xref: /openbmc/linux/drivers/infiniband/core/verbs.c (revision 4da722ca)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #include <linux/errno.h>
40 #include <linux/err.h>
41 #include <linux/export.h>
42 #include <linux/string.h>
43 #include <linux/slab.h>
44 #include <linux/in.h>
45 #include <linux/in6.h>
46 #include <net/addrconf.h>
47 #include <linux/security.h>
48 
49 #include <rdma/ib_verbs.h>
50 #include <rdma/ib_cache.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/rw.h>
53 
54 #include "core_priv.h"
55 
56 static const char * const ib_events[] = {
57 	[IB_EVENT_CQ_ERR]		= "CQ error",
58 	[IB_EVENT_QP_FATAL]		= "QP fatal error",
59 	[IB_EVENT_QP_REQ_ERR]		= "QP request error",
60 	[IB_EVENT_QP_ACCESS_ERR]	= "QP access error",
61 	[IB_EVENT_COMM_EST]		= "communication established",
62 	[IB_EVENT_SQ_DRAINED]		= "send queue drained",
63 	[IB_EVENT_PATH_MIG]		= "path migration successful",
64 	[IB_EVENT_PATH_MIG_ERR]		= "path migration error",
65 	[IB_EVENT_DEVICE_FATAL]		= "device fatal error",
66 	[IB_EVENT_PORT_ACTIVE]		= "port active",
67 	[IB_EVENT_PORT_ERR]		= "port error",
68 	[IB_EVENT_LID_CHANGE]		= "LID change",
69 	[IB_EVENT_PKEY_CHANGE]		= "P_key change",
70 	[IB_EVENT_SM_CHANGE]		= "SM change",
71 	[IB_EVENT_SRQ_ERR]		= "SRQ error",
72 	[IB_EVENT_SRQ_LIMIT_REACHED]	= "SRQ limit reached",
73 	[IB_EVENT_QP_LAST_WQE_REACHED]	= "last WQE reached",
74 	[IB_EVENT_CLIENT_REREGISTER]	= "client reregister",
75 	[IB_EVENT_GID_CHANGE]		= "GID changed",
76 };
77 
78 const char *__attribute_const__ ib_event_msg(enum ib_event_type event)
79 {
80 	size_t index = event;
81 
82 	return (index < ARRAY_SIZE(ib_events) && ib_events[index]) ?
83 			ib_events[index] : "unrecognized event";
84 }
85 EXPORT_SYMBOL(ib_event_msg);
86 
87 static const char * const wc_statuses[] = {
88 	[IB_WC_SUCCESS]			= "success",
89 	[IB_WC_LOC_LEN_ERR]		= "local length error",
90 	[IB_WC_LOC_QP_OP_ERR]		= "local QP operation error",
91 	[IB_WC_LOC_EEC_OP_ERR]		= "local EE context operation error",
92 	[IB_WC_LOC_PROT_ERR]		= "local protection error",
93 	[IB_WC_WR_FLUSH_ERR]		= "WR flushed",
94 	[IB_WC_MW_BIND_ERR]		= "memory management operation error",
95 	[IB_WC_BAD_RESP_ERR]		= "bad response error",
96 	[IB_WC_LOC_ACCESS_ERR]		= "local access error",
97 	[IB_WC_REM_INV_REQ_ERR]		= "invalid request error",
98 	[IB_WC_REM_ACCESS_ERR]		= "remote access error",
99 	[IB_WC_REM_OP_ERR]		= "remote operation error",
100 	[IB_WC_RETRY_EXC_ERR]		= "transport retry counter exceeded",
101 	[IB_WC_RNR_RETRY_EXC_ERR]	= "RNR retry counter exceeded",
102 	[IB_WC_LOC_RDD_VIOL_ERR]	= "local RDD violation error",
103 	[IB_WC_REM_INV_RD_REQ_ERR]	= "remote invalid RD request",
104 	[IB_WC_REM_ABORT_ERR]		= "operation aborted",
105 	[IB_WC_INV_EECN_ERR]		= "invalid EE context number",
106 	[IB_WC_INV_EEC_STATE_ERR]	= "invalid EE context state",
107 	[IB_WC_FATAL_ERR]		= "fatal error",
108 	[IB_WC_RESP_TIMEOUT_ERR]	= "response timeout error",
109 	[IB_WC_GENERAL_ERR]		= "general error",
110 };
111 
112 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status)
113 {
114 	size_t index = status;
115 
116 	return (index < ARRAY_SIZE(wc_statuses) && wc_statuses[index]) ?
117 			wc_statuses[index] : "unrecognized status";
118 }
119 EXPORT_SYMBOL(ib_wc_status_msg);
120 
121 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate)
122 {
123 	switch (rate) {
124 	case IB_RATE_2_5_GBPS: return  1;
125 	case IB_RATE_5_GBPS:   return  2;
126 	case IB_RATE_10_GBPS:  return  4;
127 	case IB_RATE_20_GBPS:  return  8;
128 	case IB_RATE_30_GBPS:  return 12;
129 	case IB_RATE_40_GBPS:  return 16;
130 	case IB_RATE_60_GBPS:  return 24;
131 	case IB_RATE_80_GBPS:  return 32;
132 	case IB_RATE_120_GBPS: return 48;
133 	default:	       return -1;
134 	}
135 }
136 EXPORT_SYMBOL(ib_rate_to_mult);
137 
138 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult)
139 {
140 	switch (mult) {
141 	case 1:  return IB_RATE_2_5_GBPS;
142 	case 2:  return IB_RATE_5_GBPS;
143 	case 4:  return IB_RATE_10_GBPS;
144 	case 8:  return IB_RATE_20_GBPS;
145 	case 12: return IB_RATE_30_GBPS;
146 	case 16: return IB_RATE_40_GBPS;
147 	case 24: return IB_RATE_60_GBPS;
148 	case 32: return IB_RATE_80_GBPS;
149 	case 48: return IB_RATE_120_GBPS;
150 	default: return IB_RATE_PORT_CURRENT;
151 	}
152 }
153 EXPORT_SYMBOL(mult_to_ib_rate);
154 
155 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate)
156 {
157 	switch (rate) {
158 	case IB_RATE_2_5_GBPS: return 2500;
159 	case IB_RATE_5_GBPS:   return 5000;
160 	case IB_RATE_10_GBPS:  return 10000;
161 	case IB_RATE_20_GBPS:  return 20000;
162 	case IB_RATE_30_GBPS:  return 30000;
163 	case IB_RATE_40_GBPS:  return 40000;
164 	case IB_RATE_60_GBPS:  return 60000;
165 	case IB_RATE_80_GBPS:  return 80000;
166 	case IB_RATE_120_GBPS: return 120000;
167 	case IB_RATE_14_GBPS:  return 14062;
168 	case IB_RATE_56_GBPS:  return 56250;
169 	case IB_RATE_112_GBPS: return 112500;
170 	case IB_RATE_168_GBPS: return 168750;
171 	case IB_RATE_25_GBPS:  return 25781;
172 	case IB_RATE_100_GBPS: return 103125;
173 	case IB_RATE_200_GBPS: return 206250;
174 	case IB_RATE_300_GBPS: return 309375;
175 	default:	       return -1;
176 	}
177 }
178 EXPORT_SYMBOL(ib_rate_to_mbps);
179 
180 __attribute_const__ enum rdma_transport_type
181 rdma_node_get_transport(enum rdma_node_type node_type)
182 {
183 	switch (node_type) {
184 	case RDMA_NODE_IB_CA:
185 	case RDMA_NODE_IB_SWITCH:
186 	case RDMA_NODE_IB_ROUTER:
187 		return RDMA_TRANSPORT_IB;
188 	case RDMA_NODE_RNIC:
189 		return RDMA_TRANSPORT_IWARP;
190 	case RDMA_NODE_USNIC:
191 		return RDMA_TRANSPORT_USNIC;
192 	case RDMA_NODE_USNIC_UDP:
193 		return RDMA_TRANSPORT_USNIC_UDP;
194 	default:
195 		BUG();
196 		return 0;
197 	}
198 }
199 EXPORT_SYMBOL(rdma_node_get_transport);
200 
201 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, u8 port_num)
202 {
203 	if (device->get_link_layer)
204 		return device->get_link_layer(device, port_num);
205 
206 	switch (rdma_node_get_transport(device->node_type)) {
207 	case RDMA_TRANSPORT_IB:
208 		return IB_LINK_LAYER_INFINIBAND;
209 	case RDMA_TRANSPORT_IWARP:
210 	case RDMA_TRANSPORT_USNIC:
211 	case RDMA_TRANSPORT_USNIC_UDP:
212 		return IB_LINK_LAYER_ETHERNET;
213 	default:
214 		return IB_LINK_LAYER_UNSPECIFIED;
215 	}
216 }
217 EXPORT_SYMBOL(rdma_port_get_link_layer);
218 
219 /* Protection domains */
220 
221 /**
222  * ib_alloc_pd - Allocates an unused protection domain.
223  * @device: The device on which to allocate the protection domain.
224  *
225  * A protection domain object provides an association between QPs, shared
226  * receive queues, address handles, memory regions, and memory windows.
227  *
228  * Every PD has a local_dma_lkey which can be used as the lkey value for local
229  * memory operations.
230  */
231 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
232 		const char *caller)
233 {
234 	struct ib_pd *pd;
235 	int mr_access_flags = 0;
236 
237 	pd = device->alloc_pd(device, NULL, NULL);
238 	if (IS_ERR(pd))
239 		return pd;
240 
241 	pd->device = device;
242 	pd->uobject = NULL;
243 	pd->__internal_mr = NULL;
244 	atomic_set(&pd->usecnt, 0);
245 	pd->flags = flags;
246 
247 	if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
248 		pd->local_dma_lkey = device->local_dma_lkey;
249 	else
250 		mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
251 
252 	if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
253 		pr_warn("%s: enabling unsafe global rkey\n", caller);
254 		mr_access_flags |= IB_ACCESS_REMOTE_READ | IB_ACCESS_REMOTE_WRITE;
255 	}
256 
257 	if (mr_access_flags) {
258 		struct ib_mr *mr;
259 
260 		mr = pd->device->get_dma_mr(pd, mr_access_flags);
261 		if (IS_ERR(mr)) {
262 			ib_dealloc_pd(pd);
263 			return ERR_CAST(mr);
264 		}
265 
266 		mr->device	= pd->device;
267 		mr->pd		= pd;
268 		mr->uobject	= NULL;
269 		mr->need_inval	= false;
270 
271 		pd->__internal_mr = mr;
272 
273 		if (!(device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY))
274 			pd->local_dma_lkey = pd->__internal_mr->lkey;
275 
276 		if (flags & IB_PD_UNSAFE_GLOBAL_RKEY)
277 			pd->unsafe_global_rkey = pd->__internal_mr->rkey;
278 	}
279 
280 	return pd;
281 }
282 EXPORT_SYMBOL(__ib_alloc_pd);
283 
284 /**
285  * ib_dealloc_pd - Deallocates a protection domain.
