xref: /openbmc/linux/include/rdma/ib_verbs.h (revision 711aab1d)
1 /*
2  * Copyright (c) 2004 Mellanox Technologies Ltd.  All rights reserved.
3  * Copyright (c) 2004 Infinicon Corporation.  All rights reserved.
4  * Copyright (c) 2004 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2004 Topspin Corporation.  All rights reserved.
6  * Copyright (c) 2004 Voltaire Corporation.  All rights reserved.
7  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
8  * Copyright (c) 2005, 2006, 2007 Cisco Systems.  All rights reserved.
9  *
10  * This software is available to you under a choice of one of two
11  * licenses.  You may choose to be licensed under the terms of the GNU
12  * General Public License (GPL) Version 2, available from the file
13  * COPYING in the main directory of this source tree, or the
14  * OpenIB.org BSD license below:
15  *
16  *     Redistribution and use in source and binary forms, with or
17  *     without modification, are permitted provided that the following
18  *     conditions are met:
19  *
20  *      - Redistributions of source code must retain the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer.
23  *
24  *      - Redistributions in binary form must reproduce the above
25  *        copyright notice, this list of conditions and the following
26  *        disclaimer in the documentation and/or other materials
27  *        provided with the distribution.
28  *
29  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
30  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
31  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
32  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
33  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
34  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
35  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
36  * SOFTWARE.
37  */
38 
39 #if !defined(IB_VERBS_H)
40 #define IB_VERBS_H
41 
42 #include <linux/types.h>
43 #include <linux/device.h>
44 #include <linux/mm.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/kref.h>
47 #include <linux/list.h>
48 #include <linux/rwsem.h>
49 #include <linux/scatterlist.h>
50 #include <linux/workqueue.h>
51 #include <linux/socket.h>
52 #include <linux/irq_poll.h>
53 #include <uapi/linux/if_ether.h>
54 #include <net/ipv6.h>
55 #include <net/ip.h>
56 #include <linux/string.h>
57 #include <linux/slab.h>
58 #include <linux/netdevice.h>
59 
60 #include <linux/if_link.h>
61 #include <linux/atomic.h>
62 #include <linux/mmu_notifier.h>
63 #include <linux/uaccess.h>
64 #include <linux/cgroup_rdma.h>
65 #include <uapi/rdma/ib_user_verbs.h>
66 
67 #define IB_FW_VERSION_NAME_MAX	ETHTOOL_FWVERS_LEN
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 	/* Not in use, former 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 	/* 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 	/* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */
216 	IB_DEVICE_RAW_IP_CSUM			= (1 << 26),
217 	/*
218 	 * Devices should set IB_DEVICE_CROSS_CHANNEL if they
219 	 * support execution of WQEs that involve synchronization
220 	 * of I/O operations with single completion queue managed
221 	 * by hardware.
222 	 */
223 	IB_DEVICE_CROSS_CHANNEL			= (1 << 27),
224 	IB_DEVICE_MANAGED_FLOW_STEERING		= (1 << 29),
225 	IB_DEVICE_SIGNATURE_HANDOVER		= (1 << 30),
226 	IB_DEVICE_ON_DEMAND_PAGING		= (1ULL << 31),
227 	IB_DEVICE_SG_GAPS_REG			= (1ULL << 32),
228 	IB_DEVICE_VIRTUAL_FUNCTION		= (1ULL << 33),
229 	/* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */
230 	IB_DEVICE_RAW_SCATTER_FCS		= (1ULL << 34),
231 	IB_DEVICE_RDMA_NETDEV_OPA_VNIC		= (1ULL << 35),
232 };
233 
234 enum ib_signature_prot_cap {
235 	IB_PROT_T10DIF_TYPE_1 = 1,
236 	IB_PROT_T10DIF_TYPE_2 = 1 << 1,
237 	IB_PROT_T10DIF_TYPE_3 = 1 << 2,
238 };
239 
240 enum ib_signature_guard_cap {
241 	IB_GUARD_T10DIF_CRC	= 1,
242 	IB_GUARD_T10DIF_CSUM	= 1 << 1,
243 };
244 
245 enum ib_atomic_cap {
246 	IB_ATOMIC_NONE,
247 	IB_ATOMIC_HCA,
248 	IB_ATOMIC_GLOB
249 };
250 
251 enum ib_odp_general_cap_bits {
252 	IB_ODP_SUPPORT		= 1 << 0,
253 	IB_ODP_SUPPORT_IMPLICIT = 1 << 1,
254 };
255 
256 enum ib_odp_transport_cap_bits {
257 	IB_ODP_SUPPORT_SEND	= 1 << 0,
258 	IB_ODP_SUPPORT_RECV	= 1 << 1,
259 	IB_ODP_SUPPORT_WRITE	= 1 << 2,
260 	IB_ODP_SUPPORT_READ	= 1 << 3,
261 	IB_ODP_SUPPORT_ATOMIC	= 1 << 4,
262 };
263 
264 struct ib_odp_caps {
265 	uint64_t general_caps;
266 	struct {
267 		uint32_t  rc_odp_caps;
268 		uint32_t  uc_odp_caps;
269 		uint32_t  ud_odp_caps;
270 	} per_transport_caps;
271 };
272 
273 struct ib_rss_caps {
274 	/* Corresponding bit will be set if qp type from
275 	 * 'enum ib_qp_type' is supported, e.g.
276 	 * supported_qpts |= 1 << IB_QPT_UD
277 	 */
278 	u32 supported_qpts;
279 	u32 max_rwq_indirection_tables;
280 	u32 max_rwq_indirection_table_size;
281 };
282 
283 enum ib_tm_cap_flags {
284 	/*  Support tag matching on RC transport */
285 	IB_TM_CAP_RC		    = 1 << 0,
286 };
287 
288 struct ib_xrq_caps {
289 	/* Max size of RNDV header */
290 	u32 max_rndv_hdr_size;
291 	/* Max number of entries in tag matching list */
292 	u32 max_num_tags;
293 	/* From enum ib_tm_cap_flags */
294 	u32 flags;
295 	/* Max number of outstanding list operations */
296 	u32 max_ops;
297 	/* Max number of SGE in tag matching entry */
298 	u32 max_sge;
299 };
300 
301 enum ib_cq_creation_flags {
302 	IB_CQ_FLAGS_TIMESTAMP_COMPLETION   = 1 << 0,
303 	IB_CQ_FLAGS_IGNORE_OVERRUN	   = 1 << 1,
304 };
305 
306 struct ib_cq_init_attr {
307 	unsigned int	cqe;
308 	int		comp_vector;
309 	u32		flags;
310 };
311 
312 struct ib_device_attr {
313 	u64			fw_ver;
314 	__be64			sys_image_guid;
315 	u64			max_mr_size;
316 	u64			page_size_cap;
317 	u32			vendor_id;
318 	u32			vendor_part_id;
319 	u32			hw_ver;
320 	int			max_qp;
321 	int			max_qp_wr;
322 	u64			device_cap_flags;
323 	int			max_sge;
324 	int			max_sge_rd;
325 	int			max_cq;
326 	int			max_cqe;
327 	int			max_mr;
328 	int			max_pd;
329 	int			max_qp_rd_atom;
330 	int			max_ee_rd_atom;
331 	int			max_res_rd_atom;
332 	int			max_qp_init_rd_atom;
333 	int			max_ee_init_rd_atom;
334 	enum ib_atomic_cap	atomic_cap;
335 	enum ib_atomic_cap	masked_atomic_cap;
336 	int			max_ee;
337 	int			max_rdd;
338 	int			max_mw;
339 	int			max_raw_ipv6_qp;
340 	int			max_raw_ethy_qp;
341 	int			max_mcast_grp;
342 	int			max_mcast_qp_attach;
343 	int			max_total_mcast_qp_attach;
344 	int			max_ah;
345 	int			max_fmr;
346 	int			max_map_per_fmr;
347 	int			max_srq;
348 	int			max_srq_wr;
349 	int			max_srq_sge;
350 	unsigned int		max_fast_reg_page_list_len;
351 	u16			max_pkeys;
352 	u8			local_ca_ack_delay;
353 	int			sig_prot_cap;
354 	int			sig_guard_cap;
355 	struct ib_odp_caps	odp_caps;
356 	uint64_t		timestamp_mask;
357 	uint64_t		hca_core_clock; /* in KHZ */
358 	struct ib_rss_caps	rss_caps;
359 	u32			max_wq_type_rq;
360 	u32			raw_packet_caps; /* Use ib_raw_packet_caps enum */
361 	struct ib_xrq_caps	xrq_caps;
362 };
363 
364 enum ib_mtu {
365 	IB_MTU_256  = 1,
366 	IB_MTU_512  = 2,
367 	IB_MTU_1024 = 3,
368 	IB_MTU_2048 = 4,
369 	IB_MTU_4096 = 5
370 };
371 
372 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu)
373 {
374 	switch (mtu) {
375 	case IB_MTU_256:  return  256;
376 	case IB_MTU_512:  return  512;
377 	case IB_MTU_1024: return 1024;
378 	case IB_MTU_2048: return 2048;
379 	case IB_MTU_4096: return 4096;
380 	default: 	  return -1;
381 	}
382 }
383 
384 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu)
385 {
386 	if (mtu >= 4096)
387 		return IB_MTU_4096;
388 	else if (mtu >= 2048)
389 		return IB_MTU_2048;
390 	else if (mtu >= 1024)
391 		return IB_MTU_1024;
392 	else if (mtu >= 512)
393 		return IB_MTU_512;
394 	else
395 		return IB_MTU_256;
396 }
397 
398 enum ib_port_state {
399 	IB_PORT_NOP		= 0,
400 	IB_PORT_DOWN		= 1,
401 	IB_PORT_INIT		= 2,
402 	IB_PORT_ARMED		= 3,
403 	IB_PORT_ACTIVE		= 4,
404 	IB_PORT_ACTIVE_DEFER	= 5
405 };
406 
407 enum ib_port_cap_flags {
408 	IB_PORT_SM				= 1 <<  1,
409 	IB_PORT_NOTICE_SUP			= 1 <<  2,
410 	IB_PORT_TRAP_SUP			= 1 <<  3,
411 	IB_PORT_OPT_IPD_SUP                     = 1 <<  4,
412 	IB_PORT_AUTO_MIGR_SUP			= 1 <<  5,
413 	IB_PORT_SL_MAP_SUP			= 1 <<  6,
414 	IB_PORT_MKEY_NVRAM			= 1 <<  7,
415 	IB_PORT_PKEY_NVRAM			= 1 <<  8,
416 	IB_PORT_LED_INFO_SUP			= 1 <<  9,
417 	IB_PORT_SM_DISABLED			= 1 << 10,
418 	IB_PORT_SYS_IMAGE_GUID_SUP		= 1 << 11,
419 	IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP	= 1 << 12,
420 	IB_PORT_EXTENDED_SPEEDS_SUP             = 1 << 14,
421 	IB_PORT_CM_SUP				= 1 << 16,
422 	IB_PORT_SNMP_TUNNEL_SUP			= 1 << 17,
423 	IB_PORT_REINIT_SUP			= 1 << 18,
424 	IB_PORT_DEVICE_MGMT_SUP			= 1 << 19,
425 	IB_PORT_VENDOR_CLASS_SUP		= 1 << 20,
426 	IB_PORT_DR_NOTICE_SUP			= 1 << 21,
427 	IB_PORT_CAP_MASK_NOTICE_SUP		= 1 << 22,
428 	IB_PORT_BOOT_MGMT_SUP			= 1 << 23,
429 	IB_PORT_LINK_LATENCY_SUP		= 1 << 24,
430 	IB_PORT_CLIENT_REG_SUP			= 1 << 25,
431 	IB_PORT_IP_BASED_GIDS			= 1 << 26,
432 };
433 
434 enum ib_port_width {
435 	IB_WIDTH_1X	= 1,
436 	IB_WIDTH_4X	= 2,
437 	IB_WIDTH_8X	= 4,
438 	IB_WIDTH_12X	= 8
439 };
440 
441 static inline int ib_width_enum_to_int(enum ib_port_width width)
442 {
443 	switch (width) {
444 	case IB_WIDTH_1X:  return  1;
445 	case IB_WIDTH_4X:  return  4;
446 	case IB_WIDTH_8X:  return  8;
447 	case IB_WIDTH_12X: return 12;
448 	default: 	  return -1;
449 	}
450 }
451 
452 enum ib_port_speed {
453 	IB_SPEED_SDR	= 1,
454 	IB_SPEED_DDR	= 2,
455 	IB_SPEED_QDR	= 4,
456 	IB_SPEED_FDR10	= 8,
457 	IB_SPEED_FDR	= 16,
458 	IB_SPEED_EDR	= 32,
459 	IB_SPEED_HDR	= 64
460 };
461 
462 /**
463  * struct rdma_hw_stats
464  * @timestamp - Used by the core code to track when the last update was
465  * @lifespan - Used by the core code to determine how old the counters
466  *   should be before being updated again.  Stored in jiffies, defaults
467  *   to 10 milliseconds, drivers can override the default be specifying
468  *   their own value during their allocation routine.
469  * @name - Array of pointers to static names used for the counters in
470  *   directory.
471  * @num_counters - How many hardware counters there are.  If name is
472  *   shorter than this number, a kernel oops will result.  Driver authors
473  *   are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters)
474  *   in their code to prevent this.
475  * @value - Array of u64 counters that are accessed by the sysfs code and
476  *   filled in by the drivers get_stats routine
477  */
478 struct rdma_hw_stats {
479 	unsigned long	timestamp;
480 	unsigned long	lifespan;
481 	const char * const *names;
482 	int		num_counters;
483 	u64		value[];
484 };
485 
486 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10
487 /**
488  * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct
489  *   for drivers.
490  * @names - Array of static const char *
491  * @num_counters - How many elements in array
492  * @lifespan - How many milliseconds between updates
493  */
494 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct(
495 		const char * const *names, int num_counters,
496 		unsigned long lifespan)
497 {
498 	struct rdma_hw_stats *stats;
499 
500 	stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64),
501 			GFP_KERNEL);
502 	if (!stats)
503 		return NULL;
504 	stats->names = names;
505 	stats->num_counters = num_counters;
506 	stats->lifespan = msecs_to_jiffies(lifespan);
507 
508 	return stats;
509 }
510 
511 
512 /* Define bits for the various functionality this port needs to be supported by
513  * the core.
