xref: /freebsd/sys/ofed/include/rdma/ib_verbs.h (revision 5def4c47d4bd90b209b9b4a4ba9faec15846d8fd)
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 ifnet	*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 	u32		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_2X	= 16,
402 	IB_WIDTH_4X	= 2,
403 	IB_WIDTH_8X	= 4,
404 	IB_WIDTH_12X	= 8
405 };
406 
407 static inline int ib_width_enum_to_int(enum ib_port_width width)
408 {
409 	switch (width) {
410 	case IB_WIDTH_1X:  return  1;
411 	case IB_WIDTH_2X:  return  2;
412 	case IB_WIDTH_4X:  return  4;
413 	case IB_WIDTH_8X:  return  8;
414 	case IB_WIDTH_12X: return 12;
415 	default: 	  return -1;
416 	}
417 }
418 
419 enum ib_port_speed {
420 	IB_SPEED_SDR	= 1,
421 	IB_SPEED_DDR	= 2,
422 	IB_SPEED_QDR	= 4,
423 	IB_SPEED_FDR10	= 8,
424 	IB_SPEED_FDR	= 16,
425 	IB_SPEED_EDR	= 32,
426 	IB_SPEED_HDR	= 64
427 };
428 
429 /**
430  * struct rdma_hw_stats
431  * @timestamp - Used by the core code to track when the last update was
432  * @lifespan - Used by the core code to determine how old the counters
433  *   should be before being updated again.  Stored in jiffies, defaults
434  *   to 10 milliseconds, drivers can override the default be specifying
435  *   their own value during their allocation routine.
436  * @name - Array of pointers to static names used for the counters in
437  *   directory.
438  * @num_counters - How many hardware counters there are.  If name is
439  *   shorter than this number, a kernel oops will result.  Driver authors
440  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
441  *   in their code to prevent this.
442  * @value - Array of u64 counters that are accessed by the sysfs code and
443  *   filled in by the drivers get_stats routine
444  */
445 struct rdma_hw_stats {
446 	unsigned long	timestamp;
447 	unsigned long	lifespan;
448 	const char * const *names;
449 	int		num_counters;
450 	u64		value[];
451 };
452 
453 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
454 /**
455  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
456  *   for drivers.
457  * @names - Array of static const char *
458  * @num_counters - How many elements in array
459  * @lifespan - How many milliseconds between updates
460  */
461 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
462 		const char * const *names, int num_counters,
463 		unsigned long lifespan)
464 {
465 	struct rdma_hw_stats *stats;
466 
467 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
468 			GFP_KERNEL);
469 	if (!stats)
470 		return NULL;
471 	stats->names = names;
472 	stats->num_counters = num_counters;
473 	stats->lifespan = msecs_to_jiffies(lifespan);
474 
475 	return stats;
476 }
477 
478 
479 /* Define bits for the various functionality this port needs to be supported by
480  * the core.
481  */
482 /* Management                           0x00000FFF */
483 #define RDMA_CORE_CAP_IB_MAD            0x00000001
484 #define RDMA_CORE_CAP_IB_SMI            0x00000002
485 #define RDMA_CORE_CAP_IB_CM             0x00000004
486 #define RDMA_CORE_CAP_IW_CM             0x00000008
487 #define RDMA_CORE_CAP_IB_SA             0x00000010
488 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
489 
490 /* Address format                       0x000FF000 */
491 #define RDMA_CORE_CAP_AF_IB             0x00001000
492 #define RDMA_CORE_CAP_ETH_AH            0x00002000
493 
494 /* Protocol                             0xFFF00000 */
495 #define RDMA_CORE_CAP_PROT_IB           0x00100000
496 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
497 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
498 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
499 
500 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
501 					| RDMA_CORE_CAP_IB_MAD \
502 					| RDMA_CORE_CAP_IB_SMI \
503 					| RDMA_CORE_CAP_IB_CM  \
504 					| RDMA_CORE_CAP_IB_SA  \
505 					| RDMA_CORE_CAP_AF_IB)
506 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
507 					| RDMA_CORE_CAP_IB_MAD  \
508 					| RDMA_CORE_CAP_IB_CM   \
509 					| RDMA_CORE_CAP_AF_IB   \
510 					| RDMA_CORE_CAP_ETH_AH)
511 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
512 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
513 					| RDMA_CORE_CAP_IB_MAD  \
514 					| RDMA_CORE_CAP_IB_CM   \
515 					| RDMA_CORE_CAP_AF_IB   \
516 					| RDMA_CORE_CAP_ETH_AH)
517 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
518 					| RDMA_CORE_CAP_IW_CM)
519 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
520 					| RDMA_CORE_CAP_OPA_MAD)
521 
522 struct ib_port_attr {
523 	u64			subnet_prefix;
524 	enum ib_port_state	state;
525 	enum ib_mtu		max_mtu;
526 	enum ib_mtu		active_mtu;
527 	int			gid_tbl_len;
528 	u32			port_cap_flags;
529 	u32			max_msg_sz;
530 	u32			bad_pkey_cntr;
531 	u32			qkey_viol_cntr;
532 	u16			pkey_tbl_len;
533 	u16			lid;
534 	u16			sm_lid;
535 	u8			lmc;
536 	u8			max_vl_num;
537 	u8			sm_sl;
538 	u8			subnet_timeout;
539 	u8			init_type_reply;
540 	u8			active_width;
541 	u8			active_speed;
542 	u8                      phys_state;
543 	bool			grh_required;
544 };
545 
546 enum ib_device_modify_flags {
547 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
548 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
549 };
550 
551 #define IB_DEVICE_NODE_DESC_MAX 64
552 
553 struct ib_device_modify {
554 	u64	sys_image_guid;
555 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
556 };
557 
558 enum ib_port_modify_flags {
559 	IB_PORT_SHUTDOWN		= 1,
560 	IB_PORT_INIT_TYPE		= (1<<2),
561 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
562 };
563 
564 struct ib_port_modify {
565 	u32	set_port_cap_mask;
566 	u32	clr_port_cap_mask;
567 	u8	init_type;
568 };
569 
570 enum ib_event_type {
571 	IB_EVENT_CQ_ERR,
572 	IB_EVENT_QP_FATAL,
573 	IB_EVENT_QP_REQ_ERR,
574 	IB_EVENT_QP_ACCESS_ERR,
575 	IB_EVENT_COMM_EST,
576 	IB_EVENT_SQ_DRAINED,
577 	IB_EVENT_PATH_MIG,
578 	IB_EVENT_PATH_MIG_ERR,
579 	IB_EVENT_DEVICE_FATAL,
580 	IB_EVENT_PORT_ACTIVE,
581 	IB_EVENT_PORT_ERR,
582 	IB_EVENT_LID_CHANGE,
583 	IB_EVENT_PKEY_CHANGE,
584 	IB_EVENT_SM_CHANGE,
585 	IB_EVENT_SRQ_ERR,
586 	IB_EVENT_SRQ_LIMIT_REACHED,
587 	IB_EVENT_QP_LAST_WQE_REACHED,
588 	IB_EVENT_CLIENT_REREGISTER,
589 	IB_EVENT_GID_CHANGE,
590 	IB_EVENT_WQ_FATAL,
591 };
592 
593 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
594 
595 struct ib_event {
596 	struct ib_device	*device;
597 	union {
598 		struct ib_cq	*cq;
599 		struct ib_qp	*qp;
600 		struct ib_srq	*srq;
601 		struct ib_wq	*wq;
602 		u8		port_num;
603 	} element;
604 	enum ib_event_type	event;
605 };
606 
607 struct ib_event_handler {
608 	struct ib_device *device;
609 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
610 	struct list_head  list;
611 };
612 
613 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
614 	do {							\
615 		(_ptr)->device  = _device;			\
616 		(_ptr)->handler = _handler;			\
617 		INIT_LIST_HEAD(&(_ptr)->list);			\
618 	} while (0)
619 
620 struct ib_global_route {
621 	union ib_gid	dgid;
622 	u32		flow_label;
623 	u8		sgid_index;
624 	u8		hop_limit;
625 	u8		traffic_class;
626 };
627 
628 struct ib_grh {
629 	__be32		version_tclass_flow;
630 	__be16		paylen;
631 	u8		next_hdr;
632 	u8		hop_limit;
633 	union ib_gid	sgid;
634 	union ib_gid	dgid;
635 };
636 
637 union rdma_network_hdr {
638 	struct ib_grh ibgrh;
639 	struct {
640 		/* The IB spec states that if it's IPv4, the header
641 		 * is located in the last 20 bytes of the header.
642 		 */
643 		u8		reserved[20];
644 		struct ip	roce4grh;
645 	};
646 };
647 
648 enum {
649 	IB_MULTICAST_QPN = 0xffffff
650 };
651 
652 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
653 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
654 
655 enum ib_ah_flags {
656 	IB_AH_GRH	= 1
657 };
658 
659 enum ib_rate {
660 	IB_RATE_PORT_CURRENT = 0,
661 	IB_RATE_2_5_GBPS = 2,
662 	IB_RATE_5_GBPS   = 5,
663 	IB_RATE_10_GBPS  = 3,
664 	IB_RATE_20_GBPS  = 6,
665 	IB_RATE_30_GBPS  = 4,
666 	IB_RATE_40_GBPS  = 7,
667 	IB_RATE_60_GBPS  = 8,
668 	IB_RATE_80_GBPS  = 9,
669 	IB_RATE_120_GBPS = 10,
670 	IB_RATE_14_GBPS  = 11,
671 	IB_RATE_56_GBPS  = 12,
672 	IB_RATE_112_GBPS = 13,
673 	IB_RATE_168_GBPS = 14,
674 	IB_RATE_25_GBPS  = 15,
675 	IB_RATE_100_GBPS = 16,
676 	IB_RATE_200_GBPS = 17,
677 	IB_RATE_300_GBPS = 18,
678 	IB_RATE_28_GBPS  = 19,
679 	IB_RATE_50_GBPS  = 20,
680 	IB_RATE_400_GBPS = 21,
681 	IB_RATE_600_GBPS = 22,
682 };
683 
684 /**
685  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
686  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
687  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
688  * @rate: rate to convert.
689  */
690 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
691 
692 /**
693  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
694  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
695  * @rate: rate to convert.
696  */
697 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
698 
699 
700 /**
701  * enum ib_mr_type - memory region type
702  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
703  *                            normal registration
704  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
705  *                            signature operations (data-integrity
706  *                            capable regions)
707  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
708  *                            register any arbitrary sg lists (without
709  *                            the normal mr constraints - see
710  *                            ib_map_mr_sg)
711  */
712 enum ib_mr_type {
713 	IB_MR_TYPE_MEM_REG,
714 	IB_MR_TYPE_SIGNATURE,
715 	IB_MR_TYPE_SG_GAPS,
716 };
717 
718 /**
719  * Signature types
720  * IB_SIG_TYPE_NONE: Unprotected.
721  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
722  */
723 enum ib_signature_type {
724 	IB_SIG_TYPE_NONE,
725 	IB_SIG_TYPE_T10_DIF,
726 };
727 
728 /**
729  * Signature T10-DIF block-guard types
730  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
731  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
732  */
733 enum ib_t10_dif_bg_type {
734 	IB_T10DIF_CRC,
735 	IB_T10DIF_CSUM
736 };
737 
738 /**
739  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
740  *     domain.
741  * @bg_type: T10-DIF block guard type (CRC|CSUM)
742  * @pi_interval: protection information interval.
743  * @bg: seed of guard computation.
744  * @app_tag: application tag of guard block
745  * @ref_tag: initial guard block reference tag.
746  * @ref_remap: Indicate wethear the reftag increments each block
747  * @app_escape: Indicate to skip block check if apptag=0xffff
748  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
749  * @apptag_check_mask: check bitmask of application tag.
750  */
751 struct ib_t10_dif_domain {
752 	enum ib_t10_dif_bg_type bg_type;
753 	u16			pi_interval;
754 	u16			bg;
755 	u16			app_tag;
756 	u32			ref_tag;
757 	bool			ref_remap;
758 	bool			app_escape;
759 	bool			ref_escape;
760 	u16			apptag_check_mask;
761 };
762 
763 /**
764  * struct ib_sig_domain - Parameters for signature domain
765  * @sig_type: specific signauture type
766  * @sig: union of all signature domain attributes that may
767  *     be used to set domain layout.
768  */
769 struct ib_sig_domain {
770 	enum ib_signature_type sig_type;
771 	union {
772 		struct ib_t10_dif_domain dif;
773 	} sig;
774 };
775 
776 /**
777  * struct ib_sig_attrs - Parameters for signature handover operation
778  * @check_mask: bitmask for signature byte check (8 bytes)
779  * @mem: memory domain layout desciptor.