286  * @pd: The protection domain to deallocate.
287  *
288  * It is an error to call this function while any resources in the pd still
289  * exist.  The caller is responsible to synchronously destroy them and
290  * guarantee no new allocations will happen.
291  */
292 void ib_dealloc_pd(struct ib_pd *pd)
293 {
294 	int ret;
295 
296 	if (pd->__internal_mr) {
297 		ret = pd->device->dereg_mr(pd->__internal_mr);
298 		WARN_ON(ret);
299 		pd->__internal_mr = NULL;
300 	}
301 
302 	/* uverbs manipulates usecnt with proper locking, while the kabi
303 	   requires the caller to guarantee we can't race here. */
304 	WARN_ON(atomic_read(&pd->usecnt));
305 
306 	/* Making delalloc_pd a void return is a WIP, no driver should return
307 	   an error here. */
308 	ret = pd->device->dealloc_pd(pd);
309 	WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd");
310 }
311 EXPORT_SYMBOL(ib_dealloc_pd);
312 
313 /* Address handles */
314 
315 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr)
316 {
317 	struct ib_ah *ah;
318 
319 	ah = pd->device->create_ah(pd, ah_attr, NULL);
320 
321 	if (!IS_ERR(ah)) {
322 		ah->device  = pd->device;
323 		ah->pd      = pd;
324 		ah->uobject = NULL;
325 		ah->type    = ah_attr->type;
326 		atomic_inc(&pd->usecnt);
327 	}
328 
329 	return ah;
330 }
331 EXPORT_SYMBOL(rdma_create_ah);
332 
333 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr)
334 {
335 	const struct iphdr *ip4h = (struct iphdr *)&hdr->roce4grh;
336 	struct iphdr ip4h_checked;
337 	const struct ipv6hdr *ip6h = (struct ipv6hdr *)&hdr->ibgrh;
338 
339 	/* If it's IPv6, the version must be 6, otherwise, the first
340 	 * 20 bytes (before the IPv4 header) are garbled.
341 	 */
342 	if (ip6h->version != 6)
343 		return (ip4h->version == 4) ? 4 : 0;
344 	/* version may be 6 or 4 because the first 20 bytes could be garbled */
345 
346 	/* RoCE v2 requires no options, thus header length
347 	 * must be 5 words
348 	 */
349 	if (ip4h->ihl != 5)
350 		return 6;
351 
352 	/* Verify checksum.
353 	 * We can't write on scattered buffers so we need to copy to
354 	 * temp buffer.
355 	 */
356 	memcpy(&ip4h_checked, ip4h, sizeof(ip4h_checked));
357 	ip4h_checked.check = 0;
358 	ip4h_checked.check = ip_fast_csum((u8 *)&ip4h_checked, 5);
359 	/* if IPv4 header checksum is OK, believe it */
360 	if (ip4h->check == ip4h_checked.check)
361 		return 4;
362 	return 6;
363 }
364 EXPORT_SYMBOL(ib_get_rdma_header_version);
365 
366 static enum rdma_network_type ib_get_net_type_by_grh(struct ib_device *device,
367 						     u8 port_num,
368 						     const struct ib_grh *grh)
369 {
370 	int grh_version;
371 
372 	if (rdma_protocol_ib(device, port_num))
373 		return RDMA_NETWORK_IB;
374 
375 	grh_version = ib_get_rdma_header_version((union rdma_network_hdr *)grh);
376 
377 	if (grh_version == 4)
378 		return RDMA_NETWORK_IPV4;
379 
380 	if (grh->next_hdr == IPPROTO_UDP)
381 		return RDMA_NETWORK_IPV6;
382 
383 	return RDMA_NETWORK_ROCE_V1;
384 }
385 
386 struct find_gid_index_context {
387 	u16 vlan_id;
388 	enum ib_gid_type gid_type;
389 };
390 
391 static bool find_gid_index(const union ib_gid *gid,
392 			   const struct ib_gid_attr *gid_attr,
393 			   void *context)
394 {
395 	struct find_gid_index_context *ctx =
396 		(struct find_gid_index_context *)context;
397 
398 	if (ctx->gid_type != gid_attr->gid_type)
399 		return false;
400 
401 	if ((!!(ctx->vlan_id != 0xffff) == !is_vlan_dev(gid_attr->ndev)) ||
402 	    (is_vlan_dev(gid_attr->ndev) &&
403 	     vlan_dev_vlan_id(gid_attr->ndev) != ctx->vlan_id))
404 		return false;
405 
406 	return true;
407 }
408 
409 static int get_sgid_index_from_eth(struct ib_device *device, u8 port_num,
410 				   u16 vlan_id, const union ib_gid *sgid,
411 				   enum ib_gid_type gid_type,
412 				   u16 *gid_index)
413 {
414 	struct find_gid_index_context context = {.vlan_id = vlan_id,
415 						 .gid_type = gid_type};
416 
417 	return ib_find_gid_by_filter(device, sgid, port_num, find_gid_index,
418 				     &context, gid_index);
419 }
420 
421 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
422 			      enum rdma_network_type net_type,
423 			      union ib_gid *sgid, union ib_gid *dgid)
424 {
425 	struct sockaddr_in  src_in;
426 	struct sockaddr_in  dst_in;
427 	__be32 src_saddr, dst_saddr;
428 
429 	if (!sgid || !dgid)
430 		return -EINVAL;
431 
432 	if (net_type == RDMA_NETWORK_IPV4) {
433 		memcpy(&src_in.sin_addr.s_addr,
434 		       &hdr->roce4grh.saddr, 4);
435 		memcpy(&dst_in.sin_addr.s_addr,
436 		       &hdr->roce4grh.daddr, 4);
437 		src_saddr = src_in.sin_addr.s_addr;
438 		dst_saddr = dst_in.sin_addr.s_addr;
439 		ipv6_addr_set_v4mapped(src_saddr,
440 				       (struct in6_addr *)sgid);
441 		ipv6_addr_set_v4mapped(dst_saddr,
442 				       (struct in6_addr *)dgid);
443 		return 0;
444 	} else if (net_type == RDMA_NETWORK_IPV6 ||
445 		   net_type == RDMA_NETWORK_IB) {
446 		*dgid = hdr->ibgrh.dgid;
447 		*sgid = hdr->ibgrh.sgid;
448 		return 0;
449 	} else {
450 		return -EINVAL;
451 	}
452 }
453 EXPORT_SYMBOL(ib_get_gids_from_rdma_hdr);
454 
455 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
456 		       const struct ib_wc *wc, const struct ib_grh *grh,
457 		       struct rdma_ah_attr *ah_attr)
458 {
459 	u32 flow_class;
460 	u16 gid_index;
461 	int ret;
462 	enum rdma_network_type net_type = RDMA_NETWORK_IB;
463 	enum ib_gid_type gid_type = IB_GID_TYPE_IB;
464 	int hoplimit = 0xff;
465 	union ib_gid dgid;
466 	union ib_gid sgid;
467 
468 	memset(ah_attr, 0, sizeof *ah_attr);
469 	ah_attr->type = rdma_ah_find_type(device, port_num);
470 	if (rdma_cap_eth_ah(device, port_num)) {
471 		if (wc->wc_flags & IB_WC_WITH_NETWORK_HDR_TYPE)
472 			net_type = wc->network_hdr_type;
473 		else
474 			net_type = ib_get_net_type_by_grh(device, port_num, grh);
475 		gid_type = ib_network_to_gid_type(net_type);
476 	}
477 	ret = ib_get_gids_from_rdma_hdr((union rdma_network_hdr *)grh, net_type,
478 					&sgid, &dgid);
479 	if (ret)
480 		return ret;
481 
482 	if (rdma_protocol_roce(device, port_num)) {
483 		int if_index = 0;
484 		u16 vlan_id = wc->wc_flags & IB_WC_WITH_VLAN ?
485 				wc->vlan_id : 0xffff;
486 		struct net_device *idev;
487 		struct net_device *resolved_dev;
488 
489 		if (!(wc->wc_flags & IB_WC_GRH))
490 			return -EPROTOTYPE;
491 
492 		if (!device->get_netdev)
493 			return -EOPNOTSUPP;
494 
495 		idev = device->get_netdev(device, port_num);
496 		if (!idev)
497 			return -ENODEV;
498 
499 		ret = rdma_addr_find_l2_eth_by_grh(&dgid, &sgid,
500 						   ah_attr->roce.dmac,
501 						   wc->wc_flags & IB_WC_WITH_VLAN ?