514  */
515 /* Management                           0x00000FFF */
516 #define RDMA_CORE_CAP_IB_MAD            0x00000001
517 #define RDMA_CORE_CAP_IB_SMI            0x00000002
518 #define RDMA_CORE_CAP_IB_CM             0x00000004
519 #define RDMA_CORE_CAP_IW_CM             0x00000008
520 #define RDMA_CORE_CAP_IB_SA             0x00000010
521 #define RDMA_CORE_CAP_OPA_MAD           0x00000020
522 
523 /* Address format                       0x000FF000 */
524 #define RDMA_CORE_CAP_AF_IB             0x00001000
525 #define RDMA_CORE_CAP_ETH_AH            0x00002000
526 #define RDMA_CORE_CAP_OPA_AH            0x00004000
527 
528 /* Protocol                             0xFFF00000 */
529 #define RDMA_CORE_CAP_PROT_IB           0x00100000
530 #define RDMA_CORE_CAP_PROT_ROCE         0x00200000
531 #define RDMA_CORE_CAP_PROT_IWARP        0x00400000
532 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000
533 #define RDMA_CORE_CAP_PROT_RAW_PACKET   0x01000000
534 #define RDMA_CORE_CAP_PROT_USNIC        0x02000000
535 
536 #define RDMA_CORE_PORT_IBA_IB          (RDMA_CORE_CAP_PROT_IB  \
537 					| RDMA_CORE_CAP_IB_MAD \
538 					| RDMA_CORE_CAP_IB_SMI \
539 					| RDMA_CORE_CAP_IB_CM  \
540 					| RDMA_CORE_CAP_IB_SA  \
541 					| RDMA_CORE_CAP_AF_IB)
542 #define RDMA_CORE_PORT_IBA_ROCE        (RDMA_CORE_CAP_PROT_ROCE \
543 					| RDMA_CORE_CAP_IB_MAD  \
544 					| RDMA_CORE_CAP_IB_CM   \
545 					| RDMA_CORE_CAP_AF_IB   \
546 					| RDMA_CORE_CAP_ETH_AH)
547 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP			\
548 					(RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \
549 					| RDMA_CORE_CAP_IB_MAD  \
550 					| RDMA_CORE_CAP_IB_CM   \
551 					| RDMA_CORE_CAP_AF_IB   \
552 					| RDMA_CORE_CAP_ETH_AH)
553 #define RDMA_CORE_PORT_IWARP           (RDMA_CORE_CAP_PROT_IWARP \
554 					| RDMA_CORE_CAP_IW_CM)
555 #define RDMA_CORE_PORT_INTEL_OPA       (RDMA_CORE_PORT_IBA_IB  \
556 					| RDMA_CORE_CAP_OPA_MAD)
557 
558 #define RDMA_CORE_PORT_RAW_PACKET	(RDMA_CORE_CAP_PROT_RAW_PACKET)
559 
560 #define RDMA_CORE_PORT_USNIC		(RDMA_CORE_CAP_PROT_USNIC)
561 
562 struct ib_port_attr {
563 	u64			subnet_prefix;
564 	enum ib_port_state	state;
565 	enum ib_mtu		max_mtu;
566 	enum ib_mtu		active_mtu;
567 	int			gid_tbl_len;
568 	u32			port_cap_flags;
569 	u32			max_msg_sz;
570 	u32			bad_pkey_cntr;
571 	u32			qkey_viol_cntr;
572 	u16			pkey_tbl_len;
573 	u32			sm_lid;
574 	u32			lid;
575 	u8			lmc;
576 	u8			max_vl_num;
577 	u8			sm_sl;
578 	u8			subnet_timeout;
579 	u8			init_type_reply;
580 	u8			active_width;
581 	u8			active_speed;
582 	u8                      phys_state;
583 	bool			grh_required;
584 };
585 
586 enum ib_device_modify_flags {
587 	IB_DEVICE_MODIFY_SYS_IMAGE_GUID	= 1 << 0,
588 	IB_DEVICE_MODIFY_NODE_DESC	= 1 << 1
589 };
590 
591 #define IB_DEVICE_NODE_DESC_MAX 64
592 
593 struct ib_device_modify {
594 	u64	sys_image_guid;
595 	char	node_desc[IB_DEVICE_NODE_DESC_MAX];
596 };
597 
598 enum ib_port_modify_flags {
599 	IB_PORT_SHUTDOWN		= 1,
600 	IB_PORT_INIT_TYPE		= (1<<2),
601 	IB_PORT_RESET_QKEY_CNTR		= (1<<3),
602 	IB_PORT_OPA_MASK_CHG		= (1<<4)
603 };
604 
605 struct ib_port_modify {
606 	u32	set_port_cap_mask;
607 	u32	clr_port_cap_mask;
608 	u8	init_type;
609 };
610 
611 enum ib_event_type {
612 	IB_EVENT_CQ_ERR,
613 	IB_EVENT_QP_FATAL,
614 	IB_EVENT_QP_REQ_ERR,
615 	IB_EVENT_QP_ACCESS_ERR,
616 	IB_EVENT_COMM_EST,
617 	IB_EVENT_SQ_DRAINED,
618 	IB_EVENT_PATH_MIG,
619 	IB_EVENT_PATH_MIG_ERR,
620 	IB_EVENT_DEVICE_FATAL,
621 	IB_EVENT_PORT_ACTIVE,
622 	IB_EVENT_PORT_ERR,
623 	IB_EVENT_LID_CHANGE,
624 	IB_EVENT_PKEY_CHANGE,
625 	IB_EVENT_SM_CHANGE,
626 	IB_EVENT_SRQ_ERR,
627 	IB_EVENT_SRQ_LIMIT_REACHED,
628 	IB_EVENT_QP_LAST_WQE_REACHED,
629 	IB_EVENT_CLIENT_REREGISTER,
630 	IB_EVENT_GID_CHANGE,
631 	IB_EVENT_WQ_FATAL,
632 };
633 
634 const char *__attribute_const__ ib_event_msg(enum ib_event_type event);
635 
636 struct ib_event {
637 	struct ib_device	*device;
638 	union {
639 		struct ib_cq	*cq;
640 		struct ib_qp	*qp;
641 		struct ib_srq	*srq;
642 		struct ib_wq	*wq;
643 		u8		port_num;
644 	} element;
645 	enum ib_event_type	event;
646 };
647 
648 struct ib_event_handler {
649 	struct ib_device *device;
650 	void            (*handler)(struct ib_event_handler *, struct ib_event *);
651 	struct list_head  list;
652 };
653 
654 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler)		\
655 	do {							\
656 		(_ptr)->device  = _device;			\
657 		(_ptr)->handler = _handler;			\
658 		INIT_LIST_HEAD(&(_ptr)->list);			\
659 	} while (0)
660 
661 struct ib_global_route {
662 	union ib_gid	dgid;
663 	u32		flow_label;
664 	u8		sgid_index;
665 	u8		hop_limit;
666 	u8		traffic_class;
667 };
668 
669 struct ib_grh {
670 	__be32		version_tclass_flow;
671 	__be16		paylen;
672 	u8		next_hdr;
673 	u8		hop_limit;
674 	union ib_gid	sgid;
675 	union ib_gid	dgid;
676 };
677 
678 union rdma_network_hdr {
679 	struct ib_grh ibgrh;
680 	struct {
681 		/* The IB spec states that if it's IPv4, the header
682 		 * is located in the last 20 bytes of the header.
683 		 */
684 		u8		reserved[20];
685 		struct iphdr	roce4grh;
686 	};
687 };
688 
689 #define IB_QPN_MASK		0xFFFFFF
690 
691 enum {
692 	IB_MULTICAST_QPN = 0xffffff
693 };
694 
695 #define IB_LID_PERMISSIVE	cpu_to_be16(0xFFFF)
696 #define IB_MULTICAST_LID_BASE	cpu_to_be16(0xC000)
697 
698 enum ib_ah_flags {
699 	IB_AH_GRH	= 1
700 };
701 
702 enum ib_rate {
703 	IB_RATE_PORT_CURRENT = 0,
704 	IB_RATE_2_5_GBPS = 2,
705 	IB_RATE_5_GBPS   = 5,
706 	IB_RATE_10_GBPS  = 3,
707 	IB_RATE_20_GBPS  = 6,
708 	IB_RATE_30_GBPS  = 4,
709 	IB_RATE_40_GBPS  = 7,
710 	IB_RATE_60_GBPS  = 8,
711 	IB_RATE_80_GBPS  = 9,
712 	IB_RATE_120_GBPS = 10,
713 	IB_RATE_14_GBPS  = 11,
714 	IB_RATE_56_GBPS  = 12,
715 	IB_RATE_112_GBPS = 13,
716 	IB_RATE_168_GBPS = 14,
717 	IB_RATE_25_GBPS  = 15,
718 	IB_RATE_100_GBPS = 16,
719 	IB_RATE_200_GBPS = 17,
720 	IB_RATE_300_GBPS = 18
721 };
722 
723 /**
724  * ib_rate_to_mult - Convert the IB rate enum to a multiple of the
725  * base rate of 2.5 Gbit/sec.  For example, IB_RATE_5_GBPS will be
726  * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec.
727  * @rate: rate to convert.
728  */
729 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate);
730 
731 /**
732  * ib_rate_to_mbps - Convert the IB rate enum to Mbps.
733  * For example, IB_RATE_2_5_GBPS will be converted to 2500.
734  * @rate: rate to convert.
735  */
736 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate);
737 
738 
739 /**
740  * enum ib_mr_type - memory region type
741  * @IB_MR_TYPE_MEM_REG:       memory region that is used for
742  *                            normal registration
743  * @IB_MR_TYPE_SIGNATURE:     memory region that is used for
744  *                            signature operations (data-integrity
745  *                            capable regions)
746  * @IB_MR_TYPE_SG_GAPS:       memory region that is capable to
747  *                            register any arbitrary sg lists (without
748  *                            the normal mr constraints - see
749  *                            ib_map_mr_sg)
750  */
751 enum ib_mr_type {
752 	IB_MR_TYPE_MEM_REG,
753 	IB_MR_TYPE_SIGNATURE,
754 	IB_MR_TYPE_SG_GAPS,
755 };
756 
757 /**
758  * Signature types
759  * IB_SIG_TYPE_NONE: Unprotected.
760  * IB_SIG_TYPE_T10_DIF: Type T10-DIF
761  */
762 enum ib_signature_type {
763 	IB_SIG_TYPE_NONE,
764 	IB_SIG_TYPE_T10_DIF,
765 };
766 
767 /**
768  * Signature T10-DIF block-guard types
769  * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules.
770  * IB_T10DIF_CSUM: Corresponds to IP checksum rules.
771  */
772 enum ib_t10_dif_bg_type {
773 	IB_T10DIF_CRC,
774 	IB_T10DIF_CSUM
775 };
776 
777 /**
778  * struct ib_t10_dif_domain - Parameters specific for T10-DIF
779  *     domain.
780  * @bg_type: T10-DIF block guard type (CRC|CSUM)
781  * @pi_interval: protection information interval.
782  * @bg: seed of guard computation.
783  * @app_tag: application tag of guard block
784  * @ref_tag: initial guard block reference tag.
785  * @ref_remap: Indicate wethear the reftag increments each block
786  * @app_escape: Indicate to skip block check if apptag=0xffff
787  * @ref_escape: Indicate to skip block check if reftag=0xffffffff
788  * @apptag_check_mask: check bitmask of application tag.
789  */
790 struct ib_t10_dif_domain {
791 	enum ib_t10_dif_bg_type bg_type;
792 	u16			pi_interval;
793 	u16			bg;
794 	u16			app_tag;
795 	u32			ref_tag;
796 	bool			ref_remap;
797 	bool			app_escape;
798 	bool			ref_escape;
799 	u16			apptag_check_mask;
800 };
801 
802 /**
803  * struct ib_sig_domain - Parameters for signature domain
804  * @sig_type: specific signauture type
805  * @sig: union of all signature domain attributes that may
806  *     be used to set domain layout.
807  */
808 struct ib_sig_domain {
809 	enum ib_signature_type sig_type;
810 	union {
811 		struct ib_t10_dif_domain dif;
812 	} sig;
813 };
814 
815 /**
816  * struct ib_sig_attrs - Parameters for signature handover operation
817  * @check_mask: bitmask for signature byte check (8 bytes)
818  * @mem: memory domain layout desciptor.
819  * @wire: wire domain layout desciptor.
820  */
821 struct ib_sig_attrs {
822 	u8			check_mask;
823 	struct ib_sig_domain	mem;
824 	struct ib_sig_domain	wire;
825 };
826 
827 enum ib_sig_err_type {
828 	IB_SIG_BAD_GUARD,
829 	IB_SIG_BAD_REFTAG,
830 	IB_SIG_BAD_APPTAG,
831 };
832 
833 /**
834  * struct ib_sig_err - signature error descriptor
835  */
836 struct ib_sig_err {
837 	enum ib_sig_err_type	err_type;
838 	u32			expected;
839 	u32			actual;
840 	u64			sig_err_offset;
841 	u32			key;
842 };
843 
844 enum ib_mr_status_check {
845 	IB_MR_CHECK_SIG_STATUS = 1,
846 };
847 
848 /**
849  * struct ib_mr_status - Memory region status container
850  *
851  * @fail_status: Bitmask of MR checks status. For each
852  *     failed check a corresponding status bit is set.
853  * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS
854  *     failure.
855  */
856 struct ib_mr_status {
857 	u32		    fail_status;
858 	struct ib_sig_err   sig_err;
859 };
860 
861 /**
862  * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate
863  * enum.
864  * @mult: multiple to convert.
865  */
866 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult);
867 
868 enum rdma_ah_attr_type {
869 	RDMA_AH_ATTR_TYPE_IB,
870 	RDMA_AH_ATTR_TYPE_ROCE,
871 	RDMA_AH_ATTR_TYPE_OPA,
872 };
873 
874 struct ib_ah_attr {
875 	u16			dlid;
876 	u8			src_path_bits;
877 };
878 
879 struct roce_ah_attr {
880 	u8			dmac[ETH_ALEN];
881 };
882 
883 struct opa_ah_attr {
884 	u32			dlid;
885 	u8			src_path_bits;
886 	bool			make_grd;
887 };
888 
889 struct rdma_ah_attr {
890 	struct ib_global_route	grh;
891 	u8			sl;
892 	u8			static_rate;
893 	u8			port_num;
894 	u8			ah_flags;
895 	enum rdma_ah_attr_type type;
896 	union {
897 		struct ib_ah_attr ib;
898 		struct roce_ah_attr roce;
899 		struct opa_ah_attr opa;
900 	};
901 };
902 
903 enum ib_wc_status {
904 	IB_WC_SUCCESS,
905 	IB_WC_LOC_LEN_ERR,
906 	IB_WC_LOC_QP_OP_ERR,
907 	IB_WC_LOC_EEC_OP_ERR,
908 	IB_WC_LOC_PROT_ERR,
909 	IB_WC_WR_FLUSH_ERR,
910 	IB_WC_MW_BIND_ERR,
911 	IB_WC_BAD_RESP_ERR,
912 	IB_WC_LOC_ACCESS_ERR,
913 	IB_WC_REM_INV_REQ_ERR,
914 	IB_WC_REM_ACCESS_ERR,
915 	IB_WC_REM_OP_ERR,
916 	IB_WC_RETRY_EXC_ERR,
917 	IB_WC_RNR_RETRY_EXC_ERR,
918 	IB_WC_LOC_RDD_VIOL_ERR,
919 	IB_WC_REM_INV_RD_REQ_ERR,
920 	IB_WC_REM_ABORT_ERR,
921 	IB_WC_INV_EECN_ERR,
922 	IB_WC_INV_EEC_STATE_ERR,
923 	IB_WC_FATAL_ERR,
924 	IB_WC_RESP_TIMEOUT_ERR,
925 	IB_WC_GENERAL_ERR
926 };
927 
928 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status);
929 
930 enum ib_wc_opcode {
931 	IB_WC_SEND,
932 	IB_WC_RDMA_WRITE,
933 	IB_WC_RDMA_READ,
934 	IB_WC_COMP_SWAP,
935 	IB_WC_FETCH_ADD,
936 	IB_WC_LSO,
937 	IB_WC_LOCAL_INV,
938 	IB_WC_REG_MR,
939 	IB_WC_MASKED_COMP_SWAP,
940 	IB_WC_MASKED_FETCH_ADD,
941 /*
942  * Set value of IB_WC_RECV so consumers can test if a completion is a
943  * receive by testing (opcode & IB_WC_RECV).
944  */
945 	IB_WC_RECV			= 1 << 7,
946 	IB_WC_RECV_RDMA_WITH_IMM
947 };
948 
949 enum ib_wc_flags {
950 	IB_WC_GRH		= 1,
951 	IB_WC_WITH_IMM		= (1<<1),
952 	IB_WC_WITH_INVALIDATE	= (1<<2),
953 	IB_WC_IP_CSUM_OK	= (1<<3),
954 	IB_WC_WITH_SMAC		= (1<<4),
955 	IB_WC_WITH_VLAN		= (1<<5),
956 	IB_WC_WITH_NETWORK_HDR_TYPE	= (1<<6),
957 };
958 
959 struct ib_wc {
960 	union {
961 		u64		wr_id;
962 		struct ib_cqe	*wr_cqe;
963 	};
964 	enum ib_wc_status	status;
965 	enum ib_wc_opcode	opcode;
966 	u32			vendor_err;
967 	u32			byte_len;
968 	struct ib_qp	       *qp;
969 	union {
970 		__be32		imm_data;
971 		u32		invalidate_rkey;
972 	} ex;
973 	u32			src_qp;
974 	int			wc_flags;
975 	u16			pkey_index;
976 	u32			slid;
977 	u8			sl;
978 	u8			dlid_path_bits;
979 	u8			port_num;	/* valid only for DR SMPs on switches */
980 	u8			smac[ETH_ALEN];
981 	u16			vlan_id;
982 	u8			network_hdr_type;
983 };
984 
985 enum ib_cq_notify_flags {
986 	IB_CQ_SOLICITED			= 1 << 0,
987 	IB_CQ_NEXT_COMP			= 1 << 1,
988 	IB_CQ_SOLICITED_MASK		= IB_CQ_SOLICITED | IB_CQ_NEXT_COMP,
989 	IB_CQ_REPORT_MISSED_EVENTS	= 1 << 2,
990 };
991 
992 enum ib_srq_type {
993 	IB_SRQT_BASIC,
994 	IB_SRQT_XRC,
995 	IB_SRQT_TM,
996 };
997 
998 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type)
999 {
1000 	return srq_type == IB_SRQT_XRC ||
1001 	       srq_type == IB_SRQT_TM;
1002 }
1003 
1004 enum ib_srq_attr_mask {
1005 	IB_SRQ_MAX_WR	= 1 << 0,
1006 	IB_SRQ_LIMIT	= 1 << 1,
1007 };
1008 
1009 struct ib_srq_attr {
1010 	u32	max_wr;
1011 	u32	max_sge;
1012 	u32	srq_limit;
1013 };
1014 
1015 struct ib_srq_init_attr {
1016 	void		      (*event_handler)(struct ib_event *, void *);
1017 	void		       *srq_context;
1018 	struct ib_srq_attr	attr;
1019 	enum ib_srq_type	srq_type;
1020 
1021 	struct {
1022 		struct ib_cq   *cq;
1023 		union {
1024 			struct {
1025 				struct ib_xrcd *xrcd;
1026 			} xrc;
1027 
1028 			struct {
1029 				u32		max_num_tags;
1030 			} tag_matching;
1031 		};
1032 	} ext;
1033 };
1034 
1035 struct ib_qp_cap {
1036 	u32	max_send_wr;
1037 	u32	max_recv_wr;
1038 	u32	max_send_sge;
1039 	u32	max_recv_sge;
1040 	u32	max_inline_data;
1041 
1042 	/*
1043 	 * Maximum number of rdma_rw_ctx structures in flight at a time.
1044 	 * ib_create_qp() will calculate the right amount of neededed WRs
1045 	 * and MRs based on this.
1046 	 */
1047 	u32	max_rdma_ctxs;
1048 };
1049 
1050 enum ib_sig_type {
1051 	IB_SIGNAL_ALL_WR,
1052 	IB_SIGNAL_REQ_WR
1053 };
1054 
1055 enum ib_qp_type {
1056 	/*
1057 	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
1058 	 * here (and in that order) since the MAD layer uses them as
1059 	 * indices into a 2-entry table.
