1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3 *
4 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved.
5 * Copyright (c) 2004 Infinicon Corporation. All rights reserved.
6 * Copyright (c) 2004 Intel Corporation. All rights reserved.
7 * Copyright (c) 2004 Topspin Corporation. All rights reserved.
8 * Copyright (c) 2004 Voltaire Corporation. All rights reserved.
9 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
10 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
11 *
12 * This software is available to you under a choice of one of two
13 * licenses. You may choose to be licensed under the terms of the GNU
14 * General Public License (GPL) Version 2, available from the file
15 * COPYING in the main directory of this source tree, or the
16 * OpenIB.org BSD license below:
17 *
18 * Redistribution and use in source and binary forms, with or
19 * without modification, are permitted provided that the following
20 * conditions are met:
21 *
22 * - Redistributions of source code must retain the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer.
25 *
26 * - Redistributions in binary form must reproduce the above
27 * copyright notice, this list of conditions and the following
28 * disclaimer in the documentation and/or other materials
29 * provided with the distribution.
30 *
31 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
32 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
33 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
34 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
35 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
36 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
37 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38 * SOFTWARE.
39 */
40
41 #include <sys/cdefs.h>
42 __FBSDID("$FreeBSD$");
43
44 #include <linux/errno.h>
45 #include <linux/err.h>
46 #include <linux/string.h>
47 #include <linux/slab.h>
48 #include <linux/in.h>
49 #include <linux/in6.h>
50
51 #include <rdma/ib_verbs.h>
52 #include <rdma/ib_cache.h>
53 #include <rdma/ib_addr.h>
54
55 #include <netinet/ip.h>
56 #include <netinet/ip6.h>
57
58 #include <machine/in_cksum.h>
59
60 #include "core_priv.h"
61
62 static const char * const ib_events[] = {
63 [IB_EVENT_CQ_ERR] = "CQ error",
64 [IB_EVENT_QP_FATAL] = "QP fatal error",
65 [IB_EVENT_QP_REQ_ERR] = "QP request error",
66 [IB_EVENT_QP_ACCESS_ERR] = "QP access error",
67 [IB_EVENT_COMM_EST] = "communication established",
68 [IB_EVENT_SQ_DRAINED] = "send queue drained",
69 [IB_EVENT_PATH_MIG] = "path migration successful",
70 [IB_EVENT_PATH_MIG_ERR] = "path migration error",
71 [IB_EVENT_DEVICE_FATAL] = "device fatal error",
72 [IB_EVENT_PORT_ACTIVE] = "port active",
73 [IB_EVENT_PORT_ERR] = "port error",
74 [IB_EVENT_LID_CHANGE] = "LID change",
75 [IB_EVENT_PKEY_CHANGE] = "P_key change",
76 [IB_EVENT_SM_CHANGE] = "SM change",
77 [IB_EVENT_SRQ_ERR] = "SRQ error",
78 [IB_EVENT_SRQ_LIMIT_REACHED] = "SRQ limit reached",
79 [IB_EVENT_QP_LAST_WQE_REACHED] = "last WQE reached",
80 [IB_EVENT_CLIENT_REREGISTER] = "client reregister",
81 [IB_EVENT_GID_CHANGE] = "GID changed",
82 };
83
ib_event_msg(enum ib_event_type event)84 const char *__attribute_const__ ib_event_msg(enum ib_event_type event)
85 {
86 size_t index = event;
87
88 return (index < ARRAY_SIZE(ib_events) && ib_events[index]) ?
89 ib_events[index] : "unrecognized event";
90 }
91 EXPORT_SYMBOL(ib_event_msg);
92
93 static const char * const wc_statuses[] = {
94 [IB_WC_SUCCESS] = "success",
95 [IB_WC_LOC_LEN_ERR] = "local length error",
96 [IB_WC_LOC_QP_OP_ERR] = "local QP operation error",
97 [IB_WC_LOC_EEC_OP_ERR] = "local EE context operation error",
98 [IB_WC_LOC_PROT_ERR] = "local protection error",
99 [IB_WC_WR_FLUSH_ERR] = "WR flushed",
100 [IB_WC_MW_BIND_ERR] = "memory management operation error",
101 [IB_WC_BAD_RESP_ERR] = "bad response error",
102 [IB_WC_LOC_ACCESS_ERR] = "local access error",
103 [IB_WC_REM_INV_REQ_ERR] = "invalid request error",
104 [IB_WC_REM_ACCESS_ERR] = "remote access error",
105 [IB_WC_REM_OP_ERR] = "remote operation error",
106 [IB_WC_RETRY_EXC_ERR] = "transport retry counter exceeded",
107 [IB_WC_RNR_RETRY_EXC_ERR] = "RNR retry counter exceeded",
108 [IB_WC_LOC_RDD_VIOL_ERR] = "local RDD violation error",
109 [IB_WC_REM_INV_RD_REQ_ERR] = "remote invalid RD request",
110 [IB_WC_REM_ABORT_ERR] = "operation aborted",
111 [IB_WC_INV_EECN_ERR] = "invalid EE context number",
112 [IB_WC_INV_EEC_STATE_ERR] = "invalid EE context state",
113 [IB_WC_FATAL_ERR] = "fatal error",
114 [IB_WC_RESP_TIMEOUT_ERR] = "response timeout error",
115 [IB_WC_GENERAL_ERR] = "general error",
116 };
117
ib_wc_status_msg(enum ib_wc_status status)118 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status)
119 {
120 size_t index = status;
121
122 return (index < ARRAY_SIZE(wc_statuses) && wc_statuses[index]) ?
123 wc_statuses[index] : "unrecognized status";
124 }
125 EXPORT_SYMBOL(ib_wc_status_msg);
126
ib_rate_to_mult(enum ib_rate rate)127 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate)
128 {
129 switch (rate) {
130 case IB_RATE_2_5_GBPS: return 1;
131 case IB_RATE_5_GBPS: return 2;
132 case IB_RATE_10_GBPS: return 4;
133 case IB_RATE_20_GBPS: return 8;
134 case IB_RATE_30_GBPS: return 12;
135 case IB_RATE_40_GBPS: return 16;
136 case IB_RATE_60_GBPS: return 24;
137 case IB_RATE_80_GBPS: return 32;
138 case IB_RATE_120_GBPS: return 48;
139 case IB_RATE_14_GBPS: return 6;
140 case IB_RATE_56_GBPS: return 22;
141 case IB_RATE_112_GBPS: return 45;
142 case IB_RATE_168_GBPS: return 67;
143 case IB_RATE_25_GBPS: return 10;
144 case IB_RATE_100_GBPS: return 40;
145 case IB_RATE_200_GBPS: return 80;
146 case IB_RATE_300_GBPS: return 120;
147 case IB_RATE_28_GBPS: return 11;
148 case IB_RATE_50_GBPS: return 20;
149 case IB_RATE_400_GBPS: return 160;
150 case IB_RATE_600_GBPS: return 240;
151 default: return -1;
152 }
153 }
154 EXPORT_SYMBOL(ib_rate_to_mult);
155
mult_to_ib_rate(int mult)156 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult)
157 {
158 switch (mult) {
159 case 1: return IB_RATE_2_5_GBPS;
160 case 2: return IB_RATE_5_GBPS;
161 case 4: return IB_RATE_10_GBPS;
162 case 8: return IB_RATE_20_GBPS;
163 case 12: return IB_RATE_30_GBPS;
164 case 16: return IB_RATE_40_GBPS;
165 case 24: return IB_RATE_60_GBPS;
166 case 32: return IB_RATE_80_GBPS;
167 case 48: return IB_RATE_120_GBPS;
168 case 6: return IB_RATE_14_GBPS;
169 case 22: return IB_RATE_56_GBPS;
170 case 45: return IB_RATE_112_GBPS;
171 case 67: return IB_RATE_168_GBPS;
172 case 10: return IB_RATE_25_GBPS;
173 case 40: return IB_RATE_100_GBPS;
174 case 80: return IB_RATE_200_GBPS;
175 case 120: return IB_RATE_300_GBPS;
176 case 11: return IB_RATE_28_GBPS;
177 case 20: return IB_RATE_50_GBPS;
178 case 160: return IB_RATE_400_GBPS;
179 case 240: return IB_RATE_600_GBPS;
180 default: return IB_RATE_PORT_CURRENT;
181 }
182 }
183 EXPORT_SYMBOL(mult_to_ib_rate);
184
ib_rate_to_mbps(enum ib_rate rate)185 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate)
186 {
187 switch (rate) {
188 case IB_RATE_2_5_GBPS: return 2500;
189 case IB_RATE_5_GBPS: return 5000;
190 case IB_RATE_10_GBPS: return 10000;
191 case IB_RATE_20_GBPS: return 20000;
192 case IB_RATE_30_GBPS: return 30000;
193 case IB_RATE_40_GBPS: return 40000;
194 case IB_RATE_60_GBPS: return 60000;
195 case IB_RATE_80_GBPS: return 80000;
196 case IB_RATE_120_GBPS: return 120000;
197 case IB_RATE_14_GBPS: return 14062;
198 case IB_RATE_56_GBPS: return 56250;
199 case IB_RATE_112_GBPS: return 112500;
200 case IB_RATE_168_GBPS: return 168750;
201 case IB_RATE_25_GBPS: return 25781;
202 case IB_RATE_100_GBPS: return 103125;
203 case IB_RATE_200_GBPS: return 206250;
204 case IB_RATE_300_GBPS: return 309375;
205 case IB_RATE_28_GBPS: return 28125;
206 case IB_RATE_50_GBPS: return 53125;
207 case IB_RATE_400_GBPS: return 425000;
208 case IB_RATE_600_GBPS: return 637500;
209 default: return -1;
210 }
211 }
212 EXPORT_SYMBOL(ib_rate_to_mbps);
213
214 __attribute_const__ enum rdma_transport_type
rdma_node_get_transport(enum rdma_node_type node_type)215 rdma_node_get_transport(enum rdma_node_type node_type)
216 {
217 switch (node_type) {
218 case RDMA_NODE_IB_CA:
219 case RDMA_NODE_IB_SWITCH:
220 case RDMA_NODE_IB_ROUTER:
221 return RDMA_TRANSPORT_IB;
222 case RDMA_NODE_RNIC:
223 return RDMA_TRANSPORT_IWARP;
224 case RDMA_NODE_USNIC:
225 return RDMA_TRANSPORT_USNIC;
226 case RDMA_NODE_USNIC_UDP:
227 return RDMA_TRANSPORT_USNIC_UDP;
228 default:
229 BUG();
230 return 0;
231 }
232 }
233 EXPORT_SYMBOL(rdma_node_get_transport);
234
rdma_port_get_link_layer(struct ib_device * device,u8 port_num)235 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, u8 port_num)
236 {
237 if (device->get_link_layer)
238 return device->get_link_layer(device, port_num);
239
240 switch (rdma_node_get_transport(device->node_type)) {
241 case RDMA_TRANSPORT_IB:
242 return IB_LINK_LAYER_INFINIBAND;
243 case RDMA_TRANSPORT_IWARP:
244 case RDMA_TRANSPORT_USNIC:
245 case RDMA_TRANSPORT_USNIC_UDP:
246 return IB_LINK_LAYER_ETHERNET;
247 default:
248 return IB_LINK_LAYER_UNSPECIFIED;
249 }
250 }
251 EXPORT_SYMBOL(rdma_port_get_link_layer);
252
253 /* Protection domains */
254
255 /**
256 * ib_alloc_pd - Allocates an unused protection domain.
