xref: /freebsd/sys/ofed/include/rdma/ib_verbs.h (revision b37f6c9805edb4b89f0a8c2b78f78a3dcfc0647b)
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 
41 #if !defined(IB_VERBS_H)
42 #define IB_VERBS_H
43 
44 #include <linux/types.h>
45 #include <linux/device.h>
46 #include <linux/mm.h>
47 #include <linux/dma-mapping.h>
48 #include <linux/kref.h>
49 #include <linux/list.h>
50 #include <linux/rwsem.h>
51 #include <linux/scatterlist.h>
52 #include <linux/workqueue.h>
53 #include <linux/socket.h>
54 #include <linux/if_ether.h>
55 #include <net/ipv6.h>
56 #include <net/ip.h>
57 #include <linux/string.h>
58 #include <linux/slab.h>
59 #include <linux/rcupdate.h>
60 #include <linux/netdevice.h>
61 #include <netinet/ip.h>
62 
63 #include <asm/atomic.h>
64 #include <asm/uaccess.h>
65 
66 struct ifla_vf_info;
67 struct ifla_vf_stats;
68 
69 extern struct workqueue_struct *ib_wq;
70 extern struct workqueue_struct *ib_comp_wq;
71 
72 union ib_gid {
73 	u8	raw[16];
74 	struct {
75 		__be64	subnet_prefix;
76 		__be64	interface_id;
77 	} global;
78 };
79 
80 extern union ib_gid zgid;
81 
82 enum ib_gid_type {
83 	/* If link layer is Ethernet, this is RoCE V1 */
84 	IB_GID_TYPE_IB        = 0,
85 	IB_GID_TYPE_ROCE      = 0,
86 	IB_GID_TYPE_ROCE_UDP_ENCAP = 1,
87 	IB_GID_TYPE_SIZE
88 };
89 
90 #define ROCE_V2_UDP_DPORT      4791
91 struct ib_gid_attr {
92 	enum ib_gid_type	gid_type;
93 	struct net_device	*ndev;
94 };
95 
96 enum rdma_node_type {
97 	/* IB values map to NodeInfo:NodeType. */
98 	RDMA_NODE_IB_CA 	= 1,
99 	RDMA_NODE_IB_SWITCH,
100 	RDMA_NODE_IB_ROUTER,
101 	RDMA_NODE_RNIC,
102 	RDMA_NODE_USNIC,
103 	RDMA_NODE_USNIC_UDP,
104 };
105 
106 enum {
107 	/* set the local administered indication */
108 	IB_SA_WELL_KNOWN_GUID	= BIT_ULL(57) | 2,
109 };
110 
111 enum rdma_transport_type {
112 	RDMA_TRANSPORT_IB,
113 	RDMA_TRANSPORT_IWARP,
114 	RDMA_TRANSPORT_USNIC,
115 	RDMA_TRANSPORT_USNIC_UDP
116 };
117 
118 enum rdma_protocol_type {
119 	RDMA_PROTOCOL_IB,
120 	RDMA_PROTOCOL_IBOE,
121 	RDMA_PROTOCOL_IWARP,
122 	RDMA_PROTOCOL_USNIC_UDP
123 };
124 
125 __attribute_const__ enum rdma_transport_type
126 rdma_node_get_transport(enum rdma_node_type node_type);
127 
128 enum rdma_network_type {
129 	RDMA_NETWORK_IB,
130 	RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB,
131 	RDMA_NETWORK_IPV4,
132 	RDMA_NETWORK_IPV6
133 };
134 
135 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type)
136 {
137 	if (network_type == RDMA_NETWORK_IPV4 ||
138 	    network_type == RDMA_NETWORK_IPV6)
139 		return IB_GID_TYPE_ROCE_UDP_ENCAP;
140 
141 	/* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */
142 	return IB_GID_TYPE_IB;
143 }
144 
145 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type,
146 							    union ib_gid *gid)
147 {
148 	if (gid_type == IB_GID_TYPE_IB)
149 		return RDMA_NETWORK_IB;
150 
151 	if (ipv6_addr_v4mapped((struct in6_addr *)gid))
152 		return RDMA_NETWORK_IPV4;
153 	else
154 		return RDMA_NETWORK_IPV6;
155 }
156 
157 enum rdma_link_layer {
158 	IB_LINK_LAYER_UNSPECIFIED,
159 	IB_LINK_LAYER_INFINIBAND,
160 	IB_LINK_LAYER_ETHERNET,
161 };
162 
163 enum ib_device_cap_flags {
164 	IB_DEVICE_RESIZE_MAX_WR			= (1 << 0),
165 	IB_DEVICE_BAD_PKEY_CNTR			= (1 << 1),
166 	IB_DEVICE_BAD_QKEY_CNTR			= (1 << 2),
167 	IB_DEVICE_RAW_MULTI			= (1 << 3),
168 	IB_DEVICE_AUTO_PATH_MIG			= (1 << 4),
169 	IB_DEVICE_CHANGE_PHY_PORT		= (1 << 5),
170 	IB_DEVICE_UD_AV_PORT_ENFORCE		= (1 << 6),
171 	IB_DEVICE_CURR_QP_STATE_MOD		= (1 << 7),
172 	IB_DEVICE_SHUTDOWN_PORT			= (1 << 8),
173 	IB_DEVICE_INIT_TYPE			= (1 << 9),
174 	IB_DEVICE_PORT_ACTIVE_EVENT		= (1 << 10),
175 	IB_DEVICE_SYS_IMAGE_GUID		= (1 << 11),
176 	IB_DEVICE_RC_RNR_NAK_GEN		= (1 << 12),
177 	IB_DEVICE_SRQ_RESIZE			= (1 << 13),
178 	IB_DEVICE_N_NOTIFY_CQ			= (1 << 14),
179 
180 	/*
181 	 * This device supports a per-device lkey or stag that can be
182 	 * used without performing a memory registration for the local
183 	 * memory.  Note that ULPs should never check this flag, but
184 	 * instead of use the local_dma_lkey flag in the ib_pd structure,
185 	 * which will always contain a usable lkey.
186 	 */
187 	IB_DEVICE_LOCAL_DMA_LKEY		= (1 << 15),
188 	IB_DEVICE_RESERVED /* old SEND_W_INV */	= (1 << 16),
189 	IB_DEVICE_MEM_WINDOW			= (1 << 17),
190 	/*
191 	 * Devices should set IB_DEVICE_UD_IP_SUM if they support
192 	 * insertion of UDP and TCP checksum on outgoing UD IPoIB
193 	 * messages and can verify the validity of checksum for
194 	 * incoming messages.  Setting this flag implies that the
195 	 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode.
196 	 */
197 	IB_DEVICE_UD_IP_CSUM			= (1 << 18),
198 	IB_DEVICE_UD_TSO			= (1 << 19),
199 	IB_DEVICE_XRC				= (1 << 20),
200 
201 	/*
202 	 * This device supports the IB "base memory management extension",
203 	 * which includes support for fast registrations (IB_WR_REG_MR,
204 	 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs).  This flag should
205 	 * also be set by any iWarp device which must support FRs to comply
206 	 * to the iWarp verbs spec.  iWarp devices also support the
207 	 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the
208 	 * stag.
209 	 */
210 	IB_DEVICE_MEM_MGT_EXTENSIONS		= (1 << 21),
211 	IB_DEVICE_BLOCK_MULTICAST_LOOPBACK	= (1 << 22),
212 	IB_DEVICE_MEM_WINDOW_TYPE_2A		= (1 << 23),
213 	IB_DEVICE_MEM_WINDOW_TYPE_2B		= (1 << 24),
214 	IB_DEVICE_RC_IP_CSUM			= (1 << 25),
215 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
216 	/*
217 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
218 	 * support execution of WQEs that involve synchronization
219 	 * of I/O operations with single completion queue managed
220 	 * by hardware.
221 	 */
222 	IB_DEVICE_CROSS_CHANNEL		= (1 << 27),
223 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
224 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
225 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
226 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
227 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
228 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
229 };
230 
231 enum ib_signature_prot_cap {
232 	IB_PROT_T10DIF_TYPE_1 = 1,
233 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
234 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
235 };
236 
237 enum ib_signature_guard_cap {
238 	IB_GUARD_T10DIF_CRC	= 1,
239 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
240 };
241 
242 enum ib_atomic_cap {
243 	IB_ATOMIC_NONE,
244 	IB_ATOMIC_HCA,
245 	IB_ATOMIC_GLOB
246 };
247 
248 enum ib_odp_general_cap_bits {
249 	IB_ODP_SUPPORT = 1 << 0,
250 };
251 
252 enum ib_odp_transport_cap_bits {
253 	IB_ODP_SUPPORT_SEND	= 1 << 0,
254 	IB_ODP_SUPPORT_RECV	= 1 << 1,
255 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
256 	IB_ODP_SUPPORT_READ	= 1 << 3,
257 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
258 };
259 
260 struct ib_odp_caps {
261 	uint64_t general_caps;
262 	struct {
263 		uint32_t  rc_odp_caps;
264 		uint32_t  uc_odp_caps;
265 		uint32_t  ud_odp_caps;
266 	} per_transport_caps;
267 };
268 
269 struct ib_rss_caps {
270 	/* Corresponding bit will be set if qp type from
271 	 * 'enum ib_qp_type' is supported, e.g.
272 	 * supported_qpts |= 1 << IB_QPT_UD
273 	 */
274 	u32 supported_qpts;
275 	u32 max_rwq_indirection_tables;
276 	u32 max_rwq_indirection_table_size;
277 };
278 
279 enum ib_cq_creation_flags {
280 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
281 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
282 };
283 
284 struct ib_cq_init_attr {
285 	unsigned int	cqe;
286 	int		comp_vector;
287 	u32		flags;
288 };
289 
290 struct ib_device_attr {
291 	u64			fw_ver;
292 	__be64			sys_image_guid;
293 	u64			max_mr_size;
294 	u64			page_size_cap;
295 	u32			vendor_id;
296 	u32			vendor_part_id;
297 	u32			hw_ver;
298 	int			max_qp;
299 	int			max_qp_wr;
300 	u64			device_cap_flags;
301 	int			max_sge;
302 	int			max_sge_rd;
303 	int			max_cq;
304 	int			max_cqe;
305 	int			max_mr;
306 	int			max_pd;
307 	int			max_qp_rd_atom;
308 	int			max_ee_rd_atom;
309 	int			max_res_rd_atom;
310 	int			max_qp_init_rd_atom;
311 	int			max_ee_init_rd_atom;
312 	enum ib_atomic_cap	atomic_cap;
313 	enum ib_atomic_cap	masked_atomic_cap;
314 	int			max_ee;
315 	int			max_rdd;
316 	int			max_mw;
317 	int			max_raw_ipv6_qp;
318 	int			max_raw_ethy_qp;
319 	int			max_mcast_grp;
320 	int			max_mcast_qp_attach;
321 	int			max_total_mcast_qp_attach;
322 	int			max_ah;
323 	int			max_fmr;
324 	int			max_map_per_fmr;
325 	int			max_srq;
326 	int			max_srq_wr;
327 	int			max_srq_sge;
328 	unsigned int		max_fast_reg_page_list_len;
329 	u16			max_pkeys;
330 	u8			local_ca_ack_delay;
331 	int			sig_prot_cap;
332 	int			sig_guard_cap;
333 	struct ib_odp_caps	odp_caps;
334 	uint64_t		timestamp_mask;
335 	uint64_t		hca_core_clock; /* in KHZ */
336 	struct ib_rss_caps	rss_caps;
337 	u32			max_wq_type_rq;
338 };
339 
340 enum ib_mtu {
341 	IB_MTU_256  = 1,
342 	IB_MTU_512  = 2,
343 	IB_MTU_1024 = 3,
344 	IB_MTU_2048 = 4,
345 	IB_MTU_4096 = 5
346 };
347 
348 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
349 {
350 	switch (mtu) {
351 	case IB_MTU_256:  return  256;
352 	case IB_MTU_512:  return  512;
353 	case IB_MTU_1024: return 1024;
354 	case IB_MTU_2048: return 2048;
355 	case IB_MTU_4096: return 4096;
356 	default: 	  return -1;
357 	}
358 }
359 
360 enum ib_port_state {
361 	IB_PORT_NOP		= 0,
362 	IB_PORT_DOWN		= 1,
363 	IB_PORT_INIT		= 2,
364 	IB_PORT_ARMED		= 3,
365 	IB_PORT_ACTIVE		= 4,
366 	IB_PORT_ACTIVE_DEFER	= 5,
367 	IB_PORT_DUMMY		= -1,	/* force enum signed */
368 };
369 
370 enum ib_port_cap_flags {
371 	IB_PORT_SM				= 1 <<  1,
372 	IB_PORT_NOTICE_SUP			= 1 <<  2,
373 	IB_PORT_TRAP_SUP			= 1 <<  3,
374 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
375 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
376 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
377 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
378 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
379 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
380 	IB_PORT_SM_DISABLED			= 1 << 10,
381 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
382 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
383 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
384 	IB_PORT_CM_SUP				= 1 << 16,
385 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
386 	IB_PORT_REINIT_SUP			= 1 << 18,
387 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
388 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
389 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
390 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
391 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
392 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
393 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
394 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
395 };
396 
397 enum ib_port_width {
398 	IB_WIDTH_1X	= 1,
399 	IB_WIDTH_4X	= 2,
400 	IB_WIDTH_8X	= 4,
401 	IB_WIDTH_12X	= 8
402 };
403 
404 static inline int ib_width_enum_to_int(enum ib_port_width width)
405 {
406 	switch (width) {
407 	case IB_WIDTH_1X:  return  1;
408 	case IB_WIDTH_4X:  return  4;
409 	case IB_WIDTH_8X:  return  8;
410 	case IB_WIDTH_12X: return 12;
411 	default: 	  return -1;
412 	}
413 }
414 
415 enum ib_port_speed {
416 	IB_SPEED_SDR	= 1,
417 	IB_SPEED_DDR	= 2,
418 	IB_SPEED_QDR	= 4,
419 	IB_SPEED_FDR10	= 8,
420 	IB_SPEED_FDR	= 16,
421 	IB_SPEED_EDR	= 32
422 };
423 
424 /**
425  * struct rdma_hw_stats
426  * @timestamp - Used by the core code to track when the last update was
427  * @lifespan - Used by the core code to determine how old the counters
428  *   should be before being updated again.  Stored in jiffies, defaults
429  *   to 10 milliseconds, drivers can override the default be specifying
430  *   their own value during their allocation routine.