502 						   NULL : &vlan_id,
503 						   &if_index, &hoplimit);
504 		if (ret) {
505 			dev_put(idev);
506 			return ret;
507 		}
508 
509 		resolved_dev = dev_get_by_index(&init_net, if_index);
510 		if (resolved_dev->flags & IFF_LOOPBACK) {
511 			dev_put(resolved_dev);
512 			resolved_dev = idev;
513 			dev_hold(resolved_dev);
514 		}
515 		rcu_read_lock();
516 		if (resolved_dev != idev && !rdma_is_upper_dev_rcu(idev,
517 								   resolved_dev))
518 			ret = -EHOSTUNREACH;
519 		rcu_read_unlock();
520 		dev_put(idev);
521 		dev_put(resolved_dev);
522 		if (ret)
523 			return ret;
524 
525 		ret = get_sgid_index_from_eth(device, port_num, vlan_id,
526 					      &dgid, gid_type, &gid_index);
527 		if (ret)
528 			return ret;
529 	}
530 
531 	rdma_ah_set_dlid(ah_attr, wc->slid);
532 	rdma_ah_set_sl(ah_attr, wc->sl);
533 	rdma_ah_set_path_bits(ah_attr, wc->dlid_path_bits);
534 	rdma_ah_set_port_num(ah_attr, port_num);
535 
536 	if (wc->wc_flags & IB_WC_GRH) {
537 		if (!rdma_cap_eth_ah(device, port_num)) {
538 			if (dgid.global.interface_id != cpu_to_be64(IB_SA_WELL_KNOWN_GUID)) {
539 				ret = ib_find_cached_gid_by_port(device, &dgid,
540 								 IB_GID_TYPE_IB,
541 								 port_num, NULL,
542 								 &gid_index);
543 				if (ret)
544 					return ret;
545 			} else {
546 				gid_index = 0;
547 			}
548 		}
549 
550 		flow_class = be32_to_cpu(grh->version_tclass_flow);
551 		rdma_ah_set_grh(ah_attr, &sgid,
552 				flow_class & 0xFFFFF,
553 				(u8)gid_index, hoplimit,
554 				(flow_class >> 20) & 0xFF);
555 
556 	}
557 	return 0;
558 }
559 EXPORT_SYMBOL(ib_init_ah_from_wc);
560 
561 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
562 				   const struct ib_grh *grh, u8 port_num)
563 {
564 	struct rdma_ah_attr ah_attr;
565 	int ret;
566 
567 	ret = ib_init_ah_from_wc(pd->device, port_num, wc, grh, &ah_attr);
568 	if (ret)
569 		return ERR_PTR(ret);
570 
571 	return rdma_create_ah(pd, &ah_attr);
572 }
573 EXPORT_SYMBOL(ib_create_ah_from_wc);
574 
575 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr)
576 {
577 	if (ah->type != ah_attr->type)
578 		return -EINVAL;
579 
580 	return ah->device->modify_ah ?
581 		ah->device->modify_ah(ah, ah_attr) :
582 		-ENOSYS;
583 }
584 EXPORT_SYMBOL(rdma_modify_ah);
585 
586 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr)
587 {
588 	return ah->device->query_ah ?
589 		ah->device->query_ah(ah, ah_attr) :
590 		-ENOSYS;
591 }
592 EXPORT_SYMBOL(rdma_query_ah);
593 
594 int rdma_destroy_ah(struct ib_ah *ah)
595 {
596 	struct ib_pd *pd;
597 	int ret;
598 
599 	pd = ah->pd;
600 	ret = ah->device->destroy_ah(ah);
601 	if (!ret)
602 		atomic_dec(&pd->usecnt);
603 
604 	return ret;
605 }
606 EXPORT_SYMBOL(rdma_destroy_ah);
607 
608 /* Shared receive queues */
609 
610 struct ib_srq *ib_create_srq(struct ib_pd *pd,
611 			     struct ib_srq_init_attr *srq_init_attr)
612 {
613 	struct ib_srq *srq;
614 
615 	if (!pd->device->create_srq)
616 		return ERR_PTR(-ENOSYS);
617 
618 	srq = pd->device->create_srq(pd, srq_init_attr, NULL);
619 
620 	if (!IS_ERR(srq)) {
621 		srq->device    	   = pd->device;
622 		srq->pd        	   = pd;
623 		srq->uobject       = NULL;
624 		srq->event_handler = srq_init_attr->event_handler;
625 		srq->srq_context   = srq_init_attr->srq_context;
626 		srq->srq_type      = srq_init_attr->srq_type;
627 		if (srq->srq_type == IB_SRQT_XRC) {
628 			srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
629 			srq->ext.xrc.cq   = srq_init_attr->ext.xrc.cq;
630 			atomic_inc(&srq->ext.xrc.xrcd->usecnt);
631 			atomic_inc(&srq->ext.xrc.cq->usecnt);
632 		}
633 		atomic_inc(&pd->usecnt);
634 		atomic_set(&srq->usecnt, 0);
635 	}
636 
637 	return srq;
638 }
639 EXPORT_SYMBOL(ib_create_srq);
640 
641 int ib_modify_srq(struct ib_srq *srq,
642 		  struct ib_srq_attr *srq_attr,
643 		  enum ib_srq_attr_mask srq_attr_mask)
644 {
645 	return srq->device->modify_srq ?
646 		srq->device->modify_srq(srq, srq_attr, srq_attr_mask, NULL) :
647 		-ENOSYS;
648 }
649 EXPORT_SYMBOL(ib_modify_srq);
650 
651 int ib_query_srq(struct ib_srq *srq,
652 		 struct ib_srq_attr *srq_attr)
653 {
654 	return srq->device->query_srq ?
655 		srq->device->query_srq(srq, srq_attr) : -ENOSYS;
656 }
657 EXPORT_SYMBOL(ib_query_srq);
658 
659 int ib_destroy_srq(struct ib_srq *srq)
660 {
661 	struct ib_pd *pd;
662 	enum ib_srq_type srq_type;
663 	struct ib_xrcd *uninitialized_var(xrcd);
664 	struct ib_cq *uninitialized_var(cq);
665 	int ret;
666 
667 	if (atomic_read(&srq->usecnt))
668 		return -EBUSY;
669 
670 	pd = srq->pd;
671 	srq_type = srq->srq_type;
672 	if (srq_type == IB_SRQT_XRC) {
673 		xrcd = srq->ext.xrc.xrcd;
674 		cq = srq->ext.xrc.cq;
675 	}
676 
677 	ret = srq->device->destroy_srq(srq);
678 	if (!ret) {
679 		atomic_dec(&pd->usecnt);
680 		if (srq_type == IB_SRQT_XRC) {
681 			atomic_dec(&xrcd->usecnt);
682 			atomic_dec(&cq->usecnt);
683 		}
684 	}
685 
686 	return ret;
687 }
688 EXPORT_SYMBOL(ib_destroy_srq);
689 
690 /* Queue pairs */
691 
692 static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
693 {
694 	struct ib_qp *qp = context;
695 	unsigned long flags;
696 
697 	spin_lock_irqsave(&qp->device->event_handler_lock, flags);
698 	list_for_each_entry(event->element.qp, &qp->open_list, open_list)
699 		if (event->element.qp->event_handler)
700 			event->element.qp->event_handler(event, event->element.qp->qp_context);
701 	spin_unlock_irqrestore(&qp->device->event_handler_lock, flags);
702 }
703 
704 static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
705 {
706 	mutex_lock(&xrcd->tgt_qp_mutex);
707 	list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
708 	mutex_unlock(&xrcd->tgt_qp_mutex);
709 }
710 
711 static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
712 				  void (*event_handler)(struct ib_event *, void *),
713 				  void *qp_context)
714 {
715 	struct ib_qp *qp;
716 	unsigned long flags;
717 	int err;
718 
719 	qp = kzalloc(sizeof *qp, GFP_KERNEL);
720 	if (!qp)
721 		return ERR_PTR(-ENOMEM);
722 
723 	qp->real_qp = real_qp;
724 	err = ib_open_shared_qp_security(qp, real_qp->device);
725 	if (err) {
726 		kfree(qp);
727 		return ERR_PTR(err);
728 	}
729 
730 	qp->real_qp = real_qp;
731 	atomic_inc(&real_qp->usecnt);
732 	qp->device = real_qp->device;
733 	qp->event_handler = event_handler;
734 	qp->qp_context = qp_context;
735 	qp->qp_num = real_qp->qp_num;
736 	qp->qp_type = real_qp->qp_type;
737 
738 	spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
739 	list_add(&qp->open_list, &real_qp->open_list);
740 	spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
741 
742 	return qp;
743 }
744 
745 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
746 			 struct ib_qp_open_attr *qp_open_attr)
747 {
748 	struct ib_qp *qp, *real_qp;
749 
750 	if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
751 		return ERR_PTR(-EINVAL);
752 
753 	qp = ERR_PTR(-EINVAL);
754 	mutex_lock(&xrcd->tgt_qp_mutex);
755 	list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
756 		if (real_qp->qp_num == qp_open_attr->qp_num) {
757 			qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
758 					  qp_open_attr->qp_context);
759 			break;
760 		}
761 	}
762 	mutex_unlock(&xrcd->tgt_qp_mutex);
763 	return qp;
764 }
765 EXPORT_SYMBOL(ib_open_qp);
766 
767 static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp,
768 		struct ib_qp_init_attr *qp_init_attr)
769 {
770 	struct ib_qp *real_qp = qp;
771 
772 	qp->event_handler = __ib_shared_qp_event_handler;
773 	qp->qp_context = qp;
774 	qp->pd = NULL;
775 	qp->send_cq = qp->recv_cq = NULL;
776 	qp->srq = NULL;
777 	qp->xrcd = qp_init_attr->xrcd;
778 	atomic_inc(&qp_init_attr->xrcd->usecnt);
779 	INIT_LIST_HEAD(&qp->open_list);
780 
781 	qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
782 			  qp_init_attr->qp_context);
783 	if (!IS_ERR(qp))
784 		__ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
785 	else
786 		real_qp->device->destroy_qp(real_qp);
787 	return qp;
788 }
789 
790 struct ib_qp *ib_create_qp(struct ib_pd *pd,
791 			   struct ib_qp_init_attr *qp_init_attr)
792 {
793 	struct ib_device *device = pd ? pd->device : qp_init_attr->xrcd->device;
794 	struct ib_qp *qp;
795 	int ret;
796 
797 	if (qp_init_attr->rwq_ind_tbl &&
798 	    (qp_init_attr->recv_cq ||
799 	    qp_init_attr->srq || qp_init_attr->cap.max_recv_wr ||
800 	    qp_init_attr->cap.max_recv_sge))
801 		return ERR_PTR(-EINVAL);
802 
803 	/*
804 	 * If the callers is using the RDMA API calculate the resources
805 	 * needed for the RDMA READ/WRITE operations.
806 	 *
807 	 * Note that these callers need to pass in a port number.