780  * @wire: wire domain layout desciptor.
781  */
782 struct ib_sig_attrs {
783 	u8			check_mask;
784 	struct ib_sig_domain	mem;
785 	struct ib_sig_domain	wire;
786 };
787 
788 enum ib_sig_err_type {
789 	IB_SIG_BAD_GUARD,
790 	IB_SIG_BAD_REFTAG,
791 	IB_SIG_BAD_APPTAG,
792 };
793 
794 /**
795  * struct ib_sig_err - signature error descriptor
796  */
797 struct ib_sig_err {
798 	enum ib_sig_err_type	err_type;
799 	u32			expected;
800 	u32			actual;
801 	u64			sig_err_offset;
802 	u32			key;
803 };
804 
805 enum ib_mr_status_check {
806 	IB_MR_CHECK_SIG_STATUS = 1,
807 };
808 
809 /**
810  * struct ib_mr_status - Memory region status container
811  *
812  * @fail_status: Bitmask of MR checks status. For each
813  *     failed check a corresponding status bit is set.
814  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
815  *     failure.
816  */
817 struct ib_mr_status {
818 	u32		    fail_status;
819 	struct ib_sig_err   sig_err;
820 };
821 
822 /**
823  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
824  * enum.
825  * @mult: multiple to convert.
826  */
827 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
828 
829 struct ib_ah_attr {
830 	struct ib_global_route	grh;
831 	u16			dlid;
832 	u8			sl;
833 	u8			src_path_bits;
834 	u8			static_rate;
835 	u8			ah_flags;
836 	u8			port_num;
837 	u8			dmac[ETH_ALEN];
838 };
839 
840 enum ib_wc_status {
841 	IB_WC_SUCCESS,
842 	IB_WC_LOC_LEN_ERR,
843 	IB_WC_LOC_QP_OP_ERR,
844 	IB_WC_LOC_EEC_OP_ERR,
845 	IB_WC_LOC_PROT_ERR,
846 	IB_WC_WR_FLUSH_ERR,
847 	IB_WC_MW_BIND_ERR,
848 	IB_WC_BAD_RESP_ERR,
849 	IB_WC_LOC_ACCESS_ERR,
850 	IB_WC_REM_INV_REQ_ERR,
851 	IB_WC_REM_ACCESS_ERR,
852 	IB_WC_REM_OP_ERR,
853 	IB_WC_RETRY_EXC_ERR,
854 	IB_WC_RNR_RETRY_EXC_ERR,
855 	IB_WC_LOC_RDD_VIOL_ERR,
856 	IB_WC_REM_INV_RD_REQ_ERR,
857 	IB_WC_REM_ABORT_ERR,
858 	IB_WC_INV_EECN_ERR,
859 	IB_WC_INV_EEC_STATE_ERR,
860 	IB_WC_FATAL_ERR,
861 	IB_WC_RESP_TIMEOUT_ERR,
862 	IB_WC_GENERAL_ERR
863 };
864 
865 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
866 
867 enum ib_wc_opcode {
868 	IB_WC_SEND,
869 	IB_WC_RDMA_WRITE,
870 	IB_WC_RDMA_READ,
871 	IB_WC_COMP_SWAP,
872 	IB_WC_FETCH_ADD,
873 	IB_WC_LSO,
874 	IB_WC_LOCAL_INV,
875 	IB_WC_REG_MR,
876 	IB_WC_MASKED_COMP_SWAP,
877 	IB_WC_MASKED_FETCH_ADD,
878 /*
879  * Set value of IB_WC_RECV so consumers can test if a completion is a
880  * receive by testing (opcode & IB_WC_RECV).
881  */
882 	IB_WC_RECV			= 1 << 7,
883 	IB_WC_RECV_RDMA_WITH_IMM,
884 	IB_WC_DUMMY = -1,	/* force enum signed */
885 };
886 
887 enum ib_wc_flags {
888 	IB_WC_GRH		= 1,
889 	IB_WC_WITH_IMM		= (1<<1),
890 	IB_WC_WITH_INVALIDATE	= (1<<2),
891 	IB_WC_IP_CSUM_OK	= (1<<3),
892 	IB_WC_WITH_SMAC		= (1<<4),
893 	IB_WC_WITH_VLAN		= (1<<5),
894 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
895 };
896 
897 struct ib_wc {
898 	union {
899 		u64		wr_id;
900 		struct ib_cqe	*wr_cqe;
901 	};
902 	enum ib_wc_status	status;
903 	enum ib_wc_opcode	opcode;
904 	u32			vendor_err;
905 	u32			byte_len;
906 	struct ib_qp	       *qp;
907 	union {
908 		__be32		imm_data;
909 		u32		invalidate_rkey;
910 	} ex;
911 	u32			src_qp;
912 	int			wc_flags;
913 	u16			pkey_index;
914 	u16			slid;
915 	u8			sl;
916 	u8			dlid_path_bits;
917 	u8			port_num;	/* valid only for DR SMPs on switches */
918 	u8			smac[ETH_ALEN];
919 	u16			vlan_id;
920 	u8			network_hdr_type;
921 };
922 
923 enum ib_cq_notify_flags {
924 	IB_CQ_SOLICITED			= 1 << 0,
925 	IB_CQ_NEXT_COMP			= 1 << 1,
926 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
927 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
928 };
929 
930 enum ib_srq_type {
931 	IB_SRQT_BASIC,
932 	IB_SRQT_XRC
933 };
934 
935 enum ib_srq_attr_mask {
936 	IB_SRQ_MAX_WR	= 1 << 0,
937 	IB_SRQ_LIMIT	= 1 << 1,
938 };
939 
940 struct ib_srq_attr {
941 	u32	max_wr;
942 	u32	max_sge;
943 	u32	srq_limit;
944 };
945 
946 struct ib_srq_init_attr {
947 	void		      (*event_handler)(struct ib_event *, void *);
948 	void		       *srq_context;
949 	struct ib_srq_attr	attr;
950 	enum ib_srq_type	srq_type;
951 
952 	union {
953 		struct {
954 			struct ib_xrcd *xrcd;
955 			struct ib_cq   *cq;
956 		} xrc;
957 	} ext;
958 };
959 
960 struct ib_qp_cap {
961 	u32	max_send_wr;
962 	u32	max_recv_wr;
963 	u32	max_send_sge;
964 	u32	max_recv_sge;
965 	u32	max_inline_data;
966 
967 	/*
968 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
969 	 * ib_create_qp() will calculate the right amount of neededed WRs
970 	 * and MRs based on this.
971 	 */
972 	u32	max_rdma_ctxs;
973 };
974 
975 enum ib_sig_type {
976 	IB_SIGNAL_ALL_WR,
977 	IB_SIGNAL_REQ_WR
978 };
979 
980 enum ib_qp_type {
981 	/*
982 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
983 	 * here (and in that order) since the MAD layer uses them as
984 	 * indices into a 2-entry table.
985 	 */
986 	IB_QPT_SMI,
987 	IB_QPT_GSI,
988 
989 	IB_QPT_RC,
990 	IB_QPT_UC,
991 	IB_QPT_UD,
992 	IB_QPT_RAW_IPV6,
993 	IB_QPT_RAW_ETHERTYPE,
994 	IB_QPT_RAW_PACKET = 8,
995 	IB_QPT_XRC_INI = 9,
996 	IB_QPT_XRC_TGT,
997 	IB_QPT_MAX,
998 	/* Reserve a range for qp types internal to the low level driver.
999 	 * These qp types will not be visible at the IB core layer, so the
1000 	 * IB_QPT_MAX usages should not be affected in the core layer
1001 	 */
1002 	IB_QPT_RESERVED1 = 0x1000,
1003 	IB_QPT_RESERVED2,
1004 	IB_QPT_RESERVED3,
1005 	IB_QPT_RESERVED4,
1006 	IB_QPT_RESERVED5,
1007 	IB_QPT_RESERVED6,
1008 	IB_QPT_RESERVED7,
1009 	IB_QPT_RESERVED8,
1010 	IB_QPT_RESERVED9,
1011 	IB_QPT_RESERVED10,
1012 };
1013 
1014 enum ib_qp_create_flags {
1015 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1016 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1017 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1018 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1019 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1020 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1021 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1022 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1023 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1024 	/* reserve bits 26-31 for low level drivers' internal use */
1025 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1026 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1027 };
1028 
1029 /*
1030  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1031  * callback to destroy the passed in QP.
1032  */
1033 
1034 struct ib_qp_init_attr {
1035 	void                  (*event_handler)(struct ib_event *, void *);
1036 	void		       *qp_context;
1037 	struct ib_cq	       *send_cq;
1038 	struct ib_cq	       *recv_cq;
1039 	struct ib_srq	       *srq;
1040 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1041 	struct ib_qp_cap	cap;
1042 	enum ib_sig_type	sq_sig_type;
1043 	enum ib_qp_type		qp_type;
1044 	enum ib_qp_create_flags	create_flags;
1045 
1046 	/*
1047 	 * Only needed for special QP types, or when using the RW API.
1048 	 */
1049 	u8			port_num;
1050 	struct ib_rwq_ind_table *rwq_ind_tbl;
1051 };
1052 
1053 struct ib_qp_open_attr {
1054 	void                  (*event_handler)(struct ib_event *, void *);
1055 	void		       *qp_context;
1056 	u32			qp_num;
1057 	enum ib_qp_type		qp_type;
1058 };
1059 
1060 enum ib_rnr_timeout {
1061 	IB_RNR_TIMER_655_36 =  0,
1062 	IB_RNR_TIMER_000_01 =  1,
1063 	IB_RNR_TIMER_000_02 =  2,
1064 	IB_RNR_TIMER_000_03 =  3,
1065 	IB_RNR_TIMER_000_04 =  4,
1066 	IB_RNR_TIMER_000_06 =  5,
1067 	IB_RNR_TIMER_000_08 =  6,
1068 	IB_RNR_TIMER_000_12 =  7,
1069 	IB_RNR_TIMER_000_16 =  8,
1070 	IB_RNR_TIMER_000_24 =  9,
1071 	IB_RNR_TIMER_000_32 = 10,
1072 	IB_RNR_TIMER_000_48 = 11,
1073 	IB_RNR_TIMER_000_64 = 12,
1074 	IB_RNR_TIMER_000_96 = 13,
1075 	IB_RNR_TIMER_001_28 = 14,
1076 	IB_RNR_TIMER_001_92 = 15,
1077 	IB_RNR_TIMER_002_56 = 16,
1078 	IB_RNR_TIMER_003_84 = 17,
1079 	IB_RNR_TIMER_005_12 = 18,
1080 	IB_RNR_TIMER_007_68 = 19,
1081 	IB_RNR_TIMER_010_24 = 20,
1082 	IB_RNR_TIMER_015_36 = 21,
1083 	IB_RNR_TIMER_020_48 = 22,
1084 	IB_RNR_TIMER_030_72 = 23,
1085 	IB_RNR_TIMER_040_96 = 24,
1086 	IB_RNR_TIMER_061_44 = 25,
1087 	IB_RNR_TIMER_081_92 = 26,
1088 	IB_RNR_TIMER_122_88 = 27,
1089 	IB_RNR_TIMER_163_84 = 28,
1090 	IB_RNR_TIMER_245_76 = 29,
1091 	IB_RNR_TIMER_327_68 = 30,
1092 	IB_RNR_TIMER_491_52 = 31
1093 };
1094 
1095 enum ib_qp_attr_mask {
1096 	IB_QP_STATE			= 1,
1097 	IB_QP_CUR_STATE			= (1<<1),
1098 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1099 	IB_QP_ACCESS_FLAGS		= (1<<3),
1100 	IB_QP_PKEY_INDEX		= (1<<4),
1101 	IB_QP_PORT			= (1<<5),
1102 	IB_QP_QKEY			= (1<<6),
1103 	IB_QP_AV			= (1<<7),
1104 	IB_QP_PATH_MTU			= (1<<8),
1105 	IB_QP_TIMEOUT			= (1<<9),
1106 	IB_QP_RETRY_CNT			= (1<<10),
1107 	IB_QP_RNR_RETRY			= (1<<11),
1108 	IB_QP_RQ_PSN			= (1<<12),
1109 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1110 	IB_QP_ALT_PATH			= (1<<14),
1111 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1112 	IB_QP_SQ_PSN			= (1<<16),
1113 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1114 	IB_QP_PATH_MIG_STATE		= (1<<18),
1115 	IB_QP_CAP			= (1<<19),
1116 	IB_QP_DEST_QPN			= (1<<20),
1117 	IB_QP_RESERVED1			= (1<<21),
1118 	IB_QP_RESERVED2			= (1<<22),
1119 	IB_QP_RESERVED3			= (1<<23),
1120 	IB_QP_RESERVED4			= (1<<24),
1121 };
1122 
1123 enum ib_qp_state {
1124 	IB_QPS_RESET,
1125 	IB_QPS_INIT,
1126 	IB_QPS_RTR,
1127 	IB_QPS_RTS,
1128 	IB_QPS_SQD,
1129 	IB_QPS_SQE,
1130 	IB_QPS_ERR,
1131 	IB_QPS_DUMMY = -1,	/* force enum signed */
1132 };
1133 
1134 enum ib_mig_state {
1135 	IB_MIG_MIGRATED,
1136 	IB_MIG_REARM,
1137 	IB_MIG_ARMED
1138 };
1139 
1140 enum ib_mw_type {
1141 	IB_MW_TYPE_1 = 1,
1142 	IB_MW_TYPE_2 = 2
1143 };
1144 
1145 struct ib_qp_attr {
1146 	enum ib_qp_state	qp_state;
1147 	enum ib_qp_state	cur_qp_state;
1148 	enum ib_mtu		path_mtu;
1149 	enum ib_mig_state	path_mig_state;
1150 	u32			qkey;
1151 	u32			rq_psn;
1152 	u32			sq_psn;
1153 	u32			dest_qp_num;
1154 	int			qp_access_flags;
1155 	struct ib_qp_cap	cap;
1156 	struct ib_ah_attr	ah_attr;
1157 	struct ib_ah_attr	alt_ah_attr;
1158 	u16			pkey_index;
1159 	u16			alt_pkey_index;
1160 	u8			en_sqd_async_notify;
1161 	u8			sq_draining;
1162 	u8			max_rd_atomic;
1163 	u8			max_dest_rd_atomic;
1164 	u8			min_rnr_timer;
1165 	u8			port_num;
1166 	u8			timeout;
1167 	u8			retry_cnt;
1168 	u8			rnr_retry;
1169 	u8			alt_port_num;
1170 	u8			alt_timeout;
1171 };
1172 
1173 enum ib_wr_opcode {
1174 	IB_WR_RDMA_WRITE,
1175 	IB_WR_RDMA_WRITE_WITH_IMM,
1176 	IB_WR_SEND,
1177 	IB_WR_SEND_WITH_IMM,
1178 	IB_WR_RDMA_READ,
1179 	IB_WR_ATOMIC_CMP_AND_SWP,
1180 	IB_WR_ATOMIC_FETCH_AND_ADD,
1181 	IB_WR_LSO,
1182 	IB_WR_SEND_WITH_INV,
1183 	IB_WR_RDMA_READ_WITH_INV,
1184 	IB_WR_LOCAL_INV,
1185 	IB_WR_REG_MR,
1186 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1187 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1188 	IB_WR_REG_SIG_MR,
1189 	/* reserve values for low level drivers' internal use.
