xref: /freebsd/sys/ofed/include/rdma/ib_verbs.h (revision 5bf5ca772c6de2d53344a78cf461447cc322ccea)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause OR GPL-2.0
3  *
4  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
5  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
6  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
7  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
8  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
9  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
10  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
11  *
12  * This software is available to you under a choice of one of two
13  * licenses.  You may choose to be licensed under the terms of the GNU
14  * General Public License (GPL) Version 2, available from the file
15  * COPYING in the main directory of this source tree, or the
16  * OpenIB.org BSD license below:
17  *
18  *     Redistribution and use in source and binary forms, with or
19  *     without modification, are permitted provided that the following
20  *     conditions are met:
21  *
22  *      - Redistributions of source code must retain the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer.
25  *
26  *      - Redistributions in binary form must reproduce the above
27  *        copyright notice, this list of conditions and the following
28  *        disclaimer in the documentation and/or other materials
29  *        provided with the distribution.
30  *
31  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
32  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
33  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
34  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
35  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
36  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
37  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
38  * SOFTWARE.
39  *
40  * $FreeBSD$
41  */
42 
43 #if !defined(IB_VERBS_H)
44 #define IB_VERBS_H
45 
46 #include <linux/types.h>
47 #include <linux/device.h>
48 #include <linux/mm.h>
49 #include <linux/dma-mapping.h>
50 #include <linux/kref.h>
51 #include <linux/list.h>
52 #include <linux/rwsem.h>
53 #include <linux/scatterlist.h>
54 #include <linux/workqueue.h>
55 #include <linux/socket.h>
56 #include <linux/if_ether.h>
57 #include <net/ipv6.h>
58 #include <net/ip.h>
59 #include <linux/string.h>
60 #include <linux/slab.h>
61 #include <linux/rcupdate.h>
62 #include <linux/netdevice.h>
63 #include <netinet/ip.h>
64 
65 #include <asm/atomic.h>
66 #include <asm/uaccess.h>
67 
68 struct ifla_vf_info;
69 struct ifla_vf_stats;
70 
71 extern struct workqueue_struct *ib_wq;
72 extern struct workqueue_struct *ib_comp_wq;
73 
74 union ib_gid {
75 	u8	raw[16];
76 	struct {
77 		__be64	subnet_prefix;
78 		__be64	interface_id;
79 	} global;
80 };
81 
82 extern union ib_gid zgid;
83 
84 enum ib_gid_type {
85 	/* If link layer is Ethernet, this is RoCE V1 */
86 	IB_GID_TYPE_IB        = 0,
87 	IB_GID_TYPE_ROCE      = 0,
88 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
89 	IB_GID_TYPE_SIZE
90 };
91 
92 #define ROCE_V2_UDP_DPORT      4791
93 struct ib_gid_attr {
94 	enum ib_gid_type	gid_type;
95 	struct net_device	*ndev;
96 };
97 
98 enum rdma_node_type {
99 	/* IB values map to NodeInfo:NodeType. */
100 	RDMA_NODE_IB_CA 	= 1,
101 	RDMA_NODE_IB_SWITCH,
102 	RDMA_NODE_IB_ROUTER,
103 	RDMA_NODE_RNIC,
104 	RDMA_NODE_USNIC,
105 	RDMA_NODE_USNIC_UDP,
106 };
107 
108 enum {
109 	/* set the local administered indication */
110 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
111 };
112 
113 enum rdma_transport_type {
114 	RDMA_TRANSPORT_IB,
115 	RDMA_TRANSPORT_IWARP,
116 	RDMA_TRANSPORT_USNIC,
117 	RDMA_TRANSPORT_USNIC_UDP
118 };
119 
120 enum rdma_protocol_type {
121 	RDMA_PROTOCOL_IB,
122 	RDMA_PROTOCOL_IBOE,
123 	RDMA_PROTOCOL_IWARP,
124 	RDMA_PROTOCOL_USNIC_UDP
125 };
126 
127 __attribute_const__ enum rdma_transport_type
128 rdma_node_get_transport(enum rdma_node_type node_type);
129 
130 enum rdma_network_type {
131 	RDMA_NETWORK_IB,
132 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
133 	RDMA_NETWORK_IPV4,
134 	RDMA_NETWORK_IPV6
135 };
136 
137 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
138 {
139 	if (network_type == RDMA_NETWORK_IPV4 ||
140 	    network_type == RDMA_NETWORK_IPV6)
141 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
142 
143 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
144 	return IB_GID_TYPE_IB;
145 }
146 
147 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
148 							    union ib_gid *gid)
149 {
150 	if (gid_type == IB_GID_TYPE_IB)
151 		return RDMA_NETWORK_IB;
152 
153 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
154 		return RDMA_NETWORK_IPV4;
155 	else
156 		return RDMA_NETWORK_IPV6;
157 }
158 
159 enum rdma_link_layer {
160 	IB_LINK_LAYER_UNSPECIFIED,
161 	IB_LINK_LAYER_INFINIBAND,
162 	IB_LINK_LAYER_ETHERNET,
163 };
164 
165 enum ib_device_cap_flags {
166 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
167 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
168 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
169 	IB_DEVICE_RAW_MULTI			= (1 << 3),
170 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
171 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
172 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
173 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
174 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
175 	IB_DEVICE_INIT_TYPE			= (1 << 9),
176 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
177 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
178 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
179 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
180 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
181 
182 	/*
183 	 * This device supports a per-device lkey or stag that can be
184 	 * used without performing a memory registration for the local
185 	 * memory.  Note that ULPs should never check this flag, but
186 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
187 	 * which will always contain a usable lkey.
188 	 */
189 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
190 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
191 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
192 	/*
193 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
194 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
195 	 * messages and can verify the validity of checksum for
196 	 * incoming messages.  Setting this flag implies that the
197 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
198 	 */
199 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
200 	IB_DEVICE_UD_TSO			= (1 << 19),
201 	IB_DEVICE_XRC				= (1 << 20),
202 
203 	/*
204 	 * This device supports the IB "base memory management extension",
205 	 * which includes support for fast registrations (IB_WR_REG_MR,
206 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
207 	 * also be set by any iWarp device which must support FRs to comply
208 	 * to the iWarp verbs spec.  iWarp devices also support the
209 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
210 	 * stag.
211 	 */
212 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
213 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
214 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
215 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
216 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
217 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
218 	/*
219 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
220 	 * support execution of WQEs that involve synchronization
221 	 * of I/O operations with single completion queue managed
222 	 * by hardware.
223 	 */
224 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
225 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
226 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
227 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
228 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
229 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
230 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
231 };
232 
233 enum ib_signature_prot_cap {
234 	IB_PROT_T10DIF_TYPE_1 = 1,
235 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
236 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
237 };
238 
239 enum ib_signature_guard_cap {
240 	IB_GUARD_T10DIF_CRC	= 1,
241 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
242 };
243 
244 enum ib_atomic_cap {
245 	IB_ATOMIC_NONE,
246 	IB_ATOMIC_HCA,
247 	IB_ATOMIC_GLOB
248 };
249 
250 enum ib_odp_general_cap_bits {
251 	IB_ODP_SUPPORT = 1 << 0,
252 };
253 
254 enum ib_odp_transport_cap_bits {
255 	IB_ODP_SUPPORT_SEND	= 1 << 0,
256 	IB_ODP_SUPPORT_RECV	= 1 << 1,
257 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
258 	IB_ODP_SUPPORT_READ	= 1 << 3,
259 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
260 };
261 
262 struct ib_odp_caps {
263 	uint64_t general_caps;
264 	struct {
265 		uint32_t  rc_odp_caps;
266 		uint32_t  uc_odp_caps;
267 		uint32_t  ud_odp_caps;
268 	} per_transport_caps;
269 };
270 
271 struct ib_rss_caps {
272 	/* Corresponding bit will be set if qp type from
273 	 * 'enum ib_qp_type' is supported, e.g.
274 	 * supported_qpts |= 1 << IB_QPT_UD
275 	 */
276 	u32 supported_qpts;
277 	u32 max_rwq_indirection_tables;
278 	u32 max_rwq_indirection_table_size;
279 };
280 
281 enum ib_cq_creation_flags {
282 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
283 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
284 };
285 
286 struct ib_cq_init_attr {
287 	unsigned int	cqe;
288 	int		comp_vector;
289 	u32		flags;
290 };
291 
292 struct ib_device_attr {
293 	u64			fw_ver;
294 	__be64			sys_image_guid;
295 	u64			max_mr_size;
296 	u64			page_size_cap;
297 	u32			vendor_id;
298 	u32			vendor_part_id;
299 	u32			hw_ver;
300 	int			max_qp;
301 	int			max_qp_wr;
302 	u64			device_cap_flags;
303 	int			max_sge;
304 	int			max_sge_rd;
305 	int			max_cq;
306 	int			max_cqe;
307 	int			max_mr;
308 	int			max_pd;
309 	int			max_qp_rd_atom;
310 	int			max_ee_rd_atom;
311 	int			max_res_rd_atom;
312 	int			max_qp_init_rd_atom;
313 	int			max_ee_init_rd_atom;
314 	enum ib_atomic_cap	atomic_cap;
315 	enum ib_atomic_cap	masked_atomic_cap;
316 	int			max_ee;
317 	int			max_rdd;
318 	int			max_mw;
319 	int			max_raw_ipv6_qp;
320 	int			max_raw_ethy_qp;
321 	int			max_mcast_grp;
322 	int			max_mcast_qp_attach;
323 	int			max_total_mcast_qp_attach;
324 	int			max_ah;
325 	int			max_fmr;
326 	int			max_map_per_fmr;
327 	int			max_srq;
328 	int			max_srq_wr;
329 	int			max_srq_sge;
330 	unsigned int		max_fast_reg_page_list_len;
331 	u16			max_pkeys;
332 	u8			local_ca_ack_delay;
333 	int			sig_prot_cap;
334 	int			sig_guard_cap;
335 	struct ib_odp_caps	odp_caps;
336 	uint64_t		timestamp_mask;
337 	uint64_t		hca_core_clock; /* in KHZ */
338 	struct ib_rss_caps	rss_caps;
339 	u32			max_wq_type_rq;
340 };
341 
342 enum ib_mtu {
343 	IB_MTU_256  = 1,
344 	IB_MTU_512  = 2,
345 	IB_MTU_1024 = 3,
346 	IB_MTU_2048 = 4,
347 	IB_MTU_4096 = 5
348 };
349 
350 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
351 {
352 	switch (mtu) {
353 	case IB_MTU_256:  return  256;
354 	case IB_MTU_512:  return  512;
355 	case IB_MTU_1024: return 1024;
356 	case IB_MTU_2048: return 2048;
357 	case IB_MTU_4096: return 4096;
358 	default: 	  return -1;
359 	}
360 }
361 
362 enum ib_port_state {
363 	IB_PORT_NOP		= 0,
364 	IB_PORT_DOWN		= 1,
365 	IB_PORT_INIT		= 2,
366 	IB_PORT_ARMED		= 3,
367 	IB_PORT_ACTIVE		= 4,
368 	IB_PORT_ACTIVE_DEFER	= 5,
369 	IB_PORT_DUMMY		= -1,	/* force enum signed */
370 };
371 
372 enum ib_port_cap_flags {
373 	IB_PORT_SM				= 1 <<  1,
374 	IB_PORT_NOTICE_SUP			= 1 <<  2,
375 	IB_PORT_TRAP_SUP			= 1 <<  3,
376 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
377 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
378 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
379 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
380 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
381 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
382 	IB_PORT_SM_DISABLED			= 1 << 10,
383 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
384 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
385 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
386 	IB_PORT_CM_SUP				= 1 << 16,
387 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
388 	IB_PORT_REINIT_SUP			= 1 << 18,
389 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
390 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
391 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
392 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
393 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
394 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
395 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
396 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
397 };
398 
399 enum ib_port_width {
400 	IB_WIDTH_1X	= 1,
401 	IB_WIDTH_4X	= 2,
402 	IB_WIDTH_8X	= 4,
403 	IB_WIDTH_12X	= 8
404 };
405 
406 static inline int ib_width_enum_to_int(enum ib_port_width width)
407 {
408 	switch (width) {
409 	case IB_WIDTH_1X:  return  1;
410 	case IB_WIDTH_4X:  return  4;
411 	case IB_WIDTH_8X:  return  8;
412 	case IB_WIDTH_12X: return 12;
413 	default: 	  return -1;
414 	}
415 }
416 
417 enum ib_port_speed {
418 	IB_SPEED_SDR	= 1,
419 	IB_SPEED_DDR	= 2,
420 	IB_SPEED_QDR	= 4,
421 	IB_SPEED_FDR10	= 8,
422 	IB_SPEED_FDR	= 16,
423 	IB_SPEED_EDR	= 32
424 };
425 
426 /**
427  * struct rdma_hw_stats
428  * @timestamp - Used by the core code to track when the last update was
429  * @lifespan - Used by the core code to determine how old the counters
430  *   should be before being updated again.  Stored in jiffies, defaults
431  *   to 10 milliseconds, drivers can override the default be specifying
432  *   their own value during their allocation routine.
433  * @name - Array of pointers to static names used for the counters in
434  *   directory.
435  * @num_counters - How many hardware counters there are.  If name is
436  *   shorter than this number, a kernel oops will result.  Driver authors
437  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
438  *   in their code to prevent this.
439  * @value - Array of u64 counters that are accessed by the sysfs code and
440  *   filled in by the drivers get_stats routine
441  */
442 struct rdma_hw_stats {
443 	unsigned long	timestamp;
444 	unsigned long	lifespan;
445 	const char * const *names;
446 	int		num_counters;
447 	u64		value[];
448 };
449 
450 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
451 /**
452  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
453  *   for drivers.