431  * @name - Array of pointers to static names used for the counters in
432  *   directory.
433  * @num_counters - How many hardware counters there are.  If name is
434  *   shorter than this number, a kernel oops will result.  Driver authors
435  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
436  *   in their code to prevent this.
437  * @value - Array of u64 counters that are accessed by the sysfs code and
438  *   filled in by the drivers get_stats routine
439  */
440 struct rdma_hw_stats {
441 	unsigned long	timestamp;
442 	unsigned long	lifespan;
443 	const char * const *names;
444 	int		num_counters;
445 	u64		value[];
446 };
447 
448 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
449 /**
450  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
451  *   for drivers.
452  * @names - Array of static const char *
453  * @num_counters - How many elements in array
454  * @lifespan - How many milliseconds between updates
455  */
456 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
457 		const char * const *names, int num_counters,
458 		unsigned long lifespan)
459 {
460 	struct rdma_hw_stats *stats;
461 
462 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
463 			GFP_KERNEL);
464 	if (!stats)
465 		return NULL;
466 	stats->names = names;
467 	stats->num_counters = num_counters;
468 	stats->lifespan = msecs_to_jiffies(lifespan);
469 
470 	return stats;
471 }
472 
473 
474 /* Define bits for the various functionality this port needs to be supported by
475  * the core.
476  */
477 /* Management                           0x00000FFF */
478 #define RDMA_CORE_CAP_IB_MAD            0x00000001
479 #define RDMA_CORE_CAP_IB_SMI            0x00000002
480 #define RDMA_CORE_CAP_IB_CM             0x00000004
481 #define RDMA_CORE_CAP_IW_CM             0x00000008
482 #define RDMA_CORE_CAP_IB_SA             0x00000010
483 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
484 
485 /* Address format                       0x000FF000 */
486 #define RDMA_CORE_CAP_AF_IB             0x00001000
487 #define RDMA_CORE_CAP_ETH_AH            0x00002000
488 
489 /* Protocol                             0xFFF00000 */
490 #define RDMA_CORE_CAP_PROT_IB           0x00100000
491 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
492 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
493 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
494 
495 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
496 					| RDMA_CORE_CAP_IB_MAD \
497 					| RDMA_CORE_CAP_IB_SMI \
498 					| RDMA_CORE_CAP_IB_CM  \
499 					| RDMA_CORE_CAP_IB_SA  \
500 					| RDMA_CORE_CAP_AF_IB)
501 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
502 					| RDMA_CORE_CAP_IB_MAD  \
503 					| RDMA_CORE_CAP_IB_CM   \
504 					| RDMA_CORE_CAP_AF_IB   \
505 					| RDMA_CORE_CAP_ETH_AH)
506 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
507 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
508 					| RDMA_CORE_CAP_IB_MAD  \
509 					| RDMA_CORE_CAP_IB_CM   \
510 					| RDMA_CORE_CAP_AF_IB   \
511 					| RDMA_CORE_CAP_ETH_AH)
512 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
513 					| RDMA_CORE_CAP_IW_CM)
514 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
515 					| RDMA_CORE_CAP_OPA_MAD)
516 
517 struct ib_port_attr {
518 	u64			subnet_prefix;
519 	enum ib_port_state	state;
520 	enum ib_mtu		max_mtu;
521 	enum ib_mtu		active_mtu;
522 	int			gid_tbl_len;
523 	u32			port_cap_flags;
524 	u32			max_msg_sz;
525 	u32			bad_pkey_cntr;
526 	u32			qkey_viol_cntr;
527 	u16			pkey_tbl_len;
528 	u16			lid;
529 	u16			sm_lid;
530 	u8			lmc;
531 	u8			max_vl_num;
532 	u8			sm_sl;
533 	u8			subnet_timeout;
534 	u8			init_type_reply;
535 	u8			active_width;
536 	u8			active_speed;
537 	u8                      phys_state;
538 	bool			grh_required;
539 };
540 
541 enum ib_device_modify_flags {
542 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
543 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
544 };
545 
546 #define IB_DEVICE_NODE_DESC_MAX 64
547 
548 struct ib_device_modify {
549 	u64	sys_image_guid;
550 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
551 };
552 
553 enum ib_port_modify_flags {
554 	IB_PORT_SHUTDOWN		= 1,
555 	IB_PORT_INIT_TYPE		= (1<<2),
556 	IB_PORT_RESET_QKEY_CNTR		= (1<<3)
557 };
558 
559 struct ib_port_modify {
560 	u32	set_port_cap_mask;
561 	u32	clr_port_cap_mask;
562 	u8	init_type;
563 };
564 
565 enum ib_event_type {
566 	IB_EVENT_CQ_ERR,
567 	IB_EVENT_QP_FATAL,
568 	IB_EVENT_QP_REQ_ERR,
569 	IB_EVENT_QP_ACCESS_ERR,
570 	IB_EVENT_COMM_EST,
571 	IB_EVENT_SQ_DRAINED,
572 	IB_EVENT_PATH_MIG,
573 	IB_EVENT_PATH_MIG_ERR,
574 	IB_EVENT_DEVICE_FATAL,
575 	IB_EVENT_PORT_ACTIVE,
576 	IB_EVENT_PORT_ERR,
577 	IB_EVENT_LID_CHANGE,
578 	IB_EVENT_PKEY_CHANGE,
579 	IB_EVENT_SM_CHANGE,
580 	IB_EVENT_SRQ_ERR,
581 	IB_EVENT_SRQ_LIMIT_REACHED,
582 	IB_EVENT_QP_LAST_WQE_REACHED,
583 	IB_EVENT_CLIENT_REREGISTER,
584 	IB_EVENT_GID_CHANGE,
585 	IB_EVENT_WQ_FATAL,
586 };
587 
588 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
589 
590 struct ib_event {
591 	struct ib_device	*device;
592 	union {
593 		struct ib_cq	*cq;
594 		struct ib_qp	*qp;
595 		struct ib_srq	*srq;
596 		struct ib_wq	*wq;
597 		u8		port_num;
598 	} element;
599 	enum ib_event_type	event;
600 };
601 
602 struct ib_event_handler {
603 	struct ib_device *device;
604 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
605 	struct list_head  list;
606 };
607 
608 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
609 	do {							\
610 		(_ptr)->device  = _device;			\
611 		(_ptr)->handler = _handler;			\
612 		INIT_LIST_HEAD(&(_ptr)->list);			\
613 	} while (0)
614 
615 struct ib_global_route {
616 	union ib_gid	dgid;
617 	u32		flow_label;
618 	u8		sgid_index;
619 	u8		hop_limit;
620 	u8		traffic_class;
621 };
622 
623 struct ib_grh {
624 	__be32		version_tclass_flow;
625 	__be16		paylen;
626 	u8		next_hdr;
627 	u8		hop_limit;
628 	union ib_gid	sgid;
629 	union ib_gid	dgid;
630 };
631 
632 union rdma_network_hdr {
633 	struct ib_grh ibgrh;
634 	struct {
635 		/* The IB spec states that if it's IPv4, the header
636 		 * is located in the last 20 bytes of the header.
637 		 */
638 		u8		reserved[20];
639 		struct ip	roce4grh;
640 	};
641 };
642 
643 enum {
644 	IB_MULTICAST_QPN = 0xffffff
645 };
646 
647 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
648 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
649 
650 enum ib_ah_flags {
651 	IB_AH_GRH	= 1
652 };
653 
654 enum ib_rate {
655 	IB_RATE_PORT_CURRENT = 0,
656 	IB_RATE_2_5_GBPS = 2,
657 	IB_RATE_5_GBPS   = 5,
658 	IB_RATE_10_GBPS  = 3,
659 	IB_RATE_20_GBPS  = 6,
660 	IB_RATE_30_GBPS  = 4,
661 	IB_RATE_40_GBPS  = 7,
662 	IB_RATE_60_GBPS  = 8,
663 	IB_RATE_80_GBPS  = 9,
664 	IB_RATE_120_GBPS = 10,
665 	IB_RATE_14_GBPS  = 11,
666 	IB_RATE_56_GBPS  = 12,
667 	IB_RATE_112_GBPS = 13,
668 	IB_RATE_168_GBPS = 14,
669 	IB_RATE_25_GBPS  = 15,
670 	IB_RATE_100_GBPS = 16,
671 	IB_RATE_200_GBPS = 17,
672 	IB_RATE_300_GBPS = 18
673 };
674 
675 /**
676  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
677  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
678  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
679  * @rate: rate to convert.
680  */
681 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
682 
683 /**
684  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
685  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
686  * @rate: rate to convert.
687  */
688 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
689 
690 
691 /**
692  * enum ib_mr_type - memory region type
693  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
694  *                            normal registration
695  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
696  *                            signature operations (data-integrity
697  *                            capable regions)
698  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
699  *                            register any arbitrary sg lists (without
700  *                            the normal mr constraints - see
701  *                            ib_map_mr_sg)
702  */
703 enum ib_mr_type {
704 	IB_MR_TYPE_MEM_REG,
705 	IB_MR_TYPE_SIGNATURE,
706 	IB_MR_TYPE_SG_GAPS,
707 };
708 
709 /**
710  * Signature types
711  * IB_SIG_TYPE_NONE: Unprotected.
712  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
713  */
714 enum ib_signature_type {
715 	IB_SIG_TYPE_NONE,
716 	IB_SIG_TYPE_T10_DIF,
717 };
718 
719 /**
720  * Signature T10-DIF block-guard types
721  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
722  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
723  */
724 enum ib_t10_dif_bg_type {
725 	IB_T10DIF_CRC,
726 	IB_T10DIF_CSUM
727 };
728 
729 /**
730  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
731  *     domain.
732  * @bg_type: T10-DIF block guard type (CRC|CSUM)
733  * @pi_interval: protection information interval.
734  * @bg: seed of guard computation.
735  * @app_tag: application tag of guard block
736  * @ref_tag: initial guard block reference tag.
737  * @ref_remap: Indicate wethear the reftag increments each block
738  * @app_escape: Indicate to skip block check if apptag=0xffff
739  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
740  * @apptag_check_mask: check bitmask of application tag.
741  */
742 struct ib_t10_dif_domain {
743 	enum ib_t10_dif_bg_type bg_type;
744 	u16			pi_interval;
745 	u16			bg;
746 	u16			app_tag;
747 	u32			ref_tag;
748 	bool			ref_remap;
749 	bool			app_escape;
750 	bool			ref_escape;
751 	u16			apptag_check_mask;
752 };
753 
754 /**
755  * struct ib_sig_domain - Parameters for signature domain
756  * @sig_type: specific signauture type
757  * @sig: union of all signature domain attributes that may
758  *     be used to set domain layout.
759  */
760 struct ib_sig_domain {
761 	enum ib_signature_type sig_type;
762 	union {
763 		struct ib_t10_dif_domain dif;
764 	} sig;
765 };
766 
767 /**
768  * struct ib_sig_attrs - Parameters for signature handover operation
769  * @check_mask: bitmask for signature byte check (8 bytes)
770  * @mem: memory domain layout desciptor.
771  * @wire: wire domain layout desciptor.
772  */
773 struct ib_sig_attrs {
774 	u8			check_mask;
775 	struct ib_sig_domain	mem;
776 	struct ib_sig_domain	wire;
777 };
778 
779 enum ib_sig_err_type {
780 	IB_SIG_BAD_GUARD,
781 	IB_SIG_BAD_REFTAG,
782 	IB_SIG_BAD_APPTAG,
783 };
784 
785 /**
786  * struct ib_sig_err - signature error descriptor
787  */
788 struct ib_sig_err {
789 	enum ib_sig_err_type	err_type;
790 	u32			expected;
791 	u32			actual;
792 	u64			sig_err_offset;
793 	u32			key;
794 };
795 
796 enum ib_mr_status_check {
797 	IB_MR_CHECK_SIG_STATUS = 1,
798 };
799 
800 /**
801  * struct ib_mr_status - Memory region status container
802  *
803  * @fail_status: Bitmask of MR checks status. For each
804  *     failed check a corresponding status bit is set.
805  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
806  *     failure.
807  */
808 struct ib_mr_status {
809 	u32		    fail_status;
810 	struct ib_sig_err   sig_err;
811 };
812 
813 /**
814  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
815  * enum.
816  * @mult: multiple to convert.
817  */
818 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
819 
820 struct ib_ah_attr {
821 	struct ib_global_route	grh;
822 	u16			dlid;
823 	u8			sl;
824 	u8			src_path_bits;
825 	u8			static_rate;
826 	u8			ah_flags;
827 	u8			port_num;
828 	u8			dmac[ETH_ALEN];
829 };
830 
831 enum ib_wc_status {
832 	IB_WC_SUCCESS,
833 	IB_WC_LOC_LEN_ERR,
834 	IB_WC_LOC_QP_OP_ERR,
835 	IB_WC_LOC_EEC_OP_ERR,
836 	IB_WC_LOC_PROT_ERR,
837 	IB_WC_WR_FLUSH_ERR,
838 	IB_WC_MW_BIND_ERR,
839 	IB_WC_BAD_RESP_ERR,
840 	IB_WC_LOC_ACCESS_ERR,
841 	IB_WC_REM_INV_REQ_ERR,
842 	IB_WC_REM_ACCESS_ERR,
843 	IB_WC_REM_OP_ERR,
844 	IB_WC_RETRY_EXC_ERR,
845 	IB_WC_RNR_RETRY_EXC_ERR,
846 	IB_WC_LOC_RDD_VIOL_ERR,
847 	IB_WC_REM_INV_RD_REQ_ERR,
848 	IB_WC_REM_ABORT_ERR,
849 	IB_WC_INV_EECN_ERR,
850 	IB_WC_INV_EEC_STATE_ERR,
851 	IB_WC_FATAL_ERR,
852 	IB_WC_RESP_TIMEOUT_ERR,
853 	IB_WC_GENERAL_ERR
854 };
855 
856 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
857 
858 enum ib_wc_opcode {
859 	IB_WC_SEND,
860 	IB_WC_RDMA_WRITE,
861 	IB_WC_RDMA_READ,
862 	IB_WC_COMP_SWAP,
863 	IB_WC_FETCH_ADD,
864 	IB_WC_LSO,
865 	IB_WC_LOCAL_INV,
866 	IB_WC_REG_MR,
867 	IB_WC_MASKED_COMP_SWAP,
868 	IB_WC_MASKED_FETCH_ADD,
869 /*
870  * Set value of IB_WC_RECV so consumers can test if a completion is a
871  * receive by testing (opcode & IB_WC_RECV).
