xref: /freebsd/sys/ofed/include/rdma/ib_verbs.h (revision 13ea0450a9c8742119d36f3bf8f47accdce46e54)
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 	IB_SPEED_HDR	= 64
425 };
426 
427 /**
428  * struct rdma_hw_stats
429  * @timestamp - Used by the core code to track when the last update was
430  * @lifespan - Used by the core code to determine how old the counters
431  *   should be before being updated again.  Stored in jiffies, defaults
432  *   to 10 milliseconds, drivers can override the default be specifying
433  *   their own value during their allocation routine.
434  * @name - Array of pointers to static names used for the counters in
435  *   directory.
436  * @num_counters - How many hardware counters there are.  If name is
437  *   shorter than this number, a kernel oops will result.  Driver authors
438  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
439  *   in their code to prevent this.
440  * @value - Array of u64 counters that are accessed by the sysfs code and
441  *   filled in by the drivers get_stats routine
442  */
443 struct rdma_hw_stats {
444 	unsigned long	timestamp;
445 	unsigned long	lifespan;
446 	const char * const *names;
447 	int		num_counters;
448 	u64		value[];
449 };
450 
451 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
452 /**
453  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
454  *   for drivers.
455  * @names - Array of static const char *
456  * @num_counters - How many elements in array
457  * @lifespan - How many milliseconds between updates
458  */
459 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
460 		const char * const *names, int num_counters,
461 		unsigned long lifespan)
462 {
463 	struct rdma_hw_stats *stats;
464 
465 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
466 			GFP_KERNEL);
467 	if (!stats)
468 		return NULL;
469 	stats->names = names;
470 	stats->num_counters = num_counters;
471 	stats->lifespan = msecs_to_jiffies(lifespan);
472 
473 	return stats;
474 }
475 
476 
477 /* Define bits for the various functionality this port needs to be supported by
478  * the core.
479  */
480 /* Management                           0x00000FFF */
481 #define RDMA_CORE_CAP_IB_MAD            0x00000001
482 #define RDMA_CORE_CAP_IB_SMI            0x00000002
483 #define RDMA_CORE_CAP_IB_CM             0x00000004
484 #define RDMA_CORE_CAP_IW_CM             0x00000008
485 #define RDMA_CORE_CAP_IB_SA             0x00000010
486 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
487 
488 /* Address format                       0x000FF000 */
489 #define RDMA_CORE_CAP_AF_IB             0x00001000
490 #define RDMA_CORE_CAP_ETH_AH            0x00002000
491 
492 /* Protocol                             0xFFF00000 */
493 #define RDMA_CORE_CAP_PROT_IB           0x00100000
494 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
495 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
496 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
497 
498 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
499 					| RDMA_CORE_CAP_IB_MAD \
500 					| RDMA_CORE_CAP_IB_SMI \
501 					| RDMA_CORE_CAP_IB_CM  \
502 					| RDMA_CORE_CAP_IB_SA  \
503 					| RDMA_CORE_CAP_AF_IB)
504 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
505 					| RDMA_CORE_CAP_IB_MAD  \
506 					| RDMA_CORE_CAP_IB_CM   \
507 					| RDMA_CORE_CAP_AF_IB   \
508 					| RDMA_CORE_CAP_ETH_AH)
509 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
510 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
511 					| RDMA_CORE_CAP_IB_MAD  \
512 					| RDMA_CORE_CAP_IB_CM   \
513 					| RDMA_CORE_CAP_AF_IB   \
514 					| RDMA_CORE_CAP_ETH_AH)
515 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
516 					| RDMA_CORE_CAP_IW_CM)
517 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
518 					| RDMA_CORE_CAP_OPA_MAD)
519 
520 struct ib_port_attr {
521 	u64			subnet_prefix;
522 	enum ib_port_state	state;
523 	enum ib_mtu		max_mtu;
524 	enum ib_mtu		active_mtu;
525 	int			gid_tbl_len;
526 	u32			port_cap_flags;
527 	u32			max_msg_sz;
528 	u32			bad_pkey_cntr;
529 	u32			qkey_viol_cntr;
530 	u16			pkey_tbl_len;
531 	u16			lid;
532 	u16			sm_lid;
533 	u8			lmc;
534 	u8			max_vl_num;
535 	u8			sm_sl;
536 	u8			subnet_timeout;
537 	u8			init_type_reply;
538 	u8			active_width;
539 	u8			active_speed;
540 	u8                      phys_state;
541 	bool			grh_required;
542 };
543 
544 enum ib_device_modify_flags {
545 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
546 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
547 };
548 
549 #define IB_DEVICE_NODE_DESC_MAX 64
550 
551 struct ib_device_modify {
552 	u64	sys_image_guid;
553 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
554 };
555 
556 enum ib_port_modify_flags {
557 	IB_PORT_SHUTDOWN		= 1,
558 	IB_PORT_INIT_TYPE		= (1<<2),
559 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
560 };
561 
562 struct ib_port_modify {
563 	u32	set_port_cap_mask;
564 	u32	clr_port_cap_mask;
565 	u8	init_type;
566 };
567 
568 enum ib_event_type {
569 	IB_EVENT_CQ_ERR,
570 	IB_EVENT_QP_FATAL,
571 	IB_EVENT_QP_REQ_ERR,
572 	IB_EVENT_QP_ACCESS_ERR,
573 	IB_EVENT_COMM_EST,
574 	IB_EVENT_SQ_DRAINED,
575 	IB_EVENT_PATH_MIG,
576 	IB_EVENT_PATH_MIG_ERR,
577 	IB_EVENT_DEVICE_FATAL,
578 	IB_EVENT_PORT_ACTIVE,
579 	IB_EVENT_PORT_ERR,
580 	IB_EVENT_LID_CHANGE,
581 	IB_EVENT_PKEY_CHANGE,
582 	IB_EVENT_SM_CHANGE,
583 	IB_EVENT_SRQ_ERR,
584 	IB_EVENT_SRQ_LIMIT_REACHED,
585 	IB_EVENT_QP_LAST_WQE_REACHED,
586 	IB_EVENT_CLIENT_REREGISTER,
587 	IB_EVENT_GID_CHANGE,
588 	IB_EVENT_WQ_FATAL,
589 };
590 
591 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
592 
593 struct ib_event {
594 	struct ib_device	*device;
595 	union {
596 		struct ib_cq	*cq;
597 		struct ib_qp	*qp;
598 		struct ib_srq	*srq;
599 		struct ib_wq	*wq;
600 		u8		port_num;
601 	} element;
602 	enum ib_event_type	event;
603 };
604 
605 struct ib_event_handler {
606 	struct ib_device *device;
607 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
608 	struct list_head  list;
609 };
610 
611 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
612 	do {							\
613 		(_ptr)->device  = _device;			\
614 		(_ptr)->handler = _handler;			\
615 		INIT_LIST_HEAD(&(_ptr)->list);			\
616 	} while (0)
617 
618 struct ib_global_route {
619 	union ib_gid	dgid;
620 	u32		flow_label;
621 	u8		sgid_index;
622 	u8		hop_limit;
623 	u8		traffic_class;
624 };
625 
626 struct ib_grh {
627 	__be32		version_tclass_flow;
628 	__be16		paylen;
629 	u8		next_hdr;
630 	u8		hop_limit;
631 	union ib_gid	sgid;
632 	union ib_gid	dgid;
633 };
634 
635 union rdma_network_hdr {
636 	struct ib_grh ibgrh;
637 	struct {
638 		/* The IB spec states that if it's IPv4, the header
639 		 * is located in the last 20 bytes of the header.
640 		 */
641 		u8		reserved[20];
642 		struct ip	roce4grh;
643 	};
644 };
645 
646 enum {
647 	IB_MULTICAST_QPN = 0xffffff
648 };
649 
650 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
651 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
652 
653 enum ib_ah_flags {
654 	IB_AH_GRH	= 1
655 };
656 
657 enum ib_rate {
658 	IB_RATE_PORT_CURRENT = 0,
659 	IB_RATE_2_5_GBPS = 2,
660 	IB_RATE_5_GBPS   = 5,
661 	IB_RATE_10_GBPS  = 3,
662 	IB_RATE_20_GBPS  = 6,
663 	IB_RATE_30_GBPS  = 4,
664 	IB_RATE_40_GBPS  = 7,
665 	IB_RATE_60_GBPS  = 8,
666 	IB_RATE_80_GBPS  = 9,
667 	IB_RATE_120_GBPS = 10,
668 	IB_RATE_14_GBPS  = 11,
669 	IB_RATE_56_GBPS  = 12,
670 	IB_RATE_112_GBPS = 13,
671 	IB_RATE_168_GBPS = 14,
672 	IB_RATE_25_GBPS  = 15,
673 	IB_RATE_100_GBPS = 16,
674 	IB_RATE_200_GBPS = 17,
675 	IB_RATE_300_GBPS = 18
676 };
677 
678 /**
679  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
680  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
681  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
682  * @rate: rate to convert.
683  */
684 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
685 
686 /**
687  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
688  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
689  * @rate: rate to convert.
690  */
691 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
692 
693 
694 /**
695  * enum ib_mr_type - memory region type
696  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
697  *                            normal registration
698  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
699  *                            signature operations (data-integrity
700  *                            capable regions)
701  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
702  *                            register any arbitrary sg lists (without
703  *                            the normal mr constraints - see
704  *                            ib_map_mr_sg)
705  */
706 enum ib_mr_type {
707 	IB_MR_TYPE_MEM_REG,
708 	IB_MR_TYPE_SIGNATURE,
709 	IB_MR_TYPE_SG_GAPS,
710 };
711 
712 /**
713  * Signature types
714  * IB_SIG_TYPE_NONE: Unprotected.
715  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
716  */
717 enum ib_signature_type {
718 	IB_SIG_TYPE_NONE,
719 	IB_SIG_TYPE_T10_DIF,
720 };
721 
722 /**
723  * Signature T10-DIF block-guard types
724  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
725  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
726  */
727 enum ib_t10_dif_bg_type {
728 	IB_T10DIF_CRC,
729 	IB_T10DIF_CSUM
730 };
731 
732 /**
733  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
734  *     domain.
735  * @bg_type: T10-DIF block guard type (CRC|CSUM)
736  * @pi_interval: protection information interval.
737  * @bg: seed of guard computation.
738  * @app_tag: application tag of guard block
739  * @ref_tag: initial guard block reference tag.
740  * @ref_remap: Indicate wethear the reftag increments each block
741  * @app_escape: Indicate to skip block check if apptag=0xffff
742  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
743  * @apptag_check_mask: check bitmask of application tag.
744  */
745 struct ib_t10_dif_domain {
746 	enum ib_t10_dif_bg_type bg_type;
747 	u16			pi_interval;
748 	u16			bg;
749 	u16			app_tag;
750 	u32			ref_tag;
751 	bool			ref_remap;
752 	bool			app_escape;
753 	bool			ref_escape;
754 	u16			apptag_check_mask;
755 };
756 
757 /**
758  * struct ib_sig_domain - Parameters for signature domain
759  * @sig_type: specific signauture type
760  * @sig: union of all signature domain attributes that may
761  *     be used to set domain layout.
762  */
763 struct ib_sig_domain {
764 	enum ib_signature_type sig_type;
765 	union {
766 		struct ib_t10_dif_domain dif;
767 	} sig;
768 };
769 
770 /**
771  * struct ib_sig_attrs - Parameters for signature handover operation
772  * @check_mask: bitmask for signature byte check (8 bytes)
773  * @mem: memory domain layout desciptor.
774  * @wire: wire domain layout desciptor.
775  */
776 struct ib_sig_attrs {
777 	u8			check_mask;
778 	struct ib_sig_domain	mem;
779 	struct ib_sig_domain	wire;
780 };
781 
782 enum ib_sig_err_type {
783 	IB_SIG_BAD_GUARD,
784 	IB_SIG_BAD_REFTAG,
785 	IB_SIG_BAD_APPTAG,
786 };
787 
788 /**
789  * struct ib_sig_err - signature error descriptor
790  */
791 struct ib_sig_err {
792 	enum ib_sig_err_type	err_type;
793 	u32			expected;
794 	u32			actual;
795 	u64			sig_err_offset;
796 	u32			key;
797 };
798 
799 enum ib_mr_status_check {
800 	IB_MR_CHECK_SIG_STATUS = 1,
801 };
802 
803 /**
804  * struct ib_mr_status - Memory region status container
805  *
806  * @fail_status: Bitmask of MR checks status. For each
807  *     failed check a corresponding status bit is set.