1190 	 * These values will not be used at all in the ib core layer.
1191 	 */
1192 	IB_WR_RESERVED1 = 0xf0,
1193 	IB_WR_RESERVED2,
1194 	IB_WR_RESERVED3,
1195 	IB_WR_RESERVED4,
1196 	IB_WR_RESERVED5,
1197 	IB_WR_RESERVED6,
1198 	IB_WR_RESERVED7,
1199 	IB_WR_RESERVED8,
1200 	IB_WR_RESERVED9,
1201 	IB_WR_RESERVED10,
1202 	IB_WR_DUMMY = -1,	/* force enum signed */
1203 };
1204 
1205 enum ib_send_flags {
1206 	IB_SEND_FENCE		= 1,
1207 	IB_SEND_SIGNALED	= (1<<1),
1208 	IB_SEND_SOLICITED	= (1<<2),
1209 	IB_SEND_INLINE		= (1<<3),
1210 	IB_SEND_IP_CSUM		= (1<<4),
1211 
1212 	/* reserve bits 26-31 for low level drivers' internal use */
1213 	IB_SEND_RESERVED_START	= (1 << 26),
1214 	IB_SEND_RESERVED_END	= (1 << 31),
1215 };
1216 
1217 struct ib_sge {
1218 	u64	addr;
1219 	u32	length;
1220 	u32	lkey;
1221 };
1222 
1223 struct ib_cqe {
1224 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1225 };
1226 
1227 struct ib_send_wr {
1228 	struct ib_send_wr      *next;
1229 	union {
1230 		u64		wr_id;
1231 		struct ib_cqe	*wr_cqe;
1232 	};
1233 	struct ib_sge	       *sg_list;
1234 	int			num_sge;
1235 	enum ib_wr_opcode	opcode;
1236 	int			send_flags;
1237 	union {
1238 		__be32		imm_data;
1239 		u32		invalidate_rkey;
1240 	} ex;
1241 };
1242 
1243 struct ib_rdma_wr {
1244 	struct ib_send_wr	wr;
1245 	u64			remote_addr;
1246 	u32			rkey;
1247 };
1248 
1249 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1250 {
1251 	return container_of(wr, struct ib_rdma_wr, wr);
1252 }
1253 
1254 struct ib_atomic_wr {
1255 	struct ib_send_wr	wr;
1256 	u64			remote_addr;
1257 	u64			compare_add;
1258 	u64			swap;
1259 	u64			compare_add_mask;
1260 	u64			swap_mask;
1261 	u32			rkey;
1262 };
1263 
1264 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1265 {
1266 	return container_of(wr, struct ib_atomic_wr, wr);
1267 }
1268 
1269 struct ib_ud_wr {
1270 	struct ib_send_wr	wr;
1271 	struct ib_ah		*ah;
1272 	void			*header;
1273 	int			hlen;
1274 	int			mss;
1275 	u32			remote_qpn;
1276 	u32			remote_qkey;
1277 	u16			pkey_index; /* valid for GSI only */
1278 	u8			port_num;   /* valid for DR SMPs on switch only */
1279 };
1280 
1281 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1282 {
1283 	return container_of(wr, struct ib_ud_wr, wr);
1284 }
1285 
1286 struct ib_reg_wr {
1287 	struct ib_send_wr	wr;
1288 	struct ib_mr		*mr;
1289 	u32			key;
1290 	int			access;
1291 };
1292 
1293 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1294 {
1295 	return container_of(wr, struct ib_reg_wr, wr);
1296 }
1297 
1298 struct ib_sig_handover_wr {
1299 	struct ib_send_wr	wr;
1300 	struct ib_sig_attrs    *sig_attrs;
1301 	struct ib_mr	       *sig_mr;
1302 	int			access_flags;
1303 	struct ib_sge	       *prot;
1304 };
1305 
1306 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1307 {
1308 	return container_of(wr, struct ib_sig_handover_wr, wr);
1309 }
1310 
1311 struct ib_recv_wr {
1312 	struct ib_recv_wr      *next;
1313 	union {
1314 		u64		wr_id;
1315 		struct ib_cqe	*wr_cqe;
1316 	};
1317 	struct ib_sge	       *sg_list;
1318 	int			num_sge;
1319 };
1320 
1321 enum ib_access_flags {
1322 	IB_ACCESS_LOCAL_WRITE	= 1,
1323 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1324 	IB_ACCESS_REMOTE_READ	= (1<<2),
1325 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1326 	IB_ACCESS_MW_BIND	= (1<<4),
1327 	IB_ZERO_BASED		= (1<<5),
1328 	IB_ACCESS_ON_DEMAND     = (1<<6),
1329 };
1330 
1331 /*
1332  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1333  * are hidden here instead of a uapi header!
1334  */
1335 enum ib_mr_rereg_flags {
1336 	IB_MR_REREG_TRANS	= 1,
1337 	IB_MR_REREG_PD		= (1<<1),
1338 	IB_MR_REREG_ACCESS	= (1<<2),
1339 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1340 };
1341 
1342 struct ib_fmr_attr {
1343 	int	max_pages;
1344 	int	max_maps;
1345 	u8	page_shift;
1346 };
1347 
1348 struct ib_umem;
1349 
1350 struct ib_ucontext {
1351 	struct ib_device       *device;
1352 	struct list_head	pd_list;
1353 	struct list_head	mr_list;
1354 	struct list_head	mw_list;
1355 	struct list_head	cq_list;
1356 	struct list_head	qp_list;
1357 	struct list_head	srq_list;
1358 	struct list_head	ah_list;
1359 	struct list_head	xrcd_list;
1360 	struct list_head	rule_list;
1361 	struct list_head	wq_list;
1362 	struct list_head	rwq_ind_tbl_list;
1363 	int			closing;
1364 
1365 	pid_t			tgid;
1366 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1367 	struct rb_root      umem_tree;
1368 	/*
1369 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1370 	 * mmu notifiers registration.
1371 	 */
1372 	struct rw_semaphore	umem_rwsem;
1373 	void (*invalidate_range)(struct ib_umem *umem,
1374 				 unsigned long start, unsigned long end);
1375 
1376 	struct mmu_notifier	mn;
1377 	atomic_t		notifier_count;
1378 	/* A list of umems that don't have private mmu notifier counters yet. */
1379 	struct list_head	no_private_counters;
1380 	int                     odp_mrs_count;
1381 #endif
1382 };
1383 
1384 struct ib_uobject {
1385 	u64			user_handle;	/* handle given to us by userspace */
1386 	struct ib_ucontext     *context;	/* associated user context */
1387 	void		       *object;		/* containing object */
1388 	struct list_head	list;		/* link to context's list */
1389 	int			id;		/* index into kernel idr */
1390 	struct kref		ref;
1391 	struct rw_semaphore	mutex;		/* protects .live */
1392 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1393 	int			live;
1394 };
1395 
1396 struct ib_udata {
1397 	const void __user *inbuf;
1398 	void __user *outbuf;
1399 	size_t       inlen;
1400 	size_t       outlen;
1401 };
1402 
1403 struct ib_pd {
1404 	u32			local_dma_lkey;
1405 	u32			flags;
1406 	struct ib_device       *device;
1407 	struct ib_uobject      *uobject;
1408 	atomic_t          	usecnt; /* count all resources */
1409 
1410 	u32			unsafe_global_rkey;
1411 
1412 	/*
1413 	 * Implementation details of the RDMA core, don't use in drivers:
1414 	 */
1415 	struct ib_mr	       *__internal_mr;
1416 };
1417 
1418 struct ib_xrcd {
1419 	struct ib_device       *device;
1420 	atomic_t		usecnt; /* count all exposed resources */
1421 	struct inode	       *inode;
1422 
1423 	struct mutex		tgt_qp_mutex;
1424 	struct list_head	tgt_qp_list;
1425 };
1426 
1427 struct ib_ah {
1428 	struct ib_device	*device;
1429 	struct ib_pd		*pd;
1430 	struct ib_uobject	*uobject;
1431 };
1432 
1433 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1434 
1435 enum ib_poll_context {
1436 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1437 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1438 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1439 };
1440 
1441 struct ib_cq {
1442 	struct ib_device       *device;
1443 	struct ib_uobject      *uobject;
1444 	ib_comp_handler   	comp_handler;
1445 	void                  (*event_handler)(struct ib_event *, void *);
1446 	void                   *cq_context;
1447 	int               	cqe;
1448 	atomic_t          	usecnt; /* count number of work queues */
1449 	enum ib_poll_context	poll_ctx;
1450 	struct work_struct	work;
1451 };
1452 
1453 struct ib_srq {
1454 	struct ib_device       *device;
1455 	struct ib_pd	       *pd;
1456 	struct ib_uobject      *uobject;
1457 	void		      (*event_handler)(struct ib_event *, void *);
1458 	void		       *srq_context;
1459 	enum ib_srq_type	srq_type;
1460 	atomic_t		usecnt;
1461 
1462 	union {
1463 		struct {
1464 			struct ib_xrcd *xrcd;
1465 			struct ib_cq   *cq;
1466 			u32		srq_num;
1467 		} xrc;
1468 	} ext;
1469 };
1470 
1471 enum ib_wq_type {
1472 	IB_WQT_RQ
1473 };
1474 
1475 enum ib_wq_state {
1476 	IB_WQS_RESET,
1477 	IB_WQS_RDY,
1478 	IB_WQS_ERR
1479 };
1480 
1481 struct ib_wq {
1482 	struct ib_device       *device;
1483 	struct ib_uobject      *uobject;
1484 	void		    *wq_context;
1485 	void		    (*event_handler)(struct ib_event *, void *);
1486 	struct ib_pd	       *pd;
1487 	struct ib_cq	       *cq;
1488 	u32		wq_num;
1489 	enum ib_wq_state       state;
1490 	enum ib_wq_type	wq_type;
1491 	atomic_t		usecnt;
1492 };
1493 
1494 struct ib_wq_init_attr {
1495 	void		       *wq_context;
1496 	enum ib_wq_type	wq_type;
1497 	u32		max_wr;
1498 	u32		max_sge;
1499 	struct	ib_cq	       *cq;
1500 	void		    (*event_handler)(struct ib_event *, void *);
1501 };
1502 
1503 enum ib_wq_attr_mask {
1504 	IB_WQ_STATE	= 1 << 0,
1505 	IB_WQ_CUR_STATE	= 1 << 1,
1506 };
1507 
1508 struct ib_wq_attr {
1509 	enum	ib_wq_state	wq_state;
1510 	enum	ib_wq_state	curr_wq_state;
1511 };
1512 
1513 struct ib_rwq_ind_table {
1514 	struct ib_device	*device;
1515 	struct ib_uobject      *uobject;
1516 	atomic_t		usecnt;
1517 	u32		ind_tbl_num;
1518 	u32		log_ind_tbl_size;
1519 	struct ib_wq	**ind_tbl;
1520 };
1521 
1522 struct ib_rwq_ind_table_init_attr {
1523 	u32		log_ind_tbl_size;
1524 	/* Each entry is a pointer to Receive Work Queue */
1525 	struct ib_wq	**ind_tbl;
1526 };
1527 
1528 /*
1529  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1530  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1531  */
1532 struct ib_qp {
1533 	struct ib_device       *device;
1534 	struct ib_pd	       *pd;
1535 	struct ib_cq	       *send_cq;
1536 	struct ib_cq	       *recv_cq;
1537 	spinlock_t		mr_lock;
1538 	struct ib_srq	       *srq;
1539 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1540 	struct list_head	xrcd_list;
1541 
1542 	/* count times opened, mcast attaches, flow attaches */
1543 	atomic_t		usecnt;