808 	 */
809 	if (qp_init_attr->cap.max_rdma_ctxs)
810 		rdma_rw_init_qp(device, qp_init_attr);
811 
812 	qp = device->create_qp(pd, qp_init_attr, NULL);
813 	if (IS_ERR(qp))
814 		return qp;
815 
816 	ret = ib_create_qp_security(qp, device);
817 	if (ret) {
818 		ib_destroy_qp(qp);
819 		return ERR_PTR(ret);
820 	}
821 
822 	qp->device     = device;
823 	qp->real_qp    = qp;
824 	qp->uobject    = NULL;
825 	qp->qp_type    = qp_init_attr->qp_type;
826 	qp->rwq_ind_tbl = qp_init_attr->rwq_ind_tbl;
827 
828 	atomic_set(&qp->usecnt, 0);
829 	qp->mrs_used = 0;
830 	spin_lock_init(&qp->mr_lock);
831 	INIT_LIST_HEAD(&qp->rdma_mrs);
832 	INIT_LIST_HEAD(&qp->sig_mrs);
833 
834 	if (qp_init_attr->qp_type == IB_QPT_XRC_TGT)
835 		return ib_create_xrc_qp(qp, qp_init_attr);
836 
837 	qp->event_handler = qp_init_attr->event_handler;
838 	qp->qp_context = qp_init_attr->qp_context;
839 	if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
840 		qp->recv_cq = NULL;
841 		qp->srq = NULL;
842 	} else {
843 		qp->recv_cq = qp_init_attr->recv_cq;
844 		if (qp_init_attr->recv_cq)
845 			atomic_inc(&qp_init_attr->recv_cq->usecnt);
846 		qp->srq = qp_init_attr->srq;
847 		if (qp->srq)
848 			atomic_inc(&qp_init_attr->srq->usecnt);
849 	}
850 
851 	qp->pd	    = pd;
852 	qp->send_cq = qp_init_attr->send_cq;
853 	qp->xrcd    = NULL;
854 
855 	atomic_inc(&pd->usecnt);
856 	if (qp_init_attr->send_cq)
857 		atomic_inc(&qp_init_attr->send_cq->usecnt);
858 	if (qp_init_attr->rwq_ind_tbl)
859 		atomic_inc(&qp->rwq_ind_tbl->usecnt);
860 
861 	if (qp_init_attr->cap.max_rdma_ctxs) {
862 		ret = rdma_rw_init_mrs(qp, qp_init_attr);
863 		if (ret) {
864 			pr_err("failed to init MR pool ret= %d\n", ret);
865 			ib_destroy_qp(qp);
866 			return ERR_PTR(ret);
867 		}
868 	}
869 
870 	/*
871 	 * Note: all hw drivers guarantee that max_send_sge is lower than
872 	 * the device RDMA WRITE SGE limit but not all hw drivers ensure that
873 	 * max_send_sge <= max_sge_rd.
874 	 */
875 	qp->max_write_sge = qp_init_attr->cap.max_send_sge;
876 	qp->max_read_sge = min_t(u32, qp_init_attr->cap.max_send_sge,
877 				 device->attrs.max_sge_rd);
878 
879 	return qp;
880 }
881 EXPORT_SYMBOL(ib_create_qp);
882 
883 static const struct {
884 	int			valid;
885 	enum ib_qp_attr_mask	req_param[IB_QPT_MAX];
886 	enum ib_qp_attr_mask	opt_param[IB_QPT_MAX];
887 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
888 	[IB_QPS_RESET] = {
889 		[IB_QPS_RESET] = { .valid = 1 },
890 		[IB_QPS_INIT]  = {
891 			.valid = 1,
892 			.req_param = {
893 				[IB_QPT_UD]  = (IB_QP_PKEY_INDEX		|
894 						IB_QP_PORT			|
895 						IB_QP_QKEY),
896 				[IB_QPT_RAW_PACKET] = IB_QP_PORT,
897 				[IB_QPT_UC]  = (IB_QP_PKEY_INDEX		|
898 						IB_QP_PORT			|
899 						IB_QP_ACCESS_FLAGS),
900 				[IB_QPT_RC]  = (IB_QP_PKEY_INDEX		|
901 						IB_QP_PORT			|
902 						IB_QP_ACCESS_FLAGS),
903 				[IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX		|
904 						IB_QP_PORT			|
905 						IB_QP_ACCESS_FLAGS),
906 				[IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX		|
907 						IB_QP_PORT			|
908 						IB_QP_ACCESS_FLAGS),
909 				[IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
910 						IB_QP_QKEY),
911 				[IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
912 						IB_QP_QKEY),
913 			}
914 		},
915 	},
916 	[IB_QPS_INIT]  = {
917 		[IB_QPS_RESET] = { .valid = 1 },
918 		[IB_QPS_ERR] =   { .valid = 1 },
919 		[IB_QPS_INIT]  = {
920 			.valid = 1,
921 			.opt_param = {
922 				[IB_QPT_UD]  = (IB_QP_PKEY_INDEX		|
923 						IB_QP_PORT			|
924 						IB_QP_QKEY),
925 				[IB_QPT_UC]  = (IB_QP_PKEY_INDEX		|
926 						IB_QP_PORT			|
927 						IB_QP_ACCESS_FLAGS),
928 				[IB_QPT_RC]  = (IB_QP_PKEY_INDEX		|
929 						IB_QP_PORT			|
930 						IB_QP_ACCESS_FLAGS),
931 				[IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX		|
932 						IB_QP_PORT			|
933 						IB_QP_ACCESS_FLAGS),
934 				[IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX		|
935 						IB_QP_PORT			|
936 						IB_QP_ACCESS_FLAGS),
937 				[IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
938 						IB_QP_QKEY),
939 				[IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
940 						IB_QP_QKEY),
941 			}
942 		},
943 		[IB_QPS_RTR]   = {
944 			.valid = 1,
945 			.req_param = {
946 				[IB_QPT_UC]  = (IB_QP_AV			|
947 						IB_QP_PATH_MTU			|
948 						IB_QP_DEST_QPN			|
949 						IB_QP_RQ_PSN),
950 				[IB_QPT_RC]  = (IB_QP_AV			|
951 						IB_QP_PATH_MTU			|
952 						IB_QP_DEST_QPN			|
953 						IB_QP_RQ_PSN			|
954 						IB_QP_MAX_DEST_RD_ATOMIC	|
955 						IB_QP_MIN_RNR_TIMER),
956 				[IB_QPT_XRC_INI] = (IB_QP_AV			|
957 						IB_QP_PATH_MTU			|
958 						IB_QP_DEST_QPN			|
959 						IB_QP_RQ_PSN),
960 				[IB_QPT_XRC_TGT] = (IB_QP_AV			|
961 						IB_QP_PATH_MTU			|
962 						IB_QP_DEST_QPN			|
963 						IB_QP_RQ_PSN			|
964 						IB_QP_MAX_DEST_RD_ATOMIC	|
965 						IB_QP_MIN_RNR_TIMER),
966 			},
967 			.opt_param = {
968 				 [IB_QPT_UD]  = (IB_QP_PKEY_INDEX		|
969 						 IB_QP_QKEY),
970 				 [IB_QPT_UC]  = (IB_QP_ALT_PATH			|
971 						 IB_QP_ACCESS_FLAGS		|
972 						 IB_QP_PKEY_INDEX),
973 				 [IB_QPT_RC]  = (IB_QP_ALT_PATH			|
974 						 IB_QP_ACCESS_FLAGS		|
975 						 IB_QP_PKEY_INDEX),
976 				 [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH		|
977 						 IB_QP_ACCESS_FLAGS		|
978 						 IB_QP_PKEY_INDEX),
979 				 [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH		|
980 						 IB_QP_ACCESS_FLAGS		|
981 						 IB_QP_PKEY_INDEX),
982 				 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
983 						 IB_QP_QKEY),
984 				 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
985 						 IB_QP_QKEY),
986 			 },
987 		},
988 	},
989 	[IB_QPS_RTR]   = {
990 		[IB_QPS_RESET] = { .valid = 1 },
991 		[IB_QPS_ERR] =   { .valid = 1 },
992 		[IB_QPS_RTS]   = {
993 			.valid = 1,
994 			.req_param = {
995 				[IB_QPT_UD]  = IB_QP_SQ_PSN,
996 				[IB_QPT_UC]  = IB_QP_SQ_PSN,
997 				[IB_QPT_RC]  = (IB_QP_TIMEOUT			|
998 						IB_QP_RETRY_CNT			|
999 						IB_QP_RNR_RETRY			|
1000 						IB_QP_SQ_PSN			|
1001 						IB_QP_MAX_QP_RD_ATOMIC),
1002 				[IB_QPT_XRC_INI] = (IB_QP_TIMEOUT		|
1003 						IB_QP_RETRY_CNT			|
1004 						IB_QP_RNR_RETRY			|
1005 						IB_QP_SQ_PSN			|
1006 						IB_QP_MAX_QP_RD_ATOMIC),
1007 				[IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT		|
1008 						IB_QP_SQ_PSN),
1009 				[IB_QPT_SMI] = IB_QP_SQ_PSN,
1010 				[IB_QPT_GSI] = IB_QP_SQ_PSN,
1011 			},
1012 			.opt_param = {
1013 				 [IB_QPT_UD]  = (IB_QP_CUR_STATE		|
1014 						 IB_QP_QKEY),
1015 				 [IB_QPT_UC]  = (IB_QP_CUR_STATE		|
1016 						 IB_QP_ALT_PATH			|
1017 						 IB_QP_ACCESS_FLAGS		|
1018 						 IB_QP_PATH_MIG_STATE),
1019 				 [IB_QPT_RC]  = (IB_QP_CUR_STATE		|
1020 						 IB_QP_ALT_PATH			|
1021 						 IB_QP_ACCESS_FLAGS		|
1022 						 IB_QP_MIN_RNR_TIMER		|
1023 						 IB_QP_PATH_MIG_STATE),
1024 				 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE		|
1025 						 IB_QP_ALT_PATH			|
1026 						 IB_QP_ACCESS_FLAGS		|
1027 						 IB_QP_PATH_MIG_STATE),
1028 				 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE		|
1029 						 IB_QP_ALT_PATH			|
1030 						 IB_QP_ACCESS_FLAGS		|
1031 						 IB_QP_MIN_RNR_TIMER		|
1032 						 IB_QP_PATH_MIG_STATE),
1033 				 [IB_QPT_SMI] = (IB_QP_CUR_STATE		|
1034 						 IB_QP_QKEY),
1035 				 [IB_QPT_GSI] = (IB_QP_CUR_STATE		|
1036 						 IB_QP_QKEY),
1037 				 [IB_QPT_RAW_PACKET] = IB_QP_RATE_LIMIT,
1038 			 }
1039 		}
1040 	},
1041 	[IB_QPS_RTS]   = {
1042 		[IB_QPS_RESET] = { .valid = 1 },
1043 		[IB_QPS_ERR] =   { .valid = 1 },
1044 		[IB_QPS_RTS]   = {