257 * @device: The device on which to allocate the protection domain.
258 *
259 * A protection domain object provides an association between QPs, shared
260 * receive queues, address handles, memory regions, and memory windows.
261 *
262 * Every PD has a local_dma_lkey which can be used as the lkey value for local
263 * memory operations.
264 */
__ib_alloc_pd(struct ib_device * device,unsigned int flags,const char * caller)265 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
266 const char *caller)
267 {
268 struct ib_pd *pd;
269 int mr_access_flags = 0;
270
271 pd = device->alloc_pd(device, NULL, NULL);
272 if (IS_ERR(pd))
273 return pd;
274
275 pd->device = device;
276 pd->uobject = NULL;
277 pd->__internal_mr = NULL;
278 atomic_set(&pd->usecnt, 0);
279 pd->flags = flags;
280
281 if (device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
282 pd->local_dma_lkey = device->local_dma_lkey;
283 else
284 mr_access_flags |= IB_ACCESS_LOCAL_WRITE;
285
286 if (flags & IB_PD_UNSAFE_GLOBAL_RKEY) {
287 pr_warn("%s: enabling unsafe global rkey\n", caller);
288 mr_access_flags |= IB_ACCESS_REMOTE_READ | IB_ACCESS_REMOTE_WRITE;
289 }
290
291 if (mr_access_flags) {
292 struct ib_mr *mr;
293
294 mr = pd->device->get_dma_mr(pd, mr_access_flags);
295 if (IS_ERR(mr)) {
296 ib_dealloc_pd(pd);
297 return ERR_CAST(mr);
298 }
299
300 mr->device = pd->device;
301 mr->pd = pd;
302 mr->uobject = NULL;
303 mr->need_inval = false;
304
305 pd->__internal_mr = mr;
306
307 if (!(device->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY))
308 pd->local_dma_lkey = pd->__internal_mr->lkey;
309
310 if (flags & IB_PD_UNSAFE_GLOBAL_RKEY)
311 pd->unsafe_global_rkey = pd->__internal_mr->rkey;
312 }
313
314 return pd;
315 }
316 EXPORT_SYMBOL(__ib_alloc_pd);
317
318 /**
319 * ib_dealloc_pd - Deallocates a protection domain.
320 * @pd: The protection domain to deallocate.
321 *
322 * It is an error to call this function while any resources in the pd still
323 * exist. The caller is responsible to synchronously destroy them and
324 * guarantee no new allocations will happen.
325 */
ib_dealloc_pd(struct ib_pd * pd)326 void ib_dealloc_pd(struct ib_pd *pd)
327 {
328 int ret;
329
330 if (pd->__internal_mr) {
331 ret = pd->device->dereg_mr(pd->__internal_mr);
332 WARN_ON(ret);
333 pd->__internal_mr = NULL;
334 }
335
336 /* uverbs manipulates usecnt with proper locking, while the kabi
337 requires the caller to guarantee we can't race here. */
338 WARN_ON(atomic_read(&pd->usecnt));
339
340 /* Making delalloc_pd a void return is a WIP, no driver should return
341 an error here. */
342 ret = pd->device->dealloc_pd(pd);
343 WARN_ONCE(ret, "Infiniband HW driver failed dealloc_pd");
344 }
345 EXPORT_SYMBOL(ib_dealloc_pd);
346
347 /* Address handles */
348
ib_create_ah(struct ib_pd * pd,struct ib_ah_attr * ah_attr)349 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
350 {
351 struct ib_ah *ah;
352
353 ah = pd->device->create_ah(pd, ah_attr, NULL);
354
355 if (!IS_ERR(ah)) {
356 ah->device = pd->device;
357 ah->pd = pd;
358 ah->uobject = NULL;
359 atomic_inc(&pd->usecnt);
360 }
361
362 return ah;
363 }
364 EXPORT_SYMBOL(ib_create_ah);
365
ib_get_header_version(const union rdma_network_hdr * hdr)366 static int ib_get_header_version(const union rdma_network_hdr *hdr)
367 {
368 const struct ip *ip4h = (const struct ip *)&hdr->roce4grh;
369 struct ip ip4h_checked;
370 const struct ip6_hdr *ip6h = (const struct ip6_hdr *)&hdr->ibgrh;
371
372 /* If it's IPv6, the version must be 6, otherwise, the first
373 * 20 bytes (before the IPv4 header) are garbled.
374 */
375 if ((ip6h->ip6_vfc & IPV6_VERSION_MASK) != IPV6_VERSION)
376 return (ip4h->ip_v == 4) ? 4 : 0;
377 /* version may be 6 or 4 because the first 20 bytes could be garbled */
378
379 /* RoCE v2 requires no options, thus header length
380 * must be 5 words
381 */
382 if (ip4h->ip_hl != 5)
383 return 6;
384
385 /* Verify checksum.
386 * We can't write on scattered buffers so we need to copy to
387 * temp buffer.
388 */
389 memcpy(&ip4h_checked, ip4h, sizeof(ip4h_checked));
390 ip4h_checked.ip_sum = 0;
391 #if defined(INET) || defined(INET6)
392 ip4h_checked.ip_sum = in_cksum_hdr(&ip4h_checked);
393 #endif
394 /* if IPv4 header checksum is OK, believe it */
395 if (ip4h->ip_sum == ip4h_checked.ip_sum)
396 return 4;
397 return 6;
398 }
399
ib_get_net_type_by_grh(struct ib_device * device,u8 port_num,const struct ib_grh * grh)400 static enum rdma_network_type ib_get_net_type_by_grh(struct ib_device *device,
401 u8 port_num,
402 const struct ib_grh *grh)
403 {
404 int grh_version;
405
406 if (rdma_protocol_ib(device, port_num))
407 return RDMA_NETWORK_IB;
408
409 grh_version = ib_get_header_version((const union rdma_network_hdr *)grh);
410
411 if (grh_version == 4)
412 return RDMA_NETWORK_IPV4;
413
414 if (grh->next_hdr == IPPROTO_UDP)
415 return RDMA_NETWORK_IPV6;
416
417 return RDMA_NETWORK_ROCE_V1;
418 }
419
420 struct find_gid_index_context {
421 u16 vlan_id;
422 enum ib_gid_type gid_type;
423 };
424
425
426 /*
427 * This function will return true only if a inspected GID index
428 * matches the request based on the GID type and VLAN configuration
429 */
find_gid_index(const union ib_gid * gid,const struct ib_gid_attr * gid_attr,void * context)430 static bool find_gid_index(const union ib_gid *gid,
431 const struct ib_gid_attr *gid_attr,
432 void *context)
433 {
434 u16 vlan_diff;
435 struct find_gid_index_context *ctx =
436 (struct find_gid_index_context *)context;
437
438 if (ctx->gid_type != gid_attr->gid_type)
439 return false;
440
441 /*
442 * The following will verify:
443 * 1. VLAN ID matching for VLAN tagged requests.
444 * 2. prio-tagged/untagged to prio-tagged/untagged matching.
445 *
446 * This XOR is valid, since 0x0 < vlan_id < 0x0FFF.
447 */
448 vlan_diff = rdma_vlan_dev_vlan_id(gid_attr->ndev) ^ ctx->vlan_id;
449
450 return (vlan_diff == 0x0000 || vlan_diff == 0xFFFF);
451 }
452
get_sgid_index_from_eth(struct ib_device * device,u8 port_num,u16 vlan_id,const union ib_gid * sgid,enum ib_gid_type gid_type,u16 * gid_index)453 static int get_sgid_index_from_eth(struct ib_device *device, u8 port_num,
454 u16 vlan_id, const union ib_gid *sgid,
455 enum ib_gid_type gid_type,
456 u16 *gid_index)
457 {
458 struct find_gid_index_context context = {.vlan_id = vlan_id,
459 .gid_type = gid_type};
460
461 return ib_find_gid_by_filter(device, sgid, port_num, find_gid_index,
462 &context, gid_index);
463 }
464
get_gids_from_rdma_hdr(const union rdma_network_hdr * hdr,enum rdma_network_type net_type,union ib_gid * sgid,union ib_gid * dgid)465 static int get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
466 enum rdma_network_type net_type,
467 union ib_gid *sgid, union ib_gid *dgid)
468 {
469 struct sockaddr_in src_in;
470 struct sockaddr_in dst_in;
471 __be32 src_saddr, dst_saddr;
472
473 if (!sgid || !dgid)
474 return -EINVAL;
475
476 if (net_type == RDMA_NETWORK_IPV4) {
477 memcpy(&src_in.sin_addr.s_addr,
478 &hdr->roce4grh.ip_src, 4);
479 memcpy(&dst_in.sin_addr.s_addr,
480 &hdr->roce4grh.ip_dst, 4);
481 src_saddr = src_in.sin_addr.s_addr;
482 dst_saddr = dst_in.sin_addr.s_addr;
483 ipv6_addr_set_v4mapped(src_saddr,
484 (struct in6_addr *)sgid);
485 ipv6_addr_set_v4mapped(dst_saddr,
486 (struct in6_addr *)dgid);
487 return 0;
488 } else if (net_type == RDMA_NETWORK_IPV6 ||
489 net_type == RDMA_NETWORK_IB) {
490 *dgid = hdr->ibgrh.dgid;
491 *sgid = hdr->ibgrh.sgid;
492 return 0;
493 } else {
494 return -EINVAL;
495 }
496 }
497
ib_init_ah_from_wc(struct ib_device * device,u8 port_num,const struct ib_wc * wc,const struct ib_grh * grh,struct ib_ah_attr * ah_attr)498 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
499 const struct ib_wc *wc, const struct ib_grh *grh,
500 struct ib_ah_attr *ah_attr)
501 {
502 u32 flow_class;
503 u16 gid_index = 0;
504 int ret;
505 enum rdma_network_type net_type = RDMA_NETWORK_IB;
506 enum ib_gid_type gid_type = IB_GID_TYPE_IB;
507 int hoplimit = 0xff;
508 union ib_gid dgid;
509 union ib_gid sgid;
510
511 memset(ah_attr, 0, sizeof *ah_attr);
512 if (rdma_cap_eth_ah(device, port_num)) {
513 if (wc->wc_flags & IB_WC_WITH_NETWORK_HDR_TYPE)
514 net_type = wc->network_hdr_type;
515 else
516 net_type = ib_get_net_type_by_grh(device, port_num, grh);
517 gid_type = ib_network_to_gid_type(net_type);
518 }
519 ret = get_gids_from_rdma_hdr((const union rdma_network_hdr *)grh, net_type,
520 &sgid, &dgid);
521 if (ret)
522 return ret;
523
524 if (rdma_protocol_roce(device, port_num)) {
525 struct ib_gid_attr dgid_attr;
526 const u16 vlan_id = (wc->wc_flags & IB_WC_WITH_VLAN) ?