808  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
809  *     failure.
810  */
811 struct ib_mr_status {
812 	u32		    fail_status;
813 	struct ib_sig_err   sig_err;
814 };
815 
816 /**
817  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
818  * enum.
819  * @mult: multiple to convert.
820  */
821 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
822 
823 struct ib_ah_attr {
824 	struct ib_global_route	grh;
825 	u16			dlid;
826 	u8			sl;
827 	u8			src_path_bits;
828 	u8			static_rate;
829 	u8			ah_flags;
830 	u8			port_num;
831 	u8			dmac[ETH_ALEN];
832 };
833 
834 enum ib_wc_status {
835 	IB_WC_SUCCESS,
836 	IB_WC_LOC_LEN_ERR,
837 	IB_WC_LOC_QP_OP_ERR,
838 	IB_WC_LOC_EEC_OP_ERR,
839 	IB_WC_LOC_PROT_ERR,
840 	IB_WC_WR_FLUSH_ERR,
841 	IB_WC_MW_BIND_ERR,
842 	IB_WC_BAD_RESP_ERR,
843 	IB_WC_LOC_ACCESS_ERR,
844 	IB_WC_REM_INV_REQ_ERR,
845 	IB_WC_REM_ACCESS_ERR,
846 	IB_WC_REM_OP_ERR,
847 	IB_WC_RETRY_EXC_ERR,
848 	IB_WC_RNR_RETRY_EXC_ERR,
849 	IB_WC_LOC_RDD_VIOL_ERR,
850 	IB_WC_REM_INV_RD_REQ_ERR,
851 	IB_WC_REM_ABORT_ERR,
852 	IB_WC_INV_EECN_ERR,
853 	IB_WC_INV_EEC_STATE_ERR,
854 	IB_WC_FATAL_ERR,
855 	IB_WC_RESP_TIMEOUT_ERR,
856 	IB_WC_GENERAL_ERR
857 };
858 
859 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
860 
861 enum ib_wc_opcode {
862 	IB_WC_SEND,
863 	IB_WC_RDMA_WRITE,
864 	IB_WC_RDMA_READ,
865 	IB_WC_COMP_SWAP,
866 	IB_WC_FETCH_ADD,
867 	IB_WC_LSO,
868 	IB_WC_LOCAL_INV,
869 	IB_WC_REG_MR,
870 	IB_WC_MASKED_COMP_SWAP,
871 	IB_WC_MASKED_FETCH_ADD,
872 /*
873  * Set value of IB_WC_RECV so consumers can test if a completion is a
874  * receive by testing (opcode & IB_WC_RECV).
875  */
876 	IB_WC_RECV			= 1 << 7,
877 	IB_WC_RECV_RDMA_WITH_IMM,
878 	IB_WC_DUMMY = -1,	/* force enum signed */
879 };
880 
881 enum ib_wc_flags {
882 	IB_WC_GRH		= 1,
883 	IB_WC_WITH_IMM		= (1<<1),
884 	IB_WC_WITH_INVALIDATE	= (1<<2),
885 	IB_WC_IP_CSUM_OK	= (1<<3),
886 	IB_WC_WITH_SMAC		= (1<<4),
887 	IB_WC_WITH_VLAN		= (1<<5),
888 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
889 };
890 
891 struct ib_wc {
892 	union {
893 		u64		wr_id;
894 		struct ib_cqe	*wr_cqe;
895 	};
896 	enum ib_wc_status	status;
897 	enum ib_wc_opcode	opcode;
898 	u32			vendor_err;
899 	u32			byte_len;
900 	struct ib_qp	       *qp;
901 	union {
902 		__be32		imm_data;
903 		u32		invalidate_rkey;
904 	} ex;
905 	u32			src_qp;
906 	int			wc_flags;
907 	u16			pkey_index;
908 	u16			slid;
909 	u8			sl;
910 	u8			dlid_path_bits;
911 	u8			port_num;	/* valid only for DR SMPs on switches */
912 	u8			smac[ETH_ALEN];
913 	u16			vlan_id;
914 	u8			network_hdr_type;
915 };
916 
917 enum ib_cq_notify_flags {
918 	IB_CQ_SOLICITED			= 1 << 0,
919 	IB_CQ_NEXT_COMP			= 1 << 1,
920 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
921 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
922 };
923 
924 enum ib_srq_type {
925 	IB_SRQT_BASIC,
926 	IB_SRQT_XRC
927 };
928 
929 enum ib_srq_attr_mask {
930 	IB_SRQ_MAX_WR	= 1 << 0,
931 	IB_SRQ_LIMIT	= 1 << 1,
932 };
933 
934 struct ib_srq_attr {
935 	u32	max_wr;
936 	u32	max_sge;
937 	u32	srq_limit;
938 };
939 
940 struct ib_srq_init_attr {
941 	void		      (*event_handler)(struct ib_event *, void *);
942 	void		       *srq_context;
943 	struct ib_srq_attr	attr;
944 	enum ib_srq_type	srq_type;
945 
946 	union {
947 		struct {
948 			struct ib_xrcd *xrcd;
949 			struct ib_cq   *cq;
950 		} xrc;
951 	} ext;
952 };
953 
954 struct ib_qp_cap {
955 	u32	max_send_wr;
956 	u32	max_recv_wr;
957 	u32	max_send_sge;
958 	u32	max_recv_sge;
959 	u32	max_inline_data;
960 
961 	/*
962 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
963 	 * ib_create_qp() will calculate the right amount of neededed WRs
964 	 * and MRs based on this.
965 	 */
966 	u32	max_rdma_ctxs;
967 };
968 
969 enum ib_sig_type {
970 	IB_SIGNAL_ALL_WR,
971 	IB_SIGNAL_REQ_WR
972 };
973 
974 enum ib_qp_type {
975 	/*
976 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
977 	 * here (and in that order) since the MAD layer uses them as
978 	 * indices into a 2-entry table.
979 	 */
980 	IB_QPT_SMI,
981 	IB_QPT_GSI,
982 
983 	IB_QPT_RC,
984 	IB_QPT_UC,
985 	IB_QPT_UD,
986 	IB_QPT_RAW_IPV6,
987 	IB_QPT_RAW_ETHERTYPE,
988 	IB_QPT_RAW_PACKET = 8,
989 	IB_QPT_XRC_INI = 9,
990 	IB_QPT_XRC_TGT,
991 	IB_QPT_MAX,
992 	/* Reserve a range for qp types internal to the low level driver.
993 	 * These qp types will not be visible at the IB core layer, so the
994 	 * IB_QPT_MAX usages should not be affected in the core layer
995 	 */
996 	IB_QPT_RESERVED1 = 0x1000,
997 	IB_QPT_RESERVED2,
998 	IB_QPT_RESERVED3,
999 	IB_QPT_RESERVED4,
1000 	IB_QPT_RESERVED5,
1001 	IB_QPT_RESERVED6,
1002 	IB_QPT_RESERVED7,
1003 	IB_QPT_RESERVED8,
1004 	IB_QPT_RESERVED9,
1005 	IB_QPT_RESERVED10,
1006 };
1007 
1008 enum ib_qp_create_flags {
1009 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1010 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1011 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1012 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1013 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1014 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1015 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1016 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1017 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1018 	/* reserve bits 26-31 for low level drivers' internal use */
1019 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1020 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1021 };
1022 
1023 /*
1024  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1025  * callback to destroy the passed in QP.
1026  */
1027 
1028 struct ib_qp_init_attr {
1029 	void                  (*event_handler)(struct ib_event *, void *);
1030 	void		       *qp_context;
1031 	struct ib_cq	       *send_cq;
1032 	struct ib_cq	       *recv_cq;
1033 	struct ib_srq	       *srq;
1034 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1035 	struct ib_qp_cap	cap;
1036 	enum ib_sig_type	sq_sig_type;
1037 	enum ib_qp_type		qp_type;
1038 	enum ib_qp_create_flags	create_flags;
1039 
1040 	/*
1041 	 * Only needed for special QP types, or when using the RW API.
1042 	 */
1043 	u8			port_num;
1044 	struct ib_rwq_ind_table *rwq_ind_tbl;
1045 };
1046 
1047 struct ib_qp_open_attr {
1048 	void                  (*event_handler)(struct ib_event *, void *);
1049 	void		       *qp_context;
1050 	u32			qp_num;
1051 	enum ib_qp_type		qp_type;
1052 };
1053 
1054 enum ib_rnr_timeout {
1055 	IB_RNR_TIMER_655_36 =  0,
1056 	IB_RNR_TIMER_000_01 =  1,
1057 	IB_RNR_TIMER_000_02 =  2,
1058 	IB_RNR_TIMER_000_03 =  3,
1059 	IB_RNR_TIMER_000_04 =  4,
1060 	IB_RNR_TIMER_000_06 =  5,
1061 	IB_RNR_TIMER_000_08 =  6,
1062 	IB_RNR_TIMER_000_12 =  7,
1063 	IB_RNR_TIMER_000_16 =  8,
1064 	IB_RNR_TIMER_000_24 =  9,
1065 	IB_RNR_TIMER_000_32 = 10,
1066 	IB_RNR_TIMER_000_48 = 11,
1067 	IB_RNR_TIMER_000_64 = 12,
1068 	IB_RNR_TIMER_000_96 = 13,
1069 	IB_RNR_TIMER_001_28 = 14,
1070 	IB_RNR_TIMER_001_92 = 15,
1071 	IB_RNR_TIMER_002_56 = 16,
1072 	IB_RNR_TIMER_003_84 = 17,
1073 	IB_RNR_TIMER_005_12 = 18,
1074 	IB_RNR_TIMER_007_68 = 19,
1075 	IB_RNR_TIMER_010_24 = 20,
1076 	IB_RNR_TIMER_015_36 = 21,
1077 	IB_RNR_TIMER_020_48 = 22,
1078 	IB_RNR_TIMER_030_72 = 23,
1079 	IB_RNR_TIMER_040_96 = 24,
1080 	IB_RNR_TIMER_061_44 = 25,
1081 	IB_RNR_TIMER_081_92 = 26,
1082 	IB_RNR_TIMER_122_88 = 27,
1083 	IB_RNR_TIMER_163_84 = 28,
1084 	IB_RNR_TIMER_245_76 = 29,
1085 	IB_RNR_TIMER_327_68 = 30,
1086 	IB_RNR_TIMER_491_52 = 31
1087 };
1088 
1089 enum ib_qp_attr_mask {
1090 	IB_QP_STATE			= 1,
1091 	IB_QP_CUR_STATE			= (1<<1),
1092 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1093 	IB_QP_ACCESS_FLAGS		= (1<<3),
1094 	IB_QP_PKEY_INDEX		= (1<<4),
1095 	IB_QP_PORT			= (1<<5),
1096 	IB_QP_QKEY			= (1<<6),
1097 	IB_QP_AV			= (1<<7),
1098 	IB_QP_PATH_MTU			= (1<<8),
1099 	IB_QP_TIMEOUT			= (1<<9),
1100 	IB_QP_RETRY_CNT			= (1<<10),
1101 	IB_QP_RNR_RETRY			= (1<<11),
1102 	IB_QP_RQ_PSN			= (1<<12),
1103 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1104 	IB_QP_ALT_PATH			= (1<<14),
1105 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1106 	IB_QP_SQ_PSN			= (1<<16),
1107 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1108 	IB_QP_PATH_MIG_STATE		= (1<<18),
1109 	IB_QP_CAP			= (1<<19),
1110 	IB_QP_DEST_QPN			= (1<<20),
1111 	IB_QP_RESERVED1			= (1<<21),
1112 	IB_QP_RESERVED2			= (1<<22),
1113 	IB_QP_RESERVED3			= (1<<23),
1114 	IB_QP_RESERVED4			= (1<<24),
1115 };
1116 
1117 enum ib_qp_state {
1118 	IB_QPS_RESET,
1119 	IB_QPS_INIT,
1120 	IB_QPS_RTR,
1121 	IB_QPS_RTS,
1122 	IB_QPS_SQD,
1123 	IB_QPS_SQE,
1124 	IB_QPS_ERR,
1125 	IB_QPS_DUMMY = -1,	/* force enum signed */
1126 };
1127 
1128 enum ib_mig_state {
1129 	IB_MIG_MIGRATED,
1130 	IB_MIG_REARM,
1131 	IB_MIG_ARMED
1132 };
1133 
1134 enum ib_mw_type {
1135 	IB_MW_TYPE_1 = 1,
1136 	IB_MW_TYPE_2 = 2
1137 };
1138 
1139 struct ib_qp_attr {
1140 	enum ib_qp_state	qp_state;
1141 	enum ib_qp_state	cur_qp_state;
1142 	enum ib_mtu		path_mtu;
1143 	enum ib_mig_state	path_mig_state;
1144 	u32			qkey;
1145 	u32			rq_psn;
1146 	u32			sq_psn;
1147 	u32			dest_qp_num;
1148 	int			qp_access_flags;
1149 	struct ib_qp_cap	cap;
1150 	struct ib_ah_attr	ah_attr;
1151 	struct ib_ah_attr	alt_ah_attr;
1152 	u16			pkey_index;
1153 	u16			alt_pkey_index;
1154 	u8			en_sqd_async_notify;
1155 	u8			sq_draining;
1156 	u8			max_rd_atomic;
1157 	u8			max_dest_rd_atomic;
1158 	u8			min_rnr_timer;
1159 	u8			port_num;
1160 	u8			timeout;
1161 	u8			retry_cnt;
1162 	u8			rnr_retry;
1163 	u8			alt_port_num;
1164 	u8			alt_timeout;
1165 };
1166 
1167 enum ib_wr_opcode {
1168 	IB_WR_RDMA_WRITE,
1169 	IB_WR_RDMA_WRITE_WITH_IMM,
1170 	IB_WR_SEND,
1171 	IB_WR_SEND_WITH_IMM,
1172 	IB_WR_RDMA_READ,
1173 	IB_WR_ATOMIC_CMP_AND_SWP,
1174 	IB_WR_ATOMIC_FETCH_AND_ADD,
1175 	IB_WR_LSO,
1176 	IB_WR_SEND_WITH_INV,
1177 	IB_WR_RDMA_READ_WITH_INV,
1178 	IB_WR_LOCAL_INV,
1179 	IB_WR_REG_MR,
1180 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1181 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1182 	IB_WR_REG_SIG_MR,
1183 	/* reserve values for low level drivers' internal use.
