757 			goto err_unmap_cqc_timer;
758 		}
759 	}
760 
761 	return 0;
762 
763 err_unmap_cqc_timer:
764 	if (hr_dev->caps.cqc_timer_entry_sz)
765 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cqc_timer_table);
766 
767 err_unmap_qpc_timer:
768 	if (hr_dev->caps.qpc_timer_entry_sz)
769 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
770 
771 err_unmap_ctx:
772 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
773 		hns_roce_cleanup_hem_table(hr_dev,
774 					   &hr_dev->qp_table.sccc_table);
775 err_unmap_srq:
776 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
777 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
778 
779 err_unmap_cq:
780 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
781 
782 err_unmap_trrl:
783 	if (hr_dev->caps.trrl_entry_sz)
784 		hns_roce_cleanup_hem_table(hr_dev,
785 					   &hr_dev->qp_table.trrl_table);
786 
787 err_unmap_irrl:
788 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
789 
790 err_unmap_qp:
791 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
792 
793 err_unmap_dmpt:
794 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
795 
796 	return ret;
797 }
798 
799 /**
800  * hns_roce_setup_hca - setup host channel adapter
801  * @hr_dev: pointer to hns roce device
802  * Return : int
803  */
804 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
805 {
806 	struct device *dev = hr_dev->dev;
807 	int ret;
808 
809 	spin_lock_init(&hr_dev->sm_lock);
810 
811 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
812 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
813 		INIT_LIST_HEAD(&hr_dev->pgdir_list);
814 		mutex_init(&hr_dev->pgdir_mutex);
815 	}
816 
817 	hns_roce_init_uar_table(hr_dev);
818 
819 	ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
820 	if (ret) {
821 		dev_err(dev, "Failed to allocate priv_uar.\n");
822 		goto err_uar_table_free;
823 	}
824 
825 	ret = hns_roce_init_qp_table(hr_dev);
826 	if (ret) {
827 		dev_err(dev, "Failed to init qp_table.\n");
828 		goto err_uar_table_free;
829 	}
830 
831 	hns_roce_init_pd_table(hr_dev);
832 
833 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
834 		hns_roce_init_xrcd_table(hr_dev);
835 
836 	hns_roce_init_mr_table(hr_dev);
837 
838 	hns_roce_init_cq_table(hr_dev);
839 
840 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
841 		hns_roce_init_srq_table(hr_dev);
842 	}
843 
844 	return 0;
845 
846 err_uar_table_free:
847 	ida_destroy(&hr_dev->uar_ida.ida);
848 	return ret;
849 }
850 
851 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
852 {
853 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
854 	unsigned long flags;
855 
856 	spin_lock_irqsave(&hr_cq->lock, flags);
857 	if (cq->comp_handler) {
858 		if (!hr_cq->is_armed) {
859 			hr_cq->is_armed = 1;
860 			list_add_tail(&hr_cq->node, cq_list);
861 		}
862 	}
863 	spin_unlock_irqrestore(&hr_cq->lock, flags);
864 }
865 
866 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
867 {
868 	struct hns_roce_qp *hr_qp;
869 	struct hns_roce_cq *hr_cq;
870 	struct list_head cq_list;
871 	unsigned long flags_qp;
872 	unsigned long flags;
873 
874 	INIT_LIST_HEAD(&cq_list);
875 
876 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
877 	list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
878 		spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
879 		if (hr_qp->sq.tail != hr_qp->sq.head)
880 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
881 		spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
882 
883 		spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