1045 			.valid = 1,
1046 			.opt_param = {
1047 				[IB_QPT_UD]  = (IB_QP_CUR_STATE			|
1048 						IB_QP_QKEY),
1049 				[IB_QPT_UC]  = (IB_QP_CUR_STATE			|
1050 						IB_QP_ACCESS_FLAGS		|
1051 						IB_QP_ALT_PATH			|
1052 						IB_QP_PATH_MIG_STATE),
1053 				[IB_QPT_RC]  = (IB_QP_CUR_STATE			|
1054 						IB_QP_ACCESS_FLAGS		|
1055 						IB_QP_ALT_PATH			|
1056 						IB_QP_PATH_MIG_STATE		|
1057 						IB_QP_MIN_RNR_TIMER),
1058 				[IB_QPT_XRC_INI] = (IB_QP_CUR_STATE		|
1059 						IB_QP_ACCESS_FLAGS		|
1060 						IB_QP_ALT_PATH			|
1061 						IB_QP_PATH_MIG_STATE),
1062 				[IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE		|
1063 						IB_QP_ACCESS_FLAGS		|
1064 						IB_QP_ALT_PATH			|
1065 						IB_QP_PATH_MIG_STATE		|
1066 						IB_QP_MIN_RNR_TIMER),
1067 				[IB_QPT_SMI] = (IB_QP_CUR_STATE			|
1068 						IB_QP_QKEY),
1069 				[IB_QPT_GSI] = (IB_QP_CUR_STATE			|
1070 						IB_QP_QKEY),
1071 				[IB_QPT_RAW_PACKET] = IB_QP_RATE_LIMIT,
1072 			}
1073 		},
1074 		[IB_QPS_SQD]   = {
1075 			.valid = 1,
1076 			.opt_param = {
1077 				[IB_QPT_UD]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1078 				[IB_QPT_UC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1079 				[IB_QPT_RC]  = IB_QP_EN_SQD_ASYNC_NOTIFY,
1080 				[IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1081 				[IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
1082 				[IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1083 				[IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
1084 			}
1085 		},
1086 	},
1087 	[IB_QPS_SQD]   = {
1088 		[IB_QPS_RESET] = { .valid = 1 },
1089 		[IB_QPS_ERR] =   { .valid = 1 },
1090 		[IB_QPS_RTS]   = {
1091 			.valid = 1,
1092 			.opt_param = {
1093 				[IB_QPT_UD]  = (IB_QP_CUR_STATE			|
1094 						IB_QP_QKEY),
1095 				[IB_QPT_UC]  = (IB_QP_CUR_STATE			|
1096 						IB_QP_ALT_PATH			|
1097 						IB_QP_ACCESS_FLAGS		|
1098 						IB_QP_PATH_MIG_STATE),
1099 				[IB_QPT_RC]  = (IB_QP_CUR_STATE			|
1100 						IB_QP_ALT_PATH			|
1101 						IB_QP_ACCESS_FLAGS		|
1102 						IB_QP_MIN_RNR_TIMER		|
1103 						IB_QP_PATH_MIG_STATE),
1104 				[IB_QPT_XRC_INI] = (IB_QP_CUR_STATE		|
1105 						IB_QP_ALT_PATH			|
1106 						IB_QP_ACCESS_FLAGS		|
1107 						IB_QP_PATH_MIG_STATE),
1108 				[IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE		|
1109 						IB_QP_ALT_PATH			|
1110 						IB_QP_ACCESS_FLAGS		|
1111 						IB_QP_MIN_RNR_TIMER		|
1112 						IB_QP_PATH_MIG_STATE),
1113 				[IB_QPT_SMI] = (IB_QP_CUR_STATE			|
1114 						IB_QP_QKEY),
1115 				[IB_QPT_GSI] = (IB_QP_CUR_STATE			|
1116 						IB_QP_QKEY),
1117 			}
1118 		},
1119 		[IB_QPS_SQD]   = {
1120 			.valid = 1,
1121 			.opt_param = {
1122 				[IB_QPT_UD]  = (IB_QP_PKEY_INDEX		|
1123 						IB_QP_QKEY),
1124 				[IB_QPT_UC]  = (IB_QP_AV			|
1125 						IB_QP_ALT_PATH			|
1126 						IB_QP_ACCESS_FLAGS		|
1127 						IB_QP_PKEY_INDEX		|
1128 						IB_QP_PATH_MIG_STATE),
1129 				[IB_QPT_RC]  = (IB_QP_PORT			|
1130 						IB_QP_AV			|
1131 						IB_QP_TIMEOUT			|
1132 						IB_QP_RETRY_CNT			|
1133 						IB_QP_RNR_RETRY			|
1134 						IB_QP_MAX_QP_RD_ATOMIC		|
1135 						IB_QP_MAX_DEST_RD_ATOMIC	|
1136 						IB_QP_ALT_PATH			|
1137 						IB_QP_ACCESS_FLAGS		|
1138 						IB_QP_PKEY_INDEX		|
1139 						IB_QP_MIN_RNR_TIMER		|
1140 						IB_QP_PATH_MIG_STATE),
1141 				[IB_QPT_XRC_INI] = (IB_QP_PORT			|
1142 						IB_QP_AV			|
1143 						IB_QP_TIMEOUT			|
1144 						IB_QP_RETRY_CNT			|
1145 						IB_QP_RNR_RETRY			|
1146 						IB_QP_MAX_QP_RD_ATOMIC		|
1147 						IB_QP_ALT_PATH			|
1148 						IB_QP_ACCESS_FLAGS		|
1149 						IB_QP_PKEY_INDEX		|
1150 						IB_QP_PATH_MIG_STATE),
1151 				[IB_QPT_XRC_TGT] = (IB_QP_PORT			|
1152 						IB_QP_AV			|
1153 						IB_QP_TIMEOUT			|
1154 						IB_QP_MAX_DEST_RD_ATOMIC	|
1155 						IB_QP_ALT_PATH			|
1156 						IB_QP_ACCESS_FLAGS		|
1157 						IB_QP_PKEY_INDEX		|
1158 						IB_QP_MIN_RNR_TIMER		|
1159 						IB_QP_PATH_MIG_STATE),
1160 				[IB_QPT_SMI] = (IB_QP_PKEY_INDEX		|
1161 						IB_QP_QKEY),
1162 				[IB_QPT_GSI] = (IB_QP_PKEY_INDEX		|
1163 						IB_QP_QKEY),
1164 			}
1165 		}
1166 	},
1167 	[IB_QPS_SQE]   = {
1168 		[IB_QPS_RESET] = { .valid = 1 },
1169 		[IB_QPS_ERR] =   { .valid = 1 },
1170 		[IB_QPS_RTS]   = {
1171 			.valid = 1,
1172 			.opt_param = {
1173 				[IB_QPT_UD]  = (IB_QP_CUR_STATE			|
1174 						IB_QP_QKEY),
1175 				[IB_QPT_UC]  = (IB_QP_CUR_STATE			|
1176 						IB_QP_ACCESS_FLAGS),
1177 				[IB_QPT_SMI] = (IB_QP_CUR_STATE			|
1178 						IB_QP_QKEY),
1179 				[IB_QPT_GSI] = (IB_QP_CUR_STATE			|
1180 						IB_QP_QKEY),
1181 			}
1182 		}
1183 	},
1184 	[IB_QPS_ERR] = {
1185 		[IB_QPS_RESET] = { .valid = 1 },
1186 		[IB_QPS_ERR] =   { .valid = 1 }
1187 	}
1188 };
1189 
1190 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1191 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
1192 		       enum rdma_link_layer ll)
1193 {
1194 	enum ib_qp_attr_mask req_param, opt_param;
1195 
1196 	if (cur_state  < 0 || cur_state  > IB_QPS_ERR ||
1197 	    next_state < 0 || next_state > IB_QPS_ERR)
1198 		return 0;
1199 
1200 	if (mask & IB_QP_CUR_STATE  &&
1201 	    cur_state != IB_QPS_RTR && cur_state != IB_QPS_RTS &&
1202 	    cur_state != IB_QPS_SQD && cur_state != IB_QPS_SQE)
1203 		return 0;
1204 
1205 	if (!qp_state_table[cur_state][next_state].valid)
1206 		return 0;
1207 
1208 	req_param = qp_state_table[cur_state][next_state].req_param[type];
1209 	opt_param = qp_state_table[cur_state][next_state].opt_param[type];
1210 
1211 	if ((mask & req_param) != req_param)
1212 		return 0;
1213 
1214 	if (mask & ~(req_param | opt_param | IB_QP_STATE))
1215 		return 0;
1216 
1217 	return 1;
1218 }
1219 EXPORT_SYMBOL(ib_modify_qp_is_ok);
1220 
1221 int ib_resolve_eth_dmac(struct ib_device *device,
1222 			struct rdma_ah_attr *ah_attr)
1223 {
1224 	int           ret = 0;
1225 	struct ib_global_route *grh;
1226 
1227 	if (!rdma_is_port_valid(device, rdma_ah_get_port_num(ah_attr)))
1228 		return -EINVAL;
1229 
1230 	if (ah_attr->type != RDMA_AH_ATTR_TYPE_ROCE)
1231 		return 0;
1232 
1233 	grh = rdma_ah_retrieve_grh(ah_attr);
1234 
1235 	if (rdma_link_local_addr((struct in6_addr *)grh->dgid.raw)) {
1236 		rdma_get_ll_mac((struct in6_addr *)grh->dgid.raw,
1237 				ah_attr->roce.dmac);
1238 	} else {
1239 		union ib_gid		sgid;
1240 		struct ib_gid_attr	sgid_attr;
1241 		int			ifindex;
1242 		int			hop_limit;
1243 
1244 		ret = ib_query_gid(device,
1245 				   rdma_ah_get_port_num(ah_attr),
1246 				   grh->sgid_index,
1247 				   &sgid, &sgid_attr);
1248 
1249 		if (ret || !sgid_attr.ndev) {
1250 			if (!ret)
1251 				ret = -ENXIO;
1252 			goto out;
1253 		}
1254 
1255 		ifindex = sgid_attr.ndev->ifindex;
1256 
1257 		ret =
1258 		rdma_addr_find_l2_eth_by_grh(&sgid, &grh->dgid,
1259 					     ah_attr->roce.dmac,
1260 					     NULL, &ifindex, &hop_limit);
1261 
1262 		dev_put(sgid_attr.ndev);
1263 
1264 		grh->hop_limit = hop_limit;
1265 	}
1266 out:
1267 	return ret;
1268 }
1269 EXPORT_SYMBOL(ib_resolve_eth_dmac);
1270 
1271 int ib_modify_qp(struct ib_qp *qp,
1272 		 struct ib_qp_attr *qp_attr,
1273 		 int qp_attr_mask)
1274 {
1275 
1276 	if (qp_attr_mask & IB_QP_AV) {
1277 		int ret;
1278 
1279 		ret = ib_resolve_eth_dmac(qp->device, &qp_attr->ah_attr);
1280 		if (ret)
1281 			return ret;
1282 	}
1283 
1284 	return ib_security_modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
1285 }
1286 EXPORT_SYMBOL(ib_modify_qp);
1287 
1288 int ib_query_qp(struct ib_qp *qp,
1289 		struct ib_qp_attr *qp_attr,
1290 		int qp_attr_mask,
1291 		struct ib_qp_init_attr *qp_init_attr)
1292 {
1293 	return qp->device->query_qp ?