1060 	 */
1061 	IB_QPT_SMI,
1062 	IB_QPT_GSI,
1063 
1064 	IB_QPT_RC,
1065 	IB_QPT_UC,
1066 	IB_QPT_UD,
1067 	IB_QPT_RAW_IPV6,
1068 	IB_QPT_RAW_ETHERTYPE,
1069 	IB_QPT_RAW_PACKET = 8,
1070 	IB_QPT_XRC_INI = 9,
1071 	IB_QPT_XRC_TGT,
1072 	IB_QPT_MAX,
1073 	/* Reserve a range for qp types internal to the low level driver.
1074 	 * These qp types will not be visible at the IB core layer, so the
1075 	 * IB_QPT_MAX usages should not be affected in the core layer
1076 	 */
1077 	IB_QPT_RESERVED1 = 0x1000,
1078 	IB_QPT_RESERVED2,
1079 	IB_QPT_RESERVED3,
1080 	IB_QPT_RESERVED4,
1081 	IB_QPT_RESERVED5,
1082 	IB_QPT_RESERVED6,
1083 	IB_QPT_RESERVED7,
1084 	IB_QPT_RESERVED8,
1085 	IB_QPT_RESERVED9,
1086 	IB_QPT_RESERVED10,
1087 };
1088 
1089 enum ib_qp_create_flags {
1090 	IB_QP_CREATE_IPOIB_UD_LSO		= 1 << 0,
1091 	IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
1092 	IB_QP_CREATE_CROSS_CHANNEL              = 1 << 2,
1093 	IB_QP_CREATE_MANAGED_SEND               = 1 << 3,
1094 	IB_QP_CREATE_MANAGED_RECV               = 1 << 4,
1095 	IB_QP_CREATE_NETIF_QP			= 1 << 5,
1096 	IB_QP_CREATE_SIGNATURE_EN		= 1 << 6,
1097 	/* FREE					= 1 << 7, */
1098 	IB_QP_CREATE_SCATTER_FCS		= 1 << 8,
1099 	IB_QP_CREATE_CVLAN_STRIPPING		= 1 << 9,
1100 	IB_QP_CREATE_SOURCE_QPN			= 1 << 10,
1101 	/* reserve bits 26-31 for low level drivers' internal use */
1102 	IB_QP_CREATE_RESERVED_START		= 1 << 26,
1103 	IB_QP_CREATE_RESERVED_END		= 1 << 31,
1104 };
1105 
1106 /*
1107  * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler
1108  * callback to destroy the passed in QP.
1109  */
1110 
1111 struct ib_qp_init_attr {
1112 	void                  (*event_handler)(struct ib_event *, void *);
1113 	void		       *qp_context;
1114 	struct ib_cq	       *send_cq;
1115 	struct ib_cq	       *recv_cq;
1116 	struct ib_srq	       *srq;
1117 	struct ib_xrcd	       *xrcd;     /* XRC TGT QPs only */
1118 	struct ib_qp_cap	cap;
1119 	enum ib_sig_type	sq_sig_type;
1120 	enum ib_qp_type		qp_type;
1121 	enum ib_qp_create_flags	create_flags;
1122 
1123 	/*
1124 	 * Only needed for special QP types, or when using the RW API.
1125 	 */
1126 	u8			port_num;
1127 	struct ib_rwq_ind_table *rwq_ind_tbl;
1128 	u32			source_qpn;
1129 };
1130 
1131 struct ib_qp_open_attr {
1132 	void                  (*event_handler)(struct ib_event *, void *);
1133 	void		       *qp_context;
1134 	u32			qp_num;
1135 	enum ib_qp_type		qp_type;
1136 };
1137 
1138 enum ib_rnr_timeout {
1139 	IB_RNR_TIMER_655_36 =  0,
1140 	IB_RNR_TIMER_000_01 =  1,
1141 	IB_RNR_TIMER_000_02 =  2,
1142 	IB_RNR_TIMER_000_03 =  3,
1143 	IB_RNR_TIMER_000_04 =  4,
1144 	IB_RNR_TIMER_000_06 =  5,
1145 	IB_RNR_TIMER_000_08 =  6,
1146 	IB_RNR_TIMER_000_12 =  7,
1147 	IB_RNR_TIMER_000_16 =  8,
1148 	IB_RNR_TIMER_000_24 =  9,
1149 	IB_RNR_TIMER_000_32 = 10,
1150 	IB_RNR_TIMER_000_48 = 11,
1151 	IB_RNR_TIMER_000_64 = 12,
1152 	IB_RNR_TIMER_000_96 = 13,
1153 	IB_RNR_TIMER_001_28 = 14,
1154 	IB_RNR_TIMER_001_92 = 15,
1155 	IB_RNR_TIMER_002_56 = 16,
1156 	IB_RNR_TIMER_003_84 = 17,
1157 	IB_RNR_TIMER_005_12 = 18,
1158 	IB_RNR_TIMER_007_68 = 19,
1159 	IB_RNR_TIMER_010_24 = 20,
1160 	IB_RNR_TIMER_015_36 = 21,
1161 	IB_RNR_TIMER_020_48 = 22,
1162 	IB_RNR_TIMER_030_72 = 23,
1163 	IB_RNR_TIMER_040_96 = 24,
1164 	IB_RNR_TIMER_061_44 = 25,
1165 	IB_RNR_TIMER_081_92 = 26,
1166 	IB_RNR_TIMER_122_88 = 27,
1167 	IB_RNR_TIMER_163_84 = 28,
1168 	IB_RNR_TIMER_245_76 = 29,
1169 	IB_RNR_TIMER_327_68 = 30,
1170 	IB_RNR_TIMER_491_52 = 31
1171 };
1172 
1173 enum ib_qp_attr_mask {
1174 	IB_QP_STATE			= 1,
1175 	IB_QP_CUR_STATE			= (1<<1),
1176 	IB_QP_EN_SQD_ASYNC_NOTIFY	= (1<<2),
1177 	IB_QP_ACCESS_FLAGS		= (1<<3),
1178 	IB_QP_PKEY_INDEX		= (1<<4),
1179 	IB_QP_PORT			= (1<<5),
1180 	IB_QP_QKEY			= (1<<6),
1181 	IB_QP_AV			= (1<<7),
1182 	IB_QP_PATH_MTU			= (1<<8),
1183 	IB_QP_TIMEOUT			= (1<<9),
1184 	IB_QP_RETRY_CNT			= (1<<10),
1185 	IB_QP_RNR_RETRY			= (1<<11),
1186 	IB_QP_RQ_PSN			= (1<<12),
1187 	IB_QP_MAX_QP_RD_ATOMIC		= (1<<13),
1188 	IB_QP_ALT_PATH			= (1<<14),
1189 	IB_QP_MIN_RNR_TIMER		= (1<<15),
1190 	IB_QP_SQ_PSN			= (1<<16),
1191 	IB_QP_MAX_DEST_RD_ATOMIC	= (1<<17),
1192 	IB_QP_PATH_MIG_STATE		= (1<<18),
1193 	IB_QP_CAP			= (1<<19),
1194 	IB_QP_DEST_QPN			= (1<<20),
1195 	IB_QP_RESERVED1			= (1<<21),
1196 	IB_QP_RESERVED2			= (1<<22),
1197 	IB_QP_RESERVED3			= (1<<23),
1198 	IB_QP_RESERVED4			= (1<<24),
1199 	IB_QP_RATE_LIMIT		= (1<<25),
1200 };
1201 
1202 enum ib_qp_state {
1203 	IB_QPS_RESET,
1204 	IB_QPS_INIT,
1205 	IB_QPS_RTR,
1206 	IB_QPS_RTS,
1207 	IB_QPS_SQD,
1208 	IB_QPS_SQE,
1209 	IB_QPS_ERR
1210 };
1211 
1212 enum ib_mig_state {
1213 	IB_MIG_MIGRATED,
1214 	IB_MIG_REARM,
1215 	IB_MIG_ARMED
1216 };
1217 
1218 enum ib_mw_type {
1219 	IB_MW_TYPE_1 = 1,
1220 	IB_MW_TYPE_2 = 2
1221 };
1222 
1223 struct ib_qp_attr {
1224 	enum ib_qp_state	qp_state;
1225 	enum ib_qp_state	cur_qp_state;
1226 	enum ib_mtu		path_mtu;
1227 	enum ib_mig_state	path_mig_state;
1228 	u32			qkey;
1229 	u32			rq_psn;
1230 	u32			sq_psn;
1231 	u32			dest_qp_num;
1232 	int			qp_access_flags;
1233 	struct ib_qp_cap	cap;
1234 	struct rdma_ah_attr	ah_attr;
1235 	struct rdma_ah_attr	alt_ah_attr;
1236 	u16			pkey_index;
1237 	u16			alt_pkey_index;
1238 	u8			en_sqd_async_notify;
1239 	u8			sq_draining;
1240 	u8			max_rd_atomic;
1241 	u8			max_dest_rd_atomic;
1242 	u8			min_rnr_timer;
1243 	u8			port_num;
1244 	u8			timeout;
1245 	u8			retry_cnt;
1246 	u8			rnr_retry;
1247 	u8			alt_port_num;
1248 	u8			alt_timeout;
1249 	u32			rate_limit;
1250 };
1251 
1252 enum ib_wr_opcode {
1253 	IB_WR_RDMA_WRITE,
1254 	IB_WR_RDMA_WRITE_WITH_IMM,
1255 	IB_WR_SEND,
1256 	IB_WR_SEND_WITH_IMM,
1257 	IB_WR_RDMA_READ,
1258 	IB_WR_ATOMIC_CMP_AND_SWP,
1259 	IB_WR_ATOMIC_FETCH_AND_ADD,
1260 	IB_WR_LSO,
1261 	IB_WR_SEND_WITH_INV,
1262 	IB_WR_RDMA_READ_WITH_INV,
1263 	IB_WR_LOCAL_INV,
1264 	IB_WR_REG_MR,
1265 	IB_WR_MASKED_ATOMIC_CMP_AND_SWP,
1266 	IB_WR_MASKED_ATOMIC_FETCH_AND_ADD,
1267 	IB_WR_REG_SIG_MR,
1268 	/* reserve values for low level drivers' internal use.
1269 	 * These values will not be used at all in the ib core layer.
1270 	 */
1271 	IB_WR_RESERVED1 = 0xf0,
1272 	IB_WR_RESERVED2,
1273 	IB_WR_RESERVED3,
1274 	IB_WR_RESERVED4,
1275 	IB_WR_RESERVED5,
1276 	IB_WR_RESERVED6,
1277 	IB_WR_RESERVED7,
1278 	IB_WR_RESERVED8,
1279 	IB_WR_RESERVED9,
1280 	IB_WR_RESERVED10,
1281 };
1282 
1283 enum ib_send_flags {
1284 	IB_SEND_FENCE		= 1,
1285 	IB_SEND_SIGNALED	= (1<<1),
1286 	IB_SEND_SOLICITED	= (1<<2),
1287 	IB_SEND_INLINE		= (1<<3),
1288 	IB_SEND_IP_CSUM		= (1<<4),
1289 
1290 	/* reserve bits 26-31 for low level drivers' internal use */
1291 	IB_SEND_RESERVED_START	= (1 << 26),
1292 	IB_SEND_RESERVED_END	= (1 << 31),
1293 };
1294 
1295 struct ib_sge {
1296 	u64	addr;
1297 	u32	length;
1298 	u32	lkey;
1299 };
1300 
1301 struct ib_cqe {
1302 	void (*done)(struct ib_cq *cq, struct ib_wc *wc);
1303 };
1304 
1305 struct ib_send_wr {
1306 	struct ib_send_wr      *next;
1307 	union {
1308 		u64		wr_id;
1309 		struct ib_cqe	*wr_cqe;
1310 	};
1311 	struct ib_sge	       *sg_list;
1312 	int			num_sge;
1313 	enum ib_wr_opcode	opcode;
1314 	int			send_flags;
1315 	union {
1316 		__be32		imm_data;
1317 		u32		invalidate_rkey;
1318 	} ex;
1319 };
1320 
1321 struct ib_rdma_wr {
1322 	struct ib_send_wr	wr;
1323 	u64			remote_addr;
1324 	u32			rkey;
1325 };
1326 
1327 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr)
1328 {
1329 	return container_of(wr, struct ib_rdma_wr, wr);
1330 }
1331 
1332 struct ib_atomic_wr {
1333 	struct ib_send_wr	wr;
1334 	u64			remote_addr;
1335 	u64			compare_add;
1336 	u64			swap;
1337 	u64			compare_add_mask;
1338 	u64			swap_mask;
1339 	u32			rkey;
1340 };
1341 
1342 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr)
1343 {
1344 	return container_of(wr, struct ib_atomic_wr, wr);
1345 }
1346 
1347 struct ib_ud_wr {
1348 	struct ib_send_wr	wr;
1349 	struct ib_ah		*ah;
1350 	void			*header;
1351 	int			hlen;
1352 	int			mss;
1353 	u32			remote_qpn;
1354 	u32			remote_qkey;
1355 	u16			pkey_index; /* valid for GSI only */
1356 	u8			port_num;   /* valid for DR SMPs on switch only */
1357 };
1358 
1359 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr)
1360 {
1361 	return container_of(wr, struct ib_ud_wr, wr);
1362 }
1363 
1364 struct ib_reg_wr {
1365 	struct ib_send_wr	wr;
1366 	struct ib_mr		*mr;
1367 	u32			key;
1368 	int			access;
1369 };
1370 
1371 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr)
1372 {
1373 	return container_of(wr, struct ib_reg_wr, wr);
1374 }
1375 
1376 struct ib_sig_handover_wr {
1377 	struct ib_send_wr	wr;
1378 	struct ib_sig_attrs    *sig_attrs;
1379 	struct ib_mr	       *sig_mr;
1380 	int			access_flags;
1381 	struct ib_sge	       *prot;
1382 };
1383 
1384 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr)
1385 {
1386 	return container_of(wr, struct ib_sig_handover_wr, wr);
1387 }
1388 
1389 struct ib_recv_wr {
1390 	struct ib_recv_wr      *next;
1391 	union {
1392 		u64		wr_id;
1393 		struct ib_cqe	*wr_cqe;
1394 	};
1395 	struct ib_sge	       *sg_list;
1396 	int			num_sge;
1397 };
1398 
1399 enum ib_access_flags {
1400 	IB_ACCESS_LOCAL_WRITE	= 1,
1401 	IB_ACCESS_REMOTE_WRITE	= (1<<1),
1402 	IB_ACCESS_REMOTE_READ	= (1<<2),
1403 	IB_ACCESS_REMOTE_ATOMIC	= (1<<3),
1404 	IB_ACCESS_MW_BIND	= (1<<4),
1405 	IB_ZERO_BASED		= (1<<5),
1406 	IB_ACCESS_ON_DEMAND     = (1<<6),
1407 	IB_ACCESS_HUGETLB	= (1<<7),
1408 };
1409 
1410 /*
1411  * XXX: these are apparently used for ->rereg_user_mr, no idea why they
1412  * are hidden here instead of a uapi header!
1413  */
1414 enum ib_mr_rereg_flags {
1415 	IB_MR_REREG_TRANS	= 1,
1416 	IB_MR_REREG_PD		= (1<<1),
1417 	IB_MR_REREG_ACCESS	= (1<<2),
1418 	IB_MR_REREG_SUPPORTED	= ((IB_MR_REREG_ACCESS << 1) - 1)
1419 };
1420 
1421 struct ib_fmr_attr {
1422 	int	max_pages;
1423 	int	max_maps;
1424 	u8	page_shift;
1425 };
1426 
1427 struct ib_umem;
1428 
1429 enum rdma_remove_reason {
1430 	/* Userspace requested uobject deletion. Call could fail */
1431 	RDMA_REMOVE_DESTROY,
1432 	/* Context deletion. This call should delete the actual object itself */
1433 	RDMA_REMOVE_CLOSE,
1434 	/* Driver is being hot-unplugged. This call should delete the actual object itself */
1435 	RDMA_REMOVE_DRIVER_REMOVE,
1436 	/* Context is being cleaned-up, but commit was just completed */
1437 	RDMA_REMOVE_DURING_CLEANUP,
1438 };
1439 
1440 struct ib_rdmacg_object {
1441 #ifdef CONFIG_CGROUP_RDMA
1442 	struct rdma_cgroup	*cg;		/* owner rdma cgroup */
1443 #endif
1444 };
1445 
1446 struct ib_ucontext {
1447 	struct ib_device       *device;
1448 	struct ib_uverbs_file  *ufile;
1449 	int			closing;
1450 
1451 	/* locking the uobjects_list */
1452 	struct mutex		uobjects_lock;
1453 	struct list_head	uobjects;
1454 	/* protects cleanup process from other actions */
1455 	struct rw_semaphore	cleanup_rwsem;
1456 	enum rdma_remove_reason cleanup_reason;
1457 
1458 	struct pid             *tgid;
1459 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING
1460 	struct rb_root_cached   umem_tree;
1461 	/*
1462 	 * Protects .umem_rbroot and tree, as well as odp_mrs_count and
1463 	 * mmu notifiers registration.