1544 	struct list_head	open_list;
1545 	struct ib_qp           *real_qp;
1546 	struct ib_uobject      *uobject;
1547 	void                  (*event_handler)(struct ib_event *, void *);
1548 	void		       *qp_context;
1549 	u32			qp_num;
1550 	u32			max_write_sge;
1551 	u32			max_read_sge;
1552 	enum ib_qp_type		qp_type;
1553 	struct ib_rwq_ind_table *rwq_ind_tbl;
1554 };
1555 
1556 struct ib_mr {
1557 	struct ib_device  *device;
1558 	struct ib_pd	  *pd;
1559 	u32		   lkey;
1560 	u32		   rkey;
1561 	u64		   iova;
1562 	u64		   length;
1563 	unsigned int	   page_size;
1564 	bool		   need_inval;
1565 	union {
1566 		struct ib_uobject	*uobject;	/* user */
1567 		struct list_head	qp_entry;	/* FR */
1568 	};
1569 };
1570 
1571 struct ib_mw {
1572 	struct ib_device	*device;
1573 	struct ib_pd		*pd;
1574 	struct ib_uobject	*uobject;
1575 	u32			rkey;
1576 	enum ib_mw_type         type;
1577 };
1578 
1579 struct ib_fmr {
1580 	struct ib_device	*device;
1581 	struct ib_pd		*pd;
1582 	struct list_head	list;
1583 	u32			lkey;
1584 	u32			rkey;
1585 };
1586 
1587 /* Supported steering options */
1588 enum ib_flow_attr_type {
1589 	/* steering according to rule specifications */
1590 	IB_FLOW_ATTR_NORMAL		= 0x0,
1591 	/* default unicast and multicast rule -
1592 	 * receive all Eth traffic which isn't steered to any QP
1593 	 */
1594 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1595 	/* default multicast rule -
1596 	 * receive all Eth multicast traffic which isn't steered to any QP
1597 	 */
1598 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1599 	/* sniffer rule - receive all port traffic */
1600 	IB_FLOW_ATTR_SNIFFER		= 0x3
1601 };
1602 
1603 /* Supported steering header types */
1604 enum ib_flow_spec_type {
1605 	/* L2 headers*/
1606 	IB_FLOW_SPEC_ETH	= 0x20,
1607 	IB_FLOW_SPEC_IB		= 0x22,
1608 	/* L3 header*/
1609 	IB_FLOW_SPEC_IPV4	= 0x30,
1610 	IB_FLOW_SPEC_IPV6	= 0x31,
1611 	/* L4 headers*/
1612 	IB_FLOW_SPEC_TCP	= 0x40,
1613 	IB_FLOW_SPEC_UDP	= 0x41
1614 };
1615 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1616 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1617 
1618 /* Flow steering rule priority is set according to it's domain.
1619  * Lower domain value means higher priority.
1620  */
1621 enum ib_flow_domain {
1622 	IB_FLOW_DOMAIN_USER,
1623 	IB_FLOW_DOMAIN_ETHTOOL,
1624 	IB_FLOW_DOMAIN_RFS,
1625 	IB_FLOW_DOMAIN_NIC,
1626 	IB_FLOW_DOMAIN_NUM /* Must be last */
1627 };
1628 
1629 enum ib_flow_flags {
1630 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1631 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1632 };
1633 
1634 struct ib_flow_eth_filter {
1635 	u8	dst_mac[6];
1636 	u8	src_mac[6];
1637 	__be16	ether_type;
1638 	__be16	vlan_tag;
1639 	/* Must be last */
1640 	u8	real_sz[0];
1641 };
1642 
1643 struct ib_flow_spec_eth {
1644 	enum ib_flow_spec_type	  type;
1645 	u16			  size;
1646 	struct ib_flow_eth_filter val;
1647 	struct ib_flow_eth_filter mask;
1648 };
1649 
1650 struct ib_flow_ib_filter {
1651 	__be16 dlid;
1652 	__u8   sl;
1653 	/* Must be last */
1654 	u8	real_sz[0];
1655 };
1656 
1657 struct ib_flow_spec_ib {
1658 	enum ib_flow_spec_type	 type;
1659 	u16			 size;
1660 	struct ib_flow_ib_filter val;
1661 	struct ib_flow_ib_filter mask;
1662 };
1663 
1664 /* IPv4 header flags */
1665 enum ib_ipv4_flags {
1666 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1667 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1668 				    last have this flag set */
1669 };
1670 
1671 struct ib_flow_ipv4_filter {
1672 	__be32	src_ip;
1673 	__be32	dst_ip;
1674 	u8	proto;
1675 	u8	tos;
1676 	u8	ttl;
1677 	u8	flags;
1678 	/* Must be last */
1679 	u8	real_sz[0];
1680 };
1681 
1682 struct ib_flow_spec_ipv4 {
1683 	enum ib_flow_spec_type	   type;
1684 	u16			   size;
1685 	struct ib_flow_ipv4_filter val;
1686 	struct ib_flow_ipv4_filter mask;
1687 };
1688 
1689 struct ib_flow_ipv6_filter {
1690 	u8	src_ip[16];
1691 	u8	dst_ip[16];
1692 	__be32	flow_label;
1693 	u8	next_hdr;
1694 	u8	traffic_class;
1695 	u8	hop_limit;
1696 	/* Must be last */
1697 	u8	real_sz[0];
1698 };
1699 
1700 struct ib_flow_spec_ipv6 {
1701 	enum ib_flow_spec_type	   type;
1702 	u16			   size;
1703 	struct ib_flow_ipv6_filter val;
1704 	struct ib_flow_ipv6_filter mask;
1705 };
1706 
1707 struct ib_flow_tcp_udp_filter {
1708 	__be16	dst_port;
1709 	__be16	src_port;
1710 	/* Must be last */
1711 	u8	real_sz[0];
1712 };
1713 
1714 struct ib_flow_spec_tcp_udp {
1715 	enum ib_flow_spec_type	      type;
1716 	u16			      size;
1717 	struct ib_flow_tcp_udp_filter val;
1718 	struct ib_flow_tcp_udp_filter mask;
1719 };
1720 
1721 union ib_flow_spec {
1722 	struct {
1723 		enum ib_flow_spec_type	type;
1724 		u16			size;
1725 	};
1726 	struct ib_flow_spec_eth		eth;
1727 	struct ib_flow_spec_ib		ib;
1728 	struct ib_flow_spec_ipv4        ipv4;
1729 	struct ib_flow_spec_tcp_udp	tcp_udp;
1730 	struct ib_flow_spec_ipv6        ipv6;
1731 };
1732 
1733 struct ib_flow_attr {
1734 	enum ib_flow_attr_type type;
1735 	u16	     size;
1736 	u16	     priority;
1737 	u32	     flags;
1738 	u8	     num_of_specs;
1739 	u8	     port;
1740 	/* Following are the optional layers according to user request
1741 	 * struct ib_flow_spec_xxx
1742 	 * struct ib_flow_spec_yyy
1743 	 */
1744 };
1745 
1746 struct ib_flow {
1747 	struct ib_qp		*qp;
1748 	struct ib_uobject	*uobject;
1749 };
1750 
1751 struct ib_mad_hdr;
1752 struct ib_grh;
1753 
1754 enum ib_process_mad_flags {
1755 	IB_MAD_IGNORE_MKEY	= 1,
1756 	IB_MAD_IGNORE_BKEY	= 2,
1757 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1758 };
1759 
1760 enum ib_mad_result {
1761 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1762 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1763 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1764 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1765 };
1766 
1767 #define IB_DEVICE_NAME_MAX 64
1768 
1769 struct ib_cache {
1770 	rwlock_t                lock;
1771 	struct ib_event_handler event_handler;
1772 	struct ib_pkey_cache  **pkey_cache;
1773 	struct ib_gid_table   **gid_cache;
1774 	u8                     *lmc_cache;
1775 };
1776 
1777 struct ib_dma_mapping_ops {
1778 	int		(*mapping_error)(struct ib_device *dev,
1779 					 u64 dma_addr);
1780 	u64		(*map_single)(struct ib_device *dev,
1781 				      void *ptr, size_t size,
1782 				      enum dma_data_direction direction);
1783 	void		(*unmap_single)(struct ib_device *dev,
1784 					u64 addr, size_t size,
1785 					enum dma_data_direction direction);
1786 	u64		(*map_page)(struct ib_device *dev,
1787 				    struct page *page, unsigned long offset,
1788 				    size_t size,
1789 				    enum dma_data_direction direction);
1790 	void		(*unmap_page)(struct ib_device *dev,
1791 				      u64 addr, size_t size,
1792 				      enum dma_data_direction direction);
1793 	int		(*map_sg)(struct ib_device *dev,
1794 				  struct scatterlist *sg, int nents,
1795 				  enum dma_data_direction direction);
1796 	void		(*unmap_sg)(struct ib_device *dev,
1797 				    struct scatterlist *sg, int nents,
1798 				    enum dma_data_direction direction);
1799 	int		(*map_sg_attrs)(struct ib_device *dev,
1800 					struct scatterlist *sg, int nents,
1801 					enum dma_data_direction direction,
1802 					struct dma_attrs *attrs);
1803 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1804 					  struct scatterlist *sg, int nents,
1805 					  enum dma_data_direction direction,
1806 					  struct dma_attrs *attrs);
1807 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1808 					       u64 dma_handle,
1809 					       size_t size,
1810 					       enum dma_data_direction dir);
1811 	void		(*sync_single_for_device)(struct ib_device *dev,
1812 						  u64 dma_handle,
1813 						  size_t size,
1814 						  enum dma_data_direction dir);
1815 	void		*(*alloc_coherent)(struct ib_device *dev,
1816 					   size_t size,
1817 					   u64 *dma_handle,
1818 					   gfp_t flag);
1819 	void		(*free_coherent)(struct ib_device *dev,
1820 					 size_t size, void *cpu_addr,
1821 					 u64 dma_handle);
1822 };
1823 
1824 struct iw_cm_verbs;
1825 
1826 struct ib_port_immutable {
1827 	int                           pkey_tbl_len;
1828 	int                           gid_tbl_len;
1829 	u32                           core_cap_flags;
1830 	u32                           max_mad_size;
1831 };
1832 
1833 struct ib_device {
1834 	struct device                *dma_device;
1835 
1836 	char                          name[IB_DEVICE_NAME_MAX];
1837 
1838 	struct list_head              event_handler_list;
1839 	spinlock_t                    event_handler_lock;
1840 
1841 	spinlock_t                    client_data_lock;
1842 	struct list_head              core_list;
1843 	/* Access to the client_data_list is protected by the client_data_lock
1844 	 * spinlock and the lists_rwsem read-write semaphore */
1845 	struct list_head              client_data_list;
1846 
1847 	struct ib_cache               cache;
1848 	/**
1849 	 * port_immutable is indexed by port number
1850 	 */
1851 	struct ib_port_immutable     *port_immutable;
1852 
1853 	int			      num_comp_vectors;
1854 
1855 	struct iw_cm_verbs	     *iwcm;
1856 
1857 	/**
1858 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1859 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1860 	 *   core when the device is removed.  A lifespan of -1 in the return
1861 	 *   struct tells the core to set a default lifespan.