884 		if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
885 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
886 		spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
887 	}
888 
889 	list_for_each_entry(hr_cq, &cq_list, node)
890 		hns_roce_cq_completion(hr_dev, hr_cq->cqn);
891 
892 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
893 }
894 
895 int hns_roce_init(struct hns_roce_dev *hr_dev)
896 {
897 	struct device *dev = hr_dev->dev;
898 	int ret;
899 
900 	hr_dev->is_reset = false;
901 
902 	if (hr_dev->hw->cmq_init) {
903 		ret = hr_dev->hw->cmq_init(hr_dev);
904 		if (ret) {
905 			dev_err(dev, "Init RoCE Command Queue failed!\n");
906 			return ret;
907 		}
908 	}
909 
910 	ret = hr_dev->hw->hw_profile(hr_dev);
911 	if (ret) {
912 		dev_err(dev, "Get RoCE engine profile failed!\n");
913 		goto error_failed_cmd_init;
914 	}
915 
916 	ret = hns_roce_cmd_init(hr_dev);
917 	if (ret) {
918 		dev_err(dev, "cmd init failed!\n");
919 		goto error_failed_cmd_init;
920 	}
921 
922 	/* EQ depends on poll mode, event mode depends on EQ */
923 	ret = hr_dev->hw->init_eq(hr_dev);
924 	if (ret) {
925 		dev_err(dev, "eq init failed!\n");
926 		goto error_failed_eq_table;
927 	}
928 
929 	if (hr_dev->cmd_mod) {
930 		ret = hns_roce_cmd_use_events(hr_dev);
931 		if (ret)
932 			dev_warn(dev,
933 				 "Cmd event  mode failed, set back to poll!\n");
934 	}
935 
936 	ret = hns_roce_init_hem(hr_dev);
937 	if (ret) {
938 		dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
939 		goto error_failed_init_hem;
940 	}
941 
942 	ret = hns_roce_setup_hca(hr_dev);
943 	if (ret) {
944 		dev_err(dev, "setup hca failed!\n");
945 		goto error_failed_setup_hca;
946 	}
947 
948 	if (hr_dev->hw->hw_init) {
949 		ret = hr_dev->hw->hw_init(hr_dev);
950 		if (ret) {
951 			dev_err(dev, "hw_init failed!\n");
952 			goto error_failed_engine_init;
953 		}
954 	}
955 
956 	INIT_LIST_HEAD(&hr_dev->qp_list);
957 	spin_lock_init(&hr_dev->qp_list_lock);
958 	INIT_LIST_HEAD(&hr_dev->dip_list);
959 	spin_lock_init(&hr_dev->dip_list_lock);
960 
961 	ret = hns_roce_register_device(hr_dev);
962 	if (ret)
963 		goto error_failed_register_device;
964 
965 	return 0;
966 
967 error_failed_register_device:
968 	if (hr_dev->hw->hw_exit)
969 		hr_dev->hw->hw_exit(hr_dev);
970 
971 error_failed_engine_init:
972 	hns_roce_cleanup_bitmap(hr_dev);
973 
974 error_failed_setup_hca:
975 	hns_roce_cleanup_hem(hr_dev);
976 
977 error_failed_init_hem:
978 	if (hr_dev->cmd_mod)
979 		hns_roce_cmd_use_polling(hr_dev);
980 	hr_dev->hw->cleanup_eq(hr_dev);
981 
982 error_failed_eq_table:
983 	hns_roce_cmd_cleanup(hr_dev);
984 
985 error_failed_cmd_init:
986 	if (hr_dev->hw->cmq_exit)
987 		hr_dev->hw->cmq_exit(hr_dev);
988 
989 	return ret;
990 }
991 
992 void hns_roce_exit(struct hns_roce_dev *hr_dev)
993 {
994 	hns_roce_unregister_device(hr_dev);
995 
996 	if (hr_dev->hw->hw_exit)
997 		hr_dev->hw->hw_exit(hr_dev);
998 	hns_roce_cleanup_bitmap(hr_dev);
999 	hns_roce_cleanup_hem(hr_dev);
1000 
1001 	if (hr_dev->cmd_mod)
1002 		hns_roce_cmd_use_polling(hr_dev);
1003 
1004 	hr_dev->hw->cleanup_eq(hr_dev);
1005 	hns_roce_cmd_cleanup(hr_dev);
1006 	if (hr_dev->hw->cmq_exit)
1007 		hr_dev->hw->cmq_exit(hr_dev);
1008 }
1009 
1010 MODULE_LICENSE("Dual BSD/GPL");
1011 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
1012 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
1013 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
1014 MODULE_DESCRIPTION("HNS RoCE Driver");
1015