1464 	 */
1465 	struct rw_semaphore	umem_rwsem;
1466 	void (*invalidate_range)(struct ib_umem *umem,
1467 				 unsigned long start, unsigned long end);
1468 
1469 	struct mmu_notifier	mn;
1470 	atomic_t		notifier_count;
1471 	/* A list of umems that don't have private mmu notifier counters yet. */
1472 	struct list_head	no_private_counters;
1473 	int                     odp_mrs_count;
1474 #endif
1475 
1476 	struct ib_rdmacg_object	cg_obj;
1477 };
1478 
1479 struct ib_uobject {
1480 	u64			user_handle;	/* handle given to us by userspace */
1481 	struct ib_ucontext     *context;	/* associated user context */
1482 	void		       *object;		/* containing object */
1483 	struct list_head	list;		/* link to context's list */
1484 	struct ib_rdmacg_object	cg_obj;		/* rdmacg object */
1485 	int			id;		/* index into kernel idr */
1486 	struct kref		ref;
1487 	atomic_t		usecnt;		/* protects exclusive access */
1488 	struct rcu_head		rcu;		/* kfree_rcu() overhead */
1489 
1490 	const struct uverbs_obj_type *type;
1491 };
1492 
1493 struct ib_uobject_file {
1494 	struct ib_uobject	uobj;
1495 	/* ufile contains the lock between context release and file close */
1496 	struct ib_uverbs_file	*ufile;
1497 };
1498 
1499 struct ib_udata {
1500 	const void __user *inbuf;
1501 	void __user *outbuf;
1502 	size_t       inlen;
1503 	size_t       outlen;
1504 };
1505 
1506 struct ib_pd {
1507 	u32			local_dma_lkey;
1508 	u32			flags;
1509 	struct ib_device       *device;
1510 	struct ib_uobject      *uobject;
1511 	atomic_t          	usecnt; /* count all resources */
1512 
1513 	u32			unsafe_global_rkey;
1514 
1515 	/*
1516 	 * Implementation details of the RDMA core, don't use in drivers:
1517 	 */
1518 	struct ib_mr	       *__internal_mr;
1519 };
1520 
1521 struct ib_xrcd {
1522 	struct ib_device       *device;
1523 	atomic_t		usecnt; /* count all exposed resources */
1524 	struct inode	       *inode;
1525 
1526 	struct mutex		tgt_qp_mutex;
1527 	struct list_head	tgt_qp_list;
1528 };
1529 
1530 struct ib_ah {
1531 	struct ib_device	*device;
1532 	struct ib_pd		*pd;
1533 	struct ib_uobject	*uobject;
1534 	enum rdma_ah_attr_type	type;
1535 };
1536 
1537 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context);
1538 
1539 enum ib_poll_context {
1540 	IB_POLL_DIRECT,		/* caller context, no hw completions */
1541 	IB_POLL_SOFTIRQ,	/* poll from softirq context */
1542 	IB_POLL_WORKQUEUE,	/* poll from workqueue */
1543 };
1544 
1545 struct ib_cq {
1546 	struct ib_device       *device;
1547 	struct ib_uobject      *uobject;
1548 	ib_comp_handler   	comp_handler;
1549 	void                  (*event_handler)(struct ib_event *, void *);
1550 	void                   *cq_context;
1551 	int               	cqe;
1552 	atomic_t          	usecnt; /* count number of work queues */
1553 	enum ib_poll_context	poll_ctx;
1554 	struct ib_wc		*wc;
1555 	union {
1556 		struct irq_poll		iop;
1557 		struct work_struct	work;
1558 	};
1559 };
1560 
1561 struct ib_srq {
1562 	struct ib_device       *device;
1563 	struct ib_pd	       *pd;
1564 	struct ib_uobject      *uobject;
1565 	void		      (*event_handler)(struct ib_event *, void *);
1566 	void		       *srq_context;
1567 	enum ib_srq_type	srq_type;
1568 	atomic_t		usecnt;
1569 
1570 	struct {
1571 		struct ib_cq   *cq;
1572 		union {
1573 			struct {
1574 				struct ib_xrcd *xrcd;
1575 				u32		srq_num;
1576 			} xrc;
1577 		};
1578 	} ext;
1579 };
1580 
1581 enum ib_raw_packet_caps {
1582 	/* Strip cvlan from incoming packet and report it in the matching work
1583 	 * completion is supported.
1584 	 */
1585 	IB_RAW_PACKET_CAP_CVLAN_STRIPPING	= (1 << 0),
1586 	/* Scatter FCS field of an incoming packet to host memory is supported.
1587 	 */
1588 	IB_RAW_PACKET_CAP_SCATTER_FCS		= (1 << 1),
1589 	/* Checksum offloads are supported (for both send and receive). */
1590 	IB_RAW_PACKET_CAP_IP_CSUM		= (1 << 2),
1591 	/* When a packet is received for an RQ with no receive WQEs, the
1592 	 * packet processing is delayed.
1593 	 */
1594 	IB_RAW_PACKET_CAP_DELAY_DROP		= (1 << 3),
1595 };
1596 
1597 enum ib_wq_type {
1598 	IB_WQT_RQ
1599 };
1600 
1601 enum ib_wq_state {
1602 	IB_WQS_RESET,
1603 	IB_WQS_RDY,
1604 	IB_WQS_ERR
1605 };
1606 
1607 struct ib_wq {
1608 	struct ib_device       *device;
1609 	struct ib_uobject      *uobject;
1610 	void		    *wq_context;
1611 	void		    (*event_handler)(struct ib_event *, void *);
1612 	struct ib_pd	       *pd;
1613 	struct ib_cq	       *cq;
1614 	u32		wq_num;
1615 	enum ib_wq_state       state;
1616 	enum ib_wq_type	wq_type;
1617 	atomic_t		usecnt;
1618 };
1619 
1620 enum ib_wq_flags {
1621 	IB_WQ_FLAGS_CVLAN_STRIPPING	= 1 << 0,
1622 	IB_WQ_FLAGS_SCATTER_FCS		= 1 << 1,
1623 	IB_WQ_FLAGS_DELAY_DROP		= 1 << 2,
1624 };
1625 
1626 struct ib_wq_init_attr {
1627 	void		       *wq_context;
1628 	enum ib_wq_type	wq_type;
1629 	u32		max_wr;
1630 	u32		max_sge;
1631 	struct	ib_cq	       *cq;
1632 	void		    (*event_handler)(struct ib_event *, void *);
1633 	u32		create_flags; /* Use enum ib_wq_flags */
1634 };
1635 
1636 enum ib_wq_attr_mask {
1637 	IB_WQ_STATE		= 1 << 0,
1638 	IB_WQ_CUR_STATE		= 1 << 1,
1639 	IB_WQ_FLAGS		= 1 << 2,
1640 };
1641 
1642 struct ib_wq_attr {
1643 	enum	ib_wq_state	wq_state;
1644 	enum	ib_wq_state	curr_wq_state;
1645 	u32			flags; /* Use enum ib_wq_flags */
1646 	u32			flags_mask; /* Use enum ib_wq_flags */
1647 };
1648 
1649 struct ib_rwq_ind_table {
1650 	struct ib_device	*device;
1651 	struct ib_uobject      *uobject;
1652 	atomic_t		usecnt;
1653 	u32		ind_tbl_num;
1654 	u32		log_ind_tbl_size;
1655 	struct ib_wq	**ind_tbl;
1656 };
1657 
1658 struct ib_rwq_ind_table_init_attr {
1659 	u32		log_ind_tbl_size;
1660 	/* Each entry is a pointer to Receive Work Queue */
1661 	struct ib_wq	**ind_tbl;
1662 };
1663 
1664 enum port_pkey_state {
1665 	IB_PORT_PKEY_NOT_VALID = 0,
1666 	IB_PORT_PKEY_VALID = 1,
1667 	IB_PORT_PKEY_LISTED = 2,
1668 };
1669 
1670 struct ib_qp_security;
1671 
1672 struct ib_port_pkey {
1673 	enum port_pkey_state	state;
1674 	u16			pkey_index;
1675 	u8			port_num;
1676 	struct list_head	qp_list;
1677 	struct list_head	to_error_list;
1678 	struct ib_qp_security  *sec;
1679 };
1680 
1681 struct ib_ports_pkeys {
1682 	struct ib_port_pkey	main;
1683 	struct ib_port_pkey	alt;
1684 };
1685 
1686 struct ib_qp_security {
1687 	struct ib_qp	       *qp;
1688 	struct ib_device       *dev;
1689 	/* Hold this mutex when changing port and pkey settings. */
1690 	struct mutex		mutex;
1691 	struct ib_ports_pkeys  *ports_pkeys;
1692 	/* A list of all open shared QP handles.  Required to enforce security
1693 	 * properly for all users of a shared QP.
1694 	 */
1695 	struct list_head        shared_qp_list;
1696 	void                   *security;
1697 	bool			destroying;
1698 	atomic_t		error_list_count;
1699 	struct completion	error_complete;
1700 	int			error_comps_pending;
1701 };
1702 
1703 /*
1704  * @max_write_sge: Maximum SGE elements per RDMA WRITE request.
1705  * @max_read_sge:  Maximum SGE elements per RDMA READ request.
1706  */
1707 struct ib_qp {
1708 	struct ib_device       *device;
1709 	struct ib_pd	       *pd;
1710 	struct ib_cq	       *send_cq;
1711 	struct ib_cq	       *recv_cq;
1712 	spinlock_t		mr_lock;
1713 	int			mrs_used;
1714 	struct list_head	rdma_mrs;
1715 	struct list_head	sig_mrs;
1716 	struct ib_srq	       *srq;
1717 	struct ib_xrcd	       *xrcd; /* XRC TGT QPs only */
1718 	struct list_head	xrcd_list;
1719 
1720 	/* count times opened, mcast attaches, flow attaches */
1721 	atomic_t		usecnt;
1722 	struct list_head	open_list;
1723 	struct ib_qp           *real_qp;
1724 	struct ib_uobject      *uobject;
1725 	void                  (*event_handler)(struct ib_event *, void *);
1726 	void		       *qp_context;
1727 	u32			qp_num;
1728 	u32			max_write_sge;
1729 	u32			max_read_sge;
1730 	enum ib_qp_type		qp_type;
1731 	struct ib_rwq_ind_table *rwq_ind_tbl;
1732 	struct ib_qp_security  *qp_sec;
1733 	u8			port;
1734 };
1735 
1736 struct ib_mr {
1737 	struct ib_device  *device;
1738 	struct ib_pd	  *pd;
1739 	u32		   lkey;
1740 	u32		   rkey;
1741 	u64		   iova;
1742 	u32		   length;
1743 	unsigned int	   page_size;
1744 	bool		   need_inval;
1745 	union {
1746 		struct ib_uobject	*uobject;	/* user */
1747 		struct list_head	qp_entry;	/* FR */
1748 	};
1749 };
1750 
1751 struct ib_mw {
1752 	struct ib_device	*device;
1753 	struct ib_pd		*pd;
1754 	struct ib_uobject	*uobject;
1755 	u32			rkey;
1756 	enum ib_mw_type         type;
1757 };
1758 
1759 struct ib_fmr {
1760 	struct ib_device	*device;
1761 	struct ib_pd		*pd;
1762 	struct list_head	list;
1763 	u32			lkey;
1764 	u32			rkey;
1765 };
1766 
1767 /* Supported steering options */
1768 enum ib_flow_attr_type {
1769 	/* steering according to rule specifications */
1770 	IB_FLOW_ATTR_NORMAL		= 0x0,
1771 	/* default unicast and multicast rule -
1772 	 * receive all Eth traffic which isn't steered to any QP
1773 	 */
1774 	IB_FLOW_ATTR_ALL_DEFAULT	= 0x1,
1775 	/* default multicast rule -
1776 	 * receive all Eth multicast traffic which isn't steered to any QP
1777 	 */
1778 	IB_FLOW_ATTR_MC_DEFAULT		= 0x2,
1779 	/* sniffer rule - receive all port traffic */
1780 	IB_FLOW_ATTR_SNIFFER		= 0x3
1781 };
1782 
1783 /* Supported steering header types */
1784 enum ib_flow_spec_type {
1785 	/* L2 headers*/
1786 	IB_FLOW_SPEC_ETH		= 0x20,
1787 	IB_FLOW_SPEC_IB			= 0x22,
1788 	/* L3 header*/
1789 	IB_FLOW_SPEC_IPV4		= 0x30,
1790 	IB_FLOW_SPEC_IPV6		= 0x31,
1791 	/* L4 headers*/
1792 	IB_FLOW_SPEC_TCP		= 0x40,
1793 	IB_FLOW_SPEC_UDP		= 0x41,
1794 	IB_FLOW_SPEC_VXLAN_TUNNEL	= 0x50,
1795 	IB_FLOW_SPEC_INNER		= 0x100,
1796 	/* Actions */
1797 	IB_FLOW_SPEC_ACTION_TAG         = 0x1000,
1798 	IB_FLOW_SPEC_ACTION_DROP        = 0x1001,
1799 };
1800 #define IB_FLOW_SPEC_LAYER_MASK	0xF0
1801 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8
1802 
1803 /* Flow steering rule priority is set according to it's domain.
1804  * Lower domain value means higher priority.
1805  */
1806 enum ib_flow_domain {
1807 	IB_FLOW_DOMAIN_USER,
1808 	IB_FLOW_DOMAIN_ETHTOOL,
1809 	IB_FLOW_DOMAIN_RFS,
1810 	IB_FLOW_DOMAIN_NIC,
1811 	IB_FLOW_DOMAIN_NUM /* Must be last */
1812 };
1813 
1814 enum ib_flow_flags {
1815 	IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */
1816 	IB_FLOW_ATTR_FLAGS_RESERVED  = 1UL << 2  /* Must be last */
1817 };
1818 
1819 struct ib_flow_eth_filter {
1820 	u8	dst_mac[6];
1821 	u8	src_mac[6];
1822 	__be16	ether_type;
1823 	__be16	vlan_tag;
1824 	/* Must be last */
1825 	u8	real_sz[0];
1826 };
1827 
1828 struct ib_flow_spec_eth {
1829 	u32			  type;
1830 	u16			  size;
1831 	struct ib_flow_eth_filter val;
1832 	struct ib_flow_eth_filter mask;
1833 };
1834 
1835 struct ib_flow_ib_filter {
1836 	__be16 dlid;
1837 	__u8   sl;
1838 	/* Must be last */
1839 	u8	real_sz[0];
1840 };
1841 
1842 struct ib_flow_spec_ib {
1843 	u32			 type;
1844 	u16			 size;
1845 	struct ib_flow_ib_filter val;
1846 	struct ib_flow_ib_filter mask;
1847 };
1848 
1849 /* IPv4 header flags */
1850 enum ib_ipv4_flags {
1851 	IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */
1852 	IB_IPV4_MORE_FRAG = 0X4  /* For All fragmented packets except the
1853 				    last have this flag set */
1854 };
1855 
1856 struct ib_flow_ipv4_filter {
1857 	__be32	src_ip;
1858 	__be32	dst_ip;
1859 	u8	proto;
1860 	u8	tos;
1861 	u8	ttl;
1862 	u8	flags;
1863 	/* Must be last */
1864 	u8	real_sz[0];
1865 };
1866 
1867 struct ib_flow_spec_ipv4 {
1868 	u32			   type;
1869 	u16			   size;
1870 	struct ib_flow_ipv4_filter val;
1871 	struct ib_flow_ipv4_filter mask;
1872 };
1873 
1874 struct ib_flow_ipv6_filter {
1875 	u8	src_ip[16];
1876 	u8	dst_ip[16];
1877 	__be32	flow_label;
1878 	u8	next_hdr;
1879 	u8	traffic_class;
1880 	u8	hop_limit;
1881 	/* Must be last */
1882 	u8	real_sz[0];
1883 };
1884 
1885 struct ib_flow_spec_ipv6 {
1886 	u32			   type;
1887 	u16			   size;
1888 	struct ib_flow_ipv6_filter val;
1889 	struct ib_flow_ipv6_filter mask;
1890 };
1891 
1892 struct ib_flow_tcp_udp_filter {
1893 	__be16	dst_port;
1894 	__be16	src_port;
1895 	/* Must be last */
1896 	u8	real_sz[0];
1897 };
1898 
1899 struct ib_flow_spec_tcp_udp {
1900 	u32			      type;
1901 	u16			      size;
1902 	struct ib_flow_tcp_udp_filter val;
1903 	struct ib_flow_tcp_udp_filter mask;
1904 };
1905 
1906 struct ib_flow_tunnel_filter {
1907 	__be32	tunnel_id;
1908 	u8	real_sz[0];
1909 };
1910 
1911 /* ib_flow_spec_tunnel describes the Vxlan tunnel
1912  * the tunnel_id from val has the vni value
1913  */
1914 struct ib_flow_spec_tunnel {
1915 	u32			      type;
1916 	u16			      size;
1917 	struct ib_flow_tunnel_filter  val;
1918 	struct ib_flow_tunnel_filter  mask;
1919 };
1920 
1921 struct ib_flow_spec_action_tag {
1922 	enum ib_flow_spec_type	      type;
1923 	u16			      size;
1924 	u32                           tag_id;
1925 };
1926 
1927 struct ib_flow_spec_action_drop {
1928 	enum ib_flow_spec_type	      type;
1929 	u16			      size;
1930 };
1931 
1932 union ib_flow_spec {
1933 	struct {
1934 		u32			type;
1935 		u16			size;
1936 	};
1937 	struct ib_flow_spec_eth		eth;
1938 	struct ib_flow_spec_ib		ib;
1939 	struct ib_flow_spec_ipv4        ipv4;
1940 	struct ib_flow_spec_tcp_udp	tcp_udp;
1941 	struct ib_flow_spec_ipv6        ipv6;
1942 	struct ib_flow_spec_tunnel      tunnel;
1943 	struct ib_flow_spec_action_tag  flow_tag;
1944 	struct ib_flow_spec_action_drop drop;
1945 };
1946 
1947 struct ib_flow_attr {
1948 	enum ib_flow_attr_type type;
1949 	u16	     size;
1950 	u16	     priority;
1951 	u32	     flags;
1952 	u8	     num_of_specs;
1953 	u8	     port;
1954 	/* Following are the optional layers according to user request
1955 	 * struct ib_flow_spec_xxx
1956 	 * struct ib_flow_spec_yyy
1957 	 */
1958 };
1959 
1960 struct ib_flow {
1961 	struct ib_qp		*qp;
1962 	struct ib_uobject	*uobject;
1963 };
1964 
1965 struct ib_mad_hdr;
1966 struct ib_grh;
1967 
1968 enum ib_process_mad_flags {
1969 	IB_MAD_IGNORE_MKEY	= 1,
1970 	IB_MAD_IGNORE_BKEY	= 2,
1971 	IB_MAD_IGNORE_ALL	= IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY
1972 };
1973 
1974 enum ib_mad_result {
1975 	IB_MAD_RESULT_FAILURE  = 0,      /* (!SUCCESS is the important flag) */
1976 	IB_MAD_RESULT_SUCCESS  = 1 << 0, /* MAD was successfully processed   */
1977 	IB_MAD_RESULT_REPLY    = 1 << 1, /* Reply packet needs to be sent    */
1978 	IB_MAD_RESULT_CONSUMED = 1 << 2  /* Packet consumed: stop processing */
1979 };
1980 
1981 struct ib_port_cache {
1982 	u64		      subnet_prefix;
1983 	struct ib_pkey_cache  *pkey;
1984 	struct ib_gid_table   *gid;
1985 	u8                     lmc;
1986 	enum ib_port_state     port_state;
1987 };
1988 
1989 struct ib_cache {
1990 	rwlock_t                lock;
1991 	struct ib_event_handler event_handler;
1992 	struct ib_port_cache   *ports;
1993 };
1994 
1995 struct iw_cm_verbs;
1996 
1997 struct ib_port_immutable {
1998 	int                           pkey_tbl_len;
1999 	int                           gid_tbl_len;
2000 	u32                           core_cap_flags;
2001 	u32                           max_mad_size;
2002 };
2003 
2004 /* rdma netdev type - specifies protocol type */
2005 enum rdma_netdev_t {
2006 	RDMA_NETDEV_OPA_VNIC,
2007 	RDMA_NETDEV_IPOIB,
2008 };
2009 
2010 /**
2011  * struct rdma_netdev - rdma netdev
2012  * For cases where netstack interfacing is required.