1862 	 */
1863 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1864 						     u8 port_num);
1865 	/**
1866 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1867 	 * @index - The index in the value array we wish to have updated, or
1868 	 *   num_counters if we want all stats updated
1869 	 * Return codes -
1870 	 *   < 0 - Error, no counters updated
1871 	 *   index - Updated the single counter pointed to by index
1872 	 *   num_counters - Updated all counters (will reset the timestamp
1873 	 *     and prevent further calls for lifespan milliseconds)
1874 	 * Drivers are allowed to update all counters in leiu of just the
1875 	 *   one given in index at their option
1876 	 */
1877 	int		           (*get_hw_stats)(struct ib_device *device,
1878 						   struct rdma_hw_stats *stats,
1879 						   u8 port, int index);
1880 	int		           (*query_device)(struct ib_device *device,
1881 						   struct ib_device_attr *device_attr,
1882 						   struct ib_udata *udata);
1883 	int		           (*query_port)(struct ib_device *device,
1884 						 u8 port_num,
1885 						 struct ib_port_attr *port_attr);
1886 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1887 						     u8 port_num);
1888 	/* When calling get_netdev, the HW vendor's driver should return the
1889 	 * net device of device @device at port @port_num or NULL if such
1890 	 * a net device doesn't exist. The vendor driver should call dev_hold
1891 	 * on this net device. The HW vendor's device driver must guarantee
1892 	 * that this function returns NULL before the net device reaches
1893 	 * NETDEV_UNREGISTER_FINAL state.
1894 	 */
1895 	struct ifnet		  *(*get_netdev)(struct ib_device *device,
1896 						 u8 port_num);
1897 	int		           (*query_gid)(struct ib_device *device,
1898 						u8 port_num, int index,
1899 						union ib_gid *gid);
1900 	/* When calling add_gid, the HW vendor's driver should
1901 	 * add the gid of device @device at gid index @index of
1902 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1903 	 * the network device related to this gid is available
1904 	 * at @attr. @context allows the HW vendor driver to store extra
1905 	 * information together with a GID entry. The HW vendor may allocate
1906 	 * memory to contain this information and store it in @context when a
1907 	 * new GID entry is written to. Params are consistent until the next
1908 	 * call of add_gid or delete_gid. The function should return 0 on
1909 	 * success or error otherwise. The function could be called
1910 	 * concurrently for different ports. This function is only called
1911 	 * when roce_gid_table is used.
1912 	 */
1913 	int		           (*add_gid)(struct ib_device *device,
1914 					      u8 port_num,
1915 					      unsigned int index,
1916 					      const union ib_gid *gid,
1917 					      const struct ib_gid_attr *attr,
1918 					      void **context);
1919 	/* When calling del_gid, the HW vendor's driver should delete the
1920 	 * gid of device @device at gid index @index of port @port_num.
1921 	 * Upon the deletion of a GID entry, the HW vendor must free any
1922 	 * allocated memory. The caller will clear @context afterwards.
1923 	 * This function is only called when roce_gid_table is used.
1924 	 */
1925 	int		           (*del_gid)(struct ib_device *device,
1926 					      u8 port_num,
1927 					      unsigned int index,
1928 					      void **context);
1929 	int		           (*query_pkey)(struct ib_device *device,
1930 						 u8 port_num, u16 index, u16 *pkey);
1931 	int		           (*modify_device)(struct ib_device *device,
1932 						    int device_modify_mask,
1933 						    struct ib_device_modify *device_modify);
1934 	int		           (*modify_port)(struct ib_device *device,
1935 						  u8 port_num, int port_modify_mask,
1936 						  struct ib_port_modify *port_modify);
1937 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1938 						     struct ib_udata *udata);
1939 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1940 	int                        (*mmap)(struct ib_ucontext *context,
1941 					   struct vm_area_struct *vma);
1942 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1943 					       struct ib_ucontext *context,
1944 					       struct ib_udata *udata);
1945 	int                        (*dealloc_pd)(struct ib_pd *pd);
1946 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1947 						struct ib_ah_attr *ah_attr,
1948 						struct ib_udata *udata);
1949 	int                        (*modify_ah)(struct ib_ah *ah,
1950 						struct ib_ah_attr *ah_attr);
1951 	int                        (*query_ah)(struct ib_ah *ah,
1952 					       struct ib_ah_attr *ah_attr);
1953 	int                        (*destroy_ah)(struct ib_ah *ah);
1954 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1955 						 struct ib_srq_init_attr *srq_init_attr,
1956 						 struct ib_udata *udata);
1957 	int                        (*modify_srq)(struct ib_srq *srq,
1958 						 struct ib_srq_attr *srq_attr,
1959 						 enum ib_srq_attr_mask srq_attr_mask,
1960 						 struct ib_udata *udata);
1961 	int                        (*query_srq)(struct ib_srq *srq,
1962 						struct ib_srq_attr *srq_attr);
1963 	int                        (*destroy_srq)(struct ib_srq *srq);
1964 	int                        (*post_srq_recv)(struct ib_srq *srq,
1965 						    struct ib_recv_wr *recv_wr,
1966 						    struct ib_recv_wr **bad_recv_wr);
1967 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1968 						struct ib_qp_init_attr *qp_init_attr,
1969 						struct ib_udata *udata);
1970 	int                        (*modify_qp)(struct ib_qp *qp,
1971 						struct ib_qp_attr *qp_attr,
1972 						int qp_attr_mask,
1973 						struct ib_udata *udata);
1974 	int                        (*query_qp)(struct ib_qp *qp,
1975 					       struct ib_qp_attr *qp_attr,
1976 					       int qp_attr_mask,
1977 					       struct ib_qp_init_attr *qp_init_attr);
1978 	int                        (*destroy_qp)(struct ib_qp *qp);
1979 	int                        (*post_send)(struct ib_qp *qp,
1980 						struct ib_send_wr *send_wr,
1981 						struct ib_send_wr **bad_send_wr);
1982 	int                        (*post_recv)(struct ib_qp *qp,
1983 						struct ib_recv_wr *recv_wr,
1984 						struct ib_recv_wr **bad_recv_wr);
1985 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1986 						const struct ib_cq_init_attr *attr,
1987 						struct ib_ucontext *context,
1988 						struct ib_udata *udata);
1989 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1990 						u16 cq_period);
1991 	int                        (*destroy_cq)(struct ib_cq *cq);
1992 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1993 						struct ib_udata *udata);
1994 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1995 					      struct ib_wc *wc);
1996 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1997 	int                        (*req_notify_cq)(struct ib_cq *cq,
1998 						    enum ib_cq_notify_flags flags);
1999 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2000 						      int wc_cnt);
2001 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2002 						 int mr_access_flags);
2003 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2004 						  u64 start, u64 length,
2005 						  u64 virt_addr,
2006 						  int mr_access_flags,
2007 						  struct ib_udata *udata);
2008 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2009 						    int flags,
2010 						    u64 start, u64 length,
2011 						    u64 virt_addr,
2012 						    int mr_access_flags,
2013 						    struct ib_pd *pd,
2014 						    struct ib_udata *udata);
2015 	int                        (*dereg_mr)(struct ib_mr *mr);
2016 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2017 					       enum ib_mr_type mr_type,
2018 					       u32 max_num_sg);
2019 	int                        (*map_mr_sg)(struct ib_mr *mr,
2020 						struct scatterlist *sg,
2021 						int sg_nents,
2022 						unsigned int *sg_offset);
2023 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2024 					       enum ib_mw_type type,
2025 					       struct ib_udata *udata);
2026 	int                        (*dealloc_mw)(struct ib_mw *mw);
2027 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2028 						int mr_access_flags,
2029 						struct ib_fmr_attr *fmr_attr);
2030 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2031 						   u64 *page_list, int list_len,
2032 						   u64 iova);
2033 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2034 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2035 	int                        (*attach_mcast)(struct ib_qp *qp,
2036 						   union ib_gid *gid,
2037 						   u16 lid);
2038 	int                        (*detach_mcast)(struct ib_qp *qp,
2039 						   union ib_gid *gid,
2040 						   u16 lid);
2041 	int                        (*process_mad)(struct ib_device *device,
2042 						  int process_mad_flags,
2043 						  u8 port_num,
2044 						  const struct ib_wc *in_wc,
2045 						  const struct ib_grh *in_grh,
2046 						  const struct ib_mad_hdr *in_mad,
2047 						  size_t in_mad_size,
2048 						  struct ib_mad_hdr *out_mad,
2049 						  size_t *out_mad_size,
2050 						  u16 *out_mad_pkey_index);
2051 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2052 						 struct ib_ucontext *ucontext,
2053 						 struct ib_udata *udata);
2054 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2055 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2056 						  struct ib_flow_attr
2057 						  *flow_attr,
2058 						  int domain);
2059 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2060 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2061 						      struct ib_mr_status *mr_status);
2062 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2063 	void			   (*drain_rq)(struct ib_qp *qp);
2064 	void			   (*drain_sq)(struct ib_qp *qp);
2065 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2066 							int state);
2067 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2068 						   struct ifla_vf_info *ivf);
2069 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2070 						   struct ifla_vf_stats *stats);
2071 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2072 						  int type);
2073 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2074 						struct ib_wq_init_attr *init_attr,
2075 						struct ib_udata *udata);
2076 	int			   (*destroy_wq)(struct ib_wq *wq);
2077 	int			   (*modify_wq)(struct ib_wq *wq,
2078 						struct ib_wq_attr *attr,
2079 						u32 wq_attr_mask,
2080 						struct ib_udata *udata);
2081 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2082 							   struct ib_rwq_ind_table_init_attr *init_attr,
2083 							   struct ib_udata *udata);
2084 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2085 	struct ib_dma_mapping_ops   *dma_ops;
2086 
2087 	struct module               *owner;
2088 	struct device                dev;
2089 	struct kobject               *ports_parent;
2090 	struct list_head             port_list;
2091 
2092 	enum {
2093 		IB_DEV_UNINITIALIZED,
2094 		IB_DEV_REGISTERED,
2095 		IB_DEV_UNREGISTERED
2096 	}                            reg_state;
2097 
2098 	int			     uverbs_abi_ver;
2099 	u64			     uverbs_cmd_mask;
2100 	u64			     uverbs_ex_cmd_mask;
2101 
2102 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2103 	__be64			     node_guid;
2104 	u32			     local_dma_lkey;
2105 	u16                          is_switch:1;
2106 	u8                           node_type;
2107 	u8                           phys_port_cnt;
2108 	struct ib_device_attr        attrs;
2109 	struct attribute_group	     *hw_stats_ag;
2110 	struct rdma_hw_stats         *hw_stats;
2111 
2112 	/**
2113 	 * The following mandatory functions are used only at device
2114 	 * registration.  Keep functions such as these at the end of this
2115 	 * structure to avoid cache line misses when accessing struct ib_device
2116 	 * in fast paths.
2117 	 */
2118 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2119 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2120 };
2121 
2122 struct ib_client {
2123 	char  *name;
2124 	void (*add)   (struct ib_device *);
2125 	void (*remove)(struct ib_device *, void *client_data);
2126 
2127 	/* Returns the net_dev belonging to this ib_client and matching the
2128 	 * given parameters.
2129 	 * @dev:	 An RDMA device that the net_dev use for communication.
2130 	 * @port:	 A physical port number on the RDMA device.
2131 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2132 	 * @gid:	 A GID that the net_dev uses to communicate.
2133 	 * @addr:	 An IP address the net_dev is configured with.
2134 	 * @client_data: The device's client data set by ib_set_client_data().
2135 	 *
2136 	 * An ib_client that implements a net_dev on top of RDMA devices
2137 	 * (such as IP over IB) should implement this callback, allowing the
2138 	 * rdma_cm module to find the right net_dev for a given request.
