xref: /linux/include/rdma/ib_verbs.h (revision c0c914eca7f251c70facc37dfebeaf176601918d)
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, 2007 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 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 
59 #include <linux/atomic.h>
60 #include <linux/mmu_notifier.h>
61 #include <asm/uaccess.h>
62 
63 extern struct workqueue_struct *ib_wq;
64 extern struct workqueue_struct *ib_comp_wq;
65 
66 union ib_gid {
67 	u8	raw[16];
68 	struct {
69 		__be64	subnet_prefix;
70 		__be64	interface_id;
71 	} global;
72 };
73 
74 extern union ib_gid zgid;
75 
76 enum ib_gid_type {
77 	/* If link layer is Ethernet, this is RoCE V1 */
78 	IB_GID_TYPE_IB        = 0,
79 	IB_GID_TYPE_ROCE      = 0,
80 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
81 	IB_GID_TYPE_SIZE
82 };
83 
84 #define ROCE_V2_UDP_DPORT      4791
85 struct ib_gid_attr {
86 	enum ib_gid_type	gid_type;
87 	struct net_device	*ndev;
88 };
89 
90 enum rdma_node_type {
91 	/* IB values map to NodeInfo:NodeType. */
92 	RDMA_NODE_IB_CA 	= 1,
93 	RDMA_NODE_IB_SWITCH,
94 	RDMA_NODE_IB_ROUTER,
95 	RDMA_NODE_RNIC,
96 	RDMA_NODE_USNIC,
97 	RDMA_NODE_USNIC_UDP,
98 };
99 
100 enum rdma_transport_type {
101 	RDMA_TRANSPORT_IB,
102 	RDMA_TRANSPORT_IWARP,
103 	RDMA_TRANSPORT_USNIC,
104 	RDMA_TRANSPORT_USNIC_UDP
105 };
106 
107 enum rdma_protocol_type {
108 	RDMA_PROTOCOL_IB,
109 	RDMA_PROTOCOL_IBOE,
110 	RDMA_PROTOCOL_IWARP,
111 	RDMA_PROTOCOL_USNIC_UDP
112 };
113 
114 __attribute_const__ enum rdma_transport_type
115 rdma_node_get_transport(enum rdma_node_type node_type);
116 
117 enum rdma_network_type {
118 	RDMA_NETWORK_IB,
119 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
120 	RDMA_NETWORK_IPV4,
121 	RDMA_NETWORK_IPV6
122 };
123 
124 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
125 {
126 	if (network_type == RDMA_NETWORK_IPV4 ||
127 	    network_type == RDMA_NETWORK_IPV6)
128 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
129 
130 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
131 	return IB_GID_TYPE_IB;
132 }
133 
134 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
135 							    union ib_gid *gid)
136 {
137 	if (gid_type == IB_GID_TYPE_IB)
138 		return RDMA_NETWORK_IB;
139 
140 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
141 		return RDMA_NETWORK_IPV4;
142 	else
143 		return RDMA_NETWORK_IPV6;
144 }
145 
146 enum rdma_link_layer {
147 	IB_LINK_LAYER_UNSPECIFIED,
148 	IB_LINK_LAYER_INFINIBAND,
149 	IB_LINK_LAYER_ETHERNET,
150 };
151 
152 enum ib_device_cap_flags {
153 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
154 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
155 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
156 	IB_DEVICE_RAW_MULTI			= (1 << 3),
157 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
158 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
159 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
160 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
161 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
162 	IB_DEVICE_INIT_TYPE			= (1 << 9),
163 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
164 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
165 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
166 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
167 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
168 
169 	/*
170 	 * This device supports a per-device lkey or stag that can be
171 	 * used without performing a memory registration for the local
172 	 * memory.  Note that ULPs should never check this flag, but
173 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
174 	 * which will always contain a usable lkey.
175 	 */
176 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
177 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
178 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
179 	/*
180 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
181 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
182 	 * messages and can verify the validity of checksum for
183 	 * incoming messages.  Setting this flag implies that the
184 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
185 	 */
186 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
187 	IB_DEVICE_UD_TSO			= (1 << 19),
188 	IB_DEVICE_XRC				= (1 << 20),
189 
190 	/*
191 	 * This device supports the IB "base memory management extension",
192 	 * which includes support for fast registrations (IB_WR_REG_MR,
193 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
194 	 * also be set by any iWarp device which must support FRs to comply
195 	 * to the iWarp verbs spec.  iWarp devices also support the
196 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
197 	 * stag.
198 	 */
199 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
200 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
201 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
202 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
203 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
204 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
205 	/*
206 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
207 	 * support execution of WQEs that involve synchronization
208 	 * of I/O operations with single completion queue managed
209 	 * by hardware.
210 	 */
211 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
212 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
213 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
214 	IB_DEVICE_ON_DEMAND_PAGING		= (1 << 31),
215 };
216 
217 enum ib_signature_prot_cap {
218 	IB_PROT_T10DIF_TYPE_1 = 1,
219 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
220 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
221 };
222 
223 enum ib_signature_guard_cap {
224 	IB_GUARD_T10DIF_CRC	= 1,
225 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
226 };
227 
228 enum ib_atomic_cap {
229 	IB_ATOMIC_NONE,
230 	IB_ATOMIC_HCA,
231 	IB_ATOMIC_GLOB
232 };
233 
234 enum ib_odp_general_cap_bits {
235 	IB_ODP_SUPPORT = 1 << 0,
236 };
237 
238 enum ib_odp_transport_cap_bits {
239 	IB_ODP_SUPPORT_SEND	= 1 << 0,
240 	IB_ODP_SUPPORT_RECV	= 1 << 1,
241 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
242 	IB_ODP_SUPPORT_READ	= 1 << 3,
243 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
244 };
245 
246 struct ib_odp_caps {
247 	uint64_t general_caps;
248 	struct {
249 		uint32_t  rc_odp_caps;
250 		uint32_t  uc_odp_caps;
251 		uint32_t  ud_odp_caps;
252 	} per_transport_caps;
253 };
254 
255 enum ib_cq_creation_flags {
256 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
257 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
258 };
259 
260 struct ib_cq_init_attr {
261 	unsigned int	cqe;
262 	int		comp_vector;
263 	u32		flags;
264 };
265 
266 struct ib_device_attr {
267 	u64			fw_ver;
268 	__be64			sys_image_guid;
269 	u64			max_mr_size;
270 	u64			page_size_cap;
271 	u32			vendor_id;
272 	u32			vendor_part_id;
273 	u32			hw_ver;
274 	int			max_qp;
275 	int			max_qp_wr;
276 	int			device_cap_flags;
277 	int			max_sge;
278 	int			max_sge_rd;
279 	int			max_cq;
280 	int			max_cqe;
281 	int			max_mr;
282 	int			max_pd;
283 	int			max_qp_rd_atom;
284 	int			max_ee_rd_atom;
285 	int			max_res_rd_atom;
286 	int			max_qp_init_rd_atom;
287 	int			max_ee_init_rd_atom;
288 	enum ib_atomic_cap	atomic_cap;
289 	enum ib_atomic_cap	masked_atomic_cap;
290 	int			max_ee;
291 	int			max_rdd;
292 	int			max_mw;
293 	int			max_raw_ipv6_qp;
294 	int			max_raw_ethy_qp;
295 	int			max_mcast_grp;
296 	int			max_mcast_qp_attach;
297 	int			max_total_mcast_qp_attach;
298 	int			max_ah;
299 	int			max_fmr;
300 	int			max_map_per_fmr;
301 	int			max_srq;
302 	int			max_srq_wr;
303 	int			max_srq_sge;
304 	unsigned int		max_fast_reg_page_list_len;
305 	u16			max_pkeys;
306 	u8			local_ca_ack_delay;
307 	int			sig_prot_cap;
308 	int			sig_guard_cap;
309 	struct ib_odp_caps	odp_caps;
310 	uint64_t		timestamp_mask;
311 	uint64_t		hca_core_clock; /* in KHZ */
312 };
313 
314 enum ib_mtu {
315 	IB_MTU_256  = 1,
316 	IB_MTU_512  = 2,
317 	IB_MTU_1024 = 3,
318 	IB_MTU_2048 = 4,
319 	IB_MTU_4096 = 5
320 };
321 
322 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
323 {
324 	switch (mtu) {
325 	case IB_MTU_256:  return  256;
326 	case IB_MTU_512:  return  512;
327 	case IB_MTU_1024: return 1024;
328 	case IB_MTU_2048: return 2048;
329 	case IB_MTU_4096: return 4096;
330 	default: 	  return -1;
331 	}
332 }
333 
334 enum ib_port_state {
335 	IB_PORT_NOP		= 0,
336 	IB_PORT_DOWN		= 1,
337 	IB_PORT_INIT		= 2,
338 	IB_PORT_ARMED		= 3,
339 	IB_PORT_ACTIVE		= 4,
340 	IB_PORT_ACTIVE_DEFER	= 5
341 };
342 
343 enum ib_port_cap_flags {
344 	IB_PORT_SM				= 1 <<  1,
345 	IB_PORT_NOTICE_SUP			= 1 <<  2,
346 	IB_PORT_TRAP_SUP			= 1 <<  3,
347 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
348 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
349 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
350 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
351 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
352 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
353 	IB_PORT_SM_DISABLED			= 1 << 10,
354 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
355 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
356 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
357 	IB_PORT_CM_SUP				= 1 << 16,
358 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
359 	IB_PORT_REINIT_SUP			= 1 << 18,
360 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
361 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
362 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
363 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
364 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
365 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
366 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
367 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
368 };
369 
370 enum ib_port_width {
371 	IB_WIDTH_1X	= 1,
372 	IB_WIDTH_4X	= 2,
373 	IB_WIDTH_8X	= 4,
374 	IB_WIDTH_12X	= 8
375 };
376 
377 static inline int ib_width_enum_to_int(enum ib_port_width width)
378 {
379 	switch (width) {
380 	case IB_WIDTH_1X:  return  1;
381 	case IB_WIDTH_4X:  return  4;
382 	case IB_WIDTH_8X:  return  8;
383 	case IB_WIDTH_12X: return 12;
384 	default: 	  return -1;
385 	}
386 }
387 
388 enum ib_port_speed {
389 	IB_SPEED_SDR	= 1,
390 	IB_SPEED_DDR	= 2,
391 	IB_SPEED_QDR	= 4,
392 	IB_SPEED_FDR10	= 8,
393 	IB_SPEED_FDR	= 16,
394 	IB_SPEED_EDR	= 32
395 };
396 
397 struct ib_protocol_stats {
398 	/* TBD... */
399 };
400 
401 struct iw_protocol_stats {
402 	u64	ipInReceives;
403 	u64	ipInHdrErrors;
404 	u64	ipInTooBigErrors;
405 	u64	ipInNoRoutes;
406 	u64	ipInAddrErrors;
407 	u64	ipInUnknownProtos;
408 	u64	ipInTruncatedPkts;
409 	u64	ipInDiscards;
410 	u64	ipInDelivers;
411 	u64	ipOutForwDatagrams;
412 	u64	ipOutRequests;
413 	u64	ipOutDiscards;
414 	u64	ipOutNoRoutes;
415 	u64	ipReasmTimeout;
416 	u64	ipReasmReqds;
417 	u64	ipReasmOKs;
418 	u64	ipReasmFails;
419 	u64	ipFragOKs;
420 	u64	ipFragFails;
421 	u64	ipFragCreates;
422 	u64	ipInMcastPkts;
423 	u64	ipOutMcastPkts;
424 	u64	ipInBcastPkts;
425 	u64	ipOutBcastPkts;
426 
427 	u64	tcpRtoAlgorithm;
428 	u64	tcpRtoMin;
429 	u64	tcpRtoMax;
430 	u64	tcpMaxConn;
431 	u64	tcpActiveOpens;
432 	u64	tcpPassiveOpens;
433 	u64	tcpAttemptFails;
434 	u64	tcpEstabResets;
435 	u64	tcpCurrEstab;
436 	u64	tcpInSegs;
437 	u64	tcpOutSegs;
438 	u64	tcpRetransSegs;
439 	u64	tcpInErrs;
440 	u64	tcpOutRsts;
441 };
442 
443 union rdma_protocol_stats {
444 	struct ib_protocol_stats	ib;
445 	struct iw_protocol_stats	iw;
446 };
447 
448 /* Define bits for the various functionality this port needs to be supported by
449  * the core.