2013  */
2014 struct rdma_netdev {
2015 	void              *clnt_priv;
2016 	struct ib_device  *hca;
2017 	u8                 port_num;
2018 
2019 	/* cleanup function must be specified */
2020 	void (*free_rdma_netdev)(struct net_device *netdev);
2021 
2022 	/* control functions */
2023 	void (*set_id)(struct net_device *netdev, int id);
2024 	/* send packet */
2025 	int (*send)(struct net_device *dev, struct sk_buff *skb,
2026 		    struct ib_ah *address, u32 dqpn);
2027 	/* multicast */
2028 	int (*attach_mcast)(struct net_device *dev, struct ib_device *hca,
2029 			    union ib_gid *gid, u16 mlid,
2030 			    int set_qkey, u32 qkey);
2031 	int (*detach_mcast)(struct net_device *dev, struct ib_device *hca,
2032 			    union ib_gid *gid, u16 mlid);
2033 };
2034 
2035 struct ib_port_pkey_list {
2036 	/* Lock to hold while modifying the list. */
2037 	spinlock_t                    list_lock;
2038 	struct list_head              pkey_list;
2039 };
2040 
2041 struct ib_device {
2042 	/* Do not access @dma_device directly from ULP nor from HW drivers. */
2043 	struct device                *dma_device;
2044 
2045 	char                          name[IB_DEVICE_NAME_MAX];
2046 
2047 	struct list_head              event_handler_list;
2048 	spinlock_t                    event_handler_lock;
2049 
2050 	spinlock_t                    client_data_lock;
2051 	struct list_head              core_list;
2052 	/* Access to the client_data_list is protected by the client_data_lock
2053 	 * spinlock and the lists_rwsem read-write semaphore */
2054 	struct list_head              client_data_list;
2055 
2056 	struct ib_cache               cache;
2057 	/**
2058 	 * port_immutable is indexed by port number
2059 	 */
2060 	struct ib_port_immutable     *port_immutable;
2061 
2062 	int			      num_comp_vectors;
2063 
2064 	struct ib_port_pkey_list     *port_pkey_list;
2065 
2066 	struct iw_cm_verbs	     *iwcm;
2067 
2068 	/**
2069 	 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the
2070 	 *   driver initialized data.  The struct is kfree()'ed by the sysfs
2071 	 *   core when the device is removed.  A lifespan of -1 in the return
2072 	 *   struct tells the core to set a default lifespan.
2073 	 */
2074 	struct rdma_hw_stats      *(*alloc_hw_stats)(struct ib_device *device,
2075 						     u8 port_num);
2076 	/**
2077 	 * get_hw_stats - Fill in the counter value(s) in the stats struct.
2078 	 * @index - The index in the value array we wish to have updated, or
2079 	 *   num_counters if we want all stats updated
2080 	 * Return codes -
2081 	 *   < 0 - Error, no counters updated
2082 	 *   index - Updated the single counter pointed to by index
2083 	 *   num_counters - Updated all counters (will reset the timestamp
2084 	 *     and prevent further calls for lifespan milliseconds)
2085 	 * Drivers are allowed to update all counters in leiu of just the
2086 	 *   one given in index at their option
2087 	 */
2088 	int		           (*get_hw_stats)(struct ib_device *device,
2089 						   struct rdma_hw_stats *stats,
2090 						   u8 port, int index);
2091 	int		           (*query_device)(struct ib_device *device,
2092 						   struct ib_device_attr *device_attr,
2093 						   struct ib_udata *udata);
2094 	int		           (*query_port)(struct ib_device *device,
2095 						 u8 port_num,
2096 						 struct ib_port_attr *port_attr);
2097 	enum rdma_link_layer	   (*get_link_layer)(struct ib_device *device,
2098 						     u8 port_num);
2099 	/* When calling get_netdev, the HW vendor's driver should return the
2100 	 * net device of device @device at port @port_num or NULL if such
2101 	 * a net device doesn't exist. The vendor driver should call dev_hold
2102 	 * on this net device. The HW vendor's device driver must guarantee
2103 	 * that this function returns NULL before the net device reaches
2104 	 * NETDEV_UNREGISTER_FINAL state.
2105 	 */
2106 	struct net_device	  *(*get_netdev)(struct ib_device *device,
2107 						 u8 port_num);
2108 	int		           (*query_gid)(struct ib_device *device,
2109 						u8 port_num, int index,
2110 						union ib_gid *gid);
2111 	/* When calling add_gid, the HW vendor's driver should
2112 	 * add the gid of device @device at gid index @index of
2113 	 * port @port_num to be @gid. Meta-info of that gid (for example,
2114 	 * the network device related to this gid is available
2115 	 * at @attr. @context allows the HW vendor driver to store extra
2116 	 * information together with a GID entry. The HW vendor may allocate
2117 	 * memory to contain this information and store it in @context when a
2118 	 * new GID entry is written to. Params are consistent until the next
2119 	 * call of add_gid or delete_gid. The function should return 0 on
2120 	 * success or error otherwise. The function could be called
2121 	 * concurrently for different ports. This function is only called
2122 	 * when roce_gid_table is used.
2123 	 */
2124 	int		           (*add_gid)(struct ib_device *device,
2125 					      u8 port_num,
2126 					      unsigned int index,
2127 					      const union ib_gid *gid,
2128 					      const struct ib_gid_attr *attr,
2129 					      void **context);
2130 	/* When calling del_gid, the HW vendor's driver should delete the
2131 	 * gid of device @device at gid index @index of port @port_num.
2132 	 * Upon the deletion of a GID entry, the HW vendor must free any
2133 	 * allocated memory. The caller will clear @context afterwards.
2134 	 * This function is only called when roce_gid_table is used.
2135 	 */
2136 	int		           (*del_gid)(struct ib_device *device,
2137 					      u8 port_num,
2138 					      unsigned int index,
2139 					      void **context);
2140 	int		           (*query_pkey)(struct ib_device *device,
2141 						 u8 port_num, u16 index, u16 *pkey);
2142 	int		           (*modify_device)(struct ib_device *device,
2143 						    int device_modify_mask,
2144 						    struct ib_device_modify *device_modify);
2145 	int		           (*modify_port)(struct ib_device *device,
2146 						  u8 port_num, int port_modify_mask,
2147 						  struct ib_port_modify *port_modify);
2148 	struct ib_ucontext *       (*alloc_ucontext)(struct ib_device *device,
2149 						     struct ib_udata *udata);
2150 	int                        (*dealloc_ucontext)(struct ib_ucontext *context);
2151 	int                        (*mmap)(struct ib_ucontext *context,
2152 					   struct vm_area_struct *vma);
2153 	struct ib_pd *             (*alloc_pd)(struct ib_device *device,
2154 					       struct ib_ucontext *context,
2155 					       struct ib_udata *udata);
2156 	int                        (*dealloc_pd)(struct ib_pd *pd);
2157 	struct ib_ah *             (*create_ah)(struct ib_pd *pd,
2158 						struct rdma_ah_attr *ah_attr,
2159 						struct ib_udata *udata);
2160 	int                        (*modify_ah)(struct ib_ah *ah,
2161 						struct rdma_ah_attr *ah_attr);
2162 	int                        (*query_ah)(struct ib_ah *ah,
2163 					       struct rdma_ah_attr *ah_attr);
2164 	int                        (*destroy_ah)(struct ib_ah *ah);
2165 	struct ib_srq *            (*create_srq)(struct ib_pd *pd,
2166 						 struct ib_srq_init_attr *srq_init_attr,
2167 						 struct ib_udata *udata);
2168 	int                        (*modify_srq)(struct ib_srq *srq,
2169 						 struct ib_srq_attr *srq_attr,
2170 						 enum ib_srq_attr_mask srq_attr_mask,
2171 						 struct ib_udata *udata);
2172 	int                        (*query_srq)(struct ib_srq *srq,
2173 						struct ib_srq_attr *srq_attr);
2174 	int                        (*destroy_srq)(struct ib_srq *srq);
2175 	int                        (*post_srq_recv)(struct ib_srq *srq,
2176 						    struct ib_recv_wr *recv_wr,
2177 						    struct ib_recv_wr **bad_recv_wr);
2178 	struct ib_qp *             (*create_qp)(struct ib_pd *pd,
2179 						struct ib_qp_init_attr *qp_init_attr,
2180 						struct ib_udata *udata);
2181 	int                        (*modify_qp)(struct ib_qp *qp,
2182 						struct ib_qp_attr *qp_attr,
2183 						int qp_attr_mask,
2184 						struct ib_udata *udata);
2185 	int                        (*query_qp)(struct ib_qp *qp,
2186 					       struct ib_qp_attr *qp_attr,
2187 					       int qp_attr_mask,
2188 					       struct ib_qp_init_attr *qp_init_attr);
2189 	int                        (*destroy_qp)(struct ib_qp *qp);
2190 	int                        (*post_send)(struct ib_qp *qp,
2191 						struct ib_send_wr *send_wr,
2192 						struct ib_send_wr **bad_send_wr);
2193 	int                        (*post_recv)(struct ib_qp *qp,
2194 						struct ib_recv_wr *recv_wr,
2195 						struct ib_recv_wr **bad_recv_wr);
2196 	struct ib_cq *             (*create_cq)(struct ib_device *device,
2197 						const struct ib_cq_init_attr *attr,
2198 						struct ib_ucontext *context,
2199 						struct ib_udata *udata);
2200 	int                        (*modify_cq)(struct ib_cq *cq, u16 cq_count,
2201 						u16 cq_period);
2202 	int                        (*destroy_cq)(struct ib_cq *cq);
2203 	int                        (*resize_cq)(struct ib_cq *cq, int cqe,
2204 						struct ib_udata *udata);
2205 	int                        (*poll_cq)(struct ib_cq *cq, int num_entries,
2206 					      struct ib_wc *wc);
2207 	int                        (*peek_cq)(struct ib_cq *cq, int wc_cnt);
2208 	int                        (*req_notify_cq)(struct ib_cq *cq,
2209 						    enum ib_cq_notify_flags flags);
2210 	int                        (*req_ncomp_notif)(struct ib_cq *cq,
2211 						      int wc_cnt);
2212 	struct ib_mr *             (*get_dma_mr)(struct ib_pd *pd,
2213 						 int mr_access_flags);
2214 	struct ib_mr *             (*reg_user_mr)(struct ib_pd *pd,
2215 						  u64 start, u64 length,
2216 						  u64 virt_addr,
2217 						  int mr_access_flags,
2218 						  struct ib_udata *udata);
2219 	int			   (*rereg_user_mr)(struct ib_mr *mr,
2220 						    int flags,
2221 						    u64 start, u64 length,
2222 						    u64 virt_addr,
2223 						    int mr_access_flags,
2224 						    struct ib_pd *pd,
2225 						    struct ib_udata *udata);
2226 	int                        (*dereg_mr)(struct ib_mr *mr);
2227 	struct ib_mr *		   (*alloc_mr)(struct ib_pd *pd,
2228 					       enum ib_mr_type mr_type,
2229 					       u32 max_num_sg);
2230 	int                        (*map_mr_sg)(struct ib_mr *mr,
2231 						struct scatterlist *sg,
2232 						int sg_nents,
2233 						unsigned int *sg_offset);
2234 	struct ib_mw *             (*alloc_mw)(struct ib_pd *pd,
2235 					       enum ib_mw_type type,
2236 					       struct ib_udata *udata);
2237 	int                        (*dealloc_mw)(struct ib_mw *mw);
2238 	struct ib_fmr *	           (*alloc_fmr)(struct ib_pd *pd,
2239 						int mr_access_flags,
2240 						struct ib_fmr_attr *fmr_attr);
2241 	int		           (*map_phys_fmr)(struct ib_fmr *fmr,
2242 						   u64 *page_list, int list_len,
2243 						   u64 iova);
2244 	int		           (*unmap_fmr)(struct list_head *fmr_list);
2245 	int		           (*dealloc_fmr)(struct ib_fmr *fmr);
2246 	int                        (*attach_mcast)(struct ib_qp *qp,
2247 						   union ib_gid *gid,
2248 						   u16 lid);
2249 	int                        (*detach_mcast)(struct ib_qp *qp,
2250 						   union ib_gid *gid,
2251 						   u16 lid);
2252 	int                        (*process_mad)(struct ib_device *device,
2253 						  int process_mad_flags,
2254 						  u8 port_num,
2255 						  const struct ib_wc *in_wc,
2256 						  const struct ib_grh *in_grh,
2257 						  const struct ib_mad_hdr *in_mad,
2258 						  size_t in_mad_size,
2259 						  struct ib_mad_hdr *out_mad,
2260 						  size_t *out_mad_size,
2261 						  u16 *out_mad_pkey_index);
2262 	struct ib_xrcd *	   (*alloc_xrcd)(struct ib_device *device,
2263 						 struct ib_ucontext *ucontext,
2264 						 struct ib_udata *udata);
2265 	int			   (*dealloc_xrcd)(struct ib_xrcd *xrcd);
2266 	struct ib_flow *	   (*create_flow)(struct ib_qp *qp,
2267 						  struct ib_flow_attr
2268 						  *flow_attr,
2269 						  int domain);
2270 	int			   (*destroy_flow)(struct ib_flow *flow_id);
2271 	int			   (*check_mr_status)(struct ib_mr *mr, u32 check_mask,
2272 						      struct ib_mr_status *mr_status);
2273 	void			   (*disassociate_ucontext)(struct ib_ucontext *ibcontext);
2274 	void			   (*drain_rq)(struct ib_qp *qp);
2275 	void			   (*drain_sq)(struct ib_qp *qp);
2276 	int			   (*set_vf_link_state)(struct ib_device *device, int vf, u8 port,
2277 							int state);
2278 	int			   (*get_vf_config)(struct ib_device *device, int vf, u8 port,
2279 						   struct ifla_vf_info *ivf);
2280 	int			   (*get_vf_stats)(struct ib_device *device, int vf, u8 port,
2281 						   struct ifla_vf_stats *stats);
2282 	int			   (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid,
2283 						  int type);
2284 	struct ib_wq *		   (*create_wq)(struct ib_pd *pd,
2285 						struct ib_wq_init_attr *init_attr,
2286 						struct ib_udata *udata);
2287 	int			   (*destroy_wq)(struct ib_wq *wq);
2288 	int			   (*modify_wq)(struct ib_wq *wq,
2289 						struct ib_wq_attr *attr,
2290 						u32 wq_attr_mask,
2291 						struct ib_udata *udata);
2292 	struct ib_rwq_ind_table *  (*create_rwq_ind_table)(struct ib_device *device,
2293 							   struct ib_rwq_ind_table_init_attr *init_attr,
2294 							   struct ib_udata *udata);
2295 	int                        (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table);
2296 	/**
2297 	 * rdma netdev operation
2298 	 *
2299 	 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it
2300 	 * doesn't support the specified rdma netdev type.