1294 		qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
1295 		-ENOSYS;
1296 }
1297 EXPORT_SYMBOL(ib_query_qp);
1298 
1299 int ib_close_qp(struct ib_qp *qp)
1300 {
1301 	struct ib_qp *real_qp;
1302 	unsigned long flags;
1303 
1304 	real_qp = qp->real_qp;
1305 	if (real_qp == qp)
1306 		return -EINVAL;
1307 
1308 	spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
1309 	list_del(&qp->open_list);
1310 	spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
1311 
1312 	atomic_dec(&real_qp->usecnt);
1313 	ib_close_shared_qp_security(qp->qp_sec);
1314 	kfree(qp);
1315 
1316 	return 0;
1317 }
1318 EXPORT_SYMBOL(ib_close_qp);
1319 
1320 static int __ib_destroy_shared_qp(struct ib_qp *qp)
1321 {
1322 	struct ib_xrcd *xrcd;
1323 	struct ib_qp *real_qp;
1324 	int ret;
1325 
1326 	real_qp = qp->real_qp;
1327 	xrcd = real_qp->xrcd;
1328 
1329 	mutex_lock(&xrcd->tgt_qp_mutex);
1330 	ib_close_qp(qp);
1331 	if (atomic_read(&real_qp->usecnt) == 0)
1332 		list_del(&real_qp->xrcd_list);
1333 	else
1334 		real_qp = NULL;
1335 	mutex_unlock(&xrcd->tgt_qp_mutex);
1336 
1337 	if (real_qp) {
1338 		ret = ib_destroy_qp(real_qp);
1339 		if (!ret)
1340 			atomic_dec(&xrcd->usecnt);
1341 		else
1342 			__ib_insert_xrcd_qp(xrcd, real_qp);
1343 	}
1344 
1345 	return 0;
1346 }
1347 
1348 int ib_destroy_qp(struct ib_qp *qp)
1349 {
1350 	struct ib_pd *pd;
1351 	struct ib_cq *scq, *rcq;
1352 	struct ib_srq *srq;
1353 	struct ib_rwq_ind_table *ind_tbl;
1354 	struct ib_qp_security *sec;
1355 	int ret;
1356 
1357 	WARN_ON_ONCE(qp->mrs_used > 0);
1358 
1359 	if (atomic_read(&qp->usecnt))
1360 		return -EBUSY;
1361 
1362 	if (qp->real_qp != qp)
1363 		return __ib_destroy_shared_qp(qp);
1364 
1365 	pd   = qp->pd;
1366 	scq  = qp->send_cq;
1367 	rcq  = qp->recv_cq;
1368 	srq  = qp->srq;
1369 	ind_tbl = qp->rwq_ind_tbl;
1370 	sec  = qp->qp_sec;
1371 	if (sec)
1372 		ib_destroy_qp_security_begin(sec);
1373 
1374 	if (!qp->uobject)
1375 		rdma_rw_cleanup_mrs(qp);
1376 
1377 	ret = qp->device->destroy_qp(qp);
1378 	if (!ret) {
1379 		if (pd)
1380 			atomic_dec(&pd->usecnt);
1381 		if (scq)
1382 			atomic_dec(&scq->usecnt);
1383 		if (rcq)
1384 			atomic_dec(&rcq->usecnt);
1385 		if (srq)
1386 			atomic_dec(&srq->usecnt);
1387 		if (ind_tbl)
1388 			atomic_dec(&ind_tbl->usecnt);
1389 		if (sec)
1390 			ib_destroy_qp_security_end(sec);
1391 	} else {
1392 		if (sec)
1393 			ib_destroy_qp_security_abort(sec);
1394 	}
1395 
1396 	return ret;
1397 }
1398 EXPORT_SYMBOL(ib_destroy_qp);
1399 
1400 /* Completion queues */
1401 
1402 struct ib_cq *ib_create_cq(struct ib_device *device,
1403 			   ib_comp_handler comp_handler,
1404 			   void (*event_handler)(struct ib_event *, void *),
1405 			   void *cq_context,
1406 			   const struct ib_cq_init_attr *cq_attr)
1407 {
1408 	struct ib_cq *cq;
1409 
1410 	cq = device->create_cq(device, cq_attr, NULL, NULL);
1411 
1412 	if (!IS_ERR(cq)) {
1413 		cq->device        = device;
1414 		cq->uobject       = NULL;
1415 		cq->comp_handler  = comp_handler;
1416 		cq->event_handler = event_handler;
1417 		cq->cq_context    = cq_context;
1418 		atomic_set(&cq->usecnt, 0);
1419 	}
1420 
1421 	return cq;
1422 }
1423 EXPORT_SYMBOL(ib_create_cq);
1424 
1425 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
1426 {
1427 	return cq->device->modify_cq ?
1428 		cq->device->modify_cq(cq, cq_count, cq_period) : -ENOSYS;
1429 }
1430 EXPORT_SYMBOL(ib_modify_cq);
1431 
1432 int ib_destroy_cq(struct ib_cq *cq)
1433 {
1434 	if (atomic_read(&cq->usecnt))
1435 		return -EBUSY;
1436 
1437 	return cq->device->destroy_cq(cq);
1438 }
1439 EXPORT_SYMBOL(ib_destroy_cq);
1440 
1441 int ib_resize_cq(struct ib_cq *cq, int cqe)
1442 {
1443 	return cq->device->resize_cq ?
1444 		cq->device->resize_cq(cq, cqe, NULL) : -ENOSYS;
1445 }
1446 EXPORT_SYMBOL(ib_resize_cq);
1447 
1448 /* Memory regions */
1449 
1450 int ib_dereg_mr(struct ib_mr *mr)
1451 {
1452 	struct ib_pd *pd = mr->pd;
1453 	int ret;
1454 
1455 	ret = mr->device->dereg_mr(mr);
1456 	if (!ret)
1457 		atomic_dec(&pd->usecnt);
1458 
1459 	return ret;
1460 }
1461 EXPORT_SYMBOL(ib_dereg_mr);
1462 
1463 /**
1464  * ib_alloc_mr() - Allocates a memory region
1465  * @pd:            protection domain associated with the region
1466  * @mr_type:       memory region type
1467  * @max_num_sg:    maximum sg entries available for registration.
1468  *
1469  * Notes:
1470  * Memory registeration page/sg lists must not exceed max_num_sg.
1471  * For mr_type IB_MR_TYPE_MEM_REG, the total length cannot exceed
1472  * max_num_sg * used_page_size.
1473  *
1474  */
1475 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
1476 			  enum ib_mr_type mr_type,
1477 			  u32 max_num_sg)
1478 {
1479 	struct ib_mr *mr;
1480 
1481 	if (!pd->device->alloc_mr)
1482 		return ERR_PTR(-ENOSYS);
1483 
1484 	mr = pd->device->alloc_mr(pd, mr_type, max_num_sg);
1485 	if (!IS_ERR(mr)) {
1486 		mr->device  = pd->device;
1487 		mr->pd      = pd;
1488 		mr->uobject = NULL;
1489 		atomic_inc(&pd->usecnt);
1490 		mr->need_inval = false;
1491 	}
1492 
1493 	return mr;
1494 }
1495 EXPORT_SYMBOL(ib_alloc_mr);
1496 
1497 /* "Fast" memory regions */
1498 
1499 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1500 			    int mr_access_flags,
1501 			    struct ib_fmr_attr *fmr_attr)
1502 {
1503 	struct ib_fmr *fmr;
1504 
1505 	if (!pd->device->alloc_fmr)
1506 		return ERR_PTR(-ENOSYS);
1507 
1508 	fmr = pd->device->alloc_fmr(pd, mr_access_flags, fmr_attr);
1509 	if (!IS_ERR(fmr)) {
1510 		fmr->device = pd->device;
1511 		fmr->pd     = pd;
1512 		atomic_inc(&pd->usecnt);
1513 	}
1514 
1515 	return fmr;
1516 }
1517 EXPORT_SYMBOL(ib_alloc_fmr);
1518 
1519 int ib_unmap_fmr(struct list_head *fmr_list)
1520 {
1521 	struct ib_fmr *fmr;
1522 
1523 	if (list_empty(fmr_list))
1524 		return 0;
1525 
1526 	fmr = list_entry(fmr_list->next, struct ib_fmr, list);
1527 	return fmr->device->unmap_fmr(fmr_list);
1528 }
1529 EXPORT_SYMBOL(ib_unmap_fmr);
1530 
1531 int ib_dealloc_fmr(struct ib_fmr *fmr)
1532 {
1533 	struct ib_pd *pd;
1534 	int ret;
1535 
1536 	pd = fmr->pd;
1537 	ret = fmr->device->dealloc_fmr(fmr);
1538 	if (!ret)
1539 		atomic_dec(&pd->usecnt);
1540 
1541 	return ret;
1542 }
1543 EXPORT_SYMBOL(ib_dealloc_fmr);
1544 
1545 /* Multicast groups */
1546 
1547 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1548 {
1549 	int ret;
1550 
1551 	if (!qp->device->attach_mcast)
1552 		return -ENOSYS;
1553 	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
1554 	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
1555 	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
1556 		return -EINVAL;
1557 
1558 	ret = qp->device->attach_mcast(qp, gid, lid);
1559 	if (!ret)
1560 		atomic_inc(&qp->usecnt);
1561 	return ret;
1562 }
1563 EXPORT_SYMBOL(ib_attach_mcast);
1564 
1565 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1566 {
1567 	int ret;
1568 
1569 	if (!qp->device->detach_mcast)
1570 		return -ENOSYS;
1571 	if (gid->raw[0] != 0xff || qp->qp_type != IB_QPT_UD ||
1572 	    lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
1573 	    lid == be16_to_cpu(IB_LID_PERMISSIVE))
1574 		return -EINVAL;
1575 
1576 	ret = qp->device->detach_mcast(qp, gid, lid);
1577 	if (!ret)
1578 		atomic_dec(&qp->usecnt);
1579 	return ret;
1580 }
1581 EXPORT_SYMBOL(ib_detach_mcast);
1582 
1583 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1584 {
1585 	struct ib_xrcd *xrcd;
1586 
1587 	if (!device->alloc_xrcd)
1588 		return ERR_PTR(-ENOSYS);
1589 
1590 	xrcd = device->alloc_xrcd(device, NULL, NULL);
1591 	if (!IS_ERR(xrcd)) {
1592 		xrcd->device = device;
1593 		xrcd->inode = NULL;
1594 		atomic_set(&xrcd->usecnt, 0);
1595 		mutex_init(&xrcd->tgt_qp_mutex);
1596 		INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1597 	}
1598 
1599 	return xrcd;
1600 }
1601 EXPORT_SYMBOL(ib_alloc_xrcd);
1602 
1603 int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1604 {
1605 	struct ib_qp *qp;
1606 	int ret;
1607 
1608 	if (atomic_read(&xrcd->usecnt))
1609 		return -EBUSY;
1610 
1611 	while (!list_empty(&xrcd->tgt_qp_list)) {
1612 		qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1613 		ret = ib_destroy_qp(qp);
1614 		if (ret)
1615 			return ret;
1616 	}
1617 
1618 	return xrcd->device->dealloc_xrcd(xrcd);
1619 }
1620 EXPORT_SYMBOL(ib_dealloc_xrcd);
1621 
1622 /**
1623  * ib_create_wq - Creates a WQ associated with the specified protection
1624  * domain.