1184 	 * These values will not be used at all in the ib core layer.
1185 	 */
1186 	IB_WR_RESERVED1 = 0xf0,
1187 	IB_WR_RESERVED2,
1188 	IB_WR_RESERVED3,
1189 	IB_WR_RESERVED4,
1190 	IB_WR_RESERVED5,
1191 	IB_WR_RESERVED6,
1192 	IB_WR_RESERVED7,
1193 	IB_WR_RESERVED8,
1194 	IB_WR_RESERVED9,
1195 	IB_WR_RESERVED10,
1196 	IB_WR_DUMMY = -1,	/* force enum signed */
1197 };
1198 
1199 enum ib_send_flags {
1200 	IB_SEND_FENCE		= 1,
1201 	IB_SEND_SIGNALED	= (1<<1),
1202 	IB_SEND_SOLICITED	= (1<<2),
1203 	IB_SEND_INLINE		= (1<<3),
1204 	IB_SEND_IP_CSUM		= (1<<4),
1205 
1206 	/* reserve bits 26-31 for low level drivers' internal use */
1207 	IB_SEND_RESERVED_START	= (1 << 26),
1208 	IB_SEND_RESERVED_END	= (1 << 31),
1209 };
1210 
1211 struct ib_sge {
1212 	u64	addr;
1213 	u32	length;
1214 	u32	lkey;
1215 };
1216 
1217 struct ib_cqe {
1218 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1219 };
1220 
1221 struct ib_send_wr {
1222 	struct ib_send_wr      *next;
1223 	union {
1224 		u64		wr_id;
1225 		struct ib_cqe	*wr_cqe;
1226 	};
1227 	struct ib_sge	       *sg_list;
1228 	int			num_sge;
1229 	enum ib_wr_opcode	opcode;
1230 	int			send_flags;
1231 	union {
1232 		__be32		imm_data;
1233 		u32		invalidate_rkey;
1234 	} ex;
1235 };
1236 
1237 struct ib_rdma_wr {
1238 	struct ib_send_wr	wr;
1239 	u64			remote_addr;
1240 	u32			rkey;
1241 };
1242 
1243 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1244 {
1245 	return container_of(wr, struct ib_rdma_wr, wr);
1246 }
1247 
1248 struct ib_atomic_wr {
1249 	struct ib_send_wr	wr;
1250 	u64			remote_addr;
1251 	u64			compare_add;
1252 	u64			swap;
1253 	u64			compare_add_mask;
1254 	u64			swap_mask;
1255 	u32			rkey;
1256 };
1257 
1258 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1259 {
1260 	return container_of(wr, struct ib_atomic_wr, wr);
1261 }
1262 
1263 struct ib_ud_wr {
1264 	struct ib_send_wr	wr;
1265 	struct ib_ah		*ah;
1266 	void			*header;
1267 	int			hlen;
1268 	int			mss;
1269 	u32			remote_qpn;
1270 	u32			remote_qkey;
1271 	u16			pkey_index; /* valid for GSI only */
1272 	u8			port_num;   /* valid for DR SMPs on switch only */
1273 };
1274 
1275 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1276 {
1277 	return container_of(wr, struct ib_ud_wr, wr);
1278 }
1279 
1280 struct ib_reg_wr {
1281 	struct ib_send_wr	wr;
1282 	struct ib_mr		*mr;
1283 	u32			key;
1284 	int			access;
1285 };
1286 
1287 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1288 {
1289 	return container_of(wr, struct ib_reg_wr, wr);
1290 }
1291 
1292 struct ib_sig_handover_wr {
1293 	struct ib_send_wr	wr;
1294 	struct ib_sig_attrs    *sig_attrs;
1295 	struct ib_mr	       *sig_mr;
1296 	int			access_flags;
1297 	struct ib_sge	       *prot;
1298 };
1299 
1300 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1301 {
1302 	return container_of(wr, struct ib_sig_handover_wr, wr);
1303 }
1304 
1305 struct ib_recv_wr {
1306 	struct ib_recv_wr      *next;
1307 	union {
1308 		u64		wr_id;
1309 		struct ib_cqe	*wr_cqe;
1310 	};
1311 	struct ib_sge	       *sg_list;
1312 	int			num_sge;
1313 };
1314 
1315 enum ib_access_flags {
1316 	IB_ACCESS_LOCAL_WRITE	= 1,
1317 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1318 	IB_ACCESS_REMOTE_READ	= (1<<2),
1319 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1320 	IB_ACCESS_MW_BIND	= (1<<4),
1321 	IB_ZERO_BASED		= (1<<5),
1322 	IB_ACCESS_ON_DEMAND     = (1<<6),
1323 };
1324 
1325 /*
1326  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1327  * are hidden here instead of a uapi header!
1328  */
1329 enum ib_mr_rereg_flags {
1330 	IB_MR_REREG_TRANS	= 1,
1331 	IB_MR_REREG_PD		= (1<<1),
1332 	IB_MR_REREG_ACCESS	= (1<<2),
1333 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1334 };
1335 
1336 struct ib_fmr_attr {
1337 	int	max_pages;
1338 	int	max_maps;
1339 	u8	page_shift;
1340 };
1341 
1342 struct ib_umem;
1343 
1344 struct ib_ucontext {
1345 	struct ib_device       *device;
1346 	struct list_head	pd_list;
1347 	struct list_head	mr_list;
1348 	struct list_head	mw_list;
1349 	struct list_head	cq_list;
1350 	struct list_head	qp_list;
1351 	struct list_head	srq_list;
1352 	struct list_head	ah_list;
1353 	struct list_head	xrcd_list;
1354 	struct list_head	rule_list;
1355 	struct list_head	wq_list;
1356 	struct list_head	rwq_ind_tbl_list;
1357 	int			closing;
1358 
1359 	pid_t			tgid;
1360 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1361 	struct rb_root      umem_tree;
1362 	/*
1363 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1364 	 * mmu notifiers registration.
1365 	 */
1366 	struct rw_semaphore	umem_rwsem;
1367 	void (*invalidate_range)(struct ib_umem *umem,
1368 				 unsigned long start, unsigned long end);
1369 
1370 	struct mmu_notifier	mn;
1371 	atomic_t		notifier_count;
1372 	/* A list of umems that don't have private mmu notifier counters yet. */
1373 	struct list_head	no_private_counters;
1374 	int                     odp_mrs_count;
1375 #endif
1376 };
1377 
1378 struct ib_uobject {
1379 	u64			user_handle;	/* handle given to us by userspace */
1380 	struct ib_ucontext     *context;	/* associated user context */
1381 	void		       *object;		/* containing object */
1382 	struct list_head	list;		/* link to context's list */
1383 	int			id;		/* index into kernel idr */
1384 	struct kref		ref;
1385 	struct rw_semaphore	mutex;		/* protects .live */
1386 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1387 	int			live;
1388 };
1389 
1390 struct ib_udata {
1391 	const void __user *inbuf;
1392 	void __user *outbuf;
1393 	size_t       inlen;
1394 	size_t       outlen;
1395 };
1396 
1397 struct ib_pd {
1398 	u32			local_dma_lkey;
1399 	u32			flags;
1400 	struct ib_device       *device;
1401 	struct ib_uobject      *uobject;
1402 	atomic_t          	usecnt; /* count all resources */
1403 
1404 	u32			unsafe_global_rkey;
1405 
1406 	/*
1407 	 * Implementation details of the RDMA core, don't use in drivers:
1408 	 */
1409 	struct ib_mr	       *__internal_mr;
1410 };
1411 
1412 struct ib_xrcd {
1413 	struct ib_device       *device;
1414 	atomic_t		usecnt; /* count all exposed resources */
1415 	struct inode	       *inode;
1416 
1417 	struct mutex		tgt_qp_mutex;
1418 	struct list_head	tgt_qp_list;
1419 };
1420 
1421 struct ib_ah {
1422 	struct ib_device	*device;
1423 	struct ib_pd		*pd;
1424 	struct ib_uobject	*uobject;
1425 };
1426 
1427 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1428 
1429 enum ib_poll_context {
1430 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1431 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1432 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1433 };
1434 
1435 struct ib_cq {
1436 	struct ib_device       *device;
1437 	struct ib_uobject      *uobject;
1438 	ib_comp_handler   	comp_handler;
1439 	void                  (*event_handler)(struct ib_event *, void *);
1440 	void                   *cq_context;
1441 	int               	cqe;
1442 	atomic_t          	usecnt; /* count number of work queues */
1443 	enum ib_poll_context	poll_ctx;
1444 	struct work_struct	work;
1445 };
1446 
1447 struct ib_srq {
1448 	struct ib_device       *device;
1449 	struct ib_pd	       *pd;
1450 	struct ib_uobject      *uobject;
1451 	void		      (*event_handler)(struct ib_event *, void *);
1452 	void		       *srq_context;
1453 	enum ib_srq_type	srq_type;
1454 	atomic_t		usecnt;
1455 
1456 	union {
1457 		struct {
1458 			struct ib_xrcd *xrcd;
1459 			struct ib_cq   *cq;
1460 			u32		srq_num;
1461 		} xrc;
1462 	} ext;
1463 };
1464 
1465 enum ib_wq_type {
1466 	IB_WQT_RQ
1467 };
1468 
1469 enum ib_wq_state {
1470 	IB_WQS_RESET,
1471 	IB_WQS_RDY,
1472 	IB_WQS_ERR
1473 };
1474 
1475 struct ib_wq {
1476 	struct ib_device       *device;
1477 	struct ib_uobject      *uobject;
1478 	void		    *wq_context;
1479 	void		    (*event_handler)(struct ib_event *, void *);
1480 	struct ib_pd	       *pd;
1481 	struct ib_cq	       *cq;
1482 	u32		wq_num;
1483 	enum ib_wq_state       state;
1484 	enum ib_wq_type	wq_type;
1485 	atomic_t		usecnt;
1486 };
1487 
1488 struct ib_wq_init_attr {
1489 	void		       *wq_context;
1490 	enum ib_wq_type	wq_type;
1491 	u32		max_wr;
1492 	u32		max_sge;
1493 	struct	ib_cq	       *cq;
1494 	void		    (*event_handler)(struct ib_event *, void *);
1495 };
1496 
1497 enum ib_wq_attr_mask {
1498 	IB_WQ_STATE	= 1 << 0,
1499 	IB_WQ_CUR_STATE	= 1 << 1,
1500 };
1501 
1502 struct ib_wq_attr {
1503 	enum	ib_wq_state	wq_state;
1504 	enum	ib_wq_state	curr_wq_state;
1505 };
1506 
1507 struct ib_rwq_ind_table {
1508 	struct ib_device	*device;
1509 	struct ib_uobject      *uobject;
1510 	atomic_t		usecnt;
1511 	u32		ind_tbl_num;
1512 	u32		log_ind_tbl_size;
1513 	struct ib_wq	**ind_tbl;
1514 };
1515 
1516 struct ib_rwq_ind_table_init_attr {
1517 	u32		log_ind_tbl_size;
1518 	/* Each entry is a pointer to Receive Work Queue */
1519 	struct ib_wq	**ind_tbl;
1520 };
1521 
1522 /*
1523  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1524  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1525  */
1526 struct ib_qp {
1527 	struct ib_device       *device;
1528 	struct ib_pd	       *pd;
1529 	struct ib_cq	       *send_cq;
1530 	struct ib_cq	       *recv_cq;
1531 	spinlock_t		mr_lock;
1532 	struct ib_srq	       *srq;
1533 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1534 	struct list_head	xrcd_list;
1535 
1536 	/* count times opened, mcast attaches, flow attaches */
1537 	atomic_t		usecnt;