454  * @names - Array of static const char *
455  * @num_counters - How many elements in array
456  * @lifespan - How many milliseconds between updates
457  */
458 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
459 		const char * const *names, int num_counters,
460 		unsigned long lifespan)
461 {
462 	struct rdma_hw_stats *stats;
463 
464 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
465 			GFP_KERNEL);
466 	if (!stats)
467 		return NULL;
468 	stats->names = names;
469 	stats->num_counters = num_counters;
470 	stats->lifespan = msecs_to_jiffies(lifespan);
471 
472 	return stats;
473 }
474 
475 
476 /* Define bits for the various functionality this port needs to be supported by
477  * the core.
478  */
479 /* Management                           0x00000FFF */
480 #define RDMA_CORE_CAP_IB_MAD            0x00000001
481 #define RDMA_CORE_CAP_IB_SMI            0x00000002
482 #define RDMA_CORE_CAP_IB_CM             0x00000004
483 #define RDMA_CORE_CAP_IW_CM             0x00000008
484 #define RDMA_CORE_CAP_IB_SA             0x00000010
485 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
486 
487 /* Address format                       0x000FF000 */
488 #define RDMA_CORE_CAP_AF_IB             0x00001000
489 #define RDMA_CORE_CAP_ETH_AH            0x00002000
490 
491 /* Protocol                             0xFFF00000 */
492 #define RDMA_CORE_CAP_PROT_IB           0x00100000
493 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
494 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
495 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
496 
497 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
498 					| RDMA_CORE_CAP_IB_MAD \
499 					| RDMA_CORE_CAP_IB_SMI \
500 					| RDMA_CORE_CAP_IB_CM  \
501 					| RDMA_CORE_CAP_IB_SA  \
502 					| RDMA_CORE_CAP_AF_IB)
503 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
504 					| RDMA_CORE_CAP_IB_MAD  \
505 					| RDMA_CORE_CAP_IB_CM   \
506 					| RDMA_CORE_CAP_AF_IB   \
507 					| RDMA_CORE_CAP_ETH_AH)
508 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
509 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
510 					| RDMA_CORE_CAP_IB_MAD  \
511 					| RDMA_CORE_CAP_IB_CM   \
512 					| RDMA_CORE_CAP_AF_IB   \
513 					| RDMA_CORE_CAP_ETH_AH)
514 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
515 					| RDMA_CORE_CAP_IW_CM)
516 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
517 					| RDMA_CORE_CAP_OPA_MAD)
518 
519 struct ib_port_attr {
520 	u64			subnet_prefix;
521 	enum ib_port_state	state;
522 	enum ib_mtu		max_mtu;
523 	enum ib_mtu		active_mtu;
524 	int			gid_tbl_len;
525 	u32			port_cap_flags;
526 	u32			max_msg_sz;
527 	u32			bad_pkey_cntr;
528 	u32			qkey_viol_cntr;
529 	u16			pkey_tbl_len;
530 	u16			lid;
531 	u16			sm_lid;
532 	u8			lmc;
533 	u8			max_vl_num;
534 	u8			sm_sl;
535 	u8			subnet_timeout;
536 	u8			init_type_reply;
537 	u8			active_width;
538 	u8			active_speed;
539 	u8                      phys_state;
540 	bool			grh_required;
541 };
542 
543 enum ib_device_modify_flags {
544 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
545 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
546 };
547 
548 #define IB_DEVICE_NODE_DESC_MAX 64
549 
550 struct ib_device_modify {
551 	u64	sys_image_guid;
552 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
553 };
554 
555 enum ib_port_modify_flags {
556 	IB_PORT_SHUTDOWN		= 1,
557 	IB_PORT_INIT_TYPE		= (1<<2),
558 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
559 };
560 
561 struct ib_port_modify {
562 	u32	set_port_cap_mask;
563 	u32	clr_port_cap_mask;
564 	u8	init_type;
565 };
566 
567 enum ib_event_type {
568 	IB_EVENT_CQ_ERR,
569 	IB_EVENT_QP_FATAL,
570 	IB_EVENT_QP_REQ_ERR,
571 	IB_EVENT_QP_ACCESS_ERR,
572 	IB_EVENT_COMM_EST,
573 	IB_EVENT_SQ_DRAINED,
574 	IB_EVENT_PATH_MIG,
575 	IB_EVENT_PATH_MIG_ERR,
576 	IB_EVENT_DEVICE_FATAL,
577 	IB_EVENT_PORT_ACTIVE,
578 	IB_EVENT_PORT_ERR,
579 	IB_EVENT_LID_CHANGE,
580 	IB_EVENT_PKEY_CHANGE,
581 	IB_EVENT_SM_CHANGE,
582 	IB_EVENT_SRQ_ERR,
583 	IB_EVENT_SRQ_LIMIT_REACHED,
584 	IB_EVENT_QP_LAST_WQE_REACHED,
585 	IB_EVENT_CLIENT_REREGISTER,
586 	IB_EVENT_GID_CHANGE,
587 	IB_EVENT_WQ_FATAL,
588 };
589 
590 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
591 
592 struct ib_event {
593 	struct ib_device	*device;
594 	union {
595 		struct ib_cq	*cq;
596 		struct ib_qp	*qp;
597 		struct ib_srq	*srq;
598 		struct ib_wq	*wq;
599 		u8		port_num;
600 	} element;
601 	enum ib_event_type	event;
602 };
603 
604 struct ib_event_handler {
605 	struct ib_device *device;
606 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
607 	struct list_head  list;
608 };
609 
610 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
611 	do {							\
612 		(_ptr)->device  = _device;			\
613 		(_ptr)->handler = _handler;			\
614 		INIT_LIST_HEAD(&(_ptr)->list);			\
615 	} while (0)
616 
617 struct ib_global_route {
618 	union ib_gid	dgid;
619 	u32		flow_label;
620 	u8		sgid_index;
621 	u8		hop_limit;
622 	u8		traffic_class;
623 };
624 
625 struct ib_grh {
626 	__be32		version_tclass_flow;
627 	__be16		paylen;
628 	u8		next_hdr;
629 	u8		hop_limit;
630 	union ib_gid	sgid;
631 	union ib_gid	dgid;
632 };
633 
634 union rdma_network_hdr {
635 	struct ib_grh ibgrh;
636 	struct {
637 		/* The IB spec states that if it's IPv4, the header
638 		 * is located in the last 20 bytes of the header.
639 		 */
640 		u8		reserved[20];
641 		struct ip	roce4grh;
642 	};
643 };
644 
645 enum {
646 	IB_MULTICAST_QPN = 0xffffff
647 };
648 
649 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
650 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
651 
652 enum ib_ah_flags {
653 	IB_AH_GRH	= 1
654 };
655 
656 enum ib_rate {
657 	IB_RATE_PORT_CURRENT = 0,
658 	IB_RATE_2_5_GBPS = 2,
659 	IB_RATE_5_GBPS   = 5,
660 	IB_RATE_10_GBPS  = 3,
661 	IB_RATE_20_GBPS  = 6,
662 	IB_RATE_30_GBPS  = 4,
663 	IB_RATE_40_GBPS  = 7,
664 	IB_RATE_60_GBPS  = 8,
665 	IB_RATE_80_GBPS  = 9,
666 	IB_RATE_120_GBPS = 10,
667 	IB_RATE_14_GBPS  = 11,
668 	IB_RATE_56_GBPS  = 12,
669 	IB_RATE_112_GBPS = 13,
670 	IB_RATE_168_GBPS = 14,
671 	IB_RATE_25_GBPS  = 15,
672 	IB_RATE_100_GBPS = 16,
673 	IB_RATE_200_GBPS = 17,
674 	IB_RATE_300_GBPS = 18
675 };
676 
677 /**
678  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
679  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
680  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
681  * @rate: rate to convert.
682  */
683 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
684 
685 /**
686  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
687  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
688  * @rate: rate to convert.
689  */
690 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
691 
692 
693 /**
694  * enum ib_mr_type - memory region type
695  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
696  *                            normal registration
697  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
698  *                            signature operations (data-integrity
699  *                            capable regions)
700  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
701  *                            register any arbitrary sg lists (without
702  *                            the normal mr constraints - see
703  *                            ib_map_mr_sg)
704  */
705 enum ib_mr_type {
706 	IB_MR_TYPE_MEM_REG,
707 	IB_MR_TYPE_SIGNATURE,
708 	IB_MR_TYPE_SG_GAPS,
709 };
710 
711 /**
712  * Signature types
713  * IB_SIG_TYPE_NONE: Unprotected.
714  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
715  */
716 enum ib_signature_type {
717 	IB_SIG_TYPE_NONE,
718 	IB_SIG_TYPE_T10_DIF,
719 };
720 
721 /**
722  * Signature T10-DIF block-guard types
723  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
724  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
725  */
726 enum ib_t10_dif_bg_type {
727 	IB_T10DIF_CRC,
728 	IB_T10DIF_CSUM
729 };
730 
731 /**
732  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
733  *     domain.
734  * @bg_type: T10-DIF block guard type (CRC|CSUM)
735  * @pi_interval: protection information interval.
736  * @bg: seed of guard computation.
737  * @app_tag: application tag of guard block
738  * @ref_tag: initial guard block reference tag.
739  * @ref_remap: Indicate wethear the reftag increments each block
740  * @app_escape: Indicate to skip block check if apptag=0xffff
741  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
742  * @apptag_check_mask: check bitmask of application tag.
743  */
744 struct ib_t10_dif_domain {
745 	enum ib_t10_dif_bg_type bg_type;
746 	u16			pi_interval;
747 	u16			bg;
748 	u16			app_tag;
749 	u32			ref_tag;
750 	bool			ref_remap;
751 	bool			app_escape;
752 	bool			ref_escape;
753 	u16			apptag_check_mask;
754 };
755 
756 /**
757  * struct ib_sig_domain - Parameters for signature domain
758  * @sig_type: specific signauture type
759  * @sig: union of all signature domain attributes that may
760  *     be used to set domain layout.
761  */
762 struct ib_sig_domain {
763 	enum ib_signature_type sig_type;
764 	union {
765 		struct ib_t10_dif_domain dif;
766 	} sig;
767 };
768 
769 /**
770  * struct ib_sig_attrs - Parameters for signature handover operation
771  * @check_mask: bitmask for signature byte check (8 bytes)
772  * @mem: memory domain layout desciptor.
773  * @wire: wire domain layout desciptor.
774  */
775 struct ib_sig_attrs {
776 	u8			check_mask;
777 	struct ib_sig_domain	mem;
778 	struct ib_sig_domain	wire;
779 };
780 
781 enum ib_sig_err_type {
782 	IB_SIG_BAD_GUARD,
783 	IB_SIG_BAD_REFTAG,
784 	IB_SIG_BAD_APPTAG,
785 };
786 
787 /**
788  * struct ib_sig_err - signature error descriptor
789  */
790 struct ib_sig_err {
791 	enum ib_sig_err_type	err_type;
792 	u32			expected;
793 	u32			actual;
794 	u64			sig_err_offset;
795 	u32			key;
796 };
797 
798 enum ib_mr_status_check {
799 	IB_MR_CHECK_SIG_STATUS = 1,
800 };
801 
802 /**
803  * struct ib_mr_status - Memory region status container
804  *
805  * @fail_status: Bitmask of MR checks status. For each
806  *     failed check a corresponding status bit is set.
807  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
808  *     failure.
809  */
810 struct ib_mr_status {
811 	u32		    fail_status;
812 	struct ib_sig_err   sig_err;
813 };
814 
815 /**
816  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
817  * enum.
818  * @mult: multiple to convert.
819  */
820 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
821 
822 struct ib_ah_attr {
823 	struct ib_global_route	grh;
824 	u16			dlid;
825 	u8			sl;
826 	u8			src_path_bits;
827 	u8			static_rate;
828 	u8			ah_flags;
829 	u8			port_num;
830 	u8			dmac[ETH_ALEN];
831 };
832 
833 enum ib_wc_status {
834 	IB_WC_SUCCESS,
835 	IB_WC_LOC_LEN_ERR,
836 	IB_WC_LOC_QP_OP_ERR,
837 	IB_WC_LOC_EEC_OP_ERR,
838 	IB_WC_LOC_PROT_ERR,
839 	IB_WC_WR_FLUSH_ERR,
840 	IB_WC_MW_BIND_ERR,
841 	IB_WC_BAD_RESP_ERR,
842 	IB_WC_LOC_ACCESS_ERR,
843 	IB_WC_REM_INV_REQ_ERR,
844 	IB_WC_REM_ACCESS_ERR,
845 	IB_WC_REM_OP_ERR,
846 	IB_WC_RETRY_EXC_ERR,
847 	IB_WC_RNR_RETRY_EXC_ERR,
848 	IB_WC_LOC_RDD_VIOL_ERR,
849 	IB_WC_REM_INV_RD_REQ_ERR,
850 	IB_WC_REM_ABORT_ERR,
851 	IB_WC_INV_EECN_ERR,
852 	IB_WC_INV_EEC_STATE_ERR,
853 	IB_WC_FATAL_ERR,
854 	IB_WC_RESP_TIMEOUT_ERR,
855 	IB_WC_GENERAL_ERR
856 };
857 
858 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
859 
860 enum ib_wc_opcode {
861 	IB_WC_SEND,
862 	IB_WC_RDMA_WRITE,
863 	IB_WC_RDMA_READ,
864 	IB_WC_COMP_SWAP,
865 	IB_WC_FETCH_ADD,
866 	IB_WC_LSO,
867 	IB_WC_LOCAL_INV,
868 	IB_WC_REG_MR,
869 	IB_WC_MASKED_COMP_SWAP,
870 	IB_WC_MASKED_FETCH_ADD,
871 /*
872  * Set value of IB_WC_RECV so consumers can test if a completion is a
873  * receive by testing (opcode & IB_WC_RECV).