1625  * @pd: The protection domain associated with the WQ.
1626  * @wq_init_attr: A list of initial attributes required to create the
1627  * WQ. If WQ creation succeeds, then the attributes are updated to
1628  * the actual capabilities of the created WQ.
1629  *
1630  * wq_init_attr->max_wr and wq_init_attr->max_sge determine
1631  * the requested size of the WQ, and set to the actual values allocated
1632  * on return.
1633  * If ib_create_wq() succeeds, then max_wr and max_sge will always be
1634  * at least as large as the requested values.
1635  */
1636 struct ib_wq *ib_create_wq(struct ib_pd *pd,
1637 			   struct ib_wq_init_attr *wq_attr)
1638 {
1639 	struct ib_wq *wq;
1640 
1641 	if (!pd->device->create_wq)
1642 		return ERR_PTR(-ENOSYS);
1643 
1644 	wq = pd->device->create_wq(pd, wq_attr, NULL);
1645 	if (!IS_ERR(wq)) {
1646 		wq->event_handler = wq_attr->event_handler;
1647 		wq->wq_context = wq_attr->wq_context;
1648 		wq->wq_type = wq_attr->wq_type;
1649 		wq->cq = wq_attr->cq;
1650 		wq->device = pd->device;
1651 		wq->pd = pd;
1652 		wq->uobject = NULL;
1653 		atomic_inc(&pd->usecnt);
1654 		atomic_inc(&wq_attr->cq->usecnt);
1655 		atomic_set(&wq->usecnt, 0);
1656 	}
1657 	return wq;
1658 }
1659 EXPORT_SYMBOL(ib_create_wq);
1660 
1661 /**
1662  * ib_destroy_wq - Destroys the specified WQ.
1663  * @wq: The WQ to destroy.
1664  */
1665 int ib_destroy_wq(struct ib_wq *wq)
1666 {
1667 	int err;
1668 	struct ib_cq *cq = wq->cq;
1669 	struct ib_pd *pd = wq->pd;
1670 
1671 	if (atomic_read(&wq->usecnt))
1672 		return -EBUSY;
1673 
1674 	err = wq->device->destroy_wq(wq);
1675 	if (!err) {
1676 		atomic_dec(&pd->usecnt);
1677 		atomic_dec(&cq->usecnt);
1678 	}
1679 	return err;
1680 }
1681 EXPORT_SYMBOL(ib_destroy_wq);
1682 
1683 /**
1684  * ib_modify_wq - Modifies the specified WQ.
1685  * @wq: The WQ to modify.
1686  * @wq_attr: On input, specifies the WQ attributes to modify.
1687  * @wq_attr_mask: A bit-mask used to specify which attributes of the WQ
1688  *   are being modified.
1689  * On output, the current values of selected WQ attributes are returned.
1690  */
1691 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1692 		 u32 wq_attr_mask)
1693 {
1694 	int err;
1695 
1696 	if (!wq->device->modify_wq)
1697 		return -ENOSYS;
1698 
1699 	err = wq->device->modify_wq(wq, wq_attr, wq_attr_mask, NULL);
1700 	return err;
1701 }
1702 EXPORT_SYMBOL(ib_modify_wq);
1703 
1704 /*
1705  * ib_create_rwq_ind_table - Creates a RQ Indirection Table.
1706  * @device: The device on which to create the rwq indirection table.
1707  * @ib_rwq_ind_table_init_attr: A list of initial attributes required to
1708  * create the Indirection Table.
1709  *
1710  * Note: The life time of ib_rwq_ind_table_init_attr->ind_tbl is not less
1711  *	than the created ib_rwq_ind_table object and the caller is responsible
1712  *	for its memory allocation/free.
1713  */
1714 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
1715 						 struct ib_rwq_ind_table_init_attr *init_attr)
1716 {
1717 	struct ib_rwq_ind_table *rwq_ind_table;
1718 	int i;
1719 	u32 table_size;
1720 
1721 	if (!device->create_rwq_ind_table)
1722 		return ERR_PTR(-ENOSYS);
1723 
1724 	table_size = (1 << init_attr->log_ind_tbl_size);
1725 	rwq_ind_table = device->create_rwq_ind_table(device,
1726 				init_attr, NULL);
1727 	if (IS_ERR(rwq_ind_table))
1728 		return rwq_ind_table;
1729 
1730 	rwq_ind_table->ind_tbl = init_attr->ind_tbl;
1731 	rwq_ind_table->log_ind_tbl_size = init_attr->log_ind_tbl_size;
1732 	rwq_ind_table->device = device;
1733 	rwq_ind_table->uobject = NULL;
1734 	atomic_set(&rwq_ind_table->usecnt, 0);
1735 
1736 	for (i = 0; i < table_size; i++)
1737 		atomic_inc(&rwq_ind_table->ind_tbl[i]->usecnt);
1738 
1739 	return rwq_ind_table;
1740 }
1741 EXPORT_SYMBOL(ib_create_rwq_ind_table);
1742 
1743 /*
1744  * ib_destroy_rwq_ind_table - Destroys the specified Indirection Table.
1745  * @wq_ind_table: The Indirection Table to destroy.
1746 */
1747 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *rwq_ind_table)
1748 {
1749 	int err, i;
1750 	u32 table_size = (1 << rwq_ind_table->log_ind_tbl_size);
1751 	struct ib_wq **ind_tbl = rwq_ind_table->ind_tbl;
1752 
1753 	if (atomic_read(&rwq_ind_table->usecnt))
1754 		return -EBUSY;
1755 
1756 	err = rwq_ind_table->device->destroy_rwq_ind_table(rwq_ind_table);
1757 	if (!err) {
1758 		for (i = 0; i < table_size; i++)
1759 			atomic_dec(&ind_tbl[i]->usecnt);
1760 	}
1761 
1762 	return err;
1763 }
1764 EXPORT_SYMBOL(ib_destroy_rwq_ind_table);
1765 
1766 struct ib_flow *ib_create_flow(struct ib_qp *qp,
1767 			       struct ib_flow_attr *flow_attr,
1768 			       int domain)
1769 {
1770 	struct ib_flow *flow_id;
1771 	if (!qp->device->create_flow)
1772 		return ERR_PTR(-ENOSYS);
1773 
1774 	flow_id = qp->device->create_flow(qp, flow_attr, domain);
1775 	if (!IS_ERR(flow_id)) {
1776 		atomic_inc(&qp->usecnt);
1777 		flow_id->qp = qp;
1778 	}
1779 	return flow_id;
1780 }
1781 EXPORT_SYMBOL(ib_create_flow);
1782 
1783 int ib_destroy_flow(struct ib_flow *flow_id)
1784 {
1785 	int err;
1786 	struct ib_qp *qp = flow_id->qp;
1787 
1788 	err = qp->device->destroy_flow(flow_id);
1789 	if (!err)
1790 		atomic_dec(&qp->usecnt);
1791 	return err;
1792 }
1793 EXPORT_SYMBOL(ib_destroy_flow);
1794 
1795 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
1796 		       struct ib_mr_status *mr_status)
1797 {
1798 	return mr->device->check_mr_status ?
1799 		mr->device->check_mr_status(mr, check_mask, mr_status) : -ENOSYS;
1800 }
1801 EXPORT_SYMBOL(ib_check_mr_status);
1802 
1803 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
1804 			 int state)
1805 {
1806 	if (!device->set_vf_link_state)
1807 		return -ENOSYS;
1808 
1809 	return device->set_vf_link_state(device, vf, port, state);
1810 }
1811 EXPORT_SYMBOL(ib_set_vf_link_state);
1812 
1813 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
1814 		     struct ifla_vf_info *info)
1815 {
1816 	if (!device->get_vf_config)
1817 		return -ENOSYS;
1818 
1819 	return device->get_vf_config(device, vf, port, info);
1820 }
1821 EXPORT_SYMBOL(ib_get_vf_config);
1822 
1823 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
1824 		    struct ifla_vf_stats *stats)
1825 {
1826 	if (!device->get_vf_stats)
1827 		return -ENOSYS;
1828 
1829 	return device->get_vf_stats(device, vf, port, stats);
1830 }
1831 EXPORT_SYMBOL(ib_get_vf_stats);
1832 
1833 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
1834 		   int type)
1835 {
1836 	if (!device->set_vf_guid)
1837 		return -ENOSYS;
1838 
1839 	return device->set_vf_guid(device, vf, port, guid, type);
1840 }
1841 EXPORT_SYMBOL(ib_set_vf_guid);
1842 
1843 /**
1844  * ib_map_mr_sg() - Map the largest prefix of a dma mapped SG list
1845  *     and set it the memory region.