1538 	struct list_head	open_list;
1539 	struct ib_qp           *real_qp;
1540 	struct ib_uobject      *uobject;
1541 	void                  (*event_handler)(struct ib_event *, void *);
1542 	void		       *qp_context;
1543 	u32			qp_num;
1544 	u32			max_write_sge;
1545 	u32			max_read_sge;
1546 	enum ib_qp_type		qp_type;
1547 	struct ib_rwq_ind_table *rwq_ind_tbl;
1548 };
1549 
1550 struct ib_mr {
1551 	struct ib_device  *device;
1552 	struct ib_pd	  *pd;
1553 	u32		   lkey;
1554 	u32		   rkey;
1555 	u64		   iova;
1556 	u32		   length;
1557 	unsigned int	   page_size;
1558 	bool		   need_inval;
1559 	union {
1560 		struct ib_uobject	*uobject;	/* user */
1561 		struct list_head	qp_entry;	/* FR */
1562 	};
1563 };
1564 
1565 struct ib_mw {
1566 	struct ib_device	*device;
1567 	struct ib_pd		*pd;
1568 	struct ib_uobject	*uobject;
1569 	u32			rkey;
1570 	enum ib_mw_type         type;
1571 };
1572 
1573 struct ib_fmr {
1574 	struct ib_device	*device;
1575 	struct ib_pd		*pd;
1576 	struct list_head	list;
1577 	u32			lkey;
1578 	u32			rkey;
1579 };
1580 
1581 /* Supported steering options */
1582 enum ib_flow_attr_type {
1583 	/* steering according to rule specifications */
1584 	IB_FLOW_ATTR_NORMAL		= 0x0,
1585 	/* default unicast and multicast rule -
1586 	 * receive all Eth traffic which isn't steered to any QP
1587 	 */
1588 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1589 	/* default multicast rule -
1590 	 * receive all Eth multicast traffic which isn't steered to any QP
1591 	 */
1592 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1593 	/* sniffer rule - receive all port traffic */
1594 	IB_FLOW_ATTR_SNIFFER		= 0x3
1595 };
1596 
1597 /* Supported steering header types */
1598 enum ib_flow_spec_type {
1599 	/* L2 headers*/
1600 	IB_FLOW_SPEC_ETH	= 0x20,
1601 	IB_FLOW_SPEC_IB		= 0x22,
1602 	/* L3 header*/
1603 	IB_FLOW_SPEC_IPV4	= 0x30,
1604 	IB_FLOW_SPEC_IPV6	= 0x31,
1605 	/* L4 headers*/
1606 	IB_FLOW_SPEC_TCP	= 0x40,
1607 	IB_FLOW_SPEC_UDP	= 0x41
1608 };
1609 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1610 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1611 
1612 /* Flow steering rule priority is set according to it's domain.
1613  * Lower domain value means higher priority.
1614  */
1615 enum ib_flow_domain {
1616 	IB_FLOW_DOMAIN_USER,
1617 	IB_FLOW_DOMAIN_ETHTOOL,
1618 	IB_FLOW_DOMAIN_RFS,
1619 	IB_FLOW_DOMAIN_NIC,
1620 	IB_FLOW_DOMAIN_NUM /* Must be last */
1621 };
1622 
1623 enum ib_flow_flags {
1624 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1625 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1626 };
1627 
1628 struct ib_flow_eth_filter {
1629 	u8	dst_mac[6];
1630 	u8	src_mac[6];
1631 	__be16	ether_type;
1632 	__be16	vlan_tag;
1633 	/* Must be last */
1634 	u8	real_sz[0];
1635 };
1636 
1637 struct ib_flow_spec_eth {
1638 	enum ib_flow_spec_type	  type;
1639 	u16			  size;
1640 	struct ib_flow_eth_filter val;
1641 	struct ib_flow_eth_filter mask;
1642 };
1643 
1644 struct ib_flow_ib_filter {
1645 	__be16 dlid;
1646 	__u8   sl;
1647 	/* Must be last */
1648 	u8	real_sz[0];
1649 };
1650 
1651 struct ib_flow_spec_ib {
1652 	enum ib_flow_spec_type	 type;
1653 	u16			 size;
1654 	struct ib_flow_ib_filter val;
1655 	struct ib_flow_ib_filter mask;
1656 };
1657 
1658 /* IPv4 header flags */
1659 enum ib_ipv4_flags {
1660 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1661 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1662 				    last have this flag set */
1663 };
1664 
1665 struct ib_flow_ipv4_filter {
1666 	__be32	src_ip;
1667 	__be32	dst_ip;
1668 	u8	proto;
1669 	u8	tos;
1670 	u8	ttl;
1671 	u8	flags;
1672 	/* Must be last */
1673 	u8	real_sz[0];
1674 };
1675 
1676 struct ib_flow_spec_ipv4 {
1677 	enum ib_flow_spec_type	   type;
1678 	u16			   size;
1679 	struct ib_flow_ipv4_filter val;
1680 	struct ib_flow_ipv4_filter mask;
1681 };
1682 
1683 struct ib_flow_ipv6_filter {
1684 	u8	src_ip[16];
1685 	u8	dst_ip[16];
1686 	__be32	flow_label;
1687 	u8	next_hdr;
1688 	u8	traffic_class;
1689 	u8	hop_limit;
1690 	/* Must be last */
1691 	u8	real_sz[0];
1692 };
1693 
1694 struct ib_flow_spec_ipv6 {
1695 	enum ib_flow_spec_type	   type;
1696 	u16			   size;
1697 	struct ib_flow_ipv6_filter val;
1698 	struct ib_flow_ipv6_filter mask;
1699 };
1700 
1701 struct ib_flow_tcp_udp_filter {
1702 	__be16	dst_port;
1703 	__be16	src_port;
1704 	/* Must be last */
1705 	u8	real_sz[0];
1706 };
1707 
1708 struct ib_flow_spec_tcp_udp {
1709 	enum ib_flow_spec_type	      type;
1710 	u16			      size;
1711 	struct ib_flow_tcp_udp_filter val;
1712 	struct ib_flow_tcp_udp_filter mask;
1713 };
1714 
1715 union ib_flow_spec {
1716 	struct {
1717 		enum ib_flow_spec_type	type;
1718 		u16			size;
1719 	};
1720 	struct ib_flow_spec_eth		eth;
1721 	struct ib_flow_spec_ib		ib;
1722 	struct ib_flow_spec_ipv4        ipv4;
1723 	struct ib_flow_spec_tcp_udp	tcp_udp;
1724 	struct ib_flow_spec_ipv6        ipv6;
1725 };
1726 
1727 struct ib_flow_attr {
1728 	enum ib_flow_attr_type type;
1729 	u16	     size;
1730 	u16	     priority;
1731 	u32	     flags;
1732 	u8	     num_of_specs;
1733 	u8	     port;
1734 	/* Following are the optional layers according to user request
1735 	 * struct ib_flow_spec_xxx
1736 	 * struct ib_flow_spec_yyy
1737 	 */
1738 };
1739 
1740 struct ib_flow {
1741 	struct ib_qp		*qp;
1742 	struct ib_uobject	*uobject;
1743 };
1744 
1745 struct ib_mad_hdr;
1746 struct ib_grh;
1747 
1748 enum ib_process_mad_flags {
1749 	IB_MAD_IGNORE_MKEY	= 1,
1750 	IB_MAD_IGNORE_BKEY	= 2,
1751 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1752 };
1753 
1754 enum ib_mad_result {
1755 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1756 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1757 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1758 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1759 };
1760 
1761 #define IB_DEVICE_NAME_MAX 64
1762 
1763 struct ib_cache {
1764 	rwlock_t                lock;
1765 	struct ib_event_handler event_handler;
1766 	struct ib_pkey_cache  **pkey_cache;
1767 	struct ib_gid_table   **gid_cache;
1768 	u8                     *lmc_cache;
1769 };
1770 
1771 struct ib_dma_mapping_ops {
1772 	int		(*mapping_error)(struct ib_device *dev,
1773 					 u64 dma_addr);
1774 	u64		(*map_single)(struct ib_device *dev,
1775 				      void *ptr, size_t size,
1776 				      enum dma_data_direction direction);
1777 	void		(*unmap_single)(struct ib_device *dev,
1778 					u64 addr, size_t size,
1779 					enum dma_data_direction direction);
1780 	u64		(*map_page)(struct ib_device *dev,
1781 				    struct page *page, unsigned long offset,
1782 				    size_t size,
1783 				    enum dma_data_direction direction);
1784 	void		(*unmap_page)(struct ib_device *dev,
1785 				      u64 addr, size_t size,
1786 				      enum dma_data_direction direction);
1787 	int		(*map_sg)(struct ib_device *dev,
1788 				  struct scatterlist *sg, int nents,
1789 				  enum dma_data_direction direction);
1790 	void		(*unmap_sg)(struct ib_device *dev,
1791 				    struct scatterlist *sg, int nents,
1792 				    enum dma_data_direction direction);
1793 	int		(*map_sg_attrs)(struct ib_device *dev,
1794 					struct scatterlist *sg, int nents,
1795 					enum dma_data_direction direction,
1796 					struct dma_attrs *attrs);
1797 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1798 					  struct scatterlist *sg, int nents,
1799 					  enum dma_data_direction direction,
1800 					  struct dma_attrs *attrs);
1801 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1802 					       u64 dma_handle,
1803 					       size_t size,
1804 					       enum dma_data_direction dir);
1805 	void		(*sync_single_for_device)(struct ib_device *dev,
1806 						  u64 dma_handle,
1807 						  size_t size,
1808 						  enum dma_data_direction dir);
1809 	void		*(*alloc_coherent)(struct ib_device *dev,
1810 					   size_t size,
1811 					   u64 *dma_handle,
1812 					   gfp_t flag);
1813 	void		(*free_coherent)(struct ib_device *dev,
1814 					 size_t size, void *cpu_addr,
1815 					 u64 dma_handle);
1816 };
1817 
1818 struct iw_cm_verbs;
1819 
1820 struct ib_port_immutable {
1821 	int                           pkey_tbl_len;
1822 	int                           gid_tbl_len;
1823 	u32                           core_cap_flags;
1824 	u32                           max_mad_size;
1825 };
1826 
1827 struct ib_device {
1828 	struct device                *dma_device;
1829 
1830 	char                          name[IB_DEVICE_NAME_MAX];
1831 
1832 	struct list_head              event_handler_list;
1833 	spinlock_t                    event_handler_lock;
1834 
1835 	spinlock_t                    client_data_lock;
1836 	struct list_head              core_list;
1837 	/* Access to the client_data_list is protected by the client_data_lock
1838 	 * spinlock and the lists_rwsem read-write semaphore */
1839 	struct list_head              client_data_list;
1840 
1841 	struct ib_cache               cache;
1842 	/**
1843 	 * port_immutable is indexed by port number
1844 	 */
1845 	struct ib_port_immutable     *port_immutable;
1846 
1847 	int			      num_comp_vectors;
1848 
1849 	struct iw_cm_verbs	     *iwcm;
1850 
1851 	/**
1852 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1853 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1854 	 *   core when the device is removed.  A lifespan of -1 in the return
1855 	 *   struct tells the core to set a default lifespan.