2139 	 *
2140 	 * The caller is responsible for calling dev_put on the returned
2141 	 * netdev. */
2142 	struct ifnet *(*get_net_dev_by_params)(
2143 			struct ib_device *dev,
2144 			u8 port,
2145 			u16 pkey,
2146 			const union ib_gid *gid,
2147 			const struct sockaddr *addr,
2148 			void *client_data);
2149 	struct list_head list;
2150 };
2151 
2152 struct ib_device *ib_alloc_device(size_t size);
2153 void ib_dealloc_device(struct ib_device *device);
2154 
2155 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2156 
2157 int ib_register_device(struct ib_device *device,
2158 		       int (*port_callback)(struct ib_device *,
2159 					    u8, struct kobject *));
2160 void ib_unregister_device(struct ib_device *device);
2161 
2162 int ib_register_client   (struct ib_client *client);
2163 void ib_unregister_client(struct ib_client *client);
2164 
2165 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2166 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2167 			 void *data);
2168 
2169 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2170 {
2171 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2172 }
2173 
2174 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2175 {
2176 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2177 }
2178 
2179 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2180 				       size_t offset,
2181 				       size_t len)
2182 {
2183 	const void __user *p = (const char __user *)udata->inbuf + offset;
2184 	bool ret;
2185 	u8 *buf;
2186 
2187 	if (len > USHRT_MAX)
2188 		return false;
2189 
2190 	buf = memdup_user(p, len);
2191 	if (IS_ERR(buf))
2192 		return false;
2193 
2194 	ret = !memchr_inv(buf, 0, len);
2195 	kfree(buf);
2196 	return ret;
2197 }
2198 
2199 /**
2200  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2201  * contains all required attributes and no attributes not allowed for
2202  * the given QP state transition.
2203  * @cur_state: Current QP state
2204  * @next_state: Next QP state
2205  * @type: QP type
2206  * @mask: Mask of supplied QP attributes
2207  * @ll : link layer of port
2208  *
2209  * This function is a helper function that a low-level driver's
2210  * modify_qp method can use to validate the consumer's input.  It
2211  * checks that cur_state and next_state are valid QP states, that a
2212  * transition from cur_state to next_state is allowed by the IB spec,
2213  * and that the attribute mask supplied is allowed for the transition.
2214  */
2215 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2216 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2217 		       enum rdma_link_layer ll);
2218 
2219 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2220 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2221 void ib_dispatch_event(struct ib_event *event);
2222 
2223 int ib_query_port(struct ib_device *device,
2224 		  u8 port_num, struct ib_port_attr *port_attr);
2225 
2226 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2227 					       u8 port_num);
2228 
2229 /**
2230  * rdma_cap_ib_switch - Check if the device is IB switch
2231  * @device: Device to check
2232  *
2233  * Device driver is responsible for setting is_switch bit on
2234  * in ib_device structure at init time.
2235  *
2236  * Return: true if the device is IB switch.
2237  */
2238 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2239 {
2240 	return device->is_switch;
2241 }
2242 
2243 /**
2244  * rdma_start_port - Return the first valid port number for the device
2245  * specified
2246  *
2247  * @device: Device to be checked
2248  *
2249  * Return start port number
2250  */
2251 static inline u8 rdma_start_port(const struct ib_device *device)
2252 {
2253 	return rdma_cap_ib_switch(device) ? 0 : 1;
2254 }
2255 
2256 /**
2257  * rdma_end_port - Return the last valid port number for the device
2258  * specified
2259  *
2260  * @device: Device to be checked
2261  *
2262  * Return last port number
2263  */
2264 static inline u8 rdma_end_port(const struct ib_device *device)
2265 {
2266 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2267 }
2268 
2269 static inline int rdma_is_port_valid(const struct ib_device *device,
2270 				     unsigned int port)
2271 {
2272 	return (port >= rdma_start_port(device) &&
2273 		port <= rdma_end_port(device));
2274 }
2275 
2276 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2277 {
2278 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2279 }
2280 
2281 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2282 {
2283 	return device->port_immutable[port_num].core_cap_flags &
2284 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2285 }
2286 
2287 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2288 {
2289 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2290 }
2291 
2292 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2293 {
2294 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2295 }
2296 
2297 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2298 {
2299 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2300 }
2301 
2302 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2303 {
2304 	return rdma_protocol_ib(device, port_num) ||
2305 		rdma_protocol_roce(device, port_num);
2306 }
2307 
2308 /**
2309  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2310  * Management Datagrams.
2311  * @device: Device to check
2312  * @port_num: Port number to check
2313  *
2314  * Management Datagrams (MAD) are a required part of the InfiniBand
2315  * specification and are supported on all InfiniBand devices.  A slightly
2316  * extended version are also supported on OPA interfaces.
2317  *
2318  * Return: true if the port supports sending/receiving of MAD packets.
2319  */
2320 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2321 {
2322 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2323 }
2324 
2325 /**
2326  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2327  * Management Datagrams.
2328  * @device: Device to check
2329  * @port_num: Port number to check
2330  *
2331  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2332  * datagrams with their own versions.  These OPA MADs share many but not all of
2333  * the characteristics of InfiniBand MADs.
2334  *
2335  * OPA MADs differ in the following ways:
2336  *
2337  *    1) MADs are variable size up to 2K
2338  *       IBTA defined MADs remain fixed at 256 bytes
2339  *    2) OPA SMPs must carry valid PKeys
2340  *    3) OPA SMP packets are a different format
2341  *
2342  * Return: true if the port supports OPA MAD packet formats.
2343  */
2344 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2345 {
2346 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2347 		== RDMA_CORE_CAP_OPA_MAD;
2348 }
2349 
2350 /**
2351  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2352  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2353  * @device: Device to check
2354  * @port_num: Port number to check
2355  *
2356  * Each InfiniBand node is required to provide a Subnet Management Agent
2357  * that the subnet manager can access.  Prior to the fabric being fully
2358  * configured by the subnet manager, the SMA is accessed via a well known
2359  * interface called the Subnet Management Interface (SMI).  This interface
2360  * uses directed route packets to communicate with the SM to get around the
2361  * chicken and egg problem of the SM needing to know what's on the fabric
2362  * in order to configure the fabric, and needing to configure the fabric in
2363  * order to send packets to the devices on the fabric.  These directed
2364  * route packets do not need the fabric fully configured in order to reach
2365  * their destination.  The SMI is the only method allowed to send
2366  * directed route packets on an InfiniBand fabric.
2367  *
2368  * Return: true if the port provides an SMI.
2369  */
2370 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2371 {
2372 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2373 }
2374 
2375 /**
2376  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2377  * Communication Manager.
2378  * @device: Device to check
2379  * @port_num: Port number to check
2380  *
2381  * The InfiniBand Communication Manager is one of many pre-defined General
2382  * Service Agents (GSA) that are accessed via the General Service
2383  * Interface (GSI).  It's role is to facilitate establishment of connections
2384  * between nodes as well as other management related tasks for established
2385  * connections.
2386  *
2387  * Return: true if the port supports an IB CM (this does not guarantee that
2388  * a CM is actually running however).
2389  */
2390 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2391 {
2392 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2393 }
2394 
2395 /**
2396  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2397  * Communication Manager.
2398  * @device: Device to check
2399  * @port_num: Port number to check
2400  *
2401  * Similar to above, but specific to iWARP connections which have a different
2402  * managment protocol than InfiniBand.
2403  *
2404  * Return: true if the port supports an iWARP CM (this does not guarantee that
2405  * a CM is actually running however).
2406  */
2407 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2408 {
2409 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2410 }
2411 
2412 /**
2413  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2414  * Subnet Administration.
2415  * @device: Device to check
2416  * @port_num: Port number to check
2417  *
2418  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2419  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2420  * fabrics, devices should resolve routes to other hosts by contacting the
2421  * SA to query the proper route.
2422  *
2423  * Return: true if the port should act as a client to the fabric Subnet
2424  * Administration interface.  This does not imply that the SA service is
2425  * running locally.
2426  */
2427 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2428 {
2429 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2430 }
2431 
2432 /**
2433  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2434  * Multicast.
2435  * @device: Device to check
2436  * @port_num: Port number to check
2437  *
2438  * InfiniBand multicast registration is more complex than normal IPv4 or
2439  * IPv6 multicast registration.  Each Host Channel Adapter must register
2440  * with the Subnet Manager when it wishes to join a multicast group.  It
2441  * should do so only once regardless of how many queue pairs it subscribes
2442  * to this group.  And it should leave the group only after all queue pairs
2443  * attached to the group have been detached.
2444  *
2445  * Return: true if the port must undertake the additional adminstrative
2446  * overhead of registering/unregistering with the SM and tracking of the
2447  * total number of queue pairs attached to the multicast group.
2448  */
2449 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2450 {
2451 	return rdma_cap_ib_sa(device, port_num);
2452 }
2453 
2454 /**
2455  * rdma_cap_af_ib - Check if the port of device has the capability
2456  * Native Infiniband Address.
2457  * @device: Device to check
2458  * @port_num: Port number to check
2459  *
2460  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2461  * GID.  RoCE uses a different mechanism, but still generates a GID via
2462  * a prescribed mechanism and port specific data.
2463  *
2464  * Return: true if the port uses a GID address to identify devices on the
2465  * network.
2466  */
2467 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2468 {
2469 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2470 }
2471 
2472 /**
2473  * rdma_cap_eth_ah - Check if the port of device has the capability
2474  * Ethernet Address Handle.
2475  * @device: Device to check
2476  * @port_num: Port number to check
2477  *
2478  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2479  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2480  * port.  Normally, packet headers are generated by the sending host
2481  * adapter, but when sending connectionless datagrams, we must manually
2482  * inject the proper headers for the fabric we are communicating over.
2483  *
2484  * Return: true if we are running as a RoCE port and must force the
2485  * addition of a Global Route Header built from our Ethernet Address
2486  * Handle into our header list for connectionless packets.
2487  */
2488 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2489 {
2490 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2491 }
2492 
2493 /**
2494  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2495  *
2496  * @device: Device
2497  * @port_num: Port number
2498  *
2499  * This MAD size includes the MAD headers and MAD payload.  No other headers
2500  * are included.
2501  *
2502  * Return the max MAD size required by the Port.  Will return 0 if the port
2503  * does not support MADs
2504  */
2505 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2506 {
2507 	return device->port_immutable[port_num].max_mad_size;
2508 }
2509 
2510 /**
2511  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2512  * @device: Device to check
2513  * @port_num: Port number to check
2514  *
2515  * RoCE GID table mechanism manages the various GIDs for a device.
2516  *
2517  * NOTE: if allocating the port's GID table has failed, this call will still
2518  * return true, but any RoCE GID table API will fail.
2519  *
2520  * Return: true if the port uses RoCE GID table mechanism in order to manage
2521  * its GIDs.
2522  */
2523 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2524 					   u8 port_num)
2525 {
2526 	return rdma_protocol_roce(device, port_num) &&
2527 		device->add_gid && device->del_gid;
2528 }
2529 
2530 /*
2531  * Check if the device supports READ W/ INVALIDATE.
2532  */
2533 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2534 {
2535 	/*
2536 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2537 	 * has support for it yet.
2538 	 */
2539 	return rdma_protocol_iwarp(dev, port_num);
2540 }
2541 
2542 int ib_query_gid(struct ib_device *device,
2543 		 u8 port_num, int index, union ib_gid *gid,
2544 		 struct ib_gid_attr *attr);
2545 
2546 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2547 			 int state);
2548 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2549 		     struct ifla_vf_info *info);
2550 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2551 		    struct ifla_vf_stats *stats);
2552 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2553 		   int type);
2554 
2555 int ib_query_pkey(struct ib_device *device,
2556 		  u8 port_num, u16 index, u16 *pkey);
2557 
2558 int ib_modify_device(struct ib_device *device,
2559 		     int device_modify_mask,
2560 		     struct ib_device_modify *device_modify);
2561 
2562 int ib_modify_port(struct ib_device *device,
2563 		   u8 port_num, int port_modify_mask,
2564 		   struct ib_port_modify *port_modify);
2565 
2566 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2567 		enum ib_gid_type gid_type, struct ifnet *ndev,
2568 		u8 *port_num, u16 *index);
2569 
2570 int ib_find_pkey(struct ib_device *device,
2571 		 u8 port_num, u16 pkey, u16 *index);
2572 
2573 enum ib_pd_flags {
2574 	/*
2575 	 * Create a memory registration for all memory in the system and place
2576 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2577 	 * ULPs to avoid the overhead of dynamic MRs.
2578 	 *
2579 	 * This flag is generally considered unsafe and must only be used in
2580 	 * extremly trusted environments.  Every use of it will log a warning
2581 	 * in the kernel log.
2582 	 */
2583 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2584 };
2585 
2586 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2587 		const char *caller);
2588 #define ib_alloc_pd(device, flags) \
2589 	__ib_alloc_pd((device), (flags), __func__)
2590 void ib_dealloc_pd(struct ib_pd *pd);
2591 
2592 /**
2593  * ib_create_ah - Creates an address handle for the given address vector.
2594  * @pd: The protection domain associated with the address handle.
2595  * @ah_attr: The attributes of the address vector.
2596  *
2597  * The address handle is used to reference a local or global destination
2598  * in all UD QP post sends.