2301 	 */
2302 	struct net_device *(*alloc_rdma_netdev)(
2303 					struct ib_device *device,
2304 					u8 port_num,
2305 					enum rdma_netdev_t type,
2306 					const char *name,
2307 					unsigned char name_assign_type,
2308 					void (*setup)(struct net_device *));
2309 
2310 	struct module               *owner;
2311 	struct device                dev;
2312 	struct kobject               *ports_parent;
2313 	struct list_head             port_list;
2314 
2315 	enum {
2316 		IB_DEV_UNINITIALIZED,
2317 		IB_DEV_REGISTERED,
2318 		IB_DEV_UNREGISTERED
2319 	}                            reg_state;
2320 
2321 	int			     uverbs_abi_ver;
2322 	u64			     uverbs_cmd_mask;
2323 	u64			     uverbs_ex_cmd_mask;
2324 
2325 	char			     node_desc[IB_DEVICE_NODE_DESC_MAX];
2326 	__be64			     node_guid;
2327 	u32			     local_dma_lkey;
2328 	u16                          is_switch:1;
2329 	u8                           node_type;
2330 	u8                           phys_port_cnt;
2331 	struct ib_device_attr        attrs;
2332 	struct attribute_group	     *hw_stats_ag;
2333 	struct rdma_hw_stats         *hw_stats;
2334 
2335 #ifdef CONFIG_CGROUP_RDMA
2336 	struct rdmacg_device         cg_device;
2337 #endif
2338 
2339 	u32                          index;
2340 
2341 	/**
2342 	 * The following mandatory functions are used only at device
2343 	 * registration.  Keep functions such as these at the end of this
2344 	 * structure to avoid cache line misses when accessing struct ib_device
2345 	 * in fast paths.
2346 	 */
2347 	int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *);
2348 	void (*get_dev_fw_str)(struct ib_device *, char *str);
2349 	const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev,
2350 						     int comp_vector);
2351 
2352 	struct uverbs_root_spec		*specs_root;
2353 };
2354 
2355 struct ib_client {
2356 	char  *name;
2357 	void (*add)   (struct ib_device *);
2358 	void (*remove)(struct ib_device *, void *client_data);
2359 
2360 	/* Returns the net_dev belonging to this ib_client and matching the
2361 	 * given parameters.
2362 	 * @dev:	 An RDMA device that the net_dev use for communication.
2363 	 * @port:	 A physical port number on the RDMA device.
2364 	 * @pkey:	 P_Key that the net_dev uses if applicable.
2365 	 * @gid:	 A GID that the net_dev uses to communicate.
2366 	 * @addr:	 An IP address the net_dev is configured with.
2367 	 * @client_data: The device's client data set by ib_set_client_data().
2368 	 *
2369 	 * An ib_client that implements a net_dev on top of RDMA devices
2370 	 * (such as IP over IB) should implement this callback, allowing the
2371 	 * rdma_cm module to find the right net_dev for a given request.
2372 	 *
2373 	 * The caller is responsible for calling dev_put on the returned
2374 	 * netdev. */
2375 	struct net_device *(*get_net_dev_by_params)(
2376 			struct ib_device *dev,
2377 			u8 port,
2378 			u16 pkey,
2379 			const union ib_gid *gid,
2380 			const struct sockaddr *addr,
2381 			void *client_data);
2382 	struct list_head list;
2383 };
2384 
2385 struct ib_device *ib_alloc_device(size_t size);
2386 void ib_dealloc_device(struct ib_device *device);
2387 
2388 void ib_get_device_fw_str(struct ib_device *device, char *str);
2389 
2390 int ib_register_device(struct ib_device *device,
2391 		       int (*port_callback)(struct ib_device *,
2392 					    u8, struct kobject *));
2393 void ib_unregister_device(struct ib_device *device);
2394 
2395 int ib_register_client   (struct ib_client *client);
2396 void ib_unregister_client(struct ib_client *client);
2397 
2398 void *ib_get_client_data(struct ib_device *device, struct ib_client *client);
2399 void  ib_set_client_data(struct ib_device *device, struct ib_client *client,
2400 			 void *data);
2401 
2402 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len)
2403 {
2404 	return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0;
2405 }
2406 
2407 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len)
2408 {
2409 	return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0;
2410 }
2411 
2412 static inline bool ib_is_udata_cleared(struct ib_udata *udata,
2413 				       size_t offset,
2414 				       size_t len)
2415 {
2416 	const void __user *p = udata->inbuf + offset;
2417 	bool ret;
2418 	u8 *buf;
2419 
2420 	if (len > USHRT_MAX)
2421 		return false;
2422 
2423 	buf = memdup_user(p, len);
2424 	if (IS_ERR(buf))
2425 		return false;
2426 
2427 	ret = !memchr_inv(buf, 0, len);
2428 	kfree(buf);
2429 	return ret;
2430 }
2431 
2432 /**
2433  * ib_modify_qp_is_ok - Check that the supplied attribute mask
2434  * contains all required attributes and no attributes not allowed for
2435  * the given QP state transition.
2436  * @cur_state: Current QP state
2437  * @next_state: Next QP state
2438  * @type: QP type
2439  * @mask: Mask of supplied QP attributes
2440  * @ll : link layer of port
2441  *
2442  * This function is a helper function that a low-level driver's
2443  * modify_qp method can use to validate the consumer's input.  It
2444  * checks that cur_state and next_state are valid QP states, that a
2445  * transition from cur_state to next_state is allowed by the IB spec,
2446  * and that the attribute mask supplied is allowed for the transition.
2447  */
2448 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state,
2449 		       enum ib_qp_type type, enum ib_qp_attr_mask mask,
2450 		       enum rdma_link_layer ll);
2451 
2452 void ib_register_event_handler(struct ib_event_handler *event_handler);
2453 void ib_unregister_event_handler(struct ib_event_handler *event_handler);
2454 void ib_dispatch_event(struct ib_event *event);
2455 
2456 int ib_query_port(struct ib_device *device,
2457 		  u8 port_num, struct ib_port_attr *port_attr);
2458 
2459 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device,
2460 					       u8 port_num);
2461 
2462 /**
2463  * rdma_cap_ib_switch - Check if the device is IB switch
2464  * @device: Device to check
2465  *
2466  * Device driver is responsible for setting is_switch bit on
2467  * in ib_device structure at init time.
2468  *
2469  * Return: true if the device is IB switch.
2470  */
2471 static inline bool rdma_cap_ib_switch(const struct ib_device *device)
2472 {
2473 	return device->is_switch;
2474 }
2475 
2476 /**
2477  * rdma_start_port - Return the first valid port number for the device
2478  * specified
2479  *
2480  * @device: Device to be checked
2481  *
2482  * Return start port number
2483  */
2484 static inline u8 rdma_start_port(const struct ib_device *device)
2485 {
2486 	return rdma_cap_ib_switch(device) ? 0 : 1;
2487 }
2488 
2489 /**
2490  * rdma_end_port - Return the last valid port number for the device
2491  * specified
2492  *
2493  * @device: Device to be checked
2494  *
2495  * Return last port number
2496  */
2497 static inline u8 rdma_end_port(const struct ib_device *device)
2498 {
2499 	return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt;
2500 }
2501 
2502 static inline int rdma_is_port_valid(const struct ib_device *device,
2503 				     unsigned int port)
2504 {
2505 	return (port >= rdma_start_port(device) &&
2506 		port <= rdma_end_port(device));
2507 }
2508 
2509 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num)
2510 {
2511 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB;
2512 }
2513 
2514 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num)
2515 {
2516 	return device->port_immutable[port_num].core_cap_flags &
2517 		(RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP);
2518 }
2519 
2520 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num)
2521 {
2522 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP;
2523 }
2524 
2525 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num)
2526 {
2527 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE;
2528 }
2529 
2530 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num)
2531 {
2532 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP;
2533 }
2534 
2535 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num)
2536 {
2537 	return rdma_protocol_ib(device, port_num) ||
2538 		rdma_protocol_roce(device, port_num);
2539 }
2540 
2541 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num)
2542 {
2543 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET;
2544 }
2545 
2546 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num)
2547 {
2548 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC;
2549 }
2550 
2551 /**
2552  * rdma_cap_ib_mad - Check if the port of a device supports Infiniband
2553  * Management Datagrams.
2554  * @device: Device to check
2555  * @port_num: Port number to check
2556  *
2557  * Management Datagrams (MAD) are a required part of the InfiniBand
2558  * specification and are supported on all InfiniBand devices.  A slightly
2559  * extended version are also supported on OPA interfaces.
2560  *
2561  * Return: true if the port supports sending/receiving of MAD packets.
2562  */
2563 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num)
2564 {
2565 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD;
2566 }
2567 
2568 /**
2569  * rdma_cap_opa_mad - Check if the port of device provides support for OPA
2570  * Management Datagrams.
2571  * @device: Device to check
2572  * @port_num: Port number to check
2573  *
2574  * Intel OmniPath devices extend and/or replace the InfiniBand Management
2575  * datagrams with their own versions.  These OPA MADs share many but not all of
2576  * the characteristics of InfiniBand MADs.
2577  *
2578  * OPA MADs differ in the following ways:
2579  *
2580  *    1) MADs are variable size up to 2K
2581  *       IBTA defined MADs remain fixed at 256 bytes
2582  *    2) OPA SMPs must carry valid PKeys
2583  *    3) OPA SMP packets are a different format
2584  *
2585  * Return: true if the port supports OPA MAD packet formats.
2586  */
2587 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num)
2588 {
2589 	return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD)
2590 		== RDMA_CORE_CAP_OPA_MAD;
2591 }
2592 
2593 /**
2594  * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband
2595  * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI).
2596  * @device: Device to check
2597  * @port_num: Port number to check
2598  *
2599  * Each InfiniBand node is required to provide a Subnet Management Agent
2600  * that the subnet manager can access.  Prior to the fabric being fully
2601  * configured by the subnet manager, the SMA is accessed via a well known
2602  * interface called the Subnet Management Interface (SMI).  This interface
2603  * uses directed route packets to communicate with the SM to get around the
2604  * chicken and egg problem of the SM needing to know what's on the fabric
2605  * in order to configure the fabric, and needing to configure the fabric in
2606  * order to send packets to the devices on the fabric.  These directed
2607  * route packets do not need the fabric fully configured in order to reach
2608  * their destination.  The SMI is the only method allowed to send
2609  * directed route packets on an InfiniBand fabric.
2610  *
2611  * Return: true if the port provides an SMI.
2612  */
2613 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num)
2614 {
2615 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI;
2616 }
2617 
2618 /**
2619  * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband
2620  * Communication Manager.
2621  * @device: Device to check
2622  * @port_num: Port number to check
2623  *
2624  * The InfiniBand Communication Manager is one of many pre-defined General
2625  * Service Agents (GSA) that are accessed via the General Service
2626  * Interface (GSI).  It's role is to facilitate establishment of connections
2627  * between nodes as well as other management related tasks for established
2628  * connections.
2629  *
2630  * Return: true if the port supports an IB CM (this does not guarantee that
2631  * a CM is actually running however).
2632  */
2633 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num)
2634 {
2635 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM;
2636 }
2637 
2638 /**
2639  * rdma_cap_iw_cm - Check if the port of device has the capability IWARP
2640  * Communication Manager.
2641  * @device: Device to check
2642  * @port_num: Port number to check
2643  *
2644  * Similar to above, but specific to iWARP connections which have a different
2645  * managment protocol than InfiniBand.
2646  *
2647  * Return: true if the port supports an iWARP CM (this does not guarantee that
2648  * a CM is actually running however).
2649  */
2650 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num)
2651 {
2652 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM;
2653 }
2654 
2655 /**
2656  * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband
2657  * Subnet Administration.
2658  * @device: Device to check
2659  * @port_num: Port number to check
2660  *
2661  * An InfiniBand Subnet Administration (SA) service is a pre-defined General
2662  * Service Agent (GSA) provided by the Subnet Manager (SM).  On InfiniBand
2663  * fabrics, devices should resolve routes to other hosts by contacting the
2664  * SA to query the proper route.
2665  *
2666  * Return: true if the port should act as a client to the fabric Subnet
2667  * Administration interface.  This does not imply that the SA service is
2668  * running locally.
2669  */
2670 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num)
2671 {
2672 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA;
2673 }
2674 
2675 /**
2676  * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband
2677  * Multicast.
2678  * @device: Device to check
2679  * @port_num: Port number to check
2680  *
2681  * InfiniBand multicast registration is more complex than normal IPv4 or
2682  * IPv6 multicast registration.  Each Host Channel Adapter must register
2683  * with the Subnet Manager when it wishes to join a multicast group.  It
2684  * should do so only once regardless of how many queue pairs it subscribes
2685  * to this group.  And it should leave the group only after all queue pairs
2686  * attached to the group have been detached.
2687  *
2688  * Return: true if the port must undertake the additional adminstrative
2689  * overhead of registering/unregistering with the SM and tracking of the
2690  * total number of queue pairs attached to the multicast group.
2691  */
2692 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num)
2693 {
2694 	return rdma_cap_ib_sa(device, port_num);
2695 }
2696 
2697 /**
2698  * rdma_cap_af_ib - Check if the port of device has the capability
2699  * Native Infiniband Address.
2700  * @device: Device to check
2701  * @port_num: Port number to check
2702  *
2703  * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default
2704  * GID.  RoCE uses a different mechanism, but still generates a GID via
2705  * a prescribed mechanism and port specific data.
2706  *
2707  * Return: true if the port uses a GID address to identify devices on the
2708  * network.
2709  */
2710 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num)
2711 {
2712 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB;
2713 }
2714 
2715 /**
2716  * rdma_cap_eth_ah - Check if the port of device has the capability
2717  * Ethernet Address Handle.
2718  * @device: Device to check
2719  * @port_num: Port number to check
2720  *
2721  * RoCE is InfiniBand over Ethernet, and it uses a well defined technique
2722  * to fabricate GIDs over Ethernet/IP specific addresses native to the
2723  * port.  Normally, packet headers are generated by the sending host
2724  * adapter, but when sending connectionless datagrams, we must manually
2725  * inject the proper headers for the fabric we are communicating over.
2726  *
2727  * Return: true if we are running as a RoCE port and must force the
2728  * addition of a Global Route Header built from our Ethernet Address
2729  * Handle into our header list for connectionless packets.
2730  */
2731 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num)
2732 {
2733 	return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH;
2734 }
2735 
2736 /**
2737  * rdma_cap_opa_ah - Check if the port of device supports
2738  * OPA Address handles
2739  * @device: Device to check
2740  * @port_num: Port number to check
2741  *
2742  * Return: true if we are running on an OPA device which supports
2743  * the extended OPA addressing.
2744  */
2745 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num)
2746 {
2747 	return (device->port_immutable[port_num].core_cap_flags &
2748 		RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH;
2749 }
2750 
2751 /**
2752  * rdma_max_mad_size - Return the max MAD size required by this RDMA Port.
2753  *
2754  * @device: Device
2755  * @port_num: Port number
2756  *
2757  * This MAD size includes the MAD headers and MAD payload.  No other headers
2758  * are included.
2759  *
2760  * Return the max MAD size required by the Port.  Will return 0 if the port
2761  * does not support MADs
2762  */
2763 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num)
2764 {
2765 	return device->port_immutable[port_num].max_mad_size;
2766 }
2767 
2768 /**
2769  * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table
2770  * @device: Device to check
2771  * @port_num: Port number to check
2772  *
2773  * RoCE GID table mechanism manages the various GIDs for a device.
2774  *
2775  * NOTE: if allocating the port's GID table has failed, this call will still
2776  * return true, but any RoCE GID table API will fail.
2777  *
2778  * Return: true if the port uses RoCE GID table mechanism in order to manage
2779  * its GIDs.
2780  */
2781 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device,
2782 					   u8 port_num)
2783 {
2784 	return rdma_protocol_roce(device, port_num) &&
2785 		device->add_gid && device->del_gid;
2786 }
2787 
2788 /*
2789  * Check if the device supports READ W/ INVALIDATE.
2790  */
2791 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num)
2792 {
2793 	/*
2794 	 * iWarp drivers must support READ W/ INVALIDATE.  No other protocol
2795 	 * has support for it yet.
2796 	 */
2797 	return rdma_protocol_iwarp(dev, port_num);
2798 }
2799 
2800 int ib_query_gid(struct ib_device *device,
2801 		 u8 port_num, int index, union ib_gid *gid,
2802 		 struct ib_gid_attr *attr);
2803 
2804 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port,
2805 			 int state);
2806 int ib_get_vf_config(struct ib_device *device, int vf, u8 port,
2807 		     struct ifla_vf_info *info);
2808 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port,
2809 		    struct ifla_vf_stats *stats);
2810 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid,
2811 		   int type);
2812 
2813 int ib_query_pkey(struct ib_device *device,
2814 		  u8 port_num, u16 index, u16 *pkey);
2815 
2816 int ib_modify_device(struct ib_device *device,
2817 		     int device_modify_mask,
2818 		     struct ib_device_modify *device_modify);
2819 
2820 int ib_modify_port(struct ib_device *device,
2821 		   u8 port_num, int port_modify_mask,
2822 		   struct ib_port_modify *port_modify);
2823 
2824 int ib_find_gid(struct ib_device *device, union ib_gid *gid,
2825 		enum ib_gid_type gid_type, struct net_device *ndev,
2826 		u8 *port_num, u16 *index);
2827 
2828 int ib_find_pkey(struct ib_device *device,
2829 		 u8 port_num, u16 pkey, u16 *index);
2830 
2831 enum ib_pd_flags {
2832 	/*
2833 	 * Create a memory registration for all memory in the system and place
2834 	 * the rkey for it into pd->unsafe_global_rkey.  This can be used by
2835 	 * ULPs to avoid the overhead of dynamic MRs.
2836 	 *
2837 	 * This flag is generally considered unsafe and must only be used in
2838 	 * extremly trusted environments.  Every use of it will log a warning
2839 	 * in the kernel log.