874  */
875 	IB_WC_RECV			= 1 << 7,
876 	IB_WC_RECV_RDMA_WITH_IMM,
877 	IB_WC_DUMMY = -1,	/* force enum signed */
878 };
879 
880 enum ib_wc_flags {
881 	IB_WC_GRH		= 1,
882 	IB_WC_WITH_IMM		= (1<<1),
883 	IB_WC_WITH_INVALIDATE	= (1<<2),
884 	IB_WC_IP_CSUM_OK	= (1<<3),
885 	IB_WC_WITH_SMAC		= (1<<4),
886 	IB_WC_WITH_VLAN		= (1<<5),
887 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
888 };
889 
890 struct ib_wc {
891 	union {
892 		u64		wr_id;
893 		struct ib_cqe	*wr_cqe;
894 	};
895 	enum ib_wc_status	status;
896 	enum ib_wc_opcode	opcode;
897 	u32			vendor_err;
898 	u32			byte_len;
899 	struct ib_qp	       *qp;
900 	union {
901 		__be32		imm_data;
902 		u32		invalidate_rkey;
903 	} ex;
904 	u32			src_qp;
905 	int			wc_flags;
906 	u16			pkey_index;
907 	u16			slid;
908 	u8			sl;
909 	u8			dlid_path_bits;
910 	u8			port_num;	/* valid only for DR SMPs on switches */
911 	u8			smac[ETH_ALEN];
912 	u16			vlan_id;
913 	u8			network_hdr_type;
914 };
915 
916 enum ib_cq_notify_flags {
917 	IB_CQ_SOLICITED			= 1 << 0,
918 	IB_CQ_NEXT_COMP			= 1 << 1,
919 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
920 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
921 };
922 
923 enum ib_srq_type {
924 	IB_SRQT_BASIC,
925 	IB_SRQT_XRC
926 };
927 
928 enum ib_srq_attr_mask {
929 	IB_SRQ_MAX_WR	= 1 << 0,
930 	IB_SRQ_LIMIT	= 1 << 1,
931 };
932 
933 struct ib_srq_attr {
934 	u32	max_wr;
935 	u32	max_sge;
936 	u32	srq_limit;
937 };
938 
939 struct ib_srq_init_attr {
940 	void		      (*event_handler)(struct ib_event *, void *);
941 	void		       *srq_context;
942 	struct ib_srq_attr	attr;
943 	enum ib_srq_type	srq_type;
944 
945 	union {
946 		struct {
947 			struct ib_xrcd *xrcd;
948 			struct ib_cq   *cq;
949 		} xrc;
950 	} ext;
951 };
952 
953 struct ib_qp_cap {
954 	u32	max_send_wr;
955 	u32	max_recv_wr;
956 	u32	max_send_sge;
957 	u32	max_recv_sge;
958 	u32	max_inline_data;
959 
960 	/*
961 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
962 	 * ib_create_qp() will calculate the right amount of neededed WRs
963 	 * and MRs based on this.
964 	 */
965 	u32	max_rdma_ctxs;
966 };
967 
968 enum ib_sig_type {
969 	IB_SIGNAL_ALL_WR,
970 	IB_SIGNAL_REQ_WR
971 };
972 
973 enum ib_qp_type {
974 	/*
975 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
976 	 * here (and in that order) since the MAD layer uses them as
977 	 * indices into a 2-entry table.
978 	 */
979 	IB_QPT_SMI,
980 	IB_QPT_GSI,
981 
982 	IB_QPT_RC,
983 	IB_QPT_UC,
984 	IB_QPT_UD,
985 	IB_QPT_RAW_IPV6,
986 	IB_QPT_RAW_ETHERTYPE,
987 	IB_QPT_RAW_PACKET = 8,
988 	IB_QPT_XRC_INI = 9,
989 	IB_QPT_XRC_TGT,
990 	IB_QPT_MAX,
991 	/* Reserve a range for qp types internal to the low level driver.
992 	 * These qp types will not be visible at the IB core layer, so the
993 	 * IB_QPT_MAX usages should not be affected in the core layer
994 	 */
995 	IB_QPT_RESERVED1 = 0x1000,
996 	IB_QPT_RESERVED2,
997 	IB_QPT_RESERVED3,
998 	IB_QPT_RESERVED4,
999 	IB_QPT_RESERVED5,
1000 	IB_QPT_RESERVED6,
1001 	IB_QPT_RESERVED7,
1002 	IB_QPT_RESERVED8,
1003 	IB_QPT_RESERVED9,
1004 	IB_QPT_RESERVED10,
1005 };
1006 
1007 enum ib_qp_create_flags {
1008 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1009 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1010 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1011 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1012 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1013 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1014 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1015 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1016 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1017 	/* reserve bits 26-31 for low level drivers' internal use */
1018 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1019 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1020 };
1021 
1022 /*
1023  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1024  * callback to destroy the passed in QP.
1025  */
1026 
1027 struct ib_qp_init_attr {
1028 	void                  (*event_handler)(struct ib_event *, void *);
1029 	void		       *qp_context;
1030 	struct ib_cq	       *send_cq;
1031 	struct ib_cq	       *recv_cq;
1032 	struct ib_srq	       *srq;
1033 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1034 	struct ib_qp_cap	cap;
1035 	enum ib_sig_type	sq_sig_type;
1036 	enum ib_qp_type		qp_type;
1037 	enum ib_qp_create_flags	create_flags;
1038 
1039 	/*
1040 	 * Only needed for special QP types, or when using the RW API.
1041 	 */
1042 	u8			port_num;
1043 	struct ib_rwq_ind_table *rwq_ind_tbl;
1044 };
1045 
1046 struct ib_qp_open_attr {
1047 	void                  (*event_handler)(struct ib_event *, void *);
1048 	void		       *qp_context;
1049 	u32			qp_num;
1050 	enum ib_qp_type		qp_type;
1051 };
1052 
1053 enum ib_rnr_timeout {
1054 	IB_RNR_TIMER_655_36 =  0,
1055 	IB_RNR_TIMER_000_01 =  1,
1056 	IB_RNR_TIMER_000_02 =  2,
1057 	IB_RNR_TIMER_000_03 =  3,
1058 	IB_RNR_TIMER_000_04 =  4,
1059 	IB_RNR_TIMER_000_06 =  5,
1060 	IB_RNR_TIMER_000_08 =  6,
1061 	IB_RNR_TIMER_000_12 =  7,
1062 	IB_RNR_TIMER_000_16 =  8,
1063 	IB_RNR_TIMER_000_24 =  9,
1064 	IB_RNR_TIMER_000_32 = 10,
1065 	IB_RNR_TIMER_000_48 = 11,
1066 	IB_RNR_TIMER_000_64 = 12,
1067 	IB_RNR_TIMER_000_96 = 13,
1068 	IB_RNR_TIMER_001_28 = 14,
1069 	IB_RNR_TIMER_001_92 = 15,
1070 	IB_RNR_TIMER_002_56 = 16,
1071 	IB_RNR_TIMER_003_84 = 17,
1072 	IB_RNR_TIMER_005_12 = 18,
1073 	IB_RNR_TIMER_007_68 = 19,
1074 	IB_RNR_TIMER_010_24 = 20,
1075 	IB_RNR_TIMER_015_36 = 21,
1076 	IB_RNR_TIMER_020_48 = 22,
1077 	IB_RNR_TIMER_030_72 = 23,
1078 	IB_RNR_TIMER_040_96 = 24,
1079 	IB_RNR_TIMER_061_44 = 25,
1080 	IB_RNR_TIMER_081_92 = 26,
1081 	IB_RNR_TIMER_122_88 = 27,
1082 	IB_RNR_TIMER_163_84 = 28,
1083 	IB_RNR_TIMER_245_76 = 29,
1084 	IB_RNR_TIMER_327_68 = 30,
1085 	IB_RNR_TIMER_491_52 = 31
1086 };
1087 
1088 enum ib_qp_attr_mask {
1089 	IB_QP_STATE			= 1,
1090 	IB_QP_CUR_STATE			= (1<<1),
1091 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1092 	IB_QP_ACCESS_FLAGS		= (1<<3),
1093 	IB_QP_PKEY_INDEX		= (1<<4),
1094 	IB_QP_PORT			= (1<<5),
1095 	IB_QP_QKEY			= (1<<6),
1096 	IB_QP_AV			= (1<<7),
1097 	IB_QP_PATH_MTU			= (1<<8),
1098 	IB_QP_TIMEOUT			= (1<<9),
1099 	IB_QP_RETRY_CNT			= (1<<10),
1100 	IB_QP_RNR_RETRY			= (1<<11),
1101 	IB_QP_RQ_PSN			= (1<<12),
1102 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1103 	IB_QP_ALT_PATH			= (1<<14),
1104 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1105 	IB_QP_SQ_PSN			= (1<<16),
1106 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1107 	IB_QP_PATH_MIG_STATE		= (1<<18),
1108 	IB_QP_CAP			= (1<<19),
1109 	IB_QP_DEST_QPN			= (1<<20),
1110 	IB_QP_RESERVED1			= (1<<21),
1111 	IB_QP_RESERVED2			= (1<<22),
1112 	IB_QP_RESERVED3			= (1<<23),
1113 	IB_QP_RESERVED4			= (1<<24),
1114 };
1115 
1116 enum ib_qp_state {
1117 	IB_QPS_RESET,
1118 	IB_QPS_INIT,
1119 	IB_QPS_RTR,
1120 	IB_QPS_RTS,
1121 	IB_QPS_SQD,
1122 	IB_QPS_SQE,
1123 	IB_QPS_ERR,
1124 	IB_QPS_DUMMY = -1,	/* force enum signed */
1125 };
1126 
1127 enum ib_mig_state {
1128 	IB_MIG_MIGRATED,
1129 	IB_MIG_REARM,
1130 	IB_MIG_ARMED
1131 };
1132 
1133 enum ib_mw_type {
1134 	IB_MW_TYPE_1 = 1,
1135 	IB_MW_TYPE_2 = 2
1136 };
1137 
1138 struct ib_qp_attr {
1139 	enum ib_qp_state	qp_state;
1140 	enum ib_qp_state	cur_qp_state;
1141 	enum ib_mtu		path_mtu;
1142 	enum ib_mig_state	path_mig_state;
1143 	u32			qkey;
1144 	u32			rq_psn;
1145 	u32			sq_psn;
1146 	u32			dest_qp_num;
1147 	int			qp_access_flags;
1148 	struct ib_qp_cap	cap;
1149 	struct ib_ah_attr	ah_attr;
1150 	struct ib_ah_attr	alt_ah_attr;
1151 	u16			pkey_index;
1152 	u16			alt_pkey_index;
1153 	u8			en_sqd_async_notify;
1154 	u8			sq_draining;
1155 	u8			max_rd_atomic;
1156 	u8			max_dest_rd_atomic;
1157 	u8			min_rnr_timer;
1158 	u8			port_num;
1159 	u8			timeout;
1160 	u8			retry_cnt;
1161 	u8			rnr_retry;
1162 	u8			alt_port_num;
1163 	u8			alt_timeout;
1164 };
1165 
1166 enum ib_wr_opcode {
1167 	IB_WR_RDMA_WRITE,
1168 	IB_WR_RDMA_WRITE_WITH_IMM,
1169 	IB_WR_SEND,
1170 	IB_WR_SEND_WITH_IMM,
1171 	IB_WR_RDMA_READ,
1172 	IB_WR_ATOMIC_CMP_AND_SWP,
1173 	IB_WR_ATOMIC_FETCH_AND_ADD,
1174 	IB_WR_LSO,
1175 	IB_WR_SEND_WITH_INV,
1176 	IB_WR_RDMA_READ_WITH_INV,
1177 	IB_WR_LOCAL_INV,
1178 	IB_WR_REG_MR,
1179 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1180 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1181 	IB_WR_REG_SIG_MR,
1182 	/* reserve values for low level drivers' internal use.
1183 	 * These values will not be used at all in the ib core layer.
1184 	 */
1185 	IB_WR_RESERVED1 = 0xf0,
1186 	IB_WR_RESERVED2,
1187 	IB_WR_RESERVED3,
1188 	IB_WR_RESERVED4,
1189 	IB_WR_RESERVED5,
1190 	IB_WR_RESERVED6,
1191 	IB_WR_RESERVED7,
1192 	IB_WR_RESERVED8,
1193 	IB_WR_RESERVED9,
1194 	IB_WR_RESERVED10,
1195 	IB_WR_DUMMY = -1,	/* force enum signed */
1196 };
1197 
1198 enum ib_send_flags {
1199 	IB_SEND_FENCE		= 1,
1200 	IB_SEND_SIGNALED	= (1<<1),
1201 	IB_SEND_SOLICITED	= (1<<2),
1202 	IB_SEND_INLINE		= (1<<3),
1203 	IB_SEND_IP_CSUM		= (1<<4),
1204 
1205 	/* reserve bits 26-31 for low level drivers' internal use */
1206 	IB_SEND_RESERVED_START	= (1 << 26),
1207 	IB_SEND_RESERVED_END	= (1 << 31),
1208 };
1209 
1210 struct ib_sge {
1211 	u64	addr;
1212 	u32	length;
1213 	u32	lkey;
1214 };
1215 
1216 struct ib_cqe {
1217 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1218 };
1219 
1220 struct ib_send_wr {
1221 	struct ib_send_wr      *next;
1222 	union {
1223 		u64		wr_id;
1224 		struct ib_cqe	*wr_cqe;
1225 	};
1226 	struct ib_sge	       *sg_list;
1227 	int			num_sge;
1228 	enum ib_wr_opcode	opcode;
1229 	int			send_flags;
1230 	union {
1231 		__be32		imm_data;
1232 		u32		invalidate_rkey;
1233 	} ex;
1234 };
1235 
1236 struct ib_rdma_wr {
1237 	struct ib_send_wr	wr;
1238 	u64			remote_addr;
1239 	u32			rkey;
1240 };
1241 
1242 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1243 {
1244 	return container_of(wr, struct ib_rdma_wr, wr);
1245 }
1246 
1247 struct ib_atomic_wr {
1248 	struct ib_send_wr	wr;
1249 	u64			remote_addr;
1250 	u64			compare_add;
1251 	u64			swap;
1252 	u64			compare_add_mask;
1253 	u64			swap_mask;
1254 	u32			rkey;
1255 };
1256 
1257 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1258 {
1259 	return container_of(wr, struct ib_atomic_wr, wr);
1260 }
1261 
1262 struct ib_ud_wr {
1263 	struct ib_send_wr	wr;
1264 	struct ib_ah		*ah;
1265 	void			*header;
1266 	int			hlen;
1267 	int			mss;
1268 	u32			remote_qpn;
1269 	u32			remote_qkey;
1270 	u16			pkey_index; /* valid for GSI only */
1271 	u8			port_num;   /* valid for DR SMPs on switch only */
1272 };
1273 
1274 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1275 {
1276 	return container_of(wr, struct ib_ud_wr, wr);
1277 }
1278 
1279 struct ib_reg_wr {
1280 	struct ib_send_wr	wr;
1281 	struct ib_mr		*mr;
1282 	u32			key;
1283 	int			access;
1284 };
1285 
1286 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1287 {
1288 	return container_of(wr, struct ib_reg_wr, wr);
1289 }
1290 
1291 struct ib_sig_handover_wr {
1292 	struct ib_send_wr	wr;
1293 	struct ib_sig_attrs    *sig_attrs;
1294 	struct ib_mr	       *sig_mr;
1295 	int			access_flags;
1296 	struct ib_sge	       *prot;
1297 };
1298 
1299 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1300 {
1301 	return container_of(wr, struct ib_sig_handover_wr, wr);
1302 }
1303 
1304 struct ib_recv_wr {
1305 	struct ib_recv_wr      *next;
1306 	union {
1307 		u64		wr_id;
1308 		struct ib_cqe	*wr_cqe;
1309 	};
1310 	struct ib_sge	       *sg_list;
1311 	int			num_sge;
1312 };
1313 
1314 enum ib_access_flags {
1315 	IB_ACCESS_LOCAL_WRITE	= 1,
1316 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1317 	IB_ACCESS_REMOTE_READ	= (1<<2),
1318 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1319 	IB_ACCESS_MW_BIND	= (1<<4),
1320 	IB_ZERO_BASED		= (1<<5),
1321 	IB_ACCESS_ON_DEMAND     = (1<<6),
1322 };
1323 
1324 /*
1325  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1326  * are hidden here instead of a uapi header!