872  */
873 	IB_WC_RECV			= 1 << 7,
874 	IB_WC_RECV_RDMA_WITH_IMM,
875 	IB_WC_DUMMY = -1,	/* force enum signed */
876 };
877 
878 enum ib_wc_flags {
879 	IB_WC_GRH		= 1,
880 	IB_WC_WITH_IMM		= (1<<1),
881 	IB_WC_WITH_INVALIDATE	= (1<<2),
882 	IB_WC_IP_CSUM_OK	= (1<<3),
883 	IB_WC_WITH_SMAC		= (1<<4),
884 	IB_WC_WITH_VLAN		= (1<<5),
885 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
886 };
887 
888 struct ib_wc {
889 	union {
890 		u64		wr_id;
891 		struct ib_cqe	*wr_cqe;
892 	};
893 	enum ib_wc_status	status;
894 	enum ib_wc_opcode	opcode;
895 	u32			vendor_err;
896 	u32			byte_len;
897 	struct ib_qp	       *qp;
898 	union {
899 		__be32		imm_data;
900 		u32		invalidate_rkey;
901 	} ex;
902 	u32			src_qp;
903 	int			wc_flags;
904 	u16			pkey_index;
905 	u16			slid;
906 	u8			sl;
907 	u8			dlid_path_bits;
908 	u8			port_num;	/* valid only for DR SMPs on switches */
909 	u8			smac[ETH_ALEN];
910 	u16			vlan_id;
911 	u8			network_hdr_type;
912 };
913 
914 enum ib_cq_notify_flags {
915 	IB_CQ_SOLICITED			= 1 << 0,
916 	IB_CQ_NEXT_COMP			= 1 << 1,
917 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
918 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
919 };
920 
921 enum ib_srq_type {
922 	IB_SRQT_BASIC,
923 	IB_SRQT_XRC
924 };
925 
926 enum ib_srq_attr_mask {
927 	IB_SRQ_MAX_WR	= 1 << 0,
928 	IB_SRQ_LIMIT	= 1 << 1,
929 };
930 
931 struct ib_srq_attr {
932 	u32	max_wr;
933 	u32	max_sge;
934 	u32	srq_limit;
935 };
936 
937 struct ib_srq_init_attr {
938 	void		      (*event_handler)(struct ib_event *, void *);
939 	void		       *srq_context;
940 	struct ib_srq_attr	attr;
941 	enum ib_srq_type	srq_type;
942 
943 	union {
944 		struct {
945 			struct ib_xrcd *xrcd;
946 			struct ib_cq   *cq;
947 		} xrc;
948 	} ext;
949 };
950 
951 struct ib_qp_cap {
952 	u32	max_send_wr;
953 	u32	max_recv_wr;
954 	u32	max_send_sge;
955 	u32	max_recv_sge;
956 	u32	max_inline_data;
957 
958 	/*
959 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
960 	 * ib_create_qp() will calculate the right amount of neededed WRs
961 	 * and MRs based on this.
962 	 */
963 	u32	max_rdma_ctxs;
964 };
965 
966 enum ib_sig_type {
967 	IB_SIGNAL_ALL_WR,
968 	IB_SIGNAL_REQ_WR
969 };
970 
971 enum ib_qp_type {
972 	/*
973 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
974 	 * here (and in that order) since the MAD layer uses them as
975 	 * indices into a 2-entry table.
976 	 */
977 	IB_QPT_SMI,
978 	IB_QPT_GSI,
979 
980 	IB_QPT_RC,
981 	IB_QPT_UC,
982 	IB_QPT_UD,
983 	IB_QPT_RAW_IPV6,
984 	IB_QPT_RAW_ETHERTYPE,
985 	IB_QPT_RAW_PACKET = 8,
986 	IB_QPT_XRC_INI = 9,
987 	IB_QPT_XRC_TGT,
988 	IB_QPT_MAX,
989 	/* Reserve a range for qp types internal to the low level driver.
990 	 * These qp types will not be visible at the IB core layer, so the
991 	 * IB_QPT_MAX usages should not be affected in the core layer
992 	 */
993 	IB_QPT_RESERVED1 = 0x1000,
994 	IB_QPT_RESERVED2,
995 	IB_QPT_RESERVED3,
996 	IB_QPT_RESERVED4,
997 	IB_QPT_RESERVED5,
998 	IB_QPT_RESERVED6,
999 	IB_QPT_RESERVED7,
1000 	IB_QPT_RESERVED8,
1001 	IB_QPT_RESERVED9,
1002 	IB_QPT_RESERVED10,
1003 };
1004 
1005 enum ib_qp_create_flags {
1006 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1007 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1008 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1009 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1010 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1011 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1012 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1013 	IB_QP_CREATE_USE_GFP_NOIO		= 1 << 7,
1014 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1015 	/* reserve bits 26-31 for low level drivers' internal use */
1016 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1017 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1018 };
1019 
1020 /*
1021  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1022  * callback to destroy the passed in QP.
1023  */
1024 
1025 struct ib_qp_init_attr {
1026 	void                  (*event_handler)(struct ib_event *, void *);
1027 	void		       *qp_context;
1028 	struct ib_cq	       *send_cq;
1029 	struct ib_cq	       *recv_cq;
1030 	struct ib_srq	       *srq;
1031 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1032 	struct ib_qp_cap	cap;
1033 	enum ib_sig_type	sq_sig_type;
1034 	enum ib_qp_type		qp_type;
1035 	enum ib_qp_create_flags	create_flags;
1036 
1037 	/*
1038 	 * Only needed for special QP types, or when using the RW API.
1039 	 */
1040 	u8			port_num;
1041 	struct ib_rwq_ind_table *rwq_ind_tbl;
1042 };
1043 
1044 struct ib_qp_open_attr {
1045 	void                  (*event_handler)(struct ib_event *, void *);
1046 	void		       *qp_context;
1047 	u32			qp_num;
1048 	enum ib_qp_type		qp_type;
1049 };
1050 
1051 enum ib_rnr_timeout {
1052 	IB_RNR_TIMER_655_36 =  0,
1053 	IB_RNR_TIMER_000_01 =  1,
1054 	IB_RNR_TIMER_000_02 =  2,
1055 	IB_RNR_TIMER_000_03 =  3,
1056 	IB_RNR_TIMER_000_04 =  4,
1057 	IB_RNR_TIMER_000_06 =  5,
1058 	IB_RNR_TIMER_000_08 =  6,
1059 	IB_RNR_TIMER_000_12 =  7,
1060 	IB_RNR_TIMER_000_16 =  8,
1061 	IB_RNR_TIMER_000_24 =  9,
1062 	IB_RNR_TIMER_000_32 = 10,
1063 	IB_RNR_TIMER_000_48 = 11,
1064 	IB_RNR_TIMER_000_64 = 12,
1065 	IB_RNR_TIMER_000_96 = 13,
1066 	IB_RNR_TIMER_001_28 = 14,
1067 	IB_RNR_TIMER_001_92 = 15,
1068 	IB_RNR_TIMER_002_56 = 16,
1069 	IB_RNR_TIMER_003_84 = 17,
1070 	IB_RNR_TIMER_005_12 = 18,
1071 	IB_RNR_TIMER_007_68 = 19,
1072 	IB_RNR_TIMER_010_24 = 20,
1073 	IB_RNR_TIMER_015_36 = 21,
1074 	IB_RNR_TIMER_020_48 = 22,
1075 	IB_RNR_TIMER_030_72 = 23,
1076 	IB_RNR_TIMER_040_96 = 24,
1077 	IB_RNR_TIMER_061_44 = 25,
1078 	IB_RNR_TIMER_081_92 = 26,
1079 	IB_RNR_TIMER_122_88 = 27,
1080 	IB_RNR_TIMER_163_84 = 28,
1081 	IB_RNR_TIMER_245_76 = 29,
1082 	IB_RNR_TIMER_327_68 = 30,
1083 	IB_RNR_TIMER_491_52 = 31
1084 };
1085 
1086 enum ib_qp_attr_mask {
1087 	IB_QP_STATE			= 1,
1088 	IB_QP_CUR_STATE			= (1<<1),
1089 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1090 	IB_QP_ACCESS_FLAGS		= (1<<3),
1091 	IB_QP_PKEY_INDEX		= (1<<4),
1092 	IB_QP_PORT			= (1<<5),
1093 	IB_QP_QKEY			= (1<<6),
1094 	IB_QP_AV			= (1<<7),
1095 	IB_QP_PATH_MTU			= (1<<8),
1096 	IB_QP_TIMEOUT			= (1<<9),
1097 	IB_QP_RETRY_CNT			= (1<<10),
1098 	IB_QP_RNR_RETRY			= (1<<11),
1099 	IB_QP_RQ_PSN			= (1<<12),
1100 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1101 	IB_QP_ALT_PATH			= (1<<14),
1102 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1103 	IB_QP_SQ_PSN			= (1<<16),
1104 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1105 	IB_QP_PATH_MIG_STATE		= (1<<18),
1106 	IB_QP_CAP			= (1<<19),
1107 	IB_QP_DEST_QPN			= (1<<20),
1108 	IB_QP_RESERVED1			= (1<<21),
1109 	IB_QP_RESERVED2			= (1<<22),
1110 	IB_QP_RESERVED3			= (1<<23),
1111 	IB_QP_RESERVED4			= (1<<24),
1112 };
1113 
1114 enum ib_qp_state {
1115 	IB_QPS_RESET,
1116 	IB_QPS_INIT,
1117 	IB_QPS_RTR,
1118 	IB_QPS_RTS,
1119 	IB_QPS_SQD,
1120 	IB_QPS_SQE,
1121 	IB_QPS_ERR,
1122 	IB_QPS_DUMMY = -1,	/* force enum signed */
1123 };
1124 
1125 enum ib_mig_state {
1126 	IB_MIG_MIGRATED,
1127 	IB_MIG_REARM,
1128 	IB_MIG_ARMED
1129 };
1130 
1131 enum ib_mw_type {
1132 	IB_MW_TYPE_1 = 1,
1133 	IB_MW_TYPE_2 = 2
1134 };
1135 
1136 struct ib_qp_attr {
1137 	enum ib_qp_state	qp_state;
1138 	enum ib_qp_state	cur_qp_state;
1139 	enum ib_mtu		path_mtu;
1140 	enum ib_mig_state	path_mig_state;
1141 	u32			qkey;
1142 	u32			rq_psn;
1143 	u32			sq_psn;
1144 	u32			dest_qp_num;
1145 	int			qp_access_flags;
1146 	struct ib_qp_cap	cap;
1147 	struct ib_ah_attr	ah_attr;
1148 	struct ib_ah_attr	alt_ah_attr;
1149 	u16			pkey_index;
1150 	u16			alt_pkey_index;
1151 	u8			en_sqd_async_notify;
1152 	u8			sq_draining;
1153 	u8			max_rd_atomic;
1154 	u8			max_dest_rd_atomic;
1155 	u8			min_rnr_timer;
1156 	u8			port_num;
1157 	u8			timeout;
1158 	u8			retry_cnt;
1159 	u8			rnr_retry;
1160 	u8			alt_port_num;
1161 	u8			alt_timeout;
1162 };
1163 
1164 enum ib_wr_opcode {
1165 	IB_WR_RDMA_WRITE,
1166 	IB_WR_RDMA_WRITE_WITH_IMM,
1167 	IB_WR_SEND,
1168 	IB_WR_SEND_WITH_IMM,
1169 	IB_WR_RDMA_READ,
1170 	IB_WR_ATOMIC_CMP_AND_SWP,
1171 	IB_WR_ATOMIC_FETCH_AND_ADD,
1172 	IB_WR_LSO,
1173 	IB_WR_SEND_WITH_INV,
1174 	IB_WR_RDMA_READ_WITH_INV,
1175 	IB_WR_LOCAL_INV,
1176 	IB_WR_REG_MR,
1177 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1178 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1179 	IB_WR_REG_SIG_MR,
1180 	/* reserve values for low level drivers' internal use.
1181 	 * These values will not be used at all in the ib core layer.
1182 	 */
1183 	IB_WR_RESERVED1 = 0xf0,
1184 	IB_WR_RESERVED2,
1185 	IB_WR_RESERVED3,
1186 	IB_WR_RESERVED4,
1187 	IB_WR_RESERVED5,
1188 	IB_WR_RESERVED6,
1189 	IB_WR_RESERVED7,
1190 	IB_WR_RESERVED8,
1191 	IB_WR_RESERVED9,
1192 	IB_WR_RESERVED10,
1193 	IB_WR_DUMMY = -1,	/* force enum signed */
1194 };
1195 
1196 enum ib_send_flags {
1197 	IB_SEND_FENCE		= 1,
1198 	IB_SEND_SIGNALED	= (1<<1),
1199 	IB_SEND_SOLICITED	= (1<<2),
1200 	IB_SEND_INLINE		= (1<<3),
1201 	IB_SEND_IP_CSUM		= (1<<4),
1202 
1203 	/* reserve bits 26-31 for low level drivers' internal use */
1204 	IB_SEND_RESERVED_START	= (1 << 26),
1205 	IB_SEND_RESERVED_END	= (1 << 31),
1206 };
1207 
1208 struct ib_sge {
1209 	u64	addr;
1210 	u32	length;
1211 	u32	lkey;
1212 };
1213 
1214 struct ib_cqe {
1215 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1216 };
1217 
1218 struct ib_send_wr {
1219 	struct ib_send_wr      *next;
1220 	union {
1221 		u64		wr_id;
1222 		struct ib_cqe	*wr_cqe;
1223 	};
1224 	struct ib_sge	       *sg_list;
1225 	int			num_sge;
1226 	enum ib_wr_opcode	opcode;
1227 	int			send_flags;
1228 	union {
1229 		__be32		imm_data;
1230 		u32		invalidate_rkey;
1231 	} ex;
1232 };
1233 
1234 struct ib_rdma_wr {
1235 	struct ib_send_wr	wr;
1236 	u64			remote_addr;
1237 	u32			rkey;
1238 };
1239 
1240 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1241 {
1242 	return container_of(wr, struct ib_rdma_wr, wr);
1243 }
1244 
1245 struct ib_atomic_wr {
1246 	struct ib_send_wr	wr;
1247 	u64			remote_addr;
1248 	u64			compare_add;
1249 	u64			swap;
1250 	u64			compare_add_mask;
1251 	u64			swap_mask;
1252 	u32			rkey;
1253 };
1254 
1255 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1256 {
1257 	return container_of(wr, struct ib_atomic_wr, wr);
1258 }
1259 
1260 struct ib_ud_wr {
1261 	struct ib_send_wr	wr;
1262 	struct ib_ah		*ah;
1263 	void			*header;
1264 	int			hlen;
1265 	int			mss;
1266 	u32			remote_qpn;
1267 	u32			remote_qkey;
1268 	u16			pkey_index; /* valid for GSI only */
1269 	u8			port_num;   /* valid for DR SMPs on switch only */
1270 };
1271 
1272 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1273 {
1274 	return container_of(wr, struct ib_ud_wr, wr);
1275 }
1276 
1277 struct ib_reg_wr {
1278 	struct ib_send_wr	wr;
1279 	struct ib_mr		*mr;
1280 	u32			key;
1281 	int			access;
1282 };
1283 
1284 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1285 {
1286 	return container_of(wr, struct ib_reg_wr, wr);
1287 }
1288 
1289 struct ib_sig_handover_wr {
1290 	struct ib_send_wr	wr;
1291 	struct ib_sig_attrs    *sig_attrs;
1292 	struct ib_mr	       *sig_mr;
1293 	int			access_flags;
1294 	struct ib_sge	       *prot;
1295 };
1296 
1297 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1298 {
1299 	return container_of(wr, struct ib_sig_handover_wr, wr);
1300 }
1301 
1302 struct ib_recv_wr {
1303 	struct ib_recv_wr      *next;
1304 	union {
1305 		u64		wr_id;
1306 		struct ib_cqe	*wr_cqe;
1307 	};
1308 	struct ib_sge	       *sg_list;
1309 	int			num_sge;
1310 };
1311 
1312 enum ib_access_flags {
1313 	IB_ACCESS_LOCAL_WRITE	= 1,
1314 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1315 	IB_ACCESS_REMOTE_READ	= (1<<2),
1316 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1317 	IB_ACCESS_MW_BIND	= (1<<4),
1318 	IB_ZERO_BASED		= (1<<5),
1319 	IB_ACCESS_ON_DEMAND     = (1<<6),
1320 };
1321 
1322 /*
1323  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1324  * are hidden here instead of a uapi header!