2840 	 */
2841 	IB_PD_UNSAFE_GLOBAL_RKEY	= 0x01,
2842 };
2843 
2844 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags,
2845 		const char *caller);
2846 #define ib_alloc_pd(device, flags) \
2847 	__ib_alloc_pd((device), (flags), __func__)
2848 void ib_dealloc_pd(struct ib_pd *pd);
2849 
2850 /**
2851  * rdma_create_ah - Creates an address handle for the given address vector.
2852  * @pd: The protection domain associated with the address handle.
2853  * @ah_attr: The attributes of the address vector.
2854  *
2855  * The address handle is used to reference a local or global destination
2856  * in all UD QP post sends.
2857  */
2858 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr);
2859 
2860 /**
2861  * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header
2862  *   work completion.
2863  * @hdr: the L3 header to parse
2864  * @net_type: type of header to parse
2865  * @sgid: place to store source gid
2866  * @dgid: place to store destination gid
2867  */
2868 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr,
2869 			      enum rdma_network_type net_type,
2870 			      union ib_gid *sgid, union ib_gid *dgid);
2871 
2872 /**
2873  * ib_get_rdma_header_version - Get the header version
2874  * @hdr: the L3 header to parse
2875  */
2876 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr);
2877 
2878 /**
2879  * ib_init_ah_from_wc - Initializes address handle attributes from a
2880  *   work completion.
2881  * @device: Device on which the received message arrived.
2882  * @port_num: Port on which the received message arrived.
2883  * @wc: Work completion associated with the received message.
2884  * @grh: References the received global route header.  This parameter is
2885  *   ignored unless the work completion indicates that the GRH is valid.
2886  * @ah_attr: Returned attributes that can be used when creating an address
2887  *   handle for replying to the message.
2888  */
2889 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num,
2890 		       const struct ib_wc *wc, const struct ib_grh *grh,
2891 		       struct rdma_ah_attr *ah_attr);
2892 
2893 /**
2894  * ib_create_ah_from_wc - Creates an address handle associated with the
2895  *   sender of the specified work completion.
2896  * @pd: The protection domain associated with the address handle.
2897  * @wc: Work completion information associated with a received message.
2898  * @grh: References the received global route header.  This parameter is
2899  *   ignored unless the work completion indicates that the GRH is valid.
2900  * @port_num: The outbound port number to associate with the address.
2901  *
2902  * The address handle is used to reference a local or global destination
2903  * in all UD QP post sends.
2904  */
2905 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc,
2906 				   const struct ib_grh *grh, u8 port_num);
2907 
2908 /**
2909  * rdma_modify_ah - Modifies the address vector associated with an address
2910  *   handle.
2911  * @ah: The address handle to modify.
2912  * @ah_attr: The new address vector attributes to associate with the
2913  *   address handle.
2914  */
2915 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2916 
2917 /**
2918  * rdma_query_ah - Queries the address vector associated with an address
2919  *   handle.
2920  * @ah: The address handle to query.
2921  * @ah_attr: The address vector attributes associated with the address
2922  *   handle.
2923  */
2924 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr);
2925 
2926 /**
2927  * rdma_destroy_ah - Destroys an address handle.
2928  * @ah: The address handle to destroy.
2929  */
2930 int rdma_destroy_ah(struct ib_ah *ah);
2931 
2932 /**
2933  * ib_create_srq - Creates a SRQ associated with the specified protection
2934  *   domain.
2935  * @pd: The protection domain associated with the SRQ.
2936  * @srq_init_attr: A list of initial attributes required to create the
2937  *   SRQ.  If SRQ creation succeeds, then the attributes are updated to
2938  *   the actual capabilities of the created SRQ.
2939  *
2940  * srq_attr->max_wr and srq_attr->max_sge are read the determine the
2941  * requested size of the SRQ, and set to the actual values allocated
2942  * on return.  If ib_create_srq() succeeds, then max_wr and max_sge
2943  * will always be at least as large as the requested values.
2944  */
2945 struct ib_srq *ib_create_srq(struct ib_pd *pd,
2946 			     struct ib_srq_init_attr *srq_init_attr);
2947 
2948 /**
2949  * ib_modify_srq - Modifies the attributes for the specified SRQ.
2950  * @srq: The SRQ to modify.
2951  * @srq_attr: On input, specifies the SRQ attributes to modify.  On output,
2952  *   the current values of selected SRQ attributes are returned.
2953  * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ
2954  *   are being modified.
2955  *
2956  * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or
2957  * IB_SRQ_LIMIT to set the SRQ's limit and request notification when
2958  * the number of receives queued drops below the limit.
2959  */
2960 int ib_modify_srq(struct ib_srq *srq,
2961 		  struct ib_srq_attr *srq_attr,
2962 		  enum ib_srq_attr_mask srq_attr_mask);
2963 
2964 /**
2965  * ib_query_srq - Returns the attribute list and current values for the
2966  *   specified SRQ.
2967  * @srq: The SRQ to query.
2968  * @srq_attr: The attributes of the specified SRQ.
2969  */
2970 int ib_query_srq(struct ib_srq *srq,
2971 		 struct ib_srq_attr *srq_attr);
2972 
2973 /**
2974  * ib_destroy_srq - Destroys the specified SRQ.
2975  * @srq: The SRQ to destroy.
2976  */
2977 int ib_destroy_srq(struct ib_srq *srq);
2978 
2979 /**
2980  * ib_post_srq_recv - Posts a list of work requests to the specified SRQ.
2981  * @srq: The SRQ to post the work request on.
2982  * @recv_wr: A list of work requests to post on the receive queue.
2983  * @bad_recv_wr: On an immediate failure, this parameter will reference
2984  *   the work request that failed to be posted on the QP.
2985  */
2986 static inline int ib_post_srq_recv(struct ib_srq *srq,
2987 				   struct ib_recv_wr *recv_wr,
2988 				   struct ib_recv_wr **bad_recv_wr)
2989 {
2990 	return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr);
2991 }
2992 
2993 /**
2994  * ib_create_qp - Creates a QP associated with the specified protection
2995  *   domain.
2996  * @pd: The protection domain associated with the QP.
2997  * @qp_init_attr: A list of initial attributes required to create the
2998  *   QP.  If QP creation succeeds, then the attributes are updated to
2999  *   the actual capabilities of the created QP.
3000  */
3001 struct ib_qp *ib_create_qp(struct ib_pd *pd,
3002 			   struct ib_qp_init_attr *qp_init_attr);
3003 
3004 /**
3005  * ib_modify_qp_with_udata - Modifies the attributes for the specified QP.
3006  * @qp: The QP to modify.
3007  * @attr: On input, specifies the QP attributes to modify.  On output,
3008  *   the current values of selected QP attributes are returned.
3009  * @attr_mask: A bit-mask used to specify which attributes of the QP
3010  *   are being modified.
3011  * @udata: pointer to user's input output buffer information
3012  *   are being modified.
3013  * It returns 0 on success and returns appropriate error code on error.
3014  */
3015 int ib_modify_qp_with_udata(struct ib_qp *qp,
3016 			    struct ib_qp_attr *attr,
3017 			    int attr_mask,
3018 			    struct ib_udata *udata);
3019 
3020 /**
3021  * ib_modify_qp - Modifies the attributes for the specified QP and then
3022  *   transitions the QP to the given state.
3023  * @qp: The QP to modify.
3024  * @qp_attr: On input, specifies the QP attributes to modify.  On output,
3025  *   the current values of selected QP attributes are returned.
3026  * @qp_attr_mask: A bit-mask used to specify which attributes of the QP
3027  *   are being modified.
3028  */
3029 int ib_modify_qp(struct ib_qp *qp,
3030 		 struct ib_qp_attr *qp_attr,
3031 		 int qp_attr_mask);
3032 
3033 /**
3034  * ib_query_qp - Returns the attribute list and current values for the
3035  *   specified QP.
3036  * @qp: The QP to query.
3037  * @qp_attr: The attributes of the specified QP.
3038  * @qp_attr_mask: A bit-mask used to select specific attributes to query.
3039  * @qp_init_attr: Additional attributes of the selected QP.
3040  *
3041  * The qp_attr_mask may be used to limit the query to gathering only the
3042  * selected attributes.
3043  */
3044 int ib_query_qp(struct ib_qp *qp,
3045 		struct ib_qp_attr *qp_attr,
3046 		int qp_attr_mask,
3047 		struct ib_qp_init_attr *qp_init_attr);
3048 
3049 /**
3050  * ib_destroy_qp - Destroys the specified QP.
3051  * @qp: The QP to destroy.
3052  */
3053 int ib_destroy_qp(struct ib_qp *qp);
3054 
3055 /**
3056  * ib_open_qp - Obtain a reference to an existing sharable QP.
3057  * @xrcd - XRC domain
3058  * @qp_open_attr: Attributes identifying the QP to open.
3059  *
3060  * Returns a reference to a sharable QP.
3061  */
3062 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd,
3063 			 struct ib_qp_open_attr *qp_open_attr);
3064 
3065 /**
3066  * ib_close_qp - Release an external reference to a QP.
3067  * @qp: The QP handle to release
3068  *
3069  * The opened QP handle is released by the caller.  The underlying
3070  * shared QP is not destroyed until all internal references are released.
3071  */
3072 int ib_close_qp(struct ib_qp *qp);
3073 
3074 /**
3075  * ib_post_send - Posts a list of work requests to the send queue of
3076  *   the specified QP.
3077  * @qp: The QP to post the work request on.
3078  * @send_wr: A list of work requests to post on the send queue.
3079  * @bad_send_wr: On an immediate failure, this parameter will reference
3080  *   the work request that failed to be posted on the QP.
3081  *
3082  * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate
3083  * error is returned, the QP state shall not be affected,
3084  * ib_post_send() will return an immediate error after queueing any
3085  * earlier work requests in the list.
3086  */
3087 static inline int ib_post_send(struct ib_qp *qp,
3088 			       struct ib_send_wr *send_wr,
3089 			       struct ib_send_wr **bad_send_wr)
3090 {
3091 	return qp->device->post_send(qp, send_wr, bad_send_wr);
3092 }
3093 
3094 /**
3095  * ib_post_recv - Posts a list of work requests to the receive queue of
3096  *   the specified QP.
3097  * @qp: The QP to post the work request on.
3098  * @recv_wr: A list of work requests to post on the receive queue.
3099  * @bad_recv_wr: On an immediate failure, this parameter will reference
3100  *   the work request that failed to be posted on the QP.
3101  */
3102 static inline int ib_post_recv(struct ib_qp *qp,
3103 			       struct ib_recv_wr *recv_wr,
3104 			       struct ib_recv_wr **bad_recv_wr)
3105 {
3106 	return qp->device->post_recv(qp, recv_wr, bad_recv_wr);
3107 }
3108 
3109 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private,
3110 		int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx);
3111 void ib_free_cq(struct ib_cq *cq);
3112 int ib_process_cq_direct(struct ib_cq *cq, int budget);
3113 
3114 /**
3115  * ib_create_cq - Creates a CQ on the specified device.
3116  * @device: The device on which to create the CQ.
3117  * @comp_handler: A user-specified callback that is invoked when a
3118  *   completion event occurs on the CQ.
3119  * @event_handler: A user-specified callback that is invoked when an
3120  *   asynchronous event not associated with a completion occurs on the CQ.
3121  * @cq_context: Context associated with the CQ returned to the user via
3122  *   the associated completion and event handlers.
3123  * @cq_attr: The attributes the CQ should be created upon.
3124  *
3125  * Users can examine the cq structure to determine the actual CQ size.
3126  */
3127 struct ib_cq *ib_create_cq(struct ib_device *device,
3128 			   ib_comp_handler comp_handler,
3129 			   void (*event_handler)(struct ib_event *, void *),
3130 			   void *cq_context,
3131 			   const struct ib_cq_init_attr *cq_attr);
3132 
3133 /**
3134  * ib_resize_cq - Modifies the capacity of the CQ.
3135  * @cq: The CQ to resize.
3136  * @cqe: The minimum size of the CQ.
3137  *
3138  * Users can examine the cq structure to determine the actual CQ size.
3139  */
3140 int ib_resize_cq(struct ib_cq *cq, int cqe);
3141 
3142 /**
3143  * ib_modify_cq - Modifies moderation params of the CQ
3144  * @cq: The CQ to modify.
3145  * @cq_count: number of CQEs that will trigger an event
3146  * @cq_period: max period of time in usec before triggering an event
3147  *
3148  */
3149 int ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
3150 
3151 /**
3152  * ib_destroy_cq - Destroys the specified CQ.
3153  * @cq: The CQ to destroy.
3154  */
3155 int ib_destroy_cq(struct ib_cq *cq);
3156 
3157 /**
3158  * ib_poll_cq - poll a CQ for completion(s)
3159  * @cq:the CQ being polled
3160  * @num_entries:maximum number of completions to return
3161  * @wc:array of at least @num_entries &struct ib_wc where completions
3162  *   will be returned
3163  *
3164  * Poll a CQ for (possibly multiple) completions.  If the return value
3165  * is < 0, an error occurred.  If the return value is >= 0, it is the
3166  * number of completions returned.  If the return value is
3167  * non-negative and < num_entries, then the CQ was emptied.
3168  */
3169 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries,
3170 			     struct ib_wc *wc)
3171 {
3172 	return cq->device->poll_cq(cq, num_entries, wc);
3173 }
3174 
3175 /**
3176  * ib_peek_cq - Returns the number of unreaped completions currently
3177  *   on the specified CQ.
3178  * @cq: The CQ to peek.
3179  * @wc_cnt: A minimum number of unreaped completions to check for.
3180  *
3181  * If the number of unreaped completions is greater than or equal to wc_cnt,
3182  * this function returns wc_cnt, otherwise, it returns the actual number of
3183  * unreaped completions.
3184  */
3185 int ib_peek_cq(struct ib_cq *cq, int wc_cnt);
3186 
3187 /**
3188  * ib_req_notify_cq - Request completion notification on a CQ.
3189  * @cq: The CQ to generate an event for.
3190  * @flags:
3191  *   Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP
3192  *   to request an event on the next solicited event or next work
3193  *   completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS
3194  *   may also be |ed in to request a hint about missed events, as
3195  *   described below.
3196  *
3197  * Return Value:
3198  *    < 0 means an error occurred while requesting notification
3199  *   == 0 means notification was requested successfully, and if
3200  *        IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events
3201  *        were missed and it is safe to wait for another event.  In
3202  *        this case is it guaranteed that any work completions added
3203  *        to the CQ since the last CQ poll will trigger a completion
3204  *        notification event.
3205  *    > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed
3206  *        in.  It means that the consumer must poll the CQ again to
3207  *        make sure it is empty to avoid missing an event because of a
3208  *        race between requesting notification and an entry being
3209  *        added to the CQ.  This return value means it is possible
3210  *        (but not guaranteed) that a work completion has been added
3211  *        to the CQ since the last poll without triggering a
3212  *        completion notification event.
3213  */
3214 static inline int ib_req_notify_cq(struct ib_cq *cq,
3215 				   enum ib_cq_notify_flags flags)
3216 {
3217 	return cq->device->req_notify_cq(cq, flags);
3218 }
3219 
3220 /**
3221  * ib_req_ncomp_notif - Request completion notification when there are
3222  *   at least the specified number of unreaped completions on the CQ.
3223  * @cq: The CQ to generate an event for.
3224  * @wc_cnt: The number of unreaped completions that should be on the
3225  *   CQ before an event is generated.
3226  */
3227 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt)
3228 {
3229 	return cq->device->req_ncomp_notif ?