527 wc->vlan_id : 0xffff;
528
529 if (!(wc->wc_flags & IB_WC_GRH))
530 return -EPROTOTYPE;
531
532 ret = get_sgid_index_from_eth(device, port_num, vlan_id,
533 &dgid, gid_type, &gid_index);
534 if (ret)
535 return ret;
536
537 ret = ib_get_cached_gid(device, port_num, gid_index, &dgid, &dgid_attr);
538 if (ret)
539 return ret;
540
541 if (dgid_attr.ndev == NULL)
542 return -ENODEV;
543
544 ret = rdma_addr_find_l2_eth_by_grh(&dgid, &sgid, ah_attr->dmac,
545 dgid_attr.ndev, &hoplimit);
546
547 dev_put(dgid_attr.ndev);
548 if (ret)
549 return ret;
550 }
551
552 ah_attr->dlid = wc->slid;
553 ah_attr->sl = wc->sl;
554 ah_attr->src_path_bits = wc->dlid_path_bits;
555 ah_attr->port_num = port_num;
556
557 if (wc->wc_flags & IB_WC_GRH) {
558 ah_attr->ah_flags = IB_AH_GRH;
559 ah_attr->grh.dgid = sgid;
560
561 if (!rdma_cap_eth_ah(device, port_num)) {
562 if (dgid.global.interface_id != cpu_to_be64(IB_SA_WELL_KNOWN_GUID)) {
563 ret = ib_find_cached_gid_by_port(device, &dgid,
564 IB_GID_TYPE_IB,
565 port_num, NULL,
566 &gid_index);
567 if (ret)
568 return ret;
569 }
570 }
571
572 ah_attr->grh.sgid_index = (u8) gid_index;
573 flow_class = be32_to_cpu(grh->version_tclass_flow);
574 ah_attr->grh.flow_label = flow_class & 0xFFFFF;
575 ah_attr->grh.hop_limit = hoplimit;
576 ah_attr->grh.traffic_class = (flow_class >> 20) & 0xFF;
577 }
578 return 0;
579 }
580 EXPORT_SYMBOL(ib_init_ah_from_wc);
581
ib_create_ah_from_wc(struct ib_pd * pd,const struct ib_wc * wc,const struct ib_grh * grh,u8 port_num)582 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
583 const struct ib_grh *grh, u8 port_num)
584 {
585 struct ib_ah_attr ah_attr;
586 int ret;
587
588 ret = ib_init_ah_from_wc(pd->device, port_num, wc, grh, &ah_attr);
589 if (ret)
590 return ERR_PTR(ret);
591
592 return ib_create_ah(pd, &ah_attr);
593 }
594 EXPORT_SYMBOL(ib_create_ah_from_wc);
595
ib_modify_ah(struct ib_ah * ah,struct ib_ah_attr * ah_attr)596 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
597 {
598 return ah->device->modify_ah ?
599 ah->device->modify_ah(ah, ah_attr) :
600 -ENOSYS;
601 }
602 EXPORT_SYMBOL(ib_modify_ah);
603
ib_query_ah(struct ib_ah * ah,struct ib_ah_attr * ah_attr)604 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr)
605 {
606 return ah->device->query_ah ?
607 ah->device->query_ah(ah, ah_attr) :
608 -ENOSYS;
609 }
610 EXPORT_SYMBOL(ib_query_ah);
611
ib_destroy_ah(struct ib_ah * ah)612 int ib_destroy_ah(struct ib_ah *ah)
613 {
614 struct ib_pd *pd;
615 int ret;
616
617 pd = ah->pd;
618 ret = ah->device->destroy_ah(ah);
619 if (!ret)
620 atomic_dec(&pd->usecnt);
621
622 return ret;
623 }
624 EXPORT_SYMBOL(ib_destroy_ah);
625
626 /* Shared receive queues */
627
ib_create_srq(struct ib_pd * pd,struct ib_srq_init_attr * srq_init_attr)628 struct ib_srq *ib_create_srq(struct ib_pd *pd,
629 struct ib_srq_init_attr *srq_init_attr)
630 {
631 struct ib_srq *srq;
632
633 if (!pd->device->create_srq)
634 return ERR_PTR(-ENOSYS);
635
636 srq = pd->device->create_srq(pd, srq_init_attr, NULL);
637
638 if (!IS_ERR(srq)) {
639 srq->device = pd->device;
640 srq->pd = pd;
641 srq->uobject = NULL;
642 srq->event_handler = srq_init_attr->event_handler;
643 srq->srq_context = srq_init_attr->srq_context;
644 srq->srq_type = srq_init_attr->srq_type;
645 if (srq->srq_type == IB_SRQT_XRC) {
646 srq->ext.xrc.xrcd = srq_init_attr->ext.xrc.xrcd;
647 srq->ext.xrc.cq = srq_init_attr->ext.xrc.cq;
648 atomic_inc(&srq->ext.xrc.xrcd->usecnt);
649 atomic_inc(&srq->ext.xrc.cq->usecnt);
650 }
651 atomic_inc(&pd->usecnt);
652 atomic_set(&srq->usecnt, 0);
653 }
654
655 return srq;
656 }
657 EXPORT_SYMBOL(ib_create_srq);
658
ib_modify_srq(struct ib_srq * srq,struct ib_srq_attr * srq_attr,enum ib_srq_attr_mask srq_attr_mask)659 int ib_modify_srq(struct ib_srq *srq,
660 struct ib_srq_attr *srq_attr,
661 enum ib_srq_attr_mask srq_attr_mask)
662 {
663 return srq->device->modify_srq ?
664 srq->device->modify_srq(srq, srq_attr, srq_attr_mask, NULL) :
665 -ENOSYS;
666 }
667 EXPORT_SYMBOL(ib_modify_srq);
668
ib_query_srq(struct ib_srq * srq,struct ib_srq_attr * srq_attr)669 int ib_query_srq(struct ib_srq *srq,
670 struct ib_srq_attr *srq_attr)
671 {
672 return srq->device->query_srq ?
673 srq->device->query_srq(srq, srq_attr) : -ENOSYS;
674 }
675 EXPORT_SYMBOL(ib_query_srq);
676
ib_destroy_srq(struct ib_srq * srq)677 int ib_destroy_srq(struct ib_srq *srq)
678 {
679 struct ib_pd *pd;
680 enum ib_srq_type srq_type;
681 struct ib_xrcd *uninitialized_var(xrcd);
682 struct ib_cq *uninitialized_var(cq);
683 int ret;
684
685 if (atomic_read(&srq->usecnt))
686 return -EBUSY;
687
688 pd = srq->pd;
689 srq_type = srq->srq_type;
690 if (srq_type == IB_SRQT_XRC) {
691 xrcd = srq->ext.xrc.xrcd;
692 cq = srq->ext.xrc.cq;
693 }
694
695 ret = srq->device->destroy_srq(srq);
696 if (!ret) {
697 atomic_dec(&pd->usecnt);
698 if (srq_type == IB_SRQT_XRC) {
699 atomic_dec(&xrcd->usecnt);
700 atomic_dec(&cq->usecnt);
701 }
702 }
703
704 return ret;
705 }
706 EXPORT_SYMBOL(ib_destroy_srq);
707
708 /* Queue pairs */
709
__ib_shared_qp_event_handler(struct ib_event * event,void * context)710 static void __ib_shared_qp_event_handler(struct ib_event *event, void *context)
711 {
712 struct ib_qp *qp = context;
713 unsigned long flags;
714
715 spin_lock_irqsave(&qp->device->event_handler_lock, flags);
716 list_for_each_entry(event->element.qp, &qp->open_list, open_list)
717 if (event->element.qp->event_handler)
718 event->element.qp->event_handler(event, event->element.qp->qp_context);
719 spin_unlock_irqrestore(&qp->device->event_handler_lock, flags);
720 }
721
__ib_insert_xrcd_qp(struct ib_xrcd * xrcd,struct ib_qp * qp)722 static void __ib_insert_xrcd_qp(struct ib_xrcd *xrcd, struct ib_qp *qp)
723 {
724 mutex_lock(&xrcd->tgt_qp_mutex);
725 list_add(&qp->xrcd_list, &xrcd->tgt_qp_list);
726 mutex_unlock(&xrcd->tgt_qp_mutex);
727 }
728
__ib_open_qp(struct ib_qp * real_qp,void (* event_handler)(struct ib_event *,void *),void * qp_context)729 static struct ib_qp *__ib_open_qp(struct ib_qp *real_qp,
730 void (*event_handler)(struct ib_event *, void *),
731 void *qp_context)
732 {
733 struct ib_qp *qp;
734 unsigned long flags;
735
736 qp = kzalloc(sizeof *qp, GFP_KERNEL);
737 if (!qp)
738 return ERR_PTR(-ENOMEM);
739
740 qp->real_qp = real_qp;
741 atomic_inc(&real_qp->usecnt);
742 qp->device = real_qp->device;
743 qp->event_handler = event_handler;
744 qp->qp_context = qp_context;
745 qp->qp_num = real_qp->qp_num;
746 qp->qp_type = real_qp->qp_type;
747
748 spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
749 list_add(&qp->open_list, &real_qp->open_list);
750 spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
751
752 return qp;
753 }
754
ib_open_qp(struct ib_xrcd * xrcd,struct ib_qp_open_attr * qp_open_attr)755 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
756 struct ib_qp_open_attr *qp_open_attr)
757 {
758 struct ib_qp *qp, *real_qp;
759
760 if (qp_open_attr->qp_type != IB_QPT_XRC_TGT)
761 return ERR_PTR(-EINVAL);
762
763 qp = ERR_PTR(-EINVAL);
764 mutex_lock(&xrcd->tgt_qp_mutex);
765 list_for_each_entry(real_qp, &xrcd->tgt_qp_list, xrcd_list) {
766 if (real_qp->qp_num == qp_open_attr->qp_num) {
767 qp = __ib_open_qp(real_qp, qp_open_attr->event_handler,
768 qp_open_attr->qp_context);
769 break;
770 }
771 }
772 mutex_unlock(&xrcd->tgt_qp_mutex);
773 return qp;
774 }
775 EXPORT_SYMBOL(ib_open_qp);
776
ib_create_xrc_qp(struct ib_qp * qp,struct ib_qp_init_attr * qp_init_attr)777 static struct ib_qp *ib_create_xrc_qp(struct ib_qp *qp,
778 struct ib_qp_init_attr *qp_init_attr)
779 {
780 struct ib_qp *real_qp = qp;
781
782 qp->event_handler = __ib_shared_qp_event_handler;
783 qp->qp_context = qp;
784 qp->pd = NULL;
785 qp->send_cq = qp->recv_cq = NULL;
786 qp->srq = NULL;
787 qp->xrcd = qp_init_attr->xrcd;
788 atomic_inc(&qp_init_attr->xrcd->usecnt);
789 INIT_LIST_HEAD(&qp->open_list);
790
791 qp = __ib_open_qp(real_qp, qp_init_attr->event_handler,
792 qp_init_attr->qp_context);
793 if (!IS_ERR(qp))
794 __ib_insert_xrcd_qp(qp_init_attr->xrcd, real_qp);
795 else
796 real_qp->device->destroy_qp(real_qp);
797 return qp;
798 }
799
ib_create_qp(struct ib_pd * pd,struct ib_qp_init_attr * qp_init_attr)800 struct ib_qp *ib_create_qp(struct ib_pd *pd,
801 struct ib_qp_init_attr *qp_init_attr)
802 {
803 struct ib_device *device = pd ? pd->device : qp_init_attr->xrcd->device;
804 struct ib_qp *qp;
805
806 if (qp_init_attr->rwq_ind_tbl &&
807 (qp_init_attr->recv_cq ||
808 qp_init_attr->srq || qp_init_attr->cap.max_recv_wr ||
809 qp_init_attr->cap.max_recv_sge))
810 return ERR_PTR(-EINVAL);
811
812 qp = device->create_qp(pd, qp_init_attr, NULL);
813 if (IS_ERR(qp))
814 return qp;
815
816 qp->device = device;
817 qp->real_qp = qp;
818 qp->uobject = NULL;
819 qp->qp_type = qp_init_attr->qp_type;
820 qp->rwq_ind_tbl = qp_init_attr->rwq_ind_tbl;
821
822 atomic_set(&qp->usecnt, 0);
823 spin_lock_init(&qp->mr_lock);
824
825 if (qp_init_attr->qp_type == IB_QPT_XRC_TGT)
826 return ib_create_xrc_qp(qp, qp_init_attr);
827
828 qp->event_handler = qp_init_attr->event_handler;
829 qp->qp_context = qp_init_attr->qp_context;
830 if (qp_init_attr->qp_type == IB_QPT_XRC_INI) {
831 qp->recv_cq = NULL;
832 qp->srq = NULL;
833 } else {
834 qp->recv_cq = qp_init_attr->recv_cq;
835 if (qp_init_attr->recv_cq)
836 atomic_inc(&qp_init_attr->recv_cq->usecnt);
837 qp->srq = qp_init_attr->srq;
838 if (qp->srq)
839 atomic_inc(&qp_init_attr->srq->usecnt);
840 }
841
842 qp->pd = pd;
843 qp->send_cq = qp_init_attr->send_cq;
844 qp->xrcd = NULL;
845
846 atomic_inc(&pd->usecnt);
847 if (qp_init_attr->send_cq)
848 atomic_inc(&qp_init_attr->send_cq->usecnt);
849 if (qp_init_attr->rwq_ind_tbl)
850 atomic_inc(&qp->rwq_ind_tbl->usecnt);
851
852 /*
853 * Note: all hw drivers guarantee that max_send_sge is lower than
854 * the device RDMA WRITE SGE limit but not all hw drivers ensure that
855 * max_send_sge <= max_sge_rd.