1325  */
1326 enum ib_mr_rereg_flags {
1327 	IB_MR_REREG_TRANS	= 1,
1328 	IB_MR_REREG_PD		= (1<<1),
1329 	IB_MR_REREG_ACCESS	= (1<<2),
1330 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1331 };
1332 
1333 struct ib_fmr_attr {
1334 	int	max_pages;
1335 	int	max_maps;
1336 	u8	page_shift;
1337 };
1338 
1339 struct ib_umem;
1340 
1341 struct ib_ucontext {
1342 	struct ib_device       *device;
1343 	struct list_head	pd_list;
1344 	struct list_head	mr_list;
1345 	struct list_head	mw_list;
1346 	struct list_head	cq_list;
1347 	struct list_head	qp_list;
1348 	struct list_head	srq_list;
1349 	struct list_head	ah_list;
1350 	struct list_head	xrcd_list;
1351 	struct list_head	rule_list;
1352 	struct list_head	wq_list;
1353 	struct list_head	rwq_ind_tbl_list;
1354 	int			closing;
1355 
1356 	pid_t			tgid;
1357 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1358 	struct rb_root      umem_tree;
1359 	/*
1360 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1361 	 * mmu notifiers registration.
1362 	 */
1363 	struct rw_semaphore	umem_rwsem;
1364 	void (*invalidate_range)(struct ib_umem *umem,
1365 				 unsigned long start, unsigned long end);
1366 
1367 	struct mmu_notifier	mn;
1368 	atomic_t		notifier_count;
1369 	/* A list of umems that don't have private mmu notifier counters yet. */
1370 	struct list_head	no_private_counters;
1371 	int                     odp_mrs_count;
1372 #endif
1373 };
1374 
1375 struct ib_uobject {
1376 	u64			user_handle;	/* handle given to us by userspace */
1377 	struct ib_ucontext     *context;	/* associated user context */
1378 	void		       *object;		/* containing object */
1379 	struct list_head	list;		/* link to context's list */
1380 	int			id;		/* index into kernel idr */
1381 	struct kref		ref;
1382 	struct rw_semaphore	mutex;		/* protects .live */
1383 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1384 	int			live;
1385 };
1386 
1387 struct ib_udata {
1388 	const void __user *inbuf;
1389 	void __user *outbuf;
1390 	size_t       inlen;
1391 	size_t       outlen;
1392 };
1393 
1394 struct ib_pd {
1395 	u32			local_dma_lkey;
1396 	u32			flags;
1397 	struct ib_device       *device;
1398 	struct ib_uobject      *uobject;
1399 	atomic_t          	usecnt; /* count all resources */
1400 
1401 	u32			unsafe_global_rkey;
1402 
1403 	/*
1404 	 * Implementation details of the RDMA core, don't use in drivers:
1405 	 */
1406 	struct ib_mr	       *__internal_mr;
1407 };
1408 
1409 struct ib_xrcd {
1410 	struct ib_device       *device;
1411 	atomic_t		usecnt; /* count all exposed resources */
1412 	struct inode	       *inode;
1413 
1414 	struct mutex		tgt_qp_mutex;
1415 	struct list_head	tgt_qp_list;
1416 };
1417 
1418 struct ib_ah {
1419 	struct ib_device	*device;
1420 	struct ib_pd		*pd;
1421 	struct ib_uobject	*uobject;
1422 };
1423 
1424 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1425 
1426 enum ib_poll_context {
1427 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1428 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1429 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1430 };
1431 
1432 struct ib_cq {
1433 	struct ib_device       *device;
1434 	struct ib_uobject      *uobject;
1435 	ib_comp_handler   	comp_handler;
1436 	void                  (*event_handler)(struct ib_event *, void *);
1437 	void                   *cq_context;
1438 	int               	cqe;
1439 	atomic_t          	usecnt; /* count number of work queues */
1440 	enum ib_poll_context	poll_ctx;
1441 	struct work_struct	work;
1442 };
1443 
1444 struct ib_srq {
1445 	struct ib_device       *device;
1446 	struct ib_pd	       *pd;
1447 	struct ib_uobject      *uobject;
1448 	void		      (*event_handler)(struct ib_event *, void *);
1449 	void		       *srq_context;
1450 	enum ib_srq_type	srq_type;
1451 	atomic_t		usecnt;
1452 
1453 	union {
1454 		struct {
1455 			struct ib_xrcd *xrcd;
1456 			struct ib_cq   *cq;
1457 			u32		srq_num;
1458 		} xrc;
1459 	} ext;
1460 };
1461 
1462 enum ib_wq_type {
1463 	IB_WQT_RQ
1464 };
1465 
1466 enum ib_wq_state {
1467 	IB_WQS_RESET,
1468 	IB_WQS_RDY,
1469 	IB_WQS_ERR
1470 };
1471 
1472 struct ib_wq {
1473 	struct ib_device       *device;
1474 	struct ib_uobject      *uobject;
1475 	void		    *wq_context;
1476 	void		    (*event_handler)(struct ib_event *, void *);
1477 	struct ib_pd	       *pd;
1478 	struct ib_cq	       *cq;
1479 	u32		wq_num;
1480 	enum ib_wq_state       state;
1481 	enum ib_wq_type	wq_type;
1482 	atomic_t		usecnt;
1483 };
1484 
1485 struct ib_wq_init_attr {
1486 	void		       *wq_context;
1487 	enum ib_wq_type	wq_type;
1488 	u32		max_wr;
1489 	u32		max_sge;
1490 	struct	ib_cq	       *cq;
1491 	void		    (*event_handler)(struct ib_event *, void *);
1492 };
1493 
1494 enum ib_wq_attr_mask {
1495 	IB_WQ_STATE	= 1 << 0,
1496 	IB_WQ_CUR_STATE	= 1 << 1,
1497 };
1498 
1499 struct ib_wq_attr {
1500 	enum	ib_wq_state	wq_state;
1501 	enum	ib_wq_state	curr_wq_state;
1502 };
1503 
1504 struct ib_rwq_ind_table {
1505 	struct ib_device	*device;
1506 	struct ib_uobject      *uobject;
1507 	atomic_t		usecnt;
1508 	u32		ind_tbl_num;
1509 	u32		log_ind_tbl_size;
1510 	struct ib_wq	**ind_tbl;
1511 };
1512 
1513 struct ib_rwq_ind_table_init_attr {
1514 	u32		log_ind_tbl_size;
1515 	/* Each entry is a pointer to Receive Work Queue */
1516 	struct ib_wq	**ind_tbl;
1517 };
1518 
1519 /*
1520  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1521  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1522  */
1523 struct ib_qp {
1524 	struct ib_device       *device;
1525 	struct ib_pd	       *pd;
1526 	struct ib_cq	       *send_cq;
1527 	struct ib_cq	       *recv_cq;
1528 	spinlock_t		mr_lock;
1529 	struct ib_srq	       *srq;
1530 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1531 	struct list_head	xrcd_list;
1532 
1533 	/* count times opened, mcast attaches, flow attaches */
1534 	atomic_t		usecnt;
1535 	struct list_head	open_list;
1536 	struct ib_qp           *real_qp;
1537 	struct ib_uobject      *uobject;
1538 	void                  (*event_handler)(struct ib_event *, void *);
1539 	void		       *qp_context;
1540 	u32			qp_num;
1541 	u32			max_write_sge;
1542 	u32			max_read_sge;
1543 	enum ib_qp_type		qp_type;
1544 	struct ib_rwq_ind_table *rwq_ind_tbl;
1545 };
1546 
1547 struct ib_mr {
1548 	struct ib_device  *device;
1549 	struct ib_pd	  *pd;
1550 	u32		   lkey;
1551 	u32		   rkey;
1552 	u64		   iova;
1553 	u32		   length;
1554 	unsigned int	   page_size;
1555 	bool		   need_inval;
1556 	union {
1557 		struct ib_uobject	*uobject;	/* user */
1558 		struct list_head	qp_entry;	/* FR */
1559 	};
1560 };
1561 
1562 struct ib_mw {
1563 	struct ib_device	*device;
1564 	struct ib_pd		*pd;
1565 	struct ib_uobject	*uobject;
1566 	u32			rkey;
1567 	enum ib_mw_type         type;
1568 };
1569 
1570 struct ib_fmr {
1571 	struct ib_device	*device;
1572 	struct ib_pd		*pd;
1573 	struct list_head	list;
1574 	u32			lkey;
1575 	u32			rkey;
1576 };
1577 
1578 /* Supported steering options */
1579 enum ib_flow_attr_type {
1580 	/* steering according to rule specifications */
1581 	IB_FLOW_ATTR_NORMAL		= 0x0,
1582 	/* default unicast and multicast rule -
1583 	 * receive all Eth traffic which isn't steered to any QP
1584 	 */
1585 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1586 	/* default multicast rule -
1587 	 * receive all Eth multicast traffic which isn't steered to any QP
1588 	 */
1589 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1590 	/* sniffer rule - receive all port traffic */
1591 	IB_FLOW_ATTR_SNIFFER		= 0x3
1592 };
1593 
1594 /* Supported steering header types */
1595 enum ib_flow_spec_type {
1596 	/* L2 headers*/
1597 	IB_FLOW_SPEC_ETH	= 0x20,
1598 	IB_FLOW_SPEC_IB		= 0x22,
1599 	/* L3 header*/
1600 	IB_FLOW_SPEC_IPV4	= 0x30,
1601 	IB_FLOW_SPEC_IPV6	= 0x31,
1602 	/* L4 headers*/
1603 	IB_FLOW_SPEC_TCP	= 0x40,
1604 	IB_FLOW_SPEC_UDP	= 0x41
1605 };
1606 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1607 #define IB_FLOW_SPEC_SUPPORT_LAYERS 4
1608 
1609 /* Flow steering rule priority is set according to it's domain.
1610  * Lower domain value means higher priority.