1327  */
1328 enum ib_mr_rereg_flags {
1329 	IB_MR_REREG_TRANS	= 1,
1330 	IB_MR_REREG_PD		= (1<<1),
1331 	IB_MR_REREG_ACCESS	= (1<<2),
1332 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1333 };
1334 
1335 struct ib_fmr_attr {
1336 	int	max_pages;
1337 	int	max_maps;
1338 	u8	page_shift;
1339 };
1340 
1341 struct ib_umem;
1342 
1343 struct ib_ucontext {
1344 	struct ib_device       *device;
1345 	struct list_head	pd_list;
1346 	struct list_head	mr_list;
1347 	struct list_head	mw_list;
1348 	struct list_head	cq_list;
1349 	struct list_head	qp_list;
1350 	struct list_head	srq_list;
1351 	struct list_head	ah_list;
1352 	struct list_head	xrcd_list;
1353 	struct list_head	rule_list;
1354 	struct list_head	wq_list;
1355 	struct list_head	rwq_ind_tbl_list;
1356 	int			closing;
1357 
1358 	pid_t			tgid;
1359 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1360 	struct rb_root      umem_tree;
1361 	/*
1362 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1363 	 * mmu notifiers registration.
1364 	 */
1365 	struct rw_semaphore	umem_rwsem;
1366 	void (*invalidate_range)(struct ib_umem *umem,
1367 				 unsigned long start, unsigned long end);
1368 
1369 	struct mmu_notifier	mn;
1370 	atomic_t		notifier_count;
1371 	/* A list of umems that don't have private mmu notifier counters yet. */
1372 	struct list_head	no_private_counters;
1373 	int                     odp_mrs_count;
1374 #endif
1375 };
1376 
1377 struct ib_uobject {
1378 	u64			user_handle;	/* handle given to us by userspace */
1379 	struct ib_ucontext     *context;	/* associated user context */
1380 	void		       *object;		/* containing object */
1381 	struct list_head	list;		/* link to context's list */
1382 	int			id;		/* index into kernel idr */
1383 	struct kref		ref;
1384 	struct rw_semaphore	mutex;		/* protects .live */
1385 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1386 	int			live;
1387 };
1388 
1389 struct ib_udata {
1390 	const void __user *inbuf;
1391 	void __user *outbuf;
1392 	size_t       inlen;
1393 	size_t       outlen;
1394 };
1395 
1396 struct ib_pd {
1397 	u32			local_dma_lkey;
1398 	u32			flags;
1399 	struct ib_device       *device;
1400 	struct ib_uobject      *uobject;
1401 	atomic_t          	usecnt; /* count all resources */
1402 
1403 	u32			unsafe_global_rkey;
1404 
1405 	/*
1406 	 * Implementation details of the RDMA core, don't use in drivers:
1407 	 */
1408 	struct ib_mr	       *__internal_mr;
1409 };
1410 
1411 struct ib_xrcd {
1412 	struct ib_device       *device;
1413 	atomic_t		usecnt; /* count all exposed resources */
1414 	struct inode	       *inode;
1415 
1416 	struct mutex		tgt_qp_mutex;
1417 	struct list_head	tgt_qp_list;
1418 };
1419 
1420 struct ib_ah {
1421 	struct ib_device	*device;
1422 	struct ib_pd		*pd;
1423 	struct ib_uobject	*uobject;
1424 };
1425 
1426 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1427 
1428 enum ib_poll_context {
1429 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1430 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1431 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1432 };
1433 
1434 struct ib_cq {
1435 	struct ib_device       *device;
1436 	struct ib_uobject      *uobject;
1437 	ib_comp_handler   	comp_handler;
1438 	void                  (*event_handler)(struct ib_event *, void *);
1439 	void                   *cq_context;
1440 	int               	cqe;
1441 	atomic_t          	usecnt; /* count number of work queues */
1442 	enum ib_poll_context	poll_ctx;
1443 	struct work_struct	work;
1444 };
1445 
1446 struct ib_srq {
1447 	struct ib_device       *device;
1448 	struct ib_pd	       *pd;
1449 	struct ib_uobject      *uobject;
1450 	void		      (*event_handler)(struct ib_event *, void *);
1451 	void		       *srq_context;
1452 	enum ib_srq_type	srq_type;
1453 	atomic_t		usecnt;
1454 
1455 	union {
1456 		struct {
1457 			struct ib_xrcd *xrcd;
1458 			struct ib_cq   *cq;
1459 			u32		srq_num;
1460 		} xrc;
1461 	} ext;
1462 };
1463 
1464 enum ib_wq_type {
1465 	IB_WQT_RQ
1466 };
1467 
1468 enum ib_wq_state {
1469 	IB_WQS_RESET,
1470 	IB_WQS_RDY,
1471 	IB_WQS_ERR
1472 };
1473 
1474 struct ib_wq {
1475 	struct ib_device       *device;
1476 	struct ib_uobject      *uobject;
1477 	void		    *wq_context;
1478 	void		    (*event_handler)(struct ib_event *, void *);
1479 	struct ib_pd	       *pd;
1480 	struct ib_cq	       *cq;
1481 	u32		wq_num;
1482 	enum ib_wq_state       state;
1483 	enum ib_wq_type	wq_type;
1484 	atomic_t		usecnt;
1485 };
1486 
1487 struct ib_wq_init_attr {
1488 	void		       *wq_context;
1489 	enum ib_wq_type	wq_type;
1490 	u32		max_wr;
1491 	u32		max_sge;
1492 	struct	ib_cq	       *cq;
1493 	void		    (*event_handler)(struct ib_event *, void *);
1494 };
1495 
1496 enum ib_wq_attr_mask {
1497 	IB_WQ_STATE	= 1 << 0,
1498 	IB_WQ_CUR_STATE	= 1 << 1,
1499 };
1500 
1501 struct ib_wq_attr {
1502 	enum	ib_wq_state	wq_state;
1503 	enum	ib_wq_state	curr_wq_state;
1504 };
1505 
1506 struct ib_rwq_ind_table {
1507 	struct ib_device	*device;
1508 	struct ib_uobject      *uobject;
1509 	atomic_t		usecnt;
1510 	u32		ind_tbl_num;
1511 	u32		log_ind_tbl_size;
1512 	struct ib_wq	**ind_tbl;
1513 };
1514 
1515 struct ib_rwq_ind_table_init_attr {
1516 	u32		log_ind_tbl_size;
1517 	/* Each entry is a pointer to Receive Work Queue */
1518 	struct ib_wq	**ind_tbl;
1519 };
1520 
1521 /*
1522  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1523  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1524  */
1525 struct ib_qp {
1526 	struct ib_device       *device;
1527 	struct ib_pd	       *pd;
1528 	struct ib_cq	       *send_cq;
1529 	struct ib_cq	       *recv_cq;
1530 	spinlock_t		mr_lock;
1531 	struct ib_srq	       *srq;
1532 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1533 	struct list_head	xrcd_list;
1534 
1535 	/* count times opened, mcast attaches, flow attaches */
1536 	atomic_t		usecnt;
1537 	struct list_head	open_list;
1538 	struct ib_qp           *real_qp;
1539 	struct ib_uobject      *uobject;
1540 	void                  (*event_handler)(struct ib_event *, void *);
1541 	void		       *qp_context;
1542 	u32			qp_num;
1543 	u32			max_write_sge;
1544 	u32			max_read_sge;
1545 	enum ib_qp_type		qp_type;
1546 	struct ib_rwq_ind_table *rwq_ind_tbl;
1547 };
1548 
1549 struct ib_mr {
1550 	struct ib_device  *device;
1551 	struct ib_pd	  *pd;
1552 	u32		   lkey;
1553 	u32		   rkey;
1554 	u64		   iova;
1555 	u32		   length;
1556 	unsigned int	   page_size;
1557 	bool		   need_inval;
1558 	union {
1559 		struct ib_uobject	*uobject;	/* user */
1560 		struct list_head	qp_entry;	/* FR */
1561 	};
1562 };
1563 
1564 struct ib_mw {
1565 	struct ib_device	*device;
1566 	struct ib_pd		*pd;
1567 	struct ib_uobject	*uobject;
1568 	u32			rkey;
1569 	enum ib_mw_type         type;
1570 };
1571 
1572 struct ib_fmr {
1573 	struct ib_device	*device;
1574 	struct ib_pd		*pd;
1575 	struct list_head	list;
1576 	u32			lkey;
1577 	u32			rkey;
1578 };
1579 
1580 /* Supported steering options */
1581 enum ib_flow_attr_type {
1582 	/* steering according to rule specifications */
1583 	IB_FLOW_ATTR_NORMAL		= 0x0,
1584 	/* default unicast and multicast rule -
1585 	 * receive all Eth traffic which isn't steered to any QP
1586 	 */
1587 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1588 	/* default multicast rule -
1589 	 * receive all Eth multicast traffic which isn't steered to any QP
1590 	 */
1591 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1592 	/* sniffer rule - receive all port traffic */
1593 	IB_FLOW_ATTR_SNIFFER		= 0x3
1594 };
1595 
1596 /* Supported steering header types */
1597 enum ib_flow_spec_type {
1598 	/* L2 headers*/
1599 	IB_FLOW_SPEC_ETH	= 0x20,
1600 	IB_FLOW_SPEC_IB		= 0x22,
1601 	/* L3 header*/
1602 	IB_FLOW_SPEC_IPV4	= 0x30,
1603 	IB_FLOW_SPEC_IPV6	= 0x31,
1604 	/* L4 headers*/
1605 	IB_FLOW_SPEC_TCP	= 0x40,
1606 	IB_FLOW_SPEC_UDP	= 0x41
1607 };
1608 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1609 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1610 
1611 /* Flow steering rule priority is set according to it's domain.
1612  * Lower domain value means higher priority.