3230 		cq->device->req_ncomp_notif(cq, wc_cnt) :
3231 		-ENOSYS;
3232 }
3233 
3234 /**
3235  * ib_dma_mapping_error - check a DMA addr for error
3236  * @dev: The device for which the dma_addr was created
3237  * @dma_addr: The DMA address to check
3238  */
3239 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
3240 {
3241 	return dma_mapping_error(dev->dma_device, dma_addr);
3242 }
3243 
3244 /**
3245  * ib_dma_map_single - Map a kernel virtual address to DMA address
3246  * @dev: The device for which the dma_addr is to be created
3247  * @cpu_addr: The kernel virtual address
3248  * @size: The size of the region in bytes
3249  * @direction: The direction of the DMA
3250  */
3251 static inline u64 ib_dma_map_single(struct ib_device *dev,
3252 				    void *cpu_addr, size_t size,
3253 				    enum dma_data_direction direction)
3254 {
3255 	return dma_map_single(dev->dma_device, cpu_addr, size, direction);
3256 }
3257 
3258 /**
3259  * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single()
3260  * @dev: The device for which the DMA address was created
3261  * @addr: The DMA address
3262  * @size: The size of the region in bytes
3263  * @direction: The direction of the DMA
3264  */
3265 static inline void ib_dma_unmap_single(struct ib_device *dev,
3266 				       u64 addr, size_t size,
3267 				       enum dma_data_direction direction)
3268 {
3269 	dma_unmap_single(dev->dma_device, addr, size, direction);
3270 }
3271 
3272 /**
3273  * ib_dma_map_page - Map a physical page to DMA address
3274  * @dev: The device for which the dma_addr is to be created
3275  * @page: The page to be mapped
3276  * @offset: The offset within the page
3277  * @size: The size of the region in bytes
3278  * @direction: The direction of the DMA
3279  */
3280 static inline u64 ib_dma_map_page(struct ib_device *dev,
3281 				  struct page *page,
3282 				  unsigned long offset,
3283 				  size_t size,
3284 					 enum dma_data_direction direction)
3285 {
3286 	return dma_map_page(dev->dma_device, page, offset, size, direction);
3287 }
3288 
3289 /**
3290  * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page()
3291  * @dev: The device for which the DMA address was created
3292  * @addr: The DMA address
3293  * @size: The size of the region in bytes
3294  * @direction: The direction of the DMA
3295  */
3296 static inline void ib_dma_unmap_page(struct ib_device *dev,
3297 				     u64 addr, size_t size,
3298 				     enum dma_data_direction direction)
3299 {
3300 	dma_unmap_page(dev->dma_device, addr, size, direction);
3301 }
3302 
3303 /**
3304  * ib_dma_map_sg - Map a scatter/gather list to DMA addresses
3305  * @dev: The device for which the DMA addresses are to be created
3306  * @sg: The array of scatter/gather entries
3307  * @nents: The number of scatter/gather entries
3308  * @direction: The direction of the DMA
3309  */
3310 static inline int ib_dma_map_sg(struct ib_device *dev,
3311 				struct scatterlist *sg, int nents,
3312 				enum dma_data_direction direction)
3313 {
3314 	return dma_map_sg(dev->dma_device, sg, nents, direction);
3315 }
3316 
3317 /**
3318  * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses
3319  * @dev: The device for which the DMA addresses were created
3320  * @sg: The array of scatter/gather entries
3321  * @nents: The number of scatter/gather entries
3322  * @direction: The direction of the DMA
3323  */
3324 static inline void ib_dma_unmap_sg(struct ib_device *dev,
3325 				   struct scatterlist *sg, int nents,
3326 				   enum dma_data_direction direction)
3327 {
3328 	dma_unmap_sg(dev->dma_device, sg, nents, direction);
3329 }
3330 
3331 static inline int ib_dma_map_sg_attrs(struct ib_device *dev,
3332 				      struct scatterlist *sg, int nents,
3333 				      enum dma_data_direction direction,
3334 				      unsigned long dma_attrs)
3335 {
3336 	return dma_map_sg_attrs(dev->dma_device, sg, nents, direction,
3337 				dma_attrs);
3338 }
3339 
3340 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev,
3341 					 struct scatterlist *sg, int nents,
3342 					 enum dma_data_direction direction,
3343 					 unsigned long dma_attrs)
3344 {
3345 	dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs);
3346 }
3347 /**
3348  * ib_sg_dma_address - Return the DMA address from a scatter/gather entry
3349  * @dev: The device for which the DMA addresses were created
3350  * @sg: The scatter/gather entry
3351  *
3352  * Note: this function is obsolete. To do: change all occurrences of
3353  * ib_sg_dma_address() into sg_dma_address().
3354  */
3355 static inline u64 ib_sg_dma_address(struct ib_device *dev,
3356 				    struct scatterlist *sg)
3357 {
3358 	return sg_dma_address(sg);
3359 }
3360 
3361 /**
3362  * ib_sg_dma_len - Return the DMA length from a scatter/gather entry
3363  * @dev: The device for which the DMA addresses were created
3364  * @sg: The scatter/gather entry
3365  *
3366  * Note: this function is obsolete. To do: change all occurrences of
3367  * ib_sg_dma_len() into sg_dma_len().
3368  */
3369 static inline unsigned int ib_sg_dma_len(struct ib_device *dev,
3370 					 struct scatterlist *sg)
3371 {
3372 	return sg_dma_len(sg);
3373 }
3374 
3375 /**
3376  * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU
3377  * @dev: The device for which the DMA address was created
3378  * @addr: The DMA address
3379  * @size: The size of the region in bytes
3380  * @dir: The direction of the DMA
3381  */
3382 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev,
3383 					      u64 addr,
3384 					      size_t size,
3385 					      enum dma_data_direction dir)
3386 {
3387 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
3388 }
3389 
3390 /**
3391  * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device
3392  * @dev: The device for which the DMA address was created
3393  * @addr: The DMA address
3394  * @size: The size of the region in bytes
3395  * @dir: The direction of the DMA
3396  */
3397 static inline void ib_dma_sync_single_for_device(struct ib_device *dev,
3398 						 u64 addr,
3399 						 size_t size,
3400 						 enum dma_data_direction dir)
3401 {
3402 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
3403 }
3404 
3405 /**
3406  * ib_dma_alloc_coherent - Allocate memory and map it for DMA
3407  * @dev: The device for which the DMA address is requested
3408  * @size: The size of the region to allocate in bytes
3409  * @dma_handle: A pointer for returning the DMA address of the region
3410  * @flag: memory allocator flags
3411  */
3412 static inline void *ib_dma_alloc_coherent(struct ib_device *dev,
3413 					   size_t size,
3414 					   dma_addr_t *dma_handle,
3415 					   gfp_t flag)
3416 {
3417 	return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag);
3418 }
3419 
3420 /**
3421  * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent()
3422  * @dev: The device for which the DMA addresses were allocated
3423  * @size: The size of the region
3424  * @cpu_addr: the address returned by ib_dma_alloc_coherent()
3425  * @dma_handle: the DMA address returned by ib_dma_alloc_coherent()
3426  */
3427 static inline void ib_dma_free_coherent(struct ib_device *dev,
3428 					size_t size, void *cpu_addr,
3429 					dma_addr_t dma_handle)
3430 {
3431 	dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle);
3432 }
3433 
3434 /**
3435  * ib_dereg_mr - Deregisters a memory region and removes it from the
3436  *   HCA translation table.
3437  * @mr: The memory region to deregister.
3438  *
3439  * This function can fail, if the memory region has memory windows bound to it.
3440  */
3441 int ib_dereg_mr(struct ib_mr *mr);
3442 
3443 struct ib_mr *ib_alloc_mr(struct ib_pd *pd,
3444 			  enum ib_mr_type mr_type,
3445 			  u32 max_num_sg);
3446 
3447 /**
3448  * ib_update_fast_reg_key - updates the key portion of the fast_reg MR
3449  *   R_Key and L_Key.
3450  * @mr - struct ib_mr pointer to be updated.
3451  * @newkey - new key to be used.
3452  */
3453 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey)
3454 {
3455 	mr->lkey = (mr->lkey & 0xffffff00) | newkey;
3456 	mr->rkey = (mr->rkey & 0xffffff00) | newkey;
3457 }
3458 
3459 /**
3460  * ib_inc_rkey - increments the key portion of the given rkey. Can be used
3461  * for calculating a new rkey for type 2 memory windows.
3462  * @rkey - the rkey to increment.
3463  */
3464 static inline u32 ib_inc_rkey(u32 rkey)
3465 {
3466 	const u32 mask = 0x000000ff;
3467 	return ((rkey + 1) & mask) | (rkey & ~mask);
3468 }
3469 
3470 /**
3471  * ib_alloc_fmr - Allocates a unmapped fast memory region.
3472  * @pd: The protection domain associated with the unmapped region.
3473  * @mr_access_flags: Specifies the memory access rights.
3474  * @fmr_attr: Attributes of the unmapped region.
3475  *
3476  * A fast memory region must be mapped before it can be used as part of
3477  * a work request.
3478  */
3479 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd,
3480 			    int mr_access_flags,
3481 			    struct ib_fmr_attr *fmr_attr);
3482 
3483 /**
3484  * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region.
3485  * @fmr: The fast memory region to associate with the pages.
3486  * @page_list: An array of physical pages to map to the fast memory region.
3487  * @list_len: The number of pages in page_list.
3488  * @iova: The I/O virtual address to use with the mapped region.
3489  */
3490 static inline int ib_map_phys_fmr(struct ib_fmr *fmr,
3491 				  u64 *page_list, int list_len,
3492 				  u64 iova)
3493 {
3494 	return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova);
3495 }
3496 
3497 /**
3498  * ib_unmap_fmr - Removes the mapping from a list of fast memory regions.
3499  * @fmr_list: A linked list of fast memory regions to unmap.
3500  */
3501 int ib_unmap_fmr(struct list_head *fmr_list);
3502 
3503 /**
3504  * ib_dealloc_fmr - Deallocates a fast memory region.
3505  * @fmr: The fast memory region to deallocate.
3506  */
3507 int ib_dealloc_fmr(struct ib_fmr *fmr);
3508 
3509 /**
3510  * ib_attach_mcast - Attaches the specified QP to a multicast group.
3511  * @qp: QP to attach to the multicast group.  The QP must be type
3512  *   IB_QPT_UD.
3513  * @gid: Multicast group GID.
3514  * @lid: Multicast group LID in host byte order.
3515  *
3516  * In order to send and receive multicast packets, subnet
3517  * administration must have created the multicast group and configured
3518  * the fabric appropriately.  The port associated with the specified
3519  * QP must also be a member of the multicast group.
3520  */
3521 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3522 
3523 /**
3524  * ib_detach_mcast - Detaches the specified QP from a multicast group.
3525  * @qp: QP to detach from the multicast group.
3526  * @gid: Multicast group GID.
3527  * @lid: Multicast group LID in host byte order.
3528  */
3529 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid);
3530 
3531 /**
3532  * ib_alloc_xrcd - Allocates an XRC domain.
3533  * @device: The device on which to allocate the XRC domain.
3534  */
3535 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device);
3536 
3537 /**
3538  * ib_dealloc_xrcd - Deallocates an XRC domain.
3539  * @xrcd: The XRC domain to deallocate.
3540  */
3541 int ib_dealloc_xrcd(struct ib_xrcd *xrcd);
3542 
3543 struct ib_flow *ib_create_flow(struct ib_qp *qp,
3544 			       struct ib_flow_attr *flow_attr, int domain);
3545 int ib_destroy_flow(struct ib_flow *flow_id);
3546 
3547 static inline int ib_check_mr_access(int flags)
3548 {
3549 	/*
3550 	 * Local write permission is required if remote write or
3551 	 * remote atomic permission is also requested.
3552 	 */
3553 	if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) &&
3554 	    !(flags & IB_ACCESS_LOCAL_WRITE))
3555 		return -EINVAL;
3556 
3557 	return 0;
3558 }
3559 
3560 /**
3561  * ib_check_mr_status: lightweight check of MR status.
3562  *     This routine may provide status checks on a selected
3563  *     ib_mr. first use is for signature status check.
3564  *
3565  * @mr: A memory region.
3566  * @check_mask: Bitmask of which checks to perform from
3567  *     ib_mr_status_check enumeration.
3568  * @mr_status: The container of relevant status checks.
3569  *     failed checks will be indicated in the status bitmask
3570  *     and the relevant info shall be in the error item.
3571  */
3572 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask,
3573 		       struct ib_mr_status *mr_status);
3574 
3575 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port,
3576 					    u16 pkey, const union ib_gid *gid,
3577 					    const struct sockaddr *addr);
3578 struct ib_wq *ib_create_wq(struct ib_pd *pd,
3579 			   struct ib_wq_init_attr *init_attr);
3580 int ib_destroy_wq(struct ib_wq *wq);
3581 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr,
3582 		 u32 wq_attr_mask);
3583 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device,
3584 						 struct ib_rwq_ind_table_init_attr*
3585 						 wq_ind_table_init_attr);
3586 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table);
3587 
3588 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3589 		 unsigned int *sg_offset, unsigned int page_size);
3590 
3591 static inline int
3592 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents,
3593 		  unsigned int *sg_offset, unsigned int page_size)
3594 {
3595 	int n;
3596 
3597 	n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size);
3598 	mr->iova = 0;
3599 
3600 	return n;
3601 }
3602 
3603 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents,
3604 		unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64));
3605 
3606 void ib_drain_rq(struct ib_qp *qp);
3607 void ib_drain_sq(struct ib_qp *qp);
3608 void ib_drain_qp(struct ib_qp *qp);
3609 
3610 int ib_resolve_eth_dmac(struct ib_device *device,
3611 			struct rdma_ah_attr *ah_attr);
3612 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width);
3613 
3614 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr)
3615 {
3616 	if (attr->type == RDMA_AH_ATTR_TYPE_ROCE)
3617 		return attr->roce.dmac;
3618 	return NULL;
3619 }
3620 
3621 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid)
3622 {
3623 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3624 		attr->ib.dlid = (u16)dlid;
3625 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3626 		attr->opa.dlid = dlid;
3627 }
3628 
3629 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr)
3630 {
3631 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3632 		return attr->ib.dlid;
3633 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3634 		return attr->opa.dlid;
3635 	return 0;
3636 }
3637 
3638 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl)
3639 {
3640 	attr->sl = sl;
3641 }
3642 
3643 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr)
3644 {
3645 	return attr->sl;
3646 }
3647 
3648 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr,
3649 					 u8 src_path_bits)
3650 {
3651 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3652 		attr->ib.src_path_bits = src_path_bits;
3653 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3654 		attr->opa.src_path_bits = src_path_bits;
3655 }
3656 
3657 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr)
3658 {
3659 	if (attr->type == RDMA_AH_ATTR_TYPE_IB)
3660 		return attr->ib.src_path_bits;
3661 	else if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3662 		return attr->opa.src_path_bits;
3663 	return 0;
3664 }
3665 
3666 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr,
3667 					bool make_grd)
3668 {
3669 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3670 		attr->opa.make_grd = make_grd;
3671 }
3672 
3673 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr)
3674 {
3675 	if (attr->type == RDMA_AH_ATTR_TYPE_OPA)
3676 		return attr->opa.make_grd;
3677 	return false;
3678 }
3679 
3680 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num)
3681 {
3682 	attr->port_num = port_num;
3683 }
3684 
3685 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr)
3686 {
3687 	return attr->port_num;
3688 }
3689 
3690 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr,
3691 					   u8 static_rate)
3692 {
3693 	attr->static_rate = static_rate;
3694 }
3695 
3696 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr)
3697 {
3698 	return attr->static_rate;
3699 }
3700 
3701 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr,
3702 					enum ib_ah_flags flag)
3703 {
3704 	attr->ah_flags = flag;
3705 }
3706 
3707 static inline enum ib_ah_flags
3708 		rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr)
3709 {
3710 	return attr->ah_flags;
3711 }
3712 
3713 static inline const struct ib_global_route
3714 		*rdma_ah_read_grh(const struct rdma_ah_attr *attr)
3715 {
3716 	return &attr->grh;
3717 }
3718 
3719 /*To retrieve and modify the grh */
3720 static inline struct ib_global_route
3721 		*rdma_ah_retrieve_grh(struct rdma_ah_attr *attr)
3722 {
3723 	return &attr->grh;
3724 }
3725 
3726 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid)
3727 {
3728 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3729 
3730 	memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid));
3731 }
3732 
3733 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr,
3734 					     __be64 prefix)
3735 {
3736 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3737 
3738 	grh->dgid.global.subnet_prefix = prefix;
3739 }
3740 
3741 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr,
3742 					    __be64 if_id)
3743 {
3744 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3745 
3746 	grh->dgid.global.interface_id = if_id;
3747 }
3748 
3749 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr,
3750 				   union ib_gid *dgid, u32 flow_label,
3751 				   u8 sgid_index, u8 hop_limit,
3752 				   u8 traffic_class)
3753 {
3754 	struct ib_global_route *grh = rdma_ah_retrieve_grh(attr);
3755 
3756 	attr->ah_flags = IB_AH_GRH;
3757 	if (dgid)
3758 		grh->dgid = *dgid;
3759 	grh->flow_label = flow_label;
3760 	grh->sgid_index = sgid_index;
3761 	grh->hop_limit = hop_limit;
3762 	grh->traffic_class = traffic_class;
3763 }
3764 
3765 /*Get AH type */
3766 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev,
3767 						       u32 port_num)
3768 {
3769 	if ((rdma_protocol_roce(dev, port_num)) ||
3770 	    (rdma_protocol_iwarp(dev, port_num)))
3771 		return RDMA_AH_ATTR_TYPE_ROCE;
3772 	else if ((rdma_protocol_ib(dev, port_num)) &&
3773 		 (rdma_cap_opa_ah(dev, port_num)))
3774 		return RDMA_AH_ATTR_TYPE_OPA;
3775 	else
3776 		return RDMA_AH_ATTR_TYPE_IB;
3777 }
3778 
3779 /**
3780  * ib_lid_cpu16 - Return lid in 16bit CPU encoding.
3781  *     In the current implementation the only way to get
3782  *     get the 32bit lid is from other sources for OPA.
3783  *     For IB, lids will always be 16bits so cast the
3784  *     value accordingly.
3785  *
3786  * @lid: A 32bit LID
3787  */
3788 static inline u16 ib_lid_cpu16(u32 lid)
3789 {
3790 	WARN_ON_ONCE(lid & 0xFFFF0000);
3791 	return (u16)lid;
3792 }
3793 
3794 /**
3795  * ib_lid_be16 - Return lid in 16bit BE encoding.
3796  *
3797  * @lid: A 32bit LID
3798  */
3799 static inline __be16 ib_lid_be16(u32 lid)
3800 {
3801 	WARN_ON_ONCE(lid & 0xFFFF0000);
3802 	return cpu_to_be16((u16)lid);
3803 }
3804 
3805 /**
3806  * ib_get_vector_affinity - Get the affinity mappings of a given completion
3807  *   vector
3808  * @device:         the rdma device
3809  * @comp_vector:    index of completion vector
3810  *
3811  * Returns NULL on failure, otherwise a corresponding cpu map of the
3812  * completion vector (returns all-cpus map if the device driver doesn't
3813  * implement get_vector_affinity).
3814  */
3815 static inline const struct cpumask *
3816 ib_get_vector_affinity(struct ib_device *device, int comp_vector)
3817 {
3818 	if (comp_vector < 0 || comp_vector >= device->num_comp_vectors ||
3819 	    !device->get_vector_affinity)
3820 		return NULL;
3821 
3822 	return device->get_vector_affinity(device, comp_vector);
3823 
3824 }
3825 
3826 #endif /* IB_VERBS_H */
3827