1611  */
1612 enum ib_flow_domain {
1613 	IB_FLOW_DOMAIN_USER,
1614 	IB_FLOW_DOMAIN_ETHTOOL,
1615 	IB_FLOW_DOMAIN_RFS,
1616 	IB_FLOW_DOMAIN_NIC,
1617 	IB_FLOW_DOMAIN_NUM /* Must be last */
1618 };
1619 
1620 enum ib_flow_flags {
1621 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1622 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1623 };
1624 
1625 struct ib_flow_eth_filter {
1626 	u8	dst_mac[6];
1627 	u8	src_mac[6];
1628 	__be16	ether_type;
1629 	__be16	vlan_tag;
1630 	/* Must be last */
1631 	u8	real_sz[0];
1632 };
1633 
1634 struct ib_flow_spec_eth {
1635 	enum ib_flow_spec_type	  type;
1636 	u16			  size;
1637 	struct ib_flow_eth_filter val;
1638 	struct ib_flow_eth_filter mask;
1639 };
1640 
1641 struct ib_flow_ib_filter {
1642 	__be16 dlid;
1643 	__u8   sl;
1644 	/* Must be last */
1645 	u8	real_sz[0];
1646 };
1647 
1648 struct ib_flow_spec_ib {
1649 	enum ib_flow_spec_type	 type;
1650 	u16			 size;
1651 	struct ib_flow_ib_filter val;
1652 	struct ib_flow_ib_filter mask;
1653 };
1654 
1655 /* IPv4 header flags */
1656 enum ib_ipv4_flags {
1657 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1658 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1659 				    last have this flag set */
1660 };
1661 
1662 struct ib_flow_ipv4_filter {
1663 	__be32	src_ip;
1664 	__be32	dst_ip;
1665 	u8	proto;
1666 	u8	tos;
1667 	u8	ttl;
1668 	u8	flags;
1669 	/* Must be last */
1670 	u8	real_sz[0];
1671 };
1672 
1673 struct ib_flow_spec_ipv4 {
1674 	enum ib_flow_spec_type	   type;
1675 	u16			   size;
1676 	struct ib_flow_ipv4_filter val;
1677 	struct ib_flow_ipv4_filter mask;
1678 };
1679 
1680 struct ib_flow_ipv6_filter {
1681 	u8	src_ip[16];
1682 	u8	dst_ip[16];
1683 	__be32	flow_label;
1684 	u8	next_hdr;
1685 	u8	traffic_class;
1686 	u8	hop_limit;
1687 	/* Must be last */
1688 	u8	real_sz[0];
1689 };
1690 
1691 struct ib_flow_spec_ipv6 {
1692 	enum ib_flow_spec_type	   type;
1693 	u16			   size;
1694 	struct ib_flow_ipv6_filter val;
1695 	struct ib_flow_ipv6_filter mask;
1696 };
1697 
1698 struct ib_flow_tcp_udp_filter {
1699 	__be16	dst_port;
1700 	__be16	src_port;
1701 	/* Must be last */
1702 	u8	real_sz[0];
1703 };
1704 
1705 struct ib_flow_spec_tcp_udp {
1706 	enum ib_flow_spec_type	      type;
1707 	u16			      size;
1708 	struct ib_flow_tcp_udp_filter val;
1709 	struct ib_flow_tcp_udp_filter mask;
1710 };
1711 
1712 union ib_flow_spec {
1713 	struct {
1714 		enum ib_flow_spec_type	type;
1715 		u16			size;
1716 	};
1717 	struct ib_flow_spec_eth		eth;
1718 	struct ib_flow_spec_ib		ib;
1719 	struct ib_flow_spec_ipv4        ipv4;
1720 	struct ib_flow_spec_tcp_udp	tcp_udp;
1721 	struct ib_flow_spec_ipv6        ipv6;
1722 };
1723 
1724 struct ib_flow_attr {
1725 	enum ib_flow_attr_type type;
1726 	u16	     size;
1727 	u16	     priority;
1728 	u32	     flags;
1729 	u8	     num_of_specs;
1730 	u8	     port;
1731 	/* Following are the optional layers according to user request
1732 	 * struct ib_flow_spec_xxx
1733 	 * struct ib_flow_spec_yyy
1734 	 */
1735 };
1736 
1737 struct ib_flow {
1738 	struct ib_qp		*qp;
1739 	struct ib_uobject	*uobject;
1740 };
1741 
1742 struct ib_mad_hdr;
1743 struct ib_grh;
1744 
1745 enum ib_process_mad_flags {
1746 	IB_MAD_IGNORE_MKEY	= 1,
1747 	IB_MAD_IGNORE_BKEY	= 2,
1748 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1749 };
1750 
1751 enum ib_mad_result {
1752 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1753 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1754 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1755 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1756 };
1757 
1758 #define IB_DEVICE_NAME_MAX 64
1759 
1760 struct ib_cache {
1761 	rwlock_t                lock;
1762 	struct ib_event_handler event_handler;
1763 	struct ib_pkey_cache  **pkey_cache;
1764 	struct ib_gid_table   **gid_cache;
1765 	u8                     *lmc_cache;
1766 };
1767 
1768 struct ib_dma_mapping_ops {
1769 	int		(*mapping_error)(struct ib_device *dev,
1770 					 u64 dma_addr);
1771 	u64		(*map_single)(struct ib_device *dev,
1772 				      void *ptr, size_t size,
1773 				      enum dma_data_direction direction);
1774 	void		(*unmap_single)(struct ib_device *dev,
1775 					u64 addr, size_t size,
1776 					enum dma_data_direction direction);
1777 	u64		(*map_page)(struct ib_device *dev,
1778 				    struct page *page, unsigned long offset,
1779 				    size_t size,
1780 				    enum dma_data_direction direction);
1781 	void		(*unmap_page)(struct ib_device *dev,
1782 				      u64 addr, size_t size,
1783 				      enum dma_data_direction direction);
1784 	int		(*map_sg)(struct ib_device *dev,
1785 				  struct scatterlist *sg, int nents,
1786 				  enum dma_data_direction direction);
1787 	void		(*unmap_sg)(struct ib_device *dev,
1788 				    struct scatterlist *sg, int nents,
1789 				    enum dma_data_direction direction);
1790 	int		(*map_sg_attrs)(struct ib_device *dev,
1791 					struct scatterlist *sg, int nents,
1792 					enum dma_data_direction direction,
1793 					struct dma_attrs *attrs);
1794 	void		(*unmap_sg_attrs)(struct ib_device *dev,
1795 					  struct scatterlist *sg, int nents,
1796 					  enum dma_data_direction direction,
1797 					  struct dma_attrs *attrs);
1798 	void		(*sync_single_for_cpu)(struct ib_device *dev,
1799 					       u64 dma_handle,
1800 					       size_t size,
1801 					       enum dma_data_direction dir);
1802 	void		(*sync_single_for_device)(struct ib_device *dev,
1803 						  u64 dma_handle,
1804 						  size_t size,
1805 						  enum dma_data_direction dir);
1806 	void		*(*alloc_coherent)(struct ib_device *dev,
1807 					   size_t size,
1808 					   u64 *dma_handle,
1809 					   gfp_t flag);
1810 	void		(*free_coherent)(struct ib_device *dev,
1811 					 size_t size, void *cpu_addr,
1812 					 u64 dma_handle);
1813 };
1814 
1815 struct iw_cm_verbs;
1816 
1817 struct ib_port_immutable {
1818 	int                           pkey_tbl_len;
1819 	int                           gid_tbl_len;
1820 	u32                           core_cap_flags;
1821 	u32                           max_mad_size;
1822 };
1823 
1824 struct ib_device {
1825 	struct device                *dma_device;
1826 
1827 	char                          name[IB_DEVICE_NAME_MAX];
1828 
1829 	struct list_head              event_handler_list;
1830 	spinlock_t                    event_handler_lock;
1831 
1832 	spinlock_t                    client_data_lock;
1833 	struct list_head              core_list;
1834 	/* Access to the client_data_list is protected by the client_data_lock
1835 	 * spinlock and the lists_rwsem read-write semaphore */
1836 	struct list_head              client_data_list;
1837 
1838 	struct ib_cache               cache;
1839 	/**
1840 	 * port_immutable is indexed by port number
1841 	 */
1842 	struct ib_port_immutable     *port_immutable;
1843 
1844 	int			      num_comp_vectors;
1845 
1846 	struct iw_cm_verbs	     *iwcm;
1847 
1848 	/**
1849 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
1850 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
1851 	 *   core when the device is removed.  A lifespan of -1 in the return
1852 	 *   struct tells the core to set a default lifespan.
1853 	 */
1854 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
1855 						     u8 port_num);
1856 	/**
1857 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
1858 	 * @index - The index in the value array we wish to have updated, or
1859 	 *   num_counters if we want all stats updated
1860 	 * Return codes -
1861 	 *   < 0 - Error, no counters updated
1862 	 *   index - Updated the single counter pointed to by index
1863 	 *   num_counters - Updated all counters (will reset the timestamp
1864 	 *     and prevent further calls for lifespan milliseconds)
1865 	 * Drivers are allowed to update all counters in leiu of just the
1866 	 *   one given in index at their option
1867 	 */
1868 	int		           (*get_hw_stats)(struct ib_device *device,
1869 						   struct rdma_hw_stats *stats,
1870 						   u8 port, int index);
1871 	int		           (*query_device)(struct ib_device *device,
1872 						   struct ib_device_attr *device_attr,
1873 						   struct ib_udata *udata);
1874 	int		           (*query_port)(struct ib_device *device,
1875 						 u8 port_num,
1876 						 struct ib_port_attr *port_attr);
1877 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
1878 						     u8 port_num);
1879 	/* When calling get_netdev, the HW vendor's driver should return the
1880 	 * net device of device @device at port @port_num or NULL if such
1881 	 * a net device doesn't exist. The vendor driver should call dev_hold
1882 	 * on this net device. The HW vendor's device driver must guarantee
1883 	 * that this function returns NULL before the net device reaches
1884 	 * NETDEV_UNREGISTER_FINAL state.
1885 	 */
1886 	struct net_device	  *(*get_netdev)(struct ib_device *device,
1887 						 u8 port_num);
1888 	int		           (*query_gid)(struct ib_device *device,
1889 						u8 port_num, int index,
1890 						union ib_gid *gid);
1891 	/* When calling add_gid, the HW vendor's driver should
1892 	 * add the gid of device @device at gid index @index of
1893 	 * port @port_num to be @gid. Meta-info of that gid (for example,
1894 	 * the network device related to this gid is available
1895 	 * at @attr. @context allows the HW vendor driver to store extra
1896 	 * information together with a GID entry. The HW vendor may allocate
1897 	 * memory to contain this information and store it in @context when a
1898 	 * new GID entry is written to. Params are consistent until the next
1899 	 * call of add_gid or delete_gid. The function should return 0 on
1900 	 * success or error otherwise. The function could be called
1901 	 * concurrently for different ports. This function is only called
1902 	 * when roce_gid_table is used.
1903 	 */
1904 	int		           (*add_gid)(struct ib_device *device,
1905 					      u8 port_num,
1906 					      unsigned int index,
1907 					      const union ib_gid *gid,
1908 					      const struct ib_gid_attr *attr,
1909 					      void **context);
1910 	/* When calling del_gid, the HW vendor's driver should delete the
1911 	 * gid of device @device at gid index @index of port @port_num.
1912 	 * Upon the deletion of a GID entry, the HW vendor must free any
1913 	 * allocated memory. The caller will clear @context afterwards.
1914 	 * This function is only called when roce_gid_table is used.
1915 	 */
1916 	int		           (*del_gid)(struct ib_device *device,
1917 					      u8 port_num,
1918 					      unsigned int index,
1919 					      void **context);
1920 	int		           (*query_pkey)(struct ib_device *device,
1921 						 u8 port_num, u16 index, u16 *pkey);
1922 	int		           (*modify_device)(struct ib_device *device,
1923 						    int device_modify_mask,
1924 						    struct ib_device_modify *device_modify);
1925 	int		           (*modify_port)(struct ib_device *device,
1926 						  u8 port_num, int port_modify_mask,
1927 						  struct ib_port_modify *port_modify);
1928 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
1929 						     struct ib_udata *udata);
1930 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
1931 	int                        (*mmap)(struct ib_ucontext *context,
1932 					   struct vm_area_struct *vma);
1933 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
1934 					       struct ib_ucontext *context,
1935 					       struct ib_udata *udata);
1936 	int                        (*dealloc_pd)(struct ib_pd *pd);
1937 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
1938 						struct ib_ah_attr *ah_attr);
1939 	int                        (*modify_ah)(struct ib_ah *ah,
1940 						struct ib_ah_attr *ah_attr);
1941 	int                        (*query_ah)(struct ib_ah *ah,
1942 					       struct ib_ah_attr *ah_attr);
1943 	int                        (*destroy_ah)(struct ib_ah *ah);
1944 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
1945 						 struct ib_srq_init_attr *srq_init_attr,
1946 						 struct ib_udata *udata);
1947 	int                        (*modify_srq)(struct ib_srq *srq,
1948 						 struct ib_srq_attr *srq_attr,
1949 						 enum ib_srq_attr_mask srq_attr_mask,
1950 						 struct ib_udata *udata);
1951 	int                        (*query_srq)(struct ib_srq *srq,
1952 						struct ib_srq_attr *srq_attr);
1953 	int                        (*destroy_srq)(struct ib_srq *srq);
1954 	int                        (*post_srq_recv)(struct ib_srq *srq,
1955 						    struct ib_recv_wr *recv_wr,
1956 						    struct ib_recv_wr **bad_recv_wr);
1957 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
1958 						struct ib_qp_init_attr *qp_init_attr,
1959 						struct ib_udata *udata);
1960 	int                        (*modify_qp)(struct ib_qp *qp,
1961 						struct ib_qp_attr *qp_attr,
1962 						int qp_attr_mask,
1963 						struct ib_udata *udata);
1964 	int                        (*query_qp)(struct ib_qp *qp,
1965 					       struct ib_qp_attr *qp_attr,
1966 					       int qp_attr_mask,
1967 					       struct ib_qp_init_attr *qp_init_attr);
1968 	int                        (*destroy_qp)(struct ib_qp *qp);
1969 	int                        (*post_send)(struct ib_qp *qp,
1970 						struct ib_send_wr *send_wr,
1971 						struct ib_send_wr **bad_send_wr);
1972 	int                        (*post_recv)(struct ib_qp *qp,
1973 						struct ib_recv_wr *recv_wr,
1974 						struct ib_recv_wr **bad_recv_wr);
1975 	struct ib_cq *             (*create_cq)(struct ib_device *device,
1976 						const struct ib_cq_init_attr *attr,
1977 						struct ib_ucontext *context,
1978 						struct ib_udata *udata);
1979 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
1980 						u16 cq_period);
1981 	int                        (*destroy_cq)(struct ib_cq *cq);
1982 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
1983 						struct ib_udata *udata);
1984 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
1985 					      struct ib_wc *wc);
1986 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
1987 	int                        (*req_notify_cq)(struct ib_cq *cq,
1988 						    enum ib_cq_notify_flags flags);
1989 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
1990 						      int wc_cnt);
1991 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
1992 						 int mr_access_flags);
1993 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
1994 						  u64 start, u64 length,
1995 						  u64 virt_addr,
1996 						  int mr_access_flags,
1997 						  struct ib_udata *udata);
1998 	int			   (*rereg_user_mr)(struct ib_mr *mr,
1999 						    int flags,
2000 						    u64 start, u64 length,
2001 						    u64 virt_addr,
2002 						    int mr_access_flags,
2003 						    struct ib_pd *pd,
2004 						    struct ib_udata *udata);
2005 	int                        (*dereg_mr)(struct ib_mr *mr);
2006 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2007 					       enum ib_mr_type mr_type,
2008 					       u32 max_num_sg);
2009 	int                        (*map_mr_sg)(struct ib_mr *mr,
2010 						struct scatterlist *sg,
2011 						int sg_nents,
2012 						unsigned int *sg_offset);
2013 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2014 					       enum ib_mw_type type,
2015 					       struct ib_udata *udata);
2016 	int                        (*dealloc_mw)(struct ib_mw *mw);
2017 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2018 						int mr_access_flags,
2019 						struct ib_fmr_attr *fmr_attr);
2020 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2021 						   u64 *page_list, int list_len,
2022 						   u64 iova);
2023 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2024 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2025 	int                        (*attach_mcast)(struct ib_qp *qp,
2026 						   union ib_gid *gid,
2027 						   u16 lid);
2028 	int                        (*detach_mcast)(struct ib_qp *qp,
2029 						   union ib_gid *gid,
2030 						   u16 lid);
2031 	int                        (*process_mad)(struct ib_device *device,
2032 						  int process_mad_flags,
2033 						  u8 port_num,
2034 						  const struct ib_wc *in_wc,
2035 						  const struct ib_grh *in_grh,
2036 						  const struct ib_mad_hdr *in_mad,
2037 						  size_t in_mad_size,
2038 						  struct ib_mad_hdr *out_mad,
2039 						  size_t *out_mad_size,
2040 						  u16 *out_mad_pkey_index);
2041 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2042 						 struct ib_ucontext *ucontext,
2043 						 struct ib_udata *udata);
2044 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2045 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2046 						  struct ib_flow_attr
2047 						  *flow_attr,
2048 						  int domain);
2049 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2050 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2051 						      struct ib_mr_status *mr_status);
2052 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2053 	void			   (*drain_rq)(struct ib_qp *qp);
2054 	void			   (*drain_sq)(struct ib_qp *qp);
2055 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2056 							int state);
2057 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2058 						   struct ifla_vf_info *ivf);
2059 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2060 						   struct ifla_vf_stats *stats);
2061 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2062 						  int type);
2063 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2064 						struct ib_wq_init_attr *init_attr,
2065 						struct ib_udata *udata);
2066 	int			   (*destroy_wq)(struct ib_wq *wq);
2067 	int			   (*modify_wq)(struct ib_wq *wq,
2068 						struct ib_wq_attr *attr,
2069 						u32 wq_attr_mask,
2070 						struct ib_udata *udata);
2071 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2072 							   struct ib_rwq_ind_table_init_attr *init_attr,
2073 							   struct ib_udata *udata);
2074 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2075 	struct ib_dma_mapping_ops   *dma_ops;
2076 
2077 	struct module               *owner;
2078 	struct device                dev;
2079 	struct kobject               *ports_parent;
2080 	struct list_head             port_list;
2081 
2082 	enum {
2083 		IB_DEV_UNINITIALIZED,
2084 		IB_DEV_REGISTERED,
2085 		IB_DEV_UNREGISTERED
2086 	}                            reg_state;
2087 
2088 	int			     uverbs_abi_ver;
2089 	u64			     uverbs_cmd_mask;
2090 	u64			     uverbs_ex_cmd_mask;
2091 
2092 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2093 	__be64			     node_guid;
2094 	u32			     local_dma_lkey;
2095 	u16                          is_switch:1;
2096 	u8                           node_type;
2097 	u8                           phys_port_cnt;
2098 	struct ib_device_attr        attrs;
2099 	struct attribute_group	     *hw_stats_ag;
2100 	struct rdma_hw_stats         *hw_stats;
2101 
2102 	/**
2103 	 * The following mandatory functions are used only at device
2104 	 * registration.  Keep functions such as these at the end of this
2105 	 * structure to avoid cache line misses when accessing struct ib_device
2106 	 * in fast paths.