1613  */
1614 enum ib_flow_domain {
1615 	IB_FLOW_DOMAIN_USER,
1616 	IB_FLOW_DOMAIN_ETHTOOL,
1617 	IB_FLOW_DOMAIN_RFS,
1618 	IB_FLOW_DOMAIN_NIC,
1619 	IB_FLOW_DOMAIN_NUM /* Must be last */
1620 };
1621 
1622 enum ib_flow_flags {
1623 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1624 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1625 };
1626 
1627 struct ib_flow_eth_filter {
1628 	u8	dst_mac[6];
1629 	u8	src_mac[6];
1630 	__be16	ether_type;
1631 	__be16	vlan_tag;
1632 	/* Must be last */
1633 	u8	real_sz[0];
1634 };
1635 
1636 struct ib_flow_spec_eth {
1637 	enum ib_flow_spec_type	  type;
1638 	u16			  size;
1639 	struct ib_flow_eth_filter val;
1640 	struct ib_flow_eth_filter mask;
1641 };
1642 
1643 struct ib_flow_ib_filter {
1644 	__be16 dlid;
1645 	__u8   sl;
1646 	/* Must be last */
1647 	u8	real_sz[0];
1648 };
1649 
1650 struct ib_flow_spec_ib {
1651 	enum ib_flow_spec_type	 type;
1652 	u16			 size;
1653 	struct ib_flow_ib_filter val;
1654 	struct ib_flow_ib_filter mask;
1655 };
1656 
1657 /* IPv4 header flags */
1658 enum ib_ipv4_flags {
1659 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1660 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1661 				    last have this flag set */
1662 };
1663 
1664 struct ib_flow_ipv4_filter {
1665 	__be32	src_ip;
1666 	__be32	dst_ip;
1667 	u8	proto;
1668 	u8	tos;
1669 	u8	ttl;
1670 	u8	flags;
1671 	/* Must be last */
1672 	u8	real_sz[0];
1673 };
1674 
1675 struct ib_flow_spec_ipv4 {
1676 	enum ib_flow_spec_type	   type;
1677 	u16			   size;
1678 	struct ib_flow_ipv4_filter val;
1679 	struct ib_flow_ipv4_filter mask;
1680 };
1681 
1682 struct ib_flow_ipv6_filter {
1683 	u8	src_ip[16];
1684 	u8	dst_ip[16];
1685 	__be32	flow_label;
1686 	u8	next_hdr;
1687 	u8	traffic_class;
1688 	u8	hop_limit;
1689 	/* Must be last */
1690 	u8	real_sz[0];
1691 };
1692 
1693 struct ib_flow_spec_ipv6 {
1694 	enum ib_flow_spec_type	   type;
1695 	u16			   size;
1696 	struct ib_flow_ipv6_filter val;
1697 	struct ib_flow_ipv6_filter mask;
1698 };
1699 
1700 struct ib_flow_tcp_udp_filter {
1701 	__be16	dst_port;
1702 	__be16	src_port;
1703 	/* Must be last */
1704 	u8	real_sz[0];
1705 };
1706 
1707 struct ib_flow_spec_tcp_udp {
1708 	enum ib_flow_spec_type	      type;
1709 	u16			      size;
1710 	struct ib_flow_tcp_udp_filter val;
1711 	struct ib_flow_tcp_udp_filter mask;
1712 };
1713 
1714 union ib_flow_spec {
1715 	struct {
1716 		enum ib_flow_spec_type	type;
1717 		u16			size;
1718 	};
1719 	struct ib_flow_spec_eth		eth;
1720 	struct ib_flow_spec_ib		ib;
1721 	struct ib_flow_spec_ipv4        ipv4;
1722 	struct ib_flow_spec_tcp_udp	tcp_udp;
1723 	struct ib_flow_spec_ipv6        ipv6;
1724 };
1725 
1726 struct ib_flow_attr {
1727 	enum ib_flow_attr_type type;
1728 	u16	     size;
1729 	u16	     priority;
1730 	u32	     flags;
1731 	u8	     num_of_specs;
1732 	u8	     port;
1733 	/* Following are the optional layers according to user request
1734 	 * struct ib_flow_spec_xxx
1735 	 * struct ib_flow_spec_yyy
1736 	 */
1737 };
1738 
1739 struct ib_flow {
1740 	struct ib_qp		*qp;
1741 	struct ib_uobject	*uobject;
1742 };
1743 
1744 struct ib_mad_hdr;
1745 struct ib_grh;
1746 
1747 enum ib_process_mad_flags {
1748 	IB_MAD_IGNORE_MKEY	= 1,
1749 	IB_MAD_IGNORE_BKEY	= 2,
1750 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1751 };
1752 
1753 enum ib_mad_result {
1754 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1755 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1756 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1757 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1758 };
1759 
1760 #define IB_DEVICE_NAME_MAX 64
1761 
1762 struct ib_cache {
1763 	rwlock_t                lock;
1764 	struct ib_event_handler event_handler;
1765 	struct ib_pkey_cache  **pkey_cache;
1766 	struct ib_gid_table   **gid_cache;
1767 	u8                     *lmc_cache;
1768 };
1769 
1770 struct ib_dma_mapping_ops {
1771 	int		(*mapping_error)(struct ib_device *dev,
1772 					 u64 dma_addr);
1773 	u64		(*map_single)(struct ib_device *dev,
1774 				      void *ptr, size_t size,
1775 				      enum dma_data_direction direction);
1776 	void		(*unmap_single)(struct ib_device *dev,
1777 					u64 addr, size_t size,
1778 					enum dma_data_direction direction);
1779 	u64		(*map_page)(struct ib_device *dev,
1780 				    struct page *page, unsigned long offset,
1781 				    size_t size,
1782 				    enum dma_data_direction direction);
1783 	void		(*unmap_page)(struct ib_device *dev,
1784 				      u64 addr, size_t size,
1785 				      enum dma_data_direction direction);
1786 	int		(*map_sg)(struct ib_device *dev,
1787 				  struct scatterlist *sg, int nents,
1788 				  enum dma_data_direction direction);
1789 	void		(*unmap_sg)(struct ib_device *dev,
1790 				    struct scatterlist *sg, int nents,
1791 				    enum dma_data_direction direction);
1792 	int		(*map_sg_attrs)(struct ib_device *dev,
1793 					struct scatterlist *sg, int nents,
1794 					enum dma_data_direction direction,
1795 					struct dma_attrs *attrs);
1796 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1797 					  struct scatterlist *sg, int nents,
1798 					  enum dma_data_direction direction,
1799 					  struct dma_attrs *attrs);
1800 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1801 					       u64 dma_handle,
1802 					       size_t size,
1803 					       enum dma_data_direction dir);
1804 	void		(*sync_single_for_device)(struct ib_device *dev,
1805 						  u64 dma_handle,
1806 						  size_t size,
1807 						  enum dma_data_direction dir);
1808 	void		*(*alloc_coherent)(struct ib_device *dev,
1809 					   size_t size,
1810 					   u64 *dma_handle,
1811 					   gfp_t flag);
1812 	void		(*free_coherent)(struct ib_device *dev,
1813 					 size_t size, void *cpu_addr,
1814 					 u64 dma_handle);
1815 };
1816 
1817 struct iw_cm_verbs;
1818 
1819 struct ib_port_immutable {
1820 	int                           pkey_tbl_len;
1821 	int                           gid_tbl_len;
1822 	u32                           core_cap_flags;
1823 	u32                           max_mad_size;
1824 };
1825 
1826 struct ib_device {
1827 	struct device                *dma_device;
1828 
1829 	char                          name[IB_DEVICE_NAME_MAX];
1830 
1831 	struct list_head              event_handler_list;
1832 	spinlock_t                    event_handler_lock;
1833 
1834 	spinlock_t                    client_data_lock;
1835 	struct list_head              core_list;
1836 	/* Access to the client_data_list is protected by the client_data_lock
1837 	 * spinlock and the lists_rwsem read-write semaphore */
1838 	struct list_head              client_data_list;
1839 
1840 	struct ib_cache               cache;
1841 	/**
1842 	 * port_immutable is indexed by port number
1843 	 */
1844 	struct ib_port_immutable     *port_immutable;
1845 
1846 	int			      num_comp_vectors;
1847 
1848 	struct iw_cm_verbs	     *iwcm;
1849 
1850 	/**
1851 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1852 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1853 	 *   core when the device is removed.  A lifespan of -1 in the return
1854 	 *   struct tells the core to set a default lifespan.
1855 	 */
1856 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1857 						     u8 port_num);
1858 	/**
1859 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1860 	 * @index - The index in the value array we wish to have updated, or
1861 	 *   num_counters if we want all stats updated
1862 	 * Return codes -
1863 	 *   < 0 - Error, no counters updated
1864 	 *   index - Updated the single counter pointed to by index
1865 	 *   num_counters - Updated all counters (will reset the timestamp
1866 	 *     and prevent further calls for lifespan milliseconds)
1867 	 * Drivers are allowed to update all counters in leiu of just the
1868 	 *   one given in index at their option
1869 	 */
1870 	int		           (*get_hw_stats)(struct ib_device *device,
1871 						   struct rdma_hw_stats *stats,
1872 						   u8 port, int index);
1873 	int		           (*query_device)(struct ib_device *device,
1874 						   struct ib_device_attr *device_attr,
1875 						   struct ib_udata *udata);
1876 	int		           (*query_port)(struct ib_device *device,
1877 						 u8 port_num,
1878 						 struct ib_port_attr *port_attr);
1879 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1880 						     u8 port_num);
1881 	/* When calling get_netdev, the HW vendor's driver should return the
1882 	 * net device of device @device at port @port_num or NULL if such
1883 	 * a net device doesn't exist. The vendor driver should call dev_hold
1884 	 * on this net device. The HW vendor's device driver must guarantee
1885 	 * that this function returns NULL before the net device reaches
1886 	 * NETDEV_UNREGISTER_FINAL state.
1887 	 */
1888 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1889 						 u8 port_num);
1890 	int		           (*query_gid)(struct ib_device *device,
1891 						u8 port_num, int index,
1892 						union ib_gid *gid);
1893 	/* When calling add_gid, the HW vendor's driver should
1894 	 * add the gid of device @device at gid index @index of
1895 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1896 	 * the network device related to this gid is available
1897 	 * at @attr. @context allows the HW vendor driver to store extra
1898 	 * information together with a GID entry. The HW vendor may allocate
1899 	 * memory to contain this information and store it in @context when a
1900 	 * new GID entry is written to. Params are consistent until the next
1901 	 * call of add_gid or delete_gid. The function should return 0 on
1902 	 * success or error otherwise. The function could be called
1903 	 * concurrently for different ports. This function is only called
1904 	 * when roce_gid_table is used.
1905 	 */
1906 	int		           (*add_gid)(struct ib_device *device,
1907 					      u8 port_num,
1908 					      unsigned int index,
1909 					      const union ib_gid *gid,
1910 					      const struct ib_gid_attr *attr,
1911 					      void **context);
1912 	/* When calling del_gid, the HW vendor's driver should delete the
1913 	 * gid of device @device at gid index @index of port @port_num.
1914 	 * Upon the deletion of a GID entry, the HW vendor must free any
1915 	 * allocated memory. The caller will clear @context afterwards.
1916 	 * This function is only called when roce_gid_table is used.
1917 	 */
1918 	int		           (*del_gid)(struct ib_device *device,
1919 					      u8 port_num,
1920 					      unsigned int index,
1921 					      void **context);
1922 	int		           (*query_pkey)(struct ib_device *device,
1923 						 u8 port_num, u16 index, u16 *pkey);
1924 	int		           (*modify_device)(struct ib_device *device,
1925 						    int device_modify_mask,
1926 						    struct ib_device_modify *device_modify);
1927 	int		           (*modify_port)(struct ib_device *device,
1928 						  u8 port_num, int port_modify_mask,
1929 						  struct ib_port_modify *port_modify);
1930 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1931 						     struct ib_udata *udata);
1932 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1933 	int                        (*mmap)(struct ib_ucontext *context,
1934 					   struct vm_area_struct *vma);
1935 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1936 					       struct ib_ucontext *context,
1937 					       struct ib_udata *udata);
1938 	int                        (*dealloc_pd)(struct ib_pd *pd);
1939 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1940 						struct ib_ah_attr *ah_attr,
1941 						struct ib_udata *udata);
1942 	int                        (*modify_ah)(struct ib_ah *ah,
1943 						struct ib_ah_attr *ah_attr);
1944 	int                        (*query_ah)(struct ib_ah *ah,
1945 					       struct ib_ah_attr *ah_attr);
1946 	int                        (*destroy_ah)(struct ib_ah *ah);
1947 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1948 						 struct ib_srq_init_attr *srq_init_attr,
1949 						 struct ib_udata *udata);
1950 	int                        (*modify_srq)(struct ib_srq *srq,
1951 						 struct ib_srq_attr *srq_attr,
1952 						 enum ib_srq_attr_mask srq_attr_mask,
1953 						 struct ib_udata *udata);
1954 	int                        (*query_srq)(struct ib_srq *srq,
1955 						struct ib_srq_attr *srq_attr);
1956 	int                        (*destroy_srq)(struct ib_srq *srq);
1957 	int                        (*post_srq_recv)(struct ib_srq *srq,
1958 						    struct ib_recv_wr *recv_wr,
1959 						    struct ib_recv_wr **bad_recv_wr);
1960 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1961 						struct ib_qp_init_attr *qp_init_attr,
1962 						struct ib_udata *udata);
1963 	int                        (*modify_qp)(struct ib_qp *qp,
1964 						struct ib_qp_attr *qp_attr,
1965 						int qp_attr_mask,
1966 						struct ib_udata *udata);
1967 	int                        (*query_qp)(struct ib_qp *qp,
1968 					       struct ib_qp_attr *qp_attr,
1969 					       int qp_attr_mask,
1970 					       struct ib_qp_init_attr *qp_init_attr);
1971 	int                        (*destroy_qp)(struct ib_qp *qp);
1972 	int                        (*post_send)(struct ib_qp *qp,
1973 						struct ib_send_wr *send_wr,
1974 						struct ib_send_wr **bad_send_wr);
1975 	int                        (*post_recv)(struct ib_qp *qp,
1976 						struct ib_recv_wr *recv_wr,
1977 						struct ib_recv_wr **bad_recv_wr);
1978 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1979 						const struct ib_cq_init_attr *attr,
1980 						struct ib_ucontext *context,
1981 						struct ib_udata *udata);
1982 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1983 						u16 cq_period);
1984 	int                        (*destroy_cq)(struct ib_cq *cq);
1985 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1986 						struct ib_udata *udata);
1987 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1988 					      struct ib_wc *wc);
1989 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1990 	int                        (*req_notify_cq)(struct ib_cq *cq,
1991 						    enum ib_cq_notify_flags flags);
1992 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1993 						      int wc_cnt);
1994 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1995 						 int mr_access_flags);
1996 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1997 						  u64 start, u64 length,
1998 						  u64 virt_addr,
1999 						  int mr_access_flags,
2000 						  struct ib_udata *udata);
2001 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2002 						    int flags,
2003 						    u64 start, u64 length,
2004 						    u64 virt_addr,
2005 						    int mr_access_flags,
2006 						    struct ib_pd *pd,
2007 						    struct ib_udata *udata);
2008 	int                        (*dereg_mr)(struct ib_mr *mr);
2009 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2010 					       enum ib_mr_type mr_type,
2011 					       u32 max_num_sg);
2012 	int                        (*map_mr_sg)(struct ib_mr *mr,
2013 						struct scatterlist *sg,
2014 						int sg_nents,
2015 						unsigned int *sg_offset);
2016 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2017 					       enum ib_mw_type type,
2018 					       struct ib_udata *udata);
2019 	int                        (*dealloc_mw)(struct ib_mw *mw);
2020 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2021 						int mr_access_flags,
2022 						struct ib_fmr_attr *fmr_attr);
2023 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2024 						   u64 *page_list, int list_len,
2025 						   u64 iova);
2026 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2027 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2028 	int                        (*attach_mcast)(struct ib_qp *qp,
2029 						   union ib_gid *gid,
2030 						   u16 lid);
2031 	int                        (*detach_mcast)(struct ib_qp *qp,
2032 						   union ib_gid *gid,
2033 						   u16 lid);
2034 	int                        (*process_mad)(struct ib_device *device,
2035 						  int process_mad_flags,
2036 						  u8 port_num,
2037 						  const struct ib_wc *in_wc,
2038 						  const struct ib_grh *in_grh,
2039 						  const struct ib_mad_hdr *in_mad,
2040 						  size_t in_mad_size,
2041 						  struct ib_mad_hdr *out_mad,
2042 						  size_t *out_mad_size,
2043 						  u16 *out_mad_pkey_index);
2044 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2045 						 struct ib_ucontext *ucontext,
2046 						 struct ib_udata *udata);
2047 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2048 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2049 						  struct ib_flow_attr
2050 						  *flow_attr,
2051 						  int domain);
2052 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2053 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2054 						      struct ib_mr_status *mr_status);
2055 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2056 	void			   (*drain_rq)(struct ib_qp *qp);
2057 	void			   (*drain_sq)(struct ib_qp *qp);
2058 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2059 							int state);
2060 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2061 						   struct ifla_vf_info *ivf);
2062 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2063 						   struct ifla_vf_stats *stats);
2064 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2065 						  int type);
2066 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2067 						struct ib_wq_init_attr *init_attr,
2068 						struct ib_udata *udata);
2069 	int			   (*destroy_wq)(struct ib_wq *wq);
2070 	int			   (*modify_wq)(struct ib_wq *wq,
2071 						struct ib_wq_attr *attr,
2072 						u32 wq_attr_mask,
2073 						struct ib_udata *udata);
2074 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2075 							   struct ib_rwq_ind_table_init_attr *init_attr,
2076 							   struct ib_udata *udata);
2077 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2078 	struct ib_dma_mapping_ops   *dma_ops;
2079 
2080 	struct module               *owner;
2081 	struct device                dev;
2082 	struct kobject               *ports_parent;
2083 	struct list_head             port_list;
2084 
2085 	enum {
2086 		IB_DEV_UNINITIALIZED,
2087 		IB_DEV_REGISTERED,
2088 		IB_DEV_UNREGISTERED
2089 	}                            reg_state;
2090 
2091 	int			     uverbs_abi_ver;
2092 	u64			     uverbs_cmd_mask;
2093 	u64			     uverbs_ex_cmd_mask;
2094 
2095 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2096 	__be64			     node_guid;
2097 	u32			     local_dma_lkey;
2098 	u16                          is_switch:1;
2099 	u8                           node_type;
2100 	u8                           phys_port_cnt;
2101 	struct ib_device_attr        attrs;
2102 	struct attribute_group	     *hw_stats_ag;
2103 	struct rdma_hw_stats         *hw_stats;
2104 
2105 	/**
2106 	 * The following mandatory functions are used only at device
2107 	 * registration.  Keep functions such as these at the end of this
2108 	 * structure to avoid cache line misses when accessing struct ib_device
2109 	 * in fast paths.