856 */
857 qp->max_write_sge = qp_init_attr->cap.max_send_sge;
858 qp->max_read_sge = min_t(u32, qp_init_attr->cap.max_send_sge,
859 device->attrs.max_sge_rd);
860
861 return qp;
862 }
863 EXPORT_SYMBOL(ib_create_qp);
864
865 static const struct {
866 int valid;
867 enum ib_qp_attr_mask req_param[IB_QPT_MAX];
868 enum ib_qp_attr_mask opt_param[IB_QPT_MAX];
869 } qp_state_table[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
870 [IB_QPS_RESET] = {
871 [IB_QPS_RESET] = { .valid = 1 },
872 [IB_QPS_INIT] = {
873 .valid = 1,
874 .req_param = {
875 [IB_QPT_UD] = (IB_QP_PKEY_INDEX |
876 IB_QP_PORT |
877 IB_QP_QKEY),
878 [IB_QPT_RAW_PACKET] = IB_QP_PORT,
879 [IB_QPT_UC] = (IB_QP_PKEY_INDEX |
880 IB_QP_PORT |
881 IB_QP_ACCESS_FLAGS),
882 [IB_QPT_RC] = (IB_QP_PKEY_INDEX |
883 IB_QP_PORT |
884 IB_QP_ACCESS_FLAGS),
885 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX |
886 IB_QP_PORT |
887 IB_QP_ACCESS_FLAGS),
888 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX |
889 IB_QP_PORT |
890 IB_QP_ACCESS_FLAGS),
891 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
892 IB_QP_QKEY),
893 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
894 IB_QP_QKEY),
895 }
896 },
897 },
898 [IB_QPS_INIT] = {
899 [IB_QPS_RESET] = { .valid = 1 },
900 [IB_QPS_ERR] = { .valid = 1 },
901 [IB_QPS_INIT] = {
902 .valid = 1,
903 .opt_param = {
904 [IB_QPT_UD] = (IB_QP_PKEY_INDEX |
905 IB_QP_PORT |
906 IB_QP_QKEY),
907 [IB_QPT_UC] = (IB_QP_PKEY_INDEX |
908 IB_QP_PORT |
909 IB_QP_ACCESS_FLAGS),
910 [IB_QPT_RC] = (IB_QP_PKEY_INDEX |
911 IB_QP_PORT |
912 IB_QP_ACCESS_FLAGS),
913 [IB_QPT_XRC_INI] = (IB_QP_PKEY_INDEX |
914 IB_QP_PORT |
915 IB_QP_ACCESS_FLAGS),
916 [IB_QPT_XRC_TGT] = (IB_QP_PKEY_INDEX |
917 IB_QP_PORT |
918 IB_QP_ACCESS_FLAGS),
919 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
920 IB_QP_QKEY),
921 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
922 IB_QP_QKEY),
923 }
924 },
925 [IB_QPS_RTR] = {
926 .valid = 1,
927 .req_param = {
928 [IB_QPT_UC] = (IB_QP_AV |
929 IB_QP_PATH_MTU |
930 IB_QP_DEST_QPN |
931 IB_QP_RQ_PSN),
932 [IB_QPT_RC] = (IB_QP_AV |
933 IB_QP_PATH_MTU |
934 IB_QP_DEST_QPN |
935 IB_QP_RQ_PSN |
936 IB_QP_MAX_DEST_RD_ATOMIC |
937 IB_QP_MIN_RNR_TIMER),
938 [IB_QPT_XRC_INI] = (IB_QP_AV |
939 IB_QP_PATH_MTU |
940 IB_QP_DEST_QPN |
941 IB_QP_RQ_PSN),
942 [IB_QPT_XRC_TGT] = (IB_QP_AV |
943 IB_QP_PATH_MTU |
944 IB_QP_DEST_QPN |
945 IB_QP_RQ_PSN |
946 IB_QP_MAX_DEST_RD_ATOMIC |
947 IB_QP_MIN_RNR_TIMER),
948 },
949 .opt_param = {
950 [IB_QPT_UD] = (IB_QP_PKEY_INDEX |
951 IB_QP_QKEY),
952 [IB_QPT_UC] = (IB_QP_ALT_PATH |
953 IB_QP_ACCESS_FLAGS |
954 IB_QP_PKEY_INDEX),
955 [IB_QPT_RC] = (IB_QP_ALT_PATH |
956 IB_QP_ACCESS_FLAGS |
957 IB_QP_PKEY_INDEX),
958 [IB_QPT_XRC_INI] = (IB_QP_ALT_PATH |
959 IB_QP_ACCESS_FLAGS |
960 IB_QP_PKEY_INDEX),
961 [IB_QPT_XRC_TGT] = (IB_QP_ALT_PATH |
962 IB_QP_ACCESS_FLAGS |
963 IB_QP_PKEY_INDEX),
964 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
965 IB_QP_QKEY),
966 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
967 IB_QP_QKEY),
968 },
969 },
970 },
971 [IB_QPS_RTR] = {
972 [IB_QPS_RESET] = { .valid = 1 },
973 [IB_QPS_ERR] = { .valid = 1 },
974 [IB_QPS_RTS] = {
975 .valid = 1,
976 .req_param = {
977 [IB_QPT_UD] = IB_QP_SQ_PSN,
978 [IB_QPT_UC] = IB_QP_SQ_PSN,
979 [IB_QPT_RC] = (IB_QP_TIMEOUT |
980 IB_QP_RETRY_CNT |
981 IB_QP_RNR_RETRY |
982 IB_QP_SQ_PSN |
983 IB_QP_MAX_QP_RD_ATOMIC),
984 [IB_QPT_XRC_INI] = (IB_QP_TIMEOUT |
985 IB_QP_RETRY_CNT |
986 IB_QP_RNR_RETRY |
987 IB_QP_SQ_PSN |
988 IB_QP_MAX_QP_RD_ATOMIC),
989 [IB_QPT_XRC_TGT] = (IB_QP_TIMEOUT |
990 IB_QP_SQ_PSN),
991 [IB_QPT_SMI] = IB_QP_SQ_PSN,
992 [IB_QPT_GSI] = IB_QP_SQ_PSN,
993 },
994 .opt_param = {
995 [IB_QPT_UD] = (IB_QP_CUR_STATE |
996 IB_QP_QKEY),
997 [IB_QPT_UC] = (IB_QP_CUR_STATE |
998 IB_QP_ALT_PATH |
999 IB_QP_ACCESS_FLAGS |
1000 IB_QP_PATH_MIG_STATE),
1001 [IB_QPT_RC] = (IB_QP_CUR_STATE |
1002 IB_QP_ALT_PATH |
1003 IB_QP_ACCESS_FLAGS |
1004 IB_QP_MIN_RNR_TIMER |
1005 IB_QP_PATH_MIG_STATE),
1006 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE |
1007 IB_QP_ALT_PATH |
1008 IB_QP_ACCESS_FLAGS |
1009 IB_QP_PATH_MIG_STATE),
1010 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE |
1011 IB_QP_ALT_PATH |
1012 IB_QP_ACCESS_FLAGS |
1013 IB_QP_MIN_RNR_TIMER |
1014 IB_QP_PATH_MIG_STATE),
1015 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
1016 IB_QP_QKEY),
1017 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
1018 IB_QP_QKEY),
1019 [IB_QPT_RAW_PACKET] = IB_QP_RATE_LIMIT,
1020 }
1021 }
1022 },
1023 [IB_QPS_RTS] = {
1024 [IB_QPS_RESET] = { .valid = 1 },
1025 [IB_QPS_ERR] = { .valid = 1 },
1026 [IB_QPS_RTS] = {
1027 .valid = 1,
1028 .opt_param = {
1029 [IB_QPT_UD] = (IB_QP_CUR_STATE |
1030 IB_QP_QKEY),
1031 [IB_QPT_UC] = (IB_QP_CUR_STATE |
1032 IB_QP_ACCESS_FLAGS |
1033 IB_QP_ALT_PATH |
1034 IB_QP_PATH_MIG_STATE),
1035 [IB_QPT_RC] = (IB_QP_CUR_STATE |
1036 IB_QP_ACCESS_FLAGS |
1037 IB_QP_ALT_PATH |
1038 IB_QP_PATH_MIG_STATE |
1039 IB_QP_MIN_RNR_TIMER),
1040 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE |
1041 IB_QP_ACCESS_FLAGS |
1042 IB_QP_ALT_PATH |
1043 IB_QP_PATH_MIG_STATE),
1044 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE |
1045 IB_QP_ACCESS_FLAGS |
1046 IB_QP_ALT_PATH |
1047 IB_QP_PATH_MIG_STATE |
1048 IB_QP_MIN_RNR_TIMER),
1049 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
1050 IB_QP_QKEY),
1051 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
1052 IB_QP_QKEY),
1053 [IB_QPT_RAW_PACKET] = IB_QP_RATE_LIMIT,
1054 }
1055 },
1056 [IB_QPS_SQD] = {
1057 .valid = 1,
1058 .opt_param = {
1059 [IB_QPT_UD] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1060 [IB_QPT_UC] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1061 [IB_QPT_RC] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1062 [IB_QPT_XRC_INI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1063 [IB_QPT_XRC_TGT] = IB_QP_EN_SQD_ASYNC_NOTIFY, /* ??? */