2107 	 */
2108 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2109 	void (*get_dev_fw_str)(struct ib_device *, char *str, size_t str_len);
2110 };
2111 
2112 struct ib_client {
2113 	char  *name;
2114 	void (*add)   (struct ib_device *);
2115 	void (*remove)(struct ib_device *, void *client_data);
2116 
2117 	/* Returns the net_dev belonging to this ib_client and matching the
2118 	 * given parameters.
2119 	 * @dev:	 An RDMA device that the net_dev use for communication.
2120 	 * @port:	 A physical port number on the RDMA device.
2121 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2122 	 * @gid:	 A GID that the net_dev uses to communicate.
2123 	 * @addr:	 An IP address the net_dev is configured with.
2124 	 * @client_data: The device's client data set by ib_set_client_data().
2125 	 *
2126 	 * An ib_client that implements a net_dev on top of RDMA devices
2127 	 * (such as IP over IB) should implement this callback, allowing the
2128 	 * rdma_cm module to find the right net_dev for a given request.
2129 	 *
2130 	 * The caller is responsible for calling dev_put on the returned
2131 	 * netdev. */
2132 	struct net_device *(*get_net_dev_by_params)(
2133 			struct ib_device *dev,
2134 			u8 port,
2135 			u16 pkey,
2136 			const union ib_gid *gid,
2137 			const struct sockaddr *addr,
2138 			void *client_data);
2139 	struct list_head list;
2140 };
2141 
2142 struct ib_device *ib_alloc_device(size_t size);
2143 void ib_dealloc_device(struct ib_device *device);
2144 
2145 void ib_get_device_fw_str(struct ib_device *device, char *str, size_t str_len);
2146 
2147 int ib_register_device(struct ib_device *device,
2148 		       int (*port_callback)(struct ib_device *,
2149 					    u8, struct kobject *));
2150 void ib_unregister_device(struct ib_device *device);
2151 
2152 int ib_register_client   (struct ib_client *client);
2153 void ib_unregister_client(struct ib_client *client);
2154 
2155 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2156 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2157 			 void *data);
2158 
2159 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2160 {
2161 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2162 }
2163 
2164 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2165 {
2166 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2167 }
2168 
2169 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2170 				       size_t offset,
2171 				       size_t len)
2172 {
2173 	const void __user *p = (const char __user *)udata->inbuf + offset;
2174 	bool ret;
2175 	u8 *buf;
2176 
2177 	if (len > USHRT_MAX)
2178 		return false;
2179 
2180 	buf = memdup_user(p, len);
2181 	if (IS_ERR(buf))
2182 		return false;
2183 
2184 	ret = !memchr_inv(buf, 0, len);
2185 	kfree(buf);
2186 	return ret;
2187 }
2188 
2189 /**
2190  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2191  * contains all required attributes and no attributes not allowed for
2192  * the given QP state transition.
2193  * @cur_state: Current QP state
2194  * @next_state: Next QP state
2195  * @type: QP type
2196  * @mask: Mask of supplied QP attributes
2197  * @ll : link layer of port
2198  *
2199  * This function is a helper function that a low-level driver's
2200  * modify_qp method can use to validate the consumer's input.  It
2201  * checks that cur_state and next_state are valid QP states, that a
2202  * transition from cur_state to next_state is allowed by the IB spec,
2203  * and that the attribute mask supplied is allowed for the transition.
2204  */
2205 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2206 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2207 		       enum rdma_link_layer ll);
2208 
2209 int ib_register_event_handler  (struct ib_event_handler *event_handler);
2210 int ib_unregister_event_handler(struct ib_event_handler *event_handler);
2211 void ib_dispatch_event(struct ib_event *event);
2212 
2213 int ib_query_port(struct ib_device *device,
2214 		  u8 port_num, struct ib_port_attr *port_attr);
2215 
2216 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2217 					       u8 port_num);
2218 
2219 /**
2220  * rdma_cap_ib_switch - Check if the device is IB switch
2221  * @device: Device to check
2222  *
2223  * Device driver is responsible for setting is_switch bit on
2224  * in ib_device structure at init time.
2225  *
2226  * Return: true if the device is IB switch.
2227  */
2228 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2229 {
2230 	return device->is_switch;
2231 }
2232 
2233 /**
2234  * rdma_start_port - Return the first valid port number for the device
2235  * specified
2236  *
2237  * @device: Device to be checked
2238  *
2239  * Return start port number
2240  */
2241 static inline u8 rdma_start_port(const struct ib_device *device)
2242 {
2243 	return rdma_cap_ib_switch(device) ? 0 : 1;
2244 }
2245 
2246 /**
2247  * rdma_end_port - Return the last valid port number for the device
2248  * specified
2249  *
2250  * @device: Device to be checked
2251  *
2252  * Return last port number
2253  */
2254 static inline u8 rdma_end_port(const struct ib_device *device)
2255 {
2256 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2257 }
2258 
2259 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2260 {
2261 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2262 }
2263 
2264 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2265 {
2266 	return device->port_immutable[port_num].core_cap_flags &
2267 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2268 }
2269 
2270 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2271 {
2272 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2273 }
2274 
2275 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2276 {
2277 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2278 }
2279 
2280 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2281 {
2282 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2283 }
2284 
2285 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2286 {
2287 	return rdma_protocol_ib(device, port_num) ||
2288 		rdma_protocol_roce(device, port_num);
2289 }
2290 
2291 /**
2292  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2293  * Management Datagrams.
2294  * @device: Device to check
2295  * @port_num: Port number to check
2296  *
2297  * Management Datagrams (MAD) are a required part of the InfiniBand
2298  * specification and are supported on all InfiniBand devices.  A slightly
2299  * extended version are also supported on OPA interfaces.
2300  *
2301  * Return: true if the port supports sending/receiving of MAD packets.
2302  */
2303 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2304 {
2305 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2306 }
2307 
2308 /**
2309  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2310  * Management Datagrams.
2311  * @device: Device to check
2312  * @port_num: Port number to check
2313  *
2314  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2315  * datagrams with their own versions.  These OPA MADs share many but not all of
2316  * the characteristics of InfiniBand MADs.
2317  *
2318  * OPA MADs differ in the following ways:
2319  *
2320  *    1) MADs are variable size up to 2K
2321  *       IBTA defined MADs remain fixed at 256 bytes
2322  *    2) OPA SMPs must carry valid PKeys
2323  *    3) OPA SMP packets are a different format
2324  *
2325  * Return: true if the port supports OPA MAD packet formats.
2326  */
2327 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2328 {
2329 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2330 		== RDMA_CORE_CAP_OPA_MAD;
2331 }
2332 
2333 /**
2334  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2335  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2336  * @device: Device to check
2337  * @port_num: Port number to check
2338  *
2339  * Each InfiniBand node is required to provide a Subnet Management Agent
2340  * that the subnet manager can access.  Prior to the fabric being fully
2341  * configured by the subnet manager, the SMA is accessed via a well known
2342  * interface called the Subnet Management Interface (SMI).  This interface
2343  * uses directed route packets to communicate with the SM to get around the
2344  * chicken and egg problem of the SM needing to know what's on the fabric
2345  * in order to configure the fabric, and needing to configure the fabric in
2346  * order to send packets to the devices on the fabric.  These directed
2347  * route packets do not need the fabric fully configured in order to reach
2348  * their destination.  The SMI is the only method allowed to send
2349  * directed route packets on an InfiniBand fabric.
2350  *
2351  * Return: true if the port provides an SMI.
2352  */
2353 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2354 {
2355 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2356 }
2357 
2358 /**
2359  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2360  * Communication Manager.
2361  * @device: Device to check
2362  * @port_num: Port number to check
2363  *
2364  * The InfiniBand Communication Manager is one of many pre-defined General
2365  * Service Agents (GSA) that are accessed via the General Service
2366  * Interface (GSI).  It's role is to facilitate establishment of connections
2367  * between nodes as well as other management related tasks for established
2368  * connections.
2369  *
2370  * Return: true if the port supports an IB CM (this does not guarantee that
2371  * a CM is actually running however).
2372  */
2373 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2374 {
2375 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2376 }
2377 
2378 /**
2379  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2380  * Communication Manager.
2381  * @device: Device to check
2382  * @port_num: Port number to check
2383  *
2384  * Similar to above, but specific to iWARP connections which have a different
2385  * managment protocol than InfiniBand.
2386  *
2387  * Return: true if the port supports an iWARP CM (this does not guarantee that
2388  * a CM is actually running however).
2389  */
2390 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2391 {
2392 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2393 }
2394 
2395 /**
2396  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2397  * Subnet Administration.
2398  * @device: Device to check
2399  * @port_num: Port number to check
2400  *
2401  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2402  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2403  * fabrics, devices should resolve routes to other hosts by contacting the
2404  * SA to query the proper route.
2405  *
2406  * Return: true if the port should act as a client to the fabric Subnet
2407  * Administration interface.  This does not imply that the SA service is
2408  * running locally.
2409  */
2410 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2411 {
2412 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2413 }
2414 
2415 /**
2416  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2417  * Multicast.
2418  * @device: Device to check
2419  * @port_num: Port number to check
2420  *
2421  * InfiniBand multicast registration is more complex than normal IPv4 or
2422  * IPv6 multicast registration.  Each Host Channel Adapter must register
2423  * with the Subnet Manager when it wishes to join a multicast group.  It
2424  * should do so only once regardless of how many queue pairs it subscribes
2425  * to this group.  And it should leave the group only after all queue pairs
2426  * attached to the group have been detached.
2427  *
2428  * Return: true if the port must undertake the additional adminstrative
2429  * overhead of registering/unregistering with the SM and tracking of the
2430  * total number of queue pairs attached to the multicast group.
2431  */
2432 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2433 {
2434 	return rdma_cap_ib_sa(device, port_num);
2435 }
2436 
2437 /**
2438  * rdma_cap_af_ib - Check if the port of device has the capability
2439  * Native Infiniband Address.
2440  * @device: Device to check
2441  * @port_num: Port number to check
2442  *
2443  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2444  * GID.  RoCE uses a different mechanism, but still generates a GID via
2445  * a prescribed mechanism and port specific data.
2446  *
2447  * Return: true if the port uses a GID address to identify devices on the
2448  * network.
2449  */
2450 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2451 {
2452 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2453 }
2454 
2455 /**
2456  * rdma_cap_eth_ah - Check if the port of device has the capability
2457  * Ethernet Address Handle.
2458  * @device: Device to check
2459  * @port_num: Port number to check
2460  *
2461  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2462  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2463  * port.  Normally, packet headers are generated by the sending host
2464  * adapter, but when sending connectionless datagrams, we must manually
2465  * inject the proper headers for the fabric we are communicating over.
2466  *
2467  * Return: true if we are running as a RoCE port and must force the
2468  * addition of a Global Route Header built from our Ethernet Address
2469  * Handle into our header list for connectionless packets.
2470  */
2471 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2472 {
2473 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2474 }
2475 
2476 /**
2477  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2478  *
2479  * @device: Device
2480  * @port_num: Port number
2481  *
2482  * This MAD size includes the MAD headers and MAD payload.  No other headers
2483  * are included.
2484  *
2485  * Return the max MAD size required by the Port.  Will return 0 if the port
2486  * does not support MADs
2487  */
2488 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2489 {
2490 	return device->port_immutable[port_num].max_mad_size;
2491 }
2492 
2493 /**
2494  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2495  * @device: Device to check
2496  * @port_num: Port number to check
2497  *
2498  * RoCE GID table mechanism manages the various GIDs for a device.
2499  *
2500  * NOTE: if allocating the port's GID table has failed, this call will still
2501  * return true, but any RoCE GID table API will fail.
2502  *
2503  * Return: true if the port uses RoCE GID table mechanism in order to manage
2504  * its GIDs.
2505  */
2506 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2507 					   u8 port_num)
2508 {
2509 	return rdma_protocol_roce(device, port_num) &&
2510 		device->add_gid && device->del_gid;
2511 }
2512 
2513 /*
2514  * Check if the device supports READ W/ INVALIDATE.
2515  */
2516 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2517 {
2518 	/*
2519 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2520 	 * has support for it yet.