1856 	 */
1857 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1858 						     u8 port_num);
1859 	/**
1860 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1861 	 * @index - The index in the value array we wish to have updated, or
1862 	 *   num_counters if we want all stats updated
1863 	 * Return codes -
1864 	 *   < 0 - Error, no counters updated
1865 	 *   index - Updated the single counter pointed to by index
1866 	 *   num_counters - Updated all counters (will reset the timestamp
1867 	 *     and prevent further calls for lifespan milliseconds)
1868 	 * Drivers are allowed to update all counters in leiu of just the
1869 	 *   one given in index at their option
1870 	 */
1871 	int		           (*get_hw_stats)(struct ib_device *device,
1872 						   struct rdma_hw_stats *stats,
1873 						   u8 port, int index);
1874 	int		           (*query_device)(struct ib_device *device,
1875 						   struct ib_device_attr *device_attr,
1876 						   struct ib_udata *udata);
1877 	int		           (*query_port)(struct ib_device *device,
1878 						 u8 port_num,
1879 						 struct ib_port_attr *port_attr);
1880 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1881 						     u8 port_num);
1882 	/* When calling get_netdev, the HW vendor's driver should return the
1883 	 * net device of device @device at port @port_num or NULL if such
1884 	 * a net device doesn't exist. The vendor driver should call dev_hold
1885 	 * on this net device. The HW vendor's device driver must guarantee
1886 	 * that this function returns NULL before the net device reaches
1887 	 * NETDEV_UNREGISTER_FINAL state.
1888 	 */
1889 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1890 						 u8 port_num);
1891 	int		           (*query_gid)(struct ib_device *device,
1892 						u8 port_num, int index,
1893 						union ib_gid *gid);
1894 	/* When calling add_gid, the HW vendor's driver should
1895 	 * add the gid of device @device at gid index @index of
1896 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1897 	 * the network device related to this gid is available
1898 	 * at @attr. @context allows the HW vendor driver to store extra
1899 	 * information together with a GID entry. The HW vendor may allocate
1900 	 * memory to contain this information and store it in @context when a
1901 	 * new GID entry is written to. Params are consistent until the next
1902 	 * call of add_gid or delete_gid. The function should return 0 on
1903 	 * success or error otherwise. The function could be called
1904 	 * concurrently for different ports. This function is only called
1905 	 * when roce_gid_table is used.
1906 	 */
1907 	int		           (*add_gid)(struct ib_device *device,
1908 					      u8 port_num,
1909 					      unsigned int index,
1910 					      const union ib_gid *gid,
1911 					      const struct ib_gid_attr *attr,
1912 					      void **context);
1913 	/* When calling del_gid, the HW vendor's driver should delete the
1914 	 * gid of device @device at gid index @index of port @port_num.
1915 	 * Upon the deletion of a GID entry, the HW vendor must free any
1916 	 * allocated memory. The caller will clear @context afterwards.
1917 	 * This function is only called when roce_gid_table is used.
1918 	 */
1919 	int		           (*del_gid)(struct ib_device *device,
1920 					      u8 port_num,
1921 					      unsigned int index,
1922 					      void **context);
1923 	int		           (*query_pkey)(struct ib_device *device,
1924 						 u8 port_num, u16 index, u16 *pkey);
1925 	int		           (*modify_device)(struct ib_device *device,
1926 						    int device_modify_mask,
1927 						    struct ib_device_modify *device_modify);
1928 	int		           (*modify_port)(struct ib_device *device,
1929 						  u8 port_num, int port_modify_mask,
1930 						  struct ib_port_modify *port_modify);
1931 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1932 						     struct ib_udata *udata);
1933 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1934 	int                        (*mmap)(struct ib_ucontext *context,
1935 					   struct vm_area_struct *vma);
1936 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1937 					       struct ib_ucontext *context,
1938 					       struct ib_udata *udata);
1939 	int                        (*dealloc_pd)(struct ib_pd *pd);
1940 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1941 						struct ib_ah_attr *ah_attr,
1942 						struct ib_udata *udata);
1943 	int                        (*modify_ah)(struct ib_ah *ah,
1944 						struct ib_ah_attr *ah_attr);
1945 	int                        (*query_ah)(struct ib_ah *ah,
1946 					       struct ib_ah_attr *ah_attr);
1947 	int                        (*destroy_ah)(struct ib_ah *ah);
1948 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1949 						 struct ib_srq_init_attr *srq_init_attr,
1950 						 struct ib_udata *udata);
1951 	int                        (*modify_srq)(struct ib_srq *srq,
1952 						 struct ib_srq_attr *srq_attr,
1953 						 enum ib_srq_attr_mask srq_attr_mask,
1954 						 struct ib_udata *udata);
1955 	int                        (*query_srq)(struct ib_srq *srq,
1956 						struct ib_srq_attr *srq_attr);
1957 	int                        (*destroy_srq)(struct ib_srq *srq);
1958 	int                        (*post_srq_recv)(struct ib_srq *srq,
1959 						    struct ib_recv_wr *recv_wr,
1960 						    struct ib_recv_wr **bad_recv_wr);
1961 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1962 						struct ib_qp_init_attr *qp_init_attr,
1963 						struct ib_udata *udata);
1964 	int                        (*modify_qp)(struct ib_qp *qp,
1965 						struct ib_qp_attr *qp_attr,
1966 						int qp_attr_mask,
1967 						struct ib_udata *udata);
1968 	int                        (*query_qp)(struct ib_qp *qp,
1969 					       struct ib_qp_attr *qp_attr,
1970 					       int qp_attr_mask,
1971 					       struct ib_qp_init_attr *qp_init_attr);
1972 	int                        (*destroy_qp)(struct ib_qp *qp);
1973 	int                        (*post_send)(struct ib_qp *qp,
1974 						struct ib_send_wr *send_wr,
1975 						struct ib_send_wr **bad_send_wr);
1976 	int                        (*post_recv)(struct ib_qp *qp,
1977 						struct ib_recv_wr *recv_wr,
1978 						struct ib_recv_wr **bad_recv_wr);
1979 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1980 						const struct ib_cq_init_attr *attr,
1981 						struct ib_ucontext *context,
1982 						struct ib_udata *udata);
1983 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1984 						u16 cq_period);
1985 	int                        (*destroy_cq)(struct ib_cq *cq);
1986 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1987 						struct ib_udata *udata);
1988 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1989 					      struct ib_wc *wc);
1990 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1991 	int                        (*req_notify_cq)(struct ib_cq *cq,
1992 						    enum ib_cq_notify_flags flags);
1993 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1994 						      int wc_cnt);
1995 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1996 						 int mr_access_flags);
1997 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1998 						  u64 start, u64 length,
1999 						  u64 virt_addr,
2000 						  int mr_access_flags,
2001 						  struct ib_udata *udata);
2002 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2003 						    int flags,
2004 						    u64 start, u64 length,
2005 						    u64 virt_addr,
2006 						    int mr_access_flags,
2007 						    struct ib_pd *pd,
2008 						    struct ib_udata *udata);
2009 	int                        (*dereg_mr)(struct ib_mr *mr);
2010 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2011 					       enum ib_mr_type mr_type,
2012 					       u32 max_num_sg);
2013 	int                        (*map_mr_sg)(struct ib_mr *mr,
2014 						struct scatterlist *sg,
2015 						int sg_nents,
2016 						unsigned int *sg_offset);
2017 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2018 					       enum ib_mw_type type,
2019 					       struct ib_udata *udata);
2020 	int                        (*dealloc_mw)(struct ib_mw *mw);
2021 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2022 						int mr_access_flags,
2023 						struct ib_fmr_attr *fmr_attr);
2024 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2025 						   u64 *page_list, int list_len,
2026 						   u64 iova);
2027 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2028 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2029 	int                        (*attach_mcast)(struct ib_qp *qp,
2030 						   union ib_gid *gid,
2031 						   u16 lid);
2032 	int                        (*detach_mcast)(struct ib_qp *qp,
2033 						   union ib_gid *gid,
2034 						   u16 lid);
2035 	int                        (*process_mad)(struct ib_device *device,
2036 						  int process_mad_flags,
2037 						  u8 port_num,
2038 						  const struct ib_wc *in_wc,
2039 						  const struct ib_grh *in_grh,
2040 						  const struct ib_mad_hdr *in_mad,
2041 						  size_t in_mad_size,
2042 						  struct ib_mad_hdr *out_mad,
2043 						  size_t *out_mad_size,
2044 						  u16 *out_mad_pkey_index);
2045 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2046 						 struct ib_ucontext *ucontext,
2047 						 struct ib_udata *udata);
2048 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2049 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2050 						  struct ib_flow_attr
2051 						  *flow_attr,
2052 						  int domain);
2053 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2054 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2055 						      struct ib_mr_status *mr_status);
2056 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2057 	void			   (*drain_rq)(struct ib_qp *qp);
2058 	void			   (*drain_sq)(struct ib_qp *qp);
2059 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2060 							int state);
2061 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2062 						   struct ifla_vf_info *ivf);
2063 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2064 						   struct ifla_vf_stats *stats);
2065 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2066 						  int type);
2067 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2068 						struct ib_wq_init_attr *init_attr,
2069 						struct ib_udata *udata);
2070 	int			   (*destroy_wq)(struct ib_wq *wq);
2071 	int			   (*modify_wq)(struct ib_wq *wq,
2072 						struct ib_wq_attr *attr,
2073 						u32 wq_attr_mask,
2074 						struct ib_udata *udata);
2075 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2076 							   struct ib_rwq_ind_table_init_attr *init_attr,
2077 							   struct ib_udata *udata);
2078 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2079 	struct ib_dma_mapping_ops   *dma_ops;
2080 
2081 	struct module               *owner;
2082 	struct device                dev;
2083 	struct kobject               *ports_parent;
2084 	struct list_head             port_list;
2085 
2086 	enum {
2087 		IB_DEV_UNINITIALIZED,
2088 		IB_DEV_REGISTERED,
2089 		IB_DEV_UNREGISTERED
2090 	}                            reg_state;
2091 
2092 	int			     uverbs_abi_ver;
2093 	u64			     uverbs_cmd_mask;
2094 	u64			     uverbs_ex_cmd_mask;
2095 
2096 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2097 	__be64			     node_guid;
2098 	u32			     local_dma_lkey;
2099 	u16                          is_switch:1;
2100 	u8                           node_type;
2101 	u8                           phys_port_cnt;
2102 	struct ib_device_attr        attrs;
2103 	struct attribute_group	     *hw_stats_ag;
2104 	struct rdma_hw_stats         *hw_stats;
2105 
2106 	/**
2107 	 * The following mandatory functions are used only at device
2108 	 * registration.  Keep functions such as these at the end of this
2109 	 * structure to avoid cache line misses when accessing struct ib_device
2110 	 * in fast paths.
2111 	 */
2112 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2113 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2114 };
2115 
2116 struct ib_client {
2117 	char  *name;
2118 	void (*add)   (struct ib_device *);
2119 	void (*remove)(struct ib_device *, void *client_data);
2120 
2121 	/* Returns the net_dev belonging to this ib_client and matching the
2122 	 * given parameters.
2123 	 * @dev:	 An RDMA device that the net_dev use for communication.
2124 	 * @port:	 A physical port number on the RDMA device.
2125 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2126 	 * @gid:	 A GID that the net_dev uses to communicate.
2127 	 * @addr:	 An IP address the net_dev is configured with.
2128 	 * @client_data: The device's client data set by ib_set_client_data().
2129 	 *
2130 	 * An ib_client that implements a net_dev on top of RDMA devices
2131 	 * (such as IP over IB) should implement this callback, allowing the
2132 	 * rdma_cm module to find the right net_dev for a given request.
2133 	 *
2134 	 * The caller is responsible for calling dev_put on the returned
2135 	 * netdev. */
2136 	struct net_device *(*get_net_dev_by_params)(
2137 			struct ib_device *dev,
2138 			u8 port,
2139 			u16 pkey,
2140 			const union ib_gid *gid,
2141 			const struct sockaddr *addr,
2142 			void *client_data);
2143 	struct list_head list;
2144 };
2145 
2146 struct ib_device *ib_alloc_device(size_t size);
2147 void ib_dealloc_device(struct ib_device *device);
2148 
2149 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2150 
2151 int ib_register_device(struct ib_device *device,
2152 		       int (*port_callback)(struct ib_device *,
2153 					    u8, struct kobject *));
2154 void ib_unregister_device(struct ib_device *device);
2155 
2156 int ib_register_client   (struct ib_client *client);
2157 void ib_unregister_client(struct ib_client *client);
2158 
2159 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2160 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2161 			 void *data);
2162 
2163 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2164 {
2165 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2166 }
2167 
2168 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2169 {
2170 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2171 }
2172 
2173 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2174 				       size_t offset,
2175 				       size_t len)
2176 {
2177 	const void __user *p = (const char __user *)udata->inbuf + offset;
2178 	bool ret;
2179 	u8 *buf;
2180 
2181 	if (len > USHRT_MAX)
2182 		return false;
2183 
2184 	buf = memdup_user(p, len);
2185 	if (IS_ERR(buf))
2186 		return false;
2187 
2188 	ret = !memchr_inv(buf, 0, len);
2189 	kfree(buf);
2190 	return ret;
2191 }
2192 
2193 /**
2194  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2195  * contains all required attributes and no attributes not allowed for
2196  * the given QP state transition.