2110 	 */
2111 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2112 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2113 };
2114 
2115 struct ib_client {
2116 	char  *name;
2117 	void (*add)   (struct ib_device *);
2118 	void (*remove)(struct ib_device *, void *client_data);
2119 
2120 	/* Returns the net_dev belonging to this ib_client and matching the
2121 	 * given parameters.
2122 	 * @dev:	 An RDMA device that the net_dev use for communication.
2123 	 * @port:	 A physical port number on the RDMA device.
2124 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2125 	 * @gid:	 A GID that the net_dev uses to communicate.
2126 	 * @addr:	 An IP address the net_dev is configured with.
2127 	 * @client_data: The device's client data set by ib_set_client_data().
2128 	 *
2129 	 * An ib_client that implements a net_dev on top of RDMA devices
2130 	 * (such as IP over IB) should implement this callback, allowing the
2131 	 * rdma_cm module to find the right net_dev for a given request.
2132 	 *
2133 	 * The caller is responsible for calling dev_put on the returned
2134 	 * netdev. */
2135 	struct net_device *(*get_net_dev_by_params)(
2136 			struct ib_device *dev,
2137 			u8 port,
2138 			u16 pkey,
2139 			const union ib_gid *gid,
2140 			const struct sockaddr *addr,
2141 			void *client_data);
2142 	struct list_head list;
2143 };
2144 
2145 struct ib_device *ib_alloc_device(size_t size);
2146 void ib_dealloc_device(struct ib_device *device);
2147 
2148 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2149 
2150 int ib_register_device(struct ib_device *device,
2151 		       int (*port_callback)(struct ib_device *,
2152 					    u8, struct kobject *));
2153 void ib_unregister_device(struct ib_device *device);
2154 
2155 int ib_register_client   (struct ib_client *client);
2156 void ib_unregister_client(struct ib_client *client);
2157 
2158 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2159 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2160 			 void *data);
2161 
2162 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2163 {
2164 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2165 }
2166 
2167 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2168 {
2169 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2170 }
2171 
2172 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2173 				       size_t offset,
2174 				       size_t len)
2175 {
2176 	const void __user *p = (const char __user *)udata->inbuf + offset;
2177 	bool ret;
2178 	u8 *buf;
2179 
2180 	if (len > USHRT_MAX)
2181 		return false;
2182 
2183 	buf = memdup_user(p, len);
2184 	if (IS_ERR(buf))
2185 		return false;
2186 
2187 	ret = !memchr_inv(buf, 0, len);
2188 	kfree(buf);
2189 	return ret;
2190 }
2191 
2192 /**
2193  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2194  * contains all required attributes and no attributes not allowed for
2195  * the given QP state transition.
2196  * @cur_state: Current QP state
2197  * @next_state: Next QP state
2198  * @type: QP type
2199  * @mask: Mask of supplied QP attributes
2200  * @ll : link layer of port
2201  *
2202  * This function is a helper function that a low-level driver's
2203  * modify_qp method can use to validate the consumer's input.  It
2204  * checks that cur_state and next_state are valid QP states, that a
2205  * transition from cur_state to next_state is allowed by the IB spec,
2206  * and that the attribute mask supplied is allowed for the transition.
2207  */
2208 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2209 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2210 		       enum rdma_link_layer ll);
2211 
2212 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2213 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2214 void ib_dispatch_event(struct ib_event *event);
2215 
2216 int ib_query_port(struct ib_device *device,
2217 		  u8 port_num, struct ib_port_attr *port_attr);
2218 
2219 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2220 					       u8 port_num);
2221 
2222 /**
2223  * rdma_cap_ib_switch - Check if the device is IB switch
2224  * @device: Device to check
2225  *
2226  * Device driver is responsible for setting is_switch bit on
2227  * in ib_device structure at init time.
2228  *
2229  * Return: true if the device is IB switch.
2230  */
2231 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2232 {
2233 	return device->is_switch;
2234 }
2235 
2236 /**
2237  * rdma_start_port - Return the first valid port number for the device
2238  * specified
2239  *
2240  * @device: Device to be checked
2241  *
2242  * Return start port number
2243  */
2244 static inline u8 rdma_start_port(const struct ib_device *device)
2245 {
2246 	return rdma_cap_ib_switch(device) ? 0 : 1;
2247 }
2248 
2249 /**
2250  * rdma_end_port - Return the last valid port number for the device
2251  * specified
2252  *
2253  * @device: Device to be checked
2254  *
2255  * Return last port number
2256  */
2257 static inline u8 rdma_end_port(const struct ib_device *device)
2258 {
2259 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2260 }
2261 
2262 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2263 {
2264 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2265 }
2266 
2267 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2268 {
2269 	return device->port_immutable[port_num].core_cap_flags &
2270 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2271 }
2272 
2273 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2274 {
2275 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2276 }
2277 
2278 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2279 {
2280 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2281 }
2282 
2283 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2284 {
2285 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2286 }
2287 
2288 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2289 {
2290 	return rdma_protocol_ib(device, port_num) ||
2291 		rdma_protocol_roce(device, port_num);
2292 }
2293 
2294 /**
2295  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2296  * Management Datagrams.
2297  * @device: Device to check
2298  * @port_num: Port number to check
2299  *
2300  * Management Datagrams (MAD) are a required part of the InfiniBand
2301  * specification and are supported on all InfiniBand devices.  A slightly
2302  * extended version are also supported on OPA interfaces.
2303  *
2304  * Return: true if the port supports sending/receiving of MAD packets.
2305  */
2306 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2307 {
2308 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2309 }
2310 
2311 /**
2312  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2313  * Management Datagrams.
2314  * @device: Device to check
2315  * @port_num: Port number to check
2316  *
2317  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2318  * datagrams with their own versions.  These OPA MADs share many but not all of
2319  * the characteristics of InfiniBand MADs.
2320  *
2321  * OPA MADs differ in the following ways:
2322  *
2323  *    1) MADs are variable size up to 2K
2324  *       IBTA defined MADs remain fixed at 256 bytes
2325  *    2) OPA SMPs must carry valid PKeys
2326  *    3) OPA SMP packets are a different format
2327  *
2328  * Return: true if the port supports OPA MAD packet formats.
2329  */
2330 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2331 {
2332 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2333 		== RDMA_CORE_CAP_OPA_MAD;
2334 }
2335 
2336 /**
2337  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2338  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2339  * @device: Device to check
2340  * @port_num: Port number to check
2341  *
2342  * Each InfiniBand node is required to provide a Subnet Management Agent
2343  * that the subnet manager can access.  Prior to the fabric being fully
2344  * configured by the subnet manager, the SMA is accessed via a well known
2345  * interface called the Subnet Management Interface (SMI).  This interface
2346  * uses directed route packets to communicate with the SM to get around the
2347  * chicken and egg problem of the SM needing to know what's on the fabric
2348  * in order to configure the fabric, and needing to configure the fabric in
2349  * order to send packets to the devices on the fabric.  These directed
2350  * route packets do not need the fabric fully configured in order to reach
2351  * their destination.  The SMI is the only method allowed to send
2352  * directed route packets on an InfiniBand fabric.
2353  *
2354  * Return: true if the port provides an SMI.
2355  */
2356 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2357 {
2358 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2359 }
2360 
2361 /**
2362  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2363  * Communication Manager.
2364  * @device: Device to check
2365  * @port_num: Port number to check
2366  *
2367  * The InfiniBand Communication Manager is one of many pre-defined General
2368  * Service Agents (GSA) that are accessed via the General Service
2369  * Interface (GSI).  It's role is to facilitate establishment of connections
2370  * between nodes as well as other management related tasks for established
2371  * connections.
2372  *
2373  * Return: true if the port supports an IB CM (this does not guarantee that
2374  * a CM is actually running however).
2375  */
2376 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2377 {
2378 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2379 }
2380 
2381 /**
2382  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2383  * Communication Manager.
2384  * @device: Device to check
2385  * @port_num: Port number to check
2386  *
2387  * Similar to above, but specific to iWARP connections which have a different
2388  * managment protocol than InfiniBand.
2389  *
2390  * Return: true if the port supports an iWARP CM (this does not guarantee that
2391  * a CM is actually running however).
2392  */
2393 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2394 {
2395 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2396 }
2397 
2398 /**
2399  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2400  * Subnet Administration.
2401  * @device: Device to check
2402  * @port_num: Port number to check
2403  *
2404  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2405  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2406  * fabrics, devices should resolve routes to other hosts by contacting the
2407  * SA to query the proper route.
2408  *
2409  * Return: true if the port should act as a client to the fabric Subnet
2410  * Administration interface.  This does not imply that the SA service is
2411  * running locally.
2412  */
2413 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2414 {
2415 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2416 }
2417 
2418 /**
2419  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2420  * Multicast.
2421  * @device: Device to check
2422  * @port_num: Port number to check
2423  *
2424  * InfiniBand multicast registration is more complex than normal IPv4 or
2425  * IPv6 multicast registration.  Each Host Channel Adapter must register
2426  * with the Subnet Manager when it wishes to join a multicast group.  It
2427  * should do so only once regardless of how many queue pairs it subscribes
2428  * to this group.  And it should leave the group only after all queue pairs
2429  * attached to the group have been detached.
2430  *
2431  * Return: true if the port must undertake the additional adminstrative
2432  * overhead of registering/unregistering with the SM and tracking of the
2433  * total number of queue pairs attached to the multicast group.
2434  */
2435 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2436 {
2437 	return rdma_cap_ib_sa(device, port_num);
2438 }
2439 
2440 /**
2441  * rdma_cap_af_ib - Check if the port of device has the capability
2442  * Native Infiniband Address.
2443  * @device: Device to check
2444  * @port_num: Port number to check
2445  *
2446  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2447  * GID.  RoCE uses a different mechanism, but still generates a GID via
2448  * a prescribed mechanism and port specific data.
2449  *
2450  * Return: true if the port uses a GID address to identify devices on the
2451  * network.
2452  */
2453 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2454 {
2455 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2456 }
2457 
2458 /**
2459  * rdma_cap_eth_ah - Check if the port of device has the capability
2460  * Ethernet Address Handle.
2461  * @device: Device to check
2462  * @port_num: Port number to check
2463  *
2464  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2465  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2466  * port.  Normally, packet headers are generated by the sending host
2467  * adapter, but when sending connectionless datagrams, we must manually
2468  * inject the proper headers for the fabric we are communicating over.
2469  *
2470  * Return: true if we are running as a RoCE port and must force the
2471  * addition of a Global Route Header built from our Ethernet Address
2472  * Handle into our header list for connectionless packets.
2473  */
2474 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2475 {
2476 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2477 }
2478 
2479 /**
2480  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2481  *
2482  * @device: Device
2483  * @port_num: Port number
2484  *
2485  * This MAD size includes the MAD headers and MAD payload.  No other headers
2486  * are included.
2487  *
2488  * Return the max MAD size required by the Port.  Will return 0 if the port
2489  * does not support MADs
2490  */
2491 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2492 {
2493 	return device->port_immutable[port_num].max_mad_size;
2494 }
2495 
2496 /**
2497  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2498  * @device: Device to check
2499  * @port_num: Port number to check
2500  *
2501  * RoCE GID table mechanism manages the various GIDs for a device.