450  */
451 /* Management                           0x00000FFF */
452 #define RDMA_CORE_CAP_IB_MAD            0x00000001
453 #define RDMA_CORE_CAP_IB_SMI            0x00000002
454 #define RDMA_CORE_CAP_IB_CM             0x00000004
455 #define RDMA_CORE_CAP_IW_CM             0x00000008
456 #define RDMA_CORE_CAP_IB_SA             0x00000010
457 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
458 
459 /* Address format                       0x000FF000 */
460 #define RDMA_CORE_CAP_AF_IB             0x00001000
461 #define RDMA_CORE_CAP_ETH_AH            0x00002000
462 
463 /* Protocol                             0xFFF00000 */
464 #define RDMA_CORE_CAP_PROT_IB           0x00100000
465 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
466 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
467 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
468 
469 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
470 					| RDMA_CORE_CAP_IB_MAD \
471 					| RDMA_CORE_CAP_IB_SMI \
472 					| RDMA_CORE_CAP_IB_CM  \
473 					| RDMA_CORE_CAP_IB_SA  \
474 					| RDMA_CORE_CAP_AF_IB)
475 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
476 					| RDMA_CORE_CAP_IB_MAD  \
477 					| RDMA_CORE_CAP_IB_CM   \
478 					| RDMA_CORE_CAP_AF_IB   \
479 					| RDMA_CORE_CAP_ETH_AH)
480 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
481 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
482 					| RDMA_CORE_CAP_IB_MAD  \
483 					| RDMA_CORE_CAP_IB_CM   \
484 					| RDMA_CORE_CAP_AF_IB   \
485 					| RDMA_CORE_CAP_ETH_AH)
486 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
487 					| RDMA_CORE_CAP_IW_CM)
488 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
489 					| RDMA_CORE_CAP_OPA_MAD)
490 
491 struct ib_port_attr {
492 	enum ib_port_state	state;
493 	enum ib_mtu		max_mtu;
494 	enum ib_mtu		active_mtu;
495 	int			gid_tbl_len;
496 	u32			port_cap_flags;
497 	u32			max_msg_sz;
498 	u32			bad_pkey_cntr;
499 	u32			qkey_viol_cntr;
500 	u16			pkey_tbl_len;
501 	u16			lid;
502 	u16			sm_lid;
503 	u8			lmc;
504 	u8			max_vl_num;
505 	u8			sm_sl;
506 	u8			subnet_timeout;
507 	u8			init_type_reply;
508 	u8			active_width;
509 	u8			active_speed;
510 	u8                      phys_state;
511 };
512 
513 enum ib_device_modify_flags {
514 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
515 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
516 };
517 
518 struct ib_device_modify {
519 	u64	sys_image_guid;
520 	char	node_desc[64];
521 };
522 
523 enum ib_port_modify_flags {
524 	IB_PORT_SHUTDOWN		= 1,
525 	IB_PORT_INIT_TYPE		= (1<<2),
526 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
527 };
528 
529 struct ib_port_modify {
530 	u32	set_port_cap_mask;
531 	u32	clr_port_cap_mask;
532 	u8	init_type;
533 };
534 
535 enum ib_event_type {
536 	IB_EVENT_CQ_ERR,
537 	IB_EVENT_QP_FATAL,
538 	IB_EVENT_QP_REQ_ERR,
539 	IB_EVENT_QP_ACCESS_ERR,
540 	IB_EVENT_COMM_EST,
541 	IB_EVENT_SQ_DRAINED,
542 	IB_EVENT_PATH_MIG,
543 	IB_EVENT_PATH_MIG_ERR,
544 	IB_EVENT_DEVICE_FATAL,
545 	IB_EVENT_PORT_ACTIVE,
546 	IB_EVENT_PORT_ERR,
547 	IB_EVENT_LID_CHANGE,
548 	IB_EVENT_PKEY_CHANGE,
549 	IB_EVENT_SM_CHANGE,
550 	IB_EVENT_SRQ_ERR,
551 	IB_EVENT_SRQ_LIMIT_REACHED,
552 	IB_EVENT_QP_LAST_WQE_REACHED,
553 	IB_EVENT_CLIENT_REREGISTER,
554 	IB_EVENT_GID_CHANGE,
555 };
556 
557 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
558 
559 struct ib_event {
560 	struct ib_device	*device;
561 	union {
562 		struct ib_cq	*cq;
563 		struct ib_qp	*qp;
564 		struct ib_srq	*srq;
565 		u8		port_num;
566 	} element;
567 	enum ib_event_type	event;
568 };
569 
570 struct ib_event_handler {
571 	struct ib_device *device;
572 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
573 	struct list_head  list;
574 };
575 
576 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
577 	do {							\
578 		(_ptr)->device  = _device;			\
579 		(_ptr)->handler = _handler;			\
580 		INIT_LIST_HEAD(&(_ptr)->list);			\
581 	} while (0)
582 
583 struct ib_global_route {
584 	union ib_gid	dgid;
585 	u32		flow_label;
586 	u8		sgid_index;
587 	u8		hop_limit;
588 	u8		traffic_class;
589 };
590 
591 struct ib_grh {
592 	__be32		version_tclass_flow;
593 	__be16		paylen;
594 	u8		next_hdr;
595 	u8		hop_limit;
596 	union ib_gid	sgid;
597 	union ib_gid	dgid;
598 };
599 
600 union rdma_network_hdr {
601 	struct ib_grh ibgrh;
602 	struct {
603 		/* The IB spec states that if it's IPv4, the header
604 		 * is located in the last 20 bytes of the header.
605 		 */
606 		u8		reserved[20];
607 		struct iphdr	roce4grh;
608 	};
609 };
610 
611 enum {
612 	IB_MULTICAST_QPN = 0xffffff
613 };
614 
615 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
616 
617 enum ib_ah_flags {
618 	IB_AH_GRH	= 1
619 };
620 
621 enum ib_rate {
622 	IB_RATE_PORT_CURRENT = 0,
623 	IB_RATE_2_5_GBPS = 2,
624 	IB_RATE_5_GBPS   = 5,
625 	IB_RATE_10_GBPS  = 3,
626 	IB_RATE_20_GBPS  = 6,
627 	IB_RATE_30_GBPS  = 4,
628 	IB_RATE_40_GBPS  = 7,
629 	IB_RATE_60_GBPS  = 8,
630 	IB_RATE_80_GBPS  = 9,
631 	IB_RATE_120_GBPS = 10,
632 	IB_RATE_14_GBPS  = 11,
633 	IB_RATE_56_GBPS  = 12,
634 	IB_RATE_112_GBPS = 13,
635 	IB_RATE_168_GBPS = 14,
636 	IB_RATE_25_GBPS  = 15,
637 	IB_RATE_100_GBPS = 16,
638 	IB_RATE_200_GBPS = 17,
639 	IB_RATE_300_GBPS = 18
640 };
641 
642 /**
643  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
644  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
645  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
646  * @rate: rate to convert.
647  */
648 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
649 
650 /**
651  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
652  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
653  * @rate: rate to convert.
654  */
655 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
656 
657 
658 /**
659  * enum ib_mr_type - memory region type
660  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
661  *                            normal registration
662  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
663  *                            signature operations (data-integrity
664  *                            capable regions)
665  */
666 enum ib_mr_type {
667 	IB_MR_TYPE_MEM_REG,
668 	IB_MR_TYPE_SIGNATURE,
669 };
670 
671 /**
672  * Signature types
673  * IB_SIG_TYPE_NONE: Unprotected.
674  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
675  */
676 enum ib_signature_type {
677 	IB_SIG_TYPE_NONE,
678 	IB_SIG_TYPE_T10_DIF,
679 };
680 
681 /**
682  * Signature T10-DIF block-guard types
683  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
684  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
685  */
686 enum ib_t10_dif_bg_type {
687 	IB_T10DIF_CRC,
688 	IB_T10DIF_CSUM
689 };
690 
691 /**
692  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
693  *     domain.
694  * @bg_type: T10-DIF block guard type (CRC|CSUM)
695  * @pi_interval: protection information interval.
696  * @bg: seed of guard computation.
697  * @app_tag: application tag of guard block
698  * @ref_tag: initial guard block reference tag.
699  * @ref_remap: Indicate wethear the reftag increments each block
700  * @app_escape: Indicate to skip block check if apptag=0xffff
701  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
702  * @apptag_check_mask: check bitmask of application tag.
703  */
704 struct ib_t10_dif_domain {
705 	enum ib_t10_dif_bg_type bg_type;
706 	u16			pi_interval;
707 	u16			bg;
708 	u16			app_tag;
709 	u32			ref_tag;
710 	bool			ref_remap;
711 	bool			app_escape;
712 	bool			ref_escape;
713 	u16			apptag_check_mask;
714 };
715 
716 /**
717  * struct ib_sig_domain - Parameters for signature domain
718  * @sig_type: specific signauture type
719  * @sig: union of all signature domain attributes that may
720  *     be used to set domain layout.
721  */
722 struct ib_sig_domain {
723 	enum ib_signature_type sig_type;
724 	union {
725 		struct ib_t10_dif_domain dif;
726 	} sig;
727 };
728 
729 /**
730  * struct ib_sig_attrs - Parameters for signature handover operation
731  * @check_mask: bitmask for signature byte check (8 bytes)
732  * @mem: memory domain layout desciptor.
733  * @wire: wire domain layout desciptor.
734  */
735 struct ib_sig_attrs {
736 	u8			check_mask;
737 	struct ib_sig_domain	mem;
738 	struct ib_sig_domain	wire;
739 };
740 
741 enum ib_sig_err_type {
742 	IB_SIG_BAD_GUARD,
743 	IB_SIG_BAD_REFTAG,
744 	IB_SIG_BAD_APPTAG,
745 };
746 
747 /**
748  * struct ib_sig_err - signature error descriptor
749  */
750 struct ib_sig_err {
751 	enum ib_sig_err_type	err_type;
752 	u32			expected;
753 	u32			actual;
754 	u64			sig_err_offset;
755 	u32			key;
756 };
757 
758 enum ib_mr_status_check {
759 	IB_MR_CHECK_SIG_STATUS = 1,
760 };
761 
762 /**
763  * struct ib_mr_status - Memory region status container
764  *
765  * @fail_status: Bitmask of MR checks status. For each
766  *     failed check a corresponding status bit is set.
767  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
768  *     failure.
769  */
770 struct ib_mr_status {
771 	u32		    fail_status;
772 	struct ib_sig_err   sig_err;
773 };
774 
775 /**
776  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
777  * enum.
778  * @mult: multiple to convert.
779  */
780 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
781 
782 struct ib_ah_attr {
783 	struct ib_global_route	grh;
784 	u16			dlid;
785 	u8			sl;
786 	u8			src_path_bits;
787 	u8			static_rate;
788 	u8			ah_flags;
789 	u8			port_num;
790 	u8			dmac[ETH_ALEN];
791 };
792 
793 enum ib_wc_status {
794 	IB_WC_SUCCESS,
795 	IB_WC_LOC_LEN_ERR,
796 	IB_WC_LOC_QP_OP_ERR,
797 	IB_WC_LOC_EEC_OP_ERR,
798 	IB_WC_LOC_PROT_ERR,
799 	IB_WC_WR_FLUSH_ERR,
800 	IB_WC_MW_BIND_ERR,
801 	IB_WC_BAD_RESP_ERR,
802 	IB_WC_LOC_ACCESS_ERR,
803 	IB_WC_REM_INV_REQ_ERR,
804 	IB_WC_REM_ACCESS_ERR,
805 	IB_WC_REM_OP_ERR,
806 	IB_WC_RETRY_EXC_ERR,
807 	IB_WC_RNR_RETRY_EXC_ERR,
808 	IB_WC_LOC_RDD_VIOL_ERR,
809 	IB_WC_REM_INV_RD_REQ_ERR,
810 	IB_WC_REM_ABORT_ERR,
811 	IB_WC_INV_EECN_ERR,
812 	IB_WC_INV_EEC_STATE_ERR,
813 	IB_WC_FATAL_ERR,
814 	IB_WC_RESP_TIMEOUT_ERR,
815 	IB_WC_GENERAL_ERR
816 };
817 
818 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
819 
820 enum ib_wc_opcode {
821 	IB_WC_SEND,
822 	IB_WC_RDMA_WRITE,
823 	IB_WC_RDMA_READ,
824 	IB_WC_COMP_SWAP,
825 	IB_WC_FETCH_ADD,
826 	IB_WC_LSO,
827 	IB_WC_LOCAL_INV,
828 	IB_WC_REG_MR,
829 	IB_WC_MASKED_COMP_SWAP,
830 	IB_WC_MASKED_FETCH_ADD,
831 /*
832  * Set value of IB_WC_RECV so consumers can test if a completion is a
833  * receive by testing (opcode & IB_WC_RECV).