1064 [IB_QPT_SMI] = IB_QP_EN_SQD_ASYNC_NOTIFY,
1065 [IB_QPT_GSI] = IB_QP_EN_SQD_ASYNC_NOTIFY
1066 }
1067 },
1068 },
1069 [IB_QPS_SQD] = {
1070 [IB_QPS_RESET] = { .valid = 1 },
1071 [IB_QPS_ERR] = { .valid = 1 },
1072 [IB_QPS_RTS] = {
1073 .valid = 1,
1074 .opt_param = {
1075 [IB_QPT_UD] = (IB_QP_CUR_STATE |
1076 IB_QP_QKEY),
1077 [IB_QPT_UC] = (IB_QP_CUR_STATE |
1078 IB_QP_ALT_PATH |
1079 IB_QP_ACCESS_FLAGS |
1080 IB_QP_PATH_MIG_STATE),
1081 [IB_QPT_RC] = (IB_QP_CUR_STATE |
1082 IB_QP_ALT_PATH |
1083 IB_QP_ACCESS_FLAGS |
1084 IB_QP_MIN_RNR_TIMER |
1085 IB_QP_PATH_MIG_STATE),
1086 [IB_QPT_XRC_INI] = (IB_QP_CUR_STATE |
1087 IB_QP_ALT_PATH |
1088 IB_QP_ACCESS_FLAGS |
1089 IB_QP_PATH_MIG_STATE),
1090 [IB_QPT_XRC_TGT] = (IB_QP_CUR_STATE |
1091 IB_QP_ALT_PATH |
1092 IB_QP_ACCESS_FLAGS |
1093 IB_QP_MIN_RNR_TIMER |
1094 IB_QP_PATH_MIG_STATE),
1095 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
1096 IB_QP_QKEY),
1097 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
1098 IB_QP_QKEY),
1099 }
1100 },
1101 [IB_QPS_SQD] = {
1102 .valid = 1,
1103 .opt_param = {
1104 [IB_QPT_UD] = (IB_QP_PKEY_INDEX |
1105 IB_QP_QKEY),
1106 [IB_QPT_UC] = (IB_QP_AV |
1107 IB_QP_ALT_PATH |
1108 IB_QP_ACCESS_FLAGS |
1109 IB_QP_PKEY_INDEX |
1110 IB_QP_PATH_MIG_STATE),
1111 [IB_QPT_RC] = (IB_QP_PORT |
1112 IB_QP_AV |
1113 IB_QP_TIMEOUT |
1114 IB_QP_RETRY_CNT |
1115 IB_QP_RNR_RETRY |
1116 IB_QP_MAX_QP_RD_ATOMIC |
1117 IB_QP_MAX_DEST_RD_ATOMIC |
1118 IB_QP_ALT_PATH |
1119 IB_QP_ACCESS_FLAGS |
1120 IB_QP_PKEY_INDEX |
1121 IB_QP_MIN_RNR_TIMER |
1122 IB_QP_PATH_MIG_STATE),
1123 [IB_QPT_XRC_INI] = (IB_QP_PORT |
1124 IB_QP_AV |
1125 IB_QP_TIMEOUT |
1126 IB_QP_RETRY_CNT |
1127 IB_QP_RNR_RETRY |
1128 IB_QP_MAX_QP_RD_ATOMIC |
1129 IB_QP_ALT_PATH |
1130 IB_QP_ACCESS_FLAGS |
1131 IB_QP_PKEY_INDEX |
1132 IB_QP_PATH_MIG_STATE),
1133 [IB_QPT_XRC_TGT] = (IB_QP_PORT |
1134 IB_QP_AV |
1135 IB_QP_TIMEOUT |
1136 IB_QP_MAX_DEST_RD_ATOMIC |
1137 IB_QP_ALT_PATH |
1138 IB_QP_ACCESS_FLAGS |
1139 IB_QP_PKEY_INDEX |
1140 IB_QP_MIN_RNR_TIMER |
1141 IB_QP_PATH_MIG_STATE),
1142 [IB_QPT_SMI] = (IB_QP_PKEY_INDEX |
1143 IB_QP_QKEY),
1144 [IB_QPT_GSI] = (IB_QP_PKEY_INDEX |
1145 IB_QP_QKEY),
1146 }
1147 }
1148 },
1149 [IB_QPS_SQE] = {
1150 [IB_QPS_RESET] = { .valid = 1 },
1151 [IB_QPS_ERR] = { .valid = 1 },
1152 [IB_QPS_RTS] = {
1153 .valid = 1,
1154 .opt_param = {
1155 [IB_QPT_UD] = (IB_QP_CUR_STATE |
1156 IB_QP_QKEY),
1157 [IB_QPT_UC] = (IB_QP_CUR_STATE |
1158 IB_QP_ACCESS_FLAGS),
1159 [IB_QPT_SMI] = (IB_QP_CUR_STATE |
1160 IB_QP_QKEY),
1161 [IB_QPT_GSI] = (IB_QP_CUR_STATE |
1162 IB_QP_QKEY),
1163 }
1164 }
1165 },
1166 [IB_QPS_ERR] = {
1167 [IB_QPS_RESET] = { .valid = 1 },
1168 [IB_QPS_ERR] = { .valid = 1 }
1169 }
1170 };
1171
ib_modify_qp_is_ok(enum ib_qp_state cur_state,enum ib_qp_state next_state,enum ib_qp_type type,enum ib_qp_attr_mask mask)1172 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1173 enum ib_qp_type type, enum ib_qp_attr_mask mask)
1174 {
1175 enum ib_qp_attr_mask req_param, opt_param;
1176
1177 if (mask & IB_QP_CUR_STATE &&
1178 cur_state != IB_QPS_RTR && cur_state != IB_QPS_RTS &&
1179 cur_state != IB_QPS_SQD && cur_state != IB_QPS_SQE)
1180 return false;
1181
1182 if (!qp_state_table[cur_state][next_state].valid)
1183 return false;
1184
1185 req_param = qp_state_table[cur_state][next_state].req_param[type];
1186 opt_param = qp_state_table[cur_state][next_state].opt_param[type];
1187
1188 if ((mask & req_param) != req_param)
1189 return false;
1190
1191 if (mask & ~(req_param | opt_param | IB_QP_STATE))
1192 return false;
1193
1194 return true;
1195 }
1196 EXPORT_SYMBOL(ib_modify_qp_is_ok);
1197
ib_resolve_eth_dmac(struct ib_device * device,struct ib_ah_attr * ah_attr)1198 int ib_resolve_eth_dmac(struct ib_device *device,
1199 struct ib_ah_attr *ah_attr)
1200 {
1201 struct ib_gid_attr sgid_attr;
1202 union ib_gid sgid;
1203 int hop_limit;
1204 int ret;
1205
1206 if (ah_attr->port_num < rdma_start_port(device) ||
1207 ah_attr->port_num > rdma_end_port(device))
1208 return -EINVAL;
1209
1210 if (!rdma_cap_eth_ah(device, ah_attr->port_num))
1211 return 0;
1212
1213 if (rdma_is_multicast_addr((struct in6_addr *)ah_attr->grh.dgid.raw)) {
1214 if (ipv6_addr_v4mapped((struct in6_addr *)ah_attr->grh.dgid.raw)) {
1215 __be32 addr = 0;
1216
1217 memcpy(&addr, ah_attr->grh.dgid.raw + 12, 4);
1218 ip_eth_mc_map(addr, (char *)ah_attr->dmac);
1219 } else {
1220 ipv6_eth_mc_map((struct in6_addr *)ah_attr->grh.dgid.raw,
1221 (char *)ah_attr->dmac);
1222 }
1223 return 0;
1224 }
1225
1226 ret = ib_query_gid(device,
1227 ah_attr->port_num,
1228 ah_attr->grh.sgid_index,
1229 &sgid, &sgid_attr);
1230 if (ret != 0)
1231 return (ret);
1232 if (!sgid_attr.ndev)
1233 return -ENXIO;
1234
1235 ret = rdma_addr_find_l2_eth_by_grh(&sgid,
1236 &ah_attr->grh.dgid,
1237 ah_attr->dmac,
1238 sgid_attr.ndev, &hop_limit);
1239 dev_put(sgid_attr.ndev);
1240
1241 ah_attr->grh.hop_limit = hop_limit;
1242 return ret;
1243 }
1244 EXPORT_SYMBOL(ib_resolve_eth_dmac);
1245
1246
ib_modify_qp(struct ib_qp * qp,struct ib_qp_attr * qp_attr,int qp_attr_mask)1247 int ib_modify_qp(struct ib_qp *qp,
1248 struct ib_qp_attr *qp_attr,
1249 int qp_attr_mask)
1250 {
1251 if (qp_attr_mask & IB_QP_AV) {
1252 int ret;
1253
1254 ret = ib_resolve_eth_dmac(qp->device, &qp_attr->ah_attr);
1255 if (ret)
1256 return ret;
1257 }
1258
1259 return qp->device->modify_qp(qp->real_qp, qp_attr, qp_attr_mask, NULL);
1260 }
1261 EXPORT_SYMBOL(ib_modify_qp);
1262
ib_query_qp(struct ib_qp * qp,struct ib_qp_attr * qp_attr,int qp_attr_mask,struct ib_qp_init_attr * qp_init_attr)1263 int ib_query_qp(struct ib_qp *qp,
1264 struct ib_qp_attr *qp_attr,
1265 int qp_attr_mask,
1266 struct ib_qp_init_attr *qp_init_attr)
1267 {
1268 return qp->device->query_qp ?