2502  *
2503  * NOTE: if allocating the port's GID table has failed, this call will still
2504  * return true, but any RoCE GID table API will fail.
2505  *
2506  * Return: true if the port uses RoCE GID table mechanism in order to manage
2507  * its GIDs.
2508  */
2509 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2510 					   u8 port_num)
2511 {
2512 	return rdma_protocol_roce(device, port_num) &&
2513 		device->add_gid && device->del_gid;
2514 }
2515 
2516 /*
2517  * Check if the device supports READ W/ INVALIDATE.
2518  */
2519 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2520 {
2521 	/*
2522 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2523 	 * has support for it yet.
2524 	 */
2525 	return rdma_protocol_iwarp(dev, port_num);
2526 }
2527 
2528 int ib_query_gid(struct ib_device *device,
2529 		 u8 port_num, int index, union ib_gid *gid,
2530 		 struct ib_gid_attr *attr);
2531 
2532 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2533 			 int state);
2534 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2535 		     struct ifla_vf_info *info);
2536 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2537 		    struct ifla_vf_stats *stats);
2538 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2539 		   int type);
2540 
2541 int ib_query_pkey(struct ib_device *device,
2542 		  u8 port_num, u16 index, u16 *pkey);
2543 
2544 int ib_modify_device(struct ib_device *device,
2545 		     int device_modify_mask,
2546 		     struct ib_device_modify *device_modify);
2547 
2548 int ib_modify_port(struct ib_device *device,
2549 		   u8 port_num, int port_modify_mask,
2550 		   struct ib_port_modify *port_modify);
2551 
2552 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2553 		enum ib_gid_type gid_type, struct net_device *ndev,
2554 		u8 *port_num, u16 *index);
2555 
2556 int ib_find_pkey(struct ib_device *device,
2557 		 u8 port_num, u16 pkey, u16 *index);
2558 
2559 enum ib_pd_flags {
2560 	/*
2561 	 * Create a memory registration for all memory in the system and place
2562 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2563 	 * ULPs to avoid the overhead of dynamic MRs.
2564 	 *
2565 	 * This flag is generally considered unsafe and must only be used in
2566 	 * extremly trusted environments.  Every use of it will log a warning
2567 	 * in the kernel log.
2568 	 */
2569 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2570 };
2571 
2572 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2573 		const char *caller);
2574 #define ib_alloc_pd(device, flags) \
2575 	__ib_alloc_pd((device), (flags), __func__)
2576 void ib_dealloc_pd(struct ib_pd *pd);
2577 
2578 /**
2579  * ib_create_ah - Creates an address handle for the given address vector.
2580  * @pd: The protection domain associated with the address handle.
2581  * @ah_attr: The attributes of the address vector.
2582  *
2583  * The address handle is used to reference a local or global destination
2584  * in all UD QP post sends.
2585  */
2586 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2587 
2588 /**
2589  * ib_init_ah_from_wc - Initializes address handle attributes from a
2590  *   work completion.
2591  * @device: Device on which the received message arrived.
2592  * @port_num: Port on which the received message arrived.
2593  * @wc: Work completion associated with the received message.
2594  * @grh: References the received global route header.  This parameter is
2595  *   ignored unless the work completion indicates that the GRH is valid.
2596  * @ah_attr: Returned attributes that can be used when creating an address
2597  *   handle for replying to the message.
2598  */
2599 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2600 		       const struct ib_wc *wc, const struct ib_grh *grh,
2601 		       struct ib_ah_attr *ah_attr);
2602 
2603 /**
2604  * ib_create_ah_from_wc - Creates an address handle associated with the
2605  *   sender of the specified work completion.
2606  * @pd: The protection domain associated with the address handle.
2607  * @wc: Work completion information associated with a received message.
2608  * @grh: References the received global route header.  This parameter is
2609  *   ignored unless the work completion indicates that the GRH is valid.
2610  * @port_num: The outbound port number to associate with the address.
2611  *
2612  * The address handle is used to reference a local or global destination
2613  * in all UD QP post sends.
2614  */
2615 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2616 				   const struct ib_grh *grh, u8 port_num);
2617 
2618 /**
2619  * ib_modify_ah - Modifies the address vector associated with an address
2620  *   handle.
2621  * @ah: The address handle to modify.
2622  * @ah_attr: The new address vector attributes to associate with the
2623  *   address handle.
2624  */
2625 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2626 
2627 /**
2628  * ib_query_ah - Queries the address vector associated with an address
2629  *   handle.
2630  * @ah: The address handle to query.
2631  * @ah_attr: The address vector attributes associated with the address
2632  *   handle.
2633  */
2634 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2635 
2636 /**
2637  * ib_destroy_ah - Destroys an address handle.
2638  * @ah: The address handle to destroy.
2639  */
2640 int ib_destroy_ah(struct ib_ah *ah);
2641 
2642 /**
2643  * ib_create_srq - Creates a SRQ associated with the specified protection
2644  *   domain.
2645  * @pd: The protection domain associated with the SRQ.
2646  * @srq_init_attr: A list of initial attributes required to create the
2647  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2648  *   the actual capabilities of the created SRQ.
2649  *
2650  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2651  * requested size of the SRQ, and set to the actual values allocated
2652  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2653  * will always be at least as large as the requested values.
2654  */
2655 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2656 			     struct ib_srq_init_attr *srq_init_attr);
2657 
2658 /**
2659  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2660  * @srq: The SRQ to modify.
2661  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2662  *   the current values of selected SRQ attributes are returned.
2663  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2664  *   are being modified.
2665  *
2666  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2667  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2668  * the number of receives queued drops below the limit.
2669  */
2670 int ib_modify_srq(struct ib_srq *srq,
2671 		  struct ib_srq_attr *srq_attr,
2672 		  enum ib_srq_attr_mask srq_attr_mask);
2673 
2674 /**
2675  * ib_query_srq - Returns the attribute list and current values for the
2676  *   specified SRQ.
2677  * @srq: The SRQ to query.
2678  * @srq_attr: The attributes of the specified SRQ.
2679  */
2680 int ib_query_srq(struct ib_srq *srq,
2681 		 struct ib_srq_attr *srq_attr);
2682 
2683 /**
2684  * ib_destroy_srq - Destroys the specified SRQ.
2685  * @srq: The SRQ to destroy.
2686  */
2687 int ib_destroy_srq(struct ib_srq *srq);
2688 
2689 /**
2690  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2691  * @srq: The SRQ to post the work request on.
2692  * @recv_wr: A list of work requests to post on the receive queue.
2693  * @bad_recv_wr: On an immediate failure, this parameter will reference
2694  *   the work request that failed to be posted on the QP.
2695  */
2696 static inline int ib_post_srq_recv(struct ib_srq *srq,
2697 				   struct ib_recv_wr *recv_wr,
2698 				   struct ib_recv_wr **bad_recv_wr)
2699 {
2700 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2701 }
2702 
2703 /**
2704  * ib_create_qp - Creates a QP associated with the specified protection
2705  *   domain.
2706  * @pd: The protection domain associated with the QP.
2707  * @qp_init_attr: A list of initial attributes required to create the
2708  *   QP.  If QP creation succeeds, then the attributes are updated to
2709  *   the actual capabilities of the created QP.
2710  */
2711 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2712 			   struct ib_qp_init_attr *qp_init_attr);
2713 
2714 /**
2715  * ib_modify_qp - Modifies the attributes for the specified QP and then
2716  *   transitions the QP to the given state.
2717  * @qp: The QP to modify.
2718  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2719  *   the current values of selected QP attributes are returned.
2720  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2721  *   are being modified.
2722  */
2723 int ib_modify_qp(struct ib_qp *qp,
2724 		 struct ib_qp_attr *qp_attr,
2725 		 int qp_attr_mask);
2726 
2727 /**
2728  * ib_query_qp - Returns the attribute list and current values for the
2729  *   specified QP.
2730  * @qp: The QP to query.
2731  * @qp_attr: The attributes of the specified QP.
2732  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2733  * @qp_init_attr: Additional attributes of the selected QP.
2734  *
2735  * The qp_attr_mask may be used to limit the query to gathering only the
2736  * selected attributes.
2737  */
2738 int ib_query_qp(struct ib_qp *qp,
2739 		struct ib_qp_attr *qp_attr,
2740 		int qp_attr_mask,
2741 		struct ib_qp_init_attr *qp_init_attr);
2742 
2743 /**
2744  * ib_destroy_qp - Destroys the specified QP.
2745  * @qp: The QP to destroy.
2746  */
2747 int ib_destroy_qp(struct ib_qp *qp);
2748 
2749 /**
2750  * ib_open_qp - Obtain a reference to an existing sharable QP.
2751  * @xrcd - XRC domain
2752  * @qp_open_attr: Attributes identifying the QP to open.
2753  *
2754  * Returns a reference to a sharable QP.
2755  */
2756 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2757 			 struct ib_qp_open_attr *qp_open_attr);
2758 
2759 /**
2760  * ib_close_qp - Release an external reference to a QP.
2761  * @qp: The QP handle to release
2762  *
2763  * The opened QP handle is released by the caller.  The underlying
2764  * shared QP is not destroyed until all internal references are released.
2765  */
2766 int ib_close_qp(struct ib_qp *qp);
2767 
2768 /**
2769  * ib_post_send - Posts a list of work requests to the send queue of
2770  *   the specified QP.
2771  * @qp: The QP to post the work request on.
2772  * @send_wr: A list of work requests to post on the send queue.
2773  * @bad_send_wr: On an immediate failure, this parameter will reference
2774  *   the work request that failed to be posted on the QP.
2775  *
2776  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2777  * error is returned, the QP state shall not be affected,
2778  * ib_post_send() will return an immediate error after queueing any
2779  * earlier work requests in the list.
2780  */
2781 static inline int ib_post_send(struct ib_qp *qp,
2782 			       struct ib_send_wr *send_wr,
2783 			       struct ib_send_wr **bad_send_wr)
2784 {
2785 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2786 }
2787 
2788 /**
2789  * ib_post_recv - Posts a list of work requests to the receive queue of
2790  *   the specified QP.
2791  * @qp: The QP to post the work request on.
2792  * @recv_wr: A list of work requests to post on the receive queue.
2793  * @bad_recv_wr: On an immediate failure, this parameter will reference
2794  *   the work request that failed to be posted on the QP.
2795  */
2796 static inline int ib_post_recv(struct ib_qp *qp,
2797 			       struct ib_recv_wr *recv_wr,
2798 			       struct ib_recv_wr **bad_recv_wr)
2799 {
2800 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2801 }
2802 
2803 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2804 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2805 void ib_free_cq(struct ib_cq *cq);
2806 
2807 /**
2808  * ib_create_cq - Creates a CQ on the specified device.
2809  * @device: The device on which to create the CQ.
2810  * @comp_handler: A user-specified callback that is invoked when a
2811  *   completion event occurs on the CQ.
2812  * @event_handler: A user-specified callback that is invoked when an
2813  *   asynchronous event not associated with a completion occurs on the CQ.
2814  * @cq_context: Context associated with the CQ returned to the user via
2815  *   the associated completion and event handlers.
2816  * @cq_attr: The attributes the CQ should be created upon.
2817  *
2818  * Users can examine the cq structure to determine the actual CQ size.
2819  */
2820 struct ib_cq *ib_create_cq(struct ib_device *device,
2821 			   ib_comp_handler comp_handler,
2822 			   void (*event_handler)(struct ib_event *, void *),
2823 			   void *cq_context,
2824 			   const struct ib_cq_init_attr *cq_attr);
2825 
2826 /**
2827  * ib_resize_cq - Modifies the capacity of the CQ.
2828  * @cq: The CQ to resize.
2829  * @cqe: The minimum size of the CQ.
2830  *
2831  * Users can examine the cq structure to determine the actual CQ size.
2832  */
2833 int ib_resize_cq(struct ib_cq *cq, int cqe);
2834 
2835 /**
2836  * ib_modify_cq - Modifies moderation params of the CQ
2837  * @cq: The CQ to modify.
2838  * @cq_count: number of CQEs that will trigger an event
2839  * @cq_period: max period of time in usec before triggering an event
2840  *
2841  */
2842 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2843 
2844 /**
2845  * ib_destroy_cq - Destroys the specified CQ.
2846  * @cq: The CQ to destroy.
2847  */
2848 int ib_destroy_cq(struct ib_cq *cq);
2849 
2850 /**
2851  * ib_poll_cq - poll a CQ for completion(s)
2852  * @cq:the CQ being polled
2853  * @num_entries:maximum number of completions to return
2854  * @wc:array of at least @num_entries &struct ib_wc where completions
2855  *   will be returned
2856  *
2857  * Poll a CQ for (possibly multiple) completions.  If the return value
2858  * is < 0, an error occurred.  If the return value is >= 0, it is the
2859  * number of completions returned.  If the return value is
2860  * non-negative and < num_entries, then the CQ was emptied.
2861  */
2862 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2863 			     struct ib_wc *wc)
2864 {
2865 	return cq->device->poll_cq(cq, num_entries, wc);
2866 }
2867 
2868 /**
2869  * ib_peek_cq - Returns the number of unreaped completions currently
2870  *   on the specified CQ.
2871  * @cq: The CQ to peek.
2872  * @wc_cnt: A minimum number of unreaped completions to check for.
2873  *
2874  * If the number of unreaped completions is greater than or equal to wc_cnt,
2875  * this function returns wc_cnt, otherwise, it returns the actual number of
2876  * unreaped completions.
2877  */
2878 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2879 
2880 /**
2881  * ib_req_notify_cq - Request completion notification on a CQ.
2882  * @cq: The CQ to generate an event for.
2883  * @flags:
2884  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2885  *   to request an event on the next solicited event or next work
2886  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2887  *   may also be |ed in to request a hint about missed events, as
2888  *   described below.
2889  *
2890  * Return Value:
2891  *    < 0 means an error occurred while requesting notification
2892  *   == 0 means notification was requested successfully, and if
2893  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2894  *        were missed and it is safe to wait for another event.  In
2895  *        this case is it guaranteed that any work completions added
2896  *        to the CQ since the last CQ poll will trigger a completion
2897  *        notification event.