834  */
835 	IB_WC_RECV			= 1 << 7,
836 	IB_WC_RECV_RDMA_WITH_IMM
837 };
838 
839 enum ib_wc_flags {
840 	IB_WC_GRH		= 1,
841 	IB_WC_WITH_IMM		= (1<<1),
842 	IB_WC_WITH_INVALIDATE	= (1<<2),
843 	IB_WC_IP_CSUM_OK	= (1<<3),
844 	IB_WC_WITH_SMAC		= (1<<4),
845 	IB_WC_WITH_VLAN		= (1<<5),
846 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
847 };
848 
849 struct ib_wc {
850 	union {
851 		u64		wr_id;
852 		struct ib_cqe	*wr_cqe;
853 	};
854 	enum ib_wc_status	status;
855 	enum ib_wc_opcode	opcode;
856 	u32			vendor_err;
857 	u32			byte_len;
858 	struct ib_qp	       *qp;
859 	union {
860 		__be32		imm_data;
861 		u32		invalidate_rkey;
862 	} ex;
863 	u32			src_qp;
864 	int			wc_flags;
865 	u16			pkey_index;
866 	u16			slid;
867 	u8			sl;
868 	u8			dlid_path_bits;
869 	u8			port_num;	/* valid only for DR SMPs on switches */
870 	u8			smac[ETH_ALEN];
871 	u16			vlan_id;
872 	u8			network_hdr_type;
873 };
874 
875 enum ib_cq_notify_flags {
876 	IB_CQ_SOLICITED			= 1 << 0,
877 	IB_CQ_NEXT_COMP			= 1 << 1,
878 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
879 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
880 };
881 
882 enum ib_srq_type {
883 	IB_SRQT_BASIC,
884 	IB_SRQT_XRC
885 };
886 
887 enum ib_srq_attr_mask {
888 	IB_SRQ_MAX_WR	= 1 << 0,
889 	IB_SRQ_LIMIT	= 1 << 1,
890 };
891 
892 struct ib_srq_attr {
893 	u32	max_wr;
894 	u32	max_sge;
895 	u32	srq_limit;
896 };
897 
898 struct ib_srq_init_attr {
899 	void		      (*event_handler)(struct ib_event *, void *);
900 	void		       *srq_context;
901 	struct ib_srq_attr	attr;
902 	enum ib_srq_type	srq_type;
903 
904 	union {
905 		struct {
906 			struct ib_xrcd *xrcd;
907 			struct ib_cq   *cq;
908 		} xrc;
909 	} ext;
910 };
911 
912 struct ib_qp_cap {
913 	u32	max_send_wr;
914 	u32	max_recv_wr;
915 	u32	max_send_sge;
916 	u32	max_recv_sge;
917 	u32	max_inline_data;
918 };
919 
920 enum ib_sig_type {
921 	IB_SIGNAL_ALL_WR,
922 	IB_SIGNAL_REQ_WR
923 };
924 
925 enum ib_qp_type {
926 	/*
927 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
928 	 * here (and in that order) since the MAD layer uses them as
929 	 * indices into a 2-entry table.
930 	 */
931 	IB_QPT_SMI,
932 	IB_QPT_GSI,
933 
934 	IB_QPT_RC,
935 	IB_QPT_UC,
936 	IB_QPT_UD,
937 	IB_QPT_RAW_IPV6,
938 	IB_QPT_RAW_ETHERTYPE,
939 	IB_QPT_RAW_PACKET = 8,
940 	IB_QPT_XRC_INI = 9,
941 	IB_QPT_XRC_TGT,
942 	IB_QPT_MAX,
943 	/* Reserve a range for qp types internal to the low level driver.
944 	 * These qp types will not be visible at the IB core layer, so the
945 	 * IB_QPT_MAX usages should not be affected in the core layer
946 	 */
947 	IB_QPT_RESERVED1 = 0x1000,
948 	IB_QPT_RESERVED2,
949 	IB_QPT_RESERVED3,
950 	IB_QPT_RESERVED4,
951 	IB_QPT_RESERVED5,
952 	IB_QPT_RESERVED6,
953 	IB_QPT_RESERVED7,
954 	IB_QPT_RESERVED8,
955 	IB_QPT_RESERVED9,
956 	IB_QPT_RESERVED10,
957 };
958 
959 enum ib_qp_create_flags {
960 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
961 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
962 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
963 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
964 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
965 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
966 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
967 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
968 	/* reserve bits 26-31 for low level drivers' internal use */
969 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
970 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
971 };
972 
973 /*
974  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
975  * callback to destroy the passed in QP.
976  */
977 
978 struct ib_qp_init_attr {
979 	void                  (*event_handler)(struct ib_event *, void *);
980 	void		       *qp_context;
981 	struct ib_cq	       *send_cq;
982 	struct ib_cq	       *recv_cq;
983 	struct ib_srq	       *srq;
984 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
985 	struct ib_qp_cap	cap;
986 	enum ib_sig_type	sq_sig_type;
987 	enum ib_qp_type		qp_type;
988 	enum ib_qp_create_flags	create_flags;
989 	u8			port_num; /* special QP types only */
990 };
991 
992 struct ib_qp_open_attr {
993 	void                  (*event_handler)(struct ib_event *, void *);
994 	void		       *qp_context;
995 	u32			qp_num;
996 	enum ib_qp_type		qp_type;
997 };
998 
999 enum ib_rnr_timeout {
1000 	IB_RNR_TIMER_655_36 =  0,
1001 	IB_RNR_TIMER_000_01 =  1,
1002 	IB_RNR_TIMER_000_02 =  2,
1003 	IB_RNR_TIMER_000_03 =  3,
1004 	IB_RNR_TIMER_000_04 =  4,
1005 	IB_RNR_TIMER_000_06 =  5,
1006 	IB_RNR_TIMER_000_08 =  6,
1007 	IB_RNR_TIMER_000_12 =  7,
1008 	IB_RNR_TIMER_000_16 =  8,
1009 	IB_RNR_TIMER_000_24 =  9,
1010 	IB_RNR_TIMER_000_32 = 10,
1011 	IB_RNR_TIMER_000_48 = 11,
1012 	IB_RNR_TIMER_000_64 = 12,
1013 	IB_RNR_TIMER_000_96 = 13,
1014 	IB_RNR_TIMER_001_28 = 14,
1015 	IB_RNR_TIMER_001_92 = 15,
1016 	IB_RNR_TIMER_002_56 = 16,
1017 	IB_RNR_TIMER_003_84 = 17,
1018 	IB_RNR_TIMER_005_12 = 18,
1019 	IB_RNR_TIMER_007_68 = 19,
1020 	IB_RNR_TIMER_010_24 = 20,
1021 	IB_RNR_TIMER_015_36 = 21,
1022 	IB_RNR_TIMER_020_48 = 22,
1023 	IB_RNR_TIMER_030_72 = 23,
1024 	IB_RNR_TIMER_040_96 = 24,
1025 	IB_RNR_TIMER_061_44 = 25,
1026 	IB_RNR_TIMER_081_92 = 26,
1027 	IB_RNR_TIMER_122_88 = 27,
1028 	IB_RNR_TIMER_163_84 = 28,
1029 	IB_RNR_TIMER_245_76 = 29,
1030 	IB_RNR_TIMER_327_68 = 30,
1031 	IB_RNR_TIMER_491_52 = 31
1032 };
1033 
1034 enum ib_qp_attr_mask {
1035 	IB_QP_STATE			= 1,
1036 	IB_QP_CUR_STATE			= (1<<1),
1037 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1038 	IB_QP_ACCESS_FLAGS		= (1<<3),
1039 	IB_QP_PKEY_INDEX		= (1<<4),
1040 	IB_QP_PORT			= (1<<5),
1041 	IB_QP_QKEY			= (1<<6),
1042 	IB_QP_AV			= (1<<7),
1043 	IB_QP_PATH_MTU			= (1<<8),
1044 	IB_QP_TIMEOUT			= (1<<9),
1045 	IB_QP_RETRY_CNT			= (1<<10),
1046 	IB_QP_RNR_RETRY			= (1<<11),
1047 	IB_QP_RQ_PSN			= (1<<12),
1048 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1049 	IB_QP_ALT_PATH			= (1<<14),
1050 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1051 	IB_QP_SQ_PSN			= (1<<16),
1052 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1053 	IB_QP_PATH_MIG_STATE		= (1<<18),
1054 	IB_QP_CAP			= (1<<19),
1055 	IB_QP_DEST_QPN			= (1<<20),
1056 	IB_QP_RESERVED1			= (1<<21),
1057 	IB_QP_RESERVED2			= (1<<22),
1058 	IB_QP_RESERVED3			= (1<<23),
1059 	IB_QP_RESERVED4			= (1<<24),
1060 };
1061 
1062 enum ib_qp_state {
1063 	IB_QPS_RESET,
1064 	IB_QPS_INIT,
1065 	IB_QPS_RTR,
1066 	IB_QPS_RTS,
1067 	IB_QPS_SQD,
1068 	IB_QPS_SQE,
1069 	IB_QPS_ERR
1070 };
1071 
1072 enum ib_mig_state {
1073 	IB_MIG_MIGRATED,
1074 	IB_MIG_REARM,
1075 	IB_MIG_ARMED
1076 };
1077 
1078 enum ib_mw_type {
1079 	IB_MW_TYPE_1 = 1,
1080 	IB_MW_TYPE_2 = 2
1081 };
1082 
1083 struct ib_qp_attr {
1084 	enum ib_qp_state	qp_state;
1085 	enum ib_qp_state	cur_qp_state;
1086 	enum ib_mtu		path_mtu;
1087 	enum ib_mig_state	path_mig_state;
1088 	u32			qkey;
1089 	u32			rq_psn;
1090 	u32			sq_psn;
1091 	u32			dest_qp_num;
1092 	int			qp_access_flags;
1093 	struct ib_qp_cap	cap;
1094 	struct ib_ah_attr	ah_attr;
1095 	struct ib_ah_attr	alt_ah_attr;
1096 	u16			pkey_index;
1097 	u16			alt_pkey_index;
1098 	u8			en_sqd_async_notify;
1099 	u8			sq_draining;
1100 	u8			max_rd_atomic;
1101 	u8			max_dest_rd_atomic;
1102 	u8			min_rnr_timer;
1103 	u8			port_num;
1104 	u8			timeout;
1105 	u8			retry_cnt;
1106 	u8			rnr_retry;
1107 	u8			alt_port_num;
1108 	u8			alt_timeout;
1109 };
1110 
1111 enum ib_wr_opcode {
1112 	IB_WR_RDMA_WRITE,
1113 	IB_WR_RDMA_WRITE_WITH_IMM,
1114 	IB_WR_SEND,
1115 	IB_WR_SEND_WITH_IMM,
1116 	IB_WR_RDMA_READ,
1117 	IB_WR_ATOMIC_CMP_AND_SWP,
1118 	IB_WR_ATOMIC_FETCH_AND_ADD,
1119 	IB_WR_LSO,
1120 	IB_WR_SEND_WITH_INV,
1121 	IB_WR_RDMA_READ_WITH_INV,
1122 	IB_WR_LOCAL_INV,
1123 	IB_WR_REG_MR,
1124 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1125 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1126 	IB_WR_REG_SIG_MR,
1127 	/* reserve values for low level drivers' internal use.
1128 	 * These values will not be used at all in the ib core layer.