2521 	 */
2522 	return rdma_protocol_iwarp(dev, port_num);
2523 }
2524 
2525 int ib_query_gid(struct ib_device *device,
2526 		 u8 port_num, int index, union ib_gid *gid,
2527 		 struct ib_gid_attr *attr);
2528 
2529 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2530 			 int state);
2531 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2532 		     struct ifla_vf_info *info);
2533 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2534 		    struct ifla_vf_stats *stats);
2535 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2536 		   int type);
2537 
2538 int ib_query_pkey(struct ib_device *device,
2539 		  u8 port_num, u16 index, u16 *pkey);
2540 
2541 int ib_modify_device(struct ib_device *device,
2542 		     int device_modify_mask,
2543 		     struct ib_device_modify *device_modify);
2544 
2545 int ib_modify_port(struct ib_device *device,
2546 		   u8 port_num, int port_modify_mask,
2547 		   struct ib_port_modify *port_modify);
2548 
2549 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2550 		enum ib_gid_type gid_type, struct net_device *ndev,
2551 		u8 *port_num, u16 *index);
2552 
2553 int ib_find_pkey(struct ib_device *device,
2554 		 u8 port_num, u16 pkey, u16 *index);
2555 
2556 enum ib_pd_flags {
2557 	/*
2558 	 * Create a memory registration for all memory in the system and place
2559 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2560 	 * ULPs to avoid the overhead of dynamic MRs.
2561 	 *
2562 	 * This flag is generally considered unsafe and must only be used in
2563 	 * extremly trusted environments.  Every use of it will log a warning
2564 	 * in the kernel log.
2565 	 */
2566 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2567 };
2568 
2569 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2570 		const char *caller);
2571 #define ib_alloc_pd(device, flags) \
2572 	__ib_alloc_pd((device), (flags), __func__)
2573 void ib_dealloc_pd(struct ib_pd *pd);
2574 
2575 /**
2576  * ib_create_ah - Creates an address handle for the given address vector.
2577  * @pd: The protection domain associated with the address handle.
2578  * @ah_attr: The attributes of the address vector.
2579  *
2580  * The address handle is used to reference a local or global destination
2581  * in all UD QP post sends.
2582  */
2583 struct ib_ah *ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
2584 
2585 /**
2586  * ib_init_ah_from_wc - Initializes address handle attributes from a
2587  *   work completion.
2588  * @device: Device on which the received message arrived.
2589  * @port_num: Port on which the received message arrived.
2590  * @wc: Work completion associated with the received message.
2591  * @grh: References the received global route header.  This parameter is
2592  *   ignored unless the work completion indicates that the GRH is valid.
2593  * @ah_attr: Returned attributes that can be used when creating an address
2594  *   handle for replying to the message.
2595  */
2596 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2597 		       const struct ib_wc *wc, const struct ib_grh *grh,
2598 		       struct ib_ah_attr *ah_attr);
2599 
2600 /**
2601  * ib_create_ah_from_wc - Creates an address handle associated with the
2602  *   sender of the specified work completion.
2603  * @pd: The protection domain associated with the address handle.
2604  * @wc: Work completion information associated with a received message.
2605  * @grh: References the received global route header.  This parameter is
2606  *   ignored unless the work completion indicates that the GRH is valid.
2607  * @port_num: The outbound port number to associate with the address.
2608  *
2609  * The address handle is used to reference a local or global destination
2610  * in all UD QP post sends.
2611  */
2612 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2613 				   const struct ib_grh *grh, u8 port_num);
2614 
2615 /**
2616  * ib_modify_ah - Modifies the address vector associated with an address
2617  *   handle.
2618  * @ah: The address handle to modify.
2619  * @ah_attr: The new address vector attributes to associate with the
2620  *   address handle.
2621  */
2622 int ib_modify_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2623 
2624 /**
2625  * ib_query_ah - Queries the address vector associated with an address
2626  *   handle.
2627  * @ah: The address handle to query.
2628  * @ah_attr: The address vector attributes associated with the address
2629  *   handle.
2630  */
2631 int ib_query_ah(struct ib_ah *ah, struct ib_ah_attr *ah_attr);
2632 
2633 /**
2634  * ib_destroy_ah - Destroys an address handle.
2635  * @ah: The address handle to destroy.
2636  */
2637 int ib_destroy_ah(struct ib_ah *ah);
2638 
2639 /**
2640  * ib_create_srq - Creates a SRQ associated with the specified protection
2641  *   domain.
2642  * @pd: The protection domain associated with the SRQ.
2643  * @srq_init_attr: A list of initial attributes required to create the
2644  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2645  *   the actual capabilities of the created SRQ.
2646  *
2647  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2648  * requested size of the SRQ, and set to the actual values allocated
2649  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2650  * will always be at least as large as the requested values.
2651  */
2652 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2653 			     struct ib_srq_init_attr *srq_init_attr);
2654 
2655 /**
2656  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2657  * @srq: The SRQ to modify.
2658  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2659  *   the current values of selected SRQ attributes are returned.
2660  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2661  *   are being modified.
2662  *
2663  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2664  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2665  * the number of receives queued drops below the limit.
2666  */
2667 int ib_modify_srq(struct ib_srq *srq,
2668 		  struct ib_srq_attr *srq_attr,
2669 		  enum ib_srq_attr_mask srq_attr_mask);
2670 
2671 /**
2672  * ib_query_srq - Returns the attribute list and current values for the
2673  *   specified SRQ.
2674  * @srq: The SRQ to query.
2675  * @srq_attr: The attributes of the specified SRQ.
2676  */
2677 int ib_query_srq(struct ib_srq *srq,
2678 		 struct ib_srq_attr *srq_attr);
2679 
2680 /**
2681  * ib_destroy_srq - Destroys the specified SRQ.
2682  * @srq: The SRQ to destroy.
2683  */
2684 int ib_destroy_srq(struct ib_srq *srq);
2685 
2686 /**
2687  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2688  * @srq: The SRQ to post the work request on.
2689  * @recv_wr: A list of work requests to post on the receive queue.
2690  * @bad_recv_wr: On an immediate failure, this parameter will reference
2691  *   the work request that failed to be posted on the QP.
2692  */
2693 static inline int ib_post_srq_recv(struct ib_srq *srq,
2694 				   struct ib_recv_wr *recv_wr,
2695 				   struct ib_recv_wr **bad_recv_wr)
2696 {
2697 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2698 }
2699 
2700 /**
2701  * ib_create_qp - Creates a QP associated with the specified protection
2702  *   domain.
2703  * @pd: The protection domain associated with the QP.
2704  * @qp_init_attr: A list of initial attributes required to create the
2705  *   QP.  If QP creation succeeds, then the attributes are updated to
2706  *   the actual capabilities of the created QP.
2707  */
2708 struct ib_qp *ib_create_qp(struct ib_pd *pd,
2709 			   struct ib_qp_init_attr *qp_init_attr);
2710 
2711 /**
2712  * ib_modify_qp - Modifies the attributes for the specified QP and then
2713  *   transitions the QP to the given state.
2714  * @qp: The QP to modify.
2715  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
2716  *   the current values of selected QP attributes are returned.
2717  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
2718  *   are being modified.
2719  */
2720 int ib_modify_qp(struct ib_qp *qp,
2721 		 struct ib_qp_attr *qp_attr,
2722 		 int qp_attr_mask);
2723 
2724 /**
2725  * ib_query_qp - Returns the attribute list and current values for the
2726  *   specified QP.
2727  * @qp: The QP to query.
2728  * @qp_attr: The attributes of the specified QP.
2729  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
2730  * @qp_init_attr: Additional attributes of the selected QP.
2731  *
2732  * The qp_attr_mask may be used to limit the query to gathering only the
2733  * selected attributes.
2734  */
2735 int ib_query_qp(struct ib_qp *qp,
2736 		struct ib_qp_attr *qp_attr,
2737 		int qp_attr_mask,
2738 		struct ib_qp_init_attr *qp_init_attr);
2739 
2740 /**
2741  * ib_destroy_qp - Destroys the specified QP.
2742  * @qp: The QP to destroy.
2743  */
2744 int ib_destroy_qp(struct ib_qp *qp);
2745 
2746 /**
2747  * ib_open_qp - Obtain a reference to an existing sharable QP.
2748  * @xrcd - XRC domain
2749  * @qp_open_attr: Attributes identifying the QP to open.
2750  *
2751  * Returns a reference to a sharable QP.
2752  */
2753 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
2754 			 struct ib_qp_open_attr *qp_open_attr);
2755 
2756 /**
2757  * ib_close_qp - Release an external reference to a QP.
2758  * @qp: The QP handle to release
2759  *
2760  * The opened QP handle is released by the caller.  The underlying
2761  * shared QP is not destroyed until all internal references are released.
2762  */
2763 int ib_close_qp(struct ib_qp *qp);
2764 
2765 /**
2766  * ib_post_send - Posts a list of work requests to the send queue of
2767  *   the specified QP.
2768  * @qp: The QP to post the work request on.
2769  * @send_wr: A list of work requests to post on the send queue.
2770  * @bad_send_wr: On an immediate failure, this parameter will reference
2771  *   the work request that failed to be posted on the QP.
2772  *
2773  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
2774  * error is returned, the QP state shall not be affected,
2775  * ib_post_send() will return an immediate error after queueing any
2776  * earlier work requests in the list.
2777  */
2778 static inline int ib_post_send(struct ib_qp *qp,
2779 			       struct ib_send_wr *send_wr,
2780 			       struct ib_send_wr **bad_send_wr)
2781 {
2782 	return qp->device->post_send(qp, send_wr, bad_send_wr);
2783 }
2784 
2785 /**
2786  * ib_post_recv - Posts a list of work requests to the receive queue of
2787  *   the specified QP.
2788  * @qp: The QP to post the work request on.
2789  * @recv_wr: A list of work requests to post on the receive queue.
2790  * @bad_recv_wr: On an immediate failure, this parameter will reference
2791  *   the work request that failed to be posted on the QP.
2792  */
2793 static inline int ib_post_recv(struct ib_qp *qp,
2794 			       struct ib_recv_wr *recv_wr,
2795 			       struct ib_recv_wr **bad_recv_wr)
2796 {
2797 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
2798 }
2799 
2800 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
2801 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
2802 void ib_free_cq(struct ib_cq *cq);
2803 
2804 /**
2805  * ib_create_cq - Creates a CQ on the specified device.
2806  * @device: The device on which to create the CQ.
2807  * @comp_handler: A user-specified callback that is invoked when a
2808  *   completion event occurs on the CQ.
2809  * @event_handler: A user-specified callback that is invoked when an
2810  *   asynchronous event not associated with a completion occurs on the CQ.
2811  * @cq_context: Context associated with the CQ returned to the user via
2812  *   the associated completion and event handlers.
2813  * @cq_attr: The attributes the CQ should be created upon.
2814  *
2815  * Users can examine the cq structure to determine the actual CQ size.
2816  */
2817 struct ib_cq *ib_create_cq(struct ib_device *device,
2818 			   ib_comp_handler comp_handler,
2819 			   void (*event_handler)(struct ib_event *, void *),
2820 			   void *cq_context,
2821 			   const struct ib_cq_init_attr *cq_attr);
2822 
2823 /**
2824  * ib_resize_cq - Modifies the capacity of the CQ.
2825  * @cq: The CQ to resize.
2826  * @cqe: The minimum size of the CQ.
2827  *
2828  * Users can examine the cq structure to determine the actual CQ size.
2829  */
2830 int ib_resize_cq(struct ib_cq *cq, int cqe);
2831 
2832 /**
2833  * ib_modify_cq - Modifies moderation params of the CQ
2834  * @cq: The CQ to modify.
2835  * @cq_count: number of CQEs that will trigger an event
2836  * @cq_period: max period of time in usec before triggering an event
2837  *
2838  */
2839 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
2840 
2841 /**
2842  * ib_destroy_cq - Destroys the specified CQ.
2843  * @cq: The CQ to destroy.
2844  */
2845 int ib_destroy_cq(struct ib_cq *cq);
2846 
2847 /**
2848  * ib_poll_cq - poll a CQ for completion(s)
2849  * @cq:the CQ being polled
2850  * @num_entries:maximum number of completions to return
2851  * @wc:array of at least @num_entries &struct ib_wc where completions
2852  *   will be returned
2853  *
2854  * Poll a CQ for (possibly multiple) completions.  If the return value
2855  * is < 0, an error occurred.  If the return value is >= 0, it is the
2856  * number of completions returned.  If the return value is
2857  * non-negative and < num_entries, then the CQ was emptied.
2858  */
2859 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
2860 			     struct ib_wc *wc)
2861 {
2862 	return cq->device->poll_cq(cq, num_entries, wc);
2863 }
2864 
2865 /**
2866  * ib_peek_cq - Returns the number of unreaped completions currently
2867  *   on the specified CQ.
2868  * @cq: The CQ to peek.
2869  * @wc_cnt: A minimum number of unreaped completions to check for.
2870  *
2871  * If the number of unreaped completions is greater than or equal to wc_cnt,
2872  * this function returns wc_cnt, otherwise, it returns the actual number of
2873  * unreaped completions.
2874  */
2875 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
2876 
2877 /**
2878  * ib_req_notify_cq - Request completion notification on a CQ.
2879  * @cq: The CQ to generate an event for.
2880  * @flags:
2881  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
2882  *   to request an event on the next solicited event or next work
2883  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
2884  *   may also be |ed in to request a hint about missed events, as
2885  *   described below.
2886  *
2887  * Return Value:
2888  *    < 0 means an error occurred while requesting notification
2889  *   == 0 means notification was requested successfully, and if
2890  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
2891  *        were missed and it is safe to wait for another event.  In
2892  *        this case is it guaranteed that any work completions added
2893  *        to the CQ since the last CQ poll will trigger a completion
2894  *        notification event.
2895  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
2896  *        in.  It means that the consumer must poll the CQ again to
2897  *        make sure it is empty to avoid missing an event because of a
2898  *        race between requesting notification and an entry being
2899  *        added to the CQ.  This return value means it is possible
2900  *        (but not guaranteed) that a work completion has been added
2901  *        to the CQ since the last poll without triggering a
2902  *        completion notification event.
2903  */
2904 static inline int ib_req_notify_cq(struct ib_cq *cq,
2905 				   enum ib_cq_notify_flags flags)
2906 {
2907 	return cq->device->req_notify_cq(cq, flags);
2908 }
2909 
2910 /**
2911  * ib_req_ncomp_notif - Request completion notification when there are
2912  *   at least the specified number of unreaped completions on the CQ.