1269 qp->device->query_qp(qp->real_qp, qp_attr, qp_attr_mask, qp_init_attr) :
1270 -ENOSYS;
1271 }
1272 EXPORT_SYMBOL(ib_query_qp);
1273
ib_close_qp(struct ib_qp * qp)1274 int ib_close_qp(struct ib_qp *qp)
1275 {
1276 struct ib_qp *real_qp;
1277 unsigned long flags;
1278
1279 real_qp = qp->real_qp;
1280 if (real_qp == qp)
1281 return -EINVAL;
1282
1283 spin_lock_irqsave(&real_qp->device->event_handler_lock, flags);
1284 list_del(&qp->open_list);
1285 spin_unlock_irqrestore(&real_qp->device->event_handler_lock, flags);
1286
1287 atomic_dec(&real_qp->usecnt);
1288 kfree(qp);
1289
1290 return 0;
1291 }
1292 EXPORT_SYMBOL(ib_close_qp);
1293
__ib_destroy_shared_qp(struct ib_qp * qp)1294 static int __ib_destroy_shared_qp(struct ib_qp *qp)
1295 {
1296 struct ib_xrcd *xrcd;
1297 struct ib_qp *real_qp;
1298 int ret;
1299
1300 real_qp = qp->real_qp;
1301 xrcd = real_qp->xrcd;
1302
1303 mutex_lock(&xrcd->tgt_qp_mutex);
1304 ib_close_qp(qp);
1305 if (atomic_read(&real_qp->usecnt) == 0)
1306 list_del(&real_qp->xrcd_list);
1307 else
1308 real_qp = NULL;
1309 mutex_unlock(&xrcd->tgt_qp_mutex);
1310
1311 if (real_qp) {
1312 ret = ib_destroy_qp(real_qp);
1313 if (!ret)
1314 atomic_dec(&xrcd->usecnt);
1315 else
1316 __ib_insert_xrcd_qp(xrcd, real_qp);
1317 }
1318
1319 return 0;
1320 }
1321
ib_destroy_qp(struct ib_qp * qp)1322 int ib_destroy_qp(struct ib_qp *qp)
1323 {
1324 struct ib_pd *pd;
1325 struct ib_cq *scq, *rcq;
1326 struct ib_srq *srq;
1327 struct ib_rwq_ind_table *ind_tbl;
1328 int ret;
1329
1330 if (atomic_read(&qp->usecnt))
1331 return -EBUSY;
1332
1333 if (qp->real_qp != qp)
1334 return __ib_destroy_shared_qp(qp);
1335
1336 pd = qp->pd;
1337 scq = qp->send_cq;
1338 rcq = qp->recv_cq;
1339 srq = qp->srq;
1340 ind_tbl = qp->rwq_ind_tbl;
1341
1342 ret = qp->device->destroy_qp(qp);
1343 if (!ret) {
1344 if (pd)
1345 atomic_dec(&pd->usecnt);
1346 if (scq)
1347 atomic_dec(&scq->usecnt);
1348 if (rcq)
1349 atomic_dec(&rcq->usecnt);
1350 if (srq)
1351 atomic_dec(&srq->usecnt);
1352 if (ind_tbl)
1353 atomic_dec(&ind_tbl->usecnt);
1354 }
1355
1356 return ret;
1357 }
1358 EXPORT_SYMBOL(ib_destroy_qp);
1359
1360 /* Completion queues */
1361
ib_create_cq(struct ib_device * device,ib_comp_handler comp_handler,void (* event_handler)(struct ib_event *,void *),void * cq_context,const struct ib_cq_init_attr * cq_attr)1362 struct ib_cq *ib_create_cq(struct ib_device *device,
1363 ib_comp_handler comp_handler,
1364 void (*event_handler)(struct ib_event *, void *),
1365 void *cq_context,
1366 const struct ib_cq_init_attr *cq_attr)
1367 {
1368 struct ib_cq *cq;
1369
1370 cq = device->create_cq(device, cq_attr, NULL, NULL);
1371
1372 if (!IS_ERR(cq)) {
1373 cq->device = device;
1374 cq->uobject = NULL;
1375 cq->comp_handler = comp_handler;
1376 cq->event_handler = event_handler;
1377 cq->cq_context = cq_context;
1378 atomic_set(&cq->usecnt, 0);
1379 }
1380
1381 return cq;
1382 }
1383 EXPORT_SYMBOL(ib_create_cq);
1384
ib_modify_cq(struct ib_cq * cq,u16 cq_count,u16 cq_period)1385 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
1386 {
1387 return cq->device->modify_cq ?
1388 cq->device->modify_cq(cq, cq_count, cq_period) : -ENOSYS;
1389 }
1390 EXPORT_SYMBOL(ib_modify_cq);
1391
ib_destroy_cq(struct ib_cq * cq)1392 int ib_destroy_cq(struct ib_cq *cq)
1393 {
1394 if (atomic_read(&cq->usecnt))
1395 return -EBUSY;
1396
1397 return cq->device->destroy_cq(cq);
1398 }
1399 EXPORT_SYMBOL(ib_destroy_cq);
1400
ib_resize_cq(struct ib_cq * cq,int cqe)1401 int ib_resize_cq(struct ib_cq *cq, int cqe)
1402 {
1403 return cq->device->resize_cq ?
1404 cq->device->resize_cq(cq, cqe, NULL) : -ENOSYS;
1405 }
1406 EXPORT_SYMBOL(ib_resize_cq);
1407
1408 /* Memory regions */
1409
ib_dereg_mr(struct ib_mr * mr)1410 int ib_dereg_mr(struct ib_mr *mr)
1411 {
1412 struct ib_pd *pd = mr->pd;
1413 int ret;
1414
1415 ret = mr->device->dereg_mr(mr);
1416 if (!ret)
1417 atomic_dec(&pd->usecnt);
1418
1419 return ret;
1420 }
1421 EXPORT_SYMBOL(ib_dereg_mr);
1422
1423 /**
1424 * ib_alloc_mr() - Allocates a memory region
1425 * @pd: protection domain associated with the region
1426 * @mr_type: memory region type
1427 * @max_num_sg: maximum sg entries available for registration.
1428 *
1429 * Notes:
1430 * Memory registeration page/sg lists must not exceed max_num_sg.
1431 * For mr_type IB_MR_TYPE_MEM_REG, the total length cannot exceed
1432 * max_num_sg * used_page_size.
1433 *
1434 */
ib_alloc_mr(struct ib_pd * pd,enum ib_mr_type mr_type,u32 max_num_sg)1435 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
1436 enum ib_mr_type mr_type,
1437 u32 max_num_sg)
1438 {
1439 struct ib_mr *mr;
1440
1441 if (!pd->device->alloc_mr)
1442 return ERR_PTR(-ENOSYS);
1443
1444 mr = pd->device->alloc_mr(pd, mr_type, max_num_sg);
1445 if (!IS_ERR(mr)) {
1446 mr->device = pd->device;
1447 mr->pd = pd;
1448 mr->uobject = NULL;
1449 atomic_inc(&pd->usecnt);
1450 mr->need_inval = false;
1451 }
1452
1453 return mr;
1454 }
1455 EXPORT_SYMBOL(ib_alloc_mr);
1456
1457 /* "Fast" memory regions */
1458
ib_alloc_fmr(struct ib_pd * pd,int mr_access_flags,struct ib_fmr_attr * fmr_attr)1459 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
1460 int mr_access_flags,
1461 struct ib_fmr_attr *fmr_attr)
1462 {
1463 struct ib_fmr *fmr;
1464
1465 if (!pd->device->alloc_fmr)
1466 return ERR_PTR(-ENOSYS);
1467
1468 fmr = pd->device->alloc_fmr(pd, mr_access_flags, fmr_attr);
1469 if (!IS_ERR(fmr)) {
1470 fmr->device = pd->device;
1471 fmr->pd = pd;
1472 atomic_inc(&pd->usecnt);
1473 }
1474
1475 return fmr;
1476 }
1477 EXPORT_SYMBOL(ib_alloc_fmr);
1478
ib_unmap_fmr(struct list_head * fmr_list)1479 int ib_unmap_fmr(struct list_head *fmr_list)
1480 {
1481 struct ib_fmr *fmr;
1482
1483 if (list_empty(fmr_list))
1484 return 0;
1485
1486 fmr = list_entry(fmr_list->next, struct ib_fmr, list);
1487 return fmr->device->unmap_fmr(fmr_list);
1488 }
1489 EXPORT_SYMBOL(ib_unmap_fmr);
1490
ib_dealloc_fmr(struct ib_fmr * fmr)1491 int ib_dealloc_fmr(struct ib_fmr *fmr)
1492 {
1493 struct ib_pd *pd;
1494 int ret;
1495
1496 pd = fmr->pd;
1497 ret = fmr->device->dealloc_fmr(fmr);
1498 if (!ret)
1499 atomic_dec(&pd->usecnt);
1500
1501 return ret;
1502 }
1503 EXPORT_SYMBOL(ib_dealloc_fmr);
1504
1505 /* Multicast groups */
1506
is_valid_mcast_lid(struct ib_qp * qp,u16 lid)1507 static bool is_valid_mcast_lid(struct ib_qp *qp, u16 lid)
1508 {
1509 struct ib_qp_init_attr init_attr = {};
1510 struct ib_qp_attr attr = {};
1511 int num_eth_ports = 0;
1512 int port;
1513
1514 /* If QP state >= init, it is assigned to a port and we can check this
1515 * port only.
1516 */
1517 if (!ib_query_qp(qp, &attr, IB_QP_STATE | IB_QP_PORT, &init_attr)) {
1518 if (attr.qp_state >= IB_QPS_INIT) {
1519 if (rdma_port_get_link_layer(qp->device, attr.port_num) !=
1520 IB_LINK_LAYER_INFINIBAND)
1521 return true;
1522 goto lid_check;
1523 }
1524 }
1525
1526 /* Can't get a quick answer, iterate over all ports */
1527 for (port = 0; port < qp->device->phys_port_cnt; port++)
1528 if (rdma_port_get_link_layer(qp->device, port) !=
1529 IB_LINK_LAYER_INFINIBAND)
1530 num_eth_ports++;
1531
1532 /* If we have at lease one Ethernet port, RoCE annex declares that
1533 * multicast LID should be ignored. We can't tell at this step if the
1534 * QP belongs to an IB or Ethernet port.
1535 */
1536 if (num_eth_ports)
1537 return true;
1538
1539 /* If all the ports are IB, we can check according to IB spec. */
1540 lid_check:
1541 return !(lid < be16_to_cpu(IB_MULTICAST_LID_BASE) ||
1542 lid == be16_to_cpu(IB_LID_PERMISSIVE));
1543 }
1544
ib_attach_mcast(struct ib_qp * qp,union ib_gid * gid,u16 lid)1545 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1546 {
1547 int ret;
1548
1549 if (!qp->device->attach_mcast)
1550 return -ENOSYS;
1551
1552 if (!rdma_is_multicast_addr((struct in6_addr *)gid->raw) ||
1553 qp->qp_type != IB_QPT_UD || !is_valid_mcast_lid(qp, lid))
1554 return -EINVAL;
1555
1556 ret = qp->device->attach_mcast(qp, gid, lid);
1557 if (!ret)
1558 atomic_inc(&qp->usecnt);
1559 return ret;
1560 }
1561 EXPORT_SYMBOL(ib_attach_mcast);
1562
ib_detach_mcast(struct ib_qp * qp,union ib_gid * gid,u16 lid)1563 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid)
1564 {
1565 int ret;
1566
1567 if (!qp->device->detach_mcast)
1568 return -ENOSYS;
1569
1570 if (!rdma_is_multicast_addr((struct in6_addr *)gid->raw) ||
1571 qp->qp_type != IB_QPT_UD || !is_valid_mcast_lid(qp, lid))
1572 return -EINVAL;
1573
1574 ret = qp->device->detach_mcast(qp, gid, lid);
1575 if (!ret)
1576 atomic_dec(&qp->usecnt);
1577 return ret;
1578 }
1579 EXPORT_SYMBOL(ib_detach_mcast);
1580
ib_alloc_xrcd(struct ib_device * device)1581 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device)
1582 {
1583 struct ib_xrcd *xrcd;
1584
1585 if (!device->alloc_xrcd)
1586 return ERR_PTR(-ENOSYS);
1587
1588 xrcd = device->alloc_xrcd(device, NULL, NULL);
1589 if (!IS_ERR(xrcd)) {
1590 xrcd->device = device;
1591 xrcd->inode = NULL;
1592 atomic_set(&xrcd->usecnt, 0);
1593 mutex_init(&xrcd->tgt_qp_mutex);
1594 INIT_LIST_HEAD(&xrcd->tgt_qp_list);
1595 }
1596
1597 return xrcd;
1598 }
1599 EXPORT_SYMBOL(ib_alloc_xrcd);
1600
ib_dealloc_xrcd(struct ib_xrcd * xrcd)1601 int ib_dealloc_xrcd(struct ib_xrcd *xrcd)
1602 {
1603 struct ib_qp *qp;
1604 int ret;
1605
1606 if (atomic_read(&xrcd->usecnt))
1607 return -EBUSY;
1608
1609 while (!list_empty(&xrcd->tgt_qp_list)) {
1610 qp = list_entry(xrcd->tgt_qp_list.next, struct ib_qp, xrcd_list);
1611 ret = ib_destroy_qp(qp);
1612 if (ret)
1613 return ret;
1614 }
1615
1616 return xrcd->device->dealloc_xrcd(xrcd);
1617 }
1618 EXPORT_SYMBOL(ib_dealloc_xrcd);
1619
1620 /**
1621 * ib_create_wq - Creates a WQ associated with the specified protection
1622 * domain.