1129 	 */
1130 	IB_WR_RESERVED1 = 0xf0,
1131 	IB_WR_RESERVED2,
1132 	IB_WR_RESERVED3,
1133 	IB_WR_RESERVED4,
1134 	IB_WR_RESERVED5,
1135 	IB_WR_RESERVED6,
1136 	IB_WR_RESERVED7,
1137 	IB_WR_RESERVED8,
1138 	IB_WR_RESERVED9,
1139 	IB_WR_RESERVED10,
1140 };
1141 
1142 enum ib_send_flags {
1143 	IB_SEND_FENCE		= 1,
1144 	IB_SEND_SIGNALED	= (1<<1),
1145 	IB_SEND_SOLICITED	= (1<<2),
1146 	IB_SEND_INLINE		= (1<<3),
1147 	IB_SEND_IP_CSUM		= (1<<4),
1148 
1149 	/* reserve bits 26-31 for low level drivers' internal use */
1150 	IB_SEND_RESERVED_START	= (1 << 26),
1151 	IB_SEND_RESERVED_END	= (1 << 31),
1152 };
1153 
1154 struct ib_sge {
1155 	u64	addr;
1156 	u32	length;
1157 	u32	lkey;
1158 };
1159 
1160 struct ib_cqe {
1161 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1162 };
1163 
1164 struct ib_send_wr {
1165 	struct ib_send_wr      *next;
1166 	union {
1167 		u64		wr_id;
1168 		struct ib_cqe	*wr_cqe;
1169 	};
1170 	struct ib_sge	       *sg_list;
1171 	int			num_sge;
1172 	enum ib_wr_opcode	opcode;
1173 	int			send_flags;
1174 	union {
1175 		__be32		imm_data;
1176 		u32		invalidate_rkey;
1177 	} ex;
1178 };
1179 
1180 struct ib_rdma_wr {
1181 	struct ib_send_wr	wr;
1182 	u64			remote_addr;
1183 	u32			rkey;
1184 };
1185 
1186 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1187 {
1188 	return container_of(wr, struct ib_rdma_wr, wr);
1189 }
1190 
1191 struct ib_atomic_wr {
1192 	struct ib_send_wr	wr;
1193 	u64			remote_addr;
1194 	u64			compare_add;
1195 	u64			swap;
1196 	u64			compare_add_mask;
1197 	u64			swap_mask;
1198 	u32			rkey;
1199 };
1200 
1201 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1202 {
1203 	return container_of(wr, struct ib_atomic_wr, wr);
1204 }
1205 
1206 struct ib_ud_wr {
1207 	struct ib_send_wr	wr;
1208 	struct ib_ah		*ah;
1209 	void			*header;
1210 	int			hlen;
1211 	int			mss;
1212 	u32			remote_qpn;
1213 	u32			remote_qkey;
1214 	u16			pkey_index; /* valid for GSI only */
1215 	u8			port_num;   /* valid for DR SMPs on switch only */
1216 };
1217 
1218 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1219 {
1220 	return container_of(wr, struct ib_ud_wr, wr);
1221 }
1222 
1223 struct ib_reg_wr {
1224 	struct ib_send_wr	wr;
1225 	struct ib_mr		*mr;
1226 	u32			key;
1227 	int			access;
1228 };
1229 
1230 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1231 {
1232 	return container_of(wr, struct ib_reg_wr, wr);
1233 }
1234 
1235 struct ib_sig_handover_wr {
1236 	struct ib_send_wr	wr;
1237 	struct ib_sig_attrs    *sig_attrs;
1238 	struct ib_mr	       *sig_mr;
1239 	int			access_flags;
1240 	struct ib_sge	       *prot;
1241 };
1242 
1243 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1244 {
1245 	return container_of(wr, struct ib_sig_handover_wr, wr);
1246 }
1247 
1248 struct ib_recv_wr {
1249 	struct ib_recv_wr      *next;
1250 	union {
1251 		u64		wr_id;
1252 		struct ib_cqe	*wr_cqe;
1253 	};
1254 	struct ib_sge	       *sg_list;
1255 	int			num_sge;
1256 };
1257 
1258 enum ib_access_flags {
1259 	IB_ACCESS_LOCAL_WRITE	= 1,
1260 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1261 	IB_ACCESS_REMOTE_READ	= (1<<2),
1262 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1263 	IB_ACCESS_MW_BIND	= (1<<4),
1264 	IB_ZERO_BASED		= (1<<5),
1265 	IB_ACCESS_ON_DEMAND     = (1<<6),
1266 };
1267 
1268 /*
1269  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1270  * are hidden here instead of a uapi header!
1271  */
1272 enum ib_mr_rereg_flags {
1273 	IB_MR_REREG_TRANS	= 1,
1274 	IB_MR_REREG_PD		= (1<<1),
1275 	IB_MR_REREG_ACCESS	= (1<<2),
1276 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1277 };
1278 
1279 struct ib_fmr_attr {
1280 	int	max_pages;
1281 	int	max_maps;
1282 	u8	page_shift;
1283 };
1284 
1285 struct ib_umem;
1286 
1287 struct ib_ucontext {
1288 	struct ib_device       *device;
1289 	struct list_head	pd_list;
1290 	struct list_head	mr_list;
1291 	struct list_head	mw_list;
1292 	struct list_head	cq_list;
1293 	struct list_head	qp_list;
1294 	struct list_head	srq_list;
1295 	struct list_head	ah_list;
1296 	struct list_head	xrcd_list;
1297 	struct list_head	rule_list;
1298 	int			closing;
1299 
1300 	struct pid             *tgid;
1301 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1302 	struct rb_root      umem_tree;
1303 	/*
1304 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1305 	 * mmu notifiers registration.
1306 	 */
1307 	struct rw_semaphore	umem_rwsem;
1308 	void (*invalidate_range)(struct ib_umem *umem,
1309 				 unsigned long start, unsigned long end);
1310 
1311 	struct mmu_notifier	mn;
1312 	atomic_t		notifier_count;
1313 	/* A list of umems that don't have private mmu notifier counters yet. */
1314 	struct list_head	no_private_counters;
1315 	int                     odp_mrs_count;
1316 #endif
1317 };
1318 
1319 struct ib_uobject {
1320 	u64			user_handle;	/* handle given to us by userspace */
1321 	struct ib_ucontext     *context;	/* associated user context */
1322 	void		       *object;		/* containing object */
1323 	struct list_head	list;		/* link to context's list */
1324 	int			id;		/* index into kernel idr */
1325 	struct kref		ref;
1326 	struct rw_semaphore	mutex;		/* protects .live */
1327 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1328 	int			live;
1329 };
1330 
1331 struct ib_udata {
1332 	const void __user *inbuf;
1333 	void __user *outbuf;
1334 	size_t       inlen;
1335 	size_t       outlen;
1336 };
1337 
1338 struct ib_pd {
1339 	u32			local_dma_lkey;
1340 	struct ib_device       *device;
1341 	struct ib_uobject      *uobject;
1342 	atomic_t          	usecnt; /* count all resources */
1343 	struct ib_mr	       *local_mr;
1344 };
1345 
1346 struct ib_xrcd {
1347 	struct ib_device       *device;
1348 	atomic_t		usecnt; /* count all exposed resources */
1349 	struct inode	       *inode;
1350 
1351 	struct mutex		tgt_qp_mutex;
1352 	struct list_head	tgt_qp_list;
1353 };
1354 
1355 struct ib_ah {
1356 	struct ib_device	*device;
1357 	struct ib_pd		*pd;
1358 	struct ib_uobject	*uobject;
1359 };
1360 
1361 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1362 
1363 enum ib_poll_context {
1364 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1365 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1366 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1367 };
1368 
1369 struct ib_cq {
1370 	struct ib_device       *device;
1371 	struct ib_uobject      *uobject;
1372 	ib_comp_handler   	comp_handler;
1373 	void                  (*event_handler)(struct ib_event *, void *);
1374 	void                   *cq_context;
1375 	int               	cqe;
1376 	atomic_t          	usecnt; /* count number of work queues */
1377 	enum ib_poll_context	poll_ctx;
1378 	struct ib_wc		*wc;
1379 	union {
1380 		struct irq_poll		iop;
1381 		struct work_struct	work;
1382 	};
1383 };
1384 
1385 struct ib_srq {
1386 	struct ib_device       *device;
1387 	struct ib_pd	       *pd;
1388 	struct ib_uobject      *uobject;
1389 	void		      (*event_handler)(struct ib_event *, void *);
1390 	void		       *srq_context;
1391 	enum ib_srq_type	srq_type;
1392 	atomic_t		usecnt;
1393 
1394 	union {
1395 		struct {
1396 			struct ib_xrcd *xrcd;
1397 			struct ib_cq   *cq;
1398 			u32		srq_num;
1399 		} xrc;
1400 	} ext;
1401 };
1402 
1403 struct ib_qp {
1404 	struct ib_device       *device;
1405 	struct ib_pd	       *pd;
1406 	struct ib_cq	       *send_cq;
1407 	struct ib_cq	       *recv_cq;
1408 	struct ib_srq	       *srq;
1409 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1410 	struct list_head	xrcd_list;
1411 	/* count times opened, mcast attaches, flow attaches */
1412 	atomic_t		usecnt;
1413 	struct list_head	open_list;
1414 	struct ib_qp           *real_qp;
1415 	struct ib_uobject      *uobject;
1416 	void                  (*event_handler)(struct ib_event *, void *);
1417 	void		       *qp_context;
1418 	u32			qp_num;
1419 	enum ib_qp_type		qp_type;
1420 };
1421 
1422 struct ib_mr {
1423 	struct ib_device  *device;
1424 	struct ib_pd	  *pd;
1425 	struct ib_uobject *uobject;
1426 	u32		   lkey;
1427 	u32		   rkey;
1428 	u64		   iova;
1429 	u32		   length;
1430 	unsigned int	   page_size;
1431 };
1432 
1433 struct ib_mw {
1434 	struct ib_device	*device;
1435 	struct ib_pd		*pd;
1436 	struct ib_uobject	*uobject;
1437 	u32			rkey;
1438 	enum ib_mw_type         type;
1439 };
1440 
1441 struct ib_fmr {
1442 	struct ib_device	*device;
1443 	struct ib_pd		*pd;
1444 	struct list_head	list;
1445 	u32			lkey;
1446 	u32			rkey;
1447 };
1448 
1449 /* Supported steering options */
1450 enum ib_flow_attr_type {
1451 	/* steering according to rule specifications */
1452 	IB_FLOW_ATTR_NORMAL		= 0x0,
1453 	/* default unicast and multicast rule -
1454 	 * receive all Eth traffic which isn't steered to any QP
1455 	 */
1456 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1457 	/* default multicast rule -
1458 	 * receive all Eth multicast traffic which isn't steered to any QP
1459 	 */
1460 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1461 	/* sniffer rule - receive all port traffic */
1462 	IB_FLOW_ATTR_SNIFFER		= 0x3
1463 };
1464 
1465 /* Supported steering header types */
1466 enum ib_flow_spec_type {
1467 	/* L2 headers*/
1468 	IB_FLOW_SPEC_ETH	= 0x20,
1469 	IB_FLOW_SPEC_IB		= 0x22,
1470 	/* L3 header*/
1471 	IB_FLOW_SPEC_IPV4	= 0x30,
1472 	/* L4 headers*/
1473 	IB_FLOW_SPEC_TCP	= 0x40,
1474 	IB_FLOW_SPEC_UDP	= 0x41
1475 };
1476 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1477 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1478 
1479 /* Flow steering rule priority is set according to it's domain.
1480  * Lower domain value means higher priority.