2898  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2899  *        in.  It means that the consumer must poll the CQ again to
2900  *        make sure it is empty to avoid missing an event because of a
2901  *        race between requesting notification and an entry being
2902  *        added to the CQ.  This return value means it is possible
2903  *        (but not guaranteed) that a work completion has been added
2904  *        to the CQ since the last poll without triggering a
2905  *        completion notification event.
2906  */
2907 static inline int ib_req_notify_cq(struct ib_cq *cq,
2908 				   enum ib_cq_notify_flags flags)
2909 {
2910 	return cq->device->req_notify_cq(cq, flags);
2911 }
2912 
2913 /**
2914  * ib_req_ncomp_notif - Request completion notification when there are
2915  *   at least the specified number of unreaped completions on the CQ.
2916  * @cq: The CQ to generate an event for.
2917  * @wc_cnt: The number of unreaped completions that should be on the
2918  *   CQ before an event is generated.
2919  */
2920 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2921 {
2922 	return cq->device->req_ncomp_notif ?
2923 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2924 		-ENOSYS;
2925 }
2926 
2927 /**
2928  * ib_dma_mapping_error - check a DMA addr for error
2929  * @dev: The device for which the dma_addr was created
2930  * @dma_addr: The DMA address to check
2931  */
2932 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2933 {
2934 	if (dev->dma_ops)
2935 		return dev->dma_ops->mapping_error(dev, dma_addr);
2936 	return dma_mapping_error(dev->dma_device, dma_addr);
2937 }
2938 
2939 /**
2940  * ib_dma_map_single - Map a kernel virtual address to DMA address
2941  * @dev: The device for which the dma_addr is to be created
2942  * @cpu_addr: The kernel virtual address
2943  * @size: The size of the region in bytes
2944  * @direction: The direction of the DMA
2945  */
2946 static inline u64 ib_dma_map_single(struct ib_device *dev,
2947 				    void *cpu_addr, size_t size,
2948 				    enum dma_data_direction direction)
2949 {
2950 	if (dev->dma_ops)
2951 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2952 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2953 }
2954 
2955 /**
2956  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2957  * @dev: The device for which the DMA address was created
2958  * @addr: The DMA address
2959  * @size: The size of the region in bytes
2960  * @direction: The direction of the DMA
2961  */
2962 static inline void ib_dma_unmap_single(struct ib_device *dev,
2963 				       u64 addr, size_t size,
2964 				       enum dma_data_direction direction)
2965 {
2966 	if (dev->dma_ops)
2967 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2968 	else
2969 		dma_unmap_single(dev->dma_device, addr, size, direction);
2970 }
2971 
2972 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2973 					  void *cpu_addr, size_t size,
2974 					  enum dma_data_direction direction,
2975 					  struct dma_attrs *dma_attrs)
2976 {
2977 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2978 				    direction, dma_attrs);
2979 }
2980 
2981 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2982 					     u64 addr, size_t size,
2983 					     enum dma_data_direction direction,
2984 					     struct dma_attrs *dma_attrs)
2985 {
2986 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2987 				      direction, dma_attrs);
2988 }
2989 
2990 /**
2991  * ib_dma_map_page - Map a physical page to DMA address
2992  * @dev: The device for which the dma_addr is to be created
2993  * @page: The page to be mapped
2994  * @offset: The offset within the page
2995  * @size: The size of the region in bytes
2996  * @direction: The direction of the DMA
2997  */
2998 static inline u64 ib_dma_map_page(struct ib_device *dev,
2999 				  struct page *page,
3000 				  unsigned long offset,
3001 				  size_t size,
3002 					 enum dma_data_direction direction)
3003 {
3004 	if (dev->dma_ops)
3005 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3006 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3007 }
3008 
3009 /**
3010  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3011  * @dev: The device for which the DMA address was created
3012  * @addr: The DMA address
3013  * @size: The size of the region in bytes
3014  * @direction: The direction of the DMA
3015  */
3016 static inline void ib_dma_unmap_page(struct ib_device *dev,
3017 				     u64 addr, size_t size,
3018 				     enum dma_data_direction direction)
3019 {
3020 	if (dev->dma_ops)
3021 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3022 	else
3023 		dma_unmap_page(dev->dma_device, addr, size, direction);
3024 }
3025 
3026 /**
3027  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3028  * @dev: The device for which the DMA addresses are to be created
3029  * @sg: The array of scatter/gather entries
3030  * @nents: The number of scatter/gather entries
3031  * @direction: The direction of the DMA
3032  */
3033 static inline int ib_dma_map_sg(struct ib_device *dev,
3034 				struct scatterlist *sg, int nents,
3035 				enum dma_data_direction direction)
3036 {
3037 	if (dev->dma_ops)
3038 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3039 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3040 }
3041 
3042 /**
3043  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3044  * @dev: The device for which the DMA addresses were created
3045  * @sg: The array of scatter/gather entries
3046  * @nents: The number of scatter/gather entries
3047  * @direction: The direction of the DMA
3048  */
3049 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3050 				   struct scatterlist *sg, int nents,
3051 				   enum dma_data_direction direction)
3052 {
3053 	if (dev->dma_ops)
3054 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3055 	else
3056 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3057 }
3058 
3059 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3060 				      struct scatterlist *sg, int nents,
3061 				      enum dma_data_direction direction,
3062 				      struct dma_attrs *dma_attrs)
3063 {
3064 	if (dev->dma_ops)
3065 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3066 						  dma_attrs);
3067 	else
3068 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3069 					dma_attrs);
3070 }
3071 
3072 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3073 					 struct scatterlist *sg, int nents,
3074 					 enum dma_data_direction direction,
3075 					 struct dma_attrs *dma_attrs)
3076 {
3077 	if (dev->dma_ops)
3078 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3079 						  dma_attrs);
3080 	else
3081 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3082 				   dma_attrs);
3083 }
3084 /**
3085  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3086  * @dev: The device for which the DMA addresses were created
3087  * @sg: The scatter/gather entry
3088  *
3089  * Note: this function is obsolete. To do: change all occurrences of
3090  * ib_sg_dma_address() into sg_dma_address().
3091  */
3092 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3093 				    struct scatterlist *sg)
3094 {
3095 	return sg_dma_address(sg);
3096 }
3097 
3098 /**
3099  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3100  * @dev: The device for which the DMA addresses were created
3101  * @sg: The scatter/gather entry
3102  *
3103  * Note: this function is obsolete. To do: change all occurrences of
3104  * ib_sg_dma_len() into sg_dma_len().
3105  */
3106 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3107 					 struct scatterlist *sg)
3108 {
3109 	return sg_dma_len(sg);
3110 }
3111 
3112 /**
3113  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3114  * @dev: The device for which the DMA address was created
3115  * @addr: The DMA address
3116  * @size: The size of the region in bytes
3117  * @dir: The direction of the DMA
3118  */
3119 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3120 					      u64 addr,
3121 					      size_t size,
3122 					      enum dma_data_direction dir)
3123 {
3124 	if (dev->dma_ops)
3125 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3126 	else
3127 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3128 }
3129 
3130 /**
3131  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3132  * @dev: The device for which the DMA address was created
3133  * @addr: The DMA address
3134  * @size: The size of the region in bytes
3135  * @dir: The direction of the DMA
3136  */
3137 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3138 						 u64 addr,
3139 						 size_t size,
3140 						 enum dma_data_direction dir)
3141 {
3142 	if (dev->dma_ops)
3143 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3144 	else
3145 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3146 }
3147 
3148 /**
3149  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3150  * @dev: The device for which the DMA address is requested
3151  * @size: The size of the region to allocate in bytes
3152  * @dma_handle: A pointer for returning the DMA address of the region
3153  * @flag: memory allocator flags
3154  */
3155 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3156 					   size_t size,
3157 					   u64 *dma_handle,
3158 					   gfp_t flag)
3159 {
3160 	if (dev->dma_ops)
3161 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3162 	else {
3163 		dma_addr_t handle;
3164 		void *ret;
3165 
3166 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3167 		*dma_handle = handle;
3168 		return ret;
3169 	}
3170 }
3171 
3172 /**
3173  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3174  * @dev: The device for which the DMA addresses were allocated
3175  * @size: The size of the region
3176  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3177  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3178  */
3179 static inline void ib_dma_free_coherent(struct ib_device *dev,
3180 					size_t size, void *cpu_addr,
3181 					u64 dma_handle)
3182 {
3183 	if (dev->dma_ops)
3184 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3185 	else
3186 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3187 }
3188 
3189 /**
3190  * ib_dereg_mr - Deregisters a memory region and removes it from the
3191  *   HCA translation table.
3192  * @mr: The memory region to deregister.
3193  *
3194  * This function can fail, if the memory region has memory windows bound to it.
3195  */
3196 int ib_dereg_mr(struct ib_mr *mr);
3197 
3198 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3199 			  enum ib_mr_type mr_type,
3200 			  u32 max_num_sg);
3201 
3202 /**
3203  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3204  *   R_Key and L_Key.
3205  * @mr - struct ib_mr pointer to be updated.
3206  * @newkey - new key to be used.
3207  */
3208 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3209 {
3210 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3211 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3212 }
3213 
3214 /**
3215  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3216  * for calculating a new rkey for type 2 memory windows.
3217  * @rkey - the rkey to increment.
3218  */
3219 static inline u32 ib_inc_rkey(u32 rkey)
3220 {
3221 	const u32 mask = 0x000000ff;
3222 	return ((rkey + 1) & mask) | (rkey & ~mask);
3223 }
3224 
3225 /**
3226  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3227  * @pd: The protection domain associated with the unmapped region.
3228  * @mr_access_flags: Specifies the memory access rights.
3229  * @fmr_attr: Attributes of the unmapped region.
3230  *
3231  * A fast memory region must be mapped before it can be used as part of
3232  * a work request.
3233  */
3234 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3235 			    int mr_access_flags,
3236 			    struct ib_fmr_attr *fmr_attr);
3237 
3238 /**
3239  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3240  * @fmr: The fast memory region to associate with the pages.
3241  * @page_list: An array of physical pages to map to the fast memory region.
3242  * @list_len: The number of pages in page_list.
3243  * @iova: The I/O virtual address to use with the mapped region.
3244  */
3245 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3246 				  u64 *page_list, int list_len,
3247 				  u64 iova)
3248 {
3249 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3250 }
3251 
3252 /**
3253  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3254  * @fmr_list: A linked list of fast memory regions to unmap.
3255  */
3256 int ib_unmap_fmr(struct list_head *fmr_list);
3257 
3258 /**
3259  * ib_dealloc_fmr - Deallocates a fast memory region.
3260  * @fmr: The fast memory region to deallocate.
3261  */
3262 int ib_dealloc_fmr(struct ib_fmr *fmr);
3263 
3264 /**
3265  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3266  * @qp: QP to attach to the multicast group.  The QP must be type
3267  *   IB_QPT_UD.
3268  * @gid: Multicast group GID.
3269  * @lid: Multicast group LID in host byte order.
3270  *
3271  * In order to send and receive multicast packets, subnet
3272  * administration must have created the multicast group and configured
3273  * the fabric appropriately.  The port associated with the specified
3274  * QP must also be a member of the multicast group.
3275  */
3276 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3277 
3278 /**
3279  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3280  * @qp: QP to detach from the multicast group.
3281  * @gid: Multicast group GID.
3282  * @lid: Multicast group LID in host byte order.
3283  */
3284 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3285 
3286 /**
3287  * ib_alloc_xrcd - Allocates an XRC domain.
3288  * @device: The device on which to allocate the XRC domain.
3289  */
3290 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3291 
3292 /**
3293  * ib_dealloc_xrcd - Deallocates an XRC domain.
3294  * @xrcd: The XRC domain to deallocate.
3295  */
3296 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3297 
3298 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3299 			       struct ib_flow_attr *flow_attr, int domain);
3300 int ib_destroy_flow(struct ib_flow *flow_id);
3301 
3302 static inline int ib_check_mr_access(int flags)
3303 {
3304 	/*
3305 	 * Local write permission is required if remote write or
3306 	 * remote atomic permission is also requested.
3307 	 */
3308 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3309 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3310 		return -EINVAL;
3311 
3312 	return 0;
3313 }
3314 
3315 /**
3316  * ib_check_mr_status: lightweight check of MR status.
3317  *     This routine may provide status checks on a selected
3318  *     ib_mr. first use is for signature status check.
3319  *
3320  * @mr: A memory region.
3321  * @check_mask: Bitmask of which checks to perform from
3322  *     ib_mr_status_check enumeration.
3323  * @mr_status: The container of relevant status checks.
3324  *     failed checks will be indicated in the status bitmask
3325  *     and the relevant info shall be in the error item.
3326  */
3327 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3328 		       struct ib_mr_status *mr_status);
3329 
3330 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3331 					    u16 pkey, const union ib_gid *gid,
3332 					    const struct sockaddr *addr);
3333 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3334 			   struct ib_wq_init_attr *init_attr);
3335 int ib_destroy_wq(struct ib_wq *wq);
3336 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3337 		 u32 wq_attr_mask);
3338 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3339 						 struct ib_rwq_ind_table_init_attr*
3340 						 wq_ind_table_init_attr);
3341 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3342 
3343 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3344 		 unsigned int *sg_offset, unsigned int page_size);
3345 
3346 static inline int
3347 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3348 		  unsigned int *sg_offset, unsigned int page_size)
3349 {
3350 	int n;
3351 
3352 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3353 	mr->iova = 0;
3354 
3355 	return n;
3356 }
3357 
3358 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3359 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3360 
3361 void ib_drain_rq(struct ib_qp *qp);
3362 void ib_drain_sq(struct ib_qp *qp);
3363 void ib_drain_qp(struct ib_qp *qp);
3364 
3365 int ib_resolve_eth_dmac(struct ib_device *device,
3366 			struct ib_ah_attr *ah_attr);
3367 #endif /* IB_VERBS_H */
3368