2197  * @cur_state: Current QP state
2198  * @next_state: Next QP state
2199  * @type: QP type
2200  * @mask: Mask of supplied QP attributes
2201  * @ll : link layer of port
2202  *
2203  * This function is a helper function that a low-level driver's
2204  * modify_qp method can use to validate the consumer's input.  It
2205  * checks that cur_state and next_state are valid QP states, that a
2206  * transition from cur_state to next_state is allowed by the IB spec,
2207  * and that the attribute mask supplied is allowed for the transition.
2208  */
2209 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2210 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2211 		       enum rdma_link_layer ll);
2212 
2213 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2214 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2215 void ib_dispatch_event(struct ib_event *event);
2216 
2217 int ib_query_port(struct ib_device *device,
2218 		  u8 port_num, struct ib_port_attr *port_attr);
2219 
2220 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2221 					       u8 port_num);
2222 
2223 /**
2224  * rdma_cap_ib_switch - Check if the device is IB switch
2225  * @device: Device to check
2226  *
2227  * Device driver is responsible for setting is_switch bit on
2228  * in ib_device structure at init time.
2229  *
2230  * Return: true if the device is IB switch.
2231  */
2232 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2233 {
2234 	return device->is_switch;
2235 }
2236 
2237 /**
2238  * rdma_start_port - Return the first valid port number for the device
2239  * specified
2240  *
2241  * @device: Device to be checked
2242  *
2243  * Return start port number
2244  */
2245 static inline u8 rdma_start_port(const struct ib_device *device)
2246 {
2247 	return rdma_cap_ib_switch(device) ? 0 : 1;
2248 }
2249 
2250 /**
2251  * rdma_end_port - Return the last valid port number for the device
2252  * specified
2253  *
2254  * @device: Device to be checked
2255  *
2256  * Return last port number
2257  */
2258 static inline u8 rdma_end_port(const struct ib_device *device)
2259 {
2260 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2261 }
2262 
2263 static inline int rdma_is_port_valid(const struct ib_device *device,
2264 				     unsigned int port)
2265 {
2266 	return (port >= rdma_start_port(device) &&
2267 		port <= rdma_end_port(device));
2268 }
2269 
2270 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2271 {
2272 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2273 }
2274 
2275 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2276 {
2277 	return device->port_immutable[port_num].core_cap_flags &
2278 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2279 }
2280 
2281 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2282 {
2283 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2284 }
2285 
2286 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2287 {
2288 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2289 }
2290 
2291 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2292 {
2293 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2294 }
2295 
2296 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2297 {
2298 	return rdma_protocol_ib(device, port_num) ||
2299 		rdma_protocol_roce(device, port_num);
2300 }
2301 
2302 /**
2303  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2304  * Management Datagrams.
2305  * @device: Device to check
2306  * @port_num: Port number to check
2307  *
2308  * Management Datagrams (MAD) are a required part of the InfiniBand
2309  * specification and are supported on all InfiniBand devices.  A slightly
2310  * extended version are also supported on OPA interfaces.
2311  *
2312  * Return: true if the port supports sending/receiving of MAD packets.
2313  */
2314 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2315 {
2316 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2317 }
2318 
2319 /**
2320  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2321  * Management Datagrams.
2322  * @device: Device to check
2323  * @port_num: Port number to check
2324  *
2325  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2326  * datagrams with their own versions.  These OPA MADs share many but not all of
2327  * the characteristics of InfiniBand MADs.
2328  *
2329  * OPA MADs differ in the following ways:
2330  *
2331  *    1) MADs are variable size up to 2K
2332  *       IBTA defined MADs remain fixed at 256 bytes
2333  *    2) OPA SMPs must carry valid PKeys
2334  *    3) OPA SMP packets are a different format
2335  *
2336  * Return: true if the port supports OPA MAD packet formats.
2337  */
2338 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2339 {
2340 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2341 		== RDMA_CORE_CAP_OPA_MAD;
2342 }
2343 
2344 /**
2345  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2346  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2347  * @device: Device to check
2348  * @port_num: Port number to check
2349  *
2350  * Each InfiniBand node is required to provide a Subnet Management Agent
2351  * that the subnet manager can access.  Prior to the fabric being fully
2352  * configured by the subnet manager, the SMA is accessed via a well known
2353  * interface called the Subnet Management Interface (SMI).  This interface
2354  * uses directed route packets to communicate with the SM to get around the
2355  * chicken and egg problem of the SM needing to know what's on the fabric
2356  * in order to configure the fabric, and needing to configure the fabric in
2357  * order to send packets to the devices on the fabric.  These directed
2358  * route packets do not need the fabric fully configured in order to reach
2359  * their destination.  The SMI is the only method allowed to send
2360  * directed route packets on an InfiniBand fabric.
2361  *
2362  * Return: true if the port provides an SMI.
2363  */
2364 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2365 {
2366 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2367 }
2368 
2369 /**
2370  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2371  * Communication Manager.
2372  * @device: Device to check
2373  * @port_num: Port number to check
2374  *
2375  * The InfiniBand Communication Manager is one of many pre-defined General
2376  * Service Agents (GSA) that are accessed via the General Service
2377  * Interface (GSI).  It's role is to facilitate establishment of connections
2378  * between nodes as well as other management related tasks for established
2379  * connections.
2380  *
2381  * Return: true if the port supports an IB CM (this does not guarantee that
2382  * a CM is actually running however).
2383  */
2384 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2385 {
2386 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2387 }
2388 
2389 /**
2390  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2391  * Communication Manager.
2392  * @device: Device to check
2393  * @port_num: Port number to check
2394  *
2395  * Similar to above, but specific to iWARP connections which have a different
2396  * managment protocol than InfiniBand.
2397  *
2398  * Return: true if the port supports an iWARP CM (this does not guarantee that
2399  * a CM is actually running however).
2400  */
2401 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2402 {
2403 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2404 }
2405 
2406 /**
2407  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2408  * Subnet Administration.
2409  * @device: Device to check
2410  * @port_num: Port number to check
2411  *
2412  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2413  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2414  * fabrics, devices should resolve routes to other hosts by contacting the
2415  * SA to query the proper route.
2416  *
2417  * Return: true if the port should act as a client to the fabric Subnet
2418  * Administration interface.  This does not imply that the SA service is
2419  * running locally.
2420  */
2421 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2422 {
2423 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2424 }
2425 
2426 /**
2427  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2428  * Multicast.
2429  * @device: Device to check
2430  * @port_num: Port number to check
2431  *
2432  * InfiniBand multicast registration is more complex than normal IPv4 or
2433  * IPv6 multicast registration.  Each Host Channel Adapter must register
2434  * with the Subnet Manager when it wishes to join a multicast group.  It
2435  * should do so only once regardless of how many queue pairs it subscribes
2436  * to this group.  And it should leave the group only after all queue pairs
2437  * attached to the group have been detached.
2438  *
2439  * Return: true if the port must undertake the additional adminstrative
2440  * overhead of registering/unregistering with the SM and tracking of the
2441  * total number of queue pairs attached to the multicast group.
2442  */
2443 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2444 {
2445 	return rdma_cap_ib_sa(device, port_num);
2446 }
2447 
2448 /**
2449  * rdma_cap_af_ib - Check if the port of device has the capability
2450  * Native Infiniband Address.
2451  * @device: Device to check
2452  * @port_num: Port number to check
2453  *
2454  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2455  * GID.  RoCE uses a different mechanism, but still generates a GID via
2456  * a prescribed mechanism and port specific data.
2457  *
2458  * Return: true if the port uses a GID address to identify devices on the
2459  * network.
2460  */
2461 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2462 {
2463 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2464 }
2465 
2466 /**
2467  * rdma_cap_eth_ah - Check if the port of device has the capability
2468  * Ethernet Address Handle.
2469  * @device: Device to check
2470  * @port_num: Port number to check
2471  *
2472  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2473  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2474  * port.  Normally, packet headers are generated by the sending host
2475  * adapter, but when sending connectionless datagrams, we must manually
2476  * inject the proper headers for the fabric we are communicating over.
2477  *
2478  * Return: true if we are running as a RoCE port and must force the
2479  * addition of a Global Route Header built from our Ethernet Address
2480  * Handle into our header list for connectionless packets.
2481  */
2482 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2483 {
2484 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2485 }
2486 
2487 /**
2488  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2489  *
2490  * @device: Device
2491  * @port_num: Port number
2492  *
2493  * This MAD size includes the MAD headers and MAD payload.  No other headers
2494  * are included.
2495  *
2496  * Return the max MAD size required by the Port.  Will return 0 if the port
2497  * does not support MADs
2498  */
2499 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2500 {
2501 	return device->port_immutable[port_num].max_mad_size;
2502 }
2503 
2504 /**
2505  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2506  * @device: Device to check
2507  * @port_num: Port number to check
2508  *
2509  * RoCE GID table mechanism manages the various GIDs for a device.
2510  *
2511  * NOTE: if allocating the port's GID table has failed, this call will still
2512  * return true, but any RoCE GID table API will fail.
2513  *
2514  * Return: true if the port uses RoCE GID table mechanism in order to manage
2515  * its GIDs.
2516  */
2517 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2518 					   u8 port_num)
2519 {
2520 	return rdma_protocol_roce(device, port_num) &&
2521 		device->add_gid && device->del_gid;
2522 }
2523 
2524 /*
2525  * Check if the device supports READ W/ INVALIDATE.
2526  */
2527 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2528 {
2529 	/*
2530 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2531 	 * has support for it yet.
2532 	 */
2533 	return rdma_protocol_iwarp(dev, port_num);
2534 }
2535 
2536 int ib_query_gid(struct ib_device *device,
2537 		 u8 port_num, int index, union ib_gid *gid,
2538 		 struct ib_gid_attr *attr);
2539 
2540 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2541 			 int state);
2542 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2543 		     struct ifla_vf_info *info);
2544 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2545 		    struct ifla_vf_stats *stats);
2546 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2547 		   int type);
2548 
2549 int ib_query_pkey(struct ib_device *device,
2550 		  u8 port_num, u16 index, u16 *pkey);
2551 
2552 int ib_modify_device(struct ib_device *device,
2553 		     int device_modify_mask,
2554 		     struct ib_device_modify *device_modify);
2555 
2556 int ib_modify_port(struct ib_device *device,
2557 		   u8 port_num, int port_modify_mask,
2558 		   struct ib_port_modify *port_modify);
2559 
2560 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2561 		enum ib_gid_type gid_type, struct net_device *ndev,
2562 		u8 *port_num, u16 *index);
2563 
2564 int ib_find_pkey(struct ib_device *device,
2565 		 u8 port_num, u16 pkey, u16 *index);
2566 
2567 enum ib_pd_flags {
2568 	/*
2569 	 * Create a memory registration for all memory in the system and place
2570 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2571 	 * ULPs to avoid the overhead of dynamic MRs.
2572 	 *
2573 	 * This flag is generally considered unsafe and must only be used in
2574 	 * extremly trusted environments.  Every use of it will log a warning
2575 	 * in the kernel log.
2576 	 */
2577 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2578 };
2579 
2580 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2581 		const char *caller);
2582 #define ib_alloc_pd(device, flags) \
2583 	__ib_alloc_pd((device), (flags), __func__)
2584 void ib_dealloc_pd(struct ib_pd *pd);
2585 
2586 /**
2587  * ib_create_ah - Creates an address handle for the given address vector.
2588  * @pd: The protection domain associated with the address handle.
2589  * @ah_attr: The attributes of the address vector.
2590  *
2591  * The address handle is used to reference a local or global destination
2592  * in all UD QP post sends.
2593  */
2594 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2595 
2596 /**
2597  * ib_init_ah_from_wc - Initializes address handle attributes from a
2598  *   work completion.
2599  * @device: Device on which the received message arrived.
2600  * @port_num: Port on which the received message arrived.
2601  * @wc: Work completion associated with the received message.