1481  */
1482 enum ib_flow_domain {
1483 	IB_FLOW_DOMAIN_USER,
1484 	IB_FLOW_DOMAIN_ETHTOOL,
1485 	IB_FLOW_DOMAIN_RFS,
1486 	IB_FLOW_DOMAIN_NIC,
1487 	IB_FLOW_DOMAIN_NUM /* Must be last */
1488 };
1489 
1490 struct ib_flow_eth_filter {
1491 	u8	dst_mac[6];
1492 	u8	src_mac[6];
1493 	__be16	ether_type;
1494 	__be16	vlan_tag;
1495 };
1496 
1497 struct ib_flow_spec_eth {
1498 	enum ib_flow_spec_type	  type;
1499 	u16			  size;
1500 	struct ib_flow_eth_filter val;
1501 	struct ib_flow_eth_filter mask;
1502 };
1503 
1504 struct ib_flow_ib_filter {
1505 	__be16 dlid;
1506 	__u8   sl;
1507 };
1508 
1509 struct ib_flow_spec_ib {
1510 	enum ib_flow_spec_type	 type;
1511 	u16			 size;
1512 	struct ib_flow_ib_filter val;
1513 	struct ib_flow_ib_filter mask;
1514 };
1515 
1516 struct ib_flow_ipv4_filter {
1517 	__be32	src_ip;
1518 	__be32	dst_ip;
1519 };
1520 
1521 struct ib_flow_spec_ipv4 {
1522 	enum ib_flow_spec_type	   type;
1523 	u16			   size;
1524 	struct ib_flow_ipv4_filter val;
1525 	struct ib_flow_ipv4_filter mask;
1526 };
1527 
1528 struct ib_flow_tcp_udp_filter {
1529 	__be16	dst_port;
1530 	__be16	src_port;
1531 };
1532 
1533 struct ib_flow_spec_tcp_udp {
1534 	enum ib_flow_spec_type	      type;
1535 	u16			      size;
1536 	struct ib_flow_tcp_udp_filter val;
1537 	struct ib_flow_tcp_udp_filter mask;
1538 };
1539 
1540 union ib_flow_spec {
1541 	struct {
1542 		enum ib_flow_spec_type	type;
1543 		u16			size;
1544 	};
1545 	struct ib_flow_spec_eth		eth;
1546 	struct ib_flow_spec_ib		ib;
1547 	struct ib_flow_spec_ipv4        ipv4;
1548 	struct ib_flow_spec_tcp_udp	tcp_udp;
1549 };
1550 
1551 struct ib_flow_attr {
1552 	enum ib_flow_attr_type type;
1553 	u16	     size;
1554 	u16	     priority;
1555 	u32	     flags;
1556 	u8	     num_of_specs;
1557 	u8	     port;
1558 	/* Following are the optional layers according to user request
1559 	 * struct ib_flow_spec_xxx
1560 	 * struct ib_flow_spec_yyy
1561 	 */
1562 };
1563 
1564 struct ib_flow {
1565 	struct ib_qp		*qp;
1566 	struct ib_uobject	*uobject;
1567 };
1568 
1569 struct ib_mad_hdr;
1570 struct ib_grh;
1571 
1572 enum ib_process_mad_flags {
1573 	IB_MAD_IGNORE_MKEY	= 1,
1574 	IB_MAD_IGNORE_BKEY	= 2,
1575 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1576 };
1577 
1578 enum ib_mad_result {
1579 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1580 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1581 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1582 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1583 };
1584 
1585 #define IB_DEVICE_NAME_MAX 64
1586 
1587 struct ib_cache {
1588 	rwlock_t                lock;
1589 	struct ib_event_handler event_handler;
1590 	struct ib_pkey_cache  **pkey_cache;
1591 	struct ib_gid_table   **gid_cache;
1592 	u8                     *lmc_cache;
1593 };
1594 
1595 struct ib_dma_mapping_ops {
1596 	int		(*mapping_error)(struct ib_device *dev,
1597 					 u64 dma_addr);
1598 	u64		(*map_single)(struct ib_device *dev,
1599 				      void *ptr, size_t size,
1600 				      enum dma_data_direction direction);
1601 	void		(*unmap_single)(struct ib_device *dev,
1602 					u64 addr, size_t size,
1603 					enum dma_data_direction direction);
1604 	u64		(*map_page)(struct ib_device *dev,
1605 				    struct page *page, unsigned long offset,
1606 				    size_t size,
1607 				    enum dma_data_direction direction);
1608 	void		(*unmap_page)(struct ib_device *dev,
1609 				      u64 addr, size_t size,
1610 				      enum dma_data_direction direction);
1611 	int		(*map_sg)(struct ib_device *dev,
1612 				  struct scatterlist *sg, int nents,
1613 				  enum dma_data_direction direction);
1614 	void		(*unmap_sg)(struct ib_device *dev,
1615 				    struct scatterlist *sg, int nents,
1616 				    enum dma_data_direction direction);
1617 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1618 					       u64 dma_handle,
1619 					       size_t size,
1620 					       enum dma_data_direction dir);
1621 	void		(*sync_single_for_device)(struct ib_device *dev,
1622 						  u64 dma_handle,
1623 						  size_t size,
1624 						  enum dma_data_direction dir);
1625 	void		*(*alloc_coherent)(struct ib_device *dev,
1626 					   size_t size,
1627 					   u64 *dma_handle,
1628 					   gfp_t flag);
1629 	void		(*free_coherent)(struct ib_device *dev,
1630 					 size_t size, void *cpu_addr,
1631 					 u64 dma_handle);
1632 };
1633 
1634 struct iw_cm_verbs;
1635 
1636 struct ib_port_immutable {
1637 	int                           pkey_tbl_len;
1638 	int                           gid_tbl_len;
1639 	u32                           core_cap_flags;
1640 	u32                           max_mad_size;
1641 };
1642 
1643 struct ib_device {
1644 	struct device                *dma_device;
1645 
1646 	char                          name[IB_DEVICE_NAME_MAX];
1647 
1648 	struct list_head              event_handler_list;
1649 	spinlock_t                    event_handler_lock;
1650 
1651 	spinlock_t                    client_data_lock;
1652 	struct list_head              core_list;
1653 	/* Access to the client_data_list is protected by the client_data_lock
1654 	 * spinlock and the lists_rwsem read-write semaphore */
1655 	struct list_head              client_data_list;
1656 
1657 	struct ib_cache               cache;
1658 	/**
1659 	 * port_immutable is indexed by port number
1660 	 */
1661 	struct ib_port_immutable     *port_immutable;
1662 
1663 	int			      num_comp_vectors;
1664 
1665 	struct iw_cm_verbs	     *iwcm;
1666 
1667 	int		           (*get_protocol_stats)(struct ib_device *device,
1668 							 union rdma_protocol_stats *stats);
1669 	int		           (*query_device)(struct ib_device *device,
1670 						   struct ib_device_attr *device_attr,
1671 						   struct ib_udata *udata);
1672 	int		           (*query_port)(struct ib_device *device,
1673 						 u8 port_num,
1674 						 struct ib_port_attr *port_attr);
1675 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1676 						     u8 port_num);
1677 	/* When calling get_netdev, the HW vendor's driver should return the
1678 	 * net device of device @device at port @port_num or NULL if such
1679 	 * a net device doesn't exist. The vendor driver should call dev_hold
1680 	 * on this net device. The HW vendor's device driver must guarantee
1681 	 * that this function returns NULL before the net device reaches
1682 	 * NETDEV_UNREGISTER_FINAL state.
1683 	 */
1684 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1685 						 u8 port_num);
1686 	int		           (*query_gid)(struct ib_device *device,
1687 						u8 port_num, int index,
1688 						union ib_gid *gid);
1689 	/* When calling add_gid, the HW vendor's driver should
1690 	 * add the gid of device @device at gid index @index of
1691 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1692 	 * the network device related to this gid is available
1693 	 * at @attr. @context allows the HW vendor driver to store extra
1694 	 * information together with a GID entry. The HW vendor may allocate
1695 	 * memory to contain this information and store it in @context when a
1696 	 * new GID entry is written to. Params are consistent until the next
1697 	 * call of add_gid or delete_gid. The function should return 0 on
1698 	 * success or error otherwise. The function could be called
1699 	 * concurrently for different ports. This function is only called
1700 	 * when roce_gid_table is used.
1701 	 */
1702 	int		           (*add_gid)(struct ib_device *device,
1703 					      u8 port_num,
1704 					      unsigned int index,
1705 					      const union ib_gid *gid,
1706 					      const struct ib_gid_attr *attr,
1707 					      void **context);
1708 	/* When calling del_gid, the HW vendor's driver should delete the
1709 	 * gid of device @device at gid index @index of port @port_num.
1710 	 * Upon the deletion of a GID entry, the HW vendor must free any
1711 	 * allocated memory. The caller will clear @context afterwards.
1712 	 * This function is only called when roce_gid_table is used.
1713 	 */
1714 	int		           (*del_gid)(struct ib_device *device,
1715 					      u8 port_num,
1716 					      unsigned int index,
1717 					      void **context);
1718 	int		           (*query_pkey)(struct ib_device *device,
1719 						 u8 port_num, u16 index, u16 *pkey);
1720 	int		           (*modify_device)(struct ib_device *device,
1721 						    int device_modify_mask,
1722 						    struct ib_device_modify *device_modify);
1723 	int		           (*modify_port)(struct ib_device *device,
1724 						  u8 port_num, int port_modify_mask,
1725 						  struct ib_port_modify *port_modify);
1726 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1727 						     struct ib_udata *udata);
1728 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1729 	int                        (*mmap)(struct ib_ucontext *context,
1730 					   struct vm_area_struct *vma);
1731 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1732 					       struct ib_ucontext *context,
1733 					       struct ib_udata *udata);
1734 	int                        (*dealloc_pd)(struct ib_pd *pd);
1735 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1736 						struct ib_ah_attr *ah_attr);
1737 	int                        (*modify_ah)(struct ib_ah *ah,
1738 						struct ib_ah_attr *ah_attr);
1739 	int                        (*query_ah)(struct ib_ah *ah,
1740 					       struct ib_ah_attr *ah_attr);
1741 	int                        (*destroy_ah)(struct ib_ah *ah);
1742 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1743 						 struct ib_srq_init_attr *srq_init_attr,
1744 						 struct ib_udata *udata);
1745 	int                        (*modify_srq)(struct ib_srq *srq,
1746 						 struct ib_srq_attr *srq_attr,
1747 						 enum ib_srq_attr_mask srq_attr_mask,
1748 						 struct ib_udata *udata);
1749 	int                        (*query_srq)(struct ib_srq *srq,
1750 						struct ib_srq_attr *srq_attr);
1751 	int                        (*destroy_srq)(struct ib_srq *srq);
1752 	int                        (*post_srq_recv)(struct ib_srq *srq,
1753 						    struct ib_recv_wr *recv_wr,
1754 						    struct ib_recv_wr **bad_recv_wr);
1755 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1756 						struct ib_qp_init_attr *qp_init_attr,
1757 						struct ib_udata *udata);
1758 	int                        (*modify_qp)(struct ib_qp *qp,
1759 						struct ib_qp_attr *qp_attr,
1760 						int qp_attr_mask,
1761 						struct ib_udata *udata);
1762 	int                        (*query_qp)(struct ib_qp *qp,
1763 					       struct ib_qp_attr *qp_attr,
1764 					       int qp_attr_mask,
1765 					       struct ib_qp_init_attr *qp_init_attr);
1766 	int                        (*destroy_qp)(struct ib_qp *qp);
1767 	int                        (*post_send)(struct ib_qp *qp,
1768 						struct ib_send_wr *send_wr,
1769 						struct ib_send_wr **bad_send_wr);
1770 	int                        (*post_recv)(struct ib_qp *qp,
1771 						struct ib_recv_wr *recv_wr,
1772 						struct ib_recv_wr **bad_recv_wr);
1773 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1774 						const struct ib_cq_init_attr *attr,
1775 						struct ib_ucontext *context,
1776 						struct ib_udata *udata);
1777 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1778 						u16 cq_period);
1779 	int                        (*destroy_cq)(struct ib_cq *cq);
1780 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1781 						struct ib_udata *udata);
1782 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1783 					      struct ib_wc *wc);
1784 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1785 	int                        (*req_notify_cq)(struct ib_cq *cq,
1786 						    enum ib_cq_notify_flags flags);
1787 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1788 						      int wc_cnt);
1789 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1790 						 int mr_access_flags);
1791 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1792 						  u64 start, u64 length,
1793 						  u64 virt_addr,
1794 						  int mr_access_flags,
1795 						  struct ib_udata *udata);
1796 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1797 						    int flags,
1798 						    u64 start, u64 length,
1799 						    u64 virt_addr,
1800 						    int mr_access_flags,
1801 						    struct ib_pd *pd,
1802 						    struct ib_udata *udata);
1803 	int                        (*dereg_mr)(struct ib_mr *mr);
1804 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
1805 					       enum ib_mr_type mr_type,
1806 					       u32 max_num_sg);
1807 	int                        (*map_mr_sg)(struct ib_mr *mr,
1808 						struct scatterlist *sg,
1809 						int sg_nents);
1810 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
1811 					       enum ib_mw_type type);
1812 	int                        (*dealloc_mw)(struct ib_mw *mw);
1813 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
1814 						int mr_access_flags,
1815 						struct ib_fmr_attr *fmr_attr);
1816 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
1817 						   u64 *page_list, int list_len,
1818 						   u64 iova);
1819 	int		           (*unmap_fmr)(struct list_head *fmr_list);
1820 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
1821 	int                        (*attach_mcast)(struct ib_qp *qp,
1822 						   union ib_gid *gid,
1823 						   u16 lid);
1824 	int                        (*detach_mcast)(struct ib_qp *qp,
1825 						   union ib_gid *gid,
1826 						   u16 lid);
1827 	int                        (*process_mad)(struct ib_device *device,
1828 						  int process_mad_flags,
1829 						  u8 port_num,
1830 						  const struct ib_wc *in_wc,
1831 						  const struct ib_grh *in_grh,
1832 						  const struct ib_mad_hdr *in_mad,
1833 						  size_t in_mad_size,
1834 						  struct ib_mad_hdr *out_mad,
1835 						  size_t *out_mad_size,
1836 						  u16 *out_mad_pkey_index);
1837 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
1838 						 struct ib_ucontext *ucontext,
1839 						 struct ib_udata *udata);
1840 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
1841 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
1842 						  struct ib_flow_attr
1843 						  *flow_attr,
1844 						  int domain);
1845 	int			   (*destroy_flow)(struct ib_flow *flow_id);
1846 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
1847 						      struct ib_mr_status *mr_status);
1848 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
1849 
1850 	struct ib_dma_mapping_ops   *dma_ops;
1851 
1852 	struct module               *owner;
1853 	struct device                dev;
1854 	struct kobject               *ports_parent;
1855 	struct list_head             port_list;
1856 
1857 	enum {
1858 		IB_DEV_UNINITIALIZED,
1859 		IB_DEV_REGISTERED,
1860 		IB_DEV_UNREGISTERED
1861 	}                            reg_state;
1862 
1863 	int			     uverbs_abi_ver;
1864 	u64			     uverbs_cmd_mask;
1865 	u64			     uverbs_ex_cmd_mask;
1866 
1867 	char			     node_desc[64];
1868 	__be64			     node_guid;
1869 	u32			     local_dma_lkey;
1870 	u16                          is_switch:1;
1871 	u8                           node_type;
1872 	u8                           phys_port_cnt;
1873 	struct ib_device_attr        attrs;
1874 
1875 	/**
1876 	 * The following mandatory functions are used only at device
1877 	 * registration.  Keep functions such as these at the end of this
1878 	 * structure to avoid cache line misses when accessing struct ib_device
1879 	 * in fast paths.
1880 	 */
1881 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
1882 };
1883 
1884 struct ib_client {
1885 	char  *name;
1886 	void (*add)   (struct ib_device *);
1887 	void (*remove)(struct ib_device *, void *client_data);
1888 
1889 	/* Returns the net_dev belonging to this ib_client and matching the
1890 	 * given parameters.
1891 	 * @dev:	 An RDMA device that the net_dev use for communication.
1892 	 * @port:	 A physical port number on the RDMA device.
1893 	 * @pkey:	 P_Key that the net_dev uses if applicable.