2599  */
2600 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2601 
2602 /**
2603  * ib_init_ah_from_wc - Initializes address handle attributes from a
2604  *   work completion.
2605  * @device: Device on which the received message arrived.
2606  * @port_num: Port on which the received message arrived.
2607  * @wc: Work completion associated with the received message.
2608  * @grh: References the received global route header.  This parameter is
2609  *   ignored unless the work completion indicates that the GRH is valid.
2610  * @ah_attr: Returned attributes that can be used when creating an address
2611  *   handle for replying to the message.
2612  */
2613 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2614 		       const struct ib_wc *wc, const struct ib_grh *grh,
2615 		       struct ib_ah_attr *ah_attr);
2616 
2617 /**
2618  * ib_create_ah_from_wc - Creates an address handle associated with the
2619  *   sender of the specified work completion.
2620  * @pd: The protection domain associated with the address handle.
2621  * @wc: Work completion information associated with a received message.
2622  * @grh: References the received global route header.  This parameter is
2623  *   ignored unless the work completion indicates that the GRH is valid.
2624  * @port_num: The outbound port number to associate with the address.
2625  *
2626  * The address handle is used to reference a local or global destination
2627  * in all UD QP post sends.
2628  */
2629 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2630 				   const struct ib_grh *grh, u8 port_num);
2631 
2632 /**
2633  * ib_modify_ah - Modifies the address vector associated with an address
2634  *   handle.
2635  * @ah: The address handle to modify.
2636  * @ah_attr: The new address vector attributes to associate with the
2637  *   address handle.
2638  */
2639 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2640 
2641 /**
2642  * ib_query_ah - Queries the address vector associated with an address
2643  *   handle.
2644  * @ah: The address handle to query.
2645  * @ah_attr: The address vector attributes associated with the address
2646  *   handle.
2647  */
2648 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2649 
2650 /**
2651  * ib_destroy_ah - Destroys an address handle.
2652  * @ah: The address handle to destroy.
2653  */
2654 int ib_destroy_ah(struct ib_ah *ah);
2655 
2656 /**
2657  * ib_create_srq - Creates a SRQ associated with the specified protection
2658  *   domain.
2659  * @pd: The protection domain associated with the SRQ.
2660  * @srq_init_attr: A list of initial attributes required to create the
2661  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2662  *   the actual capabilities of the created SRQ.
2663  *
2664  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2665  * requested size of the SRQ, and set to the actual values allocated
2666  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2667  * will always be at least as large as the requested values.
2668  */
2669 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2670 			     struct ib_srq_init_attr *srq_init_attr);
2671 
2672 /**
2673  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2674  * @srq: The SRQ to modify.
2675  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2676  *   the current values of selected SRQ attributes are returned.
2677  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2678  *   are being modified.
2679  *
2680  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2681  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2682  * the number of receives queued drops below the limit.
2683  */
2684 int ib_modify_srq(struct ib_srq *srq,
2685 		  struct ib_srq_attr *srq_attr,
2686 		  enum ib_srq_attr_mask srq_attr_mask);
2687 
2688 /**
2689  * ib_query_srq - Returns the attribute list and current values for the
2690  *   specified SRQ.
2691  * @srq: The SRQ to query.
2692  * @srq_attr: The attributes of the specified SRQ.
2693  */
2694 int ib_query_srq(struct ib_srq *srq,
2695 		 struct ib_srq_attr *srq_attr);
2696 
2697 /**
2698  * ib_destroy_srq - Destroys the specified SRQ.
2699  * @srq: The SRQ to destroy.
2700  */
2701 int ib_destroy_srq(struct ib_srq *srq);
2702 
2703 /**
2704  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2705  * @srq: The SRQ to post the work request on.
2706  * @recv_wr: A list of work requests to post on the receive queue.
2707  * @bad_recv_wr: On an immediate failure, this parameter will reference
2708  *   the work request that failed to be posted on the QP.
2709  */
2710 static inline int ib_post_srq_recv(struct ib_srq *srq,
2711 				   struct ib_recv_wr *recv_wr,
2712 				   struct ib_recv_wr **bad_recv_wr)
2713 {
2714 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2715 }
2716 
2717 /**
2718  * ib_create_qp - Creates a QP associated with the specified protection
2719  *   domain.
2720  * @pd: The protection domain associated with the QP.
2721  * @qp_init_attr: A list of initial attributes required to create the
2722  *   QP.  If QP creation succeeds, then the attributes are updated to
2723  *   the actual capabilities of the created QP.
2724  */
2725 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2726 			   struct ib_qp_init_attr *qp_init_attr);
2727 
2728 /**
2729  * ib_modify_qp - Modifies the attributes for the specified QP and then
2730  *   transitions the QP to the given state.
2731  * @qp: The QP to modify.
2732  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2733  *   the current values of selected QP attributes are returned.
2734  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2735  *   are being modified.
2736  */
2737 int ib_modify_qp(struct ib_qp *qp,
2738 		 struct ib_qp_attr *qp_attr,
2739 		 int qp_attr_mask);
2740 
2741 /**
2742  * ib_query_qp - Returns the attribute list and current values for the
2743  *   specified QP.
2744  * @qp: The QP to query.
2745  * @qp_attr: The attributes of the specified QP.
2746  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2747  * @qp_init_attr: Additional attributes of the selected QP.
2748  *
2749  * The qp_attr_mask may be used to limit the query to gathering only the
2750  * selected attributes.
2751  */
2752 int ib_query_qp(struct ib_qp *qp,
2753 		struct ib_qp_attr *qp_attr,
2754 		int qp_attr_mask,
2755 		struct ib_qp_init_attr *qp_init_attr);
2756 
2757 /**
2758  * ib_destroy_qp - Destroys the specified QP.
2759  * @qp: The QP to destroy.
2760  */
2761 int ib_destroy_qp(struct ib_qp *qp);
2762 
2763 /**
2764  * ib_open_qp - Obtain a reference to an existing sharable QP.
2765  * @xrcd - XRC domain
2766  * @qp_open_attr: Attributes identifying the QP to open.
2767  *
2768  * Returns a reference to a sharable QP.
2769  */
2770 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2771 			 struct ib_qp_open_attr *qp_open_attr);
2772 
2773 /**
2774  * ib_close_qp - Release an external reference to a QP.
2775  * @qp: The QP handle to release
2776  *
2777  * The opened QP handle is released by the caller.  The underlying
2778  * shared QP is not destroyed until all internal references are released.
2779  */
2780 int ib_close_qp(struct ib_qp *qp);
2781 
2782 /**
2783  * ib_post_send - Posts a list of work requests to the send queue of
2784  *   the specified QP.
2785  * @qp: The QP to post the work request on.
2786  * @send_wr: A list of work requests to post on the send queue.
2787  * @bad_send_wr: On an immediate failure, this parameter will reference
2788  *   the work request that failed to be posted on the QP.
2789  *
2790  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2791  * error is returned, the QP state shall not be affected,
2792  * ib_post_send() will return an immediate error after queueing any
2793  * earlier work requests in the list.
2794  */
2795 static inline int ib_post_send(struct ib_qp *qp,
2796 			       struct ib_send_wr *send_wr,
2797 			       struct ib_send_wr **bad_send_wr)
2798 {
2799 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2800 }
2801 
2802 /**
2803  * ib_post_recv - Posts a list of work requests to the receive queue of
2804  *   the specified QP.
2805  * @qp: The QP to post the work request on.
2806  * @recv_wr: A list of work requests to post on the receive queue.
2807  * @bad_recv_wr: On an immediate failure, this parameter will reference
2808  *   the work request that failed to be posted on the QP.
2809  */
2810 static inline int ib_post_recv(struct ib_qp *qp,
2811 			       struct ib_recv_wr *recv_wr,
2812 			       struct ib_recv_wr **bad_recv_wr)
2813 {
2814 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2815 }
2816 
2817 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2818 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2819 void ib_free_cq(struct ib_cq *cq);
2820 
2821 /**
2822  * ib_create_cq - Creates a CQ on the specified device.
2823  * @device: The device on which to create the CQ.
2824  * @comp_handler: A user-specified callback that is invoked when a
2825  *   completion event occurs on the CQ.
2826  * @event_handler: A user-specified callback that is invoked when an
2827  *   asynchronous event not associated with a completion occurs on the CQ.
2828  * @cq_context: Context associated with the CQ returned to the user via
2829  *   the associated completion and event handlers.
2830  * @cq_attr: The attributes the CQ should be created upon.
2831  *
2832  * Users can examine the cq structure to determine the actual CQ size.
2833  */
2834 struct ib_cq *ib_create_cq(struct ib_device *device,
2835 			   ib_comp_handler comp_handler,
2836 			   void (*event_handler)(struct ib_event *, void *),
2837 			   void *cq_context,
2838 			   const struct ib_cq_init_attr *cq_attr);
2839 
2840 /**
2841  * ib_resize_cq - Modifies the capacity of the CQ.
2842  * @cq: The CQ to resize.
2843  * @cqe: The minimum size of the CQ.
2844  *
2845  * Users can examine the cq structure to determine the actual CQ size.
2846  */
2847 int ib_resize_cq(struct ib_cq *cq, int cqe);
2848 
2849 /**
2850  * ib_modify_cq - Modifies moderation params of the CQ
2851  * @cq: The CQ to modify.
2852  * @cq_count: number of CQEs that will trigger an event
2853  * @cq_period: max period of time in usec before triggering an event
2854  *
2855  */
2856 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2857 
2858 /**
2859  * ib_destroy_cq - Destroys the specified CQ.
2860  * @cq: The CQ to destroy.
2861  */
2862 int ib_destroy_cq(struct ib_cq *cq);
2863 
2864 /**
2865  * ib_poll_cq - poll a CQ for completion(s)
2866  * @cq:the CQ being polled
2867  * @num_entries:maximum number of completions to return
2868  * @wc:array of at least @num_entries &struct ib_wc where completions
2869  *   will be returned
2870  *
2871  * Poll a CQ for (possibly multiple) completions.  If the return value
2872  * is < 0, an error occurred.  If the return value is >= 0, it is the
2873  * number of completions returned.  If the return value is
2874  * non-negative and < num_entries, then the CQ was emptied.
2875  */
2876 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2877 			     struct ib_wc *wc)
2878 {
2879 	return cq->device->poll_cq(cq, num_entries, wc);
2880 }
2881 
2882 /**
2883  * ib_peek_cq - Returns the number of unreaped completions currently
2884  *   on the specified CQ.
2885  * @cq: The CQ to peek.
2886  * @wc_cnt: A minimum number of unreaped completions to check for.
2887  *
2888  * If the number of unreaped completions is greater than or equal to wc_cnt,
2889  * this function returns wc_cnt, otherwise, it returns the actual number of
2890  * unreaped completions.
2891  */
2892 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2893 
2894 /**
2895  * ib_req_notify_cq - Request completion notification on a CQ.
2896  * @cq: The CQ to generate an event for.
2897  * @flags:
2898  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2899  *   to request an event on the next solicited event or next work
2900  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2901  *   may also be |ed in to request a hint about missed events, as
2902  *   described below.
2903  *
2904  * Return Value:
2905  *    < 0 means an error occurred while requesting notification
2906  *   == 0 means notification was requested successfully, and if
2907  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2908  *        were missed and it is safe to wait for another event.  In
2909  *        this case is it guaranteed that any work completions added
2910  *        to the CQ since the last CQ poll will trigger a completion
2911  *        notification event.
2912  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2913  *        in.  It means that the consumer must poll the CQ again to
2914  *        make sure it is empty to avoid missing an event because of a
2915  *        race between requesting notification and an entry being
2916  *        added to the CQ.  This return value means it is possible
2917  *        (but not guaranteed) that a work completion has been added
2918  *        to the CQ since the last poll without triggering a
2919  *        completion notification event.
2920  */
2921 static inline int ib_req_notify_cq(struct ib_cq *cq,
2922 				   enum ib_cq_notify_flags flags)
2923 {
2924 	return cq->device->req_notify_cq(cq, flags);
2925 }
2926 
2927 /**
2928  * ib_req_ncomp_notif - Request completion notification when there are
2929  *   at least the specified number of unreaped completions on the CQ.
2930  * @cq: The CQ to generate an event for.
2931  * @wc_cnt: The number of unreaped completions that should be on the
2932  *   CQ before an event is generated.
2933  */
2934 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2935 {
2936 	return cq->device->req_ncomp_notif ?