1846  * @mr:            memory region
1847  * @sg:            dma mapped scatterlist
1848  * @sg_nents:      number of entries in sg
1849  * @sg_offset:     offset in bytes into sg
1850  * @page_size:     page vector desired page size
1851  *
1852  * Constraints:
1853  * - The first sg element is allowed to have an offset.
1854  * - Each sg element must either be aligned to page_size or virtually
1855  *   contiguous to the previous element. In case an sg element has a
1856  *   non-contiguous offset, the mapping prefix will not include it.
1857  * - The last sg element is allowed to have length less than page_size.
1858  * - If sg_nents total byte length exceeds the mr max_num_sge * page_size
1859  *   then only max_num_sg entries will be mapped.
1860  * - If the MR was allocated with type IB_MR_TYPE_SG_GAPS, none of these
1861  *   constraints holds and the page_size argument is ignored.
1862  *
1863  * Returns the number of sg elements that were mapped to the memory region.
1864  *
1865  * After this completes successfully, the  memory region
1866  * is ready for registration.
1867  */
1868 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
1869 		 unsigned int *sg_offset, unsigned int page_size)
1870 {
1871 	if (unlikely(!mr->device->map_mr_sg))
1872 		return -ENOSYS;
1873 
1874 	mr->page_size = page_size;
1875 
1876 	return mr->device->map_mr_sg(mr, sg, sg_nents, sg_offset);
1877 }
1878 EXPORT_SYMBOL(ib_map_mr_sg);
1879 
1880 /**
1881  * ib_sg_to_pages() - Convert the largest prefix of a sg list
1882  *     to a page vector
1883  * @mr:            memory region
1884  * @sgl:           dma mapped scatterlist
1885  * @sg_nents:      number of entries in sg
1886  * @sg_offset_p:   IN:  start offset in bytes into sg
1887  *                 OUT: offset in bytes for element n of the sg of the first
1888  *                      byte that has not been processed where n is the return
1889  *                      value of this function.
1890  * @set_page:      driver page assignment function pointer
1891  *
1892  * Core service helper for drivers to convert the largest
1893  * prefix of given sg list to a page vector. The sg list
1894  * prefix converted is the prefix that meet the requirements
1895  * of ib_map_mr_sg.
1896  *
1897  * Returns the number of sg elements that were assigned to
1898  * a page vector.
1899  */
1900 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
1901 		unsigned int *sg_offset_p, int (*set_page)(struct ib_mr *, u64))
1902 {
1903 	struct scatterlist *sg;
1904 	u64 last_end_dma_addr = 0;
1905 	unsigned int sg_offset = sg_offset_p ? *sg_offset_p : 0;
1906 	unsigned int last_page_off = 0;
1907 	u64 page_mask = ~((u64)mr->page_size - 1);
1908 	int i, ret;
1909 
1910 	if (unlikely(sg_nents <= 0 || sg_offset > sg_dma_len(&sgl[0])))
1911 		return -EINVAL;
1912 
1913 	mr->iova = sg_dma_address(&sgl[0]) + sg_offset;
1914 	mr->length = 0;
1915 
1916 	for_each_sg(sgl, sg, sg_nents, i) {
1917 		u64 dma_addr = sg_dma_address(sg) + sg_offset;
1918 		u64 prev_addr = dma_addr;
1919 		unsigned int dma_len = sg_dma_len(sg) - sg_offset;
1920 		u64 end_dma_addr = dma_addr + dma_len;
1921 		u64 page_addr = dma_addr & page_mask;
1922 
1923 		/*
1924 		 * For the second and later elements, check whether either the
1925 		 * end of element i-1 or the start of element i is not aligned
1926 		 * on a page boundary.
1927 		 */
1928 		if (i && (last_page_off != 0 || page_addr != dma_addr)) {
1929 			/* Stop mapping if there is a gap. */
1930 			if (last_end_dma_addr != dma_addr)
1931 				break;
1932 
1933 			/*
1934 			 * Coalesce this element with the last. If it is small
1935 			 * enough just update mr->length. Otherwise start
1936 			 * mapping from the next page.
1937 			 */
1938 			goto next_page;
1939 		}
1940 
1941 		do {
1942 			ret = set_page(mr, page_addr);
1943 			if (unlikely(ret < 0)) {
1944 				sg_offset = prev_addr - sg_dma_address(sg);
1945 				mr->length += prev_addr - dma_addr;
1946 				if (sg_offset_p)
1947 					*sg_offset_p = sg_offset;
1948 				return i || sg_offset ? i : ret;
1949 			}
1950 			prev_addr = page_addr;
1951 next_page:
1952 			page_addr += mr->page_size;
1953 		} while (page_addr < end_dma_addr);
1954 
1955 		mr->length += dma_len;
1956 		last_end_dma_addr = end_dma_addr;
1957 		last_page_off = end_dma_addr & ~page_mask;
1958 
1959 		sg_offset = 0;
1960 	}
1961 
1962 	if (sg_offset_p)
1963 		*sg_offset_p = 0;
1964 	return i;
1965 }
1966 EXPORT_SYMBOL(ib_sg_to_pages);
1967 
1968 struct ib_drain_cqe {
1969 	struct ib_cqe cqe;
1970 	struct completion done;
1971 };
1972 
1973 static void ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
1974 {
1975 	struct ib_drain_cqe *cqe = container_of(wc->wr_cqe, struct ib_drain_cqe,
1976 						cqe);
1977 
1978 	complete(&cqe->done);
1979 }
1980 
1981 /*
1982  * Post a WR and block until its completion is reaped for the SQ.
1983  */
1984 static void __ib_drain_sq(struct ib_qp *qp)
1985 {
1986 	struct ib_cq *cq = qp->send_cq;
1987 	struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1988 	struct ib_drain_cqe sdrain;
1989 	struct ib_send_wr swr = {}, *bad_swr;
1990 	int ret;
1991 
1992 	swr.wr_cqe = &sdrain.cqe;
1993 	sdrain.cqe.done = ib_drain_qp_done;
1994 	init_completion(&sdrain.done);
1995 
1996 	ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
1997 	if (ret) {
1998 		WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
1999 		return;
2000 	}
2001 
2002 	ret = ib_post_send(qp, &swr, &bad_swr);
2003 	if (ret) {
2004 		WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
2005 		return;
2006 	}
2007 
2008 	if (cq->poll_ctx == IB_POLL_DIRECT)
2009 		while (wait_for_completion_timeout(&sdrain.done, HZ / 10) <= 0)
2010 			ib_process_cq_direct(cq, -1);
2011 	else
2012 		wait_for_completion(&sdrain.done);
2013 }
2014 
2015 /*
2016  * Post a WR and block until its completion is reaped for the RQ.
2017  */
2018 static void __ib_drain_rq(struct ib_qp *qp)
2019 {
2020 	struct ib_cq *cq = qp->recv_cq;
2021 	struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
2022 	struct ib_drain_cqe rdrain;
2023 	struct ib_recv_wr rwr = {}, *bad_rwr;
2024 	int ret;
2025 
2026 	rwr.wr_cqe = &rdrain.cqe;
2027 	rdrain.cqe.done = ib_drain_qp_done;
2028 	init_completion(&rdrain.done);
2029 
2030 	ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
2031 	if (ret) {
2032 		WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
2033 		return;
2034 	}
2035 
2036 	ret = ib_post_recv(qp, &rwr, &bad_rwr);
2037 	if (ret) {
2038 		WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
2039 		return;
2040 	}
2041 
2042 	if (cq->poll_ctx == IB_POLL_DIRECT)
2043 		while (wait_for_completion_timeout(&rdrain.done, HZ / 10) <= 0)
2044 			ib_process_cq_direct(cq, -1);
2045 	else
2046 		wait_for_completion(&rdrain.done);
2047 }
2048 
2049 /**
2050  * ib_drain_sq() - Block until all SQ CQEs have been consumed by the
2051  *		   application.
2052  * @qp:            queue pair to drain
2053  *
2054  * If the device has a provider-specific drain function, then
2055  * call that.  Otherwise call the generic drain function
2056  * __ib_drain_sq().
2057  *
2058  * The caller must:
2059  *
2060  * ensure there is room in the CQ and SQ for the drain work request and
2061  * completion.
2062  *
2063  * allocate the CQ using ib_alloc_cq().
2064  *
2065  * ensure that there are no other contexts that are posting WRs concurrently.
2066  * Otherwise the drain is not guaranteed.
2067  */
2068 void ib_drain_sq(struct ib_qp *qp)
2069 {
2070 	if (qp->device->drain_sq)
2071 		qp->device->drain_sq(qp);
2072 	else
2073 		__ib_drain_sq(qp);
2074 }
2075 EXPORT_SYMBOL(ib_drain_sq);
2076 
2077 /**
2078  * ib_drain_rq() - Block until all RQ CQEs have been consumed by the
2079  *		   application.
2080  * @qp:            queue pair to drain
2081  *
2082  * If the device has a provider-specific drain function, then
2083  * call that.  Otherwise call the generic drain function
2084  * __ib_drain_rq().
2085  *
2086  * The caller must:
2087  *
2088  * ensure there is room in the CQ and RQ for the drain work request and
2089  * completion.
2090  *
2091  * allocate the CQ using ib_alloc_cq().
2092  *
2093  * ensure that there are no other contexts that are posting WRs concurrently.
2094  * Otherwise the drain is not guaranteed.
2095  */
2096 void ib_drain_rq(struct ib_qp *qp)
2097 {
2098 	if (qp->device->drain_rq)
2099 		qp->device->drain_rq(qp);
2100 	else
2101 		__ib_drain_rq(qp);
2102 }
2103 EXPORT_SYMBOL(ib_drain_rq);
2104 
2105 /**
2106  * ib_drain_qp() - Block until all CQEs have been consumed by the
2107  *		   application on both the RQ and SQ.
2108  * @qp:            queue pair to drain
2109  *
2110  * The caller must:
2111  *
2112  * ensure there is room in the CQ(s), SQ, and RQ for drain work requests
2113  * and completions.
2114  *
2115  * allocate the CQs using ib_alloc_cq().
2116  *
2117  * ensure that there are no other contexts that are posting WRs concurrently.
2118  * Otherwise the drain is not guaranteed.
2119  */
2120 void ib_drain_qp(struct ib_qp *qp)
2121 {
2122 	ib_drain_sq(qp);
2123 	if (!qp->srq)
2124 		ib_drain_rq(qp);
2125 }
2126 EXPORT_SYMBOL(ib_drain_qp);
2127