1894 	 * @gid:	 A GID that the net_dev uses to communicate.
1895 	 * @addr:	 An IP address the net_dev is configured with.
1896 	 * @client_data: The device's client data set by ib_set_client_data().
1897 	 *
1898 	 * An ib_client that implements a net_dev on top of RDMA devices
1899 	 * (such as IP over IB) should implement this callback, allowing the
1900 	 * rdma_cm module to find the right net_dev for a given request.
1901 	 *
1902 	 * The caller is responsible for calling dev_put on the returned
1903 	 * netdev. */
1904 	struct net_device *(*get_net_dev_by_params)(
1905 			struct ib_device *dev,
1906 			u8 port,
1907 			u16 pkey,
1908 			const union ib_gid *gid,
1909 			const struct sockaddr *addr,
1910 			void *client_data);
1911 	struct list_head list;
1912 };
1913 
1914 struct ib_device *ib_alloc_device(size_t size);
1915 void ib_dealloc_device(struct ib_device *device);
1916 
1917 int ib_register_device(struct ib_device *device,
1918 		       int (*port_callback)(struct ib_device *,
1919 					    u8, struct kobject *));
1920 void ib_unregister_device(struct ib_device *device);
1921 
1922 int ib_register_client   (struct ib_client *client);
1923 void ib_unregister_client(struct ib_client *client);
1924 
1925 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
1926 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
1927 			 void *data);
1928 
1929 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
1930 {
1931 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
1932 }
1933 
1934 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
1935 {
1936 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
1937 }
1938 
1939 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
1940 				       size_t offset,
1941 				       size_t len)
1942 {
1943 	const void __user *p = udata->inbuf + offset;
1944 	bool ret = false;
1945 	u8 *buf;
1946 
1947 	if (len > USHRT_MAX)
1948 		return false;
1949 
1950 	buf = kmalloc(len, GFP_KERNEL);
1951 	if (!buf)
1952 		return false;
1953 
1954 	if (copy_from_user(buf, p, len))
1955 		goto free;
1956 
1957 	ret = !memchr_inv(buf, 0, len);
1958 
1959 free:
1960 	kfree(buf);
1961 	return ret;
1962 }
1963 
1964 /**
1965  * ib_modify_qp_is_ok - Check that the supplied attribute mask
1966  * contains all required attributes and no attributes not allowed for
1967  * the given QP state transition.
1968  * @cur_state: Current QP state
1969  * @next_state: Next QP state
1970  * @type: QP type
1971  * @mask: Mask of supplied QP attributes
1972  * @ll : link layer of port
1973  *
1974  * This function is a helper function that a low-level driver's
1975  * modify_qp method can use to validate the consumer's input.  It
1976  * checks that cur_state and next_state are valid QP states, that a
1977  * transition from cur_state to next_state is allowed by the IB spec,
1978  * and that the attribute mask supplied is allowed for the transition.
1979  */
1980 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
1981 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
1982 		       enum rdma_link_layer ll);
1983 
1984 int ib_register_event_handler  (struct ib_event_handler *event_handler);
1985 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
1986 void ib_dispatch_event(struct ib_event *event);
1987 
1988 int ib_query_port(struct ib_device *device,
1989 		  u8 port_num, struct ib_port_attr *port_attr);
1990 
1991 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
1992 					       u8 port_num);
1993 
1994 /**
1995  * rdma_cap_ib_switch - Check if the device is IB switch
1996  * @device: Device to check
1997  *
1998  * Device driver is responsible for setting is_switch bit on
1999  * in ib_device structure at init time.
2000  *
2001  * Return: true if the device is IB switch.
2002  */
2003 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2004 {
2005 	return device->is_switch;
2006 }
2007 
2008 /**
2009  * rdma_start_port - Return the first valid port number for the device
2010  * specified
2011  *
2012  * @device: Device to be checked
2013  *
2014  * Return start port number
2015  */
2016 static inline u8 rdma_start_port(const struct ib_device *device)
2017 {
2018 	return rdma_cap_ib_switch(device) ? 0 : 1;
2019 }
2020 
2021 /**
2022  * rdma_end_port - Return the last valid port number for the device
2023  * specified
2024  *
2025  * @device: Device to be checked
2026  *
2027  * Return last port number
2028  */
2029 static inline u8 rdma_end_port(const struct ib_device *device)
2030 {
2031 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2032 }
2033 
2034 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2035 {
2036 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2037 }
2038 
2039 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2040 {
2041 	return device->port_immutable[port_num].core_cap_flags &
2042 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2043 }
2044 
2045 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2046 {
2047 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2048 }
2049 
2050 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2051 {
2052 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2053 }
2054 
2055 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2056 {
2057 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2058 }
2059 
2060 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2061 {
2062 	return rdma_protocol_ib(device, port_num) ||
2063 		rdma_protocol_roce(device, port_num);
2064 }
2065 
2066 /**
2067  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2068  * Management Datagrams.
2069  * @device: Device to check
2070  * @port_num: Port number to check
2071  *
2072  * Management Datagrams (MAD) are a required part of the InfiniBand
2073  * specification and are supported on all InfiniBand devices.  A slightly
2074  * extended version are also supported on OPA interfaces.
2075  *
2076  * Return: true if the port supports sending/receiving of MAD packets.
2077  */
2078 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2079 {
2080 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2081 }
2082 
2083 /**
2084  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2085  * Management Datagrams.
2086  * @device: Device to check
2087  * @port_num: Port number to check
2088  *
2089  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2090  * datagrams with their own versions.  These OPA MADs share many but not all of
2091  * the characteristics of InfiniBand MADs.
2092  *
2093  * OPA MADs differ in the following ways:
2094  *
2095  *    1) MADs are variable size up to 2K
2096  *       IBTA defined MADs remain fixed at 256 bytes
2097  *    2) OPA SMPs must carry valid PKeys
2098  *    3) OPA SMP packets are a different format
2099  *
2100  * Return: true if the port supports OPA MAD packet formats.
2101  */
2102 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2103 {
2104 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2105 		== RDMA_CORE_CAP_OPA_MAD;
2106 }
2107 
2108 /**
2109  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2110  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2111  * @device: Device to check
2112  * @port_num: Port number to check
2113  *
2114  * Each InfiniBand node is required to provide a Subnet Management Agent
2115  * that the subnet manager can access.  Prior to the fabric being fully
2116  * configured by the subnet manager, the SMA is accessed via a well known
2117  * interface called the Subnet Management Interface (SMI).  This interface
2118  * uses directed route packets to communicate with the SM to get around the
2119  * chicken and egg problem of the SM needing to know what's on the fabric
2120  * in order to configure the fabric, and needing to configure the fabric in
2121  * order to send packets to the devices on the fabric.  These directed
2122  * route packets do not need the fabric fully configured in order to reach
2123  * their destination.  The SMI is the only method allowed to send
2124  * directed route packets on an InfiniBand fabric.
2125  *
2126  * Return: true if the port provides an SMI.
2127  */
2128 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2129 {
2130 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2131 }
2132 
2133 /**
2134  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2135  * Communication Manager.
2136  * @device: Device to check
2137  * @port_num: Port number to check
2138  *
2139  * The InfiniBand Communication Manager is one of many pre-defined General
2140  * Service Agents (GSA) that are accessed via the General Service
2141  * Interface (GSI).  It's role is to facilitate establishment of connections
2142  * between nodes as well as other management related tasks for established
2143  * connections.
2144  *
2145  * Return: true if the port supports an IB CM (this does not guarantee that
2146  * a CM is actually running however).
2147  */
2148 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2149 {
2150 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2151 }
2152 
2153 /**
2154  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2155  * Communication Manager.
2156  * @device: Device to check
2157  * @port_num: Port number to check
2158  *
2159  * Similar to above, but specific to iWARP connections which have a different
2160  * managment protocol than InfiniBand.
2161  *
2162  * Return: true if the port supports an iWARP CM (this does not guarantee that
2163  * a CM is actually running however).
2164  */
2165 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2166 {
2167 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2168 }
2169 
2170 /**
2171  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2172  * Subnet Administration.
2173  * @device: Device to check
2174  * @port_num: Port number to check
2175  *
2176  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2177  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2178  * fabrics, devices should resolve routes to other hosts by contacting the
2179  * SA to query the proper route.
2180  *
2181  * Return: true if the port should act as a client to the fabric Subnet
2182  * Administration interface.  This does not imply that the SA service is
2183  * running locally.
2184  */
2185 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2186 {
2187 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2188 }
2189 
2190 /**
2191  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2192  * Multicast.
2193  * @device: Device to check
2194  * @port_num: Port number to check
2195  *
2196  * InfiniBand multicast registration is more complex than normal IPv4 or
2197  * IPv6 multicast registration.  Each Host Channel Adapter must register
2198  * with the Subnet Manager when it wishes to join a multicast group.  It
2199  * should do so only once regardless of how many queue pairs it subscribes
2200  * to this group.  And it should leave the group only after all queue pairs
2201  * attached to the group have been detached.
2202  *
2203  * Return: true if the port must undertake the additional adminstrative
2204  * overhead of registering/unregistering with the SM and tracking of the
2205  * total number of queue pairs attached to the multicast group.
2206  */
2207 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2208 {
2209 	return rdma_cap_ib_sa(device, port_num);
2210 }
2211 
2212 /**
2213  * rdma_cap_af_ib - Check if the port of device has the capability
2214  * Native Infiniband Address.
2215  * @device: Device to check
2216  * @port_num: Port number to check
2217  *
2218  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2219  * GID.  RoCE uses a different mechanism, but still generates a GID via
2220  * a prescribed mechanism and port specific data.
2221  *
2222  * Return: true if the port uses a GID address to identify devices on the
2223  * network.
2224  */
2225 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2226 {
2227 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2228 }
2229 
2230 /**
2231  * rdma_cap_eth_ah - Check if the port of device has the capability
2232  * Ethernet Address Handle.
2233  * @device: Device to check
2234  * @port_num: Port number to check
2235  *
2236  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2237  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2238  * port.  Normally, packet headers are generated by the sending host
2239  * adapter, but when sending connectionless datagrams, we must manually
2240  * inject the proper headers for the fabric we are communicating over.
2241  *
2242  * Return: true if we are running as a RoCE port and must force the
2243  * addition of a Global Route Header built from our Ethernet Address
2244  * Handle into our header list for connectionless packets.
2245  */
2246 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2247 {
2248 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2249 }
2250 
2251 /**
2252  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2253  *
2254  * @device: Device
2255  * @port_num: Port number
2256  *
2257  * This MAD size includes the MAD headers and MAD payload.  No other headers
2258  * are included.
2259  *
2260  * Return the max MAD size required by the Port.  Will return 0 if the port
2261  * does not support MADs
2262  */
2263 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2264 {
2265 	return device->port_immutable[port_num].max_mad_size;
2266 }
2267 
2268 /**
2269  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2270  * @device: Device to check
2271  * @port_num: Port number to check
2272  *
2273  * RoCE GID table mechanism manages the various GIDs for a device.
2274  *
2275  * NOTE: if allocating the port's GID table has failed, this call will still
2276  * return true, but any RoCE GID table API will fail.
2277  *
2278  * Return: true if the port uses RoCE GID table mechanism in order to manage
2279  * its GIDs.
2280  */
2281 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2282 					   u8 port_num)
2283 {
2284 	return rdma_protocol_roce(device, port_num) &&
2285 		device->add_gid && device->del_gid;
2286 }
2287 
2288 int ib_query_gid(struct ib_device *device,
2289 		 u8 port_num, int index, union ib_gid *gid,
2290 		 struct ib_gid_attr *attr);
2291 
2292 int ib_query_pkey(struct ib_device *device,
2293 		  u8 port_num, u16 index, u16 *pkey);
2294 
2295 int ib_modify_device(struct ib_device *device,
2296 		     int device_modify_mask,
2297 		     struct ib_device_modify *device_modify);
2298 
2299 int ib_modify_port(struct ib_device *device,
2300 		   u8 port_num, int port_modify_mask,
2301 		   struct ib_port_modify *port_modify);
2302 
2303 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2304 		enum ib_gid_type gid_type, struct net_device *ndev,
2305 		u8 *port_num, u16 *index);
2306 
2307 int ib_find_pkey(struct ib_device *device,
2308 		 u8 port_num, u16 pkey, u16 *index);
2309 
2310 struct ib_pd *ib_alloc_pd(struct ib_device *device);
2311 
2312 void ib_dealloc_pd(struct ib_pd *pd);
2313 
2314 /**
2315  * ib_create_ah - Creates an address handle for the given address vector.
2316  * @pd: The protection domain associated with the address handle.
2317  * @ah_attr: The attributes of the address vector.
2318  *
2319  * The address handle is used to reference a local or global destination
2320  * in all UD QP post sends.
2321  */
2322 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2323 
2324 /**
2325  * ib_init_ah_from_wc - Initializes address handle attributes from a
2326  *   work completion.
2327  * @device: Device on which the received message arrived.