1623 * @pd: The protection domain associated with the WQ.
1624 * @wq_init_attr: A list of initial attributes required to create the
1625 * WQ. If WQ creation succeeds, then the attributes are updated to
1626 * the actual capabilities of the created WQ.
1627 *
1628 * wq_init_attr->max_wr and wq_init_attr->max_sge determine
1629 * the requested size of the WQ, and set to the actual values allocated
1630 * on return.
1631 * If ib_create_wq() succeeds, then max_wr and max_sge will always be
1632 * at least as large as the requested values.
1633 */
ib_create_wq(struct ib_pd * pd,struct ib_wq_init_attr * wq_attr)1634 struct ib_wq *ib_create_wq(struct ib_pd *pd,
1635 struct ib_wq_init_attr *wq_attr)
1636 {
1637 struct ib_wq *wq;
1638
1639 if (!pd->device->create_wq)
1640 return ERR_PTR(-ENOSYS);
1641
1642 wq = pd->device->create_wq(pd, wq_attr, NULL);
1643 if (!IS_ERR(wq)) {
1644 wq->event_handler = wq_attr->event_handler;
1645 wq->wq_context = wq_attr->wq_context;
1646 wq->wq_type = wq_attr->wq_type;
1647 wq->cq = wq_attr->cq;
1648 wq->device = pd->device;
1649 wq->pd = pd;
1650 wq->uobject = NULL;
1651 atomic_inc(&pd->usecnt);
1652 atomic_inc(&wq_attr->cq->usecnt);
1653 atomic_set(&wq->usecnt, 0);
1654 }
1655 return wq;
1656 }
1657 EXPORT_SYMBOL(ib_create_wq);
1658
1659 /**
1660 * ib_destroy_wq - Destroys the specified WQ.
1661 * @wq: The WQ to destroy.
1662 */
ib_destroy_wq(struct ib_wq * wq)1663 int ib_destroy_wq(struct ib_wq *wq)
1664 {
1665 int err;
1666 struct ib_cq *cq = wq->cq;
1667 struct ib_pd *pd = wq->pd;
1668
1669 if (atomic_read(&wq->usecnt))
1670 return -EBUSY;
1671
1672 err = wq->device->destroy_wq(wq);
1673 if (!err) {
1674 atomic_dec(&pd->usecnt);
1675 atomic_dec(&cq->usecnt);
1676 }
1677 return err;
1678 }
1679 EXPORT_SYMBOL(ib_destroy_wq);
1680
1681 /**
1682 * ib_modify_wq - Modifies the specified WQ.
1683 * @wq: The WQ to modify.
1684 * @wq_attr: On input, specifies the WQ attributes to modify.
1685 * @wq_attr_mask: A bit-mask used to specify which attributes of the WQ
1686 * are being modified.
1687 * On output, the current values of selected WQ attributes are returned.
1688 */
ib_modify_wq(struct ib_wq * wq,struct ib_wq_attr * wq_attr,u32 wq_attr_mask)1689 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *wq_attr,
1690 u32 wq_attr_mask)
1691 {
1692 int err;
1693
1694 if (!wq->device->modify_wq)
1695 return -ENOSYS;
1696
1697 err = wq->device->modify_wq(wq, wq_attr, wq_attr_mask, NULL);
1698 return err;
1699 }
1700 EXPORT_SYMBOL(ib_modify_wq);
1701
1702 /*
1703 * ib_create_rwq_ind_table - Creates a RQ Indirection Table.
1704 * @device: The device on which to create the rwq indirection table.
1705 * @ib_rwq_ind_table_init_attr: A list of initial attributes required to
1706 * create the Indirection Table.
1707 *
1708 * Note: The life time of ib_rwq_ind_table_init_attr->ind_tbl is not less
1709 * than the created ib_rwq_ind_table object and the caller is responsible
1710 * for its memory allocation/free.
1711 */
ib_create_rwq_ind_table(struct ib_device * device,struct ib_rwq_ind_table_init_attr * init_attr)1712 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
1713 struct ib_rwq_ind_table_init_attr *init_attr)
1714 {
1715 struct ib_rwq_ind_table *rwq_ind_table;
1716 int i;
1717 u32 table_size;
1718
1719 if (!device->create_rwq_ind_table)
1720 return ERR_PTR(-ENOSYS);
1721
1722 table_size = (1 << init_attr->log_ind_tbl_size);
1723 rwq_ind_table = device->create_rwq_ind_table(device,
1724 init_attr, NULL);
1725 if (IS_ERR(rwq_ind_table))
1726 return rwq_ind_table;
1727
1728 rwq_ind_table->ind_tbl = init_attr->ind_tbl;
1729 rwq_ind_table->log_ind_tbl_size = init_attr->log_ind_tbl_size;
1730 rwq_ind_table->device = device;
1731 rwq_ind_table->uobject = NULL;
1732 atomic_set(&rwq_ind_table->usecnt, 0);
1733
1734 for (i = 0; i < table_size; i++)
1735 atomic_inc(&rwq_ind_table->ind_tbl[i]->usecnt);
1736
1737 return rwq_ind_table;
1738 }
1739 EXPORT_SYMBOL(ib_create_rwq_ind_table);
1740
1741 /*
1742 * ib_destroy_rwq_ind_table - Destroys the specified Indirection Table.
1743 * @wq_ind_table: The Indirection Table to destroy.
1744 */
ib_destroy_rwq_ind_table(struct ib_rwq_ind_table * rwq_ind_table)1745 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *rwq_ind_table)
1746 {
1747 int err, i;
1748 u32 table_size = (1 << rwq_ind_table->log_ind_tbl_size);
1749 struct ib_wq **ind_tbl = rwq_ind_table->ind_tbl;
1750
1751 if (atomic_read(&rwq_ind_table->usecnt))
1752 return -EBUSY;
1753
1754 err = rwq_ind_table->device->destroy_rwq_ind_table(rwq_ind_table);
1755 if (!err) {
1756 for (i = 0; i < table_size; i++)
1757 atomic_dec(&ind_tbl[i]->usecnt);
1758 }
1759
1760 return err;
1761 }
1762 EXPORT_SYMBOL(ib_destroy_rwq_ind_table);
1763
ib_create_flow(struct ib_qp * qp,struct ib_flow_attr * flow_attr,int domain)1764 struct ib_flow *ib_create_flow(struct ib_qp *qp,
1765 struct ib_flow_attr *flow_attr,
1766 int domain)
1767 {
1768 struct ib_flow *flow_id;
1769 if (!qp->device->create_flow)
1770 return ERR_PTR(-ENOSYS);
1771
1772 flow_id = qp->device->create_flow(qp, flow_attr, domain);
1773 if (!IS_ERR(flow_id))
1774 atomic_inc(&qp->usecnt);
1775 return flow_id;
1776 }
1777 EXPORT_SYMBOL(ib_create_flow);
1778
ib_destroy_flow(struct ib_flow * flow_id)1779 int ib_destroy_flow(struct ib_flow *flow_id)
1780 {
1781 int err;
1782 struct ib_qp *qp = flow_id->qp;
1783
1784 err = qp->device->destroy_flow(flow_id);
1785 if (!err)
1786 atomic_dec(&qp->usecnt);
1787 return err;
1788 }
1789 EXPORT_SYMBOL(ib_destroy_flow);
1790
ib_check_mr_status(struct ib_mr * mr,u32 check_mask,struct ib_mr_status * mr_status)1791 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
1792 struct ib_mr_status *mr_status)
1793 {
1794 return mr->device->check_mr_status ?
1795 mr->device->check_mr_status(mr, check_mask, mr_status) : -ENOSYS;
1796 }
1797 EXPORT_SYMBOL(ib_check_mr_status);
1798
ib_set_vf_link_state(struct ib_device * device,int vf,u8 port,int state)1799 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
1800 int state)
1801 {
1802 if (!device->set_vf_link_state)
1803 return -ENOSYS;
1804
1805 return device->set_vf_link_state(device, vf, port, state);
1806 }
1807 EXPORT_SYMBOL(ib_set_vf_link_state);
1808
ib_get_vf_config(struct ib_device * device,int vf,u8 port,struct ifla_vf_info * info)1809 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
1810 struct ifla_vf_info *info)
1811 {
1812 if (!device->get_vf_config)
1813 return -ENOSYS;
1814
1815 return device->get_vf_config(device, vf, port, info);
1816 }
1817 EXPORT_SYMBOL(ib_get_vf_config);
1818
ib_get_vf_stats(struct ib_device * device,int vf,u8 port,struct ifla_vf_stats * stats)1819 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
1820 struct ifla_vf_stats *stats)
1821 {
1822 if (!device->get_vf_stats)
1823 return -ENOSYS;
1824
1825 return device->get_vf_stats(device, vf, port, stats);
1826 }
1827 EXPORT_SYMBOL(ib_get_vf_stats);
1828
ib_set_vf_guid(struct ib_device * device,int vf,u8 port,u64 guid,int type)1829 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
1830 int type)
1831 {
1832 if (!device->set_vf_guid)
1833 return -ENOSYS;
1834
1835 return device->set_vf_guid(device, vf, port, guid, type);
1836 }
1837 EXPORT_SYMBOL(ib_set_vf_guid);
1838
1839 /**
1840 * ib_map_mr_sg() - Map the largest prefix of a dma mapped SG list
1841 * and set it the memory region.
1842 * @mr: memory region
1843 * @sg: dma mapped scatterlist
1844 * @sg_nents: number of entries in sg
1845 * @sg_offset: offset in bytes into sg
1846 * @page_size: page vector desired page size
1847 *
1848 * Constraints:
1849 * - The first sg element is allowed to have an offset.