2937 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2938 		-ENOSYS;
2939 }
2940 
2941 /**
2942  * ib_dma_mapping_error - check a DMA addr for error
2943  * @dev: The device for which the dma_addr was created
2944  * @dma_addr: The DMA address to check
2945  */
2946 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2947 {
2948 	if (dev->dma_ops)
2949 		return dev->dma_ops->mapping_error(dev, dma_addr);
2950 	return dma_mapping_error(dev->dma_device, dma_addr);
2951 }
2952 
2953 /**
2954  * ib_dma_map_single - Map a kernel virtual address to DMA address
2955  * @dev: The device for which the dma_addr is to be created
2956  * @cpu_addr: The kernel virtual address
2957  * @size: The size of the region in bytes
2958  * @direction: The direction of the DMA
2959  */
2960 static inline u64 ib_dma_map_single(struct ib_device *dev,
2961 				    void *cpu_addr, size_t size,
2962 				    enum dma_data_direction direction)
2963 {
2964 	if (dev->dma_ops)
2965 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2966 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2967 }
2968 
2969 /**
2970  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2971  * @dev: The device for which the DMA address was created
2972  * @addr: The DMA address
2973  * @size: The size of the region in bytes
2974  * @direction: The direction of the DMA
2975  */
2976 static inline void ib_dma_unmap_single(struct ib_device *dev,
2977 				       u64 addr, size_t size,
2978 				       enum dma_data_direction direction)
2979 {
2980 	if (dev->dma_ops)
2981 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2982 	else
2983 		dma_unmap_single(dev->dma_device, addr, size, direction);
2984 }
2985 
2986 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2987 					  void *cpu_addr, size_t size,
2988 					  enum dma_data_direction direction,
2989 					  struct dma_attrs *dma_attrs)
2990 {
2991 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2992 				    direction, dma_attrs);
2993 }
2994 
2995 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2996 					     u64 addr, size_t size,
2997 					     enum dma_data_direction direction,
2998 					     struct dma_attrs *dma_attrs)
2999 {
3000 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
3001 				      direction, dma_attrs);
3002 }
3003 
3004 /**
3005  * ib_dma_map_page - Map a physical page to DMA address
3006  * @dev: The device for which the dma_addr is to be created
3007  * @page: The page to be mapped
3008  * @offset: The offset within the page
3009  * @size: The size of the region in bytes
3010  * @direction: The direction of the DMA
3011  */
3012 static inline u64 ib_dma_map_page(struct ib_device *dev,
3013 				  struct page *page,
3014 				  unsigned long offset,
3015 				  size_t size,
3016 					 enum dma_data_direction direction)
3017 {
3018 	if (dev->dma_ops)
3019 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3020 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3021 }
3022 
3023 /**
3024  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3025  * @dev: The device for which the DMA address was created
3026  * @addr: The DMA address
3027  * @size: The size of the region in bytes
3028  * @direction: The direction of the DMA
3029  */
3030 static inline void ib_dma_unmap_page(struct ib_device *dev,
3031 				     u64 addr, size_t size,
3032 				     enum dma_data_direction direction)
3033 {
3034 	if (dev->dma_ops)
3035 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3036 	else
3037 		dma_unmap_page(dev->dma_device, addr, size, direction);
3038 }
3039 
3040 /**
3041  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3042  * @dev: The device for which the DMA addresses are to be created
3043  * @sg: The array of scatter/gather entries
3044  * @nents: The number of scatter/gather entries
3045  * @direction: The direction of the DMA
3046  */
3047 static inline int ib_dma_map_sg(struct ib_device *dev,
3048 				struct scatterlist *sg, int nents,
3049 				enum dma_data_direction direction)
3050 {
3051 	if (dev->dma_ops)
3052 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3053 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3054 }
3055 
3056 /**
3057  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3058  * @dev: The device for which the DMA addresses were created
3059  * @sg: The array of scatter/gather entries
3060  * @nents: The number of scatter/gather entries
3061  * @direction: The direction of the DMA
3062  */
3063 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3064 				   struct scatterlist *sg, int nents,
3065 				   enum dma_data_direction direction)
3066 {
3067 	if (dev->dma_ops)
3068 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3069 	else
3070 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3071 }
3072 
3073 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3074 				      struct scatterlist *sg, int nents,
3075 				      enum dma_data_direction direction,
3076 				      struct dma_attrs *dma_attrs)
3077 {
3078 	if (dev->dma_ops)
3079 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3080 						  dma_attrs);
3081 	else
3082 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3083 					dma_attrs);
3084 }
3085 
3086 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3087 					 struct scatterlist *sg, int nents,
3088 					 enum dma_data_direction direction,
3089 					 struct dma_attrs *dma_attrs)
3090 {
3091 	if (dev->dma_ops)
3092 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3093 						  dma_attrs);
3094 	else
3095 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3096 				   dma_attrs);
3097 }
3098 /**
3099  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3100  * @dev: The device for which the DMA addresses were created
3101  * @sg: The scatter/gather entry
3102  *
3103  * Note: this function is obsolete. To do: change all occurrences of
3104  * ib_sg_dma_address() into sg_dma_address().
3105  */
3106 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3107 				    struct scatterlist *sg)
3108 {
3109 	return sg_dma_address(sg);
3110 }
3111 
3112 /**
3113  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3114  * @dev: The device for which the DMA addresses were created
3115  * @sg: The scatter/gather entry
3116  *
3117  * Note: this function is obsolete. To do: change all occurrences of
3118  * ib_sg_dma_len() into sg_dma_len().
3119  */
3120 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3121 					 struct scatterlist *sg)
3122 {
3123 	return sg_dma_len(sg);
3124 }
3125 
3126 /**
3127  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3128  * @dev: The device for which the DMA address was created
3129  * @addr: The DMA address
3130  * @size: The size of the region in bytes
3131  * @dir: The direction of the DMA
3132  */
3133 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3134 					      u64 addr,
3135 					      size_t size,
3136 					      enum dma_data_direction dir)
3137 {
3138 	if (dev->dma_ops)
3139 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3140 	else
3141 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3142 }
3143 
3144 /**
3145  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3146  * @dev: The device for which the DMA address was created
3147  * @addr: The DMA address
3148  * @size: The size of the region in bytes
3149  * @dir: The direction of the DMA
3150  */
3151 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3152 						 u64 addr,
3153 						 size_t size,
3154 						 enum dma_data_direction dir)
3155 {
3156 	if (dev->dma_ops)
3157 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3158 	else
3159 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3160 }
3161 
3162 /**
3163  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3164  * @dev: The device for which the DMA address is requested
3165  * @size: The size of the region to allocate in bytes
3166  * @dma_handle: A pointer for returning the DMA address of the region
3167  * @flag: memory allocator flags
3168  */
3169 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3170 					   size_t size,
3171 					   u64 *dma_handle,
3172 					   gfp_t flag)
3173 {
3174 	if (dev->dma_ops)
3175 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3176 	else {
3177 		dma_addr_t handle;
3178 		void *ret;
3179 
3180 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3181 		*dma_handle = handle;
3182 		return ret;
3183 	}
3184 }
3185 
3186 /**
3187  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3188  * @dev: The device for which the DMA addresses were allocated
3189  * @size: The size of the region
3190  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3191  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3192  */
3193 static inline void ib_dma_free_coherent(struct ib_device *dev,
3194 					size_t size, void *cpu_addr,
3195 					u64 dma_handle)
3196 {
3197 	if (dev->dma_ops)
3198 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3199 	else
3200 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3201 }
3202 
3203 /**
3204  * ib_dereg_mr - Deregisters a memory region and removes it from the
3205  *   HCA translation table.
3206  * @mr: The memory region to deregister.
3207  *
3208  * This function can fail, if the memory region has memory windows bound to it.
3209  */
3210 int ib_dereg_mr(struct ib_mr *mr);
3211 
3212 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3213 			  enum ib_mr_type mr_type,
3214 			  u32 max_num_sg);
3215 
3216 /**
3217  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3218  *   R_Key and L_Key.
3219  * @mr - struct ib_mr pointer to be updated.
3220  * @newkey - new key to be used.
3221  */
3222 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3223 {
3224 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3225 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3226 }
3227 
3228 /**
3229  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3230  * for calculating a new rkey for type 2 memory windows.
3231  * @rkey - the rkey to increment.
3232  */
3233 static inline u32 ib_inc_rkey(u32 rkey)
3234 {
3235 	const u32 mask = 0x000000ff;
3236 	return ((rkey + 1) & mask) | (rkey & ~mask);
3237 }
3238 
3239 /**
3240  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3241  * @pd: The protection domain associated with the unmapped region.
3242  * @mr_access_flags: Specifies the memory access rights.
3243  * @fmr_attr: Attributes of the unmapped region.
3244  *
3245  * A fast memory region must be mapped before it can be used as part of
3246  * a work request.
3247  */
3248 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3249 			    int mr_access_flags,
3250 			    struct ib_fmr_attr *fmr_attr);
3251 
3252 /**
3253  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3254  * @fmr: The fast memory region to associate with the pages.
3255  * @page_list: An array of physical pages to map to the fast memory region.
3256  * @list_len: The number of pages in page_list.
3257  * @iova: The I/O virtual address to use with the mapped region.
3258  */
3259 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3260 				  u64 *page_list, int list_len,
3261 				  u64 iova)
3262 {
3263 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3264 }
3265 
3266 /**
3267  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3268  * @fmr_list: A linked list of fast memory regions to unmap.
3269  */
3270 int ib_unmap_fmr(struct list_head *fmr_list);
3271 
3272 /**
3273  * ib_dealloc_fmr - Deallocates a fast memory region.
3274  * @fmr: The fast memory region to deallocate.
3275  */
3276 int ib_dealloc_fmr(struct ib_fmr *fmr);
3277 
3278 /**
3279  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3280  * @qp: QP to attach to the multicast group.  The QP must be type
3281  *   IB_QPT_UD.
3282  * @gid: Multicast group GID.
3283  * @lid: Multicast group LID in host byte order.
3284  *
3285  * In order to send and receive multicast packets, subnet
3286  * administration must have created the multicast group and configured
3287  * the fabric appropriately.  The port associated with the specified
3288  * QP must also be a member of the multicast group.
3289  */
3290 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3291 
3292 /**
3293  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3294  * @qp: QP to detach from the multicast group.
3295  * @gid: Multicast group GID.
3296  * @lid: Multicast group LID in host byte order.
3297  */
3298 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3299 
3300 /**
3301  * ib_alloc_xrcd - Allocates an XRC domain.
3302  * @device: The device on which to allocate the XRC domain.
3303  */
3304 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3305 
3306 /**
3307  * ib_dealloc_xrcd - Deallocates an XRC domain.
3308  * @xrcd: The XRC domain to deallocate.
3309  */
3310 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3311 
3312 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3313 			       struct ib_flow_attr *flow_attr, int domain);
3314 int ib_destroy_flow(struct ib_flow *flow_id);
3315 
3316 static inline int ib_check_mr_access(int flags)
3317 {
3318 	/*
3319 	 * Local write permission is required if remote write or
3320 	 * remote atomic permission is also requested.
3321 	 */
3322 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3323 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3324 		return -EINVAL;
3325 
3326 	return 0;
3327 }
3328 
3329 /**
3330  * ib_check_mr_status: lightweight check of MR status.
3331  *     This routine may provide status checks on a selected
3332  *     ib_mr. first use is for signature status check.
3333  *
3334  * @mr: A memory region.
3335  * @check_mask: Bitmask of which checks to perform from
3336  *     ib_mr_status_check enumeration.
3337  * @mr_status: The container of relevant status checks.
3338  *     failed checks will be indicated in the status bitmask
3339  *     and the relevant info shall be in the error item.
3340  */
3341 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3342 		       struct ib_mr_status *mr_status);
3343 
3344 struct ifnet *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3345 					    u16 pkey, const union ib_gid *gid,
3346 					    const struct sockaddr *addr);
3347 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3348 			   struct ib_wq_init_attr *init_attr);
3349 int ib_destroy_wq(struct ib_wq *wq);
3350 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3351 		 u32 wq_attr_mask);
3352 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3353 						 struct ib_rwq_ind_table_init_attr*
3354 						 wq_ind_table_init_attr);
3355 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3356 
3357 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3358 		 unsigned int *sg_offset, unsigned int page_size);
3359 
3360 static inline int
3361 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3362 		  unsigned int *sg_offset, unsigned int page_size)
3363 {
3364 	int n;
3365 
3366 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3367 	mr->iova = 0;
3368 
3369 	return n;
3370 }
3371 
3372 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3373 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3374 
3375 void ib_drain_rq(struct ib_qp *qp);
3376 void ib_drain_sq(struct ib_qp *qp);
3377 void ib_drain_qp(struct ib_qp *qp);
3378 
3379 int ib_resolve_eth_dmac(struct ib_device *device,
3380 			struct ib_ah_attr *ah_attr);
3381 #endif /* IB_VERBS_H */
3382