2328  * @port_num: Port on which the received message arrived.
2329  * @wc: Work completion associated with the received message.
2330  * @grh: References the received global route header.  This parameter is
2331  *   ignored unless the work completion indicates that the GRH is valid.
2332  * @ah_attr: Returned attributes that can be used when creating an address
2333  *   handle for replying to the message.
2334  */
2335 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2336 		       const struct ib_wc *wc, const struct ib_grh *grh,
2337 		       struct ib_ah_attr *ah_attr);
2338 
2339 /**
2340  * ib_create_ah_from_wc - Creates an address handle associated with the
2341  *   sender of the specified work completion.
2342  * @pd: The protection domain associated with the address handle.
2343  * @wc: Work completion information associated with a received message.
2344  * @grh: References the received global route header.  This parameter is
2345  *   ignored unless the work completion indicates that the GRH is valid.
2346  * @port_num: The outbound port number to associate with the address.
2347  *
2348  * The address handle is used to reference a local or global destination
2349  * in all UD QP post sends.
2350  */
2351 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2352 				   const struct ib_grh *grh, u8 port_num);
2353 
2354 /**
2355  * ib_modify_ah - Modifies the address vector associated with an address
2356  *   handle.
2357  * @ah: The address handle to modify.
2358  * @ah_attr: The new address vector attributes to associate with the
2359  *   address handle.
2360  */
2361 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2362 
2363 /**
2364  * ib_query_ah - Queries the address vector associated with an address
2365  *   handle.
2366  * @ah: The address handle to query.
2367  * @ah_attr: The address vector attributes associated with the address
2368  *   handle.
2369  */
2370 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2371 
2372 /**
2373  * ib_destroy_ah - Destroys an address handle.
2374  * @ah: The address handle to destroy.
2375  */
2376 int ib_destroy_ah(struct ib_ah *ah);
2377 
2378 /**
2379  * ib_create_srq - Creates a SRQ associated with the specified protection
2380  *   domain.
2381  * @pd: The protection domain associated with the SRQ.
2382  * @srq_init_attr: A list of initial attributes required to create the
2383  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2384  *   the actual capabilities of the created SRQ.
2385  *
2386  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2387  * requested size of the SRQ, and set to the actual values allocated
2388  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2389  * will always be at least as large as the requested values.
2390  */
2391 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2392 			     struct ib_srq_init_attr *srq_init_attr);
2393 
2394 /**
2395  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2396  * @srq: The SRQ to modify.
2397  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2398  *   the current values of selected SRQ attributes are returned.
2399  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2400  *   are being modified.
2401  *
2402  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2403  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2404  * the number of receives queued drops below the limit.
2405  */
2406 int ib_modify_srq(struct ib_srq *srq,
2407 		  struct ib_srq_attr *srq_attr,
2408 		  enum ib_srq_attr_mask srq_attr_mask);
2409 
2410 /**
2411  * ib_query_srq - Returns the attribute list and current values for the
2412  *   specified SRQ.
2413  * @srq: The SRQ to query.
2414  * @srq_attr: The attributes of the specified SRQ.
2415  */
2416 int ib_query_srq(struct ib_srq *srq,
2417 		 struct ib_srq_attr *srq_attr);
2418 
2419 /**
2420  * ib_destroy_srq - Destroys the specified SRQ.
2421  * @srq: The SRQ to destroy.
2422  */
2423 int ib_destroy_srq(struct ib_srq *srq);
2424 
2425 /**
2426  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2427  * @srq: The SRQ to post the work request on.
2428  * @recv_wr: A list of work requests to post on the receive queue.
2429  * @bad_recv_wr: On an immediate failure, this parameter will reference
2430  *   the work request that failed to be posted on the QP.
2431  */
2432 static inline int ib_post_srq_recv(struct ib_srq *srq,
2433 				   struct ib_recv_wr *recv_wr,
2434 				   struct ib_recv_wr **bad_recv_wr)
2435 {
2436 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2437 }
2438 
2439 /**
2440  * ib_create_qp - Creates a QP associated with the specified protection
2441  *   domain.
2442  * @pd: The protection domain associated with the QP.
2443  * @qp_init_attr: A list of initial attributes required to create the
2444  *   QP.  If QP creation succeeds, then the attributes are updated to
2445  *   the actual capabilities of the created QP.
2446  */
2447 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2448 			   struct ib_qp_init_attr *qp_init_attr);
2449 
2450 /**
2451  * ib_modify_qp - Modifies the attributes for the specified QP and then
2452  *   transitions the QP to the given state.
2453  * @qp: The QP to modify.
2454  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2455  *   the current values of selected QP attributes are returned.
2456  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2457  *   are being modified.
2458  */
2459 int ib_modify_qp(struct ib_qp *qp,
2460 		 struct ib_qp_attr *qp_attr,
2461 		 int qp_attr_mask);
2462 
2463 /**
2464  * ib_query_qp - Returns the attribute list and current values for the
2465  *   specified QP.
2466  * @qp: The QP to query.
2467  * @qp_attr: The attributes of the specified QP.
2468  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2469  * @qp_init_attr: Additional attributes of the selected QP.
2470  *
2471  * The qp_attr_mask may be used to limit the query to gathering only the
2472  * selected attributes.
2473  */
2474 int ib_query_qp(struct ib_qp *qp,
2475 		struct ib_qp_attr *qp_attr,
2476 		int qp_attr_mask,
2477 		struct ib_qp_init_attr *qp_init_attr);
2478 
2479 /**
2480  * ib_destroy_qp - Destroys the specified QP.
2481  * @qp: The QP to destroy.
2482  */
2483 int ib_destroy_qp(struct ib_qp *qp);
2484 
2485 /**
2486  * ib_open_qp - Obtain a reference to an existing sharable QP.
2487  * @xrcd - XRC domain
2488  * @qp_open_attr: Attributes identifying the QP to open.
2489  *
2490  * Returns a reference to a sharable QP.
2491  */
2492 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2493 			 struct ib_qp_open_attr *qp_open_attr);
2494 
2495 /**
2496  * ib_close_qp - Release an external reference to a QP.
2497  * @qp: The QP handle to release
2498  *
2499  * The opened QP handle is released by the caller.  The underlying
2500  * shared QP is not destroyed until all internal references are released.
2501  */
2502 int ib_close_qp(struct ib_qp *qp);
2503 
2504 /**
2505  * ib_post_send - Posts a list of work requests to the send queue of
2506  *   the specified QP.
2507  * @qp: The QP to post the work request on.
2508  * @send_wr: A list of work requests to post on the send queue.
2509  * @bad_send_wr: On an immediate failure, this parameter will reference
2510  *   the work request that failed to be posted on the QP.
2511  *
2512  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2513  * error is returned, the QP state shall not be affected,
2514  * ib_post_send() will return an immediate error after queueing any
2515  * earlier work requests in the list.
2516  */
2517 static inline int ib_post_send(struct ib_qp *qp,
2518 			       struct ib_send_wr *send_wr,
2519 			       struct ib_send_wr **bad_send_wr)
2520 {
2521 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2522 }
2523 
2524 /**
2525  * ib_post_recv - Posts a list of work requests to the receive queue of
2526  *   the specified QP.
2527  * @qp: The QP to post the work request on.
2528  * @recv_wr: A list of work requests to post on the receive queue.
2529  * @bad_recv_wr: On an immediate failure, this parameter will reference
2530  *   the work request that failed to be posted on the QP.
2531  */
2532 static inline int ib_post_recv(struct ib_qp *qp,
2533 			       struct ib_recv_wr *recv_wr,
2534 			       struct ib_recv_wr **bad_recv_wr)
2535 {
2536 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2537 }
2538 
2539 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2540 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2541 void ib_free_cq(struct ib_cq *cq);
2542 int ib_process_cq_direct(struct ib_cq *cq, int budget);
2543 
2544 /**
2545  * ib_create_cq - Creates a CQ on the specified device.
2546  * @device: The device on which to create the CQ.
2547  * @comp_handler: A user-specified callback that is invoked when a
2548  *   completion event occurs on the CQ.
2549  * @event_handler: A user-specified callback that is invoked when an
2550  *   asynchronous event not associated with a completion occurs on the CQ.
2551  * @cq_context: Context associated with the CQ returned to the user via
2552  *   the associated completion and event handlers.
2553  * @cq_attr: The attributes the CQ should be created upon.
2554  *
2555  * Users can examine the cq structure to determine the actual CQ size.
2556  */
2557 struct ib_cq *ib_create_cq(struct ib_device *device,
2558 			   ib_comp_handler comp_handler,
2559 			   void (*event_handler)(struct ib_event *, void *),
2560 			   void *cq_context,
2561 			   const struct ib_cq_init_attr *cq_attr);
2562 
2563 /**
2564  * ib_resize_cq - Modifies the capacity of the CQ.
2565  * @cq: The CQ to resize.
2566  * @cqe: The minimum size of the CQ.
2567  *
2568  * Users can examine the cq structure to determine the actual CQ size.
2569  */
2570 int ib_resize_cq(struct ib_cq *cq, int cqe);
2571 
2572 /**
2573  * ib_modify_cq - Modifies moderation params of the CQ
2574  * @cq: The CQ to modify.
2575  * @cq_count: number of CQEs that will trigger an event
2576  * @cq_period: max period of time in usec before triggering an event
2577  *
2578  */
2579 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2580 
2581 /**
2582  * ib_destroy_cq - Destroys the specified CQ.
2583  * @cq: The CQ to destroy.
2584  */
2585 int ib_destroy_cq(struct ib_cq *cq);
2586 
2587 /**
2588  * ib_poll_cq - poll a CQ for completion(s)
2589  * @cq:the CQ being polled
2590  * @num_entries:maximum number of completions to return
2591  * @wc:array of at least @num_entries &struct ib_wc where completions
2592  *   will be returned
2593  *
2594  * Poll a CQ for (possibly multiple) completions.  If the return value
2595  * is < 0, an error occurred.  If the return value is >= 0, it is the
2596  * number of completions returned.  If the return value is
2597  * non-negative and < num_entries, then the CQ was emptied.
2598  */
2599 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2600 			     struct ib_wc *wc)
2601 {
2602 	return cq->device->poll_cq(cq, num_entries, wc);
2603 }
2604 
2605 /**
2606  * ib_peek_cq - Returns the number of unreaped completions currently
2607  *   on the specified CQ.
2608  * @cq: The CQ to peek.
2609  * @wc_cnt: A minimum number of unreaped completions to check for.
2610  *
2611  * If the number of unreaped completions is greater than or equal to wc_cnt,
2612  * this function returns wc_cnt, otherwise, it returns the actual number of
2613  * unreaped completions.
2614  */
2615 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2616 
2617 /**
2618  * ib_req_notify_cq - Request completion notification on a CQ.
2619  * @cq: The CQ to generate an event for.
2620  * @flags:
2621  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2622  *   to request an event on the next solicited event or next work
2623  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2624  *   may also be |ed in to request a hint about missed events, as
2625  *   described below.
2626  *
2627  * Return Value:
2628  *    < 0 means an error occurred while requesting notification
2629  *   == 0 means notification was requested successfully, and if
2630  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2631  *        were missed and it is safe to wait for another event.  In
2632  *        this case is it guaranteed that any work completions added
2633  *        to the CQ since the last CQ poll will trigger a completion
2634  *        notification event.
2635  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2636  *        in.  It means that the consumer must poll the CQ again to
2637  *        make sure it is empty to avoid missing an event because of a
2638  *        race between requesting notification and an entry being
2639  *        added to the CQ.  This return value means it is possible
2640  *        (but not guaranteed) that a work completion has been added
2641  *        to the CQ since the last poll without triggering a
2642  *        completion notification event.
2643  */
2644 static inline int ib_req_notify_cq(struct ib_cq *cq,
2645 				   enum ib_cq_notify_flags flags)
2646 {
2647 	return cq->device->req_notify_cq(cq, flags);
2648 }
2649 
2650 /**
2651  * ib_req_ncomp_notif - Request completion notification when there are
2652  *   at least the specified number of unreaped completions on the CQ.
2653  * @cq: The CQ to generate an event for.
2654  * @wc_cnt: The number of unreaped completions that should be on the
2655  *   CQ before an event is generated.
2656  */
2657 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2658 {
2659 	return cq->device->req_ncomp_notif ?
2660 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2661 		-ENOSYS;
2662 }
2663 
2664 /**
2665  * ib_get_dma_mr - Returns a memory region for system memory that is
2666  *   usable for DMA.
2667  * @pd: The protection domain associated with the memory region.
2668  * @mr_access_flags: Specifies the memory access rights.
2669  *
2670  * Note that the ib_dma_*() functions defined below must be used
2671  * to create/destroy addresses used with the Lkey or Rkey returned
2672  * by ib_get_dma_mr().