1850 * - Each sg element must either be aligned to page_size or virtually
1851 * contiguous to the previous element. In case an sg element has a
1852 * non-contiguous offset, the mapping prefix will not include it.
1853 * - The last sg element is allowed to have length less than page_size.
1854 * - If sg_nents total byte length exceeds the mr max_num_sge * page_size
1855 * then only max_num_sg entries will be mapped.
1856 * - If the MR was allocated with type IB_MR_TYPE_SG_GAPS, none of these
1857 * constraints holds and the page_size argument is ignored.
1858 *
1859 * Returns the number of sg elements that were mapped to the memory region.
1860 *
1861 * After this completes successfully, the memory region
1862 * is ready for registration.
1863 */
ib_map_mr_sg(struct ib_mr * mr,struct scatterlist * sg,int sg_nents,unsigned int * sg_offset,unsigned int page_size)1864 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
1865 unsigned int *sg_offset, unsigned int page_size)
1866 {
1867 if (unlikely(!mr->device->map_mr_sg))
1868 return -ENOSYS;
1869
1870 mr->page_size = page_size;
1871
1872 return mr->device->map_mr_sg(mr, sg, sg_nents, sg_offset);
1873 }
1874 EXPORT_SYMBOL(ib_map_mr_sg);
1875
1876 /**
1877 * ib_sg_to_pages() - Convert the largest prefix of a sg list
1878 * to a page vector
1879 * @mr: memory region
1880 * @sgl: dma mapped scatterlist
1881 * @sg_nents: number of entries in sg
1882 * @sg_offset_p: IN: start offset in bytes into sg
1883 * OUT: offset in bytes for element n of the sg of the first
1884 * byte that has not been processed where n is the return
1885 * value of this function.
1886 * @set_page: driver page assignment function pointer
1887 *
1888 * Core service helper for drivers to convert the largest
1889 * prefix of given sg list to a page vector. The sg list
1890 * prefix converted is the prefix that meet the requirements
1891 * of ib_map_mr_sg.
1892 *
1893 * Returns the number of sg elements that were assigned to
1894 * a page vector.
1895 */
ib_sg_to_pages(struct ib_mr * mr,struct scatterlist * sgl,int sg_nents,unsigned int * sg_offset_p,int (* set_page)(struct ib_mr *,u64))1896 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
1897 unsigned int *sg_offset_p, int (*set_page)(struct ib_mr *, u64))
1898 {
1899 struct scatterlist *sg;
1900 u64 last_end_dma_addr = 0;
1901 unsigned int sg_offset = sg_offset_p ? *sg_offset_p : 0;
1902 unsigned int last_page_off = 0;
1903 u64 page_mask = ~((u64)mr->page_size - 1);
1904 int i, ret;
1905
1906 if (unlikely(sg_nents <= 0 || sg_offset > sg_dma_len(&sgl[0])))
1907 return -EINVAL;
1908
1909 mr->iova = sg_dma_address(&sgl[0]) + sg_offset;
1910 mr->length = 0;
1911
1912 for_each_sg(sgl, sg, sg_nents, i) {
1913 u64 dma_addr = sg_dma_address(sg) + sg_offset;
1914 u64 prev_addr = dma_addr;
1915 unsigned int dma_len = sg_dma_len(sg) - sg_offset;
1916 u64 end_dma_addr = dma_addr + dma_len;
1917 u64 page_addr = dma_addr & page_mask;
1918
1919 /*
1920 * For the second and later elements, check whether either the
1921 * end of element i-1 or the start of element i is not aligned
1922 * on a page boundary.
1923 */
1924 if (i && (last_page_off != 0 || page_addr != dma_addr)) {
1925 /* Stop mapping if there is a gap. */
1926 if (last_end_dma_addr != dma_addr)
1927 break;
1928
1929 /*
1930 * Coalesce this element with the last. If it is small
1931 * enough just update mr->length. Otherwise start
1932 * mapping from the next page.
1933 */
1934 goto next_page;
1935 }
1936
1937 do {
1938 ret = set_page(mr, page_addr);
1939 if (unlikely(ret < 0)) {
1940 sg_offset = prev_addr - sg_dma_address(sg);
1941 mr->length += prev_addr - dma_addr;
1942 if (sg_offset_p)
1943 *sg_offset_p = sg_offset;
1944 return i || sg_offset ? i : ret;
1945 }
1946 prev_addr = page_addr;
1947 next_page:
1948 page_addr += mr->page_size;
1949 } while (page_addr < end_dma_addr);
1950
1951 mr->length += dma_len;
1952 last_end_dma_addr = end_dma_addr;
1953 last_page_off = end_dma_addr & ~page_mask;
1954
1955 sg_offset = 0;
1956 }
1957
1958 if (sg_offset_p)
1959 *sg_offset_p = 0;
1960 return i;
1961 }
1962 EXPORT_SYMBOL(ib_sg_to_pages);
1963
1964 struct ib_drain_cqe {
1965 struct ib_cqe cqe;
1966 struct completion done;
1967 };
1968
ib_drain_qp_done(struct ib_cq * cq,struct ib_wc * wc)1969 static void ib_drain_qp_done(struct ib_cq *cq, struct ib_wc *wc)
1970 {
1971 struct ib_drain_cqe *cqe = container_of(wc->wr_cqe, struct ib_drain_cqe,
1972 cqe);
1973
1974 complete(&cqe->done);
1975 }
1976
1977 /*
1978 * Post a WR and block until its completion is reaped for the SQ.
1979 */
__ib_drain_sq(struct ib_qp * qp)1980 static void __ib_drain_sq(struct ib_qp *qp)
1981 {
1982 struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
1983 struct ib_drain_cqe sdrain;
1984 const struct ib_send_wr *bad_swr;
1985 struct ib_rdma_wr swr = {
1986 .wr = {
1987 .opcode = IB_WR_RDMA_WRITE,
1988 .wr_cqe = &sdrain.cqe,
1989 },
1990 };
1991 int ret;
1992
1993 if (qp->send_cq->poll_ctx == IB_POLL_DIRECT) {
1994 WARN_ONCE(qp->send_cq->poll_ctx == IB_POLL_DIRECT,
1995 "IB_POLL_DIRECT poll_ctx not supported for drain\n");
1996 return;
1997 }
1998
1999 sdrain.cqe.done = ib_drain_qp_done;
2000 init_completion(&sdrain.done);
2001
2002 ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
2003 if (ret) {
2004 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
2005 return;
2006 }
2007
2008 ret = ib_post_send(qp, &swr.wr, &bad_swr);
2009 if (ret) {
2010 WARN_ONCE(ret, "failed to drain send queue: %d\n", ret);
2011 return;
2012 }
2013
2014 wait_for_completion(&sdrain.done);
2015 }
2016
2017 /*
2018 * Post a WR and block until its completion is reaped for the RQ.
2019 */
__ib_drain_rq(struct ib_qp * qp)2020 static void __ib_drain_rq(struct ib_qp *qp)
2021 {
2022 struct ib_qp_attr attr = { .qp_state = IB_QPS_ERR };
2023 struct ib_drain_cqe rdrain;
2024 struct ib_recv_wr rwr = {};
2025 const struct ib_recv_wr *bad_rwr;
2026 int ret;
2027
2028 if (qp->recv_cq->poll_ctx == IB_POLL_DIRECT) {
2029 WARN_ONCE(qp->recv_cq->poll_ctx == IB_POLL_DIRECT,
2030 "IB_POLL_DIRECT poll_ctx not supported for drain\n");
2031 return;
2032 }
2033
2034 rwr.wr_cqe = &rdrain.cqe;
2035 rdrain.cqe.done = ib_drain_qp_done;
2036 init_completion(&rdrain.done);
2037
2038 ret = ib_modify_qp(qp, &attr, IB_QP_STATE);
2039 if (ret) {
2040 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
2041 return;
2042 }
2043
2044 ret = ib_post_recv(qp, &rwr, &bad_rwr);
2045 if (ret) {
2046 WARN_ONCE(ret, "failed to drain recv queue: %d\n", ret);
2047 return;
2048 }
2049
2050 wait_for_completion(&rdrain.done);
2051 }
2052
2053 /**
2054 * ib_drain_sq() - Block until all SQ CQEs have been consumed by the
2055 * application.
2056 * @qp: queue pair to drain
2057 *
2058 * If the device has a provider-specific drain function, then
2059 * call that. Otherwise call the generic drain function
2060 * __ib_drain_sq().
2061 *
2062 * The caller must:
2063 *
2064 * ensure there is room in the CQ and SQ for the drain work request and
2065 * completion.
2066 *
2067 * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
2068 * IB_POLL_DIRECT.
2069 *
2070 * ensure that there are no other contexts that are posting WRs concurrently.
2071 * Otherwise the drain is not guaranteed.
2072 */
ib_drain_sq(struct ib_qp * qp)2073 void ib_drain_sq(struct ib_qp *qp)
2074 {
2075 if (qp->device->drain_sq)
2076 qp->device->drain_sq(qp);
2077 else
2078 __ib_drain_sq(qp);
2079 }
2080 EXPORT_SYMBOL(ib_drain_sq);
2081
2082 /**
2083 * ib_drain_rq() - Block until all RQ CQEs have been consumed by the
2084 * application.
2085 * @qp: queue pair to drain
2086 *
2087 * If the device has a provider-specific drain function, then
2088 * call that. Otherwise call the generic drain function
2089 * __ib_drain_rq().
2090 *
2091 * The caller must:
2092 *
2093 * ensure there is room in the CQ and RQ for the drain work request and
2094 * completion.
2095 *
2096 * allocate the CQ using ib_alloc_cq() and the CQ poll context cannot be
2097 * IB_POLL_DIRECT.
2098 *
2099 * ensure that there are no other contexts that are posting WRs concurrently.
2100 * Otherwise the drain is not guaranteed.
2101 */
ib_drain_rq(struct ib_qp * qp)2102 void ib_drain_rq(struct ib_qp *qp)
2103 {
2104 if (qp->device->drain_rq)
2105 qp->device->drain_rq(qp);
2106 else
2107 __ib_drain_rq(qp);
2108 }
2109 EXPORT_SYMBOL(ib_drain_rq);
2110
2111 /**
2112 * ib_drain_qp() - Block until all CQEs have been consumed by the
2113 * application on both the RQ and SQ.
2114 * @qp: queue pair to drain
2115 *
2116 * The caller must:
2117 *
2118 * ensure there is room in the CQ(s), SQ, and RQ for drain work requests
2119 * and completions.
2120 *
2121 * allocate the CQs using ib_alloc_cq() and the CQ poll context cannot be
2122 * IB_POLL_DIRECT.
2123 *
2124 * ensure that there are no other contexts that are posting WRs concurrently.
2125 * Otherwise the drain is not guaranteed.
2126 */
ib_drain_qp(struct ib_qp * qp)2127 void ib_drain_qp(struct ib_qp *qp)
2128 {
2129 ib_drain_sq(qp);
2130 if (!qp->srq)
2131 ib_drain_rq(qp);
2132 }
2133 EXPORT_SYMBOL(ib_drain_qp);
2134