2913  * @cq: The CQ to generate an event for.
2914  * @wc_cnt: The number of unreaped completions that should be on the
2915  *   CQ before an event is generated.
2916  */
2917 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
2918 {
2919 	return cq->device->req_ncomp_notif ?
2920 		cq->device->req_ncomp_notif(cq, wc_cnt) :
2921 		-ENOSYS;
2922 }
2923 
2924 /**
2925  * ib_dma_mapping_error - check a DMA addr for error
2926  * @dev: The device for which the dma_addr was created
2927  * @dma_addr: The DMA address to check
2928  */
2929 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2930 {
2931 	if (dev->dma_ops)
2932 		return dev->dma_ops->mapping_error(dev, dma_addr);
2933 	return dma_mapping_error(dev->dma_device, dma_addr);
2934 }
2935 
2936 /**
2937  * ib_dma_map_single - Map a kernel virtual address to DMA address
2938  * @dev: The device for which the dma_addr is to be created
2939  * @cpu_addr: The kernel virtual address
2940  * @size: The size of the region in bytes
2941  * @direction: The direction of the DMA
2942  */
2943 static inline u64 ib_dma_map_single(struct ib_device *dev,
2944 				    void *cpu_addr, size_t size,
2945 				    enum dma_data_direction direction)
2946 {
2947 	if (dev->dma_ops)
2948 		return dev->dma_ops->map_single(dev, cpu_addr, size, direction);
2949 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
2950 }
2951 
2952 /**
2953  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
2954  * @dev: The device for which the DMA address was created
2955  * @addr: The DMA address
2956  * @size: The size of the region in bytes
2957  * @direction: The direction of the DMA
2958  */
2959 static inline void ib_dma_unmap_single(struct ib_device *dev,
2960 				       u64 addr, size_t size,
2961 				       enum dma_data_direction direction)
2962 {
2963 	if (dev->dma_ops)
2964 		dev->dma_ops->unmap_single(dev, addr, size, direction);
2965 	else
2966 		dma_unmap_single(dev->dma_device, addr, size, direction);
2967 }
2968 
2969 static inline u64 ib_dma_map_single_attrs(struct ib_device *dev,
2970 					  void *cpu_addr, size_t size,
2971 					  enum dma_data_direction direction,
2972 					  struct dma_attrs *dma_attrs)
2973 {
2974 	return dma_map_single_attrs(dev->dma_device, cpu_addr, size,
2975 				    direction, dma_attrs);
2976 }
2977 
2978 static inline void ib_dma_unmap_single_attrs(struct ib_device *dev,
2979 					     u64 addr, size_t size,
2980 					     enum dma_data_direction direction,
2981 					     struct dma_attrs *dma_attrs)
2982 {
2983 	return dma_unmap_single_attrs(dev->dma_device, addr, size,
2984 				      direction, dma_attrs);
2985 }
2986 
2987 /**
2988  * ib_dma_map_page - Map a physical page to DMA address
2989  * @dev: The device for which the dma_addr is to be created
2990  * @page: The page to be mapped
2991  * @offset: The offset within the page
2992  * @size: The size of the region in bytes
2993  * @direction: The direction of the DMA
2994  */
2995 static inline u64 ib_dma_map_page(struct ib_device *dev,
2996 				  struct page *page,
2997 				  unsigned long offset,
2998 				  size_t size,
2999 					 enum dma_data_direction direction)
3000 {
3001 	if (dev->dma_ops)
3002 		return dev->dma_ops->map_page(dev, page, offset, size, direction);
3003 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3004 }
3005 
3006 /**
3007  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3008  * @dev: The device for which the DMA address was created
3009  * @addr: The DMA address
3010  * @size: The size of the region in bytes
3011  * @direction: The direction of the DMA
3012  */
3013 static inline void ib_dma_unmap_page(struct ib_device *dev,
3014 				     u64 addr, size_t size,
3015 				     enum dma_data_direction direction)
3016 {
3017 	if (dev->dma_ops)
3018 		dev->dma_ops->unmap_page(dev, addr, size, direction);
3019 	else
3020 		dma_unmap_page(dev->dma_device, addr, size, direction);
3021 }
3022 
3023 /**
3024  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3025  * @dev: The device for which the DMA addresses are to be created
3026  * @sg: The array of scatter/gather entries
3027  * @nents: The number of scatter/gather entries
3028  * @direction: The direction of the DMA
3029  */
3030 static inline int ib_dma_map_sg(struct ib_device *dev,
3031 				struct scatterlist *sg, int nents,
3032 				enum dma_data_direction direction)
3033 {
3034 	if (dev->dma_ops)
3035 		return dev->dma_ops->map_sg(dev, sg, nents, direction);
3036 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3037 }
3038 
3039 /**
3040  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3041  * @dev: The device for which the DMA addresses were created
3042  * @sg: The array of scatter/gather entries
3043  * @nents: The number of scatter/gather entries
3044  * @direction: The direction of the DMA
3045  */
3046 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3047 				   struct scatterlist *sg, int nents,
3048 				   enum dma_data_direction direction)
3049 {
3050 	if (dev->dma_ops)
3051 		dev->dma_ops->unmap_sg(dev, sg, nents, direction);
3052 	else
3053 		dma_unmap_sg(dev->dma_device, sg, nents, direction);
3054 }
3055 
3056 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3057 				      struct scatterlist *sg, int nents,
3058 				      enum dma_data_direction direction,
3059 				      struct dma_attrs *dma_attrs)
3060 {
3061 	if (dev->dma_ops)
3062 		return dev->dma_ops->map_sg_attrs(dev, sg, nents, direction,
3063 						  dma_attrs);
3064 	else
3065 		return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3066 					dma_attrs);
3067 }
3068 
3069 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3070 					 struct scatterlist *sg, int nents,
3071 					 enum dma_data_direction direction,
3072 					 struct dma_attrs *dma_attrs)
3073 {
3074 	if (dev->dma_ops)
3075 		return dev->dma_ops->unmap_sg_attrs(dev, sg, nents, direction,
3076 						  dma_attrs);
3077 	else
3078 		dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction,
3079 				   dma_attrs);
3080 }
3081 /**
3082  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3083  * @dev: The device for which the DMA addresses were created
3084  * @sg: The scatter/gather entry
3085  *
3086  * Note: this function is obsolete. To do: change all occurrences of
3087  * ib_sg_dma_address() into sg_dma_address().
3088  */
3089 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3090 				    struct scatterlist *sg)
3091 {
3092 	return sg_dma_address(sg);
3093 }
3094 
3095 /**
3096  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3097  * @dev: The device for which the DMA addresses were created
3098  * @sg: The scatter/gather entry
3099  *
3100  * Note: this function is obsolete. To do: change all occurrences of
3101  * ib_sg_dma_len() into sg_dma_len().
3102  */
3103 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3104 					 struct scatterlist *sg)
3105 {
3106 	return sg_dma_len(sg);
3107 }
3108 
3109 /**
3110  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3111  * @dev: The device for which the DMA address was created
3112  * @addr: The DMA address
3113  * @size: The size of the region in bytes
3114  * @dir: The direction of the DMA
3115  */
3116 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3117 					      u64 addr,
3118 					      size_t size,
3119 					      enum dma_data_direction dir)
3120 {
3121 	if (dev->dma_ops)
3122 		dev->dma_ops->sync_single_for_cpu(dev, addr, size, dir);
3123 	else
3124 		dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3125 }
3126 
3127 /**
3128  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3129  * @dev: The device for which the DMA address was created
3130  * @addr: The DMA address
3131  * @size: The size of the region in bytes
3132  * @dir: The direction of the DMA
3133  */
3134 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3135 						 u64 addr,
3136 						 size_t size,
3137 						 enum dma_data_direction dir)
3138 {
3139 	if (dev->dma_ops)
3140 		dev->dma_ops->sync_single_for_device(dev, addr, size, dir);
3141 	else
3142 		dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3143 }
3144 
3145 /**
3146  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3147  * @dev: The device for which the DMA address is requested
3148  * @size: The size of the region to allocate in bytes
3149  * @dma_handle: A pointer for returning the DMA address of the region
3150  * @flag: memory allocator flags
3151  */
3152 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3153 					   size_t size,
3154 					   u64 *dma_handle,
3155 					   gfp_t flag)
3156 {
3157 	if (dev->dma_ops)
3158 		return dev->dma_ops->alloc_coherent(dev, size, dma_handle, flag);
3159 	else {
3160 		dma_addr_t handle;
3161 		void *ret;
3162 
3163 		ret = dma_alloc_coherent(dev->dma_device, size, &handle, flag);
3164 		*dma_handle = handle;
3165 		return ret;
3166 	}
3167 }
3168 
3169 /**
3170  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3171  * @dev: The device for which the DMA addresses were allocated
3172  * @size: The size of the region
3173  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3174  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3175  */
3176 static inline void ib_dma_free_coherent(struct ib_device *dev,
3177 					size_t size, void *cpu_addr,
3178 					u64 dma_handle)
3179 {
3180 	if (dev->dma_ops)
3181 		dev->dma_ops->free_coherent(dev, size, cpu_addr, dma_handle);
3182 	else
3183 		dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3184 }
3185 
3186 /**
3187  * ib_dereg_mr - Deregisters a memory region and removes it from the
3188  *   HCA translation table.
3189  * @mr: The memory region to deregister.
3190  *
3191  * This function can fail, if the memory region has memory windows bound to it.
3192  */
3193 int ib_dereg_mr(struct ib_mr *mr);
3194 
3195 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3196 			  enum ib_mr_type mr_type,
3197 			  u32 max_num_sg);
3198 
3199 /**
3200  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3201  *   R_Key and L_Key.
3202  * @mr - struct ib_mr pointer to be updated.
3203  * @newkey - new key to be used.
3204  */
3205 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3206 {
3207 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3208 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3209 }
3210 
3211 /**
3212  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3213  * for calculating a new rkey for type 2 memory windows.
3214  * @rkey - the rkey to increment.
3215  */
3216 static inline u32 ib_inc_rkey(u32 rkey)
3217 {
3218 	const u32 mask = 0x000000ff;
3219 	return ((rkey + 1) & mask) | (rkey & ~mask);
3220 }
3221 
3222 /**
3223  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3224  * @pd: The protection domain associated with the unmapped region.
3225  * @mr_access_flags: Specifies the memory access rights.
3226  * @fmr_attr: Attributes of the unmapped region.
3227  *
3228  * A fast memory region must be mapped before it can be used as part of
3229  * a work request.
3230  */
3231 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3232 			    int mr_access_flags,
3233 			    struct ib_fmr_attr *fmr_attr);
3234 
3235 /**
3236  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3237  * @fmr: The fast memory region to associate with the pages.
3238  * @page_list: An array of physical pages to map to the fast memory region.
3239  * @list_len: The number of pages in page_list.
3240  * @iova: The I/O virtual address to use with the mapped region.
3241  */
3242 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3243 				  u64 *page_list, int list_len,
3244 				  u64 iova)
3245 {
3246 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3247 }
3248 
3249 /**
3250  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3251  * @fmr_list: A linked list of fast memory regions to unmap.
3252  */
3253 int ib_unmap_fmr(struct list_head *fmr_list);
3254 
3255 /**
3256  * ib_dealloc_fmr - Deallocates a fast memory region.
3257  * @fmr: The fast memory region to deallocate.
3258  */
3259 int ib_dealloc_fmr(struct ib_fmr *fmr);
3260 
3261 /**
3262  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3263  * @qp: QP to attach to the multicast group.  The QP must be type
3264  *   IB_QPT_UD.
3265  * @gid: Multicast group GID.
3266  * @lid: Multicast group LID in host byte order.
3267  *
3268  * In order to send and receive multicast packets, subnet
3269  * administration must have created the multicast group and configured
3270  * the fabric appropriately.  The port associated with the specified
3271  * QP must also be a member of the multicast group.
3272  */
3273 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3274 
3275 /**
3276  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3277  * @qp: QP to detach from the multicast group.
3278  * @gid: Multicast group GID.
3279  * @lid: Multicast group LID in host byte order.
3280  */
3281 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3282 
3283 /**
3284  * ib_alloc_xrcd - Allocates an XRC domain.
3285  * @device: The device on which to allocate the XRC domain.
3286  */
3287 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3288 
3289 /**
3290  * ib_dealloc_xrcd - Deallocates an XRC domain.
3291  * @xrcd: The XRC domain to deallocate.
3292  */
3293 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3294 
3295 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3296 			       struct ib_flow_attr *flow_attr, int domain);
3297 int ib_destroy_flow(struct ib_flow *flow_id);
3298 
3299 static inline int ib_check_mr_access(int flags)
3300 {
3301 	/*
3302 	 * Local write permission is required if remote write or
3303 	 * remote atomic permission is also requested.
3304 	 */
3305 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3306 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3307 		return -EINVAL;
3308 
3309 	return 0;
3310 }
3311 
3312 /**
3313  * ib_check_mr_status: lightweight check of MR status.
3314  *     This routine may provide status checks on a selected
3315  *     ib_mr. first use is for signature status check.
3316  *
3317  * @mr: A memory region.
3318  * @check_mask: Bitmask of which checks to perform from
3319  *     ib_mr_status_check enumeration.
3320  * @mr_status: The container of relevant status checks.
3321  *     failed checks will be indicated in the status bitmask
3322  *     and the relevant info shall be in the error item.
3323  */
3324 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3325 		       struct ib_mr_status *mr_status);
3326 
3327 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3328 					    u16 pkey, const union ib_gid *gid,
3329 					    const struct sockaddr *addr);
3330 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3331 			   struct ib_wq_init_attr *init_attr);
3332 int ib_destroy_wq(struct ib_wq *wq);
3333 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3334 		 u32 wq_attr_mask);
3335 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3336 						 struct ib_rwq_ind_table_init_attr*
3337 						 wq_ind_table_init_attr);
3338 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3339 
3340 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3341 		 unsigned int *sg_offset, unsigned int page_size);
3342 
3343 static inline int
3344 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3345 		  unsigned int *sg_offset, unsigned int page_size)
3346 {
3347 	int n;
3348 
3349 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3350 	mr->iova = 0;
3351 
3352 	return n;
3353 }
3354 
3355 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3356 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3357 
3358 void ib_drain_rq(struct ib_qp *qp);
3359 void ib_drain_sq(struct ib_qp *qp);
3360 void ib_drain_qp(struct ib_qp *qp);
3361 #endif /* IB_VERBS_H */
3362