2602  * @grh: References the received global route header.  This parameter is
2603  *   ignored unless the work completion indicates that the GRH is valid.
2604  * @ah_attr: Returned attributes that can be used when creating an address
2605  *   handle for replying to the message.
2606  */
2607 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2608 		       const struct ib_wc *wc, const struct ib_grh *grh,
2609 		       struct ib_ah_attr *ah_attr);
2610 
2611 /**
2612  * ib_create_ah_from_wc - Creates an address handle associated with the
2613  *   sender of the specified work completion.
2614  * @pd: The protection domain associated with the address handle.
2615  * @wc: Work completion information associated with a received message.
2616  * @grh: References the received global route header.  This parameter is
2617  *   ignored unless the work completion indicates that the GRH is valid.
2618  * @port_num: The outbound port number to associate with the address.
2619  *
2620  * The address handle is used to reference a local or global destination
2621  * in all UD QP post sends.
2622  */
2623 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2624 				   const struct ib_grh *grh, u8 port_num);
2625 
2626 /**
2627  * ib_modify_ah - Modifies the address vector associated with an address
2628  *   handle.
2629  * @ah: The address handle to modify.
2630  * @ah_attr: The new address vector attributes to associate with the
2631  *   address handle.
2632  */
2633 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2634 
2635 /**
2636  * ib_query_ah - Queries the address vector associated with an address
2637  *   handle.
2638  * @ah: The address handle to query.
2639  * @ah_attr: The address vector attributes associated with the address
2640  *   handle.
2641  */
2642 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2643 
2644 /**
2645  * ib_destroy_ah - Destroys an address handle.
2646  * @ah: The address handle to destroy.
2647  */
2648 int ib_destroy_ah(struct ib_ah *ah);
2649 
2650 /**
2651  * ib_create_srq - Creates a SRQ associated with the specified protection
2652  *   domain.
2653  * @pd: The protection domain associated with the SRQ.
2654  * @srq_init_attr: A list of initial attributes required to create the
2655  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2656  *   the actual capabilities of the created SRQ.
2657  *
2658  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2659  * requested size of the SRQ, and set to the actual values allocated
2660  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2661  * will always be at least as large as the requested values.
2662  */
2663 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2664 			     struct ib_srq_init_attr *srq_init_attr);
2665 
2666 /**
2667  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2668  * @srq: The SRQ to modify.
2669  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2670  *   the current values of selected SRQ attributes are returned.
2671  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2672  *   are being modified.
2673  *
2674  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2675  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2676  * the number of receives queued drops below the limit.
2677  */
2678 int ib_modify_srq(struct ib_srq *srq,
2679 		  struct ib_srq_attr *srq_attr,
2680 		  enum ib_srq_attr_mask srq_attr_mask);
2681 
2682 /**
2683  * ib_query_srq - Returns the attribute list and current values for the
2684  *   specified SRQ.
2685  * @srq: The SRQ to query.
2686  * @srq_attr: The attributes of the specified SRQ.
2687  */
2688 int ib_query_srq(struct ib_srq *srq,
2689 		 struct ib_srq_attr *srq_attr);
2690 
2691 /**
2692  * ib_destroy_srq - Destroys the specified SRQ.
2693  * @srq: The SRQ to destroy.
2694  */
2695 int ib_destroy_srq(struct ib_srq *srq);
2696 
2697 /**
2698  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2699  * @srq: The SRQ to post the work request on.
2700  * @recv_wr: A list of work requests to post on the receive queue.
2701  * @bad_recv_wr: On an immediate failure, this parameter will reference
2702  *   the work request that failed to be posted on the QP.
2703  */
2704 static inline int ib_post_srq_recv(struct ib_srq *srq,
2705 				   struct ib_recv_wr *recv_wr,
2706 				   struct ib_recv_wr **bad_recv_wr)
2707 {
2708 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2709 }
2710 
2711 /**
2712  * ib_create_qp - Creates a QP associated with the specified protection
2713  *   domain.
2714  * @pd: The protection domain associated with the QP.
2715  * @qp_init_attr: A list of initial attributes required to create the
2716  *   QP.  If QP creation succeeds, then the attributes are updated to
2717  *   the actual capabilities of the created QP.
2718  */
2719 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2720 			   struct ib_qp_init_attr *qp_init_attr);
2721 
2722 /**
2723  * ib_modify_qp - Modifies the attributes for the specified QP and then
2724  *   transitions the QP to the given state.
2725  * @qp: The QP to modify.
2726  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2727  *   the current values of selected QP attributes are returned.
2728  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2729  *   are being modified.
2730  */
2731 int ib_modify_qp(struct ib_qp *qp,
2732 		 struct ib_qp_attr *qp_attr,
2733 		 int qp_attr_mask);
2734 
2735 /**
2736  * ib_query_qp - Returns the attribute list and current values for the
2737  *   specified QP.
2738  * @qp: The QP to query.
2739  * @qp_attr: The attributes of the specified QP.
2740  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2741  * @qp_init_attr: Additional attributes of the selected QP.
2742  *
2743  * The qp_attr_mask may be used to limit the query to gathering only the
2744  * selected attributes.
2745  */
2746 int ib_query_qp(struct ib_qp *qp,
2747 		struct ib_qp_attr *qp_attr,
2748 		int qp_attr_mask,
2749 		struct ib_qp_init_attr *qp_init_attr);
2750 
2751 /**
2752  * ib_destroy_qp - Destroys the specified QP.
2753  * @qp: The QP to destroy.
2754  */
2755 int ib_destroy_qp(struct ib_qp *qp);
2756 
2757 /**
2758  * ib_open_qp - Obtain a reference to an existing sharable QP.
2759  * @xrcd - XRC domain
2760  * @qp_open_attr: Attributes identifying the QP to open.
2761  *
2762  * Returns a reference to a sharable QP.
2763  */
2764 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2765 			 struct ib_qp_open_attr *qp_open_attr);
2766 
2767 /**
2768  * ib_close_qp - Release an external reference to a QP.
2769  * @qp: The QP handle to release
2770  *
2771  * The opened QP handle is released by the caller.  The underlying
2772  * shared QP is not destroyed until all internal references are released.
2773  */
2774 int ib_close_qp(struct ib_qp *qp);
2775 
2776 /**
2777  * ib_post_send - Posts a list of work requests to the send queue of
2778  *   the specified QP.
2779  * @qp: The QP to post the work request on.
2780  * @send_wr: A list of work requests to post on the send queue.
2781  * @bad_send_wr: On an immediate failure, this parameter will reference
2782  *   the work request that failed to be posted on the QP.
2783  *
2784  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2785  * error is returned, the QP state shall not be affected,
2786  * ib_post_send() will return an immediate error after queueing any
2787  * earlier work requests in the list.
2788  */
2789 static inline int ib_post_send(struct ib_qp *qp,
2790 			       struct ib_send_wr *send_wr,
2791 			       struct ib_send_wr **bad_send_wr)
2792 {
2793 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2794 }
2795 
2796 /**
2797  * ib_post_recv - Posts a list of work requests to the receive queue of
2798  *   the specified QP.
2799  * @qp: The QP to post the work request on.
2800  * @recv_wr: A list of work requests to post on the receive queue.
2801  * @bad_recv_wr: On an immediate failure, this parameter will reference
2802  *   the work request that failed to be posted on the QP.
2803  */
2804 static inline int ib_post_recv(struct ib_qp *qp,
2805 			       struct ib_recv_wr *recv_wr,
2806 			       struct ib_recv_wr **bad_recv_wr)
2807 {
2808 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2809 }
2810 
2811 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2812 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2813 void ib_free_cq(struct ib_cq *cq);
2814 
2815 /**
2816  * ib_create_cq - Creates a CQ on the specified device.
2817  * @device: The device on which to create the CQ.
2818  * @comp_handler: A user-specified callback that is invoked when a
2819  *   completion event occurs on the CQ.
2820  * @event_handler: A user-specified callback that is invoked when an
2821  *   asynchronous event not associated with a completion occurs on the CQ.
2822  * @cq_context: Context associated with the CQ returned to the user via
2823  *   the associated completion and event handlers.
2824  * @cq_attr: The attributes the CQ should be created upon.
2825  *
2826  * Users can examine the cq structure to determine the actual CQ size.
2827  */
2828 struct ib_cq *ib_create_cq(struct ib_device *device,
2829 			   ib_comp_handler comp_handler,
2830 			   void (*event_handler)(struct ib_event *, void *),
2831 			   void *cq_context,
2832 			   const struct ib_cq_init_attr *cq_attr);
2833 
2834 /**
2835  * ib_resize_cq - Modifies the capacity of the CQ.
2836  * @cq: The CQ to resize.
2837  * @cqe: The minimum size of the CQ.
2838  *
2839  * Users can examine the cq structure to determine the actual CQ size.
2840  */
2841 int ib_resize_cq(struct ib_cq *cq, int cqe);
2842 
2843 /**
2844  * ib_modify_cq - Modifies moderation params of the CQ
2845  * @cq: The CQ to modify.
2846  * @cq_count: number of CQEs that will trigger an event
2847  * @cq_period: max period of time in usec before triggering an event
2848  *
2849  */
2850 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2851 
2852 /**
2853  * ib_destroy_cq - Destroys the specified CQ.
2854  * @cq: The CQ to destroy.
2855  */
2856 int ib_destroy_cq(struct ib_cq *cq);
2857 
2858 /**
2859  * ib_poll_cq - poll a CQ for completion(s)
2860  * @cq:the CQ being polled
2861  * @num_entries:maximum number of completions to return
2862  * @wc:array of at least @num_entries &struct ib_wc where completions
2863  *   will be returned
2864  *
2865  * Poll a CQ for (possibly multiple) completions.  If the return value
2866  * is < 0, an error occurred.  If the return value is >= 0, it is the
2867  * number of completions returned.  If the return value is
2868  * non-negative and < num_entries, then the CQ was emptied.
2869  */
2870 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2871 			     struct ib_wc *wc)
2872 {
2873 	return cq->device->poll_cq(cq, num_entries, wc);
2874 }
2875 
2876 /**
2877  * ib_peek_cq - Returns the number of unreaped completions currently
2878  *   on the specified CQ.
2879  * @cq: The CQ to peek.
2880  * @wc_cnt: A minimum number of unreaped completions to check for.
2881  *
2882  * If the number of unreaped completions is greater than or equal to wc_cnt,
2883  * this function returns wc_cnt, otherwise, it returns the actual number of
2884  * unreaped completions.
2885  */
2886 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2887 
2888 /**
2889  * ib_req_notify_cq - Request completion notification on a CQ.
2890  * @cq: The CQ to generate an event for.
2891  * @flags:
2892  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2893  *   to request an event on the next solicited event or next work
2894  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2895  *   may also be |ed in to request a hint about missed events, as
2896  *   described below.
2897  *
2898  * Return Value:
2899  *    < 0 means an error occurred while requesting notification
2900  *   == 0 means notification was requested successfully, and if
2901  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2902  *        were missed and it is safe to wait for another event.  In
2903  *        this case is it guaranteed that any work completions added
2904  *        to the CQ since the last CQ poll will trigger a completion
2905  *        notification event.
2906  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2907  *        in.  It means that the consumer must poll the CQ again to
2908  *        make sure it is empty to avoid missing an event because of a
2909  *        race between requesting notification and an entry being
2910  *        added to the CQ.  This return value means it is possible
2911  *        (but not guaranteed) that a work completion has been added
2912  *        to the CQ since the last poll without triggering a
2913  *        completion notification event.
2914  */
2915 static inline int ib_req_notify_cq(struct ib_cq *cq,
2916 				   enum ib_cq_notify_flags flags)
2917 {
2918 	return cq->device->req_notify_cq(cq, flags);
2919 }
2920 
2921 /**
2922  * ib_req_ncomp_notif - Request completion notification when there are
2923  *   at least the specified number of unreaped completions on the CQ.
2924  * @cq: The CQ to generate an event for.
2925  * @wc_cnt: The number of unreaped completions that should be on the
2926  *   CQ before an event is generated.
2927  */
2928 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2929 {
2930 	return cq->device->req_ncomp_notif ?