2673  */
2674 struct ib_mr *ib_get_dma_mr(struct ib_pd *pd, int mr_access_flags);
2675 
2676 /**
2677  * ib_dma_mapping_error - check a DMA addr for error
2678  * @dev: The device for which the dma_addr was created
2679  * @dma_addr: The DMA address to check
2680  */
2681 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2682 {
2683 	if (dev->dma_ops)
2684 		return dev->dma_ops->mapping_error(dev, dma_addr);
2685 	return dma_mapping_error(dev->dma_device, dma_addr);
2686 }
2687 
2688 /**
2689  * ib_dma_map_single - Map a kernel virtual address to DMA address
2690  * @dev: The device for which the dma_addr is to be created
2691  * @cpu_addr: The kernel virtual address
2692  * @size: The size of the region in bytes
2693  * @direction: The direction of the DMA
2694  */
2695 static inline u64 ib_dma_map_single(struct ib_device *dev,
2696 				    void *cpu_addr, size_t size,
2697 				    enum dma_data_direction direction)
2698 {
2699 	if (dev->dma_ops)
2700 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2701 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2702 }
2703 
2704 /**
2705  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2706  * @dev: The device for which the DMA address was created
2707  * @addr: The DMA address
2708  * @size: The size of the region in bytes
2709  * @direction: The direction of the DMA
2710  */
2711 static inline void ib_dma_unmap_single(struct ib_device *dev,
2712 				       u64 addr, size_t size,
2713 				       enum dma_data_direction direction)
2714 {
2715 	if (dev->dma_ops)
2716 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2717 	else
2718 		dma_unmap_single(dev->dma_device, addr, size, direction);
2719 }
2720 
2721 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2722 					  void *cpu_addr, size_t size,
2723 					  enum dma_data_direction direction,
2724 					  struct dma_attrs *attrs)
2725 {
2726 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2727 				    direction, attrs);
2728 }
2729 
2730 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2731 					     u64 addr, size_t size,
2732 					     enum dma_data_direction direction,
2733 					     struct dma_attrs *attrs)
2734 {
2735 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2736 				      direction, attrs);
2737 }
2738 
2739 /**
2740  * ib_dma_map_page - Map a physical page to DMA address
2741  * @dev: The device for which the dma_addr is to be created
2742  * @page: The page to be mapped
2743  * @offset: The offset within the page
2744  * @size: The size of the region in bytes
2745  * @direction: The direction of the DMA
2746  */
2747 static inline u64 ib_dma_map_page(struct ib_device *dev,
2748 				  struct page *page,
2749 				  unsigned long offset,
2750 				  size_t size,
2751 					 enum dma_data_direction direction)
2752 {
2753 	if (dev->dma_ops)
2754 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
2755 	return dma_map_page(dev->dma_device, page, offset, size, direction);
2756 }
2757 
2758 /**
2759  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
2760  * @dev: The device for which the DMA address was created
2761  * @addr: The DMA address
2762  * @size: The size of the region in bytes
2763  * @direction: The direction of the DMA
2764  */
2765 static inline void ib_dma_unmap_page(struct ib_device *dev,
2766 				     u64 addr, size_t size,
2767 				     enum dma_data_direction direction)
2768 {
2769 	if (dev->dma_ops)
2770 		dev->dma_ops->unmap_page(dev, addr, size, direction);
2771 	else
2772 		dma_unmap_page(dev->dma_device, addr, size, direction);
2773 }
2774 
2775 /**
2776  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
2777  * @dev: The device for which the DMA addresses are to be created
2778  * @sg: The array of scatter/gather entries
2779  * @nents: The number of scatter/gather entries
2780  * @direction: The direction of the DMA
2781  */
2782 static inline int ib_dma_map_sg(struct ib_device *dev,
2783 				struct scatterlist *sg, int nents,
2784 				enum dma_data_direction direction)
2785 {
2786 	if (dev->dma_ops)
2787 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
2788 	return dma_map_sg(dev->dma_device, sg, nents, direction);
2789 }
2790 
2791 /**
2792  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
2793  * @dev: The device for which the DMA addresses were created
2794  * @sg: The array of scatter/gather entries
2795  * @nents: The number of scatter/gather entries
2796  * @direction: The direction of the DMA
2797  */
2798 static inline void ib_dma_unmap_sg(struct ib_device *dev,
2799 				   struct scatterlist *sg, int nents,
2800 				   enum dma_data_direction direction)
2801 {
2802 	if (dev->dma_ops)
2803 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
2804 	else
2805 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
2806 }
2807 
2808 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
2809 				      struct scatterlist *sg, int nents,
2810 				      enum dma_data_direction direction,
2811 				      struct dma_attrs *attrs)
2812 {
2813 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2814 }
2815 
2816 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
2817 					 struct scatterlist *sg, int nents,
2818 					 enum dma_data_direction direction,
2819 					 struct dma_attrs *attrs)
2820 {
2821 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, attrs);
2822 }
2823 /**
2824  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
2825  * @dev: The device for which the DMA addresses were created
2826  * @sg: The scatter/gather entry
2827  *
2828  * Note: this function is obsolete. To do: change all occurrences of
2829  * ib_sg_dma_address() into sg_dma_address().
2830  */
2831 static inline u64 ib_sg_dma_address(struct ib_device *dev,
2832 				    struct scatterlist *sg)
2833 {
2834 	return sg_dma_address(sg);
2835 }
2836 
2837 /**
2838  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
2839  * @dev: The device for which the DMA addresses were created
2840  * @sg: The scatter/gather entry
2841  *
2842  * Note: this function is obsolete. To do: change all occurrences of
2843  * ib_sg_dma_len() into sg_dma_len().
2844  */
2845 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
2846 					 struct scatterlist *sg)
2847 {
2848 	return sg_dma_len(sg);
2849 }
2850 
2851 /**
2852  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
2853  * @dev: The device for which the DMA address was created
2854  * @addr: The DMA address
2855  * @size: The size of the region in bytes
2856  * @dir: The direction of the DMA
2857  */
2858 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
2859 					      u64 addr,
2860 					      size_t size,
2861 					      enum dma_data_direction dir)
2862 {
2863 	if (dev->dma_ops)
2864 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
2865 	else
2866 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2867 }
2868 
2869 /**
2870  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
2871  * @dev: The device for which the DMA address was created
2872  * @addr: The DMA address
2873  * @size: The size of the region in bytes
2874  * @dir: The direction of the DMA
2875  */
2876 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
2877 						 u64 addr,
2878 						 size_t size,
2879 						 enum dma_data_direction dir)
2880 {
2881 	if (dev->dma_ops)
2882 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
2883 	else
2884 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2885 }
2886 
2887 /**
2888  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
2889  * @dev: The device for which the DMA address is requested
2890  * @size: The size of the region to allocate in bytes
2891  * @dma_handle: A pointer for returning the DMA address of the region
2892  * @flag: memory allocator flags
2893  */
2894 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
2895 					   size_t size,
2896 					   u64 *dma_handle,
2897 					   gfp_t flag)
2898 {
2899 	if (dev->dma_ops)
2900 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
2901 	else {
2902 		dma_addr_t handle;
2903 		void *ret;
2904 
2905 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
2906 		*dma_handle = handle;
2907 		return ret;
2908 	}
2909 }
2910 
2911 /**
2912  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
2913  * @dev: The device for which the DMA addresses were allocated
2914  * @size: The size of the region
2915  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
2916  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
2917  */
2918 static inline void ib_dma_free_coherent(struct ib_device *dev,
2919 					size_t size, void *cpu_addr,
2920 					u64 dma_handle)
2921 {
2922 	if (dev->dma_ops)
2923 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
2924 	else
2925 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
2926 }
2927 
2928 /**
2929  * ib_dereg_mr - Deregisters a memory region and removes it from the
2930  *   HCA translation table.
2931  * @mr: The memory region to deregister.
2932  *
2933  * This function can fail, if the memory region has memory windows bound to it.
2934  */
2935 int ib_dereg_mr(struct ib_mr *mr);
2936 
2937 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
2938 			  enum ib_mr_type mr_type,
2939 			  u32 max_num_sg);
2940 
2941 /**
2942  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
2943  *   R_Key and L_Key.
2944  * @mr - struct ib_mr pointer to be updated.
2945  * @newkey - new key to be used.
2946  */
2947 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
2948 {
2949 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
2950 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
2951 }
2952 
2953 /**
2954  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
2955  * for calculating a new rkey for type 2 memory windows.
2956  * @rkey - the rkey to increment.
2957  */
2958 static inline u32 ib_inc_rkey(u32 rkey)
2959 {
2960 	const u32 mask = 0x000000ff;
2961 	return ((rkey + 1) & mask) | (rkey & ~mask);
2962 }
2963 
2964 /**
2965  * ib_alloc_fmr - Allocates a unmapped fast memory region.
2966  * @pd: The protection domain associated with the unmapped region.
2967  * @mr_access_flags: Specifies the memory access rights.
2968  * @fmr_attr: Attributes of the unmapped region.
2969  *
2970  * A fast memory region must be mapped before it can be used as part of
2971  * a work request.
2972  */
2973 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
2974 			    int mr_access_flags,
2975 			    struct ib_fmr_attr *fmr_attr);
2976 
2977 /**
2978  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
2979  * @fmr: The fast memory region to associate with the pages.
2980  * @page_list: An array of physical pages to map to the fast memory region.
2981  * @list_len: The number of pages in page_list.
2982  * @iova: The I/O virtual address to use with the mapped region.
2983  */
2984 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
2985 				  u64 *page_list, int list_len,
2986 				  u64 iova)
2987 {
2988 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
2989 }
2990 
2991 /**
2992  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
2993  * @fmr_list: A linked list of fast memory regions to unmap.
2994  */
2995 int ib_unmap_fmr(struct list_head *fmr_list);
2996 
2997 /**
2998  * ib_dealloc_fmr - Deallocates a fast memory region.
2999  * @fmr: The fast memory region to deallocate.
3000  */
3001 int ib_dealloc_fmr(struct ib_fmr *fmr);
3002 
3003 /**
3004  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3005  * @qp: QP to attach to the multicast group.  The QP must be type
3006  *   IB_QPT_UD.
3007  * @gid: Multicast group GID.
3008  * @lid: Multicast group LID in host byte order.
3009  *
3010  * In order to send and receive multicast packets, subnet
3011  * administration must have created the multicast group and configured
3012  * the fabric appropriately.  The port associated with the specified
3013  * QP must also be a member of the multicast group.
3014  */
3015 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3016 
3017 /**
3018  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3019  * @qp: QP to detach from the multicast group.
3020  * @gid: Multicast group GID.
3021  * @lid: Multicast group LID in host byte order.
3022  */
3023 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3024 
3025 /**
3026  * ib_alloc_xrcd - Allocates an XRC domain.
3027  * @device: The device on which to allocate the XRC domain.
3028  */
3029 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3030 
3031 /**
3032  * ib_dealloc_xrcd - Deallocates an XRC domain.
3033  * @xrcd: The XRC domain to deallocate.
3034  */
3035 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3036 
3037 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3038 			       struct ib_flow_attr *flow_attr, int domain);
3039 int ib_destroy_flow(struct ib_flow *flow_id);
3040 
3041 static inline int ib_check_mr_access(int flags)
3042 {
3043 	/*
3044 	 * Local write permission is required if remote write or
3045 	 * remote atomic permission is also requested.
3046 	 */
3047 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3048 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3049 		return -EINVAL;
3050 
3051 	return 0;
3052 }
3053 
3054 /**
3055  * ib_check_mr_status: lightweight check of MR status.
3056  *     This routine may provide status checks on a selected
3057  *     ib_mr. first use is for signature status check.
3058  *
3059  * @mr: A memory region.
3060  * @check_mask: Bitmask of which checks to perform from
3061  *     ib_mr_status_check enumeration.
3062  * @mr_status: The container of relevant status checks.
3063  *     failed checks will be indicated in the status bitmask
3064  *     and the relevant info shall be in the error item.
3065  */
3066 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3067 		       struct ib_mr_status *mr_status);
3068 
3069 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3070 					    u16 pkey, const union ib_gid *gid,
3071 					    const struct sockaddr *addr);
3072 
3073 int ib_map_mr_sg(struct ib_mr *mr,
3074 		 struct scatterlist *sg,
3075 		 int sg_nents,
3076 		 unsigned int page_size);
3077 
3078 static inline int
3079 ib_map_mr_sg_zbva(struct ib_mr *mr,
3080 		  struct scatterlist *sg,
3081 		  int sg_nents,
3082 		  unsigned int page_size)
3083 {
3084 	int n;
3085 
3086 	n = ib_map_mr_sg(mr, sg, sg_nents, page_size);
3087 	mr->iova = 0;
3088 
3089 	return n;
3090 }
3091 
3092 int ib_sg_to_pages(struct ib_mr *mr,
3093 		   struct scatterlist *sgl,
3094 		   int sg_nents,
3095 		   int (*set_page)(struct ib_mr *, u64));
3096 
3097 #endif /* IB_VERBS_H */
3098