2931 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2932 		-ENOSYS;
2933 }
2934 
2935 /**
2936  * ib_dma_mapping_error - check a DMA addr for error
2937  * @dev: The device for which the dma_addr was created
2938  * @dma_addr: The DMA address to check
2939  */
2940 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2941 {
2942 	if (dev->dma_ops)
2943 		return dev->dma_ops->mapping_error(dev, dma_addr);
2944 	return dma_mapping_error(dev->dma_device, dma_addr);
2945 }
2946 
2947 /**
2948  * ib_dma_map_single - Map a kernel virtual address to DMA address
2949  * @dev: The device for which the dma_addr is to be created
2950  * @cpu_addr: The kernel virtual address
2951  * @size: The size of the region in bytes
2952  * @direction: The direction of the DMA
2953  */
2954 static inline u64 ib_dma_map_single(struct ib_device *dev,
2955 				    void *cpu_addr, size_t size,
2956 				    enum dma_data_direction direction)
2957 {
2958 	if (dev->dma_ops)
2959 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2960 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2961 }
2962 
2963 /**
2964  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2965  * @dev: The device for which the DMA address was created
2966  * @addr: The DMA address
2967  * @size: The size of the region in bytes
2968  * @direction: The direction of the DMA
2969  */
2970 static inline void ib_dma_unmap_single(struct ib_device *dev,
2971 				       u64 addr, size_t size,
2972 				       enum dma_data_direction direction)
2973 {
2974 	if (dev->dma_ops)
2975 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2976 	else
2977 		dma_unmap_single(dev->dma_device, addr, size, direction);
2978 }
2979 
2980 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2981 					  void *cpu_addr, size_t size,
2982 					  enum dma_data_direction direction,
2983 					  struct dma_attrs *dma_attrs)
2984 {
2985 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2986 				    direction, dma_attrs);
2987 }
2988 
2989 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2990 					     u64 addr, size_t size,
2991 					     enum dma_data_direction direction,
2992 					     struct dma_attrs *dma_attrs)
2993 {
2994 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2995 				      direction, dma_attrs);
2996 }
2997 
2998 /**
2999  * ib_dma_map_page - Map a physical page to DMA address
3000  * @dev: The device for which the dma_addr is to be created
3001  * @page: The page to be mapped
3002  * @offset: The offset within the page
3003  * @size: The size of the region in bytes
3004  * @direction: The direction of the DMA
3005  */
3006 static inline u64 ib_dma_map_page(struct ib_device *dev,
3007 				  struct page *page,
3008 				  unsigned long offset,
3009 				  size_t size,
3010 					 enum dma_data_direction direction)
3011 {
3012 	if (dev->dma_ops)
3013 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3014 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3015 }
3016 
3017 /**
3018  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3019  * @dev: The device for which the DMA address was created
3020  * @addr: The DMA address
3021  * @size: The size of the region in bytes
3022  * @direction: The direction of the DMA
3023  */
3024 static inline void ib_dma_unmap_page(struct ib_device *dev,
3025 				     u64 addr, size_t size,
3026 				     enum dma_data_direction direction)
3027 {
3028 	if (dev->dma_ops)
3029 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3030 	else
3031 		dma_unmap_page(dev->dma_device, addr, size, direction);
3032 }
3033 
3034 /**
3035  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3036  * @dev: The device for which the DMA addresses are to be created
3037  * @sg: The array of scatter/gather entries
3038  * @nents: The number of scatter/gather entries
3039  * @direction: The direction of the DMA
3040  */
3041 static inline int ib_dma_map_sg(struct ib_device *dev,
3042 				struct scatterlist *sg, int nents,
3043 				enum dma_data_direction direction)
3044 {
3045 	if (dev->dma_ops)
3046 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3047 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3048 }
3049 
3050 /**
3051  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3052  * @dev: The device for which the DMA addresses were created
3053  * @sg: The array of scatter/gather entries
3054  * @nents: The number of scatter/gather entries
3055  * @direction: The direction of the DMA
3056  */
3057 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3058 				   struct scatterlist *sg, int nents,
3059 				   enum dma_data_direction direction)
3060 {
3061 	if (dev->dma_ops)
3062 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3063 	else
3064 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3065 }
3066 
3067 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3068 				      struct scatterlist *sg, int nents,
3069 				      enum dma_data_direction direction,
3070 				      struct dma_attrs *dma_attrs)
3071 {
3072 	if (dev->dma_ops)
3073 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3074 						  dma_attrs);
3075 	else
3076 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3077 					dma_attrs);
3078 }
3079 
3080 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3081 					 struct scatterlist *sg, int nents,
3082 					 enum dma_data_direction direction,
3083 					 struct dma_attrs *dma_attrs)
3084 {
3085 	if (dev->dma_ops)
3086 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3087 						  dma_attrs);
3088 	else
3089 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3090 				   dma_attrs);
3091 }
3092 /**
3093  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3094  * @dev: The device for which the DMA addresses were created
3095  * @sg: The scatter/gather entry
3096  *
3097  * Note: this function is obsolete. To do: change all occurrences of
3098  * ib_sg_dma_address() into sg_dma_address().
3099  */
3100 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3101 				    struct scatterlist *sg)
3102 {
3103 	return sg_dma_address(sg);
3104 }
3105 
3106 /**
3107  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3108  * @dev: The device for which the DMA addresses were created
3109  * @sg: The scatter/gather entry
3110  *
3111  * Note: this function is obsolete. To do: change all occurrences of
3112  * ib_sg_dma_len() into sg_dma_len().
3113  */
3114 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3115 					 struct scatterlist *sg)
3116 {
3117 	return sg_dma_len(sg);
3118 }
3119 
3120 /**
3121  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3122  * @dev: The device for which the DMA address was created
3123  * @addr: The DMA address
3124  * @size: The size of the region in bytes
3125  * @dir: The direction of the DMA
3126  */
3127 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3128 					      u64 addr,
3129 					      size_t size,
3130 					      enum dma_data_direction dir)
3131 {
3132 	if (dev->dma_ops)
3133 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3134 	else
3135 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3136 }
3137 
3138 /**
3139  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3140  * @dev: The device for which the DMA address was created
3141  * @addr: The DMA address
3142  * @size: The size of the region in bytes
3143  * @dir: The direction of the DMA
3144  */
3145 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3146 						 u64 addr,
3147 						 size_t size,
3148 						 enum dma_data_direction dir)
3149 {
3150 	if (dev->dma_ops)
3151 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3152 	else
3153 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3154 }
3155 
3156 /**
3157  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3158  * @dev: The device for which the DMA address is requested
3159  * @size: The size of the region to allocate in bytes
3160  * @dma_handle: A pointer for returning the DMA address of the region
3161  * @flag: memory allocator flags
3162  */
3163 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3164 					   size_t size,
3165 					   u64 *dma_handle,
3166 					   gfp_t flag)
3167 {
3168 	if (dev->dma_ops)
3169 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3170 	else {
3171 		dma_addr_t handle;
3172 		void *ret;
3173 
3174 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3175 		*dma_handle = handle;
3176 		return ret;
3177 	}
3178 }
3179 
3180 /**
3181  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3182  * @dev: The device for which the DMA addresses were allocated
3183  * @size: The size of the region
3184  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3185  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3186  */
3187 static inline void ib_dma_free_coherent(struct ib_device *dev,
3188 					size_t size, void *cpu_addr,
3189 					u64 dma_handle)
3190 {
3191 	if (dev->dma_ops)
3192 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3193 	else
3194 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3195 }
3196 
3197 /**
3198  * ib_dereg_mr - Deregisters a memory region and removes it from the
3199  *   HCA translation table.
3200  * @mr: The memory region to deregister.
3201  *
3202  * This function can fail, if the memory region has memory windows bound to it.
3203  */
3204 int ib_dereg_mr(struct ib_mr *mr);
3205 
3206 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3207 			  enum ib_mr_type mr_type,
3208 			  u32 max_num_sg);
3209 
3210 /**
3211  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3212  *   R_Key and L_Key.
3213  * @mr - struct ib_mr pointer to be updated.
3214  * @newkey - new key to be used.
3215  */
3216 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3217 {
3218 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3219 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3220 }
3221 
3222 /**
3223  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3224  * for calculating a new rkey for type 2 memory windows.
3225  * @rkey - the rkey to increment.
3226  */
3227 static inline u32 ib_inc_rkey(u32 rkey)
3228 {
3229 	const u32 mask = 0x000000ff;
3230 	return ((rkey + 1) & mask) | (rkey & ~mask);
3231 }
3232 
3233 /**
3234  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3235  * @pd: The protection domain associated with the unmapped region.
3236  * @mr_access_flags: Specifies the memory access rights.
3237  * @fmr_attr: Attributes of the unmapped region.
3238  *
3239  * A fast memory region must be mapped before it can be used as part of
3240  * a work request.
3241  */
3242 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3243 			    int mr_access_flags,
3244 			    struct ib_fmr_attr *fmr_attr);
3245 
3246 /**
3247  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3248  * @fmr: The fast memory region to associate with the pages.
3249  * @page_list: An array of physical pages to map to the fast memory region.
3250  * @list_len: The number of pages in page_list.
3251  * @iova: The I/O virtual address to use with the mapped region.
3252  */
3253 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3254 				  u64 *page_list, int list_len,
3255 				  u64 iova)
3256 {
3257 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3258 }
3259 
3260 /**
3261  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3262  * @fmr_list: A linked list of fast memory regions to unmap.
3263  */
3264 int ib_unmap_fmr(struct list_head *fmr_list);
3265 
3266 /**
3267  * ib_dealloc_fmr - Deallocates a fast memory region.
3268  * @fmr: The fast memory region to deallocate.
3269  */
3270 int ib_dealloc_fmr(struct ib_fmr *fmr);
3271 
3272 /**
3273  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3274  * @qp: QP to attach to the multicast group.  The QP must be type
3275  *   IB_QPT_UD.
3276  * @gid: Multicast group GID.
3277  * @lid: Multicast group LID in host byte order.
3278  *
3279  * In order to send and receive multicast packets, subnet
3280  * administration must have created the multicast group and configured
3281  * the fabric appropriately.  The port associated with the specified
3282  * QP must also be a member of the multicast group.
3283  */
3284 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3285 
3286 /**
3287  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3288  * @qp: QP to detach from the multicast group.
3289  * @gid: Multicast group GID.
3290  * @lid: Multicast group LID in host byte order.
3291  */
3292 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3293 
3294 /**
3295  * ib_alloc_xrcd - Allocates an XRC domain.
3296  * @device: The device on which to allocate the XRC domain.
3297  */
3298 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3299 
3300 /**
3301  * ib_dealloc_xrcd - Deallocates an XRC domain.
3302  * @xrcd: The XRC domain to deallocate.
3303  */
3304 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3305 
3306 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3307 			       struct ib_flow_attr *flow_attr, int domain);
3308 int ib_destroy_flow(struct ib_flow *flow_id);
3309 
3310 static inline int ib_check_mr_access(int flags)
3311 {
3312 	/*
3313 	 * Local write permission is required if remote write or
3314 	 * remote atomic permission is also requested.
3315 	 */
3316 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3317 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3318 		return -EINVAL;
3319 
3320 	return 0;
3321 }
3322 
3323 /**
3324  * ib_check_mr_status: lightweight check of MR status.
3325  *     This routine may provide status checks on a selected
3326  *     ib_mr. first use is for signature status check.
3327  *
3328  * @mr: A memory region.
3329  * @check_mask: Bitmask of which checks to perform from
3330  *     ib_mr_status_check enumeration.
3331  * @mr_status: The container of relevant status checks.
3332  *     failed checks will be indicated in the status bitmask
3333  *     and the relevant info shall be in the error item.
3334  */
3335 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3336 		       struct ib_mr_status *mr_status);
3337 
3338 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3339 					    u16 pkey, const union ib_gid *gid,
3340 					    const struct sockaddr *addr);
3341 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3342 			   struct ib_wq_init_attr *init_attr);
3343 int ib_destroy_wq(struct ib_wq *wq);
3344 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3345 		 u32 wq_attr_mask);
3346 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3347 						 struct ib_rwq_ind_table_init_attr*
3348 						 wq_ind_table_init_attr);
3349 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3350 
3351 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3352 		 unsigned int *sg_offset, unsigned int page_size);
3353 
3354 static inline int
3355 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3356 		  unsigned int *sg_offset, unsigned int page_size)
3357 {
3358 	int n;
3359 
3360 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3361 	mr->iova = 0;
3362 
3363 	return n;
3364 }
3365 
3366 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3367 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3368 
3369 void ib_drain_rq(struct ib_qp *qp);
3370 void ib_drain_sq(struct ib_qp *qp);
3371 void ib_drain_qp(struct ib_qp *qp);
3372 
3373 int ib_resolve_eth_dmac(struct ib_device *device,
3374 			struct ib_ah_attr *ah_attr);
3375 #endif /* IB_VERBS_H */
3376